/* Portable arc4random.c based on arc4random.c from OpenBSD. * Portable version by Chris Davis, adapted for Libevent by Nick Mathewson * Copyright (c) 2010 Chris Davis, Niels Provos, and Nick Mathewson * Copyright (c) 2010-2012 Niels Provos and Nick Mathewson * * Note that in Libevent, this file isn't compiled directly. Instead, * it's included from evutil_rand.c */ /* * Copyright (c) 1996, David Mazieres * Copyright (c) 2008, Damien Miller * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ /* * Arc4 random number generator for OpenBSD. * * This code is derived from section 17.1 of Applied Cryptography, * second edition, which describes a stream cipher allegedly * compatible with RSA Labs "RC4" cipher (the actual description of * which is a trade secret). The same algorithm is used as a stream * cipher called "arcfour" in Tatu Ylonen's ssh package. * * Here the stream cipher has been modified always to include the time * when initializing the state. That makes it impossible to * regenerate the same random sequence twice, so this can't be used * for encryption, but will generate good random numbers. * * RC4 is a registered trademark of RSA Laboratories. */ #ifndef ARC4RANDOM_EXPORT #define ARC4RANDOM_EXPORT #endif #ifndef ARC4RANDOM_UINT32 #define ARC4RANDOM_UINT32 uint32_t #endif #ifndef ARC4RANDOM_NO_INCLUDES #include "evconfig-private.h" #ifdef _WIN32 #include #include #include #else #include #include #include #include #ifdef EVENT__HAVE_SYS_SYSCTL_H #include #endif #ifdef EVENT__HAVE_SYS_RANDOM_H #include #endif #endif #include #include #include #endif /* Add platform entropy 32 bytes (256 bits) at a time. */ #define ADD_ENTROPY 32 /* Re-seed from the platform RNG after generating this many bytes. */ #define BYTES_BEFORE_RESEED 1600000 struct arc4_stream { unsigned char i; unsigned char j; unsigned char s[256]; }; #ifdef _WIN32 #define getpid _getpid #define pid_t int #endif static int rs_initialized; static struct arc4_stream rs; static pid_t arc4_stir_pid; static int arc4_count; static inline unsigned char arc4_getbyte(void); static inline void arc4_init(void) { int n; for (n = 0; n < 256; n++) rs.s[n] = n; rs.i = 0; rs.j = 0; } static inline void arc4_addrandom(const unsigned char *dat, int datlen) { int n; unsigned char si; rs.i--; for (n = 0; n < 256; n++) { rs.i = (rs.i + 1); si = rs.s[rs.i]; rs.j = (rs.j + si + dat[n % datlen]); rs.s[rs.i] = rs.s[rs.j]; rs.s[rs.j] = si; } rs.j = rs.i; } #ifndef _WIN32 static ssize_t read_all(int fd, unsigned char *buf, size_t count) { size_t numread = 0; ssize_t result; while (numread < count) { result = read(fd, buf+numread, count-numread); if (result<0) return -1; else if (result == 0) break; numread += result; } return (ssize_t)numread; } #endif #ifdef _WIN32 #define TRY_SEED_WIN32 static int arc4_seed_win32(void) { /* This is adapted from Tor's crypto_seed_rng() */ static int provider_set = 0; static HCRYPTPROV provider; unsigned char buf[ADD_ENTROPY]; if (!provider_set) { if (!CryptAcquireContext(&provider, NULL, NULL, PROV_RSA_FULL, CRYPT_VERIFYCONTEXT)) { if (GetLastError() != (DWORD)NTE_BAD_KEYSET) return -1; } provider_set = 1; } if (!CryptGenRandom(provider, sizeof(buf), buf)) return -1; arc4_addrandom(buf, sizeof(buf)); evutil_memclear_(buf, sizeof(buf)); return 0; } #endif #if defined(EVENT__HAVE_GETRANDOM) #define TRY_SEED_GETRANDOM static int arc4_seed_getrandom(void) { unsigned char buf[ADD_ENTROPY]; size_t len, n; unsigned i; int any_set; memset(buf, 0, sizeof(buf)); for (len = 0; len < sizeof(buf); len += n) { n = sizeof(buf) - len; if (0 == getrandom(&buf[len], n, 0)) return -1; } /* make sure that the buffer actually got set. */ for (i=0,any_set=0; i sizeof(buf)) n = len - sizeof(buf); if (sysctl(mib, 2, &buf[len], &n, NULL, 0) == -1) return -1; } } /* make sure that the buffer actually got set. */ for (i=any_set=0; i 0xffffffffUL) min = 0x100000000UL % upper_bound; #else /* Calculate (2**32 % upper_bound) avoiding 64-bit math */ if (upper_bound > 0x80000000) min = 1 + ~upper_bound; /* 2**32 - upper_bound */ else { /* (2**32 - (x * 2)) % x == 2**32 % x when x <= 2**31 */ min = ((0xffffffff - (upper_bound * 2)) + 1) % upper_bound; } #endif /* * This could theoretically loop forever but each retry has * p > 0.5 (worst case, usually far better) of selecting a * number inside the range we need, so it should rarely need * to re-roll. */ for (;;) { r = arc4random(); if (r >= min) break; } return r % upper_bound; } #endif /* * Copyright (c) 2002-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #include #include #endif #ifdef EVENT__HAVE_VASPRINTF /* If we have vasprintf, we need to define _GNU_SOURCE before we include * stdio.h. This comes from evconfig-private.h. */ #endif #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #ifdef EVENT__HAVE_SYS_UIO_H #include #endif #ifdef EVENT__HAVE_SYS_IOCTL_H #include #endif #ifdef EVENT__HAVE_SYS_MMAN_H #include #endif #ifdef EVENT__HAVE_SYS_SENDFILE_H #include #endif #ifdef EVENT__HAVE_SYS_STAT_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #include #include "event2/event.h" #include "event2/buffer.h" #include "event2/buffer_compat.h" #include "event2/bufferevent.h" #include "event2/bufferevent_compat.h" #include "event2/bufferevent_struct.h" #include "event2/thread.h" #include "log-internal.h" #include "mm-internal.h" #include "util-internal.h" #include "evthread-internal.h" #include "evbuffer-internal.h" #include "bufferevent-internal.h" #include "event-internal.h" /* some systems do not have MAP_FAILED */ #ifndef MAP_FAILED #define MAP_FAILED ((void *)-1) #endif /* send file support */ #if defined(EVENT__HAVE_SYS_SENDFILE_H) && defined(EVENT__HAVE_SENDFILE) && defined(__linux__) #define USE_SENDFILE 1 #define SENDFILE_IS_LINUX 1 #elif defined(EVENT__HAVE_SENDFILE) && defined(__FreeBSD__) #define USE_SENDFILE 1 #define SENDFILE_IS_FREEBSD 1 #elif defined(EVENT__HAVE_SENDFILE) && defined(__APPLE__) #define USE_SENDFILE 1 #define SENDFILE_IS_MACOSX 1 #elif defined(EVENT__HAVE_SENDFILE) && defined(__sun__) && defined(__svr4__) #define USE_SENDFILE 1 #define SENDFILE_IS_SOLARIS 1 #endif /* Mask of user-selectable callback flags. */ #define EVBUFFER_CB_USER_FLAGS 0xffff /* Mask of all internal-use-only flags. */ #define EVBUFFER_CB_INTERNAL_FLAGS 0xffff0000 /* Flag set if the callback is using the cb_obsolete function pointer */ #define EVBUFFER_CB_OBSOLETE 0x00040000 /* evbuffer_chain support */ #define CHAIN_SPACE_PTR(ch) ((ch)->buffer + (ch)->misalign + (ch)->off) #define CHAIN_SPACE_LEN(ch) ((ch)->flags & EVBUFFER_IMMUTABLE ? \ 0 : (ch)->buffer_len - ((ch)->misalign + (ch)->off)) #define CHAIN_PINNED(ch) (((ch)->flags & EVBUFFER_MEM_PINNED_ANY) != 0) #define CHAIN_PINNED_R(ch) (((ch)->flags & EVBUFFER_MEM_PINNED_R) != 0) /* evbuffer_ptr support */ #define PTR_NOT_FOUND(ptr) do { \ (ptr)->pos = -1; \ (ptr)->internal_.chain = NULL; \ (ptr)->internal_.pos_in_chain = 0; \ } while (0) static void evbuffer_chain_align(struct evbuffer_chain *chain); static int evbuffer_chain_should_realign(struct evbuffer_chain *chain, size_t datalen); static void evbuffer_deferred_callback(struct event_callback *cb, void *arg); static int evbuffer_ptr_memcmp(const struct evbuffer *buf, const struct evbuffer_ptr *pos, const char *mem, size_t len); static struct evbuffer_chain *evbuffer_expand_singlechain(struct evbuffer *buf, size_t datlen); static int evbuffer_ptr_subtract(struct evbuffer *buf, struct evbuffer_ptr *pos, size_t howfar); static int evbuffer_file_segment_materialize(struct evbuffer_file_segment *seg); static inline void evbuffer_chain_incref(struct evbuffer_chain *chain); static struct evbuffer_chain * evbuffer_chain_new(size_t size) { struct evbuffer_chain *chain; size_t to_alloc; if (size > EVBUFFER_CHAIN_MAX - EVBUFFER_CHAIN_SIZE) return (NULL); size += EVBUFFER_CHAIN_SIZE; /* get the next largest memory that can hold the buffer */ if (size < EVBUFFER_CHAIN_MAX / 2) { to_alloc = MIN_BUFFER_SIZE; while (to_alloc < size) { to_alloc <<= 1; } } else { to_alloc = size; } /* we get everything in one chunk */ if ((chain = mm_malloc(to_alloc)) == NULL) return (NULL); memset(chain, 0, EVBUFFER_CHAIN_SIZE); chain->buffer_len = to_alloc - EVBUFFER_CHAIN_SIZE; /* this way we can manipulate the buffer to different addresses, * which is required for mmap for example. */ chain->buffer = EVBUFFER_CHAIN_EXTRA(unsigned char, chain); chain->refcnt = 1; return (chain); } static inline void evbuffer_chain_free(struct evbuffer_chain *chain) { EVUTIL_ASSERT(chain->refcnt > 0); if (--chain->refcnt > 0) { /* chain is still referenced by other chains */ return; } if (CHAIN_PINNED(chain)) { /* will get freed once no longer dangling */ chain->refcnt++; chain->flags |= EVBUFFER_DANGLING; return; } /* safe to release chain, it's either a referencing * chain or all references to it have been freed */ if (chain->flags & EVBUFFER_REFERENCE) { struct evbuffer_chain_reference *info = EVBUFFER_CHAIN_EXTRA( struct evbuffer_chain_reference, chain); if (info->cleanupfn) (*info->cleanupfn)(chain->buffer, chain->buffer_len, info->extra); } if (chain->flags & EVBUFFER_FILESEGMENT) { struct evbuffer_chain_file_segment *info = EVBUFFER_CHAIN_EXTRA( struct evbuffer_chain_file_segment, chain); if (info->segment) { #ifdef _WIN32 if (info->segment->is_mapping) UnmapViewOfFile(chain->buffer); #endif evbuffer_file_segment_free(info->segment); } } if (chain->flags & EVBUFFER_MULTICAST) { struct evbuffer_multicast_parent *info = EVBUFFER_CHAIN_EXTRA( struct evbuffer_multicast_parent, chain); /* referencing chain is being freed, decrease * refcounts of source chain and associated * evbuffer (which get freed once both reach * zero) */ EVUTIL_ASSERT(info->source != NULL); EVUTIL_ASSERT(info->parent != NULL); EVBUFFER_LOCK(info->source); evbuffer_chain_free(info->parent); evbuffer_decref_and_unlock_(info->source); } mm_free(chain); } static void evbuffer_free_all_chains(struct evbuffer_chain *chain) { struct evbuffer_chain *next; for (; chain; chain = next) { next = chain->next; evbuffer_chain_free(chain); } } #ifndef NDEBUG static int evbuffer_chains_all_empty(struct evbuffer_chain *chain) { for (; chain; chain = chain->next) { if (chain->off) return 0; } return 1; } #else /* The definition is needed for EVUTIL_ASSERT, which uses sizeof to avoid "unused variable" warnings. */ static inline int evbuffer_chains_all_empty(struct evbuffer_chain *chain) { return 1; } #endif /* Free all trailing chains in 'buf' that are neither pinned nor empty, prior * to replacing them all with a new chain. Return a pointer to the place * where the new chain will go. * * Internal; requires lock. The caller must fix up buf->last and buf->first * as needed; they might have been freed. */ static struct evbuffer_chain ** evbuffer_free_trailing_empty_chains(struct evbuffer *buf) { struct evbuffer_chain **ch = buf->last_with_datap; /* Find the first victim chain. It might be *last_with_datap */ while ((*ch) && ((*ch)->off != 0 || CHAIN_PINNED(*ch))) ch = &(*ch)->next; if (*ch) { EVUTIL_ASSERT(evbuffer_chains_all_empty(*ch)); evbuffer_free_all_chains(*ch); *ch = NULL; } return ch; } /* Add a single chain 'chain' to the end of 'buf', freeing trailing empty * chains as necessary. Requires lock. Does not schedule callbacks. */ static void evbuffer_chain_insert(struct evbuffer *buf, struct evbuffer_chain *chain) { ASSERT_EVBUFFER_LOCKED(buf); if (*buf->last_with_datap == NULL) { /* There are no chains data on the buffer at all. */ EVUTIL_ASSERT(buf->last_with_datap == &buf->first); EVUTIL_ASSERT(buf->first == NULL); buf->first = buf->last = chain; } else { struct evbuffer_chain **chp; chp = evbuffer_free_trailing_empty_chains(buf); *chp = chain; if (chain->off) buf->last_with_datap = chp; buf->last = chain; } buf->total_len += chain->off; } static inline struct evbuffer_chain * evbuffer_chain_insert_new(struct evbuffer *buf, size_t datlen) { struct evbuffer_chain *chain; if ((chain = evbuffer_chain_new(datlen)) == NULL) return NULL; evbuffer_chain_insert(buf, chain); return chain; } void evbuffer_chain_pin_(struct evbuffer_chain *chain, unsigned flag) { EVUTIL_ASSERT((chain->flags & flag) == 0); chain->flags |= flag; } void evbuffer_chain_unpin_(struct evbuffer_chain *chain, unsigned flag) { EVUTIL_ASSERT((chain->flags & flag) != 0); chain->flags &= ~flag; if (chain->flags & EVBUFFER_DANGLING) evbuffer_chain_free(chain); } static inline void evbuffer_chain_incref(struct evbuffer_chain *chain) { ++chain->refcnt; } struct evbuffer * evbuffer_new(void) { struct evbuffer *buffer; buffer = mm_calloc(1, sizeof(struct evbuffer)); if (buffer == NULL) return (NULL); LIST_INIT(&buffer->callbacks); buffer->refcnt = 1; buffer->last_with_datap = &buffer->first; return (buffer); } int evbuffer_set_flags(struct evbuffer *buf, ev_uint64_t flags) { EVBUFFER_LOCK(buf); buf->flags |= (ev_uint32_t)flags; EVBUFFER_UNLOCK(buf); return 0; } int evbuffer_clear_flags(struct evbuffer *buf, ev_uint64_t flags) { EVBUFFER_LOCK(buf); buf->flags &= ~(ev_uint32_t)flags; EVBUFFER_UNLOCK(buf); return 0; } void evbuffer_incref_(struct evbuffer *buf) { EVBUFFER_LOCK(buf); ++buf->refcnt; EVBUFFER_UNLOCK(buf); } void evbuffer_incref_and_lock_(struct evbuffer *buf) { EVBUFFER_LOCK(buf); ++buf->refcnt; } int evbuffer_defer_callbacks(struct evbuffer *buffer, struct event_base *base) { EVBUFFER_LOCK(buffer); buffer->cb_queue = base; buffer->deferred_cbs = 1; event_deferred_cb_init_(&buffer->deferred, event_base_get_npriorities(base) / 2, evbuffer_deferred_callback, buffer); EVBUFFER_UNLOCK(buffer); return 0; } int evbuffer_enable_locking(struct evbuffer *buf, void *lock) { #ifdef EVENT__DISABLE_THREAD_SUPPORT return -1; #else if (buf->lock) return -1; if (!lock) { EVTHREAD_ALLOC_LOCK(lock, EVTHREAD_LOCKTYPE_RECURSIVE); if (!lock) return -1; buf->lock = lock; buf->own_lock = 1; } else { buf->lock = lock; buf->own_lock = 0; } return 0; #endif } void evbuffer_set_parent_(struct evbuffer *buf, struct bufferevent *bev) { EVBUFFER_LOCK(buf); buf->parent = bev; EVBUFFER_UNLOCK(buf); } static void evbuffer_run_callbacks(struct evbuffer *buffer, int running_deferred) { struct evbuffer_cb_entry *cbent, *next; struct evbuffer_cb_info info; size_t new_size; ev_uint32_t mask, masked_val; int clear = 1; if (running_deferred) { mask = EVBUFFER_CB_NODEFER|EVBUFFER_CB_ENABLED; masked_val = EVBUFFER_CB_ENABLED; } else if (buffer->deferred_cbs) { mask = EVBUFFER_CB_NODEFER|EVBUFFER_CB_ENABLED; masked_val = EVBUFFER_CB_NODEFER|EVBUFFER_CB_ENABLED; /* Don't zero-out n_add/n_del, since the deferred callbacks will want to see them. */ clear = 0; } else { mask = EVBUFFER_CB_ENABLED; masked_val = EVBUFFER_CB_ENABLED; } ASSERT_EVBUFFER_LOCKED(buffer); if (LIST_EMPTY(&buffer->callbacks)) { buffer->n_add_for_cb = buffer->n_del_for_cb = 0; return; } if (buffer->n_add_for_cb == 0 && buffer->n_del_for_cb == 0) return; new_size = buffer->total_len; info.orig_size = new_size + buffer->n_del_for_cb - buffer->n_add_for_cb; info.n_added = buffer->n_add_for_cb; info.n_deleted = buffer->n_del_for_cb; if (clear) { buffer->n_add_for_cb = 0; buffer->n_del_for_cb = 0; } for (cbent = LIST_FIRST(&buffer->callbacks); cbent != LIST_END(&buffer->callbacks); cbent = next) { /* Get the 'next' pointer now in case this callback decides * to remove itself or something. */ next = LIST_NEXT(cbent, next); if ((cbent->flags & mask) != masked_val) continue; if ((cbent->flags & EVBUFFER_CB_OBSOLETE)) cbent->cb.cb_obsolete(buffer, info.orig_size, new_size, cbent->cbarg); else cbent->cb.cb_func(buffer, &info, cbent->cbarg); } } void evbuffer_invoke_callbacks_(struct evbuffer *buffer) { if (LIST_EMPTY(&buffer->callbacks)) { buffer->n_add_for_cb = buffer->n_del_for_cb = 0; return; } if (buffer->deferred_cbs) { if (event_deferred_cb_schedule_(buffer->cb_queue, &buffer->deferred)) { evbuffer_incref_and_lock_(buffer); if (buffer->parent) bufferevent_incref_(buffer->parent); EVBUFFER_UNLOCK(buffer); } } evbuffer_run_callbacks(buffer, 0); } static void evbuffer_deferred_callback(struct event_callback *cb, void *arg) { struct bufferevent *parent = NULL; struct evbuffer *buffer = arg; /* XXXX It would be better to run these callbacks without holding the * lock */ EVBUFFER_LOCK(buffer); parent = buffer->parent; evbuffer_run_callbacks(buffer, 1); evbuffer_decref_and_unlock_(buffer); if (parent) bufferevent_decref_(parent); } static void evbuffer_remove_all_callbacks(struct evbuffer *buffer) { struct evbuffer_cb_entry *cbent; while ((cbent = LIST_FIRST(&buffer->callbacks))) { LIST_REMOVE(cbent, next); mm_free(cbent); } } void evbuffer_decref_and_unlock_(struct evbuffer *buffer) { struct evbuffer_chain *chain, *next; ASSERT_EVBUFFER_LOCKED(buffer); EVUTIL_ASSERT(buffer->refcnt > 0); if (--buffer->refcnt > 0) { EVBUFFER_UNLOCK(buffer); return; } for (chain = buffer->first; chain != NULL; chain = next) { next = chain->next; evbuffer_chain_free(chain); } evbuffer_remove_all_callbacks(buffer); if (buffer->deferred_cbs) event_deferred_cb_cancel_(buffer->cb_queue, &buffer->deferred); EVBUFFER_UNLOCK(buffer); if (buffer->own_lock) EVTHREAD_FREE_LOCK(buffer->lock, EVTHREAD_LOCKTYPE_RECURSIVE); mm_free(buffer); } void evbuffer_free(struct evbuffer *buffer) { EVBUFFER_LOCK(buffer); evbuffer_decref_and_unlock_(buffer); } void evbuffer_lock(struct evbuffer *buf) { EVBUFFER_LOCK(buf); } void evbuffer_unlock(struct evbuffer *buf) { EVBUFFER_UNLOCK(buf); } size_t evbuffer_get_length(const struct evbuffer *buffer) { size_t result; EVBUFFER_LOCK(buffer); result = (buffer->total_len); EVBUFFER_UNLOCK(buffer); return result; } size_t evbuffer_get_contiguous_space(const struct evbuffer *buf) { struct evbuffer_chain *chain; size_t result; EVBUFFER_LOCK(buf); chain = buf->first; result = (chain != NULL ? chain->off : 0); EVBUFFER_UNLOCK(buf); return result; } size_t evbuffer_add_iovec(struct evbuffer * buf, struct evbuffer_iovec * vec, int n_vec) { int n; size_t res; size_t to_alloc; EVBUFFER_LOCK(buf); res = to_alloc = 0; for (n = 0; n < n_vec; n++) { to_alloc += vec[n].iov_len; } if (evbuffer_expand_fast_(buf, to_alloc, 2) < 0) { goto done; } for (n = 0; n < n_vec; n++) { /* XXX each 'add' call here does a bunch of setup that's * obviated by evbuffer_expand_fast_, and some cleanup that we * would like to do only once. Instead we should just extract * the part of the code that's needed. */ if (evbuffer_add(buf, vec[n].iov_base, vec[n].iov_len) < 0) { goto done; } res += vec[n].iov_len; } done: EVBUFFER_UNLOCK(buf); return res; } int evbuffer_reserve_space(struct evbuffer *buf, ev_ssize_t size, struct evbuffer_iovec *vec, int n_vecs) { struct evbuffer_chain *chain, **chainp; int n = -1; EVBUFFER_LOCK(buf); if (buf->freeze_end) goto done; if (n_vecs < 1) goto done; if (n_vecs == 1) { if ((chain = evbuffer_expand_singlechain(buf, size)) == NULL) goto done; vec[0].iov_base = (void *)CHAIN_SPACE_PTR(chain); vec[0].iov_len = (size_t)CHAIN_SPACE_LEN(chain); EVUTIL_ASSERT(size<0 || (size_t)vec[0].iov_len >= (size_t)size); n = 1; } else { if (evbuffer_expand_fast_(buf, size, n_vecs)<0) goto done; n = evbuffer_read_setup_vecs_(buf, size, vec, n_vecs, &chainp, 0); } done: EVBUFFER_UNLOCK(buf); return n; } static int advance_last_with_data(struct evbuffer *buf) { int n = 0; struct evbuffer_chain **chainp = buf->last_with_datap; ASSERT_EVBUFFER_LOCKED(buf); if (!*chainp) return 0; while ((*chainp)->next) { chainp = &(*chainp)->next; if ((*chainp)->off) buf->last_with_datap = chainp; ++n; } return n; } int evbuffer_commit_space(struct evbuffer *buf, struct evbuffer_iovec *vec, int n_vecs) { struct evbuffer_chain *chain, **firstchainp, **chainp; int result = -1; size_t added = 0; int i; EVBUFFER_LOCK(buf); if (buf->freeze_end) goto done; if (n_vecs == 0) { result = 0; goto done; } else if (n_vecs == 1 && (buf->last && vec[0].iov_base == (void *)CHAIN_SPACE_PTR(buf->last))) { /* The user only got or used one chain; it might not * be the first one with space in it. */ if ((size_t)vec[0].iov_len > (size_t)CHAIN_SPACE_LEN(buf->last)) goto done; buf->last->off += vec[0].iov_len; added = vec[0].iov_len; if (added) advance_last_with_data(buf); goto okay; } /* Advance 'firstchain' to the first chain with space in it. */ firstchainp = buf->last_with_datap; if (!*firstchainp) goto done; if (CHAIN_SPACE_LEN(*firstchainp) == 0) { firstchainp = &(*firstchainp)->next; } chain = *firstchainp; /* pass 1: make sure that the pointers and lengths of vecs[] are in * bounds before we try to commit anything. */ for (i=0; i CHAIN_SPACE_LEN(chain)) goto done; chain = chain->next; } /* pass 2: actually adjust all the chains. */ chainp = firstchainp; for (i=0; ioff += vec[i].iov_len; added += vec[i].iov_len; if (vec[i].iov_len) { buf->last_with_datap = chainp; } chainp = &(*chainp)->next; } okay: buf->total_len += added; buf->n_add_for_cb += added; result = 0; evbuffer_invoke_callbacks_(buf); done: EVBUFFER_UNLOCK(buf); return result; } static inline int HAS_PINNED_R(struct evbuffer *buf) { return (buf->last && CHAIN_PINNED_R(buf->last)); } static inline void ZERO_CHAIN(struct evbuffer *dst) { ASSERT_EVBUFFER_LOCKED(dst); dst->first = NULL; dst->last = NULL; dst->last_with_datap = &(dst)->first; dst->total_len = 0; } /* Prepares the contents of src to be moved to another buffer by removing * read-pinned chains. The first pinned chain is saved in first, and the * last in last. If src has no read-pinned chains, first and last are set * to NULL. */ static int PRESERVE_PINNED(struct evbuffer *src, struct evbuffer_chain **first, struct evbuffer_chain **last) { struct evbuffer_chain *chain, **pinned; ASSERT_EVBUFFER_LOCKED(src); if (!HAS_PINNED_R(src)) { *first = *last = NULL; return 0; } pinned = src->last_with_datap; if (!CHAIN_PINNED_R(*pinned)) pinned = &(*pinned)->next; EVUTIL_ASSERT(CHAIN_PINNED_R(*pinned)); chain = *first = *pinned; *last = src->last; /* If there's data in the first pinned chain, we need to allocate * a new chain and copy the data over. */ if (chain->off) { struct evbuffer_chain *tmp; EVUTIL_ASSERT(pinned == src->last_with_datap); tmp = evbuffer_chain_new(chain->off); if (!tmp) return -1; memcpy(tmp->buffer, chain->buffer + chain->misalign, chain->off); tmp->off = chain->off; *src->last_with_datap = tmp; src->last = tmp; chain->misalign += chain->off; chain->off = 0; } else { src->last = *src->last_with_datap; *pinned = NULL; } return 0; } static inline void RESTORE_PINNED(struct evbuffer *src, struct evbuffer_chain *pinned, struct evbuffer_chain *last) { ASSERT_EVBUFFER_LOCKED(src); if (!pinned) { ZERO_CHAIN(src); return; } src->first = pinned; src->last = last; src->last_with_datap = &src->first; src->total_len = 0; } static inline void COPY_CHAIN(struct evbuffer *dst, struct evbuffer *src) { ASSERT_EVBUFFER_LOCKED(dst); ASSERT_EVBUFFER_LOCKED(src); dst->first = src->first; if (src->last_with_datap == &src->first) dst->last_with_datap = &dst->first; else dst->last_with_datap = src->last_with_datap; dst->last = src->last; dst->total_len = src->total_len; } static void APPEND_CHAIN(struct evbuffer *dst, struct evbuffer *src) { struct evbuffer_chain **chp; ASSERT_EVBUFFER_LOCKED(dst); ASSERT_EVBUFFER_LOCKED(src); chp = evbuffer_free_trailing_empty_chains(dst); *chp = src->first; if (src->last_with_datap == &src->first) dst->last_with_datap = chp; else dst->last_with_datap = src->last_with_datap; dst->last = src->last; dst->total_len += src->total_len; } static inline void APPEND_CHAIN_MULTICAST(struct evbuffer *dst, struct evbuffer *src) { struct evbuffer_chain *tmp; struct evbuffer_chain *chain = src->first; struct evbuffer_multicast_parent *extra; ASSERT_EVBUFFER_LOCKED(dst); ASSERT_EVBUFFER_LOCKED(src); for (; chain; chain = chain->next) { if (!chain->off || chain->flags & EVBUFFER_DANGLING) { /* skip empty chains */ continue; } tmp = evbuffer_chain_new(sizeof(struct evbuffer_multicast_parent)); if (!tmp) { event_warn("%s: out of memory", __func__); return; } extra = EVBUFFER_CHAIN_EXTRA(struct evbuffer_multicast_parent, tmp); /* reference evbuffer containing source chain so it * doesn't get released while the chain is still * being referenced to */ evbuffer_incref_(src); extra->source = src; /* reference source chain which now becomes immutable */ evbuffer_chain_incref(chain); extra->parent = chain; chain->flags |= EVBUFFER_IMMUTABLE; tmp->buffer_len = chain->buffer_len; tmp->misalign = chain->misalign; tmp->off = chain->off; tmp->flags |= EVBUFFER_MULTICAST|EVBUFFER_IMMUTABLE; tmp->buffer = chain->buffer; evbuffer_chain_insert(dst, tmp); } } static void PREPEND_CHAIN(struct evbuffer *dst, struct evbuffer *src) { ASSERT_EVBUFFER_LOCKED(dst); ASSERT_EVBUFFER_LOCKED(src); src->last->next = dst->first; dst->first = src->first; dst->total_len += src->total_len; if (*dst->last_with_datap == NULL) { if (src->last_with_datap == &(src)->first) dst->last_with_datap = &dst->first; else dst->last_with_datap = src->last_with_datap; } else if (dst->last_with_datap == &dst->first) { dst->last_with_datap = &src->last->next; } } int evbuffer_add_buffer(struct evbuffer *outbuf, struct evbuffer *inbuf) { struct evbuffer_chain *pinned, *last; size_t in_total_len, out_total_len; int result = 0; EVBUFFER_LOCK2(inbuf, outbuf); in_total_len = inbuf->total_len; out_total_len = outbuf->total_len; if (in_total_len == 0 || outbuf == inbuf) goto done; if (outbuf->freeze_end || inbuf->freeze_start) { result = -1; goto done; } if (PRESERVE_PINNED(inbuf, &pinned, &last) < 0) { result = -1; goto done; } if (out_total_len == 0) { /* There might be an empty chain at the start of outbuf; free * it. */ evbuffer_free_all_chains(outbuf->first); COPY_CHAIN(outbuf, inbuf); } else { APPEND_CHAIN(outbuf, inbuf); } RESTORE_PINNED(inbuf, pinned, last); inbuf->n_del_for_cb += in_total_len; outbuf->n_add_for_cb += in_total_len; evbuffer_invoke_callbacks_(inbuf); evbuffer_invoke_callbacks_(outbuf); done: EVBUFFER_UNLOCK2(inbuf, outbuf); return result; } int evbuffer_add_buffer_reference(struct evbuffer *outbuf, struct evbuffer *inbuf) { size_t in_total_len, out_total_len; struct evbuffer_chain *chain; int result = 0; EVBUFFER_LOCK2(inbuf, outbuf); in_total_len = inbuf->total_len; out_total_len = outbuf->total_len; chain = inbuf->first; if (in_total_len == 0) goto done; if (outbuf->freeze_end || outbuf == inbuf) { result = -1; goto done; } for (; chain; chain = chain->next) { if ((chain->flags & (EVBUFFER_FILESEGMENT|EVBUFFER_SENDFILE|EVBUFFER_MULTICAST)) != 0) { /* chain type can not be referenced */ result = -1; goto done; } } if (out_total_len == 0) { /* There might be an empty chain at the start of outbuf; free * it. */ evbuffer_free_all_chains(outbuf->first); } APPEND_CHAIN_MULTICAST(outbuf, inbuf); outbuf->n_add_for_cb += in_total_len; evbuffer_invoke_callbacks_(outbuf); done: EVBUFFER_UNLOCK2(inbuf, outbuf); return result; } int evbuffer_prepend_buffer(struct evbuffer *outbuf, struct evbuffer *inbuf) { struct evbuffer_chain *pinned, *last; size_t in_total_len, out_total_len; int result = 0; EVBUFFER_LOCK2(inbuf, outbuf); in_total_len = inbuf->total_len; out_total_len = outbuf->total_len; if (!in_total_len || inbuf == outbuf) goto done; if (outbuf->freeze_start || inbuf->freeze_start) { result = -1; goto done; } if (PRESERVE_PINNED(inbuf, &pinned, &last) < 0) { result = -1; goto done; } if (out_total_len == 0) { /* There might be an empty chain at the start of outbuf; free * it. */ evbuffer_free_all_chains(outbuf->first); COPY_CHAIN(outbuf, inbuf); } else { PREPEND_CHAIN(outbuf, inbuf); } RESTORE_PINNED(inbuf, pinned, last); inbuf->n_del_for_cb += in_total_len; outbuf->n_add_for_cb += in_total_len; evbuffer_invoke_callbacks_(inbuf); evbuffer_invoke_callbacks_(outbuf); done: EVBUFFER_UNLOCK2(inbuf, outbuf); return result; } int evbuffer_drain(struct evbuffer *buf, size_t len) { struct evbuffer_chain *chain, *next; size_t remaining, old_len; int result = 0; EVBUFFER_LOCK(buf); old_len = buf->total_len; if (old_len == 0) goto done; if (buf->freeze_start) { result = -1; goto done; } if (len >= old_len && !HAS_PINNED_R(buf)) { len = old_len; for (chain = buf->first; chain != NULL; chain = next) { next = chain->next; evbuffer_chain_free(chain); } ZERO_CHAIN(buf); } else { if (len >= old_len) len = old_len; buf->total_len -= len; remaining = len; for (chain = buf->first; remaining >= chain->off; chain = next) { next = chain->next; remaining -= chain->off; if (chain == *buf->last_with_datap) { buf->last_with_datap = &buf->first; } if (&chain->next == buf->last_with_datap) buf->last_with_datap = &buf->first; if (CHAIN_PINNED_R(chain)) { EVUTIL_ASSERT(remaining == 0); chain->misalign += chain->off; chain->off = 0; break; } else evbuffer_chain_free(chain); } buf->first = chain; EVUTIL_ASSERT(remaining <= chain->off); chain->misalign += remaining; chain->off -= remaining; } buf->n_del_for_cb += len; /* Tell someone about changes in this buffer */ evbuffer_invoke_callbacks_(buf); done: EVBUFFER_UNLOCK(buf); return result; } /* Reads data from an event buffer and drains the bytes read */ int evbuffer_remove(struct evbuffer *buf, void *data_out, size_t datlen) { ev_ssize_t n; EVBUFFER_LOCK(buf); n = evbuffer_copyout_from(buf, NULL, data_out, datlen); if (n > 0) { if (evbuffer_drain(buf, n)<0) n = -1; } EVBUFFER_UNLOCK(buf); return (int)n; } ev_ssize_t evbuffer_copyout(struct evbuffer *buf, void *data_out, size_t datlen) { return evbuffer_copyout_from(buf, NULL, data_out, datlen); } ev_ssize_t evbuffer_copyout_from(struct evbuffer *buf, const struct evbuffer_ptr *pos, void *data_out, size_t datlen) { /*XXX fails badly on sendfile case. */ struct evbuffer_chain *chain; char *data = data_out; size_t nread; ev_ssize_t result = 0; size_t pos_in_chain; EVBUFFER_LOCK(buf); if (pos) { if (datlen > (size_t)(EV_SSIZE_MAX - pos->pos)) { result = -1; goto done; } chain = pos->internal_.chain; pos_in_chain = pos->internal_.pos_in_chain; if (datlen + pos->pos > buf->total_len) datlen = buf->total_len - pos->pos; } else { chain = buf->first; pos_in_chain = 0; if (datlen > buf->total_len) datlen = buf->total_len; } if (datlen == 0) goto done; if (buf->freeze_start) { result = -1; goto done; } nread = datlen; while (datlen && datlen >= chain->off - pos_in_chain) { size_t copylen = chain->off - pos_in_chain; memcpy(data, chain->buffer + chain->misalign + pos_in_chain, copylen); data += copylen; datlen -= copylen; chain = chain->next; pos_in_chain = 0; EVUTIL_ASSERT(chain || datlen==0); } if (datlen) { EVUTIL_ASSERT(chain); EVUTIL_ASSERT(datlen+pos_in_chain <= chain->off); memcpy(data, chain->buffer + chain->misalign + pos_in_chain, datlen); } result = nread; done: EVBUFFER_UNLOCK(buf); return result; } /* reads data from the src buffer to the dst buffer, avoids memcpy as * possible. */ /* XXXX should return ev_ssize_t */ int evbuffer_remove_buffer(struct evbuffer *src, struct evbuffer *dst, size_t datlen) { /*XXX We should have an option to force this to be zero-copy.*/ /*XXX can fail badly on sendfile case. */ struct evbuffer_chain *chain, *previous; size_t nread = 0; int result; EVBUFFER_LOCK2(src, dst); chain = previous = src->first; if (datlen == 0 || dst == src) { result = 0; goto done; } if (dst->freeze_end || src->freeze_start) { result = -1; goto done; } /* short-cut if there is no more data buffered */ if (datlen >= src->total_len) { datlen = src->total_len; evbuffer_add_buffer(dst, src); result = (int)datlen; /*XXXX should return ev_ssize_t*/ goto done; } /* removes chains if possible */ while (chain->off <= datlen) { /* We can't remove the last with data from src unless we * remove all chains, in which case we would have done the if * block above */ EVUTIL_ASSERT(chain != *src->last_with_datap); nread += chain->off; datlen -= chain->off; previous = chain; if (src->last_with_datap == &chain->next) src->last_with_datap = &src->first; chain = chain->next; } if (chain != src->first) { /* we can remove the chain */ struct evbuffer_chain **chp; chp = evbuffer_free_trailing_empty_chains(dst); if (dst->first == NULL) { dst->first = src->first; } else { *chp = src->first; } dst->last = previous; previous->next = NULL; src->first = chain; advance_last_with_data(dst); dst->total_len += nread; dst->n_add_for_cb += nread; } /* we know that there is more data in the src buffer than * we want to read, so we manually drain the chain */ evbuffer_add(dst, chain->buffer + chain->misalign, datlen); chain->misalign += datlen; chain->off -= datlen; nread += datlen; /* You might think we would want to increment dst->n_add_for_cb * here too. But evbuffer_add above already took care of that. */ src->total_len -= nread; src->n_del_for_cb += nread; if (nread) { evbuffer_invoke_callbacks_(dst); evbuffer_invoke_callbacks_(src); } result = (int)nread;/*XXXX should change return type */ done: EVBUFFER_UNLOCK2(src, dst); return result; } unsigned char * evbuffer_pullup(struct evbuffer *buf, ev_ssize_t size) { struct evbuffer_chain *chain, *next, *tmp, *last_with_data; unsigned char *buffer, *result = NULL; ev_ssize_t remaining; int removed_last_with_data = 0; int removed_last_with_datap = 0; EVBUFFER_LOCK(buf); chain = buf->first; if (size < 0) size = buf->total_len; /* if size > buf->total_len, we cannot guarantee to the user that she * is going to have a long enough buffer afterwards; so we return * NULL */ if (size == 0 || (size_t)size > buf->total_len) goto done; /* No need to pull up anything; the first size bytes are * already here. */ if (chain->off >= (size_t)size) { result = chain->buffer + chain->misalign; goto done; } /* Make sure that none of the chains we need to copy from is pinned. */ remaining = size - chain->off; EVUTIL_ASSERT(remaining >= 0); for (tmp=chain->next; tmp; tmp=tmp->next) { if (CHAIN_PINNED(tmp)) goto done; if (tmp->off >= (size_t)remaining) break; remaining -= tmp->off; } if (CHAIN_PINNED(chain)) { size_t old_off = chain->off; if (CHAIN_SPACE_LEN(chain) < size - chain->off) { /* not enough room at end of chunk. */ goto done; } buffer = CHAIN_SPACE_PTR(chain); tmp = chain; tmp->off = size; size -= old_off; chain = chain->next; } else if (chain->buffer_len - chain->misalign >= (size_t)size) { /* already have enough space in the first chain */ size_t old_off = chain->off; buffer = chain->buffer + chain->misalign + chain->off; tmp = chain; tmp->off = size; size -= old_off; chain = chain->next; } else { if ((tmp = evbuffer_chain_new(size)) == NULL) { event_warn("%s: out of memory", __func__); goto done; } buffer = tmp->buffer; tmp->off = size; buf->first = tmp; } /* TODO(niels): deal with buffers that point to NULL like sendfile */ /* Copy and free every chunk that will be entirely pulled into tmp */ last_with_data = *buf->last_with_datap; for (; chain != NULL && (size_t)size >= chain->off; chain = next) { next = chain->next; if (chain->buffer) { memcpy(buffer, chain->buffer + chain->misalign, chain->off); size -= chain->off; buffer += chain->off; } if (chain == last_with_data) removed_last_with_data = 1; if (&chain->next == buf->last_with_datap) removed_last_with_datap = 1; evbuffer_chain_free(chain); } if (chain != NULL) { memcpy(buffer, chain->buffer + chain->misalign, size); chain->misalign += size; chain->off -= size; } else { buf->last = tmp; } tmp->next = chain; if (removed_last_with_data) { buf->last_with_datap = &buf->first; } else if (removed_last_with_datap) { if (buf->first->next && buf->first->next->off) buf->last_with_datap = &buf->first->next; else buf->last_with_datap = &buf->first; } result = (tmp->buffer + tmp->misalign); done: EVBUFFER_UNLOCK(buf); return result; } /* * Reads a line terminated by either '\r\n', '\n\r' or '\r' or '\n'. * The returned buffer needs to be freed by the called. */ char * evbuffer_readline(struct evbuffer *buffer) { return evbuffer_readln(buffer, NULL, EVBUFFER_EOL_ANY); } static inline ev_ssize_t evbuffer_strchr(struct evbuffer_ptr *it, const char chr) { struct evbuffer_chain *chain = it->internal_.chain; size_t i = it->internal_.pos_in_chain; while (chain != NULL) { char *buffer = (char *)chain->buffer + chain->misalign; char *cp = memchr(buffer+i, chr, chain->off-i); if (cp) { it->internal_.chain = chain; it->internal_.pos_in_chain = cp - buffer; it->pos += (cp - buffer - i); return it->pos; } it->pos += chain->off - i; i = 0; chain = chain->next; } return (-1); } static inline char * find_eol_char(char *s, size_t len) { #define CHUNK_SZ 128 /* Lots of benchmarking found this approach to be faster in practice * than doing two memchrs over the whole buffer, doin a memchr on each * char of the buffer, or trying to emulate memchr by hand. */ char *s_end, *cr, *lf; s_end = s+len; while (s < s_end) { size_t chunk = (s + CHUNK_SZ < s_end) ? CHUNK_SZ : (s_end - s); cr = memchr(s, '\r', chunk); lf = memchr(s, '\n', chunk); if (cr) { if (lf && lf < cr) return lf; return cr; } else if (lf) { return lf; } s += CHUNK_SZ; } return NULL; #undef CHUNK_SZ } static ev_ssize_t evbuffer_find_eol_char(struct evbuffer_ptr *it) { struct evbuffer_chain *chain = it->internal_.chain; size_t i = it->internal_.pos_in_chain; while (chain != NULL) { char *buffer = (char *)chain->buffer + chain->misalign; char *cp = find_eol_char(buffer+i, chain->off-i); if (cp) { it->internal_.chain = chain; it->internal_.pos_in_chain = cp - buffer; it->pos += (cp - buffer) - i; return it->pos; } it->pos += chain->off - i; i = 0; chain = chain->next; } return (-1); } static inline size_t evbuffer_strspn( struct evbuffer_ptr *ptr, const char *chrset) { size_t count = 0; struct evbuffer_chain *chain = ptr->internal_.chain; size_t i = ptr->internal_.pos_in_chain; if (!chain) return 0; while (1) { char *buffer = (char *)chain->buffer + chain->misalign; for (; i < chain->off; ++i) { const char *p = chrset; while (*p) { if (buffer[i] == *p++) goto next; } ptr->internal_.chain = chain; ptr->internal_.pos_in_chain = i; ptr->pos += count; return count; next: ++count; } i = 0; if (! chain->next) { ptr->internal_.chain = chain; ptr->internal_.pos_in_chain = i; ptr->pos += count; return count; } chain = chain->next; } } static inline int evbuffer_getchr(struct evbuffer_ptr *it) { struct evbuffer_chain *chain = it->internal_.chain; size_t off = it->internal_.pos_in_chain; if (chain == NULL) return -1; return (unsigned char)chain->buffer[chain->misalign + off]; } struct evbuffer_ptr evbuffer_search_eol(struct evbuffer *buffer, struct evbuffer_ptr *start, size_t *eol_len_out, enum evbuffer_eol_style eol_style) { struct evbuffer_ptr it, it2; size_t extra_drain = 0; int ok = 0; /* Avoid locking in trivial edge cases */ if (start && start->internal_.chain == NULL) { PTR_NOT_FOUND(&it); if (eol_len_out) *eol_len_out = extra_drain; return it; } EVBUFFER_LOCK(buffer); if (start) { memcpy(&it, start, sizeof(it)); } else { it.pos = 0; it.internal_.chain = buffer->first; it.internal_.pos_in_chain = 0; } /* the eol_style determines our first stop character and how many * characters we are going to drain afterwards. */ switch (eol_style) { case EVBUFFER_EOL_ANY: if (evbuffer_find_eol_char(&it) < 0) goto done; memcpy(&it2, &it, sizeof(it)); extra_drain = evbuffer_strspn(&it2, "\r\n"); break; case EVBUFFER_EOL_CRLF_STRICT: { it = evbuffer_search(buffer, "\r\n", 2, &it); if (it.pos < 0) goto done; extra_drain = 2; break; } case EVBUFFER_EOL_CRLF: { ev_ssize_t start_pos = it.pos; /* Look for a LF ... */ if (evbuffer_strchr(&it, '\n') < 0) goto done; extra_drain = 1; /* ... optionally preceeded by a CR. */ if (it.pos == start_pos) break; /* If the first character is \n, don't back up */ /* This potentially does an extra linear walk over the first * few chains. Probably, that's not too expensive unless you * have a really pathological setup. */ memcpy(&it2, &it, sizeof(it)); if (evbuffer_ptr_subtract(buffer, &it2, 1)<0) break; if (evbuffer_getchr(&it2) == '\r') { memcpy(&it, &it2, sizeof(it)); extra_drain = 2; } break; } case EVBUFFER_EOL_LF: if (evbuffer_strchr(&it, '\n') < 0) goto done; extra_drain = 1; break; case EVBUFFER_EOL_NUL: if (evbuffer_strchr(&it, '\0') < 0) goto done; extra_drain = 1; break; default: goto done; } ok = 1; done: EVBUFFER_UNLOCK(buffer); if (!ok) PTR_NOT_FOUND(&it); if (eol_len_out) *eol_len_out = extra_drain; return it; } char * evbuffer_readln(struct evbuffer *buffer, size_t *n_read_out, enum evbuffer_eol_style eol_style) { struct evbuffer_ptr it; char *line; size_t n_to_copy=0, extra_drain=0; char *result = NULL; EVBUFFER_LOCK(buffer); if (buffer->freeze_start) { goto done; } it = evbuffer_search_eol(buffer, NULL, &extra_drain, eol_style); if (it.pos < 0) goto done; n_to_copy = it.pos; if ((line = mm_malloc(n_to_copy+1)) == NULL) { event_warn("%s: out of memory", __func__); goto done; } evbuffer_remove(buffer, line, n_to_copy); line[n_to_copy] = '\0'; evbuffer_drain(buffer, extra_drain); result = line; done: EVBUFFER_UNLOCK(buffer); if (n_read_out) *n_read_out = result ? n_to_copy : 0; return result; } #define EVBUFFER_CHAIN_MAX_AUTO_SIZE 4096 /* Adds data to an event buffer */ int evbuffer_add(struct evbuffer *buf, const void *data_in, size_t datlen) { struct evbuffer_chain *chain, *tmp; const unsigned char *data = data_in; size_t remain, to_alloc; int result = -1; EVBUFFER_LOCK(buf); if (buf->freeze_end) { goto done; } /* Prevent buf->total_len overflow */ if (datlen > EV_SIZE_MAX - buf->total_len) { goto done; } if (*buf->last_with_datap == NULL) { chain = buf->last; } else { chain = *buf->last_with_datap; } /* If there are no chains allocated for this buffer, allocate one * big enough to hold all the data. */ if (chain == NULL) { chain = evbuffer_chain_new(datlen); if (!chain) goto done; evbuffer_chain_insert(buf, chain); } if ((chain->flags & EVBUFFER_IMMUTABLE) == 0) { /* Always true for mutable buffers */ EVUTIL_ASSERT(chain->misalign >= 0 && (ev_uint64_t)chain->misalign <= EVBUFFER_CHAIN_MAX); remain = chain->buffer_len - (size_t)chain->misalign - chain->off; if (remain >= datlen) { /* there's enough space to hold all the data in the * current last chain */ memcpy(chain->buffer + chain->misalign + chain->off, data, datlen); chain->off += datlen; buf->total_len += datlen; buf->n_add_for_cb += datlen; goto out; } else if (!CHAIN_PINNED(chain) && evbuffer_chain_should_realign(chain, datlen)) { /* we can fit the data into the misalignment */ evbuffer_chain_align(chain); memcpy(chain->buffer + chain->off, data, datlen); chain->off += datlen; buf->total_len += datlen; buf->n_add_for_cb += datlen; goto out; } } else { /* we cannot write any data to the last chain */ remain = 0; } /* we need to add another chain */ to_alloc = chain->buffer_len; if (to_alloc <= EVBUFFER_CHAIN_MAX_AUTO_SIZE/2) to_alloc <<= 1; if (datlen > to_alloc) to_alloc = datlen; tmp = evbuffer_chain_new(to_alloc); if (tmp == NULL) goto done; if (remain) { memcpy(chain->buffer + chain->misalign + chain->off, data, remain); chain->off += remain; buf->total_len += remain; buf->n_add_for_cb += remain; } data += remain; datlen -= remain; memcpy(tmp->buffer, data, datlen); tmp->off = datlen; evbuffer_chain_insert(buf, tmp); buf->n_add_for_cb += datlen; out: evbuffer_invoke_callbacks_(buf); result = 0; done: EVBUFFER_UNLOCK(buf); return result; } int evbuffer_prepend(struct evbuffer *buf, const void *data, size_t datlen) { struct evbuffer_chain *chain, *tmp; int result = -1; EVBUFFER_LOCK(buf); if (datlen == 0) { result = 0; goto done; } if (buf->freeze_start) { goto done; } if (datlen > EV_SIZE_MAX - buf->total_len) { goto done; } chain = buf->first; if (chain == NULL) { chain = evbuffer_chain_new(datlen); if (!chain) goto done; evbuffer_chain_insert(buf, chain); } /* we cannot touch immutable buffers */ if ((chain->flags & EVBUFFER_IMMUTABLE) == 0) { /* Always true for mutable buffers */ EVUTIL_ASSERT(chain->misalign >= 0 && (ev_uint64_t)chain->misalign <= EVBUFFER_CHAIN_MAX); /* If this chain is empty, we can treat it as * 'empty at the beginning' rather than 'empty at the end' */ if (chain->off == 0) chain->misalign = chain->buffer_len; if ((size_t)chain->misalign >= datlen) { /* we have enough space to fit everything */ memcpy(chain->buffer + chain->misalign - datlen, data, datlen); chain->off += datlen; chain->misalign -= datlen; buf->total_len += datlen; buf->n_add_for_cb += datlen; goto out; } else if (chain->misalign) { /* we can only fit some of the data. */ memcpy(chain->buffer, (char*)data + datlen - chain->misalign, (size_t)chain->misalign); chain->off += (size_t)chain->misalign; buf->total_len += (size_t)chain->misalign; buf->n_add_for_cb += (size_t)chain->misalign; datlen -= (size_t)chain->misalign; chain->misalign = 0; } } /* we need to add another chain */ if ((tmp = evbuffer_chain_new(datlen)) == NULL) goto done; buf->first = tmp; if (buf->last_with_datap == &buf->first && chain->off) buf->last_with_datap = &tmp->next; tmp->next = chain; tmp->off = datlen; EVUTIL_ASSERT(datlen <= tmp->buffer_len); tmp->misalign = tmp->buffer_len - datlen; memcpy(tmp->buffer + tmp->misalign, data, datlen); buf->total_len += datlen; buf->n_add_for_cb += datlen; out: evbuffer_invoke_callbacks_(buf); result = 0; done: EVBUFFER_UNLOCK(buf); return result; } /** Helper: realigns the memory in chain->buffer so that misalign is 0. */ static void evbuffer_chain_align(struct evbuffer_chain *chain) { EVUTIL_ASSERT(!(chain->flags & EVBUFFER_IMMUTABLE)); EVUTIL_ASSERT(!(chain->flags & EVBUFFER_MEM_PINNED_ANY)); memmove(chain->buffer, chain->buffer + chain->misalign, chain->off); chain->misalign = 0; } #define MAX_TO_COPY_IN_EXPAND 4096 #define MAX_TO_REALIGN_IN_EXPAND 2048 /** Helper: return true iff we should realign chain to fit datalen bytes of data in it. */ static int evbuffer_chain_should_realign(struct evbuffer_chain *chain, size_t datlen) { return chain->buffer_len - chain->off >= datlen && (chain->off < chain->buffer_len / 2) && (chain->off <= MAX_TO_REALIGN_IN_EXPAND); } /* Expands the available space in the event buffer to at least datlen, all in * a single chunk. Return that chunk. */ static struct evbuffer_chain * evbuffer_expand_singlechain(struct evbuffer *buf, size_t datlen) { struct evbuffer_chain *chain, **chainp; struct evbuffer_chain *result = NULL; ASSERT_EVBUFFER_LOCKED(buf); chainp = buf->last_with_datap; /* XXX If *chainp is no longer writeable, but has enough space in its * misalign, this might be a bad idea: we could still use *chainp, not * (*chainp)->next. */ if (*chainp && CHAIN_SPACE_LEN(*chainp) == 0) chainp = &(*chainp)->next; /* 'chain' now points to the first chain with writable space (if any) * We will either use it, realign it, replace it, or resize it. */ chain = *chainp; if (chain == NULL || (chain->flags & (EVBUFFER_IMMUTABLE|EVBUFFER_MEM_PINNED_ANY))) { /* We can't use the last_with_data chain at all. Just add a * new one that's big enough. */ goto insert_new; } /* If we can fit all the data, then we don't have to do anything */ if (CHAIN_SPACE_LEN(chain) >= datlen) { result = chain; goto ok; } /* If the chain is completely empty, just replace it by adding a new * empty chain. */ if (chain->off == 0) { goto insert_new; } /* If the misalignment plus the remaining space fulfills our data * needs, we could just force an alignment to happen. Afterwards, we * have enough space. But only do this if we're saving a lot of space * and not moving too much data. Otherwise the space savings are * probably offset by the time lost in copying. */ if (evbuffer_chain_should_realign(chain, datlen)) { evbuffer_chain_align(chain); result = chain; goto ok; } /* At this point, we can either resize the last chunk with space in * it, use the next chunk after it, or If we add a new chunk, we waste * CHAIN_SPACE_LEN(chain) bytes in the former last chunk. If we * resize, we have to copy chain->off bytes. */ /* Would expanding this chunk be affordable and worthwhile? */ if (CHAIN_SPACE_LEN(chain) < chain->buffer_len / 8 || chain->off > MAX_TO_COPY_IN_EXPAND || datlen >= (EVBUFFER_CHAIN_MAX - chain->off)) { /* It's not worth resizing this chain. Can the next one be * used? */ if (chain->next && CHAIN_SPACE_LEN(chain->next) >= datlen) { /* Yes, we can just use the next chain (which should * be empty. */ result = chain->next; goto ok; } else { /* No; append a new chain (which will free all * terminal empty chains.) */ goto insert_new; } } else { /* Okay, we're going to try to resize this chain: Not doing so * would waste at least 1/8 of its current allocation, and we * can do so without having to copy more than * MAX_TO_COPY_IN_EXPAND bytes. */ /* figure out how much space we need */ size_t length = chain->off + datlen; struct evbuffer_chain *tmp = evbuffer_chain_new(length); if (tmp == NULL) goto err; /* copy the data over that we had so far */ tmp->off = chain->off; memcpy(tmp->buffer, chain->buffer + chain->misalign, chain->off); /* fix up the list */ EVUTIL_ASSERT(*chainp == chain); result = *chainp = tmp; if (buf->last == chain) buf->last = tmp; tmp->next = chain->next; evbuffer_chain_free(chain); goto ok; } insert_new: result = evbuffer_chain_insert_new(buf, datlen); if (!result) goto err; ok: EVUTIL_ASSERT(result); EVUTIL_ASSERT(CHAIN_SPACE_LEN(result) >= datlen); err: return result; } /* Make sure that datlen bytes are available for writing in the last n * chains. Never copies or moves data. */ int evbuffer_expand_fast_(struct evbuffer *buf, size_t datlen, int n) { struct evbuffer_chain *chain = buf->last, *tmp, *next; size_t avail; int used; ASSERT_EVBUFFER_LOCKED(buf); EVUTIL_ASSERT(n >= 2); if (chain == NULL || (chain->flags & EVBUFFER_IMMUTABLE)) { /* There is no last chunk, or we can't touch the last chunk. * Just add a new chunk. */ chain = evbuffer_chain_new(datlen); if (chain == NULL) return (-1); evbuffer_chain_insert(buf, chain); return (0); } used = 0; /* number of chains we're using space in. */ avail = 0; /* how much space they have. */ /* How many bytes can we stick at the end of buffer as it is? Iterate * over the chains at the end of the buffer, tring to see how much * space we have in the first n. */ for (chain = *buf->last_with_datap; chain; chain = chain->next) { if (chain->off) { size_t space = (size_t) CHAIN_SPACE_LEN(chain); EVUTIL_ASSERT(chain == *buf->last_with_datap); if (space) { avail += space; ++used; } } else { /* No data in chain; realign it. */ chain->misalign = 0; avail += chain->buffer_len; ++used; } if (avail >= datlen) { /* There is already enough space. Just return */ return (0); } if (used == n) break; } /* There wasn't enough space in the first n chains with space in * them. Either add a new chain with enough space, or replace all * empty chains with one that has enough space, depending on n. */ if (used < n) { /* The loop ran off the end of the chains before it hit n * chains; we can add another. */ EVUTIL_ASSERT(chain == NULL); tmp = evbuffer_chain_new(datlen - avail); if (tmp == NULL) return (-1); buf->last->next = tmp; buf->last = tmp; /* (we would only set last_with_data if we added the first * chain. But if the buffer had no chains, we would have * just allocated a new chain earlier) */ return (0); } else { /* Nuke _all_ the empty chains. */ int rmv_all = 0; /* True iff we removed last_with_data. */ chain = *buf->last_with_datap; if (!chain->off) { EVUTIL_ASSERT(chain == buf->first); rmv_all = 1; avail = 0; } else { /* can't overflow, since only mutable chains have * huge misaligns. */ avail = (size_t) CHAIN_SPACE_LEN(chain); chain = chain->next; } for (; chain; chain = next) { next = chain->next; EVUTIL_ASSERT(chain->off == 0); evbuffer_chain_free(chain); } EVUTIL_ASSERT(datlen >= avail); tmp = evbuffer_chain_new(datlen - avail); if (tmp == NULL) { if (rmv_all) { ZERO_CHAIN(buf); } else { buf->last = *buf->last_with_datap; (*buf->last_with_datap)->next = NULL; } return (-1); } if (rmv_all) { buf->first = buf->last = tmp; buf->last_with_datap = &buf->first; } else { (*buf->last_with_datap)->next = tmp; buf->last = tmp; } return (0); } } int evbuffer_expand(struct evbuffer *buf, size_t datlen) { struct evbuffer_chain *chain; EVBUFFER_LOCK(buf); chain = evbuffer_expand_singlechain(buf, datlen); EVBUFFER_UNLOCK(buf); return chain ? 0 : -1; } /* * Reads data from a file descriptor into a buffer. */ #if defined(EVENT__HAVE_SYS_UIO_H) || defined(_WIN32) #define USE_IOVEC_IMPL #endif #ifdef USE_IOVEC_IMPL #ifdef EVENT__HAVE_SYS_UIO_H /* number of iovec we use for writev, fragmentation is going to determine * how much we end up writing */ #define DEFAULT_WRITE_IOVEC 128 #if defined(UIO_MAXIOV) && UIO_MAXIOV < DEFAULT_WRITE_IOVEC #define NUM_WRITE_IOVEC UIO_MAXIOV #elif defined(IOV_MAX) && IOV_MAX < DEFAULT_WRITE_IOVEC #define NUM_WRITE_IOVEC IOV_MAX #else #define NUM_WRITE_IOVEC DEFAULT_WRITE_IOVEC #endif #define IOV_TYPE struct iovec #define IOV_PTR_FIELD iov_base #define IOV_LEN_FIELD iov_len #define IOV_LEN_TYPE size_t #else #define NUM_WRITE_IOVEC 16 #define IOV_TYPE WSABUF #define IOV_PTR_FIELD buf #define IOV_LEN_FIELD len #define IOV_LEN_TYPE unsigned long #endif #endif #define NUM_READ_IOVEC 4 #define EVBUFFER_MAX_READ 4096 /** Helper function to figure out which space to use for reading data into an evbuffer. Internal use only. @param buf The buffer to read into @param howmuch How much we want to read. @param vecs An array of two or more iovecs or WSABUFs. @param n_vecs_avail The length of vecs @param chainp A pointer to a variable to hold the first chain we're reading into. @param exact Boolean: if true, we do not provide more than 'howmuch' space in the vectors, even if more space is available. @return The number of buffers we're using. */ int evbuffer_read_setup_vecs_(struct evbuffer *buf, ev_ssize_t howmuch, struct evbuffer_iovec *vecs, int n_vecs_avail, struct evbuffer_chain ***chainp, int exact) { struct evbuffer_chain *chain; struct evbuffer_chain **firstchainp; size_t so_far; int i; ASSERT_EVBUFFER_LOCKED(buf); if (howmuch < 0) return -1; so_far = 0; /* Let firstchain be the first chain with any space on it */ firstchainp = buf->last_with_datap; EVUTIL_ASSERT(*firstchainp); if (CHAIN_SPACE_LEN(*firstchainp) == 0) { firstchainp = &(*firstchainp)->next; } chain = *firstchainp; EVUTIL_ASSERT(chain); for (i = 0; i < n_vecs_avail && so_far < (size_t)howmuch; ++i) { size_t avail = (size_t) CHAIN_SPACE_LEN(chain); if (avail > (howmuch - so_far) && exact) avail = howmuch - so_far; vecs[i].iov_base = (void *)CHAIN_SPACE_PTR(chain); vecs[i].iov_len = avail; so_far += avail; chain = chain->next; } *chainp = firstchainp; return i; } static int get_n_bytes_readable_on_socket(evutil_socket_t fd) { #if defined(FIONREAD) && defined(_WIN32) unsigned long lng = EVBUFFER_MAX_READ; if (ioctlsocket(fd, FIONREAD, &lng) < 0) return -1; /* Can overflow, but mostly harmlessly. XXXX */ return (int)lng; #elif defined(FIONREAD) int n = EVBUFFER_MAX_READ; if (ioctl(fd, FIONREAD, &n) < 0) return -1; return n; #else return EVBUFFER_MAX_READ; #endif } /* TODO(niels): should this function return ev_ssize_t and take ev_ssize_t * as howmuch? */ int evbuffer_read(struct evbuffer *buf, evutil_socket_t fd, int howmuch) { struct evbuffer_chain **chainp; int n; int result; #ifdef USE_IOVEC_IMPL int nvecs, i, remaining; #else struct evbuffer_chain *chain; unsigned char *p; #endif EVBUFFER_LOCK(buf); if (buf->freeze_end) { result = -1; goto done; } n = get_n_bytes_readable_on_socket(fd); if (n <= 0 || n > EVBUFFER_MAX_READ) n = EVBUFFER_MAX_READ; if (howmuch < 0 || howmuch > n) howmuch = n; #ifdef USE_IOVEC_IMPL /* Since we can use iovecs, we're willing to use the last * NUM_READ_IOVEC chains. */ if (evbuffer_expand_fast_(buf, howmuch, NUM_READ_IOVEC) == -1) { result = -1; goto done; } else { IOV_TYPE vecs[NUM_READ_IOVEC]; #ifdef EVBUFFER_IOVEC_IS_NATIVE_ nvecs = evbuffer_read_setup_vecs_(buf, howmuch, vecs, NUM_READ_IOVEC, &chainp, 1); #else /* We aren't using the native struct iovec. Therefore, we are on win32. */ struct evbuffer_iovec ev_vecs[NUM_READ_IOVEC]; nvecs = evbuffer_read_setup_vecs_(buf, howmuch, ev_vecs, 2, &chainp, 1); for (i=0; i < nvecs; ++i) WSABUF_FROM_EVBUFFER_IOV(&vecs[i], &ev_vecs[i]); #endif #ifdef _WIN32 { DWORD bytesRead; DWORD flags=0; if (WSARecv(fd, vecs, nvecs, &bytesRead, &flags, NULL, NULL)) { /* The read failed. It might be a close, * or it might be an error. */ if (WSAGetLastError() == WSAECONNABORTED) n = 0; else n = -1; } else n = bytesRead; } #else n = readv(fd, vecs, nvecs); #endif } #else /*!USE_IOVEC_IMPL*/ /* If we don't have FIONREAD, we might waste some space here */ /* XXX we _will_ waste some space here if there is any space left * over on buf->last. */ if ((chain = evbuffer_expand_singlechain(buf, howmuch)) == NULL) { result = -1; goto done; } /* We can append new data at this point */ p = chain->buffer + chain->misalign + chain->off; #ifndef _WIN32 n = read(fd, p, howmuch); #else n = recv(fd, p, howmuch, 0); #endif #endif /* USE_IOVEC_IMPL */ if (n == -1) { result = -1; goto done; } if (n == 0) { result = 0; goto done; } #ifdef USE_IOVEC_IMPL remaining = n; for (i=0; i < nvecs; ++i) { /* can't overflow, since only mutable chains have * huge misaligns. */ size_t space = (size_t) CHAIN_SPACE_LEN(*chainp); /* XXXX This is a kludge that can waste space in perverse * situations. */ if (space > EVBUFFER_CHAIN_MAX) space = EVBUFFER_CHAIN_MAX; if ((ev_ssize_t)space < remaining) { (*chainp)->off += space; remaining -= (int)space; } else { (*chainp)->off += remaining; buf->last_with_datap = chainp; break; } chainp = &(*chainp)->next; } #else chain->off += n; advance_last_with_data(buf); #endif buf->total_len += n; buf->n_add_for_cb += n; /* Tell someone about changes in this buffer */ evbuffer_invoke_callbacks_(buf); result = n; done: EVBUFFER_UNLOCK(buf); return result; } #ifdef USE_IOVEC_IMPL static inline int evbuffer_write_iovec(struct evbuffer *buffer, evutil_socket_t fd, ev_ssize_t howmuch) { IOV_TYPE iov[NUM_WRITE_IOVEC]; struct evbuffer_chain *chain = buffer->first; int n, i = 0; if (howmuch < 0) return -1; ASSERT_EVBUFFER_LOCKED(buffer); /* XXX make this top out at some maximal data length? if the * buffer has (say) 1MB in it, split over 128 chains, there's * no way it all gets written in one go. */ while (chain != NULL && i < NUM_WRITE_IOVEC && howmuch) { #ifdef USE_SENDFILE /* we cannot write the file info via writev */ if (chain->flags & EVBUFFER_SENDFILE) break; #endif iov[i].IOV_PTR_FIELD = (void *) (chain->buffer + chain->misalign); if ((size_t)howmuch >= chain->off) { /* XXXcould be problematic when windows supports mmap*/ iov[i++].IOV_LEN_FIELD = (IOV_LEN_TYPE)chain->off; howmuch -= chain->off; } else { /* XXXcould be problematic when windows supports mmap*/ iov[i++].IOV_LEN_FIELD = (IOV_LEN_TYPE)howmuch; break; } chain = chain->next; } if (! i) return 0; #ifdef _WIN32 { DWORD bytesSent; if (WSASend(fd, iov, i, &bytesSent, 0, NULL, NULL)) n = -1; else n = bytesSent; } #else n = writev(fd, iov, i); #endif return (n); } #endif #ifdef USE_SENDFILE static inline int evbuffer_write_sendfile(struct evbuffer *buffer, evutil_socket_t dest_fd, ev_ssize_t howmuch) { struct evbuffer_chain *chain = buffer->first; struct evbuffer_chain_file_segment *info = EVBUFFER_CHAIN_EXTRA(struct evbuffer_chain_file_segment, chain); const int source_fd = info->segment->fd; #if defined(SENDFILE_IS_MACOSX) || defined(SENDFILE_IS_FREEBSD) int res; ev_off_t len = chain->off; #elif defined(SENDFILE_IS_LINUX) || defined(SENDFILE_IS_SOLARIS) ev_ssize_t res; off_t offset = chain->misalign; #endif ASSERT_EVBUFFER_LOCKED(buffer); #if defined(SENDFILE_IS_MACOSX) res = sendfile(source_fd, dest_fd, chain->misalign, &len, NULL, 0); if (res == -1 && !EVUTIL_ERR_RW_RETRIABLE(errno)) return (-1); return (len); #elif defined(SENDFILE_IS_FREEBSD) res = sendfile(source_fd, dest_fd, chain->misalign, chain->off, NULL, &len, 0); if (res == -1 && !EVUTIL_ERR_RW_RETRIABLE(errno)) return (-1); return (len); #elif defined(SENDFILE_IS_LINUX) /* TODO(niels): implement splice */ res = sendfile(dest_fd, source_fd, &offset, chain->off); if (res == -1 && EVUTIL_ERR_RW_RETRIABLE(errno)) { /* if this is EAGAIN or EINTR return 0; otherwise, -1 */ return (0); } return (res); #elif defined(SENDFILE_IS_SOLARIS) { const off_t offset_orig = offset; res = sendfile(dest_fd, source_fd, &offset, chain->off); if (res == -1 && EVUTIL_ERR_RW_RETRIABLE(errno)) { if (offset - offset_orig) return offset - offset_orig; /* if this is EAGAIN or EINTR and no bytes were * written, return 0 */ return (0); } return (res); } #endif } #endif int evbuffer_write_atmost(struct evbuffer *buffer, evutil_socket_t fd, ev_ssize_t howmuch) { int n = -1; EVBUFFER_LOCK(buffer); if (buffer->freeze_start) { goto done; } if (howmuch < 0 || (size_t)howmuch > buffer->total_len) howmuch = buffer->total_len; if (howmuch > 0) { #ifdef USE_SENDFILE struct evbuffer_chain *chain = buffer->first; if (chain != NULL && (chain->flags & EVBUFFER_SENDFILE)) n = evbuffer_write_sendfile(buffer, fd, howmuch); else { #endif #ifdef USE_IOVEC_IMPL n = evbuffer_write_iovec(buffer, fd, howmuch); #elif defined(_WIN32) /* XXX(nickm) Don't disable this code until we know if * the WSARecv code above works. */ void *p = evbuffer_pullup(buffer, howmuch); EVUTIL_ASSERT(p || !howmuch); n = send(fd, p, howmuch, 0); #else void *p = evbuffer_pullup(buffer, howmuch); EVUTIL_ASSERT(p || !howmuch); n = write(fd, p, howmuch); #endif #ifdef USE_SENDFILE } #endif } if (n > 0) evbuffer_drain(buffer, n); done: EVBUFFER_UNLOCK(buffer); return (n); } int evbuffer_write(struct evbuffer *buffer, evutil_socket_t fd) { return evbuffer_write_atmost(buffer, fd, -1); } unsigned char * evbuffer_find(struct evbuffer *buffer, const unsigned char *what, size_t len) { unsigned char *search; struct evbuffer_ptr ptr; EVBUFFER_LOCK(buffer); ptr = evbuffer_search(buffer, (const char *)what, len, NULL); if (ptr.pos < 0) { search = NULL; } else { search = evbuffer_pullup(buffer, ptr.pos + len); if (search) search += ptr.pos; } EVBUFFER_UNLOCK(buffer); return search; } /* Subract howfar from the position of pos within * buf. Returns 0 on success, -1 on failure. * * This isn't exposed yet, because of potential inefficiency issues. * Maybe it should be. */ static int evbuffer_ptr_subtract(struct evbuffer *buf, struct evbuffer_ptr *pos, size_t howfar) { if (pos->pos < 0) return -1; if (howfar > (size_t)pos->pos) return -1; if (pos->internal_.chain && howfar <= pos->internal_.pos_in_chain) { pos->internal_.pos_in_chain -= howfar; pos->pos -= howfar; return 0; } else { const size_t newpos = pos->pos - howfar; /* Here's the inefficient part: it walks over the * chains until we hit newpos. */ return evbuffer_ptr_set(buf, pos, newpos, EVBUFFER_PTR_SET); } } int evbuffer_ptr_set(struct evbuffer *buf, struct evbuffer_ptr *pos, size_t position, enum evbuffer_ptr_how how) { size_t left = position; struct evbuffer_chain *chain = NULL; int result = 0; EVBUFFER_LOCK(buf); switch (how) { case EVBUFFER_PTR_SET: chain = buf->first; pos->pos = position; position = 0; break; case EVBUFFER_PTR_ADD: /* this avoids iterating over all previous chains if we just want to advance the position */ if (pos->pos < 0 || EV_SIZE_MAX - position < (size_t)pos->pos) { EVBUFFER_UNLOCK(buf); return -1; } chain = pos->internal_.chain; pos->pos += position; position = pos->internal_.pos_in_chain; break; } EVUTIL_ASSERT(EV_SIZE_MAX - left >= position); while (chain && position + left >= chain->off) { left -= chain->off - position; chain = chain->next; position = 0; } if (chain) { pos->internal_.chain = chain; pos->internal_.pos_in_chain = position + left; } else if (left == 0) { /* The first byte in the (nonexistent) chain after the last chain */ pos->internal_.chain = NULL; pos->internal_.pos_in_chain = 0; } else { PTR_NOT_FOUND(pos); result = -1; } EVBUFFER_UNLOCK(buf); return result; } /** Compare the bytes in buf at position pos to the len bytes in mem. Return less than 0, 0, or greater than 0 as memcmp. */ static int evbuffer_ptr_memcmp(const struct evbuffer *buf, const struct evbuffer_ptr *pos, const char *mem, size_t len) { struct evbuffer_chain *chain; size_t position; int r; ASSERT_EVBUFFER_LOCKED(buf); if (pos->pos < 0 || EV_SIZE_MAX - len < (size_t)pos->pos || pos->pos + len > buf->total_len) return -1; chain = pos->internal_.chain; position = pos->internal_.pos_in_chain; while (len && chain) { size_t n_comparable; if (len + position > chain->off) n_comparable = chain->off - position; else n_comparable = len; r = memcmp(chain->buffer + chain->misalign + position, mem, n_comparable); if (r) return r; mem += n_comparable; len -= n_comparable; position = 0; chain = chain->next; } return 0; } struct evbuffer_ptr evbuffer_search(struct evbuffer *buffer, const char *what, size_t len, const struct evbuffer_ptr *start) { return evbuffer_search_range(buffer, what, len, start, NULL); } struct evbuffer_ptr evbuffer_search_range(struct evbuffer *buffer, const char *what, size_t len, const struct evbuffer_ptr *start, const struct evbuffer_ptr *end) { struct evbuffer_ptr pos; struct evbuffer_chain *chain, *last_chain = NULL; const unsigned char *p; char first; EVBUFFER_LOCK(buffer); if (start) { memcpy(&pos, start, sizeof(pos)); chain = pos.internal_.chain; } else { pos.pos = 0; chain = pos.internal_.chain = buffer->first; pos.internal_.pos_in_chain = 0; } if (end) last_chain = end->internal_.chain; if (!len || len > EV_SSIZE_MAX) goto done; first = what[0]; while (chain) { const unsigned char *start_at = chain->buffer + chain->misalign + pos.internal_.pos_in_chain; p = memchr(start_at, first, chain->off - pos.internal_.pos_in_chain); if (p) { pos.pos += p - start_at; pos.internal_.pos_in_chain += p - start_at; if (!evbuffer_ptr_memcmp(buffer, &pos, what, len)) { if (end && pos.pos + (ev_ssize_t)len > end->pos) goto not_found; else goto done; } ++pos.pos; ++pos.internal_.pos_in_chain; if (pos.internal_.pos_in_chain == chain->off) { chain = pos.internal_.chain = chain->next; pos.internal_.pos_in_chain = 0; } } else { if (chain == last_chain) goto not_found; pos.pos += chain->off - pos.internal_.pos_in_chain; chain = pos.internal_.chain = chain->next; pos.internal_.pos_in_chain = 0; } } not_found: PTR_NOT_FOUND(&pos); done: EVBUFFER_UNLOCK(buffer); return pos; } int evbuffer_peek(struct evbuffer *buffer, ev_ssize_t len, struct evbuffer_ptr *start_at, struct evbuffer_iovec *vec, int n_vec) { struct evbuffer_chain *chain; int idx = 0; ev_ssize_t len_so_far = 0; /* Avoid locking in trivial edge cases */ if (start_at && start_at->internal_.chain == NULL) return 0; EVBUFFER_LOCK(buffer); if (start_at) { chain = start_at->internal_.chain; len_so_far = chain->off - start_at->internal_.pos_in_chain; idx = 1; if (n_vec > 0) { vec[0].iov_base = (void *)(chain->buffer + chain->misalign + start_at->internal_.pos_in_chain); vec[0].iov_len = len_so_far; } chain = chain->next; } else { chain = buffer->first; } if (n_vec == 0 && len < 0) { /* If no vectors are provided and they asked for "everything", * pretend they asked for the actual available amount. */ len = buffer->total_len; if (start_at) { len -= start_at->pos; } } while (chain) { if (len >= 0 && len_so_far >= len) break; if (idxbuffer + chain->misalign); vec[idx].iov_len = chain->off; } else if (len<0) { break; } ++idx; len_so_far += chain->off; chain = chain->next; } EVBUFFER_UNLOCK(buffer); return idx; } int evbuffer_add_vprintf(struct evbuffer *buf, const char *fmt, va_list ap) { char *buffer; size_t space; int sz, result = -1; va_list aq; struct evbuffer_chain *chain; EVBUFFER_LOCK(buf); if (buf->freeze_end) { goto done; } /* make sure that at least some space is available */ if ((chain = evbuffer_expand_singlechain(buf, 64)) == NULL) goto done; for (;;) { #if 0 size_t used = chain->misalign + chain->off; buffer = (char *)chain->buffer + chain->misalign + chain->off; EVUTIL_ASSERT(chain->buffer_len >= used); space = chain->buffer_len - used; #endif buffer = (char*) CHAIN_SPACE_PTR(chain); space = (size_t) CHAIN_SPACE_LEN(chain); #ifndef va_copy #define va_copy(dst, src) memcpy(&(dst), &(src), sizeof(va_list)) #endif va_copy(aq, ap); sz = evutil_vsnprintf(buffer, space, fmt, aq); va_end(aq); if (sz < 0) goto done; if (INT_MAX >= EVBUFFER_CHAIN_MAX && (size_t)sz >= EVBUFFER_CHAIN_MAX) goto done; if ((size_t)sz < space) { chain->off += sz; buf->total_len += sz; buf->n_add_for_cb += sz; advance_last_with_data(buf); evbuffer_invoke_callbacks_(buf); result = sz; goto done; } if ((chain = evbuffer_expand_singlechain(buf, sz + 1)) == NULL) goto done; } /* NOTREACHED */ done: EVBUFFER_UNLOCK(buf); return result; } int evbuffer_add_printf(struct evbuffer *buf, const char *fmt, ...) { int res = -1; va_list ap; va_start(ap, fmt); res = evbuffer_add_vprintf(buf, fmt, ap); va_end(ap); return (res); } int evbuffer_add_reference(struct evbuffer *outbuf, const void *data, size_t datlen, evbuffer_ref_cleanup_cb cleanupfn, void *extra) { struct evbuffer_chain *chain; struct evbuffer_chain_reference *info; int result = -1; chain = evbuffer_chain_new(sizeof(struct evbuffer_chain_reference)); if (!chain) return (-1); chain->flags |= EVBUFFER_REFERENCE | EVBUFFER_IMMUTABLE; chain->buffer = (unsigned char *)data; chain->buffer_len = datlen; chain->off = datlen; info = EVBUFFER_CHAIN_EXTRA(struct evbuffer_chain_reference, chain); info->cleanupfn = cleanupfn; info->extra = extra; EVBUFFER_LOCK(outbuf); if (outbuf->freeze_end) { /* don't call chain_free; we do not want to actually invoke * the cleanup function */ mm_free(chain); goto done; } evbuffer_chain_insert(outbuf, chain); outbuf->n_add_for_cb += datlen; evbuffer_invoke_callbacks_(outbuf); result = 0; done: EVBUFFER_UNLOCK(outbuf); return result; } /* TODO(niels): we may want to add to automagically convert to mmap, in * case evbuffer_remove() or evbuffer_pullup() are being used. */ struct evbuffer_file_segment * evbuffer_file_segment_new( int fd, ev_off_t offset, ev_off_t length, unsigned flags) { struct evbuffer_file_segment *seg = mm_calloc(sizeof(struct evbuffer_file_segment), 1); if (!seg) return NULL; seg->refcnt = 1; seg->fd = fd; seg->flags = flags; seg->file_offset = offset; seg->cleanup_cb = NULL; seg->cleanup_cb_arg = NULL; #ifdef _WIN32 #ifndef lseek #define lseek _lseeki64 #endif #ifndef fstat #define fstat _fstat #endif #ifndef stat #define stat _stat #endif #endif if (length == -1) { struct stat st; if (fstat(fd, &st) < 0) goto err; length = st.st_size; } seg->length = length; if (offset < 0 || length < 0 || ((ev_uint64_t)length > EVBUFFER_CHAIN_MAX) || (ev_uint64_t)offset > (ev_uint64_t)(EVBUFFER_CHAIN_MAX - length)) goto err; #if defined(USE_SENDFILE) if (!(flags & EVBUF_FS_DISABLE_SENDFILE)) { seg->can_sendfile = 1; goto done; } #endif if (evbuffer_file_segment_materialize(seg)<0) goto err; #if defined(USE_SENDFILE) done: #endif if (!(flags & EVBUF_FS_DISABLE_LOCKING)) { EVTHREAD_ALLOC_LOCK(seg->lock, 0); } return seg; err: mm_free(seg); return NULL; } #ifdef EVENT__HAVE_MMAP static long get_page_size(void) { #ifdef SC_PAGE_SIZE return sysconf(SC_PAGE_SIZE); #elif defined(_SC_PAGE_SIZE) return sysconf(_SC_PAGE_SIZE); #else return 1; #endif } #endif /* DOCDOC */ /* Requires lock */ static int evbuffer_file_segment_materialize(struct evbuffer_file_segment *seg) { const unsigned flags = seg->flags; const int fd = seg->fd; const ev_off_t length = seg->length; const ev_off_t offset = seg->file_offset; if (seg->contents) return 0; /* already materialized */ #if defined(EVENT__HAVE_MMAP) if (!(flags & EVBUF_FS_DISABLE_MMAP)) { off_t offset_rounded = 0, offset_leftover = 0; void *mapped; if (offset) { /* mmap implementations don't generally like us * to have an offset that isn't a round */ long page_size = get_page_size(); if (page_size == -1) goto err; offset_leftover = offset % page_size; offset_rounded = offset - offset_leftover; } mapped = mmap(NULL, length + offset_leftover, PROT_READ, #ifdef MAP_NOCACHE MAP_NOCACHE | /* ??? */ #endif #ifdef MAP_FILE MAP_FILE | #endif MAP_PRIVATE, fd, offset_rounded); if (mapped == MAP_FAILED) { event_warn("%s: mmap(%d, %d, %zu) failed", __func__, fd, 0, (size_t)(offset + length)); } else { seg->mapping = mapped; seg->contents = (char*)mapped+offset_leftover; seg->mmap_offset = 0; seg->is_mapping = 1; goto done; } } #endif #ifdef _WIN32 if (!(flags & EVBUF_FS_DISABLE_MMAP)) { intptr_t h = _get_osfhandle(fd); HANDLE m; ev_uint64_t total_size = length+offset; if ((HANDLE)h == INVALID_HANDLE_VALUE) goto err; m = CreateFileMapping((HANDLE)h, NULL, PAGE_READONLY, (total_size >> 32), total_size & 0xfffffffful, NULL); if (m != INVALID_HANDLE_VALUE) { /* Does h leak? */ seg->mapping_handle = m; seg->mmap_offset = offset; seg->is_mapping = 1; goto done; } } #endif { ev_off_t start_pos = lseek(fd, 0, SEEK_CUR), pos; ev_off_t read_so_far = 0; char *mem; int e; ev_ssize_t n = 0; if (!(mem = mm_malloc(length))) goto err; if (start_pos < 0) { mm_free(mem); goto err; } if (lseek(fd, offset, SEEK_SET) < 0) { mm_free(mem); goto err; } while (read_so_far < length) { n = read(fd, mem+read_so_far, length-read_so_far); if (n <= 0) break; read_so_far += n; } e = errno; pos = lseek(fd, start_pos, SEEK_SET); if (n < 0 || (n == 0 && length > read_so_far)) { mm_free(mem); errno = e; goto err; } else if (pos < 0) { mm_free(mem); goto err; } seg->contents = mem; } done: return 0; err: return -1; } void evbuffer_file_segment_add_cleanup_cb(struct evbuffer_file_segment *seg, evbuffer_file_segment_cleanup_cb cb, void* arg) { EVUTIL_ASSERT(seg->refcnt > 0); seg->cleanup_cb = cb; seg->cleanup_cb_arg = arg; } void evbuffer_file_segment_free(struct evbuffer_file_segment *seg) { int refcnt; EVLOCK_LOCK(seg->lock, 0); refcnt = --seg->refcnt; EVLOCK_UNLOCK(seg->lock, 0); if (refcnt > 0) return; EVUTIL_ASSERT(refcnt == 0); if (seg->is_mapping) { #ifdef _WIN32 CloseHandle(seg->mapping_handle); #elif defined (EVENT__HAVE_MMAP) off_t offset_leftover; offset_leftover = seg->file_offset % get_page_size(); if (munmap(seg->mapping, seg->length + offset_leftover) == -1) event_warn("%s: munmap failed", __func__); #endif } else if (seg->contents) { mm_free(seg->contents); } if ((seg->flags & EVBUF_FS_CLOSE_ON_FREE) && seg->fd >= 0) { close(seg->fd); } if (seg->cleanup_cb) { (*seg->cleanup_cb)((struct evbuffer_file_segment const*)seg, seg->flags, seg->cleanup_cb_arg); seg->cleanup_cb = NULL; seg->cleanup_cb_arg = NULL; } EVTHREAD_FREE_LOCK(seg->lock, 0); mm_free(seg); } int evbuffer_add_file_segment(struct evbuffer *buf, struct evbuffer_file_segment *seg, ev_off_t offset, ev_off_t length) { struct evbuffer_chain *chain; struct evbuffer_chain_file_segment *extra; int can_use_sendfile = 0; EVBUFFER_LOCK(buf); EVLOCK_LOCK(seg->lock, 0); if (buf->flags & EVBUFFER_FLAG_DRAINS_TO_FD) { can_use_sendfile = 1; } else { if (!seg->contents) { if (evbuffer_file_segment_materialize(seg)<0) { EVLOCK_UNLOCK(seg->lock, 0); EVBUFFER_UNLOCK(buf); return -1; } } } EVLOCK_UNLOCK(seg->lock, 0); if (buf->freeze_end) goto err; if (length < 0) { if (offset > seg->length) goto err; length = seg->length - offset; } /* Can we actually add this? */ if (offset+length > seg->length) goto err; chain = evbuffer_chain_new(sizeof(struct evbuffer_chain_file_segment)); if (!chain) goto err; extra = EVBUFFER_CHAIN_EXTRA(struct evbuffer_chain_file_segment, chain); chain->flags |= EVBUFFER_IMMUTABLE|EVBUFFER_FILESEGMENT; if (can_use_sendfile && seg->can_sendfile) { chain->flags |= EVBUFFER_SENDFILE; chain->misalign = seg->file_offset + offset; chain->off = length; chain->buffer_len = chain->misalign + length; } else if (seg->is_mapping) { #ifdef _WIN32 ev_uint64_t total_offset = seg->mmap_offset+offset; ev_uint64_t offset_rounded=0, offset_remaining=0; LPVOID data; if (total_offset) { SYSTEM_INFO si; memset(&si, 0, sizeof(si)); /* cargo cult */ GetSystemInfo(&si); offset_remaining = total_offset % si.dwAllocationGranularity; offset_rounded = total_offset - offset_remaining; } data = MapViewOfFile( seg->mapping_handle, FILE_MAP_READ, offset_rounded >> 32, offset_rounded & 0xfffffffful, length + offset_remaining); if (data == NULL) { mm_free(chain); goto err; } chain->buffer = (unsigned char*) data; chain->buffer_len = length+offset_remaining; chain->misalign = offset_remaining; chain->off = length; #else chain->buffer = (unsigned char*)(seg->contents + offset); chain->buffer_len = length; chain->off = length; #endif } else { chain->buffer = (unsigned char*)(seg->contents + offset); chain->buffer_len = length; chain->off = length; } EVLOCK_LOCK(seg->lock, 0); ++seg->refcnt; EVLOCK_UNLOCK(seg->lock, 0); extra->segment = seg; buf->n_add_for_cb += length; evbuffer_chain_insert(buf, chain); evbuffer_invoke_callbacks_(buf); EVBUFFER_UNLOCK(buf); return 0; err: EVBUFFER_UNLOCK(buf); evbuffer_file_segment_free(seg); /* Lowers the refcount */ return -1; } int evbuffer_add_file(struct evbuffer *buf, int fd, ev_off_t offset, ev_off_t length) { struct evbuffer_file_segment *seg; unsigned flags = EVBUF_FS_CLOSE_ON_FREE; int r; seg = evbuffer_file_segment_new(fd, offset, length, flags); if (!seg) return -1; r = evbuffer_add_file_segment(buf, seg, 0, length); if (r == 0) evbuffer_file_segment_free(seg); return r; } int evbuffer_setcb(struct evbuffer *buffer, evbuffer_cb cb, void *cbarg) { EVBUFFER_LOCK(buffer); if (!LIST_EMPTY(&buffer->callbacks)) evbuffer_remove_all_callbacks(buffer); if (cb) { struct evbuffer_cb_entry *ent = evbuffer_add_cb(buffer, NULL, cbarg); if (!ent) { EVBUFFER_UNLOCK(buffer); return -1; } ent->cb.cb_obsolete = cb; ent->flags |= EVBUFFER_CB_OBSOLETE; } EVBUFFER_UNLOCK(buffer); return 0; } struct evbuffer_cb_entry * evbuffer_add_cb(struct evbuffer *buffer, evbuffer_cb_func cb, void *cbarg) { struct evbuffer_cb_entry *e; if (! (e = mm_calloc(1, sizeof(struct evbuffer_cb_entry)))) return NULL; EVBUFFER_LOCK(buffer); e->cb.cb_func = cb; e->cbarg = cbarg; e->flags = EVBUFFER_CB_ENABLED; LIST_INSERT_HEAD(&buffer->callbacks, e, next); EVBUFFER_UNLOCK(buffer); return e; } int evbuffer_remove_cb_entry(struct evbuffer *buffer, struct evbuffer_cb_entry *ent) { EVBUFFER_LOCK(buffer); LIST_REMOVE(ent, next); EVBUFFER_UNLOCK(buffer); mm_free(ent); return 0; } int evbuffer_remove_cb(struct evbuffer *buffer, evbuffer_cb_func cb, void *cbarg) { struct evbuffer_cb_entry *cbent; int result = -1; EVBUFFER_LOCK(buffer); LIST_FOREACH(cbent, &buffer->callbacks, next) { if (cb == cbent->cb.cb_func && cbarg == cbent->cbarg) { result = evbuffer_remove_cb_entry(buffer, cbent); goto done; } } done: EVBUFFER_UNLOCK(buffer); return result; } int evbuffer_cb_set_flags(struct evbuffer *buffer, struct evbuffer_cb_entry *cb, ev_uint32_t flags) { /* the user isn't allowed to mess with these. */ flags &= ~EVBUFFER_CB_INTERNAL_FLAGS; EVBUFFER_LOCK(buffer); cb->flags |= flags; EVBUFFER_UNLOCK(buffer); return 0; } int evbuffer_cb_clear_flags(struct evbuffer *buffer, struct evbuffer_cb_entry *cb, ev_uint32_t flags) { /* the user isn't allowed to mess with these. */ flags &= ~EVBUFFER_CB_INTERNAL_FLAGS; EVBUFFER_LOCK(buffer); cb->flags &= ~flags; EVBUFFER_UNLOCK(buffer); return 0; } int evbuffer_freeze(struct evbuffer *buffer, int start) { EVBUFFER_LOCK(buffer); if (start) buffer->freeze_start = 1; else buffer->freeze_end = 1; EVBUFFER_UNLOCK(buffer); return 0; } int evbuffer_unfreeze(struct evbuffer *buffer, int start) { EVBUFFER_LOCK(buffer); if (start) buffer->freeze_start = 0; else buffer->freeze_end = 0; EVBUFFER_UNLOCK(buffer); return 0; } #if 0 void evbuffer_cb_suspend(struct evbuffer *buffer, struct evbuffer_cb_entry *cb) { if (!(cb->flags & EVBUFFER_CB_SUSPENDED)) { cb->size_before_suspend = evbuffer_get_length(buffer); cb->flags |= EVBUFFER_CB_SUSPENDED; } } void evbuffer_cb_unsuspend(struct evbuffer *buffer, struct evbuffer_cb_entry *cb) { if ((cb->flags & EVBUFFER_CB_SUSPENDED)) { unsigned call = (cb->flags & EVBUFFER_CB_CALL_ON_UNSUSPEND); size_t sz = cb->size_before_suspend; cb->flags &= ~(EVBUFFER_CB_SUSPENDED| EVBUFFER_CB_CALL_ON_UNSUSPEND); cb->size_before_suspend = 0; if (call && (cb->flags & EVBUFFER_CB_ENABLED)) { cb->cb(buffer, sz, evbuffer_get_length(buffer), cb->cbarg); } } } #endif int evbuffer_get_callbacks_(struct evbuffer *buffer, struct event_callback **cbs, int max_cbs) { int r = 0; EVBUFFER_LOCK(buffer); if (buffer->deferred_cbs) { if (max_cbs < 1) { r = -1; goto done; } cbs[0] = &buffer->deferred; r = 1; } done: EVBUFFER_UNLOCK(buffer); return r; } /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #ifdef _WIN32 #include #include #include #endif #include #include "event2/util.h" #include "event2/bufferevent.h" #include "event2/buffer.h" #include "event2/bufferevent_struct.h" #include "event2/event.h" #include "event2/util.h" #include "event-internal.h" #include "log-internal.h" #include "mm-internal.h" #include "bufferevent-internal.h" #include "util-internal.h" #include "iocp-internal.h" #ifndef SO_UPDATE_CONNECT_CONTEXT /* Mingw is sometimes missing this */ #define SO_UPDATE_CONNECT_CONTEXT 0x7010 #endif /* prototypes */ static int be_async_enable(struct bufferevent *, short); static int be_async_disable(struct bufferevent *, short); static void be_async_destruct(struct bufferevent *); static int be_async_flush(struct bufferevent *, short, enum bufferevent_flush_mode); static int be_async_ctrl(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *); struct bufferevent_async { struct bufferevent_private bev; struct event_overlapped connect_overlapped; struct event_overlapped read_overlapped; struct event_overlapped write_overlapped; size_t read_in_progress; size_t write_in_progress; unsigned ok : 1; unsigned read_added : 1; unsigned write_added : 1; }; const struct bufferevent_ops bufferevent_ops_async = { "socket_async", evutil_offsetof(struct bufferevent_async, bev.bev), be_async_enable, be_async_disable, NULL, /* Unlink */ be_async_destruct, bufferevent_generic_adj_timeouts_, be_async_flush, be_async_ctrl, }; static inline void be_async_run_eventcb(struct bufferevent *bev, short what, int options) { bufferevent_run_eventcb_(bev, what, options|BEV_TRIG_DEFER_CALLBACKS); } static inline void be_async_trigger_nolock(struct bufferevent *bev, short what, int options) { bufferevent_trigger_nolock_(bev, what, options|BEV_TRIG_DEFER_CALLBACKS); } static inline int fatal_error(int err) { switch (err) { /* We may have already associated this fd with a port. * Let's hope it's this port, and that the error code * for doing this neer changes. */ case ERROR_INVALID_PARAMETER: return 0; } return 1; } static inline struct bufferevent_async * upcast(struct bufferevent *bev) { struct bufferevent_async *bev_a; if (!BEV_IS_ASYNC(bev)) return NULL; bev_a = EVUTIL_UPCAST(bev, struct bufferevent_async, bev.bev); return bev_a; } static inline struct bufferevent_async * upcast_connect(struct event_overlapped *eo) { struct bufferevent_async *bev_a; bev_a = EVUTIL_UPCAST(eo, struct bufferevent_async, connect_overlapped); EVUTIL_ASSERT(BEV_IS_ASYNC(&bev_a->bev.bev)); return bev_a; } static inline struct bufferevent_async * upcast_read(struct event_overlapped *eo) { struct bufferevent_async *bev_a; bev_a = EVUTIL_UPCAST(eo, struct bufferevent_async, read_overlapped); EVUTIL_ASSERT(BEV_IS_ASYNC(&bev_a->bev.bev)); return bev_a; } static inline struct bufferevent_async * upcast_write(struct event_overlapped *eo) { struct bufferevent_async *bev_a; bev_a = EVUTIL_UPCAST(eo, struct bufferevent_async, write_overlapped); EVUTIL_ASSERT(BEV_IS_ASYNC(&bev_a->bev.bev)); return bev_a; } static void bev_async_del_write(struct bufferevent_async *beva) { struct bufferevent *bev = &beva->bev.bev; if (beva->write_added) { beva->write_added = 0; event_base_del_virtual_(bev->ev_base); } } static void bev_async_del_read(struct bufferevent_async *beva) { struct bufferevent *bev = &beva->bev.bev; if (beva->read_added) { beva->read_added = 0; event_base_del_virtual_(bev->ev_base); } } static void bev_async_add_write(struct bufferevent_async *beva) { struct bufferevent *bev = &beva->bev.bev; if (!beva->write_added) { beva->write_added = 1; event_base_add_virtual_(bev->ev_base); } } static void bev_async_add_read(struct bufferevent_async *beva) { struct bufferevent *bev = &beva->bev.bev; if (!beva->read_added) { beva->read_added = 1; event_base_add_virtual_(bev->ev_base); } } static void bev_async_consider_writing(struct bufferevent_async *beva) { size_t at_most; int limit; struct bufferevent *bev = &beva->bev.bev; /* Don't write if there's a write in progress, or we do not * want to write, or when there's nothing left to write. */ if (beva->write_in_progress || beva->bev.connecting) return; if (!beva->ok || !(bev->enabled&EV_WRITE) || !evbuffer_get_length(bev->output)) { bev_async_del_write(beva); return; } at_most = evbuffer_get_length(bev->output); /* This is safe so long as bufferevent_get_write_max never returns * more than INT_MAX. That's true for now. XXXX */ limit = (int)bufferevent_get_write_max_(&beva->bev); if (at_most >= (size_t)limit && limit >= 0) at_most = limit; if (beva->bev.write_suspended) { bev_async_del_write(beva); return; } /* XXXX doesn't respect low-water mark very well. */ bufferevent_incref_(bev); if (evbuffer_launch_write_(bev->output, at_most, &beva->write_overlapped)) { bufferevent_decref_(bev); beva->ok = 0; be_async_run_eventcb(bev, BEV_EVENT_ERROR, 0); } else { beva->write_in_progress = at_most; bufferevent_decrement_write_buckets_(&beva->bev, at_most); bev_async_add_write(beva); } } static void bev_async_consider_reading(struct bufferevent_async *beva) { size_t cur_size; size_t read_high; size_t at_most; int limit; struct bufferevent *bev = &beva->bev.bev; /* Don't read if there is a read in progress, or we do not * want to read. */ if (beva->read_in_progress || beva->bev.connecting) return; if (!beva->ok || !(bev->enabled&EV_READ)) { bev_async_del_read(beva); return; } /* Don't read if we're full */ cur_size = evbuffer_get_length(bev->input); read_high = bev->wm_read.high; if (read_high) { if (cur_size >= read_high) { bev_async_del_read(beva); return; } at_most = read_high - cur_size; } else { at_most = 16384; /* FIXME totally magic. */ } /* XXXX This over-commits. */ /* XXXX see also not above on cast on bufferevent_get_write_max_() */ limit = (int)bufferevent_get_read_max_(&beva->bev); if (at_most >= (size_t)limit && limit >= 0) at_most = limit; if (beva->bev.read_suspended) { bev_async_del_read(beva); return; } bufferevent_incref_(bev); if (evbuffer_launch_read_(bev->input, at_most, &beva->read_overlapped)) { beva->ok = 0; be_async_run_eventcb(bev, BEV_EVENT_ERROR, 0); bufferevent_decref_(bev); } else { beva->read_in_progress = at_most; bufferevent_decrement_read_buckets_(&beva->bev, at_most); bev_async_add_read(beva); } return; } static void be_async_outbuf_callback(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent *bev = arg; struct bufferevent_async *bev_async = upcast(bev); /* If we added data to the outbuf and were not writing before, * we may want to write now. */ bufferevent_incref_and_lock_(bev); if (cbinfo->n_added) bev_async_consider_writing(bev_async); bufferevent_decref_and_unlock_(bev); } static void be_async_inbuf_callback(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent *bev = arg; struct bufferevent_async *bev_async = upcast(bev); /* If we drained data from the inbuf and were not reading before, * we may want to read now */ bufferevent_incref_and_lock_(bev); if (cbinfo->n_deleted) bev_async_consider_reading(bev_async); bufferevent_decref_and_unlock_(bev); } static int be_async_enable(struct bufferevent *buf, short what) { struct bufferevent_async *bev_async = upcast(buf); if (!bev_async->ok) return -1; if (bev_async->bev.connecting) { /* Don't launch anything during connection attempts. */ return 0; } if (what & EV_READ) BEV_RESET_GENERIC_READ_TIMEOUT(buf); if (what & EV_WRITE) BEV_RESET_GENERIC_WRITE_TIMEOUT(buf); /* If we newly enable reading or writing, and we aren't reading or writing already, consider launching a new read or write. */ if (what & EV_READ) bev_async_consider_reading(bev_async); if (what & EV_WRITE) bev_async_consider_writing(bev_async); return 0; } static int be_async_disable(struct bufferevent *bev, short what) { struct bufferevent_async *bev_async = upcast(bev); /* XXXX If we disable reading or writing, we may want to consider * canceling any in-progress read or write operation, though it might * not work. */ if (what & EV_READ) { BEV_DEL_GENERIC_READ_TIMEOUT(bev); bev_async_del_read(bev_async); } if (what & EV_WRITE) { BEV_DEL_GENERIC_WRITE_TIMEOUT(bev); bev_async_del_write(bev_async); } return 0; } static void be_async_destruct(struct bufferevent *bev) { struct bufferevent_async *bev_async = upcast(bev); struct bufferevent_private *bev_p = BEV_UPCAST(bev); evutil_socket_t fd; EVUTIL_ASSERT(!upcast(bev)->write_in_progress && !upcast(bev)->read_in_progress); bev_async_del_read(bev_async); bev_async_del_write(bev_async); fd = evbuffer_overlapped_get_fd_(bev->input); if (fd != (evutil_socket_t)EVUTIL_INVALID_SOCKET && (bev_p->options & BEV_OPT_CLOSE_ON_FREE)) { evutil_closesocket(fd); evbuffer_overlapped_set_fd_(bev->input, EVUTIL_INVALID_SOCKET); } } /* GetQueuedCompletionStatus doesn't reliably yield WSA error codes, so * we use WSAGetOverlappedResult to translate. */ static void bev_async_set_wsa_error(struct bufferevent *bev, struct event_overlapped *eo) { DWORD bytes, flags; evutil_socket_t fd; fd = evbuffer_overlapped_get_fd_(bev->input); WSAGetOverlappedResult(fd, &eo->overlapped, &bytes, FALSE, &flags); } static int be_async_flush(struct bufferevent *bev, short what, enum bufferevent_flush_mode mode) { return 0; } static void connect_complete(struct event_overlapped *eo, ev_uintptr_t key, ev_ssize_t nbytes, int ok) { struct bufferevent_async *bev_a = upcast_connect(eo); struct bufferevent *bev = &bev_a->bev.bev; evutil_socket_t sock; BEV_LOCK(bev); EVUTIL_ASSERT(bev_a->bev.connecting); bev_a->bev.connecting = 0; sock = evbuffer_overlapped_get_fd_(bev_a->bev.bev.input); /* XXXX Handle error? */ setsockopt(sock, SOL_SOCKET, SO_UPDATE_CONNECT_CONTEXT, NULL, 0); if (ok) bufferevent_async_set_connected_(bev); else bev_async_set_wsa_error(bev, eo); be_async_run_eventcb(bev, ok ? BEV_EVENT_CONNECTED : BEV_EVENT_ERROR, 0); event_base_del_virtual_(bev->ev_base); bufferevent_decref_and_unlock_(bev); } static void read_complete(struct event_overlapped *eo, ev_uintptr_t key, ev_ssize_t nbytes, int ok) { struct bufferevent_async *bev_a = upcast_read(eo); struct bufferevent *bev = &bev_a->bev.bev; short what = BEV_EVENT_READING; ev_ssize_t amount_unread; BEV_LOCK(bev); EVUTIL_ASSERT(bev_a->read_in_progress); amount_unread = bev_a->read_in_progress - nbytes; evbuffer_commit_read_(bev->input, nbytes); bev_a->read_in_progress = 0; if (amount_unread) bufferevent_decrement_read_buckets_(&bev_a->bev, -amount_unread); if (!ok) bev_async_set_wsa_error(bev, eo); if (bev_a->ok) { if (ok && nbytes) { BEV_RESET_GENERIC_READ_TIMEOUT(bev); be_async_trigger_nolock(bev, EV_READ, 0); bev_async_consider_reading(bev_a); } else if (!ok) { what |= BEV_EVENT_ERROR; bev_a->ok = 0; be_async_run_eventcb(bev, what, 0); } else if (!nbytes) { what |= BEV_EVENT_EOF; bev_a->ok = 0; be_async_run_eventcb(bev, what, 0); } } bufferevent_decref_and_unlock_(bev); } static void write_complete(struct event_overlapped *eo, ev_uintptr_t key, ev_ssize_t nbytes, int ok) { struct bufferevent_async *bev_a = upcast_write(eo); struct bufferevent *bev = &bev_a->bev.bev; short what = BEV_EVENT_WRITING; ev_ssize_t amount_unwritten; BEV_LOCK(bev); EVUTIL_ASSERT(bev_a->write_in_progress); amount_unwritten = bev_a->write_in_progress - nbytes; evbuffer_commit_write_(bev->output, nbytes); bev_a->write_in_progress = 0; if (amount_unwritten) bufferevent_decrement_write_buckets_(&bev_a->bev, -amount_unwritten); if (!ok) bev_async_set_wsa_error(bev, eo); if (bev_a->ok) { if (ok && nbytes) { BEV_RESET_GENERIC_WRITE_TIMEOUT(bev); be_async_trigger_nolock(bev, EV_WRITE, 0); bev_async_consider_writing(bev_a); } else if (!ok) { what |= BEV_EVENT_ERROR; bev_a->ok = 0; be_async_run_eventcb(bev, what, 0); } else if (!nbytes) { what |= BEV_EVENT_EOF; bev_a->ok = 0; be_async_run_eventcb(bev, what, 0); } } bufferevent_decref_and_unlock_(bev); } struct bufferevent * bufferevent_async_new_(struct event_base *base, evutil_socket_t fd, int options) { struct bufferevent_async *bev_a; struct bufferevent *bev; struct event_iocp_port *iocp; options |= BEV_OPT_THREADSAFE; if (!(iocp = event_base_get_iocp_(base))) return NULL; if (fd >= 0 && event_iocp_port_associate_(iocp, fd, 1)<0) { if (fatal_error(GetLastError())) return NULL; } if (!(bev_a = mm_calloc(1, sizeof(struct bufferevent_async)))) return NULL; bev = &bev_a->bev.bev; if (!(bev->input = evbuffer_overlapped_new_(fd))) { mm_free(bev_a); return NULL; } if (!(bev->output = evbuffer_overlapped_new_(fd))) { evbuffer_free(bev->input); mm_free(bev_a); return NULL; } if (bufferevent_init_common_(&bev_a->bev, base, &bufferevent_ops_async, options)<0) goto err; evbuffer_add_cb(bev->input, be_async_inbuf_callback, bev); evbuffer_add_cb(bev->output, be_async_outbuf_callback, bev); event_overlapped_init_(&bev_a->connect_overlapped, connect_complete); event_overlapped_init_(&bev_a->read_overlapped, read_complete); event_overlapped_init_(&bev_a->write_overlapped, write_complete); bufferevent_init_generic_timeout_cbs_(bev); bev_a->ok = fd >= 0; return bev; err: bufferevent_free(&bev_a->bev.bev); return NULL; } void bufferevent_async_set_connected_(struct bufferevent *bev) { struct bufferevent_async *bev_async = upcast(bev); bev_async->ok = 1; /* Now's a good time to consider reading/writing */ be_async_enable(bev, bev->enabled); } int bufferevent_async_can_connect_(struct bufferevent *bev) { const struct win32_extension_fns *ext = event_get_win32_extension_fns_(); if (BEV_IS_ASYNC(bev) && event_base_get_iocp_(bev->ev_base) && ext && ext->ConnectEx) return 1; return 0; } int bufferevent_async_connect_(struct bufferevent *bev, evutil_socket_t fd, const struct sockaddr *sa, int socklen) { BOOL rc; struct bufferevent_async *bev_async = upcast(bev); struct sockaddr_storage ss; const struct win32_extension_fns *ext = event_get_win32_extension_fns_(); EVUTIL_ASSERT(ext && ext->ConnectEx && fd >= 0 && sa != NULL); /* ConnectEx() requires that the socket be bound to an address * with bind() before using, otherwise it will fail. We attempt * to issue a bind() here, taking into account that the error * code is set to WSAEINVAL when the socket is already bound. */ memset(&ss, 0, sizeof(ss)); if (sa->sa_family == AF_INET) { struct sockaddr_in *sin = (struct sockaddr_in *)&ss; sin->sin_family = AF_INET; sin->sin_addr.s_addr = INADDR_ANY; } else if (sa->sa_family == AF_INET6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ss; sin6->sin6_family = AF_INET6; sin6->sin6_addr = in6addr_any; } else { /* Well, the user will have to bind() */ return -1; } if (bind(fd, (struct sockaddr *)&ss, sizeof(ss)) < 0 && WSAGetLastError() != WSAEINVAL) return -1; event_base_add_virtual_(bev->ev_base); bufferevent_incref_(bev); rc = ext->ConnectEx(fd, sa, socklen, NULL, 0, NULL, &bev_async->connect_overlapped.overlapped); if (rc || WSAGetLastError() == ERROR_IO_PENDING) return 0; event_base_del_virtual_(bev->ev_base); bufferevent_decref_(bev); return -1; } static int be_async_ctrl(struct bufferevent *bev, enum bufferevent_ctrl_op op, union bufferevent_ctrl_data *data) { switch (op) { case BEV_CTRL_GET_FD: data->fd = evbuffer_overlapped_get_fd_(bev->input); return 0; case BEV_CTRL_SET_FD: { struct bufferevent_async *bev_a = upcast(bev); struct event_iocp_port *iocp; if (data->fd == evbuffer_overlapped_get_fd_(bev->input)) return 0; if (!(iocp = event_base_get_iocp_(bev->ev_base))) return -1; if (event_iocp_port_associate_(iocp, data->fd, 1) < 0) { if (fatal_error(GetLastError())) return -1; } evbuffer_overlapped_set_fd_(bev->input, data->fd); evbuffer_overlapped_set_fd_(bev->output, data->fd); bev_a->ok = data->fd >= 0; return 0; } case BEV_CTRL_CANCEL_ALL: { struct bufferevent_async *bev_a = upcast(bev); evutil_socket_t fd = evbuffer_overlapped_get_fd_(bev->input); if (fd != (evutil_socket_t)EVUTIL_INVALID_SOCKET && (bev_a->bev.options & BEV_OPT_CLOSE_ON_FREE)) { closesocket(fd); evbuffer_overlapped_set_fd_(bev->input, EVUTIL_INVALID_SOCKET); } bev_a->ok = 0; return 0; } case BEV_CTRL_GET_UNDERLYING: default: return -1; } } /* * Copyright (c) 2002-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef _WIN32 #include #endif #include "event2/util.h" #include "event2/buffer.h" #include "event2/buffer_compat.h" #include "event2/bufferevent.h" #include "event2/bufferevent_struct.h" #include "event2/bufferevent_compat.h" #include "event2/event.h" #include "event-internal.h" #include "log-internal.h" #include "mm-internal.h" #include "bufferevent-internal.h" #include "evbuffer-internal.h" #include "util-internal.h" static void bufferevent_cancel_all_(struct bufferevent *bev); static void bufferevent_finalize_cb_(struct event_callback *evcb, void *arg_); void bufferevent_suspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); if (!bufev_private->read_suspended) bufev->be_ops->disable(bufev, EV_READ); bufev_private->read_suspended |= what; BEV_UNLOCK(bufev); } void bufferevent_unsuspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); bufev_private->read_suspended &= ~what; if (!bufev_private->read_suspended && (bufev->enabled & EV_READ)) bufev->be_ops->enable(bufev, EV_READ); BEV_UNLOCK(bufev); } void bufferevent_suspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); if (!bufev_private->write_suspended) bufev->be_ops->disable(bufev, EV_WRITE); bufev_private->write_suspended |= what; BEV_UNLOCK(bufev); } void bufferevent_unsuspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); bufev_private->write_suspended &= ~what; if (!bufev_private->write_suspended && (bufev->enabled & EV_WRITE)) bufev->be_ops->enable(bufev, EV_WRITE); BEV_UNLOCK(bufev); } /** * Sometimes bufferevent's implementation can overrun high watermarks * (one of examples is openssl) and in this case if the read callback * will not handle enough data do over condition above the read * callback will never be called again (due to suspend above). * * To avoid this we are scheduling read callback again here, but only * from the user callback to avoid multiple scheduling: * - when the data had been added to it * - when the data had been drained from it (user specified read callback) */ static void bufferevent_inbuf_wm_check(struct bufferevent *bev) { if (!bev->wm_read.high) return; if (!(bev->enabled & EV_READ)) return; if (evbuffer_get_length(bev->input) < bev->wm_read.high) return; bufferevent_trigger(bev, EV_READ, BEV_OPT_DEFER_CALLBACKS); } /* Callback to implement watermarks on the input buffer. Only enabled * if the watermark is set. */ static void bufferevent_inbuf_wm_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent *bufev = arg; size_t size; size = evbuffer_get_length(buf); if (size >= bufev->wm_read.high) bufferevent_wm_suspend_read(bufev); else bufferevent_wm_unsuspend_read(bufev); } static void bufferevent_run_deferred_callbacks_locked(struct event_callback *cb, void *arg) { struct bufferevent_private *bufev_private = arg; struct bufferevent *bufev = &bufev_private->bev; BEV_LOCK(bufev); if ((bufev_private->eventcb_pending & BEV_EVENT_CONNECTED) && bufev->errorcb) { /* The "connected" happened before any reads or writes, so send it first. */ bufev_private->eventcb_pending &= ~BEV_EVENT_CONNECTED; bufev->errorcb(bufev, BEV_EVENT_CONNECTED, bufev->cbarg); } if (bufev_private->readcb_pending && bufev->readcb) { bufev_private->readcb_pending = 0; bufev->readcb(bufev, bufev->cbarg); bufferevent_inbuf_wm_check(bufev); } if (bufev_private->writecb_pending && bufev->writecb) { bufev_private->writecb_pending = 0; bufev->writecb(bufev, bufev->cbarg); } if (bufev_private->eventcb_pending && bufev->errorcb) { short what = bufev_private->eventcb_pending; int err = bufev_private->errno_pending; bufev_private->eventcb_pending = 0; bufev_private->errno_pending = 0; EVUTIL_SET_SOCKET_ERROR(err); bufev->errorcb(bufev, what, bufev->cbarg); } bufferevent_decref_and_unlock_(bufev); } static void bufferevent_run_deferred_callbacks_unlocked(struct event_callback *cb, void *arg) { struct bufferevent_private *bufev_private = arg; struct bufferevent *bufev = &bufev_private->bev; BEV_LOCK(bufev); #define UNLOCKED(stmt) \ do { BEV_UNLOCK(bufev); stmt; BEV_LOCK(bufev); } while(0) if ((bufev_private->eventcb_pending & BEV_EVENT_CONNECTED) && bufev->errorcb) { /* The "connected" happened before any reads or writes, so send it first. */ bufferevent_event_cb errorcb = bufev->errorcb; void *cbarg = bufev->cbarg; bufev_private->eventcb_pending &= ~BEV_EVENT_CONNECTED; UNLOCKED(errorcb(bufev, BEV_EVENT_CONNECTED, cbarg)); } if (bufev_private->readcb_pending && bufev->readcb) { bufferevent_data_cb readcb = bufev->readcb; void *cbarg = bufev->cbarg; bufev_private->readcb_pending = 0; UNLOCKED(readcb(bufev, cbarg)); bufferevent_inbuf_wm_check(bufev); } if (bufev_private->writecb_pending && bufev->writecb) { bufferevent_data_cb writecb = bufev->writecb; void *cbarg = bufev->cbarg; bufev_private->writecb_pending = 0; UNLOCKED(writecb(bufev, cbarg)); } if (bufev_private->eventcb_pending && bufev->errorcb) { bufferevent_event_cb errorcb = bufev->errorcb; void *cbarg = bufev->cbarg; short what = bufev_private->eventcb_pending; int err = bufev_private->errno_pending; bufev_private->eventcb_pending = 0; bufev_private->errno_pending = 0; EVUTIL_SET_SOCKET_ERROR(err); UNLOCKED(errorcb(bufev,what,cbarg)); } bufferevent_decref_and_unlock_(bufev); #undef UNLOCKED } #define SCHEDULE_DEFERRED(bevp) \ do { \ if (event_deferred_cb_schedule_( \ (bevp)->bev.ev_base, \ &(bevp)->deferred)) \ bufferevent_incref_(&(bevp)->bev); \ } while (0) void bufferevent_run_readcb_(struct bufferevent *bufev, int options) { /* Requires that we hold the lock and a reference */ struct bufferevent_private *p = BEV_UPCAST(bufev); if (bufev->readcb == NULL) return; if ((p->options|options) & BEV_OPT_DEFER_CALLBACKS) { p->readcb_pending = 1; SCHEDULE_DEFERRED(p); } else { bufev->readcb(bufev, bufev->cbarg); bufferevent_inbuf_wm_check(bufev); } } void bufferevent_run_writecb_(struct bufferevent *bufev, int options) { /* Requires that we hold the lock and a reference */ struct bufferevent_private *p = BEV_UPCAST(bufev); if (bufev->writecb == NULL) return; if ((p->options|options) & BEV_OPT_DEFER_CALLBACKS) { p->writecb_pending = 1; SCHEDULE_DEFERRED(p); } else { bufev->writecb(bufev, bufev->cbarg); } } #define BEV_TRIG_ALL_OPTS ( \ BEV_TRIG_IGNORE_WATERMARKS| \ BEV_TRIG_DEFER_CALLBACKS \ ) void bufferevent_trigger(struct bufferevent *bufev, short iotype, int options) { bufferevent_incref_and_lock_(bufev); bufferevent_trigger_nolock_(bufev, iotype, options&BEV_TRIG_ALL_OPTS); bufferevent_decref_and_unlock_(bufev); } void bufferevent_run_eventcb_(struct bufferevent *bufev, short what, int options) { /* Requires that we hold the lock and a reference */ struct bufferevent_private *p = BEV_UPCAST(bufev); if (bufev->errorcb == NULL) return; if ((p->options|options) & BEV_OPT_DEFER_CALLBACKS) { p->eventcb_pending |= what; p->errno_pending = EVUTIL_SOCKET_ERROR(); SCHEDULE_DEFERRED(p); } else { bufev->errorcb(bufev, what, bufev->cbarg); } } void bufferevent_trigger_event(struct bufferevent *bufev, short what, int options) { bufferevent_incref_and_lock_(bufev); bufferevent_run_eventcb_(bufev, what, options&BEV_TRIG_ALL_OPTS); bufferevent_decref_and_unlock_(bufev); } int bufferevent_init_common_(struct bufferevent_private *bufev_private, struct event_base *base, const struct bufferevent_ops *ops, enum bufferevent_options options) { struct bufferevent *bufev = &bufev_private->bev; if (!bufev->input) { if ((bufev->input = evbuffer_new()) == NULL) goto err; } if (!bufev->output) { if ((bufev->output = evbuffer_new()) == NULL) goto err; } bufev_private->refcnt = 1; bufev->ev_base = base; /* Disable timeouts. */ evutil_timerclear(&bufev->timeout_read); evutil_timerclear(&bufev->timeout_write); bufev->be_ops = ops; if (bufferevent_ratelim_init_(bufev_private)) goto err; /* * Set to EV_WRITE so that using bufferevent_write is going to * trigger a callback. Reading needs to be explicitly enabled * because otherwise no data will be available. */ bufev->enabled = EV_WRITE; #ifndef EVENT__DISABLE_THREAD_SUPPORT if (options & BEV_OPT_THREADSAFE) { if (bufferevent_enable_locking_(bufev, NULL) < 0) goto err; } #endif if ((options & (BEV_OPT_DEFER_CALLBACKS|BEV_OPT_UNLOCK_CALLBACKS)) == BEV_OPT_UNLOCK_CALLBACKS) { event_warnx("UNLOCK_CALLBACKS requires DEFER_CALLBACKS"); goto err; } if (options & BEV_OPT_UNLOCK_CALLBACKS) event_deferred_cb_init_( &bufev_private->deferred, event_base_get_npriorities(base) / 2, bufferevent_run_deferred_callbacks_unlocked, bufev_private); else event_deferred_cb_init_( &bufev_private->deferred, event_base_get_npriorities(base) / 2, bufferevent_run_deferred_callbacks_locked, bufev_private); bufev_private->options = options; evbuffer_set_parent_(bufev->input, bufev); evbuffer_set_parent_(bufev->output, bufev); return 0; err: if (bufev->input) { evbuffer_free(bufev->input); bufev->input = NULL; } if (bufev->output) { evbuffer_free(bufev->output); bufev->output = NULL; } return -1; } void bufferevent_setcb(struct bufferevent *bufev, bufferevent_data_cb readcb, bufferevent_data_cb writecb, bufferevent_event_cb eventcb, void *cbarg) { BEV_LOCK(bufev); bufev->readcb = readcb; bufev->writecb = writecb; bufev->errorcb = eventcb; bufev->cbarg = cbarg; BEV_UNLOCK(bufev); } void bufferevent_getcb(struct bufferevent *bufev, bufferevent_data_cb *readcb_ptr, bufferevent_data_cb *writecb_ptr, bufferevent_event_cb *eventcb_ptr, void **cbarg_ptr) { BEV_LOCK(bufev); if (readcb_ptr) *readcb_ptr = bufev->readcb; if (writecb_ptr) *writecb_ptr = bufev->writecb; if (eventcb_ptr) *eventcb_ptr = bufev->errorcb; if (cbarg_ptr) *cbarg_ptr = bufev->cbarg; BEV_UNLOCK(bufev); } struct evbuffer * bufferevent_get_input(struct bufferevent *bufev) { return bufev->input; } struct evbuffer * bufferevent_get_output(struct bufferevent *bufev) { return bufev->output; } struct event_base * bufferevent_get_base(struct bufferevent *bufev) { return bufev->ev_base; } int bufferevent_get_priority(const struct bufferevent *bufev) { if (event_initialized(&bufev->ev_read)) { return event_get_priority(&bufev->ev_read); } else { return event_base_get_npriorities(bufev->ev_base) / 2; } } int bufferevent_write(struct bufferevent *bufev, const void *data, size_t size) { if (evbuffer_add(bufev->output, data, size) == -1) return (-1); return 0; } int bufferevent_write_buffer(struct bufferevent *bufev, struct evbuffer *buf) { if (evbuffer_add_buffer(bufev->output, buf) == -1) return (-1); return 0; } size_t bufferevent_read(struct bufferevent *bufev, void *data, size_t size) { return (evbuffer_remove(bufev->input, data, size)); } int bufferevent_read_buffer(struct bufferevent *bufev, struct evbuffer *buf) { return (evbuffer_add_buffer(buf, bufev->input)); } int bufferevent_enable(struct bufferevent *bufev, short event) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); short impl_events = event; int r = 0; bufferevent_incref_and_lock_(bufev); if (bufev_private->read_suspended) impl_events &= ~EV_READ; if (bufev_private->write_suspended) impl_events &= ~EV_WRITE; bufev->enabled |= event; if (impl_events && bufev->be_ops->enable(bufev, impl_events) < 0) r = -1; if (r) event_debug(("%s: cannot enable 0x%hx on %p", __func__, event, bufev)); bufferevent_decref_and_unlock_(bufev); return r; } int bufferevent_set_timeouts(struct bufferevent *bufev, const struct timeval *tv_read, const struct timeval *tv_write) { int r = 0; BEV_LOCK(bufev); if (tv_read) { bufev->timeout_read = *tv_read; } else { evutil_timerclear(&bufev->timeout_read); } if (tv_write) { bufev->timeout_write = *tv_write; } else { evutil_timerclear(&bufev->timeout_write); } if (bufev->be_ops->adj_timeouts) r = bufev->be_ops->adj_timeouts(bufev); BEV_UNLOCK(bufev); return r; } /* Obsolete; use bufferevent_set_timeouts */ void bufferevent_settimeout(struct bufferevent *bufev, int timeout_read, int timeout_write) { struct timeval tv_read, tv_write; struct timeval *ptv_read = NULL, *ptv_write = NULL; memset(&tv_read, 0, sizeof(tv_read)); memset(&tv_write, 0, sizeof(tv_write)); if (timeout_read) { tv_read.tv_sec = timeout_read; ptv_read = &tv_read; } if (timeout_write) { tv_write.tv_sec = timeout_write; ptv_write = &tv_write; } bufferevent_set_timeouts(bufev, ptv_read, ptv_write); } int bufferevent_disable_hard_(struct bufferevent *bufev, short event) { int r = 0; struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); bufev->enabled &= ~event; bufev_private->connecting = 0; if (bufev->be_ops->disable(bufev, event) < 0) r = -1; BEV_UNLOCK(bufev); return r; } int bufferevent_disable(struct bufferevent *bufev, short event) { int r = 0; BEV_LOCK(bufev); bufev->enabled &= ~event; if (bufev->be_ops->disable(bufev, event) < 0) r = -1; if (r) event_debug(("%s: cannot disable 0x%hx on %p", __func__, event, bufev)); BEV_UNLOCK(bufev); return r; } /* * Sets the water marks */ void bufferevent_setwatermark(struct bufferevent *bufev, short events, size_t lowmark, size_t highmark) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); if (events & EV_WRITE) { bufev->wm_write.low = lowmark; bufev->wm_write.high = highmark; } if (events & EV_READ) { bufev->wm_read.low = lowmark; bufev->wm_read.high = highmark; if (highmark) { /* There is now a new high-water mark for read. enable the callback if needed, and see if we should suspend/bufferevent_wm_unsuspend. */ if (bufev_private->read_watermarks_cb == NULL) { bufev_private->read_watermarks_cb = evbuffer_add_cb(bufev->input, bufferevent_inbuf_wm_cb, bufev); } evbuffer_cb_set_flags(bufev->input, bufev_private->read_watermarks_cb, EVBUFFER_CB_ENABLED|EVBUFFER_CB_NODEFER); if (evbuffer_get_length(bufev->input) >= highmark) bufferevent_wm_suspend_read(bufev); else if (evbuffer_get_length(bufev->input) < highmark) bufferevent_wm_unsuspend_read(bufev); } else { /* There is now no high-water mark for read. */ if (bufev_private->read_watermarks_cb) evbuffer_cb_clear_flags(bufev->input, bufev_private->read_watermarks_cb, EVBUFFER_CB_ENABLED); bufferevent_wm_unsuspend_read(bufev); } } BEV_UNLOCK(bufev); } int bufferevent_getwatermark(struct bufferevent *bufev, short events, size_t *lowmark, size_t *highmark) { if (events == EV_WRITE) { BEV_LOCK(bufev); if (lowmark) *lowmark = bufev->wm_write.low; if (highmark) *highmark = bufev->wm_write.high; BEV_UNLOCK(bufev); return 0; } if (events == EV_READ) { BEV_LOCK(bufev); if (lowmark) *lowmark = bufev->wm_read.low; if (highmark) *highmark = bufev->wm_read.high; BEV_UNLOCK(bufev); return 0; } return -1; } int bufferevent_flush(struct bufferevent *bufev, short iotype, enum bufferevent_flush_mode mode) { int r = -1; BEV_LOCK(bufev); if (bufev->be_ops->flush) r = bufev->be_ops->flush(bufev, iotype, mode); BEV_UNLOCK(bufev); return r; } void bufferevent_incref_and_lock_(struct bufferevent *bufev) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); ++bufev_private->refcnt; } #if 0 static void bufferevent_transfer_lock_ownership_(struct bufferevent *donor, struct bufferevent *recipient) { struct bufferevent_private *d = BEV_UPCAST(donor); struct bufferevent_private *r = BEV_UPCAST(recipient); if (d->lock != r->lock) return; if (r->own_lock) return; if (d->own_lock) { d->own_lock = 0; r->own_lock = 1; } } #endif int bufferevent_decref_and_unlock_(struct bufferevent *bufev) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); int n_cbs = 0; #define MAX_CBS 16 struct event_callback *cbs[MAX_CBS]; EVUTIL_ASSERT(bufev_private->refcnt > 0); if (--bufev_private->refcnt) { BEV_UNLOCK(bufev); return 0; } if (bufev->be_ops->unlink) bufev->be_ops->unlink(bufev); /* Okay, we're out of references. Let's finalize this once all the * callbacks are done running. */ cbs[0] = &bufev->ev_read.ev_evcallback; cbs[1] = &bufev->ev_write.ev_evcallback; cbs[2] = &bufev_private->deferred; n_cbs = 3; if (bufev_private->rate_limiting) { struct event *e = &bufev_private->rate_limiting->refill_bucket_event; if (event_initialized(e)) cbs[n_cbs++] = &e->ev_evcallback; } n_cbs += evbuffer_get_callbacks_(bufev->input, cbs+n_cbs, MAX_CBS-n_cbs); n_cbs += evbuffer_get_callbacks_(bufev->output, cbs+n_cbs, MAX_CBS-n_cbs); event_callback_finalize_many_(bufev->ev_base, n_cbs, cbs, bufferevent_finalize_cb_); #undef MAX_CBS BEV_UNLOCK(bufev); return 1; } static void bufferevent_finalize_cb_(struct event_callback *evcb, void *arg_) { struct bufferevent *bufev = arg_; struct bufferevent *underlying; struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); BEV_LOCK(bufev); underlying = bufferevent_get_underlying(bufev); /* Clean up the shared info */ if (bufev->be_ops->destruct) bufev->be_ops->destruct(bufev); /* XXX what happens if refcnt for these buffers is > 1? * The buffers can share a lock with this bufferevent object, * but the lock might be destroyed below. */ /* evbuffer will free the callbacks */ evbuffer_free(bufev->input); evbuffer_free(bufev->output); if (bufev_private->rate_limiting) { if (bufev_private->rate_limiting->group) bufferevent_remove_from_rate_limit_group_internal_(bufev,0); mm_free(bufev_private->rate_limiting); bufev_private->rate_limiting = NULL; } BEV_UNLOCK(bufev); if (bufev_private->own_lock) EVTHREAD_FREE_LOCK(bufev_private->lock, EVTHREAD_LOCKTYPE_RECURSIVE); /* Free the actual allocated memory. */ mm_free(((char*)bufev) - bufev->be_ops->mem_offset); /* Release the reference to underlying now that we no longer need the * reference to it. We wait this long mainly in case our lock is * shared with underlying. * * The 'destruct' function will also drop a reference to underlying * if BEV_OPT_CLOSE_ON_FREE is set. * * XXX Should we/can we just refcount evbuffer/bufferevent locks? * It would probably save us some headaches. */ if (underlying) bufferevent_decref_(underlying); } int bufferevent_decref(struct bufferevent *bufev) { BEV_LOCK(bufev); return bufferevent_decref_and_unlock_(bufev); } void bufferevent_free(struct bufferevent *bufev) { BEV_LOCK(bufev); bufferevent_setcb(bufev, NULL, NULL, NULL, NULL); bufferevent_cancel_all_(bufev); bufferevent_decref_and_unlock_(bufev); } void bufferevent_incref(struct bufferevent *bufev) { struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); /* XXX: now that this function is public, we might want to * - return the count from this function * - create a new function to atomically grab the current refcount */ BEV_LOCK(bufev); ++bufev_private->refcnt; BEV_UNLOCK(bufev); } int bufferevent_enable_locking_(struct bufferevent *bufev, void *lock) { #ifdef EVENT__DISABLE_THREAD_SUPPORT return -1; #else struct bufferevent *underlying; if (BEV_UPCAST(bufev)->lock) return -1; underlying = bufferevent_get_underlying(bufev); if (!lock && underlying && BEV_UPCAST(underlying)->lock) { lock = BEV_UPCAST(underlying)->lock; BEV_UPCAST(bufev)->lock = lock; BEV_UPCAST(bufev)->own_lock = 0; } else if (!lock) { EVTHREAD_ALLOC_LOCK(lock, EVTHREAD_LOCKTYPE_RECURSIVE); if (!lock) return -1; BEV_UPCAST(bufev)->lock = lock; BEV_UPCAST(bufev)->own_lock = 1; } else { BEV_UPCAST(bufev)->lock = lock; BEV_UPCAST(bufev)->own_lock = 0; } evbuffer_enable_locking(bufev->input, lock); evbuffer_enable_locking(bufev->output, lock); if (underlying && !BEV_UPCAST(underlying)->lock) bufferevent_enable_locking_(underlying, lock); return 0; #endif } int bufferevent_setfd(struct bufferevent *bev, evutil_socket_t fd) { union bufferevent_ctrl_data d; int res = -1; d.fd = fd; BEV_LOCK(bev); if (bev->be_ops->ctrl) res = bev->be_ops->ctrl(bev, BEV_CTRL_SET_FD, &d); if (res) event_debug(("%s: cannot set fd for %p to "EV_SOCK_FMT, __func__, bev, fd)); BEV_UNLOCK(bev); return res; } evutil_socket_t bufferevent_getfd(struct bufferevent *bev) { union bufferevent_ctrl_data d; int res = -1; d.fd = -1; BEV_LOCK(bev); if (bev->be_ops->ctrl) res = bev->be_ops->ctrl(bev, BEV_CTRL_GET_FD, &d); if (res) event_debug(("%s: cannot get fd for %p", __func__, bev)); BEV_UNLOCK(bev); return (res<0) ? -1 : d.fd; } enum bufferevent_options bufferevent_get_options_(struct bufferevent *bev) { struct bufferevent_private *bev_p = BEV_UPCAST(bev); enum bufferevent_options options; BEV_LOCK(bev); options = bev_p->options; BEV_UNLOCK(bev); return options; } static void bufferevent_cancel_all_(struct bufferevent *bev) { union bufferevent_ctrl_data d; memset(&d, 0, sizeof(d)); BEV_LOCK(bev); if (bev->be_ops->ctrl) bev->be_ops->ctrl(bev, BEV_CTRL_CANCEL_ALL, &d); BEV_UNLOCK(bev); } short bufferevent_get_enabled(struct bufferevent *bufev) { short r; BEV_LOCK(bufev); r = bufev->enabled; BEV_UNLOCK(bufev); return r; } struct bufferevent * bufferevent_get_underlying(struct bufferevent *bev) { union bufferevent_ctrl_data d; int res = -1; d.ptr = NULL; BEV_LOCK(bev); if (bev->be_ops->ctrl) res = bev->be_ops->ctrl(bev, BEV_CTRL_GET_UNDERLYING, &d); BEV_UNLOCK(bev); return (res<0) ? NULL : d.ptr; } static void bufferevent_generic_read_timeout_cb(evutil_socket_t fd, short event, void *ctx) { struct bufferevent *bev = ctx; bufferevent_incref_and_lock_(bev); bufferevent_disable(bev, EV_READ); bufferevent_run_eventcb_(bev, BEV_EVENT_TIMEOUT|BEV_EVENT_READING, 0); bufferevent_decref_and_unlock_(bev); } static void bufferevent_generic_write_timeout_cb(evutil_socket_t fd, short event, void *ctx) { struct bufferevent *bev = ctx; bufferevent_incref_and_lock_(bev); bufferevent_disable(bev, EV_WRITE); bufferevent_run_eventcb_(bev, BEV_EVENT_TIMEOUT|BEV_EVENT_WRITING, 0); bufferevent_decref_and_unlock_(bev); } void bufferevent_init_generic_timeout_cbs_(struct bufferevent *bev) { event_assign(&bev->ev_read, bev->ev_base, -1, EV_FINALIZE, bufferevent_generic_read_timeout_cb, bev); event_assign(&bev->ev_write, bev->ev_base, -1, EV_FINALIZE, bufferevent_generic_write_timeout_cb, bev); } int bufferevent_generic_adj_timeouts_(struct bufferevent *bev) { const short enabled = bev->enabled; struct bufferevent_private *bev_p = BEV_UPCAST(bev); int r1=0, r2=0; if ((enabled & EV_READ) && !bev_p->read_suspended && evutil_timerisset(&bev->timeout_read)) r1 = event_add(&bev->ev_read, &bev->timeout_read); else r1 = event_del(&bev->ev_read); if ((enabled & EV_WRITE) && !bev_p->write_suspended && evutil_timerisset(&bev->timeout_write) && evbuffer_get_length(bev->output)) r2 = event_add(&bev->ev_write, &bev->timeout_write); else r2 = event_del(&bev->ev_write); if (r1 < 0 || r2 < 0) return -1; return 0; } int bufferevent_generic_adj_existing_timeouts_(struct bufferevent *bev) { int r = 0; if (event_pending(&bev->ev_read, EV_READ, NULL)) { if (evutil_timerisset(&bev->timeout_read)) { if (bufferevent_add_event_(&bev->ev_read, &bev->timeout_read) < 0) r = -1; } else { event_remove_timer(&bev->ev_read); } } if (event_pending(&bev->ev_write, EV_WRITE, NULL)) { if (evutil_timerisset(&bev->timeout_write)) { if (bufferevent_add_event_(&bev->ev_write, &bev->timeout_write) < 0) r = -1; } else { event_remove_timer(&bev->ev_write); } } return r; } int bufferevent_add_event_(struct event *ev, const struct timeval *tv) { if (!evutil_timerisset(tv)) return event_add(ev, NULL); else return event_add(ev, tv); } /* For use by user programs only; internally, we should be calling either bufferevent_incref_and_lock_(), or BEV_LOCK. */ void bufferevent_lock(struct bufferevent *bev) { bufferevent_incref_and_lock_(bev); } void bufferevent_unlock(struct bufferevent *bev) { bufferevent_decref_and_unlock_(bev); } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * Copyright (c) 2002-2006 Niels Provos * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "evconfig-private.h" #include #include "event2/event-config.h" #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef _WIN32 #include #endif #include "event2/util.h" #include "event2/bufferevent.h" #include "event2/buffer.h" #include "event2/bufferevent_struct.h" #include "event2/event.h" #include "log-internal.h" #include "mm-internal.h" #include "bufferevent-internal.h" #include "util-internal.h" /* prototypes */ static int be_filter_enable(struct bufferevent *, short); static int be_filter_disable(struct bufferevent *, short); static void be_filter_unlink(struct bufferevent *); static void be_filter_destruct(struct bufferevent *); static void be_filter_readcb(struct bufferevent *, void *); static void be_filter_writecb(struct bufferevent *, void *); static void be_filter_eventcb(struct bufferevent *, short, void *); static int be_filter_flush(struct bufferevent *bufev, short iotype, enum bufferevent_flush_mode mode); static int be_filter_ctrl(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *); static void bufferevent_filtered_inbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg); static void bufferevent_filtered_outbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *info, void *arg); struct bufferevent_filtered { struct bufferevent_private bev; /** The bufferevent that we read/write filtered data from/to. */ struct bufferevent *underlying; /** A callback on our inbuf to notice somebory removes data */ struct evbuffer_cb_entry *inbuf_cb; /** A callback on our outbuf to notice when somebody adds data */ struct evbuffer_cb_entry *outbuf_cb; /** True iff we have received an EOF callback from the underlying * bufferevent. */ unsigned got_eof; /** Function to free context when we're done. */ void (*free_context)(void *); /** Input filter */ bufferevent_filter_cb process_in; /** Output filter */ bufferevent_filter_cb process_out; /** User-supplied argument to the filters. */ void *context; }; const struct bufferevent_ops bufferevent_ops_filter = { "filter", evutil_offsetof(struct bufferevent_filtered, bev.bev), be_filter_enable, be_filter_disable, be_filter_unlink, be_filter_destruct, bufferevent_generic_adj_timeouts_, be_filter_flush, be_filter_ctrl, }; /* Given a bufferevent that's really the bev filter of a bufferevent_filtered, * return that bufferevent_filtered. Returns NULL otherwise.*/ static inline struct bufferevent_filtered * upcast(struct bufferevent *bev) { struct bufferevent_filtered *bev_f; if (!BEV_IS_FILTER(bev)) return NULL; bev_f = (void*)( ((char*)bev) - evutil_offsetof(struct bufferevent_filtered, bev.bev)); EVUTIL_ASSERT(BEV_IS_FILTER(&bev_f->bev.bev)); return bev_f; } #define downcast(bev_f) (&(bev_f)->bev.bev) /** Return 1 iff bevf's underlying bufferevent's output buffer is at or * over its high watermark such that we should not write to it in a given * flush mode. */ static int be_underlying_writebuf_full(struct bufferevent_filtered *bevf, enum bufferevent_flush_mode state) { struct bufferevent *u = bevf->underlying; return state == BEV_NORMAL && u->wm_write.high && evbuffer_get_length(u->output) >= u->wm_write.high; } /** Return 1 if our input buffer is at or over its high watermark such that we * should not write to it in a given flush mode. */ static int be_readbuf_full(struct bufferevent_filtered *bevf, enum bufferevent_flush_mode state) { struct bufferevent *bufev = downcast(bevf); return state == BEV_NORMAL && bufev->wm_read.high && evbuffer_get_length(bufev->input) >= bufev->wm_read.high; } /* Filter to use when we're created with a NULL filter. */ static enum bufferevent_filter_result be_null_filter(struct evbuffer *src, struct evbuffer *dst, ev_ssize_t lim, enum bufferevent_flush_mode state, void *ctx) { (void)state; if (evbuffer_remove_buffer(src, dst, lim) >= 0) return BEV_OK; else return BEV_ERROR; } struct bufferevent * bufferevent_filter_new(struct bufferevent *underlying, bufferevent_filter_cb input_filter, bufferevent_filter_cb output_filter, int options, void (*free_context)(void *), void *ctx) { struct bufferevent_filtered *bufev_f; int tmp_options = options & ~BEV_OPT_THREADSAFE; if (!underlying) return NULL; if (!input_filter) input_filter = be_null_filter; if (!output_filter) output_filter = be_null_filter; bufev_f = mm_calloc(1, sizeof(struct bufferevent_filtered)); if (!bufev_f) return NULL; if (bufferevent_init_common_(&bufev_f->bev, underlying->ev_base, &bufferevent_ops_filter, tmp_options) < 0) { mm_free(bufev_f); return NULL; } if (options & BEV_OPT_THREADSAFE) { bufferevent_enable_locking_(downcast(bufev_f), NULL); } bufev_f->underlying = underlying; bufev_f->process_in = input_filter; bufev_f->process_out = output_filter; bufev_f->free_context = free_context; bufev_f->context = ctx; bufferevent_setcb(bufev_f->underlying, be_filter_readcb, be_filter_writecb, be_filter_eventcb, bufev_f); bufev_f->inbuf_cb = evbuffer_add_cb(downcast(bufev_f)->input, bufferevent_filtered_inbuf_cb, bufev_f); evbuffer_cb_clear_flags(downcast(bufev_f)->input, bufev_f->inbuf_cb, EVBUFFER_CB_ENABLED); bufev_f->outbuf_cb = evbuffer_add_cb(downcast(bufev_f)->output, bufferevent_filtered_outbuf_cb, bufev_f); bufferevent_init_generic_timeout_cbs_(downcast(bufev_f)); bufferevent_incref_(underlying); bufferevent_enable(underlying, EV_READ|EV_WRITE); bufferevent_suspend_read_(underlying, BEV_SUSPEND_FILT_READ); return downcast(bufev_f); } static void be_filter_unlink(struct bufferevent *bev) { struct bufferevent_filtered *bevf = upcast(bev); EVUTIL_ASSERT(bevf); if (bevf->bev.options & BEV_OPT_CLOSE_ON_FREE) { /* Yes, there is also a decref in bufferevent_decref_. * That decref corresponds to the incref when we set * underlying for the first time. This decref is an * extra one to remove the last reference. */ if (BEV_UPCAST(bevf->underlying)->refcnt < 2) { event_warnx("BEV_OPT_CLOSE_ON_FREE set on an " "bufferevent with too few references"); } else { bufferevent_free(bevf->underlying); } } else { if (bevf->underlying) { if (bevf->underlying->errorcb == be_filter_eventcb) bufferevent_setcb(bevf->underlying, NULL, NULL, NULL, NULL); bufferevent_unsuspend_read_(bevf->underlying, BEV_SUSPEND_FILT_READ); } } } static void be_filter_destruct(struct bufferevent *bev) { struct bufferevent_filtered *bevf = upcast(bev); EVUTIL_ASSERT(bevf); if (bevf->free_context) bevf->free_context(bevf->context); if (bevf->inbuf_cb) evbuffer_remove_cb_entry(bev->input, bevf->inbuf_cb); if (bevf->outbuf_cb) evbuffer_remove_cb_entry(bev->output, bevf->outbuf_cb); } static int be_filter_enable(struct bufferevent *bev, short event) { struct bufferevent_filtered *bevf = upcast(bev); if (event & EV_WRITE) BEV_RESET_GENERIC_WRITE_TIMEOUT(bev); if (event & EV_READ) { BEV_RESET_GENERIC_READ_TIMEOUT(bev); bufferevent_unsuspend_read_(bevf->underlying, BEV_SUSPEND_FILT_READ); } return 0; } static int be_filter_disable(struct bufferevent *bev, short event) { struct bufferevent_filtered *bevf = upcast(bev); if (event & EV_WRITE) BEV_DEL_GENERIC_WRITE_TIMEOUT(bev); if (event & EV_READ) { BEV_DEL_GENERIC_READ_TIMEOUT(bev); bufferevent_suspend_read_(bevf->underlying, BEV_SUSPEND_FILT_READ); } return 0; } static enum bufferevent_filter_result be_filter_process_input(struct bufferevent_filtered *bevf, enum bufferevent_flush_mode state, int *processed_out) { enum bufferevent_filter_result res; struct bufferevent *bev = downcast(bevf); if (state == BEV_NORMAL) { /* If we're in 'normal' mode, don't urge data on the filter * unless we're reading data and under our high-water mark.*/ if (!(bev->enabled & EV_READ) || be_readbuf_full(bevf, state)) return BEV_OK; } do { ev_ssize_t limit = -1; if (state == BEV_NORMAL && bev->wm_read.high) limit = bev->wm_read.high - evbuffer_get_length(bev->input); res = bevf->process_in(bevf->underlying->input, bev->input, limit, state, bevf->context); if (res == BEV_OK) *processed_out = 1; } while (res == BEV_OK && (bev->enabled & EV_READ) && evbuffer_get_length(bevf->underlying->input) && !be_readbuf_full(bevf, state)); if (*processed_out) BEV_RESET_GENERIC_READ_TIMEOUT(bev); return res; } static enum bufferevent_filter_result be_filter_process_output(struct bufferevent_filtered *bevf, enum bufferevent_flush_mode state, int *processed_out) { /* Requires references and lock: might call writecb */ enum bufferevent_filter_result res = BEV_OK; struct bufferevent *bufev = downcast(bevf); int again = 0; if (state == BEV_NORMAL) { /* If we're in 'normal' mode, don't urge data on the * filter unless we're writing data, and the underlying * bufferevent is accepting data, and we have data to * give the filter. If we're in 'flush' or 'finish', * call the filter no matter what. */ if (!(bufev->enabled & EV_WRITE) || be_underlying_writebuf_full(bevf, state) || !evbuffer_get_length(bufev->output)) return BEV_OK; } /* disable the callback that calls this function when the user adds to the output buffer. */ evbuffer_cb_clear_flags(bufev->output, bevf->outbuf_cb, EVBUFFER_CB_ENABLED); do { int processed = 0; again = 0; do { ev_ssize_t limit = -1; if (state == BEV_NORMAL && bevf->underlying->wm_write.high) limit = bevf->underlying->wm_write.high - evbuffer_get_length(bevf->underlying->output); res = bevf->process_out(downcast(bevf)->output, bevf->underlying->output, limit, state, bevf->context); if (res == BEV_OK) processed = *processed_out = 1; } while (/* Stop if the filter wasn't successful...*/ res == BEV_OK && /* Or if we aren't writing any more. */ (bufev->enabled & EV_WRITE) && /* Of if we have nothing more to write and we are * not flushing. */ evbuffer_get_length(bufev->output) && /* Or if we have filled the underlying output buffer. */ !be_underlying_writebuf_full(bevf,state)); if (processed) { /* call the write callback.*/ bufferevent_trigger_nolock_(bufev, EV_WRITE, 0); if (res == BEV_OK && (bufev->enabled & EV_WRITE) && evbuffer_get_length(bufev->output) && !be_underlying_writebuf_full(bevf, state)) { again = 1; } } } while (again); /* reenable the outbuf_cb */ evbuffer_cb_set_flags(bufev->output,bevf->outbuf_cb, EVBUFFER_CB_ENABLED); if (*processed_out) BEV_RESET_GENERIC_WRITE_TIMEOUT(bufev); return res; } /* Called when the size of our outbuf changes. */ static void bufferevent_filtered_outbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent_filtered *bevf = arg; struct bufferevent *bev = downcast(bevf); if (cbinfo->n_added) { int processed_any = 0; /* Somebody added more data to the output buffer. Try to * process it, if we should. */ bufferevent_incref_and_lock_(bev); be_filter_process_output(bevf, BEV_NORMAL, &processed_any); bufferevent_decref_and_unlock_(bev); } } static void be_filter_read_nolock_(struct bufferevent *underlying, void *me_) { struct bufferevent_filtered *bevf = me_; enum bufferevent_filter_result res; enum bufferevent_flush_mode state; struct bufferevent *bufev = downcast(bevf); struct bufferevent_private *bufev_private = BEV_UPCAST(bufev); int processed_any = 0; // It's possible our refcount is 0 at this point if another thread free'd our filterevent EVUTIL_ASSERT(bufev_private->refcnt >= 0); // If our refcount is > 0 if (bufev_private->refcnt > 0) { if (bevf->got_eof) state = BEV_FINISHED; else state = BEV_NORMAL; /* XXXX use return value */ res = be_filter_process_input(bevf, state, &processed_any); (void)res; /* XXX This should be in process_input, not here. There are * other places that can call process-input, and they should * force readcb calls as needed. */ if (processed_any) { bufferevent_trigger_nolock_(bufev, EV_READ, 0); if (evbuffer_get_length(underlying->input) > 0 && be_readbuf_full(bevf, state)) { /* data left in underlying buffer and filter input buffer * hit its read high watermark. * Schedule callback to avoid data gets stuck in underlying * input buffer. */ evbuffer_cb_set_flags(bufev->input, bevf->inbuf_cb, EVBUFFER_CB_ENABLED); } } } } /* Called when the size of our inbuf changes. */ static void bufferevent_filtered_inbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent_filtered *bevf = arg; enum bufferevent_flush_mode state; struct bufferevent *bev = downcast(bevf); BEV_LOCK(bev); if (bevf->got_eof) state = BEV_FINISHED; else state = BEV_NORMAL; if (!be_readbuf_full(bevf, state)) { /* opportunity to read data which was left in underlying * input buffer because filter input buffer hit read * high watermark. */ evbuffer_cb_clear_flags(bev->input, bevf->inbuf_cb, EVBUFFER_CB_ENABLED); if (evbuffer_get_length(bevf->underlying->input) > 0) be_filter_read_nolock_(bevf->underlying, bevf); } BEV_UNLOCK(bev); } /* Called when the underlying socket has read. */ static void be_filter_readcb(struct bufferevent *underlying, void *me_) { struct bufferevent_filtered *bevf = me_; struct bufferevent *bev = downcast(bevf); BEV_LOCK(bev); be_filter_read_nolock_(underlying, me_); BEV_UNLOCK(bev); } /* Called when the underlying socket has drained enough that we can write to it. */ static void be_filter_writecb(struct bufferevent *underlying, void *me_) { struct bufferevent_filtered *bevf = me_; struct bufferevent *bev = downcast(bevf); struct bufferevent_private *bufev_private = BEV_UPCAST(bev); int processed_any = 0; BEV_LOCK(bev); // It's possible our refcount is 0 at this point if another thread free'd our filterevent EVUTIL_ASSERT(bufev_private->refcnt >= 0); // If our refcount is > 0 if (bufev_private->refcnt > 0) { be_filter_process_output(bevf, BEV_NORMAL, &processed_any); } BEV_UNLOCK(bev); } /* Called when the underlying socket has given us an error */ static void be_filter_eventcb(struct bufferevent *underlying, short what, void *me_) { struct bufferevent_filtered *bevf = me_; struct bufferevent *bev = downcast(bevf); struct bufferevent_private *bufev_private = BEV_UPCAST(bev); BEV_LOCK(bev); // It's possible our refcount is 0 at this point if another thread free'd our filterevent EVUTIL_ASSERT(bufev_private->refcnt >= 0); // If our refcount is > 0 if (bufev_private->refcnt > 0) { /* All we can really to is tell our own eventcb. */ bufferevent_run_eventcb_(bev, what, 0); } BEV_UNLOCK(bev); } static int be_filter_flush(struct bufferevent *bufev, short iotype, enum bufferevent_flush_mode mode) { struct bufferevent_filtered *bevf = upcast(bufev); int processed_any = 0; EVUTIL_ASSERT(bevf); bufferevent_incref_and_lock_(bufev); if (iotype & EV_READ) { be_filter_process_input(bevf, mode, &processed_any); } if (iotype & EV_WRITE) { be_filter_process_output(bevf, mode, &processed_any); } /* XXX check the return value? */ /* XXX does this want to recursively call lower-level flushes? */ bufferevent_flush(bevf->underlying, iotype, mode); bufferevent_decref_and_unlock_(bufev); return processed_any; } static int be_filter_ctrl(struct bufferevent *bev, enum bufferevent_ctrl_op op, union bufferevent_ctrl_data *data) { struct bufferevent_filtered *bevf; switch (op) { case BEV_CTRL_GET_UNDERLYING: bevf = upcast(bev); data->ptr = bevf->underlying; return 0; case BEV_CTRL_SET_FD: case BEV_CTRL_GET_FD: bevf = upcast(bev); if (bevf->underlying && bevf->underlying->be_ops && bevf->underlying->be_ops->ctrl) { return (bevf->underlying->be_ops->ctrl)(bevf->underlying, op, data); } EVUTIL_FALLTHROUGH; case BEV_CTRL_CANCEL_ALL: EVUTIL_FALLTHROUGH; default: return -1; } return -1; } /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ // Get rid of OSX 10.7 and greater deprecation warnings. #if defined(__APPLE__) && defined(__clang__) #pragma clang diagnostic ignored "-Wdeprecated-declarations" #endif #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #ifdef _WIN32 #include #endif #include "event2/bufferevent.h" #include "event2/bufferevent_struct.h" #include "event2/bufferevent_ssl.h" #include "event2/buffer.h" #include "event2/event.h" #include "mm-internal.h" #include "bufferevent-internal.h" #include "log-internal.h" #include #include #include "openssl-compat.h" /* * Define an OpenSSL bio that targets a bufferevent. */ /* -------------------- A BIO is an OpenSSL abstraction that handles reading and writing data. The library will happily speak SSL over anything that implements a BIO interface. Here we define a BIO implementation that directs its output to a bufferevent. We'll want to use this only when none of OpenSSL's built-in IO mechanisms work for us. -------------------- */ /* every BIO type needs its own integer type value. */ #define BIO_TYPE_LIBEVENT 57 /* ???? Arguably, we should set BIO_TYPE_FILTER or BIO_TYPE_SOURCE_SINK on * this. */ #if 0 static void print_err(int val) { int err; printf("Error was %d\n", val); while ((err = ERR_get_error())) { const char *msg = (const char*)ERR_reason_error_string(err); const char *lib = (const char*)ERR_lib_error_string(err); const char *func = (const char*)ERR_func_error_string(err); printf("%s in %s %s\n", msg, lib, func); } } #else #define print_err(v) ((void)0) #endif /* Called to initialize a new BIO */ static int bio_bufferevent_new(BIO *b) { BIO_set_init(b, 0); BIO_set_data(b, NULL); /* We'll be putting the bufferevent in this field.*/ return 1; } /* Called to uninitialize the BIO. */ static int bio_bufferevent_free(BIO *b) { if (!b) return 0; if (BIO_get_shutdown(b)) { if (BIO_get_init(b) && BIO_get_data(b)) bufferevent_free(BIO_get_data(b)); BIO_free(b); } return 1; } /* Called to extract data from the BIO. */ static int bio_bufferevent_read(BIO *b, char *out, int outlen) { int r = 0; struct evbuffer *input; BIO_clear_retry_flags(b); if (!out) return 0; if (!BIO_get_data(b)) return -1; input = bufferevent_get_input(BIO_get_data(b)); if (evbuffer_get_length(input) == 0) { /* If there's no data to read, say so. */ BIO_set_retry_read(b); return -1; } else { r = evbuffer_remove(input, out, outlen); } return r; } /* Called to write data into the BIO */ static int bio_bufferevent_write(BIO *b, const char *in, int inlen) { struct bufferevent *bufev = BIO_get_data(b); struct evbuffer *output; size_t outlen; BIO_clear_retry_flags(b); if (!BIO_get_data(b)) return -1; output = bufferevent_get_output(bufev); outlen = evbuffer_get_length(output); /* Copy only as much data onto the output buffer as can fit under the * high-water mark. */ if (bufev->wm_write.high && bufev->wm_write.high <= (outlen+inlen)) { if (bufev->wm_write.high <= outlen) { /* If no data can fit, we'll need to retry later. */ BIO_set_retry_write(b); return -1; } inlen = bufev->wm_write.high - outlen; } EVUTIL_ASSERT(inlen > 0); evbuffer_add(output, in, inlen); return inlen; } /* Called to handle various requests */ static long bio_bufferevent_ctrl(BIO *b, int cmd, long num, void *ptr) { struct bufferevent *bufev = BIO_get_data(b); long ret = 1; switch (cmd) { case BIO_CTRL_GET_CLOSE: ret = BIO_get_shutdown(b); break; case BIO_CTRL_SET_CLOSE: BIO_set_shutdown(b, (int)num); break; case BIO_CTRL_PENDING: ret = evbuffer_get_length(bufferevent_get_input(bufev)) != 0; break; case BIO_CTRL_WPENDING: ret = evbuffer_get_length(bufferevent_get_output(bufev)) != 0; break; /* XXXX These two are given a special-case treatment because * of cargo-cultism. I should come up with a better reason. */ case BIO_CTRL_DUP: case BIO_CTRL_FLUSH: ret = 1; break; default: ret = 0; break; } return ret; } /* Called to write a string to the BIO */ static int bio_bufferevent_puts(BIO *b, const char *s) { return bio_bufferevent_write(b, s, strlen(s)); } /* Method table for the bufferevent BIO */ static BIO_METHOD *methods_bufferevent; /* Return the method table for the bufferevents BIO */ static BIO_METHOD * BIO_s_bufferevent(void) { if (methods_bufferevent == NULL) { methods_bufferevent = BIO_meth_new(BIO_TYPE_LIBEVENT, "bufferevent"); if (methods_bufferevent == NULL) return NULL; BIO_meth_set_write(methods_bufferevent, bio_bufferevent_write); BIO_meth_set_read(methods_bufferevent, bio_bufferevent_read); BIO_meth_set_puts(methods_bufferevent, bio_bufferevent_puts); BIO_meth_set_ctrl(methods_bufferevent, bio_bufferevent_ctrl); BIO_meth_set_create(methods_bufferevent, bio_bufferevent_new); BIO_meth_set_destroy(methods_bufferevent, bio_bufferevent_free); } return methods_bufferevent; } /* Create a new BIO to wrap communication around a bufferevent. If close_flag * is true, the bufferevent will be freed when the BIO is closed. */ static BIO * BIO_new_bufferevent(struct bufferevent *bufferevent) { BIO *result; if (!bufferevent) return NULL; if (!(result = BIO_new(BIO_s_bufferevent()))) return NULL; BIO_set_init(result, 1); BIO_set_data(result, bufferevent); /* We don't tell the BIO to close the bufferevent; we do it ourselves on * be_openssl_destruct() */ BIO_set_shutdown(result, 0); return result; } /* -------------------- Now, here's the OpenSSL-based implementation of bufferevent. The implementation comes in two flavors: one that connects its SSL object to an underlying bufferevent using a BIO_bufferevent, and one that has the SSL object connect to a socket directly. The latter should generally be faster, except on Windows, where your best bet is using a bufferevent_async. (OpenSSL supports many other BIO types, too. But we can't use any unless we have a good way to get notified when they become readable/writable.) -------------------- */ struct bio_data_counts { unsigned long n_written; unsigned long n_read; }; struct bufferevent_openssl { /* Shared fields with common bufferevent implementation code. If we were set up with an underlying bufferevent, we use the events here as timers only. If we have an SSL, then we use the events as socket events. */ struct bufferevent_private bev; /* An underlying bufferevent that we're directing our output to. If it's NULL, then we're connected to an fd, not an evbuffer. */ struct bufferevent *underlying; /* The SSL object doing our encryption. */ SSL *ssl; /* A callback that's invoked when data arrives on our outbuf so we know to write data to the SSL. */ struct evbuffer_cb_entry *outbuf_cb; /* A count of how much data the bios have read/written total. Used for rate-limiting. */ struct bio_data_counts counts; /* If this value is greater than 0, then the last SSL_write blocked, * and we need to try it again with this many bytes. */ ev_ssize_t last_write; #define NUM_ERRORS 3 ev_uint32_t errors[NUM_ERRORS]; /* When we next get available space, we should say "read" instead of "write". This can happen if there's a renegotiation during a read operation. */ unsigned read_blocked_on_write : 1; /* When we next get data, we should say "write" instead of "read". */ unsigned write_blocked_on_read : 1; /* Treat TCP close before SSL close on SSL >= v3 as clean EOF. */ unsigned allow_dirty_shutdown : 1; /* XXX */ unsigned n_errors : 2; /* Are we currently connecting, accepting, or doing IO? */ unsigned state : 2; /* If we reset fd, we sould reset state too */ unsigned old_state : 2; }; static int be_openssl_enable(struct bufferevent *, short); static int be_openssl_disable(struct bufferevent *, short); static void be_openssl_unlink(struct bufferevent *); static void be_openssl_destruct(struct bufferevent *); static int be_openssl_adj_timeouts(struct bufferevent *); static int be_openssl_flush(struct bufferevent *bufev, short iotype, enum bufferevent_flush_mode mode); static int be_openssl_ctrl(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *); const struct bufferevent_ops bufferevent_ops_openssl = { "ssl", evutil_offsetof(struct bufferevent_openssl, bev.bev), be_openssl_enable, be_openssl_disable, be_openssl_unlink, be_openssl_destruct, be_openssl_adj_timeouts, be_openssl_flush, be_openssl_ctrl, }; /* Given a bufferevent, return a pointer to the bufferevent_openssl that * contains it, if any. */ static inline struct bufferevent_openssl * upcast(struct bufferevent *bev) { struct bufferevent_openssl *bev_o; if (!BEV_IS_OPENSSL(bev)) return NULL; bev_o = (void*)( ((char*)bev) - evutil_offsetof(struct bufferevent_openssl, bev.bev)); EVUTIL_ASSERT(BEV_IS_OPENSSL(&bev_o->bev.bev)); return bev_o; } static inline void put_error(struct bufferevent_openssl *bev_ssl, unsigned long err) { if (bev_ssl->n_errors == NUM_ERRORS) return; /* The error type according to openssl is "unsigned long", but openssl never uses more than 32 bits of it. It _can't_ use more than 32 bits of it, since it needs to report errors on systems where long is only 32 bits. */ bev_ssl->errors[bev_ssl->n_errors++] = (ev_uint32_t) err; } /* Have the base communications channel (either the underlying bufferevent or * ev_read and ev_write) start reading. Take the read-blocked-on-write flag * into account. */ static int start_reading(struct bufferevent_openssl *bev_ssl) { if (bev_ssl->underlying) { bufferevent_unsuspend_read_(bev_ssl->underlying, BEV_SUSPEND_FILT_READ); return 0; } else { struct bufferevent *bev = &bev_ssl->bev.bev; int r; r = bufferevent_add_event_(&bev->ev_read, &bev->timeout_read); if (r == 0 && bev_ssl->read_blocked_on_write) r = bufferevent_add_event_(&bev->ev_write, &bev->timeout_write); return r; } } /* Have the base communications channel (either the underlying bufferevent or * ev_read and ev_write) start writing. Take the write-blocked-on-read flag * into account. */ static int start_writing(struct bufferevent_openssl *bev_ssl) { int r = 0; if (bev_ssl->underlying) { if (bev_ssl->write_blocked_on_read) { bufferevent_unsuspend_read_(bev_ssl->underlying, BEV_SUSPEND_FILT_READ); } } else { struct bufferevent *bev = &bev_ssl->bev.bev; r = bufferevent_add_event_(&bev->ev_write, &bev->timeout_write); if (!r && bev_ssl->write_blocked_on_read) r = bufferevent_add_event_(&bev->ev_read, &bev->timeout_read); } return r; } static void stop_reading(struct bufferevent_openssl *bev_ssl) { if (bev_ssl->write_blocked_on_read) return; if (bev_ssl->underlying) { bufferevent_suspend_read_(bev_ssl->underlying, BEV_SUSPEND_FILT_READ); } else { struct bufferevent *bev = &bev_ssl->bev.bev; event_del(&bev->ev_read); } } static void stop_writing(struct bufferevent_openssl *bev_ssl) { if (bev_ssl->read_blocked_on_write) return; if (bev_ssl->underlying) { bufferevent_unsuspend_read_(bev_ssl->underlying, BEV_SUSPEND_FILT_READ); } else { struct bufferevent *bev = &bev_ssl->bev.bev; event_del(&bev->ev_write); } } static int set_rbow(struct bufferevent_openssl *bev_ssl) { if (!bev_ssl->underlying) stop_reading(bev_ssl); bev_ssl->read_blocked_on_write = 1; return start_writing(bev_ssl); } static int set_wbor(struct bufferevent_openssl *bev_ssl) { if (!bev_ssl->underlying) stop_writing(bev_ssl); bev_ssl->write_blocked_on_read = 1; return start_reading(bev_ssl); } static int clear_rbow(struct bufferevent_openssl *bev_ssl) { struct bufferevent *bev = &bev_ssl->bev.bev; int r = 0; bev_ssl->read_blocked_on_write = 0; if (!(bev->enabled & EV_WRITE)) stop_writing(bev_ssl); if (bev->enabled & EV_READ) r = start_reading(bev_ssl); return r; } static int clear_wbor(struct bufferevent_openssl *bev_ssl) { struct bufferevent *bev = &bev_ssl->bev.bev; int r = 0; bev_ssl->write_blocked_on_read = 0; if (!(bev->enabled & EV_READ)) stop_reading(bev_ssl); if (bev->enabled & EV_WRITE) r = start_writing(bev_ssl); return r; } static void conn_closed(struct bufferevent_openssl *bev_ssl, int when, int errcode, int ret) { int event = BEV_EVENT_ERROR; int dirty_shutdown = 0; unsigned long err; switch (errcode) { case SSL_ERROR_ZERO_RETURN: /* Possibly a clean shutdown. */ if (SSL_get_shutdown(bev_ssl->ssl) & SSL_RECEIVED_SHUTDOWN) event = BEV_EVENT_EOF; else dirty_shutdown = 1; break; case SSL_ERROR_SYSCALL: /* IO error; possibly a dirty shutdown. */ if ((ret == 0 || ret == -1) && ERR_peek_error() == 0) dirty_shutdown = 1; put_error(bev_ssl, errcode); break; case SSL_ERROR_SSL: /* Protocol error. */ put_error(bev_ssl, errcode); break; case SSL_ERROR_WANT_X509_LOOKUP: /* XXXX handle this. */ put_error(bev_ssl, errcode); break; case SSL_ERROR_NONE: case SSL_ERROR_WANT_READ: case SSL_ERROR_WANT_WRITE: case SSL_ERROR_WANT_CONNECT: case SSL_ERROR_WANT_ACCEPT: default: /* should be impossible; treat as normal error. */ event_warnx("BUG: Unexpected OpenSSL error code %d", errcode); break; } while ((err = ERR_get_error())) { put_error(bev_ssl, err); } if (dirty_shutdown && bev_ssl->allow_dirty_shutdown) event = BEV_EVENT_EOF; stop_reading(bev_ssl); stop_writing(bev_ssl); /* when is BEV_EVENT_{READING|WRITING} */ event = when | event; bufferevent_run_eventcb_(&bev_ssl->bev.bev, event, 0); } static void init_bio_counts(struct bufferevent_openssl *bev_ssl) { BIO *rbio, *wbio; wbio = SSL_get_wbio(bev_ssl->ssl); bev_ssl->counts.n_written = wbio ? BIO_number_written(wbio) : 0; rbio = SSL_get_rbio(bev_ssl->ssl); bev_ssl->counts.n_read = rbio ? BIO_number_read(rbio) : 0; } static inline void decrement_buckets(struct bufferevent_openssl *bev_ssl) { unsigned long num_w = BIO_number_written(SSL_get_wbio(bev_ssl->ssl)); unsigned long num_r = BIO_number_read(SSL_get_rbio(bev_ssl->ssl)); /* These next two subtractions can wrap around. That's okay. */ unsigned long w = num_w - bev_ssl->counts.n_written; unsigned long r = num_r - bev_ssl->counts.n_read; if (w) bufferevent_decrement_write_buckets_(&bev_ssl->bev, w); if (r) bufferevent_decrement_read_buckets_(&bev_ssl->bev, r); bev_ssl->counts.n_written = num_w; bev_ssl->counts.n_read = num_r; } #define OP_MADE_PROGRESS 1 #define OP_BLOCKED 2 #define OP_ERR 4 /* Return a bitmask of OP_MADE_PROGRESS (if we read anything); OP_BLOCKED (if we're now blocked); and OP_ERR (if an error occurred). */ static int do_read(struct bufferevent_openssl *bev_ssl, int n_to_read) { /* Requires lock */ struct bufferevent *bev = &bev_ssl->bev.bev; struct evbuffer *input = bev->input; int r, n, i, n_used = 0, atmost; struct evbuffer_iovec space[2]; int result = 0; if (bev_ssl->bev.read_suspended) return 0; atmost = bufferevent_get_read_max_(&bev_ssl->bev); if (n_to_read > atmost) n_to_read = atmost; n = evbuffer_reserve_space(input, n_to_read, space, 2); if (n < 0) return OP_ERR; for (i=0; ibev.read_suspended) break; ERR_clear_error(); r = SSL_read(bev_ssl->ssl, space[i].iov_base, space[i].iov_len); if (r>0) { result |= OP_MADE_PROGRESS; if (bev_ssl->read_blocked_on_write) if (clear_rbow(bev_ssl) < 0) return OP_ERR | result; ++n_used; space[i].iov_len = r; decrement_buckets(bev_ssl); } else { int err = SSL_get_error(bev_ssl->ssl, r); print_err(err); switch (err) { case SSL_ERROR_WANT_READ: /* Can't read until underlying has more data. */ if (bev_ssl->read_blocked_on_write) if (clear_rbow(bev_ssl) < 0) return OP_ERR | result; break; case SSL_ERROR_WANT_WRITE: /* This read operation requires a write, and the * underlying is full */ if (!bev_ssl->read_blocked_on_write) if (set_rbow(bev_ssl) < 0) return OP_ERR | result; break; default: conn_closed(bev_ssl, BEV_EVENT_READING, err, r); break; } result |= OP_BLOCKED; break; /* out of the loop */ } } if (n_used) { evbuffer_commit_space(input, space, n_used); if (bev_ssl->underlying) BEV_RESET_GENERIC_READ_TIMEOUT(bev); } return result; } /* Return a bitmask of OP_MADE_PROGRESS (if we wrote anything); OP_BLOCKED (if we're now blocked); and OP_ERR (if an error occurred). */ static int do_write(struct bufferevent_openssl *bev_ssl, int atmost) { int i, r, n, n_written = 0; struct bufferevent *bev = &bev_ssl->bev.bev; struct evbuffer *output = bev->output; struct evbuffer_iovec space[8]; int result = 0; if (bev_ssl->last_write > 0) atmost = bev_ssl->last_write; else atmost = bufferevent_get_write_max_(&bev_ssl->bev); n = evbuffer_peek(output, atmost, NULL, space, 8); if (n < 0) return OP_ERR | result; if (n > 8) n = 8; for (i=0; i < n; ++i) { if (bev_ssl->bev.write_suspended) break; /* SSL_write will (reasonably) return 0 if we tell it to send 0 data. Skip this case so we don't interpret the result as an error */ if (space[i].iov_len == 0) continue; ERR_clear_error(); r = SSL_write(bev_ssl->ssl, space[i].iov_base, space[i].iov_len); if (r > 0) { result |= OP_MADE_PROGRESS; if (bev_ssl->write_blocked_on_read) if (clear_wbor(bev_ssl) < 0) return OP_ERR | result; n_written += r; bev_ssl->last_write = -1; decrement_buckets(bev_ssl); } else { int err = SSL_get_error(bev_ssl->ssl, r); print_err(err); switch (err) { case SSL_ERROR_WANT_WRITE: /* Can't read until underlying has more data. */ if (bev_ssl->write_blocked_on_read) if (clear_wbor(bev_ssl) < 0) return OP_ERR | result; bev_ssl->last_write = space[i].iov_len; break; case SSL_ERROR_WANT_READ: /* This read operation requires a write, and the * underlying is full */ if (!bev_ssl->write_blocked_on_read) if (set_wbor(bev_ssl) < 0) return OP_ERR | result; bev_ssl->last_write = space[i].iov_len; break; default: conn_closed(bev_ssl, BEV_EVENT_WRITING, err, r); bev_ssl->last_write = -1; break; } result |= OP_BLOCKED; break; } } if (n_written) { evbuffer_drain(output, n_written); if (bev_ssl->underlying) BEV_RESET_GENERIC_WRITE_TIMEOUT(bev); bufferevent_trigger_nolock_(bev, EV_WRITE, BEV_OPT_DEFER_CALLBACKS); } return result; } #define WRITE_FRAME 15000 #define READ_DEFAULT 4096 /* Try to figure out how many bytes to read; return 0 if we shouldn't be * reading. */ static int bytes_to_read(struct bufferevent_openssl *bev) { struct evbuffer *input = bev->bev.bev.input; struct event_watermark *wm = &bev->bev.bev.wm_read; int result = READ_DEFAULT; ev_ssize_t limit; /* XXX 99% of this is generic code that nearly all bufferevents will * want. */ if (bev->write_blocked_on_read) { return 0; } if (! (bev->bev.bev.enabled & EV_READ)) { return 0; } if (bev->bev.read_suspended) { return 0; } if (wm->high) { if (evbuffer_get_length(input) >= wm->high) { return 0; } result = wm->high - evbuffer_get_length(input); } else { result = READ_DEFAULT; } /* Respect the rate limit */ limit = bufferevent_get_read_max_(&bev->bev); if (result > limit) { result = limit; } return result; } /* Things look readable. If write is blocked on read, write till it isn't. * Read from the underlying buffer until we block or we hit our high-water * mark. */ static void consider_reading(struct bufferevent_openssl *bev_ssl) { int r; int n_to_read; int all_result_flags = 0; while (bev_ssl->write_blocked_on_read) { r = do_write(bev_ssl, WRITE_FRAME); if (r & (OP_BLOCKED|OP_ERR)) break; } if (bev_ssl->write_blocked_on_read) return; n_to_read = bytes_to_read(bev_ssl); while (n_to_read) { r = do_read(bev_ssl, n_to_read); all_result_flags |= r; if (r & (OP_BLOCKED|OP_ERR)) break; if (bev_ssl->bev.read_suspended) break; /* Read all pending data. This won't hit the network * again, and will (most importantly) put us in a state * where we don't need to read anything else until the * socket is readable again. It'll potentially make us * overrun our read high-watermark (somewhat * regrettable). The damage to the rate-limit has * already been done, since OpenSSL went and read a * whole SSL record anyway. */ n_to_read = SSL_pending(bev_ssl->ssl); /* XXX This if statement is actually a bad bug, added to avoid * XXX a worse bug. * * The bad bug: It can potentially cause resource unfairness * by reading too much data from the underlying bufferevent; * it can potentially cause read looping if the underlying * bufferevent is a bufferevent_pair and deferred callbacks * aren't used. * * The worse bug: If we didn't do this, then we would * potentially not read any more from bev_ssl->underlying * until more data arrived there, which could lead to us * waiting forever. */ if (!n_to_read && bev_ssl->underlying) n_to_read = bytes_to_read(bev_ssl); } if (all_result_flags & OP_MADE_PROGRESS) { struct bufferevent *bev = &bev_ssl->bev.bev; bufferevent_trigger_nolock_(bev, EV_READ, 0); } if (!bev_ssl->underlying) { /* Should be redundant, but let's avoid busy-looping */ if (bev_ssl->bev.read_suspended || !(bev_ssl->bev.bev.enabled & EV_READ)) { event_del(&bev_ssl->bev.bev.ev_read); } } } static void consider_writing(struct bufferevent_openssl *bev_ssl) { int r; struct evbuffer *output = bev_ssl->bev.bev.output; struct evbuffer *target = NULL; struct event_watermark *wm = NULL; while (bev_ssl->read_blocked_on_write) { r = do_read(bev_ssl, 1024); /* XXXX 1024 is a hack */ if (r & OP_MADE_PROGRESS) { struct bufferevent *bev = &bev_ssl->bev.bev; bufferevent_trigger_nolock_(bev, EV_READ, 0); } if (r & (OP_ERR|OP_BLOCKED)) break; } if (bev_ssl->read_blocked_on_write) return; if (bev_ssl->underlying) { target = bev_ssl->underlying->output; wm = &bev_ssl->underlying->wm_write; } while ((bev_ssl->bev.bev.enabled & EV_WRITE) && (! bev_ssl->bev.write_suspended) && evbuffer_get_length(output) && (!target || (! wm->high || evbuffer_get_length(target) < wm->high))) { int n_to_write; if (wm && wm->high) n_to_write = wm->high - evbuffer_get_length(target); else n_to_write = WRITE_FRAME; r = do_write(bev_ssl, n_to_write); if (r & (OP_BLOCKED|OP_ERR)) break; } if (!bev_ssl->underlying) { if (evbuffer_get_length(output) == 0) { event_del(&bev_ssl->bev.bev.ev_write); } else if (bev_ssl->bev.write_suspended || !(bev_ssl->bev.bev.enabled & EV_WRITE)) { /* Should be redundant, but let's avoid busy-looping */ event_del(&bev_ssl->bev.bev.ev_write); } } } static void be_openssl_readcb(struct bufferevent *bev_base, void *ctx) { struct bufferevent_openssl *bev_ssl = ctx; consider_reading(bev_ssl); } static void be_openssl_writecb(struct bufferevent *bev_base, void *ctx) { struct bufferevent_openssl *bev_ssl = ctx; consider_writing(bev_ssl); } static void be_openssl_eventcb(struct bufferevent *bev_base, short what, void *ctx) { struct bufferevent_openssl *bev_ssl = ctx; int event = 0; if (what & BEV_EVENT_EOF) { if (bev_ssl->allow_dirty_shutdown) event = BEV_EVENT_EOF; else event = BEV_EVENT_ERROR; } else if (what & BEV_EVENT_TIMEOUT) { /* We sure didn't set this. Propagate it to the user. */ event = what; } else if (what & BEV_EVENT_ERROR) { /* An error occurred on the connection. Propagate it to the user. */ event = what; } else if (what & BEV_EVENT_CONNECTED) { /* Ignore it. We're saying SSL_connect() already, which will eat it. */ } if (event) bufferevent_run_eventcb_(&bev_ssl->bev.bev, event, 0); } static void be_openssl_readeventcb(evutil_socket_t fd, short what, void *ptr) { struct bufferevent_openssl *bev_ssl = ptr; bufferevent_incref_and_lock_(&bev_ssl->bev.bev); if (what == EV_TIMEOUT) { bufferevent_run_eventcb_(&bev_ssl->bev.bev, BEV_EVENT_TIMEOUT|BEV_EVENT_READING, 0); } else { consider_reading(bev_ssl); } bufferevent_decref_and_unlock_(&bev_ssl->bev.bev); } static void be_openssl_writeeventcb(evutil_socket_t fd, short what, void *ptr) { struct bufferevent_openssl *bev_ssl = ptr; bufferevent_incref_and_lock_(&bev_ssl->bev.bev); if (what == EV_TIMEOUT) { bufferevent_run_eventcb_(&bev_ssl->bev.bev, BEV_EVENT_TIMEOUT|BEV_EVENT_WRITING, 0); } else { consider_writing(bev_ssl); } bufferevent_decref_and_unlock_(&bev_ssl->bev.bev); } static evutil_socket_t be_openssl_auto_fd(struct bufferevent_openssl *bev_ssl, evutil_socket_t fd) { if (!bev_ssl->underlying) { struct bufferevent *bev = &bev_ssl->bev.bev; if (event_initialized(&bev->ev_read) && fd < 0) { fd = event_get_fd(&bev->ev_read); } } return fd; } static int set_open_callbacks(struct bufferevent_openssl *bev_ssl, evutil_socket_t fd) { if (bev_ssl->underlying) { bufferevent_setcb(bev_ssl->underlying, be_openssl_readcb, be_openssl_writecb, be_openssl_eventcb, bev_ssl); return 0; } else { struct bufferevent *bev = &bev_ssl->bev.bev; int rpending=0, wpending=0, r1=0, r2=0; if (event_initialized(&bev->ev_read)) { rpending = event_pending(&bev->ev_read, EV_READ, NULL); wpending = event_pending(&bev->ev_write, EV_WRITE, NULL); event_del(&bev->ev_read); event_del(&bev->ev_write); } event_assign(&bev->ev_read, bev->ev_base, fd, EV_READ|EV_PERSIST|EV_FINALIZE, be_openssl_readeventcb, bev_ssl); event_assign(&bev->ev_write, bev->ev_base, fd, EV_WRITE|EV_PERSIST|EV_FINALIZE, be_openssl_writeeventcb, bev_ssl); if (rpending) r1 = bufferevent_add_event_(&bev->ev_read, &bev->timeout_read); if (wpending) r2 = bufferevent_add_event_(&bev->ev_write, &bev->timeout_write); return (r1<0 || r2<0) ? -1 : 0; } } static int do_handshake(struct bufferevent_openssl *bev_ssl) { int r; switch (bev_ssl->state) { default: case BUFFEREVENT_SSL_OPEN: EVUTIL_ASSERT(0); return -1; case BUFFEREVENT_SSL_CONNECTING: case BUFFEREVENT_SSL_ACCEPTING: ERR_clear_error(); r = SSL_do_handshake(bev_ssl->ssl); break; } decrement_buckets(bev_ssl); if (r==1) { evutil_socket_t fd = event_get_fd(&bev_ssl->bev.bev.ev_read); /* We're done! */ bev_ssl->state = BUFFEREVENT_SSL_OPEN; set_open_callbacks(bev_ssl, fd); /* XXXX handle failure */ /* Call do_read and do_write as needed */ bufferevent_enable(&bev_ssl->bev.bev, bev_ssl->bev.bev.enabled); bufferevent_run_eventcb_(&bev_ssl->bev.bev, BEV_EVENT_CONNECTED, 0); return 1; } else { int err = SSL_get_error(bev_ssl->ssl, r); print_err(err); switch (err) { case SSL_ERROR_WANT_WRITE: stop_reading(bev_ssl); return start_writing(bev_ssl); case SSL_ERROR_WANT_READ: stop_writing(bev_ssl); return start_reading(bev_ssl); default: conn_closed(bev_ssl, BEV_EVENT_READING, err, r); return -1; } } } static void be_openssl_handshakecb(struct bufferevent *bev_base, void *ctx) { struct bufferevent_openssl *bev_ssl = ctx; do_handshake(bev_ssl);/* XXX handle failure */ } static void be_openssl_handshakeeventcb(evutil_socket_t fd, short what, void *ptr) { struct bufferevent_openssl *bev_ssl = ptr; bufferevent_incref_and_lock_(&bev_ssl->bev.bev); if (what & EV_TIMEOUT) { bufferevent_run_eventcb_(&bev_ssl->bev.bev, BEV_EVENT_TIMEOUT, 0); } else do_handshake(bev_ssl);/* XXX handle failure */ bufferevent_decref_and_unlock_(&bev_ssl->bev.bev); } static int set_handshake_callbacks(struct bufferevent_openssl *bev_ssl, evutil_socket_t fd) { if (bev_ssl->underlying) { bufferevent_setcb(bev_ssl->underlying, be_openssl_handshakecb, be_openssl_handshakecb, be_openssl_eventcb, bev_ssl); if (fd < 0) return 0; if (bufferevent_setfd(bev_ssl->underlying, fd)) return 1; return do_handshake(bev_ssl); } else { struct bufferevent *bev = &bev_ssl->bev.bev; if (event_initialized(&bev->ev_read)) { event_del(&bev->ev_read); event_del(&bev->ev_write); } event_assign(&bev->ev_read, bev->ev_base, fd, EV_READ|EV_PERSIST|EV_FINALIZE, be_openssl_handshakeeventcb, bev_ssl); event_assign(&bev->ev_write, bev->ev_base, fd, EV_WRITE|EV_PERSIST|EV_FINALIZE, be_openssl_handshakeeventcb, bev_ssl); if (fd >= 0) bufferevent_enable(bev, bev->enabled); return 0; } } int bufferevent_ssl_renegotiate(struct bufferevent *bev) { struct bufferevent_openssl *bev_ssl = upcast(bev); if (!bev_ssl) return -1; if (SSL_renegotiate(bev_ssl->ssl) < 0) return -1; bev_ssl->state = BUFFEREVENT_SSL_CONNECTING; if (set_handshake_callbacks(bev_ssl, be_openssl_auto_fd(bev_ssl, -1)) < 0) return -1; if (!bev_ssl->underlying) return do_handshake(bev_ssl); return 0; } static void be_openssl_outbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent_openssl *bev_ssl = arg; int r = 0; /* XXX need to hold a reference here. */ if (cbinfo->n_added && bev_ssl->state == BUFFEREVENT_SSL_OPEN) { if (cbinfo->orig_size == 0) r = bufferevent_add_event_(&bev_ssl->bev.bev.ev_write, &bev_ssl->bev.bev.timeout_write); if (bev_ssl->underlying) consider_writing(bev_ssl); } /* XXX Handle r < 0 */ (void)r; } static int be_openssl_enable(struct bufferevent *bev, short events) { struct bufferevent_openssl *bev_ssl = upcast(bev); int r1 = 0, r2 = 0; if (events & EV_READ) r1 = start_reading(bev_ssl); if (events & EV_WRITE) r2 = start_writing(bev_ssl); if (bev_ssl->underlying) { if (events & EV_READ) BEV_RESET_GENERIC_READ_TIMEOUT(bev); if (events & EV_WRITE) BEV_RESET_GENERIC_WRITE_TIMEOUT(bev); if (events & EV_READ) consider_reading(bev_ssl); if (events & EV_WRITE) consider_writing(bev_ssl); } return (r1 < 0 || r2 < 0) ? -1 : 0; } static int be_openssl_disable(struct bufferevent *bev, short events) { struct bufferevent_openssl *bev_ssl = upcast(bev); if (events & EV_READ) stop_reading(bev_ssl); if (events & EV_WRITE) stop_writing(bev_ssl); if (bev_ssl->underlying) { if (events & EV_READ) BEV_DEL_GENERIC_READ_TIMEOUT(bev); if (events & EV_WRITE) BEV_DEL_GENERIC_WRITE_TIMEOUT(bev); } return 0; } static void be_openssl_unlink(struct bufferevent *bev) { struct bufferevent_openssl *bev_ssl = upcast(bev); if (bev_ssl->bev.options & BEV_OPT_CLOSE_ON_FREE) { if (bev_ssl->underlying) { if (BEV_UPCAST(bev_ssl->underlying)->refcnt < 2) { event_warnx("BEV_OPT_CLOSE_ON_FREE set on an " "bufferevent with too few references"); } else { bufferevent_free(bev_ssl->underlying); /* We still have a reference to it, via our * BIO. So we don't drop this. */ // bev_ssl->underlying = NULL; } } } else { if (bev_ssl->underlying) { if (bev_ssl->underlying->errorcb == be_openssl_eventcb) bufferevent_setcb(bev_ssl->underlying, NULL,NULL,NULL,NULL); bufferevent_unsuspend_read_(bev_ssl->underlying, BEV_SUSPEND_FILT_READ); } } } static void be_openssl_destruct(struct bufferevent *bev) { struct bufferevent_openssl *bev_ssl = upcast(bev); if (bev_ssl->bev.options & BEV_OPT_CLOSE_ON_FREE) { if (! bev_ssl->underlying) { evutil_socket_t fd = EVUTIL_INVALID_SOCKET; BIO *bio = SSL_get_wbio(bev_ssl->ssl); if (bio) fd = BIO_get_fd(bio, NULL); if (fd >= 0) evutil_closesocket(fd); } SSL_free(bev_ssl->ssl); } } static int be_openssl_adj_timeouts(struct bufferevent *bev) { struct bufferevent_openssl *bev_ssl = upcast(bev); if (bev_ssl->underlying) { return bufferevent_generic_adj_timeouts_(bev); } else { return bufferevent_generic_adj_existing_timeouts_(bev); } } static int be_openssl_flush(struct bufferevent *bufev, short iotype, enum bufferevent_flush_mode mode) { /* XXXX Implement this. */ return 0; } static int be_openssl_set_fd(struct bufferevent_openssl *bev_ssl, enum bufferevent_ssl_state state, evutil_socket_t fd) { bev_ssl->state = state; switch (state) { case BUFFEREVENT_SSL_ACCEPTING: if (!SSL_clear(bev_ssl->ssl)) return -1; SSL_set_accept_state(bev_ssl->ssl); if (set_handshake_callbacks(bev_ssl, fd) < 0) return -1; break; case BUFFEREVENT_SSL_CONNECTING: if (!SSL_clear(bev_ssl->ssl)) return -1; SSL_set_connect_state(bev_ssl->ssl); if (set_handshake_callbacks(bev_ssl, fd) < 0) return -1; break; case BUFFEREVENT_SSL_OPEN: if (set_open_callbacks(bev_ssl, fd) < 0) return -1; break; default: return -1; } return 0; } static int be_openssl_ctrl(struct bufferevent *bev, enum bufferevent_ctrl_op op, union bufferevent_ctrl_data *data) { struct bufferevent_openssl *bev_ssl = upcast(bev); switch (op) { case BEV_CTRL_SET_FD: if (!bev_ssl->underlying) { BIO *bio; bio = BIO_new_socket((int)data->fd, 0); SSL_set_bio(bev_ssl->ssl, bio, bio); } else { BIO *bio; if (!(bio = BIO_new_bufferevent(bev_ssl->underlying))) return -1; SSL_set_bio(bev_ssl->ssl, bio, bio); } return be_openssl_set_fd(bev_ssl, bev_ssl->old_state, data->fd); case BEV_CTRL_GET_FD: if (bev_ssl->underlying) { data->fd = event_get_fd(&bev_ssl->underlying->ev_read); } else { data->fd = event_get_fd(&bev->ev_read); } return 0; case BEV_CTRL_GET_UNDERLYING: data->ptr = bev_ssl->underlying; return 0; case BEV_CTRL_CANCEL_ALL: default: return -1; } } SSL * bufferevent_openssl_get_ssl(struct bufferevent *bufev) { struct bufferevent_openssl *bev_ssl = upcast(bufev); if (!bev_ssl) return NULL; return bev_ssl->ssl; } static struct bufferevent * bufferevent_openssl_new_impl(struct event_base *base, struct bufferevent *underlying, evutil_socket_t fd, SSL *ssl, enum bufferevent_ssl_state state, int options) { struct bufferevent_openssl *bev_ssl = NULL; struct bufferevent_private *bev_p = NULL; int tmp_options = options & ~BEV_OPT_THREADSAFE; /* Only one can be set. */ if (underlying != NULL && fd >= 0) goto err; if (!(bev_ssl = mm_calloc(1, sizeof(struct bufferevent_openssl)))) goto err; bev_p = &bev_ssl->bev; if (bufferevent_init_common_(bev_p, base, &bufferevent_ops_openssl, tmp_options) < 0) goto err; /* Don't explode if we decide to realloc a chunk we're writing from in * the output buffer. */ SSL_set_mode(ssl, SSL_MODE_ACCEPT_MOVING_WRITE_BUFFER); bev_ssl->underlying = underlying; bev_ssl->ssl = ssl; bev_ssl->outbuf_cb = evbuffer_add_cb(bev_p->bev.output, be_openssl_outbuf_cb, bev_ssl); if (options & BEV_OPT_THREADSAFE) bufferevent_enable_locking_(&bev_ssl->bev.bev, NULL); if (underlying) { bufferevent_init_generic_timeout_cbs_(&bev_ssl->bev.bev); bufferevent_incref_(underlying); } bev_ssl->old_state = state; bev_ssl->last_write = -1; init_bio_counts(bev_ssl); fd = be_openssl_auto_fd(bev_ssl, fd); if (be_openssl_set_fd(bev_ssl, state, fd)) goto err; if (underlying) { bufferevent_setwatermark(underlying, EV_READ, 0, 0); bufferevent_enable(underlying, EV_READ|EV_WRITE); if (state == BUFFEREVENT_SSL_OPEN) bufferevent_suspend_read_(underlying, BEV_SUSPEND_FILT_READ); } return &bev_ssl->bev.bev; err: if (options & BEV_OPT_CLOSE_ON_FREE) SSL_free(ssl); if (bev_ssl) { bev_ssl->ssl = NULL; bufferevent_free(&bev_ssl->bev.bev); } return NULL; } struct bufferevent * bufferevent_openssl_filter_new(struct event_base *base, struct bufferevent *underlying, SSL *ssl, enum bufferevent_ssl_state state, int options) { BIO *bio; struct bufferevent *bev; if (!underlying) goto err; if (!(bio = BIO_new_bufferevent(underlying))) goto err; SSL_set_bio(ssl, bio, bio); bev = bufferevent_openssl_new_impl( base, underlying, -1, ssl, state, options); return bev; err: if (options & BEV_OPT_CLOSE_ON_FREE) SSL_free(ssl); return NULL; } struct bufferevent * bufferevent_openssl_socket_new(struct event_base *base, evutil_socket_t fd, SSL *ssl, enum bufferevent_ssl_state state, int options) { /* Does the SSL already have an fd? */ BIO *bio = SSL_get_wbio(ssl); long have_fd = -1; if (bio) have_fd = BIO_get_fd(bio, NULL); if (have_fd >= 0) { /* The SSL is already configured with an fd. */ if (fd < 0) { /* We should learn the fd from the SSL. */ fd = (evutil_socket_t) have_fd; } else if (have_fd == (long)fd) { /* We already know the fd from the SSL; do nothing */ } else { /* We specified an fd different from that of the SSL. This is probably an error on our part. Fail. */ goto err; } BIO_set_close(bio, 0); } else { /* The SSL isn't configured with a BIO with an fd. */ if (fd >= 0) { /* ... and we have an fd we want to use. */ bio = BIO_new_socket((int)fd, 0); SSL_set_bio(ssl, bio, bio); } else { /* Leave the fd unset. */ } } return bufferevent_openssl_new_impl( base, NULL, fd, ssl, state, options); err: if (options & BEV_OPT_CLOSE_ON_FREE) SSL_free(ssl); return NULL; } int bufferevent_openssl_get_allow_dirty_shutdown(struct bufferevent *bev) { int allow_dirty_shutdown = -1; struct bufferevent_openssl *bev_ssl; BEV_LOCK(bev); bev_ssl = upcast(bev); if (bev_ssl) allow_dirty_shutdown = bev_ssl->allow_dirty_shutdown; BEV_UNLOCK(bev); return allow_dirty_shutdown; } void bufferevent_openssl_set_allow_dirty_shutdown(struct bufferevent *bev, int allow_dirty_shutdown) { struct bufferevent_openssl *bev_ssl; BEV_LOCK(bev); bev_ssl = upcast(bev); if (bev_ssl) bev_ssl->allow_dirty_shutdown = !!allow_dirty_shutdown; BEV_UNLOCK(bev); } unsigned long bufferevent_get_openssl_error(struct bufferevent *bev) { unsigned long err = 0; struct bufferevent_openssl *bev_ssl; BEV_LOCK(bev); bev_ssl = upcast(bev); if (bev_ssl && bev_ssl->n_errors) { err = bev_ssl->errors[--bev_ssl->n_errors]; } BEV_UNLOCK(bev); return err; } /* * Copyright (c) 2009-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef _WIN32 #include #endif #include "event2/util.h" #include "event2/buffer.h" #include "event2/bufferevent.h" #include "event2/bufferevent_struct.h" #include "event2/event.h" #include "defer-internal.h" #include "bufferevent-internal.h" #include "mm-internal.h" #include "util-internal.h" struct bufferevent_pair { struct bufferevent_private bev; struct bufferevent_pair *partner; /* For ->destruct() lock checking */ struct bufferevent_pair *unlinked_partner; }; /* Given a bufferevent that's really a bev part of a bufferevent_pair, * return that bufferevent_filtered. Returns NULL otherwise.*/ static inline struct bufferevent_pair * upcast(struct bufferevent *bev) { struct bufferevent_pair *bev_p; if (!BEV_IS_PAIR(bev)) return NULL; bev_p = EVUTIL_UPCAST(bev, struct bufferevent_pair, bev.bev); EVUTIL_ASSERT(BEV_IS_PAIR(&bev_p->bev.bev)); return bev_p; } #define downcast(bev_pair) (&(bev_pair)->bev.bev) static inline void incref_and_lock(struct bufferevent *b) { struct bufferevent_pair *bevp; bufferevent_incref_and_lock_(b); bevp = upcast(b); if (bevp->partner) bufferevent_incref_and_lock_(downcast(bevp->partner)); } static inline void decref_and_unlock(struct bufferevent *b) { struct bufferevent_pair *bevp = upcast(b); if (bevp->partner) bufferevent_decref_and_unlock_(downcast(bevp->partner)); bufferevent_decref_and_unlock_(b); } /* XXX Handle close */ static void be_pair_outbuf_cb(struct evbuffer *, const struct evbuffer_cb_info *, void *); static struct bufferevent_pair * bufferevent_pair_elt_new(struct event_base *base, int options) { struct bufferevent_pair *bufev; if (! (bufev = mm_calloc(1, sizeof(struct bufferevent_pair)))) return NULL; if (bufferevent_init_common_(&bufev->bev, base, &bufferevent_ops_pair, options)) { mm_free(bufev); return NULL; } if (!evbuffer_add_cb(bufev->bev.bev.output, be_pair_outbuf_cb, bufev)) { bufferevent_free(downcast(bufev)); return NULL; } bufferevent_init_generic_timeout_cbs_(&bufev->bev.bev); return bufev; } int bufferevent_pair_new(struct event_base *base, int options, struct bufferevent *pair[2]) { struct bufferevent_pair *bufev1 = NULL, *bufev2 = NULL; int tmp_options; options |= BEV_OPT_DEFER_CALLBACKS; tmp_options = options & ~BEV_OPT_THREADSAFE; bufev1 = bufferevent_pair_elt_new(base, options); if (!bufev1) return -1; bufev2 = bufferevent_pair_elt_new(base, tmp_options); if (!bufev2) { bufferevent_free(downcast(bufev1)); return -1; } if (options & BEV_OPT_THREADSAFE) { /*XXXX check return */ bufferevent_enable_locking_(downcast(bufev2), bufev1->bev.lock); } bufev1->partner = bufev2; bufev2->partner = bufev1; evbuffer_freeze(downcast(bufev1)->input, 0); evbuffer_freeze(downcast(bufev1)->output, 1); evbuffer_freeze(downcast(bufev2)->input, 0); evbuffer_freeze(downcast(bufev2)->output, 1); pair[0] = downcast(bufev1); pair[1] = downcast(bufev2); return 0; } static void be_pair_transfer(struct bufferevent *src, struct bufferevent *dst, int ignore_wm) { size_t dst_size; size_t n; evbuffer_unfreeze(src->output, 1); evbuffer_unfreeze(dst->input, 0); if (dst->wm_read.high) { dst_size = evbuffer_get_length(dst->input); if (dst_size < dst->wm_read.high) { n = dst->wm_read.high - dst_size; evbuffer_remove_buffer(src->output, dst->input, n); } else { if (!ignore_wm) goto done; n = evbuffer_get_length(src->output); evbuffer_add_buffer(dst->input, src->output); } } else { n = evbuffer_get_length(src->output); evbuffer_add_buffer(dst->input, src->output); } if (n) { BEV_RESET_GENERIC_READ_TIMEOUT(dst); if (evbuffer_get_length(dst->output)) BEV_RESET_GENERIC_WRITE_TIMEOUT(dst); else BEV_DEL_GENERIC_WRITE_TIMEOUT(dst); } bufferevent_trigger_nolock_(dst, EV_READ, 0); bufferevent_trigger_nolock_(src, EV_WRITE, 0); done: evbuffer_freeze(src->output, 1); evbuffer_freeze(dst->input, 0); } static inline int be_pair_wants_to_talk(struct bufferevent_pair *src, struct bufferevent_pair *dst) { return (downcast(src)->enabled & EV_WRITE) && (downcast(dst)->enabled & EV_READ) && !dst->bev.read_suspended && evbuffer_get_length(downcast(src)->output); } static void be_pair_outbuf_cb(struct evbuffer *outbuf, const struct evbuffer_cb_info *info, void *arg) { struct bufferevent_pair *bev_pair = arg; struct bufferevent_pair *partner = bev_pair->partner; incref_and_lock(downcast(bev_pair)); if (info->n_added > info->n_deleted && partner) { /* We got more data. If the other side's reading, then hand it over. */ if (be_pair_wants_to_talk(bev_pair, partner)) { be_pair_transfer(downcast(bev_pair), downcast(partner), 0); } } decref_and_unlock(downcast(bev_pair)); } static int be_pair_enable(struct bufferevent *bufev, short events) { struct bufferevent_pair *bev_p = upcast(bufev); struct bufferevent_pair *partner = bev_p->partner; incref_and_lock(bufev); if (events & EV_READ) { BEV_RESET_GENERIC_READ_TIMEOUT(bufev); } if ((events & EV_WRITE) && evbuffer_get_length(bufev->output)) BEV_RESET_GENERIC_WRITE_TIMEOUT(bufev); /* We're starting to read! Does the other side have anything to write?*/ if ((events & EV_READ) && partner && be_pair_wants_to_talk(partner, bev_p)) { be_pair_transfer(downcast(partner), bufev, 0); } /* We're starting to write! Does the other side want to read? */ if ((events & EV_WRITE) && partner && be_pair_wants_to_talk(bev_p, partner)) { be_pair_transfer(bufev, downcast(partner), 0); } decref_and_unlock(bufev); return 0; } static int be_pair_disable(struct bufferevent *bev, short events) { if (events & EV_READ) { BEV_DEL_GENERIC_READ_TIMEOUT(bev); } if (events & EV_WRITE) { BEV_DEL_GENERIC_WRITE_TIMEOUT(bev); } return 0; } static void be_pair_unlink(struct bufferevent *bev) { struct bufferevent_pair *bev_p = upcast(bev); if (bev_p->partner) { bev_p->unlinked_partner = bev_p->partner; bev_p->partner->partner = NULL; bev_p->partner = NULL; } } /* Free *shared* lock in the latest be (since we share it between two of them). */ static void be_pair_destruct(struct bufferevent *bev) { struct bufferevent_pair *bev_p = upcast(bev); /* Transfer ownership of the lock into partner, otherwise we will use * already free'd lock during freeing second bev, see next example: * * bev1->own_lock = 1 * bev2->own_lock = 0 * bev2->lock = bev1->lock * * bufferevent_free(bev1) # refcnt == 0 -> unlink * bufferevent_free(bev2) # refcnt == 0 -> unlink * * event_base_free() -> finilizers -> EVTHREAD_FREE_LOCK(bev1->lock) * -> BEV_LOCK(bev2->lock) <-- already freed * * Where bev1 == pair[0], bev2 == pair[1]. */ if (bev_p->unlinked_partner && bev_p->bev.own_lock) { bev_p->unlinked_partner->bev.own_lock = 1; bev_p->bev.own_lock = 0; } bev_p->unlinked_partner = NULL; } static int be_pair_flush(struct bufferevent *bev, short iotype, enum bufferevent_flush_mode mode) { struct bufferevent_pair *bev_p = upcast(bev); struct bufferevent *partner; if (!bev_p->partner) return -1; if (mode == BEV_NORMAL) return 0; incref_and_lock(bev); partner = downcast(bev_p->partner); if ((iotype & EV_READ) != 0) be_pair_transfer(partner, bev, 1); if ((iotype & EV_WRITE) != 0) be_pair_transfer(bev, partner, 1); if (mode == BEV_FINISHED) { short what = BEV_EVENT_EOF; if (iotype & EV_READ) what |= BEV_EVENT_WRITING; if (iotype & EV_WRITE) what |= BEV_EVENT_READING; bufferevent_run_eventcb_(partner, what, 0); } decref_and_unlock(bev); return 0; } struct bufferevent * bufferevent_pair_get_partner(struct bufferevent *bev) { struct bufferevent_pair *bev_p; struct bufferevent *partner = NULL; bev_p = upcast(bev); if (! bev_p) return NULL; incref_and_lock(bev); if (bev_p->partner) partner = downcast(bev_p->partner); decref_and_unlock(bev); return partner; } const struct bufferevent_ops bufferevent_ops_pair = { "pair_elt", evutil_offsetof(struct bufferevent_pair, bev.bev), be_pair_enable, be_pair_disable, be_pair_unlink, be_pair_destruct, bufferevent_generic_adj_timeouts_, be_pair_flush, NULL, /* ctrl */ }; /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * Copyright (c) 2002-2006 Niels Provos * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "evconfig-private.h" #include #include #include #include #include "event2/event.h" #include "event2/event_struct.h" #include "event2/util.h" #include "event2/bufferevent.h" #include "event2/bufferevent_struct.h" #include "event2/buffer.h" #include "ratelim-internal.h" #include "bufferevent-internal.h" #include "mm-internal.h" #include "util-internal.h" #include "event-internal.h" int ev_token_bucket_init_(struct ev_token_bucket *bucket, const struct ev_token_bucket_cfg *cfg, ev_uint32_t current_tick, int reinitialize) { if (reinitialize) { /* on reinitialization, we only clip downwards, since we've already used who-knows-how-much bandwidth this tick. We leave "last_updated" as it is; the next update will add the appropriate amount of bandwidth to the bucket. */ if (bucket->read_limit > (ev_int64_t) cfg->read_maximum) bucket->read_limit = cfg->read_maximum; if (bucket->write_limit > (ev_int64_t) cfg->write_maximum) bucket->write_limit = cfg->write_maximum; } else { bucket->read_limit = cfg->read_rate; bucket->write_limit = cfg->write_rate; bucket->last_updated = current_tick; } return 0; } int ev_token_bucket_update_(struct ev_token_bucket *bucket, const struct ev_token_bucket_cfg *cfg, ev_uint32_t current_tick) { /* It's okay if the tick number overflows, since we'll just * wrap around when we do the unsigned substraction. */ unsigned n_ticks = current_tick - bucket->last_updated; /* Make sure some ticks actually happened, and that time didn't * roll back. */ if (n_ticks == 0 || n_ticks > INT_MAX) return 0; /* Naively, we would say bucket->limit += n_ticks * cfg->rate; if (bucket->limit > cfg->maximum) bucket->limit = cfg->maximum; But we're worried about overflow, so we do it like this: */ if ((cfg->read_maximum - bucket->read_limit) / n_ticks < cfg->read_rate) bucket->read_limit = cfg->read_maximum; else bucket->read_limit += n_ticks * cfg->read_rate; if ((cfg->write_maximum - bucket->write_limit) / n_ticks < cfg->write_rate) bucket->write_limit = cfg->write_maximum; else bucket->write_limit += n_ticks * cfg->write_rate; bucket->last_updated = current_tick; return 1; } static inline void bufferevent_update_buckets(struct bufferevent_private *bev) { /* Must hold lock on bev. */ struct timeval now; unsigned tick; event_base_gettimeofday_cached(bev->bev.ev_base, &now); tick = ev_token_bucket_get_tick_(&now, bev->rate_limiting->cfg); if (tick != bev->rate_limiting->limit.last_updated) ev_token_bucket_update_(&bev->rate_limiting->limit, bev->rate_limiting->cfg, tick); } ev_uint32_t ev_token_bucket_get_tick_(const struct timeval *tv, const struct ev_token_bucket_cfg *cfg) { /* This computation uses two multiplies and a divide. We could do * fewer if we knew that the tick length was an integer number of * seconds, or if we knew it divided evenly into a second. We should * investigate that more. */ /* We cast to an ev_uint64_t first, since we don't want to overflow * before we do the final divide. */ ev_uint64_t msec = (ev_uint64_t)tv->tv_sec * 1000 + tv->tv_usec / 1000; return (unsigned)(msec / cfg->msec_per_tick); } struct ev_token_bucket_cfg * ev_token_bucket_cfg_new(size_t read_rate, size_t read_burst, size_t write_rate, size_t write_burst, const struct timeval *tick_len) { struct ev_token_bucket_cfg *r; struct timeval g; if (! tick_len) { g.tv_sec = 1; g.tv_usec = 0; tick_len = &g; } if (read_rate > read_burst || write_rate > write_burst || read_rate < 1 || write_rate < 1) return NULL; if (read_rate > EV_RATE_LIMIT_MAX || write_rate > EV_RATE_LIMIT_MAX || read_burst > EV_RATE_LIMIT_MAX || write_burst > EV_RATE_LIMIT_MAX) return NULL; r = mm_calloc(1, sizeof(struct ev_token_bucket_cfg)); if (!r) return NULL; r->read_rate = read_rate; r->write_rate = write_rate; r->read_maximum = read_burst; r->write_maximum = write_burst; memcpy(&r->tick_timeout, tick_len, sizeof(struct timeval)); r->msec_per_tick = (tick_len->tv_sec * 1000) + (tick_len->tv_usec & COMMON_TIMEOUT_MICROSECONDS_MASK)/1000; return r; } void ev_token_bucket_cfg_free(struct ev_token_bucket_cfg *cfg) { mm_free(cfg); } /* Default values for max_single_read & max_single_write variables. */ #define MAX_SINGLE_READ_DEFAULT 16384 #define MAX_SINGLE_WRITE_DEFAULT 16384 #define LOCK_GROUP(g) EVLOCK_LOCK((g)->lock, 0) #define UNLOCK_GROUP(g) EVLOCK_UNLOCK((g)->lock, 0) static int bev_group_suspend_reading_(struct bufferevent_rate_limit_group *g); static int bev_group_suspend_writing_(struct bufferevent_rate_limit_group *g); static void bev_group_unsuspend_reading_(struct bufferevent_rate_limit_group *g); static void bev_group_unsuspend_writing_(struct bufferevent_rate_limit_group *g); /** Helper: figure out the maximum amount we should write if is_write, or the maximum amount we should read if is_read. Return that maximum, or 0 if our bucket is wholly exhausted. */ static inline ev_ssize_t bufferevent_get_rlim_max_(struct bufferevent_private *bev, int is_write) { /* needs lock on bev. */ ev_ssize_t max_so_far = is_write?bev->max_single_write:bev->max_single_read; #define LIM(x) \ (is_write ? (x).write_limit : (x).read_limit) #define GROUP_SUSPENDED(g) \ (is_write ? (g)->write_suspended : (g)->read_suspended) /* Sets max_so_far to MIN(x, max_so_far) */ #define CLAMPTO(x) \ do { \ if (max_so_far > (x)) \ max_so_far = (x); \ } while (0); if (!bev->rate_limiting) return max_so_far; /* If rate-limiting is enabled at all, update the appropriate bucket, and take the smaller of our rate limit and the group rate limit. */ if (bev->rate_limiting->cfg) { bufferevent_update_buckets(bev); max_so_far = LIM(bev->rate_limiting->limit); } if (bev->rate_limiting->group) { struct bufferevent_rate_limit_group *g = bev->rate_limiting->group; ev_ssize_t share; LOCK_GROUP(g); if (GROUP_SUSPENDED(g)) { /* We can get here if we failed to lock this * particular bufferevent while suspending the whole * group. */ if (is_write) bufferevent_suspend_write_(&bev->bev, BEV_SUSPEND_BW_GROUP); else bufferevent_suspend_read_(&bev->bev, BEV_SUSPEND_BW_GROUP); share = 0; } else { /* XXXX probably we should divide among the active * members, not the total members. */ share = LIM(g->rate_limit) / g->n_members; if (share < g->min_share) share = g->min_share; } UNLOCK_GROUP(g); CLAMPTO(share); } if (max_so_far < 0) max_so_far = 0; return max_so_far; } ev_ssize_t bufferevent_get_read_max_(struct bufferevent_private *bev) { return bufferevent_get_rlim_max_(bev, 0); } ev_ssize_t bufferevent_get_write_max_(struct bufferevent_private *bev) { return bufferevent_get_rlim_max_(bev, 1); } int bufferevent_decrement_read_buckets_(struct bufferevent_private *bev, ev_ssize_t bytes) { /* XXXXX Make sure all users of this function check its return value */ int r = 0; /* need to hold lock on bev */ if (!bev->rate_limiting) return 0; if (bev->rate_limiting->cfg) { bev->rate_limiting->limit.read_limit -= bytes; if (bev->rate_limiting->limit.read_limit <= 0) { bufferevent_suspend_read_(&bev->bev, BEV_SUSPEND_BW); if (event_add(&bev->rate_limiting->refill_bucket_event, &bev->rate_limiting->cfg->tick_timeout) < 0) r = -1; } else if (bev->read_suspended & BEV_SUSPEND_BW) { if (!(bev->write_suspended & BEV_SUSPEND_BW)) event_del(&bev->rate_limiting->refill_bucket_event); bufferevent_unsuspend_read_(&bev->bev, BEV_SUSPEND_BW); } } if (bev->rate_limiting->group) { LOCK_GROUP(bev->rate_limiting->group); bev->rate_limiting->group->rate_limit.read_limit -= bytes; bev->rate_limiting->group->total_read += bytes; if (bev->rate_limiting->group->rate_limit.read_limit <= 0) { bev_group_suspend_reading_(bev->rate_limiting->group); } else if (bev->rate_limiting->group->read_suspended) { bev_group_unsuspend_reading_(bev->rate_limiting->group); } UNLOCK_GROUP(bev->rate_limiting->group); } return r; } int bufferevent_decrement_write_buckets_(struct bufferevent_private *bev, ev_ssize_t bytes) { /* XXXXX Make sure all users of this function check its return value */ int r = 0; /* need to hold lock */ if (!bev->rate_limiting) return 0; if (bev->rate_limiting->cfg) { bev->rate_limiting->limit.write_limit -= bytes; if (bev->rate_limiting->limit.write_limit <= 0) { bufferevent_suspend_write_(&bev->bev, BEV_SUSPEND_BW); if (event_add(&bev->rate_limiting->refill_bucket_event, &bev->rate_limiting->cfg->tick_timeout) < 0) r = -1; } else if (bev->write_suspended & BEV_SUSPEND_BW) { if (!(bev->read_suspended & BEV_SUSPEND_BW)) event_del(&bev->rate_limiting->refill_bucket_event); bufferevent_unsuspend_write_(&bev->bev, BEV_SUSPEND_BW); } } if (bev->rate_limiting->group) { LOCK_GROUP(bev->rate_limiting->group); bev->rate_limiting->group->rate_limit.write_limit -= bytes; bev->rate_limiting->group->total_written += bytes; if (bev->rate_limiting->group->rate_limit.write_limit <= 0) { bev_group_suspend_writing_(bev->rate_limiting->group); } else if (bev->rate_limiting->group->write_suspended) { bev_group_unsuspend_writing_(bev->rate_limiting->group); } UNLOCK_GROUP(bev->rate_limiting->group); } return r; } /** Stop reading on every bufferevent in g */ static int bev_group_suspend_reading_(struct bufferevent_rate_limit_group *g) { /* Needs group lock */ struct bufferevent_private *bev; g->read_suspended = 1; g->pending_unsuspend_read = 0; /* Note that in this loop we call EVLOCK_TRY_LOCK_ instead of BEV_LOCK, to prevent a deadlock. (Ordinarily, the group lock nests inside the bufferevent locks. If we are unable to lock any individual bufferevent, it will find out later when it looks at its limit and sees that its group is suspended.) */ LIST_FOREACH(bev, &g->members, rate_limiting->next_in_group) { if (EVLOCK_TRY_LOCK_(bev->lock)) { bufferevent_suspend_read_(&bev->bev, BEV_SUSPEND_BW_GROUP); EVLOCK_UNLOCK(bev->lock, 0); } } return 0; } /** Stop writing on every bufferevent in g */ static int bev_group_suspend_writing_(struct bufferevent_rate_limit_group *g) { /* Needs group lock */ struct bufferevent_private *bev; g->write_suspended = 1; g->pending_unsuspend_write = 0; LIST_FOREACH(bev, &g->members, rate_limiting->next_in_group) { if (EVLOCK_TRY_LOCK_(bev->lock)) { bufferevent_suspend_write_(&bev->bev, BEV_SUSPEND_BW_GROUP); EVLOCK_UNLOCK(bev->lock, 0); } } return 0; } /** Timer callback invoked on a single bufferevent with one or more exhausted buckets when they are ready to refill. */ static void bev_refill_callback_(evutil_socket_t fd, short what, void *arg) { unsigned tick; struct timeval now; struct bufferevent_private *bev = arg; int again = 0; BEV_LOCK(&bev->bev); if (!bev->rate_limiting || !bev->rate_limiting->cfg) { BEV_UNLOCK(&bev->bev); return; } /* First, update the bucket */ event_base_gettimeofday_cached(bev->bev.ev_base, &now); tick = ev_token_bucket_get_tick_(&now, bev->rate_limiting->cfg); ev_token_bucket_update_(&bev->rate_limiting->limit, bev->rate_limiting->cfg, tick); /* Now unsuspend any read/write operations as appropriate. */ if ((bev->read_suspended & BEV_SUSPEND_BW)) { if (bev->rate_limiting->limit.read_limit > 0) bufferevent_unsuspend_read_(&bev->bev, BEV_SUSPEND_BW); else again = 1; } if ((bev->write_suspended & BEV_SUSPEND_BW)) { if (bev->rate_limiting->limit.write_limit > 0) bufferevent_unsuspend_write_(&bev->bev, BEV_SUSPEND_BW); else again = 1; } if (again) { /* One or more of the buckets may need another refill if they started negative. XXXX if we need to be quiet for more ticks, we should maybe figure out what timeout we really want. */ /* XXXX Handle event_add failure somehow */ event_add(&bev->rate_limiting->refill_bucket_event, &bev->rate_limiting->cfg->tick_timeout); } BEV_UNLOCK(&bev->bev); } /** Helper: grab a random element from a bufferevent group. * * Requires that we hold the lock on the group. */ static struct bufferevent_private * bev_group_random_element_(struct bufferevent_rate_limit_group *group) { int which; struct bufferevent_private *bev; /* requires group lock */ if (!group->n_members) return NULL; EVUTIL_ASSERT(! LIST_EMPTY(&group->members)); which = evutil_weakrand_range_(&group->weakrand_seed, group->n_members); bev = LIST_FIRST(&group->members); while (which--) bev = LIST_NEXT(bev, rate_limiting->next_in_group); return bev; } /** Iterate over the elements of a rate-limiting group 'g' with a random starting point, assigning each to the variable 'bev', and executing the block 'block'. We do this in a half-baked effort to get fairness among group members. XXX Round-robin or some kind of priority queue would be even more fair. */ #define FOREACH_RANDOM_ORDER(block) \ do { \ first = bev_group_random_element_(g); \ for (bev = first; bev != LIST_END(&g->members); \ bev = LIST_NEXT(bev, rate_limiting->next_in_group)) { \ block ; \ } \ for (bev = LIST_FIRST(&g->members); bev && bev != first; \ bev = LIST_NEXT(bev, rate_limiting->next_in_group)) { \ block ; \ } \ } while (0) static void bev_group_unsuspend_reading_(struct bufferevent_rate_limit_group *g) { int again = 0; struct bufferevent_private *bev, *first; g->read_suspended = 0; FOREACH_RANDOM_ORDER({ if (EVLOCK_TRY_LOCK_(bev->lock)) { bufferevent_unsuspend_read_(&bev->bev, BEV_SUSPEND_BW_GROUP); EVLOCK_UNLOCK(bev->lock, 0); } else { again = 1; } }); g->pending_unsuspend_read = again; } static void bev_group_unsuspend_writing_(struct bufferevent_rate_limit_group *g) { int again = 0; struct bufferevent_private *bev, *first; g->write_suspended = 0; FOREACH_RANDOM_ORDER({ if (EVLOCK_TRY_LOCK_(bev->lock)) { bufferevent_unsuspend_write_(&bev->bev, BEV_SUSPEND_BW_GROUP); EVLOCK_UNLOCK(bev->lock, 0); } else { again = 1; } }); g->pending_unsuspend_write = again; } /** Callback invoked every tick to add more elements to the group bucket and unsuspend group members as needed. */ static void bev_group_refill_callback_(evutil_socket_t fd, short what, void *arg) { struct bufferevent_rate_limit_group *g = arg; unsigned tick; struct timeval now; event_base_gettimeofday_cached(event_get_base(&g->master_refill_event), &now); LOCK_GROUP(g); tick = ev_token_bucket_get_tick_(&now, &g->rate_limit_cfg); ev_token_bucket_update_(&g->rate_limit, &g->rate_limit_cfg, tick); if (g->pending_unsuspend_read || (g->read_suspended && (g->rate_limit.read_limit >= g->min_share))) { bev_group_unsuspend_reading_(g); } if (g->pending_unsuspend_write || (g->write_suspended && (g->rate_limit.write_limit >= g->min_share))){ bev_group_unsuspend_writing_(g); } /* XXXX Rather than waiting to the next tick to unsuspend stuff * with pending_unsuspend_write/read, we should do it on the * next iteration of the mainloop. */ UNLOCK_GROUP(g); } int bufferevent_set_rate_limit(struct bufferevent *bev, struct ev_token_bucket_cfg *cfg) { struct bufferevent_private *bevp = BEV_UPCAST(bev); int r = -1; struct bufferevent_rate_limit *rlim; struct timeval now; ev_uint32_t tick; int reinit = 0, suspended = 0; /* XXX reference-count cfg */ BEV_LOCK(bev); if (cfg == NULL) { if (bevp->rate_limiting) { rlim = bevp->rate_limiting; rlim->cfg = NULL; bufferevent_unsuspend_read_(bev, BEV_SUSPEND_BW); bufferevent_unsuspend_write_(bev, BEV_SUSPEND_BW); if (event_initialized(&rlim->refill_bucket_event)) event_del(&rlim->refill_bucket_event); } r = 0; goto done; } event_base_gettimeofday_cached(bev->ev_base, &now); tick = ev_token_bucket_get_tick_(&now, cfg); if (bevp->rate_limiting && bevp->rate_limiting->cfg == cfg) { /* no-op */ r = 0; goto done; } if (bevp->rate_limiting == NULL) { rlim = mm_calloc(1, sizeof(struct bufferevent_rate_limit)); if (!rlim) goto done; bevp->rate_limiting = rlim; } else { rlim = bevp->rate_limiting; } reinit = rlim->cfg != NULL; rlim->cfg = cfg; ev_token_bucket_init_(&rlim->limit, cfg, tick, reinit); if (reinit) { EVUTIL_ASSERT(event_initialized(&rlim->refill_bucket_event)); event_del(&rlim->refill_bucket_event); } event_assign(&rlim->refill_bucket_event, bev->ev_base, -1, EV_FINALIZE, bev_refill_callback_, bevp); if (rlim->limit.read_limit > 0) { bufferevent_unsuspend_read_(bev, BEV_SUSPEND_BW); } else { bufferevent_suspend_read_(bev, BEV_SUSPEND_BW); suspended=1; } if (rlim->limit.write_limit > 0) { bufferevent_unsuspend_write_(bev, BEV_SUSPEND_BW); } else { bufferevent_suspend_write_(bev, BEV_SUSPEND_BW); suspended = 1; } if (suspended) event_add(&rlim->refill_bucket_event, &cfg->tick_timeout); r = 0; done: BEV_UNLOCK(bev); return r; } struct bufferevent_rate_limit_group * bufferevent_rate_limit_group_new(struct event_base *base, const struct ev_token_bucket_cfg *cfg) { struct bufferevent_rate_limit_group *g; struct timeval now; ev_uint32_t tick; event_base_gettimeofday_cached(base, &now); tick = ev_token_bucket_get_tick_(&now, cfg); g = mm_calloc(1, sizeof(struct bufferevent_rate_limit_group)); if (!g) return NULL; memcpy(&g->rate_limit_cfg, cfg, sizeof(g->rate_limit_cfg)); LIST_INIT(&g->members); ev_token_bucket_init_(&g->rate_limit, cfg, tick, 0); event_assign(&g->master_refill_event, base, -1, EV_PERSIST|EV_FINALIZE, bev_group_refill_callback_, g); /*XXXX handle event_add failure */ event_add(&g->master_refill_event, &cfg->tick_timeout); EVTHREAD_ALLOC_LOCK(g->lock, EVTHREAD_LOCKTYPE_RECURSIVE); bufferevent_rate_limit_group_set_min_share(g, 64); evutil_weakrand_seed_(&g->weakrand_seed, (ev_uint32_t) ((now.tv_sec + now.tv_usec) + (ev_intptr_t)g)); return g; } int bufferevent_rate_limit_group_set_cfg( struct bufferevent_rate_limit_group *g, const struct ev_token_bucket_cfg *cfg) { int same_tick; if (!g || !cfg) return -1; LOCK_GROUP(g); same_tick = evutil_timercmp( &g->rate_limit_cfg.tick_timeout, &cfg->tick_timeout, ==); memcpy(&g->rate_limit_cfg, cfg, sizeof(g->rate_limit_cfg)); if (g->rate_limit.read_limit > (ev_ssize_t)cfg->read_maximum) g->rate_limit.read_limit = cfg->read_maximum; if (g->rate_limit.write_limit > (ev_ssize_t)cfg->write_maximum) g->rate_limit.write_limit = cfg->write_maximum; if (!same_tick) { /* This can cause a hiccup in the schedule */ event_add(&g->master_refill_event, &cfg->tick_timeout); } /* The new limits might force us to adjust min_share differently. */ bufferevent_rate_limit_group_set_min_share(g, g->configured_min_share); UNLOCK_GROUP(g); return 0; } int bufferevent_rate_limit_group_set_min_share( struct bufferevent_rate_limit_group *g, size_t share) { if (share > EV_SSIZE_MAX) return -1; g->configured_min_share = share; /* Can't set share to less than the one-tick maximum. IOW, at steady * state, at least one connection can go per tick. */ if (share > g->rate_limit_cfg.read_rate) share = g->rate_limit_cfg.read_rate; if (share > g->rate_limit_cfg.write_rate) share = g->rate_limit_cfg.write_rate; g->min_share = share; return 0; } void bufferevent_rate_limit_group_free(struct bufferevent_rate_limit_group *g) { LOCK_GROUP(g); EVUTIL_ASSERT(0 == g->n_members); event_del(&g->master_refill_event); UNLOCK_GROUP(g); EVTHREAD_FREE_LOCK(g->lock, EVTHREAD_LOCKTYPE_RECURSIVE); mm_free(g); } int bufferevent_add_to_rate_limit_group(struct bufferevent *bev, struct bufferevent_rate_limit_group *g) { int wsuspend, rsuspend; struct bufferevent_private *bevp = BEV_UPCAST(bev); BEV_LOCK(bev); if (!bevp->rate_limiting) { struct bufferevent_rate_limit *rlim; rlim = mm_calloc(1, sizeof(struct bufferevent_rate_limit)); if (!rlim) { BEV_UNLOCK(bev); return -1; } event_assign(&rlim->refill_bucket_event, bev->ev_base, -1, EV_FINALIZE, bev_refill_callback_, bevp); bevp->rate_limiting = rlim; } if (bevp->rate_limiting->group == g) { BEV_UNLOCK(bev); return 0; } if (bevp->rate_limiting->group) bufferevent_remove_from_rate_limit_group(bev); LOCK_GROUP(g); bevp->rate_limiting->group = g; ++g->n_members; LIST_INSERT_HEAD(&g->members, bevp, rate_limiting->next_in_group); rsuspend = g->read_suspended; wsuspend = g->write_suspended; UNLOCK_GROUP(g); if (rsuspend) bufferevent_suspend_read_(bev, BEV_SUSPEND_BW_GROUP); if (wsuspend) bufferevent_suspend_write_(bev, BEV_SUSPEND_BW_GROUP); BEV_UNLOCK(bev); return 0; } int bufferevent_remove_from_rate_limit_group(struct bufferevent *bev) { return bufferevent_remove_from_rate_limit_group_internal_(bev, 1); } int bufferevent_remove_from_rate_limit_group_internal_(struct bufferevent *bev, int unsuspend) { struct bufferevent_private *bevp = BEV_UPCAST(bev); BEV_LOCK(bev); if (bevp->rate_limiting && bevp->rate_limiting->group) { struct bufferevent_rate_limit_group *g = bevp->rate_limiting->group; LOCK_GROUP(g); bevp->rate_limiting->group = NULL; --g->n_members; LIST_REMOVE(bevp, rate_limiting->next_in_group); UNLOCK_GROUP(g); } if (unsuspend) { bufferevent_unsuspend_read_(bev, BEV_SUSPEND_BW_GROUP); bufferevent_unsuspend_write_(bev, BEV_SUSPEND_BW_GROUP); } BEV_UNLOCK(bev); return 0; } /* === * API functions to expose rate limits. * * Don't use these from inside Libevent; they're meant to be for use by * the program. * === */ /* Mostly you don't want to use this function from inside libevent; * bufferevent_get_read_max_() is more likely what you want*/ ev_ssize_t bufferevent_get_read_limit(struct bufferevent *bev) { ev_ssize_t r; struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); if (bevp->rate_limiting && bevp->rate_limiting->cfg) { bufferevent_update_buckets(bevp); r = bevp->rate_limiting->limit.read_limit; } else { r = EV_SSIZE_MAX; } BEV_UNLOCK(bev); return r; } /* Mostly you don't want to use this function from inside libevent; * bufferevent_get_write_max_() is more likely what you want*/ ev_ssize_t bufferevent_get_write_limit(struct bufferevent *bev) { ev_ssize_t r; struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); if (bevp->rate_limiting && bevp->rate_limiting->cfg) { bufferevent_update_buckets(bevp); r = bevp->rate_limiting->limit.write_limit; } else { r = EV_SSIZE_MAX; } BEV_UNLOCK(bev); return r; } int bufferevent_set_max_single_read(struct bufferevent *bev, size_t size) { struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); if (size == 0 || size > EV_SSIZE_MAX) bevp->max_single_read = MAX_SINGLE_READ_DEFAULT; else bevp->max_single_read = size; BEV_UNLOCK(bev); return 0; } int bufferevent_set_max_single_write(struct bufferevent *bev, size_t size) { struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); if (size == 0 || size > EV_SSIZE_MAX) bevp->max_single_write = MAX_SINGLE_WRITE_DEFAULT; else bevp->max_single_write = size; BEV_UNLOCK(bev); return 0; } ev_ssize_t bufferevent_get_max_single_read(struct bufferevent *bev) { ev_ssize_t r; BEV_LOCK(bev); r = BEV_UPCAST(bev)->max_single_read; BEV_UNLOCK(bev); return r; } ev_ssize_t bufferevent_get_max_single_write(struct bufferevent *bev) { ev_ssize_t r; BEV_LOCK(bev); r = BEV_UPCAST(bev)->max_single_write; BEV_UNLOCK(bev); return r; } ev_ssize_t bufferevent_get_max_to_read(struct bufferevent *bev) { ev_ssize_t r; BEV_LOCK(bev); r = bufferevent_get_read_max_(BEV_UPCAST(bev)); BEV_UNLOCK(bev); return r; } ev_ssize_t bufferevent_get_max_to_write(struct bufferevent *bev) { ev_ssize_t r; BEV_LOCK(bev); r = bufferevent_get_write_max_(BEV_UPCAST(bev)); BEV_UNLOCK(bev); return r; } const struct ev_token_bucket_cfg * bufferevent_get_token_bucket_cfg(const struct bufferevent *bev) { struct bufferevent_private *bufev_private = BEV_UPCAST(bev); struct ev_token_bucket_cfg *cfg; BEV_LOCK(bev); if (bufev_private->rate_limiting) { cfg = bufev_private->rate_limiting->cfg; } else { cfg = NULL; } BEV_UNLOCK(bev); return cfg; } /* Mostly you don't want to use this function from inside libevent; * bufferevent_get_read_max_() is more likely what you want*/ ev_ssize_t bufferevent_rate_limit_group_get_read_limit( struct bufferevent_rate_limit_group *grp) { ev_ssize_t r; LOCK_GROUP(grp); r = grp->rate_limit.read_limit; UNLOCK_GROUP(grp); return r; } /* Mostly you don't want to use this function from inside libevent; * bufferevent_get_write_max_() is more likely what you want. */ ev_ssize_t bufferevent_rate_limit_group_get_write_limit( struct bufferevent_rate_limit_group *grp) { ev_ssize_t r; LOCK_GROUP(grp); r = grp->rate_limit.write_limit; UNLOCK_GROUP(grp); return r; } int bufferevent_decrement_read_limit(struct bufferevent *bev, ev_ssize_t decr) { int r = 0; ev_ssize_t old_limit, new_limit; struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); EVUTIL_ASSERT(bevp->rate_limiting && bevp->rate_limiting->cfg); old_limit = bevp->rate_limiting->limit.read_limit; new_limit = (bevp->rate_limiting->limit.read_limit -= decr); if (old_limit > 0 && new_limit <= 0) { bufferevent_suspend_read_(bev, BEV_SUSPEND_BW); if (event_add(&bevp->rate_limiting->refill_bucket_event, &bevp->rate_limiting->cfg->tick_timeout) < 0) r = -1; } else if (old_limit <= 0 && new_limit > 0) { if (!(bevp->write_suspended & BEV_SUSPEND_BW)) event_del(&bevp->rate_limiting->refill_bucket_event); bufferevent_unsuspend_read_(bev, BEV_SUSPEND_BW); } BEV_UNLOCK(bev); return r; } int bufferevent_decrement_write_limit(struct bufferevent *bev, ev_ssize_t decr) { /* XXXX this is mostly copy-and-paste from * bufferevent_decrement_read_limit */ int r = 0; ev_ssize_t old_limit, new_limit; struct bufferevent_private *bevp; BEV_LOCK(bev); bevp = BEV_UPCAST(bev); EVUTIL_ASSERT(bevp->rate_limiting && bevp->rate_limiting->cfg); old_limit = bevp->rate_limiting->limit.write_limit; new_limit = (bevp->rate_limiting->limit.write_limit -= decr); if (old_limit > 0 && new_limit <= 0) { bufferevent_suspend_write_(bev, BEV_SUSPEND_BW); if (event_add(&bevp->rate_limiting->refill_bucket_event, &bevp->rate_limiting->cfg->tick_timeout) < 0) r = -1; } else if (old_limit <= 0 && new_limit > 0) { if (!(bevp->read_suspended & BEV_SUSPEND_BW)) event_del(&bevp->rate_limiting->refill_bucket_event); bufferevent_unsuspend_write_(bev, BEV_SUSPEND_BW); } BEV_UNLOCK(bev); return r; } int bufferevent_rate_limit_group_decrement_read( struct bufferevent_rate_limit_group *grp, ev_ssize_t decr) { int r = 0; ev_ssize_t old_limit, new_limit; LOCK_GROUP(grp); old_limit = grp->rate_limit.read_limit; new_limit = (grp->rate_limit.read_limit -= decr); if (old_limit > 0 && new_limit <= 0) { bev_group_suspend_reading_(grp); } else if (old_limit <= 0 && new_limit > 0) { bev_group_unsuspend_reading_(grp); } UNLOCK_GROUP(grp); return r; } int bufferevent_rate_limit_group_decrement_write( struct bufferevent_rate_limit_group *grp, ev_ssize_t decr) { int r = 0; ev_ssize_t old_limit, new_limit; LOCK_GROUP(grp); old_limit = grp->rate_limit.write_limit; new_limit = (grp->rate_limit.write_limit -= decr); if (old_limit > 0 && new_limit <= 0) { bev_group_suspend_writing_(grp); } else if (old_limit <= 0 && new_limit > 0) { bev_group_unsuspend_writing_(grp); } UNLOCK_GROUP(grp); return r; } void bufferevent_rate_limit_group_get_totals(struct bufferevent_rate_limit_group *grp, ev_uint64_t *total_read_out, ev_uint64_t *total_written_out) { EVUTIL_ASSERT(grp != NULL); if (total_read_out) *total_read_out = grp->total_read; if (total_written_out) *total_written_out = grp->total_written; } void bufferevent_rate_limit_group_reset_totals(struct bufferevent_rate_limit_group *grp) { grp->total_read = grp->total_written = 0; } int bufferevent_ratelim_init_(struct bufferevent_private *bev) { bev->rate_limiting = NULL; bev->max_single_read = MAX_SINGLE_READ_DEFAULT; bev->max_single_write = MAX_SINGLE_WRITE_DEFAULT; return 0; } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * Copyright (c) 2002-2006 Niels Provos * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_STDARG_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #ifdef _WIN32 #include #include #endif #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #ifdef EVENT__HAVE_NETINET_IN_H #include #endif #ifdef EVENT__HAVE_NETINET_IN6_H #include #endif #include "event2/util.h" #include "event2/bufferevent.h" #include "event2/buffer.h" #include "event2/bufferevent_struct.h" #include "event2/bufferevent_compat.h" #include "event2/event.h" #include "log-internal.h" #include "mm-internal.h" #include "bufferevent-internal.h" #include "util-internal.h" #ifdef _WIN32 #include "iocp-internal.h" #endif /* prototypes */ static int be_socket_enable(struct bufferevent *, short); static int be_socket_disable(struct bufferevent *, short); static void be_socket_destruct(struct bufferevent *); static int be_socket_flush(struct bufferevent *, short, enum bufferevent_flush_mode); static int be_socket_ctrl(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *); static void be_socket_setfd(struct bufferevent *, evutil_socket_t); const struct bufferevent_ops bufferevent_ops_socket = { "socket", evutil_offsetof(struct bufferevent_private, bev), be_socket_enable, be_socket_disable, NULL, /* unlink */ be_socket_destruct, bufferevent_generic_adj_existing_timeouts_, be_socket_flush, be_socket_ctrl, }; const struct sockaddr* bufferevent_socket_get_conn_address_(struct bufferevent *bev) { struct bufferevent_private *bev_p = BEV_UPCAST(bev); return (struct sockaddr *)&bev_p->conn_address; } void bufferevent_socket_set_conn_address_fd_(struct bufferevent *bev, evutil_socket_t fd) { struct bufferevent_private *bev_p = BEV_UPCAST(bev); socklen_t len = sizeof(bev_p->conn_address); struct sockaddr *addr = (struct sockaddr *)&bev_p->conn_address; if (addr->sa_family != AF_UNSPEC) getpeername(fd, addr, &len); } void bufferevent_socket_set_conn_address_(struct bufferevent *bev, struct sockaddr *addr, size_t addrlen) { struct bufferevent_private *bev_p = BEV_UPCAST(bev); EVUTIL_ASSERT(addrlen <= sizeof(bev_p->conn_address)); memcpy(&bev_p->conn_address, addr, addrlen); } static void bufferevent_socket_outbuf_cb(struct evbuffer *buf, const struct evbuffer_cb_info *cbinfo, void *arg) { struct bufferevent *bufev = arg; struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); if (cbinfo->n_added && (bufev->enabled & EV_WRITE) && !event_pending(&bufev->ev_write, EV_WRITE, NULL) && !bufev_p->write_suspended) { /* Somebody added data to the buffer, and we would like to * write, and we were not writing. So, start writing. */ if (bufferevent_add_event_(&bufev->ev_write, &bufev->timeout_write) == -1) { /* Should we log this? */ } } } static void bufferevent_readcb(evutil_socket_t fd, short event, void *arg) { struct bufferevent *bufev = arg; struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); struct evbuffer *input; int res = 0; short what = BEV_EVENT_READING; ev_ssize_t howmuch = -1, readmax=-1; bufferevent_incref_and_lock_(bufev); if (event == EV_TIMEOUT) { /* Note that we only check for event==EV_TIMEOUT. If * event==EV_TIMEOUT|EV_READ, we can safely ignore the * timeout, since a read has occurred */ what |= BEV_EVENT_TIMEOUT; goto error; } input = bufev->input; /* * If we have a high watermark configured then we don't want to * read more data than would make us reach the watermark. */ if (bufev->wm_read.high != 0) { howmuch = bufev->wm_read.high - evbuffer_get_length(input); /* we somehow lowered the watermark, stop reading */ if (howmuch <= 0) { bufferevent_wm_suspend_read(bufev); goto done; } } readmax = bufferevent_get_read_max_(bufev_p); if (howmuch < 0 || howmuch > readmax) /* The use of -1 for "unlimited" * uglifies this code. XXXX */ howmuch = readmax; if (bufev_p->read_suspended) goto done; evbuffer_unfreeze(input, 0); res = evbuffer_read(input, fd, (int)howmuch); /* XXXX evbuffer_read would do better to take and return ev_ssize_t */ evbuffer_freeze(input, 0); if (res == -1) { int err = evutil_socket_geterror(fd); if (EVUTIL_ERR_RW_RETRIABLE(err)) goto reschedule; if (EVUTIL_ERR_CONNECT_REFUSED(err)) { bufev_p->connection_refused = 1; goto done; } /* error case */ what |= BEV_EVENT_ERROR; } else if (res == 0) { /* eof case */ what |= BEV_EVENT_EOF; } if (res <= 0) goto error; bufferevent_decrement_read_buckets_(bufev_p, res); /* Invoke the user callback - must always be called last */ bufferevent_trigger_nolock_(bufev, EV_READ, 0); goto done; reschedule: goto done; error: bufferevent_disable(bufev, EV_READ); bufferevent_run_eventcb_(bufev, what, 0); done: bufferevent_decref_and_unlock_(bufev); } static void bufferevent_writecb(evutil_socket_t fd, short event, void *arg) { struct bufferevent *bufev = arg; struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); int res = 0; short what = BEV_EVENT_WRITING; int connected = 0; ev_ssize_t atmost = -1; bufferevent_incref_and_lock_(bufev); if (event == EV_TIMEOUT) { /* Note that we only check for event==EV_TIMEOUT. If * event==EV_TIMEOUT|EV_WRITE, we can safely ignore the * timeout, since a read has occurred */ what |= BEV_EVENT_TIMEOUT; goto error; } if (bufev_p->connecting) { int c = evutil_socket_finished_connecting_(fd); /* we need to fake the error if the connection was refused * immediately - usually connection to localhost on BSD */ if (bufev_p->connection_refused) { bufev_p->connection_refused = 0; c = -1; } if (c == 0) goto done; bufev_p->connecting = 0; if (c < 0) { event_del(&bufev->ev_write); event_del(&bufev->ev_read); bufferevent_run_eventcb_(bufev, BEV_EVENT_ERROR, 0); goto done; } else { connected = 1; bufferevent_socket_set_conn_address_fd_(bufev, fd); #ifdef _WIN32 if (BEV_IS_ASYNC(bufev)) { event_del(&bufev->ev_write); bufferevent_async_set_connected_(bufev); bufferevent_run_eventcb_(bufev, BEV_EVENT_CONNECTED, 0); goto done; } #endif bufferevent_run_eventcb_(bufev, BEV_EVENT_CONNECTED, 0); if (!(bufev->enabled & EV_WRITE) || bufev_p->write_suspended) { event_del(&bufev->ev_write); goto done; } } } atmost = bufferevent_get_write_max_(bufev_p); if (bufev_p->write_suspended) goto done; if (evbuffer_get_length(bufev->output)) { evbuffer_unfreeze(bufev->output, 1); res = evbuffer_write_atmost(bufev->output, fd, atmost); evbuffer_freeze(bufev->output, 1); if (res == -1) { int err = evutil_socket_geterror(fd); if (EVUTIL_ERR_RW_RETRIABLE(err)) goto reschedule; what |= BEV_EVENT_ERROR; } else if (res == 0) { /* eof case XXXX Actually, a 0 on write doesn't indicate an EOF. An ECONNRESET might be more typical. */ what |= BEV_EVENT_EOF; } if (res <= 0) goto error; bufferevent_decrement_write_buckets_(bufev_p, res); } if (evbuffer_get_length(bufev->output) == 0) { event_del(&bufev->ev_write); } /* * Invoke the user callback if our buffer is drained or below the * low watermark. */ if (res || !connected) { bufferevent_trigger_nolock_(bufev, EV_WRITE, 0); } goto done; reschedule: if (evbuffer_get_length(bufev->output) == 0) { event_del(&bufev->ev_write); } goto done; error: bufferevent_disable(bufev, EV_WRITE); bufferevent_run_eventcb_(bufev, what, 0); done: bufferevent_decref_and_unlock_(bufev); } struct bufferevent * bufferevent_socket_new(struct event_base *base, evutil_socket_t fd, int options) { struct bufferevent_private *bufev_p; struct bufferevent *bufev; #ifdef _WIN32 if (base && event_base_get_iocp_(base)) return bufferevent_async_new_(base, fd, options); #endif if ((bufev_p = mm_calloc(1, sizeof(struct bufferevent_private)))== NULL) return NULL; if (bufferevent_init_common_(bufev_p, base, &bufferevent_ops_socket, options) < 0) { mm_free(bufev_p); return NULL; } bufev = &bufev_p->bev; evbuffer_set_flags(bufev->output, EVBUFFER_FLAG_DRAINS_TO_FD); event_assign(&bufev->ev_read, bufev->ev_base, fd, EV_READ|EV_PERSIST|EV_FINALIZE, bufferevent_readcb, bufev); event_assign(&bufev->ev_write, bufev->ev_base, fd, EV_WRITE|EV_PERSIST|EV_FINALIZE, bufferevent_writecb, bufev); evbuffer_add_cb(bufev->output, bufferevent_socket_outbuf_cb, bufev); evbuffer_freeze(bufev->input, 0); evbuffer_freeze(bufev->output, 1); return bufev; } int bufferevent_socket_connect(struct bufferevent *bev, const struct sockaddr *sa, int socklen) { struct bufferevent_private *bufev_p = BEV_UPCAST(bev); evutil_socket_t fd; int r = 0; int result=-1; int ownfd = 0; bufferevent_incref_and_lock_(bev); fd = bufferevent_getfd(bev); if (fd < 0) { if (!sa) goto done; fd = evutil_socket_(sa->sa_family, SOCK_STREAM|EVUTIL_SOCK_NONBLOCK, 0); if (fd < 0) goto freesock; ownfd = 1; } if (sa) { #ifdef _WIN32 if (bufferevent_async_can_connect_(bev)) { bufferevent_setfd(bev, fd); r = bufferevent_async_connect_(bev, fd, sa, socklen); if (r < 0) goto freesock; bufev_p->connecting = 1; result = 0; goto done; } else #endif r = evutil_socket_connect_(&fd, sa, socklen); if (r < 0) goto freesock; } #ifdef _WIN32 /* ConnectEx() isn't always around, even when IOCP is enabled. * Here, we borrow the socket object's write handler to fall back * on a non-blocking connect() when ConnectEx() is unavailable. */ if (BEV_IS_ASYNC(bev)) { event_assign(&bev->ev_write, bev->ev_base, fd, EV_WRITE|EV_PERSIST|EV_FINALIZE, bufferevent_writecb, bev); } #endif bufferevent_setfd(bev, fd); if (r == 0) { if (! be_socket_enable(bev, EV_WRITE)) { bufev_p->connecting = 1; result = 0; goto done; } } else if (r == 1) { /* The connect succeeded already. How very BSD of it. */ result = 0; bufev_p->connecting = 1; bufferevent_trigger_nolock_(bev, EV_WRITE, BEV_OPT_DEFER_CALLBACKS); } else { /* The connect failed already. How very BSD of it. */ result = 0; bufferevent_run_eventcb_(bev, BEV_EVENT_ERROR, BEV_OPT_DEFER_CALLBACKS); bufferevent_disable(bev, EV_WRITE|EV_READ); } goto done; freesock: if (ownfd) evutil_closesocket(fd); done: bufferevent_decref_and_unlock_(bev); return result; } static void bufferevent_connect_getaddrinfo_cb(int result, struct evutil_addrinfo *ai, void *arg) { struct bufferevent *bev = arg; struct bufferevent_private *bev_p = BEV_UPCAST(bev); int r; BEV_LOCK(bev); bufferevent_unsuspend_write_(bev, BEV_SUSPEND_LOOKUP); bufferevent_unsuspend_read_(bev, BEV_SUSPEND_LOOKUP); bev_p->dns_request = NULL; if (result == EVUTIL_EAI_CANCEL) { bev_p->dns_error = result; bufferevent_decref_and_unlock_(bev); return; } if (result != 0) { bev_p->dns_error = result; bufferevent_run_eventcb_(bev, BEV_EVENT_ERROR, 0); bufferevent_decref_and_unlock_(bev); if (ai) evutil_freeaddrinfo(ai); return; } /* XXX use the other addrinfos? */ bufferevent_socket_set_conn_address_(bev, ai->ai_addr, (int)ai->ai_addrlen); r = bufferevent_socket_connect(bev, ai->ai_addr, (int)ai->ai_addrlen); if (r < 0) bufferevent_run_eventcb_(bev, BEV_EVENT_ERROR, 0); bufferevent_decref_and_unlock_(bev); evutil_freeaddrinfo(ai); } int bufferevent_socket_connect_hostname(struct bufferevent *bev, struct evdns_base *evdns_base, int family, const char *hostname, int port) { char portbuf[10]; struct evutil_addrinfo hint; struct bufferevent_private *bev_p = BEV_UPCAST(bev); if (family != AF_INET && family != AF_INET6 && family != AF_UNSPEC) return -1; if (port < 1 || port > 65535) return -1; memset(&hint, 0, sizeof(hint)); hint.ai_family = family; hint.ai_protocol = IPPROTO_TCP; hint.ai_socktype = SOCK_STREAM; evutil_snprintf(portbuf, sizeof(portbuf), "%d", port); BEV_LOCK(bev); bev_p->dns_error = 0; bufferevent_suspend_write_(bev, BEV_SUSPEND_LOOKUP); bufferevent_suspend_read_(bev, BEV_SUSPEND_LOOKUP); bufferevent_incref_(bev); bev_p->dns_request = evutil_getaddrinfo_async_(evdns_base, hostname, portbuf, &hint, bufferevent_connect_getaddrinfo_cb, bev); BEV_UNLOCK(bev); return 0; } int bufferevent_socket_get_dns_error(struct bufferevent *bev) { int rv; struct bufferevent_private *bev_p = BEV_UPCAST(bev); BEV_LOCK(bev); rv = bev_p->dns_error; BEV_UNLOCK(bev); return rv; } /* * Create a new buffered event object. * * The read callback is invoked whenever we read new data. * The write callback is invoked whenever the output buffer is drained. * The error callback is invoked on a write/read error or on EOF. * * Both read and write callbacks maybe NULL. The error callback is not * allowed to be NULL and have to be provided always. */ struct bufferevent * bufferevent_new(evutil_socket_t fd, bufferevent_data_cb readcb, bufferevent_data_cb writecb, bufferevent_event_cb eventcb, void *cbarg) { struct bufferevent *bufev; if (!(bufev = bufferevent_socket_new(NULL, fd, 0))) return NULL; bufferevent_setcb(bufev, readcb, writecb, eventcb, cbarg); return bufev; } static int be_socket_enable(struct bufferevent *bufev, short event) { if (event & EV_READ && bufferevent_add_event_(&bufev->ev_read, &bufev->timeout_read) == -1) return -1; if (event & EV_WRITE && bufferevent_add_event_(&bufev->ev_write, &bufev->timeout_write) == -1) return -1; return 0; } static int be_socket_disable(struct bufferevent *bufev, short event) { struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); if (event & EV_READ) { if (event_del(&bufev->ev_read) == -1) return -1; } /* Don't actually disable the write if we are trying to connect. */ if ((event & EV_WRITE) && ! bufev_p->connecting) { if (event_del(&bufev->ev_write) == -1) return -1; } return 0; } static void be_socket_destruct(struct bufferevent *bufev) { struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); evutil_socket_t fd; EVUTIL_ASSERT(BEV_IS_SOCKET(bufev)); fd = event_get_fd(&bufev->ev_read); if ((bufev_p->options & BEV_OPT_CLOSE_ON_FREE) && fd >= 0) EVUTIL_CLOSESOCKET(fd); evutil_getaddrinfo_cancel_async_(bufev_p->dns_request); } static int be_socket_flush(struct bufferevent *bev, short iotype, enum bufferevent_flush_mode mode) { return 0; } static void be_socket_setfd(struct bufferevent *bufev, evutil_socket_t fd) { struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); BEV_LOCK(bufev); EVUTIL_ASSERT(BEV_IS_SOCKET(bufev)); event_del(&bufev->ev_read); event_del(&bufev->ev_write); evbuffer_unfreeze(bufev->input, 0); evbuffer_unfreeze(bufev->output, 1); event_assign(&bufev->ev_read, bufev->ev_base, fd, EV_READ|EV_PERSIST|EV_FINALIZE, bufferevent_readcb, bufev); event_assign(&bufev->ev_write, bufev->ev_base, fd, EV_WRITE|EV_PERSIST|EV_FINALIZE, bufferevent_writecb, bufev); if (fd >= 0) bufferevent_enable(bufev, bufev->enabled); evutil_getaddrinfo_cancel_async_(bufev_p->dns_request); BEV_UNLOCK(bufev); } /* XXXX Should non-socket bufferevents support this? */ int bufferevent_priority_set(struct bufferevent *bufev, int priority) { int r = -1; struct bufferevent_private *bufev_p = BEV_UPCAST(bufev); BEV_LOCK(bufev); if (BEV_IS_ASYNC(bufev) || BEV_IS_FILTER(bufev) || BEV_IS_PAIR(bufev)) goto done; if (event_priority_set(&bufev->ev_read, priority) == -1) goto done; if (event_priority_set(&bufev->ev_write, priority) == -1) goto done; event_deferred_cb_set_priority_(&bufev_p->deferred, priority); r = 0; done: BEV_UNLOCK(bufev); return r; } /* XXXX Should non-socket bufferevents support this? */ int bufferevent_base_set(struct event_base *base, struct bufferevent *bufev) { int res = -1; BEV_LOCK(bufev); if (!BEV_IS_SOCKET(bufev)) goto done; bufev->ev_base = base; res = event_base_set(base, &bufev->ev_read); if (res == -1) goto done; res = event_base_set(base, &bufev->ev_write); done: BEV_UNLOCK(bufev); return res; } static int be_socket_ctrl(struct bufferevent *bev, enum bufferevent_ctrl_op op, union bufferevent_ctrl_data *data) { switch (op) { case BEV_CTRL_SET_FD: be_socket_setfd(bev, data->fd); return 0; case BEV_CTRL_GET_FD: data->fd = event_get_fd(&bev->ev_read); return 0; case BEV_CTRL_GET_UNDERLYING: case BEV_CTRL_CANCEL_ALL: default: return -1; } } /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /** @file buffer_iocp.c This module implements overlapped read and write functions for evbuffer objects on Windows. */ #include "event2/event-config.h" #include "evconfig-private.h" #include "event2/buffer.h" #include "event2/buffer_compat.h" #include "event2/util.h" #include "event2/thread.h" #include "util-internal.h" #include "evthread-internal.h" #include "evbuffer-internal.h" #include "iocp-internal.h" #include "mm-internal.h" #include #include #include #include #define MAX_WSABUFS 16 /** An evbuffer that can handle overlapped IO. */ struct evbuffer_overlapped { struct evbuffer buffer; /** The socket that we're doing overlapped IO on. */ evutil_socket_t fd; /** pending I/O type */ unsigned read_in_progress : 1; unsigned write_in_progress : 1; /** The first pinned chain in the buffer. */ struct evbuffer_chain *first_pinned; /** How many chains are pinned; how many of the fields in buffers * are we using. */ int n_buffers; WSABUF buffers[MAX_WSABUFS]; }; /** Given an evbuffer, return the correponding evbuffer structure, or NULL if * the evbuffer isn't overlapped. */ static inline struct evbuffer_overlapped * upcast_evbuffer(struct evbuffer *buf) { if (!buf || !buf->is_overlapped) return NULL; return EVUTIL_UPCAST(buf, struct evbuffer_overlapped, buffer); } /** Unpin all the chains noted as pinned in 'eo'. */ static void pin_release(struct evbuffer_overlapped *eo, unsigned flag) { int i; struct evbuffer_chain *next, *chain = eo->first_pinned; for (i = 0; i < eo->n_buffers; ++i) { EVUTIL_ASSERT(chain); next = chain->next; evbuffer_chain_unpin_(chain, flag); chain = next; } } void evbuffer_commit_read_(struct evbuffer *evbuf, ev_ssize_t nBytes) { struct evbuffer_overlapped *buf = upcast_evbuffer(evbuf); struct evbuffer_chain **chainp; size_t remaining, len; unsigned i; EVBUFFER_LOCK(evbuf); EVUTIL_ASSERT(buf->read_in_progress && !buf->write_in_progress); EVUTIL_ASSERT(nBytes >= 0); /* XXXX Can this be false? */ evbuffer_unfreeze(evbuf, 0); chainp = evbuf->last_with_datap; if (!((*chainp)->flags & EVBUFFER_MEM_PINNED_R)) chainp = &(*chainp)->next; remaining = nBytes; for (i = 0; remaining > 0 && i < (unsigned)buf->n_buffers; ++i) { EVUTIL_ASSERT(*chainp); len = buf->buffers[i].len; if (remaining < len) len = remaining; (*chainp)->off += len; evbuf->last_with_datap = chainp; remaining -= len; chainp = &(*chainp)->next; } pin_release(buf, EVBUFFER_MEM_PINNED_R); buf->read_in_progress = 0; evbuf->total_len += nBytes; evbuf->n_add_for_cb += nBytes; evbuffer_invoke_callbacks_(evbuf); evbuffer_decref_and_unlock_(evbuf); } void evbuffer_commit_write_(struct evbuffer *evbuf, ev_ssize_t nBytes) { struct evbuffer_overlapped *buf = upcast_evbuffer(evbuf); EVBUFFER_LOCK(evbuf); EVUTIL_ASSERT(buf->write_in_progress && !buf->read_in_progress); evbuffer_unfreeze(evbuf, 1); evbuffer_drain(evbuf, nBytes); pin_release(buf,EVBUFFER_MEM_PINNED_W); buf->write_in_progress = 0; evbuffer_decref_and_unlock_(evbuf); } struct evbuffer * evbuffer_overlapped_new_(evutil_socket_t fd) { struct evbuffer_overlapped *evo; evo = mm_calloc(1, sizeof(struct evbuffer_overlapped)); if (!evo) return NULL; LIST_INIT(&evo->buffer.callbacks); evo->buffer.refcnt = 1; evo->buffer.last_with_datap = &evo->buffer.first; evo->buffer.is_overlapped = 1; evo->fd = fd; return &evo->buffer; } int evbuffer_launch_write_(struct evbuffer *buf, ev_ssize_t at_most, struct event_overlapped *ol) { struct evbuffer_overlapped *buf_o = upcast_evbuffer(buf); int r = -1; int i; struct evbuffer_chain *chain; DWORD bytesSent; if (!buf) { /* No buffer, or it isn't overlapped */ return -1; } EVBUFFER_LOCK(buf); EVUTIL_ASSERT(!buf_o->read_in_progress); if (buf->freeze_start || buf_o->write_in_progress) goto done; if (!buf->total_len) { /* Nothing to write */ r = 0; goto done; } else if (at_most < 0 || (size_t)at_most > buf->total_len) { at_most = buf->total_len; } evbuffer_freeze(buf, 1); buf_o->first_pinned = NULL; buf_o->n_buffers = 0; memset(buf_o->buffers, 0, sizeof(buf_o->buffers)); chain = buf_o->first_pinned = buf->first; for (i=0; i < MAX_WSABUFS && chain; ++i, chain=chain->next) { WSABUF *b = &buf_o->buffers[i]; b->buf = (char*)( chain->buffer + chain->misalign ); evbuffer_chain_pin_(chain, EVBUFFER_MEM_PINNED_W); if ((size_t)at_most > chain->off) { /* XXXX Cast is safe for now, since win32 has no mmaped chains. But later, we need to have this add more WSAbufs if chain->off is greater than ULONG_MAX */ b->len = (unsigned long)chain->off; at_most -= chain->off; } else { b->len = (unsigned long)at_most; ++i; break; } } buf_o->n_buffers = i; evbuffer_incref_(buf); if (WSASend(buf_o->fd, buf_o->buffers, i, &bytesSent, 0, &ol->overlapped, NULL)) { int error = WSAGetLastError(); if (error != WSA_IO_PENDING) { /* An actual error. */ pin_release(buf_o, EVBUFFER_MEM_PINNED_W); evbuffer_unfreeze(buf, 1); evbuffer_free(buf); /* decref */ goto done; } } buf_o->write_in_progress = 1; r = 0; done: EVBUFFER_UNLOCK(buf); return r; } int evbuffer_launch_read_(struct evbuffer *buf, size_t at_most, struct event_overlapped *ol) { struct evbuffer_overlapped *buf_o = upcast_evbuffer(buf); int r = -1, i; int nvecs; int npin=0; struct evbuffer_chain *chain=NULL, **chainp; DWORD bytesRead; DWORD flags = 0; struct evbuffer_iovec vecs[MAX_WSABUFS]; if (!buf_o) return -1; EVBUFFER_LOCK(buf); EVUTIL_ASSERT(!buf_o->write_in_progress); if (buf->freeze_end || buf_o->read_in_progress) goto done; buf_o->first_pinned = NULL; buf_o->n_buffers = 0; memset(buf_o->buffers, 0, sizeof(buf_o->buffers)); if (evbuffer_expand_fast_(buf, at_most, MAX_WSABUFS) == -1) goto done; evbuffer_freeze(buf, 0); nvecs = evbuffer_read_setup_vecs_(buf, at_most, vecs, MAX_WSABUFS, &chainp, 1); for (i=0;ibuffers[i], &vecs[i]); } buf_o->n_buffers = nvecs; buf_o->first_pinned = chain = *chainp; npin=0; for ( ; chain; chain = chain->next) { evbuffer_chain_pin_(chain, EVBUFFER_MEM_PINNED_R); ++npin; } EVUTIL_ASSERT(npin == nvecs); evbuffer_incref_(buf); if (WSARecv(buf_o->fd, buf_o->buffers, nvecs, &bytesRead, &flags, &ol->overlapped, NULL)) { int error = WSAGetLastError(); if (error != WSA_IO_PENDING) { /* An actual error. */ pin_release(buf_o, EVBUFFER_MEM_PINNED_R); evbuffer_unfreeze(buf, 0); evbuffer_free(buf); /* decref */ goto done; } } buf_o->read_in_progress = 1; r = 0; done: EVBUFFER_UNLOCK(buf); return r; } evutil_socket_t evbuffer_overlapped_get_fd_(struct evbuffer *buf) { struct evbuffer_overlapped *buf_o = upcast_evbuffer(buf); return buf_o ? buf_o->fd : -1; } void evbuffer_overlapped_set_fd_(struct evbuffer *buf, evutil_socket_t fd) { struct evbuffer_overlapped *buf_o = upcast_evbuffer(buf); EVBUFFER_LOCK(buf); /* XXX is this right?, should it cancel current I/O operations? */ if (buf_o) buf_o->fd = fd; EVBUFFER_UNLOCK(buf); } /* * Copyright 2000-2009 Niels Provos * Copyright 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_DEVPOLL #include #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #include #include #include #include #include #include "event2/event.h" #include "event2/event_struct.h" #include "event2/thread.h" #include "event-internal.h" #include "evsignal-internal.h" #include "log-internal.h" #include "evmap-internal.h" #include "evthread-internal.h" struct devpollop { struct pollfd *events; int nevents; int dpfd; struct pollfd *changes; int nchanges; }; static void *devpoll_init(struct event_base *); static int devpoll_add(struct event_base *, int fd, short old, short events, void *); static int devpoll_del(struct event_base *, int fd, short old, short events, void *); static int devpoll_dispatch(struct event_base *, struct timeval *); static void devpoll_dealloc(struct event_base *); const struct eventop devpollops = { "devpoll", devpoll_init, devpoll_add, devpoll_del, devpoll_dispatch, devpoll_dealloc, 1, /* need reinit */ EV_FEATURE_FDS|EV_FEATURE_O1, 0 }; #define NEVENT 32000 static int devpoll_commit(struct devpollop *devpollop) { /* * Due to a bug in Solaris, we have to use pwrite with an offset of 0. * Write is limited to 2GB of data, until it will fail. */ if (pwrite(devpollop->dpfd, devpollop->changes, sizeof(struct pollfd) * devpollop->nchanges, 0) == -1) return (-1); devpollop->nchanges = 0; return (0); } static int devpoll_queue(struct devpollop *devpollop, int fd, int events) { struct pollfd *pfd; if (devpollop->nchanges >= devpollop->nevents) { /* * Change buffer is full, must commit it to /dev/poll before * adding more */ if (devpoll_commit(devpollop) != 0) return (-1); } pfd = &devpollop->changes[devpollop->nchanges++]; pfd->fd = fd; pfd->events = events; pfd->revents = 0; return (0); } static void * devpoll_init(struct event_base *base) { int dpfd, nfiles = NEVENT; struct rlimit rl; struct devpollop *devpollop; if (!(devpollop = mm_calloc(1, sizeof(struct devpollop)))) return (NULL); if (getrlimit(RLIMIT_NOFILE, &rl) == 0 && rl.rlim_cur != RLIM_INFINITY) nfiles = rl.rlim_cur; /* Initialize the kernel queue */ if ((dpfd = evutil_open_closeonexec_("/dev/poll", O_RDWR, 0)) == -1) { event_warn("open: /dev/poll"); mm_free(devpollop); return (NULL); } devpollop->dpfd = dpfd; /* Initialize fields */ /* FIXME: allocating 'nfiles' worth of space here can be * expensive and unnecessary. See how epoll.c does it instead. */ devpollop->events = mm_calloc(nfiles, sizeof(struct pollfd)); if (devpollop->events == NULL) { mm_free(devpollop); close(dpfd); return (NULL); } devpollop->nevents = nfiles; devpollop->changes = mm_calloc(nfiles, sizeof(struct pollfd)); if (devpollop->changes == NULL) { mm_free(devpollop->events); mm_free(devpollop); close(dpfd); return (NULL); } evsig_init_(base); return (devpollop); } static int devpoll_dispatch(struct event_base *base, struct timeval *tv) { struct devpollop *devpollop = base->evbase; struct pollfd *events = devpollop->events; struct dvpoll dvp; int i, res, timeout = -1; if (devpollop->nchanges) devpoll_commit(devpollop); if (tv != NULL) timeout = tv->tv_sec * 1000 + (tv->tv_usec + 999) / 1000; dvp.dp_fds = devpollop->events; dvp.dp_nfds = devpollop->nevents; dvp.dp_timeout = timeout; EVBASE_RELEASE_LOCK(base, th_base_lock); res = ioctl(devpollop->dpfd, DP_POLL, &dvp); EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (res == -1) { if (errno != EINTR) { event_warn("ioctl: DP_POLL"); return (-1); } return (0); } event_debug(("%s: devpoll_wait reports %d", __func__, res)); for (i = 0; i < res; i++) { int which = 0; int what = events[i].revents; if (what & POLLHUP) what |= POLLIN | POLLOUT; else if (what & POLLERR) what |= POLLIN | POLLOUT; if (what & POLLIN) which |= EV_READ; if (what & POLLOUT) which |= EV_WRITE; if (!which) continue; /* XXX(niels): not sure if this works for devpoll */ evmap_io_active_(base, events[i].fd, which); } return (0); } static int devpoll_add(struct event_base *base, int fd, short old, short events, void *p) { struct devpollop *devpollop = base->evbase; int res; (void)p; /* * It's not necessary to OR the existing read/write events that we * are currently interested in with the new event we are adding. * The /dev/poll driver ORs any new events with the existing events * that it has cached for the fd. */ res = 0; if (events & EV_READ) res |= POLLIN; if (events & EV_WRITE) res |= POLLOUT; if (devpoll_queue(devpollop, fd, res) != 0) return (-1); return (0); } static int devpoll_del(struct event_base *base, int fd, short old, short events, void *p) { struct devpollop *devpollop = base->evbase; int res; (void)p; res = 0; if (events & EV_READ) res |= POLLIN; if (events & EV_WRITE) res |= POLLOUT; /* * The only way to remove an fd from the /dev/poll monitored set is * to use POLLREMOVE by itself. This removes ALL events for the fd * provided so if we care about two events and are only removing one * we must re-add the other event after POLLREMOVE. */ if (devpoll_queue(devpollop, fd, POLLREMOVE) != 0) return (-1); if ((res & (POLLIN|POLLOUT)) != (POLLIN|POLLOUT)) { /* * We're not deleting all events, so we must resubmit the * event that we are still interested in if one exists. */ if ((res & POLLIN) && (old & EV_WRITE)) { /* Deleting read, still care about write */ devpoll_queue(devpollop, fd, POLLOUT); } else if ((res & POLLOUT) && (old & EV_READ)) { /* Deleting write, still care about read */ devpoll_queue(devpollop, fd, POLLIN); } } return (0); } static void devpoll_dealloc(struct event_base *base) { struct devpollop *devpollop = base->evbase; evsig_dealloc_(base); if (devpollop->events) mm_free(devpollop->events); if (devpollop->changes) mm_free(devpollop->changes); if (devpollop->dpfd >= 0) close(devpollop->dpfd); memset(devpollop, 0, sizeof(struct devpollop)); mm_free(devpollop); } #endif /* EVENT__HAVE_DEVPOLL */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_EPOLL #include #include #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #include #include #include #include #include #ifdef EVENT__HAVE_FCNTL_H #include #endif #ifdef EVENT__HAVE_SYS_TIMERFD_H #include #endif #include "event-internal.h" #include "evsignal-internal.h" #include "event2/thread.h" #include "evthread-internal.h" #include "log-internal.h" #include "evmap-internal.h" #include "changelist-internal.h" #include "time-internal.h" /* Since Linux 2.6.17, epoll is able to report about peer half-closed connection using special EPOLLRDHUP flag on a read event. */ #if !defined(EPOLLRDHUP) #define EPOLLRDHUP 0 #define EARLY_CLOSE_IF_HAVE_RDHUP 0 #else #define EARLY_CLOSE_IF_HAVE_RDHUP EV_FEATURE_EARLY_CLOSE #endif #include "epolltable-internal.h" #if defined(EVENT__HAVE_SYS_TIMERFD_H) && \ defined(EVENT__HAVE_TIMERFD_CREATE) && \ defined(HAVE_POSIX_MONOTONIC) && defined(TFD_NONBLOCK) && \ defined(TFD_CLOEXEC) /* Note that we only use timerfd if TFD_NONBLOCK and TFD_CLOEXEC are available and working. This means that we can't support it on 2.6.25 (where timerfd was introduced) or 2.6.26, since 2.6.27 introduced those flags. */ #define USING_TIMERFD #endif struct epollop { struct epoll_event *events; int nevents; int epfd; #ifdef USING_TIMERFD int timerfd; #endif }; static void *epoll_init(struct event_base *); static int epoll_dispatch(struct event_base *, struct timeval *); static void epoll_dealloc(struct event_base *); static const struct eventop epollops_changelist = { "epoll (with changelist)", epoll_init, event_changelist_add_, event_changelist_del_, epoll_dispatch, epoll_dealloc, 1, /* need reinit */ EV_FEATURE_ET|EV_FEATURE_O1| EARLY_CLOSE_IF_HAVE_RDHUP, EVENT_CHANGELIST_FDINFO_SIZE }; static int epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd, short old, short events, void *p); static int epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd, short old, short events, void *p); const struct eventop epollops = { "epoll", epoll_init, epoll_nochangelist_add, epoll_nochangelist_del, epoll_dispatch, epoll_dealloc, 1, /* need reinit */ EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_EARLY_CLOSE, 0 }; #define INITIAL_NEVENT 32 #define MAX_NEVENT 4096 /* On Linux kernels at least up to 2.6.24.4, epoll can't handle timeout * values bigger than (LONG_MAX - 999ULL)/HZ. HZ in the wild can be * as big as 1000, and LONG_MAX can be as small as (1<<31)-1, so the * largest number of msec we can support here is 2147482. Let's * round that down by 47 seconds. */ #define MAX_EPOLL_TIMEOUT_MSEC (35*60*1000) static void * epoll_init(struct event_base *base) { int epfd = -1; struct epollop *epollop; #ifdef EVENT__HAVE_EPOLL_CREATE1 /* First, try the shiny new epoll_create1 interface, if we have it. */ epfd = epoll_create1(EPOLL_CLOEXEC); #endif if (epfd == -1) { /* Initialize the kernel queue using the old interface. (The size field is ignored since 2.6.8.) */ if ((epfd = epoll_create(32000)) == -1) { if (errno != ENOSYS) event_warn("epoll_create"); return (NULL); } evutil_make_socket_closeonexec(epfd); } if (!(epollop = mm_calloc(1, sizeof(struct epollop)))) { close(epfd); return (NULL); } epollop->epfd = epfd; /* Initialize fields */ epollop->events = mm_calloc(INITIAL_NEVENT, sizeof(struct epoll_event)); if (epollop->events == NULL) { mm_free(epollop); close(epfd); return (NULL); } epollop->nevents = INITIAL_NEVENT; if ((base->flags & EVENT_BASE_FLAG_EPOLL_USE_CHANGELIST) != 0 || ((base->flags & EVENT_BASE_FLAG_IGNORE_ENV) == 0 && evutil_getenv_("EVENT_EPOLL_USE_CHANGELIST") != NULL)) { base->evsel = &epollops_changelist; } #ifdef USING_TIMERFD /* The epoll interface ordinarily gives us one-millisecond precision, so on Linux it makes perfect sense to use the CLOCK_MONOTONIC_COARSE timer. But when the user has set the new PRECISE_TIMER flag for an event_base, we can try to use timerfd to give them finer granularity. */ if ((base->flags & EVENT_BASE_FLAG_PRECISE_TIMER) && base->monotonic_timer.monotonic_clock == CLOCK_MONOTONIC) { int fd; fd = epollop->timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); if (epollop->timerfd >= 0) { struct epoll_event epev; memset(&epev, 0, sizeof(epev)); epev.data.fd = epollop->timerfd; epev.events = EPOLLIN; if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, fd, &epev) < 0) { event_warn("epoll_ctl(timerfd)"); close(fd); epollop->timerfd = -1; } } else { if (errno != EINVAL && errno != ENOSYS) { /* These errors probably mean that we were * compiled with timerfd/TFD_* support, but * we're running on a kernel that lacks those. */ event_warn("timerfd_create"); } epollop->timerfd = -1; } } else { epollop->timerfd = -1; } #endif evsig_init_(base); return (epollop); } static const char * change_to_string(int change) { change &= (EV_CHANGE_ADD|EV_CHANGE_DEL); if (change == EV_CHANGE_ADD) { return "add"; } else if (change == EV_CHANGE_DEL) { return "del"; } else if (change == 0) { return "none"; } else { return "???"; } } static const char * epoll_op_to_string(int op) { return op == EPOLL_CTL_ADD?"ADD": op == EPOLL_CTL_DEL?"DEL": op == EPOLL_CTL_MOD?"MOD": "???"; } #define PRINT_CHANGES(op, events, ch, status) \ "Epoll %s(%d) on fd %d " status ". " \ "Old events were %d; " \ "read change was %d (%s); " \ "write change was %d (%s); " \ "close change was %d (%s)", \ epoll_op_to_string(op), \ events, \ ch->fd, \ ch->old_events, \ ch->read_change, \ change_to_string(ch->read_change), \ ch->write_change, \ change_to_string(ch->write_change), \ ch->close_change, \ change_to_string(ch->close_change) static int epoll_apply_one_change(struct event_base *base, struct epollop *epollop, const struct event_change *ch) { struct epoll_event epev; int op, events = 0; int idx; idx = EPOLL_OP_TABLE_INDEX(ch); op = epoll_op_table[idx].op; events = epoll_op_table[idx].events; if (!events) { EVUTIL_ASSERT(op == 0); return 0; } if ((ch->read_change|ch->write_change|ch->close_change) & EV_CHANGE_ET) events |= EPOLLET; memset(&epev, 0, sizeof(epev)); epev.data.fd = ch->fd; epev.events = events; if (epoll_ctl(epollop->epfd, op, ch->fd, &epev) == 0) { event_debug((PRINT_CHANGES(op, epev.events, ch, "okay"))); return 0; } switch (op) { case EPOLL_CTL_MOD: if (errno == ENOENT) { /* If a MOD operation fails with ENOENT, the * fd was probably closed and re-opened. We * should retry the operation as an ADD. */ if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, ch->fd, &epev) == -1) { event_warn("Epoll MOD(%d) on %d retried as ADD; that failed too", (int)epev.events, ch->fd); return -1; } else { event_debug(("Epoll MOD(%d) on %d retried as ADD; succeeded.", (int)epev.events, ch->fd)); return 0; } } break; case EPOLL_CTL_ADD: if (errno == EEXIST) { /* If an ADD operation fails with EEXIST, * either the operation was redundant (as with a * precautionary add), or we ran into a fun * kernel bug where using dup*() to duplicate the * same file into the same fd gives you the same epitem * rather than a fresh one. For the second case, * we must retry with MOD. */ if (epoll_ctl(epollop->epfd, EPOLL_CTL_MOD, ch->fd, &epev) == -1) { event_warn("Epoll ADD(%d) on %d retried as MOD; that failed too", (int)epev.events, ch->fd); return -1; } else { event_debug(("Epoll ADD(%d) on %d retried as MOD; succeeded.", (int)epev.events, ch->fd)); return 0; } } break; case EPOLL_CTL_DEL: if (errno == ENOENT || errno == EBADF || errno == EPERM) { /* If a delete fails with one of these errors, * that's fine too: we closed the fd before we * got around to calling epoll_dispatch. */ event_debug(("Epoll DEL(%d) on fd %d gave %s: DEL was unnecessary.", (int)epev.events, ch->fd, strerror(errno))); return 0; } break; default: break; } event_warn(PRINT_CHANGES(op, epev.events, ch, "failed")); return -1; } static int epoll_apply_changes(struct event_base *base) { struct event_changelist *changelist = &base->changelist; struct epollop *epollop = base->evbase; struct event_change *ch; int r = 0; int i; for (i = 0; i < changelist->n_changes; ++i) { ch = &changelist->changes[i]; if (epoll_apply_one_change(base, epollop, ch) < 0) r = -1; } return (r); } static int epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd, short old, short events, void *p) { struct event_change ch; ch.fd = fd; ch.old_events = old; ch.read_change = ch.write_change = ch.close_change = 0; if (events & EV_WRITE) ch.write_change = EV_CHANGE_ADD | (events & EV_ET); if (events & EV_READ) ch.read_change = EV_CHANGE_ADD | (events & EV_ET); if (events & EV_CLOSED) ch.close_change = EV_CHANGE_ADD | (events & EV_ET); return epoll_apply_one_change(base, base->evbase, &ch); } static int epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd, short old, short events, void *p) { struct event_change ch; ch.fd = fd; ch.old_events = old; ch.read_change = ch.write_change = ch.close_change = 0; if (events & EV_WRITE) ch.write_change = EV_CHANGE_DEL | (events & EV_ET); if (events & EV_READ) ch.read_change = EV_CHANGE_DEL | (events & EV_ET); if (events & EV_CLOSED) ch.close_change = EV_CHANGE_DEL | (events & EV_ET); return epoll_apply_one_change(base, base->evbase, &ch); } static int epoll_dispatch(struct event_base *base, struct timeval *tv) { struct epollop *epollop = base->evbase; struct epoll_event *events = epollop->events; int i, res; long timeout = -1; #ifdef USING_TIMERFD if (epollop->timerfd >= 0) { struct itimerspec is; is.it_interval.tv_sec = 0; is.it_interval.tv_nsec = 0; if (tv == NULL) { /* No timeout; disarm the timer. */ is.it_value.tv_sec = 0; is.it_value.tv_nsec = 0; } else { if (tv->tv_sec == 0 && tv->tv_usec == 0) { /* we need to exit immediately; timerfd can't * do that. */ timeout = 0; } is.it_value.tv_sec = tv->tv_sec; is.it_value.tv_nsec = tv->tv_usec * 1000; } /* TODO: we could avoid unnecessary syscalls here by only calling timerfd_settime when the top timeout changes, or when we're called with a different timeval. */ if (timerfd_settime(epollop->timerfd, 0, &is, NULL) < 0) { event_warn("timerfd_settime"); } } else #endif if (tv != NULL) { timeout = evutil_tv_to_msec_(tv); if (timeout < 0 || timeout > MAX_EPOLL_TIMEOUT_MSEC) { /* Linux kernels can wait forever if the timeout is * too big; see comment on MAX_EPOLL_TIMEOUT_MSEC. */ timeout = MAX_EPOLL_TIMEOUT_MSEC; } } epoll_apply_changes(base); event_changelist_remove_all_(&base->changelist, base); EVBASE_RELEASE_LOCK(base, th_base_lock); res = epoll_wait(epollop->epfd, events, epollop->nevents, timeout); EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (res == -1) { if (errno != EINTR) { event_warn("epoll_wait"); return (-1); } return (0); } event_debug(("%s: epoll_wait reports %d", __func__, res)); EVUTIL_ASSERT(res <= epollop->nevents); for (i = 0; i < res; i++) { int what = events[i].events; short ev = 0; #ifdef USING_TIMERFD if (events[i].data.fd == epollop->timerfd) continue; #endif if (what & EPOLLERR) { ev = EV_READ | EV_WRITE; } else if ((what & EPOLLHUP) && !(what & EPOLLRDHUP)) { ev = EV_READ | EV_WRITE; } else { if (what & EPOLLIN) ev |= EV_READ; if (what & EPOLLOUT) ev |= EV_WRITE; if (what & EPOLLRDHUP) ev |= EV_CLOSED; } if (!ev) continue; evmap_io_active_(base, events[i].data.fd, ev | EV_ET); } if (res == epollop->nevents && epollop->nevents < MAX_NEVENT) { /* We used all of the event space this time. We should be ready for more events next time. */ int new_nevents = epollop->nevents * 2; struct epoll_event *new_events; new_events = mm_realloc(epollop->events, new_nevents * sizeof(struct epoll_event)); if (new_events) { epollop->events = new_events; epollop->nevents = new_nevents; } } return (0); } static void epoll_dealloc(struct event_base *base) { struct epollop *epollop = base->evbase; evsig_dealloc_(base); if (epollop->events) mm_free(epollop->events); if (epollop->epfd >= 0) close(epollop->epfd); #ifdef USING_TIMERFD if (epollop->timerfd >= 0) close(epollop->timerfd); #endif memset(epollop, 0, sizeof(struct epollop)); mm_free(epollop); } #endif /* EVENT__HAVE_EPOLL */ /* * Copyright 2003-2009 Niels Provos * Copyright 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "evconfig-private.h" #include #include #include #include #include #include #include int epoll_create(int size) { #if !defined(__NR_epoll_create) && defined(__NR_epoll_create1) if (size <= 0) { errno = EINVAL; return -1; } return (syscall(__NR_epoll_create1, 0)); #else return (syscall(__NR_epoll_create, size)); #endif } int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event) { return (syscall(__NR_epoll_ctl, epfd, op, fd, event)); } int epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout) { #if !defined(__NR_epoll_wait) && defined(__NR_epoll_pwait) return (syscall(__NR_epoll_pwait, epfd, events, maxevents, timeout, NULL, 0)); #else return (syscall(__NR_epoll_wait, epfd, events, maxevents, timeout)); #endif } /* Copyright 2006-2007 Niels Provos * Copyright 2007-2012 Nick Mathewson and Niels Provos * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* Based on software by Adam Langly. Adam's original message: * * Async DNS Library * Adam Langley * http://www.imperialviolet.org/eventdns.html * Public Domain code * * This software is Public Domain. To view a copy of the public domain dedication, * visit http://creativecommons.org/licenses/publicdomain/ or send a letter to * Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA. * * I ask and expect, but do not require, that all derivative works contain an * attribution similar to: * Parts developed by Adam Langley * * You may wish to replace the word "Parts" with something else depending on * the amount of original code. * * (Derivative works does not include programs which link against, run or include * the source verbatim in their source distributions) * * Version: 0.1b */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifndef _FORTIFY_SOURCE #define _FORTIFY_SOURCE 3 #endif #include #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #ifdef EVENT__HAVE_STDINT_H #include #endif #include #include #include #ifdef EVENT__HAVE_UNISTD_H #include #endif #include #include #include #include #ifdef _WIN32 #include #include #include #ifndef _WIN32_IE #define _WIN32_IE 0x400 #endif #include #endif #include "event2/dns.h" #include "event2/dns_struct.h" #include "event2/dns_compat.h" #include "event2/util.h" #include "event2/event.h" #include "event2/event_struct.h" #include "event2/thread.h" #include "defer-internal.h" #include "log-internal.h" #include "mm-internal.h" #include "strlcpy-internal.h" #include "ipv6-internal.h" #include "util-internal.h" #include "evthread-internal.h" #ifdef _WIN32 #include #include #include #include #include #else #include #include #include #endif #ifdef EVENT__HAVE_NETINET_IN6_H #include #endif #define EVDNS_LOG_DEBUG EVENT_LOG_DEBUG #define EVDNS_LOG_WARN EVENT_LOG_WARN #define EVDNS_LOG_MSG EVENT_LOG_MSG #ifndef HOST_NAME_MAX #define HOST_NAME_MAX 255 #endif #include #undef MIN #define MIN(a,b) ((a)<(b)?(a):(b)) #define ASSERT_VALID_REQUEST(req) \ EVUTIL_ASSERT((req)->handle && (req)->handle->current_req == (req)) #define u64 ev_uint64_t #define u32 ev_uint32_t #define u16 ev_uint16_t #define u8 ev_uint8_t /* maximum number of addresses from a single packet */ /* that we bother recording */ #define MAX_V4_ADDRS 32 #define MAX_V6_ADDRS 32 #define TYPE_A EVDNS_TYPE_A #define TYPE_CNAME 5 #define TYPE_PTR EVDNS_TYPE_PTR #define TYPE_SOA EVDNS_TYPE_SOA #define TYPE_AAAA EVDNS_TYPE_AAAA #define CLASS_INET EVDNS_CLASS_INET /* Persistent handle. We keep this separate from 'struct request' since we * need some object to last for as long as an evdns_request is outstanding so * that it can be canceled, whereas a search request can lead to multiple * 'struct request' instances being created over its lifetime. */ struct evdns_request { struct request *current_req; struct evdns_base *base; int pending_cb; /* Waiting for its callback to be invoked; not * owned by event base any more. */ /* elements used by the searching code */ int search_index; struct search_state *search_state; char *search_origname; /* needs to be free()ed */ int search_flags; }; struct request { u8 *request; /* the dns packet data */ u8 request_type; /* TYPE_PTR or TYPE_A or TYPE_AAAA */ unsigned int request_len; int reissue_count; int tx_count; /* the number of times that this packet has been sent */ void *user_pointer; /* the pointer given to us for this request */ evdns_callback_type user_callback; struct nameserver *ns; /* the server which we last sent it */ /* these objects are kept in a circular list */ /* XXX We could turn this into a CIRCLEQ. */ struct request *next, *prev; struct event timeout_event; u16 trans_id; /* the transaction id */ unsigned request_appended :1; /* true if the request pointer is data which follows this struct */ unsigned transmit_me :1; /* needs to be transmitted */ /* XXXX This is a horrible hack. */ char **put_cname_in_ptr; /* store the cname here if we get one. */ struct evdns_base *base; struct evdns_request *handle; }; struct reply { unsigned int type; unsigned int have_answer : 1; union { struct { u32 addrcount; u32 addresses[MAX_V4_ADDRS]; } a; struct { u32 addrcount; struct in6_addr addresses[MAX_V6_ADDRS]; } aaaa; struct { char name[HOST_NAME_MAX]; } ptr; } data; }; struct nameserver { evutil_socket_t socket; /* a connected UDP socket */ struct sockaddr_storage address; ev_socklen_t addrlen; int failed_times; /* number of times which we have given this server a chance */ int timedout; /* number of times in a row a request has timed out */ struct event event; /* these objects are kept in a circular list */ struct nameserver *next, *prev; struct event timeout_event; /* used to keep the timeout for */ /* when we next probe this server. */ /* Valid if state == 0 */ /* Outstanding probe request for this nameserver, if any */ struct evdns_request *probe_request; char state; /* zero if we think that this server is down */ char choked; /* true if we have an EAGAIN from this server's socket */ char write_waiting; /* true if we are waiting for EV_WRITE events */ struct evdns_base *base; /* Number of currently inflight requests: used * to track when we should add/del the event. */ int requests_inflight; }; /* Represents a local port where we're listening for DNS requests. Right now, */ /* only UDP is supported. */ struct evdns_server_port { evutil_socket_t socket; /* socket we use to read queries and write replies. */ int refcnt; /* reference count. */ char choked; /* Are we currently blocked from writing? */ char closing; /* Are we trying to close this port, pending writes? */ evdns_request_callback_fn_type user_callback; /* Fn to handle requests */ void *user_data; /* Opaque pointer passed to user_callback */ struct event event; /* Read/write event */ /* circular list of replies that we want to write. */ struct server_request *pending_replies; struct event_base *event_base; #ifndef EVENT__DISABLE_THREAD_SUPPORT void *lock; #endif }; /* Represents part of a reply being built. (That is, a single RR.) */ struct server_reply_item { struct server_reply_item *next; /* next item in sequence. */ char *name; /* name part of the RR */ u16 type; /* The RR type */ u16 class; /* The RR class (usually CLASS_INET) */ u32 ttl; /* The RR TTL */ char is_name; /* True iff data is a label */ u16 datalen; /* Length of data; -1 if data is a label */ void *data; /* The contents of the RR */ }; /* Represents a request that we've received as a DNS server, and holds */ /* the components of the reply as we're constructing it. */ struct server_request { /* Pointers to the next and previous entries on the list of replies */ /* that we're waiting to write. Only set if we have tried to respond */ /* and gotten EAGAIN. */ struct server_request *next_pending; struct server_request *prev_pending; u16 trans_id; /* Transaction id. */ struct evdns_server_port *port; /* Which port received this request on? */ struct sockaddr_storage addr; /* Where to send the response */ ev_socklen_t addrlen; /* length of addr */ int n_answer; /* how many answer RRs have been set? */ int n_authority; /* how many authority RRs have been set? */ int n_additional; /* how many additional RRs have been set? */ struct server_reply_item *answer; /* linked list of answer RRs */ struct server_reply_item *authority; /* linked list of authority RRs */ struct server_reply_item *additional; /* linked list of additional RRs */ /* Constructed response. Only set once we're ready to send a reply. */ /* Once this is set, the RR fields are cleared, and no more should be set. */ char *response; size_t response_len; /* Caller-visible fields: flags, questions. */ struct evdns_server_request base; }; struct evdns_base { /* An array of n_req_heads circular lists for inflight requests. * Each inflight request req is in req_heads[req->trans_id % n_req_heads]. */ struct request **req_heads; /* A circular list of requests that we're waiting to send, but haven't * sent yet because there are too many requests inflight */ struct request *req_waiting_head; /* A circular list of nameservers. */ struct nameserver *server_head; int n_req_heads; struct event_base *event_base; /* The number of good nameservers that we have */ int global_good_nameservers; /* inflight requests are contained in the req_head list */ /* and are actually going out across the network */ int global_requests_inflight; /* requests which aren't inflight are in the waiting list */ /* and are counted here */ int global_requests_waiting; int global_max_requests_inflight; struct timeval global_timeout; /* 5 seconds by default */ int global_max_reissues; /* a reissue occurs when we get some errors from the server */ int global_max_retransmits; /* number of times we'll retransmit a request which timed out */ /* number of timeouts in a row before we consider this server to be down */ int global_max_nameserver_timeout; /* true iff we will use the 0x20 hack to prevent poisoning attacks. */ int global_randomize_case; /* The first time that a nameserver fails, how long do we wait before * probing to see if it has returned? */ struct timeval global_nameserver_probe_initial_timeout; /** Port to bind to for outgoing DNS packets. */ struct sockaddr_storage global_outgoing_address; /** ev_socklen_t for global_outgoing_address. 0 if it isn't set. */ ev_socklen_t global_outgoing_addrlen; struct timeval global_getaddrinfo_allow_skew; int so_rcvbuf; int so_sndbuf; int getaddrinfo_ipv4_timeouts; int getaddrinfo_ipv6_timeouts; int getaddrinfo_ipv4_answered; int getaddrinfo_ipv6_answered; struct search_state *global_search_state; TAILQ_HEAD(hosts_list, hosts_entry) hostsdb; #ifndef EVENT__DISABLE_THREAD_SUPPORT void *lock; #endif int disable_when_inactive; }; struct hosts_entry { TAILQ_ENTRY(hosts_entry) next; union { struct sockaddr sa; struct sockaddr_in sin; struct sockaddr_in6 sin6; } addr; int addrlen; char hostname[1]; }; static struct evdns_base *current_base = NULL; struct evdns_base * evdns_get_global_base(void) { return current_base; } /* Given a pointer to an evdns_server_request, get the corresponding */ /* server_request. */ #define TO_SERVER_REQUEST(base_ptr) \ ((struct server_request*) \ (((char*)(base_ptr) - evutil_offsetof(struct server_request, base)))) #define REQ_HEAD(base, id) ((base)->req_heads[id % (base)->n_req_heads]) static struct nameserver *nameserver_pick(struct evdns_base *base); static void evdns_request_insert(struct request *req, struct request **head); static void evdns_request_remove(struct request *req, struct request **head); static void nameserver_ready_callback(evutil_socket_t fd, short events, void *arg); static int evdns_transmit(struct evdns_base *base); static int evdns_request_transmit(struct request *req); static void nameserver_send_probe(struct nameserver *const ns); static void search_request_finished(struct evdns_request *const); static int search_try_next(struct evdns_request *const req); static struct request *search_request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg); static void evdns_requests_pump_waiting_queue(struct evdns_base *base); static u16 transaction_id_pick(struct evdns_base *base); static struct request *request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *name, int flags, evdns_callback_type callback, void *ptr); static void request_submit(struct request *const req); static int server_request_free(struct server_request *req); static void server_request_free_answers(struct server_request *req); static void server_port_free(struct evdns_server_port *port); static void server_port_ready_callback(evutil_socket_t fd, short events, void *arg); static int evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename); static int evdns_base_set_option_impl(struct evdns_base *base, const char *option, const char *val, int flags); static void evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests); static void evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg); static int strtoint(const char *const str); #ifdef EVENT__DISABLE_THREAD_SUPPORT #define EVDNS_LOCK(base) EVUTIL_NIL_STMT_ #define EVDNS_UNLOCK(base) EVUTIL_NIL_STMT_ #define ASSERT_LOCKED(base) EVUTIL_NIL_STMT_ #else #define EVDNS_LOCK(base) \ EVLOCK_LOCK((base)->lock, 0) #define EVDNS_UNLOCK(base) \ EVLOCK_UNLOCK((base)->lock, 0) #define ASSERT_LOCKED(base) \ EVLOCK_ASSERT_LOCKED((base)->lock) #endif static evdns_debug_log_fn_type evdns_log_fn = NULL; void evdns_set_log_fn(evdns_debug_log_fn_type fn) { evdns_log_fn = fn; } #ifdef __GNUC__ #define EVDNS_LOG_CHECK __attribute__ ((format(printf, 2, 3))) #else #define EVDNS_LOG_CHECK #endif static void evdns_log_(int severity, const char *fmt, ...) EVDNS_LOG_CHECK; static void evdns_log_(int severity, const char *fmt, ...) { va_list args; va_start(args,fmt); if (evdns_log_fn) { char buf[512]; int is_warn = (severity == EVDNS_LOG_WARN); evutil_vsnprintf(buf, sizeof(buf), fmt, args); evdns_log_fn(is_warn, buf); } else { event_logv_(severity, NULL, fmt, args); } va_end(args); } #define log evdns_log_ /* This walks the list of inflight requests to find the */ /* one with a matching transaction id. Returns NULL on */ /* failure */ static struct request * request_find_from_trans_id(struct evdns_base *base, u16 trans_id) { struct request *req = REQ_HEAD(base, trans_id); struct request *const started_at = req; ASSERT_LOCKED(base); if (req) { do { if (req->trans_id == trans_id) return req; req = req->next; } while (req != started_at); } return NULL; } /* a libevent callback function which is called when a nameserver */ /* has gone down and we want to test if it has came back to life yet */ static void nameserver_prod_callback(evutil_socket_t fd, short events, void *arg) { struct nameserver *const ns = (struct nameserver *) arg; (void)fd; (void)events; EVDNS_LOCK(ns->base); nameserver_send_probe(ns); EVDNS_UNLOCK(ns->base); } /* a libevent callback which is called when a nameserver probe (to see if */ /* it has come back to life) times out. We increment the count of failed_times */ /* and wait longer to send the next probe packet. */ static void nameserver_probe_failed(struct nameserver *const ns) { struct timeval timeout; int i; ASSERT_LOCKED(ns->base); (void) evtimer_del(&ns->timeout_event); if (ns->state == 1) { /* This can happen if the nameserver acts in a way which makes us mark */ /* it as bad and then starts sending good replies. */ return; } #define MAX_PROBE_TIMEOUT 3600 #define TIMEOUT_BACKOFF_FACTOR 3 memcpy(&timeout, &ns->base->global_nameserver_probe_initial_timeout, sizeof(struct timeval)); for (i=ns->failed_times; i > 0 && timeout.tv_sec < MAX_PROBE_TIMEOUT; --i) { timeout.tv_sec *= TIMEOUT_BACKOFF_FACTOR; timeout.tv_usec *= TIMEOUT_BACKOFF_FACTOR; if (timeout.tv_usec > 1000000) { timeout.tv_sec += timeout.tv_usec / 1000000; timeout.tv_usec %= 1000000; } } if (timeout.tv_sec > MAX_PROBE_TIMEOUT) { timeout.tv_sec = MAX_PROBE_TIMEOUT; timeout.tv_usec = 0; } ns->failed_times++; if (evtimer_add(&ns->timeout_event, &timeout) < 0) { char addrbuf[128]; log(EVDNS_LOG_WARN, "Error from libevent when adding timer event for %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf))); } } static void request_swap_ns(struct request *req, struct nameserver *ns) { if (ns && req->ns != ns) { EVUTIL_ASSERT(req->ns->requests_inflight > 0); req->ns->requests_inflight--; ns->requests_inflight++; req->ns = ns; } } /* called when a nameserver has been deemed to have failed. For example, too */ /* many packets have timed out etc */ static void nameserver_failed(struct nameserver *const ns, const char *msg) { struct request *req, *started_at; struct evdns_base *base = ns->base; int i; char addrbuf[128]; ASSERT_LOCKED(base); /* if this nameserver has already been marked as failed */ /* then don't do anything */ if (!ns->state) return; log(EVDNS_LOG_MSG, "Nameserver %s has failed: %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf)), msg); base->global_good_nameservers--; EVUTIL_ASSERT(base->global_good_nameservers >= 0); if (base->global_good_nameservers == 0) { log(EVDNS_LOG_MSG, "All nameservers have failed"); } ns->state = 0; ns->failed_times = 1; if (evtimer_add(&ns->timeout_event, &base->global_nameserver_probe_initial_timeout) < 0) { log(EVDNS_LOG_WARN, "Error from libevent when adding timer event for %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf))); /* ???? Do more? */ } /* walk the list of inflight requests to see if any can be reassigned to */ /* a different server. Requests in the waiting queue don't have a */ /* nameserver assigned yet */ /* if we don't have *any* good nameservers then there's no point */ /* trying to reassign requests to one */ if (!base->global_good_nameservers) return; for (i = 0; i < base->n_req_heads; ++i) { req = started_at = base->req_heads[i]; if (req) { do { if (req->tx_count == 0 && req->ns == ns) { /* still waiting to go out, can be moved */ /* to another server */ request_swap_ns(req, nameserver_pick(base)); } req = req->next; } while (req != started_at); } } } static void nameserver_up(struct nameserver *const ns) { char addrbuf[128]; ASSERT_LOCKED(ns->base); if (ns->state) return; log(EVDNS_LOG_MSG, "Nameserver %s is back up", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf))); evtimer_del(&ns->timeout_event); if (ns->probe_request) { evdns_cancel_request(ns->base, ns->probe_request); ns->probe_request = NULL; } ns->state = 1; ns->failed_times = 0; ns->timedout = 0; ns->base->global_good_nameservers++; } static void request_trans_id_set(struct request *const req, const u16 trans_id) { req->trans_id = trans_id; *((u16 *) req->request) = htons(trans_id); } /* Called to remove a request from a list and dealloc it. */ /* head is a pointer to the head of the list it should be */ /* removed from or NULL if the request isn't in a list. */ /* when free_handle is one, free the handle as well. */ static void request_finished(struct request *const req, struct request **head, int free_handle) { struct evdns_base *base = req->base; int was_inflight = (head != &base->req_waiting_head); EVDNS_LOCK(base); ASSERT_VALID_REQUEST(req); if (head) evdns_request_remove(req, head); log(EVDNS_LOG_DEBUG, "Removing timeout for request %p", req); if (was_inflight) { evtimer_del(&req->timeout_event); base->global_requests_inflight--; req->ns->requests_inflight--; } else { base->global_requests_waiting--; } /* it was initialized during request_new / evtimer_assign */ event_debug_unassign(&req->timeout_event); if (req->ns && req->ns->requests_inflight == 0 && req->base->disable_when_inactive) { event_del(&req->ns->event); evtimer_del(&req->ns->timeout_event); } if (!req->request_appended) { /* need to free the request data on it's own */ mm_free(req->request); } else { /* the request data is appended onto the header */ /* so everything gets free()ed when we: */ } if (req->handle) { EVUTIL_ASSERT(req->handle->current_req == req); if (free_handle) { search_request_finished(req->handle); req->handle->current_req = NULL; if (! req->handle->pending_cb) { /* If we're planning to run the callback, * don't free the handle until later. */ mm_free(req->handle); } req->handle = NULL; /* If we have a bug, let's crash * early */ } else { req->handle->current_req = NULL; } } mm_free(req); evdns_requests_pump_waiting_queue(base); EVDNS_UNLOCK(base); } /* This is called when a server returns a funny error code. */ /* We try the request again with another server. */ /* */ /* return: */ /* 0 ok */ /* 1 failed/reissue is pointless */ static int request_reissue(struct request *req) { const struct nameserver *const last_ns = req->ns; ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); /* the last nameserver should have been marked as failing */ /* by the caller of this function, therefore pick will try */ /* not to return it */ request_swap_ns(req, nameserver_pick(req->base)); if (req->ns == last_ns) { /* ... but pick did return it */ /* not a lot of point in trying again with the */ /* same server */ return 1; } req->reissue_count++; req->tx_count = 0; req->transmit_me = 1; return 0; } /* this function looks for space on the inflight queue and promotes */ /* requests from the waiting queue if it can. */ /* */ /* TODO: */ /* add return code, see at nameserver_pick() and other functions. */ static void evdns_requests_pump_waiting_queue(struct evdns_base *base) { ASSERT_LOCKED(base); while (base->global_requests_inflight < base->global_max_requests_inflight && base->global_requests_waiting) { struct request *req; EVUTIL_ASSERT(base->req_waiting_head); req = base->req_waiting_head; req->ns = nameserver_pick(base); if (!req->ns) return; /* move a request from the waiting queue to the inflight queue */ req->ns->requests_inflight++; evdns_request_remove(req, &base->req_waiting_head); base->global_requests_waiting--; base->global_requests_inflight++; request_trans_id_set(req, transaction_id_pick(base)); evdns_request_insert(req, &REQ_HEAD(base, req->trans_id)); evdns_request_transmit(req); evdns_transmit(base); } } /* TODO(nickm) document */ struct deferred_reply_callback { struct event_callback deferred; struct evdns_request *handle; u8 request_type; u8 have_reply; u32 ttl; u32 err; evdns_callback_type user_callback; struct reply reply; }; static void reply_run_callback(struct event_callback *d, void *user_pointer) { struct deferred_reply_callback *cb = EVUTIL_UPCAST(d, struct deferred_reply_callback, deferred); switch (cb->request_type) { case TYPE_A: if (cb->have_reply) cb->user_callback(DNS_ERR_NONE, DNS_IPv4_A, cb->reply.data.a.addrcount, cb->ttl, cb->reply.data.a.addresses, user_pointer); else cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer); break; case TYPE_PTR: if (cb->have_reply) { char *name = cb->reply.data.ptr.name; cb->user_callback(DNS_ERR_NONE, DNS_PTR, 1, cb->ttl, &name, user_pointer); } else { cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer); } break; case TYPE_AAAA: if (cb->have_reply) cb->user_callback(DNS_ERR_NONE, DNS_IPv6_AAAA, cb->reply.data.aaaa.addrcount, cb->ttl, cb->reply.data.aaaa.addresses, user_pointer); else cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer); break; default: EVUTIL_ASSERT(0); } if (cb->handle && cb->handle->pending_cb) { mm_free(cb->handle); } mm_free(cb); } static void reply_schedule_callback(struct request *const req, u32 ttl, u32 err, struct reply *reply) { struct deferred_reply_callback *d = mm_calloc(1, sizeof(*d)); if (!d) { event_warn("%s: Couldn't allocate space for deferred callback.", __func__); return; } ASSERT_LOCKED(req->base); d->request_type = req->request_type; d->user_callback = req->user_callback; d->ttl = ttl; d->err = err; if (reply) { d->have_reply = 1; memcpy(&d->reply, reply, sizeof(struct reply)); } if (req->handle) { req->handle->pending_cb = 1; d->handle = req->handle; } event_deferred_cb_init_( &d->deferred, event_get_priority(&req->timeout_event), reply_run_callback, req->user_pointer); event_deferred_cb_schedule_( req->base->event_base, &d->deferred); } #define _QR_MASK 0x8000U #define _OP_MASK 0x7800U #define _AA_MASK 0x0400U #define _TC_MASK 0x0200U #define _RD_MASK 0x0100U #define _RA_MASK 0x0080U #define _Z_MASK 0x0040U #define _AD_MASK 0x0020U #define _CD_MASK 0x0010U #define _RCODE_MASK 0x000fU #define _Z_MASK_DEPRECATED 0x0070U /* this processes a parsed reply packet */ static void reply_handle(struct request *const req, u16 flags, u32 ttl, struct reply *reply) { int error; char addrbuf[128]; static const int error_codes[] = { DNS_ERR_FORMAT, DNS_ERR_SERVERFAILED, DNS_ERR_NOTEXIST, DNS_ERR_NOTIMPL, DNS_ERR_REFUSED }; ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); if (flags & (_RCODE_MASK | _TC_MASK) || !reply || !reply->have_answer) { /* there was an error */ if (flags & _TC_MASK) { error = DNS_ERR_TRUNCATED; } else if (flags & _RCODE_MASK) { u16 error_code = (flags & _RCODE_MASK) - 1; if (error_code > 4) { error = DNS_ERR_UNKNOWN; } else { error = error_codes[error_code]; } } else if (reply && !reply->have_answer) { error = DNS_ERR_NODATA; } else { error = DNS_ERR_UNKNOWN; } switch (error) { case DNS_ERR_NOTIMPL: case DNS_ERR_REFUSED: /* we regard these errors as marking a bad nameserver */ if (req->reissue_count < req->base->global_max_reissues) { char msg[64]; evutil_snprintf(msg, sizeof(msg), "Bad response %d (%s)", error, evdns_err_to_string(error)); nameserver_failed(req->ns, msg); if (!request_reissue(req)) return; } break; case DNS_ERR_SERVERFAILED: /* rcode 2 (servfailed) sometimes means "we * are broken" and sometimes (with some binds) * means "that request was very confusing." * Treat this as a timeout, not a failure. */ log(EVDNS_LOG_DEBUG, "Got a SERVERFAILED from nameserver" "at %s; will allow the request to time out.", evutil_format_sockaddr_port_( (struct sockaddr *)&req->ns->address, addrbuf, sizeof(addrbuf))); /* Call the timeout function */ evdns_request_timeout_callback(0, 0, req); return; default: /* we got a good reply from the nameserver: it is up. */ if (req->handle == req->ns->probe_request) { /* Avoid double-free */ req->ns->probe_request = NULL; } nameserver_up(req->ns); } if (req->handle->search_state && req->request_type != TYPE_PTR) { /* if we have a list of domains to search in, * try the next one */ if (!search_try_next(req->handle)) { /* a new request was issued so this * request is finished and */ /* the user callback will be made when * that request (or a */ /* child of it) finishes. */ return; } } /* all else failed. Pass the failure up */ reply_schedule_callback(req, ttl, error, NULL); request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1); } else { /* all ok, tell the user */ reply_schedule_callback(req, ttl, 0, reply); if (req->handle == req->ns->probe_request) req->ns->probe_request = NULL; /* Avoid double-free */ nameserver_up(req->ns); request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1); } } static int name_parse(u8 *packet, int length, int *idx, char *name_out, int name_out_len) { int name_end = -1; int j = *idx; int ptr_count = 0; #define GET32(x) do { if (j + 4 > length) goto err; memcpy(&t32_, packet + j, 4); j += 4; x = ntohl(t32_); } while (0) #define GET16(x) do { if (j + 2 > length) goto err; memcpy(&t_, packet + j, 2); j += 2; x = ntohs(t_); } while (0) #define GET8(x) do { if (j >= length) goto err; x = packet[j++]; } while (0) char *cp = name_out; const char *const end = name_out + name_out_len; /* Normally, names are a series of length prefixed strings terminated */ /* with a length of 0 (the lengths are u8's < 63). */ /* However, the length can start with a pair of 1 bits and that */ /* means that the next 14 bits are a pointer within the current */ /* packet. */ for (;;) { u8 label_len; GET8(label_len); if (!label_len) break; if (label_len & 0xc0) { u8 ptr_low; GET8(ptr_low); if (name_end < 0) name_end = j; j = (((int)label_len & 0x3f) << 8) + ptr_low; /* Make sure that the target offset is in-bounds. */ if (j < 0 || j >= length) return -1; /* If we've jumped more times than there are characters in the * message, we must have a loop. */ if (++ptr_count > length) return -1; continue; } if (label_len > 63) return -1; if (cp != name_out) { if (cp + 1 >= end) return -1; *cp++ = '.'; } if (cp + label_len >= end) return -1; if (j + label_len > length) return -1; memcpy(cp, packet + j, label_len); cp += label_len; j += label_len; } if (cp >= end) return -1; *cp = '\0'; if (name_end < 0) *idx = j; else *idx = name_end; return 0; err: return -1; } /* parses a raw request from a nameserver */ static int reply_parse(struct evdns_base *base, u8 *packet, int length) { int j = 0, k = 0; /* index into packet */ u16 t_; /* used by the macros */ u32 t32_; /* used by the macros */ char tmp_name[256], cmp_name[256]; /* used by the macros */ int name_matches = 0; u16 trans_id, questions, answers, authority, additional, datalength; u16 flags = 0; u32 ttl, ttl_r = 0xffffffff; struct reply reply; struct request *req = NULL; unsigned int i; ASSERT_LOCKED(base); GET16(trans_id); GET16(flags); GET16(questions); GET16(answers); GET16(authority); GET16(additional); (void) authority; /* suppress "unused variable" warnings. */ (void) additional; /* suppress "unused variable" warnings. */ req = request_find_from_trans_id(base, trans_id); if (!req) return -1; EVUTIL_ASSERT(req->base == base); memset(&reply, 0, sizeof(reply)); /* If it's not an answer, it doesn't correspond to any request. */ if (!(flags & _QR_MASK)) return -1; /* must be an answer */ if ((flags & (_RCODE_MASK|_TC_MASK)) && (flags & (_RCODE_MASK|_TC_MASK)) != DNS_ERR_NOTEXIST) { /* there was an error and it's not NXDOMAIN */ goto err; } /* if (!answers) return; */ /* must have an answer of some form */ /* This macro skips a name in the DNS reply. */ #define SKIP_NAME \ do { tmp_name[0] = '\0'; \ if (name_parse(packet, length, &j, tmp_name, \ sizeof(tmp_name))<0) \ goto err; \ } while (0) reply.type = req->request_type; /* skip over each question in the reply */ for (i = 0; i < questions; ++i) { /* the question looks like * */ tmp_name[0] = '\0'; cmp_name[0] = '\0'; k = j; if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name)) < 0) goto err; if (name_parse(req->request, req->request_len, &k, cmp_name, sizeof(cmp_name))<0) goto err; if (!base->global_randomize_case) { if (strcmp(tmp_name, cmp_name) == 0) name_matches = 1; } else { if (evutil_ascii_strcasecmp(tmp_name, cmp_name) == 0) name_matches = 1; } j += 4; if (j > length) goto err; } if (!name_matches) goto err; /* now we have the answer section which looks like * */ for (i = 0; i < answers; ++i) { u16 type, class; SKIP_NAME; GET16(type); GET16(class); GET32(ttl); GET16(datalength); if (type == TYPE_A && class == CLASS_INET) { int addrcount, addrtocopy; if (req->request_type != TYPE_A) { j += datalength; continue; } if ((datalength & 3) != 0) /* not an even number of As. */ goto err; addrcount = datalength >> 2; addrtocopy = MIN(MAX_V4_ADDRS - reply.data.a.addrcount, (unsigned)addrcount); ttl_r = MIN(ttl_r, ttl); /* we only bother with the first four addresses. */ if (j + 4*addrtocopy > length) goto err; memcpy(&reply.data.a.addresses[reply.data.a.addrcount], packet + j, 4*addrtocopy); j += 4*addrtocopy; reply.data.a.addrcount += addrtocopy; reply.have_answer = 1; if (reply.data.a.addrcount == MAX_V4_ADDRS) break; } else if (type == TYPE_PTR && class == CLASS_INET) { if (req->request_type != TYPE_PTR) { j += datalength; continue; } if (name_parse(packet, length, &j, reply.data.ptr.name, sizeof(reply.data.ptr.name))<0) goto err; ttl_r = MIN(ttl_r, ttl); reply.have_answer = 1; break; } else if (type == TYPE_CNAME) { char cname[HOST_NAME_MAX]; if (!req->put_cname_in_ptr || *req->put_cname_in_ptr) { j += datalength; continue; } if (name_parse(packet, length, &j, cname, sizeof(cname))<0) goto err; *req->put_cname_in_ptr = mm_strdup(cname); } else if (type == TYPE_AAAA && class == CLASS_INET) { int addrcount, addrtocopy; if (req->request_type != TYPE_AAAA) { j += datalength; continue; } if ((datalength & 15) != 0) /* not an even number of AAAAs. */ goto err; addrcount = datalength >> 4; /* each address is 16 bytes long */ addrtocopy = MIN(MAX_V6_ADDRS - reply.data.aaaa.addrcount, (unsigned)addrcount); ttl_r = MIN(ttl_r, ttl); /* we only bother with the first four addresses. */ if (j + 16*addrtocopy > length) goto err; memcpy(&reply.data.aaaa.addresses[reply.data.aaaa.addrcount], packet + j, 16*addrtocopy); reply.data.aaaa.addrcount += addrtocopy; j += 16*addrtocopy; reply.have_answer = 1; if (reply.data.aaaa.addrcount == MAX_V6_ADDRS) break; } else { /* skip over any other type of resource */ j += datalength; } } if (!reply.have_answer) { for (i = 0; i < authority; ++i) { u16 type, class; SKIP_NAME; GET16(type); GET16(class); GET32(ttl); GET16(datalength); if (type == TYPE_SOA && class == CLASS_INET) { u32 serial, refresh, retry, expire, minimum; SKIP_NAME; SKIP_NAME; GET32(serial); GET32(refresh); GET32(retry); GET32(expire); GET32(minimum); (void)expire; (void)retry; (void)refresh; (void)serial; ttl_r = MIN(ttl_r, ttl); ttl_r = MIN(ttl_r, minimum); } else { /* skip over any other type of resource */ j += datalength; } } } if (ttl_r == 0xffffffff) ttl_r = 0; reply_handle(req, flags, ttl_r, &reply); return 0; err: if (req) reply_handle(req, flags, 0, NULL); return -1; } /* Parse a raw request (packet,length) sent to a nameserver port (port) from */ /* a DNS client (addr,addrlen), and if it's well-formed, call the corresponding */ /* callback. */ static int request_parse(u8 *packet, int length, struct evdns_server_port *port, struct sockaddr *addr, ev_socklen_t addrlen) { int j = 0; /* index into packet */ u16 t_; /* used by the macros */ char tmp_name[256]; /* used by the macros */ int i; u16 trans_id, flags, questions, answers, authority, additional; struct server_request *server_req = NULL; ASSERT_LOCKED(port); /* Get the header fields */ GET16(trans_id); GET16(flags); GET16(questions); GET16(answers); GET16(authority); GET16(additional); (void)answers; (void)additional; (void)authority; if (flags & _QR_MASK) return -1; /* Must not be an answer. */ flags &= (_RD_MASK|_CD_MASK); /* Only RD and CD get preserved. */ server_req = mm_malloc(sizeof(struct server_request)); if (server_req == NULL) return -1; memset(server_req, 0, sizeof(struct server_request)); server_req->trans_id = trans_id; memcpy(&server_req->addr, addr, addrlen); server_req->addrlen = addrlen; server_req->base.flags = flags; server_req->base.nquestions = 0; server_req->base.questions = mm_calloc(sizeof(struct evdns_server_question *), questions); if (server_req->base.questions == NULL) goto err; for (i = 0; i < questions; ++i) { u16 type, class; struct evdns_server_question *q; int namelen; if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0) goto err; GET16(type); GET16(class); namelen = (int)strlen(tmp_name); q = mm_malloc(sizeof(struct evdns_server_question) + namelen); if (!q) goto err; q->type = type; q->dns_question_class = class; memcpy(q->name, tmp_name, namelen+1); server_req->base.questions[server_req->base.nquestions++] = q; } /* Ignore answers, authority, and additional. */ server_req->port = port; port->refcnt++; /* Only standard queries are supported. */ if (flags & _OP_MASK) { evdns_server_request_respond(&(server_req->base), DNS_ERR_NOTIMPL); return -1; } port->user_callback(&(server_req->base), port->user_data); return 0; err: if (server_req->base.questions) { for (i = 0; i < server_req->base.nquestions; ++i) mm_free(server_req->base.questions[i]); mm_free(server_req->base.questions); } mm_free(server_req); return -1; #undef SKIP_NAME #undef GET32 #undef GET16 #undef GET8 } void evdns_set_transaction_id_fn(ev_uint16_t (*fn)(void)) { } void evdns_set_random_bytes_fn(void (*fn)(char *, size_t)) { } /* Try to choose a strong transaction id which isn't already in flight */ static u16 transaction_id_pick(struct evdns_base *base) { ASSERT_LOCKED(base); for (;;) { u16 trans_id; evutil_secure_rng_get_bytes(&trans_id, sizeof(trans_id)); if (trans_id == 0xffff) continue; /* now check to see if that id is already inflight */ if (request_find_from_trans_id(base, trans_id) == NULL) return trans_id; } } /* choose a namesever to use. This function will try to ignore */ /* nameservers which we think are down and load balance across the rest */ /* by updating the server_head global each time. */ static struct nameserver * nameserver_pick(struct evdns_base *base) { struct nameserver *started_at = base->server_head, *picked; ASSERT_LOCKED(base); if (!base->server_head) return NULL; /* if we don't have any good nameservers then there's no */ /* point in trying to find one. */ if (!base->global_good_nameservers) { base->server_head = base->server_head->next; return base->server_head; } /* remember that nameservers are in a circular list */ for (;;) { if (base->server_head->state) { /* we think this server is currently good */ picked = base->server_head; base->server_head = base->server_head->next; return picked; } base->server_head = base->server_head->next; if (base->server_head == started_at) { /* all the nameservers seem to be down */ /* so we just return this one and hope for the */ /* best */ EVUTIL_ASSERT(base->global_good_nameservers == 0); picked = base->server_head; base->server_head = base->server_head->next; return picked; } } } /* this is called when a namesever socket is ready for reading */ static void nameserver_read(struct nameserver *ns) { struct sockaddr_storage ss; ev_socklen_t addrlen = sizeof(ss); u8 packet[1500]; char addrbuf[128]; ASSERT_LOCKED(ns->base); for (;;) { const int r = recvfrom(ns->socket, (void*)packet, sizeof(packet), 0, (struct sockaddr*)&ss, &addrlen); if (r < 0) { int err = evutil_socket_geterror(ns->socket); if (EVUTIL_ERR_RW_RETRIABLE(err)) return; nameserver_failed(ns, evutil_socket_error_to_string(err)); return; } if (evutil_sockaddr_cmp((struct sockaddr*)&ss, (struct sockaddr*)&ns->address, 0)) { log(EVDNS_LOG_WARN, "Address mismatch on received " "DNS packet. Apparent source was %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ss, addrbuf, sizeof(addrbuf))); return; } ns->timedout = 0; reply_parse(ns->base, packet, r); } } /* Read a packet from a DNS client on a server port s, parse it, and */ /* act accordingly. */ static void server_port_read(struct evdns_server_port *s) { u8 packet[1500]; struct sockaddr_storage addr; ev_socklen_t addrlen; int r; ASSERT_LOCKED(s); for (;;) { addrlen = sizeof(struct sockaddr_storage); r = recvfrom(s->socket, (void*)packet, sizeof(packet), 0, (struct sockaddr*) &addr, &addrlen); if (r < 0) { int err = evutil_socket_geterror(s->socket); if (EVUTIL_ERR_RW_RETRIABLE(err)) return; log(EVDNS_LOG_WARN, "Error %s (%d) while reading request.", evutil_socket_error_to_string(err), err); return; } request_parse(packet, r, s, (struct sockaddr*) &addr, addrlen); } } /* Try to write all pending replies on a given DNS server port. */ static void server_port_flush(struct evdns_server_port *port) { struct server_request *req = port->pending_replies; ASSERT_LOCKED(port); while (req) { int r = sendto(port->socket, req->response, (int)req->response_len, 0, (struct sockaddr*) &req->addr, (ev_socklen_t)req->addrlen); if (r < 0) { int err = evutil_socket_geterror(port->socket); if (EVUTIL_ERR_RW_RETRIABLE(err)) return; log(EVDNS_LOG_WARN, "Error %s (%d) while writing response to port; dropping", evutil_socket_error_to_string(err), err); } if (server_request_free(req)) { /* we released the last reference to req->port. */ return; } else { EVUTIL_ASSERT(req != port->pending_replies); req = port->pending_replies; } } /* We have no more pending requests; stop listening for 'writeable' events. */ (void) event_del(&port->event); event_assign(&port->event, port->event_base, port->socket, EV_READ | EV_PERSIST, server_port_ready_callback, port); if (event_add(&port->event, NULL) < 0) { log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server."); /* ???? Do more? */ } } /* set if we are waiting for the ability to write to this server. */ /* if waiting is true then we ask libevent for EV_WRITE events, otherwise */ /* we stop these events. */ static void nameserver_write_waiting(struct nameserver *ns, char waiting) { ASSERT_LOCKED(ns->base); if (ns->write_waiting == waiting) return; ns->write_waiting = waiting; (void) event_del(&ns->event); event_assign(&ns->event, ns->base->event_base, ns->socket, EV_READ | (waiting ? EV_WRITE : 0) | EV_PERSIST, nameserver_ready_callback, ns); if (event_add(&ns->event, NULL) < 0) { char addrbuf[128]; log(EVDNS_LOG_WARN, "Error from libevent when adding event for %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf))); /* ???? Do more? */ } } /* a callback function. Called by libevent when the kernel says that */ /* a nameserver socket is ready for writing or reading */ static void nameserver_ready_callback(evutil_socket_t fd, short events, void *arg) { struct nameserver *ns = (struct nameserver *) arg; (void)fd; EVDNS_LOCK(ns->base); if (events & EV_WRITE) { ns->choked = 0; if (!evdns_transmit(ns->base)) { nameserver_write_waiting(ns, 0); } } if (events & EV_READ) { nameserver_read(ns); } EVDNS_UNLOCK(ns->base); } /* a callback function. Called by libevent when the kernel says that */ /* a server socket is ready for writing or reading. */ static void server_port_ready_callback(evutil_socket_t fd, short events, void *arg) { struct evdns_server_port *port = (struct evdns_server_port *) arg; (void) fd; EVDNS_LOCK(port); if (events & EV_WRITE) { port->choked = 0; server_port_flush(port); } if (events & EV_READ) { server_port_read(port); } EVDNS_UNLOCK(port); } /* This is an inefficient representation; only use it via the dnslabel_table_* * functions, so that is can be safely replaced with something smarter later. */ #define MAX_LABELS 128 /* Structures used to implement name compression */ struct dnslabel_entry { char *v; off_t pos; }; struct dnslabel_table { int n_labels; /* number of current entries */ /* map from name to position in message */ struct dnslabel_entry labels[MAX_LABELS]; }; /* Initialize dnslabel_table. */ static void dnslabel_table_init(struct dnslabel_table *table) { table->n_labels = 0; } /* Free all storage held by table, but not the table itself. */ static void dnslabel_clear(struct dnslabel_table *table) { int i; for (i = 0; i < table->n_labels; ++i) mm_free(table->labels[i].v); table->n_labels = 0; } /* return the position of the label in the current message, or -1 if the label */ /* hasn't been used yet. */ static int dnslabel_table_get_pos(const struct dnslabel_table *table, const char *label) { int i; for (i = 0; i < table->n_labels; ++i) { if (!strcmp(label, table->labels[i].v)) return table->labels[i].pos; } return -1; } /* remember that we've used the label at position pos */ static int dnslabel_table_add(struct dnslabel_table *table, const char *label, off_t pos) { char *v; int p; if (table->n_labels == MAX_LABELS) return (-1); v = mm_strdup(label); if (v == NULL) return (-1); p = table->n_labels++; table->labels[p].v = v; table->labels[p].pos = pos; return (0); } /* Converts a string to a length-prefixed set of DNS labels, starting */ /* at buf[j]. name and buf must not overlap. name_len should be the length */ /* of name. table is optional, and is used for compression. */ /* */ /* Input: abc.def */ /* Output: <3>abc<3>def<0> */ /* */ /* Returns the first index after the encoded name, or negative on error. */ /* -1 label was > 63 bytes */ /* -2 name too long to fit in buffer. */ /* */ static off_t dnsname_to_labels(u8 *const buf, size_t buf_len, off_t j, const char *name, const size_t name_len, struct dnslabel_table *table) { const char *end = name + name_len; int ref = 0; u16 t_; #define APPEND16(x) do { \ if (j + 2 > (off_t)buf_len) \ goto overflow; \ t_ = htons(x); \ memcpy(buf + j, &t_, 2); \ j += 2; \ } while (0) #define APPEND32(x) do { \ if (j + 4 > (off_t)buf_len) \ goto overflow; \ t32_ = htonl(x); \ memcpy(buf + j, &t32_, 4); \ j += 4; \ } while (0) if (name_len > 255) return -2; for (;;) { const char *const start = name; if (table && (ref = dnslabel_table_get_pos(table, name)) >= 0) { APPEND16(ref | 0xc000); return j; } name = strchr(name, '.'); if (!name) { const size_t label_len = end - start; if (label_len > 63) return -1; if ((size_t)(j+label_len+1) > buf_len) return -2; if (table) dnslabel_table_add(table, start, j); buf[j++] = (ev_uint8_t)label_len; memcpy(buf + j, start, label_len); j += (int) label_len; break; } else { /* append length of the label. */ const size_t label_len = name - start; if (label_len > 63) return -1; if ((size_t)(j+label_len+1) > buf_len) return -2; if (table) dnslabel_table_add(table, start, j); buf[j++] = (ev_uint8_t)label_len; memcpy(buf + j, start, label_len); j += (int) label_len; /* hop over the '.' */ name++; } } /* the labels must be terminated by a 0. */ /* It's possible that the name ended in a . */ /* in which case the zero is already there */ if (!j || buf[j-1]) buf[j++] = 0; return j; overflow: return (-2); } /* Finds the length of a dns request for a DNS name of the given */ /* length. The actual request may be smaller than the value returned */ /* here */ static size_t evdns_request_len(const size_t name_len) { return 96 + /* length of the DNS standard header */ name_len + 2 + 4; /* space for the resource type */ } /* build a dns request packet into buf. buf should be at least as long */ /* as evdns_request_len told you it should be. */ /* */ /* Returns the amount of space used. Negative on error. */ static int evdns_request_data_build(const char *const name, const size_t name_len, const u16 trans_id, const u16 type, const u16 class, u8 *const buf, size_t buf_len) { off_t j = 0; /* current offset into buf */ u16 t_; /* used by the macros */ APPEND16(trans_id); APPEND16(0x0100); /* standard query, recusion needed */ APPEND16(1); /* one question */ APPEND16(0); /* no answers */ APPEND16(0); /* no authority */ APPEND16(0); /* no additional */ j = dnsname_to_labels(buf, buf_len, j, name, name_len, NULL); if (j < 0) { return (int)j; } APPEND16(type); APPEND16(class); return (int)j; overflow: return (-1); } /* exported function */ struct evdns_server_port * evdns_add_server_port_with_base(struct event_base *base, evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data) { struct evdns_server_port *port; if (flags) return NULL; /* flags not yet implemented */ if (!(port = mm_malloc(sizeof(struct evdns_server_port)))) return NULL; memset(port, 0, sizeof(struct evdns_server_port)); port->socket = socket; port->refcnt = 1; port->choked = 0; port->closing = 0; port->user_callback = cb; port->user_data = user_data; port->pending_replies = NULL; port->event_base = base; event_assign(&port->event, port->event_base, port->socket, EV_READ | EV_PERSIST, server_port_ready_callback, port); if (event_add(&port->event, NULL) < 0) { mm_free(port); return NULL; } EVTHREAD_ALLOC_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE); return port; } struct evdns_server_port * evdns_add_server_port(evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data) { return evdns_add_server_port_with_base(NULL, socket, flags, cb, user_data); } /* exported function */ void evdns_close_server_port(struct evdns_server_port *port) { EVDNS_LOCK(port); if (--port->refcnt == 0) { EVDNS_UNLOCK(port); server_port_free(port); } else { port->closing = 1; EVDNS_UNLOCK(port); } } /* exported function */ int evdns_server_request_add_reply(struct evdns_server_request *req_, int section, const char *name, int type, int class, int ttl, int datalen, int is_name, const char *data) { struct server_request *req = TO_SERVER_REQUEST(req_); struct server_reply_item **itemp, *item; int *countp; int result = -1; EVDNS_LOCK(req->port); if (req->response) /* have we already answered? */ goto done; switch (section) { case EVDNS_ANSWER_SECTION: itemp = &req->answer; countp = &req->n_answer; break; case EVDNS_AUTHORITY_SECTION: itemp = &req->authority; countp = &req->n_authority; break; case EVDNS_ADDITIONAL_SECTION: itemp = &req->additional; countp = &req->n_additional; break; default: goto done; } while (*itemp) { itemp = &((*itemp)->next); } item = mm_malloc(sizeof(struct server_reply_item)); if (!item) goto done; item->next = NULL; if (!(item->name = mm_strdup(name))) { mm_free(item); goto done; } item->type = type; item->dns_question_class = class; item->ttl = ttl; item->is_name = is_name != 0; item->datalen = 0; item->data = NULL; if (data) { if (item->is_name) { if (!(item->data = mm_strdup(data))) { mm_free(item->name); mm_free(item); goto done; } item->datalen = (u16)-1; } else { if (!(item->data = mm_malloc(datalen))) { mm_free(item->name); mm_free(item); goto done; } item->datalen = datalen; memcpy(item->data, data, datalen); } } *itemp = item; ++(*countp); result = 0; done: EVDNS_UNLOCK(req->port); return result; } /* exported function */ int evdns_server_request_add_a_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl) { return evdns_server_request_add_reply( req, EVDNS_ANSWER_SECTION, name, TYPE_A, CLASS_INET, ttl, n*4, 0, addrs); } /* exported function */ int evdns_server_request_add_aaaa_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl) { return evdns_server_request_add_reply( req, EVDNS_ANSWER_SECTION, name, TYPE_AAAA, CLASS_INET, ttl, n*16, 0, addrs); } /* exported function */ int evdns_server_request_add_ptr_reply(struct evdns_server_request *req, struct in_addr *in, const char *inaddr_name, const char *hostname, int ttl) { u32 a; char buf[32]; if (in && inaddr_name) return -1; else if (!in && !inaddr_name) return -1; if (in) { a = ntohl(in->s_addr); evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa", (int)(u8)((a )&0xff), (int)(u8)((a>>8 )&0xff), (int)(u8)((a>>16)&0xff), (int)(u8)((a>>24)&0xff)); inaddr_name = buf; } return evdns_server_request_add_reply( req, EVDNS_ANSWER_SECTION, inaddr_name, TYPE_PTR, CLASS_INET, ttl, -1, 1, hostname); } /* exported function */ int evdns_server_request_add_cname_reply(struct evdns_server_request *req, const char *name, const char *cname, int ttl) { return evdns_server_request_add_reply( req, EVDNS_ANSWER_SECTION, name, TYPE_CNAME, CLASS_INET, ttl, -1, 1, cname); } /* exported function */ void evdns_server_request_set_flags(struct evdns_server_request *exreq, int flags) { struct server_request *req = TO_SERVER_REQUEST(exreq); req->base.flags &= ~(EVDNS_FLAGS_AA|EVDNS_FLAGS_RD); req->base.flags |= flags; } static int evdns_server_request_format_response(struct server_request *req, int err) { unsigned char buf[1500]; size_t buf_len = sizeof(buf); off_t j = 0, r; u16 t_; u32 t32_; int i; u16 flags; struct dnslabel_table table; if (err < 0 || err > 15) return -1; /* Set response bit and error code; copy OPCODE and RD fields from * question; copy RA and AA if set by caller. */ flags = req->base.flags; flags |= (_QR_MASK | err); dnslabel_table_init(&table); APPEND16(req->trans_id); APPEND16(flags); APPEND16(req->base.nquestions); APPEND16(req->n_answer); APPEND16(req->n_authority); APPEND16(req->n_additional); /* Add questions. */ for (i=0; i < req->base.nquestions; ++i) { const char *s = req->base.questions[i]->name; j = dnsname_to_labels(buf, buf_len, j, s, strlen(s), &table); if (j < 0) { dnslabel_clear(&table); return (int) j; } APPEND16(req->base.questions[i]->type); APPEND16(req->base.questions[i]->dns_question_class); } /* Add answer, authority, and additional sections. */ for (i=0; i<3; ++i) { struct server_reply_item *item; if (i==0) item = req->answer; else if (i==1) item = req->authority; else item = req->additional; while (item) { r = dnsname_to_labels(buf, buf_len, j, item->name, strlen(item->name), &table); if (r < 0) goto overflow; j = r; APPEND16(item->type); APPEND16(item->dns_question_class); APPEND32(item->ttl); if (item->is_name) { off_t len_idx = j, name_start; j += 2; name_start = j; r = dnsname_to_labels(buf, buf_len, j, item->data, strlen(item->data), &table); if (r < 0) goto overflow; j = r; t_ = htons( (short) (j-name_start) ); memcpy(buf+len_idx, &t_, 2); } else { APPEND16(item->datalen); if (j+item->datalen > (off_t)buf_len) goto overflow; memcpy(buf+j, item->data, item->datalen); j += item->datalen; } item = item->next; } } if (j > 512) { overflow: j = 512; buf[2] |= 0x02; /* set the truncated bit. */ } req->response_len = j; if (!(req->response = mm_malloc(req->response_len))) { server_request_free_answers(req); dnslabel_clear(&table); return (-1); } memcpy(req->response, buf, req->response_len); server_request_free_answers(req); dnslabel_clear(&table); return (0); } /* exported function */ int evdns_server_request_respond(struct evdns_server_request *req_, int err) { struct server_request *req = TO_SERVER_REQUEST(req_); struct evdns_server_port *port = req->port; int r = -1; EVDNS_LOCK(port); if (!req->response) { if ((r = evdns_server_request_format_response(req, err))<0) goto done; } r = sendto(port->socket, req->response, (int)req->response_len, 0, (struct sockaddr*) &req->addr, (ev_socklen_t)req->addrlen); if (r<0) { int sock_err = evutil_socket_geterror(port->socket); if (EVUTIL_ERR_RW_RETRIABLE(sock_err)) goto done; if (port->pending_replies) { req->prev_pending = port->pending_replies->prev_pending; req->next_pending = port->pending_replies; req->prev_pending->next_pending = req->next_pending->prev_pending = req; } else { req->prev_pending = req->next_pending = req; port->pending_replies = req; port->choked = 1; (void) event_del(&port->event); event_assign(&port->event, port->event_base, port->socket, (port->closing?0:EV_READ) | EV_WRITE | EV_PERSIST, server_port_ready_callback, port); if (event_add(&port->event, NULL) < 0) { log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server"); } } r = 1; goto done; } if (server_request_free(req)) { r = 0; goto done; } if (port->pending_replies) server_port_flush(port); r = 0; done: EVDNS_UNLOCK(port); return r; } /* Free all storage held by RRs in req. */ static void server_request_free_answers(struct server_request *req) { struct server_reply_item *victim, *next, **list; int i; for (i = 0; i < 3; ++i) { if (i==0) list = &req->answer; else if (i==1) list = &req->authority; else list = &req->additional; victim = *list; while (victim) { next = victim->next; mm_free(victim->name); if (victim->data) mm_free(victim->data); mm_free(victim); victim = next; } *list = NULL; } } /* Free all storage held by req, and remove links to it. */ /* return true iff we just wound up freeing the server_port. */ static int server_request_free(struct server_request *req) { int i, rc=1, lock=0; if (req->base.questions) { for (i = 0; i < req->base.nquestions; ++i) mm_free(req->base.questions[i]); mm_free(req->base.questions); } if (req->port) { EVDNS_LOCK(req->port); lock=1; if (req->port->pending_replies == req) { if (req->next_pending && req->next_pending != req) req->port->pending_replies = req->next_pending; else req->port->pending_replies = NULL; } rc = --req->port->refcnt; } if (req->response) { mm_free(req->response); } server_request_free_answers(req); if (req->next_pending && req->next_pending != req) { req->next_pending->prev_pending = req->prev_pending; req->prev_pending->next_pending = req->next_pending; } if (rc == 0) { EVDNS_UNLOCK(req->port); /* ????? nickm */ server_port_free(req->port); mm_free(req); return (1); } if (lock) EVDNS_UNLOCK(req->port); mm_free(req); return (0); } /* Free all storage held by an evdns_server_port. Only called when */ static void server_port_free(struct evdns_server_port *port) { EVUTIL_ASSERT(port); EVUTIL_ASSERT(!port->refcnt); EVUTIL_ASSERT(!port->pending_replies); if (port->socket > 0) { evutil_closesocket(port->socket); port->socket = -1; } (void) event_del(&port->event); event_debug_unassign(&port->event); EVTHREAD_FREE_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE); mm_free(port); } /* exported function */ int evdns_server_request_drop(struct evdns_server_request *req_) { struct server_request *req = TO_SERVER_REQUEST(req_); server_request_free(req); return 0; } /* exported function */ int evdns_server_request_get_requesting_addr(struct evdns_server_request *req_, struct sockaddr *sa, int addr_len) { struct server_request *req = TO_SERVER_REQUEST(req_); if (addr_len < (int)req->addrlen) return -1; memcpy(sa, &(req->addr), req->addrlen); return req->addrlen; } #undef APPEND16 #undef APPEND32 /* this is a libevent callback function which is called when a request */ /* has timed out. */ static void evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg) { struct request *const req = (struct request *) arg; struct evdns_base *base = req->base; (void) fd; (void) events; log(EVDNS_LOG_DEBUG, "Request %p timed out", arg); EVDNS_LOCK(base); if (req->tx_count >= req->base->global_max_retransmits) { struct nameserver *ns = req->ns; /* this request has failed */ log(EVDNS_LOG_DEBUG, "Giving up on request %p; tx_count==%d", arg, req->tx_count); reply_schedule_callback(req, 0, DNS_ERR_TIMEOUT, NULL); request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1); nameserver_failed(ns, "request timed out."); } else { /* retransmit it */ log(EVDNS_LOG_DEBUG, "Retransmitting request %p; tx_count==%d", arg, req->tx_count); (void) evtimer_del(&req->timeout_event); request_swap_ns(req, nameserver_pick(base)); evdns_request_transmit(req); req->ns->timedout++; if (req->ns->timedout > req->base->global_max_nameserver_timeout) { req->ns->timedout = 0; nameserver_failed(req->ns, "request timed out."); } } EVDNS_UNLOCK(base); } /* try to send a request to a given server. */ /* */ /* return: */ /* 0 ok */ /* 1 temporary failure */ /* 2 other failure */ static int evdns_request_transmit_to(struct request *req, struct nameserver *server) { int r; ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); if (server->requests_inflight == 1 && req->base->disable_when_inactive && event_add(&server->event, NULL) < 0) { return 1; } r = sendto(server->socket, (void*)req->request, req->request_len, 0, (struct sockaddr *)&server->address, server->addrlen); if (r < 0) { int err = evutil_socket_geterror(server->socket); if (EVUTIL_ERR_RW_RETRIABLE(err)) return 1; nameserver_failed(req->ns, evutil_socket_error_to_string(err)); return 2; } else if (r != (int)req->request_len) { return 1; /* short write */ } else { return 0; } } /* try to send a request, updating the fields of the request */ /* as needed */ /* */ /* return: */ /* 0 ok */ /* 1 failed */ static int evdns_request_transmit(struct request *req) { int retcode = 0, r; ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); /* if we fail to send this packet then this flag marks it */ /* for evdns_transmit */ req->transmit_me = 1; EVUTIL_ASSERT(req->trans_id != 0xffff); if (!req->ns) { /* unable to transmit request if no nameservers */ return 1; } if (req->ns->choked) { /* don't bother trying to write to a socket */ /* which we have had EAGAIN from */ return 1; } r = evdns_request_transmit_to(req, req->ns); switch (r) { case 1: /* temp failure */ req->ns->choked = 1; nameserver_write_waiting(req->ns, 1); return 1; case 2: /* failed to transmit the request entirely. we can fallthrough since * we'll set a timeout, which will time out, and make us retransmit the * request anyway. */ retcode = 1; EVUTIL_FALLTHROUGH; default: /* all ok */ log(EVDNS_LOG_DEBUG, "Setting timeout for request %p, sent to nameserver %p", req, req->ns); if (evtimer_add(&req->timeout_event, &req->base->global_timeout) < 0) { log(EVDNS_LOG_WARN, "Error from libevent when adding timer for request %p", req); /* ???? Do more? */ } req->tx_count++; req->transmit_me = 0; return retcode; } } static void nameserver_probe_callback(int result, char type, int count, int ttl, void *addresses, void *arg) { struct nameserver *const ns = (struct nameserver *) arg; (void) type; (void) count; (void) ttl; (void) addresses; if (result == DNS_ERR_CANCEL) { /* We canceled this request because the nameserver came up * for some other reason. Do not change our opinion about * the nameserver. */ return; } EVDNS_LOCK(ns->base); ns->probe_request = NULL; if (result == DNS_ERR_NONE || result == DNS_ERR_NOTEXIST) { /* this is a good reply */ nameserver_up(ns); } else { nameserver_probe_failed(ns); } EVDNS_UNLOCK(ns->base); } static void nameserver_send_probe(struct nameserver *const ns) { struct evdns_request *handle; struct request *req; char addrbuf[128]; /* here we need to send a probe to a given nameserver */ /* in the hope that it is up now. */ ASSERT_LOCKED(ns->base); log(EVDNS_LOG_DEBUG, "Sending probe to %s", evutil_format_sockaddr_port_( (struct sockaddr *)&ns->address, addrbuf, sizeof(addrbuf))); handle = mm_calloc(1, sizeof(*handle)); if (!handle) return; req = request_new(ns->base, handle, TYPE_A, "google.com", DNS_QUERY_NO_SEARCH, nameserver_probe_callback, ns); if (!req) { mm_free(handle); return; } ns->probe_request = handle; /* we force this into the inflight queue no matter what */ request_trans_id_set(req, transaction_id_pick(ns->base)); req->ns = ns; request_submit(req); } /* returns: */ /* 0 didn't try to transmit anything */ /* 1 tried to transmit something */ static int evdns_transmit(struct evdns_base *base) { char did_try_to_transmit = 0; int i; ASSERT_LOCKED(base); for (i = 0; i < base->n_req_heads; ++i) { if (base->req_heads[i]) { struct request *const started_at = base->req_heads[i], *req = started_at; /* first transmit all the requests which are currently waiting */ do { if (req->transmit_me) { did_try_to_transmit = 1; evdns_request_transmit(req); } req = req->next; } while (req != started_at); } } return did_try_to_transmit; } /* exported function */ int evdns_base_count_nameservers(struct evdns_base *base) { const struct nameserver *server; int n = 0; EVDNS_LOCK(base); server = base->server_head; if (!server) goto done; do { ++n; server = server->next; } while (server != base->server_head); done: EVDNS_UNLOCK(base); return n; } int evdns_count_nameservers(void) { return evdns_base_count_nameservers(current_base); } /* exported function */ int evdns_base_clear_nameservers_and_suspend(struct evdns_base *base) { struct nameserver *server, *started_at; int i; EVDNS_LOCK(base); server = base->server_head; started_at = base->server_head; if (!server) { EVDNS_UNLOCK(base); return 0; } while (1) { struct nameserver *next = server->next; (void) event_del(&server->event); if (evtimer_initialized(&server->timeout_event)) (void) evtimer_del(&server->timeout_event); if (server->probe_request) { evdns_cancel_request(server->base, server->probe_request); server->probe_request = NULL; } if (server->socket >= 0) evutil_closesocket(server->socket); mm_free(server); if (next == started_at) break; server = next; } base->server_head = NULL; base->global_good_nameservers = 0; for (i = 0; i < base->n_req_heads; ++i) { struct request *req, *req_started_at; req = req_started_at = base->req_heads[i]; while (req) { struct request *next = req->next; req->tx_count = req->reissue_count = 0; req->ns = NULL; /* ???? What to do about searches? */ (void) evtimer_del(&req->timeout_event); req->trans_id = 0; req->transmit_me = 0; base->global_requests_waiting++; evdns_request_insert(req, &base->req_waiting_head); /* We want to insert these suspended elements at the front of * the waiting queue, since they were pending before any of * the waiting entries were added. This is a circular list, * so we can just shift the start back by one.*/ base->req_waiting_head = base->req_waiting_head->prev; if (next == req_started_at) break; req = next; } base->req_heads[i] = NULL; } base->global_requests_inflight = 0; EVDNS_UNLOCK(base); return 0; } int evdns_clear_nameservers_and_suspend(void) { return evdns_base_clear_nameservers_and_suspend(current_base); } /* exported function */ int evdns_base_resume(struct evdns_base *base) { EVDNS_LOCK(base); evdns_requests_pump_waiting_queue(base); EVDNS_UNLOCK(base); return 0; } int evdns_resume(void) { return evdns_base_resume(current_base); } static int evdns_nameserver_add_impl_(struct evdns_base *base, const struct sockaddr *address, int addrlen) { /* first check to see if we already have this nameserver */ const struct nameserver *server = base->server_head, *const started_at = base->server_head; struct nameserver *ns; int err = 0; char addrbuf[128]; ASSERT_LOCKED(base); if (server) { do { if (!evutil_sockaddr_cmp((struct sockaddr*)&server->address, address, 1)) return 3; server = server->next; } while (server != started_at); } if (addrlen > (int)sizeof(ns->address)) { log(EVDNS_LOG_DEBUG, "Addrlen %d too long.", (int)addrlen); return 2; } ns = (struct nameserver *) mm_malloc(sizeof(struct nameserver)); if (!ns) return -1; memset(ns, 0, sizeof(struct nameserver)); ns->base = base; evtimer_assign(&ns->timeout_event, ns->base->event_base, nameserver_prod_callback, ns); ns->socket = evutil_socket_(address->sa_family, SOCK_DGRAM|EVUTIL_SOCK_NONBLOCK|EVUTIL_SOCK_CLOEXEC, 0); if (ns->socket < 0) { err = 1; goto out1; } if (base->global_outgoing_addrlen && !evutil_sockaddr_is_loopback_(address)) { if (bind(ns->socket, (struct sockaddr*)&base->global_outgoing_address, base->global_outgoing_addrlen) < 0) { log(EVDNS_LOG_WARN,"Couldn't bind to outgoing address"); err = 2; goto out2; } } if (base->so_rcvbuf) { if (setsockopt(ns->socket, SOL_SOCKET, SO_RCVBUF, (void *)&base->so_rcvbuf, sizeof(base->so_rcvbuf))) { log(EVDNS_LOG_WARN, "Couldn't set SO_RCVBUF to %i", base->so_rcvbuf); err = -SO_RCVBUF; goto out2; } } if (base->so_sndbuf) { if (setsockopt(ns->socket, SOL_SOCKET, SO_SNDBUF, (void *)&base->so_sndbuf, sizeof(base->so_sndbuf))) { log(EVDNS_LOG_WARN, "Couldn't set SO_SNDBUF to %i", base->so_sndbuf); err = -SO_SNDBUF; goto out2; } } memcpy(&ns->address, address, addrlen); ns->addrlen = addrlen; ns->state = 1; event_assign(&ns->event, ns->base->event_base, ns->socket, EV_READ | EV_PERSIST, nameserver_ready_callback, ns); if (!base->disable_when_inactive && event_add(&ns->event, NULL) < 0) { err = 2; goto out2; } log(EVDNS_LOG_DEBUG, "Added nameserver %s as %p", evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), ns); /* insert this nameserver into the list of them */ if (!base->server_head) { ns->next = ns->prev = ns; base->server_head = ns; } else { ns->next = base->server_head->next; ns->prev = base->server_head; base->server_head->next = ns; ns->next->prev = ns; } base->global_good_nameservers++; return 0; out2: evutil_closesocket(ns->socket); out1: event_debug_unassign(&ns->event); mm_free(ns); log(EVDNS_LOG_WARN, "Unable to add nameserver %s: error %d", evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), err); return err; } /* exported function */ int evdns_base_nameserver_add(struct evdns_base *base, unsigned long int address) { struct sockaddr_in sin; int res; memset(&sin, 0, sizeof(sin)); sin.sin_addr.s_addr = address; sin.sin_port = htons(53); sin.sin_family = AF_INET; EVDNS_LOCK(base); res = evdns_nameserver_add_impl_(base, (struct sockaddr*)&sin, sizeof(sin)); EVDNS_UNLOCK(base); return res; } int evdns_nameserver_add(unsigned long int address) { if (!current_base) current_base = evdns_base_new(NULL, 0); return evdns_base_nameserver_add(current_base, address); } static void sockaddr_setport(struct sockaddr *sa, ev_uint16_t port) { if (sa->sa_family == AF_INET) { ((struct sockaddr_in *)sa)->sin_port = htons(port); } else if (sa->sa_family == AF_INET6) { ((struct sockaddr_in6 *)sa)->sin6_port = htons(port); } } static ev_uint16_t sockaddr_getport(struct sockaddr *sa) { if (sa->sa_family == AF_INET) { return ntohs(((struct sockaddr_in *)sa)->sin_port); } else if (sa->sa_family == AF_INET6) { return ntohs(((struct sockaddr_in6 *)sa)->sin6_port); } else { return 0; } } /* exported function */ int evdns_base_nameserver_ip_add(struct evdns_base *base, const char *ip_as_string) { struct sockaddr_storage ss; struct sockaddr *sa; int len = sizeof(ss); int res; if (evutil_parse_sockaddr_port(ip_as_string, (struct sockaddr *)&ss, &len)) { log(EVDNS_LOG_WARN, "Unable to parse nameserver address %s", ip_as_string); return 4; } sa = (struct sockaddr *) &ss; if (sockaddr_getport(sa) == 0) sockaddr_setport(sa, 53); EVDNS_LOCK(base); res = evdns_nameserver_add_impl_(base, sa, len); EVDNS_UNLOCK(base); return res; } int evdns_nameserver_ip_add(const char *ip_as_string) { if (!current_base) current_base = evdns_base_new(NULL, 0); return evdns_base_nameserver_ip_add(current_base, ip_as_string); } int evdns_base_nameserver_sockaddr_add(struct evdns_base *base, const struct sockaddr *sa, ev_socklen_t len, unsigned flags) { int res; EVUTIL_ASSERT(base); EVDNS_LOCK(base); res = evdns_nameserver_add_impl_(base, sa, len); EVDNS_UNLOCK(base); return res; } int evdns_base_get_nameserver_addr(struct evdns_base *base, int idx, struct sockaddr *sa, ev_socklen_t len) { int result = -1; int i; struct nameserver *server; EVDNS_LOCK(base); server = base->server_head; for (i = 0; i < idx && server; ++i, server = server->next) { if (server->next == base->server_head) goto done; } if (! server) goto done; if (server->addrlen > len) { result = (int) server->addrlen; goto done; } memcpy(sa, &server->address, server->addrlen); result = (int) server->addrlen; done: EVDNS_UNLOCK(base); return result; } /* remove from the queue */ static void evdns_request_remove(struct request *req, struct request **head) { ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); #if 0 { struct request *ptr; int found = 0; EVUTIL_ASSERT(*head != NULL); ptr = *head; do { if (ptr == req) { found = 1; break; } ptr = ptr->next; } while (ptr != *head); EVUTIL_ASSERT(found); EVUTIL_ASSERT(req->next); } #endif if (req->next == req) { /* only item in the list */ *head = NULL; } else { req->next->prev = req->prev; req->prev->next = req->next; if (*head == req) *head = req->next; } req->next = req->prev = NULL; } /* insert into the tail of the queue */ static void evdns_request_insert(struct request *req, struct request **head) { ASSERT_LOCKED(req->base); ASSERT_VALID_REQUEST(req); if (!*head) { *head = req; req->next = req->prev = req; return; } req->prev = (*head)->prev; req->prev->next = req; req->next = *head; (*head)->prev = req; } static int string_num_dots(const char *s) { int count = 0; while ((s = strchr(s, '.'))) { s++; count++; } return count; } static struct request * request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *name, int flags, evdns_callback_type callback, void *user_ptr) { const char issuing_now = (base->global_requests_inflight < base->global_max_requests_inflight) ? 1 : 0; const size_t name_len = strlen(name); const size_t request_max_len = evdns_request_len(name_len); const u16 trans_id = issuing_now ? transaction_id_pick(base) : 0xffff; /* the request data is alloced in a single block with the header */ struct request *const req = mm_malloc(sizeof(struct request) + request_max_len); int rlen; char namebuf[256]; (void) flags; ASSERT_LOCKED(base); if (!req) return NULL; if (name_len >= sizeof(namebuf)) { mm_free(req); return NULL; } memset(req, 0, sizeof(struct request)); req->base = base; evtimer_assign(&req->timeout_event, req->base->event_base, evdns_request_timeout_callback, req); if (base->global_randomize_case) { unsigned i; char randbits[(sizeof(namebuf)+7)/8]; strlcpy(namebuf, name, sizeof(namebuf)); evutil_secure_rng_get_bytes(randbits, (name_len+7)/8); for (i = 0; i < name_len; ++i) { if (EVUTIL_ISALPHA_(namebuf[i])) { if ((randbits[i >> 3] & (1<<(i & 7)))) namebuf[i] |= 0x20; else namebuf[i] &= ~0x20; } } name = namebuf; } /* request data lives just after the header */ req->request = ((u8 *) req) + sizeof(struct request); /* denotes that the request data shouldn't be free()ed */ req->request_appended = 1; rlen = evdns_request_data_build(name, name_len, trans_id, type, CLASS_INET, req->request, request_max_len); if (rlen < 0) goto err1; req->request_len = rlen; req->trans_id = trans_id; req->tx_count = 0; req->request_type = type; req->user_pointer = user_ptr; req->user_callback = callback; req->ns = issuing_now ? nameserver_pick(base) : NULL; req->next = req->prev = NULL; req->handle = handle; if (handle) { handle->current_req = req; handle->base = base; } return req; err1: mm_free(req); return NULL; } static void request_submit(struct request *const req) { struct evdns_base *base = req->base; ASSERT_LOCKED(base); ASSERT_VALID_REQUEST(req); if (req->ns) { /* if it has a nameserver assigned then this is going */ /* straight into the inflight queue */ evdns_request_insert(req, &REQ_HEAD(base, req->trans_id)); base->global_requests_inflight++; req->ns->requests_inflight++; evdns_request_transmit(req); } else { evdns_request_insert(req, &base->req_waiting_head); base->global_requests_waiting++; } } /* exported function */ void evdns_cancel_request(struct evdns_base *base, struct evdns_request *handle) { struct request *req; if (!handle->current_req) return; if (!base) { /* This redundancy is silly; can we fix it? (Not for 2.0) XXXX */ base = handle->base; if (!base) base = handle->current_req->base; } EVDNS_LOCK(base); if (handle->pending_cb) { EVDNS_UNLOCK(base); return; } req = handle->current_req; ASSERT_VALID_REQUEST(req); reply_schedule_callback(req, 0, DNS_ERR_CANCEL, NULL); if (req->ns) { /* remove from inflight queue */ request_finished(req, &REQ_HEAD(base, req->trans_id), 1); } else { /* remove from global_waiting head */ request_finished(req, &base->req_waiting_head, 1); } EVDNS_UNLOCK(base); } /* exported function */ struct evdns_request * evdns_base_resolve_ipv4(struct evdns_base *base, const char *name, int flags, evdns_callback_type callback, void *ptr) { struct evdns_request *handle; struct request *req; log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name); handle = mm_calloc(1, sizeof(*handle)); if (handle == NULL) return NULL; EVDNS_LOCK(base); if (flags & DNS_QUERY_NO_SEARCH) { req = request_new(base, handle, TYPE_A, name, flags, callback, ptr); if (req) request_submit(req); } else { search_request_new(base, handle, TYPE_A, name, flags, callback, ptr); } if (handle->current_req == NULL) { mm_free(handle); handle = NULL; } EVDNS_UNLOCK(base); return handle; } int evdns_resolve_ipv4(const char *name, int flags, evdns_callback_type callback, void *ptr) { return evdns_base_resolve_ipv4(current_base, name, flags, callback, ptr) ? 0 : -1; } /* exported function */ struct evdns_request * evdns_base_resolve_ipv6(struct evdns_base *base, const char *name, int flags, evdns_callback_type callback, void *ptr) { struct evdns_request *handle; struct request *req; log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name); handle = mm_calloc(1, sizeof(*handle)); if (handle == NULL) return NULL; EVDNS_LOCK(base); if (flags & DNS_QUERY_NO_SEARCH) { req = request_new(base, handle, TYPE_AAAA, name, flags, callback, ptr); if (req) request_submit(req); } else { search_request_new(base, handle, TYPE_AAAA, name, flags, callback, ptr); } if (handle->current_req == NULL) { mm_free(handle); handle = NULL; } EVDNS_UNLOCK(base); return handle; } int evdns_resolve_ipv6(const char *name, int flags, evdns_callback_type callback, void *ptr) { return evdns_base_resolve_ipv6(current_base, name, flags, callback, ptr) ? 0 : -1; } struct evdns_request * evdns_base_resolve_reverse(struct evdns_base *base, const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) { char buf[32]; struct evdns_request *handle; struct request *req; u32 a; EVUTIL_ASSERT(in); a = ntohl(in->s_addr); evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa", (int)(u8)((a )&0xff), (int)(u8)((a>>8 )&0xff), (int)(u8)((a>>16)&0xff), (int)(u8)((a>>24)&0xff)); handle = mm_calloc(1, sizeof(*handle)); if (handle == NULL) return NULL; log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf); EVDNS_LOCK(base); req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr); if (req) request_submit(req); if (handle->current_req == NULL) { mm_free(handle); handle = NULL; } EVDNS_UNLOCK(base); return (handle); } int evdns_resolve_reverse(const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) { return evdns_base_resolve_reverse(current_base, in, flags, callback, ptr) ? 0 : -1; } struct evdns_request * evdns_base_resolve_reverse_ipv6(struct evdns_base *base, const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) { /* 32 nybbles, 32 periods, "ip6.arpa", NUL. */ char buf[73]; char *cp; struct evdns_request *handle; struct request *req; int i; EVUTIL_ASSERT(in); cp = buf; for (i=15; i >= 0; --i) { u8 byte = in->s6_addr[i]; *cp++ = "0123456789abcdef"[byte & 0x0f]; *cp++ = '.'; *cp++ = "0123456789abcdef"[byte >> 4]; *cp++ = '.'; } EVUTIL_ASSERT(cp + strlen("ip6.arpa") < buf+sizeof(buf)); memcpy(cp, "ip6.arpa", strlen("ip6.arpa")+1); handle = mm_calloc(1, sizeof(*handle)); if (handle == NULL) return NULL; log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf); EVDNS_LOCK(base); req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr); if (req) request_submit(req); if (handle->current_req == NULL) { mm_free(handle); handle = NULL; } EVDNS_UNLOCK(base); return (handle); } int evdns_resolve_reverse_ipv6(const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) { return evdns_base_resolve_reverse_ipv6(current_base, in, flags, callback, ptr) ? 0 : -1; } /* ================================================================= */ /* Search support */ /* */ /* the libc resolver has support for searching a number of domains */ /* to find a name. If nothing else then it takes the single domain */ /* from the gethostname() call. */ /* */ /* It can also be configured via the domain and search options in a */ /* resolv.conf. */ /* */ /* The ndots option controls how many dots it takes for the resolver */ /* to decide that a name is non-local and so try a raw lookup first. */ struct search_domain { int len; struct search_domain *next; /* the text string is appended to this structure */ }; struct search_state { int refcount; int ndots; int num_domains; struct search_domain *head; }; static void search_state_decref(struct search_state *const state) { if (!state) return; state->refcount--; if (!state->refcount) { struct search_domain *next, *dom; for (dom = state->head; dom; dom = next) { next = dom->next; mm_free(dom); } mm_free(state); } } static struct search_state * search_state_new(void) { struct search_state *state = (struct search_state *) mm_malloc(sizeof(struct search_state)); if (!state) return NULL; memset(state, 0, sizeof(struct search_state)); state->refcount = 1; state->ndots = 1; return state; } static void search_postfix_clear(struct evdns_base *base) { search_state_decref(base->global_search_state); base->global_search_state = search_state_new(); } /* exported function */ void evdns_base_search_clear(struct evdns_base *base) { EVDNS_LOCK(base); search_postfix_clear(base); EVDNS_UNLOCK(base); } void evdns_search_clear(void) { evdns_base_search_clear(current_base); } static void search_postfix_add(struct evdns_base *base, const char *domain) { size_t domain_len; struct search_domain *sdomain; while (domain[0] == '.') domain++; domain_len = strlen(domain); ASSERT_LOCKED(base); if (!base->global_search_state) base->global_search_state = search_state_new(); if (!base->global_search_state) return; base->global_search_state->num_domains++; sdomain = (struct search_domain *) mm_malloc(sizeof(struct search_domain) + domain_len); if (!sdomain) return; memcpy( ((u8 *) sdomain) + sizeof(struct search_domain), domain, domain_len); sdomain->next = base->global_search_state->head; sdomain->len = (int) domain_len; base->global_search_state->head = sdomain; } /* reverse the order of members in the postfix list. This is needed because, */ /* when parsing resolv.conf we push elements in the wrong order */ static void search_reverse(struct evdns_base *base) { struct search_domain *cur, *prev = NULL, *next; ASSERT_LOCKED(base); cur = base->global_search_state->head; while (cur) { next = cur->next; cur->next = prev; prev = cur; cur = next; } base->global_search_state->head = prev; } /* exported function */ void evdns_base_search_add(struct evdns_base *base, const char *domain) { EVDNS_LOCK(base); search_postfix_add(base, domain); EVDNS_UNLOCK(base); } void evdns_search_add(const char *domain) { evdns_base_search_add(current_base, domain); } /* exported function */ void evdns_base_search_ndots_set(struct evdns_base *base, const int ndots) { EVDNS_LOCK(base); if (!base->global_search_state) base->global_search_state = search_state_new(); if (base->global_search_state) base->global_search_state->ndots = ndots; EVDNS_UNLOCK(base); } void evdns_search_ndots_set(const int ndots) { evdns_base_search_ndots_set(current_base, ndots); } static void search_set_from_hostname(struct evdns_base *base) { char hostname[HOST_NAME_MAX + 1], *domainname; ASSERT_LOCKED(base); search_postfix_clear(base); if (gethostname(hostname, sizeof(hostname))) return; domainname = strchr(hostname, '.'); if (!domainname) return; search_postfix_add(base, domainname); } /* warning: returns malloced string */ static char * search_make_new(const struct search_state *const state, int n, const char *const base_name) { const size_t base_len = strlen(base_name); char need_to_append_dot; struct search_domain *dom; if (!base_len) return NULL; need_to_append_dot = base_name[base_len - 1] == '.' ? 0 : 1; for (dom = state->head; dom; dom = dom->next) { if (!n--) { /* this is the postfix we want */ /* the actual postfix string is kept at the end of the structure */ const u8 *const postfix = ((u8 *) dom) + sizeof(struct search_domain); const int postfix_len = dom->len; char *const newname = (char *) mm_malloc(base_len + need_to_append_dot + postfix_len + 1); if (!newname) return NULL; memcpy(newname, base_name, base_len); if (need_to_append_dot) newname[base_len] = '.'; memcpy(newname + base_len + need_to_append_dot, postfix, postfix_len); newname[base_len + need_to_append_dot + postfix_len] = 0; return newname; } } /* we ran off the end of the list and still didn't find the requested string */ EVUTIL_ASSERT(0); return NULL; /* unreachable; stops warnings in some compilers. */ } static struct request * search_request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg) { ASSERT_LOCKED(base); EVUTIL_ASSERT(type == TYPE_A || type == TYPE_AAAA); EVUTIL_ASSERT(handle->current_req == NULL); if ( ((flags & DNS_QUERY_NO_SEARCH) == 0) && base->global_search_state && base->global_search_state->num_domains) { /* we have some domains to search */ struct request *req; if (string_num_dots(name) >= base->global_search_state->ndots) { req = request_new(base, handle, type, name, flags, user_callback, user_arg); if (!req) return NULL; handle->search_index = -1; } else { char *const new_name = search_make_new(base->global_search_state, 0, name); if (!new_name) return NULL; req = request_new(base, handle, type, new_name, flags, user_callback, user_arg); mm_free(new_name); if (!req) return NULL; handle->search_index = 0; } EVUTIL_ASSERT(handle->search_origname == NULL); handle->search_origname = mm_strdup(name); if (handle->search_origname == NULL) { /* XXX Should we dealloc req? If yes, how? */ if (req) mm_free(req); return NULL; } handle->search_state = base->global_search_state; handle->search_flags = flags; base->global_search_state->refcount++; request_submit(req); return req; } else { struct request *const req = request_new(base, handle, type, name, flags, user_callback, user_arg); if (!req) return NULL; request_submit(req); return req; } } /* this is called when a request has failed to find a name. We need to check */ /* if it is part of a search and, if so, try the next name in the list */ /* returns: */ /* 0 another request has been submitted */ /* 1 no more requests needed */ static int search_try_next(struct evdns_request *const handle) { struct request *req = handle->current_req; struct evdns_base *base = req->base; struct request *newreq; ASSERT_LOCKED(base); if (handle->search_state) { /* it is part of a search */ char *new_name; handle->search_index++; if (handle->search_index >= handle->search_state->num_domains) { /* no more postfixes to try, however we may need to try */ /* this name without a postfix */ if (string_num_dots(handle->search_origname) < handle->search_state->ndots) { /* yep, we need to try it raw */ newreq = request_new(base, NULL, req->request_type, handle->search_origname, handle->search_flags, req->user_callback, req->user_pointer); log(EVDNS_LOG_DEBUG, "Search: trying raw query %s", handle->search_origname); if (newreq) { search_request_finished(handle); goto submit_next; } } return 1; } new_name = search_make_new(handle->search_state, handle->search_index, handle->search_origname); if (!new_name) return 1; log(EVDNS_LOG_DEBUG, "Search: now trying %s (%d)", new_name, handle->search_index); newreq = request_new(base, NULL, req->request_type, new_name, handle->search_flags, req->user_callback, req->user_pointer); mm_free(new_name); if (!newreq) return 1; goto submit_next; } return 1; submit_next: request_finished(req, &REQ_HEAD(req->base, req->trans_id), 0); handle->current_req = newreq; newreq->handle = handle; request_submit(newreq); return 0; } static void search_request_finished(struct evdns_request *const handle) { ASSERT_LOCKED(handle->current_req->base); if (handle->search_state) { search_state_decref(handle->search_state); handle->search_state = NULL; } if (handle->search_origname) { mm_free(handle->search_origname); handle->search_origname = NULL; } } /* ================================================================= */ /* Parsing resolv.conf files */ static void evdns_resolv_set_defaults(struct evdns_base *base, int flags) { int add_default = flags & DNS_OPTION_NAMESERVERS; if (flags & DNS_OPTION_NAMESERVERS_NO_DEFAULT) add_default = 0; /* if the file isn't found then we assume a local resolver */ ASSERT_LOCKED(base); if (flags & DNS_OPTION_SEARCH) search_set_from_hostname(base); if (add_default) evdns_base_nameserver_ip_add(base, "127.0.0.1"); } #ifndef EVENT__HAVE_STRTOK_R static char * strtok_r(char *s, const char *delim, char **state) { char *cp, *start; start = cp = s ? s : *state; if (!cp) return NULL; while (*cp && !strchr(delim, *cp)) ++cp; if (!*cp) { if (cp == start) return NULL; *state = NULL; return start; } else { *cp++ = '\0'; *state = cp; return start; } } #endif /* helper version of atoi which returns -1 on error */ static int strtoint(const char *const str) { char *endptr; const int r = strtol(str, &endptr, 10); if (*endptr) return -1; return r; } /* Parse a number of seconds into a timeval; return -1 on error. */ static int evdns_strtotimeval(const char *const str, struct timeval *out) { double d; char *endptr; d = strtod(str, &endptr); if (*endptr) return -1; if (d < 0) return -1; out->tv_sec = (int) d; out->tv_usec = (int) ((d - (int) d)*1000000); if (out->tv_sec == 0 && out->tv_usec < 1000) /* less than 1 msec */ return -1; return 0; } /* helper version of atoi that returns -1 on error and clips to bounds. */ static int strtoint_clipped(const char *const str, int min, int max) { int r = strtoint(str); if (r == -1) return r; else if (rmax) return max; else return r; } static int evdns_base_set_max_requests_inflight(struct evdns_base *base, int maxinflight) { int old_n_heads = base->n_req_heads, n_heads; struct request **old_heads = base->req_heads, **new_heads, *req; int i; ASSERT_LOCKED(base); if (maxinflight < 1) maxinflight = 1; n_heads = (maxinflight+4) / 5; EVUTIL_ASSERT(n_heads > 0); new_heads = mm_calloc(n_heads, sizeof(struct request*)); if (!new_heads) return (-1); if (old_heads) { for (i = 0; i < old_n_heads; ++i) { while (old_heads[i]) { req = old_heads[i]; evdns_request_remove(req, &old_heads[i]); evdns_request_insert(req, &new_heads[req->trans_id % n_heads]); } } mm_free(old_heads); } base->req_heads = new_heads; base->n_req_heads = n_heads; base->global_max_requests_inflight = maxinflight; return (0); } /* exported function */ int evdns_base_set_option(struct evdns_base *base, const char *option, const char *val) { int res; EVDNS_LOCK(base); res = evdns_base_set_option_impl(base, option, val, DNS_OPTIONS_ALL); EVDNS_UNLOCK(base); return res; } static inline int str_matches_option(const char *s1, const char *optionname) { /* Option names are given as "option:" We accept either 'option' in * s1, or 'option:randomjunk'. The latter form is to implement the * resolv.conf parser. */ size_t optlen = strlen(optionname); size_t slen = strlen(s1); if (slen == optlen || slen == optlen - 1) return !strncmp(s1, optionname, slen); else if (slen > optlen) return !strncmp(s1, optionname, optlen); else return 0; } static int evdns_base_set_option_impl(struct evdns_base *base, const char *option, const char *val, int flags) { ASSERT_LOCKED(base); if (str_matches_option(option, "ndots:")) { const int ndots = strtoint(val); if (ndots == -1) return -1; if (!(flags & DNS_OPTION_SEARCH)) return 0; log(EVDNS_LOG_DEBUG, "Setting ndots to %d", ndots); if (!base->global_search_state) base->global_search_state = search_state_new(); if (!base->global_search_state) return -1; base->global_search_state->ndots = ndots; } else if (str_matches_option(option, "timeout:")) { struct timeval tv; if (evdns_strtotimeval(val, &tv) == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting timeout to %s", val); memcpy(&base->global_timeout, &tv, sizeof(struct timeval)); } else if (str_matches_option(option, "getaddrinfo-allow-skew:")) { struct timeval tv; if (evdns_strtotimeval(val, &tv) == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting getaddrinfo-allow-skew to %s", val); memcpy(&base->global_getaddrinfo_allow_skew, &tv, sizeof(struct timeval)); } else if (str_matches_option(option, "max-timeouts:")) { const int maxtimeout = strtoint_clipped(val, 1, 255); if (maxtimeout == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting maximum allowed timeouts to %d", maxtimeout); base->global_max_nameserver_timeout = maxtimeout; } else if (str_matches_option(option, "max-inflight:")) { const int maxinflight = strtoint_clipped(val, 1, 65000); if (maxinflight == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting maximum inflight requests to %d", maxinflight); evdns_base_set_max_requests_inflight(base, maxinflight); } else if (str_matches_option(option, "attempts:")) { int retries = strtoint(val); if (retries == -1) return -1; if (retries > 255) retries = 255; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting retries to %d", retries); base->global_max_retransmits = retries; } else if (str_matches_option(option, "randomize-case:")) { int randcase = strtoint(val); if (randcase == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; base->global_randomize_case = randcase; } else if (str_matches_option(option, "bind-to:")) { /* XXX This only applies to successive nameservers, not * to already-configured ones. We might want to fix that. */ int len = sizeof(base->global_outgoing_address); if (!(flags & DNS_OPTION_NAMESERVERS)) return 0; if (evutil_parse_sockaddr_port(val, (struct sockaddr*)&base->global_outgoing_address, &len)) return -1; base->global_outgoing_addrlen = len; } else if (str_matches_option(option, "initial-probe-timeout:")) { struct timeval tv; if (evdns_strtotimeval(val, &tv) == -1) return -1; if (tv.tv_sec > 3600) tv.tv_sec = 3600; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting initial probe timeout to %s", val); memcpy(&base->global_nameserver_probe_initial_timeout, &tv, sizeof(tv)); } else if (str_matches_option(option, "so-rcvbuf:")) { int buf = strtoint(val); if (buf == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting SO_RCVBUF to %s", val); base->so_rcvbuf = buf; } else if (str_matches_option(option, "so-sndbuf:")) { int buf = strtoint(val); if (buf == -1) return -1; if (!(flags & DNS_OPTION_MISC)) return 0; log(EVDNS_LOG_DEBUG, "Setting SO_SNDBUF to %s", val); base->so_sndbuf = buf; } return 0; } int evdns_set_option(const char *option, const char *val, int flags) { if (!current_base) current_base = evdns_base_new(NULL, 0); return evdns_base_set_option(current_base, option, val); } static void resolv_conf_parse_line(struct evdns_base *base, char *const start, int flags) { char *strtok_state; static const char *const delims = " \t"; #define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state) char *const first_token = strtok_r(start, delims, &strtok_state); ASSERT_LOCKED(base); if (!first_token) return; if (!strcmp(first_token, "nameserver") && (flags & DNS_OPTION_NAMESERVERS)) { const char *const nameserver = NEXT_TOKEN; if (nameserver) evdns_base_nameserver_ip_add(base, nameserver); } else if (!strcmp(first_token, "domain") && (flags & DNS_OPTION_SEARCH)) { const char *const domain = NEXT_TOKEN; if (domain) { search_postfix_clear(base); search_postfix_add(base, domain); } } else if (!strcmp(first_token, "search") && (flags & DNS_OPTION_SEARCH)) { const char *domain; search_postfix_clear(base); while ((domain = NEXT_TOKEN)) { search_postfix_add(base, domain); } search_reverse(base); } else if (!strcmp(first_token, "options")) { const char *option; while ((option = NEXT_TOKEN)) { const char *val = strchr(option, ':'); evdns_base_set_option_impl(base, option, val ? val+1 : "", flags); } } #undef NEXT_TOKEN } /* exported function */ /* returns: */ /* 0 no errors */ /* 1 failed to open file */ /* 2 failed to stat file */ /* 3 file too large */ /* 4 out of memory */ /* 5 short read from file */ int evdns_base_resolv_conf_parse(struct evdns_base *base, int flags, const char *const filename) { int res; EVDNS_LOCK(base); res = evdns_base_resolv_conf_parse_impl(base, flags, filename); EVDNS_UNLOCK(base); return res; } static char * evdns_get_default_hosts_filename(void) { #ifdef _WIN32 /* Windows is a little coy about where it puts its configuration * files. Sure, they're _usually_ in C:\windows\system32, but * there's no reason in principle they couldn't be in * W:\hoboken chicken emergency\ */ char path[MAX_PATH+1]; static const char hostfile[] = "\\drivers\\etc\\hosts"; char *path_out; size_t len_out; if (! SHGetSpecialFolderPathA(NULL, path, CSIDL_SYSTEM, 0)) return NULL; len_out = strlen(path)+strlen(hostfile)+1; path_out = mm_malloc(len_out); evutil_snprintf(path_out, len_out, "%s%s", path, hostfile); return path_out; #else return mm_strdup("/etc/hosts"); #endif } static int evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename) { size_t n; char *resolv; char *start; int err = 0; int add_default; log(EVDNS_LOG_DEBUG, "Parsing resolv.conf file %s", filename); add_default = flags & DNS_OPTION_NAMESERVERS; if (flags & DNS_OPTION_NAMESERVERS_NO_DEFAULT) add_default = 0; if (flags & DNS_OPTION_HOSTSFILE) { char *fname = evdns_get_default_hosts_filename(); evdns_base_load_hosts(base, fname); if (fname) mm_free(fname); } if (!filename) { evdns_resolv_set_defaults(base, flags); return 1; } if ((err = evutil_read_file_(filename, &resolv, &n, 0)) < 0) { if (err == -1) { /* No file. */ evdns_resolv_set_defaults(base, flags); return 1; } else { return 2; } } start = resolv; for (;;) { char *const newline = strchr(start, '\n'); if (!newline) { resolv_conf_parse_line(base, start, flags); break; } else { *newline = 0; resolv_conf_parse_line(base, start, flags); start = newline + 1; } } if (!base->server_head && add_default) { /* no nameservers were configured. */ evdns_base_nameserver_ip_add(base, "127.0.0.1"); err = 6; } if (flags & DNS_OPTION_SEARCH && (!base->global_search_state || base->global_search_state->num_domains == 0)) { search_set_from_hostname(base); } mm_free(resolv); return err; } int evdns_resolv_conf_parse(int flags, const char *const filename) { if (!current_base) current_base = evdns_base_new(NULL, 0); return evdns_base_resolv_conf_parse(current_base, flags, filename); } #ifdef _WIN32 /* Add multiple nameservers from a space-or-comma-separated list. */ static int evdns_nameserver_ip_add_line(struct evdns_base *base, const char *ips) { const char *addr; char *buf; int r; ASSERT_LOCKED(base); while (*ips) { while (isspace(*ips) || *ips == ',' || *ips == '\t') ++ips; addr = ips; while (isdigit(*ips) || *ips == '.' || *ips == ':' || *ips=='[' || *ips==']') ++ips; buf = mm_malloc(ips-addr+1); if (!buf) return 4; memcpy(buf, addr, ips-addr); buf[ips-addr] = '\0'; r = evdns_base_nameserver_ip_add(base, buf); mm_free(buf); if (r) return r; } return 0; } typedef DWORD(WINAPI *GetNetworkParams_fn_t)(FIXED_INFO *, DWORD*); /* Use the windows GetNetworkParams interface in iphlpapi.dll to */ /* figure out what our nameservers are. */ static int load_nameservers_with_getnetworkparams(struct evdns_base *base) { /* Based on MSDN examples and inspection of c-ares code. */ FIXED_INFO *fixed; HMODULE handle = 0; ULONG size = sizeof(FIXED_INFO); void *buf = NULL; int status = 0, r, added_any; IP_ADDR_STRING *ns; GetNetworkParams_fn_t fn; ASSERT_LOCKED(base); if (!(handle = evutil_load_windows_system_library_( TEXT("iphlpapi.dll")))) { log(EVDNS_LOG_WARN, "Could not open iphlpapi.dll"); status = -1; goto done; } if (!(fn = (GetNetworkParams_fn_t) GetProcAddress(handle, "GetNetworkParams"))) { log(EVDNS_LOG_WARN, "Could not get address of function."); status = -1; goto done; } buf = mm_malloc(size); if (!buf) { status = 4; goto done; } fixed = buf; r = fn(fixed, &size); if (r != ERROR_SUCCESS && r != ERROR_BUFFER_OVERFLOW) { status = -1; goto done; } if (r != ERROR_SUCCESS) { mm_free(buf); buf = mm_malloc(size); if (!buf) { status = 4; goto done; } fixed = buf; r = fn(fixed, &size); if (r != ERROR_SUCCESS) { log(EVDNS_LOG_DEBUG, "fn() failed."); status = -1; goto done; } } EVUTIL_ASSERT(fixed); added_any = 0; ns = &(fixed->DnsServerList); while (ns) { r = evdns_nameserver_ip_add_line(base, ns->IpAddress.String); if (r) { log(EVDNS_LOG_DEBUG,"Could not add nameserver %s to list,error: %d", (ns->IpAddress.String),(int)GetLastError()); status = r; } else { ++added_any; log(EVDNS_LOG_DEBUG,"Successfully added %s as nameserver",ns->IpAddress.String); } ns = ns->Next; } if (!added_any) { log(EVDNS_LOG_DEBUG, "No nameservers added."); if (status == 0) status = -1; } else { status = 0; } done: if (buf) mm_free(buf); if (handle) FreeLibrary(handle); return status; } static int config_nameserver_from_reg_key(struct evdns_base *base, HKEY key, const TCHAR *subkey) { char *buf; DWORD bufsz = 0, type = 0; int status = 0; ASSERT_LOCKED(base); if (RegQueryValueEx(key, subkey, 0, &type, NULL, &bufsz) != ERROR_MORE_DATA) return -1; if (!(buf = mm_malloc(bufsz))) return -1; if (RegQueryValueEx(key, subkey, 0, &type, (LPBYTE)buf, &bufsz) == ERROR_SUCCESS && bufsz > 1) { status = evdns_nameserver_ip_add_line(base,buf); } mm_free(buf); return status; } #define SERVICES_KEY TEXT("System\\CurrentControlSet\\Services\\") #define WIN_NS_9X_KEY SERVICES_KEY TEXT("VxD\\MSTCP") #define WIN_NS_NT_KEY SERVICES_KEY TEXT("Tcpip\\Parameters") static int load_nameservers_from_registry(struct evdns_base *base) { int found = 0; int r; #define TRY(k, name) \ if (!found && config_nameserver_from_reg_key(base,k,TEXT(name)) == 0) { \ log(EVDNS_LOG_DEBUG,"Found nameservers in %s/%s",#k,name); \ found = 1; \ } else if (!found) { \ log(EVDNS_LOG_DEBUG,"Didn't find nameservers in %s/%s", \ #k,#name); \ } ASSERT_LOCKED(base); if (((int)GetVersion()) > 0) { /* NT */ HKEY nt_key = 0, interfaces_key = 0; if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_NT_KEY, 0, KEY_READ, &nt_key) != ERROR_SUCCESS) { log(EVDNS_LOG_DEBUG,"Couldn't open nt key, %d",(int)GetLastError()); return -1; } r = RegOpenKeyEx(nt_key, TEXT("Interfaces"), 0, KEY_QUERY_VALUE|KEY_ENUMERATE_SUB_KEYS, &interfaces_key); if (r != ERROR_SUCCESS) { log(EVDNS_LOG_DEBUG,"Couldn't open interfaces key, %d",(int)GetLastError()); return -1; } TRY(nt_key, "NameServer"); TRY(nt_key, "DhcpNameServer"); TRY(interfaces_key, "NameServer"); TRY(interfaces_key, "DhcpNameServer"); RegCloseKey(interfaces_key); RegCloseKey(nt_key); } else { HKEY win_key = 0; if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_9X_KEY, 0, KEY_READ, &win_key) != ERROR_SUCCESS) { log(EVDNS_LOG_DEBUG, "Couldn't open registry key, %d", (int)GetLastError()); return -1; } TRY(win_key, "NameServer"); RegCloseKey(win_key); } if (found == 0) { log(EVDNS_LOG_WARN,"Didn't find any nameservers."); } return found ? 0 : -1; #undef TRY } int evdns_base_config_windows_nameservers(struct evdns_base *base) { int r; char *fname; if (base == NULL) base = current_base; if (base == NULL) return -1; EVDNS_LOCK(base); fname = evdns_get_default_hosts_filename(); log(EVDNS_LOG_DEBUG, "Loading hosts entries from %s", fname); evdns_base_load_hosts(base, fname); if (fname) mm_free(fname); if (load_nameservers_with_getnetworkparams(base) == 0) { EVDNS_UNLOCK(base); return 0; } r = load_nameservers_from_registry(base); EVDNS_UNLOCK(base); return r; } int evdns_config_windows_nameservers(void) { if (!current_base) { current_base = evdns_base_new(NULL, 1); return current_base == NULL ? -1 : 0; } else { return evdns_base_config_windows_nameservers(current_base); } } #endif struct evdns_base * evdns_base_new(struct event_base *event_base, int flags) { struct evdns_base *base; if (evutil_secure_rng_init() < 0) { log(EVDNS_LOG_WARN, "Unable to seed random number generator; " "DNS can't run."); return NULL; } /* Give the evutil library a hook into its evdns-enabled * functionality. We can't just call evdns_getaddrinfo directly or * else libevent-core will depend on libevent-extras. */ evutil_set_evdns_getaddrinfo_fn_(evdns_getaddrinfo); evutil_set_evdns_getaddrinfo_cancel_fn_(evdns_getaddrinfo_cancel); base = mm_malloc(sizeof(struct evdns_base)); if (base == NULL) return (NULL); memset(base, 0, sizeof(struct evdns_base)); base->req_waiting_head = NULL; EVTHREAD_ALLOC_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE); EVDNS_LOCK(base); /* Set max requests inflight and allocate req_heads. */ base->req_heads = NULL; evdns_base_set_max_requests_inflight(base, 64); base->server_head = NULL; base->event_base = event_base; base->global_good_nameservers = base->global_requests_inflight = base->global_requests_waiting = 0; base->global_timeout.tv_sec = 5; base->global_timeout.tv_usec = 0; base->global_max_reissues = 1; base->global_max_retransmits = 3; base->global_max_nameserver_timeout = 3; base->global_search_state = NULL; base->global_randomize_case = 1; base->global_getaddrinfo_allow_skew.tv_sec = 3; base->global_getaddrinfo_allow_skew.tv_usec = 0; base->global_nameserver_probe_initial_timeout.tv_sec = 10; base->global_nameserver_probe_initial_timeout.tv_usec = 0; TAILQ_INIT(&base->hostsdb); #define EVDNS_BASE_ALL_FLAGS ( \ EVDNS_BASE_INITIALIZE_NAMESERVERS | \ EVDNS_BASE_DISABLE_WHEN_INACTIVE | \ EVDNS_BASE_NAMESERVERS_NO_DEFAULT | \ 0) if (flags & ~EVDNS_BASE_ALL_FLAGS) { flags = EVDNS_BASE_INITIALIZE_NAMESERVERS; log(EVDNS_LOG_WARN, "Unrecognized flag passed to evdns_base_new(). Assuming " "you meant EVDNS_BASE_INITIALIZE_NAMESERVERS."); } #undef EVDNS_BASE_ALL_FLAGS if (flags & EVDNS_BASE_INITIALIZE_NAMESERVERS) { int r; int opts = DNS_OPTIONS_ALL; if (flags & EVDNS_BASE_NAMESERVERS_NO_DEFAULT) { opts |= DNS_OPTION_NAMESERVERS_NO_DEFAULT; } #ifdef _WIN32 r = evdns_base_config_windows_nameservers(base); #else r = evdns_base_resolv_conf_parse(base, opts, "/etc/resolv.conf"); #endif if (r) { evdns_base_free_and_unlock(base, 0); return NULL; } } if (flags & EVDNS_BASE_DISABLE_WHEN_INACTIVE) { base->disable_when_inactive = 1; } EVDNS_UNLOCK(base); return base; } int evdns_init(void) { struct evdns_base *base = evdns_base_new(NULL, 1); if (base) { current_base = base; return 0; } else { return -1; } } const char * evdns_err_to_string(int err) { switch (err) { case DNS_ERR_NONE: return "no error"; case DNS_ERR_FORMAT: return "misformatted query"; case DNS_ERR_SERVERFAILED: return "server failed"; case DNS_ERR_NOTEXIST: return "name does not exist"; case DNS_ERR_NOTIMPL: return "query not implemented"; case DNS_ERR_REFUSED: return "refused"; case DNS_ERR_TRUNCATED: return "reply truncated or ill-formed"; case DNS_ERR_UNKNOWN: return "unknown"; case DNS_ERR_TIMEOUT: return "request timed out"; case DNS_ERR_SHUTDOWN: return "dns subsystem shut down"; case DNS_ERR_CANCEL: return "dns request canceled"; case DNS_ERR_NODATA: return "no records in the reply"; default: return "[Unknown error code]"; } } static void evdns_nameserver_free(struct nameserver *server) { if (server->socket >= 0) evutil_closesocket(server->socket); (void) event_del(&server->event); event_debug_unassign(&server->event); if (server->state == 0) (void) event_del(&server->timeout_event); if (server->probe_request) { evdns_cancel_request(server->base, server->probe_request); server->probe_request = NULL; } event_debug_unassign(&server->timeout_event); mm_free(server); } static void evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests) { struct nameserver *server, *server_next; struct search_domain *dom, *dom_next; int i; /* Requires that we hold the lock. */ /* TODO(nickm) we might need to refcount here. */ while (base->req_waiting_head) { if (fail_requests) reply_schedule_callback(base->req_waiting_head, 0, DNS_ERR_SHUTDOWN, NULL); request_finished(base->req_waiting_head, &base->req_waiting_head, 1); } for (i = 0; i < base->n_req_heads; ++i) { while (base->req_heads[i]) { if (fail_requests) reply_schedule_callback(base->req_heads[i], 0, DNS_ERR_SHUTDOWN, NULL); request_finished(base->req_heads[i], &REQ_HEAD(base, base->req_heads[i]->trans_id), 1); } } base->global_requests_inflight = base->global_requests_waiting = 0; for (server = base->server_head; server; server = server_next) { server_next = server->next; /** already done something before */ server->probe_request = NULL; evdns_nameserver_free(server); if (server_next == base->server_head) break; } base->server_head = NULL; base->global_good_nameservers = 0; if (base->global_search_state) { for (dom = base->global_search_state->head; dom; dom = dom_next) { dom_next = dom->next; mm_free(dom); } mm_free(base->global_search_state); base->global_search_state = NULL; } { struct hosts_entry *victim; while ((victim = TAILQ_FIRST(&base->hostsdb))) { TAILQ_REMOVE(&base->hostsdb, victim, next); mm_free(victim); } } mm_free(base->req_heads); EVDNS_UNLOCK(base); EVTHREAD_FREE_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE); mm_free(base); } void evdns_base_free(struct evdns_base *base, int fail_requests) { EVDNS_LOCK(base); evdns_base_free_and_unlock(base, fail_requests); } void evdns_base_clear_host_addresses(struct evdns_base *base) { struct hosts_entry *victim; EVDNS_LOCK(base); while ((victim = TAILQ_FIRST(&base->hostsdb))) { TAILQ_REMOVE(&base->hostsdb, victim, next); mm_free(victim); } EVDNS_UNLOCK(base); } void evdns_shutdown(int fail_requests) { if (current_base) { struct evdns_base *b = current_base; current_base = NULL; evdns_base_free(b, fail_requests); } evdns_log_fn = NULL; } static int evdns_base_parse_hosts_line(struct evdns_base *base, char *line) { char *strtok_state; static const char *const delims = " \t"; char *const addr = strtok_r(line, delims, &strtok_state); char *hostname, *hash; struct sockaddr_storage ss; int socklen = sizeof(ss); ASSERT_LOCKED(base); #define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state) if (!addr || *addr == '#') return 0; memset(&ss, 0, sizeof(ss)); if (evutil_parse_sockaddr_port(addr, (struct sockaddr*)&ss, &socklen)<0) return -1; if (socklen > (int)sizeof(struct sockaddr_in6)) return -1; if (sockaddr_getport((struct sockaddr*)&ss)) return -1; while ((hostname = NEXT_TOKEN)) { struct hosts_entry *he; size_t namelen; if ((hash = strchr(hostname, '#'))) { if (hash == hostname) return 0; *hash = '\0'; } namelen = strlen(hostname); he = mm_calloc(1, sizeof(struct hosts_entry)+namelen); if (!he) return -1; EVUTIL_ASSERT(socklen <= (int)sizeof(he->addr)); memcpy(&he->addr, &ss, socklen); memcpy(he->hostname, hostname, namelen+1); he->addrlen = socklen; TAILQ_INSERT_TAIL(&base->hostsdb, he, next); if (hash) return 0; } return 0; #undef NEXT_TOKEN } static int evdns_base_load_hosts_impl(struct evdns_base *base, const char *hosts_fname) { char *str=NULL, *cp, *eol; size_t len; int err=0; ASSERT_LOCKED(base); if (hosts_fname == NULL || (err = evutil_read_file_(hosts_fname, &str, &len, 0)) < 0) { char tmp[64]; strlcpy(tmp, "127.0.0.1 localhost", sizeof(tmp)); evdns_base_parse_hosts_line(base, tmp); strlcpy(tmp, "::1 localhost", sizeof(tmp)); evdns_base_parse_hosts_line(base, tmp); return err ? -1 : 0; } /* This will break early if there is a NUL in the hosts file. * Probably not a problem.*/ cp = str; for (;;) { eol = strchr(cp, '\n'); if (eol) { *eol = '\0'; evdns_base_parse_hosts_line(base, cp); cp = eol+1; } else { evdns_base_parse_hosts_line(base, cp); break; } } mm_free(str); return 0; } int evdns_base_load_hosts(struct evdns_base *base, const char *hosts_fname) { int res; if (!base) base = current_base; EVDNS_LOCK(base); res = evdns_base_load_hosts_impl(base, hosts_fname); EVDNS_UNLOCK(base); return res; } /* A single request for a getaddrinfo, either v4 or v6. */ struct getaddrinfo_subrequest { struct evdns_request *r; ev_uint32_t type; }; /* State data used to implement an in-progress getaddrinfo. */ struct evdns_getaddrinfo_request { struct evdns_base *evdns_base; /* Copy of the modified 'hints' data that we'll use to build * answers. */ struct evutil_addrinfo hints; /* The callback to invoke when we're done */ evdns_getaddrinfo_cb user_cb; /* User-supplied data to give to the callback. */ void *user_data; /* The port to use when building sockaddrs. */ ev_uint16_t port; /* The sub_request for an A record (if any) */ struct getaddrinfo_subrequest ipv4_request; /* The sub_request for an AAAA record (if any) */ struct getaddrinfo_subrequest ipv6_request; /* The cname result that we were told (if any) */ char *cname_result; /* If we have one request answered and one request still inflight, * then this field holds the answer from the first request... */ struct evutil_addrinfo *pending_result; /* And this event is a timeout that will tell us to cancel the second * request if it's taking a long time. */ struct event timeout; /* And this field holds the error code from the first request... */ int pending_error; /* If this is set, the user canceled this request. */ unsigned user_canceled : 1; /* If this is set, the user can no longer cancel this request; we're * just waiting for the free. */ unsigned request_done : 1; }; /* Convert an evdns errors to the equivalent getaddrinfo error. */ static int evdns_err_to_getaddrinfo_err(int e1) { /* XXX Do this better! */ if (e1 == DNS_ERR_NONE) return 0; else if (e1 == DNS_ERR_NOTEXIST) return EVUTIL_EAI_NONAME; else return EVUTIL_EAI_FAIL; } /* Return the more informative of two getaddrinfo errors. */ static int getaddrinfo_merge_err(int e1, int e2) { /* XXXX be cleverer here. */ if (e1 == 0) return e2; else return e1; } static void free_getaddrinfo_request(struct evdns_getaddrinfo_request *data) { /* DO NOT CALL this if either of the requests is pending. Only once * both callbacks have been invoked is it safe to free the request */ if (data->pending_result) evutil_freeaddrinfo(data->pending_result); if (data->cname_result) mm_free(data->cname_result); event_del(&data->timeout); mm_free(data); return; } static void add_cname_to_reply(struct evdns_getaddrinfo_request *data, struct evutil_addrinfo *ai) { if (data->cname_result && ai) { ai->ai_canonname = data->cname_result; data->cname_result = NULL; } } /* Callback: invoked when one request in a mixed-format A/AAAA getaddrinfo * request has finished, but the other one took too long to answer. Pass * along the answer we got, and cancel the other request. */ static void evdns_getaddrinfo_timeout_cb(evutil_socket_t fd, short what, void *ptr) { int v4_timedout = 0, v6_timedout = 0; struct evdns_getaddrinfo_request *data = ptr; /* Cancel any pending requests, and note which one */ if (data->ipv4_request.r) { /* XXXX This does nothing if the request's callback is already * running (pending_cb is set). */ evdns_cancel_request(NULL, data->ipv4_request.r); v4_timedout = 1; EVDNS_LOCK(data->evdns_base); ++data->evdns_base->getaddrinfo_ipv4_timeouts; EVDNS_UNLOCK(data->evdns_base); } if (data->ipv6_request.r) { /* XXXX This does nothing if the request's callback is already * running (pending_cb is set). */ evdns_cancel_request(NULL, data->ipv6_request.r); v6_timedout = 1; EVDNS_LOCK(data->evdns_base); ++data->evdns_base->getaddrinfo_ipv6_timeouts; EVDNS_UNLOCK(data->evdns_base); } /* We only use this timeout callback when we have an answer for * one address. */ EVUTIL_ASSERT(!v4_timedout || !v6_timedout); /* Report the outcome of the other request that didn't time out. */ if (data->pending_result) { add_cname_to_reply(data, data->pending_result); data->user_cb(0, data->pending_result, data->user_data); data->pending_result = NULL; } else { int e = data->pending_error; if (!e) e = EVUTIL_EAI_AGAIN; data->user_cb(e, NULL, data->user_data); } data->user_cb = NULL; /* prevent double-call if evdns callbacks are * in-progress. XXXX It would be better if this * weren't necessary. */ if (!v4_timedout && !v6_timedout) { /* should be impossible? XXXX */ free_getaddrinfo_request(data); } } static int evdns_getaddrinfo_set_timeout(struct evdns_base *evdns_base, struct evdns_getaddrinfo_request *data) { return event_add(&data->timeout, &evdns_base->global_getaddrinfo_allow_skew); } static inline int evdns_result_is_answer(int result) { return (result != DNS_ERR_NOTIMPL && result != DNS_ERR_REFUSED && result != DNS_ERR_SERVERFAILED && result != DNS_ERR_CANCEL); } static void evdns_getaddrinfo_gotresolve(int result, char type, int count, int ttl, void *addresses, void *arg) { int i; struct getaddrinfo_subrequest *req = arg; struct getaddrinfo_subrequest *other_req; struct evdns_getaddrinfo_request *data; struct evutil_addrinfo *res; struct sockaddr_in sin; struct sockaddr_in6 sin6; struct sockaddr *sa; int socklen, addrlen; void *addrp; int err; int user_canceled; EVUTIL_ASSERT(req->type == DNS_IPv4_A || req->type == DNS_IPv6_AAAA); if (req->type == DNS_IPv4_A) { data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv4_request); other_req = &data->ipv6_request; } else { data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv6_request); other_req = &data->ipv4_request; } /** Called from evdns_base_free() with @fail_requests == 1 */ if (result != DNS_ERR_SHUTDOWN) { EVDNS_LOCK(data->evdns_base); if (evdns_result_is_answer(result)) { if (req->type == DNS_IPv4_A) ++data->evdns_base->getaddrinfo_ipv4_answered; else ++data->evdns_base->getaddrinfo_ipv6_answered; } user_canceled = data->user_canceled; if (other_req->r == NULL) data->request_done = 1; EVDNS_UNLOCK(data->evdns_base); } else { data->evdns_base = NULL; user_canceled = data->user_canceled; } req->r = NULL; if (result == DNS_ERR_CANCEL && ! user_canceled) { /* Internal cancel request from timeout or internal error. * we already answered the user. */ if (other_req->r == NULL) free_getaddrinfo_request(data); return; } if (data->user_cb == NULL) { /* We already answered. XXXX This shouldn't be needed; see * comments in evdns_getaddrinfo_timeout_cb */ free_getaddrinfo_request(data); return; } if (result == DNS_ERR_NONE) { if (count == 0) err = EVUTIL_EAI_NODATA; else err = 0; } else { err = evdns_err_to_getaddrinfo_err(result); } if (err) { /* Looks like we got an error. */ if (other_req->r) { /* The other request is still working; maybe it will * succeed. */ /* XXXX handle failure from set_timeout */ if (result != DNS_ERR_SHUTDOWN) { evdns_getaddrinfo_set_timeout(data->evdns_base, data); } data->pending_error = err; return; } if (user_canceled) { data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data); } else if (data->pending_result) { /* If we have an answer waiting, and we weren't * canceled, ignore this error. */ add_cname_to_reply(data, data->pending_result); data->user_cb(0, data->pending_result, data->user_data); data->pending_result = NULL; } else { if (data->pending_error) err = getaddrinfo_merge_err(err, data->pending_error); data->user_cb(err, NULL, data->user_data); } free_getaddrinfo_request(data); return; } else if (user_canceled) { if (other_req->r) { /* The other request is still working; let it hit this * callback with EVUTIL_EAI_CANCEL callback and report * the failure. */ return; } data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data); free_getaddrinfo_request(data); return; } /* Looks like we got some answers. We should turn them into addrinfos * and then either queue those or return them all. */ EVUTIL_ASSERT(type == DNS_IPv4_A || type == DNS_IPv6_AAAA); if (type == DNS_IPv4_A) { memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_port = htons(data->port); sa = (struct sockaddr *)&sin; socklen = sizeof(sin); addrlen = 4; addrp = &sin.sin_addr.s_addr; } else { memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(data->port); sa = (struct sockaddr *)&sin6; socklen = sizeof(sin6); addrlen = 16; addrp = &sin6.sin6_addr.s6_addr; } res = NULL; for (i=0; i < count; ++i) { struct evutil_addrinfo *ai; memcpy(addrp, ((char*)addresses)+i*addrlen, addrlen); ai = evutil_new_addrinfo_(sa, socklen, &data->hints); if (!ai) { if (other_req->r) { evdns_cancel_request(NULL, other_req->r); } data->user_cb(EVUTIL_EAI_MEMORY, NULL, data->user_data); if (res) evutil_freeaddrinfo(res); if (other_req->r == NULL) free_getaddrinfo_request(data); return; } res = evutil_addrinfo_append_(res, ai); } if (other_req->r) { /* The other request is still in progress; wait for it */ /* XXXX handle failure from set_timeout */ evdns_getaddrinfo_set_timeout(data->evdns_base, data); data->pending_result = res; return; } else { /* The other request is done or never started; append its * results (if any) and return them. */ if (data->pending_result) { if (req->type == DNS_IPv4_A) res = evutil_addrinfo_append_(res, data->pending_result); else res = evutil_addrinfo_append_( data->pending_result, res); data->pending_result = NULL; } /* Call the user callback. */ add_cname_to_reply(data, res); data->user_cb(0, res, data->user_data); /* Free data. */ free_getaddrinfo_request(data); } } static struct hosts_entry * find_hosts_entry(struct evdns_base *base, const char *hostname, struct hosts_entry *find_after) { struct hosts_entry *e; if (find_after) e = TAILQ_NEXT(find_after, next); else e = TAILQ_FIRST(&base->hostsdb); for (; e; e = TAILQ_NEXT(e, next)) { if (!evutil_ascii_strcasecmp(e->hostname, hostname)) return e; } return NULL; } static int evdns_getaddrinfo_fromhosts(struct evdns_base *base, const char *nodename, struct evutil_addrinfo *hints, ev_uint16_t port, struct evutil_addrinfo **res) { int n_found = 0; struct hosts_entry *e; struct evutil_addrinfo *ai=NULL; int f = hints->ai_family; EVDNS_LOCK(base); for (e = find_hosts_entry(base, nodename, NULL); e; e = find_hosts_entry(base, nodename, e)) { struct evutil_addrinfo *ai_new; ++n_found; if ((e->addr.sa.sa_family == AF_INET && f == PF_INET6) || (e->addr.sa.sa_family == AF_INET6 && f == PF_INET)) continue; ai_new = evutil_new_addrinfo_(&e->addr.sa, e->addrlen, hints); if (!ai_new) { n_found = 0; goto out; } sockaddr_setport(ai_new->ai_addr, port); ai = evutil_addrinfo_append_(ai, ai_new); } EVDNS_UNLOCK(base); out: if (n_found) { /* Note that we return an empty answer if we found entries for * this hostname but none were of the right address type. */ *res = ai; return 0; } else { if (ai) evutil_freeaddrinfo(ai); return -1; } } struct evdns_getaddrinfo_request * evdns_getaddrinfo(struct evdns_base *dns_base, const char *nodename, const char *servname, const struct evutil_addrinfo *hints_in, evdns_getaddrinfo_cb cb, void *arg) { struct evdns_getaddrinfo_request *data; struct evutil_addrinfo hints; struct evutil_addrinfo *res = NULL; int err; int port = 0; int want_cname = 0; int started = 0; if (!dns_base) { dns_base = current_base; if (!dns_base) { log(EVDNS_LOG_WARN, "Call to getaddrinfo_async with no " "evdns_base configured."); cb(EVUTIL_EAI_FAIL, NULL, arg); /* ??? better error? */ return NULL; } } /* If we _must_ answer this immediately, do so. */ if ((hints_in && (hints_in->ai_flags & EVUTIL_AI_NUMERICHOST))) { res = NULL; err = evutil_getaddrinfo(nodename, servname, hints_in, &res); cb(err, res, arg); return NULL; } if (hints_in) { memcpy(&hints, hints_in, sizeof(hints)); } else { memset(&hints, 0, sizeof(hints)); hints.ai_family = PF_UNSPEC; } evutil_adjust_hints_for_addrconfig_(&hints); /* Now try to see if we _can_ answer immediately. */ /* (It would be nice to do this by calling getaddrinfo directly, with * AI_NUMERICHOST, on plaforms that have it, but we can't: there isn't * a reliable way to distinguish the "that wasn't a numeric host!" case * from any other EAI_NONAME cases.) */ err = evutil_getaddrinfo_common_(nodename, servname, &hints, &res, &port); if (err != EVUTIL_EAI_NEED_RESOLVE) { cb(err, res, arg); return NULL; } /* If there is an entry in the hosts file, we should give it now. */ if (!evdns_getaddrinfo_fromhosts(dns_base, nodename, &hints, port, &res)) { cb(0, res, arg); return NULL; } /* Okay, things are serious now. We're going to need to actually * launch a request. */ data = mm_calloc(1,sizeof(struct evdns_getaddrinfo_request)); if (!data) { cb(EVUTIL_EAI_MEMORY, NULL, arg); return NULL; } memcpy(&data->hints, &hints, sizeof(data->hints)); data->port = (ev_uint16_t)port; data->ipv4_request.type = DNS_IPv4_A; data->ipv6_request.type = DNS_IPv6_AAAA; data->user_cb = cb; data->user_data = arg; data->evdns_base = dns_base; want_cname = (hints.ai_flags & EVUTIL_AI_CANONNAME); /* If we are asked for a PF_UNSPEC address, we launch two requests in * parallel: one for an A address and one for an AAAA address. We * can't send just one request, since many servers only answer one * question per DNS request. * * Once we have the answer to one request, we allow for a short * timeout before we report it, to see if the other one arrives. If * they both show up in time, then we report both the answers. * * If too many addresses of one type time out or fail, we should stop * launching those requests. (XXX we don't do that yet.) */ EVDNS_LOCK(dns_base); if (hints.ai_family != PF_INET6) { log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv4 as %p", nodename, &data->ipv4_request); data->ipv4_request.r = evdns_base_resolve_ipv4(dns_base, nodename, 0, evdns_getaddrinfo_gotresolve, &data->ipv4_request); if (want_cname && data->ipv4_request.r) data->ipv4_request.r->current_req->put_cname_in_ptr = &data->cname_result; } if (hints.ai_family != PF_INET) { log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv6 as %p", nodename, &data->ipv6_request); data->ipv6_request.r = evdns_base_resolve_ipv6(dns_base, nodename, 0, evdns_getaddrinfo_gotresolve, &data->ipv6_request); if (want_cname && data->ipv6_request.r) data->ipv6_request.r->current_req->put_cname_in_ptr = &data->cname_result; } evtimer_assign(&data->timeout, dns_base->event_base, evdns_getaddrinfo_timeout_cb, data); started = (data->ipv4_request.r || data->ipv6_request.r); EVDNS_UNLOCK(dns_base); if (started) { return data; } else { mm_free(data); cb(EVUTIL_EAI_FAIL, NULL, arg); return NULL; } } void evdns_getaddrinfo_cancel(struct evdns_getaddrinfo_request *data) { EVDNS_LOCK(data->evdns_base); if (data->request_done) { EVDNS_UNLOCK(data->evdns_base); return; } event_del(&data->timeout); data->user_canceled = 1; if (data->ipv4_request.r) evdns_cancel_request(data->evdns_base, data->ipv4_request.r); if (data->ipv6_request.r) evdns_cancel_request(data->evdns_base, data->ipv6_request.r); EVDNS_UNLOCK(data->evdns_base); } /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #endif #include #if !defined(_WIN32) && defined(EVENT__HAVE_SYS_TIME_H) #include #endif #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #include #include #ifdef EVENT__HAVE_UNISTD_H #include #endif #include #include #include #include #include #include #ifdef EVENT__HAVE_FCNTL_H #include #endif #include "event2/event.h" #include "event2/event_struct.h" #include "event2/event_compat.h" #include "event-internal.h" #include "defer-internal.h" #include "evthread-internal.h" #include "event2/thread.h" #include "event2/util.h" #include "log-internal.h" #include "evmap-internal.h" #include "iocp-internal.h" #include "changelist-internal.h" #define HT_NO_CACHE_HASH_VALUES #include "ht-internal.h" #include "util-internal.h" #ifdef EVENT__HAVE_WORKING_KQUEUE #include "kqueue-internal.h" #endif #ifdef EVENT__HAVE_EVENT_PORTS extern const struct eventop evportops; #endif #ifdef EVENT__HAVE_SELECT extern const struct eventop selectops; #endif #ifdef EVENT__HAVE_POLL extern const struct eventop pollops; #endif #ifdef EVENT__HAVE_EPOLL extern const struct eventop epollops; #endif #ifdef EVENT__HAVE_WORKING_KQUEUE extern const struct eventop kqops; #endif #ifdef EVENT__HAVE_DEVPOLL extern const struct eventop devpollops; #endif #ifdef _WIN32 extern const struct eventop win32ops; #endif /* Array of backends in order of preference. */ static const struct eventop *eventops[] = { #ifdef EVENT__HAVE_EVENT_PORTS &evportops, #endif #ifdef EVENT__HAVE_WORKING_KQUEUE &kqops, #endif #ifdef EVENT__HAVE_EPOLL &epollops, #endif #ifdef EVENT__HAVE_DEVPOLL &devpollops, #endif #ifdef EVENT__HAVE_POLL &pollops, #endif #ifdef EVENT__HAVE_SELECT &selectops, #endif #ifdef _WIN32 &win32ops, #endif NULL }; /* Global state; deprecated */ EVENT2_EXPORT_SYMBOL struct event_base *event_global_current_base_ = NULL; #define current_base event_global_current_base_ /* Global state */ static void *event_self_cbarg_ptr_ = NULL; /* Prototypes */ static void event_queue_insert_active(struct event_base *, struct event_callback *); static void event_queue_insert_active_later(struct event_base *, struct event_callback *); static void event_queue_insert_timeout(struct event_base *, struct event *); static void event_queue_insert_inserted(struct event_base *, struct event *); static void event_queue_remove_active(struct event_base *, struct event_callback *); static void event_queue_remove_active_later(struct event_base *, struct event_callback *); static void event_queue_remove_timeout(struct event_base *, struct event *); static void event_queue_remove_inserted(struct event_base *, struct event *); static void event_queue_make_later_events_active(struct event_base *base); static int evthread_make_base_notifiable_nolock_(struct event_base *base); static int event_del_(struct event *ev, int blocking); #ifdef USE_REINSERT_TIMEOUT /* This code seems buggy; only turn it on if we find out what the trouble is. */ static void event_queue_reinsert_timeout(struct event_base *,struct event *, int was_common, int is_common, int old_timeout_idx); #endif static int event_haveevents(struct event_base *); static int event_process_active(struct event_base *); static int timeout_next(struct event_base *, struct timeval **); static void timeout_process(struct event_base *); static inline void event_signal_closure(struct event_base *, struct event *ev); static inline void event_persist_closure(struct event_base *, struct event *ev); static int evthread_notify_base(struct event_base *base); static void insert_common_timeout_inorder(struct common_timeout_list *ctl, struct event *ev); #ifndef EVENT__DISABLE_DEBUG_MODE /* These functions implement a hashtable of which 'struct event *' structures * have been setup or added. We don't want to trust the content of the struct * event itself, since we're trying to work through cases where an event gets * clobbered or freed. Instead, we keep a hashtable indexed by the pointer. */ struct event_debug_entry { HT_ENTRY(event_debug_entry) node; const struct event *ptr; unsigned added : 1; }; static inline unsigned hash_debug_entry(const struct event_debug_entry *e) { /* We need to do this silliness to convince compilers that we * honestly mean to cast e->ptr to an integer, and discard any * part of it that doesn't fit in an unsigned. */ unsigned u = (unsigned) ((ev_uintptr_t) e->ptr); /* Our hashtable implementation is pretty sensitive to low bits, * and every struct event is over 64 bytes in size, so we can * just say >>6. */ return (u >> 6); } static inline int eq_debug_entry(const struct event_debug_entry *a, const struct event_debug_entry *b) { return a->ptr == b->ptr; } int event_debug_mode_on_ = 0; #if !defined(EVENT__DISABLE_THREAD_SUPPORT) && !defined(EVENT__DISABLE_DEBUG_MODE) /** * @brief debug mode variable which is set for any function/structure that needs * to be shared across threads (if thread support is enabled). * * When and if evthreads are initialized, this variable will be evaluated, * and if set to something other than zero, this means the evthread setup * functions were called out of order. * * See: "Locks and threading" in the documentation. */ int event_debug_created_threadable_ctx_ = 0; #endif /* Set if it's too late to enable event_debug_mode. */ static int event_debug_mode_too_late = 0; #ifndef EVENT__DISABLE_THREAD_SUPPORT static void *event_debug_map_lock_ = NULL; #endif static HT_HEAD(event_debug_map, event_debug_entry) global_debug_map = HT_INITIALIZER(); HT_PROTOTYPE(event_debug_map, event_debug_entry, node, hash_debug_entry, eq_debug_entry) HT_GENERATE(event_debug_map, event_debug_entry, node, hash_debug_entry, eq_debug_entry, 0.5, mm_malloc, mm_realloc, mm_free) /* record that ev is now setup (that is, ready for an add) */ static void event_debug_note_setup_(const struct event *ev) { struct event_debug_entry *dent, find; if (!event_debug_mode_on_) goto out; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_FIND(event_debug_map, &global_debug_map, &find); if (dent) { dent->added = 0; } else { dent = mm_malloc(sizeof(*dent)); if (!dent) event_err(1, "Out of memory in debugging code"); dent->ptr = ev; dent->added = 0; HT_INSERT(event_debug_map, &global_debug_map, dent); } EVLOCK_UNLOCK(event_debug_map_lock_, 0); out: event_debug_mode_too_late = 1; } /* record that ev is no longer setup */ static void event_debug_note_teardown_(const struct event *ev) { struct event_debug_entry *dent, find; if (!event_debug_mode_on_) goto out; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_REMOVE(event_debug_map, &global_debug_map, &find); if (dent) mm_free(dent); EVLOCK_UNLOCK(event_debug_map_lock_, 0); out: event_debug_mode_too_late = 1; } /* Macro: record that ev is now added */ static void event_debug_note_add_(const struct event *ev) { struct event_debug_entry *dent,find; if (!event_debug_mode_on_) goto out; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_FIND(event_debug_map, &global_debug_map, &find); if (dent) { dent->added = 1; } else { event_errx(EVENT_ERR_ABORT_, "%s: noting an add on a non-setup event %p" " (events: 0x%x, fd: "EV_SOCK_FMT ", flags: 0x%x)", __func__, ev, ev->ev_events, EV_SOCK_ARG(ev->ev_fd), ev->ev_flags); } EVLOCK_UNLOCK(event_debug_map_lock_, 0); out: event_debug_mode_too_late = 1; } /* record that ev is no longer added */ static void event_debug_note_del_(const struct event *ev) { struct event_debug_entry *dent, find; if (!event_debug_mode_on_) goto out; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_FIND(event_debug_map, &global_debug_map, &find); if (dent) { dent->added = 0; } else { event_errx(EVENT_ERR_ABORT_, "%s: noting a del on a non-setup event %p" " (events: 0x%x, fd: "EV_SOCK_FMT ", flags: 0x%x)", __func__, ev, ev->ev_events, EV_SOCK_ARG(ev->ev_fd), ev->ev_flags); } EVLOCK_UNLOCK(event_debug_map_lock_, 0); out: event_debug_mode_too_late = 1; } /* assert that ev is setup (i.e., okay to add or inspect) */ static void event_debug_assert_is_setup_(const struct event *ev) { struct event_debug_entry *dent, find; if (!event_debug_mode_on_) return; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_FIND(event_debug_map, &global_debug_map, &find); if (!dent) { event_errx(EVENT_ERR_ABORT_, "%s called on a non-initialized event %p" " (events: 0x%x, fd: "EV_SOCK_FMT ", flags: 0x%x)", __func__, ev, ev->ev_events, EV_SOCK_ARG(ev->ev_fd), ev->ev_flags); } EVLOCK_UNLOCK(event_debug_map_lock_, 0); } /* assert that ev is not added (i.e., okay to tear down or set up again) */ static void event_debug_assert_not_added_(const struct event *ev) { struct event_debug_entry *dent, find; if (!event_debug_mode_on_) return; find.ptr = ev; EVLOCK_LOCK(event_debug_map_lock_, 0); dent = HT_FIND(event_debug_map, &global_debug_map, &find); if (dent && dent->added) { event_errx(EVENT_ERR_ABORT_, "%s called on an already added event %p" " (events: 0x%x, fd: "EV_SOCK_FMT", " "flags: 0x%x)", __func__, ev, ev->ev_events, EV_SOCK_ARG(ev->ev_fd), ev->ev_flags); } EVLOCK_UNLOCK(event_debug_map_lock_, 0); } static void event_debug_assert_socket_nonblocking_(evutil_socket_t fd) { if (!event_debug_mode_on_) return; if (fd < 0) return; #ifndef _WIN32 { int flags; if ((flags = fcntl(fd, F_GETFL, NULL)) >= 0) { EVUTIL_ASSERT(flags & O_NONBLOCK); } } #endif } #else static void event_debug_note_setup_(const struct event *ev) { (void)ev; } static void event_debug_note_teardown_(const struct event *ev) { (void)ev; } static void event_debug_note_add_(const struct event *ev) { (void)ev; } static void event_debug_note_del_(const struct event *ev) { (void)ev; } static void event_debug_assert_is_setup_(const struct event *ev) { (void)ev; } static void event_debug_assert_not_added_(const struct event *ev) { (void)ev; } static void event_debug_assert_socket_nonblocking_(evutil_socket_t fd) { (void)fd; } #endif #define EVENT_BASE_ASSERT_LOCKED(base) \ EVLOCK_ASSERT_LOCKED((base)->th_base_lock) /* How often (in seconds) do we check for changes in wall clock time relative * to monotonic time? Set this to -1 for 'never.' */ #define CLOCK_SYNC_INTERVAL 5 /** Set 'tp' to the current time according to 'base'. We must hold the lock * on 'base'. If there is a cached time, return it. Otherwise, use * clock_gettime or gettimeofday as appropriate to find out the right time. * Return 0 on success, -1 on failure. */ static int gettime(struct event_base *base, struct timeval *tp) { EVENT_BASE_ASSERT_LOCKED(base); if (base->tv_cache.tv_sec) { *tp = base->tv_cache; return (0); } if (evutil_gettime_monotonic_(&base->monotonic_timer, tp) == -1) { return -1; } if (base->last_updated_clock_diff + CLOCK_SYNC_INTERVAL < tp->tv_sec) { struct timeval tv; evutil_gettimeofday(&tv,NULL); evutil_timersub(&tv, tp, &base->tv_clock_diff); base->last_updated_clock_diff = tp->tv_sec; } return 0; } int event_base_gettimeofday_cached(struct event_base *base, struct timeval *tv) { int r; if (!base) { base = current_base; if (!current_base) return evutil_gettimeofday(tv, NULL); } EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (base->tv_cache.tv_sec == 0) { r = evutil_gettimeofday(tv, NULL); } else { evutil_timeradd(&base->tv_cache, &base->tv_clock_diff, tv); r = 0; } EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } /** Make 'base' have no current cached time. */ static inline void clear_time_cache(struct event_base *base) { base->tv_cache.tv_sec = 0; } /** Replace the cached time in 'base' with the current time. */ static inline void update_time_cache(struct event_base *base) { base->tv_cache.tv_sec = 0; if (!(base->flags & EVENT_BASE_FLAG_NO_CACHE_TIME)) gettime(base, &base->tv_cache); } int event_base_update_cache_time(struct event_base *base) { if (!base) { base = current_base; if (!current_base) return -1; } EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (base->running_loop) update_time_cache(base); EVBASE_RELEASE_LOCK(base, th_base_lock); return 0; } static inline struct event * event_callback_to_event(struct event_callback *evcb) { EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_INIT)); return EVUTIL_UPCAST(evcb, struct event, ev_evcallback); } static inline struct event_callback * event_to_event_callback(struct event *ev) { return &ev->ev_evcallback; } struct event_base * event_init(void) { struct event_base *base = event_base_new_with_config(NULL); if (base == NULL) { event_errx(1, "%s: Unable to construct event_base", __func__); return NULL; } current_base = base; return (base); } struct event_base * event_base_new(void) { struct event_base *base = NULL; struct event_config *cfg = event_config_new(); if (cfg) { base = event_base_new_with_config(cfg); event_config_free(cfg); } return base; } /** Return true iff 'method' is the name of a method that 'cfg' tells us to * avoid. */ static int event_config_is_avoided_method(const struct event_config *cfg, const char *method) { struct event_config_entry *entry; TAILQ_FOREACH(entry, &cfg->entries, next) { if (entry->avoid_method != NULL && strcmp(entry->avoid_method, method) == 0) return (1); } return (0); } /** Return true iff 'method' is disabled according to the environment. */ static int event_is_method_disabled(const char *name) { char environment[64]; int i; evutil_snprintf(environment, sizeof(environment), "EVENT_NO%s", name); for (i = 8; environment[i] != '\0'; ++i) environment[i] = EVUTIL_TOUPPER_(environment[i]); /* Note that evutil_getenv_() ignores the environment entirely if * we're setuid */ return (evutil_getenv_(environment) != NULL); } int event_base_get_features(const struct event_base *base) { return base->evsel->features; } void event_enable_debug_mode(void) { #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_mode_on_) event_errx(1, "%s was called twice!", __func__); if (event_debug_mode_too_late) event_errx(1, "%s must be called *before* creating any events " "or event_bases",__func__); event_debug_mode_on_ = 1; HT_INIT(event_debug_map, &global_debug_map); #endif } void event_disable_debug_mode(void) { #ifndef EVENT__DISABLE_DEBUG_MODE struct event_debug_entry **ent, *victim; EVLOCK_LOCK(event_debug_map_lock_, 0); for (ent = HT_START(event_debug_map, &global_debug_map); ent; ) { victim = *ent; ent = HT_NEXT_RMV(event_debug_map, &global_debug_map, ent); mm_free(victim); } HT_CLEAR(event_debug_map, &global_debug_map); EVLOCK_UNLOCK(event_debug_map_lock_ , 0); event_debug_mode_on_ = 0; #endif } struct event_base * event_base_new_with_config(const struct event_config *cfg) { int i; struct event_base *base; int should_check_environment; #ifndef EVENT__DISABLE_DEBUG_MODE event_debug_mode_too_late = 1; #endif if ((base = mm_calloc(1, sizeof(struct event_base))) == NULL) { event_warn("%s: calloc", __func__); return NULL; } if (cfg) base->flags = cfg->flags; should_check_environment = !(cfg && (cfg->flags & EVENT_BASE_FLAG_IGNORE_ENV)); { struct timeval tmp; int precise_time = cfg && (cfg->flags & EVENT_BASE_FLAG_PRECISE_TIMER); int flags; if (should_check_environment && !precise_time) { precise_time = evutil_getenv_("EVENT_PRECISE_TIMER") != NULL; if (precise_time) { base->flags |= EVENT_BASE_FLAG_PRECISE_TIMER; } } flags = precise_time ? EV_MONOT_PRECISE : 0; evutil_configure_monotonic_time_(&base->monotonic_timer, flags); gettime(base, &tmp); } min_heap_ctor_(&base->timeheap); base->sig.ev_signal_pair[0] = -1; base->sig.ev_signal_pair[1] = -1; base->th_notify_fd[0] = -1; base->th_notify_fd[1] = -1; TAILQ_INIT(&base->active_later_queue); evmap_io_initmap_(&base->io); evmap_signal_initmap_(&base->sigmap); event_changelist_init_(&base->changelist); base->evbase = NULL; if (cfg) { memcpy(&base->max_dispatch_time, &cfg->max_dispatch_interval, sizeof(struct timeval)); base->limit_callbacks_after_prio = cfg->limit_callbacks_after_prio; } else { base->max_dispatch_time.tv_sec = -1; base->limit_callbacks_after_prio = 1; } if (cfg && cfg->max_dispatch_callbacks >= 0) { base->max_dispatch_callbacks = cfg->max_dispatch_callbacks; } else { base->max_dispatch_callbacks = INT_MAX; } if (base->max_dispatch_callbacks == INT_MAX && base->max_dispatch_time.tv_sec == -1) base->limit_callbacks_after_prio = INT_MAX; for (i = 0; eventops[i] && !base->evbase; i++) { if (cfg != NULL) { /* determine if this backend should be avoided */ if (event_config_is_avoided_method(cfg, eventops[i]->name)) continue; if ((eventops[i]->features & cfg->require_features) != cfg->require_features) continue; } /* also obey the environment variables */ if (should_check_environment && event_is_method_disabled(eventops[i]->name)) continue; base->evsel = eventops[i]; base->evbase = base->evsel->init(base); } if (base->evbase == NULL) { event_warnx("%s: no event mechanism available", __func__); base->evsel = NULL; event_base_free(base); return NULL; } if (evutil_getenv_("EVENT_SHOW_METHOD")) event_msgx("libevent using: %s", base->evsel->name); /* allocate a single active event queue */ if (event_base_priority_init(base, 1) < 0) { event_base_free(base); return NULL; } /* prepare for threading */ #if !defined(EVENT__DISABLE_THREAD_SUPPORT) && !defined(EVENT__DISABLE_DEBUG_MODE) event_debug_created_threadable_ctx_ = 1; #endif #ifndef EVENT__DISABLE_THREAD_SUPPORT if (EVTHREAD_LOCKING_ENABLED() && (!cfg || !(cfg->flags & EVENT_BASE_FLAG_NOLOCK))) { int r; EVTHREAD_ALLOC_LOCK(base->th_base_lock, 0); EVTHREAD_ALLOC_COND(base->current_event_cond); r = evthread_make_base_notifiable(base); if (r<0) { event_warnx("%s: Unable to make base notifiable.", __func__); event_base_free(base); return NULL; } } #endif #ifdef _WIN32 if (cfg && (cfg->flags & EVENT_BASE_FLAG_STARTUP_IOCP)) event_base_start_iocp_(base, cfg->n_cpus_hint); #endif return (base); } int event_base_start_iocp_(struct event_base *base, int n_cpus) { #ifdef _WIN32 if (base->iocp) return 0; base->iocp = event_iocp_port_launch_(n_cpus); if (!base->iocp) { event_warnx("%s: Couldn't launch IOCP", __func__); return -1; } return 0; #else return -1; #endif } void event_base_stop_iocp_(struct event_base *base) { #ifdef _WIN32 int rv; if (!base->iocp) return; rv = event_iocp_shutdown_(base->iocp, -1); EVUTIL_ASSERT(rv >= 0); base->iocp = NULL; #endif } static int event_base_cancel_single_callback_(struct event_base *base, struct event_callback *evcb, int run_finalizers) { int result = 0; if (evcb->evcb_flags & EVLIST_INIT) { struct event *ev = event_callback_to_event(evcb); if (!(ev->ev_flags & EVLIST_INTERNAL)) { event_del_(ev, EVENT_DEL_EVEN_IF_FINALIZING); result = 1; } } else { EVBASE_ACQUIRE_LOCK(base, th_base_lock); event_callback_cancel_nolock_(base, evcb, 1); EVBASE_RELEASE_LOCK(base, th_base_lock); result = 1; } if (run_finalizers && (evcb->evcb_flags & EVLIST_FINALIZING)) { switch (evcb->evcb_closure) { case EV_CLOSURE_EVENT_FINALIZE: case EV_CLOSURE_EVENT_FINALIZE_FREE: { struct event *ev = event_callback_to_event(evcb); ev->ev_evcallback.evcb_cb_union.evcb_evfinalize(ev, ev->ev_arg); if (evcb->evcb_closure == EV_CLOSURE_EVENT_FINALIZE_FREE) mm_free(ev); break; } case EV_CLOSURE_CB_FINALIZE: evcb->evcb_cb_union.evcb_cbfinalize(evcb, evcb->evcb_arg); break; default: break; } } return result; } static int event_base_free_queues_(struct event_base *base, int run_finalizers) { int deleted = 0, i; for (i = 0; i < base->nactivequeues; ++i) { struct event_callback *evcb, *next; for (evcb = TAILQ_FIRST(&base->activequeues[i]); evcb; ) { next = TAILQ_NEXT(evcb, evcb_active_next); deleted += event_base_cancel_single_callback_(base, evcb, run_finalizers); evcb = next; } } { struct event_callback *evcb; while ((evcb = TAILQ_FIRST(&base->active_later_queue))) { deleted += event_base_cancel_single_callback_(base, evcb, run_finalizers); } } return deleted; } static void event_base_free_(struct event_base *base, int run_finalizers) { int i, n_deleted=0; struct event *ev; /* XXXX grab the lock? If there is contention when one thread frees * the base, then the contending thread will be very sad soon. */ /* event_base_free(NULL) is how to free the current_base if we * made it with event_init and forgot to hold a reference to it. */ if (base == NULL && current_base) base = current_base; /* Don't actually free NULL. */ if (base == NULL) { event_warnx("%s: no base to free", __func__); return; } /* XXX(niels) - check for internal events first */ #ifdef _WIN32 event_base_stop_iocp_(base); #endif /* threading fds if we have them */ if (base->th_notify_fd[0] != -1) { event_del(&base->th_notify); EVUTIL_CLOSESOCKET(base->th_notify_fd[0]); if (base->th_notify_fd[1] != -1) EVUTIL_CLOSESOCKET(base->th_notify_fd[1]); base->th_notify_fd[0] = -1; base->th_notify_fd[1] = -1; event_debug_unassign(&base->th_notify); } /* Delete all non-internal events. */ evmap_delete_all_(base); while ((ev = min_heap_top_(&base->timeheap)) != NULL) { event_del(ev); ++n_deleted; } for (i = 0; i < base->n_common_timeouts; ++i) { struct common_timeout_list *ctl = base->common_timeout_queues[i]; event_del(&ctl->timeout_event); /* Internal; doesn't count */ event_debug_unassign(&ctl->timeout_event); for (ev = TAILQ_FIRST(&ctl->events); ev; ) { struct event *next = TAILQ_NEXT(ev, ev_timeout_pos.ev_next_with_common_timeout); if (!(ev->ev_flags & EVLIST_INTERNAL)) { event_del(ev); ++n_deleted; } ev = next; } mm_free(ctl); } if (base->common_timeout_queues) mm_free(base->common_timeout_queues); for (;;) { /* For finalizers we can register yet another finalizer out from * finalizer, and iff finalizer will be in active_later_queue we can * add finalizer to activequeues, and we will have events in * activequeues after this function returns, which is not what we want * (we even have an assertion for this). * * A simple case is bufferevent with underlying (i.e. filters). */ int i = event_base_free_queues_(base, run_finalizers); event_debug(("%s: %d events freed", __func__, i)); if (!i) { break; } n_deleted += i; } if (n_deleted) event_debug(("%s: %d events were still set in base", __func__, n_deleted)); while (LIST_FIRST(&base->once_events)) { struct event_once *eonce = LIST_FIRST(&base->once_events); LIST_REMOVE(eonce, next_once); mm_free(eonce); } if (base->evsel != NULL && base->evsel->dealloc != NULL) base->evsel->dealloc(base); for (i = 0; i < base->nactivequeues; ++i) EVUTIL_ASSERT(TAILQ_EMPTY(&base->activequeues[i])); EVUTIL_ASSERT(min_heap_empty_(&base->timeheap)); min_heap_dtor_(&base->timeheap); mm_free(base->activequeues); evmap_io_clear_(&base->io); evmap_signal_clear_(&base->sigmap); event_changelist_freemem_(&base->changelist); EVTHREAD_FREE_LOCK(base->th_base_lock, 0); EVTHREAD_FREE_COND(base->current_event_cond); /* If we're freeing current_base, there won't be a current_base. */ if (base == current_base) current_base = NULL; mm_free(base); } void event_base_free_nofinalize(struct event_base *base) { event_base_free_(base, 0); } void event_base_free(struct event_base *base) { event_base_free_(base, 1); } /* Fake eventop; used to disable the backend temporarily inside event_reinit * so that we can call event_del() on an event without telling the backend. */ static int nil_backend_del(struct event_base *b, evutil_socket_t fd, short old, short events, void *fdinfo) { return 0; } const struct eventop nil_eventop = { "nil", NULL, /* init: unused. */ NULL, /* add: unused. */ nil_backend_del, /* del: used, so needs to be killed. */ NULL, /* dispatch: unused. */ NULL, /* dealloc: unused. */ 0, 0, 0 }; /* reinitialize the event base after a fork */ int event_reinit(struct event_base *base) { const struct eventop *evsel; int res = 0; int was_notifiable = 0; int had_signal_added = 0; EVBASE_ACQUIRE_LOCK(base, th_base_lock); evsel = base->evsel; /* check if this event mechanism requires reinit on the backend */ if (evsel->need_reinit) { /* We're going to call event_del() on our notify events (the * ones that tell about signals and wakeup events). But we * don't actually want to tell the backend to change its * state, since it might still share some resource (a kqueue, * an epoll fd) with the parent process, and we don't want to * delete the fds from _that_ backend, we temporarily stub out * the evsel with a replacement. */ base->evsel = &nil_eventop; } /* We need to re-create a new signal-notification fd and a new * thread-notification fd. Otherwise, we'll still share those with * the parent process, which would make any notification sent to them * get received by one or both of the event loops, more or less at * random. */ if (base->sig.ev_signal_added) { event_del_nolock_(&base->sig.ev_signal, EVENT_DEL_AUTOBLOCK); event_debug_unassign(&base->sig.ev_signal); memset(&base->sig.ev_signal, 0, sizeof(base->sig.ev_signal)); had_signal_added = 1; base->sig.ev_signal_added = 0; } if (base->sig.ev_signal_pair[0] != -1) EVUTIL_CLOSESOCKET(base->sig.ev_signal_pair[0]); if (base->sig.ev_signal_pair[1] != -1) EVUTIL_CLOSESOCKET(base->sig.ev_signal_pair[1]); if (base->th_notify_fn != NULL) { was_notifiable = 1; base->th_notify_fn = NULL; } if (base->th_notify_fd[0] != -1) { event_del_nolock_(&base->th_notify, EVENT_DEL_AUTOBLOCK); EVUTIL_CLOSESOCKET(base->th_notify_fd[0]); if (base->th_notify_fd[1] != -1) EVUTIL_CLOSESOCKET(base->th_notify_fd[1]); base->th_notify_fd[0] = -1; base->th_notify_fd[1] = -1; event_debug_unassign(&base->th_notify); } /* Replace the original evsel. */ base->evsel = evsel; if (evsel->need_reinit) { /* Reconstruct the backend through brute-force, so that we do * not share any structures with the parent process. For some * backends, this is necessary: epoll and kqueue, for * instance, have events associated with a kernel * structure. If didn't reinitialize, we'd share that * structure with the parent process, and any changes made by * the parent would affect our backend's behavior (and vice * versa). */ if (base->evsel->dealloc != NULL) base->evsel->dealloc(base); base->evbase = evsel->init(base); if (base->evbase == NULL) { event_errx(1, "%s: could not reinitialize event mechanism", __func__); res = -1; goto done; } /* Empty out the changelist (if any): we are starting from a * blank slate. */ event_changelist_freemem_(&base->changelist); /* Tell the event maps to re-inform the backend about all * pending events. This will make the signal notification * event get re-created if necessary. */ if (evmap_reinit_(base) < 0) res = -1; } else { res = evsig_init_(base); if (res == 0 && had_signal_added) { res = event_add_nolock_(&base->sig.ev_signal, NULL, 0); if (res == 0) base->sig.ev_signal_added = 1; } } /* If we were notifiable before, and nothing just exploded, become * notifiable again. */ if (was_notifiable && res == 0) res = evthread_make_base_notifiable_nolock_(base); done: EVBASE_RELEASE_LOCK(base, th_base_lock); return (res); } /* Get the monotonic time for this event_base' timer */ int event_gettime_monotonic(struct event_base *base, struct timeval *tv) { int rv = -1; if (base && tv) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); rv = evutil_gettime_monotonic_(&(base->monotonic_timer), tv); EVBASE_RELEASE_LOCK(base, th_base_lock); } return rv; } const char ** event_get_supported_methods(void) { static const char **methods = NULL; const struct eventop **method; const char **tmp; int i = 0, k; /* count all methods */ for (method = &eventops[0]; *method != NULL; ++method) { ++i; } /* allocate one more than we need for the NULL pointer */ tmp = mm_calloc((i + 1), sizeof(char *)); if (tmp == NULL) return (NULL); /* populate the array with the supported methods */ for (k = 0, i = 0; eventops[k] != NULL; ++k) { tmp[i++] = eventops[k]->name; } tmp[i] = NULL; if (methods != NULL) mm_free((char**)methods); methods = tmp; return (methods); } struct event_config * event_config_new(void) { struct event_config *cfg = mm_calloc(1, sizeof(*cfg)); if (cfg == NULL) return (NULL); TAILQ_INIT(&cfg->entries); cfg->max_dispatch_interval.tv_sec = -1; cfg->max_dispatch_callbacks = INT_MAX; cfg->limit_callbacks_after_prio = 1; return (cfg); } static void event_config_entry_free(struct event_config_entry *entry) { if (entry->avoid_method != NULL) mm_free((char *)entry->avoid_method); mm_free(entry); } void event_config_free(struct event_config *cfg) { struct event_config_entry *entry; while ((entry = TAILQ_FIRST(&cfg->entries)) != NULL) { TAILQ_REMOVE(&cfg->entries, entry, next); event_config_entry_free(entry); } mm_free(cfg); } int event_config_set_flag(struct event_config *cfg, int flag) { if (!cfg) return -1; cfg->flags |= flag; return 0; } int event_config_avoid_method(struct event_config *cfg, const char *method) { struct event_config_entry *entry = mm_malloc(sizeof(*entry)); if (entry == NULL) return (-1); if ((entry->avoid_method = mm_strdup(method)) == NULL) { mm_free(entry); return (-1); } TAILQ_INSERT_TAIL(&cfg->entries, entry, next); return (0); } int event_config_require_features(struct event_config *cfg, int features) { if (!cfg) return (-1); cfg->require_features = features; return (0); } int event_config_set_num_cpus_hint(struct event_config *cfg, int cpus) { if (!cfg) return (-1); cfg->n_cpus_hint = cpus; return (0); } int event_config_set_max_dispatch_interval(struct event_config *cfg, const struct timeval *max_interval, int max_callbacks, int min_priority) { if (max_interval) memcpy(&cfg->max_dispatch_interval, max_interval, sizeof(struct timeval)); else cfg->max_dispatch_interval.tv_sec = -1; cfg->max_dispatch_callbacks = max_callbacks >= 0 ? max_callbacks : INT_MAX; if (min_priority < 0) min_priority = 0; cfg->limit_callbacks_after_prio = min_priority; return (0); } int event_priority_init(int npriorities) { return event_base_priority_init(current_base, npriorities); } int event_base_priority_init(struct event_base *base, int npriorities) { int i, r; r = -1; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (N_ACTIVE_CALLBACKS(base) || npriorities < 1 || npriorities >= EVENT_MAX_PRIORITIES) goto err; if (npriorities == base->nactivequeues) goto ok; if (base->nactivequeues) { mm_free(base->activequeues); base->nactivequeues = 0; } /* Allocate our priority queues */ base->activequeues = (struct evcallback_list *) mm_calloc(npriorities, sizeof(struct evcallback_list)); if (base->activequeues == NULL) { event_warn("%s: calloc", __func__); goto err; } base->nactivequeues = npriorities; for (i = 0; i < base->nactivequeues; ++i) { TAILQ_INIT(&base->activequeues[i]); } ok: r = 0; err: EVBASE_RELEASE_LOCK(base, th_base_lock); return (r); } int event_base_get_npriorities(struct event_base *base) { int n; if (base == NULL) base = current_base; EVBASE_ACQUIRE_LOCK(base, th_base_lock); n = base->nactivequeues; EVBASE_RELEASE_LOCK(base, th_base_lock); return (n); } int event_base_get_num_events(struct event_base *base, unsigned int type) { int r = 0; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (type & EVENT_BASE_COUNT_ACTIVE) r += base->event_count_active; if (type & EVENT_BASE_COUNT_VIRTUAL) r += base->virtual_event_count; if (type & EVENT_BASE_COUNT_ADDED) r += base->event_count; EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } int event_base_get_max_events(struct event_base *base, unsigned int type, int clear) { int r = 0; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (type & EVENT_BASE_COUNT_ACTIVE) { r += base->event_count_active_max; if (clear) base->event_count_active_max = 0; } if (type & EVENT_BASE_COUNT_VIRTUAL) { r += base->virtual_event_count_max; if (clear) base->virtual_event_count_max = 0; } if (type & EVENT_BASE_COUNT_ADDED) { r += base->event_count_max; if (clear) base->event_count_max = 0; } EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } /* Returns true iff we're currently watching any events. */ static int event_haveevents(struct event_base *base) { /* Caller must hold th_base_lock */ return (base->virtual_event_count > 0 || base->event_count > 0); } /* "closure" function called when processing active signal events */ static inline void event_signal_closure(struct event_base *base, struct event *ev) { short ncalls; int should_break; /* Allows deletes to work */ ncalls = ev->ev_ncalls; if (ncalls != 0) ev->ev_pncalls = &ncalls; EVBASE_RELEASE_LOCK(base, th_base_lock); while (ncalls) { ncalls--; ev->ev_ncalls = ncalls; if (ncalls == 0) ev->ev_pncalls = NULL; (*ev->ev_callback)(ev->ev_fd, ev->ev_res, ev->ev_arg); EVBASE_ACQUIRE_LOCK(base, th_base_lock); should_break = base->event_break; EVBASE_RELEASE_LOCK(base, th_base_lock); if (should_break) { if (ncalls != 0) ev->ev_pncalls = NULL; return; } } } /* Common timeouts are special timeouts that are handled as queues rather than * in the minheap. This is more efficient than the minheap if we happen to * know that we're going to get several thousands of timeout events all with * the same timeout value. * * Since all our timeout handling code assumes timevals can be copied, * assigned, etc, we can't use "magic pointer" to encode these common * timeouts. Searching through a list to see if every timeout is common could * also get inefficient. Instead, we take advantage of the fact that tv_usec * is 32 bits long, but only uses 20 of those bits (since it can never be over * 999999.) We use the top bits to encode 4 bites of magic number, and 8 bits * of index into the event_base's aray of common timeouts. */ #define MICROSECONDS_MASK COMMON_TIMEOUT_MICROSECONDS_MASK #define COMMON_TIMEOUT_IDX_MASK 0x0ff00000 #define COMMON_TIMEOUT_IDX_SHIFT 20 #define COMMON_TIMEOUT_MASK 0xf0000000 #define COMMON_TIMEOUT_MAGIC 0x50000000 #define COMMON_TIMEOUT_IDX(tv) \ (((tv)->tv_usec & COMMON_TIMEOUT_IDX_MASK)>>COMMON_TIMEOUT_IDX_SHIFT) /** Return true iff if 'tv' is a common timeout in 'base' */ static inline int is_common_timeout(const struct timeval *tv, const struct event_base *base) { int idx; if ((tv->tv_usec & COMMON_TIMEOUT_MASK) != COMMON_TIMEOUT_MAGIC) return 0; idx = COMMON_TIMEOUT_IDX(tv); return idx < base->n_common_timeouts; } /* True iff tv1 and tv2 have the same common-timeout index, or if neither * one is a common timeout. */ static inline int is_same_common_timeout(const struct timeval *tv1, const struct timeval *tv2) { return (tv1->tv_usec & ~MICROSECONDS_MASK) == (tv2->tv_usec & ~MICROSECONDS_MASK); } /** Requires that 'tv' is a common timeout. Return the corresponding * common_timeout_list. */ static inline struct common_timeout_list * get_common_timeout_list(struct event_base *base, const struct timeval *tv) { return base->common_timeout_queues[COMMON_TIMEOUT_IDX(tv)]; } #if 0 static inline int common_timeout_ok(const struct timeval *tv, struct event_base *base) { const struct timeval *expect = &get_common_timeout_list(base, tv)->duration; return tv->tv_sec == expect->tv_sec && tv->tv_usec == expect->tv_usec; } #endif /* Add the timeout for the first event in given common timeout list to the * event_base's minheap. */ static void common_timeout_schedule(struct common_timeout_list *ctl, const struct timeval *now, struct event *head) { struct timeval timeout = head->ev_timeout; timeout.tv_usec &= MICROSECONDS_MASK; event_add_nolock_(&ctl->timeout_event, &timeout, 1); } /* Callback: invoked when the timeout for a common timeout queue triggers. * This means that (at least) the first event in that queue should be run, * and the timeout should be rescheduled if there are more events. */ static void common_timeout_callback(evutil_socket_t fd, short what, void *arg) { struct timeval now; struct common_timeout_list *ctl = arg; struct event_base *base = ctl->base; struct event *ev = NULL; EVBASE_ACQUIRE_LOCK(base, th_base_lock); gettime(base, &now); while (1) { ev = TAILQ_FIRST(&ctl->events); if (!ev || ev->ev_timeout.tv_sec > now.tv_sec || (ev->ev_timeout.tv_sec == now.tv_sec && (ev->ev_timeout.tv_usec&MICROSECONDS_MASK) > now.tv_usec)) break; event_del_nolock_(ev, EVENT_DEL_NOBLOCK); event_active_nolock_(ev, EV_TIMEOUT, 1); } if (ev) common_timeout_schedule(ctl, &now, ev); EVBASE_RELEASE_LOCK(base, th_base_lock); } #define MAX_COMMON_TIMEOUTS 256 const struct timeval * event_base_init_common_timeout(struct event_base *base, const struct timeval *duration) { int i; struct timeval tv; const struct timeval *result=NULL; struct common_timeout_list *new_ctl; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (duration->tv_usec > 1000000) { memcpy(&tv, duration, sizeof(struct timeval)); if (is_common_timeout(duration, base)) tv.tv_usec &= MICROSECONDS_MASK; tv.tv_sec += tv.tv_usec / 1000000; tv.tv_usec %= 1000000; duration = &tv; } for (i = 0; i < base->n_common_timeouts; ++i) { const struct common_timeout_list *ctl = base->common_timeout_queues[i]; if (duration->tv_sec == ctl->duration.tv_sec && duration->tv_usec == (ctl->duration.tv_usec & MICROSECONDS_MASK)) { EVUTIL_ASSERT(is_common_timeout(&ctl->duration, base)); result = &ctl->duration; goto done; } } if (base->n_common_timeouts == MAX_COMMON_TIMEOUTS) { event_warnx("%s: Too many common timeouts already in use; " "we only support %d per event_base", __func__, MAX_COMMON_TIMEOUTS); goto done; } if (base->n_common_timeouts_allocated == base->n_common_timeouts) { int n = base->n_common_timeouts < 16 ? 16 : base->n_common_timeouts*2; struct common_timeout_list **newqueues = mm_realloc(base->common_timeout_queues, n*sizeof(struct common_timeout_queue *)); if (!newqueues) { event_warn("%s: realloc",__func__); goto done; } base->n_common_timeouts_allocated = n; base->common_timeout_queues = newqueues; } new_ctl = mm_calloc(1, sizeof(struct common_timeout_list)); if (!new_ctl) { event_warn("%s: calloc",__func__); goto done; } TAILQ_INIT(&new_ctl->events); new_ctl->duration.tv_sec = duration->tv_sec; new_ctl->duration.tv_usec = duration->tv_usec | COMMON_TIMEOUT_MAGIC | (base->n_common_timeouts << COMMON_TIMEOUT_IDX_SHIFT); evtimer_assign(&new_ctl->timeout_event, base, common_timeout_callback, new_ctl); new_ctl->timeout_event.ev_flags |= EVLIST_INTERNAL; event_priority_set(&new_ctl->timeout_event, 0); new_ctl->base = base; base->common_timeout_queues[base->n_common_timeouts++] = new_ctl; result = &new_ctl->duration; done: if (result) EVUTIL_ASSERT(is_common_timeout(result, base)); EVBASE_RELEASE_LOCK(base, th_base_lock); return result; } /* Closure function invoked when we're activating a persistent event. */ static inline void event_persist_closure(struct event_base *base, struct event *ev) { void (*evcb_callback)(evutil_socket_t, short, void *); // Other fields of *ev that must be stored before executing evutil_socket_t evcb_fd; short evcb_res; void *evcb_arg; /* reschedule the persistent event if we have a timeout. */ if (ev->ev_io_timeout.tv_sec || ev->ev_io_timeout.tv_usec) { /* If there was a timeout, we want it to run at an interval of * ev_io_timeout after the last time it was _scheduled_ for, * not ev_io_timeout after _now_. If it fired for another * reason, though, the timeout ought to start ticking _now_. */ struct timeval run_at, relative_to, delay, now; ev_uint32_t usec_mask = 0; EVUTIL_ASSERT(is_same_common_timeout(&ev->ev_timeout, &ev->ev_io_timeout)); gettime(base, &now); if (is_common_timeout(&ev->ev_timeout, base)) { delay = ev->ev_io_timeout; usec_mask = delay.tv_usec & ~MICROSECONDS_MASK; delay.tv_usec &= MICROSECONDS_MASK; if (ev->ev_res & EV_TIMEOUT) { relative_to = ev->ev_timeout; relative_to.tv_usec &= MICROSECONDS_MASK; } else { relative_to = now; } } else { delay = ev->ev_io_timeout; if (ev->ev_res & EV_TIMEOUT) { relative_to = ev->ev_timeout; } else { relative_to = now; } } evutil_timeradd(&relative_to, &delay, &run_at); if (evutil_timercmp(&run_at, &now, <)) { /* Looks like we missed at least one invocation due to * a clock jump, not running the event loop for a * while, really slow callbacks, or * something. Reschedule relative to now. */ evutil_timeradd(&now, &delay, &run_at); } run_at.tv_usec |= usec_mask; event_add_nolock_(ev, &run_at, 1); } // Save our callback before we release the lock evcb_callback = ev->ev_callback; evcb_fd = ev->ev_fd; evcb_res = ev->ev_res; evcb_arg = ev->ev_arg; // Release the lock EVBASE_RELEASE_LOCK(base, th_base_lock); // Execute the callback (evcb_callback)(evcb_fd, evcb_res, evcb_arg); } /* Helper for event_process_active to process all the events in a single queue, releasing the lock as we go. This function requires that the lock be held when it's invoked. Returns -1 if we get a signal or an event_break that means we should stop processing any active events now. Otherwise returns the number of non-internal event_callbacks that we processed. */ static int event_process_active_single_queue(struct event_base *base, struct evcallback_list *activeq, int max_to_process, const struct timeval *endtime) { struct event_callback *evcb; int count = 0; EVUTIL_ASSERT(activeq != NULL); for (evcb = TAILQ_FIRST(activeq); evcb; evcb = TAILQ_FIRST(activeq)) { struct event *ev=NULL; if (evcb->evcb_flags & EVLIST_INIT) { ev = event_callback_to_event(evcb); if (ev->ev_events & EV_PERSIST || ev->ev_flags & EVLIST_FINALIZING) event_queue_remove_active(base, evcb); else event_del_nolock_(ev, EVENT_DEL_NOBLOCK); event_debug(( "event_process_active: event: %p, %s%s%scall %p", ev, ev->ev_res & EV_READ ? "EV_READ " : " ", ev->ev_res & EV_WRITE ? "EV_WRITE " : " ", ev->ev_res & EV_CLOSED ? "EV_CLOSED " : " ", ev->ev_callback)); } else { event_queue_remove_active(base, evcb); event_debug(("event_process_active: event_callback %p, " "closure %d, call %p", evcb, evcb->evcb_closure, evcb->evcb_cb_union.evcb_callback)); } if (!(evcb->evcb_flags & EVLIST_INTERNAL)) ++count; base->current_event = evcb; #ifndef EVENT__DISABLE_THREAD_SUPPORT base->current_event_waiters = 0; #endif switch (evcb->evcb_closure) { case EV_CLOSURE_EVENT_SIGNAL: EVUTIL_ASSERT(ev != NULL); event_signal_closure(base, ev); break; case EV_CLOSURE_EVENT_PERSIST: EVUTIL_ASSERT(ev != NULL); event_persist_closure(base, ev); break; case EV_CLOSURE_EVENT: { void (*evcb_callback)(evutil_socket_t, short, void *); short res; EVUTIL_ASSERT(ev != NULL); evcb_callback = *ev->ev_callback; res = ev->ev_res; EVBASE_RELEASE_LOCK(base, th_base_lock); evcb_callback(ev->ev_fd, res, ev->ev_arg); } break; case EV_CLOSURE_CB_SELF: { void (*evcb_selfcb)(struct event_callback *, void *) = evcb->evcb_cb_union.evcb_selfcb; EVBASE_RELEASE_LOCK(base, th_base_lock); evcb_selfcb(evcb, evcb->evcb_arg); } break; case EV_CLOSURE_EVENT_FINALIZE: case EV_CLOSURE_EVENT_FINALIZE_FREE: { void (*evcb_evfinalize)(struct event *, void *); int evcb_closure = evcb->evcb_closure; EVUTIL_ASSERT(ev != NULL); base->current_event = NULL; evcb_evfinalize = ev->ev_evcallback.evcb_cb_union.evcb_evfinalize; EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_FINALIZING)); EVBASE_RELEASE_LOCK(base, th_base_lock); event_debug_note_teardown_(ev); evcb_evfinalize(ev, ev->ev_arg); if (evcb_closure == EV_CLOSURE_EVENT_FINALIZE_FREE) mm_free(ev); } break; case EV_CLOSURE_CB_FINALIZE: { void (*evcb_cbfinalize)(struct event_callback *, void *) = evcb->evcb_cb_union.evcb_cbfinalize; base->current_event = NULL; EVUTIL_ASSERT((evcb->evcb_flags & EVLIST_FINALIZING)); EVBASE_RELEASE_LOCK(base, th_base_lock); evcb_cbfinalize(evcb, evcb->evcb_arg); } break; default: EVUTIL_ASSERT(0); } EVBASE_ACQUIRE_LOCK(base, th_base_lock); base->current_event = NULL; #ifndef EVENT__DISABLE_THREAD_SUPPORT if (base->current_event_waiters) { base->current_event_waiters = 0; EVTHREAD_COND_BROADCAST(base->current_event_cond); } #endif if (base->event_break) return -1; if (count >= max_to_process) return count; if (count && endtime) { struct timeval now; update_time_cache(base); gettime(base, &now); if (evutil_timercmp(&now, endtime, >=)) return count; } if (base->event_continue) break; } return count; } /* * Active events are stored in priority queues. Lower priorities are always * process before higher priorities. Low priority events can starve high * priority ones. */ static int event_process_active(struct event_base *base) { /* Caller must hold th_base_lock */ struct evcallback_list *activeq = NULL; int i, c = 0; const struct timeval *endtime; struct timeval tv; const int maxcb = base->max_dispatch_callbacks; const int limit_after_prio = base->limit_callbacks_after_prio; if (base->max_dispatch_time.tv_sec >= 0) { update_time_cache(base); gettime(base, &tv); evutil_timeradd(&base->max_dispatch_time, &tv, &tv); endtime = &tv; } else { endtime = NULL; } for (i = 0; i < base->nactivequeues; ++i) { if (TAILQ_FIRST(&base->activequeues[i]) != NULL) { base->event_running_priority = i; activeq = &base->activequeues[i]; if (i < limit_after_prio) c = event_process_active_single_queue(base, activeq, INT_MAX, NULL); else c = event_process_active_single_queue(base, activeq, maxcb, endtime); if (c < 0) { goto done; } else if (c > 0) break; /* Processed a real event; do not * consider lower-priority events */ /* If we get here, all of the events we processed * were internal. Continue. */ } } done: base->event_running_priority = -1; return c; } /* * Wait continuously for events. We exit only if no events are left. */ int event_dispatch(void) { return (event_loop(0)); } int event_base_dispatch(struct event_base *event_base) { return (event_base_loop(event_base, 0)); } const char * event_base_get_method(const struct event_base *base) { EVUTIL_ASSERT(base); return (base->evsel->name); } /** Callback: used to implement event_base_loopexit by telling the event_base * that it's time to exit its loop. */ static void event_loopexit_cb(evutil_socket_t fd, short what, void *arg) { struct event_base *base = arg; base->event_gotterm = 1; } int event_loopexit(const struct timeval *tv) { return (event_once(-1, EV_TIMEOUT, event_loopexit_cb, current_base, tv)); } int event_base_loopexit(struct event_base *event_base, const struct timeval *tv) { return (event_base_once(event_base, -1, EV_TIMEOUT, event_loopexit_cb, event_base, tv)); } int event_loopbreak(void) { return (event_base_loopbreak(current_base)); } int event_base_loopbreak(struct event_base *event_base) { int r = 0; if (event_base == NULL) return (-1); EVBASE_ACQUIRE_LOCK(event_base, th_base_lock); event_base->event_break = 1; if (EVBASE_NEED_NOTIFY(event_base)) { r = evthread_notify_base(event_base); } else { r = (0); } EVBASE_RELEASE_LOCK(event_base, th_base_lock); return r; } int event_base_loopcontinue(struct event_base *event_base) { int r = 0; if (event_base == NULL) return (-1); EVBASE_ACQUIRE_LOCK(event_base, th_base_lock); event_base->event_continue = 1; if (EVBASE_NEED_NOTIFY(event_base)) { r = evthread_notify_base(event_base); } else { r = (0); } EVBASE_RELEASE_LOCK(event_base, th_base_lock); return r; } int event_base_got_break(struct event_base *event_base) { int res; EVBASE_ACQUIRE_LOCK(event_base, th_base_lock); res = event_base->event_break; EVBASE_RELEASE_LOCK(event_base, th_base_lock); return res; } int event_base_got_exit(struct event_base *event_base) { int res; EVBASE_ACQUIRE_LOCK(event_base, th_base_lock); res = event_base->event_gotterm; EVBASE_RELEASE_LOCK(event_base, th_base_lock); return res; } /* not thread safe */ int event_loop(int flags) { return event_base_loop(current_base, flags); } int event_base_loop(struct event_base *base, int flags) { const struct eventop *evsel = base->evsel; struct timeval tv; struct timeval *tv_p; int res, done, retval = 0; /* Grab the lock. We will release it inside evsel.dispatch, and again * as we invoke user callbacks. */ EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (base->running_loop) { event_warnx("%s: reentrant invocation. Only one event_base_loop" " can run on each event_base at once.", __func__); EVBASE_RELEASE_LOCK(base, th_base_lock); return -1; } base->running_loop = 1; clear_time_cache(base); if (base->sig.ev_signal_added && base->sig.ev_n_signals_added) evsig_set_base_(base); done = 0; #ifndef EVENT__DISABLE_THREAD_SUPPORT base->th_owner_id = EVTHREAD_GET_ID(); #endif base->event_gotterm = base->event_break = 0; while (!done) { base->event_continue = 0; base->n_deferreds_queued = 0; /* Terminate the loop if we have been asked to */ if (base->event_gotterm) { break; } if (base->event_break) { break; } tv_p = &tv; if (!N_ACTIVE_CALLBACKS(base) && !(flags & EVLOOP_NONBLOCK)) { timeout_next(base, &tv_p); } else { /* * if we have active events, we just poll new events * without waiting. */ evutil_timerclear(&tv); } /* If we have no events, we just exit */ if (0==(flags&EVLOOP_NO_EXIT_ON_EMPTY) && !event_haveevents(base) && !N_ACTIVE_CALLBACKS(base)) { event_debug(("%s: no events registered.", __func__)); retval = 1; goto done; } event_queue_make_later_events_active(base); clear_time_cache(base); res = evsel->dispatch(base, tv_p); if (res == -1) { event_debug(("%s: dispatch returned unsuccessfully.", __func__)); retval = -1; goto done; } update_time_cache(base); timeout_process(base); if (N_ACTIVE_CALLBACKS(base)) { int n = event_process_active(base); if ((flags & EVLOOP_ONCE) && N_ACTIVE_CALLBACKS(base) == 0 && n != 0) done = 1; } else if (flags & EVLOOP_NONBLOCK) done = 1; } event_debug(("%s: asked to terminate loop.", __func__)); done: clear_time_cache(base); base->running_loop = 0; EVBASE_RELEASE_LOCK(base, th_base_lock); return (retval); } /* One-time callback to implement event_base_once: invokes the user callback, * then deletes the allocated storage */ static void event_once_cb(evutil_socket_t fd, short events, void *arg) { struct event_once *eonce = arg; (*eonce->cb)(fd, events, eonce->arg); EVBASE_ACQUIRE_LOCK(eonce->ev.ev_base, th_base_lock); LIST_REMOVE(eonce, next_once); EVBASE_RELEASE_LOCK(eonce->ev.ev_base, th_base_lock); event_debug_unassign(&eonce->ev); mm_free(eonce); } /* not threadsafe, event scheduled once. */ int event_once(evutil_socket_t fd, short events, void (*callback)(evutil_socket_t, short, void *), void *arg, const struct timeval *tv) { return event_base_once(current_base, fd, events, callback, arg, tv); } /* Schedules an event once */ int event_base_once(struct event_base *base, evutil_socket_t fd, short events, void (*callback)(evutil_socket_t, short, void *), void *arg, const struct timeval *tv) { struct event_once *eonce; int res = 0; int activate = 0; if (!base) return (-1); /* We cannot support signals that just fire once, or persistent * events. */ if (events & (EV_SIGNAL|EV_PERSIST)) return (-1); if ((eonce = mm_calloc(1, sizeof(struct event_once))) == NULL) return (-1); eonce->cb = callback; eonce->arg = arg; if ((events & (EV_TIMEOUT|EV_SIGNAL|EV_READ|EV_WRITE|EV_CLOSED)) == EV_TIMEOUT) { evtimer_assign(&eonce->ev, base, event_once_cb, eonce); if (tv == NULL || ! evutil_timerisset(tv)) { /* If the event is going to become active immediately, * don't put it on the timeout queue. This is one * idiom for scheduling a callback, so let's make * it fast (and order-preserving). */ activate = 1; } } else if (events & (EV_READ|EV_WRITE|EV_CLOSED)) { events &= EV_READ|EV_WRITE|EV_CLOSED; event_assign(&eonce->ev, base, fd, events, event_once_cb, eonce); } else { /* Bad event combination */ mm_free(eonce); return (-1); } if (res == 0) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (activate) event_active_nolock_(&eonce->ev, EV_TIMEOUT, 1); else res = event_add_nolock_(&eonce->ev, tv, 0); if (res != 0) { mm_free(eonce); return (res); } else { LIST_INSERT_HEAD(&base->once_events, eonce, next_once); } EVBASE_RELEASE_LOCK(base, th_base_lock); } return (0); } int event_assign(struct event *ev, struct event_base *base, evutil_socket_t fd, short events, void (*callback)(evutil_socket_t, short, void *), void *arg) { if (!base) base = current_base; if (arg == &event_self_cbarg_ptr_) arg = ev; if (!(events & EV_SIGNAL)) event_debug_assert_socket_nonblocking_(fd); event_debug_assert_not_added_(ev); ev->ev_base = base; ev->ev_callback = callback; ev->ev_arg = arg; ev->ev_fd = fd; ev->ev_events = events; ev->ev_res = 0; ev->ev_flags = EVLIST_INIT; ev->ev_ncalls = 0; ev->ev_pncalls = NULL; if (events & EV_SIGNAL) { if ((events & (EV_READ|EV_WRITE|EV_CLOSED)) != 0) { event_warnx("%s: EV_SIGNAL is not compatible with " "EV_READ, EV_WRITE or EV_CLOSED", __func__); return -1; } ev->ev_closure = EV_CLOSURE_EVENT_SIGNAL; } else { if (events & EV_PERSIST) { evutil_timerclear(&ev->ev_io_timeout); ev->ev_closure = EV_CLOSURE_EVENT_PERSIST; } else { ev->ev_closure = EV_CLOSURE_EVENT; } } min_heap_elem_init_(ev); if (base != NULL) { /* by default, we put new events into the middle priority */ ev->ev_pri = base->nactivequeues / 2; } event_debug_note_setup_(ev); return 0; } int event_base_set(struct event_base *base, struct event *ev) { /* Only innocent events may be assigned to a different base */ if (ev->ev_flags != EVLIST_INIT) return (-1); event_debug_assert_is_setup_(ev); ev->ev_base = base; ev->ev_pri = base->nactivequeues/2; return (0); } void event_set(struct event *ev, evutil_socket_t fd, short events, void (*callback)(evutil_socket_t, short, void *), void *arg) { int r; r = event_assign(ev, current_base, fd, events, callback, arg); EVUTIL_ASSERT(r == 0); } void * event_self_cbarg(void) { return &event_self_cbarg_ptr_; } struct event * event_base_get_running_event(struct event_base *base) { struct event *ev = NULL; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (EVBASE_IN_THREAD(base)) { struct event_callback *evcb = base->current_event; if (evcb->evcb_flags & EVLIST_INIT) ev = event_callback_to_event(evcb); } EVBASE_RELEASE_LOCK(base, th_base_lock); return ev; } struct event * event_new(struct event_base *base, evutil_socket_t fd, short events, void (*cb)(evutil_socket_t, short, void *), void *arg) { struct event *ev; ev = mm_malloc(sizeof(struct event)); if (ev == NULL) return (NULL); if (event_assign(ev, base, fd, events, cb, arg) < 0) { mm_free(ev); return (NULL); } return (ev); } void event_free(struct event *ev) { /* This is disabled, so that events which have been finalized be a * valid target for event_free(). That's */ // event_debug_assert_is_setup_(ev); /* make sure that this event won't be coming back to haunt us. */ event_del(ev); event_debug_note_teardown_(ev); mm_free(ev); } void event_debug_unassign(struct event *ev) { event_debug_assert_not_added_(ev); event_debug_note_teardown_(ev); ev->ev_flags &= ~EVLIST_INIT; } #define EVENT_FINALIZE_FREE_ 0x10000 static int event_finalize_nolock_(struct event_base *base, unsigned flags, struct event *ev, event_finalize_callback_fn cb) { ev_uint8_t closure = (flags & EVENT_FINALIZE_FREE_) ? EV_CLOSURE_EVENT_FINALIZE_FREE : EV_CLOSURE_EVENT_FINALIZE; event_del_nolock_(ev, EVENT_DEL_NOBLOCK); ev->ev_closure = closure; ev->ev_evcallback.evcb_cb_union.evcb_evfinalize = cb; event_active_nolock_(ev, EV_FINALIZE, 1); ev->ev_flags |= EVLIST_FINALIZING; return 0; } static int event_finalize_impl_(unsigned flags, struct event *ev, event_finalize_callback_fn cb) { int r; struct event_base *base = ev->ev_base; if (EVUTIL_FAILURE_CHECK(!base)) { event_warnx("%s: event has no event_base set.", __func__); return -1; } EVBASE_ACQUIRE_LOCK(base, th_base_lock); r = event_finalize_nolock_(base, flags, ev, cb); EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } int event_finalize(unsigned flags, struct event *ev, event_finalize_callback_fn cb) { return event_finalize_impl_(flags, ev, cb); } int event_free_finalize(unsigned flags, struct event *ev, event_finalize_callback_fn cb) { return event_finalize_impl_(flags|EVENT_FINALIZE_FREE_, ev, cb); } void event_callback_finalize_nolock_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *)) { struct event *ev = NULL; if (evcb->evcb_flags & EVLIST_INIT) { ev = event_callback_to_event(evcb); event_del_nolock_(ev, EVENT_DEL_NOBLOCK); } else { event_callback_cancel_nolock_(base, evcb, 0); /*XXX can this fail?*/ } evcb->evcb_closure = EV_CLOSURE_CB_FINALIZE; evcb->evcb_cb_union.evcb_cbfinalize = cb; event_callback_activate_nolock_(base, evcb); /* XXX can this really fail?*/ evcb->evcb_flags |= EVLIST_FINALIZING; } void event_callback_finalize_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *)) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); event_callback_finalize_nolock_(base, flags, evcb, cb); EVBASE_RELEASE_LOCK(base, th_base_lock); } /** Internal: Finalize all of the n_cbs callbacks in evcbs. The provided * callback will be invoked on *one of them*, after they have *all* been * finalized. */ int event_callback_finalize_many_(struct event_base *base, int n_cbs, struct event_callback **evcbs, void (*cb)(struct event_callback *, void *)) { int n_pending = 0, i; if (base == NULL) base = current_base; EVBASE_ACQUIRE_LOCK(base, th_base_lock); event_debug(("%s: %d events finalizing", __func__, n_cbs)); /* At most one can be currently executing; the rest we just * cancel... But we always make sure that the finalize callback * runs. */ for (i = 0; i < n_cbs; ++i) { struct event_callback *evcb = evcbs[i]; if (evcb == base->current_event) { event_callback_finalize_nolock_(base, 0, evcb, cb); ++n_pending; } else { event_callback_cancel_nolock_(base, evcb, 0); } } if (n_pending == 0) { /* Just do the first one. */ event_callback_finalize_nolock_(base, 0, evcbs[0], cb); } EVBASE_RELEASE_LOCK(base, th_base_lock); return 0; } /* * Set's the priority of an event - if an event is already scheduled * changing the priority is going to fail. */ int event_priority_set(struct event *ev, int pri) { event_debug_assert_is_setup_(ev); if (ev->ev_flags & EVLIST_ACTIVE) return (-1); if (pri < 0 || pri >= ev->ev_base->nactivequeues) return (-1); ev->ev_pri = pri; return (0); } /* * Checks if a specific event is pending or scheduled. */ int event_pending(const struct event *ev, short event, struct timeval *tv) { int flags = 0; if (EVUTIL_FAILURE_CHECK(ev->ev_base == NULL)) { event_warnx("%s: event has no event_base set.", __func__); return 0; } EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock); event_debug_assert_is_setup_(ev); if (ev->ev_flags & EVLIST_INSERTED) flags |= (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL)); if (ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) flags |= ev->ev_res; if (ev->ev_flags & EVLIST_TIMEOUT) flags |= EV_TIMEOUT; event &= (EV_TIMEOUT|EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL); /* See if there is a timeout that we should report */ if (tv != NULL && (flags & event & EV_TIMEOUT)) { struct timeval tmp = ev->ev_timeout; tmp.tv_usec &= MICROSECONDS_MASK; /* correctly remamp to real time */ evutil_timeradd(&ev->ev_base->tv_clock_diff, &tmp, tv); } EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock); return (flags & event); } int event_initialized(const struct event *ev) { if (!(ev->ev_flags & EVLIST_INIT)) return 0; return 1; } void event_get_assignment(const struct event *event, struct event_base **base_out, evutil_socket_t *fd_out, short *events_out, event_callback_fn *callback_out, void **arg_out) { event_debug_assert_is_setup_(event); if (base_out) *base_out = event->ev_base; if (fd_out) *fd_out = event->ev_fd; if (events_out) *events_out = event->ev_events; if (callback_out) *callback_out = event->ev_callback; if (arg_out) *arg_out = event->ev_arg; } size_t event_get_struct_event_size(void) { return sizeof(struct event); } evutil_socket_t event_get_fd(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_fd; } struct event_base * event_get_base(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_base; } short event_get_events(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_events; } event_callback_fn event_get_callback(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_callback; } void * event_get_callback_arg(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_arg; } int event_get_priority(const struct event *ev) { event_debug_assert_is_setup_(ev); return ev->ev_pri; } int event_add(struct event *ev, const struct timeval *tv) { int res; if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) { event_warnx("%s: event has no event_base set.", __func__); return -1; } EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock); res = event_add_nolock_(ev, tv, 0); EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock); return (res); } /* Helper callback: wake an event_base from another thread. This version * works by writing a byte to one end of a socketpair, so that the event_base * listening on the other end will wake up as the corresponding event * triggers */ static int evthread_notify_base_default(struct event_base *base) { char buf[1]; int r; buf[0] = (char) 0; #ifdef _WIN32 r = send(base->th_notify_fd[1], buf, 1, 0); #else r = write(base->th_notify_fd[1], buf, 1); #endif return (r < 0 && ! EVUTIL_ERR_IS_EAGAIN(errno)) ? -1 : 0; } #ifdef EVENT__HAVE_EVENTFD /* Helper callback: wake an event_base from another thread. This version * assumes that you have a working eventfd() implementation. */ static int evthread_notify_base_eventfd(struct event_base *base) { ev_uint64_t msg = 1; int r; do { r = write(base->th_notify_fd[0], (void*) &msg, sizeof(msg)); } while (r < 0 && errno == EAGAIN); return (r < 0) ? -1 : 0; } #endif /** Tell the thread currently running the event_loop for base (if any) that it * needs to stop waiting in its dispatch function (if it is) and process all * active callbacks. */ static int evthread_notify_base(struct event_base *base) { EVENT_BASE_ASSERT_LOCKED(base); if (!base->th_notify_fn) return -1; if (base->is_notify_pending) return 0; base->is_notify_pending = 1; return base->th_notify_fn(base); } /* Implementation function to remove a timeout on a currently pending event. */ int event_remove_timer_nolock_(struct event *ev) { struct event_base *base = ev->ev_base; EVENT_BASE_ASSERT_LOCKED(base); event_debug_assert_is_setup_(ev); event_debug(("event_remove_timer_nolock: event: %p", ev)); /* If it's not pending on a timeout, we don't need to do anything. */ if (ev->ev_flags & EVLIST_TIMEOUT) { event_queue_remove_timeout(base, ev); evutil_timerclear(&ev->ev_.ev_io.ev_timeout); } return (0); } int event_remove_timer(struct event *ev) { int res; if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) { event_warnx("%s: event has no event_base set.", __func__); return -1; } EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock); res = event_remove_timer_nolock_(ev); EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock); return (res); } /* Implementation function to add an event. Works just like event_add, * except: 1) it requires that we have the lock. 2) if tv_is_absolute is set, * we treat tv as an absolute time, not as an interval to add to the current * time */ int event_add_nolock_(struct event *ev, const struct timeval *tv, int tv_is_absolute) { struct event_base *base = ev->ev_base; int res = 0; int notify = 0; EVENT_BASE_ASSERT_LOCKED(base); event_debug_assert_is_setup_(ev); event_debug(( "event_add: event: %p (fd "EV_SOCK_FMT"), %s%s%s%scall %p", ev, EV_SOCK_ARG(ev->ev_fd), ev->ev_events & EV_READ ? "EV_READ " : " ", ev->ev_events & EV_WRITE ? "EV_WRITE " : " ", ev->ev_events & EV_CLOSED ? "EV_CLOSED " : " ", tv ? "EV_TIMEOUT " : " ", ev->ev_callback)); EVUTIL_ASSERT(!(ev->ev_flags & ~EVLIST_ALL)); if (ev->ev_flags & EVLIST_FINALIZING) { /* XXXX debug */ return (-1); } /* * prepare for timeout insertion further below, if we get a * failure on any step, we should not change any state. */ if (tv != NULL && !(ev->ev_flags & EVLIST_TIMEOUT)) { if (min_heap_reserve_(&base->timeheap, 1 + min_heap_size_(&base->timeheap)) == -1) return (-1); /* ENOMEM == errno */ } /* If the main thread is currently executing a signal event's * callback, and we are not the main thread, then we want to wait * until the callback is done before we mess with the event, or else * we can race on ev_ncalls and ev_pncalls below. */ #ifndef EVENT__DISABLE_THREAD_SUPPORT if (base->current_event == event_to_event_callback(ev) && (ev->ev_events & EV_SIGNAL) && !EVBASE_IN_THREAD(base)) { ++base->current_event_waiters; EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock); } #endif if ((ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED|EV_SIGNAL)) && !(ev->ev_flags & (EVLIST_INSERTED|EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) { if (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED)) res = evmap_io_add_(base, ev->ev_fd, ev); else if (ev->ev_events & EV_SIGNAL) res = evmap_signal_add_(base, (int)ev->ev_fd, ev); if (res != -1) event_queue_insert_inserted(base, ev); if (res == 1) { /* evmap says we need to notify the main thread. */ notify = 1; res = 0; } } /* * we should change the timeout state only if the previous event * addition succeeded. */ if (res != -1 && tv != NULL) { struct timeval now; int common_timeout; #ifdef USE_REINSERT_TIMEOUT int was_common; int old_timeout_idx; #endif /* * for persistent timeout events, we remember the * timeout value and re-add the event. * * If tv_is_absolute, this was already set. */ if (ev->ev_closure == EV_CLOSURE_EVENT_PERSIST && !tv_is_absolute) ev->ev_io_timeout = *tv; #ifndef USE_REINSERT_TIMEOUT if (ev->ev_flags & EVLIST_TIMEOUT) { event_queue_remove_timeout(base, ev); } #endif /* Check if it is active due to a timeout. Rescheduling * this timeout before the callback can be executed * removes it from the active list. */ if ((ev->ev_flags & EVLIST_ACTIVE) && (ev->ev_res & EV_TIMEOUT)) { if (ev->ev_events & EV_SIGNAL) { /* See if we are just active executing * this event in a loop */ if (ev->ev_ncalls && ev->ev_pncalls) { /* Abort loop */ *ev->ev_pncalls = 0; } } event_queue_remove_active(base, event_to_event_callback(ev)); } gettime(base, &now); common_timeout = is_common_timeout(tv, base); #ifdef USE_REINSERT_TIMEOUT was_common = is_common_timeout(&ev->ev_timeout, base); old_timeout_idx = COMMON_TIMEOUT_IDX(&ev->ev_timeout); #endif if (tv_is_absolute) { ev->ev_timeout = *tv; } else if (common_timeout) { struct timeval tmp = *tv; tmp.tv_usec &= MICROSECONDS_MASK; evutil_timeradd(&now, &tmp, &ev->ev_timeout); ev->ev_timeout.tv_usec |= (tv->tv_usec & ~MICROSECONDS_MASK); } else { evutil_timeradd(&now, tv, &ev->ev_timeout); } event_debug(( "event_add: event %p, timeout in %d seconds %d useconds, call %p", ev, (int)tv->tv_sec, (int)tv->tv_usec, ev->ev_callback)); #ifdef USE_REINSERT_TIMEOUT event_queue_reinsert_timeout(base, ev, was_common, common_timeout, old_timeout_idx); #else event_queue_insert_timeout(base, ev); #endif if (common_timeout) { struct common_timeout_list *ctl = get_common_timeout_list(base, &ev->ev_timeout); if (ev == TAILQ_FIRST(&ctl->events)) { common_timeout_schedule(ctl, &now, ev); } } else { struct event* top = NULL; /* See if the earliest timeout is now earlier than it * was before: if so, we will need to tell the main * thread to wake up earlier than it would otherwise. * We double check the timeout of the top element to * handle time distortions due to system suspension. */ if (min_heap_elt_is_top_(ev)) notify = 1; else if ((top = min_heap_top_(&base->timeheap)) != NULL && evutil_timercmp(&top->ev_timeout, &now, <)) notify = 1; } } /* if we are not in the right thread, we need to wake up the loop */ if (res != -1 && notify && EVBASE_NEED_NOTIFY(base)) evthread_notify_base(base); event_debug_note_add_(ev); return (res); } static int event_del_(struct event *ev, int blocking) { int res; struct event_base *base = ev->ev_base; if (EVUTIL_FAILURE_CHECK(!base)) { event_warnx("%s: event has no event_base set.", __func__); return -1; } EVBASE_ACQUIRE_LOCK(base, th_base_lock); res = event_del_nolock_(ev, blocking); EVBASE_RELEASE_LOCK(base, th_base_lock); return (res); } int event_del(struct event *ev) { return event_del_(ev, EVENT_DEL_AUTOBLOCK); } int event_del_block(struct event *ev) { return event_del_(ev, EVENT_DEL_BLOCK); } int event_del_noblock(struct event *ev) { return event_del_(ev, EVENT_DEL_NOBLOCK); } /** Helper for event_del: always called with th_base_lock held. * * "blocking" must be one of the EVENT_DEL_{BLOCK, NOBLOCK, AUTOBLOCK, * EVEN_IF_FINALIZING} values. See those for more information. */ int event_del_nolock_(struct event *ev, int blocking) { struct event_base *base; int res = 0, notify = 0; event_debug(("event_del: %p (fd "EV_SOCK_FMT"), callback %p", ev, EV_SOCK_ARG(ev->ev_fd), ev->ev_callback)); /* An event without a base has not been added */ if (ev->ev_base == NULL) return (-1); EVENT_BASE_ASSERT_LOCKED(ev->ev_base); if (blocking != EVENT_DEL_EVEN_IF_FINALIZING) { if (ev->ev_flags & EVLIST_FINALIZING) { /* XXXX Debug */ return 0; } } base = ev->ev_base; EVUTIL_ASSERT(!(ev->ev_flags & ~EVLIST_ALL)); /* See if we are just active executing this event in a loop */ if (ev->ev_events & EV_SIGNAL) { if (ev->ev_ncalls && ev->ev_pncalls) { /* Abort loop */ *ev->ev_pncalls = 0; } } if (ev->ev_flags & EVLIST_TIMEOUT) { /* NOTE: We never need to notify the main thread because of a * deleted timeout event: all that could happen if we don't is * that the dispatch loop might wake up too early. But the * point of notifying the main thread _is_ to wake up the * dispatch loop early anyway, so we wouldn't gain anything by * doing it. */ event_queue_remove_timeout(base, ev); } if (ev->ev_flags & EVLIST_ACTIVE) event_queue_remove_active(base, event_to_event_callback(ev)); else if (ev->ev_flags & EVLIST_ACTIVE_LATER) event_queue_remove_active_later(base, event_to_event_callback(ev)); if (ev->ev_flags & EVLIST_INSERTED) { event_queue_remove_inserted(base, ev); if (ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED)) res = evmap_io_del_(base, ev->ev_fd, ev); else res = evmap_signal_del_(base, (int)ev->ev_fd, ev); if (res == 1) { /* evmap says we need to notify the main thread. */ notify = 1; res = 0; } /* If we do not have events, let's notify event base so it can * exit without waiting */ if (!event_haveevents(base) && !N_ACTIVE_CALLBACKS(base)) notify = 1; } /* if we are not in the right thread, we need to wake up the loop */ if (res != -1 && notify && EVBASE_NEED_NOTIFY(base)) evthread_notify_base(base); event_debug_note_del_(ev); /* If the main thread is currently executing this event's callback, * and we are not the main thread, then we want to wait until the * callback is done before returning. That way, when this function * returns, it will be safe to free the user-supplied argument. */ #ifndef EVENT__DISABLE_THREAD_SUPPORT if (blocking != EVENT_DEL_NOBLOCK && base->current_event == event_to_event_callback(ev) && !EVBASE_IN_THREAD(base) && (blocking == EVENT_DEL_BLOCK || !(ev->ev_events & EV_FINALIZE))) { ++base->current_event_waiters; EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock); } #endif return (res); } void event_active(struct event *ev, int res, short ncalls) { if (EVUTIL_FAILURE_CHECK(!ev->ev_base)) { event_warnx("%s: event has no event_base set.", __func__); return; } EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock); event_debug_assert_is_setup_(ev); event_active_nolock_(ev, res, ncalls); EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock); } void event_active_nolock_(struct event *ev, int res, short ncalls) { struct event_base *base; event_debug(("event_active: %p (fd "EV_SOCK_FMT"), res %d, callback %p", ev, EV_SOCK_ARG(ev->ev_fd), (int)res, ev->ev_callback)); base = ev->ev_base; EVENT_BASE_ASSERT_LOCKED(base); if (ev->ev_flags & EVLIST_FINALIZING) { /* XXXX debug */ return; } switch ((ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) { default: case EVLIST_ACTIVE|EVLIST_ACTIVE_LATER: EVUTIL_ASSERT(0); break; case EVLIST_ACTIVE: /* We get different kinds of events, add them together */ ev->ev_res |= res; return; case EVLIST_ACTIVE_LATER: ev->ev_res |= res; break; case 0: ev->ev_res = res; break; } if (ev->ev_pri < base->event_running_priority) base->event_continue = 1; if (ev->ev_events & EV_SIGNAL) { #ifndef EVENT__DISABLE_THREAD_SUPPORT if (base->current_event == event_to_event_callback(ev) && !EVBASE_IN_THREAD(base)) { ++base->current_event_waiters; EVTHREAD_COND_WAIT(base->current_event_cond, base->th_base_lock); } #endif ev->ev_ncalls = ncalls; ev->ev_pncalls = NULL; } event_callback_activate_nolock_(base, event_to_event_callback(ev)); } void event_active_later_(struct event *ev, int res) { EVBASE_ACQUIRE_LOCK(ev->ev_base, th_base_lock); event_active_later_nolock_(ev, res); EVBASE_RELEASE_LOCK(ev->ev_base, th_base_lock); } void event_active_later_nolock_(struct event *ev, int res) { struct event_base *base = ev->ev_base; EVENT_BASE_ASSERT_LOCKED(base); if (ev->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) { /* We get different kinds of events, add them together */ ev->ev_res |= res; return; } ev->ev_res = res; event_callback_activate_later_nolock_(base, event_to_event_callback(ev)); } int event_callback_activate_(struct event_base *base, struct event_callback *evcb) { int r; EVBASE_ACQUIRE_LOCK(base, th_base_lock); r = event_callback_activate_nolock_(base, evcb); EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } int event_callback_activate_nolock_(struct event_base *base, struct event_callback *evcb) { int r = 1; if (evcb->evcb_flags & EVLIST_FINALIZING) return 0; switch (evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) { default: EVUTIL_ASSERT(0); EVUTIL_FALLTHROUGH; case EVLIST_ACTIVE_LATER: event_queue_remove_active_later(base, evcb); r = 0; break; case EVLIST_ACTIVE: return 0; case 0: break; } event_queue_insert_active(base, evcb); if (EVBASE_NEED_NOTIFY(base)) evthread_notify_base(base); return r; } int event_callback_activate_later_nolock_(struct event_base *base, struct event_callback *evcb) { if (evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) return 0; event_queue_insert_active_later(base, evcb); if (EVBASE_NEED_NOTIFY(base)) evthread_notify_base(base); return 1; } void event_callback_init_(struct event_base *base, struct event_callback *cb) { memset(cb, 0, sizeof(*cb)); cb->evcb_pri = base->nactivequeues - 1; } int event_callback_cancel_(struct event_base *base, struct event_callback *evcb) { int r; EVBASE_ACQUIRE_LOCK(base, th_base_lock); r = event_callback_cancel_nolock_(base, evcb, 0); EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } int event_callback_cancel_nolock_(struct event_base *base, struct event_callback *evcb, int even_if_finalizing) { if ((evcb->evcb_flags & EVLIST_FINALIZING) && !even_if_finalizing) return 0; if (evcb->evcb_flags & EVLIST_INIT) return event_del_nolock_(event_callback_to_event(evcb), even_if_finalizing ? EVENT_DEL_EVEN_IF_FINALIZING : EVENT_DEL_AUTOBLOCK); switch ((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) { default: case EVLIST_ACTIVE|EVLIST_ACTIVE_LATER: EVUTIL_ASSERT(0); break; case EVLIST_ACTIVE: /* We get different kinds of events, add them together */ event_queue_remove_active(base, evcb); return 0; case EVLIST_ACTIVE_LATER: event_queue_remove_active_later(base, evcb); break; case 0: break; } return 0; } void event_deferred_cb_init_(struct event_callback *cb, ev_uint8_t priority, deferred_cb_fn fn, void *arg) { memset(cb, 0, sizeof(*cb)); cb->evcb_cb_union.evcb_selfcb = fn; cb->evcb_arg = arg; cb->evcb_pri = priority; cb->evcb_closure = EV_CLOSURE_CB_SELF; } void event_deferred_cb_set_priority_(struct event_callback *cb, ev_uint8_t priority) { cb->evcb_pri = priority; } void event_deferred_cb_cancel_(struct event_base *base, struct event_callback *cb) { if (!base) base = current_base; event_callback_cancel_(base, cb); } #define MAX_DEFERREDS_QUEUED 32 int event_deferred_cb_schedule_(struct event_base *base, struct event_callback *cb) { int r = 1; if (!base) base = current_base; EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (base->n_deferreds_queued > MAX_DEFERREDS_QUEUED) { r = event_callback_activate_later_nolock_(base, cb); } else { r = event_callback_activate_nolock_(base, cb); if (r) { ++base->n_deferreds_queued; } } EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } static int timeout_next(struct event_base *base, struct timeval **tv_p) { /* Caller must hold th_base_lock */ struct timeval now; struct event *ev; struct timeval *tv = *tv_p; int res = 0; ev = min_heap_top_(&base->timeheap); if (ev == NULL) { /* if no time-based events are active wait for I/O */ *tv_p = NULL; goto out; } if (gettime(base, &now) == -1) { res = -1; goto out; } if (evutil_timercmp(&ev->ev_timeout, &now, <=)) { evutil_timerclear(tv); goto out; } evutil_timersub(&ev->ev_timeout, &now, tv); EVUTIL_ASSERT(tv->tv_sec >= 0); EVUTIL_ASSERT(tv->tv_usec >= 0); event_debug(("timeout_next: event: %p, in %d seconds, %d useconds", ev, (int)tv->tv_sec, (int)tv->tv_usec)); out: return (res); } /* Activate every event whose timeout has elapsed. */ static void timeout_process(struct event_base *base) { /* Caller must hold lock. */ struct timeval now; struct event *ev; if (min_heap_empty_(&base->timeheap)) { return; } gettime(base, &now); while ((ev = min_heap_top_(&base->timeheap))) { if (evutil_timercmp(&ev->ev_timeout, &now, >)) break; /* delete this event from the I/O queues */ event_del_nolock_(ev, EVENT_DEL_NOBLOCK); event_debug(("timeout_process: event: %p, call %p", ev, ev->ev_callback)); event_active_nolock_(ev, EV_TIMEOUT, 1); } } #ifndef MAX #define MAX(a,b) (((a)>(b))?(a):(b)) #endif #define MAX_EVENT_COUNT(var, v) var = MAX(var, v) /* These are a fancy way to spell if (~flags & EVLIST_INTERNAL) base->event_count--/++; */ #define DECR_EVENT_COUNT(base,flags) \ ((base)->event_count -= !((flags) & EVLIST_INTERNAL)) #define INCR_EVENT_COUNT(base,flags) do { \ ((base)->event_count += !((flags) & EVLIST_INTERNAL)); \ MAX_EVENT_COUNT((base)->event_count_max, (base)->event_count); \ } while (0) static void event_queue_remove_inserted(struct event_base *base, struct event *ev) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(!(ev->ev_flags & EVLIST_INSERTED))) { event_errx(1, "%s: %p(fd "EV_SOCK_FMT") not on queue %x", __func__, ev, EV_SOCK_ARG(ev->ev_fd), EVLIST_INSERTED); return; } DECR_EVENT_COUNT(base, ev->ev_flags); ev->ev_flags &= ~EVLIST_INSERTED; } static void event_queue_remove_active(struct event_base *base, struct event_callback *evcb) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(!(evcb->evcb_flags & EVLIST_ACTIVE))) { event_errx(1, "%s: %p not on queue %x", __func__, evcb, EVLIST_ACTIVE); return; } DECR_EVENT_COUNT(base, evcb->evcb_flags); evcb->evcb_flags &= ~EVLIST_ACTIVE; base->event_count_active--; TAILQ_REMOVE(&base->activequeues[evcb->evcb_pri], evcb, evcb_active_next); } static void event_queue_remove_active_later(struct event_base *base, struct event_callback *evcb) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(!(evcb->evcb_flags & EVLIST_ACTIVE_LATER))) { event_errx(1, "%s: %p not on queue %x", __func__, evcb, EVLIST_ACTIVE_LATER); return; } DECR_EVENT_COUNT(base, evcb->evcb_flags); evcb->evcb_flags &= ~EVLIST_ACTIVE_LATER; base->event_count_active--; TAILQ_REMOVE(&base->active_later_queue, evcb, evcb_active_next); } static void event_queue_remove_timeout(struct event_base *base, struct event *ev) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(!(ev->ev_flags & EVLIST_TIMEOUT))) { event_errx(1, "%s: %p(fd "EV_SOCK_FMT") not on queue %x", __func__, ev, EV_SOCK_ARG(ev->ev_fd), EVLIST_TIMEOUT); return; } DECR_EVENT_COUNT(base, ev->ev_flags); ev->ev_flags &= ~EVLIST_TIMEOUT; if (is_common_timeout(&ev->ev_timeout, base)) { struct common_timeout_list *ctl = get_common_timeout_list(base, &ev->ev_timeout); TAILQ_REMOVE(&ctl->events, ev, ev_timeout_pos.ev_next_with_common_timeout); } else { min_heap_erase_(&base->timeheap, ev); } } #ifdef USE_REINSERT_TIMEOUT /* Remove and reinsert 'ev' into the timeout queue. */ static void event_queue_reinsert_timeout(struct event_base *base, struct event *ev, int was_common, int is_common, int old_timeout_idx) { struct common_timeout_list *ctl; if (!(ev->ev_flags & EVLIST_TIMEOUT)) { event_queue_insert_timeout(base, ev); return; } switch ((was_common<<1) | is_common) { case 3: /* Changing from one common timeout to another */ ctl = base->common_timeout_queues[old_timeout_idx]; TAILQ_REMOVE(&ctl->events, ev, ev_timeout_pos.ev_next_with_common_timeout); ctl = get_common_timeout_list(base, &ev->ev_timeout); insert_common_timeout_inorder(ctl, ev); break; case 2: /* Was common; is no longer common */ ctl = base->common_timeout_queues[old_timeout_idx]; TAILQ_REMOVE(&ctl->events, ev, ev_timeout_pos.ev_next_with_common_timeout); min_heap_push_(&base->timeheap, ev); break; case 1: /* Wasn't common; has become common. */ min_heap_erase_(&base->timeheap, ev); ctl = get_common_timeout_list(base, &ev->ev_timeout); insert_common_timeout_inorder(ctl, ev); break; case 0: /* was in heap; is still on heap. */ min_heap_adjust_(&base->timeheap, ev); break; default: EVUTIL_ASSERT(0); /* unreachable */ break; } } #endif /* Add 'ev' to the common timeout list in 'ev'. */ static void insert_common_timeout_inorder(struct common_timeout_list *ctl, struct event *ev) { struct event *e; /* By all logic, we should just be able to append 'ev' to the end of * ctl->events, since the timeout on each 'ev' is set to {the common * timeout} + {the time when we add the event}, and so the events * should arrive in order of their timeeouts. But just in case * there's some wacky threading issue going on, we do a search from * the end of 'ev' to find the right insertion point. */ TAILQ_FOREACH_REVERSE(e, &ctl->events, event_list, ev_timeout_pos.ev_next_with_common_timeout) { /* This timercmp is a little sneaky, since both ev and e have * magic values in tv_usec. Fortunately, they ought to have * the _same_ magic values in tv_usec. Let's assert for that. */ EVUTIL_ASSERT( is_same_common_timeout(&e->ev_timeout, &ev->ev_timeout)); if (evutil_timercmp(&ev->ev_timeout, &e->ev_timeout, >=)) { TAILQ_INSERT_AFTER(&ctl->events, e, ev, ev_timeout_pos.ev_next_with_common_timeout); return; } } TAILQ_INSERT_HEAD(&ctl->events, ev, ev_timeout_pos.ev_next_with_common_timeout); } static void event_queue_insert_inserted(struct event_base *base, struct event *ev) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(ev->ev_flags & EVLIST_INSERTED)) { event_errx(1, "%s: %p(fd "EV_SOCK_FMT") already inserted", __func__, ev, EV_SOCK_ARG(ev->ev_fd)); return; } INCR_EVENT_COUNT(base, ev->ev_flags); ev->ev_flags |= EVLIST_INSERTED; } static void event_queue_insert_active(struct event_base *base, struct event_callback *evcb) { EVENT_BASE_ASSERT_LOCKED(base); if (evcb->evcb_flags & EVLIST_ACTIVE) { /* Double insertion is possible for active events */ return; } INCR_EVENT_COUNT(base, evcb->evcb_flags); evcb->evcb_flags |= EVLIST_ACTIVE; base->event_count_active++; MAX_EVENT_COUNT(base->event_count_active_max, base->event_count_active); EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues); TAILQ_INSERT_TAIL(&base->activequeues[evcb->evcb_pri], evcb, evcb_active_next); } static void event_queue_insert_active_later(struct event_base *base, struct event_callback *evcb) { EVENT_BASE_ASSERT_LOCKED(base); if (evcb->evcb_flags & (EVLIST_ACTIVE_LATER|EVLIST_ACTIVE)) { /* Double insertion is possible */ return; } INCR_EVENT_COUNT(base, evcb->evcb_flags); evcb->evcb_flags |= EVLIST_ACTIVE_LATER; base->event_count_active++; MAX_EVENT_COUNT(base->event_count_active_max, base->event_count_active); EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues); TAILQ_INSERT_TAIL(&base->active_later_queue, evcb, evcb_active_next); } static void event_queue_insert_timeout(struct event_base *base, struct event *ev) { EVENT_BASE_ASSERT_LOCKED(base); if (EVUTIL_FAILURE_CHECK(ev->ev_flags & EVLIST_TIMEOUT)) { event_errx(1, "%s: %p(fd "EV_SOCK_FMT") already on timeout", __func__, ev, EV_SOCK_ARG(ev->ev_fd)); return; } INCR_EVENT_COUNT(base, ev->ev_flags); ev->ev_flags |= EVLIST_TIMEOUT; if (is_common_timeout(&ev->ev_timeout, base)) { struct common_timeout_list *ctl = get_common_timeout_list(base, &ev->ev_timeout); insert_common_timeout_inorder(ctl, ev); } else { min_heap_push_(&base->timeheap, ev); } } static void event_queue_make_later_events_active(struct event_base *base) { struct event_callback *evcb; EVENT_BASE_ASSERT_LOCKED(base); while ((evcb = TAILQ_FIRST(&base->active_later_queue))) { TAILQ_REMOVE(&base->active_later_queue, evcb, evcb_active_next); evcb->evcb_flags = (evcb->evcb_flags & ~EVLIST_ACTIVE_LATER) | EVLIST_ACTIVE; EVUTIL_ASSERT(evcb->evcb_pri < base->nactivequeues); TAILQ_INSERT_TAIL(&base->activequeues[evcb->evcb_pri], evcb, evcb_active_next); base->n_deferreds_queued += (evcb->evcb_closure == EV_CLOSURE_CB_SELF); } } /* Functions for debugging */ const char * event_get_version(void) { return (EVENT__VERSION); } ev_uint32_t event_get_version_number(void) { return (EVENT__NUMERIC_VERSION); } /* * No thread-safe interface needed - the information should be the same * for all threads. */ const char * event_get_method(void) { return (current_base->evsel->name); } #ifndef EVENT__DISABLE_MM_REPLACEMENT static void *(*mm_malloc_fn_)(size_t sz) = NULL; static void *(*mm_realloc_fn_)(void *p, size_t sz) = NULL; static void (*mm_free_fn_)(void *p) = NULL; void * event_mm_malloc_(size_t sz) { if (sz == 0) return NULL; if (mm_malloc_fn_) return mm_malloc_fn_(sz); else return malloc(sz); } void * event_mm_calloc_(size_t count, size_t size) { if (count == 0 || size == 0) return NULL; if (mm_malloc_fn_) { size_t sz = count * size; void *p = NULL; if (count > EV_SIZE_MAX / size) goto error; p = mm_malloc_fn_(sz); if (p) return memset(p, 0, sz); } else { void *p = calloc(count, size); #ifdef _WIN32 /* Windows calloc doesn't reliably set ENOMEM */ if (p == NULL) goto error; #endif return p; } error: errno = ENOMEM; return NULL; } char * event_mm_strdup_(const char *str) { if (!str) { errno = EINVAL; return NULL; } if (mm_malloc_fn_) { size_t ln = strlen(str); void *p = NULL; if (ln == EV_SIZE_MAX) goto error; p = mm_malloc_fn_(ln+1); if (p) return memcpy(p, str, ln+1); } else #ifdef _WIN32 return _strdup(str); #else return strdup(str); #endif error: errno = ENOMEM; return NULL; } void * event_mm_realloc_(void *ptr, size_t sz) { if (mm_realloc_fn_) return mm_realloc_fn_(ptr, sz); else return realloc(ptr, sz); } void event_mm_free_(void *ptr) { if (mm_free_fn_) mm_free_fn_(ptr); else free(ptr); } void event_set_mem_functions(void *(*malloc_fn)(size_t sz), void *(*realloc_fn)(void *ptr, size_t sz), void (*free_fn)(void *ptr)) { mm_malloc_fn_ = malloc_fn; mm_realloc_fn_ = realloc_fn; mm_free_fn_ = free_fn; } #endif #ifdef EVENT__HAVE_EVENTFD static void evthread_notify_drain_eventfd(evutil_socket_t fd, short what, void *arg) { ev_uint64_t msg; ev_ssize_t r; struct event_base *base = arg; r = read(fd, (void*) &msg, sizeof(msg)); if (r<0 && errno != EAGAIN) { event_sock_warn(fd, "Error reading from eventfd"); } EVBASE_ACQUIRE_LOCK(base, th_base_lock); base->is_notify_pending = 0; EVBASE_RELEASE_LOCK(base, th_base_lock); } #endif static void evthread_notify_drain_default(evutil_socket_t fd, short what, void *arg) { unsigned char buf[1024]; struct event_base *base = arg; #ifdef _WIN32 while (recv(fd, (char*)buf, sizeof(buf), 0) > 0) ; #else while (read(fd, (char*)buf, sizeof(buf)) > 0) ; #endif EVBASE_ACQUIRE_LOCK(base, th_base_lock); base->is_notify_pending = 0; EVBASE_RELEASE_LOCK(base, th_base_lock); } int evthread_make_base_notifiable(struct event_base *base) { int r; if (!base) return -1; EVBASE_ACQUIRE_LOCK(base, th_base_lock); r = evthread_make_base_notifiable_nolock_(base); EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } static int evthread_make_base_notifiable_nolock_(struct event_base *base) { void (*cb)(evutil_socket_t, short, void *); int (*notify)(struct event_base *); if (base->th_notify_fn != NULL) { /* The base is already notifiable: we're doing fine. */ return 0; } #if defined(EVENT__HAVE_WORKING_KQUEUE) if (base->evsel == &kqops && event_kq_add_notify_event_(base) == 0) { base->th_notify_fn = event_kq_notify_base_; /* No need to add an event here; the backend can wake * itself up just fine. */ return 0; } #endif #ifdef EVENT__HAVE_EVENTFD base->th_notify_fd[0] = evutil_eventfd_(0, EVUTIL_EFD_CLOEXEC|EVUTIL_EFD_NONBLOCK); if (base->th_notify_fd[0] >= 0) { base->th_notify_fd[1] = -1; notify = evthread_notify_base_eventfd; cb = evthread_notify_drain_eventfd; } else #endif if (evutil_make_internal_pipe_(base->th_notify_fd) == 0) { notify = evthread_notify_base_default; cb = evthread_notify_drain_default; } else { return -1; } base->th_notify_fn = notify; /* prepare an event that we can use for wakeup */ event_assign(&base->th_notify, base, base->th_notify_fd[0], EV_READ|EV_PERSIST, cb, base); /* we need to mark this as internal event */ base->th_notify.ev_flags |= EVLIST_INTERNAL; event_priority_set(&base->th_notify, 0); return event_add_nolock_(&base->th_notify, NULL, 0); } int event_base_foreach_event_nolock_(struct event_base *base, event_base_foreach_event_cb fn, void *arg) { int r, i; unsigned u; struct event *ev; /* Start out with all the EVLIST_INSERTED events. */ if ((r = evmap_foreach_event_(base, fn, arg))) return r; /* Okay, now we deal with those events that have timeouts and are in * the min-heap. */ for (u = 0; u < base->timeheap.n; ++u) { ev = base->timeheap.p[u]; if (ev->ev_flags & EVLIST_INSERTED) { /* we already processed this one */ continue; } if ((r = fn(base, ev, arg))) return r; } /* Now for the events in one of the timeout queues. * the min-heap. */ for (i = 0; i < base->n_common_timeouts; ++i) { struct common_timeout_list *ctl = base->common_timeout_queues[i]; TAILQ_FOREACH(ev, &ctl->events, ev_timeout_pos.ev_next_with_common_timeout) { if (ev->ev_flags & EVLIST_INSERTED) { /* we already processed this one */ continue; } if ((r = fn(base, ev, arg))) return r; } } /* Finally, we deal wit all the active events that we haven't touched * yet. */ for (i = 0; i < base->nactivequeues; ++i) { struct event_callback *evcb; TAILQ_FOREACH(evcb, &base->activequeues[i], evcb_active_next) { if ((evcb->evcb_flags & (EVLIST_INIT|EVLIST_INSERTED|EVLIST_TIMEOUT)) != EVLIST_INIT) { /* This isn't an event (evlist_init clear), or * we already processed it. (inserted or * timeout set */ continue; } ev = event_callback_to_event(evcb); if ((r = fn(base, ev, arg))) return r; } } return 0; } /* Helper for event_base_dump_events: called on each event in the event base; * dumps only the inserted events. */ static int dump_inserted_event_fn(const struct event_base *base, const struct event *e, void *arg) { FILE *output = arg; const char *gloss = (e->ev_events & EV_SIGNAL) ? "sig" : "fd "; if (! (e->ev_flags & (EVLIST_INSERTED|EVLIST_TIMEOUT))) return 0; fprintf(output, " %p [%s "EV_SOCK_FMT"]%s%s%s%s%s%s%s", (void*)e, gloss, EV_SOCK_ARG(e->ev_fd), (e->ev_events&EV_READ)?" Read":"", (e->ev_events&EV_WRITE)?" Write":"", (e->ev_events&EV_CLOSED)?" EOF":"", (e->ev_events&EV_SIGNAL)?" Signal":"", (e->ev_events&EV_PERSIST)?" Persist":"", (e->ev_events&EV_ET)?" ET":"", (e->ev_flags&EVLIST_INTERNAL)?" Internal":""); if (e->ev_flags & EVLIST_TIMEOUT) { struct timeval tv; tv.tv_sec = e->ev_timeout.tv_sec; tv.tv_usec = e->ev_timeout.tv_usec & MICROSECONDS_MASK; evutil_timeradd(&tv, &base->tv_clock_diff, &tv); fprintf(output, " Timeout=%ld.%06d", (long)tv.tv_sec, (int)(tv.tv_usec & MICROSECONDS_MASK)); } fputc('\n', output); return 0; } /* Helper for event_base_dump_events: called on each event in the event base; * dumps only the active events. */ static int dump_active_event_fn(const struct event_base *base, const struct event *e, void *arg) { FILE *output = arg; const char *gloss = (e->ev_events & EV_SIGNAL) ? "sig" : "fd "; if (! (e->ev_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER))) return 0; fprintf(output, " %p [%s "EV_SOCK_FMT", priority=%d]%s%s%s%s%s active%s%s\n", (void*)e, gloss, EV_SOCK_ARG(e->ev_fd), e->ev_pri, (e->ev_res&EV_READ)?" Read":"", (e->ev_res&EV_WRITE)?" Write":"", (e->ev_res&EV_CLOSED)?" EOF":"", (e->ev_res&EV_SIGNAL)?" Signal":"", (e->ev_res&EV_TIMEOUT)?" Timeout":"", (e->ev_flags&EVLIST_INTERNAL)?" [Internal]":"", (e->ev_flags&EVLIST_ACTIVE_LATER)?" [NextTime]":""); return 0; } int event_base_foreach_event(struct event_base *base, event_base_foreach_event_cb fn, void *arg) { int r; if ((!fn) || (!base)) { return -1; } EVBASE_ACQUIRE_LOCK(base, th_base_lock); r = event_base_foreach_event_nolock_(base, fn, arg); EVBASE_RELEASE_LOCK(base, th_base_lock); return r; } void event_base_dump_events(struct event_base *base, FILE *output) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); fprintf(output, "Inserted events:\n"); event_base_foreach_event_nolock_(base, dump_inserted_event_fn, output); fprintf(output, "Active events:\n"); event_base_foreach_event_nolock_(base, dump_active_event_fn, output); EVBASE_RELEASE_LOCK(base, th_base_lock); } void event_base_active_by_fd(struct event_base *base, evutil_socket_t fd, short events) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); /* Activate any non timer events */ if (!(events & EV_TIMEOUT)) { evmap_io_active_(base, fd, events & (EV_READ|EV_WRITE|EV_CLOSED)); } else { /* If we want to activate timer events, loop and activate each event with * the same fd in both the timeheap and common timeouts list */ int i; unsigned u; struct event *ev; for (u = 0; u < base->timeheap.n; ++u) { ev = base->timeheap.p[u]; if (ev->ev_fd == fd) { event_active_nolock_(ev, EV_TIMEOUT, 1); } } for (i = 0; i < base->n_common_timeouts; ++i) { struct common_timeout_list *ctl = base->common_timeout_queues[i]; TAILQ_FOREACH(ev, &ctl->events, ev_timeout_pos.ev_next_with_common_timeout) { if (ev->ev_fd == fd) { event_active_nolock_(ev, EV_TIMEOUT, 1); } } } } EVBASE_RELEASE_LOCK(base, th_base_lock); } void event_base_active_by_signal(struct event_base *base, int sig) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); evmap_signal_active_(base, sig, 1); EVBASE_RELEASE_LOCK(base, th_base_lock); } void event_base_add_virtual_(struct event_base *base) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); base->virtual_event_count++; MAX_EVENT_COUNT(base->virtual_event_count_max, base->virtual_event_count); EVBASE_RELEASE_LOCK(base, th_base_lock); } void event_base_del_virtual_(struct event_base *base) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); EVUTIL_ASSERT(base->virtual_event_count > 0); base->virtual_event_count--; if (base->virtual_event_count == 0 && EVBASE_NEED_NOTIFY(base)) evthread_notify_base(base); EVBASE_RELEASE_LOCK(base, th_base_lock); } static void event_free_debug_globals_locks(void) { #ifndef EVENT__DISABLE_THREAD_SUPPORT #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_map_lock_ != NULL) { EVTHREAD_FREE_LOCK(event_debug_map_lock_, 0); event_debug_map_lock_ = NULL; evthreadimpl_disable_lock_debugging_(); } #endif /* EVENT__DISABLE_DEBUG_MODE */ #endif /* EVENT__DISABLE_THREAD_SUPPORT */ return; } static void event_free_debug_globals(void) { event_free_debug_globals_locks(); } static void event_free_evsig_globals(void) { evsig_free_globals_(); } static void event_free_evutil_globals(void) { evutil_free_globals_(); } static void event_free_globals(void) { event_free_debug_globals(); event_free_evsig_globals(); event_free_evutil_globals(); } void libevent_global_shutdown(void) { event_disable_debug_mode(); event_free_globals(); } #ifndef EVENT__DISABLE_THREAD_SUPPORT int event_global_setup_locks_(const int enable_locks) { #ifndef EVENT__DISABLE_DEBUG_MODE EVTHREAD_SETUP_GLOBAL_LOCK(event_debug_map_lock_, 0); #endif if (evsig_global_setup_locks_(enable_locks) < 0) return -1; if (evutil_global_setup_locks_(enable_locks) < 0) return -1; if (evutil_secure_rng_global_setup_locks_(enable_locks) < 0) return -1; return 0; } #endif void event_base_assert_ok_(struct event_base *base) { EVBASE_ACQUIRE_LOCK(base, th_base_lock); event_base_assert_ok_nolock_(base); EVBASE_RELEASE_LOCK(base, th_base_lock); } void event_base_assert_ok_nolock_(struct event_base *base) { int i; int count; /* First do checks on the per-fd and per-signal lists */ evmap_check_integrity_(base); /* Check the heap property */ for (i = 1; i < (int)base->timeheap.n; ++i) { int parent = (i - 1) / 2; struct event *ev, *p_ev; ev = base->timeheap.p[i]; p_ev = base->timeheap.p[parent]; EVUTIL_ASSERT(ev->ev_flags & EVLIST_TIMEOUT); EVUTIL_ASSERT(evutil_timercmp(&p_ev->ev_timeout, &ev->ev_timeout, <=)); EVUTIL_ASSERT(ev->ev_timeout_pos.min_heap_idx == i); } /* Check that the common timeouts are fine */ for (i = 0; i < base->n_common_timeouts; ++i) { struct common_timeout_list *ctl = base->common_timeout_queues[i]; struct event *last=NULL, *ev; EVUTIL_ASSERT_TAILQ_OK(&ctl->events, event, ev_timeout_pos.ev_next_with_common_timeout); TAILQ_FOREACH(ev, &ctl->events, ev_timeout_pos.ev_next_with_common_timeout) { if (last) EVUTIL_ASSERT(evutil_timercmp(&last->ev_timeout, &ev->ev_timeout, <=)); EVUTIL_ASSERT(ev->ev_flags & EVLIST_TIMEOUT); EVUTIL_ASSERT(is_common_timeout(&ev->ev_timeout,base)); EVUTIL_ASSERT(COMMON_TIMEOUT_IDX(&ev->ev_timeout) == i); last = ev; } } /* Check the active queues. */ count = 0; for (i = 0; i < base->nactivequeues; ++i) { struct event_callback *evcb; EVUTIL_ASSERT_TAILQ_OK(&base->activequeues[i], event_callback, evcb_active_next); TAILQ_FOREACH(evcb, &base->activequeues[i], evcb_active_next) { EVUTIL_ASSERT((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) == EVLIST_ACTIVE); EVUTIL_ASSERT(evcb->evcb_pri == i); ++count; } } { struct event_callback *evcb; TAILQ_FOREACH(evcb, &base->active_later_queue, evcb_active_next) { EVUTIL_ASSERT((evcb->evcb_flags & (EVLIST_ACTIVE|EVLIST_ACTIVE_LATER)) == EVLIST_ACTIVE_LATER); ++count; } } EVUTIL_ASSERT(count == base->event_count_active); } /* * Copyright (c) 2009-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "evconfig-private.h" #ifndef _WIN32_WINNT /* Minimum required for InitializeCriticalSectionAndSpinCount */ #define _WIN32_WINNT 0x0403 #endif #include #include #include #include #include #include "event2/util.h" #include "util-internal.h" #include "iocp-internal.h" #include "log-internal.h" #include "mm-internal.h" #include "event-internal.h" #include "evthread-internal.h" #define NOTIFICATION_KEY ((ULONG_PTR)-1) void event_overlapped_init_(struct event_overlapped *o, iocp_callback cb) { memset(o, 0, sizeof(struct event_overlapped)); o->cb = cb; } static void handle_entry(OVERLAPPED *o, ULONG_PTR completion_key, DWORD nBytes, int ok) { struct event_overlapped *eo = EVUTIL_UPCAST(o, struct event_overlapped, overlapped); eo->cb(eo, completion_key, nBytes, ok); } static void loop(void *port_) { struct event_iocp_port *port = port_; long ms = port->ms; HANDLE p = port->port; if (ms <= 0) ms = INFINITE; while (1) { OVERLAPPED *overlapped=NULL; ULONG_PTR key=0; DWORD bytes=0; int ok = GetQueuedCompletionStatus(p, &bytes, &key, &overlapped, ms); EnterCriticalSection(&port->lock); if (port->shutdown) { if (--port->n_live_threads == 0) ReleaseSemaphore(port->shutdownSemaphore, 1, NULL); LeaveCriticalSection(&port->lock); return; } LeaveCriticalSection(&port->lock); if (key != NOTIFICATION_KEY && overlapped) handle_entry(overlapped, key, bytes, ok); else if (!overlapped) break; } event_warnx("GetQueuedCompletionStatus exited with no event."); EnterCriticalSection(&port->lock); if (--port->n_live_threads == 0) ReleaseSemaphore(port->shutdownSemaphore, 1, NULL); LeaveCriticalSection(&port->lock); } int event_iocp_port_associate_(struct event_iocp_port *port, evutil_socket_t fd, ev_uintptr_t key) { HANDLE h; h = CreateIoCompletionPort((HANDLE)fd, port->port, key, port->n_threads); if (!h) return -1; return 0; } static void * get_extension_function(SOCKET s, const GUID *which_fn) { void *ptr = NULL; DWORD bytes=0; WSAIoctl(s, SIO_GET_EXTENSION_FUNCTION_POINTER, (GUID*)which_fn, sizeof(*which_fn), &ptr, sizeof(ptr), &bytes, NULL, NULL); /* No need to detect errors here: if ptr is set, then we have a good function pointer. Otherwise, we should behave as if we had no function pointer. */ return ptr; } /* Mingw doesn't have these in its mswsock.h. The values are copied from wine.h. Perhaps if we copy them exactly, the cargo will come again. */ #ifndef WSAID_ACCEPTEX #define WSAID_ACCEPTEX \ {0xb5367df1,0xcbac,0x11cf,{0x95,0xca,0x00,0x80,0x5f,0x48,0xa1,0x92}} #endif #ifndef WSAID_CONNECTEX #define WSAID_CONNECTEX \ {0x25a207b9,0xddf3,0x4660,{0x8e,0xe9,0x76,0xe5,0x8c,0x74,0x06,0x3e}} #endif #ifndef WSAID_GETACCEPTEXSOCKADDRS #define WSAID_GETACCEPTEXSOCKADDRS \ {0xb5367df2,0xcbac,0x11cf,{0x95,0xca,0x00,0x80,0x5f,0x48,0xa1,0x92}} #endif static int extension_fns_initialized = 0; static void init_extension_functions(struct win32_extension_fns *ext) { const GUID acceptex = WSAID_ACCEPTEX; const GUID connectex = WSAID_CONNECTEX; const GUID getacceptexsockaddrs = WSAID_GETACCEPTEXSOCKADDRS; SOCKET s = socket(AF_INET, SOCK_STREAM, 0); if (s == EVUTIL_INVALID_SOCKET) return; ext->AcceptEx = get_extension_function(s, &acceptex); ext->ConnectEx = get_extension_function(s, &connectex); ext->GetAcceptExSockaddrs = get_extension_function(s, &getacceptexsockaddrs); closesocket(s); extension_fns_initialized = 1; } static struct win32_extension_fns the_extension_fns; const struct win32_extension_fns * event_get_win32_extension_fns_(void) { return &the_extension_fns; } #define N_CPUS_DEFAULT 2 struct event_iocp_port * event_iocp_port_launch_(int n_cpus) { struct event_iocp_port *port; int i; if (!extension_fns_initialized) init_extension_functions(&the_extension_fns); if (!(port = mm_calloc(1, sizeof(struct event_iocp_port)))) return NULL; if (n_cpus <= 0) n_cpus = N_CPUS_DEFAULT; port->n_threads = n_cpus * 2; port->threads = mm_calloc(port->n_threads, sizeof(HANDLE)); if (!port->threads) goto err; port->port = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, n_cpus); port->ms = -1; if (!port->port) goto err; port->shutdownSemaphore = CreateSemaphore(NULL, 0, 1, NULL); if (!port->shutdownSemaphore) goto err; for (i=0; in_threads; ++i) { ev_uintptr_t th = _beginthread(loop, 0, port); if (th == (ev_uintptr_t)-1) goto err; port->threads[i] = (HANDLE)th; ++port->n_live_threads; } InitializeCriticalSectionAndSpinCount(&port->lock, 1000); return port; err: if (port->port) CloseHandle(port->port); if (port->threads) mm_free(port->threads); if (port->shutdownSemaphore) CloseHandle(port->shutdownSemaphore); mm_free(port); return NULL; } static void event_iocp_port_unlock_and_free_(struct event_iocp_port *port) { DeleteCriticalSection(&port->lock); CloseHandle(port->port); CloseHandle(port->shutdownSemaphore); mm_free(port->threads); mm_free(port); } static int event_iocp_notify_all(struct event_iocp_port *port) { int i, r, ok=1; for (i=0; in_threads; ++i) { r = PostQueuedCompletionStatus(port->port, 0, NOTIFICATION_KEY, NULL); if (!r) ok = 0; } return ok ? 0 : -1; } int event_iocp_shutdown_(struct event_iocp_port *port, long waitMsec) { DWORD ms = INFINITE; int n; EnterCriticalSection(&port->lock); port->shutdown = 1; LeaveCriticalSection(&port->lock); event_iocp_notify_all(port); if (waitMsec >= 0) ms = waitMsec; WaitForSingleObject(port->shutdownSemaphore, ms); EnterCriticalSection(&port->lock); n = port->n_live_threads; LeaveCriticalSection(&port->lock); if (n == 0) { event_iocp_port_unlock_and_free_(port); return 0; } else { return -1; } } int event_iocp_activate_overlapped_( struct event_iocp_port *port, struct event_overlapped *o, ev_uintptr_t key, ev_uint32_t n) { BOOL r; r = PostQueuedCompletionStatus(port->port, n, key, &o->overlapped); return (r==0) ? -1 : 0; } struct event_iocp_port * event_base_get_iocp_(struct event_base *base) { #ifdef _WIN32 return base->iocp; #else return NULL; #endif } /* * Copyright (c) 2003-2009 Niels Provos * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_SYS_TYPES_H #include #endif #ifdef EVENT__HAVE_SYS_PARAM_H #include #endif #ifdef _WIN32 #define WIN32_LEAN_AND_MEAN #include #include #undef WIN32_LEAN_AND_MEAN #endif #ifdef EVENT__HAVE_SYS_IOCTL_H #include #endif #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #ifndef _WIN32 #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #include #include "event2/event.h" #include "event2/tag.h" #include "event2/buffer.h" #include "log-internal.h" #include "mm-internal.h" #include "util-internal.h" /* Here's our wire format: Stream = TaggedData* TaggedData = Tag Length Data where the integer value of 'Length' is the length of 'data'. Tag = HByte* LByte where HByte is a byte with the high bit set, and LByte is a byte with the high bit clear. The integer value of the tag is taken by concatenating the lower 7 bits from all the tags. So for example, the tag 0x66 is encoded as [66], whereas the tag 0x166 is encoded as [82 66] Length = Integer Integer = NNibbles Nibble* Padding? where NNibbles is a 4-bit value encoding the number of nibbles-1, and each Nibble is 4 bits worth of encoded integer, in big-endian order. If the total encoded integer size is an odd number of nibbles, a final padding nibble with value 0 is appended. */ EVENT2_EXPORT_SYMBOL int evtag_decode_int(ev_uint32_t *pnumber, struct evbuffer *evbuf); EVENT2_EXPORT_SYMBOL int evtag_decode_int64(ev_uint64_t *pnumber, struct evbuffer *evbuf); EVENT2_EXPORT_SYMBOL int evtag_encode_tag(struct evbuffer *evbuf, ev_uint32_t tag); EVENT2_EXPORT_SYMBOL int evtag_decode_tag(ev_uint32_t *ptag, struct evbuffer *evbuf); void evtag_init(void) { } /* * We encode integers by nibbles; the first nibble contains the number * of significant nibbles - 1; this allows us to encode up to 64-bit * integers. This function is byte-order independent. * * @param number a 32-bit unsigned integer to encode * @param data a pointer to where the data should be written. Must * have at least 5 bytes free. * @return the number of bytes written into data. */ #define ENCODE_INT_INTERNAL(data, number) do { \ int off = 1, nibbles = 0; \ \ memset(data, 0, sizeof(number)+1); \ while (number) { \ if (off & 0x1) \ data[off/2] = (data[off/2] & 0xf0) | (number & 0x0f); \ else \ data[off/2] = (data[off/2] & 0x0f) | \ ((number & 0x0f) << 4); \ number >>= 4; \ off++; \ } \ \ if (off > 2) \ nibbles = off - 2; \ \ /* Off - 1 is the number of encoded nibbles */ \ data[0] = (data[0] & 0x0f) | ((nibbles & 0x0f) << 4); \ \ return ((off + 1) / 2); \ } while (0) static inline int encode_int_internal(ev_uint8_t *data, ev_uint32_t number) { ENCODE_INT_INTERNAL(data, number); } static inline int encode_int64_internal(ev_uint8_t *data, ev_uint64_t number) { ENCODE_INT_INTERNAL(data, number); } void evtag_encode_int(struct evbuffer *evbuf, ev_uint32_t number) { ev_uint8_t data[5]; int len = encode_int_internal(data, number); evbuffer_add(evbuf, data, len); } void evtag_encode_int64(struct evbuffer *evbuf, ev_uint64_t number) { ev_uint8_t data[9]; int len = encode_int64_internal(data, number); evbuffer_add(evbuf, data, len); } /* * Support variable length encoding of tags; we use the high bit in each * octet as a continuation signal. */ int evtag_encode_tag(struct evbuffer *evbuf, ev_uint32_t tag) { int bytes = 0; ev_uint8_t data[5]; memset(data, 0, sizeof(data)); do { ev_uint8_t lower = tag & 0x7f; tag >>= 7; if (tag) lower |= 0x80; data[bytes++] = lower; } while (tag); if (evbuf != NULL) evbuffer_add(evbuf, data, bytes); return (bytes); } static int decode_tag_internal(ev_uint32_t *ptag, struct evbuffer *evbuf, int dodrain) { ev_uint32_t number = 0; size_t len = evbuffer_get_length(evbuf); ev_uint8_t *data; size_t count = 0; int shift = 0, done = 0; /* * the encoding of a number is at most one byte more than its * storage size. however, it may also be much smaller. */ data = evbuffer_pullup( evbuf, len < sizeof(number) + 1 ? len : sizeof(number) + 1); if (!data) return (-1); while (count++ < len) { ev_uint8_t lower = *data++; if (shift >= 28) { /* Make sure it fits into 32 bits */ if (shift > 28) return (-1); if ((lower & 0x7f) > 15) return (-1); } number |= (lower & (unsigned)0x7f) << shift; shift += 7; if (!(lower & 0x80)) { done = 1; break; } } if (!done) return (-1); if (dodrain) evbuffer_drain(evbuf, count); if (ptag != NULL) *ptag = number; return count > INT_MAX ? INT_MAX : (int)(count); } int evtag_decode_tag(ev_uint32_t *ptag, struct evbuffer *evbuf) { return (decode_tag_internal(ptag, evbuf, 1 /* dodrain */)); } /* * Marshal a data type, the general format is as follows: * * tag number: one byte; length: var bytes; payload: var bytes */ void evtag_marshal(struct evbuffer *evbuf, ev_uint32_t tag, const void *data, ev_uint32_t len) { evtag_encode_tag(evbuf, tag); evtag_encode_int(evbuf, len); evbuffer_add(evbuf, (void *)data, len); } void evtag_marshal_buffer(struct evbuffer *evbuf, ev_uint32_t tag, struct evbuffer *data) { evtag_encode_tag(evbuf, tag); /* XXX support more than UINT32_MAX data */ evtag_encode_int(evbuf, (ev_uint32_t)evbuffer_get_length(data)); evbuffer_add_buffer(evbuf, data); } /* Marshaling for integers */ void evtag_marshal_int(struct evbuffer *evbuf, ev_uint32_t tag, ev_uint32_t integer) { ev_uint8_t data[5]; int len = encode_int_internal(data, integer); evtag_encode_tag(evbuf, tag); evtag_encode_int(evbuf, len); evbuffer_add(evbuf, data, len); } void evtag_marshal_int64(struct evbuffer *evbuf, ev_uint32_t tag, ev_uint64_t integer) { ev_uint8_t data[9]; int len = encode_int64_internal(data, integer); evtag_encode_tag(evbuf, tag); evtag_encode_int(evbuf, len); evbuffer_add(evbuf, data, len); } void evtag_marshal_string(struct evbuffer *buf, ev_uint32_t tag, const char *string) { /* TODO support strings longer than UINT32_MAX ? */ evtag_marshal(buf, tag, string, (ev_uint32_t)strlen(string)); } void evtag_marshal_timeval(struct evbuffer *evbuf, ev_uint32_t tag, struct timeval *tv) { ev_uint8_t data[10]; int len = encode_int_internal(data, tv->tv_sec); len += encode_int_internal(data + len, tv->tv_usec); evtag_marshal(evbuf, tag, data, len); } #define DECODE_INT_INTERNAL(number, maxnibbles, pnumber, evbuf, offset) \ do { \ ev_uint8_t *data; \ ev_ssize_t len = evbuffer_get_length(evbuf) - offset; \ int nibbles = 0; \ \ if (len <= 0) \ return (-1); \ \ /* XXX(niels): faster? */ \ data = evbuffer_pullup(evbuf, offset + 1) + offset; \ if (!data) \ return (-1); \ \ nibbles = ((data[0] & 0xf0) >> 4) + 1; \ if (nibbles > maxnibbles || (nibbles >> 1) + 1 > len) \ return (-1); \ len = (nibbles >> 1) + 1; \ \ data = evbuffer_pullup(evbuf, offset + len) + offset; \ if (!data) \ return (-1); \ \ while (nibbles > 0) { \ number <<= 4; \ if (nibbles & 0x1) \ number |= data[nibbles >> 1] & 0x0f; \ else \ number |= (data[nibbles >> 1] & 0xf0) >> 4; \ nibbles--; \ } \ \ *pnumber = number; \ \ return (int)(len); \ } while (0) /* Internal: decode an integer from an evbuffer, without draining it. * Only integers up to 32-bits are supported. * * @param evbuf the buffer to read from * @param offset an index into the buffer at which we should start reading. * @param pnumber a pointer to receive the integer. * @return The length of the number as encoded, or -1 on error. */ static int decode_int_internal(ev_uint32_t *pnumber, struct evbuffer *evbuf, int offset) { ev_uint32_t number = 0; DECODE_INT_INTERNAL(number, 8, pnumber, evbuf, offset); } static int decode_int64_internal(ev_uint64_t *pnumber, struct evbuffer *evbuf, int offset) { ev_uint64_t number = 0; DECODE_INT_INTERNAL(number, 16, pnumber, evbuf, offset); } int evtag_decode_int(ev_uint32_t *pnumber, struct evbuffer *evbuf) { int res = decode_int_internal(pnumber, evbuf, 0); if (res != -1) evbuffer_drain(evbuf, res); return (res == -1 ? -1 : 0); } int evtag_decode_int64(ev_uint64_t *pnumber, struct evbuffer *evbuf) { int res = decode_int64_internal(pnumber, evbuf, 0); if (res != -1) evbuffer_drain(evbuf, res); return (res == -1 ? -1 : 0); } int evtag_peek(struct evbuffer *evbuf, ev_uint32_t *ptag) { return (decode_tag_internal(ptag, evbuf, 0 /* dodrain */)); } int evtag_peek_length(struct evbuffer *evbuf, ev_uint32_t *plength) { int res, len; len = decode_tag_internal(NULL, evbuf, 0 /* dodrain */); if (len == -1) return (-1); res = decode_int_internal(plength, evbuf, len); if (res == -1) return (-1); *plength += res + len; return (0); } int evtag_payload_length(struct evbuffer *evbuf, ev_uint32_t *plength) { int res, len; len = decode_tag_internal(NULL, evbuf, 0 /* dodrain */); if (len == -1) return (-1); res = decode_int_internal(plength, evbuf, len); if (res == -1) return (-1); return (0); } /* just unmarshals the header and returns the length of the remaining data */ int evtag_unmarshal_header(struct evbuffer *evbuf, ev_uint32_t *ptag) { ev_uint32_t len; if (decode_tag_internal(ptag, evbuf, 1 /* dodrain */) == -1) return (-1); if (evtag_decode_int(&len, evbuf) == -1) return (-1); if (evbuffer_get_length(evbuf) < len) return (-1); return (len); } int evtag_consume(struct evbuffer *evbuf) { int len; if ((len = evtag_unmarshal_header(evbuf, NULL)) == -1) return (-1); evbuffer_drain(evbuf, len); return (0); } /* Reads the data type from an event buffer */ int evtag_unmarshal(struct evbuffer *src, ev_uint32_t *ptag, struct evbuffer *dst) { int len; if ((len = evtag_unmarshal_header(src, ptag)) == -1) return (-1); if (evbuffer_add(dst, evbuffer_pullup(src, len), len) == -1) return (-1); evbuffer_drain(src, len); return (len); } /* Marshaling for integers */ int evtag_unmarshal_int(struct evbuffer *evbuf, ev_uint32_t need_tag, ev_uint32_t *pinteger) { ev_uint32_t tag; ev_uint32_t len; int result; if (decode_tag_internal(&tag, evbuf, 1 /* dodrain */) == -1) return (-1); if (need_tag != tag) return (-1); if (evtag_decode_int(&len, evbuf) == -1) return (-1); if (evbuffer_get_length(evbuf) < len) return (-1); result = decode_int_internal(pinteger, evbuf, 0); evbuffer_drain(evbuf, len); if (result < 0 || (size_t)result > len) /* XXX Should this be != rather than > ?*/ return (-1); else return result; } int evtag_unmarshal_int64(struct evbuffer *evbuf, ev_uint32_t need_tag, ev_uint64_t *pinteger) { ev_uint32_t tag; ev_uint32_t len; int result; if (decode_tag_internal(&tag, evbuf, 1 /* dodrain */) == -1) return (-1); if (need_tag != tag) return (-1); if (evtag_decode_int(&len, evbuf) == -1) return (-1); if (evbuffer_get_length(evbuf) < len) return (-1); result = decode_int64_internal(pinteger, evbuf, 0); evbuffer_drain(evbuf, len); if (result < 0 || (size_t)result > len) /* XXX Should this be != rather than > ?*/ return (-1); else return result; } /* Unmarshal a fixed length tag */ int evtag_unmarshal_fixed(struct evbuffer *src, ev_uint32_t need_tag, void *data, size_t len) { ev_uint32_t tag; int tag_len; /* Now unmarshal a tag and check that it matches the tag we want */ if ((tag_len = evtag_unmarshal_header(src, &tag)) < 0 || tag != need_tag) return (-1); if ((size_t)tag_len != len) return (-1); evbuffer_remove(src, data, len); return (0); } int evtag_unmarshal_string(struct evbuffer *evbuf, ev_uint32_t need_tag, char **pstring) { ev_uint32_t tag; int tag_len; if ((tag_len = evtag_unmarshal_header(evbuf, &tag)) == -1 || tag != need_tag) return (-1); *pstring = mm_malloc(tag_len + 1); if (*pstring == NULL) { event_warn("%s: malloc", __func__); return -1; } evbuffer_remove(evbuf, *pstring, tag_len); (*pstring)[tag_len] = '\0'; return (0); } int evtag_unmarshal_timeval(struct evbuffer *evbuf, ev_uint32_t need_tag, struct timeval *ptv) { ev_uint32_t tag; ev_uint32_t integer; int len, offset, offset2; int result = -1; if ((len = evtag_unmarshal_header(evbuf, &tag)) == -1) return (-1); if (tag != need_tag) goto done; if ((offset = decode_int_internal(&integer, evbuf, 0)) == -1) goto done; ptv->tv_sec = integer; if ((offset2 = decode_int_internal(&integer, evbuf, offset)) == -1) goto done; ptv->tv_usec = integer; if (offset + offset2 > len) /* XXX Should this be != instead of > ? */ goto done; result = 0; done: evbuffer_drain(evbuf, len); return result; } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #endif #include #if !defined(_WIN32) && defined(EVENT__HAVE_SYS_TIME_H) #include #endif #include #include #include #ifndef _WIN32 #include #endif #include #include #include #include #include #include "event-internal.h" #include "evmap-internal.h" #include "mm-internal.h" #include "changelist-internal.h" /** An entry for an evmap_io list: notes all the events that want to read or write on a given fd, and the number of each. */ struct evmap_io { struct event_dlist events; ev_uint16_t nread; ev_uint16_t nwrite; ev_uint16_t nclose; }; /* An entry for an evmap_signal list: notes all the events that want to know when a signal triggers. */ struct evmap_signal { struct event_dlist events; }; /* On some platforms, fds start at 0 and increment by 1 as they are allocated, and old numbers get used. For these platforms, we implement io maps just like signal maps: as an array of pointers to struct evmap_io. But on other platforms (windows), sockets are not 0-indexed, not necessarily consecutive, and not necessarily reused. There, we use a hashtable to implement evmap_io. */ #ifdef EVMAP_USE_HT struct event_map_entry { HT_ENTRY(event_map_entry) map_node; evutil_socket_t fd; union { /* This is a union in case we need to make more things that can be in the hashtable. */ struct evmap_io evmap_io; } ent; }; /* Helper used by the event_io_map hashtable code; tries to return a good hash * of the fd in e->fd. */ static inline unsigned hashsocket(struct event_map_entry *e) { /* On win32, in practice, the low 2-3 bits of a SOCKET seem not to * matter. Our hashtable implementation really likes low-order bits, * though, so let's do the rotate-and-add trick. */ unsigned h = (unsigned) e->fd; h += (h >> 2) | (h << 30); return h; } /* Helper used by the event_io_map hashtable code; returns true iff e1 and e2 * have the same e->fd. */ static inline int eqsocket(struct event_map_entry *e1, struct event_map_entry *e2) { return e1->fd == e2->fd; } HT_PROTOTYPE(event_io_map, event_map_entry, map_node, hashsocket, eqsocket) HT_GENERATE(event_io_map, event_map_entry, map_node, hashsocket, eqsocket, 0.5, mm_malloc, mm_realloc, mm_free) #define GET_IO_SLOT(x, map, slot, type) \ do { \ struct event_map_entry key_, *ent_; \ key_.fd = slot; \ ent_ = HT_FIND(event_io_map, map, &key_); \ (x) = ent_ ? &ent_->ent.type : NULL; \ } while (0); #define GET_IO_SLOT_AND_CTOR(x, map, slot, type, ctor, fdinfo_len) \ do { \ struct event_map_entry key_, *ent_; \ key_.fd = slot; \ HT_FIND_OR_INSERT_(event_io_map, map_node, hashsocket, map, \ event_map_entry, &key_, ptr, \ { \ ent_ = *ptr; \ }, \ { \ ent_ = mm_calloc(1,sizeof(struct event_map_entry)+fdinfo_len); \ if (EVUTIL_UNLIKELY(ent_ == NULL)) \ return (-1); \ ent_->fd = slot; \ (ctor)(&ent_->ent.type); \ HT_FOI_INSERT_(map_node, map, &key_, ent_, ptr) \ }); \ (x) = &ent_->ent.type; \ } while (0) void evmap_io_initmap_(struct event_io_map *ctx) { HT_INIT(event_io_map, ctx); } void evmap_io_clear_(struct event_io_map *ctx) { struct event_map_entry **ent, **next, *this; for (ent = HT_START(event_io_map, ctx); ent; ent = next) { this = *ent; next = HT_NEXT_RMV(event_io_map, ctx, ent); mm_free(this); } HT_CLEAR(event_io_map, ctx); /* remove all storage held by the ctx. */ } #endif /* Set the variable 'x' to the field in event_map 'map' with fields of type 'struct type *' corresponding to the fd or signal 'slot'. Set 'x' to NULL if there are no entries for 'slot'. Does no bounds-checking. */ #define GET_SIGNAL_SLOT(x, map, slot, type) \ (x) = (struct type *)((map)->entries[slot]) /* As GET_SLOT, but construct the entry for 'slot' if it is not present, by allocating enough memory for a 'struct type', and initializing the new value by calling the function 'ctor' on it. Makes the function return -1 on allocation failure. */ #define GET_SIGNAL_SLOT_AND_CTOR(x, map, slot, type, ctor, fdinfo_len) \ do { \ if ((map)->entries[slot] == NULL) { \ (map)->entries[slot] = \ mm_calloc(1,sizeof(struct type)+fdinfo_len); \ if (EVUTIL_UNLIKELY((map)->entries[slot] == NULL)) \ return (-1); \ (ctor)((struct type *)(map)->entries[slot]); \ } \ (x) = (struct type *)((map)->entries[slot]); \ } while (0) /* If we aren't using hashtables, then define the IO_SLOT macros and functions as thin aliases over the SIGNAL_SLOT versions. */ #ifndef EVMAP_USE_HT #define GET_IO_SLOT(x,map,slot,type) GET_SIGNAL_SLOT(x,map,slot,type) #define GET_IO_SLOT_AND_CTOR(x,map,slot,type,ctor,fdinfo_len) \ GET_SIGNAL_SLOT_AND_CTOR(x,map,slot,type,ctor,fdinfo_len) #define FDINFO_OFFSET sizeof(struct evmap_io) void evmap_io_initmap_(struct event_io_map* ctx) { evmap_signal_initmap_(ctx); } void evmap_io_clear_(struct event_io_map* ctx) { evmap_signal_clear_(ctx); } #endif /** Expand 'map' with new entries of width 'msize' until it is big enough to store a value in 'slot'. */ static int evmap_make_space(struct event_signal_map *map, int slot, int msize) { if (map->nentries <= slot) { int nentries = map->nentries ? map->nentries : 32; void **tmp; if (slot > INT_MAX / 2) return (-1); while (nentries <= slot) nentries <<= 1; if (nentries > INT_MAX / msize) return (-1); tmp = (void **)mm_realloc(map->entries, nentries * msize); if (tmp == NULL) return (-1); memset(&tmp[map->nentries], 0, (nentries - map->nentries) * msize); map->nentries = nentries; map->entries = tmp; } return (0); } void evmap_signal_initmap_(struct event_signal_map *ctx) { ctx->nentries = 0; ctx->entries = NULL; } void evmap_signal_clear_(struct event_signal_map *ctx) { if (ctx->entries != NULL) { int i; for (i = 0; i < ctx->nentries; ++i) { if (ctx->entries[i] != NULL) mm_free(ctx->entries[i]); } mm_free(ctx->entries); ctx->entries = NULL; } ctx->nentries = 0; } /* code specific to file descriptors */ /** Constructor for struct evmap_io */ static void evmap_io_init(struct evmap_io *entry) { LIST_INIT(&entry->events); entry->nread = 0; entry->nwrite = 0; entry->nclose = 0; } /* return -1 on error, 0 on success if nothing changed in the event backend, * and 1 on success if something did. */ int evmap_io_add_(struct event_base *base, evutil_socket_t fd, struct event *ev) { const struct eventop *evsel = base->evsel; struct event_io_map *io = &base->io; struct evmap_io *ctx = NULL; int nread, nwrite, nclose, retval = 0; short res = 0, old = 0; struct event *old_ev; EVUTIL_ASSERT(fd == ev->ev_fd); if (fd < 0) return 0; #ifndef EVMAP_USE_HT if (fd >= io->nentries) { if (evmap_make_space(io, fd, sizeof(struct evmap_io *)) == -1) return (-1); } #endif GET_IO_SLOT_AND_CTOR(ctx, io, fd, evmap_io, evmap_io_init, evsel->fdinfo_len); nread = ctx->nread; nwrite = ctx->nwrite; nclose = ctx->nclose; if (nread) old |= EV_READ; if (nwrite) old |= EV_WRITE; if (nclose) old |= EV_CLOSED; if (ev->ev_events & EV_READ) { if (++nread == 1) res |= EV_READ; } if (ev->ev_events & EV_WRITE) { if (++nwrite == 1) res |= EV_WRITE; } if (ev->ev_events & EV_CLOSED) { if (++nclose == 1) res |= EV_CLOSED; } if (EVUTIL_UNLIKELY(nread > 0xffff || nwrite > 0xffff || nclose > 0xffff)) { event_warnx("Too many events reading or writing on fd %d", (int)fd); return -1; } if (EVENT_DEBUG_MODE_IS_ON() && (old_ev = LIST_FIRST(&ctx->events)) && (old_ev->ev_events&EV_ET) != (ev->ev_events&EV_ET)) { event_warnx("Tried to mix edge-triggered and non-edge-triggered" " events on fd %d", (int)fd); return -1; } if (res) { void *extra = ((char*)ctx) + sizeof(struct evmap_io); /* XXX(niels): we cannot mix edge-triggered and * level-triggered, we should probably assert on * this. */ if (evsel->add(base, ev->ev_fd, old, (ev->ev_events & EV_ET) | res, extra) == -1) return (-1); retval = 1; } ctx->nread = (ev_uint16_t) nread; ctx->nwrite = (ev_uint16_t) nwrite; ctx->nclose = (ev_uint16_t) nclose; LIST_INSERT_HEAD(&ctx->events, ev, ev_io_next); return (retval); } /* return -1 on error, 0 on success if nothing changed in the event backend, * and 1 on success if something did. */ int evmap_io_del_(struct event_base *base, evutil_socket_t fd, struct event *ev) { const struct eventop *evsel = base->evsel; struct event_io_map *io = &base->io; struct evmap_io *ctx; int nread, nwrite, nclose, retval = 0; short res = 0, old = 0; if (fd < 0) return 0; EVUTIL_ASSERT(fd == ev->ev_fd); #ifndef EVMAP_USE_HT if (fd >= io->nentries) return (-1); #endif GET_IO_SLOT(ctx, io, fd, evmap_io); nread = ctx->nread; nwrite = ctx->nwrite; nclose = ctx->nclose; if (nread) old |= EV_READ; if (nwrite) old |= EV_WRITE; if (nclose) old |= EV_CLOSED; if (ev->ev_events & EV_READ) { if (--nread == 0) res |= EV_READ; EVUTIL_ASSERT(nread >= 0); } if (ev->ev_events & EV_WRITE) { if (--nwrite == 0) res |= EV_WRITE; EVUTIL_ASSERT(nwrite >= 0); } if (ev->ev_events & EV_CLOSED) { if (--nclose == 0) res |= EV_CLOSED; EVUTIL_ASSERT(nclose >= 0); } if (res) { void *extra = ((char*)ctx) + sizeof(struct evmap_io); if (evsel->del(base, ev->ev_fd, old, (ev->ev_events & EV_ET) | res, extra) == -1) { retval = -1; } else { retval = 1; } } ctx->nread = nread; ctx->nwrite = nwrite; ctx->nclose = nclose; LIST_REMOVE(ev, ev_io_next); return (retval); } void evmap_io_active_(struct event_base *base, evutil_socket_t fd, short events) { struct event_io_map *io = &base->io; struct evmap_io *ctx; struct event *ev; #ifndef EVMAP_USE_HT if (fd < 0 || fd >= io->nentries) return; #endif GET_IO_SLOT(ctx, io, fd, evmap_io); if (NULL == ctx) return; LIST_FOREACH(ev, &ctx->events, ev_io_next) { if (ev->ev_events & (events & ~EV_ET)) event_active_nolock_(ev, ev->ev_events & events, 1); } } /* code specific to signals */ static void evmap_signal_init(struct evmap_signal *entry) { LIST_INIT(&entry->events); } int evmap_signal_add_(struct event_base *base, int sig, struct event *ev) { const struct eventop *evsel = base->evsigsel; struct event_signal_map *map = &base->sigmap; struct evmap_signal *ctx = NULL; if (sig < 0 || sig >= NSIG) return (-1); if (sig >= map->nentries) { if (evmap_make_space( map, sig, sizeof(struct evmap_signal *)) == -1) return (-1); } GET_SIGNAL_SLOT_AND_CTOR(ctx, map, sig, evmap_signal, evmap_signal_init, base->evsigsel->fdinfo_len); if (LIST_EMPTY(&ctx->events)) { if (evsel->add(base, ev->ev_fd, 0, EV_SIGNAL, NULL) == -1) return (-1); } LIST_INSERT_HEAD(&ctx->events, ev, ev_signal_next); return (1); } int evmap_signal_del_(struct event_base *base, int sig, struct event *ev) { const struct eventop *evsel = base->evsigsel; struct event_signal_map *map = &base->sigmap; struct evmap_signal *ctx; if (sig < 0 || sig >= map->nentries) return (-1); GET_SIGNAL_SLOT(ctx, map, sig, evmap_signal); LIST_REMOVE(ev, ev_signal_next); if (LIST_FIRST(&ctx->events) == NULL) { if (evsel->del(base, ev->ev_fd, 0, EV_SIGNAL, NULL) == -1) return (-1); } return (1); } void evmap_signal_active_(struct event_base *base, evutil_socket_t sig, int ncalls) { struct event_signal_map *map = &base->sigmap; struct evmap_signal *ctx; struct event *ev; if (sig < 0 || sig >= map->nentries) return; GET_SIGNAL_SLOT(ctx, map, sig, evmap_signal); if (!ctx) return; LIST_FOREACH(ev, &ctx->events, ev_signal_next) event_active_nolock_(ev, EV_SIGNAL, ncalls); } void * evmap_io_get_fdinfo_(struct event_io_map *map, evutil_socket_t fd) { struct evmap_io *ctx; GET_IO_SLOT(ctx, map, fd, evmap_io); if (ctx) return ((char*)ctx) + sizeof(struct evmap_io); else return NULL; } /* Callback type for evmap_io_foreach_fd */ typedef int (*evmap_io_foreach_fd_cb)( struct event_base *, evutil_socket_t, struct evmap_io *, void *); /* Multipurpose helper function: Iterate over every file descriptor event_base * for which we could have EV_READ or EV_WRITE events. For each such fd, call * fn(base, signum, evmap_io, arg), where fn is the user-provided * function, base is the event_base, signum is the signal number, evmap_io * is an evmap_io structure containing a list of events pending on the * file descriptor, and arg is the user-supplied argument. * * If fn returns 0, continue on to the next signal. Otherwise, return the same * value that fn returned. * * Note that there is no guarantee that the file descriptors will be processed * in any particular order. */ static int evmap_io_foreach_fd(struct event_base *base, evmap_io_foreach_fd_cb fn, void *arg) { evutil_socket_t fd; struct event_io_map *iomap = &base->io; int r = 0; #ifdef EVMAP_USE_HT struct event_map_entry **mapent; HT_FOREACH(mapent, event_io_map, iomap) { struct evmap_io *ctx = &(*mapent)->ent.evmap_io; fd = (*mapent)->fd; #else for (fd = 0; fd < iomap->nentries; ++fd) { struct evmap_io *ctx = iomap->entries[fd]; if (!ctx) continue; #endif if ((r = fn(base, fd, ctx, arg))) break; } return r; } /* Callback type for evmap_signal_foreach_signal */ typedef int (*evmap_signal_foreach_signal_cb)( struct event_base *, int, struct evmap_signal *, void *); /* Multipurpose helper function: Iterate over every signal number in the * event_base for which we could have signal events. For each such signal, * call fn(base, signum, evmap_signal, arg), where fn is the user-provided * function, base is the event_base, signum is the signal number, evmap_signal * is an evmap_signal structure containing a list of events pending on the * signal, and arg is the user-supplied argument. * * If fn returns 0, continue on to the next signal. Otherwise, return the same * value that fn returned. */ static int evmap_signal_foreach_signal(struct event_base *base, evmap_signal_foreach_signal_cb fn, void *arg) { struct event_signal_map *sigmap = &base->sigmap; int r = 0; int signum; for (signum = 0; signum < sigmap->nentries; ++signum) { struct evmap_signal *ctx = sigmap->entries[signum]; if (!ctx) continue; if ((r = fn(base, signum, ctx, arg))) break; } return r; } /* Helper for evmap_reinit_: tell the backend to add every fd for which we have * pending events, with the appropriate combination of EV_READ, EV_WRITE, and * EV_ET. */ static int evmap_io_reinit_iter_fn(struct event_base *base, evutil_socket_t fd, struct evmap_io *ctx, void *arg) { const struct eventop *evsel = base->evsel; void *extra; int *result = arg; short events = 0; struct event *ev; EVUTIL_ASSERT(ctx); extra = ((char*)ctx) + sizeof(struct evmap_io); if (ctx->nread) events |= EV_READ; if (ctx->nwrite) events |= EV_WRITE; if (ctx->nclose) events |= EV_CLOSED; if (evsel->fdinfo_len) memset(extra, 0, evsel->fdinfo_len); if (events && (ev = LIST_FIRST(&ctx->events)) && (ev->ev_events & EV_ET)) events |= EV_ET; if (evsel->add(base, fd, 0, events, extra) == -1) *result = -1; return 0; } /* Helper for evmap_reinit_: tell the backend to add every signal for which we * have pending events. */ static int evmap_signal_reinit_iter_fn(struct event_base *base, int signum, struct evmap_signal *ctx, void *arg) { const struct eventop *evsel = base->evsigsel; int *result = arg; if (!LIST_EMPTY(&ctx->events)) { if (evsel->add(base, signum, 0, EV_SIGNAL, NULL) == -1) *result = -1; } return 0; } int evmap_reinit_(struct event_base *base) { int result = 0; evmap_io_foreach_fd(base, evmap_io_reinit_iter_fn, &result); if (result < 0) return -1; evmap_signal_foreach_signal(base, evmap_signal_reinit_iter_fn, &result); if (result < 0) return -1; return 0; } /* Helper for evmap_delete_all_: delete every event in an event_dlist. */ static int delete_all_in_dlist(struct event_dlist *dlist) { struct event *ev; while ((ev = LIST_FIRST(dlist))) event_del(ev); return 0; } /* Helper for evmap_delete_all_: delete every event pending on an fd. */ static int evmap_io_delete_all_iter_fn(struct event_base *base, evutil_socket_t fd, struct evmap_io *io_info, void *arg) { return delete_all_in_dlist(&io_info->events); } /* Helper for evmap_delete_all_: delete every event pending on a signal. */ static int evmap_signal_delete_all_iter_fn(struct event_base *base, int signum, struct evmap_signal *sig_info, void *arg) { return delete_all_in_dlist(&sig_info->events); } void evmap_delete_all_(struct event_base *base) { evmap_signal_foreach_signal(base, evmap_signal_delete_all_iter_fn, NULL); evmap_io_foreach_fd(base, evmap_io_delete_all_iter_fn, NULL); } /** Per-fd structure for use with changelists. It keeps track, for each fd or * signal using the changelist, of where its entry in the changelist is. */ struct event_changelist_fdinfo { int idxplus1; /* this is the index +1, so that memset(0) will make it * a no-such-element */ }; void event_changelist_init_(struct event_changelist *changelist) { changelist->changes = NULL; changelist->changes_size = 0; changelist->n_changes = 0; } /** Helper: return the changelist_fdinfo corresponding to a given change. */ static inline struct event_changelist_fdinfo * event_change_get_fdinfo(struct event_base *base, const struct event_change *change) { char *ptr; if (change->read_change & EV_CHANGE_SIGNAL) { struct evmap_signal *ctx; GET_SIGNAL_SLOT(ctx, &base->sigmap, change->fd, evmap_signal); ptr = ((char*)ctx) + sizeof(struct evmap_signal); } else { struct evmap_io *ctx; GET_IO_SLOT(ctx, &base->io, change->fd, evmap_io); ptr = ((char*)ctx) + sizeof(struct evmap_io); } return (void*)ptr; } /** Callback helper for event_changelist_assert_ok */ static int event_changelist_assert_ok_foreach_iter_fn( struct event_base *base, evutil_socket_t fd, struct evmap_io *io, void *arg) { struct event_changelist *changelist = &base->changelist; struct event_changelist_fdinfo *f; f = (void*) ( ((char*)io) + sizeof(struct evmap_io) ); if (f->idxplus1) { struct event_change *c = &changelist->changes[f->idxplus1 - 1]; EVUTIL_ASSERT(c->fd == fd); } return 0; } /** Make sure that the changelist is consistent with the evmap structures. */ static void event_changelist_assert_ok(struct event_base *base) { int i; struct event_changelist *changelist = &base->changelist; EVUTIL_ASSERT(changelist->changes_size >= changelist->n_changes); for (i = 0; i < changelist->n_changes; ++i) { struct event_change *c = &changelist->changes[i]; struct event_changelist_fdinfo *f; EVUTIL_ASSERT(c->fd >= 0); f = event_change_get_fdinfo(base, c); EVUTIL_ASSERT(f); EVUTIL_ASSERT(f->idxplus1 == i + 1); } evmap_io_foreach_fd(base, event_changelist_assert_ok_foreach_iter_fn, NULL); } #ifdef DEBUG_CHANGELIST #define event_changelist_check(base) event_changelist_assert_ok((base)) #else #define event_changelist_check(base) ((void)0) #endif void event_changelist_remove_all_(struct event_changelist *changelist, struct event_base *base) { int i; event_changelist_check(base); for (i = 0; i < changelist->n_changes; ++i) { struct event_change *ch = &changelist->changes[i]; struct event_changelist_fdinfo *fdinfo = event_change_get_fdinfo(base, ch); EVUTIL_ASSERT(fdinfo->idxplus1 == i + 1); fdinfo->idxplus1 = 0; } changelist->n_changes = 0; event_changelist_check(base); } void event_changelist_freemem_(struct event_changelist *changelist) { if (changelist->changes) mm_free(changelist->changes); event_changelist_init_(changelist); /* zero it all out. */ } /** Increase the size of 'changelist' to hold more changes. */ static int event_changelist_grow(struct event_changelist *changelist) { int new_size; struct event_change *new_changes; if (changelist->changes_size < 64) new_size = 64; else new_size = changelist->changes_size * 2; new_changes = mm_realloc(changelist->changes, new_size * sizeof(struct event_change)); if (EVUTIL_UNLIKELY(new_changes == NULL)) return (-1); changelist->changes = new_changes; changelist->changes_size = new_size; return (0); } /** Return a pointer to the changelist entry for the file descriptor or signal * 'fd', whose fdinfo is 'fdinfo'. If none exists, construct it, setting its * old_events field to old_events. */ static struct event_change * event_changelist_get_or_construct(struct event_changelist *changelist, evutil_socket_t fd, short old_events, struct event_changelist_fdinfo *fdinfo) { struct event_change *change; if (fdinfo->idxplus1 == 0) { int idx; EVUTIL_ASSERT(changelist->n_changes <= changelist->changes_size); if (changelist->n_changes == changelist->changes_size) { if (event_changelist_grow(changelist) < 0) return NULL; } idx = changelist->n_changes++; change = &changelist->changes[idx]; fdinfo->idxplus1 = idx + 1; memset(change, 0, sizeof(struct event_change)); change->fd = fd; change->old_events = old_events; } else { change = &changelist->changes[fdinfo->idxplus1 - 1]; EVUTIL_ASSERT(change->fd == fd); } return change; } int event_changelist_add_(struct event_base *base, evutil_socket_t fd, short old, short events, void *p) { struct event_changelist *changelist = &base->changelist; struct event_changelist_fdinfo *fdinfo = p; struct event_change *change; ev_uint8_t evchange = EV_CHANGE_ADD | (events & (EV_ET|EV_PERSIST|EV_SIGNAL)); event_changelist_check(base); change = event_changelist_get_or_construct(changelist, fd, old, fdinfo); if (!change) return -1; /* An add replaces any previous delete, but doesn't result in a no-op, * since the delete might fail (because the fd had been closed since * the last add, for instance. */ if (events & (EV_READ|EV_SIGNAL)) change->read_change = evchange; if (events & EV_WRITE) change->write_change = evchange; if (events & EV_CLOSED) change->close_change = evchange; event_changelist_check(base); return (0); } int event_changelist_del_(struct event_base *base, evutil_socket_t fd, short old, short events, void *p) { struct event_changelist *changelist = &base->changelist; struct event_changelist_fdinfo *fdinfo = p; struct event_change *change; ev_uint8_t del = EV_CHANGE_DEL | (events & EV_ET); event_changelist_check(base); change = event_changelist_get_or_construct(changelist, fd, old, fdinfo); event_changelist_check(base); if (!change) return -1; /* A delete on an event set that doesn't contain the event to be deleted produces a no-op. This effectively emoves any previous uncommitted add, rather than replacing it: on those platforms where "add, delete, dispatch" is not the same as "no-op, dispatch", we want the no-op behavior. If we have a no-op item, we could remove it it from the list entirely, but really there's not much point: skipping the no-op change when we do the dispatch later is far cheaper than rejuggling the array now. As this stands, it also lets through deletions of events that are not currently set. */ if (events & (EV_READ|EV_SIGNAL)) { if (!(change->old_events & (EV_READ | EV_SIGNAL))) change->read_change = 0; else change->read_change = del; } if (events & EV_WRITE) { if (!(change->old_events & EV_WRITE)) change->write_change = 0; else change->write_change = del; } if (events & EV_CLOSED) { if (!(change->old_events & EV_CLOSED)) change->close_change = 0; else change->close_change = del; } event_changelist_check(base); return (0); } /* Helper for evmap_check_integrity_: verify that all of the events pending on * given fd are set up correctly, and that the nread and nwrite counts on that * fd are correct. */ static int evmap_io_check_integrity_fn(struct event_base *base, evutil_socket_t fd, struct evmap_io *io_info, void *arg) { struct event *ev; int n_read = 0, n_write = 0, n_close = 0; /* First, make sure the list itself isn't corrupt. Otherwise, * running LIST_FOREACH could be an exciting adventure. */ EVUTIL_ASSERT_LIST_OK(&io_info->events, event, ev_io_next); LIST_FOREACH(ev, &io_info->events, ev_io_next) { EVUTIL_ASSERT(ev->ev_flags & EVLIST_INSERTED); EVUTIL_ASSERT(ev->ev_fd == fd); EVUTIL_ASSERT(!(ev->ev_events & EV_SIGNAL)); EVUTIL_ASSERT((ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED))); if (ev->ev_events & EV_READ) ++n_read; if (ev->ev_events & EV_WRITE) ++n_write; if (ev->ev_events & EV_CLOSED) ++n_close; } EVUTIL_ASSERT(n_read == io_info->nread); EVUTIL_ASSERT(n_write == io_info->nwrite); EVUTIL_ASSERT(n_close == io_info->nclose); return 0; } /* Helper for evmap_check_integrity_: verify that all of the events pending * on given signal are set up correctly. */ static int evmap_signal_check_integrity_fn(struct event_base *base, int signum, struct evmap_signal *sig_info, void *arg) { struct event *ev; /* First, make sure the list itself isn't corrupt. */ EVUTIL_ASSERT_LIST_OK(&sig_info->events, event, ev_signal_next); LIST_FOREACH(ev, &sig_info->events, ev_io_next) { EVUTIL_ASSERT(ev->ev_flags & EVLIST_INSERTED); EVUTIL_ASSERT(ev->ev_fd == signum); EVUTIL_ASSERT((ev->ev_events & EV_SIGNAL)); EVUTIL_ASSERT(!(ev->ev_events & (EV_READ|EV_WRITE|EV_CLOSED))); } return 0; } void evmap_check_integrity_(struct event_base *base) { evmap_io_foreach_fd(base, evmap_io_check_integrity_fn, NULL); evmap_signal_foreach_signal(base, evmap_signal_check_integrity_fn, NULL); if (base->evsel->add == event_changelist_add_) event_changelist_assert_ok(base); } /* Helper type for evmap_foreach_event_: Bundles a function to call on every * event, and the user-provided void* to use as its third argument. */ struct evmap_foreach_event_helper { event_base_foreach_event_cb fn; void *arg; }; /* Helper for evmap_foreach_event_: calls a provided function on every event * pending on a given fd. */ static int evmap_io_foreach_event_fn(struct event_base *base, evutil_socket_t fd, struct evmap_io *io_info, void *arg) { struct evmap_foreach_event_helper *h = arg; struct event *ev; int r; LIST_FOREACH(ev, &io_info->events, ev_io_next) { if ((r = h->fn(base, ev, h->arg))) return r; } return 0; } /* Helper for evmap_foreach_event_: calls a provided function on every event * pending on a given signal. */ static int evmap_signal_foreach_event_fn(struct event_base *base, int signum, struct evmap_signal *sig_info, void *arg) { struct event *ev; struct evmap_foreach_event_helper *h = arg; int r; LIST_FOREACH(ev, &sig_info->events, ev_signal_next) { if ((r = h->fn(base, ev, h->arg))) return r; } return 0; } int evmap_foreach_event_(struct event_base *base, event_base_foreach_event_cb fn, void *arg) { struct evmap_foreach_event_helper h; int r; h.fn = fn; h.arg = arg; if ((r = evmap_io_foreach_fd(base, evmap_io_foreach_event_fn, &h))) return r; return evmap_signal_foreach_signal(base, evmap_signal_foreach_event_fn, &h); } /* * Submitted by David Pacheco (dp.spambait@gmail.com) * * Copyright 2006-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY SUN MICROSYSTEMS, INC. ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL SUN MICROSYSTEMS, INC. BE LIABLE FOR ANY * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* * Copyright (c) 2007 Sun Microsystems. All rights reserved. * Use is subject to license terms. */ /* * evport.c: event backend using Solaris 10 event ports. See port_create(3C). * This implementation is loosely modeled after the one used for select(2) (in * select.c). * * The outstanding events are tracked in a data structure called evport_data. * Each entry in the ed_fds array corresponds to a file descriptor, and contains * pointers to the read and write events that correspond to that fd. (That is, * when the file is readable, the "read" event should handle it, etc.) * * evport_add and evport_del update this data structure. evport_dispatch uses it * to determine where to callback when an event occurs (which it gets from * port_getn). * * Helper functions are used: grow() grows the file descriptor array as * necessary when large fd's come in. reassociate() takes care of maintaining * the proper file-descriptor/event-port associations. * * As in the select(2) implementation, signals are handled by evsignal. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_EVENT_PORTS #include #include #include #include #include #include #include #include #include #include #include #include "event2/thread.h" #include "evthread-internal.h" #include "event-internal.h" #include "log-internal.h" #include "evsignal-internal.h" #include "evmap-internal.h" #define INITIAL_EVENTS_PER_GETN 8 #define MAX_EVENTS_PER_GETN 4096 /* * Per-file-descriptor information about what events we're subscribed to. These * fields are NULL if no event is subscribed to either of them. */ struct fd_info { /* combinations of EV_READ and EV_WRITE */ short fdi_what; /* Index of this fd within ed_pending, plus 1. Zero if this fd is * not in ed_pending. (The +1 is a hack so that memset(0) will set * it to a nil index. */ int pending_idx_plus_1; }; #define FDI_HAS_READ(fdi) ((fdi)->fdi_what & EV_READ) #define FDI_HAS_WRITE(fdi) ((fdi)->fdi_what & EV_WRITE) #define FDI_HAS_EVENTS(fdi) (FDI_HAS_READ(fdi) || FDI_HAS_WRITE(fdi)) #define FDI_TO_SYSEVENTS(fdi) (FDI_HAS_READ(fdi) ? POLLIN : 0) | \ (FDI_HAS_WRITE(fdi) ? POLLOUT : 0) struct evport_data { int ed_port; /* event port for system events */ /* How many elements of ed_pending should we look at? */ int ed_npending; /* How many elements are allocated in ed_pending and pevtlist? */ int ed_maxevents; /* fdi's that we need to reassoc */ int *ed_pending; /* storage space for incoming events. */ port_event_t *ed_pevtlist; }; static void* evport_init(struct event_base *); static int evport_add(struct event_base *, int fd, short old, short events, void *); static int evport_del(struct event_base *, int fd, short old, short events, void *); static int evport_dispatch(struct event_base *, struct timeval *); static void evport_dealloc(struct event_base *); static int grow(struct evport_data *, int min_events); const struct eventop evportops = { "evport", evport_init, evport_add, evport_del, evport_dispatch, evport_dealloc, 1, /* need reinit */ 0, /* features */ sizeof(struct fd_info), /* fdinfo length */ }; /* * Initialize the event port implementation. */ static void* evport_init(struct event_base *base) { struct evport_data *evpd; if (!(evpd = mm_calloc(1, sizeof(struct evport_data)))) return (NULL); if ((evpd->ed_port = port_create()) == -1) { mm_free(evpd); return (NULL); } if (grow(evpd, INITIAL_EVENTS_PER_GETN) < 0) { close(evpd->ed_port); mm_free(evpd); return NULL; } evpd->ed_npending = 0; evsig_init_(base); return (evpd); } static int grow(struct evport_data *data, int min_events) { int newsize; int *new_pending; port_event_t *new_pevtlist; if (data->ed_maxevents) { newsize = data->ed_maxevents; do { newsize *= 2; } while (newsize < min_events); } else { newsize = min_events; } new_pending = mm_realloc(data->ed_pending, sizeof(int)*newsize); if (new_pending == NULL) return -1; data->ed_pending = new_pending; new_pevtlist = mm_realloc(data->ed_pevtlist, sizeof(port_event_t)*newsize); if (new_pevtlist == NULL) return -1; data->ed_pevtlist = new_pevtlist; data->ed_maxevents = newsize; return 0; } #ifdef CHECK_INVARIANTS /* * Checks some basic properties about the evport_data structure. Because it * checks all file descriptors, this function can be expensive when the maximum * file descriptor ever used is rather large. */ static void check_evportop(struct evport_data *evpd) { EVUTIL_ASSERT(evpd); EVUTIL_ASSERT(evpd->ed_port > 0); } /* * Verifies very basic integrity of a given port_event. */ static void check_event(port_event_t* pevt) { /* * We've only registered for PORT_SOURCE_FD events. The only * other thing we can legitimately receive is PORT_SOURCE_ALERT, * but since we're not using port_alert either, we can assume * PORT_SOURCE_FD. */ EVUTIL_ASSERT(pevt->portev_source == PORT_SOURCE_FD); } #else #define check_evportop(epop) #define check_event(pevt) #endif /* CHECK_INVARIANTS */ /* * (Re)associates the given file descriptor with the event port. The OS events * are specified (implicitly) from the fd_info struct. */ static int reassociate(struct evport_data *epdp, struct fd_info *fdip, int fd) { int sysevents = FDI_TO_SYSEVENTS(fdip); if (sysevents != 0) { if (port_associate(epdp->ed_port, PORT_SOURCE_FD, fd, sysevents, fdip) == -1) { event_warn("port_associate"); return (-1); } } check_evportop(epdp); return (0); } /* * Main event loop - polls port_getn for some number of events, and processes * them. */ static int evport_dispatch(struct event_base *base, struct timeval *tv) { int i, res; struct evport_data *epdp = base->evbase; port_event_t *pevtlist = epdp->ed_pevtlist; /* * port_getn will block until it has at least nevents events. It will * also return how many it's given us (which may be more than we asked * for, as long as it's less than our maximum (ed_maxevents)) in * nevents. */ int nevents = 1; /* * We have to convert a struct timeval to a struct timespec * (only difference is nanoseconds vs. microseconds). If no time-based * events are active, we should wait for I/O (and tv == NULL). */ struct timespec ts; struct timespec *ts_p = NULL; if (tv != NULL) { ts.tv_sec = tv->tv_sec; ts.tv_nsec = tv->tv_usec * 1000; ts_p = &ts; } /* * Before doing anything else, we need to reassociate the events we hit * last time which need reassociation. See comment at the end of the * loop below. */ for (i = 0; i < epdp->ed_npending; ++i) { struct fd_info *fdi = NULL; const int fd = epdp->ed_pending[i]; if (fd != -1) { /* We might have cleared out this event; we need * to be sure that it's still set. */ fdi = evmap_io_get_fdinfo_(&base->io, fd); } if (fdi != NULL && FDI_HAS_EVENTS(fdi)) { reassociate(epdp, fdi, fd); /* epdp->ed_pending[i] = -1; */ fdi->pending_idx_plus_1 = 0; } } EVBASE_RELEASE_LOCK(base, th_base_lock); res = port_getn(epdp->ed_port, pevtlist, epdp->ed_maxevents, (unsigned int *) &nevents, ts_p); EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (res == -1) { if (errno == EINTR || errno == EAGAIN) { return (0); } else if (errno == ETIME) { if (nevents == 0) return (0); } else { event_warn("port_getn"); return (-1); } } event_debug(("%s: port_getn reports %d events", __func__, nevents)); for (i = 0; i < nevents; ++i) { port_event_t *pevt = &pevtlist[i]; int fd = (int) pevt->portev_object; struct fd_info *fdi = pevt->portev_user; /*EVUTIL_ASSERT(evmap_io_get_fdinfo_(&base->io, fd) == fdi);*/ check_evportop(epdp); check_event(pevt); epdp->ed_pending[i] = fd; fdi->pending_idx_plus_1 = i + 1; /* * Figure out what kind of event it was * (because we have to pass this to the callback) */ res = 0; if (pevt->portev_events & (POLLERR|POLLHUP)) { res = EV_READ | EV_WRITE; } else { if (pevt->portev_events & POLLIN) res |= EV_READ; if (pevt->portev_events & POLLOUT) res |= EV_WRITE; } /* * Check for the error situations or a hangup situation */ if (pevt->portev_events & (POLLERR|POLLHUP|POLLNVAL)) res |= EV_READ|EV_WRITE; evmap_io_active_(base, fd, res); } /* end of all events gotten */ epdp->ed_npending = nevents; if (nevents == epdp->ed_maxevents && epdp->ed_maxevents < MAX_EVENTS_PER_GETN) { /* we used all the space this time. We should be ready * for more events next time around. */ grow(epdp, epdp->ed_maxevents * 2); } check_evportop(epdp); return (0); } /* * Adds the given event (so that you will be notified when it happens via * the callback function). */ static int evport_add(struct event_base *base, int fd, short old, short events, void *p) { struct evport_data *evpd = base->evbase; struct fd_info *fdi = p; check_evportop(evpd); fdi->fdi_what |= events; return reassociate(evpd, fdi, fd); } /* * Removes the given event from the list of events to wait for. */ static int evport_del(struct event_base *base, int fd, short old, short events, void *p) { struct evport_data *evpd = base->evbase; struct fd_info *fdi = p; int associated = ! fdi->pending_idx_plus_1; check_evportop(evpd); fdi->fdi_what &= ~(events &(EV_READ|EV_WRITE)); if (associated) { if (!FDI_HAS_EVENTS(fdi) && port_dissociate(evpd->ed_port, PORT_SOURCE_FD, fd) == -1) { /* * Ignore EBADFD error the fd could have been closed * before event_del() was called. */ if (errno != EBADFD) { event_warn("port_dissociate"); return (-1); } } else { if (FDI_HAS_EVENTS(fdi)) { return (reassociate(evpd, fdi, fd)); } } } else { if ((fdi->fdi_what & (EV_READ|EV_WRITE)) == 0) { const int i = fdi->pending_idx_plus_1 - 1; EVUTIL_ASSERT(evpd->ed_pending[i] == fd); evpd->ed_pending[i] = -1; fdi->pending_idx_plus_1 = 0; } } return 0; } static void evport_dealloc(struct event_base *base) { struct evport_data *evpd = base->evbase; evsig_dealloc_(base); close(evpd->ed_port); if (evpd->ed_pending) mm_free(evpd->ed_pending); if (evpd->ed_pevtlist) mm_free(evpd->ed_pevtlist); mm_free(evpd); } #endif /* EVENT__HAVE_EVENT_PORTS */ /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #define WIN32_LEAN_AND_MEAN #include #include #undef WIN32_LEAN_AND_MEAN #endif #include #ifndef _WIN32 #include #endif #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #ifndef _WIN32 #include #endif #include #include #include #include #include "event2/event.h" #include "event2/event_struct.h" #include "event2/rpc.h" #include "event2/rpc_struct.h" #include "evrpc-internal.h" #include "event2/http.h" #include "event2/buffer.h" #include "event2/tag.h" #include "event2/http_struct.h" #include "event2/http_compat.h" #include "event2/util.h" #include "util-internal.h" #include "log-internal.h" #include "mm-internal.h" struct evrpc_base * evrpc_init(struct evhttp *http_server) { struct evrpc_base* base = mm_calloc(1, sizeof(struct evrpc_base)); if (base == NULL) return (NULL); /* we rely on the tagging sub system */ evtag_init(); TAILQ_INIT(&base->registered_rpcs); TAILQ_INIT(&base->input_hooks); TAILQ_INIT(&base->output_hooks); TAILQ_INIT(&base->paused_requests); base->http_server = http_server; return (base); } void evrpc_free(struct evrpc_base *base) { struct evrpc *rpc; struct evrpc_hook *hook; struct evrpc_hook_ctx *pause; int r; while ((rpc = TAILQ_FIRST(&base->registered_rpcs)) != NULL) { r = evrpc_unregister_rpc(base, rpc->uri); EVUTIL_ASSERT(r == 0); } while ((pause = TAILQ_FIRST(&base->paused_requests)) != NULL) { TAILQ_REMOVE(&base->paused_requests, pause, next); mm_free(pause); } while ((hook = TAILQ_FIRST(&base->input_hooks)) != NULL) { r = evrpc_remove_hook(base, EVRPC_INPUT, hook); EVUTIL_ASSERT(r); } while ((hook = TAILQ_FIRST(&base->output_hooks)) != NULL) { r = evrpc_remove_hook(base, EVRPC_OUTPUT, hook); EVUTIL_ASSERT(r); } mm_free(base); } void * evrpc_add_hook(void *vbase, enum EVRPC_HOOK_TYPE hook_type, int (*cb)(void *, struct evhttp_request *, struct evbuffer *, void *), void *cb_arg) { struct evrpc_hooks_ *base = vbase; struct evrpc_hook_list *head = NULL; struct evrpc_hook *hook = NULL; switch (hook_type) { case EVRPC_INPUT: head = &base->in_hooks; break; case EVRPC_OUTPUT: head = &base->out_hooks; break; default: EVUTIL_ASSERT(hook_type == EVRPC_INPUT || hook_type == EVRPC_OUTPUT); } hook = mm_calloc(1, sizeof(struct evrpc_hook)); EVUTIL_ASSERT(hook != NULL); hook->process = cb; hook->process_arg = cb_arg; TAILQ_INSERT_TAIL(head, hook, next); return (hook); } static int evrpc_remove_hook_internal(struct evrpc_hook_list *head, void *handle) { struct evrpc_hook *hook = NULL; TAILQ_FOREACH(hook, head, next) { if (hook == handle) { TAILQ_REMOVE(head, hook, next); mm_free(hook); return (1); } } return (0); } /* * remove the hook specified by the handle */ int evrpc_remove_hook(void *vbase, enum EVRPC_HOOK_TYPE hook_type, void *handle) { struct evrpc_hooks_ *base = vbase; struct evrpc_hook_list *head = NULL; switch (hook_type) { case EVRPC_INPUT: head = &base->in_hooks; break; case EVRPC_OUTPUT: head = &base->out_hooks; break; default: EVUTIL_ASSERT(hook_type == EVRPC_INPUT || hook_type == EVRPC_OUTPUT); } return (evrpc_remove_hook_internal(head, handle)); } static int evrpc_process_hooks(struct evrpc_hook_list *head, void *ctx, struct evhttp_request *req, struct evbuffer *evbuf) { struct evrpc_hook *hook; TAILQ_FOREACH(hook, head, next) { int res = hook->process(ctx, req, evbuf, hook->process_arg); if (res != EVRPC_CONTINUE) return (res); } return (EVRPC_CONTINUE); } static void evrpc_pool_schedule(struct evrpc_pool *pool); static void evrpc_request_cb(struct evhttp_request *, void *); /* * Registers a new RPC with the HTTP server. The evrpc object is expected * to have been filled in via the EVRPC_REGISTER_OBJECT macro which in turn * calls this function. */ static char * evrpc_construct_uri(const char *uri) { char *constructed_uri; size_t constructed_uri_len; constructed_uri_len = strlen(EVRPC_URI_PREFIX) + strlen(uri) + 1; if ((constructed_uri = mm_malloc(constructed_uri_len)) == NULL) event_err(1, "%s: failed to register rpc at %s", __func__, uri); memcpy(constructed_uri, EVRPC_URI_PREFIX, strlen(EVRPC_URI_PREFIX)); memcpy(constructed_uri + strlen(EVRPC_URI_PREFIX), uri, strlen(uri)); constructed_uri[constructed_uri_len - 1] = '\0'; return (constructed_uri); } int evrpc_register_rpc(struct evrpc_base *base, struct evrpc *rpc, void (*cb)(struct evrpc_req_generic *, void *), void *cb_arg) { char *constructed_uri = evrpc_construct_uri(rpc->uri); rpc->base = base; rpc->cb = cb; rpc->cb_arg = cb_arg; TAILQ_INSERT_TAIL(&base->registered_rpcs, rpc, next); evhttp_set_cb(base->http_server, constructed_uri, evrpc_request_cb, rpc); mm_free(constructed_uri); return (0); } int evrpc_unregister_rpc(struct evrpc_base *base, const char *name) { char *registered_uri = NULL; struct evrpc *rpc; int r; /* find the right rpc; linear search might be slow */ TAILQ_FOREACH(rpc, &base->registered_rpcs, next) { if (strcmp(rpc->uri, name) == 0) break; } if (rpc == NULL) { /* We did not find an RPC with this name */ return (-1); } TAILQ_REMOVE(&base->registered_rpcs, rpc, next); registered_uri = evrpc_construct_uri(name); /* remove the http server callback */ r = evhttp_del_cb(base->http_server, registered_uri); EVUTIL_ASSERT(r == 0); mm_free(registered_uri); mm_free((char *)rpc->uri); mm_free(rpc); return (0); } static int evrpc_pause_request(void *vbase, void *ctx, void (*cb)(void *, enum EVRPC_HOOK_RESULT)); static void evrpc_request_cb_closure(void *, enum EVRPC_HOOK_RESULT); static void evrpc_request_cb(struct evhttp_request *req, void *arg) { struct evrpc *rpc = arg; struct evrpc_req_generic *rpc_state = NULL; /* let's verify the outside parameters */ if (req->type != EVHTTP_REQ_POST || evbuffer_get_length(req->input_buffer) <= 0) goto error; rpc_state = mm_calloc(1, sizeof(struct evrpc_req_generic)); if (rpc_state == NULL) goto error; rpc_state->rpc = rpc; rpc_state->http_req = req; rpc_state->rpc_data = NULL; if (TAILQ_FIRST(&rpc->base->input_hooks) != NULL) { int hook_res; evrpc_hook_associate_meta_(&rpc_state->hook_meta, req->evcon); /* * allow hooks to modify the outgoing request */ hook_res = evrpc_process_hooks(&rpc->base->input_hooks, rpc_state, req, req->input_buffer); switch (hook_res) { case EVRPC_TERMINATE: goto error; case EVRPC_PAUSE: evrpc_pause_request(rpc->base, rpc_state, evrpc_request_cb_closure); return; case EVRPC_CONTINUE: break; default: EVUTIL_ASSERT(hook_res == EVRPC_TERMINATE || hook_res == EVRPC_CONTINUE || hook_res == EVRPC_PAUSE); } } evrpc_request_cb_closure(rpc_state, EVRPC_CONTINUE); return; error: if (rpc_state) evrpc_reqstate_free_(rpc_state); evhttp_send_error(req, HTTP_SERVUNAVAIL, NULL); return; } static void evrpc_request_cb_closure(void *arg, enum EVRPC_HOOK_RESULT hook_res) { struct evrpc_req_generic *rpc_state = arg; struct evrpc *rpc; struct evhttp_request *req; EVUTIL_ASSERT(rpc_state); rpc = rpc_state->rpc; req = rpc_state->http_req; if (hook_res == EVRPC_TERMINATE) goto error; /* let's check that we can parse the request */ rpc_state->request = rpc->request_new(rpc->request_new_arg); if (rpc_state->request == NULL) goto error; if (rpc->request_unmarshal( rpc_state->request, req->input_buffer) == -1) { /* we failed to parse the request; that's a bummer */ goto error; } /* at this point, we have a well formed request, prepare the reply */ rpc_state->reply = rpc->reply_new(rpc->reply_new_arg); if (rpc_state->reply == NULL) goto error; /* give the rpc to the user; they can deal with it */ rpc->cb(rpc_state, rpc->cb_arg); return; error: evrpc_reqstate_free_(rpc_state); evhttp_send_error(req, HTTP_SERVUNAVAIL, NULL); return; } void evrpc_reqstate_free_(struct evrpc_req_generic* rpc_state) { struct evrpc *rpc; EVUTIL_ASSERT(rpc_state != NULL); rpc = rpc_state->rpc; /* clean up all memory */ if (rpc_state->hook_meta != NULL) evrpc_hook_context_free_(rpc_state->hook_meta); if (rpc_state->request != NULL) rpc->request_free(rpc_state->request); if (rpc_state->reply != NULL) rpc->reply_free(rpc_state->reply); if (rpc_state->rpc_data != NULL) evbuffer_free(rpc_state->rpc_data); mm_free(rpc_state); } static void evrpc_request_done_closure(void *, enum EVRPC_HOOK_RESULT); void evrpc_request_done(struct evrpc_req_generic *rpc_state) { struct evhttp_request *req; struct evrpc *rpc; EVUTIL_ASSERT(rpc_state); req = rpc_state->http_req; rpc = rpc_state->rpc; if (rpc->reply_complete(rpc_state->reply) == -1) { /* the reply was not completely filled in. error out */ goto error; } if ((rpc_state->rpc_data = evbuffer_new()) == NULL) { /* out of memory */ goto error; } /* serialize the reply */ rpc->reply_marshal(rpc_state->rpc_data, rpc_state->reply); if (TAILQ_FIRST(&rpc->base->output_hooks) != NULL) { int hook_res; evrpc_hook_associate_meta_(&rpc_state->hook_meta, req->evcon); /* do hook based tweaks to the request */ hook_res = evrpc_process_hooks(&rpc->base->output_hooks, rpc_state, req, rpc_state->rpc_data); switch (hook_res) { case EVRPC_TERMINATE: goto error; case EVRPC_PAUSE: if (evrpc_pause_request(rpc->base, rpc_state, evrpc_request_done_closure) == -1) goto error; return; case EVRPC_CONTINUE: break; default: EVUTIL_ASSERT(hook_res == EVRPC_TERMINATE || hook_res == EVRPC_CONTINUE || hook_res == EVRPC_PAUSE); } } evrpc_request_done_closure(rpc_state, EVRPC_CONTINUE); return; error: evrpc_reqstate_free_(rpc_state); evhttp_send_error(req, HTTP_SERVUNAVAIL, NULL); return; } void * evrpc_get_request(struct evrpc_req_generic *req) { return req->request; } void * evrpc_get_reply(struct evrpc_req_generic *req) { return req->reply; } static void evrpc_request_done_closure(void *arg, enum EVRPC_HOOK_RESULT hook_res) { struct evrpc_req_generic *rpc_state = arg; struct evhttp_request *req; EVUTIL_ASSERT(rpc_state); req = rpc_state->http_req; if (hook_res == EVRPC_TERMINATE) goto error; /* on success, we are going to transmit marshaled binary data */ if (evhttp_find_header(req->output_headers, "Content-Type") == NULL) { evhttp_add_header(req->output_headers, "Content-Type", "application/octet-stream"); } evhttp_send_reply(req, HTTP_OK, "OK", rpc_state->rpc_data); evrpc_reqstate_free_(rpc_state); return; error: evrpc_reqstate_free_(rpc_state); evhttp_send_error(req, HTTP_SERVUNAVAIL, NULL); return; } /* Client implementation of RPC site */ static int evrpc_schedule_request(struct evhttp_connection *connection, struct evrpc_request_wrapper *ctx); struct evrpc_pool * evrpc_pool_new(struct event_base *base) { struct evrpc_pool *pool = mm_calloc(1, sizeof(struct evrpc_pool)); if (pool == NULL) return (NULL); TAILQ_INIT(&pool->connections); TAILQ_INIT(&pool->requests); TAILQ_INIT(&pool->paused_requests); TAILQ_INIT(&pool->input_hooks); TAILQ_INIT(&pool->output_hooks); pool->base = base; pool->timeout = -1; return (pool); } static void evrpc_request_wrapper_free(struct evrpc_request_wrapper *request) { if (request->hook_meta != NULL) evrpc_hook_context_free_(request->hook_meta); mm_free(request->name); mm_free(request); } void evrpc_pool_free(struct evrpc_pool *pool) { struct evhttp_connection *connection; struct evrpc_request_wrapper *request; struct evrpc_hook_ctx *pause; struct evrpc_hook *hook; int r; while ((request = TAILQ_FIRST(&pool->requests)) != NULL) { TAILQ_REMOVE(&pool->requests, request, next); evrpc_request_wrapper_free(request); } while ((pause = TAILQ_FIRST(&pool->paused_requests)) != NULL) { TAILQ_REMOVE(&pool->paused_requests, pause, next); mm_free(pause); } while ((connection = TAILQ_FIRST(&pool->connections)) != NULL) { TAILQ_REMOVE(&pool->connections, connection, next); evhttp_connection_free(connection); } while ((hook = TAILQ_FIRST(&pool->input_hooks)) != NULL) { r = evrpc_remove_hook(pool, EVRPC_INPUT, hook); EVUTIL_ASSERT(r); } while ((hook = TAILQ_FIRST(&pool->output_hooks)) != NULL) { r = evrpc_remove_hook(pool, EVRPC_OUTPUT, hook); EVUTIL_ASSERT(r); } mm_free(pool); } /* * Add a connection to the RPC pool. A request scheduled on the pool * may use any available connection. */ void evrpc_pool_add_connection(struct evrpc_pool *pool, struct evhttp_connection *connection) { EVUTIL_ASSERT(connection->http_server == NULL); TAILQ_INSERT_TAIL(&pool->connections, connection, next); /* * associate an event base with this connection */ if (pool->base != NULL) evhttp_connection_set_base(connection, pool->base); /* * unless a timeout was specifically set for a connection, * the connection inherits the timeout from the pool. */ if (!evutil_timerisset(&connection->timeout)) evhttp_connection_set_timeout(connection, pool->timeout); /* * if we have any requests pending, schedule them with the new * connections. */ if (TAILQ_FIRST(&pool->requests) != NULL) { struct evrpc_request_wrapper *request = TAILQ_FIRST(&pool->requests); TAILQ_REMOVE(&pool->requests, request, next); evrpc_schedule_request(connection, request); } } void evrpc_pool_remove_connection(struct evrpc_pool *pool, struct evhttp_connection *connection) { TAILQ_REMOVE(&pool->connections, connection, next); } void evrpc_pool_set_timeout(struct evrpc_pool *pool, int timeout_in_secs) { struct evhttp_connection *evcon; TAILQ_FOREACH(evcon, &pool->connections, next) { evhttp_connection_set_timeout(evcon, timeout_in_secs); } pool->timeout = timeout_in_secs; } static void evrpc_reply_done(struct evhttp_request *, void *); static void evrpc_request_timeout(evutil_socket_t, short, void *); /* * Finds a connection object associated with the pool that is currently * idle and can be used to make a request. */ static struct evhttp_connection * evrpc_pool_find_connection(struct evrpc_pool *pool) { struct evhttp_connection *connection; TAILQ_FOREACH(connection, &pool->connections, next) { if (TAILQ_FIRST(&connection->requests) == NULL) return (connection); } return (NULL); } /* * Prototypes responsible for evrpc scheduling and hooking */ static void evrpc_schedule_request_closure(void *ctx, enum EVRPC_HOOK_RESULT); /* * We assume that the ctx is no longer queued on the pool. */ static int evrpc_schedule_request(struct evhttp_connection *connection, struct evrpc_request_wrapper *ctx) { struct evhttp_request *req = NULL; struct evrpc_pool *pool = ctx->pool; struct evrpc_status status; if ((req = evhttp_request_new(evrpc_reply_done, ctx)) == NULL) goto error; /* serialize the request data into the output buffer */ ctx->request_marshal(req->output_buffer, ctx->request); /* we need to know the connection that we might have to abort */ ctx->evcon = connection; /* if we get paused we also need to know the request */ ctx->req = req; if (TAILQ_FIRST(&pool->output_hooks) != NULL) { int hook_res; evrpc_hook_associate_meta_(&ctx->hook_meta, connection); /* apply hooks to the outgoing request */ hook_res = evrpc_process_hooks(&pool->output_hooks, ctx, req, req->output_buffer); switch (hook_res) { case EVRPC_TERMINATE: goto error; case EVRPC_PAUSE: /* we need to be explicitly resumed */ if (evrpc_pause_request(pool, ctx, evrpc_schedule_request_closure) == -1) goto error; return (0); case EVRPC_CONTINUE: /* we can just continue */ break; default: EVUTIL_ASSERT(hook_res == EVRPC_TERMINATE || hook_res == EVRPC_CONTINUE || hook_res == EVRPC_PAUSE); } } evrpc_schedule_request_closure(ctx, EVRPC_CONTINUE); return (0); error: memset(&status, 0, sizeof(status)); status.error = EVRPC_STATUS_ERR_UNSTARTED; (*ctx->cb)(&status, ctx->request, ctx->reply, ctx->cb_arg); evrpc_request_wrapper_free(ctx); return (-1); } static void evrpc_schedule_request_closure(void *arg, enum EVRPC_HOOK_RESULT hook_res) { struct evrpc_request_wrapper *ctx = arg; struct evhttp_connection *connection = ctx->evcon; struct evhttp_request *req = ctx->req; struct evrpc_pool *pool = ctx->pool; struct evrpc_status status; char *uri = NULL; int res = 0; if (hook_res == EVRPC_TERMINATE) goto error; uri = evrpc_construct_uri(ctx->name); if (uri == NULL) goto error; if (pool->timeout > 0) { /* * a timeout after which the whole rpc is going to be aborted. */ struct timeval tv; evutil_timerclear(&tv); tv.tv_sec = pool->timeout; evtimer_add(&ctx->ev_timeout, &tv); } /* start the request over the connection */ res = evhttp_make_request(connection, req, EVHTTP_REQ_POST, uri); mm_free(uri); if (res == -1) goto error; return; error: memset(&status, 0, sizeof(status)); status.error = EVRPC_STATUS_ERR_UNSTARTED; (*ctx->cb)(&status, ctx->request, ctx->reply, ctx->cb_arg); evrpc_request_wrapper_free(ctx); } /* we just queue the paused request on the pool under the req object */ static int evrpc_pause_request(void *vbase, void *ctx, void (*cb)(void *, enum EVRPC_HOOK_RESULT)) { struct evrpc_hooks_ *base = vbase; struct evrpc_hook_ctx *pause = mm_malloc(sizeof(*pause)); if (pause == NULL) return (-1); pause->ctx = ctx; pause->cb = cb; TAILQ_INSERT_TAIL(&base->pause_requests, pause, next); return (0); } int evrpc_resume_request(void *vbase, void *ctx, enum EVRPC_HOOK_RESULT res) { struct evrpc_hooks_ *base = vbase; struct evrpc_pause_list *head = &base->pause_requests; struct evrpc_hook_ctx *pause; TAILQ_FOREACH(pause, head, next) { if (pause->ctx == ctx) break; } if (pause == NULL) return (-1); (*pause->cb)(pause->ctx, res); TAILQ_REMOVE(head, pause, next); mm_free(pause); return (0); } int evrpc_make_request(struct evrpc_request_wrapper *ctx) { struct evrpc_pool *pool = ctx->pool; /* initialize the event structure for this rpc */ evtimer_assign(&ctx->ev_timeout, pool->base, evrpc_request_timeout, ctx); /* we better have some available connections on the pool */ EVUTIL_ASSERT(TAILQ_FIRST(&pool->connections) != NULL); /* * if no connection is available, we queue the request on the pool, * the next time a connection is empty, the rpc will be send on that. */ TAILQ_INSERT_TAIL(&pool->requests, ctx, next); evrpc_pool_schedule(pool); return (0); } struct evrpc_request_wrapper * evrpc_make_request_ctx( struct evrpc_pool *pool, void *request, void *reply, const char *rpcname, void (*req_marshal)(struct evbuffer*, void *), void (*rpl_clear)(void *), int (*rpl_unmarshal)(void *, struct evbuffer *), void (*cb)(struct evrpc_status *, void *, void *, void *), void *cbarg) { struct evrpc_request_wrapper *ctx = (struct evrpc_request_wrapper *) mm_malloc(sizeof(struct evrpc_request_wrapper)); if (ctx == NULL) return (NULL); ctx->pool = pool; ctx->hook_meta = NULL; ctx->evcon = NULL; ctx->name = mm_strdup(rpcname); if (ctx->name == NULL) { mm_free(ctx); return (NULL); } ctx->cb = cb; ctx->cb_arg = cbarg; ctx->request = request; ctx->reply = reply; ctx->request_marshal = req_marshal; ctx->reply_clear = rpl_clear; ctx->reply_unmarshal = rpl_unmarshal; return (ctx); } static void evrpc_reply_done_closure(void *, enum EVRPC_HOOK_RESULT); static void evrpc_reply_done(struct evhttp_request *req, void *arg) { struct evrpc_request_wrapper *ctx = arg; struct evrpc_pool *pool = ctx->pool; int hook_res = EVRPC_CONTINUE; /* cancel any timeout we might have scheduled */ event_del(&ctx->ev_timeout); ctx->req = req; /* we need to get the reply now */ if (req == NULL) { evrpc_reply_done_closure(ctx, EVRPC_CONTINUE); return; } if (TAILQ_FIRST(&pool->input_hooks) != NULL) { evrpc_hook_associate_meta_(&ctx->hook_meta, ctx->evcon); /* apply hooks to the incoming request */ hook_res = evrpc_process_hooks(&pool->input_hooks, ctx, req, req->input_buffer); switch (hook_res) { case EVRPC_TERMINATE: case EVRPC_CONTINUE: break; case EVRPC_PAUSE: /* * if we get paused we also need to know the * request. unfortunately, the underlying * layer is going to free it. we need to * request ownership explicitly */ evhttp_request_own(req); evrpc_pause_request(pool, ctx, evrpc_reply_done_closure); return; default: EVUTIL_ASSERT(hook_res == EVRPC_TERMINATE || hook_res == EVRPC_CONTINUE || hook_res == EVRPC_PAUSE); } } evrpc_reply_done_closure(ctx, hook_res); /* http request is being freed by underlying layer */ } static void evrpc_reply_done_closure(void *arg, enum EVRPC_HOOK_RESULT hook_res) { struct evrpc_request_wrapper *ctx = arg; struct evhttp_request *req = ctx->req; struct evrpc_pool *pool = ctx->pool; struct evrpc_status status; int res = -1; memset(&status, 0, sizeof(status)); status.http_req = req; /* we need to get the reply now */ if (req == NULL) { status.error = EVRPC_STATUS_ERR_TIMEOUT; } else if (hook_res == EVRPC_TERMINATE) { status.error = EVRPC_STATUS_ERR_HOOKABORTED; } else { res = ctx->reply_unmarshal(ctx->reply, req->input_buffer); if (res == -1) status.error = EVRPC_STATUS_ERR_BADPAYLOAD; } if (res == -1) { /* clear everything that we might have written previously */ ctx->reply_clear(ctx->reply); } (*ctx->cb)(&status, ctx->request, ctx->reply, ctx->cb_arg); evrpc_request_wrapper_free(ctx); /* the http layer owned the original request structure, but if we * got paused, we asked for ownership and need to free it here. */ if (req != NULL && evhttp_request_is_owned(req)) evhttp_request_free(req); /* see if we can schedule another request */ evrpc_pool_schedule(pool); } static void evrpc_pool_schedule(struct evrpc_pool *pool) { struct evrpc_request_wrapper *ctx = TAILQ_FIRST(&pool->requests); struct evhttp_connection *evcon; /* if no requests are pending, we have no work */ if (ctx == NULL) return; if ((evcon = evrpc_pool_find_connection(pool)) != NULL) { TAILQ_REMOVE(&pool->requests, ctx, next); evrpc_schedule_request(evcon, ctx); } } static void evrpc_request_timeout(evutil_socket_t fd, short what, void *arg) { struct evrpc_request_wrapper *ctx = arg; struct evhttp_connection *evcon = ctx->evcon; EVUTIL_ASSERT(evcon != NULL); evhttp_connection_fail_(evcon, EVREQ_HTTP_TIMEOUT); } /* * frees potential meta data associated with a request. */ static void evrpc_meta_data_free(struct evrpc_meta_list *meta_data) { struct evrpc_meta *entry; EVUTIL_ASSERT(meta_data != NULL); while ((entry = TAILQ_FIRST(meta_data)) != NULL) { TAILQ_REMOVE(meta_data, entry, next); mm_free(entry->key); mm_free(entry->data); mm_free(entry); } } static struct evrpc_hook_meta * evrpc_hook_meta_new_(void) { struct evrpc_hook_meta *ctx; ctx = mm_malloc(sizeof(struct evrpc_hook_meta)); EVUTIL_ASSERT(ctx != NULL); TAILQ_INIT(&ctx->meta_data); ctx->evcon = NULL; return (ctx); } static void evrpc_hook_associate_meta_(struct evrpc_hook_meta **pctx, struct evhttp_connection *evcon) { struct evrpc_hook_meta *ctx = *pctx; if (ctx == NULL) *pctx = ctx = evrpc_hook_meta_new_(); ctx->evcon = evcon; } static void evrpc_hook_context_free_(struct evrpc_hook_meta *ctx) { evrpc_meta_data_free(&ctx->meta_data); mm_free(ctx); } /* Adds meta data */ void evrpc_hook_add_meta(void *ctx, const char *key, const void *data, size_t data_size) { struct evrpc_request_wrapper *req = ctx; struct evrpc_hook_meta *store = NULL; struct evrpc_meta *meta = NULL; if ((store = req->hook_meta) == NULL) store = req->hook_meta = evrpc_hook_meta_new_(); meta = mm_malloc(sizeof(struct evrpc_meta)); EVUTIL_ASSERT(meta != NULL); meta->key = mm_strdup(key); EVUTIL_ASSERT(meta->key != NULL); meta->data_size = data_size; meta->data = mm_malloc(data_size); EVUTIL_ASSERT(meta->data != NULL); memcpy(meta->data, data, data_size); TAILQ_INSERT_TAIL(&store->meta_data, meta, next); } int evrpc_hook_find_meta(void *ctx, const char *key, void **data, size_t *data_size) { struct evrpc_request_wrapper *req = ctx; struct evrpc_meta *meta = NULL; if (req->hook_meta == NULL) return (-1); TAILQ_FOREACH(meta, &req->hook_meta->meta_data, next) { if (strcmp(meta->key, key) == 0) { *data = meta->data; *data_size = meta->data_size; return (0); } } return (-1); } struct evhttp_connection * evrpc_hook_get_connection(void *ctx) { struct evrpc_request_wrapper *req = ctx; return (req->hook_meta != NULL ? req->hook_meta->evcon : NULL); } int evrpc_send_request_generic(struct evrpc_pool *pool, void *request, void *reply, void (*cb)(struct evrpc_status *, void *, void *, void *), void *cb_arg, const char *rpcname, void (*req_marshal)(struct evbuffer *, void *), void (*rpl_clear)(void *), int (*rpl_unmarshal)(void *, struct evbuffer *)) { struct evrpc_status status; struct evrpc_request_wrapper *ctx; ctx = evrpc_make_request_ctx(pool, request, reply, rpcname, req_marshal, rpl_clear, rpl_unmarshal, cb, cb_arg); if (ctx == NULL) goto error; return (evrpc_make_request(ctx)); error: memset(&status, 0, sizeof(status)); status.error = EVRPC_STATUS_ERR_UNSTARTED; (*(cb))(&status, request, reply, cb_arg); return (-1); } /** Takes a request object and fills it in with the right magic */ static struct evrpc * evrpc_register_object(const char *name, void *(*req_new)(void*), void *req_new_arg, void (*req_free)(void *), int (*req_unmarshal)(void *, struct evbuffer *), void *(*rpl_new)(void*), void *rpl_new_arg, void (*rpl_free)(void *), int (*rpl_complete)(void *), void (*rpl_marshal)(struct evbuffer *, void *)) { struct evrpc* rpc = (struct evrpc *)mm_calloc(1, sizeof(struct evrpc)); if (rpc == NULL) return (NULL); rpc->uri = mm_strdup(name); if (rpc->uri == NULL) { mm_free(rpc); return (NULL); } rpc->request_new = req_new; rpc->request_new_arg = req_new_arg; rpc->request_free = req_free; rpc->request_unmarshal = req_unmarshal; rpc->reply_new = rpl_new; rpc->reply_new_arg = rpl_new_arg; rpc->reply_free = rpl_free; rpc->reply_complete = rpl_complete; rpc->reply_marshal = rpl_marshal; return (rpc); } int evrpc_register_generic(struct evrpc_base *base, const char *name, void (*callback)(struct evrpc_req_generic *, void *), void *cbarg, void *(*req_new)(void *), void *req_new_arg, void (*req_free)(void *), int (*req_unmarshal)(void *, struct evbuffer *), void *(*rpl_new)(void *), void *rpl_new_arg, void (*rpl_free)(void *), int (*rpl_complete)(void *), void (*rpl_marshal)(struct evbuffer *, void *)) { struct evrpc* rpc = evrpc_register_object(name, req_new, req_new_arg, req_free, req_unmarshal, rpl_new, rpl_new_arg, rpl_free, rpl_complete, rpl_marshal); if (rpc == NULL) return (-1); evrpc_register_rpc(base, rpc, (void (*)(struct evrpc_req_generic*, void *))callback, cbarg); return (0); } /** accessors for obscure and undocumented functionality */ struct evrpc_pool * evrpc_request_get_pool(struct evrpc_request_wrapper *ctx) { return (ctx->pool); } void evrpc_request_set_pool(struct evrpc_request_wrapper *ctx, struct evrpc_pool *pool) { ctx->pool = pool; } void evrpc_request_set_cb(struct evrpc_request_wrapper *ctx, void (*cb)(struct evrpc_status*, void *request, void *reply, void *arg), void *cb_arg) { ctx->cb = cb; ctx->cb_arg = cb_arg; } /* * Copyright (c) 2008-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifndef EVENT__DISABLE_THREAD_SUPPORT #include "event2/thread.h" #include #include #include "log-internal.h" #include "mm-internal.h" #include "util-internal.h" #include "evthread-internal.h" #ifdef EVTHREAD_EXPOSE_STRUCTS #define GLOBAL #else #define GLOBAL static #endif #ifndef EVENT__DISABLE_DEBUG_MODE extern int event_debug_created_threadable_ctx_; extern int event_debug_mode_on_; #endif /* globals */ GLOBAL int evthread_lock_debugging_enabled_ = 0; GLOBAL struct evthread_lock_callbacks evthread_lock_fns_ = { 0, 0, NULL, NULL, NULL, NULL }; GLOBAL unsigned long (*evthread_id_fn_)(void) = NULL; GLOBAL struct evthread_condition_callbacks evthread_cond_fns_ = { 0, NULL, NULL, NULL, NULL }; /* Used for debugging */ static struct evthread_lock_callbacks original_lock_fns_ = { 0, 0, NULL, NULL, NULL, NULL }; static struct evthread_condition_callbacks original_cond_fns_ = { 0, NULL, NULL, NULL, NULL }; void evthread_set_id_callback(unsigned long (*id_fn)(void)) { evthread_id_fn_ = id_fn; } struct evthread_lock_callbacks *evthread_get_lock_callbacks() { return evthread_lock_debugging_enabled_ ? &original_lock_fns_ : &evthread_lock_fns_; } struct evthread_condition_callbacks *evthread_get_condition_callbacks() { return evthread_lock_debugging_enabled_ ? &original_cond_fns_ : &evthread_cond_fns_; } void evthreadimpl_disable_lock_debugging_(void) { evthread_lock_debugging_enabled_ = 0; } int evthread_set_lock_callbacks(const struct evthread_lock_callbacks *cbs) { struct evthread_lock_callbacks *target = evthread_get_lock_callbacks(); #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_mode_on_) { if (event_debug_created_threadable_ctx_) { event_errx(1, "evthread initialization must be called BEFORE anything else!"); } } #endif if (!cbs) { if (target->alloc) event_warnx("Trying to disable lock functions after " "they have been set up will probaby not work."); memset(target, 0, sizeof(evthread_lock_fns_)); return 0; } if (target->alloc) { /* Uh oh; we already had locking callbacks set up.*/ if (target->lock_api_version == cbs->lock_api_version && target->supported_locktypes == cbs->supported_locktypes && target->alloc == cbs->alloc && target->free == cbs->free && target->lock == cbs->lock && target->unlock == cbs->unlock) { /* no change -- allow this. */ return 0; } event_warnx("Can't change lock callbacks once they have been " "initialized."); return -1; } if (cbs->alloc && cbs->free && cbs->lock && cbs->unlock) { memcpy(target, cbs, sizeof(evthread_lock_fns_)); return event_global_setup_locks_(1); } else { return -1; } } int evthread_set_condition_callbacks(const struct evthread_condition_callbacks *cbs) { struct evthread_condition_callbacks *target = evthread_get_condition_callbacks(); #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_mode_on_) { if (event_debug_created_threadable_ctx_) { event_errx(1, "evthread initialization must be called BEFORE anything else!"); } } #endif if (!cbs) { if (target->alloc_condition) event_warnx("Trying to disable condition functions " "after they have been set up will probaby not " "work."); memset(target, 0, sizeof(evthread_cond_fns_)); return 0; } if (target->alloc_condition) { /* Uh oh; we already had condition callbacks set up.*/ if (target->condition_api_version == cbs->condition_api_version && target->alloc_condition == cbs->alloc_condition && target->free_condition == cbs->free_condition && target->signal_condition == cbs->signal_condition && target->wait_condition == cbs->wait_condition) { /* no change -- allow this. */ return 0; } event_warnx("Can't change condition callbacks once they " "have been initialized."); return -1; } if (cbs->alloc_condition && cbs->free_condition && cbs->signal_condition && cbs->wait_condition) { memcpy(target, cbs, sizeof(evthread_cond_fns_)); } if (evthread_lock_debugging_enabled_) { evthread_cond_fns_.alloc_condition = cbs->alloc_condition; evthread_cond_fns_.free_condition = cbs->free_condition; evthread_cond_fns_.signal_condition = cbs->signal_condition; } return 0; } #define DEBUG_LOCK_SIG 0xdeb0b10c struct debug_lock { unsigned signature; unsigned locktype; unsigned long held_by; /* XXXX if we ever use read-write locks, we will need a separate * lock to protect count. */ int count; void *lock; }; static void * debug_lock_alloc(unsigned locktype) { struct debug_lock *result = mm_malloc(sizeof(struct debug_lock)); if (!result) return NULL; if (original_lock_fns_.alloc) { if (!(result->lock = original_lock_fns_.alloc( locktype|EVTHREAD_LOCKTYPE_RECURSIVE))) { mm_free(result); return NULL; } } else { result->lock = NULL; } result->signature = DEBUG_LOCK_SIG; result->locktype = locktype; result->count = 0; result->held_by = 0; return result; } static void debug_lock_free(void *lock_, unsigned locktype) { struct debug_lock *lock = lock_; EVUTIL_ASSERT(lock->count == 0); EVUTIL_ASSERT(locktype == lock->locktype); EVUTIL_ASSERT(DEBUG_LOCK_SIG == lock->signature); if (original_lock_fns_.free) { original_lock_fns_.free(lock->lock, lock->locktype|EVTHREAD_LOCKTYPE_RECURSIVE); } lock->lock = NULL; lock->count = -100; lock->signature = 0x12300fda; mm_free(lock); } static void evthread_debug_lock_mark_locked(unsigned mode, struct debug_lock *lock) { EVUTIL_ASSERT(DEBUG_LOCK_SIG == lock->signature); ++lock->count; if (!(lock->locktype & EVTHREAD_LOCKTYPE_RECURSIVE)) EVUTIL_ASSERT(lock->count == 1); if (evthread_id_fn_) { unsigned long me; me = evthread_id_fn_(); if (lock->count > 1) EVUTIL_ASSERT(lock->held_by == me); lock->held_by = me; } } static int debug_lock_lock(unsigned mode, void *lock_) { struct debug_lock *lock = lock_; int res = 0; if (lock->locktype & EVTHREAD_LOCKTYPE_READWRITE) EVUTIL_ASSERT(mode & (EVTHREAD_READ|EVTHREAD_WRITE)); else EVUTIL_ASSERT((mode & (EVTHREAD_READ|EVTHREAD_WRITE)) == 0); if (original_lock_fns_.lock) res = original_lock_fns_.lock(mode, lock->lock); if (!res) { evthread_debug_lock_mark_locked(mode, lock); } return res; } static void evthread_debug_lock_mark_unlocked(unsigned mode, struct debug_lock *lock) { EVUTIL_ASSERT(DEBUG_LOCK_SIG == lock->signature); if (lock->locktype & EVTHREAD_LOCKTYPE_READWRITE) EVUTIL_ASSERT(mode & (EVTHREAD_READ|EVTHREAD_WRITE)); else EVUTIL_ASSERT((mode & (EVTHREAD_READ|EVTHREAD_WRITE)) == 0); if (evthread_id_fn_) { unsigned long me; me = evthread_id_fn_(); EVUTIL_ASSERT(lock->held_by == me); if (lock->count == 1) lock->held_by = 0; } --lock->count; EVUTIL_ASSERT(lock->count >= 0); } static int debug_lock_unlock(unsigned mode, void *lock_) { struct debug_lock *lock = lock_; int res = 0; evthread_debug_lock_mark_unlocked(mode, lock); if (original_lock_fns_.unlock) res = original_lock_fns_.unlock(mode, lock->lock); return res; } static int debug_cond_wait(void *cond_, void *lock_, const struct timeval *tv) { int r; struct debug_lock *lock = lock_; EVUTIL_ASSERT(lock); EVUTIL_ASSERT(DEBUG_LOCK_SIG == lock->signature); EVLOCK_ASSERT_LOCKED(lock_); evthread_debug_lock_mark_unlocked(0, lock); r = original_cond_fns_.wait_condition(cond_, lock->lock, tv); evthread_debug_lock_mark_locked(0, lock); return r; } /* misspelled version for backward compatibility */ void evthread_enable_lock_debuging(void) { evthread_enable_lock_debugging(); } void evthread_enable_lock_debugging(void) { struct evthread_lock_callbacks cbs = { EVTHREAD_LOCK_API_VERSION, EVTHREAD_LOCKTYPE_RECURSIVE, debug_lock_alloc, debug_lock_free, debug_lock_lock, debug_lock_unlock }; if (evthread_lock_debugging_enabled_) return; memcpy(&original_lock_fns_, &evthread_lock_fns_, sizeof(struct evthread_lock_callbacks)); memcpy(&evthread_lock_fns_, &cbs, sizeof(struct evthread_lock_callbacks)); memcpy(&original_cond_fns_, &evthread_cond_fns_, sizeof(struct evthread_condition_callbacks)); evthread_cond_fns_.wait_condition = debug_cond_wait; evthread_lock_debugging_enabled_ = 1; /* XXX return value should get checked. */ event_global_setup_locks_(0); } int evthread_is_debug_lock_held_(void *lock_) { struct debug_lock *lock = lock_; if (! lock->count) return 0; if (evthread_id_fn_) { unsigned long me = evthread_id_fn_(); if (lock->held_by != me) return 0; } return 1; } void * evthread_debug_get_real_lock_(void *lock_) { struct debug_lock *lock = lock_; return lock->lock; } void * evthread_setup_global_lock_(void *lock_, unsigned locktype, int enable_locks) { /* there are four cases here: 1) we're turning on debugging; locking is not on. 2) we're turning on debugging; locking is on. 3) we're turning on locking; debugging is not on. 4) we're turning on locking; debugging is on. */ if (!enable_locks && original_lock_fns_.alloc == NULL) { /* Case 1: allocate a debug lock. */ EVUTIL_ASSERT(lock_ == NULL); return debug_lock_alloc(locktype); } else if (!enable_locks && original_lock_fns_.alloc != NULL) { /* Case 2: wrap the lock in a debug lock. */ struct debug_lock *lock; EVUTIL_ASSERT(lock_ != NULL); if (!(locktype & EVTHREAD_LOCKTYPE_RECURSIVE)) { /* We can't wrap it: We need a recursive lock */ original_lock_fns_.free(lock_, locktype); return debug_lock_alloc(locktype); } lock = mm_malloc(sizeof(struct debug_lock)); if (!lock) { original_lock_fns_.free(lock_, locktype); return NULL; } lock->lock = lock_; lock->locktype = locktype; lock->count = 0; lock->held_by = 0; return lock; } else if (enable_locks && ! evthread_lock_debugging_enabled_) { /* Case 3: allocate a regular lock */ EVUTIL_ASSERT(lock_ == NULL); return evthread_lock_fns_.alloc(locktype); } else { /* Case 4: Fill in a debug lock with a real lock */ struct debug_lock *lock = lock_ ? lock_ : debug_lock_alloc(locktype); EVUTIL_ASSERT(enable_locks && evthread_lock_debugging_enabled_); EVUTIL_ASSERT(lock->locktype == locktype); if (!lock->lock) { lock->lock = original_lock_fns_.alloc( locktype|EVTHREAD_LOCKTYPE_RECURSIVE); if (!lock->lock) { lock->count = -200; mm_free(lock); return NULL; } } return lock; } } #ifndef EVTHREAD_EXPOSE_STRUCTS unsigned long evthreadimpl_get_id_() { return evthread_id_fn_ ? evthread_id_fn_() : 1; } void * evthreadimpl_lock_alloc_(unsigned locktype) { #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_mode_on_) { event_debug_created_threadable_ctx_ = 1; } #endif return evthread_lock_fns_.alloc ? evthread_lock_fns_.alloc(locktype) : NULL; } void evthreadimpl_lock_free_(void *lock, unsigned locktype) { if (evthread_lock_fns_.free) evthread_lock_fns_.free(lock, locktype); } int evthreadimpl_lock_lock_(unsigned mode, void *lock) { if (evthread_lock_fns_.lock) return evthread_lock_fns_.lock(mode, lock); else return 0; } int evthreadimpl_lock_unlock_(unsigned mode, void *lock) { if (evthread_lock_fns_.unlock) return evthread_lock_fns_.unlock(mode, lock); else return 0; } void * evthreadimpl_cond_alloc_(unsigned condtype) { #ifndef EVENT__DISABLE_DEBUG_MODE if (event_debug_mode_on_) { event_debug_created_threadable_ctx_ = 1; } #endif return evthread_cond_fns_.alloc_condition ? evthread_cond_fns_.alloc_condition(condtype) : NULL; } void evthreadimpl_cond_free_(void *cond) { if (evthread_cond_fns_.free_condition) evthread_cond_fns_.free_condition(cond); } int evthreadimpl_cond_signal_(void *cond, int broadcast) { if (evthread_cond_fns_.signal_condition) return evthread_cond_fns_.signal_condition(cond, broadcast); else return 0; } int evthreadimpl_cond_wait_(void *cond, void *lock, const struct timeval *tv) { if (evthread_cond_fns_.wait_condition) return evthread_cond_fns_.wait_condition(cond, lock, tv); else return 0; } int evthreadimpl_is_lock_debugging_enabled_(void) { return evthread_lock_debugging_enabled_; } int evthreadimpl_locking_enabled_(void) { return evthread_lock_fns_.lock != NULL; } #endif #endif /* * Copyright 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" /* With glibc we need to define _GNU_SOURCE to get PTHREAD_MUTEX_RECURSIVE. * This comes from evconfig-private.h */ #include struct event_base; #include "event2/thread.h" #include #include #include "mm-internal.h" #include "evthread-internal.h" static pthread_mutexattr_t attr_recursive; static void * evthread_posix_lock_alloc(unsigned locktype) { pthread_mutexattr_t *attr = NULL; pthread_mutex_t *lock = mm_malloc(sizeof(pthread_mutex_t)); if (!lock) return NULL; if (locktype & EVTHREAD_LOCKTYPE_RECURSIVE) attr = &attr_recursive; if (pthread_mutex_init(lock, attr)) { mm_free(lock); return NULL; } return lock; } static void evthread_posix_lock_free(void *lock_, unsigned locktype) { pthread_mutex_t *lock = lock_; pthread_mutex_destroy(lock); mm_free(lock); } static int evthread_posix_lock(unsigned mode, void *lock_) { pthread_mutex_t *lock = lock_; if (mode & EVTHREAD_TRY) return pthread_mutex_trylock(lock); else return pthread_mutex_lock(lock); } static int evthread_posix_unlock(unsigned mode, void *lock_) { pthread_mutex_t *lock = lock_; return pthread_mutex_unlock(lock); } static unsigned long evthread_posix_get_id(void) { union { pthread_t thr; #if EVENT__SIZEOF_PTHREAD_T > EVENT__SIZEOF_LONG ev_uint64_t id; #else unsigned long id; #endif } r; #if EVENT__SIZEOF_PTHREAD_T < EVENT__SIZEOF_LONG memset(&r, 0, sizeof(r)); #endif r.thr = pthread_self(); return (unsigned long)r.id; } static void * evthread_posix_cond_alloc(unsigned condflags) { pthread_cond_t *cond = mm_malloc(sizeof(pthread_cond_t)); if (!cond) return NULL; if (pthread_cond_init(cond, NULL)) { mm_free(cond); return NULL; } return cond; } static void evthread_posix_cond_free(void *cond_) { pthread_cond_t *cond = cond_; pthread_cond_destroy(cond); mm_free(cond); } static int evthread_posix_cond_signal(void *cond_, int broadcast) { pthread_cond_t *cond = cond_; int r; if (broadcast) r = pthread_cond_broadcast(cond); else r = pthread_cond_signal(cond); return r ? -1 : 0; } static int evthread_posix_cond_wait(void *cond_, void *lock_, const struct timeval *tv) { int r; pthread_cond_t *cond = cond_; pthread_mutex_t *lock = lock_; if (tv) { struct timeval now, abstime; struct timespec ts; evutil_gettimeofday(&now, NULL); evutil_timeradd(&now, tv, &abstime); ts.tv_sec = abstime.tv_sec; ts.tv_nsec = abstime.tv_usec*1000; r = pthread_cond_timedwait(cond, lock, &ts); if (r == ETIMEDOUT) return 1; else if (r) return -1; else return 0; } else { r = pthread_cond_wait(cond, lock); return r ? -1 : 0; } } int evthread_use_pthreads(void) { struct evthread_lock_callbacks cbs = { EVTHREAD_LOCK_API_VERSION, EVTHREAD_LOCKTYPE_RECURSIVE, evthread_posix_lock_alloc, evthread_posix_lock_free, evthread_posix_lock, evthread_posix_unlock }; struct evthread_condition_callbacks cond_cbs = { EVTHREAD_CONDITION_API_VERSION, evthread_posix_cond_alloc, evthread_posix_cond_free, evthread_posix_cond_signal, evthread_posix_cond_wait }; /* Set ourselves up to get recursive locks. */ if (pthread_mutexattr_init(&attr_recursive)) return -1; if (pthread_mutexattr_settype(&attr_recursive, PTHREAD_MUTEX_RECURSIVE)) return -1; evthread_set_lock_callbacks(&cbs); evthread_set_condition_callbacks(&cond_cbs); evthread_set_id_callback(evthread_posix_get_id); return 0; } /* * Copyright 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #ifndef _WIN32_WINNT /* Minimum required for InitializeCriticalSectionAndSpinCount */ #define _WIN32_WINNT 0x0403 #endif #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #include #endif struct event_base; #include "event2/thread.h" #include "mm-internal.h" #include "evthread-internal.h" #include "time-internal.h" #define SPIN_COUNT 2000 static void * evthread_win32_lock_create(unsigned locktype) { CRITICAL_SECTION *lock = mm_malloc(sizeof(CRITICAL_SECTION)); if (!lock) return NULL; if (InitializeCriticalSectionAndSpinCount(lock, SPIN_COUNT) == 0) { mm_free(lock); return NULL; } return lock; } static void evthread_win32_lock_free(void *lock_, unsigned locktype) { CRITICAL_SECTION *lock = lock_; DeleteCriticalSection(lock); mm_free(lock); } static int evthread_win32_lock(unsigned mode, void *lock_) { CRITICAL_SECTION *lock = lock_; if ((mode & EVTHREAD_TRY)) { return ! TryEnterCriticalSection(lock); } else { EnterCriticalSection(lock); return 0; } } static int evthread_win32_unlock(unsigned mode, void *lock_) { CRITICAL_SECTION *lock = lock_; LeaveCriticalSection(lock); return 0; } static unsigned long evthread_win32_get_id(void) { return (unsigned long) GetCurrentThreadId(); } #ifdef WIN32_HAVE_CONDITION_VARIABLES static void WINAPI (*InitializeConditionVariable_fn)(PCONDITION_VARIABLE) = NULL; static BOOL WINAPI (*SleepConditionVariableCS_fn)( PCONDITION_VARIABLE, PCRITICAL_SECTION, DWORD) = NULL; static void WINAPI (*WakeAllConditionVariable_fn)(PCONDITION_VARIABLE) = NULL; static void WINAPI (*WakeConditionVariable_fn)(PCONDITION_VARIABLE) = NULL; static int evthread_win32_condvar_init(void) { HANDLE lib; lib = GetModuleHandle(TEXT("kernel32.dll")); if (lib == NULL) return 0; #define LOAD(name) \ name##_fn = GetProcAddress(lib, #name) LOAD(InitializeConditionVariable); LOAD(SleepConditionVariableCS); LOAD(WakeAllConditionVariable); LOAD(WakeConditionVariable); return InitializeConditionVariable_fn && SleepConditionVariableCS_fn && WakeAllConditionVariable_fn && WakeConditionVariable_fn; } /* XXXX Even if we can build this, we don't necessarily want to: the functions * in question didn't exist before Vista, so we'd better LoadProc them. */ static void * evthread_win32_condvar_alloc(unsigned condflags) { CONDITION_VARIABLE *cond = mm_malloc(sizeof(CONDITION_VARIABLE)); if (!cond) return NULL; InitializeConditionVariable_fn(cond); return cond; } static void evthread_win32_condvar_free(void *cond_) { CONDITION_VARIABLE *cond = cond_; /* There doesn't _seem_ to be a cleaup fn here... */ mm_free(cond); } static int evthread_win32_condvar_signal(void *cond, int broadcast) { CONDITION_VARIABLE *cond = cond_; if (broadcast) WakeAllConditionVariable_fn(cond); else WakeConditionVariable_fn(cond); return 0; } static int evthread_win32_condvar_wait(void *cond_, void *lock_, const struct timeval *tv) { CONDITION_VARIABLE *cond = cond_; CRITICAL_SECTION *lock = lock_; DWORD ms, err; BOOL result; if (tv) ms = evutil_tv_to_msec_(tv); else ms = INFINITE; result = SleepConditionVariableCS_fn(cond, lock, ms); if (result) { if (GetLastError() == WAIT_TIMEOUT) return 1; else return -1; } else { return 0; } } #endif struct evthread_win32_cond { HANDLE event; CRITICAL_SECTION lock; int n_waiting; int n_to_wake; int generation; }; static void * evthread_win32_cond_alloc(unsigned flags) { struct evthread_win32_cond *cond; if (!(cond = mm_malloc(sizeof(struct evthread_win32_cond)))) return NULL; if (InitializeCriticalSectionAndSpinCount(&cond->lock, SPIN_COUNT)==0) { mm_free(cond); return NULL; } if ((cond->event = CreateEvent(NULL,TRUE,FALSE,NULL)) == NULL) { DeleteCriticalSection(&cond->lock); mm_free(cond); return NULL; } cond->n_waiting = cond->n_to_wake = cond->generation = 0; return cond; } static void evthread_win32_cond_free(void *cond_) { struct evthread_win32_cond *cond = cond_; DeleteCriticalSection(&cond->lock); CloseHandle(cond->event); mm_free(cond); } static int evthread_win32_cond_signal(void *cond_, int broadcast) { struct evthread_win32_cond *cond = cond_; EnterCriticalSection(&cond->lock); if (broadcast) cond->n_to_wake = cond->n_waiting; else ++cond->n_to_wake; cond->generation++; SetEvent(cond->event); LeaveCriticalSection(&cond->lock); return 0; } static int evthread_win32_cond_wait(void *cond_, void *lock_, const struct timeval *tv) { struct evthread_win32_cond *cond = cond_; CRITICAL_SECTION *lock = lock_; int generation_at_start; int waiting = 1; int result = -1; DWORD ms = INFINITE, ms_orig = INFINITE, startTime, endTime; if (tv) ms_orig = ms = evutil_tv_to_msec_(tv); EnterCriticalSection(&cond->lock); ++cond->n_waiting; generation_at_start = cond->generation; LeaveCriticalSection(&cond->lock); LeaveCriticalSection(lock); startTime = GetTickCount(); do { DWORD res; res = WaitForSingleObject(cond->event, ms); EnterCriticalSection(&cond->lock); if (cond->n_to_wake && cond->generation != generation_at_start) { --cond->n_to_wake; --cond->n_waiting; result = 0; waiting = 0; goto out; } else if (res != WAIT_OBJECT_0) { result = (res==WAIT_TIMEOUT) ? 1 : -1; --cond->n_waiting; waiting = 0; goto out; } else if (ms != INFINITE) { endTime = GetTickCount(); if (startTime + ms_orig <= endTime) { result = 1; /* Timeout */ --cond->n_waiting; waiting = 0; goto out; } else { ms = startTime + ms_orig - endTime; } } /* If we make it here, we are still waiting. */ if (cond->n_to_wake == 0) { /* There is nobody else who should wake up; reset * the event. */ ResetEvent(cond->event); } out: LeaveCriticalSection(&cond->lock); } while (waiting); EnterCriticalSection(lock); EnterCriticalSection(&cond->lock); if (!cond->n_waiting) ResetEvent(cond->event); LeaveCriticalSection(&cond->lock); return result; } int evthread_use_windows_threads(void) { struct evthread_lock_callbacks cbs = { EVTHREAD_LOCK_API_VERSION, EVTHREAD_LOCKTYPE_RECURSIVE, evthread_win32_lock_create, evthread_win32_lock_free, evthread_win32_lock, evthread_win32_unlock }; struct evthread_condition_callbacks cond_cbs = { EVTHREAD_CONDITION_API_VERSION, evthread_win32_cond_alloc, evthread_win32_cond_free, evthread_win32_cond_signal, evthread_win32_cond_wait }; #ifdef WIN32_HAVE_CONDITION_VARIABLES struct evthread_condition_callbacks condvar_cbs = { EVTHREAD_CONDITION_API_VERSION, evthread_win32_condvar_alloc, evthread_win32_condvar_free, evthread_win32_condvar_signal, evthread_win32_condvar_wait }; #endif evthread_set_lock_callbacks(&cbs); evthread_set_id_callback(evthread_win32_get_id); #ifdef WIN32_HAVE_CONDITION_VARIABLES if (evthread_win32_condvar_init()) { evthread_set_condition_callbacks(&condvar_cbs); return 0; } #endif evthread_set_condition_callbacks(&cond_cbs); return 0; } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #include #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #include #include #include #undef _WIN32_WINNT /* For structs needed by GetAdaptersAddresses */ #define _WIN32_WINNT 0x0501 #include #include #endif #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #ifdef EVENT__HAVE_FCNTL_H #include #endif #ifdef EVENT__HAVE_STDLIB_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_NETINET_IN_H #include #endif #ifdef EVENT__HAVE_NETINET_IN6_H #include #endif #ifdef EVENT__HAVE_NETINET_TCP_H #include #endif #ifdef EVENT__HAVE_ARPA_INET_H #include #endif #include #include #ifndef _WIN32 #include #endif #ifdef EVENT__HAVE_IFADDRS_H #include #endif #include "event2/util.h" #include "util-internal.h" #include "log-internal.h" #include "mm-internal.h" #include "evthread-internal.h" #include "strlcpy-internal.h" #include "ipv6-internal.h" #ifdef _WIN32 #define HT_NO_CACHE_HASH_VALUES #include "ht-internal.h" #define open _open #define read _read #define close _close #ifndef fstat #define fstat _fstati64 #endif #ifndef stat #define stat _stati64 #endif #define mode_t int #endif int evutil_open_closeonexec_(const char *pathname, int flags, unsigned mode) { int fd; #ifdef O_CLOEXEC fd = open(pathname, flags|O_CLOEXEC, (mode_t)mode); if (fd >= 0 || errno == EINVAL) return fd; /* If we got an EINVAL, fall through and try without O_CLOEXEC */ #endif fd = open(pathname, flags, (mode_t)mode); if (fd < 0) return -1; #if defined(FD_CLOEXEC) if (fcntl(fd, F_SETFD, FD_CLOEXEC) < 0) { close(fd); return -1; } #endif return fd; } /** Read the contents of 'filename' into a newly allocated NUL-terminated string. Set *content_out to hold this string, and *len_out to hold its length (not including the appended NUL). If 'is_binary', open the file in binary mode. Returns 0 on success, -1 if the open fails, and -2 for all other failures. Used internally only; may go away in a future version. */ int evutil_read_file_(const char *filename, char **content_out, size_t *len_out, int is_binary) { int fd, r; struct stat st; char *mem; size_t read_so_far=0; int mode = O_RDONLY; EVUTIL_ASSERT(content_out); EVUTIL_ASSERT(len_out); *content_out = NULL; *len_out = 0; #ifdef O_BINARY if (is_binary) mode |= O_BINARY; #endif fd = evutil_open_closeonexec_(filename, mode, 0); if (fd < 0) return -1; if (fstat(fd, &st) || st.st_size < 0 || st.st_size > EV_SSIZE_MAX-1 ) { close(fd); return -2; } mem = mm_malloc((size_t)st.st_size + 1); if (!mem) { close(fd); return -2; } read_so_far = 0; #ifdef _WIN32 #define N_TO_READ(x) ((x) > INT_MAX) ? INT_MAX : ((int)(x)) #else #define N_TO_READ(x) (x) #endif while ((r = read(fd, mem+read_so_far, N_TO_READ(st.st_size - read_so_far))) > 0) { read_so_far += r; if (read_so_far >= (size_t)st.st_size) break; EVUTIL_ASSERT(read_so_far < (size_t)st.st_size); } close(fd); if (r < 0) { mm_free(mem); return -2; } mem[read_so_far] = 0; *len_out = read_so_far; *content_out = mem; return 0; } int evutil_socketpair(int family, int type, int protocol, evutil_socket_t fd[2]) { #ifndef _WIN32 return socketpair(family, type, protocol, fd); #else return evutil_ersatz_socketpair_(family, type, protocol, fd); #endif } int evutil_ersatz_socketpair_(int family, int type, int protocol, evutil_socket_t fd[2]) { /* This code is originally from Tor. Used with permission. */ /* This socketpair does not work when localhost is down. So * it's really not the same thing at all. But it's close enough * for now, and really, when localhost is down sometimes, we * have other problems too. */ #ifdef _WIN32 #define ERR(e) WSA##e #else #define ERR(e) e #endif evutil_socket_t listener = -1; evutil_socket_t connector = -1; evutil_socket_t acceptor = -1; struct sockaddr_in listen_addr; struct sockaddr_in connect_addr; ev_socklen_t size; int saved_errno = -1; int family_test; family_test = family != AF_INET; #ifdef AF_UNIX family_test = family_test && (family != AF_UNIX); #endif if (protocol || family_test) { EVUTIL_SET_SOCKET_ERROR(ERR(EAFNOSUPPORT)); return -1; } if (!fd) { EVUTIL_SET_SOCKET_ERROR(ERR(EINVAL)); return -1; } listener = socket(AF_INET, type, 0); if (listener < 0) return -1; memset(&listen_addr, 0, sizeof(listen_addr)); listen_addr.sin_family = AF_INET; listen_addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); listen_addr.sin_port = 0; /* kernel chooses port. */ if (bind(listener, (struct sockaddr *) &listen_addr, sizeof (listen_addr)) == -1) goto tidy_up_and_fail; if (listen(listener, 1) == -1) goto tidy_up_and_fail; connector = socket(AF_INET, type, 0); if (connector < 0) goto tidy_up_and_fail; memset(&connect_addr, 0, sizeof(connect_addr)); /* We want to find out the port number to connect to. */ size = sizeof(connect_addr); if (getsockname(listener, (struct sockaddr *) &connect_addr, &size) == -1) goto tidy_up_and_fail; if (size != sizeof (connect_addr)) goto abort_tidy_up_and_fail; if (connect(connector, (struct sockaddr *) &connect_addr, sizeof(connect_addr)) == -1) goto tidy_up_and_fail; size = sizeof(listen_addr); acceptor = accept(listener, (struct sockaddr *) &listen_addr, &size); if (acceptor < 0) goto tidy_up_and_fail; if (size != sizeof(listen_addr)) goto abort_tidy_up_and_fail; /* Now check we are talking to ourself by matching port and host on the two sockets. */ if (getsockname(connector, (struct sockaddr *) &connect_addr, &size) == -1) goto tidy_up_and_fail; if (size != sizeof (connect_addr) || listen_addr.sin_family != connect_addr.sin_family || listen_addr.sin_addr.s_addr != connect_addr.sin_addr.s_addr || listen_addr.sin_port != connect_addr.sin_port) goto abort_tidy_up_and_fail; evutil_closesocket(listener); fd[0] = connector; fd[1] = acceptor; return 0; abort_tidy_up_and_fail: saved_errno = ERR(ECONNABORTED); tidy_up_and_fail: if (saved_errno < 0) saved_errno = EVUTIL_SOCKET_ERROR(); if (listener != -1) evutil_closesocket(listener); if (connector != -1) evutil_closesocket(connector); if (acceptor != -1) evutil_closesocket(acceptor); EVUTIL_SET_SOCKET_ERROR(saved_errno); return -1; #undef ERR } int evutil_make_socket_nonblocking(evutil_socket_t fd) { #ifdef _WIN32 { unsigned long nonblocking = 1; if (ioctlsocket(fd, FIONBIO, &nonblocking) == SOCKET_ERROR) { event_sock_warn(fd, "fcntl(%d, F_GETFL)", (int)fd); return -1; } } #else { int flags; if ((flags = fcntl(fd, F_GETFL, NULL)) < 0) { event_warn("fcntl(%d, F_GETFL)", fd); return -1; } if (!(flags & O_NONBLOCK)) { if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { event_warn("fcntl(%d, F_SETFL)", fd); return -1; } } } #endif return 0; } /* Faster version of evutil_make_socket_nonblocking for internal use. * * Requires that no F_SETFL flags were previously set on the fd. */ static int evutil_fast_socket_nonblocking(evutil_socket_t fd) { #ifdef _WIN32 return evutil_make_socket_nonblocking(fd); #else if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) { event_warn("fcntl(%d, F_SETFL)", fd); return -1; } return 0; #endif } int evutil_make_listen_socket_reuseable(evutil_socket_t sock) { #if defined(SO_REUSEADDR) && !defined(_WIN32) int one = 1; /* REUSEADDR on Unix means, "don't hang on to this address after the * listener is closed." On Windows, though, it means "don't keep other * processes from binding to this address while we're using it. */ return setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, (void*) &one, (ev_socklen_t)sizeof(one)); #else return 0; #endif } int evutil_make_listen_socket_reuseable_port(evutil_socket_t sock) { #if defined __linux__ && defined(SO_REUSEPORT) int one = 1; /* REUSEPORT on Linux 3.9+ means, "Multiple servers (processes or * threads) can bind to the same port if they each set the option. */ return setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, (void*) &one, (ev_socklen_t)sizeof(one)); #else return 0; #endif } int evutil_make_listen_socket_ipv6only(evutil_socket_t sock) { #if defined(IPV6_V6ONLY) int one = 1; return setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, (void*) &one, (ev_socklen_t)sizeof(one)); #endif return 0; } int evutil_make_tcp_listen_socket_deferred(evutil_socket_t sock) { #if defined(EVENT__HAVE_NETINET_TCP_H) && defined(TCP_DEFER_ACCEPT) int one = 1; /* TCP_DEFER_ACCEPT tells the kernel to call defer accept() only after data * has arrived and ready to read */ return setsockopt(sock, IPPROTO_TCP, TCP_DEFER_ACCEPT, &one, (ev_socklen_t)sizeof(one)); #endif return 0; } int evutil_make_socket_closeonexec(evutil_socket_t fd) { #if !defined(_WIN32) && defined(EVENT__HAVE_SETFD) int flags; if ((flags = fcntl(fd, F_GETFD, NULL)) < 0) { event_warn("fcntl(%d, F_GETFD)", fd); return -1; } if (!(flags & FD_CLOEXEC)) { if (fcntl(fd, F_SETFD, flags | FD_CLOEXEC) == -1) { event_warn("fcntl(%d, F_SETFD)", fd); return -1; } } #endif return 0; } /* Faster version of evutil_make_socket_closeonexec for internal use. * * Requires that no F_SETFD flags were previously set on the fd. */ static int evutil_fast_socket_closeonexec(evutil_socket_t fd) { #if !defined(_WIN32) && defined(EVENT__HAVE_SETFD) if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1) { event_warn("fcntl(%d, F_SETFD)", fd); return -1; } #endif return 0; } int evutil_closesocket(evutil_socket_t sock) { #ifndef _WIN32 return close(sock); #else return closesocket(sock); #endif } ev_int64_t evutil_strtoll(const char *s, char **endptr, int base) { #ifdef EVENT__HAVE_STRTOLL return (ev_int64_t)strtoll(s, endptr, base); #elif EVENT__SIZEOF_LONG == 8 return (ev_int64_t)strtol(s, endptr, base); #elif defined(_WIN32) && defined(_MSC_VER) && _MSC_VER < 1300 /* XXXX on old versions of MS APIs, we only support base * 10. */ ev_int64_t r; if (base != 10) return 0; r = (ev_int64_t) _atoi64(s); while (isspace(*s)) ++s; if (*s == '-') ++s; while (isdigit(*s)) ++s; if (endptr) *endptr = (char*) s; return r; #elif defined(_WIN32) return (ev_int64_t) _strtoi64(s, endptr, base); #elif defined(EVENT__SIZEOF_LONG_LONG) && EVENT__SIZEOF_LONG_LONG == 8 long long r; int n; if (base != 10 && base != 16) return 0; if (base == 10) { n = sscanf(s, "%lld", &r); } else { unsigned long long ru=0; n = sscanf(s, "%llx", &ru); if (ru > EV_INT64_MAX) return 0; r = (long long) ru; } if (n != 1) return 0; while (EVUTIL_ISSPACE_(*s)) ++s; if (*s == '-') ++s; if (base == 10) { while (EVUTIL_ISDIGIT_(*s)) ++s; } else { while (EVUTIL_ISXDIGIT_(*s)) ++s; } if (endptr) *endptr = (char*) s; return r; #else #error "I don't know how to parse 64-bit integers." #endif } #ifdef _WIN32 int evutil_socket_geterror(evutil_socket_t sock) { int optval, optvallen=sizeof(optval); int err = WSAGetLastError(); if (err == WSAEWOULDBLOCK && sock >= 0) { if (getsockopt(sock, SOL_SOCKET, SO_ERROR, (void*)&optval, &optvallen)) return err; if (optval) return optval; } return err; } #endif /* XXX we should use an enum here. */ /* 2 for connection refused, 1 for connected, 0 for not yet, -1 for error. */ int evutil_socket_connect_(evutil_socket_t *fd_ptr, const struct sockaddr *sa, int socklen) { int made_fd = 0; if (*fd_ptr < 0) { if ((*fd_ptr = socket(sa->sa_family, SOCK_STREAM, 0)) < 0) goto err; made_fd = 1; if (evutil_make_socket_nonblocking(*fd_ptr) < 0) { goto err; } } if (connect(*fd_ptr, sa, socklen) < 0) { int e = evutil_socket_geterror(*fd_ptr); if (EVUTIL_ERR_CONNECT_RETRIABLE(e)) return 0; if (EVUTIL_ERR_CONNECT_REFUSED(e)) return 2; goto err; } else { return 1; } err: if (made_fd) { evutil_closesocket(*fd_ptr); *fd_ptr = -1; } return -1; } /* Check whether a socket on which we called connect() is done connecting. Return 1 for connected, 0 for not yet, -1 for error. In the error case, set the current socket errno to the error that happened during the connect operation. */ int evutil_socket_finished_connecting_(evutil_socket_t fd) { int e; ev_socklen_t elen = sizeof(e); if (getsockopt(fd, SOL_SOCKET, SO_ERROR, (void*)&e, &elen) < 0) return -1; if (e) { if (EVUTIL_ERR_CONNECT_RETRIABLE(e)) return 0; EVUTIL_SET_SOCKET_ERROR(e); return -1; } return 1; } #if (EVUTIL_AI_PASSIVE|EVUTIL_AI_CANONNAME|EVUTIL_AI_NUMERICHOST| \ EVUTIL_AI_NUMERICSERV|EVUTIL_AI_V4MAPPED|EVUTIL_AI_ALL| \ EVUTIL_AI_ADDRCONFIG) != \ (EVUTIL_AI_PASSIVE^EVUTIL_AI_CANONNAME^EVUTIL_AI_NUMERICHOST^ \ EVUTIL_AI_NUMERICSERV^EVUTIL_AI_V4MAPPED^EVUTIL_AI_ALL^ \ EVUTIL_AI_ADDRCONFIG) #error "Some of our EVUTIL_AI_* flags seem to overlap with system AI_* flags" #endif /* We sometimes need to know whether we have an ipv4 address and whether we have an ipv6 address. If 'have_checked_interfaces', then we've already done the test. If 'had_ipv4_address', then it turns out we had an ipv4 address. If 'had_ipv6_address', then it turns out we had an ipv6 address. These are set by evutil_check_interfaces. */ static int have_checked_interfaces, had_ipv4_address, had_ipv6_address; /* True iff the IPv4 address 'addr', in host order, is in 127.0.0.0/8 */ static inline int evutil_v4addr_is_localhost(ev_uint32_t addr) { return addr>>24 == 127; } /* True iff the IPv4 address 'addr', in host order, is link-local * 169.254.0.0/16 (RFC3927) */ static inline int evutil_v4addr_is_linklocal(ev_uint32_t addr) { return ((addr & 0xffff0000U) == 0xa9fe0000U); } /* True iff the IPv4 address 'addr', in host order, is a class D * (multiclass) address. */ static inline int evutil_v4addr_is_classd(ev_uint32_t addr) { return ((addr>>24) & 0xf0) == 0xe0; } int evutil_v4addr_is_local_(const struct in_addr *in) { const ev_uint32_t addr = ntohl(in->s_addr); return addr == INADDR_ANY || evutil_v4addr_is_localhost(addr) || evutil_v4addr_is_linklocal(addr) || evutil_v4addr_is_classd(addr); } int evutil_v6addr_is_local_(const struct in6_addr *in) { static const char ZEROES[] = "\x00\x00\x00\x00\x00\x00\x00\x00" "\x00\x00\x00\x00\x00\x00\x00\x00"; const unsigned char *addr = (const unsigned char *)in->s6_addr; return !memcmp(addr, ZEROES, 8) || ((addr[0] & 0xfe) == 0xfc) || (addr[0] == 0xfe && (addr[1] & 0xc0) == 0x80) || (addr[0] == 0xfe && (addr[1] & 0xc0) == 0xc0) || (addr[0] == 0xff); } static void evutil_found_ifaddr(const struct sockaddr *sa) { if (sa->sa_family == AF_INET) { const struct sockaddr_in *sin = (struct sockaddr_in *)sa; if (!evutil_v4addr_is_local_(&sin->sin_addr)) { event_debug(("Detected an IPv4 interface")); had_ipv4_address = 1; } } else if (sa->sa_family == AF_INET6) { const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa; if (!evutil_v6addr_is_local_(&sin6->sin6_addr)) { event_debug(("Detected an IPv6 interface")); had_ipv6_address = 1; } } } #ifdef _WIN32 typedef ULONG (WINAPI *GetAdaptersAddresses_fn_t)( ULONG, ULONG, PVOID, PIP_ADAPTER_ADDRESSES, PULONG); #endif static int evutil_check_ifaddrs(void) { #if defined(EVENT__HAVE_GETIFADDRS) /* Most free Unixy systems provide getifaddrs, which gives us a linked list * of struct ifaddrs. */ struct ifaddrs *ifa = NULL; const struct ifaddrs *i; if (getifaddrs(&ifa) < 0) { event_warn("Unable to call getifaddrs()"); return -1; } for (i = ifa; i; i = i->ifa_next) { if (!i->ifa_addr) continue; evutil_found_ifaddr(i->ifa_addr); } freeifaddrs(ifa); return 0; #elif defined(_WIN32) /* Windows XP began to provide GetAdaptersAddresses. Windows 2000 had a "GetAdaptersInfo", but that's deprecated; let's just try GetAdaptersAddresses and fall back to connect+getsockname. */ HMODULE lib = evutil_load_windows_system_library_(TEXT("iphlpapi.dll")); GetAdaptersAddresses_fn_t fn; ULONG size, res; IP_ADAPTER_ADDRESSES *addresses = NULL, *address; int result = -1; #define FLAGS (GAA_FLAG_SKIP_ANYCAST | \ GAA_FLAG_SKIP_MULTICAST | \ GAA_FLAG_SKIP_DNS_SERVER) if (!lib) goto done; if (!(fn = (GetAdaptersAddresses_fn_t) GetProcAddress(lib, "GetAdaptersAddresses"))) goto done; /* Guess how much space we need. */ size = 15*1024; addresses = mm_malloc(size); if (!addresses) goto done; res = fn(AF_UNSPEC, FLAGS, NULL, addresses, &size); if (res == ERROR_BUFFER_OVERFLOW) { /* we didn't guess that we needed enough space; try again */ mm_free(addresses); addresses = mm_malloc(size); if (!addresses) goto done; res = fn(AF_UNSPEC, FLAGS, NULL, addresses, &size); } if (res != NO_ERROR) goto done; for (address = addresses; address; address = address->Next) { IP_ADAPTER_UNICAST_ADDRESS *a; for (a = address->FirstUnicastAddress; a; a = a->Next) { /* Yes, it's a linked list inside a linked list */ struct sockaddr *sa = a->Address.lpSockaddr; evutil_found_ifaddr(sa); } } result = 0; done: if (lib) FreeLibrary(lib); if (addresses) mm_free(addresses); return result; #else return -1; #endif } /* Test whether we have an ipv4 interface and an ipv6 interface. Return 0 if * the test seemed successful. */ static int evutil_check_interfaces(void) { evutil_socket_t fd = -1; struct sockaddr_in sin, sin_out; struct sockaddr_in6 sin6, sin6_out; ev_socklen_t sin_out_len = sizeof(sin_out); ev_socklen_t sin6_out_len = sizeof(sin6_out); int r; if (have_checked_interfaces) return 0; /* From this point on we have done the ipv4/ipv6 interface check */ have_checked_interfaces = 1; if (evutil_check_ifaddrs() == 0) { /* Use a nice sane interface, if this system has one. */ return 0; } /* Ugh. There was no nice sane interface. So to check whether we have * an interface open for a given protocol, will try to make a UDP * 'connection' to a remote host on the internet. We don't actually * use it, so the address doesn't matter, but we want to pick one that * keep us from using a host- or link-local interface. */ memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_port = htons(53); r = evutil_inet_pton(AF_INET, "18.244.0.188", &sin.sin_addr); EVUTIL_ASSERT(r); memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(53); r = evutil_inet_pton(AF_INET6, "2001:4860:b002::68", &sin6.sin6_addr); EVUTIL_ASSERT(r); memset(&sin_out, 0, sizeof(sin_out)); memset(&sin6_out, 0, sizeof(sin6_out)); /* XXX some errnos mean 'no address'; some mean 'not enough sockets'. */ if ((fd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP)) >= 0 && connect(fd, (struct sockaddr*)&sin, sizeof(sin)) == 0 && getsockname(fd, (struct sockaddr*)&sin_out, &sin_out_len) == 0) { /* We might have an IPv4 interface. */ evutil_found_ifaddr((struct sockaddr*) &sin_out); } if (fd >= 0) evutil_closesocket(fd); if ((fd = socket(AF_INET6, SOCK_DGRAM, IPPROTO_UDP)) >= 0 && connect(fd, (struct sockaddr*)&sin6, sizeof(sin6)) == 0 && getsockname(fd, (struct sockaddr*)&sin6_out, &sin6_out_len) == 0) { /* We might have an IPv6 interface. */ evutil_found_ifaddr((struct sockaddr*) &sin6_out); } if (fd >= 0) evutil_closesocket(fd); return 0; } /* Internal addrinfo flag. This one is set when we allocate the addrinfo from * inside libevent. Otherwise, the built-in getaddrinfo() function allocated * it, and we should trust what they said. **/ #define EVUTIL_AI_LIBEVENT_ALLOCATED 0x80000000 /* Helper: construct a new addrinfo containing the socket address in * 'sa', which must be a sockaddr_in or a sockaddr_in6. Take the * socktype and protocol info from hints. If they weren't set, then * allocate both a TCP and a UDP addrinfo. */ struct evutil_addrinfo * evutil_new_addrinfo_(struct sockaddr *sa, ev_socklen_t socklen, const struct evutil_addrinfo *hints) { struct evutil_addrinfo *res; EVUTIL_ASSERT(hints); if (hints->ai_socktype == 0 && hints->ai_protocol == 0) { /* Indecisive user! Give them a UDP and a TCP. */ struct evutil_addrinfo *r1, *r2; struct evutil_addrinfo tmp; memcpy(&tmp, hints, sizeof(tmp)); tmp.ai_socktype = SOCK_STREAM; tmp.ai_protocol = IPPROTO_TCP; r1 = evutil_new_addrinfo_(sa, socklen, &tmp); if (!r1) return NULL; tmp.ai_socktype = SOCK_DGRAM; tmp.ai_protocol = IPPROTO_UDP; r2 = evutil_new_addrinfo_(sa, socklen, &tmp); if (!r2) { evutil_freeaddrinfo(r1); return NULL; } r1->ai_next = r2; return r1; } /* We're going to allocate extra space to hold the sockaddr. */ res = mm_calloc(1,sizeof(struct evutil_addrinfo)+socklen); if (!res) return NULL; res->ai_addr = (struct sockaddr*) (((char*)res) + sizeof(struct evutil_addrinfo)); memcpy(res->ai_addr, sa, socklen); res->ai_addrlen = socklen; res->ai_family = sa->sa_family; /* Same or not? XXX */ res->ai_flags = EVUTIL_AI_LIBEVENT_ALLOCATED; res->ai_socktype = hints->ai_socktype; res->ai_protocol = hints->ai_protocol; return res; } /* Append the addrinfo 'append' to the end of 'first', and return the start of * the list. Either element can be NULL, in which case we return the element * that is not NULL. */ struct evutil_addrinfo * evutil_addrinfo_append_(struct evutil_addrinfo *first, struct evutil_addrinfo *append) { struct evutil_addrinfo *ai = first; if (!ai) return append; while (ai->ai_next) ai = ai->ai_next; ai->ai_next = append; return first; } static int parse_numeric_servname(const char *servname) { int n; char *endptr=NULL; n = (int) strtol(servname, &endptr, 10); if (n>=0 && n <= 65535 && servname[0] && endptr && !endptr[0]) return n; else return -1; } /** Parse a service name in 'servname', which can be a decimal port. * Return the port number, or -1 on error. */ static int evutil_parse_servname(const char *servname, const char *protocol, const struct evutil_addrinfo *hints) { int n = parse_numeric_servname(servname); if (n>=0) return n; #if defined(EVENT__HAVE_GETSERVBYNAME) || defined(_WIN32) if (!(hints->ai_flags & EVUTIL_AI_NUMERICSERV)) { struct servent *ent = getservbyname(servname, protocol); if (ent) { return ntohs(ent->s_port); } } #endif return -1; } /* Return a string corresponding to a protocol number that we can pass to * getservyname. */ static const char * evutil_unparse_protoname(int proto) { switch (proto) { case 0: return NULL; case IPPROTO_TCP: return "tcp"; case IPPROTO_UDP: return "udp"; #ifdef IPPROTO_SCTP case IPPROTO_SCTP: return "sctp"; #endif default: #ifdef EVENT__HAVE_GETPROTOBYNUMBER { struct protoent *ent = getprotobynumber(proto); if (ent) return ent->p_name; } #endif return NULL; } } static void evutil_getaddrinfo_infer_protocols(struct evutil_addrinfo *hints) { /* If we can guess the protocol from the socktype, do so. */ if (!hints->ai_protocol && hints->ai_socktype) { if (hints->ai_socktype == SOCK_DGRAM) hints->ai_protocol = IPPROTO_UDP; else if (hints->ai_socktype == SOCK_STREAM) hints->ai_protocol = IPPROTO_TCP; } /* Set the socktype if it isn't set. */ if (!hints->ai_socktype && hints->ai_protocol) { if (hints->ai_protocol == IPPROTO_UDP) hints->ai_socktype = SOCK_DGRAM; else if (hints->ai_protocol == IPPROTO_TCP) hints->ai_socktype = SOCK_STREAM; #ifdef IPPROTO_SCTP else if (hints->ai_protocol == IPPROTO_SCTP) hints->ai_socktype = SOCK_STREAM; #endif } } #if AF_UNSPEC != PF_UNSPEC #error "I cannot build on a system where AF_UNSPEC != PF_UNSPEC" #endif /** Implements the part of looking up hosts by name that's common to both * the blocking and nonblocking resolver: * - Adjust 'hints' to have a reasonable socktype and protocol. * - Look up the port based on 'servname', and store it in *portnum, * - Handle the nodename==NULL case * - Handle some invalid arguments cases. * - Handle the cases where nodename is an IPv4 or IPv6 address. * * If we need the resolver to look up the hostname, we return * EVUTIL_EAI_NEED_RESOLVE. Otherwise, we can completely implement * getaddrinfo: we return 0 or an appropriate EVUTIL_EAI_* error, and * set *res as getaddrinfo would. */ int evutil_getaddrinfo_common_(const char *nodename, const char *servname, struct evutil_addrinfo *hints, struct evutil_addrinfo **res, int *portnum) { int port = 0; unsigned int if_index; const char *pname; if (nodename == NULL && servname == NULL) return EVUTIL_EAI_NONAME; /* We only understand 3 families */ if (hints->ai_family != PF_UNSPEC && hints->ai_family != PF_INET && hints->ai_family != PF_INET6) return EVUTIL_EAI_FAMILY; evutil_getaddrinfo_infer_protocols(hints); /* Look up the port number and protocol, if possible. */ pname = evutil_unparse_protoname(hints->ai_protocol); if (servname) { /* XXXX We could look at the protocol we got back from * getservbyname, but it doesn't seem too useful. */ port = evutil_parse_servname(servname, pname, hints); if (port < 0) { return EVUTIL_EAI_NONAME; } } /* If we have no node name, then we're supposed to bind to 'any' and * connect to localhost. */ if (nodename == NULL) { struct evutil_addrinfo *res4=NULL, *res6=NULL; if (hints->ai_family != PF_INET) { /* INET6 or UNSPEC. */ struct sockaddr_in6 sin6; memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(port); if (hints->ai_flags & EVUTIL_AI_PASSIVE) { /* Bind to :: */ } else { /* connect to ::1 */ sin6.sin6_addr.s6_addr[15] = 1; } res6 = evutil_new_addrinfo_((struct sockaddr*)&sin6, sizeof(sin6), hints); if (!res6) return EVUTIL_EAI_MEMORY; } if (hints->ai_family != PF_INET6) { /* INET or UNSPEC */ struct sockaddr_in sin; memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_port = htons(port); if (hints->ai_flags & EVUTIL_AI_PASSIVE) { /* Bind to 0.0.0.0 */ } else { /* connect to 127.0.0.1 */ sin.sin_addr.s_addr = htonl(0x7f000001); } res4 = evutil_new_addrinfo_((struct sockaddr*)&sin, sizeof(sin), hints); if (!res4) { if (res6) evutil_freeaddrinfo(res6); return EVUTIL_EAI_MEMORY; } } *res = evutil_addrinfo_append_(res4, res6); return 0; } /* If we can, we should try to parse the hostname without resolving * it. */ /* Try ipv6. */ if (hints->ai_family == PF_INET6 || hints->ai_family == PF_UNSPEC) { struct sockaddr_in6 sin6; memset(&sin6, 0, sizeof(sin6)); if (1 == evutil_inet_pton_scope( AF_INET6, nodename, &sin6.sin6_addr, &if_index)) { /* Got an ipv6 address. */ sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(port); sin6.sin6_scope_id = if_index; *res = evutil_new_addrinfo_((struct sockaddr*)&sin6, sizeof(sin6), hints); if (!*res) return EVUTIL_EAI_MEMORY; return 0; } } /* Try ipv4. */ if (hints->ai_family == PF_INET || hints->ai_family == PF_UNSPEC) { struct sockaddr_in sin; memset(&sin, 0, sizeof(sin)); if (1==evutil_inet_pton(AF_INET, nodename, &sin.sin_addr)) { /* Got an ipv4 address. */ sin.sin_family = AF_INET; sin.sin_port = htons(port); *res = evutil_new_addrinfo_((struct sockaddr*)&sin, sizeof(sin), hints); if (!*res) return EVUTIL_EAI_MEMORY; return 0; } } /* If we have reached this point, we definitely need to do a DNS * lookup. */ if ((hints->ai_flags & EVUTIL_AI_NUMERICHOST)) { /* If we're not allowed to do one, then say so. */ return EVUTIL_EAI_NONAME; } *portnum = port; return EVUTIL_EAI_NEED_RESOLVE; } #ifdef EVENT__HAVE_GETADDRINFO #define USE_NATIVE_GETADDRINFO #endif #ifdef USE_NATIVE_GETADDRINFO /* A mask of all the flags that we declare, so we can clear them before calling * the native getaddrinfo */ static const unsigned int ALL_NONNATIVE_AI_FLAGS = #ifndef AI_PASSIVE EVUTIL_AI_PASSIVE | #endif #ifndef AI_CANONNAME EVUTIL_AI_CANONNAME | #endif #ifndef AI_NUMERICHOST EVUTIL_AI_NUMERICHOST | #endif #ifndef AI_NUMERICSERV EVUTIL_AI_NUMERICSERV | #endif #ifndef AI_ADDRCONFIG EVUTIL_AI_ADDRCONFIG | #endif #ifndef AI_ALL EVUTIL_AI_ALL | #endif #ifndef AI_V4MAPPED EVUTIL_AI_V4MAPPED | #endif EVUTIL_AI_LIBEVENT_ALLOCATED; static const unsigned int ALL_NATIVE_AI_FLAGS = #ifdef AI_PASSIVE AI_PASSIVE | #endif #ifdef AI_CANONNAME AI_CANONNAME | #endif #ifdef AI_NUMERICHOST AI_NUMERICHOST | #endif #ifdef AI_NUMERICSERV AI_NUMERICSERV | #endif #ifdef AI_ADDRCONFIG AI_ADDRCONFIG | #endif #ifdef AI_ALL AI_ALL | #endif #ifdef AI_V4MAPPED AI_V4MAPPED | #endif 0; #endif #ifndef USE_NATIVE_GETADDRINFO /* Helper for systems with no getaddrinfo(): make one or more addrinfos out of * a struct hostent. */ static struct evutil_addrinfo * addrinfo_from_hostent(const struct hostent *ent, int port, const struct evutil_addrinfo *hints) { int i; struct sockaddr_in sin; struct sockaddr_in6 sin6; struct sockaddr *sa; int socklen; struct evutil_addrinfo *res=NULL, *ai; void *addrp; if (ent->h_addrtype == PF_INET) { memset(&sin, 0, sizeof(sin)); sin.sin_family = AF_INET; sin.sin_port = htons(port); sa = (struct sockaddr *)&sin; socklen = sizeof(struct sockaddr_in); addrp = &sin.sin_addr; if (ent->h_length != sizeof(sin.sin_addr)) { event_warnx("Weird h_length from gethostbyname"); return NULL; } } else if (ent->h_addrtype == PF_INET6) { memset(&sin6, 0, sizeof(sin6)); sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(port); sa = (struct sockaddr *)&sin6; socklen = sizeof(struct sockaddr_in6); addrp = &sin6.sin6_addr; if (ent->h_length != sizeof(sin6.sin6_addr)) { event_warnx("Weird h_length from gethostbyname"); return NULL; } } else return NULL; for (i = 0; ent->h_addr_list[i]; ++i) { memcpy(addrp, ent->h_addr_list[i], ent->h_length); ai = evutil_new_addrinfo_(sa, socklen, hints); if (!ai) { evutil_freeaddrinfo(res); return NULL; } res = evutil_addrinfo_append_(res, ai); } if (res && ((hints->ai_flags & EVUTIL_AI_CANONNAME) && ent->h_name)) { res->ai_canonname = mm_strdup(ent->h_name); if (res->ai_canonname == NULL) { evutil_freeaddrinfo(res); return NULL; } } return res; } #endif /* If the EVUTIL_AI_ADDRCONFIG flag is set on hints->ai_flags, and * hints->ai_family is PF_UNSPEC, then revise the value of hints->ai_family so * that we'll only get addresses we could maybe connect to. */ void evutil_adjust_hints_for_addrconfig_(struct evutil_addrinfo *hints) { if (!(hints->ai_flags & EVUTIL_AI_ADDRCONFIG)) return; if (hints->ai_family != PF_UNSPEC) return; evutil_check_interfaces(); if (had_ipv4_address && !had_ipv6_address) { hints->ai_family = PF_INET; } else if (!had_ipv4_address && had_ipv6_address) { hints->ai_family = PF_INET6; } } #ifdef USE_NATIVE_GETADDRINFO static int need_numeric_port_hack_=0; static int need_socktype_protocol_hack_=0; static int tested_for_getaddrinfo_hacks=0; /* Some older BSDs (like OpenBSD up to 4.6) used to believe that giving a numeric port without giving an ai_socktype was verboten. We test for this so we can apply an appropriate workaround. If it turns out that the bug is present, then: - If nodename==NULL and servname is numeric, we build an answer ourselves using evutil_getaddrinfo_common_(). - If nodename!=NULL and servname is numeric, then we set servname=NULL when calling getaddrinfo, and post-process the result to set the ports on it. We test for this bug at runtime, since otherwise we can't have the same binary run on multiple BSD versions. - Some versions of Solaris believe that it's nice to leave to protocol field set to 0. We test for this so we can apply an appropriate workaround. */ static struct evutil_addrinfo *ai_find_protocol(struct evutil_addrinfo *ai) { while (ai) { if (ai->ai_protocol) return ai; ai = ai->ai_next; } return NULL; } static void test_for_getaddrinfo_hacks(void) { int r, r2; struct evutil_addrinfo *ai=NULL, *ai2=NULL, *ai3=NULL; struct evutil_addrinfo hints; memset(&hints,0,sizeof(hints)); hints.ai_family = PF_UNSPEC; hints.ai_flags = #ifdef AI_NUMERICHOST AI_NUMERICHOST | #endif #ifdef AI_NUMERICSERV AI_NUMERICSERV | #endif 0; r = getaddrinfo("1.2.3.4", "80", &hints, &ai); getaddrinfo("1.2.3.4", NULL, &hints, &ai3); hints.ai_socktype = SOCK_STREAM; r2 = getaddrinfo("1.2.3.4", "80", &hints, &ai2); if (r2 == 0 && r != 0) { need_numeric_port_hack_=1; } if (!ai_find_protocol(ai2) || !ai_find_protocol(ai3)) { need_socktype_protocol_hack_=1; } if (ai) freeaddrinfo(ai); if (ai2) freeaddrinfo(ai2); if (ai3) freeaddrinfo(ai3); tested_for_getaddrinfo_hacks=1; } static inline int need_numeric_port_hack(void) { if (!tested_for_getaddrinfo_hacks) test_for_getaddrinfo_hacks(); return need_numeric_port_hack_; } static inline int need_socktype_protocol_hack(void) { if (!tested_for_getaddrinfo_hacks) test_for_getaddrinfo_hacks(); return need_socktype_protocol_hack_; } static void apply_numeric_port_hack(int port, struct evutil_addrinfo **ai) { /* Now we run through the list and set the ports on all of the * results where ports would make sense. */ for ( ; *ai; ai = &(*ai)->ai_next) { struct sockaddr *sa = (*ai)->ai_addr; if (sa && sa->sa_family == AF_INET) { struct sockaddr_in *sin = (struct sockaddr_in*)sa; sin->sin_port = htons(port); } else if (sa && sa->sa_family == AF_INET6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)sa; sin6->sin6_port = htons(port); } else { /* A numeric port makes no sense here; remove this one * from the list. */ struct evutil_addrinfo *victim = *ai; *ai = victim->ai_next; victim->ai_next = NULL; freeaddrinfo(victim); } } } static int apply_socktype_protocol_hack(struct evutil_addrinfo *ai) { struct evutil_addrinfo *ai_new; for (; ai; ai = ai->ai_next) { evutil_getaddrinfo_infer_protocols(ai); if (ai->ai_socktype || ai->ai_protocol) continue; ai_new = mm_malloc(sizeof(*ai_new)); if (!ai_new) return -1; memcpy(ai_new, ai, sizeof(*ai_new)); ai->ai_socktype = SOCK_STREAM; ai->ai_protocol = IPPROTO_TCP; ai_new->ai_socktype = SOCK_DGRAM; ai_new->ai_protocol = IPPROTO_UDP; ai_new->ai_next = ai->ai_next; ai->ai_next = ai_new; } return 0; } #endif int evutil_getaddrinfo(const char *nodename, const char *servname, const struct evutil_addrinfo *hints_in, struct evutil_addrinfo **res) { #ifdef USE_NATIVE_GETADDRINFO struct evutil_addrinfo hints; int portnum=-1, need_np_hack, err; if (hints_in) { memcpy(&hints, hints_in, sizeof(hints)); } else { memset(&hints, 0, sizeof(hints)); hints.ai_family = PF_UNSPEC; } #ifndef AI_ADDRCONFIG /* Not every system has AI_ADDRCONFIG, so fake it. */ if (hints.ai_family == PF_UNSPEC && (hints.ai_flags & EVUTIL_AI_ADDRCONFIG)) { evutil_adjust_hints_for_addrconfig_(&hints); } #endif #ifndef AI_NUMERICSERV /* Not every system has AI_NUMERICSERV, so fake it. */ if (hints.ai_flags & EVUTIL_AI_NUMERICSERV) { if (servname && parse_numeric_servname(servname)<0) return EVUTIL_EAI_NONAME; } #endif /* Enough operating systems handle enough common non-resolve * cases here weirdly enough that we are better off just * overriding them. For example: * * - Windows doesn't like to infer the protocol from the * socket type, or fill in socket or protocol types much at * all. It also seems to do its own broken implicit * always-on version of AI_ADDRCONFIG that keeps it from * ever resolving even a literal IPv6 address when * ai_addrtype is PF_UNSPEC. */ #ifdef _WIN32 { int tmp_port; err = evutil_getaddrinfo_common_(nodename,servname,&hints, res, &tmp_port); if (err == 0 || err == EVUTIL_EAI_MEMORY || err == EVUTIL_EAI_NONAME) return err; /* If we make it here, the system getaddrinfo can * have a crack at it. */ } #endif /* See documentation for need_numeric_port_hack above.*/ need_np_hack = need_numeric_port_hack() && servname && !hints.ai_socktype && ((portnum=parse_numeric_servname(servname)) >= 0); if (need_np_hack) { if (!nodename) return evutil_getaddrinfo_common_( NULL,servname,&hints, res, &portnum); servname = NULL; } if (need_socktype_protocol_hack()) { evutil_getaddrinfo_infer_protocols(&hints); } /* Make sure that we didn't actually steal any AI_FLAGS values that * the system is using. (This is a constant expression, and should ge * optimized out.) * * XXXX Turn this into a compile-time failure rather than a run-time * failure. */ EVUTIL_ASSERT((ALL_NONNATIVE_AI_FLAGS & ALL_NATIVE_AI_FLAGS) == 0); /* Clear any flags that only libevent understands. */ hints.ai_flags &= ~ALL_NONNATIVE_AI_FLAGS; err = getaddrinfo(nodename, servname, &hints, res); if (need_np_hack) apply_numeric_port_hack(portnum, res); if (need_socktype_protocol_hack()) { if (apply_socktype_protocol_hack(*res) < 0) { evutil_freeaddrinfo(*res); *res = NULL; return EVUTIL_EAI_MEMORY; } } return err; #else int port=0, err; struct hostent *ent = NULL; struct evutil_addrinfo hints; if (hints_in) { memcpy(&hints, hints_in, sizeof(hints)); } else { memset(&hints, 0, sizeof(hints)); hints.ai_family = PF_UNSPEC; } evutil_adjust_hints_for_addrconfig_(&hints); err = evutil_getaddrinfo_common_(nodename, servname, &hints, res, &port); if (err != EVUTIL_EAI_NEED_RESOLVE) { /* We either succeeded or failed. No need to continue */ return err; } err = 0; /* Use any of the various gethostbyname_r variants as available. */ { #ifdef EVENT__HAVE_GETHOSTBYNAME_R_6_ARG /* This one is what glibc provides. */ char buf[2048]; struct hostent hostent; int r; r = gethostbyname_r(nodename, &hostent, buf, sizeof(buf), &ent, &err); #elif defined(EVENT__HAVE_GETHOSTBYNAME_R_5_ARG) char buf[2048]; struct hostent hostent; ent = gethostbyname_r(nodename, &hostent, buf, sizeof(buf), &err); #elif defined(EVENT__HAVE_GETHOSTBYNAME_R_3_ARG) struct hostent_data data; struct hostent hostent; memset(&data, 0, sizeof(data)); err = gethostbyname_r(nodename, &hostent, &data); ent = err ? NULL : &hostent; #else /* fall back to gethostbyname. */ /* XXXX This needs a lock everywhere but Windows. */ ent = gethostbyname(nodename); #ifdef _WIN32 err = WSAGetLastError(); #else err = h_errno; #endif #endif /* Now we have either ent or err set. */ if (!ent) { /* XXX is this right for windows ? */ switch (err) { case TRY_AGAIN: return EVUTIL_EAI_AGAIN; case NO_RECOVERY: default: return EVUTIL_EAI_FAIL; case HOST_NOT_FOUND: return EVUTIL_EAI_NONAME; case NO_ADDRESS: #if NO_DATA != NO_ADDRESS case NO_DATA: #endif return EVUTIL_EAI_NODATA; } } if (ent->h_addrtype != hints.ai_family && hints.ai_family != PF_UNSPEC) { /* This wasn't the type we were hoping for. Too bad * we never had a chance to ask gethostbyname for what * we wanted. */ return EVUTIL_EAI_NONAME; } /* Make sure we got _some_ answers. */ if (ent->h_length == 0) return EVUTIL_EAI_NODATA; /* If we got an address type we don't know how to make a sockaddr for, give up. */ if (ent->h_addrtype != PF_INET && ent->h_addrtype != PF_INET6) return EVUTIL_EAI_FAMILY; *res = addrinfo_from_hostent(ent, port, &hints); if (! *res) return EVUTIL_EAI_MEMORY; } return 0; #endif } void evutil_freeaddrinfo(struct evutil_addrinfo *ai) { #ifdef EVENT__HAVE_GETADDRINFO if (!(ai->ai_flags & EVUTIL_AI_LIBEVENT_ALLOCATED)) { freeaddrinfo(ai); return; } #endif while (ai) { struct evutil_addrinfo *next = ai->ai_next; if (ai->ai_canonname) mm_free(ai->ai_canonname); mm_free(ai); ai = next; } } static evdns_getaddrinfo_fn evdns_getaddrinfo_impl = NULL; static evdns_getaddrinfo_cancel_fn evdns_getaddrinfo_cancel_impl = NULL; void evutil_set_evdns_getaddrinfo_fn_(evdns_getaddrinfo_fn fn) { if (!evdns_getaddrinfo_impl) evdns_getaddrinfo_impl = fn; } void evutil_set_evdns_getaddrinfo_cancel_fn_(evdns_getaddrinfo_cancel_fn fn) { if (!evdns_getaddrinfo_cancel_impl) evdns_getaddrinfo_cancel_impl = fn; } /* Internal helper function: act like evdns_getaddrinfo if dns_base is set; * otherwise do a blocking resolve and pass the result to the callback in the * way that evdns_getaddrinfo would. */ struct evdns_getaddrinfo_request *evutil_getaddrinfo_async_( struct evdns_base *dns_base, const char *nodename, const char *servname, const struct evutil_addrinfo *hints_in, void (*cb)(int, struct evutil_addrinfo *, void *), void *arg) { if (dns_base && evdns_getaddrinfo_impl) { return evdns_getaddrinfo_impl( dns_base, nodename, servname, hints_in, cb, arg); } else { struct evutil_addrinfo *ai=NULL; int err; err = evutil_getaddrinfo(nodename, servname, hints_in, &ai); cb(err, ai, arg); return NULL; } } void evutil_getaddrinfo_cancel_async_(struct evdns_getaddrinfo_request *data) { if (evdns_getaddrinfo_cancel_impl && data) { evdns_getaddrinfo_cancel_impl(data); } } const char * evutil_gai_strerror(int err) { /* As a sneaky side-benefit, this case statement will get most * compilers to tell us if any of the error codes we defined * conflict with the platform's native error codes. */ switch (err) { case EVUTIL_EAI_CANCEL: return "Request canceled"; case 0: return "No error"; case EVUTIL_EAI_ADDRFAMILY: return "address family for nodename not supported"; case EVUTIL_EAI_AGAIN: return "temporary failure in name resolution"; case EVUTIL_EAI_BADFLAGS: return "invalid value for ai_flags"; case EVUTIL_EAI_FAIL: return "non-recoverable failure in name resolution"; case EVUTIL_EAI_FAMILY: return "ai_family not supported"; case EVUTIL_EAI_MEMORY: return "memory allocation failure"; case EVUTIL_EAI_NODATA: return "no address associated with nodename"; case EVUTIL_EAI_NONAME: return "nodename nor servname provided, or not known"; case EVUTIL_EAI_SERVICE: return "servname not supported for ai_socktype"; case EVUTIL_EAI_SOCKTYPE: return "ai_socktype not supported"; case EVUTIL_EAI_SYSTEM: return "system error"; default: #if defined(USE_NATIVE_GETADDRINFO) && defined(_WIN32) return gai_strerrorA(err); #elif defined(USE_NATIVE_GETADDRINFO) return gai_strerror(err); #else return "Unknown error code"; #endif } } #ifdef _WIN32 /* destructively remove a trailing line terminator from s */ static void chomp (char *s) { size_t len; if (s && (len = strlen (s)) > 0 && s[len - 1] == '\n') { s[--len] = 0; if (len > 0 && s[len - 1] == '\r') s[--len] = 0; } } /* FormatMessage returns allocated strings, but evutil_socket_error_to_string * is supposed to return a string which is good indefinitely without having * to be freed. To make this work without leaking memory, we cache the * string the first time FormatMessage is called on a particular error * code, and then return the cached string on subsequent calls with the * same code. The strings aren't freed until libevent_global_shutdown * (or never). We use a linked list to cache the errors, because we * only expect there to be a few dozen, and that should be fast enough. */ struct cached_sock_errs_entry { HT_ENTRY(cached_sock_errs_entry) node; DWORD code; char *msg; /* allocated with LocalAlloc; free with LocalFree */ }; static inline unsigned hash_cached_sock_errs(const struct cached_sock_errs_entry *e) { /* Use Murmur3's 32-bit finalizer as an integer hash function */ DWORD h = e->code; h ^= h >> 16; h *= 0x85ebca6b; h ^= h >> 13; h *= 0xc2b2ae35; h ^= h >> 16; return h; } static inline int eq_cached_sock_errs(const struct cached_sock_errs_entry *a, const struct cached_sock_errs_entry *b) { return a->code == b->code; } #ifndef EVENT__DISABLE_THREAD_SUPPORT static void *windows_socket_errors_lock_ = NULL; #endif static HT_HEAD(cached_sock_errs_map, cached_sock_errs_entry) windows_socket_errors = HT_INITIALIZER(); HT_PROTOTYPE(cached_sock_errs_map, cached_sock_errs_entry, node, hash_cached_sock_errs, eq_cached_sock_errs); HT_GENERATE(cached_sock_errs_map, cached_sock_errs_entry, node, hash_cached_sock_errs, eq_cached_sock_errs, 0.5, mm_malloc, mm_realloc, mm_free); /** Equivalent to strerror, but for windows socket errors. */ const char * evutil_socket_error_to_string(int errcode) { struct cached_sock_errs_entry *errs, *newerr, find; char *msg = NULL; EVLOCK_LOCK(windows_socket_errors_lock_, 0); find.code = errcode; errs = HT_FIND(cached_sock_errs_map, &windows_socket_errors, &find); if (errs) { msg = errs->msg; goto done; } if (0 != FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS | FORMAT_MESSAGE_ALLOCATE_BUFFER, NULL, errcode, 0, (char *)&msg, 0, NULL)) chomp (msg); /* because message has trailing newline */ else { size_t len = 50; /* use LocalAlloc because FormatMessage does */ msg = LocalAlloc(LMEM_FIXED, len); if (!msg) { msg = (char *)"LocalAlloc failed during Winsock error"; goto done; } evutil_snprintf(msg, len, "winsock error 0x%08x", errcode); } newerr = (struct cached_sock_errs_entry *) mm_malloc(sizeof (struct cached_sock_errs_entry)); if (!newerr) { LocalFree(msg); msg = (char *)"malloc failed during Winsock error"; goto done; } newerr->code = errcode; newerr->msg = msg; HT_INSERT(cached_sock_errs_map, &windows_socket_errors, newerr); done: EVLOCK_UNLOCK(windows_socket_errors_lock_, 0); return msg; } #ifndef EVENT__DISABLE_THREAD_SUPPORT int evutil_global_setup_locks_(const int enable_locks) { EVTHREAD_SETUP_GLOBAL_LOCK(windows_socket_errors_lock_, 0); return 0; } #endif static void evutil_free_sock_err_globals(void) { struct cached_sock_errs_entry **errs, *tofree; for (errs = HT_START(cached_sock_errs_map, &windows_socket_errors) ; errs; ) { tofree = *errs; errs = HT_NEXT_RMV(cached_sock_errs_map, &windows_socket_errors, errs); LocalFree(tofree->msg); mm_free(tofree); } HT_CLEAR(cached_sock_errs_map, &windows_socket_errors); #ifndef EVENT__DISABLE_THREAD_SUPPORT if (windows_socket_errors_lock_ != NULL) { EVTHREAD_FREE_LOCK(windows_socket_errors_lock_, 0); windows_socket_errors_lock_ = NULL; } #endif } #else #ifndef EVENT__DISABLE_THREAD_SUPPORT int evutil_global_setup_locks_(const int enable_locks) { return 0; } #endif static void evutil_free_sock_err_globals(void) { } #endif int evutil_snprintf(char *buf, size_t buflen, const char *format, ...) { int r; va_list ap; va_start(ap, format); r = evutil_vsnprintf(buf, buflen, format, ap); va_end(ap); return r; } int evutil_vsnprintf(char *buf, size_t buflen, const char *format, va_list ap) { int r; if (!buflen) return 0; #if defined(_MSC_VER) || defined(_WIN32) r = _vsnprintf(buf, buflen, format, ap); if (r < 0) r = _vscprintf(format, ap); #elif defined(sgi) /* Make sure we always use the correct vsnprintf on IRIX */ extern int _xpg5_vsnprintf(char * __restrict, __SGI_LIBC_NAMESPACE_QUALIFIER size_t, const char * __restrict, /* va_list */ char *); r = _xpg5_vsnprintf(buf, buflen, format, ap); #else r = vsnprintf(buf, buflen, format, ap); #endif buf[buflen-1] = '\0'; return r; } #define USE_INTERNAL_NTOP #define USE_INTERNAL_PTON const char * evutil_inet_ntop(int af, const void *src, char *dst, size_t len) { #if defined(EVENT__HAVE_INET_NTOP) && !defined(USE_INTERNAL_NTOP) return inet_ntop(af, src, dst, len); #else if (af == AF_INET) { const struct in_addr *in = src; const ev_uint32_t a = ntohl(in->s_addr); int r; r = evutil_snprintf(dst, len, "%d.%d.%d.%d", (int)(ev_uint8_t)((a>>24)&0xff), (int)(ev_uint8_t)((a>>16)&0xff), (int)(ev_uint8_t)((a>>8 )&0xff), (int)(ev_uint8_t)((a )&0xff)); if (r<0||(size_t)r>=len) return NULL; else return dst; #ifdef AF_INET6 } else if (af == AF_INET6) { const struct in6_addr *addr = src; char buf[64], *cp; int longestGapLen = 0, longestGapPos = -1, i, curGapPos = -1, curGapLen = 0; ev_uint16_t words[8]; for (i = 0; i < 8; ++i) { words[i] = (((ev_uint16_t)addr->s6_addr[2*i])<<8) + addr->s6_addr[2*i+1]; } if (words[0] == 0 && words[1] == 0 && words[2] == 0 && words[3] == 0 && words[4] == 0 && ((words[5] == 0 && words[6] && words[7]) || (words[5] == 0xffff))) { /* This is an IPv4 address. */ if (words[5] == 0) { evutil_snprintf(buf, sizeof(buf), "::%d.%d.%d.%d", addr->s6_addr[12], addr->s6_addr[13], addr->s6_addr[14], addr->s6_addr[15]); } else { evutil_snprintf(buf, sizeof(buf), "::%x:%d.%d.%d.%d", words[5], addr->s6_addr[12], addr->s6_addr[13], addr->s6_addr[14], addr->s6_addr[15]); } if (strlen(buf) > len) return NULL; strlcpy(dst, buf, len); return dst; } i = 0; while (i < 8) { if (words[i] == 0) { curGapPos = i++; curGapLen = 1; while (i<8 && words[i] == 0) { ++i; ++curGapLen; } if (curGapLen > longestGapLen) { longestGapPos = curGapPos; longestGapLen = curGapLen; } } else { ++i; } } if (longestGapLen<=1) longestGapPos = -1; cp = buf; for (i = 0; i < 8; ++i) { if (words[i] == 0 && longestGapPos == i) { if (i == 0) *cp++ = ':'; *cp++ = ':'; while (i < 8 && words[i] == 0) ++i; --i; /* to compensate for loop increment. */ } else { evutil_snprintf(cp, sizeof(buf)-(cp-buf), "%x", (unsigned)words[i]); cp += strlen(cp); if (i != 7) *cp++ = ':'; } } *cp = '\0'; if (strlen(buf) > len) return NULL; strlcpy(dst, buf, len); return dst; #endif } else { return NULL; } #endif } int evutil_inet_pton_scope(int af, const char *src, void *dst, unsigned *indexp) { int r; unsigned if_index; char *check, *cp, *tmp_src; *indexp = 0; /* Reasonable default */ /* Bail out if not IPv6 */ if (af != AF_INET6) return evutil_inet_pton(af, src, dst); cp = strchr(src, '%'); /* Bail out if no zone ID */ if (cp == NULL) return evutil_inet_pton(af, src, dst); if_index = if_nametoindex(cp + 1); if (if_index == 0) { /* Could be numeric */ if_index = strtoul(cp + 1, &check, 10); if (check[0] != '\0') return 0; } *indexp = if_index; tmp_src = mm_strdup(src); cp = strchr(tmp_src, '%'); *cp = '\0'; r = evutil_inet_pton(af, tmp_src, dst); free(tmp_src); return r; } int evutil_inet_pton(int af, const char *src, void *dst) { #if defined(EVENT__HAVE_INET_PTON) && !defined(USE_INTERNAL_PTON) return inet_pton(af, src, dst); #else if (af == AF_INET) { unsigned a,b,c,d; char more; struct in_addr *addr = dst; if (sscanf(src, "%u.%u.%u.%u%c", &a,&b,&c,&d,&more) != 4) return 0; if (a > 255) return 0; if (b > 255) return 0; if (c > 255) return 0; if (d > 255) return 0; addr->s_addr = htonl((a<<24) | (b<<16) | (c<<8) | d); return 1; #ifdef AF_INET6 } else if (af == AF_INET6) { struct in6_addr *out = dst; ev_uint16_t words[8]; int gapPos = -1, i, setWords=0; const char *dot = strchr(src, '.'); const char *eow; /* end of words. */ if (dot == src) return 0; else if (!dot) eow = src+strlen(src); else { unsigned byte1,byte2,byte3,byte4; char more; for (eow = dot-1; eow >= src && EVUTIL_ISDIGIT_(*eow); --eow) ; ++eow; /* We use "scanf" because some platform inet_aton()s are too lax * about IPv4 addresses of the form "1.2.3" */ if (sscanf(eow, "%u.%u.%u.%u%c", &byte1,&byte2,&byte3,&byte4,&more) != 4) return 0; if (byte1 > 255 || byte2 > 255 || byte3 > 255 || byte4 > 255) return 0; words[6] = (byte1<<8) | byte2; words[7] = (byte3<<8) | byte4; setWords += 2; } i = 0; while (src < eow) { if (i > 7) return 0; if (EVUTIL_ISXDIGIT_(*src)) { char *next; long r = strtol(src, &next, 16); if (next > 4+src) return 0; if (next == src) return 0; if (r<0 || r>65536) return 0; words[i++] = (ev_uint16_t)r; setWords++; src = next; if (*src != ':' && src != eow) return 0; ++src; } else if (*src == ':' && i > 0 && gapPos==-1) { gapPos = i; ++src; } else if (*src == ':' && i == 0 && src[1] == ':' && gapPos==-1) { gapPos = i; src += 2; } else { return 0; } } if (setWords > 8 || (setWords == 8 && gapPos != -1) || (setWords < 8 && gapPos == -1)) return 0; if (gapPos >= 0) { int nToMove = setWords - (dot ? 2 : 0) - gapPos; int gapLen = 8 - setWords; /* assert(nToMove >= 0); */ if (nToMove < 0) return -1; /* should be impossible */ memmove(&words[gapPos+gapLen], &words[gapPos], sizeof(ev_uint16_t)*nToMove); memset(&words[gapPos], 0, sizeof(ev_uint16_t)*gapLen); } for (i = 0; i < 8; ++i) { out->s6_addr[2*i ] = words[i] >> 8; out->s6_addr[2*i+1] = words[i] & 0xff; } return 1; #endif } else { return -1; } #endif } int evutil_parse_sockaddr_port(const char *ip_as_string, struct sockaddr *out, int *outlen) { int port; unsigned int if_index; char buf[128]; const char *cp, *addr_part, *port_part; int is_ipv6; /* recognized formats are: * [ipv6]:port * ipv6 * [ipv6] * ipv4:port * ipv4 */ cp = strchr(ip_as_string, ':'); if (*ip_as_string == '[') { size_t len; if (!(cp = strchr(ip_as_string, ']'))) { return -1; } len = ( cp-(ip_as_string + 1) ); if (len > sizeof(buf)-1) { return -1; } memcpy(buf, ip_as_string+1, len); buf[len] = '\0'; addr_part = buf; if (cp[1] == ':') port_part = cp+2; else port_part = NULL; is_ipv6 = 1; } else if (cp && strchr(cp+1, ':')) { is_ipv6 = 1; addr_part = ip_as_string; port_part = NULL; } else if (cp) { is_ipv6 = 0; if (cp - ip_as_string > (int)sizeof(buf)-1) { return -1; } memcpy(buf, ip_as_string, cp-ip_as_string); buf[cp-ip_as_string] = '\0'; addr_part = buf; port_part = cp+1; } else { addr_part = ip_as_string; port_part = NULL; is_ipv6 = 0; } if (port_part == NULL) { port = 0; } else { port = atoi(port_part); if (port <= 0 || port > 65535) { return -1; } } if (!addr_part) return -1; /* Should be impossible. */ #ifdef AF_INET6 if (is_ipv6) { struct sockaddr_in6 sin6; memset(&sin6, 0, sizeof(sin6)); #ifdef EVENT__HAVE_STRUCT_SOCKADDR_IN6_SIN6_LEN sin6.sin6_len = sizeof(sin6); #endif sin6.sin6_family = AF_INET6; sin6.sin6_port = htons(port); if (1 != evutil_inet_pton_scope( AF_INET6, addr_part, &sin6.sin6_addr, &if_index)) { return -1; } if ((int)sizeof(sin6) > *outlen) return -1; sin6.sin6_scope_id = if_index; memset(out, 0, *outlen); memcpy(out, &sin6, sizeof(sin6)); *outlen = sizeof(sin6); return 0; } else #endif { struct sockaddr_in sin; memset(&sin, 0, sizeof(sin)); #ifdef EVENT__HAVE_STRUCT_SOCKADDR_IN_SIN_LEN sin.sin_len = sizeof(sin); #endif sin.sin_family = AF_INET; sin.sin_port = htons(port); if (1 != evutil_inet_pton(AF_INET, addr_part, &sin.sin_addr)) return -1; if ((int)sizeof(sin) > *outlen) return -1; memset(out, 0, *outlen); memcpy(out, &sin, sizeof(sin)); *outlen = sizeof(sin); return 0; } } const char * evutil_format_sockaddr_port_(const struct sockaddr *sa, char *out, size_t outlen) { char b[128]; const char *res=NULL; int port; if (sa->sa_family == AF_INET) { const struct sockaddr_in *sin = (const struct sockaddr_in*)sa; res = evutil_inet_ntop(AF_INET, &sin->sin_addr,b,sizeof(b)); port = ntohs(sin->sin_port); if (res) { evutil_snprintf(out, outlen, "%s:%d", b, port); return out; } } else if (sa->sa_family == AF_INET6) { const struct sockaddr_in6 *sin6 = (const struct sockaddr_in6*)sa; res = evutil_inet_ntop(AF_INET6, &sin6->sin6_addr,b,sizeof(b)); port = ntohs(sin6->sin6_port); if (res) { evutil_snprintf(out, outlen, "[%s]:%d", b, port); return out; } } evutil_snprintf(out, outlen, "", (int)sa->sa_family); return out; } int evutil_sockaddr_cmp(const struct sockaddr *sa1, const struct sockaddr *sa2, int include_port) { int r; if (0 != (r = (sa1->sa_family - sa2->sa_family))) return r; if (sa1->sa_family == AF_INET) { const struct sockaddr_in *sin1, *sin2; sin1 = (const struct sockaddr_in *)sa1; sin2 = (const struct sockaddr_in *)sa2; if (sin1->sin_addr.s_addr < sin2->sin_addr.s_addr) return -1; else if (sin1->sin_addr.s_addr > sin2->sin_addr.s_addr) return 1; else if (include_port && (r = ((int)sin1->sin_port - (int)sin2->sin_port))) return r; else return 0; } #ifdef AF_INET6 else if (sa1->sa_family == AF_INET6) { const struct sockaddr_in6 *sin1, *sin2; sin1 = (const struct sockaddr_in6 *)sa1; sin2 = (const struct sockaddr_in6 *)sa2; if ((r = memcmp(sin1->sin6_addr.s6_addr, sin2->sin6_addr.s6_addr, 16))) return r; else if (include_port && (r = ((int)sin1->sin6_port - (int)sin2->sin6_port))) return r; else return 0; } #endif return 1; } /* Tables to implement ctypes-replacement EVUTIL_IS*() functions. Each table * has 256 bits to look up whether a character is in some set or not. This * fails on non-ASCII platforms, but so does every other place where we * take a char and write it onto the network. **/ static const ev_uint32_t EVUTIL_ISALPHA_TABLE[8] = { 0, 0, 0x7fffffe, 0x7fffffe, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISALNUM_TABLE[8] = { 0, 0x3ff0000, 0x7fffffe, 0x7fffffe, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISSPACE_TABLE[8] = { 0x3e00, 0x1, 0, 0, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISXDIGIT_TABLE[8] = { 0, 0x3ff0000, 0x7e, 0x7e, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISDIGIT_TABLE[8] = { 0, 0x3ff0000, 0, 0, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISPRINT_TABLE[8] = { 0, 0xffffffff, 0xffffffff, 0x7fffffff, 0, 0, 0, 0x0 }; static const ev_uint32_t EVUTIL_ISUPPER_TABLE[8] = { 0, 0, 0x7fffffe, 0, 0, 0, 0, 0 }; static const ev_uint32_t EVUTIL_ISLOWER_TABLE[8] = { 0, 0, 0, 0x7fffffe, 0, 0, 0, 0 }; /* Upper-casing and lowercasing tables to map characters to upper/lowercase * equivalents. */ static const unsigned char EVUTIL_TOUPPER_TABLE[256] = { 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31, 32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47, 48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63, 64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79, 80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95, 96,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79, 80,81,82,83,84,85,86,87,88,89,90,123,124,125,126,127, 128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143, 144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159, 160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175, 176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191, 192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207, 208,209,210,211,212,213,214,215,216,217,218,219,220,221,222,223, 224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,239, 240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255, }; static const unsigned char EVUTIL_TOLOWER_TABLE[256] = { 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31, 32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47, 48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63, 64,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111, 112,113,114,115,116,117,118,119,120,121,122,91,92,93,94,95, 96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111, 112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127, 128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143, 144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159, 160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175, 176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191, 192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207, 208,209,210,211,212,213,214,215,216,217,218,219,220,221,222,223, 224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,239, 240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255, }; #define IMPL_CTYPE_FN(name) \ int EVUTIL_##name##_(char c) { \ ev_uint8_t u = c; \ return !!(EVUTIL_##name##_TABLE[(u >> 5) & 7] & (1U << (u & 31))); \ } IMPL_CTYPE_FN(ISALPHA) IMPL_CTYPE_FN(ISALNUM) IMPL_CTYPE_FN(ISSPACE) IMPL_CTYPE_FN(ISDIGIT) IMPL_CTYPE_FN(ISXDIGIT) IMPL_CTYPE_FN(ISPRINT) IMPL_CTYPE_FN(ISLOWER) IMPL_CTYPE_FN(ISUPPER) char EVUTIL_TOLOWER_(char c) { return ((char)EVUTIL_TOLOWER_TABLE[(ev_uint8_t)c]); } char EVUTIL_TOUPPER_(char c) { return ((char)EVUTIL_TOUPPER_TABLE[(ev_uint8_t)c]); } int evutil_ascii_strcasecmp(const char *s1, const char *s2) { char c1, c2; while (1) { c1 = EVUTIL_TOLOWER_(*s1++); c2 = EVUTIL_TOLOWER_(*s2++); if (c1 < c2) return -1; else if (c1 > c2) return 1; else if (c1 == 0) return 0; } } int evutil_ascii_strncasecmp(const char *s1, const char *s2, size_t n) { char c1, c2; while (n--) { c1 = EVUTIL_TOLOWER_(*s1++); c2 = EVUTIL_TOLOWER_(*s2++); if (c1 < c2) return -1; else if (c1 > c2) return 1; else if (c1 == 0) return 0; } return 0; } void evutil_rtrim_lws_(char *str) { char *cp; if (str == NULL) return; if ((cp = strchr(str, '\0')) == NULL || (cp == str)) return; --cp; while (*cp == ' ' || *cp == '\t') { *cp = '\0'; if (cp == str) break; --cp; } } static int evutil_issetugid(void) { #ifdef EVENT__HAVE_ISSETUGID return issetugid(); #else #ifdef EVENT__HAVE_GETEUID if (getuid() != geteuid()) return 1; #endif #ifdef EVENT__HAVE_GETEGID if (getgid() != getegid()) return 1; #endif return 0; #endif } const char * evutil_getenv_(const char *varname) { if (evutil_issetugid()) return NULL; return getenv(varname); } ev_uint32_t evutil_weakrand_seed_(struct evutil_weakrand_state *state, ev_uint32_t seed) { if (seed == 0) { struct timeval tv; evutil_gettimeofday(&tv, NULL); seed = (ev_uint32_t)tv.tv_sec + (ev_uint32_t)tv.tv_usec; #ifdef _WIN32 seed += (ev_uint32_t) _getpid(); #else seed += (ev_uint32_t) getpid(); #endif } state->seed = seed; return seed; } ev_int32_t evutil_weakrand_(struct evutil_weakrand_state *state) { /* This RNG implementation is a linear congruential generator, with * modulus 2^31, multiplier 1103515245, and addend 12345. It's also * used by OpenBSD, and by Glibc's TYPE_0 RNG. * * The linear congruential generator is not an industrial-strength * RNG! It's fast, but it can have higher-order patterns. Notably, * the low bits tend to have periodicity. */ state->seed = ((state->seed) * 1103515245 + 12345) & 0x7fffffff; return (ev_int32_t)(state->seed); } ev_int32_t evutil_weakrand_range_(struct evutil_weakrand_state *state, ev_int32_t top) { ev_int32_t divisor, result; /* We can't just do weakrand() % top, since the low bits of the LCG * are less random than the high ones. (Specifically, since the LCG * modulus is 2^N, every 2^m for m= top); return result; } /** * Volatile pointer to memset: we use this to keep the compiler from * eliminating our call to memset. */ void * (*volatile evutil_memset_volatile_)(void *, int, size_t) = memset; void evutil_memclear_(void *mem, size_t len) { evutil_memset_volatile_(mem, 0, len); } int evutil_sockaddr_is_loopback_(const struct sockaddr *addr) { static const char LOOPBACK_S6[16] = "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\1"; if (addr->sa_family == AF_INET) { struct sockaddr_in *sin = (struct sockaddr_in *)addr; return (ntohl(sin->sin_addr.s_addr) & 0xff000000) == 0x7f000000; } else if (addr->sa_family == AF_INET6) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr; return !memcmp(sin6->sin6_addr.s6_addr, LOOPBACK_S6, 16); } return 0; } int evutil_hex_char_to_int_(char c) { switch(c) { case '0': return 0; case '1': return 1; case '2': return 2; case '3': return 3; case '4': return 4; case '5': return 5; case '6': return 6; case '7': return 7; case '8': return 8; case '9': return 9; case 'A': case 'a': return 10; case 'B': case 'b': return 11; case 'C': case 'c': return 12; case 'D': case 'd': return 13; case 'E': case 'e': return 14; case 'F': case 'f': return 15; } return -1; } #ifdef _WIN32 HMODULE evutil_load_windows_system_library_(const TCHAR *library_name) { TCHAR path[MAX_PATH]; unsigned n; n = GetSystemDirectory(path, MAX_PATH); if (n == 0 || n + _tcslen(library_name) + 2 >= MAX_PATH) return 0; _tcscat(path, TEXT("\\")); _tcscat(path, library_name); return LoadLibrary(path); } #endif /* Internal wrapper around 'socket' to provide Linux-style support for * syscall-saving methods where available. * * In addition to regular socket behavior, you can use a bitwise or to set the * flags EVUTIL_SOCK_NONBLOCK and EVUTIL_SOCK_CLOEXEC in the 'type' argument, * to make the socket nonblocking or close-on-exec with as few syscalls as * possible. */ evutil_socket_t evutil_socket_(int domain, int type, int protocol) { evutil_socket_t r; #if defined(SOCK_NONBLOCK) && defined(SOCK_CLOEXEC) r = socket(domain, type, protocol); if (r >= 0) return r; else if ((type & (SOCK_NONBLOCK|SOCK_CLOEXEC)) == 0) return -1; #endif #define SOCKET_TYPE_MASK (~(EVUTIL_SOCK_NONBLOCK|EVUTIL_SOCK_CLOEXEC)) r = socket(domain, type & SOCKET_TYPE_MASK, protocol); if (r < 0) return -1; if (type & EVUTIL_SOCK_NONBLOCK) { if (evutil_fast_socket_nonblocking(r) < 0) { evutil_closesocket(r); return -1; } } if (type & EVUTIL_SOCK_CLOEXEC) { if (evutil_fast_socket_closeonexec(r) < 0) { evutil_closesocket(r); return -1; } } return r; } /* Internal wrapper around 'accept' or 'accept4' to provide Linux-style * support for syscall-saving methods where available. * * In addition to regular accept behavior, you can set one or more of flags * EVUTIL_SOCK_NONBLOCK and EVUTIL_SOCK_CLOEXEC in the 'flags' argument, to * make the socket nonblocking or close-on-exec with as few syscalls as * possible. */ evutil_socket_t evutil_accept4_(evutil_socket_t sockfd, struct sockaddr *addr, ev_socklen_t *addrlen, int flags) { evutil_socket_t result; #if defined(EVENT__HAVE_ACCEPT4) && defined(SOCK_CLOEXEC) && defined(SOCK_NONBLOCK) result = accept4(sockfd, addr, addrlen, flags); if (result >= 0 || (errno != EINVAL && errno != ENOSYS)) { /* A nonnegative result means that we succeeded, so return. * Failing with EINVAL means that an option wasn't supported, * and failing with ENOSYS means that the syscall wasn't * there: in those cases we want to fall back. Otherwise, we * got a real error, and we should return. */ return result; } #endif result = accept(sockfd, addr, addrlen); if (result < 0) return result; if (flags & EVUTIL_SOCK_CLOEXEC) { if (evutil_fast_socket_closeonexec(result) < 0) { evutil_closesocket(result); return -1; } } if (flags & EVUTIL_SOCK_NONBLOCK) { if (evutil_fast_socket_nonblocking(result) < 0) { evutil_closesocket(result); return -1; } } return result; } /* Internal function: Set fd[0] and fd[1] to a pair of fds such that writes on * fd[1] get read from fd[0]. Make both fds nonblocking and close-on-exec. * Return 0 on success, -1 on failure. */ int evutil_make_internal_pipe_(evutil_socket_t fd[2]) { /* Making the second socket nonblocking is a bit subtle, given that we ignore any EAGAIN returns when writing to it, and you don't usally do that for a nonblocking socket. But if the kernel gives us EAGAIN, then there's no need to add any more data to the buffer, since the main thread is already either about to wake up and drain it, or woken up and in the process of draining it. */ #if defined(EVENT__HAVE_PIPE2) if (pipe2(fd, O_NONBLOCK|O_CLOEXEC) == 0) return 0; #endif #if defined(EVENT__HAVE_PIPE) if (pipe(fd) == 0) { if (evutil_fast_socket_nonblocking(fd[0]) < 0 || evutil_fast_socket_nonblocking(fd[1]) < 0 || evutil_fast_socket_closeonexec(fd[0]) < 0 || evutil_fast_socket_closeonexec(fd[1]) < 0) { close(fd[0]); close(fd[1]); fd[0] = fd[1] = -1; return -1; } return 0; } else { event_warn("%s: pipe", __func__); } #endif #ifdef _WIN32 #define LOCAL_SOCKETPAIR_AF AF_INET #else #define LOCAL_SOCKETPAIR_AF AF_UNIX #endif if (evutil_socketpair(LOCAL_SOCKETPAIR_AF, SOCK_STREAM, 0, fd) == 0) { if (evutil_fast_socket_nonblocking(fd[0]) < 0 || evutil_fast_socket_nonblocking(fd[1]) < 0 || evutil_fast_socket_closeonexec(fd[0]) < 0 || evutil_fast_socket_closeonexec(fd[1]) < 0) { evutil_closesocket(fd[0]); evutil_closesocket(fd[1]); fd[0] = fd[1] = -1; return -1; } return 0; } fd[0] = fd[1] = -1; return -1; } /* Wrapper around eventfd on systems that provide it. Unlike the system * eventfd, it always supports EVUTIL_EFD_CLOEXEC and EVUTIL_EFD_NONBLOCK as * flags. Returns -1 on error or if eventfd is not supported. */ evutil_socket_t evutil_eventfd_(unsigned initval, int flags) { #if defined(EVENT__HAVE_EVENTFD) && defined(EVENT__HAVE_SYS_EVENTFD_H) int r; #if defined(EFD_CLOEXEC) && defined(EFD_NONBLOCK) r = eventfd(initval, flags); if (r >= 0 || flags == 0) return r; #endif r = eventfd(initval, 0); if (r < 0) return r; if (flags & EVUTIL_EFD_CLOEXEC) { if (evutil_fast_socket_closeonexec(r) < 0) { evutil_closesocket(r); return -1; } } if (flags & EVUTIL_EFD_NONBLOCK) { if (evutil_fast_socket_nonblocking(r) < 0) { evutil_closesocket(r); return -1; } } return r; #else return -1; #endif } void evutil_free_globals_(void) { evutil_free_secure_rng_globals_(); evutil_free_sock_err_globals(); } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* This file has our secure PRNG code. On platforms that have arc4random(), * we just use that. Otherwise, we include arc4random.c as a bunch of static * functions, and wrap it lightly. We don't expose the arc4random*() APIs * because A) they aren't in our namespace, and B) it's not nice to name your * APIs after their implementations. We keep them in a separate file * so that other people can rip it out and use it for whatever. */ #include "event2/event-config.h" #include "evconfig-private.h" #include #include "util-internal.h" #include "evthread-internal.h" #ifdef EVENT__HAVE_ARC4RANDOM #include #include int evutil_secure_rng_set_urandom_device_file(char *fname) { (void) fname; return -1; } int evutil_secure_rng_init(void) { /* call arc4random() now to force it to self-initialize */ (void) arc4random(); return 0; } #ifndef EVENT__DISABLE_THREAD_SUPPORT int evutil_secure_rng_global_setup_locks_(const int enable_locks) { return 0; } #endif static void evutil_free_secure_rng_globals_locks(void) { } static void ev_arc4random_buf(void *buf, size_t n) { #if defined(EVENT__HAVE_ARC4RANDOM_BUF) && !defined(__APPLE__) arc4random_buf(buf, n); return; #else unsigned char *b = buf; #if defined(EVENT__HAVE_ARC4RANDOM_BUF) /* OSX 10.7 introducd arc4random_buf, so if you build your program * there, you'll get surprised when older versions of OSX fail to run. * To solve this, we can check whether the function pointer is set, * and fall back otherwise. (OSX does this using some linker * trickery.) */ { void (*tptr)(void *,size_t) = (void (*)(void*,size_t))arc4random_buf; if (tptr != NULL) { arc4random_buf(buf, n); return; } } #endif /* Make sure that we start out with b at a 4-byte alignment; plenty * of CPUs care about this for 32-bit access. */ if (n >= 4 && ((ev_uintptr_t)b) & 3) { ev_uint32_t u = arc4random(); int n_bytes = 4 - (((ev_uintptr_t)b) & 3); memcpy(b, &u, n_bytes); b += n_bytes; n -= n_bytes; } while (n >= 4) { *(ev_uint32_t*)b = arc4random(); b += 4; n -= 4; } if (n) { ev_uint32_t u = arc4random(); memcpy(b, &u, n); } #endif } #else /* !EVENT__HAVE_ARC4RANDOM { */ #ifdef EVENT__ssize_t #define ssize_t EVENT__ssize_t #endif #define ARC4RANDOM_EXPORT static #define ARC4_LOCK_() EVLOCK_LOCK(arc4rand_lock, 0) #define ARC4_UNLOCK_() EVLOCK_UNLOCK(arc4rand_lock, 0) #ifndef EVENT__DISABLE_THREAD_SUPPORT static void *arc4rand_lock; #endif #define ARC4RANDOM_UINT32 ev_uint32_t #define ARC4RANDOM_NOSTIR #define ARC4RANDOM_NORANDOM #define ARC4RANDOM_NOUNIFORM #include "./arc4random.c" #ifndef EVENT__DISABLE_THREAD_SUPPORT int evutil_secure_rng_global_setup_locks_(const int enable_locks) { EVTHREAD_SETUP_GLOBAL_LOCK(arc4rand_lock, 0); return 0; } #endif static void evutil_free_secure_rng_globals_locks(void) { #ifndef EVENT__DISABLE_THREAD_SUPPORT if (arc4rand_lock != NULL) { EVTHREAD_FREE_LOCK(arc4rand_lock, 0); arc4rand_lock = NULL; } #endif return; } int evutil_secure_rng_set_urandom_device_file(char *fname) { #ifdef TRY_SEED_URANDOM ARC4_LOCK_(); arc4random_urandom_filename = fname; ARC4_UNLOCK_(); #endif return 0; } int evutil_secure_rng_init(void) { int val; ARC4_LOCK_(); val = (!arc4_stir()) ? 0 : -1; ARC4_UNLOCK_(); return val; } static void ev_arc4random_buf(void *buf, size_t n) { arc4random_buf(buf, n); } #endif /* } !EVENT__HAVE_ARC4RANDOM */ void evutil_secure_rng_get_bytes(void *buf, size_t n) { ev_arc4random_buf(buf, n); } #if !defined(EVENT__HAVE_ARC4RANDOM) || defined(EVENT__HAVE_ARC4RANDOM_ADDRANDOM) void evutil_secure_rng_add_bytes(const char *buf, size_t n) { arc4random_addrandom((unsigned char*)buf, n>(size_t)INT_MAX ? INT_MAX : (int)n); } #endif void evutil_free_secure_rng_globals_(void) { evutil_free_secure_rng_globals_locks(); } /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #endif #include #ifdef EVENT__HAVE_STDLIB_H #include #endif #include #include #ifndef EVENT__HAVE_GETTIMEOFDAY #include #endif #if !defined(EVENT__HAVE_NANOSLEEP) && !defined(EVENT__HAVE_USLEEP) && \ !defined(_WIN32) #include #endif #include #include #include /** evutil_usleep_() */ #if defined(_WIN32) #elif defined(EVENT__HAVE_NANOSLEEP) #elif defined(EVENT__HAVE_USLEEP) #include #endif #include "event2/util.h" #include "util-internal.h" #include "log-internal.h" #include "mm-internal.h" #ifndef EVENT__HAVE_GETTIMEOFDAY /* No gettimeofday; this must be windows. */ typedef void (WINAPI *GetSystemTimePreciseAsFileTime_fn_t) (LPFILETIME); int evutil_gettimeofday(struct timeval *tv, struct timezone *tz) { #ifdef _MSC_VER #define U64_LITERAL(n) n##ui64 #else #define U64_LITERAL(n) n##llu #endif /* Conversion logic taken from Tor, which in turn took it * from Perl. GetSystemTimeAsFileTime returns its value as * an unaligned (!) 64-bit value containing the number of * 100-nanosecond intervals since 1 January 1601 UTC. */ #define EPOCH_BIAS U64_LITERAL(116444736000000000) #define UNITS_PER_SEC U64_LITERAL(10000000) #define USEC_PER_SEC U64_LITERAL(1000000) #define UNITS_PER_USEC U64_LITERAL(10) union { FILETIME ft_ft; ev_uint64_t ft_64; } ft; if (tv == NULL) return -1; static GetSystemTimePreciseAsFileTime_fn_t GetSystemTimePreciseAsFileTime_fn = NULL; static int check_precise = 1; if (EVUTIL_UNLIKELY(check_precise)) { HMODULE h = evutil_load_windows_system_library_(TEXT("kernel32.dll")); if (h != NULL) GetSystemTimePreciseAsFileTime_fn = (GetSystemTimePreciseAsFileTime_fn_t) GetProcAddress(h, "GetSystemTimePreciseAsFileTime"); check_precise = 0; } if (GetSystemTimePreciseAsFileTime_fn != NULL) GetSystemTimePreciseAsFileTime_fn(&ft.ft_ft); else GetSystemTimeAsFileTime(&ft.ft_ft); if (EVUTIL_UNLIKELY(ft.ft_64 < EPOCH_BIAS)) { /* Time before the unix epoch. */ return -1; } ft.ft_64 -= EPOCH_BIAS; tv->tv_sec = (long) (ft.ft_64 / UNITS_PER_SEC); tv->tv_usec = (long) ((ft.ft_64 / UNITS_PER_USEC) % USEC_PER_SEC); return 0; } #endif #define MAX_SECONDS_IN_MSEC_LONG \ (((LONG_MAX) - 999) / 1000) long evutil_tv_to_msec_(const struct timeval *tv) { if (tv->tv_usec > 1000000 || tv->tv_sec > MAX_SECONDS_IN_MSEC_LONG) return -1; return (tv->tv_sec * 1000) + ((tv->tv_usec + 999) / 1000); } /* Replacement for usleep on platforms that don't have one. Not guaranteed to be any more finegrained than 1 msec. */ void evutil_usleep_(const struct timeval *tv) { if (!tv) return; #if defined(_WIN32) { __int64 usec; LARGE_INTEGER li; HANDLE timer; usec = tv->tv_sec * 1000000LL + tv->tv_usec; if (!usec) return; li.QuadPart = -10LL * usec; timer = CreateWaitableTimer(NULL, TRUE, NULL); if (!timer) return; SetWaitableTimer(timer, &li, 0, NULL, NULL, 0); WaitForSingleObject(timer, INFINITE); CloseHandle(timer); } #elif defined(EVENT__HAVE_NANOSLEEP) { struct timespec ts; ts.tv_sec = tv->tv_sec; ts.tv_nsec = tv->tv_usec*1000; nanosleep(&ts, NULL); } #elif defined(EVENT__HAVE_USLEEP) /* Some systems don't like to usleep more than 999999 usec */ sleep(tv->tv_sec); usleep(tv->tv_usec); #else { struct timeval tv2 = *tv; select(0, NULL, NULL, NULL, &tv2); } #endif } int evutil_date_rfc1123(char *date, const size_t datelen, const struct tm *tm) { static const char *DAYS[] = { "Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat" }; static const char *MONTHS[] = { "Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec" }; time_t t = time(NULL); #if defined(EVENT__HAVE__GMTIME64_S) || !defined(_WIN32) struct tm sys; #endif /* If `tm` is null, set system's current time. */ if (tm == NULL) { #if !defined(_WIN32) gmtime_r(&t, &sys); tm = &sys; /** detect _gmtime64()/_gmtime64_s() */ #elif defined(EVENT__HAVE__GMTIME64_S) errno_t err; err = _gmtime64_s(&sys, &t); if (err) { event_errx(1, "Invalid argument to _gmtime64_s"); } else { tm = &sys; } #elif defined(EVENT__HAVE__GMTIME64) tm = _gmtime64(&t); #else tm = gmtime(&t); #endif } return evutil_snprintf( date, datelen, "%s, %02d %s %4d %02d:%02d:%02d GMT", DAYS[tm->tm_wday], tm->tm_mday, MONTHS[tm->tm_mon], 1900 + tm->tm_year, tm->tm_hour, tm->tm_min, tm->tm_sec); } /* This function assumes it's called repeatedly with a not-actually-so-monotonic time source whose outputs are in 'tv'. It implements a trivial ratcheting mechanism so that the values never go backwards. */ static void adjust_monotonic_time(struct evutil_monotonic_timer *base, struct timeval *tv) { evutil_timeradd(tv, &base->adjust_monotonic_clock, tv); if (evutil_timercmp(tv, &base->last_time, <)) { /* Guess it wasn't monotonic after all. */ struct timeval adjust; evutil_timersub(&base->last_time, tv, &adjust); evutil_timeradd(&adjust, &base->adjust_monotonic_clock, &base->adjust_monotonic_clock); *tv = base->last_time; } base->last_time = *tv; } /* Allocate a new struct evutil_monotonic_timer */ struct evutil_monotonic_timer * evutil_monotonic_timer_new(void) { struct evutil_monotonic_timer *p = NULL; p = mm_malloc(sizeof(*p)); if (!p) goto done; memset(p, 0, sizeof(*p)); done: return p; } /* Free a struct evutil_monotonic_timer */ void evutil_monotonic_timer_free(struct evutil_monotonic_timer *timer) { if (timer) { mm_free(timer); } } /* Set up a struct evutil_monotonic_timer for initial use */ int evutil_configure_monotonic_time(struct evutil_monotonic_timer *timer, int flags) { return evutil_configure_monotonic_time_(timer, flags); } /* Query the current monotonic time */ int evutil_gettime_monotonic(struct evutil_monotonic_timer *timer, struct timeval *tp) { return evutil_gettime_monotonic_(timer, tp); } #if defined(HAVE_POSIX_MONOTONIC) /* ===== The POSIX clock_gettime() interface provides a few ways to get at a monotonic clock. CLOCK_MONOTONIC is most widely supported. Linux also provides a CLOCK_MONOTONIC_COARSE with accuracy of about 1-4 msec. On all platforms I'm aware of, CLOCK_MONOTONIC really is monotonic. Platforms don't agree about whether it should jump on a sleep/resume. */ int evutil_configure_monotonic_time_(struct evutil_monotonic_timer *base, int flags) { /* CLOCK_MONOTONIC exists on FreeBSD, Linux, and Solaris. You need to * check for it at runtime, because some older kernel versions won't * have it working. */ #ifdef CLOCK_MONOTONIC_COARSE const int precise = flags & EV_MONOT_PRECISE; #endif const int fallback = flags & EV_MONOT_FALLBACK; struct timespec ts; #ifdef CLOCK_MONOTONIC_COARSE if (CLOCK_MONOTONIC_COARSE < 0) { /* Technically speaking, nothing keeps CLOCK_* from being * negative (as far as I know). This check and the one below * make sure that it's safe for us to use -1 as an "unset" * value. */ event_errx(1,"I didn't expect CLOCK_MONOTONIC_COARSE to be < 0"); } if (! precise && ! fallback) { if (clock_gettime(CLOCK_MONOTONIC_COARSE, &ts) == 0) { base->monotonic_clock = CLOCK_MONOTONIC_COARSE; return 0; } } #endif if (!fallback && clock_gettime(CLOCK_MONOTONIC, &ts) == 0) { base->monotonic_clock = CLOCK_MONOTONIC; return 0; } if (CLOCK_MONOTONIC < 0) { event_errx(1,"I didn't expect CLOCK_MONOTONIC to be < 0"); } base->monotonic_clock = -1; return 0; } int evutil_gettime_monotonic_(struct evutil_monotonic_timer *base, struct timeval *tp) { struct timespec ts; if (base->monotonic_clock < 0) { if (evutil_gettimeofday(tp, NULL) < 0) return -1; adjust_monotonic_time(base, tp); return 0; } if (clock_gettime(base->monotonic_clock, &ts) == -1) return -1; tp->tv_sec = ts.tv_sec; tp->tv_usec = ts.tv_nsec / 1000; return 0; } #endif #if defined(HAVE_MACH_MONOTONIC) /* ====== Apple is a little late to the POSIX party. And why not? Instead of clock_gettime(), they provide mach_absolute_time(). Its units are not fixed; we need to use mach_timebase_info() to get the right functions to convert its units into nanoseconds. To all appearances, mach_absolute_time() seems to be honest-to-goodness monotonic. Whether it stops during sleep or not is unspecified in principle, and dependent on CPU architecture in practice. */ int evutil_configure_monotonic_time_(struct evutil_monotonic_timer *base, int flags) { const int fallback = flags & EV_MONOT_FALLBACK; struct mach_timebase_info mi; memset(base, 0, sizeof(*base)); /* OSX has mach_absolute_time() */ if (!fallback && mach_timebase_info(&mi) == 0 && mach_absolute_time() != 0) { /* mach_timebase_info tells us how to convert * mach_absolute_time() into nanoseconds, but we * want to use microseconds instead. */ mi.denom *= 1000; memcpy(&base->mach_timebase_units, &mi, sizeof(mi)); } else { base->mach_timebase_units.numer = 0; } return 0; } int evutil_gettime_monotonic_(struct evutil_monotonic_timer *base, struct timeval *tp) { ev_uint64_t abstime, usec; if (base->mach_timebase_units.numer == 0) { if (evutil_gettimeofday(tp, NULL) < 0) return -1; adjust_monotonic_time(base, tp); return 0; } abstime = mach_absolute_time(); usec = (abstime * base->mach_timebase_units.numer) / (base->mach_timebase_units.denom); tp->tv_sec = usec / 1000000; tp->tv_usec = usec % 1000000; return 0; } #endif #if defined(HAVE_WIN32_MONOTONIC) /* ===== Turn we now to Windows. Want monontonic time on Windows? Windows has QueryPerformanceCounter(), which gives time most high- resolution time. It's a pity it's not so monotonic in practice; it's also got some fun bugs, especially: with older Windowses, under virtualizations, with funny hardware, on multiprocessor systems, and so on. PEP418 [1] has a nice roundup of the issues here. There's GetTickCount64() on Vista and later, which gives a number of 1-msec ticks since startup. The accuracy here might be as bad as 10-20 msec, I hear. There's an undocumented function (NtSetTimerResolution) that allegedly increases the accuracy. Good luck! There's also GetTickCount(), which is only 32 bits, but seems to be supported on pre-Vista versions of Windows. Apparently, you can coax another 14 bits out of it, giving you 2231 years before rollover. The less said about timeGetTime() the better. "We don't care. We don't have to. We're the Phone Company." -- Lily Tomlin, SNL Our strategy, if precise timers are turned off, is to just use the best GetTickCount equivalent available. If we've been asked for precise timing, then we mostly[2] assume that GetTickCount is monotonic, and correct GetPerformanceCounter to approximate it. [1] http://www.python.org/dev/peps/pep-0418 [2] Of course, we feed the Windows stuff into adjust_monotonic_time() anyway, just in case it isn't. */ /* Parts of our logic in the win32 timer code here are closely based on BitTorrent's libUTP library. That code is subject to the following license: Copyright (c) 2010 BitTorrent, Inc. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ static ev_uint64_t evutil_GetTickCount_(struct evutil_monotonic_timer *base) { if (base->GetTickCount64_fn) { /* Let's just use GetTickCount64 if we can. */ return base->GetTickCount64_fn(); } else if (base->GetTickCount_fn) { /* Greg Hazel assures me that this works, that BitTorrent has * done it for years, and this it won't turn around and * bite us. He says they found it on some game programmers' * forum some time around 2007. */ ev_uint64_t v = base->GetTickCount_fn(); return (DWORD)v | ((v >> 18) & 0xFFFFFFFF00000000); } else { /* Here's the fallback implementation. We have to use * GetTickCount() with its given signature, so we only get * 32 bits worth of milliseconds, which will roll ove every * 49 days or so. */ DWORD ticks = GetTickCount(); if (ticks < base->last_tick_count) { base->adjust_tick_count += ((ev_uint64_t)1) << 32; } base->last_tick_count = ticks; return ticks + base->adjust_tick_count; } } int evutil_configure_monotonic_time_(struct evutil_monotonic_timer *base, int flags) { const int precise = flags & EV_MONOT_PRECISE; const int fallback = flags & EV_MONOT_FALLBACK; HANDLE h; memset(base, 0, sizeof(*base)); h = evutil_load_windows_system_library_(TEXT("kernel32.dll")); if (h != NULL && !fallback) { base->GetTickCount64_fn = (ev_GetTickCount_func)GetProcAddress(h, "GetTickCount64"); base->GetTickCount_fn = (ev_GetTickCount_func)GetProcAddress(h, "GetTickCount"); } base->first_tick = base->last_tick_count = evutil_GetTickCount_(base); if (precise && !fallback) { LARGE_INTEGER freq; if (QueryPerformanceFrequency(&freq)) { LARGE_INTEGER counter; QueryPerformanceCounter(&counter); base->first_counter = counter.QuadPart; base->usec_per_count = 1.0e6 / freq.QuadPart; base->use_performance_counter = 1; } } return 0; } static inline ev_int64_t abs64(ev_int64_t i) { return i < 0 ? -i : i; } int evutil_gettime_monotonic_(struct evutil_monotonic_timer *base, struct timeval *tp) { ev_uint64_t ticks = evutil_GetTickCount_(base); if (base->use_performance_counter) { /* Here's a trick we took from BitTorrent's libutp, at Greg * Hazel's recommendation. We use QueryPerformanceCounter for * our high-resolution timer, but use GetTickCount*() to keep * it sane, and adjust_monotonic_time() to keep it monotonic. */ LARGE_INTEGER counter; ev_int64_t counter_elapsed, counter_usec_elapsed, ticks_elapsed; QueryPerformanceCounter(&counter); counter_elapsed = (ev_int64_t) (counter.QuadPart - base->first_counter); ticks_elapsed = ticks - base->first_tick; /* TODO: This may upset VC6. If you need this to work with * VC6, please supply an appropriate patch. */ counter_usec_elapsed = (ev_int64_t) (counter_elapsed * base->usec_per_count); if (abs64(ticks_elapsed*1000 - counter_usec_elapsed) > 1000000) { /* It appears that the QueryPerformanceCounter() * result is more than 1 second away from * GetTickCount() result. Let's adjust it to be as * accurate as we can; adjust_monotnonic_time() below * will keep it monotonic. */ counter_usec_elapsed = ticks_elapsed * 1000; base->first_counter = (ev_uint64_t) (counter.QuadPart - counter_usec_elapsed / base->usec_per_count); } tp->tv_sec = (time_t) (counter_usec_elapsed / 1000000); tp->tv_usec = counter_usec_elapsed % 1000000; } else { /* We're just using GetTickCount(). */ tp->tv_sec = (time_t) (ticks / 1000); tp->tv_usec = (ticks % 1000) * 1000; } adjust_monotonic_time(base, tp); return 0; } #endif #if defined(HAVE_FALLBACK_MONOTONIC) /* ===== And if none of the other options work, let's just use gettimeofday(), and ratchet it forward so that it acts like a monotonic timer, whether it wants to or not. */ int evutil_configure_monotonic_time_(struct evutil_monotonic_timer *base, int precise) { memset(base, 0, sizeof(*base)); return 0; } int evutil_gettime_monotonic_(struct evutil_monotonic_timer *base, struct timeval *tp) { if (evutil_gettimeofday(tp, NULL) < 0) return -1; adjust_monotonic_time(base, tp); return 0; } #endif /* * Copyright (c) 2002-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_SYS_PARAM_H #include #endif #ifdef EVENT__HAVE_SYS_TYPES_H #include #endif #ifdef HAVE_SYS_IOCCOM_H #include #endif #ifdef EVENT__HAVE_SYS_RESOURCE_H #include #endif #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #ifdef EVENT__HAVE_SYS_WAIT_H #include #endif #ifndef _WIN32 #include #include #else /* _WIN32 */ #include #include #endif /* _WIN32 */ #ifdef EVENT__HAVE_SYS_UN_H #include #endif #ifdef EVENT__HAVE_AFUNIX_H #include #endif #include #ifdef EVENT__HAVE_NETINET_IN_H #include #endif #ifdef EVENT__HAVE_ARPA_INET_H #include #endif #ifdef EVENT__HAVE_NETDB_H #include #endif #ifdef _WIN32 #include #endif #include #include #include #include #ifndef _WIN32 #include #endif /* !_WIN32 */ #include #ifdef EVENT__HAVE_UNISTD_H #include #endif #ifdef EVENT__HAVE_FCNTL_H #include #endif #undef timeout_pending #undef timeout_initialized #include "strlcpy-internal.h" #include "event2/http.h" #include "event2/event.h" #include "event2/buffer.h" #include "event2/bufferevent.h" #include "event2/http_struct.h" #include "event2/http_compat.h" #include "event2/util.h" #include "event2/listener.h" #include "log-internal.h" #include "util-internal.h" #include "http-internal.h" #include "mm-internal.h" #include "bufferevent-internal.h" #ifndef EVENT__HAVE_GETNAMEINFO #define NI_MAXSERV 32 #define NI_MAXHOST 1025 #ifndef NI_NUMERICHOST #define NI_NUMERICHOST 1 #endif #ifndef NI_NUMERICSERV #define NI_NUMERICSERV 2 #endif static int fake_getnameinfo(const struct sockaddr *sa, size_t salen, char *host, size_t hostlen, char *serv, size_t servlen, int flags) { struct sockaddr_in *sin = (struct sockaddr_in *)sa; if (serv != NULL) { char tmpserv[16]; evutil_snprintf(tmpserv, sizeof(tmpserv), "%d", ntohs(sin->sin_port)); if (strlcpy(serv, tmpserv, servlen) >= servlen) return (-1); } if (host != NULL) { if (flags & NI_NUMERICHOST) { if (strlcpy(host, inet_ntoa(sin->sin_addr), hostlen) >= hostlen) return (-1); else return (0); } else { struct hostent *hp; hp = gethostbyaddr((char *)&sin->sin_addr, sizeof(struct in_addr), AF_INET); if (hp == NULL) return (-2); if (strlcpy(host, hp->h_name, hostlen) >= hostlen) return (-1); else return (0); } } return (0); } #endif #define REQ_VERSION_BEFORE(req, major_v, minor_v) \ ((req)->major < (major_v) || \ ((req)->major == (major_v) && (req)->minor < (minor_v))) #define REQ_VERSION_ATLEAST(req, major_v, minor_v) \ ((req)->major > (major_v) || \ ((req)->major == (major_v) && (req)->minor >= (minor_v))) #ifndef MIN #define MIN(a,b) (((a)<(b))?(a):(b)) #endif extern int debug; static evutil_socket_t create_bind_socket_nonblock(struct evutil_addrinfo *, int reuse); static evutil_socket_t bind_socket(const char *, ev_uint16_t, int reuse); static void name_from_addr(struct sockaddr *, ev_socklen_t, char **, char **); static struct evhttp_uri *evhttp_uri_parse_authority(char *source_uri); static int evhttp_associate_new_request_with_connection( struct evhttp_connection *evcon); static void evhttp_connection_start_detectclose( struct evhttp_connection *evcon); static void evhttp_connection_stop_detectclose( struct evhttp_connection *evcon); static void evhttp_request_dispatch(struct evhttp_connection* evcon); static void evhttp_read_firstline(struct evhttp_connection *evcon, struct evhttp_request *req); static void evhttp_read_header(struct evhttp_connection *evcon, struct evhttp_request *req); static int evhttp_add_header_internal(struct evkeyvalq *headers, const char *key, const char *value); static const char *evhttp_response_phrase_internal(int code); static void evhttp_get_request(struct evhttp *, evutil_socket_t, struct sockaddr *, ev_socklen_t); static void evhttp_write_buffer(struct evhttp_connection *, void (*)(struct evhttp_connection *, void *), void *); static void evhttp_make_header(struct evhttp_connection *, struct evhttp_request *); /* callbacks for bufferevent */ static void evhttp_read_cb(struct bufferevent *, void *); static void evhttp_write_cb(struct bufferevent *, void *); static void evhttp_error_cb(struct bufferevent *bufev, short what, void *arg); static int evhttp_find_vhost(struct evhttp *http, struct evhttp **outhttp, const char *hostname); #ifndef EVENT__HAVE_STRSEP /* strsep replacement for platforms that lack it. Only works if * del is one character long. */ static char * strsep(char **s, const char *del) { char *d, *tok; EVUTIL_ASSERT(strlen(del) == 1); if (!s || !*s) return NULL; tok = *s; d = strstr(tok, del); if (d) { *d = '\0'; *s = d + 1; } else *s = NULL; return tok; } #endif static size_t html_replace(const char ch, const char **escaped) { switch (ch) { case '<': *escaped = "<"; return 4; case '>': *escaped = ">"; return 4; case '"': *escaped = """; return 6; case '\'': *escaped = "'"; return 6; case '&': *escaped = "&"; return 5; default: break; } return 1; } /* * Replaces <, >, ", ' and & with <, >, ", * ' and & correspondingly. * * The returned string needs to be freed by the caller. */ char * evhttp_htmlescape(const char *html) { size_t i; size_t new_size = 0, old_size = 0; char *escaped_html, *p; if (html == NULL) return (NULL); old_size = strlen(html); for (i = 0; i < old_size; ++i) { const char *replaced = NULL; const size_t replace_size = html_replace(html[i], &replaced); if (replace_size > EV_SIZE_MAX - new_size) { event_warn("%s: html_replace overflow", __func__); return (NULL); } new_size += replace_size; } if (new_size == EV_SIZE_MAX) return (NULL); p = escaped_html = mm_malloc(new_size + 1); if (escaped_html == NULL) { event_warn("%s: malloc(%lu)", __func__, (unsigned long)(new_size + 1)); return (NULL); } for (i = 0; i < old_size; ++i) { const char *replaced = &html[i]; const size_t len = html_replace(html[i], &replaced); memcpy(p, replaced, len); p += len; } *p = '\0'; return (escaped_html); } /** Given an evhttp_cmd_type, returns a constant string containing the * equivalent HTTP command, or NULL if the evhttp_command_type is * unrecognized. */ static const char * evhttp_method(enum evhttp_cmd_type type) { const char *method; switch (type) { case EVHTTP_REQ_GET: method = "GET"; break; case EVHTTP_REQ_POST: method = "POST"; break; case EVHTTP_REQ_HEAD: method = "HEAD"; break; case EVHTTP_REQ_PUT: method = "PUT"; break; case EVHTTP_REQ_DELETE: method = "DELETE"; break; case EVHTTP_REQ_OPTIONS: method = "OPTIONS"; break; case EVHTTP_REQ_TRACE: method = "TRACE"; break; case EVHTTP_REQ_CONNECT: method = "CONNECT"; break; case EVHTTP_REQ_PATCH: method = "PATCH"; break; default: method = NULL; break; } return (method); } /** * Determines if a response should have a body. * Follows the rules in RFC 2616 section 4.3. * @return 1 if the response MUST have a body; 0 if the response MUST NOT have * a body. */ static int evhttp_response_needs_body(struct evhttp_request *req) { return (req->response_code != HTTP_NOCONTENT && req->response_code != HTTP_NOTMODIFIED && (req->response_code < 100 || req->response_code >= 200) && req->type != EVHTTP_REQ_CONNECT && req->type != EVHTTP_REQ_HEAD); } /** Helper: called after we've added some data to an evcon's bufferevent's * output buffer. Sets the evconn's writing-is-done callback, and puts * the bufferevent into writing mode. */ static void evhttp_write_buffer(struct evhttp_connection *evcon, void (*cb)(struct evhttp_connection *, void *), void *arg) { event_debug(("%s: preparing to write buffer\n", __func__)); /* Set call back */ evcon->cb = cb; evcon->cb_arg = arg; /* Disable the read callback: we don't actually care about data; * we only care about close detection. (We don't disable reading -- * EV_READ, since we *do* want to learn about any close events.) */ bufferevent_setcb(evcon->bufev, NULL, /*read*/ evhttp_write_cb, evhttp_error_cb, evcon); bufferevent_enable(evcon->bufev, EV_READ|EV_WRITE); } static void evhttp_send_continue_done(struct evhttp_connection *evcon, void *arg) { bufferevent_disable(evcon->bufev, EV_WRITE); } static void evhttp_send_continue(struct evhttp_connection *evcon, struct evhttp_request *req) { bufferevent_enable(evcon->bufev, EV_WRITE); evbuffer_add_printf(bufferevent_get_output(evcon->bufev), "HTTP/%d.%d 100 Continue\r\n\r\n", req->major, req->minor); evcon->cb = evhttp_send_continue_done; evcon->cb_arg = NULL; bufferevent_setcb(evcon->bufev, evhttp_read_cb, evhttp_write_cb, evhttp_error_cb, evcon); } /** Helper: returns true iff evconn is in any connected state. */ static int evhttp_connected(struct evhttp_connection *evcon) { switch (evcon->state) { case EVCON_DISCONNECTED: case EVCON_CONNECTING: return (0); case EVCON_IDLE: case EVCON_READING_FIRSTLINE: case EVCON_READING_HEADERS: case EVCON_READING_BODY: case EVCON_READING_TRAILER: case EVCON_WRITING: default: return (1); } } /* Create the headers needed for an outgoing HTTP request, adds them to * the request's header list, and writes the request line to the * connection's output buffer. */ static void evhttp_make_header_request(struct evhttp_connection *evcon, struct evhttp_request *req) { const char *method; evhttp_remove_header(req->output_headers, "Proxy-Connection"); /* Generate request line */ if (!(method = evhttp_method(req->type))) { method = "NULL"; } evbuffer_add_printf(bufferevent_get_output(evcon->bufev), "%s %s HTTP/%d.%d\r\n", method, req->uri, req->major, req->minor); /* Add the content length on a post or put request if missing */ if ((req->type == EVHTTP_REQ_POST || req->type == EVHTTP_REQ_PUT) && evhttp_find_header(req->output_headers, "Content-Length") == NULL){ char size[22]; evutil_snprintf(size, sizeof(size), EV_SIZE_FMT, EV_SIZE_ARG(evbuffer_get_length(req->output_buffer))); evhttp_add_header(req->output_headers, "Content-Length", size); } } /** Return true if the list of headers in 'headers', intepreted with respect * to flags, means that we should send a "connection: close" when the request * is done. */ static int evhttp_is_connection_close(int flags, struct evkeyvalq* headers) { if (flags & EVHTTP_PROXY_REQUEST) { /* proxy connection */ const char *connection = evhttp_find_header(headers, "Proxy-Connection"); return (connection == NULL || evutil_ascii_strcasecmp(connection, "keep-alive") != 0); } else { const char *connection = evhttp_find_header(headers, "Connection"); return (connection != NULL && evutil_ascii_strcasecmp(connection, "close") == 0); } } static int evhttp_is_request_connection_close(struct evhttp_request *req) { if (req->type == EVHTTP_REQ_CONNECT) return 0; return evhttp_is_connection_close(req->flags, req->input_headers) || evhttp_is_connection_close(req->flags, req->output_headers); } /* Return true iff 'headers' contains 'Connection: keep-alive' */ static int evhttp_is_connection_keepalive(struct evkeyvalq* headers) { const char *connection = evhttp_find_header(headers, "Connection"); return (connection != NULL && evutil_ascii_strncasecmp(connection, "keep-alive", 10) == 0); } /* Add a correct "Date" header to headers, unless it already has one. */ static void evhttp_maybe_add_date_header(struct evkeyvalq *headers) { if (evhttp_find_header(headers, "Date") == NULL) { char date[50]; if (sizeof(date) - evutil_date_rfc1123(date, sizeof(date), NULL) > 0) { evhttp_add_header(headers, "Date", date); } } } /* Add a "Content-Length" header with value 'content_length' to headers, * unless it already has a content-length or transfer-encoding header. */ static void evhttp_maybe_add_content_length_header(struct evkeyvalq *headers, size_t content_length) { if (evhttp_find_header(headers, "Transfer-Encoding") == NULL && evhttp_find_header(headers, "Content-Length") == NULL) { char len[22]; evutil_snprintf(len, sizeof(len), EV_SIZE_FMT, EV_SIZE_ARG(content_length)); evhttp_add_header(headers, "Content-Length", len); } } /* * Create the headers needed for an HTTP reply in req->output_headers, * and write the first HTTP response for req line to evcon. */ static void evhttp_make_header_response(struct evhttp_connection *evcon, struct evhttp_request *req) { int is_keepalive = evhttp_is_connection_keepalive(req->input_headers); evbuffer_add_printf(bufferevent_get_output(evcon->bufev), "HTTP/%d.%d %d %s\r\n", req->major, req->minor, req->response_code, req->response_code_line); if (req->major == 1) { if (req->minor >= 1) evhttp_maybe_add_date_header(req->output_headers); /* * if the protocol is 1.0; and the connection was keep-alive * we need to add a keep-alive header, too. */ if (req->minor == 0 && is_keepalive) evhttp_add_header(req->output_headers, "Connection", "keep-alive"); if ((req->minor >= 1 || is_keepalive) && evhttp_response_needs_body(req)) { /* * we need to add the content length if the * user did not give it, this is required for * persistent connections to work. */ evhttp_maybe_add_content_length_header( req->output_headers, evbuffer_get_length(req->output_buffer)); } } /* Potentially add headers for unidentified content. */ if (evhttp_response_needs_body(req)) { if (evhttp_find_header(req->output_headers, "Content-Type") == NULL && evcon->http_server->default_content_type) { evhttp_add_header(req->output_headers, "Content-Type", evcon->http_server->default_content_type); } } /* if the request asked for a close, we send a close, too */ if (evhttp_is_connection_close(req->flags, req->input_headers)) { evhttp_remove_header(req->output_headers, "Connection"); if (!(req->flags & EVHTTP_PROXY_REQUEST)) evhttp_add_header(req->output_headers, "Connection", "close"); evhttp_remove_header(req->output_headers, "Proxy-Connection"); } } enum expect { NO, CONTINUE, OTHER }; static enum expect evhttp_have_expect(struct evhttp_request *req, int input) { const char *expect; struct evkeyvalq *h = input ? req->input_headers : req->output_headers; if (!(req->kind == EVHTTP_REQUEST) || !REQ_VERSION_ATLEAST(req, 1, 1)) return NO; expect = evhttp_find_header(h, "Expect"); if (!expect) return NO; return !evutil_ascii_strcasecmp(expect, "100-continue") ? CONTINUE : OTHER; } /** Generate all headers appropriate for sending the http request in req (or * the response, if we're sending a response), and write them to evcon's * bufferevent. Also writes all data from req->output_buffer */ static void evhttp_make_header(struct evhttp_connection *evcon, struct evhttp_request *req) { struct evkeyval *header; struct evbuffer *output = bufferevent_get_output(evcon->bufev); /* * Depending if this is a HTTP request or response, we might need to * add some new headers or remove existing headers. */ if (req->kind == EVHTTP_REQUEST) { evhttp_make_header_request(evcon, req); } else { evhttp_make_header_response(evcon, req); } TAILQ_FOREACH(header, req->output_headers, next) { evbuffer_add_printf(output, "%s: %s\r\n", header->key, header->value); } evbuffer_add(output, "\r\n", 2); if (evhttp_have_expect(req, 0) != CONTINUE && evbuffer_get_length(req->output_buffer)) { /* * For a request, we add the POST data, for a reply, this * is the regular data. */ evbuffer_add_buffer(output, req->output_buffer); } } void evhttp_connection_set_max_headers_size(struct evhttp_connection *evcon, ev_ssize_t new_max_headers_size) { if (new_max_headers_size<0) evcon->max_headers_size = EV_SIZE_MAX; else evcon->max_headers_size = new_max_headers_size; } void evhttp_connection_set_max_body_size(struct evhttp_connection* evcon, ev_ssize_t new_max_body_size) { if (new_max_body_size<0) evcon->max_body_size = EV_UINT64_MAX; else evcon->max_body_size = new_max_body_size; } static int evhttp_connection_incoming_fail(struct evhttp_request *req, enum evhttp_request_error error) { switch (error) { case EVREQ_HTTP_DATA_TOO_LONG: req->response_code = HTTP_ENTITYTOOLARGE; break; default: req->response_code = HTTP_BADREQUEST; } switch (error) { case EVREQ_HTTP_TIMEOUT: case EVREQ_HTTP_EOF: /* * these are cases in which we probably should just * close the connection and not send a reply. this * case may happen when a browser keeps a persistent * connection open and we timeout on the read. when * the request is still being used for sending, we * need to disassociated it from the connection here. */ if (!req->userdone) { /* remove it so that it will not be freed */ TAILQ_REMOVE(&req->evcon->requests, req, next); /* indicate that this request no longer has a * connection object */ req->evcon = NULL; } return (-1); case EVREQ_HTTP_INVALID_HEADER: case EVREQ_HTTP_BUFFER_ERROR: case EVREQ_HTTP_REQUEST_CANCEL: case EVREQ_HTTP_DATA_TOO_LONG: default: /* xxx: probably should just error on default */ /* the callback looks at the uri to determine errors */ if (req->uri) { mm_free(req->uri); req->uri = NULL; } if (req->uri_elems) { evhttp_uri_free(req->uri_elems); req->uri_elems = NULL; } /* * the callback needs to send a reply, once the reply has * been send, the connection should get freed. */ (*req->cb)(req, req->cb_arg); } return (0); } /* Free connection ownership of which can be acquired by user using * evhttp_request_own(). */ static inline void evhttp_request_free_auto(struct evhttp_request *req) { if (!(req->flags & EVHTTP_USER_OWNED)) evhttp_request_free(req); } static void evhttp_request_free_(struct evhttp_connection *evcon, struct evhttp_request *req) { TAILQ_REMOVE(&evcon->requests, req, next); evhttp_request_free_auto(req); } /* Called when evcon has experienced a (non-recoverable? -NM) error, as * given in error. If it's an outgoing connection, reset the connection, * retry any pending requests, and inform the user. If it's incoming, * delegates to evhttp_connection_incoming_fail(). */ void evhttp_connection_fail_(struct evhttp_connection *evcon, enum evhttp_request_error error) { const int errsave = EVUTIL_SOCKET_ERROR(); struct evhttp_request* req = TAILQ_FIRST(&evcon->requests); void (*cb)(struct evhttp_request *, void *); void *cb_arg; void (*error_cb)(enum evhttp_request_error, void *); void *error_cb_arg; EVUTIL_ASSERT(req != NULL); bufferevent_disable(evcon->bufev, EV_READ|EV_WRITE); if (evcon->flags & EVHTTP_CON_INCOMING) { /* * for incoming requests, there are two different * failure cases. it's either a network level error * or an http layer error. for problems on the network * layer like timeouts we just drop the connections. * For HTTP problems, we might have to send back a * reply before the connection can be freed. */ if (evhttp_connection_incoming_fail(req, error) == -1) evhttp_connection_free(evcon); return; } error_cb = req->error_cb; error_cb_arg = req->cb_arg; /* when the request was canceled, the callback is not executed */ if (error != EVREQ_HTTP_REQUEST_CANCEL) { /* save the callback for later; the cb might free our object */ cb = req->cb; cb_arg = req->cb_arg; } else { cb = NULL; cb_arg = NULL; } /* do not fail all requests; the next request is going to get * send over a new connection. when a user cancels a request, * all other pending requests should be processed as normal */ evhttp_request_free_(evcon, req); /* reset the connection */ evhttp_connection_reset_(evcon); /* We are trying the next request that was queued on us */ if (TAILQ_FIRST(&evcon->requests) != NULL) evhttp_connection_connect_(evcon); else if ((evcon->flags & EVHTTP_CON_OUTGOING) && (evcon->flags & EVHTTP_CON_AUTOFREE)) { evhttp_connection_free(evcon); } /* The call to evhttp_connection_reset_ overwrote errno. * Let's restore the original errno, so that the user's * callback can have a better idea of what the error was. */ EVUTIL_SET_SOCKET_ERROR(errsave); /* inform the user */ if (error_cb != NULL) error_cb(error, error_cb_arg); if (cb != NULL) (*cb)(NULL, cb_arg); } /* Bufferevent callback: invoked when any data has been written from an * http connection's bufferevent */ static void evhttp_write_cb(struct bufferevent *bufev, void *arg) { struct evhttp_connection *evcon = arg; /* Activate our call back */ if (evcon->cb != NULL) (*evcon->cb)(evcon, evcon->cb_arg); } /** * Advance the connection state. * - If this is an outgoing connection, we've just processed the response; * idle or close the connection. * - If this is an incoming connection, we've just processed the request; * respond. */ static void evhttp_connection_done(struct evhttp_connection *evcon) { struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); int con_outgoing = evcon->flags & EVHTTP_CON_OUTGOING; int free_evcon = 0; if (con_outgoing) { /* idle or close the connection */ int need_close = evhttp_is_request_connection_close(req); TAILQ_REMOVE(&evcon->requests, req, next); req->evcon = NULL; evcon->state = EVCON_IDLE; /* check if we got asked to close the connection */ if (need_close) evhttp_connection_reset_(evcon); if (TAILQ_FIRST(&evcon->requests) != NULL) { /* * We have more requests; reset the connection * and deal with the next request. */ if (!evhttp_connected(evcon)) evhttp_connection_connect_(evcon); else evhttp_request_dispatch(evcon); } else if (!need_close) { /* * The connection is going to be persistent, but we * need to detect if the other side closes it. */ evhttp_connection_start_detectclose(evcon); } else if ((evcon->flags & EVHTTP_CON_AUTOFREE)) { /* * If we have no more requests that need completion * and we're not waiting for the connection to close */ free_evcon = 1; } } else { /* * incoming connection - we need to leave the request on the * connection so that we can reply to it. */ evcon->state = EVCON_WRITING; } /* notify the user of the request */ (*req->cb)(req, req->cb_arg); /* if this was an outgoing request, we own and it's done. so free it. */ if (con_outgoing) { evhttp_request_free_auto(req); } /* If this was the last request of an outgoing connection and we're * not waiting to receive a connection close event and we want to * automatically free the connection. We check to ensure our request * list is empty one last time just in case our callback added a * new request. */ if (free_evcon && TAILQ_FIRST(&evcon->requests) == NULL) { evhttp_connection_free(evcon); } } /* * Handles reading from a chunked request. * return ALL_DATA_READ: * all data has been read * return MORE_DATA_EXPECTED: * more data is expected * return DATA_CORRUPTED: * data is corrupted * return REQUEST_CANCELED: * request was canceled by the user calling evhttp_cancel_request * return DATA_TOO_LONG: * ran over the maximum limit */ static enum message_read_status evhttp_handle_chunked_read(struct evhttp_request *req, struct evbuffer *buf) { if (req == NULL || buf == NULL) { return DATA_CORRUPTED; } while (1) { size_t buflen; if ((buflen = evbuffer_get_length(buf)) == 0) { break; } /* evbuffer_get_length returns size_t, but len variable is ssize_t, * check for overflow conditions */ if (buflen > EV_SSIZE_MAX) { return DATA_CORRUPTED; } if (req->ntoread < 0) { /* Read chunk size */ ev_int64_t ntoread; char *p = evbuffer_readln(buf, NULL, EVBUFFER_EOL_CRLF); char *endp; int error; if (p == NULL) break; /* the last chunk is on a new line? */ if (strlen(p) == 0) { mm_free(p); continue; } ntoread = evutil_strtoll(p, &endp, 16); error = (*p == '\0' || (*endp != '\0' && *endp != ' ') || ntoread < 0); mm_free(p); if (error) { /* could not get chunk size */ return (DATA_CORRUPTED); } /* ntoread is signed int64, body_size is unsigned size_t, check for under/overflow conditions */ if ((ev_uint64_t)ntoread > EV_SIZE_MAX - req->body_size) { return DATA_CORRUPTED; } if (req->body_size + (size_t)ntoread > req->evcon->max_body_size) { /* failed body length test */ event_debug(("Request body is too long")); return (DATA_TOO_LONG); } req->body_size += (size_t)ntoread; req->ntoread = ntoread; if (req->ntoread == 0) { /* Last chunk */ return (ALL_DATA_READ); } continue; } /* req->ntoread is signed int64, len is ssize_t, based on arch, * ssize_t could only be 32b, check for these conditions */ if (req->ntoread > EV_SSIZE_MAX) { return DATA_CORRUPTED; } /* don't have enough to complete a chunk; wait for more */ if (req->ntoread > 0 && buflen < (ev_uint64_t)req->ntoread) return (MORE_DATA_EXPECTED); /* Completed chunk */ evbuffer_remove_buffer(buf, req->input_buffer, (size_t)req->ntoread); req->ntoread = -1; if (req->chunk_cb != NULL) { req->flags |= EVHTTP_REQ_DEFER_FREE; (*req->chunk_cb)(req, req->cb_arg); evbuffer_drain(req->input_buffer, evbuffer_get_length(req->input_buffer)); req->flags &= ~EVHTTP_REQ_DEFER_FREE; if ((req->flags & EVHTTP_REQ_NEEDS_FREE) != 0) { return (REQUEST_CANCELED); } } } return (MORE_DATA_EXPECTED); } static void evhttp_read_trailer(struct evhttp_connection *evcon, struct evhttp_request *req) { struct evbuffer *buf = bufferevent_get_input(evcon->bufev); switch (evhttp_parse_headers_(req, buf)) { case DATA_CORRUPTED: case DATA_TOO_LONG: evhttp_connection_fail_(evcon, EVREQ_HTTP_DATA_TOO_LONG); break; case ALL_DATA_READ: bufferevent_disable(evcon->bufev, EV_READ); evhttp_connection_done(evcon); break; case MORE_DATA_EXPECTED: case REQUEST_CANCELED: /* ??? */ default: break; } } static void evhttp_lingering_close(struct evhttp_connection *evcon, struct evhttp_request *req) { struct evbuffer *buf = bufferevent_get_input(evcon->bufev); size_t n = evbuffer_get_length(buf); if (n > (size_t) req->ntoread) n = (size_t) req->ntoread; req->ntoread -= n; req->body_size += n; event_debug(("Request body is too long, left " EV_I64_FMT, EV_I64_ARG(req->ntoread))); evbuffer_drain(buf, n); if (!req->ntoread) evhttp_connection_fail_(evcon, EVREQ_HTTP_DATA_TOO_LONG); } static void evhttp_lingering_fail(struct evhttp_connection *evcon, struct evhttp_request *req) { if (evcon->flags & EVHTTP_CON_LINGERING_CLOSE) evhttp_lingering_close(evcon, req); else evhttp_connection_fail_(evcon, EVREQ_HTTP_DATA_TOO_LONG); } static void evhttp_read_body(struct evhttp_connection *evcon, struct evhttp_request *req) { struct evbuffer *buf = bufferevent_get_input(evcon->bufev); if (req->chunked) { switch (evhttp_handle_chunked_read(req, buf)) { case ALL_DATA_READ: /* finished last chunk */ evcon->state = EVCON_READING_TRAILER; evhttp_read_trailer(evcon, req); return; case DATA_CORRUPTED: case DATA_TOO_LONG: /* corrupted data */ evhttp_connection_fail_(evcon, EVREQ_HTTP_DATA_TOO_LONG); return; case REQUEST_CANCELED: /* request canceled */ evhttp_request_free_auto(req); return; case MORE_DATA_EXPECTED: default: break; } } else if (req->ntoread < 0) { /* Read until connection close. */ if ((size_t)(req->body_size + evbuffer_get_length(buf)) < req->body_size) { evhttp_connection_fail_(evcon, EVREQ_HTTP_INVALID_HEADER); return; } req->body_size += evbuffer_get_length(buf); evbuffer_add_buffer(req->input_buffer, buf); } else if (req->chunk_cb != NULL || evbuffer_get_length(buf) >= (size_t)req->ntoread) { /* XXX: the above get_length comparison has to be fixed for overflow conditions! */ /* We've postponed moving the data until now, but we're * about to use it. */ size_t n = evbuffer_get_length(buf); if (n > (size_t) req->ntoread) n = (size_t) req->ntoread; req->ntoread -= n; req->body_size += n; evbuffer_remove_buffer(buf, req->input_buffer, n); } if (req->body_size > req->evcon->max_body_size || (!req->chunked && req->ntoread >= 0 && (size_t)req->ntoread > req->evcon->max_body_size)) { /* XXX: The above casted comparison must checked for overflow */ /* failed body length test */ evhttp_lingering_fail(evcon, req); return; } if (evbuffer_get_length(req->input_buffer) > 0 && req->chunk_cb != NULL) { req->flags |= EVHTTP_REQ_DEFER_FREE; (*req->chunk_cb)(req, req->cb_arg); req->flags &= ~EVHTTP_REQ_DEFER_FREE; evbuffer_drain(req->input_buffer, evbuffer_get_length(req->input_buffer)); if ((req->flags & EVHTTP_REQ_NEEDS_FREE) != 0) { evhttp_request_free_auto(req); return; } } if (!req->ntoread) { bufferevent_disable(evcon->bufev, EV_READ); /* Completed content length */ evhttp_connection_done(evcon); return; } } #define get_deferred_queue(evcon) \ ((evcon)->base) /* * Gets called when more data becomes available */ static void evhttp_read_cb(struct bufferevent *bufev, void *arg) { struct evhttp_connection *evcon = arg; struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); /* Cancel if it's pending. */ event_deferred_cb_cancel_(get_deferred_queue(evcon), &evcon->read_more_deferred_cb); switch (evcon->state) { case EVCON_READING_FIRSTLINE: evhttp_read_firstline(evcon, req); /* note the request may have been freed in * evhttp_read_body */ break; case EVCON_READING_HEADERS: evhttp_read_header(evcon, req); /* note the request may have been freed in * evhttp_read_body */ break; case EVCON_READING_BODY: evhttp_read_body(evcon, req); /* note the request may have been freed in * evhttp_read_body */ break; case EVCON_READING_TRAILER: evhttp_read_trailer(evcon, req); break; case EVCON_IDLE: { #ifdef USE_DEBUG struct evbuffer *input; size_t total_len; input = bufferevent_get_input(evcon->bufev); total_len = evbuffer_get_length(input); event_debug(("%s: read "EV_SIZE_FMT " bytes in EVCON_IDLE state," " resetting connection", __func__, EV_SIZE_ARG(total_len))); #endif evhttp_connection_reset_(evcon); } break; case EVCON_DISCONNECTED: case EVCON_CONNECTING: case EVCON_WRITING: default: event_errx(1, "%s: illegal connection state %d", __func__, evcon->state); } } static void evhttp_deferred_read_cb(struct event_callback *cb, void *data) { struct evhttp_connection *evcon = data; struct bufferevent *bev = evcon->bufev; if (bev->readcb) (bev->readcb)(evcon->bufev, evcon); } static void evhttp_write_connectioncb(struct evhttp_connection *evcon, void *arg) { /* This is after writing the request to the server */ struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); struct evbuffer *output = bufferevent_get_output(evcon->bufev); EVUTIL_ASSERT(req != NULL); EVUTIL_ASSERT(evcon->state == EVCON_WRITING); /* We need to wait until we've written all of our output data before we can * continue */ if (evbuffer_get_length(output) > 0) return; /* We are done writing our header and are now expecting the response */ req->kind = EVHTTP_RESPONSE; evhttp_start_read_(evcon); } /* * Clean up a connection object */ void evhttp_connection_free(struct evhttp_connection *evcon) { struct evhttp_request *req; int need_close = 0; /* notify interested parties that this connection is going down */ if (evcon->fd != -1) { if (evhttp_connected(evcon) && evcon->closecb != NULL) (*evcon->closecb)(evcon, evcon->closecb_arg); } /* remove all requests that might be queued on this * connection. for server connections, this should be empty. * because it gets dequeued either in evhttp_connection_done or * evhttp_connection_fail_. */ while ((req = TAILQ_FIRST(&evcon->requests)) != NULL) { evhttp_request_free_(evcon, req); } if (evcon->http_server != NULL) { struct evhttp *http = evcon->http_server; TAILQ_REMOVE(&http->connections, evcon, next); } if (event_initialized(&evcon->retry_ev)) { event_del(&evcon->retry_ev); event_debug_unassign(&evcon->retry_ev); } event_deferred_cb_cancel_(get_deferred_queue(evcon), &evcon->read_more_deferred_cb); if (evcon->bufev != NULL) { need_close = !(bufferevent_get_options_(evcon->bufev) & BEV_OPT_CLOSE_ON_FREE); if (evcon->fd == -1) evcon->fd = bufferevent_getfd(evcon->bufev); bufferevent_free(evcon->bufev); } if (evcon->fd != -1) { shutdown(evcon->fd, EVUTIL_SHUT_WR); if (need_close) evutil_closesocket(evcon->fd); } if (evcon->bind_address != NULL) mm_free(evcon->bind_address); if (evcon->address != NULL) mm_free(evcon->address); mm_free(evcon); } void evhttp_connection_free_on_completion(struct evhttp_connection *evcon) { evcon->flags |= EVHTTP_CON_AUTOFREE; } void evhttp_connection_set_local_address(struct evhttp_connection *evcon, const char *address) { EVUTIL_ASSERT(evcon->state == EVCON_DISCONNECTED); if (evcon->bind_address) mm_free(evcon->bind_address); if ((evcon->bind_address = mm_strdup(address)) == NULL) event_warn("%s: strdup", __func__); } void evhttp_connection_set_local_port(struct evhttp_connection *evcon, ev_uint16_t port) { EVUTIL_ASSERT(evcon->state == EVCON_DISCONNECTED); evcon->bind_port = port; } static void evhttp_request_dispatch(struct evhttp_connection* evcon) { struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); /* this should not usually happy but it's possible */ if (req == NULL) return; EVUTIL_ASSERT(req->kind == EVHTTP_REQUEST); /* delete possible close detection events */ evhttp_connection_stop_detectclose(evcon); /* we assume that the connection is connected already */ EVUTIL_ASSERT(evcon->state == EVCON_IDLE); evcon->state = EVCON_WRITING; /* Create the header from the store arguments */ evhttp_make_header(evcon, req); evhttp_write_buffer(evcon, evhttp_write_connectioncb, NULL); } /* Reset our connection state: disables reading/writing, closes our fd (if * any), clears out buffers, and puts us in state DISCONNECTED. */ void evhttp_connection_reset_(struct evhttp_connection *evcon) { struct evbuffer *tmp; int err; bufferevent_setcb(evcon->bufev, NULL, NULL, NULL, NULL); /* XXXX This is not actually an optimal fix. Instead we ought to have an API for "stop connecting", or use bufferevent_setfd to turn off connecting. But for Libevent 2.0, this seems like a minimal change least likely to disrupt the rest of the bufferevent and http code. Why is this here? If the fd is set in the bufferevent, and the bufferevent is connecting, then you can't actually stop the bufferevent from trying to connect with bufferevent_disable(). The connect will never trigger, since we close the fd, but the timeout might. That caused an assertion failure in evhttp_connection_fail_. */ bufferevent_disable_hard_(evcon->bufev, EV_READ|EV_WRITE); if (evcon->fd == -1) evcon->fd = bufferevent_getfd(evcon->bufev); if (evcon->fd != -1) { /* inform interested parties about connection close */ if (evhttp_connected(evcon) && evcon->closecb != NULL) (*evcon->closecb)(evcon, evcon->closecb_arg); shutdown(evcon->fd, EVUTIL_SHUT_WR); evutil_closesocket(evcon->fd); evcon->fd = -1; } err = bufferevent_setfd(evcon->bufev, -1); EVUTIL_ASSERT(!err && "setfd"); /* we need to clean up any buffered data */ tmp = bufferevent_get_output(evcon->bufev); err = evbuffer_drain(tmp, -1); EVUTIL_ASSERT(!err && "drain output"); tmp = bufferevent_get_input(evcon->bufev); err = evbuffer_drain(tmp, -1); EVUTIL_ASSERT(!err && "drain input"); evcon->flags &= ~EVHTTP_CON_READING_ERROR; evcon->state = EVCON_DISCONNECTED; } static void evhttp_connection_start_detectclose(struct evhttp_connection *evcon) { evcon->flags |= EVHTTP_CON_CLOSEDETECT; bufferevent_enable(evcon->bufev, EV_READ); } static void evhttp_connection_stop_detectclose(struct evhttp_connection *evcon) { evcon->flags &= ~EVHTTP_CON_CLOSEDETECT; bufferevent_disable(evcon->bufev, EV_READ); } static void evhttp_connection_retry(evutil_socket_t fd, short what, void *arg) { struct evhttp_connection *evcon = arg; evcon->state = EVCON_DISCONNECTED; evhttp_connection_connect_(evcon); } static void evhttp_connection_cb_cleanup(struct evhttp_connection *evcon) { struct evcon_requestq requests; evhttp_connection_reset_(evcon); if (evcon->retry_max < 0 || evcon->retry_cnt < evcon->retry_max) { struct timeval tv_retry = evcon->initial_retry_timeout; int i; evtimer_assign(&evcon->retry_ev, evcon->base, evhttp_connection_retry, evcon); /* XXXX handle failure from evhttp_add_event */ for (i=0; i < evcon->retry_cnt; ++i) { tv_retry.tv_usec *= 2; if (tv_retry.tv_usec > 1000000) { tv_retry.tv_usec -= 1000000; tv_retry.tv_sec += 1; } tv_retry.tv_sec *= 2; if (tv_retry.tv_sec > 3600) { tv_retry.tv_sec = 3600; tv_retry.tv_usec = 0; } } event_add(&evcon->retry_ev, &tv_retry); evcon->retry_cnt++; return; } /* * User callback can do evhttp_make_request() on the same * evcon so new request will be added to evcon->requests. To * avoid freeing it prematurely we iterate over the copy of * the queue. */ TAILQ_INIT(&requests); while (TAILQ_FIRST(&evcon->requests) != NULL) { struct evhttp_request *request = TAILQ_FIRST(&evcon->requests); TAILQ_REMOVE(&evcon->requests, request, next); TAILQ_INSERT_TAIL(&requests, request, next); } /* for now, we just signal all requests by executing their callbacks */ while (TAILQ_FIRST(&requests) != NULL) { struct evhttp_request *request = TAILQ_FIRST(&requests); TAILQ_REMOVE(&requests, request, next); request->evcon = NULL; /* we might want to set an error here */ request->cb(request, request->cb_arg); evhttp_request_free_auto(request); } } static void evhttp_connection_read_on_write_error(struct evhttp_connection *evcon, struct evhttp_request *req) { struct evbuffer *buf; /** Second time, we can't read anything */ if (evcon->flags & EVHTTP_CON_READING_ERROR) { evcon->flags &= ~EVHTTP_CON_READING_ERROR; evhttp_connection_fail_(evcon, EVREQ_HTTP_EOF); return; } req->kind = EVHTTP_RESPONSE; buf = bufferevent_get_output(evcon->bufev); evbuffer_unfreeze(buf, 1); evbuffer_drain(buf, evbuffer_get_length(buf)); evbuffer_freeze(buf, 1); evhttp_start_read_(evcon); evcon->flags |= EVHTTP_CON_READING_ERROR; } static void evhttp_error_cb(struct bufferevent *bufev, short what, void *arg) { struct evhttp_connection *evcon = arg; struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); if (evcon->fd == -1) evcon->fd = bufferevent_getfd(bufev); switch (evcon->state) { case EVCON_CONNECTING: if (what & BEV_EVENT_TIMEOUT) { event_debug(("%s: connection timeout for \"%s:%d\" on " EV_SOCK_FMT, __func__, evcon->address, evcon->port, EV_SOCK_ARG(evcon->fd))); evhttp_connection_cb_cleanup(evcon); return; } break; case EVCON_READING_BODY: if (!req->chunked && req->ntoread < 0 && what == (BEV_EVENT_READING|BEV_EVENT_EOF)) { /* EOF on read can be benign */ evhttp_connection_done(evcon); return; } break; case EVCON_DISCONNECTED: case EVCON_IDLE: case EVCON_READING_FIRSTLINE: case EVCON_READING_HEADERS: case EVCON_READING_TRAILER: case EVCON_WRITING: default: break; } /* when we are in close detect mode, a read error means that * the other side closed their connection. */ if (evcon->flags & EVHTTP_CON_CLOSEDETECT) { evcon->flags &= ~EVHTTP_CON_CLOSEDETECT; EVUTIL_ASSERT(evcon->http_server == NULL); /* For connections from the client, we just * reset the connection so that it becomes * disconnected. */ EVUTIL_ASSERT(evcon->state == EVCON_IDLE); evhttp_connection_reset_(evcon); /* * If we have no more requests that need completion * and we want to auto-free the connection when all * requests have been completed. */ if (TAILQ_FIRST(&evcon->requests) == NULL && (evcon->flags & EVHTTP_CON_OUTGOING) && (evcon->flags & EVHTTP_CON_AUTOFREE)) { evhttp_connection_free(evcon); } return; } if (what & BEV_EVENT_TIMEOUT) { evhttp_connection_fail_(evcon, EVREQ_HTTP_TIMEOUT); } else if (what & (BEV_EVENT_EOF|BEV_EVENT_ERROR)) { if (what & BEV_EVENT_WRITING && evcon->flags & EVHTTP_CON_READ_ON_WRITE_ERROR) { evhttp_connection_read_on_write_error(evcon, req); return; } if (what & BEV_EVENT_READING && evcon->flags & EVHTTP_CON_READ_ON_WRITE_ERROR && evbuffer_get_length(bufferevent_get_input(bufev))) { event_deferred_cb_schedule_(get_deferred_queue(evcon), &evcon->read_more_deferred_cb); return; } evhttp_connection_fail_(evcon, EVREQ_HTTP_EOF); } else if (what == BEV_EVENT_CONNECTED) { } else { evhttp_connection_fail_(evcon, EVREQ_HTTP_BUFFER_ERROR); } } /* * Event callback for asynchronous connection attempt. */ static void evhttp_connection_cb(struct bufferevent *bufev, short what, void *arg) { struct evhttp_connection *evcon = arg; int error; ev_socklen_t errsz = sizeof(error); if (evcon->fd == -1) evcon->fd = bufferevent_getfd(bufev); if (!(what & BEV_EVENT_CONNECTED)) { /* some operating systems return ECONNREFUSED immediately * when connecting to a local address. the cleanup is going * to reschedule this function call. */ #ifndef _WIN32 if (errno == ECONNREFUSED) goto cleanup; #endif evhttp_error_cb(bufev, what, arg); return; } if (evcon->fd == -1) { event_debug(("%s: bufferevent_getfd returned -1", __func__)); goto cleanup; } /* Check if the connection completed */ if (getsockopt(evcon->fd, SOL_SOCKET, SO_ERROR, (void*)&error, &errsz) == -1) { event_debug(("%s: getsockopt for \"%s:%d\" on "EV_SOCK_FMT, __func__, evcon->address, evcon->port, EV_SOCK_ARG(evcon->fd))); goto cleanup; } if (error) { event_debug(("%s: connect failed for \"%s:%d\" on " EV_SOCK_FMT": %s", __func__, evcon->address, evcon->port, EV_SOCK_ARG(evcon->fd), evutil_socket_error_to_string(error))); goto cleanup; } /* We are connected to the server now */ event_debug(("%s: connected to \"%s:%d\" on "EV_SOCK_FMT"\n", __func__, evcon->address, evcon->port, EV_SOCK_ARG(evcon->fd))); /* Reset the retry count as we were successful in connecting */ evcon->retry_cnt = 0; evcon->state = EVCON_IDLE; /* reset the bufferevent cbs */ bufferevent_setcb(evcon->bufev, evhttp_read_cb, evhttp_write_cb, evhttp_error_cb, evcon); if (!evutil_timerisset(&evcon->timeout)) { const struct timeval read_tv = { HTTP_READ_TIMEOUT, 0 }; const struct timeval write_tv = { HTTP_WRITE_TIMEOUT, 0 }; bufferevent_set_timeouts(evcon->bufev, &read_tv, &write_tv); } else { bufferevent_set_timeouts(evcon->bufev, &evcon->timeout, &evcon->timeout); } /* try to start requests that have queued up on this connection */ evhttp_request_dispatch(evcon); return; cleanup: evhttp_connection_cb_cleanup(evcon); } /* * Check if we got a valid response code. */ static int evhttp_valid_response_code(int code) { if (code == 0) return (0); return (1); } static int evhttp_parse_http_version(const char *version, struct evhttp_request *req) { int major, minor; char ch; int n = sscanf(version, "HTTP/%d.%d%c", &major, &minor, &ch); if (n != 2 || major > 1) { event_debug(("%s: bad version %s on message %p from %s", __func__, version, req, req->remote_host)); return (-1); } req->major = major; req->minor = minor; return (0); } /* Parses the status line of a web server */ static int evhttp_parse_response_line(struct evhttp_request *req, char *line) { char *protocol; char *number; const char *readable = ""; protocol = strsep(&line, " "); if (line == NULL) return (-1); number = strsep(&line, " "); if (line != NULL) readable = line; if (evhttp_parse_http_version(protocol, req) < 0) return (-1); req->response_code = atoi(number); if (!evhttp_valid_response_code(req->response_code)) { event_debug(("%s: bad response code \"%s\"", __func__, number)); return (-1); } if (req->response_code_line != NULL) mm_free(req->response_code_line); if ((req->response_code_line = mm_strdup(readable)) == NULL) { event_warn("%s: strdup", __func__); return (-1); } return (0); } /* Parse the first line of a HTTP request */ static int evhttp_parse_request_line(struct evhttp_request *req, char *line, size_t len) { char *eos = line + len; char *method; char *uri; char *version; const char *hostname; const char *scheme; size_t method_len; enum evhttp_cmd_type type; while (eos > line && *(eos-1) == ' ') { *(eos-1) = '\0'; --eos; --len; } if (len < strlen("GET / HTTP/1.0")) return -1; /* Parse the request line */ method = strsep(&line, " "); if (!line) return -1; uri = line; version = strrchr(uri, ' '); if (!version || uri == version) return -1; *version = '\0'; version++; method_len = (uri - method) - 1; type = EVHTTP_REQ_UNKNOWN_; /* First line */ switch (method_len) { case 3: /* The length of the method string is 3, meaning it can only be one of two methods: GET or PUT */ /* Since both GET and PUT share the same character 'T' at the end, * if the string doesn't have 'T', we can immediately determine this * is an invalid HTTP method */ if (method[2] != 'T') { break; } switch (*method) { case 'G': /* This first byte is 'G', so make sure the next byte is * 'E', if it isn't then this isn't a valid method */ if (method[1] == 'E') { type = EVHTTP_REQ_GET; } break; case 'P': /* First byte is P, check second byte for 'U', if not, * we know it's an invalid method */ if (method[1] == 'U') { type = EVHTTP_REQ_PUT; } break; default: break; } break; case 4: /* The method length is 4 bytes, leaving only the methods "POST" and "HEAD" */ switch (*method) { case 'P': if (method[3] == 'T' && method[2] == 'S' && method[1] == 'O') { type = EVHTTP_REQ_POST; } break; case 'H': if (method[3] == 'D' && method[2] == 'A' && method[1] == 'E') { type = EVHTTP_REQ_HEAD; } break; default: break; } break; case 5: /* Method length is 5 bytes, which can only encompass PATCH and TRACE */ switch (*method) { case 'P': if (method[4] == 'H' && method[3] == 'C' && method[2] == 'T' && method[1] == 'A') { type = EVHTTP_REQ_PATCH; } break; case 'T': if (method[4] == 'E' && method[3] == 'C' && method[2] == 'A' && method[1] == 'R') { type = EVHTTP_REQ_TRACE; } break; default: break; } break; case 6: /* Method length is 6, only valid method 6 bytes in length is DELEte */ /* If the first byte isn't 'D' then it's invalid */ if (*method != 'D') { break; } if (method[5] == 'E' && method[4] == 'T' && method[3] == 'E' && method[2] == 'L' && method[1] == 'E') { type = EVHTTP_REQ_DELETE; } break; case 7: /* Method length is 7, only valid methods are "OPTIONS" and "CONNECT" */ switch (*method) { case 'O': if (method[6] == 'S' && method[5] == 'N' && method[4] == 'O' && method[3] == 'I' && method[2] == 'T' && method[1] == 'P') { type = EVHTTP_REQ_OPTIONS; } break; case 'C': if (method[6] == 'T' && method[5] == 'C' && method[4] == 'E' && method[3] == 'N' && method[2] == 'N' && method[1] == 'O') { type = EVHTTP_REQ_CONNECT; } break; default: break; } break; } /* switch */ if ((int)type == EVHTTP_REQ_UNKNOWN_) { event_debug(("%s: bad method %s on request %p from %s", __func__, method, req, req->remote_host)); /* No error yet; we'll give a better error later when * we see that req->type is unsupported. */ } req->type = type; if (evhttp_parse_http_version(version, req) < 0) return -1; if ((req->uri = mm_strdup(uri)) == NULL) { event_debug(("%s: mm_strdup", __func__)); return -1; } if (type == EVHTTP_REQ_CONNECT) { if ((req->uri_elems = evhttp_uri_parse_authority(req->uri)) == NULL) { return -1; } } else { if ((req->uri_elems = evhttp_uri_parse_with_flags(req->uri, EVHTTP_URI_NONCONFORMANT)) == NULL) { return -1; } } /* If we have an absolute-URI, check to see if it is an http request for a known vhost or server alias. If we don't know about this host, we consider it a proxy request. */ scheme = evhttp_uri_get_scheme(req->uri_elems); hostname = evhttp_uri_get_host(req->uri_elems); if (scheme && (!evutil_ascii_strcasecmp(scheme, "http") || !evutil_ascii_strcasecmp(scheme, "https")) && hostname && !evhttp_find_vhost(req->evcon->http_server, NULL, hostname)) req->flags |= EVHTTP_PROXY_REQUEST; return 0; } const char * evhttp_find_header(const struct evkeyvalq *headers, const char *key) { struct evkeyval *header; TAILQ_FOREACH(header, headers, next) { if (evutil_ascii_strcasecmp(header->key, key) == 0) return (header->value); } return (NULL); } void evhttp_clear_headers(struct evkeyvalq *headers) { struct evkeyval *header; for (header = TAILQ_FIRST(headers); header != NULL; header = TAILQ_FIRST(headers)) { TAILQ_REMOVE(headers, header, next); mm_free(header->key); mm_free(header->value); mm_free(header); } } /* * Returns 0, if the header was successfully removed. * Returns -1, if the header could not be found. */ int evhttp_remove_header(struct evkeyvalq *headers, const char *key) { struct evkeyval *header; TAILQ_FOREACH(header, headers, next) { if (evutil_ascii_strcasecmp(header->key, key) == 0) break; } if (header == NULL) return (-1); /* Free and remove the header that we found */ TAILQ_REMOVE(headers, header, next); mm_free(header->key); mm_free(header->value); mm_free(header); return (0); } static int evhttp_header_is_valid_value(const char *value) { const char *p = value; while ((p = strpbrk(p, "\r\n")) != NULL) { /* we really expect only one new line */ p += strspn(p, "\r\n"); /* we expect a space or tab for continuation */ if (*p != ' ' && *p != '\t') return (0); } return (1); } int evhttp_add_header(struct evkeyvalq *headers, const char *key, const char *value) { event_debug(("%s: key: %s val: %s\n", __func__, key, value)); if (strchr(key, '\r') != NULL || strchr(key, '\n') != NULL) { /* drop illegal headers */ event_debug(("%s: dropping illegal header key\n", __func__)); return (-1); } if (!evhttp_header_is_valid_value(value)) { event_debug(("%s: dropping illegal header value\n", __func__)); return (-1); } return (evhttp_add_header_internal(headers, key, value)); } static int evhttp_add_header_internal(struct evkeyvalq *headers, const char *key, const char *value) { struct evkeyval *header = mm_calloc(1, sizeof(struct evkeyval)); if (header == NULL) { event_warn("%s: calloc", __func__); return (-1); } if ((header->key = mm_strdup(key)) == NULL) { mm_free(header); event_warn("%s: strdup", __func__); return (-1); } if ((header->value = mm_strdup(value)) == NULL) { mm_free(header->key); mm_free(header); event_warn("%s: strdup", __func__); return (-1); } TAILQ_INSERT_TAIL(headers, header, next); return (0); } /* * Parses header lines from a request or a response into the specified * request object given an event buffer. * * Returns * DATA_CORRUPTED on error * MORE_DATA_EXPECTED when we need to read more headers * ALL_DATA_READ when all headers have been read. */ enum message_read_status evhttp_parse_firstline_(struct evhttp_request *req, struct evbuffer *buffer) { char *line; enum message_read_status status = ALL_DATA_READ; size_t len; /* XXX try */ line = evbuffer_readln(buffer, &len, EVBUFFER_EOL_CRLF); if (line == NULL) { if (req->evcon != NULL && evbuffer_get_length(buffer) > req->evcon->max_headers_size) return (DATA_TOO_LONG); else return (MORE_DATA_EXPECTED); } if (req->evcon != NULL && len > req->evcon->max_headers_size) { mm_free(line); return (DATA_TOO_LONG); } req->headers_size = len; switch (req->kind) { case EVHTTP_REQUEST: if (evhttp_parse_request_line(req, line, len) == -1) status = DATA_CORRUPTED; break; case EVHTTP_RESPONSE: if (evhttp_parse_response_line(req, line) == -1) status = DATA_CORRUPTED; break; default: status = DATA_CORRUPTED; } mm_free(line); return (status); } static int evhttp_append_to_last_header(struct evkeyvalq *headers, char *line) { struct evkeyval *header = TAILQ_LAST(headers, evkeyvalq); char *newval; size_t old_len, line_len; if (header == NULL) return (-1); old_len = strlen(header->value); /* Strip space from start and end of line. */ while (*line == ' ' || *line == '\t') ++line; evutil_rtrim_lws_(line); line_len = strlen(line); newval = mm_realloc(header->value, old_len + line_len + 2); if (newval == NULL) return (-1); newval[old_len] = ' '; memcpy(newval + old_len + 1, line, line_len + 1); header->value = newval; return (0); } enum message_read_status evhttp_parse_headers_(struct evhttp_request *req, struct evbuffer* buffer) { enum message_read_status errcode = DATA_CORRUPTED; char *line; enum message_read_status status = MORE_DATA_EXPECTED; struct evkeyvalq* headers = req->input_headers; size_t len; while ((line = evbuffer_readln(buffer, &len, EVBUFFER_EOL_CRLF)) != NULL) { char *skey, *svalue; req->headers_size += len; if (req->evcon != NULL && req->headers_size > req->evcon->max_headers_size) { errcode = DATA_TOO_LONG; goto error; } if (*line == '\0') { /* Last header - Done */ status = ALL_DATA_READ; mm_free(line); break; } /* Check if this is a continuation line */ if (*line == ' ' || *line == '\t') { if (evhttp_append_to_last_header(headers, line) == -1) goto error; mm_free(line); continue; } /* Processing of header lines */ svalue = line; skey = strsep(&svalue, ":"); if (svalue == NULL) goto error; svalue += strspn(svalue, " "); evutil_rtrim_lws_(svalue); if (evhttp_add_header(headers, skey, svalue) == -1) goto error; mm_free(line); } if (status == MORE_DATA_EXPECTED) { if (req->evcon != NULL && req->headers_size + evbuffer_get_length(buffer) > req->evcon->max_headers_size) return (DATA_TOO_LONG); } return (status); error: mm_free(line); return (errcode); } static int evhttp_get_body_length(struct evhttp_request *req) { struct evkeyvalq *headers = req->input_headers; const char *content_length; const char *connection; content_length = evhttp_find_header(headers, "Content-Length"); connection = evhttp_find_header(headers, "Connection"); if (content_length == NULL && connection == NULL) req->ntoread = -1; else if (content_length == NULL && evutil_ascii_strcasecmp(connection, "Close") != 0) { req->ntoread = 0; } else if (content_length == NULL) { req->ntoread = -1; } else { char *endp; ev_int64_t ntoread = evutil_strtoll(content_length, &endp, 10); if (*content_length == '\0' || *endp != '\0' || ntoread < 0) { event_debug(("%s: illegal content length: %s", __func__, content_length)); return (-1); } req->ntoread = ntoread; } event_debug(("%s: bytes to read: "EV_I64_FMT" (in buffer "EV_SIZE_FMT")\n", __func__, EV_I64_ARG(req->ntoread), EV_SIZE_ARG(evbuffer_get_length(bufferevent_get_input(req->evcon->bufev))))); return (0); } static int evhttp_method_may_have_body(enum evhttp_cmd_type type) { switch (type) { case EVHTTP_REQ_POST: case EVHTTP_REQ_PUT: case EVHTTP_REQ_PATCH: case EVHTTP_REQ_GET: case EVHTTP_REQ_DELETE: case EVHTTP_REQ_OPTIONS: case EVHTTP_REQ_CONNECT: return 1; case EVHTTP_REQ_TRACE: case EVHTTP_REQ_HEAD: default: return 0; } } static void evhttp_get_body(struct evhttp_connection *evcon, struct evhttp_request *req) { const char *xfer_enc; /* If this is a request without a body, then we are done */ if (req->kind == EVHTTP_REQUEST && !evhttp_method_may_have_body(req->type)) { evhttp_connection_done(evcon); return; } evcon->state = EVCON_READING_BODY; xfer_enc = evhttp_find_header(req->input_headers, "Transfer-Encoding"); if (xfer_enc != NULL && evutil_ascii_strcasecmp(xfer_enc, "chunked") == 0) { req->chunked = 1; req->ntoread = -1; } else { if (evhttp_get_body_length(req) == -1) { evhttp_connection_fail_(evcon, EVREQ_HTTP_INVALID_HEADER); return; } if (req->kind == EVHTTP_REQUEST && req->ntoread < 1) { /* An incoming request with no content-length and no * transfer-encoding has no body. */ evhttp_connection_done(evcon); return; } } /* Should we send a 100 Continue status line? */ switch (evhttp_have_expect(req, 1)) { case CONTINUE: /* XXX It would be nice to do some sanity checking here. Does the resource exist? Should the resource accept post requests? If no, we should respond with an error. For now, just optimistically tell the client to send their message body. */ if (req->ntoread > 0) { /* ntoread is ev_int64_t, max_body_size is ev_uint64_t */ if ((req->evcon->max_body_size <= EV_INT64_MAX) && (ev_uint64_t)req->ntoread > req->evcon->max_body_size) { evhttp_lingering_fail(evcon, req); return; } } if (!evbuffer_get_length(bufferevent_get_input(evcon->bufev))) evhttp_send_continue(evcon, req); break; case OTHER: evhttp_send_error(req, HTTP_EXPECTATIONFAILED, NULL); return; case NO: break; } evhttp_read_body(evcon, req); /* note the request may have been freed in evhttp_read_body */ } static void evhttp_read_firstline(struct evhttp_connection *evcon, struct evhttp_request *req) { enum message_read_status res; res = evhttp_parse_firstline_(req, bufferevent_get_input(evcon->bufev)); if (res == DATA_CORRUPTED || res == DATA_TOO_LONG) { /* Error while reading, terminate */ event_debug(("%s: bad header lines on "EV_SOCK_FMT"\n", __func__, EV_SOCK_ARG(evcon->fd))); evhttp_connection_fail_(evcon, EVREQ_HTTP_INVALID_HEADER); return; } else if (res == MORE_DATA_EXPECTED) { /* Need more header lines */ return; } evcon->state = EVCON_READING_HEADERS; evhttp_read_header(evcon, req); } static void evhttp_read_header(struct evhttp_connection *evcon, struct evhttp_request *req) { enum message_read_status res; evutil_socket_t fd = evcon->fd; res = evhttp_parse_headers_(req, bufferevent_get_input(evcon->bufev)); if (res == DATA_CORRUPTED || res == DATA_TOO_LONG) { /* Error while reading, terminate */ event_debug(("%s: bad header lines on "EV_SOCK_FMT"\n", __func__, EV_SOCK_ARG(fd))); evhttp_connection_fail_(evcon, EVREQ_HTTP_INVALID_HEADER); return; } else if (res == MORE_DATA_EXPECTED) { /* Need more header lines */ return; } /* Callback can shut down connection with negative return value */ if (req->header_cb != NULL) { if ((*req->header_cb)(req, req->cb_arg) < 0) { evhttp_connection_fail_(evcon, EVREQ_HTTP_EOF); return; } } /* Done reading headers, do the real work */ switch (req->kind) { case EVHTTP_REQUEST: event_debug(("%s: checking for post data on "EV_SOCK_FMT"\n", __func__, EV_SOCK_ARG(fd))); evhttp_get_body(evcon, req); /* note the request may have been freed in evhttp_get_body */ break; case EVHTTP_RESPONSE: /* Start over if we got a 100 Continue response. */ if (req->response_code == 100) { struct evbuffer *output = bufferevent_get_output(evcon->bufev); evbuffer_add_buffer(output, req->output_buffer); evhttp_start_write_(evcon); return; } if (!evhttp_response_needs_body(req)) { event_debug(("%s: skipping body for code %d\n", __func__, req->response_code)); evhttp_connection_done(evcon); } else { event_debug(("%s: start of read body for %s on " EV_SOCK_FMT"\n", __func__, req->remote_host, EV_SOCK_ARG(fd))); evhttp_get_body(evcon, req); /* note the request may have been freed in * evhttp_get_body */ } break; default: event_warnx("%s: bad header on "EV_SOCK_FMT, __func__, EV_SOCK_ARG(fd)); evhttp_connection_fail_(evcon, EVREQ_HTTP_INVALID_HEADER); break; } /* request may have been freed above */ } /* * Creates a TCP connection to the specified port and executes a callback * when finished. Failure or success is indicate by the passed connection * object. * * Although this interface accepts a hostname, it is intended to take * only numeric hostnames so that non-blocking DNS resolution can * happen elsewhere. */ struct evhttp_connection * evhttp_connection_new(const char *address, ev_uint16_t port) { return (evhttp_connection_base_new(NULL, NULL, address, port)); } struct evhttp_connection * evhttp_connection_base_bufferevent_new(struct event_base *base, struct evdns_base *dnsbase, struct bufferevent* bev, const char *address, ev_uint16_t port) { struct evhttp_connection *evcon = NULL; event_debug(("Attempting connection to %s:%d\n", address, port)); if ((evcon = mm_calloc(1, sizeof(struct evhttp_connection))) == NULL) { event_warn("%s: calloc failed", __func__); goto error; } evcon->fd = -1; evcon->port = port; evcon->max_headers_size = EV_SIZE_MAX; evcon->max_body_size = EV_SIZE_MAX; evutil_timerclear(&evcon->timeout); evcon->retry_cnt = evcon->retry_max = 0; if ((evcon->address = mm_strdup(address)) == NULL) { event_warn("%s: strdup failed", __func__); goto error; } if (bev == NULL) { if (!(bev = bufferevent_socket_new(base, -1, 0))) { event_warn("%s: bufferevent_socket_new failed", __func__); goto error; } } bufferevent_setcb(bev, evhttp_read_cb, evhttp_write_cb, evhttp_error_cb, evcon); evcon->bufev = bev; evcon->state = EVCON_DISCONNECTED; TAILQ_INIT(&evcon->requests); evcon->initial_retry_timeout.tv_sec = 2; evcon->initial_retry_timeout.tv_usec = 0; if (base != NULL) { evcon->base = base; if (bufferevent_get_base(bev) != base) bufferevent_base_set(base, evcon->bufev); } event_deferred_cb_init_( &evcon->read_more_deferred_cb, bufferevent_get_priority(bev), evhttp_deferred_read_cb, evcon); evcon->dns_base = dnsbase; evcon->ai_family = AF_UNSPEC; return (evcon); error: if (evcon != NULL) evhttp_connection_free(evcon); return (NULL); } struct bufferevent* evhttp_connection_get_bufferevent(struct evhttp_connection *evcon) { return evcon->bufev; } struct evhttp * evhttp_connection_get_server(struct evhttp_connection *evcon) { return evcon->http_server; } struct evhttp_connection * evhttp_connection_base_new(struct event_base *base, struct evdns_base *dnsbase, const char *address, ev_uint16_t port) { return evhttp_connection_base_bufferevent_new(base, dnsbase, NULL, address, port); } void evhttp_connection_set_family(struct evhttp_connection *evcon, int family) { evcon->ai_family = family; } int evhttp_connection_set_flags(struct evhttp_connection *evcon, int flags) { int avail_flags = 0; avail_flags |= EVHTTP_CON_REUSE_CONNECTED_ADDR; avail_flags |= EVHTTP_CON_READ_ON_WRITE_ERROR; if (flags & ~avail_flags || flags > EVHTTP_CON_PUBLIC_FLAGS_END) return 1; evcon->flags &= ~avail_flags; evcon->flags |= flags; return 0; } void evhttp_connection_set_base(struct evhttp_connection *evcon, struct event_base *base) { EVUTIL_ASSERT(evcon->base == NULL); EVUTIL_ASSERT(evcon->state == EVCON_DISCONNECTED); evcon->base = base; bufferevent_base_set(base, evcon->bufev); } void evhttp_connection_set_timeout(struct evhttp_connection *evcon, int timeout_in_secs) { if (timeout_in_secs == -1) evhttp_connection_set_timeout_tv(evcon, NULL); else { struct timeval tv; tv.tv_sec = timeout_in_secs; tv.tv_usec = 0; evhttp_connection_set_timeout_tv(evcon, &tv); } } void evhttp_connection_set_timeout_tv(struct evhttp_connection *evcon, const struct timeval* tv) { if (tv) { evcon->timeout = *tv; bufferevent_set_timeouts(evcon->bufev, &evcon->timeout, &evcon->timeout); } else { const struct timeval read_tv = { HTTP_READ_TIMEOUT, 0 }; const struct timeval write_tv = { HTTP_WRITE_TIMEOUT, 0 }; evutil_timerclear(&evcon->timeout); bufferevent_set_timeouts(evcon->bufev, &read_tv, &write_tv); } } void evhttp_connection_set_initial_retry_tv(struct evhttp_connection *evcon, const struct timeval *tv) { if (tv) { evcon->initial_retry_timeout = *tv; } else { evutil_timerclear(&evcon->initial_retry_timeout); evcon->initial_retry_timeout.tv_sec = 2; } } void evhttp_connection_set_retries(struct evhttp_connection *evcon, int retry_max) { evcon->retry_max = retry_max; } void evhttp_connection_set_closecb(struct evhttp_connection *evcon, void (*cb)(struct evhttp_connection *, void *), void *cbarg) { evcon->closecb = cb; evcon->closecb_arg = cbarg; } void evhttp_connection_get_peer(struct evhttp_connection *evcon, char **address, ev_uint16_t *port) { *address = evcon->address; *port = evcon->port; } const struct sockaddr* evhttp_connection_get_addr(struct evhttp_connection *evcon) { return bufferevent_socket_get_conn_address_(evcon->bufev); } int evhttp_connection_connect_(struct evhttp_connection *evcon) { int old_state = evcon->state; const char *address = evcon->address; const struct sockaddr *sa = evhttp_connection_get_addr(evcon); int ret; if (evcon->state == EVCON_CONNECTING) return (0); evhttp_connection_reset_(evcon); EVUTIL_ASSERT(!(evcon->flags & EVHTTP_CON_INCOMING)); evcon->flags |= EVHTTP_CON_OUTGOING; if (evcon->bind_address || evcon->bind_port) { evcon->fd = bind_socket( evcon->bind_address, evcon->bind_port, 0 /*reuse*/); if (evcon->fd == -1) { event_debug(("%s: failed to bind to \"%s\"", __func__, evcon->bind_address)); return (-1); } if (bufferevent_setfd(evcon->bufev, evcon->fd)) return (-1); } else { if (bufferevent_setfd(evcon->bufev, -1)) return (-1); } /* Set up a callback for successful connection setup */ bufferevent_setcb(evcon->bufev, NULL /* evhttp_read_cb */, NULL /* evhttp_write_cb */, evhttp_connection_cb, evcon); if (!evutil_timerisset(&evcon->timeout)) { const struct timeval conn_tv = { HTTP_CONNECT_TIMEOUT, 0 }; bufferevent_set_timeouts(evcon->bufev, &conn_tv, &conn_tv); } else { bufferevent_set_timeouts(evcon->bufev, &evcon->timeout, &evcon->timeout); } /* make sure that we get a write callback */ if (bufferevent_enable(evcon->bufev, EV_WRITE)) return (-1); evcon->state = EVCON_CONNECTING; if (evcon->flags & EVHTTP_CON_REUSE_CONNECTED_ADDR && sa && (sa->sa_family == AF_INET || sa->sa_family == AF_INET6)) { int socklen = sizeof(struct sockaddr_in); if (sa->sa_family == AF_INET6) { socklen = sizeof(struct sockaddr_in6); } ret = bufferevent_socket_connect(evcon->bufev, sa, socklen); } else { ret = bufferevent_socket_connect_hostname(evcon->bufev, evcon->dns_base, evcon->ai_family, address, evcon->port); } if (ret < 0) { evcon->state = old_state; event_sock_warn(evcon->fd, "%s: connection to \"%s\" failed", __func__, evcon->address); /* some operating systems return ECONNREFUSED immediately * when connecting to a local address. the cleanup is going * to reschedule this function call. */ evhttp_connection_cb_cleanup(evcon); return (0); } return (0); } /* * Starts an HTTP request on the provided evhttp_connection object. * If the connection object is not connected to the web server already, * this will start the connection. */ int evhttp_make_request(struct evhttp_connection *evcon, struct evhttp_request *req, enum evhttp_cmd_type type, const char *uri) { /* We are making a request */ req->kind = EVHTTP_REQUEST; req->type = type; if (req->uri != NULL) mm_free(req->uri); if ((req->uri = mm_strdup(uri)) == NULL) { event_warn("%s: strdup", __func__); evhttp_request_free_auto(req); return (-1); } /* Set the protocol version if it is not supplied */ if (!req->major && !req->minor) { req->major = 1; req->minor = 1; } EVUTIL_ASSERT(req->evcon == NULL); req->evcon = evcon; EVUTIL_ASSERT(!(req->flags & EVHTTP_REQ_OWN_CONNECTION)); TAILQ_INSERT_TAIL(&evcon->requests, req, next); /* We do not want to conflict with retry_ev */ if (evcon->retry_cnt) return (0); /* If the connection object is not connected; make it so */ if (!evhttp_connected(evcon)) { int res = evhttp_connection_connect_(evcon); /* evhttp_connection_fail_(), which is called through * evhttp_connection_connect_(), assumes that req lies in * evcon->requests. Thus, enqueue the request in advance and * remove it in the error case. */ if (res != 0) TAILQ_REMOVE(&evcon->requests, req, next); return (res); } /* * If it's connected already and we are the first in the queue, * then we can dispatch this request immediately. Otherwise, it * will be dispatched once the pending requests are completed. */ if (TAILQ_FIRST(&evcon->requests) == req) evhttp_request_dispatch(evcon); return (0); } void evhttp_cancel_request(struct evhttp_request *req) { struct evhttp_connection *evcon = req->evcon; if (evcon != NULL) { /* We need to remove it from the connection */ if (TAILQ_FIRST(&evcon->requests) == req) { /* it's currently being worked on, so reset * the connection. */ evhttp_connection_fail_(evcon, EVREQ_HTTP_REQUEST_CANCEL); /* connection fail freed the request */ return; } else { /* otherwise, we can just remove it from the * queue */ TAILQ_REMOVE(&evcon->requests, req, next); } } evhttp_request_free_auto(req); } /* * Reads data from file descriptor into request structure * Request structure needs to be set up correctly. */ void evhttp_start_read_(struct evhttp_connection *evcon) { bufferevent_disable(evcon->bufev, EV_WRITE); bufferevent_enable(evcon->bufev, EV_READ); evcon->state = EVCON_READING_FIRSTLINE; /* Reset the bufferevent callbacks */ bufferevent_setcb(evcon->bufev, evhttp_read_cb, evhttp_write_cb, evhttp_error_cb, evcon); /* If there's still data pending, process it next time through the * loop. Don't do it now; that could get recusive. */ if (evbuffer_get_length(bufferevent_get_input(evcon->bufev))) { event_deferred_cb_schedule_(get_deferred_queue(evcon), &evcon->read_more_deferred_cb); } } void evhttp_start_write_(struct evhttp_connection *evcon) { bufferevent_disable(evcon->bufev, EV_WRITE); bufferevent_enable(evcon->bufev, EV_READ); evcon->state = EVCON_WRITING; evhttp_write_buffer(evcon, evhttp_write_connectioncb, NULL); } static void evhttp_send_done(struct evhttp_connection *evcon, void *arg) { int need_close; struct evhttp_request *req = TAILQ_FIRST(&evcon->requests); TAILQ_REMOVE(&evcon->requests, req, next); if (req->on_complete_cb != NULL) { req->on_complete_cb(req, req->on_complete_cb_arg); } need_close = (REQ_VERSION_BEFORE(req, 1, 1) && !evhttp_is_connection_keepalive(req->input_headers)) || evhttp_is_request_connection_close(req); EVUTIL_ASSERT(req->flags & EVHTTP_REQ_OWN_CONNECTION); evhttp_request_free(req); if (need_close) { evhttp_connection_free(evcon); return; } /* we have a persistent connection; try to accept another request. */ if (evhttp_associate_new_request_with_connection(evcon) == -1) { evhttp_connection_free(evcon); } } /* * Returns an error page. */ void evhttp_send_error(struct evhttp_request *req, int error, const char *reason) { #define ERR_FORMAT "\n" \ "%d %s\n" \ "\n" \ "

%s

\n" \ "\n" struct evbuffer *buf = evbuffer_new(); if (buf == NULL) { /* if we cannot allocate memory; we just drop the connection */ evhttp_connection_free(req->evcon); return; } if (reason == NULL) { reason = evhttp_response_phrase_internal(error); } evhttp_response_code_(req, error, reason); evbuffer_add_printf(buf, ERR_FORMAT, error, reason, reason); evhttp_send_page_(req, buf); evbuffer_free(buf); #undef ERR_FORMAT } /* Requires that headers and response code are already set up */ static inline void evhttp_send(struct evhttp_request *req, struct evbuffer *databuf) { struct evhttp_connection *evcon = req->evcon; if (evcon == NULL) { evhttp_request_free(req); return; } EVUTIL_ASSERT(TAILQ_FIRST(&evcon->requests) == req); /* we expect no more calls form the user on this request */ req->userdone = 1; /* xxx: not sure if we really should expose the data buffer this way */ if (databuf != NULL) evbuffer_add_buffer(req->output_buffer, databuf); /* Adds headers to the response */ evhttp_make_header(evcon, req); evhttp_write_buffer(evcon, evhttp_send_done, NULL); } void evhttp_send_reply(struct evhttp_request *req, int code, const char *reason, struct evbuffer *databuf) { evhttp_response_code_(req, code, reason); evhttp_send(req, databuf); } void evhttp_send_reply_start(struct evhttp_request *req, int code, const char *reason) { evhttp_response_code_(req, code, reason); if (req->evcon == NULL) return; if (evhttp_find_header(req->output_headers, "Content-Length") == NULL && REQ_VERSION_ATLEAST(req, 1, 1) && evhttp_response_needs_body(req)) { /* * prefer HTTP/1.1 chunked encoding to closing the connection; * note RFC 2616 section 4.4 forbids it with Content-Length: * and it's not necessary then anyway. */ evhttp_add_header(req->output_headers, "Transfer-Encoding", "chunked"); req->chunked = 1; } else { req->chunked = 0; } evhttp_make_header(req->evcon, req); evhttp_write_buffer(req->evcon, NULL, NULL); } void evhttp_send_reply_chunk_with_cb(struct evhttp_request *req, struct evbuffer *databuf, void (*cb)(struct evhttp_connection *, void *), void *arg) { struct evhttp_connection *evcon = req->evcon; struct evbuffer *output; if (evcon == NULL) return; output = bufferevent_get_output(evcon->bufev); if (evbuffer_get_length(databuf) == 0) return; if (!evhttp_response_needs_body(req)) return; if (req->chunked) { evbuffer_add_printf(output, "%x\r\n", (unsigned)evbuffer_get_length(databuf)); } evbuffer_add_buffer(output, databuf); if (req->chunked) { evbuffer_add(output, "\r\n", 2); } evhttp_write_buffer(evcon, cb, arg); } void evhttp_send_reply_chunk(struct evhttp_request *req, struct evbuffer *databuf) { evhttp_send_reply_chunk_with_cb(req, databuf, NULL, NULL); } void evhttp_send_reply_end(struct evhttp_request *req) { struct evhttp_connection *evcon = req->evcon; struct evbuffer *output; if (evcon == NULL) { evhttp_request_free(req); return; } output = bufferevent_get_output(evcon->bufev); /* we expect no more calls form the user on this request */ req->userdone = 1; if (req->chunked) { evbuffer_add(output, "0\r\n\r\n", 5); evhttp_write_buffer(req->evcon, evhttp_send_done, NULL); req->chunked = 0; } else if (evbuffer_get_length(output) == 0) { /* let the connection know that we are done with the request */ evhttp_send_done(evcon, NULL); } else { /* make the callback execute after all data has been written */ evcon->cb = evhttp_send_done; evcon->cb_arg = NULL; } } static const char *informational_phrases[] = { /* 100 */ "Continue", /* 101 */ "Switching Protocols" }; static const char *success_phrases[] = { /* 200 */ "OK", /* 201 */ "Created", /* 202 */ "Accepted", /* 203 */ "Non-Authoritative Information", /* 204 */ "No Content", /* 205 */ "Reset Content", /* 206 */ "Partial Content" }; static const char *redirection_phrases[] = { /* 300 */ "Multiple Choices", /* 301 */ "Moved Permanently", /* 302 */ "Found", /* 303 */ "See Other", /* 304 */ "Not Modified", /* 305 */ "Use Proxy", /* 307 */ "Temporary Redirect" }; static const char *client_error_phrases[] = { /* 400 */ "Bad Request", /* 401 */ "Unauthorized", /* 402 */ "Payment Required", /* 403 */ "Forbidden", /* 404 */ "Not Found", /* 405 */ "Method Not Allowed", /* 406 */ "Not Acceptable", /* 407 */ "Proxy Authentication Required", /* 408 */ "Request Time-out", /* 409 */ "Conflict", /* 410 */ "Gone", /* 411 */ "Length Required", /* 412 */ "Precondition Failed", /* 413 */ "Request Entity Too Large", /* 414 */ "Request-URI Too Large", /* 415 */ "Unsupported Media Type", /* 416 */ "Requested range not satisfiable", /* 417 */ "Expectation Failed" }; static const char *server_error_phrases[] = { /* 500 */ "Internal Server Error", /* 501 */ "Not Implemented", /* 502 */ "Bad Gateway", /* 503 */ "Service Unavailable", /* 504 */ "Gateway Time-out", /* 505 */ "HTTP Version not supported" }; struct response_class { const char *name; size_t num_responses; const char **responses; }; #ifndef MEMBERSOF #define MEMBERSOF(x) (sizeof(x)/sizeof(x[0])) #endif static const struct response_class response_classes[] = { /* 1xx */ { "Informational", MEMBERSOF(informational_phrases), informational_phrases }, /* 2xx */ { "Success", MEMBERSOF(success_phrases), success_phrases }, /* 3xx */ { "Redirection", MEMBERSOF(redirection_phrases), redirection_phrases }, /* 4xx */ { "Client Error", MEMBERSOF(client_error_phrases), client_error_phrases }, /* 5xx */ { "Server Error", MEMBERSOF(server_error_phrases), server_error_phrases } }; static const char * evhttp_response_phrase_internal(int code) { int klass = code / 100 - 1; int subcode = code % 100; /* Unknown class - can't do any better here */ if (klass < 0 || klass >= (int) MEMBERSOF(response_classes)) return "Unknown Status Class"; /* Unknown sub-code, return class name at least */ if (subcode >= (int) response_classes[klass].num_responses) return response_classes[klass].name; return response_classes[klass].responses[subcode]; } void evhttp_response_code_(struct evhttp_request *req, int code, const char *reason) { req->kind = EVHTTP_RESPONSE; req->response_code = code; if (req->response_code_line != NULL) mm_free(req->response_code_line); if (reason == NULL) reason = evhttp_response_phrase_internal(code); req->response_code_line = mm_strdup(reason); if (req->response_code_line == NULL) { event_warn("%s: strdup", __func__); /* XXX what else can we do? */ } } void evhttp_send_page_(struct evhttp_request *req, struct evbuffer *databuf) { if (!req->major || !req->minor) { req->major = 1; req->minor = 1; } if (req->kind != EVHTTP_RESPONSE) evhttp_response_code_(req, 200, "OK"); evhttp_clear_headers(req->output_headers); evhttp_add_header(req->output_headers, "Content-Type", "text/html"); evhttp_add_header(req->output_headers, "Connection", "close"); evhttp_send(req, databuf); } static const char uri_chars[256] = { /* 0 */ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 64 */ 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, /* 128 */ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 192 */ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }; #define CHAR_IS_UNRESERVED(c) \ (uri_chars[(unsigned char)(c)]) /* * Helper functions to encode/decode a string for inclusion in a URI. * The returned string must be freed by the caller. */ char * evhttp_uriencode(const char *uri, ev_ssize_t len, int space_as_plus) { struct evbuffer *buf = evbuffer_new(); const char *p, *end; char *result = NULL; if (!buf) { goto out; } if (len >= 0) { if (uri + len < uri) { goto out; } end = uri + len; } else { size_t slen = strlen(uri); if (slen >= EV_SSIZE_MAX) { /* we don't want to mix signed and unsigned */ goto out; } if (uri + slen < uri) { goto out; } end = uri + slen; } for (p = uri; p < end; p++) { if (CHAR_IS_UNRESERVED(*p)) { evbuffer_add(buf, p, 1); } else if (*p == ' ' && space_as_plus) { evbuffer_add(buf, "+", 1); } else { evbuffer_add_printf(buf, "%%%02X", (unsigned char)(*p)); } } evbuffer_add(buf, "", 1); /* NUL-terminator. */ result = mm_malloc(evbuffer_get_length(buf)); if (result) evbuffer_remove(buf, result, evbuffer_get_length(buf)); out: if (buf) evbuffer_free(buf); return result; } char * evhttp_encode_uri(const char *str) { return evhttp_uriencode(str, -1, 0); } /* * @param decode_plus_ctl: if 1, we decode plus into space. If 0, we don't. * If -1, when true we transform plus to space only after we've seen * a ?. -1 is deprecated. * @return the number of bytes written to 'ret'. */ int evhttp_decode_uri_internal( const char *uri, size_t length, char *ret, int decode_plus_ctl) { char c; int j; int decode_plus = (decode_plus_ctl == 1) ? 1: 0; unsigned i; for (i = j = 0; i < length; i++) { c = uri[i]; if (c == '?') { if (decode_plus_ctl < 0) decode_plus = 1; } else if (c == '+' && decode_plus) { c = ' '; } else if ((i + 2) < length && c == '%' && EVUTIL_ISXDIGIT_(uri[i+1]) && EVUTIL_ISXDIGIT_(uri[i+2])) { char tmp[3]; tmp[0] = uri[i+1]; tmp[1] = uri[i+2]; tmp[2] = '\0'; c = (char)strtol(tmp, NULL, 16); i += 2; } ret[j++] = c; } ret[j] = '\0'; return (j); } /* deprecated */ char * evhttp_decode_uri(const char *uri) { char *ret; if ((ret = mm_malloc(strlen(uri) + 1)) == NULL) { event_warn("%s: malloc(%lu)", __func__, (unsigned long)(strlen(uri) + 1)); return (NULL); } evhttp_decode_uri_internal(uri, strlen(uri), ret, -1 /*always_decode_plus*/); return (ret); } char * evhttp_uridecode(const char *uri, int decode_plus, size_t *size_out) { char *ret; int n; if ((ret = mm_malloc(strlen(uri) + 1)) == NULL) { event_warn("%s: malloc(%lu)", __func__, (unsigned long)(strlen(uri) + 1)); return (NULL); } n = evhttp_decode_uri_internal(uri, strlen(uri), ret, !!decode_plus/*always_decode_plus*/); if (size_out) { EVUTIL_ASSERT(n >= 0); *size_out = (size_t)n; } return (ret); } /* * Helper function to parse out arguments in a query. * The arguments are separated by key and value. */ static int evhttp_parse_query_impl(const char *str, struct evkeyvalq *headers, int is_whole_uri) { char *line=NULL; char *argument; char *p; const char *query_part; int result = -1; struct evhttp_uri *uri=NULL; TAILQ_INIT(headers); if (is_whole_uri) { uri = evhttp_uri_parse(str); if (!uri) goto error; query_part = evhttp_uri_get_query(uri); } else { query_part = str; } /* No arguments - we are done */ if (!query_part || !strlen(query_part)) { result = 0; goto done; } if ((line = mm_strdup(query_part)) == NULL) { event_warn("%s: strdup", __func__); goto error; } p = argument = line; while (p != NULL && *p != '\0') { char *key, *value, *decoded_value; int err; argument = strsep(&p, "&"); value = argument; key = strsep(&value, "="); if (value == NULL || *key == '\0') { goto error; } if ((decoded_value = mm_malloc(strlen(value) + 1)) == NULL) { event_warn("%s: mm_malloc", __func__); goto error; } evhttp_decode_uri_internal(value, strlen(value), decoded_value, 1 /*always_decode_plus*/); event_debug(("Query Param: %s -> %s\n", key, decoded_value)); err = evhttp_add_header_internal(headers, key, decoded_value); mm_free(decoded_value); if (err) goto error; } result = 0; goto done; error: evhttp_clear_headers(headers); done: if (line) mm_free(line); if (uri) evhttp_uri_free(uri); return result; } int evhttp_parse_query(const char *uri, struct evkeyvalq *headers) { return evhttp_parse_query_impl(uri, headers, 1); } int evhttp_parse_query_str(const char *uri, struct evkeyvalq *headers) { return evhttp_parse_query_impl(uri, headers, 0); } static struct evhttp_cb * evhttp_dispatch_callback(struct httpcbq *callbacks, struct evhttp_request *req) { struct evhttp_cb *cb; size_t offset = 0; char *translated; const char *path; /* Test for different URLs */ path = evhttp_uri_get_path(req->uri_elems); offset = strlen(path); if ((translated = mm_malloc(offset + 1)) == NULL) return (NULL); evhttp_decode_uri_internal(path, offset, translated, 0 /* decode_plus */); TAILQ_FOREACH(cb, callbacks, next) { if (!strcmp(cb->what, translated)) { mm_free(translated); return (cb); } } mm_free(translated); return (NULL); } static int prefix_suffix_match(const char *pattern, const char *name, int ignorecase) { char c; while (1) { switch (c = *pattern++) { case '\0': return *name == '\0'; case '*': while (*name != '\0') { if (prefix_suffix_match(pattern, name, ignorecase)) return (1); ++name; } return (0); default: if (c != *name) { if (!ignorecase || EVUTIL_TOLOWER_(c) != EVUTIL_TOLOWER_(*name)) return (0); } ++name; } } /* NOTREACHED */ } /* Search the vhost hierarchy beginning with http for a server alias matching hostname. If a match is found, and outhttp is non-null, outhttp is set to the matching http object and 1 is returned. */ static int evhttp_find_alias(struct evhttp *http, struct evhttp **outhttp, const char *hostname) { struct evhttp_server_alias *alias; struct evhttp *vhost; TAILQ_FOREACH(alias, &http->aliases, next) { /* XXX Do we need to handle IP addresses? */ if (!evutil_ascii_strcasecmp(alias->alias, hostname)) { if (outhttp) *outhttp = http; return 1; } } /* XXX It might be good to avoid recursion here, but I don't see a way to do that w/o a list. */ TAILQ_FOREACH(vhost, &http->virtualhosts, next_vhost) { if (evhttp_find_alias(vhost, outhttp, hostname)) return 1; } return 0; } /* Attempts to find the best http object to handle a request for a hostname. All aliases for the root http object and vhosts are searched for an exact match. Then, the vhost hierarchy is traversed again for a matching pattern. If an alias or vhost is matched, 1 is returned, and outhttp, if non-null, is set with the best matching http object. If there are no matches, the root http object is stored in outhttp and 0 is returned. */ static int evhttp_find_vhost(struct evhttp *http, struct evhttp **outhttp, const char *hostname) { struct evhttp *vhost; struct evhttp *oldhttp; int match_found = 0; if (evhttp_find_alias(http, outhttp, hostname)) return 1; do { oldhttp = http; TAILQ_FOREACH(vhost, &http->virtualhosts, next_vhost) { if (prefix_suffix_match(vhost->vhost_pattern, hostname, 1 /* ignorecase */)) { http = vhost; match_found = 1; break; } } } while (oldhttp != http); if (outhttp) *outhttp = http; return match_found; } static void evhttp_handle_request(struct evhttp_request *req, void *arg) { struct evhttp *http = arg; struct evhttp_cb *cb = NULL; const char *hostname; /* we have a new request on which the user needs to take action */ req->userdone = 0; bufferevent_disable(req->evcon->bufev, EV_READ); if (req->type == 0 || req->uri == NULL) { evhttp_send_error(req, req->response_code, NULL); return; } if ((http->allowed_methods & req->type) == 0) { event_debug(("Rejecting disallowed method %x (allowed: %x)\n", (unsigned)req->type, (unsigned)http->allowed_methods)); evhttp_send_error(req, HTTP_NOTIMPLEMENTED, NULL); return; } /* handle potential virtual hosts */ hostname = evhttp_request_get_host(req); if (hostname != NULL) { evhttp_find_vhost(http, &http, hostname); } if ((cb = evhttp_dispatch_callback(&http->callbacks, req)) != NULL) { (*cb->cb)(req, cb->cbarg); return; } /* Generic call back */ if (http->gencb) { (*http->gencb)(req, http->gencbarg); return; } else { /* We need to send a 404 here */ #define ERR_FORMAT "" \ "404 Not Found" \ "" \ "

Not Found

" \ "

The requested URL %s was not found on this server.

"\ "\n" char *escaped_html; struct evbuffer *buf; if ((escaped_html = evhttp_htmlescape(req->uri)) == NULL) { evhttp_connection_free(req->evcon); return; } if ((buf = evbuffer_new()) == NULL) { mm_free(escaped_html); evhttp_connection_free(req->evcon); return; } evhttp_response_code_(req, HTTP_NOTFOUND, "Not Found"); evbuffer_add_printf(buf, ERR_FORMAT, escaped_html); mm_free(escaped_html); evhttp_send_page_(req, buf); evbuffer_free(buf); #undef ERR_FORMAT } } /* Listener callback when a connection arrives at a server. */ static void accept_socket_cb(struct evconnlistener *listener, evutil_socket_t nfd, struct sockaddr *peer_sa, int peer_socklen, void *arg) { struct evhttp *http = arg; evhttp_get_request(http, nfd, peer_sa, peer_socklen); } int evhttp_bind_socket(struct evhttp *http, const char *address, ev_uint16_t port) { struct evhttp_bound_socket *bound = evhttp_bind_socket_with_handle(http, address, port); if (bound == NULL) return (-1); return (0); } struct evhttp_bound_socket * evhttp_bind_socket_with_handle(struct evhttp *http, const char *address, ev_uint16_t port) { evutil_socket_t fd; struct evhttp_bound_socket *bound; int serrno; if ((fd = bind_socket(address, port, 1 /*reuse*/)) == -1) return (NULL); if (listen(fd, 128) == -1) { serrno = EVUTIL_SOCKET_ERROR(); event_sock_warn(fd, "%s: listen", __func__); evutil_closesocket(fd); EVUTIL_SET_SOCKET_ERROR(serrno); return (NULL); } bound = evhttp_accept_socket_with_handle(http, fd); if (bound != NULL) { event_debug(("Bound to port %d - Awaiting connections ... ", port)); return (bound); } return (NULL); } int evhttp_accept_socket(struct evhttp *http, evutil_socket_t fd) { struct evhttp_bound_socket *bound = evhttp_accept_socket_with_handle(http, fd); if (bound == NULL) return (-1); return (0); } void evhttp_foreach_bound_socket(struct evhttp *http, evhttp_bound_socket_foreach_fn *function, void *argument) { struct evhttp_bound_socket *bound; TAILQ_FOREACH(bound, &http->sockets, next) function(bound, argument); } struct evhttp_bound_socket * evhttp_accept_socket_with_handle(struct evhttp *http, evutil_socket_t fd) { struct evhttp_bound_socket *bound; struct evconnlistener *listener; const int flags = LEV_OPT_REUSEABLE|LEV_OPT_CLOSE_ON_EXEC|LEV_OPT_CLOSE_ON_FREE; listener = evconnlistener_new(http->base, NULL, NULL, flags, 0, /* Backlog is '0' because we already said 'listen' */ fd); if (!listener) return (NULL); bound = evhttp_bind_listener(http, listener); if (!bound) { evconnlistener_free(listener); return (NULL); } return (bound); } struct evhttp_bound_socket * evhttp_bind_listener(struct evhttp *http, struct evconnlistener *listener) { struct evhttp_bound_socket *bound; bound = mm_malloc(sizeof(struct evhttp_bound_socket)); if (bound == NULL) return (NULL); bound->listener = listener; TAILQ_INSERT_TAIL(&http->sockets, bound, next); evconnlistener_set_cb(listener, accept_socket_cb, http); return bound; } evutil_socket_t evhttp_bound_socket_get_fd(struct evhttp_bound_socket *bound) { return evconnlistener_get_fd(bound->listener); } struct evconnlistener * evhttp_bound_socket_get_listener(struct evhttp_bound_socket *bound) { return bound->listener; } void evhttp_del_accept_socket(struct evhttp *http, struct evhttp_bound_socket *bound) { TAILQ_REMOVE(&http->sockets, bound, next); evconnlistener_free(bound->listener); mm_free(bound); } static struct evhttp* evhttp_new_object(void) { struct evhttp *http = NULL; if ((http = mm_calloc(1, sizeof(struct evhttp))) == NULL) { event_warn("%s: calloc", __func__); return (NULL); } evutil_timerclear(&http->timeout); evhttp_set_max_headers_size(http, EV_SIZE_MAX); evhttp_set_max_body_size(http, EV_SIZE_MAX); evhttp_set_default_content_type(http, "text/html; charset=ISO-8859-1"); evhttp_set_allowed_methods(http, EVHTTP_REQ_GET | EVHTTP_REQ_POST | EVHTTP_REQ_HEAD | EVHTTP_REQ_PUT | EVHTTP_REQ_DELETE); TAILQ_INIT(&http->sockets); TAILQ_INIT(&http->callbacks); TAILQ_INIT(&http->connections); TAILQ_INIT(&http->virtualhosts); TAILQ_INIT(&http->aliases); return (http); } struct evhttp * evhttp_new(struct event_base *base) { struct evhttp *http = NULL; http = evhttp_new_object(); if (http == NULL) return (NULL); http->base = base; return (http); } /* * Start a web server on the specified address and port. */ struct evhttp * evhttp_start(const char *address, ev_uint16_t port) { struct evhttp *http = NULL; http = evhttp_new_object(); if (http == NULL) return (NULL); if (evhttp_bind_socket(http, address, port) == -1) { mm_free(http); return (NULL); } return (http); } void evhttp_free(struct evhttp* http) { struct evhttp_cb *http_cb; struct evhttp_connection *evcon; struct evhttp_bound_socket *bound; struct evhttp* vhost; struct evhttp_server_alias *alias; /* Remove the accepting part */ while ((bound = TAILQ_FIRST(&http->sockets)) != NULL) { TAILQ_REMOVE(&http->sockets, bound, next); evconnlistener_free(bound->listener); mm_free(bound); } while ((evcon = TAILQ_FIRST(&http->connections)) != NULL) { /* evhttp_connection_free removes the connection */ evhttp_connection_free(evcon); } while ((http_cb = TAILQ_FIRST(&http->callbacks)) != NULL) { TAILQ_REMOVE(&http->callbacks, http_cb, next); mm_free(http_cb->what); mm_free(http_cb); } while ((vhost = TAILQ_FIRST(&http->virtualhosts)) != NULL) { TAILQ_REMOVE(&http->virtualhosts, vhost, next_vhost); evhttp_free(vhost); } if (http->vhost_pattern != NULL) mm_free(http->vhost_pattern); while ((alias = TAILQ_FIRST(&http->aliases)) != NULL) { TAILQ_REMOVE(&http->aliases, alias, next); mm_free(alias->alias); mm_free(alias); } mm_free(http); } int evhttp_add_virtual_host(struct evhttp* http, const char *pattern, struct evhttp* vhost) { /* a vhost can only be a vhost once and should not have bound sockets */ if (vhost->vhost_pattern != NULL || TAILQ_FIRST(&vhost->sockets) != NULL) return (-1); vhost->vhost_pattern = mm_strdup(pattern); if (vhost->vhost_pattern == NULL) return (-1); TAILQ_INSERT_TAIL(&http->virtualhosts, vhost, next_vhost); return (0); } int evhttp_remove_virtual_host(struct evhttp* http, struct evhttp* vhost) { if (vhost->vhost_pattern == NULL) return (-1); TAILQ_REMOVE(&http->virtualhosts, vhost, next_vhost); mm_free(vhost->vhost_pattern); vhost->vhost_pattern = NULL; return (0); } int evhttp_add_server_alias(struct evhttp *http, const char *alias) { struct evhttp_server_alias *evalias; evalias = mm_calloc(1, sizeof(*evalias)); if (!evalias) return -1; evalias->alias = mm_strdup(alias); if (!evalias->alias) { mm_free(evalias); return -1; } TAILQ_INSERT_TAIL(&http->aliases, evalias, next); return 0; } int evhttp_remove_server_alias(struct evhttp *http, const char *alias) { struct evhttp_server_alias *evalias; TAILQ_FOREACH(evalias, &http->aliases, next) { if (evutil_ascii_strcasecmp(evalias->alias, alias) == 0) { TAILQ_REMOVE(&http->aliases, evalias, next); mm_free(evalias->alias); mm_free(evalias); return 0; } } return -1; } void evhttp_set_timeout(struct evhttp* http, int timeout_in_secs) { if (timeout_in_secs == -1) { evhttp_set_timeout_tv(http, NULL); } else { struct timeval tv; tv.tv_sec = timeout_in_secs; tv.tv_usec = 0; evhttp_set_timeout_tv(http, &tv); } } void evhttp_set_timeout_tv(struct evhttp* http, const struct timeval* tv) { if (tv) { http->timeout = *tv; } else { evutil_timerclear(&http->timeout); } } int evhttp_set_flags(struct evhttp *http, int flags) { int avail_flags = 0; avail_flags |= EVHTTP_SERVER_LINGERING_CLOSE; if (flags & ~avail_flags) return 1; http->flags &= ~avail_flags; http->flags |= flags; return 0; } void evhttp_set_max_headers_size(struct evhttp* http, ev_ssize_t max_headers_size) { if (max_headers_size < 0) http->default_max_headers_size = EV_SIZE_MAX; else http->default_max_headers_size = max_headers_size; } void evhttp_set_max_body_size(struct evhttp* http, ev_ssize_t max_body_size) { if (max_body_size < 0) http->default_max_body_size = EV_UINT64_MAX; else http->default_max_body_size = max_body_size; } void evhttp_set_default_content_type(struct evhttp *http, const char *content_type) { http->default_content_type = content_type; } void evhttp_set_allowed_methods(struct evhttp* http, ev_uint16_t methods) { http->allowed_methods = methods; } int evhttp_set_cb(struct evhttp *http, const char *uri, void (*cb)(struct evhttp_request *, void *), void *cbarg) { struct evhttp_cb *http_cb; TAILQ_FOREACH(http_cb, &http->callbacks, next) { if (strcmp(http_cb->what, uri) == 0) return (-1); } if ((http_cb = mm_calloc(1, sizeof(struct evhttp_cb))) == NULL) { event_warn("%s: calloc", __func__); return (-2); } http_cb->what = mm_strdup(uri); if (http_cb->what == NULL) { event_warn("%s: strdup", __func__); mm_free(http_cb); return (-3); } http_cb->cb = cb; http_cb->cbarg = cbarg; TAILQ_INSERT_TAIL(&http->callbacks, http_cb, next); return (0); } int evhttp_del_cb(struct evhttp *http, const char *uri) { struct evhttp_cb *http_cb; TAILQ_FOREACH(http_cb, &http->callbacks, next) { if (strcmp(http_cb->what, uri) == 0) break; } if (http_cb == NULL) return (-1); TAILQ_REMOVE(&http->callbacks, http_cb, next); mm_free(http_cb->what); mm_free(http_cb); return (0); } void evhttp_set_gencb(struct evhttp *http, void (*cb)(struct evhttp_request *, void *), void *cbarg) { http->gencb = cb; http->gencbarg = cbarg; } void evhttp_set_bevcb(struct evhttp *http, struct bufferevent* (*cb)(struct event_base *, void *), void *cbarg) { http->bevcb = cb; http->bevcbarg = cbarg; } /* * Request related functions */ struct evhttp_request * evhttp_request_new(void (*cb)(struct evhttp_request *, void *), void *arg) { struct evhttp_request *req = NULL; /* Allocate request structure */ if ((req = mm_calloc(1, sizeof(struct evhttp_request))) == NULL) { event_warn("%s: calloc", __func__); goto error; } req->headers_size = 0; req->body_size = 0; req->kind = EVHTTP_RESPONSE; req->input_headers = mm_calloc(1, sizeof(struct evkeyvalq)); if (req->input_headers == NULL) { event_warn("%s: calloc", __func__); goto error; } TAILQ_INIT(req->input_headers); req->output_headers = mm_calloc(1, sizeof(struct evkeyvalq)); if (req->output_headers == NULL) { event_warn("%s: calloc", __func__); goto error; } TAILQ_INIT(req->output_headers); if ((req->input_buffer = evbuffer_new()) == NULL) { event_warn("%s: evbuffer_new", __func__); goto error; } if ((req->output_buffer = evbuffer_new()) == NULL) { event_warn("%s: evbuffer_new", __func__); goto error; } req->cb = cb; req->cb_arg = arg; return (req); error: if (req != NULL) evhttp_request_free(req); return (NULL); } void evhttp_request_free(struct evhttp_request *req) { if ((req->flags & EVHTTP_REQ_DEFER_FREE) != 0) { req->flags |= EVHTTP_REQ_NEEDS_FREE; return; } if (req->remote_host != NULL) mm_free(req->remote_host); if (req->uri != NULL) mm_free(req->uri); if (req->uri_elems != NULL) evhttp_uri_free(req->uri_elems); if (req->response_code_line != NULL) mm_free(req->response_code_line); if (req->host_cache != NULL) mm_free(req->host_cache); evhttp_clear_headers(req->input_headers); mm_free(req->input_headers); evhttp_clear_headers(req->output_headers); mm_free(req->output_headers); if (req->input_buffer != NULL) evbuffer_free(req->input_buffer); if (req->output_buffer != NULL) evbuffer_free(req->output_buffer); mm_free(req); } void evhttp_request_own(struct evhttp_request *req) { req->flags |= EVHTTP_USER_OWNED; } int evhttp_request_is_owned(struct evhttp_request *req) { return (req->flags & EVHTTP_USER_OWNED) != 0; } struct evhttp_connection * evhttp_request_get_connection(struct evhttp_request *req) { return req->evcon; } struct event_base * evhttp_connection_get_base(struct evhttp_connection *conn) { return conn->base; } void evhttp_request_set_chunked_cb(struct evhttp_request *req, void (*cb)(struct evhttp_request *, void *)) { req->chunk_cb = cb; } void evhttp_request_set_header_cb(struct evhttp_request *req, int (*cb)(struct evhttp_request *, void *)) { req->header_cb = cb; } void evhttp_request_set_error_cb(struct evhttp_request *req, void (*cb)(enum evhttp_request_error, void *)) { req->error_cb = cb; } void evhttp_request_set_on_complete_cb(struct evhttp_request *req, void (*cb)(struct evhttp_request *, void *), void *cb_arg) { req->on_complete_cb = cb; req->on_complete_cb_arg = cb_arg; } /* * Allows for inspection of the request URI */ const char * evhttp_request_get_uri(const struct evhttp_request *req) { if (req->uri == NULL) event_debug(("%s: request %p has no uri\n", __func__, req)); return (req->uri); } const struct evhttp_uri * evhttp_request_get_evhttp_uri(const struct evhttp_request *req) { if (req->uri_elems == NULL) event_debug(("%s: request %p has no uri elems\n", __func__, req)); return (req->uri_elems); } const char * evhttp_request_get_host(struct evhttp_request *req) { const char *host = NULL; if (req->host_cache) return req->host_cache; if (req->uri_elems) host = evhttp_uri_get_host(req->uri_elems); if (!host && req->input_headers) { const char *p; size_t len; host = evhttp_find_header(req->input_headers, "Host"); /* The Host: header may include a port. Remove it here to be consistent with uri_elems case above. */ if (host) { p = host + strlen(host) - 1; while (p > host && EVUTIL_ISDIGIT_(*p)) --p; if (p > host && *p == ':') { len = p - host; req->host_cache = mm_malloc(len + 1); if (!req->host_cache) { event_warn("%s: malloc", __func__); return NULL; } memcpy(req->host_cache, host, len); req->host_cache[len] = '\0'; host = req->host_cache; } } } return host; } enum evhttp_cmd_type evhttp_request_get_command(const struct evhttp_request *req) { return (req->type); } int evhttp_request_get_response_code(const struct evhttp_request *req) { return req->response_code; } const char * evhttp_request_get_response_code_line(const struct evhttp_request *req) { return req->response_code_line; } /** Returns the input headers */ struct evkeyvalq *evhttp_request_get_input_headers(struct evhttp_request *req) { return (req->input_headers); } /** Returns the output headers */ struct evkeyvalq *evhttp_request_get_output_headers(struct evhttp_request *req) { return (req->output_headers); } /** Returns the input buffer */ struct evbuffer *evhttp_request_get_input_buffer(struct evhttp_request *req) { return (req->input_buffer); } /** Returns the output buffer */ struct evbuffer *evhttp_request_get_output_buffer(struct evhttp_request *req) { return (req->output_buffer); } /* * Takes a file descriptor to read a request from. * The callback is executed once the whole request has been read. */ static struct evhttp_connection* evhttp_get_request_connection( struct evhttp* http, evutil_socket_t fd, struct sockaddr *sa, ev_socklen_t salen) { struct evhttp_connection *evcon; char *hostname = NULL, *portname = NULL; struct bufferevent* bev = NULL; #ifdef EVENT__HAVE_STRUCT_SOCKADDR_UN if (sa->sa_family == AF_UNIX) { struct sockaddr_un *sa_un = (struct sockaddr_un *)sa; sa_un->sun_path[0] = '\0'; } #endif name_from_addr(sa, salen, &hostname, &portname); if (hostname == NULL || portname == NULL) { if (hostname) mm_free(hostname); if (portname) mm_free(portname); return (NULL); } event_debug(("%s: new request from %s:%s on "EV_SOCK_FMT"\n", __func__, hostname, portname, EV_SOCK_ARG(fd))); /* we need a connection object to put the http request on */ if (http->bevcb != NULL) { bev = (*http->bevcb)(http->base, http->bevcbarg); } evcon = evhttp_connection_base_bufferevent_new( http->base, NULL, bev, hostname, atoi(portname)); mm_free(hostname); mm_free(portname); if (evcon == NULL) return (NULL); evcon->max_headers_size = http->default_max_headers_size; evcon->max_body_size = http->default_max_body_size; if (http->flags & EVHTTP_SERVER_LINGERING_CLOSE) evcon->flags |= EVHTTP_CON_LINGERING_CLOSE; evcon->flags |= EVHTTP_CON_INCOMING; evcon->state = EVCON_READING_FIRSTLINE; evcon->fd = fd; if (bufferevent_setfd(evcon->bufev, fd)) goto err; if (bufferevent_enable(evcon->bufev, EV_READ)) goto err; if (bufferevent_disable(evcon->bufev, EV_WRITE)) goto err; bufferevent_socket_set_conn_address_(evcon->bufev, sa, salen); return (evcon); err: evhttp_connection_free(evcon); return (NULL); } static int evhttp_associate_new_request_with_connection(struct evhttp_connection *evcon) { struct evhttp *http = evcon->http_server; struct evhttp_request *req; if ((req = evhttp_request_new(evhttp_handle_request, http)) == NULL) return (-1); if ((req->remote_host = mm_strdup(evcon->address)) == NULL) { event_warn("%s: strdup", __func__); evhttp_request_free(req); return (-1); } req->remote_port = evcon->port; req->evcon = evcon; /* the request ends up owning the connection */ req->flags |= EVHTTP_REQ_OWN_CONNECTION; /* We did not present the request to the user user yet, so treat it as * if the user was done with the request. This allows us to free the * request on a persistent connection if the client drops it without * sending a request. */ req->userdone = 1; TAILQ_INSERT_TAIL(&evcon->requests, req, next); req->kind = EVHTTP_REQUEST; evhttp_start_read_(evcon); return (0); } static void evhttp_get_request(struct evhttp *http, evutil_socket_t fd, struct sockaddr *sa, ev_socklen_t salen) { struct evhttp_connection *evcon; evcon = evhttp_get_request_connection(http, fd, sa, salen); if (evcon == NULL) { event_sock_warn(fd, "%s: cannot get connection on "EV_SOCK_FMT, __func__, EV_SOCK_ARG(fd)); evutil_closesocket(fd); return; } /* the timeout can be used by the server to close idle connections */ if (evutil_timerisset(&http->timeout)) evhttp_connection_set_timeout_tv(evcon, &http->timeout); /* * if we want to accept more than one request on a connection, * we need to know which http server it belongs to. */ evcon->http_server = http; TAILQ_INSERT_TAIL(&http->connections, evcon, next); if (evhttp_associate_new_request_with_connection(evcon) == -1) evhttp_connection_free(evcon); } /* * Network helper functions that we do not want to export to the rest of * the world. */ static void name_from_addr(struct sockaddr *sa, ev_socklen_t salen, char **phost, char **pport) { char ntop[NI_MAXHOST]; char strport[NI_MAXSERV]; int ni_result; #ifdef EVENT__HAVE_GETNAMEINFO ni_result = getnameinfo(sa, salen, ntop, sizeof(ntop), strport, sizeof(strport), NI_NUMERICHOST|NI_NUMERICSERV); if (ni_result != 0) { #ifdef EAI_SYSTEM /* Windows doesn't have an EAI_SYSTEM. */ if (ni_result == EAI_SYSTEM) event_err(1, "getnameinfo failed"); else #endif event_errx(1, "getnameinfo failed: %s", gai_strerror(ni_result)); return; } #else ni_result = fake_getnameinfo(sa, salen, ntop, sizeof(ntop), strport, sizeof(strport), NI_NUMERICHOST|NI_NUMERICSERV); if (ni_result != 0) return; #endif *phost = mm_strdup(ntop); *pport = mm_strdup(strport); } /* Create a non-blocking socket and bind it */ static evutil_socket_t create_bind_socket_nonblock(struct evutil_addrinfo *ai, int reuse) { evutil_socket_t fd; int on = 1, r; int serrno; /* Create listen socket */ fd = evutil_socket_(ai ? ai->ai_family : AF_INET, SOCK_STREAM|EVUTIL_SOCK_NONBLOCK|EVUTIL_SOCK_CLOEXEC, 0); if (fd == -1) { event_sock_warn(-1, "socket"); return (-1); } if (setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, (void *)&on, sizeof(on))<0) goto out; if (reuse) { if (evutil_make_listen_socket_reuseable(fd) < 0) goto out; } if (ai != NULL) { r = bind(fd, ai->ai_addr, (ev_socklen_t)ai->ai_addrlen); if (r == -1) goto out; } return (fd); out: serrno = EVUTIL_SOCKET_ERROR(); evutil_closesocket(fd); EVUTIL_SET_SOCKET_ERROR(serrno); return (-1); } static struct evutil_addrinfo * make_addrinfo(const char *address, ev_uint16_t port) { struct evutil_addrinfo *ai = NULL; struct evutil_addrinfo hints; char strport[NI_MAXSERV]; int ai_result; memset(&hints, 0, sizeof(hints)); hints.ai_family = AF_UNSPEC; hints.ai_socktype = SOCK_STREAM; /* turn NULL hostname into INADDR_ANY, and skip looking up any address * types we don't have an interface to connect to. */ hints.ai_flags = EVUTIL_AI_PASSIVE|EVUTIL_AI_ADDRCONFIG; evutil_snprintf(strport, sizeof(strport), "%d", port); if ((ai_result = evutil_getaddrinfo(address, strport, &hints, &ai)) != 0) { if (ai_result == EVUTIL_EAI_SYSTEM) event_warn("getaddrinfo"); else event_warnx("getaddrinfo: %s", evutil_gai_strerror(ai_result)); return (NULL); } return (ai); } static evutil_socket_t bind_socket(const char *address, ev_uint16_t port, int reuse) { evutil_socket_t fd; struct evutil_addrinfo *aitop = NULL; /* just create an unbound socket */ if (address == NULL && port == 0) return create_bind_socket_nonblock(NULL, 0); aitop = make_addrinfo(address, port); if (aitop == NULL) return (-1); fd = create_bind_socket_nonblock(aitop, reuse); evutil_freeaddrinfo(aitop); return (fd); } struct evhttp_uri { unsigned flags; char *scheme; /* scheme; e.g http, ftp etc */ char *userinfo; /* userinfo (typically username:pass), or NULL */ char *host; /* hostname, IP address, or NULL */ int port; /* port, or zero */ char *path; /* path, or "". */ char *query; /* query, or NULL */ char *fragment; /* fragment or NULL */ }; struct evhttp_uri * evhttp_uri_new(void) { struct evhttp_uri *uri = mm_calloc(sizeof(struct evhttp_uri), 1); if (uri) uri->port = -1; return uri; } void evhttp_uri_set_flags(struct evhttp_uri *uri, unsigned flags) { uri->flags = flags; } /* Return true if the string starting at s and ending immediately before eos * is a valid URI scheme according to RFC3986 */ static int scheme_ok(const char *s, const char *eos) { /* scheme = ALPHA *( ALPHA / DIGIT / "+" / "-" / "." ) */ EVUTIL_ASSERT(eos >= s); if (s == eos) return 0; if (!EVUTIL_ISALPHA_(*s)) return 0; while (++s < eos) { if (! EVUTIL_ISALNUM_(*s) && *s != '+' && *s != '-' && *s != '.') return 0; } return 1; } #define SUBDELIMS "!$&'()*+,;=" /* Return true iff [s..eos) is a valid userinfo */ static int userinfo_ok(const char *s, const char *eos) { while (s < eos) { if (CHAR_IS_UNRESERVED(*s) || strchr(SUBDELIMS, *s) || *s == ':') ++s; else if (*s == '%' && s+2 < eos && EVUTIL_ISXDIGIT_(s[1]) && EVUTIL_ISXDIGIT_(s[2])) s += 3; else return 0; } return 1; } static int regname_ok(const char *s, const char *eos) { while (s && s 65535) return -1; ++s; } return portnum; } /* returns 0 for bad, 1 for ipv6, 2 for IPvFuture */ static int bracket_addr_ok(const char *s, const char *eos) { if (s + 3 > eos || *s != '[' || *(eos-1) != ']') return 0; if (s[1] == 'v') { /* IPvFuture, or junk. "v" 1*HEXDIG "." 1*( unreserved / sub-delims / ":" ) */ s += 2; /* skip [v */ --eos; if (!EVUTIL_ISXDIGIT_(*s)) /*require at least one*/ return 0; while (s < eos && *s != '.') { if (EVUTIL_ISXDIGIT_(*s)) ++s; else return 0; } if (*s != '.') return 0; ++s; while (s < eos) { if (CHAR_IS_UNRESERVED(*s) || strchr(SUBDELIMS, *s) || *s == ':') ++s; else return 0; } return 2; } else { /* IPv6, or junk */ char buf[64]; ev_ssize_t n_chars = eos-s-2; struct in6_addr in6; if (n_chars >= 64) /* way too long */ return 0; memcpy(buf, s+1, n_chars); buf[n_chars]='\0'; return (evutil_inet_pton(AF_INET6,buf,&in6)==1) ? 1 : 0; } } static int parse_authority(struct evhttp_uri *uri, char *s, char *eos) { char *cp, *port; EVUTIL_ASSERT(eos); if (eos == s) { uri->host = mm_strdup(""); if (uri->host == NULL) { event_warn("%s: strdup", __func__); return -1; } return 0; } /* Optionally, we start with "userinfo@" */ cp = strchr(s, '@'); if (cp && cp < eos) { if (! userinfo_ok(s,cp)) return -1; *cp++ = '\0'; uri->userinfo = mm_strdup(s); if (uri->userinfo == NULL) { event_warn("%s: strdup", __func__); return -1; } } else { cp = s; } /* Optionally, we end with ":port" */ for (port=eos-1; port >= cp && EVUTIL_ISDIGIT_(*port); --port) ; if (port >= cp && *port == ':') { if (port+1 == eos) /* Leave port unspecified; the RFC allows a * nil port */ uri->port = -1; else if ((uri->port = parse_port(port+1, eos))<0) return -1; eos = port; } /* Now, cp..eos holds the "host" port, which can be an IPv4Address, * an IP-Literal, or a reg-name */ EVUTIL_ASSERT(eos >= cp); if (*cp == '[' && eos >= cp+2 && *(eos-1) == ']') { /* IPv6address, IP-Literal, or junk. */ if (! bracket_addr_ok(cp, eos)) return -1; } else { /* Make sure the host part is ok. */ if (! regname_ok(cp,eos)) /* Match IPv4Address or reg-name */ return -1; } uri->host = mm_malloc(eos-cp+1); if (uri->host == NULL) { event_warn("%s: malloc", __func__); return -1; } memcpy(uri->host, cp, eos-cp); uri->host[eos-cp] = '\0'; return 0; } static char * end_of_authority(char *cp) { while (*cp) { if (*cp == '?' || *cp == '#' || *cp == '/') return cp; ++cp; } return cp; } enum uri_part { PART_PATH, PART_QUERY, PART_FRAGMENT }; /* Return the character after the longest prefix of 'cp' that matches... * *pchar / "/" if allow_qchars is false, or * *(pchar / "/" / "?") if allow_qchars is true. */ static char * end_of_path(char *cp, enum uri_part part, unsigned flags) { if (flags & EVHTTP_URI_NONCONFORMANT) { /* If NONCONFORMANT: * Path is everything up to a # or ? or nul. * Query is everything up a # or nul * Fragment is everything up to a nul. */ switch (part) { case PART_PATH: while (*cp && *cp != '#' && *cp != '?') ++cp; break; case PART_QUERY: while (*cp && *cp != '#') ++cp; break; case PART_FRAGMENT: cp += strlen(cp); break; }; return cp; } while (*cp) { if (CHAR_IS_UNRESERVED(*cp) || strchr(SUBDELIMS, *cp) || *cp == ':' || *cp == '@' || *cp == '/') ++cp; else if (*cp == '%' && EVUTIL_ISXDIGIT_(cp[1]) && EVUTIL_ISXDIGIT_(cp[2])) cp += 3; else if (*cp == '?' && part != PART_PATH) ++cp; else return cp; } return cp; } static int path_matches_noscheme(const char *cp) { while (*cp) { if (*cp == ':') return 0; else if (*cp == '/') return 1; ++cp; } return 1; } struct evhttp_uri * evhttp_uri_parse(const char *source_uri) { return evhttp_uri_parse_with_flags(source_uri, 0); } struct evhttp_uri * evhttp_uri_parse_with_flags(const char *source_uri, unsigned flags) { char *readbuf = NULL, *readp = NULL, *token = NULL, *query = NULL; char *path = NULL, *fragment = NULL; int got_authority = 0; struct evhttp_uri *uri = mm_calloc(1, sizeof(struct evhttp_uri)); if (uri == NULL) { event_warn("%s: calloc", __func__); goto err; } uri->port = -1; uri->flags = flags; readbuf = mm_strdup(source_uri); if (readbuf == NULL) { event_warn("%s: strdup", __func__); goto err; } readp = readbuf; token = NULL; /* We try to follow RFC3986 here as much as we can, and match the productions URI = scheme ":" hier-part [ "?" query ] [ "#" fragment ] relative-ref = relative-part [ "?" query ] [ "#" fragment ] */ /* 1. scheme: */ token = strchr(readp, ':'); if (token && scheme_ok(readp,token)) { *token = '\0'; uri->scheme = mm_strdup(readp); if (uri->scheme == NULL) { event_warn("%s: strdup", __func__); goto err; } readp = token+1; /* eat : */ } /* 2. Optionally, "//" then an 'authority' part. */ if (readp[0]=='/' && readp[1] == '/') { char *authority; readp += 2; authority = readp; path = end_of_authority(readp); if (parse_authority(uri, authority, path) < 0) goto err; readp = path; got_authority = 1; } /* 3. Query: path-abempty, path-absolute, path-rootless, or path-empty */ path = readp; readp = end_of_path(path, PART_PATH, flags); /* Query */ if (*readp == '?') { *readp = '\0'; ++readp; query = readp; readp = end_of_path(readp, PART_QUERY, flags); } /* fragment */ if (*readp == '#') { *readp = '\0'; ++readp; fragment = readp; readp = end_of_path(readp, PART_FRAGMENT, flags); } if (*readp != '\0') { goto err; } /* These next two cases may be unreachable; I'm leaving them * in to be defensive. */ /* If you didn't get an authority, the path can't begin with "//" */ if (!got_authority && path[0]=='/' && path[1]=='/') goto err; /* If you did get an authority, the path must begin with "/" or be * empty. */ if (got_authority && path[0] != '/' && path[0] != '\0') goto err; /* (End of maybe-unreachable cases) */ /* If there was no scheme, the first part of the path (if any) must * have no colon in it. */ if (! uri->scheme && !path_matches_noscheme(path)) goto err; EVUTIL_ASSERT(path); uri->path = mm_strdup(path); if (uri->path == NULL) { event_warn("%s: strdup", __func__); goto err; } if (query) { uri->query = mm_strdup(query); if (uri->query == NULL) { event_warn("%s: strdup", __func__); goto err; } } if (fragment) { uri->fragment = mm_strdup(fragment); if (uri->fragment == NULL) { event_warn("%s: strdup", __func__); goto err; } } mm_free(readbuf); return uri; err: if (uri) evhttp_uri_free(uri); if (readbuf) mm_free(readbuf); return NULL; } static struct evhttp_uri * evhttp_uri_parse_authority(char *source_uri) { struct evhttp_uri *uri = mm_calloc(1, sizeof(struct evhttp_uri)); char *end; if (uri == NULL) { event_warn("%s: calloc", __func__); goto err; } uri->port = -1; uri->flags = 0; end = end_of_authority(source_uri); if (parse_authority(uri, source_uri, end) < 0) goto err; uri->path = mm_strdup(""); if (uri->path == NULL) { event_warn("%s: strdup", __func__); goto err; } return uri; err: if (uri) evhttp_uri_free(uri); return NULL; } void evhttp_uri_free(struct evhttp_uri *uri) { #define URI_FREE_STR_(f) \ if (uri->f) { \ mm_free(uri->f); \ } URI_FREE_STR_(scheme); URI_FREE_STR_(userinfo); URI_FREE_STR_(host); URI_FREE_STR_(path); URI_FREE_STR_(query); URI_FREE_STR_(fragment); mm_free(uri); #undef URI_FREE_STR_ } char * evhttp_uri_join(struct evhttp_uri *uri, char *buf, size_t limit) { struct evbuffer *tmp = 0; size_t joined_size = 0; char *output = NULL; #define URI_ADD_(f) evbuffer_add(tmp, uri->f, strlen(uri->f)) if (!uri || !buf || !limit) return NULL; tmp = evbuffer_new(); if (!tmp) return NULL; if (uri->scheme) { URI_ADD_(scheme); evbuffer_add(tmp, ":", 1); } if (uri->host) { evbuffer_add(tmp, "//", 2); if (uri->userinfo) evbuffer_add_printf(tmp,"%s@", uri->userinfo); URI_ADD_(host); if (uri->port >= 0) evbuffer_add_printf(tmp,":%d", uri->port); if (uri->path && uri->path[0] != '/' && uri->path[0] != '\0') goto err; } if (uri->path) URI_ADD_(path); if (uri->query) { evbuffer_add(tmp, "?", 1); URI_ADD_(query); } if (uri->fragment) { evbuffer_add(tmp, "#", 1); URI_ADD_(fragment); } evbuffer_add(tmp, "\0", 1); /* NUL */ joined_size = evbuffer_get_length(tmp); if (joined_size > limit) { /* It doesn't fit. */ evbuffer_free(tmp); return NULL; } evbuffer_remove(tmp, buf, joined_size); output = buf; err: evbuffer_free(tmp); return output; #undef URI_ADD_ } const char * evhttp_uri_get_scheme(const struct evhttp_uri *uri) { return uri->scheme; } const char * evhttp_uri_get_userinfo(const struct evhttp_uri *uri) { return uri->userinfo; } const char * evhttp_uri_get_host(const struct evhttp_uri *uri) { return uri->host; } int evhttp_uri_get_port(const struct evhttp_uri *uri) { return uri->port; } const char * evhttp_uri_get_path(const struct evhttp_uri *uri) { return uri->path; } const char * evhttp_uri_get_query(const struct evhttp_uri *uri) { return uri->query; } const char * evhttp_uri_get_fragment(const struct evhttp_uri *uri) { return uri->fragment; } #define URI_SET_STR_(f) do { \ if (uri->f) \ mm_free(uri->f); \ if (f) { \ if ((uri->f = mm_strdup(f)) == NULL) { \ event_warn("%s: strdup()", __func__); \ return -1; \ } \ } else { \ uri->f = NULL; \ } \ } while(0) int evhttp_uri_set_scheme(struct evhttp_uri *uri, const char *scheme) { if (scheme && !scheme_ok(scheme, scheme+strlen(scheme))) return -1; URI_SET_STR_(scheme); return 0; } int evhttp_uri_set_userinfo(struct evhttp_uri *uri, const char *userinfo) { if (userinfo && !userinfo_ok(userinfo, userinfo+strlen(userinfo))) return -1; URI_SET_STR_(userinfo); return 0; } int evhttp_uri_set_host(struct evhttp_uri *uri, const char *host) { if (host) { if (host[0] == '[') { if (! bracket_addr_ok(host, host+strlen(host))) return -1; } else { if (! regname_ok(host, host+strlen(host))) return -1; } } URI_SET_STR_(host); return 0; } int evhttp_uri_set_port(struct evhttp_uri *uri, int port) { if (port < -1) return -1; uri->port = port; return 0; } #define end_of_cpath(cp,p,f) \ ((const char*)(end_of_path(((char*)(cp)), (p), (f)))) int evhttp_uri_set_path(struct evhttp_uri *uri, const char *path) { if (path && end_of_cpath(path, PART_PATH, uri->flags) != path+strlen(path)) return -1; URI_SET_STR_(path); return 0; } int evhttp_uri_set_query(struct evhttp_uri *uri, const char *query) { if (query && end_of_cpath(query, PART_QUERY, uri->flags) != query+strlen(query)) return -1; URI_SET_STR_(query); return 0; } int evhttp_uri_set_fragment(struct evhttp_uri *uri, const char *fragment) { if (fragment && end_of_cpath(fragment, PART_FRAGMENT, uri->flags) != fragment+strlen(fragment)) return -1; URI_SET_STR_(fragment); return 0; } /* $OpenBSD: kqueue.c,v 1.5 2002/07/10 14:41:31 art Exp $ */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_KQUEUE #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #include #include #include #include #include #ifdef EVENT__HAVE_INTTYPES_H #include #endif /* Some platforms apparently define the udata field of struct kevent as * intptr_t, whereas others define it as void*. There doesn't seem to be an * easy way to tell them apart via autoconf, so we need to use OS macros. */ #if defined(__NetBSD__) #define PTR_TO_UDATA(x) ((typeof(((struct kevent *)0)->udata))(x)) #define INT_TO_UDATA(x) ((typeof(((struct kevent *)0)->udata))(intptr_t)(x)) #elif defined(EVENT__HAVE_INTTYPES_H) && !defined(__OpenBSD__) && !defined(__FreeBSD__) && !defined(__darwin__) && !defined(__APPLE__) && !defined(__CloudABI__) #define PTR_TO_UDATA(x) ((intptr_t)(x)) #define INT_TO_UDATA(x) ((intptr_t)(x)) #else #define PTR_TO_UDATA(x) (x) #define INT_TO_UDATA(x) ((void*)(x)) #endif #include "event-internal.h" #include "log-internal.h" #include "evmap-internal.h" #include "event2/thread.h" #include "event2/util.h" #include "evthread-internal.h" #include "changelist-internal.h" #include "kqueue-internal.h" #define NEVENT 64 struct kqop { struct kevent *changes; int changes_size; struct kevent *events; int events_size; int kq; int notify_event_added; pid_t pid; }; static void kqop_free(struct kqop *kqop); static void *kq_init(struct event_base *); static int kq_sig_add(struct event_base *, int, short, short, void *); static int kq_sig_del(struct event_base *, int, short, short, void *); static int kq_dispatch(struct event_base *, struct timeval *); static void kq_dealloc(struct event_base *); const struct eventop kqops = { "kqueue", kq_init, event_changelist_add_, event_changelist_del_, kq_dispatch, kq_dealloc, 1 /* need reinit */, EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_FDS, EVENT_CHANGELIST_FDINFO_SIZE }; static const struct eventop kqsigops = { "kqueue_signal", NULL, kq_sig_add, kq_sig_del, NULL, NULL, 1 /* need reinit */, 0, 0 }; static void * kq_init(struct event_base *base) { int kq = -1; struct kqop *kqueueop = NULL; if (!(kqueueop = mm_calloc(1, sizeof(struct kqop)))) return (NULL); /* Initialize the kernel queue */ if ((kq = kqueue()) == -1) { event_warn("kqueue"); goto err; } kqueueop->kq = kq; kqueueop->pid = getpid(); /* Initialize fields */ kqueueop->changes = mm_calloc(NEVENT, sizeof(struct kevent)); if (kqueueop->changes == NULL) goto err; kqueueop->events = mm_calloc(NEVENT, sizeof(struct kevent)); if (kqueueop->events == NULL) goto err; kqueueop->events_size = kqueueop->changes_size = NEVENT; /* Check for Mac OS X kqueue bug. */ memset(&kqueueop->changes[0], 0, sizeof kqueueop->changes[0]); kqueueop->changes[0].ident = -1; kqueueop->changes[0].filter = EVFILT_READ; kqueueop->changes[0].flags = EV_ADD; /* * If kqueue works, then kevent will succeed, and it will * stick an error in events[0]. If kqueue is broken, then * kevent will fail. */ if (kevent(kq, kqueueop->changes, 1, kqueueop->events, NEVENT, NULL) != 1 || (int)kqueueop->events[0].ident != -1 || !(kqueueop->events[0].flags & EV_ERROR)) { event_warn("%s: detected broken kqueue; not using.", __func__); goto err; } base->evsigsel = &kqsigops; return (kqueueop); err: if (kqueueop) kqop_free(kqueueop); return (NULL); } #define ADD_UDATA 0x30303 static void kq_setup_kevent(struct kevent *out, evutil_socket_t fd, int filter, short change) { memset(out, 0, sizeof(struct kevent)); out->ident = fd; out->filter = filter; if (change & EV_CHANGE_ADD) { out->flags = EV_ADD; /* We set a magic number here so that we can tell 'add' * errors from 'del' errors. */ out->udata = INT_TO_UDATA(ADD_UDATA); if (change & EV_ET) out->flags |= EV_CLEAR; #ifdef NOTE_EOF /* Make it behave like select() and poll() */ if (filter == EVFILT_READ) out->fflags = NOTE_EOF; #endif } else { EVUTIL_ASSERT(change & EV_CHANGE_DEL); out->flags = EV_DELETE; } } static int kq_build_changes_list(const struct event_changelist *changelist, struct kqop *kqop) { int i; int n_changes = 0; for (i = 0; i < changelist->n_changes; ++i) { struct event_change *in_ch = &changelist->changes[i]; struct kevent *out_ch; if (n_changes >= kqop->changes_size - 1) { int newsize; struct kevent *newchanges; if (kqop->changes_size > INT_MAX / 2 || (size_t)kqop->changes_size * 2 > EV_SIZE_MAX / sizeof(struct kevent)) { event_warnx("%s: int overflow", __func__); return (-1); } newsize = kqop->changes_size * 2; newchanges = mm_realloc(kqop->changes, newsize * sizeof(struct kevent)); if (newchanges == NULL) { event_warn("%s: realloc", __func__); return (-1); } kqop->changes = newchanges; kqop->changes_size = newsize; } if (in_ch->read_change) { out_ch = &kqop->changes[n_changes++]; kq_setup_kevent(out_ch, in_ch->fd, EVFILT_READ, in_ch->read_change); } if (in_ch->write_change) { out_ch = &kqop->changes[n_changes++]; kq_setup_kevent(out_ch, in_ch->fd, EVFILT_WRITE, in_ch->write_change); } } return n_changes; } static int kq_grow_events(struct kqop *kqop, size_t new_size) { struct kevent *newresult; newresult = mm_realloc(kqop->events, new_size * sizeof(struct kevent)); if (newresult) { kqop->events = newresult; kqop->events_size = new_size; return 0; } else { return -1; } } static int kq_dispatch(struct event_base *base, struct timeval *tv) { struct kqop *kqop = base->evbase; struct kevent *events = kqop->events; struct kevent *changes; struct timespec ts, *ts_p = NULL; int i, n_changes, res; if (tv != NULL) { ts.tv_sec = tv->tv_sec; ts.tv_nsec = tv->tv_usec * 1000; ts_p = &ts; } /* Build "changes" from "base->changes" */ EVUTIL_ASSERT(kqop->changes); n_changes = kq_build_changes_list(&base->changelist, kqop); if (n_changes < 0) return -1; event_changelist_remove_all_(&base->changelist, base); /* steal the changes array in case some broken code tries to call * dispatch twice at once. */ changes = kqop->changes; kqop->changes = NULL; /* Make sure that 'events' is at least as long as the list of changes: * otherwise errors in the changes can get reported as a -1 return * value from kevent() rather than as EV_ERROR events in the events * array. * * (We could instead handle -1 return values from kevent() by * retrying with a smaller changes array or a larger events array, * but this approach seems less risky for now.) */ if (kqop->events_size < n_changes) { int new_size = kqop->events_size; do { new_size *= 2; } while (new_size < n_changes); kq_grow_events(kqop, new_size); events = kqop->events; } EVBASE_RELEASE_LOCK(base, th_base_lock); res = kevent(kqop->kq, changes, n_changes, events, kqop->events_size, ts_p); EVBASE_ACQUIRE_LOCK(base, th_base_lock); EVUTIL_ASSERT(kqop->changes == NULL); kqop->changes = changes; if (res == -1) { if (errno != EINTR) { event_warn("kevent"); return (-1); } return (0); } event_debug(("%s: kevent reports %d", __func__, res)); for (i = 0; i < res; i++) { int which = 0; if (events[i].flags & EV_ERROR) { switch (events[i].data) { /* Can occur on delete if we are not currently * watching any events on this fd. That can * happen when the fd was closed and another * file was opened with that fd. */ case ENOENT: /* Can occur for reasons not fully understood * on FreeBSD. */ case EINVAL: continue; #if defined(__FreeBSD__) /* * This currently occurs if an FD is closed * before the EV_DELETE makes it out via kevent(). * The FreeBSD capabilities code sees the blank * capability set and rejects the request to * modify an event. * * To be strictly correct - when an FD is closed, * all the registered events are also removed. * Queuing EV_DELETE to a closed FD is wrong. * The event(s) should just be deleted from * the pending changelist. */ case ENOTCAPABLE: continue; #endif /* Can occur on a delete if the fd is closed. */ case EBADF: /* XXXX On NetBSD, we can also get EBADF if we * try to add the write side of a pipe, but * the read side has already been closed. * Other BSDs call this situation 'EPIPE'. It * would be good if we had a way to report * this situation. */ continue; /* These two can occur on an add if the fd was one side * of a pipe, and the other side was closed. */ case EPERM: case EPIPE: /* Report read events, if we're listening for * them, so that the user can learn about any * add errors. (If the operation was a * delete, then udata should be cleared.) */ if (events[i].udata) { /* The operation was an add: * report the error as a read. */ which |= EV_READ; break; } else { /* The operation was a del: * report nothing. */ continue; } /* Other errors shouldn't occur. */ default: errno = events[i].data; return (-1); } } else if (events[i].filter == EVFILT_READ) { which |= EV_READ; } else if (events[i].filter == EVFILT_WRITE) { which |= EV_WRITE; } else if (events[i].filter == EVFILT_SIGNAL) { which |= EV_SIGNAL; #ifdef EVFILT_USER } else if (events[i].filter == EVFILT_USER) { base->is_notify_pending = 0; #endif } if (!which) continue; if (events[i].filter == EVFILT_SIGNAL) { evmap_signal_active_(base, events[i].ident, 1); } else { evmap_io_active_(base, events[i].ident, which | EV_ET); } } if (res == kqop->events_size) { /* We used all the events space that we have. Maybe we should make it bigger. */ kq_grow_events(kqop, kqop->events_size * 2); } return (0); } static void kqop_free(struct kqop *kqop) { if (kqop->changes) mm_free(kqop->changes); if (kqop->events) mm_free(kqop->events); if (kqop->kq >= 0 && kqop->pid == getpid()) close(kqop->kq); memset(kqop, 0, sizeof(struct kqop)); mm_free(kqop); } static void kq_dealloc(struct event_base *base) { struct kqop *kqop = base->evbase; evsig_dealloc_(base); kqop_free(kqop); } /* signal handling */ static int kq_sig_add(struct event_base *base, int nsignal, short old, short events, void *p) { struct kqop *kqop = base->evbase; struct kevent kev; struct timespec timeout = { 0, 0 }; (void)p; EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); memset(&kev, 0, sizeof(kev)); kev.ident = nsignal; kev.filter = EVFILT_SIGNAL; kev.flags = EV_ADD; /* Be ready for the signal if it is sent any * time between now and the next call to * kq_dispatch. */ if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) return (-1); /* We can set the handler for most signals to SIG_IGN and * still have them reported to us in the queue. However, * if the handler for SIGCHLD is SIG_IGN, the system reaps * zombie processes for us, and we don't get any notification. * This appears to be the only signal with this quirk. */ if (evsig_set_handler_(base, nsignal, nsignal == SIGCHLD ? SIG_DFL : SIG_IGN) == -1) return (-1); return (0); } static int kq_sig_del(struct event_base *base, int nsignal, short old, short events, void *p) { struct kqop *kqop = base->evbase; struct kevent kev; struct timespec timeout = { 0, 0 }; (void)p; EVUTIL_ASSERT(nsignal >= 0 && nsignal < NSIG); memset(&kev, 0, sizeof(kev)); kev.ident = nsignal; kev.filter = EVFILT_SIGNAL; kev.flags = EV_DELETE; /* Because we insert signal events * immediately, we need to delete them * immediately, too */ if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) return (-1); if (evsig_restore_handler_(base, nsignal) == -1) return (-1); return (0); } /* OSX 10.6 and FreeBSD 8.1 add support for EVFILT_USER, which we can use * to wake up the event loop from another thread. */ /* Magic number we use for our filter ID. */ #define NOTIFY_IDENT 42 int event_kq_add_notify_event_(struct event_base *base) { struct kqop *kqop = base->evbase; #if defined(EVFILT_USER) && defined(NOTE_TRIGGER) struct kevent kev; struct timespec timeout = { 0, 0 }; #endif if (kqop->notify_event_added) return 0; #if defined(EVFILT_USER) && defined(NOTE_TRIGGER) memset(&kev, 0, sizeof(kev)); kev.ident = NOTIFY_IDENT; kev.filter = EVFILT_USER; kev.flags = EV_ADD | EV_CLEAR; if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) { event_warn("kevent: adding EVFILT_USER event"); return -1; } kqop->notify_event_added = 1; return 0; #else return -1; #endif } int event_kq_notify_base_(struct event_base *base) { struct kqop *kqop = base->evbase; #if defined(EVFILT_USER) && defined(NOTE_TRIGGER) struct kevent kev; struct timespec timeout = { 0, 0 }; #endif if (! kqop->notify_event_added) return -1; #if defined(EVFILT_USER) && defined(NOTE_TRIGGER) memset(&kev, 0, sizeof(kev)); kev.ident = NOTIFY_IDENT; kev.filter = EVFILT_USER; kev.fflags = NOTE_TRIGGER; if (kevent(kqop->kq, &kev, 1, NULL, 0, &timeout) == -1) { event_warn("kevent: triggering EVFILT_USER event"); return -1; } return 0; #else return -1; #endif } #endif /* EVENT__HAVE_KQUEUE */ /* * Copyright (c) 2009-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef _WIN32 #ifndef _WIN32_WINNT /* Minimum required for InitializeCriticalSectionAndSpinCount */ #define _WIN32_WINNT 0x0403 #endif #include #include #include #include #endif #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #ifdef EVENT__HAVE_FCNTL_H #include #endif #ifdef EVENT__HAVE_UNISTD_H #include #endif #include "event2/listener.h" #include "event2/util.h" #include "event2/event.h" #include "event2/event_struct.h" #include "mm-internal.h" #include "util-internal.h" #include "log-internal.h" #include "evthread-internal.h" #ifdef _WIN32 #include "iocp-internal.h" #include "defer-internal.h" #include "event-internal.h" #endif struct evconnlistener_ops { int (*enable)(struct evconnlistener *); int (*disable)(struct evconnlistener *); void (*destroy)(struct evconnlistener *); void (*shutdown)(struct evconnlistener *); evutil_socket_t (*getfd)(struct evconnlistener *); struct event_base *(*getbase)(struct evconnlistener *); }; struct evconnlistener { const struct evconnlistener_ops *ops; void *lock; evconnlistener_cb cb; evconnlistener_errorcb errorcb; void *user_data; unsigned flags; short refcnt; int accept4_flags; unsigned enabled : 1; }; struct evconnlistener_event { struct evconnlistener base; struct event listener; }; #ifdef _WIN32 struct evconnlistener_iocp { struct evconnlistener base; evutil_socket_t fd; struct event_base *event_base; struct event_iocp_port *port; short n_accepting; unsigned shutting_down : 1; unsigned event_added : 1; struct accepting_socket **accepting; }; #endif #define LOCK(listener) EVLOCK_LOCK((listener)->lock, 0) #define UNLOCK(listener) EVLOCK_UNLOCK((listener)->lock, 0) struct evconnlistener * evconnlistener_new_async(struct event_base *base, evconnlistener_cb cb, void *ptr, unsigned flags, int backlog, evutil_socket_t fd); /* XXXX export this? */ static int event_listener_enable(struct evconnlistener *); static int event_listener_disable(struct evconnlistener *); static void event_listener_destroy(struct evconnlistener *); static evutil_socket_t event_listener_getfd(struct evconnlistener *); static struct event_base *event_listener_getbase(struct evconnlistener *); #if 0 static void listener_incref_and_lock(struct evconnlistener *listener) { LOCK(listener); ++listener->refcnt; } #endif static int listener_decref_and_unlock(struct evconnlistener *listener) { int refcnt = --listener->refcnt; if (refcnt == 0) { listener->ops->destroy(listener); UNLOCK(listener); EVTHREAD_FREE_LOCK(listener->lock, EVTHREAD_LOCKTYPE_RECURSIVE); mm_free(listener); return 1; } else { UNLOCK(listener); return 0; } } static const struct evconnlistener_ops evconnlistener_event_ops = { event_listener_enable, event_listener_disable, event_listener_destroy, NULL, /* shutdown */ event_listener_getfd, event_listener_getbase }; static void listener_read_cb(evutil_socket_t, short, void *); struct evconnlistener * evconnlistener_new(struct event_base *base, evconnlistener_cb cb, void *ptr, unsigned flags, int backlog, evutil_socket_t fd) { struct evconnlistener_event *lev; #ifdef _WIN32 if (base && event_base_get_iocp_(base)) { const struct win32_extension_fns *ext = event_get_win32_extension_fns_(); if (ext->AcceptEx && ext->GetAcceptExSockaddrs) return evconnlistener_new_async(base, cb, ptr, flags, backlog, fd); } #endif if (backlog > 0) { if (listen(fd, backlog) < 0) return NULL; } else if (backlog < 0) { if (listen(fd, 128) < 0) return NULL; } lev = mm_calloc(1, sizeof(struct evconnlistener_event)); if (!lev) return NULL; lev->base.ops = &evconnlistener_event_ops; lev->base.cb = cb; lev->base.user_data = ptr; lev->base.flags = flags; lev->base.refcnt = 1; lev->base.accept4_flags = 0; if (!(flags & LEV_OPT_LEAVE_SOCKETS_BLOCKING)) lev->base.accept4_flags |= EVUTIL_SOCK_NONBLOCK; if (flags & LEV_OPT_CLOSE_ON_EXEC) lev->base.accept4_flags |= EVUTIL_SOCK_CLOEXEC; if (flags & LEV_OPT_THREADSAFE) { EVTHREAD_ALLOC_LOCK(lev->base.lock, EVTHREAD_LOCKTYPE_RECURSIVE); } event_assign(&lev->listener, base, fd, EV_READ|EV_PERSIST, listener_read_cb, lev); if (!(flags & LEV_OPT_DISABLED)) evconnlistener_enable(&lev->base); return &lev->base; } struct evconnlistener * evconnlistener_new_bind(struct event_base *base, evconnlistener_cb cb, void *ptr, unsigned flags, int backlog, const struct sockaddr *sa, int socklen) { struct evconnlistener *listener; evutil_socket_t fd; int on = 1; int family = sa ? sa->sa_family : AF_UNSPEC; int socktype = SOCK_STREAM | EVUTIL_SOCK_NONBLOCK; if (backlog == 0) return NULL; if (flags & LEV_OPT_CLOSE_ON_EXEC) socktype |= EVUTIL_SOCK_CLOEXEC; fd = evutil_socket_(family, socktype, 0); if (fd == -1) return NULL; if (setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, (void*)&on, sizeof(on))<0) goto err; if (flags & LEV_OPT_REUSEABLE) { if (evutil_make_listen_socket_reuseable(fd) < 0) goto err; } if (flags & LEV_OPT_REUSEABLE_PORT) { if (evutil_make_listen_socket_reuseable_port(fd) < 0) goto err; } if (flags & LEV_OPT_DEFERRED_ACCEPT) { if (evutil_make_tcp_listen_socket_deferred(fd) < 0) goto err; } if (flags & LEV_OPT_BIND_IPV6ONLY) { if (evutil_make_listen_socket_ipv6only(fd) < 0) goto err; } if (sa) { if (bind(fd, sa, socklen)<0) goto err; } listener = evconnlistener_new(base, cb, ptr, flags, backlog, fd); if (!listener) goto err; return listener; err: evutil_closesocket(fd); return NULL; } void evconnlistener_free(struct evconnlistener *lev) { LOCK(lev); lev->cb = NULL; lev->errorcb = NULL; if (lev->ops->shutdown) lev->ops->shutdown(lev); listener_decref_and_unlock(lev); } static void event_listener_destroy(struct evconnlistener *lev) { struct evconnlistener_event *lev_e = EVUTIL_UPCAST(lev, struct evconnlistener_event, base); event_del(&lev_e->listener); if (lev->flags & LEV_OPT_CLOSE_ON_FREE) evutil_closesocket(event_get_fd(&lev_e->listener)); event_debug_unassign(&lev_e->listener); } int evconnlistener_enable(struct evconnlistener *lev) { int r; LOCK(lev); lev->enabled = 1; if (lev->cb) r = lev->ops->enable(lev); else r = 0; UNLOCK(lev); return r; } int evconnlistener_disable(struct evconnlistener *lev) { int r; LOCK(lev); lev->enabled = 0; r = lev->ops->disable(lev); UNLOCK(lev); return r; } static int event_listener_enable(struct evconnlistener *lev) { struct evconnlistener_event *lev_e = EVUTIL_UPCAST(lev, struct evconnlistener_event, base); return event_add(&lev_e->listener, NULL); } static int event_listener_disable(struct evconnlistener *lev) { struct evconnlistener_event *lev_e = EVUTIL_UPCAST(lev, struct evconnlistener_event, base); return event_del(&lev_e->listener); } evutil_socket_t evconnlistener_get_fd(struct evconnlistener *lev) { evutil_socket_t fd; LOCK(lev); fd = lev->ops->getfd(lev); UNLOCK(lev); return fd; } static evutil_socket_t event_listener_getfd(struct evconnlistener *lev) { struct evconnlistener_event *lev_e = EVUTIL_UPCAST(lev, struct evconnlistener_event, base); return event_get_fd(&lev_e->listener); } struct event_base * evconnlistener_get_base(struct evconnlistener *lev) { struct event_base *base; LOCK(lev); base = lev->ops->getbase(lev); UNLOCK(lev); return base; } static struct event_base * event_listener_getbase(struct evconnlistener *lev) { struct evconnlistener_event *lev_e = EVUTIL_UPCAST(lev, struct evconnlistener_event, base); return event_get_base(&lev_e->listener); } void evconnlistener_set_cb(struct evconnlistener *lev, evconnlistener_cb cb, void *arg) { int enable = 0; LOCK(lev); if (lev->enabled && !lev->cb) enable = 1; lev->cb = cb; lev->user_data = arg; if (enable) evconnlistener_enable(lev); UNLOCK(lev); } void evconnlistener_set_error_cb(struct evconnlistener *lev, evconnlistener_errorcb errorcb) { LOCK(lev); lev->errorcb = errorcb; UNLOCK(lev); } static void listener_read_cb(evutil_socket_t fd, short what, void *p) { struct evconnlistener *lev = p; int err; evconnlistener_cb cb; evconnlistener_errorcb errorcb; void *user_data; LOCK(lev); while (1) { struct sockaddr_storage ss; ev_socklen_t socklen = sizeof(ss); evutil_socket_t new_fd = evutil_accept4_(fd, (struct sockaddr*)&ss, &socklen, lev->accept4_flags); if (new_fd < 0) break; if (socklen == 0) { /* This can happen with some older linux kernels in * response to nmap. */ evutil_closesocket(new_fd); continue; } if (lev->cb == NULL) { evutil_closesocket(new_fd); UNLOCK(lev); return; } ++lev->refcnt; cb = lev->cb; user_data = lev->user_data; UNLOCK(lev); cb(lev, new_fd, (struct sockaddr*)&ss, (int)socklen, user_data); LOCK(lev); if (lev->refcnt == 1) { int freed = listener_decref_and_unlock(lev); EVUTIL_ASSERT(freed); return; } --lev->refcnt; if (!lev->enabled) { /* the callback could have disabled the listener */ UNLOCK(lev); return; } } err = evutil_socket_geterror(fd); if (EVUTIL_ERR_ACCEPT_RETRIABLE(err)) { UNLOCK(lev); return; } if (lev->errorcb != NULL) { ++lev->refcnt; errorcb = lev->errorcb; user_data = lev->user_data; UNLOCK(lev); errorcb(lev, user_data); LOCK(lev); listener_decref_and_unlock(lev); } else { event_sock_warn(fd, "Error from accept() call"); UNLOCK(lev); } } #ifdef _WIN32 struct accepting_socket { CRITICAL_SECTION lock; struct event_overlapped overlapped; SOCKET s; int error; struct event_callback deferred; struct evconnlistener_iocp *lev; ev_uint8_t buflen; ev_uint8_t family; unsigned free_on_cb:1; char addrbuf[1]; }; static void accepted_socket_cb(struct event_overlapped *o, ev_uintptr_t key, ev_ssize_t n, int ok); static void accepted_socket_invoke_user_cb(struct event_callback *cb, void *arg); static void iocp_listener_event_add(struct evconnlistener_iocp *lev) { if (lev->event_added) return; lev->event_added = 1; event_base_add_virtual_(lev->event_base); } static void iocp_listener_event_del(struct evconnlistener_iocp *lev) { if (!lev->event_added) return; lev->event_added = 0; event_base_del_virtual_(lev->event_base); } static struct accepting_socket * new_accepting_socket(struct evconnlistener_iocp *lev, int family) { struct accepting_socket *res; int addrlen; int buflen; if (family == AF_INET) addrlen = sizeof(struct sockaddr_in); else if (family == AF_INET6) addrlen = sizeof(struct sockaddr_in6); else return NULL; buflen = (addrlen+16)*2; res = mm_calloc(1,sizeof(struct accepting_socket)-1+buflen); if (!res) return NULL; event_overlapped_init_(&res->overlapped, accepted_socket_cb); res->s = EVUTIL_INVALID_SOCKET; res->lev = lev; res->buflen = buflen; res->family = family; event_deferred_cb_init_(&res->deferred, event_base_get_npriorities(lev->event_base) / 2, accepted_socket_invoke_user_cb, res); InitializeCriticalSectionAndSpinCount(&res->lock, 1000); return res; } static void free_and_unlock_accepting_socket(struct accepting_socket *as) { /* requires lock. */ if (as->s != EVUTIL_INVALID_SOCKET) closesocket(as->s); LeaveCriticalSection(&as->lock); DeleteCriticalSection(&as->lock); mm_free(as); } static int start_accepting(struct accepting_socket *as) { /* requires lock */ const struct win32_extension_fns *ext = event_get_win32_extension_fns_(); DWORD pending = 0; SOCKET s = socket(as->family, SOCK_STREAM, 0); int error = 0; if (!as->lev->base.enabled) return 0; if (s == EVUTIL_INVALID_SOCKET) { error = WSAGetLastError(); goto report_err; } /* XXXX It turns out we need to do this again later. Does this call * have any effect? */ setsockopt(s, SOL_SOCKET, SO_UPDATE_ACCEPT_CONTEXT, (char *)&as->lev->fd, sizeof(&as->lev->fd)); if (!(as->lev->base.flags & LEV_OPT_LEAVE_SOCKETS_BLOCKING)) evutil_make_socket_nonblocking(s); if (event_iocp_port_associate_(as->lev->port, s, 1) < 0) { closesocket(s); return -1; } as->s = s; if (ext->AcceptEx(as->lev->fd, s, as->addrbuf, 0, as->buflen/2, as->buflen/2, &pending, &as->overlapped.overlapped)) { /* Immediate success! */ accepted_socket_cb(&as->overlapped, 1, 0, 1); } else { error = WSAGetLastError(); if (error != ERROR_IO_PENDING) { goto report_err; } } return 0; report_err: as->error = error; event_deferred_cb_schedule_( as->lev->event_base, &as->deferred); return 0; } static void stop_accepting(struct accepting_socket *as) { /* requires lock. */ SOCKET s = as->s; as->s = EVUTIL_INVALID_SOCKET; closesocket(s); } static void accepted_socket_invoke_user_cb(struct event_callback *dcb, void *arg) { struct accepting_socket *as = arg; struct sockaddr *sa_local=NULL, *sa_remote=NULL; int socklen_local=0, socklen_remote=0; const struct win32_extension_fns *ext = event_get_win32_extension_fns_(); struct evconnlistener *lev = &as->lev->base; evutil_socket_t sock=-1; void *data; evconnlistener_cb cb=NULL; evconnlistener_errorcb errorcb=NULL; int error; EVUTIL_ASSERT(ext->GetAcceptExSockaddrs); LOCK(lev); EnterCriticalSection(&as->lock); if (as->free_on_cb) { free_and_unlock_accepting_socket(as); listener_decref_and_unlock(lev); return; } ++lev->refcnt; error = as->error; if (error) { as->error = 0; errorcb = lev->errorcb; } else { ext->GetAcceptExSockaddrs( as->addrbuf, 0, as->buflen/2, as->buflen/2, &sa_local, &socklen_local, &sa_remote, &socklen_remote); sock = as->s; cb = lev->cb; as->s = EVUTIL_INVALID_SOCKET; /* We need to call this so getsockname, getpeername, and * shutdown work correctly on the accepted socket. */ /* XXXX handle error? */ setsockopt(sock, SOL_SOCKET, SO_UPDATE_ACCEPT_CONTEXT, (char *)&as->lev->fd, sizeof(&as->lev->fd)); } data = lev->user_data; LeaveCriticalSection(&as->lock); UNLOCK(lev); if (errorcb) { WSASetLastError(error); errorcb(lev, data); } else if (cb) { cb(lev, sock, sa_remote, socklen_remote, data); } LOCK(lev); if (listener_decref_and_unlock(lev)) return; EnterCriticalSection(&as->lock); start_accepting(as); LeaveCriticalSection(&as->lock); } static void accepted_socket_cb(struct event_overlapped *o, ev_uintptr_t key, ev_ssize_t n, int ok) { struct accepting_socket *as = EVUTIL_UPCAST(o, struct accepting_socket, overlapped); LOCK(&as->lev->base); EnterCriticalSection(&as->lock); if (ok) { /* XXXX Don't do this if some EV_MT flag is set. */ event_deferred_cb_schedule_( as->lev->event_base, &as->deferred); LeaveCriticalSection(&as->lock); } else if (as->free_on_cb) { struct evconnlistener *lev = &as->lev->base; free_and_unlock_accepting_socket(as); listener_decref_and_unlock(lev); return; } else if (as->s == EVUTIL_INVALID_SOCKET) { /* This is okay; we were disabled by iocp_listener_disable. */ LeaveCriticalSection(&as->lock); } else { /* Some error on accept that we couldn't actually handle. */ BOOL ok; DWORD transfer = 0, flags=0; event_sock_warn(as->s, "Unexpected error on AcceptEx"); ok = WSAGetOverlappedResult(as->s, &o->overlapped, &transfer, FALSE, &flags); if (ok) { /* well, that was confusing! */ as->error = 1; } else { as->error = WSAGetLastError(); } event_deferred_cb_schedule_( as->lev->event_base, &as->deferred); LeaveCriticalSection(&as->lock); } UNLOCK(&as->lev->base); } static int iocp_listener_enable(struct evconnlistener *lev) { int i; struct evconnlistener_iocp *lev_iocp = EVUTIL_UPCAST(lev, struct evconnlistener_iocp, base); LOCK(lev); iocp_listener_event_add(lev_iocp); for (i = 0; i < lev_iocp->n_accepting; ++i) { struct accepting_socket *as = lev_iocp->accepting[i]; if (!as) continue; EnterCriticalSection(&as->lock); if (!as->free_on_cb && as->s == EVUTIL_INVALID_SOCKET) start_accepting(as); LeaveCriticalSection(&as->lock); } UNLOCK(lev); return 0; } static int iocp_listener_disable_impl(struct evconnlistener *lev, int shutdown) { int i; struct evconnlistener_iocp *lev_iocp = EVUTIL_UPCAST(lev, struct evconnlistener_iocp, base); LOCK(lev); iocp_listener_event_del(lev_iocp); for (i = 0; i < lev_iocp->n_accepting; ++i) { struct accepting_socket *as = lev_iocp->accepting[i]; if (!as) continue; EnterCriticalSection(&as->lock); if (!as->free_on_cb && as->s != EVUTIL_INVALID_SOCKET) { if (shutdown) as->free_on_cb = 1; stop_accepting(as); } LeaveCriticalSection(&as->lock); } if (shutdown && lev->flags & LEV_OPT_CLOSE_ON_FREE) evutil_closesocket(lev_iocp->fd); UNLOCK(lev); return 0; } static int iocp_listener_disable(struct evconnlistener *lev) { return iocp_listener_disable_impl(lev,0); } static void iocp_listener_destroy(struct evconnlistener *lev) { struct evconnlistener_iocp *lev_iocp = EVUTIL_UPCAST(lev, struct evconnlistener_iocp, base); if (! lev_iocp->shutting_down) { lev_iocp->shutting_down = 1; iocp_listener_disable_impl(lev,1); } } static evutil_socket_t iocp_listener_getfd(struct evconnlistener *lev) { struct evconnlistener_iocp *lev_iocp = EVUTIL_UPCAST(lev, struct evconnlistener_iocp, base); return lev_iocp->fd; } static struct event_base * iocp_listener_getbase(struct evconnlistener *lev) { struct evconnlistener_iocp *lev_iocp = EVUTIL_UPCAST(lev, struct evconnlistener_iocp, base); return lev_iocp->event_base; } static const struct evconnlistener_ops evconnlistener_iocp_ops = { iocp_listener_enable, iocp_listener_disable, iocp_listener_destroy, iocp_listener_destroy, /* shutdown */ iocp_listener_getfd, iocp_listener_getbase }; /* XXX define some way to override this. */ #define N_SOCKETS_PER_LISTENER 4 struct evconnlistener * evconnlistener_new_async(struct event_base *base, evconnlistener_cb cb, void *ptr, unsigned flags, int backlog, evutil_socket_t fd) { struct sockaddr_storage ss; int socklen = sizeof(ss); struct evconnlistener_iocp *lev; int i; flags |= LEV_OPT_THREADSAFE; if (!base || !event_base_get_iocp_(base)) goto err; /* XXXX duplicate code */ if (backlog > 0) { if (listen(fd, backlog) < 0) goto err; } else if (backlog < 0) { if (listen(fd, 128) < 0) goto err; } if (getsockname(fd, (struct sockaddr*)&ss, &socklen)) { event_sock_warn(fd, "getsockname"); goto err; } lev = mm_calloc(1, sizeof(struct evconnlistener_iocp)); if (!lev) { event_warn("calloc"); goto err; } lev->base.ops = &evconnlistener_iocp_ops; lev->base.cb = cb; lev->base.user_data = ptr; lev->base.flags = flags; lev->base.refcnt = 1; lev->base.enabled = 1; lev->port = event_base_get_iocp_(base); lev->fd = fd; lev->event_base = base; if (event_iocp_port_associate_(lev->port, fd, 1) < 0) goto err_free_lev; EVTHREAD_ALLOC_LOCK(lev->base.lock, EVTHREAD_LOCKTYPE_RECURSIVE); lev->n_accepting = N_SOCKETS_PER_LISTENER; lev->accepting = mm_calloc(lev->n_accepting, sizeof(struct accepting_socket *)); if (!lev->accepting) { event_warn("calloc"); goto err_delete_lock; } for (i = 0; i < lev->n_accepting; ++i) { lev->accepting[i] = new_accepting_socket(lev, ss.ss_family); if (!lev->accepting[i]) { event_warnx("Couldn't create accepting socket"); goto err_free_accepting; } if (cb && start_accepting(lev->accepting[i]) < 0) { event_warnx("Couldn't start accepting on socket"); EnterCriticalSection(&lev->accepting[i]->lock); free_and_unlock_accepting_socket(lev->accepting[i]); goto err_free_accepting; } ++lev->base.refcnt; } iocp_listener_event_add(lev); return &lev->base; err_free_accepting: mm_free(lev->accepting); /* XXXX free the other elements. */ err_delete_lock: EVTHREAD_FREE_LOCK(lev->base.lock, EVTHREAD_LOCKTYPE_RECURSIVE); err_free_lev: mm_free(lev); err: /* Don't close the fd, it is caller's responsibility. */ return NULL; } #endif /* $OpenBSD: err.c,v 1.2 2002/06/25 15:50:15 mickey Exp $ */ /* * log.c * * Based on err.c, which was adapted from OpenBSD libc *err* *warn* code. * * Copyright (c) 2005-2012 Niels Provos and Nick Mathewson * * Copyright (c) 2000 Dug Song * * Copyright (c) 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #define WIN32_LEAN_AND_MEAN #include #undef WIN32_LEAN_AND_MEAN #endif #include #include #include #include #include #include #include "event2/event.h" #include "event2/util.h" #include "log-internal.h" static void event_log(int severity, const char *msg); static void event_exit(int errcode) EV_NORETURN; static event_fatal_cb fatal_fn = NULL; #ifdef EVENT_DEBUG_LOGGING_ENABLED #ifdef USE_DEBUG #define DEFAULT_MASK EVENT_DBG_ALL #else #define DEFAULT_MASK 0 #endif EVENT2_EXPORT_SYMBOL ev_uint32_t event_debug_logging_mask_ = DEFAULT_MASK; #endif /* EVENT_DEBUG_LOGGING_ENABLED */ void event_enable_debug_logging(ev_uint32_t which) { #ifdef EVENT_DEBUG_LOGGING_ENABLED event_debug_logging_mask_ = which; #endif } void event_set_fatal_callback(event_fatal_cb cb) { fatal_fn = cb; } static void event_exit(int errcode) { if (fatal_fn) { fatal_fn(errcode); exit(errcode); /* should never be reached */ } else if (errcode == EVENT_ERR_ABORT_) abort(); else exit(errcode); } void event_err(int eval, const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_ERR, strerror(errno), fmt, ap); va_end(ap); event_exit(eval); } void event_warn(const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_WARN, strerror(errno), fmt, ap); va_end(ap); } void event_sock_err(int eval, evutil_socket_t sock, const char *fmt, ...) { va_list ap; int err = evutil_socket_geterror(sock); va_start(ap, fmt); event_logv_(EVENT_LOG_ERR, evutil_socket_error_to_string(err), fmt, ap); va_end(ap); event_exit(eval); } void event_sock_warn(evutil_socket_t sock, const char *fmt, ...) { va_list ap; int err = evutil_socket_geterror(sock); va_start(ap, fmt); event_logv_(EVENT_LOG_WARN, evutil_socket_error_to_string(err), fmt, ap); va_end(ap); } void event_errx(int eval, const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_ERR, NULL, fmt, ap); va_end(ap); event_exit(eval); } void event_warnx(const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_WARN, NULL, fmt, ap); va_end(ap); } void event_msgx(const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_MSG, NULL, fmt, ap); va_end(ap); } void event_debugx_(const char *fmt, ...) { va_list ap; va_start(ap, fmt); event_logv_(EVENT_LOG_DEBUG, NULL, fmt, ap); va_end(ap); } void event_logv_(int severity, const char *errstr, const char *fmt, va_list ap) { char buf[1024]; size_t len; if (severity == EVENT_LOG_DEBUG && !event_debug_get_logging_mask_()) return; if (fmt != NULL) evutil_vsnprintf(buf, sizeof(buf), fmt, ap); else buf[0] = '\0'; if (errstr) { len = strlen(buf); if (len < sizeof(buf) - 3) { evutil_snprintf(buf + len, sizeof(buf) - len, ": %s", errstr); } } event_log(severity, buf); } static event_log_cb log_fn = NULL; void event_set_log_callback(event_log_cb cb) { log_fn = cb; } static void event_log(int severity, const char *msg) { if (log_fn) log_fn(severity, msg); else { const char *severity_str; switch (severity) { case EVENT_LOG_DEBUG: severity_str = "debug"; break; case EVENT_LOG_MSG: severity_str = "msg"; break; case EVENT_LOG_WARN: severity_str = "warn"; break; case EVENT_LOG_ERR: severity_str = "err"; break; default: severity_str = "???"; break; } (void)fprintf(stderr, "[%s] %s\n", severity_str, msg); } } /* $OpenBSD: poll.c,v 1.2 2002/06/25 15:50:15 mickey Exp $ */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_POLL #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #include #include #include #include #include #include #include #include #include "event-internal.h" #include "evsignal-internal.h" #include "log-internal.h" #include "evmap-internal.h" #include "event2/thread.h" #include "evthread-internal.h" #include "time-internal.h" /* Since Linux 2.6.17, poll is able to report about peer half-closed connection using special POLLRDHUP flag on a read event. */ #if !defined(POLLRDHUP) #define POLLRDHUP 0 #define EARLY_CLOSE_IF_HAVE_RDHUP 0 #else #define EARLY_CLOSE_IF_HAVE_RDHUP EV_FEATURE_EARLY_CLOSE #endif struct pollidx { int idxplus1; }; struct pollop { int event_count; /* Highest number alloc */ int nfds; /* Highest number used */ int realloc_copy; /* True iff we must realloc * event_set_copy */ struct pollfd *event_set; struct pollfd *event_set_copy; }; static void *poll_init(struct event_base *); static int poll_add(struct event_base *, int, short old, short events, void *idx); static int poll_del(struct event_base *, int, short old, short events, void *idx); static int poll_dispatch(struct event_base *, struct timeval *); static void poll_dealloc(struct event_base *); const struct eventop pollops = { "poll", poll_init, poll_add, poll_del, poll_dispatch, poll_dealloc, 1, /* need_reinit */ EV_FEATURE_FDS|EARLY_CLOSE_IF_HAVE_RDHUP, sizeof(struct pollidx), }; static void * poll_init(struct event_base *base) { struct pollop *pollop; if (!(pollop = mm_calloc(1, sizeof(struct pollop)))) return (NULL); evsig_init_(base); evutil_weakrand_seed_(&base->weakrand_seed, 0); return (pollop); } #ifdef CHECK_INVARIANTS static void poll_check_ok(struct pollop *pop) { int i, idx; struct event *ev; for (i = 0; i < pop->fd_count; ++i) { idx = pop->idxplus1_by_fd[i]-1; if (idx < 0) continue; EVUTIL_ASSERT(pop->event_set[idx].fd == i); } for (i = 0; i < pop->nfds; ++i) { struct pollfd *pfd = &pop->event_set[i]; EVUTIL_ASSERT(pop->idxplus1_by_fd[pfd->fd] == i+1); } } #else #define poll_check_ok(pop) #endif static int poll_dispatch(struct event_base *base, struct timeval *tv) { int res, i, j, nfds; long msec = -1; struct pollop *pop = base->evbase; struct pollfd *event_set; poll_check_ok(pop); nfds = pop->nfds; #ifndef EVENT__DISABLE_THREAD_SUPPORT if (base->th_base_lock) { /* If we're using this backend in a multithreaded setting, * then we need to work on a copy of event_set, so that we can * let other threads modify the main event_set while we're * polling. If we're not multithreaded, then we'll skip the * copy step here to save memory and time. */ if (pop->realloc_copy) { struct pollfd *tmp = mm_realloc(pop->event_set_copy, pop->event_count * sizeof(struct pollfd)); if (tmp == NULL) { event_warn("realloc"); return -1; } pop->event_set_copy = tmp; pop->realloc_copy = 0; } memcpy(pop->event_set_copy, pop->event_set, sizeof(struct pollfd)*nfds); event_set = pop->event_set_copy; } else { event_set = pop->event_set; } #else event_set = pop->event_set; #endif if (tv != NULL) { msec = evutil_tv_to_msec_(tv); if (msec < 0 || msec > INT_MAX) msec = INT_MAX; } EVBASE_RELEASE_LOCK(base, th_base_lock); res = poll(event_set, nfds, msec); EVBASE_ACQUIRE_LOCK(base, th_base_lock); if (res == -1) { if (errno != EINTR) { event_warn("poll"); return (-1); } return (0); } event_debug(("%s: poll reports %d", __func__, res)); if (res == 0 || nfds == 0) return (0); i = evutil_weakrand_range_(&base->weakrand_seed, nfds); for (j = 0; j < nfds; j++) { int what; if (++i == nfds) i = 0; what = event_set[i].revents; if (!what) continue; res = 0; /* If the file gets closed notify */ if (what & (POLLHUP|POLLERR|POLLNVAL)) what |= POLLIN|POLLOUT; if (what & POLLIN) res |= EV_READ; if (what & POLLOUT) res |= EV_WRITE; if (what & POLLRDHUP) res |= EV_CLOSED; if (res == 0) continue; evmap_io_active_(base, event_set[i].fd, res); } return (0); } static int poll_add(struct event_base *base, int fd, short old, short events, void *idx_) { struct pollop *pop = base->evbase; struct pollfd *pfd = NULL; struct pollidx *idx = idx_; int i; EVUTIL_ASSERT((events & EV_SIGNAL) == 0); if (!(events & (EV_READ|EV_WRITE|EV_CLOSED))) return (0); poll_check_ok(pop); if (pop->nfds + 1 >= pop->event_count) { struct pollfd *tmp_event_set; int tmp_event_count; if (pop->event_count < 32) tmp_event_count = 32; else tmp_event_count = pop->event_count * 2; /* We need more file descriptors */ tmp_event_set = mm_realloc(pop->event_set, tmp_event_count * sizeof(struct pollfd)); if (tmp_event_set == NULL) { event_warn("realloc"); return (-1); } pop->event_set = tmp_event_set; pop->event_count = tmp_event_count; pop->realloc_copy = 1; } i = idx->idxplus1 - 1; if (i >= 0) { pfd = &pop->event_set[i]; } else { i = pop->nfds++; pfd = &pop->event_set[i]; pfd->events = 0; pfd->fd = fd; idx->idxplus1 = i + 1; } pfd->revents = 0; if (events & EV_WRITE) pfd->events |= POLLOUT; if (events & EV_READ) pfd->events |= POLLIN; if (events & EV_CLOSED) pfd->events |= POLLRDHUP; poll_check_ok(pop); return (0); } /* * Nothing to be done here. */ static int poll_del(struct event_base *base, int fd, short old, short events, void *idx_) { struct pollop *pop = base->evbase; struct pollfd *pfd = NULL; struct pollidx *idx = idx_; int i; EVUTIL_ASSERT((events & EV_SIGNAL) == 0); if (!(events & (EV_READ|EV_WRITE|EV_CLOSED))) return (0); poll_check_ok(pop); i = idx->idxplus1 - 1; if (i < 0) return (-1); /* Do we still want to read or write? */ pfd = &pop->event_set[i]; if (events & EV_READ) pfd->events &= ~POLLIN; if (events & EV_WRITE) pfd->events &= ~POLLOUT; if (events & EV_CLOSED) pfd->events &= ~POLLRDHUP; poll_check_ok(pop); if (pfd->events) /* Another event cares about that fd. */ return (0); /* Okay, so we aren't interested in that fd anymore. */ idx->idxplus1 = 0; --pop->nfds; if (i != pop->nfds) { /* * Shift the last pollfd down into the now-unoccupied * position. */ memcpy(&pop->event_set[i], &pop->event_set[pop->nfds], sizeof(struct pollfd)); idx = evmap_io_get_fdinfo_(&base->io, pop->event_set[i].fd); EVUTIL_ASSERT(idx); EVUTIL_ASSERT(idx->idxplus1 == pop->nfds + 1); idx->idxplus1 = i + 1; } poll_check_ok(pop); return (0); } static void poll_dealloc(struct event_base *base) { struct pollop *pop = base->evbase; evsig_dealloc_(base); if (pop->event_set) mm_free(pop->event_set); if (pop->event_set_copy) mm_free(pop->event_set_copy); memset(pop, 0, sizeof(struct pollop)); mm_free(pop); } #endif /* EVENT__HAVE_POLL */ /* $OpenBSD: select.c,v 1.2 2002/06/25 15:50:15 mickey Exp $ */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_SELECT #ifdef __APPLE__ /* Apple wants us to define this if we might ever pass more than * FD_SETSIZE bits to select(). */ #define _DARWIN_UNLIMITED_SELECT #endif #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #ifdef EVENT__HAVE_SYS_SELECT_H #include #endif #include #include #include #include #include #include #include #include "event-internal.h" #include "evsignal-internal.h" #include "event2/thread.h" #include "evthread-internal.h" #include "log-internal.h" #include "evmap-internal.h" #ifndef EVENT__HAVE_FD_MASK /* This type is mandatory, but Android doesn't define it. */ typedef unsigned long fd_mask; #endif #ifndef NFDBITS #define NFDBITS (sizeof(fd_mask)*8) #endif /* Divide positive x by y, rounding up. */ #define DIV_ROUNDUP(x, y) (((x)+((y)-1))/(y)) /* How many bytes to allocate for N fds? */ #define SELECT_ALLOC_SIZE(n) \ (DIV_ROUNDUP(n, NFDBITS) * sizeof(fd_mask)) struct selectop { int event_fds; /* Highest fd in fd set */ int event_fdsz; int resize_out_sets; fd_set *event_readset_in; fd_set *event_writeset_in; fd_set *event_readset_out; fd_set *event_writeset_out; }; static void *select_init(struct event_base *); static int select_add(struct event_base *, int, short old, short events, void*); static int select_del(struct event_base *, int, short old, short events, void*); static int select_dispatch(struct event_base *, struct timeval *); static void select_dealloc(struct event_base *); const struct eventop selectops = { "select", select_init, select_add, select_del, select_dispatch, select_dealloc, 1, /* need_reinit. */ EV_FEATURE_FDS, 0, }; static int select_resize(struct selectop *sop, int fdsz); static void select_free_selectop(struct selectop *sop); static void * select_init(struct event_base *base) { struct selectop *sop; if (!(sop = mm_calloc(1, sizeof(struct selectop)))) return (NULL); if (select_resize(sop, SELECT_ALLOC_SIZE(32 + 1))) { select_free_selectop(sop); return (NULL); } evsig_init_(base); evutil_weakrand_seed_(&base->weakrand_seed, 0); return (sop); } #ifdef CHECK_INVARIANTS static void check_selectop(struct selectop *sop) { /* nothing to be done here */ } #else #define check_selectop(sop) do { (void) sop; } while (0) #endif static int select_dispatch(struct event_base *base, struct timeval *tv) { int res=0, i, j, nfds; struct selectop *sop = base->evbase; check_selectop(sop); if (sop->resize_out_sets) { fd_set *readset_out=NULL, *writeset_out=NULL; size_t sz = sop->event_fdsz; if (!(readset_out = mm_realloc(sop->event_readset_out, sz))) return (-1); sop->event_readset_out = readset_out; if (!(writeset_out = mm_realloc(sop->event_writeset_out, sz))) { /* We don't free readset_out here, since it was * already successfully reallocated. The next time * we call select_dispatch, the realloc will be a * no-op. */ return (-1); } sop->event_writeset_out = writeset_out; sop->resize_out_sets = 0; } memcpy(sop->event_readset_out, sop->event_readset_in, sop->event_fdsz); memcpy(sop->event_writeset_out, sop->event_writeset_in, sop->event_fdsz); nfds = sop->event_fds+1; EVBASE_RELEASE_LOCK(base, th_base_lock); res = select(nfds, sop->event_readset_out, sop->event_writeset_out, NULL, tv); EVBASE_ACQUIRE_LOCK(base, th_base_lock); check_selectop(sop); if (res == -1) { if (errno != EINTR) { event_warn("select"); return (-1); } return (0); } event_debug(("%s: select reports %d", __func__, res)); check_selectop(sop); i = evutil_weakrand_range_(&base->weakrand_seed, nfds); for (j = 0; j < nfds; ++j) { if (++i >= nfds) i = 0; res = 0; if (FD_ISSET(i, sop->event_readset_out)) res |= EV_READ; if (FD_ISSET(i, sop->event_writeset_out)) res |= EV_WRITE; if (res == 0) continue; evmap_io_active_(base, i, res); } check_selectop(sop); return (0); } static int select_resize(struct selectop *sop, int fdsz) { fd_set *readset_in = NULL; fd_set *writeset_in = NULL; if (sop->event_readset_in) check_selectop(sop); if ((readset_in = mm_realloc(sop->event_readset_in, fdsz)) == NULL) goto error; sop->event_readset_in = readset_in; if ((writeset_in = mm_realloc(sop->event_writeset_in, fdsz)) == NULL) { /* Note that this will leave event_readset_in expanded. * That's okay; we wouldn't want to free it, since that would * change the semantics of select_resize from "expand the * readset_in and writeset_in, or return -1" to "expand the * *set_in members, or trash them and return -1." */ goto error; } sop->event_writeset_in = writeset_in; sop->resize_out_sets = 1; memset((char *)sop->event_readset_in + sop->event_fdsz, 0, fdsz - sop->event_fdsz); memset((char *)sop->event_writeset_in + sop->event_fdsz, 0, fdsz - sop->event_fdsz); sop->event_fdsz = fdsz; check_selectop(sop); return (0); error: event_warn("malloc"); return (-1); } static int select_add(struct event_base *base, int fd, short old, short events, void *p) { struct selectop *sop = base->evbase; (void) p; EVUTIL_ASSERT((events & EV_SIGNAL) == 0); check_selectop(sop); /* * Keep track of the highest fd, so that we can calculate the size * of the fd_sets for select(2) */ if (sop->event_fds < fd) { int fdsz = sop->event_fdsz; if (fdsz < (int)sizeof(fd_mask)) fdsz = (int)sizeof(fd_mask); /* In theory we should worry about overflow here. In * reality, though, the highest fd on a unixy system will * not overflow here. XXXX */ while (fdsz < (int) SELECT_ALLOC_SIZE(fd + 1)) fdsz *= 2; if (fdsz != sop->event_fdsz) { if (select_resize(sop, fdsz)) { check_selectop(sop); return (-1); } } sop->event_fds = fd; } if (events & EV_READ) FD_SET(fd, sop->event_readset_in); if (events & EV_WRITE) FD_SET(fd, sop->event_writeset_in); check_selectop(sop); return (0); } /* * Nothing to be done here. */ static int select_del(struct event_base *base, int fd, short old, short events, void *p) { struct selectop *sop = base->evbase; (void)p; EVUTIL_ASSERT((events & EV_SIGNAL) == 0); check_selectop(sop); if (sop->event_fds < fd) { check_selectop(sop); return (0); } if (events & EV_READ) FD_CLR(fd, sop->event_readset_in); if (events & EV_WRITE) FD_CLR(fd, sop->event_writeset_in); check_selectop(sop); return (0); } static void select_free_selectop(struct selectop *sop) { if (sop->event_readset_in) mm_free(sop->event_readset_in); if (sop->event_writeset_in) mm_free(sop->event_writeset_in); if (sop->event_readset_out) mm_free(sop->event_readset_out); if (sop->event_writeset_out) mm_free(sop->event_writeset_out); memset(sop, 0, sizeof(struct selectop)); mm_free(sop); } static void select_dealloc(struct event_base *base) { evsig_dealloc_(base); select_free_selectop(base->evbase); } #endif /* EVENT__HAVE_SELECT */ /* $OpenBSD: select.c,v 1.2 2002/06/25 15:50:15 mickey Exp $ */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #define WIN32_LEAN_AND_MEAN #include #include #undef WIN32_LEAN_AND_MEAN #endif #include #ifdef EVENT__HAVE_SYS_TIME_H #include #endif #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #include #include #include #include #ifdef EVENT__HAVE_UNISTD_H #include #endif #include #ifdef EVENT__HAVE_FCNTL_H #include #endif #include "event2/event.h" #include "event2/event_struct.h" #include "event-internal.h" #include "event2/util.h" #include "evsignal-internal.h" #include "log-internal.h" #include "evmap-internal.h" #include "evthread-internal.h" /* signal.c This is the signal-handling implementation we use for backends that don't have a better way to do signal handling. It uses sigaction() or signal() to set a signal handler, and a socket pair to tell the event base when Note that I said "the event base" : only one event base can be set up to use this at a time. For historical reasons and backward compatibility, if you add an event for a signal to event_base A, then add an event for a signal (any signal!) to event_base B, event_base B will get informed about the signal, but event_base A won't. It would be neat to change this behavior in some future version of Libevent. kqueue already does something far more sensible. We can make all backends on Linux do a reasonable thing using signalfd. */ #ifndef _WIN32 /* Windows wants us to call our signal handlers as __cdecl. Nobody else * expects you to do anything crazy like this. */ #ifndef __cdecl #define __cdecl #endif #endif static int evsig_add(struct event_base *, evutil_socket_t, short, short, void *); static int evsig_del(struct event_base *, evutil_socket_t, short, short, void *); static const struct eventop evsigops = { "signal", NULL, evsig_add, evsig_del, NULL, NULL, 0, 0, 0 }; #ifndef EVENT__DISABLE_THREAD_SUPPORT /* Lock for evsig_base and evsig_base_n_signals_added fields. */ static void *evsig_base_lock = NULL; #endif /* The event base that's currently getting informed about signals. */ static struct event_base *evsig_base = NULL; /* A copy of evsig_base->sigev_n_signals_added. */ static int evsig_base_n_signals_added = 0; static evutil_socket_t evsig_base_fd = -1; static void __cdecl evsig_handler(int sig); #define EVSIGBASE_LOCK() EVLOCK_LOCK(evsig_base_lock, 0) #define EVSIGBASE_UNLOCK() EVLOCK_UNLOCK(evsig_base_lock, 0) void evsig_set_base_(struct event_base *base) { EVSIGBASE_LOCK(); evsig_base = base; evsig_base_n_signals_added = base->sig.ev_n_signals_added; evsig_base_fd = base->sig.ev_signal_pair[1]; EVSIGBASE_UNLOCK(); } /* Callback for when the signal handler write a byte to our signaling socket */ static void evsig_cb(evutil_socket_t fd, short what, void *arg) { static char signals[1024]; ev_ssize_t n; int i; int ncaught[NSIG]; struct event_base *base; base = arg; memset(&ncaught, 0, sizeof(ncaught)); while (1) { #ifdef _WIN32 n = recv(fd, signals, sizeof(signals), 0); #else n = read(fd, signals, sizeof(signals)); #endif if (n == -1) { int err = evutil_socket_geterror(fd); if (! EVUTIL_ERR_RW_RETRIABLE(err)) event_sock_err(1, fd, "%s: recv", __func__); break; } else if (n == 0) { /* XXX warn? */ break; } for (i = 0; i < n; ++i) { ev_uint8_t sig = signals[i]; if (sig < NSIG) ncaught[sig]++; } } EVBASE_ACQUIRE_LOCK(base, th_base_lock); for (i = 0; i < NSIG; ++i) { if (ncaught[i]) evmap_signal_active_(base, i, ncaught[i]); } EVBASE_RELEASE_LOCK(base, th_base_lock); } int evsig_init_(struct event_base *base) { /* * Our signal handler is going to write to one end of the socket * pair to wake up our event loop. The event loop then scans for * signals that got delivered. */ if (evutil_make_internal_pipe_(base->sig.ev_signal_pair) == -1) { #ifdef _WIN32 /* Make this nonfatal on win32, where sometimes people have localhost firewalled. */ event_sock_warn(-1, "%s: socketpair", __func__); #else event_sock_err(1, -1, "%s: socketpair", __func__); #endif return -1; } if (base->sig.sh_old) { mm_free(base->sig.sh_old); } base->sig.sh_old = NULL; base->sig.sh_old_max = 0; event_assign(&base->sig.ev_signal, base, base->sig.ev_signal_pair[0], EV_READ | EV_PERSIST, evsig_cb, base); base->sig.ev_signal.ev_flags |= EVLIST_INTERNAL; event_priority_set(&base->sig.ev_signal, 0); base->evsigsel = &evsigops; return 0; } /* Helper: set the signal handler for evsignal to handler in base, so that * we can restore the original handler when we clear the current one. */ int evsig_set_handler_(struct event_base *base, int evsignal, void (__cdecl *handler)(int)) { #ifdef EVENT__HAVE_SIGACTION struct sigaction sa; #else ev_sighandler_t sh; #endif struct evsig_info *sig = &base->sig; void *p; /* * resize saved signal handler array up to the highest signal number. * a dynamic array is used to keep footprint on the low side. */ if (evsignal >= sig->sh_old_max) { int new_max = evsignal + 1; event_debug(("%s: evsignal (%d) >= sh_old_max (%d), resizing", __func__, evsignal, sig->sh_old_max)); p = mm_realloc(sig->sh_old, new_max * sizeof(*sig->sh_old)); if (p == NULL) { event_warn("realloc"); return (-1); } memset((char *)p + sig->sh_old_max * sizeof(*sig->sh_old), 0, (new_max - sig->sh_old_max) * sizeof(*sig->sh_old)); sig->sh_old_max = new_max; sig->sh_old = p; } /* allocate space for previous handler out of dynamic array */ sig->sh_old[evsignal] = mm_malloc(sizeof *sig->sh_old[evsignal]); if (sig->sh_old[evsignal] == NULL) { event_warn("malloc"); return (-1); } /* save previous handler and setup new handler */ #ifdef EVENT__HAVE_SIGACTION memset(&sa, 0, sizeof(sa)); sa.sa_handler = handler; sa.sa_flags |= SA_RESTART; sigfillset(&sa.sa_mask); if (sigaction(evsignal, &sa, sig->sh_old[evsignal]) == -1) { event_warn("sigaction"); mm_free(sig->sh_old[evsignal]); sig->sh_old[evsignal] = NULL; return (-1); } #else if ((sh = signal(evsignal, handler)) == SIG_ERR) { event_warn("signal"); mm_free(sig->sh_old[evsignal]); sig->sh_old[evsignal] = NULL; return (-1); } *sig->sh_old[evsignal] = sh; #endif return (0); } static int evsig_add(struct event_base *base, evutil_socket_t evsignal, short old, short events, void *p) { struct evsig_info *sig = &base->sig; (void)p; EVUTIL_ASSERT(evsignal >= 0 && evsignal < NSIG); /* catch signals if they happen quickly */ EVSIGBASE_LOCK(); if (evsig_base != base && evsig_base_n_signals_added) { event_warnx("Added a signal to event base %p with signals " "already added to event_base %p. Only one can have " "signals at a time with the %s backend. The base with " "the most recently added signal or the most recent " "event_base_loop() call gets preference; do " "not rely on this behavior in future Libevent versions.", base, evsig_base, base->evsel->name); } evsig_base = base; evsig_base_n_signals_added = ++sig->ev_n_signals_added; evsig_base_fd = base->sig.ev_signal_pair[1]; EVSIGBASE_UNLOCK(); event_debug(("%s: %d: changing signal handler", __func__, (int)evsignal)); if (evsig_set_handler_(base, (int)evsignal, evsig_handler) == -1) { goto err; } if (!sig->ev_signal_added) { if (event_add_nolock_(&sig->ev_signal, NULL, 0)) goto err; sig->ev_signal_added = 1; } return (0); err: EVSIGBASE_LOCK(); --evsig_base_n_signals_added; --sig->ev_n_signals_added; EVSIGBASE_UNLOCK(); return (-1); } int evsig_restore_handler_(struct event_base *base, int evsignal) { int ret = 0; struct evsig_info *sig = &base->sig; #ifdef EVENT__HAVE_SIGACTION struct sigaction *sh; #else ev_sighandler_t *sh; #endif if (evsignal >= sig->sh_old_max) { /* Can't actually restore. */ /* XXXX.*/ return 0; } /* restore previous handler */ sh = sig->sh_old[evsignal]; sig->sh_old[evsignal] = NULL; #ifdef EVENT__HAVE_SIGACTION if (sigaction(evsignal, sh, NULL) == -1) { event_warn("sigaction"); ret = -1; } #else if (signal(evsignal, *sh) == SIG_ERR) { event_warn("signal"); ret = -1; } #endif mm_free(sh); return ret; } static int evsig_del(struct event_base *base, evutil_socket_t evsignal, short old, short events, void *p) { EVUTIL_ASSERT(evsignal >= 0 && evsignal < NSIG); event_debug(("%s: "EV_SOCK_FMT": restoring signal handler", __func__, EV_SOCK_ARG(evsignal))); EVSIGBASE_LOCK(); --evsig_base_n_signals_added; --base->sig.ev_n_signals_added; EVSIGBASE_UNLOCK(); return (evsig_restore_handler_(base, (int)evsignal)); } static void __cdecl evsig_handler(int sig) { int save_errno = errno; #ifdef _WIN32 int socket_errno = EVUTIL_SOCKET_ERROR(); #endif ev_uint8_t msg; if (evsig_base == NULL) { event_warnx( "%s: received signal %d, but have no base configured", __func__, sig); return; } #ifndef EVENT__HAVE_SIGACTION signal(sig, evsig_handler); #endif /* Wake up our notification mechanism */ msg = sig; #ifdef _WIN32 send(evsig_base_fd, (char*)&msg, 1, 0); #else { int r = write(evsig_base_fd, (char*)&msg, 1); (void)r; /* Suppress 'unused return value' and 'unused var' */ } #endif errno = save_errno; #ifdef _WIN32 EVUTIL_SET_SOCKET_ERROR(socket_errno); #endif } void evsig_dealloc_(struct event_base *base) { int i = 0; if (base->sig.ev_signal_added) { event_del(&base->sig.ev_signal); base->sig.ev_signal_added = 0; } /* debug event is created in evsig_init_/event_assign even when * ev_signal_added == 0, so unassign is required */ event_debug_unassign(&base->sig.ev_signal); for (i = 0; i < NSIG; ++i) { if (i < base->sig.sh_old_max && base->sig.sh_old[i] != NULL) evsig_restore_handler_(base, i); } EVSIGBASE_LOCK(); if (base == evsig_base) { evsig_base = NULL; evsig_base_n_signals_added = 0; evsig_base_fd = -1; } EVSIGBASE_UNLOCK(); if (base->sig.ev_signal_pair[0] != -1) { evutil_closesocket(base->sig.ev_signal_pair[0]); base->sig.ev_signal_pair[0] = -1; } if (base->sig.ev_signal_pair[1] != -1) { evutil_closesocket(base->sig.ev_signal_pair[1]); base->sig.ev_signal_pair[1] = -1; } base->sig.sh_old_max = 0; /* per index frees are handled in evsig_del() */ if (base->sig.sh_old) { mm_free(base->sig.sh_old); base->sig.sh_old = NULL; } } static void evsig_free_globals_locks(void) { #ifndef EVENT__DISABLE_THREAD_SUPPORT if (evsig_base_lock != NULL) { EVTHREAD_FREE_LOCK(evsig_base_lock, 0); evsig_base_lock = NULL; } #endif return; } void evsig_free_globals_(void) { evsig_free_globals_locks(); } #ifndef EVENT__DISABLE_THREAD_SUPPORT int evsig_global_setup_locks_(const int enable_locks) { EVTHREAD_SETUP_GLOBAL_LOCK(evsig_base_lock, 0); return 0; } #endif /* $OpenBSD: strlcpy.c,v 1.5 2001/05/13 15:40:16 deraadt Exp $ */ /* * Copyright (c) 1998 Todd C. Miller * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #if defined(LIBC_SCCS) && !defined(lint) static char *rcsid = "$OpenBSD: strlcpy.c,v 1.5 2001/05/13 15:40:16 deraadt Exp $"; #endif /* LIBC_SCCS and not lint */ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifndef EVENT__HAVE_STRLCPY #include "strlcpy-internal.h" /* * Copy src to string dst of size siz. At most siz-1 characters * will be copied. Always NUL terminates (unless siz == 0). * Returns strlen(src); if retval >= siz, truncation occurred. */ size_t event_strlcpy_(dst, src, siz) char *dst; const char *src; size_t siz; { register char *d = dst; register const char *s = src; register size_t n = siz; /* Copy as many bytes as will fit */ if (n != 0 && --n != 0) { do { if ((*d++ = *s++) == 0) break; } while (--n != 0); } /* Not enough room in dst, add NUL and traverse rest of src */ if (n == 0) { if (siz != 0) *d = '\0'; /* NUL-terminate dst */ while (*s++) ; } return (s - src - 1); /* count does not include NUL */ } #endif /* * Copyright 2007-2012 Niels Provos and Nick Mathewson * Copyright 2000-2007 Niels Provos * Copyright 2003 Michael A. Davis * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef _WIN32 #include #include #include #include #include #include #include #include #include #include #include "event2/util.h" #include "util-internal.h" #include "log-internal.h" #include "event2/event.h" #include "event-internal.h" #include "evmap-internal.h" #include "event2/thread.h" #include "evthread-internal.h" #include "time-internal.h" #define XFREE(ptr) do { if (ptr) mm_free(ptr); } while (0) extern struct event_list timequeue; extern struct event_list addqueue; struct win_fd_set { unsigned int fd_count; SOCKET fd_array[1]; }; /* MSDN says this is required to handle SIGFPE */ volatile double SIGFPE_REQ = 0.0f; struct idx_info { int read_pos_plus1; int write_pos_plus1; }; struct win32op { unsigned num_fds_in_fd_sets; int resize_out_sets; struct win_fd_set *readset_in; struct win_fd_set *writeset_in; struct win_fd_set *readset_out; struct win_fd_set *writeset_out; struct win_fd_set *exset_out; unsigned signals_are_broken : 1; }; static void *win32_init(struct event_base *); static int win32_add(struct event_base *, evutil_socket_t, short old, short events, void *idx_); static int win32_del(struct event_base *, evutil_socket_t, short old, short events, void *idx_); static int win32_dispatch(struct event_base *base, struct timeval *); static void win32_dealloc(struct event_base *); struct eventop win32ops = { "win32", win32_init, win32_add, win32_del, win32_dispatch, win32_dealloc, 0, /* doesn't need reinit */ 0, /* No features supported. */ sizeof(struct idx_info), }; #define FD_SET_ALLOC_SIZE(n) ((sizeof(struct win_fd_set) + ((n)-1)*sizeof(SOCKET))) static int grow_fd_sets(struct win32op *op, unsigned new_num_fds) { size_t size; EVUTIL_ASSERT(new_num_fds >= op->readset_in->fd_count && new_num_fds >= op->writeset_in->fd_count); EVUTIL_ASSERT(new_num_fds >= 1); size = FD_SET_ALLOC_SIZE(new_num_fds); if (!(op->readset_in = mm_realloc(op->readset_in, size))) return (-1); if (!(op->writeset_in = mm_realloc(op->writeset_in, size))) return (-1); op->resize_out_sets = 1; op->num_fds_in_fd_sets = new_num_fds; return (0); } static int do_fd_set(struct win32op *op, struct idx_info *ent, evutil_socket_t s, int read) { struct win_fd_set *set = read ? op->readset_in : op->writeset_in; if (read) { if (ent->read_pos_plus1 > 0) return (0); } else { if (ent->write_pos_plus1 > 0) return (0); } if (set->fd_count == op->num_fds_in_fd_sets) { if (grow_fd_sets(op, op->num_fds_in_fd_sets*2)) return (-1); /* set pointer will have changed and needs reiniting! */ set = read ? op->readset_in : op->writeset_in; } set->fd_array[set->fd_count] = s; if (read) ent->read_pos_plus1 = set->fd_count+1; else ent->write_pos_plus1 = set->fd_count+1; return (set->fd_count++); } static int do_fd_clear(struct event_base *base, struct win32op *op, struct idx_info *ent, int read) { int i; struct win_fd_set *set = read ? op->readset_in : op->writeset_in; if (read) { i = ent->read_pos_plus1 - 1; ent->read_pos_plus1 = 0; } else { i = ent->write_pos_plus1 - 1; ent->write_pos_plus1 = 0; } if (i < 0) return (0); if (--set->fd_count != (unsigned)i) { struct idx_info *ent2; SOCKET s2; s2 = set->fd_array[i] = set->fd_array[set->fd_count]; ent2 = evmap_io_get_fdinfo_(&base->io, s2); if (!ent2) /* This indicates a bug. */ return (0); if (read) ent2->read_pos_plus1 = i+1; else ent2->write_pos_plus1 = i+1; } return (0); } #define NEVENT 32 void * win32_init(struct event_base *base) { struct win32op *winop; size_t size; if (!(winop = mm_calloc(1, sizeof(struct win32op)))) return NULL; winop->num_fds_in_fd_sets = NEVENT; size = FD_SET_ALLOC_SIZE(NEVENT); if (!(winop->readset_in = mm_malloc(size))) goto err; if (!(winop->writeset_in = mm_malloc(size))) goto err; if (!(winop->readset_out = mm_malloc(size))) goto err; if (!(winop->writeset_out = mm_malloc(size))) goto err; if (!(winop->exset_out = mm_malloc(size))) goto err; winop->readset_in->fd_count = winop->writeset_in->fd_count = 0; winop->readset_out->fd_count = winop->writeset_out->fd_count = winop->exset_out->fd_count = 0; if (evsig_init_(base) < 0) winop->signals_are_broken = 1; evutil_weakrand_seed_(&base->weakrand_seed, 0); return (winop); err: XFREE(winop->readset_in); XFREE(winop->writeset_in); XFREE(winop->readset_out); XFREE(winop->writeset_out); XFREE(winop->exset_out); XFREE(winop); return (NULL); } int win32_add(struct event_base *base, evutil_socket_t fd, short old, short events, void *idx_) { struct win32op *win32op = base->evbase; struct idx_info *idx = idx_; if ((events & EV_SIGNAL) && win32op->signals_are_broken) return (-1); if (!(events & (EV_READ|EV_WRITE))) return (0); event_debug(("%s: adding event for %d", __func__, (int)fd)); if (events & EV_READ) { if (do_fd_set(win32op, idx, fd, 1)<0) return (-1); } if (events & EV_WRITE) { if (do_fd_set(win32op, idx, fd, 0)<0) return (-1); } return (0); } int win32_del(struct event_base *base, evutil_socket_t fd, short old, short events, void *idx_) { struct win32op *win32op = base->evbase; struct idx_info *idx = idx_; event_debug(("%s: Removing event for "EV_SOCK_FMT, __func__, EV_SOCK_ARG(fd))); if (events & EV_READ) do_fd_clear(base, win32op, idx, 1); if (events & EV_WRITE) do_fd_clear(base, win32op, idx, 0); return 0; } static void fd_set_copy(struct win_fd_set *out, const struct win_fd_set *in) { out->fd_count = in->fd_count; memcpy(out->fd_array, in->fd_array, in->fd_count * (sizeof(SOCKET))); } /* static void dump_fd_set(struct win_fd_set *s) { unsigned int i; printf("[ "); for(i=0;ifd_count;++i) printf("%d ",(int)s->fd_array[i]); printf("]\n"); } */ int win32_dispatch(struct event_base *base, struct timeval *tv) { struct win32op *win32op = base->evbase; int res = 0; unsigned j, i; int fd_count; SOCKET s; if (win32op->resize_out_sets) { size_t size = FD_SET_ALLOC_SIZE(win32op->num_fds_in_fd_sets); if (!(win32op->readset_out = mm_realloc(win32op->readset_out, size))) return (-1); if (!(win32op->exset_out = mm_realloc(win32op->exset_out, size))) return (-1); if (!(win32op->writeset_out = mm_realloc(win32op->writeset_out, size))) return (-1); win32op->resize_out_sets = 0; } fd_set_copy(win32op->readset_out, win32op->readset_in); fd_set_copy(win32op->exset_out, win32op->writeset_in); fd_set_copy(win32op->writeset_out, win32op->writeset_in); fd_count = (win32op->readset_out->fd_count > win32op->writeset_out->fd_count) ? win32op->readset_out->fd_count : win32op->writeset_out->fd_count; if (!fd_count) { long msec = tv ? evutil_tv_to_msec_(tv) : LONG_MAX; /* Sleep's DWORD argument is unsigned long */ if (msec < 0) msec = LONG_MAX; /* Windows doesn't like you to call select() with no sockets */ Sleep(msec); return (0); } EVBASE_RELEASE_LOCK(base, th_base_lock); res = select(fd_count, (struct fd_set*)win32op->readset_out, (struct fd_set*)win32op->writeset_out, (struct fd_set*)win32op->exset_out, tv); EVBASE_ACQUIRE_LOCK(base, th_base_lock); event_debug(("%s: select returned %d", __func__, res)); if (res <= 0) { event_debug(("%s: %s", __func__, evutil_socket_error_to_string(EVUTIL_SOCKET_ERROR()))); return res; } if (win32op->readset_out->fd_count) { i = evutil_weakrand_range_(&base->weakrand_seed, win32op->readset_out->fd_count); for (j=0; jreadset_out->fd_count; ++j) { if (++i >= win32op->readset_out->fd_count) i = 0; s = win32op->readset_out->fd_array[i]; evmap_io_active_(base, s, EV_READ); } } if (win32op->exset_out->fd_count) { i = evutil_weakrand_range_(&base->weakrand_seed, win32op->exset_out->fd_count); for (j=0; jexset_out->fd_count; ++j) { if (++i >= win32op->exset_out->fd_count) i = 0; s = win32op->exset_out->fd_array[i]; evmap_io_active_(base, s, EV_WRITE); } } if (win32op->writeset_out->fd_count) { i = evutil_weakrand_range_(&base->weakrand_seed, win32op->writeset_out->fd_count); for (j=0; jwriteset_out->fd_count; ++j) { if (++i >= win32op->writeset_out->fd_count) i = 0; s = win32op->writeset_out->fd_array[i]; evmap_io_active_(base, s, EV_WRITE); } } return (0); } void win32_dealloc(struct event_base *base) { struct win32op *win32op = base->evbase; evsig_dealloc_(base); if (win32op->readset_in) mm_free(win32op->readset_in); if (win32op->writeset_in) mm_free(win32op->writeset_in); if (win32op->readset_out) mm_free(win32op->readset_out); if (win32op->writeset_out) mm_free(win32op->writeset_out); if (win32op->exset_out) mm_free(win32op->exset_out); /* XXXXX free the tree. */ memset(win32op, 0, sizeof(*win32op)); mm_free(win32op); } #endif /* * Copyright (c) 2008-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef BUFFEREVENT_INTERNAL_H_INCLUDED_ #define BUFFEREVENT_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "event2/event_struct.h" #include "evconfig-private.h" #include "event2/util.h" #include "defer-internal.h" #include "evthread-internal.h" #include "event2/thread.h" #include "ratelim-internal.h" #include "event2/bufferevent_struct.h" #include "ipv6-internal.h" #ifdef _WIN32 #include #endif #ifdef EVENT__HAVE_NETINET_IN_H #include #endif #ifdef EVENT__HAVE_NETINET_IN6_H #include #endif /* These flags are reasons that we might be declining to actually enable reading or writing on a bufferevent. */ /* On a all bufferevents, for reading: used when we have read up to the watermark value. On a filtering bufferevent, for writing: used when the underlying bufferevent's write buffer has been filled up to its watermark value. */ #define BEV_SUSPEND_WM 0x01 /* On a base bufferevent: when we have emptied a bandwidth buckets */ #define BEV_SUSPEND_BW 0x02 /* On a base bufferevent: when we have emptied the group's bandwidth bucket. */ #define BEV_SUSPEND_BW_GROUP 0x04 /* On a socket bufferevent: can't do any operations while we're waiting for * name lookup to finish. */ #define BEV_SUSPEND_LOOKUP 0x08 /* On a base bufferevent, for reading: used when a filter has choked this * (underlying) bufferevent because it has stopped reading from it. */ #define BEV_SUSPEND_FILT_READ 0x10 typedef ev_uint16_t bufferevent_suspend_flags; struct bufferevent_rate_limit_group { /** List of all members in the group */ LIST_HEAD(rlim_group_member_list, bufferevent_private) members; /** Current limits for the group. */ struct ev_token_bucket rate_limit; struct ev_token_bucket_cfg rate_limit_cfg; /** True iff we don't want to read from any member of the group.until * the token bucket refills. */ unsigned read_suspended : 1; /** True iff we don't want to write from any member of the group.until * the token bucket refills. */ unsigned write_suspended : 1; /** True iff we were unable to suspend one of the bufferevents in the * group for reading the last time we tried, and we should try * again. */ unsigned pending_unsuspend_read : 1; /** True iff we were unable to suspend one of the bufferevents in the * group for writing the last time we tried, and we should try * again. */ unsigned pending_unsuspend_write : 1; /*@{*/ /** Total number of bytes read or written in this group since last * reset. */ ev_uint64_t total_read; ev_uint64_t total_written; /*@}*/ /** The number of bufferevents in the group. */ int n_members; /** The smallest number of bytes that any member of the group should * be limited to read or write at a time. */ ev_ssize_t min_share; ev_ssize_t configured_min_share; /** Timeout event that goes off once a tick, when the bucket is ready * to refill. */ struct event master_refill_event; /** Seed for weak random number generator. Protected by 'lock' */ struct evutil_weakrand_state weakrand_seed; /** Lock to protect the members of this group. This lock should nest * within every bufferevent lock: if you are holding this lock, do * not assume you can lock another bufferevent. */ void *lock; }; /** Fields for rate-limiting a single bufferevent. */ struct bufferevent_rate_limit { /* Linked-list elements for storing this bufferevent_private in a * group. * * Note that this field is supposed to be protected by the group * lock */ LIST_ENTRY(bufferevent_private) next_in_group; /** The rate-limiting group for this bufferevent, or NULL if it is * only rate-limited on its own. */ struct bufferevent_rate_limit_group *group; /* This bufferevent's current limits. */ struct ev_token_bucket limit; /* Pointer to the rate-limit configuration for this bufferevent. * Can be shared. XXX reference-count this? */ struct ev_token_bucket_cfg *cfg; /* Timeout event used when one this bufferevent's buckets are * empty. */ struct event refill_bucket_event; }; /** Parts of the bufferevent structure that are shared among all bufferevent * types, but not exposed in bufferevent_struct.h. */ struct bufferevent_private { /** The underlying bufferevent structure. */ struct bufferevent bev; /** Evbuffer callback to enforce watermarks on input. */ struct evbuffer_cb_entry *read_watermarks_cb; /** If set, we should free the lock when we free the bufferevent. */ unsigned own_lock : 1; /** Flag: set if we have deferred callbacks and a read callback is * pending. */ unsigned readcb_pending : 1; /** Flag: set if we have deferred callbacks and a write callback is * pending. */ unsigned writecb_pending : 1; /** Flag: set if we are currently busy connecting. */ unsigned connecting : 1; /** Flag: set if a connect failed prematurely; this is a hack for * getting around the bufferevent abstraction. */ unsigned connection_refused : 1; /** Set to the events pending if we have deferred callbacks and * an events callback is pending. */ short eventcb_pending; /** If set, read is suspended until one or more conditions are over. * The actual value here is a bitfield of those conditions; see the * BEV_SUSPEND_* flags above. */ bufferevent_suspend_flags read_suspended; /** If set, writing is suspended until one or more conditions are over. * The actual value here is a bitfield of those conditions; see the * BEV_SUSPEND_* flags above. */ bufferevent_suspend_flags write_suspended; /** Set to the current socket errno if we have deferred callbacks and * an events callback is pending. */ int errno_pending; /** The DNS error code for bufferevent_socket_connect_hostname */ int dns_error; /** Used to implement deferred callbacks */ struct event_callback deferred; /** The options this bufferevent was constructed with */ enum bufferevent_options options; /** Current reference count for this bufferevent. */ int refcnt; /** Lock for this bufferevent. Shared by the inbuf and the outbuf. * If NULL, locking is disabled. */ void *lock; /** No matter how big our bucket gets, don't try to read more than this * much in a single read operation. */ ev_ssize_t max_single_read; /** No matter how big our bucket gets, don't try to write more than this * much in a single write operation. */ ev_ssize_t max_single_write; /** Rate-limiting information for this bufferevent */ struct bufferevent_rate_limit *rate_limiting; /* Saved conn_addr, to extract IP address from it. * * Because some servers may reset/close connection without waiting clients, * in that case we can't extract IP address even in close_cb. * So we need to save it, just after we connected to remote server, or * after resolving (to avoid extra dns requests during retrying, since UDP * is slow) */ union { struct sockaddr_in6 in6; struct sockaddr_in in; } conn_address; struct evdns_getaddrinfo_request *dns_request; }; /** Possible operations for a control callback. */ enum bufferevent_ctrl_op { BEV_CTRL_SET_FD, BEV_CTRL_GET_FD, BEV_CTRL_GET_UNDERLYING, BEV_CTRL_CANCEL_ALL }; /** Possible data types for a control callback */ union bufferevent_ctrl_data { void *ptr; evutil_socket_t fd; }; /** Implementation table for a bufferevent: holds function pointers and other information to make the various bufferevent types work. */ struct bufferevent_ops { /** The name of the bufferevent's type. */ const char *type; /** At what offset into the implementation type will we find a bufferevent structure? Example: if the type is implemented as struct bufferevent_x { int extra_data; struct bufferevent bev; } then mem_offset should be offsetof(struct bufferevent_x, bev) */ off_t mem_offset; /** Enables one or more of EV_READ|EV_WRITE on a bufferevent. Does not need to adjust the 'enabled' field. Returns 0 on success, -1 on failure. */ int (*enable)(struct bufferevent *, short); /** Disables one or more of EV_READ|EV_WRITE on a bufferevent. Does not need to adjust the 'enabled' field. Returns 0 on success, -1 on failure. */ int (*disable)(struct bufferevent *, short); /** Detatches the bufferevent from related data structures. Called as * soon as its reference count reaches 0. */ void (*unlink)(struct bufferevent *); /** Free any storage and deallocate any extra data or structures used in this implementation. Called when the bufferevent is finalized. */ void (*destruct)(struct bufferevent *); /** Called when the timeouts on the bufferevent have changed.*/ int (*adj_timeouts)(struct bufferevent *); /** Called to flush data. */ int (*flush)(struct bufferevent *, short, enum bufferevent_flush_mode); /** Called to access miscellaneous fields. */ int (*ctrl)(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *); }; extern const struct bufferevent_ops bufferevent_ops_socket; extern const struct bufferevent_ops bufferevent_ops_filter; extern const struct bufferevent_ops bufferevent_ops_pair; #define BEV_IS_SOCKET(bevp) ((bevp)->be_ops == &bufferevent_ops_socket) #define BEV_IS_FILTER(bevp) ((bevp)->be_ops == &bufferevent_ops_filter) #define BEV_IS_PAIR(bevp) ((bevp)->be_ops == &bufferevent_ops_pair) #if defined(EVENT__HAVE_OPENSSL) extern const struct bufferevent_ops bufferevent_ops_openssl; #define BEV_IS_OPENSSL(bevp) ((bevp)->be_ops == &bufferevent_ops_openssl) #else #define BEV_IS_OPENSSL(bevp) 0 #endif #ifdef _WIN32 extern const struct bufferevent_ops bufferevent_ops_async; #define BEV_IS_ASYNC(bevp) ((bevp)->be_ops == &bufferevent_ops_async) #else #define BEV_IS_ASYNC(bevp) 0 #endif /** Initialize the shared parts of a bufferevent. */ EVENT2_EXPORT_SYMBOL int bufferevent_init_common_(struct bufferevent_private *, struct event_base *, const struct bufferevent_ops *, enum bufferevent_options options); /** For internal use: temporarily stop all reads on bufev, until the conditions * in 'what' are over. */ EVENT2_EXPORT_SYMBOL void bufferevent_suspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what); /** For internal use: clear the conditions 'what' on bufev, and re-enable * reading if there are no conditions left. */ EVENT2_EXPORT_SYMBOL void bufferevent_unsuspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what); /** For internal use: temporarily stop all writes on bufev, until the conditions * in 'what' are over. */ void bufferevent_suspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what); /** For internal use: clear the conditions 'what' on bufev, and re-enable * writing if there are no conditions left. */ void bufferevent_unsuspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what); #define bufferevent_wm_suspend_read(b) \ bufferevent_suspend_read_((b), BEV_SUSPEND_WM) #define bufferevent_wm_unsuspend_read(b) \ bufferevent_unsuspend_read_((b), BEV_SUSPEND_WM) /* Disable a bufferevent. Equivalent to bufferevent_disable(), but first resets 'connecting' flag to force EV_WRITE down for sure. XXXX this method will go away in the future; try not to add new users. See comment in evhttp_connection_reset_() for discussion. @param bufev the bufferevent to be disabled @param event any combination of EV_READ | EV_WRITE. @return 0 if successful, or -1 if an error occurred @see bufferevent_disable() */ EVENT2_EXPORT_SYMBOL int bufferevent_disable_hard_(struct bufferevent *bufev, short event); /** Internal: Set up locking on a bufferevent. If lock is set, use it. * Otherwise, use a new lock. */ EVENT2_EXPORT_SYMBOL int bufferevent_enable_locking_(struct bufferevent *bufev, void *lock); /** Internal: backwards compat macro for the now public function * Increment the reference count on bufev. */ #define bufferevent_incref_(bufev) bufferevent_incref(bufev) /** Internal: Lock bufev and increase its reference count. * unlocking it otherwise. */ EVENT2_EXPORT_SYMBOL void bufferevent_incref_and_lock_(struct bufferevent *bufev); /** Internal: backwards compat macro for the now public function * Decrement the reference count on bufev. Returns 1 if it freed * the bufferevent.*/ #define bufferevent_decref_(bufev) bufferevent_decref(bufev) /** Internal: Drop the reference count on bufev, freeing as necessary, and * unlocking it otherwise. Returns 1 if it freed the bufferevent. */ EVENT2_EXPORT_SYMBOL int bufferevent_decref_and_unlock_(struct bufferevent *bufev); /** Internal: If callbacks are deferred and we have a read callback, schedule * a readcb. Otherwise just run the readcb. Ignores watermarks. */ EVENT2_EXPORT_SYMBOL void bufferevent_run_readcb_(struct bufferevent *bufev, int options); /** Internal: If callbacks are deferred and we have a write callback, schedule * a writecb. Otherwise just run the writecb. Ignores watermarks. */ EVENT2_EXPORT_SYMBOL void bufferevent_run_writecb_(struct bufferevent *bufev, int options); /** Internal: If callbacks are deferred and we have an eventcb, schedule * it to run with events "what". Otherwise just run the eventcb. * See bufferevent_trigger_event for meaning of "options". */ EVENT2_EXPORT_SYMBOL void bufferevent_run_eventcb_(struct bufferevent *bufev, short what, int options); /** Internal: Run or schedule (if deferred or options contain * BEV_TRIG_DEFER_CALLBACKS) I/O callbacks specified in iotype. * Must already hold the bufev lock. Honors watermarks unless * BEV_TRIG_IGNORE_WATERMARKS is in options. */ static inline void bufferevent_trigger_nolock_(struct bufferevent *bufev, short iotype, int options); /* Making this inline since all of the common-case calls to this function in * libevent use constant arguments. */ static inline void bufferevent_trigger_nolock_(struct bufferevent *bufev, short iotype, int options) { if ((iotype & EV_READ) && ((options & BEV_TRIG_IGNORE_WATERMARKS) || evbuffer_get_length(bufev->input) >= bufev->wm_read.low)) bufferevent_run_readcb_(bufev, options); if ((iotype & EV_WRITE) && ((options & BEV_TRIG_IGNORE_WATERMARKS) || evbuffer_get_length(bufev->output) <= bufev->wm_write.low)) bufferevent_run_writecb_(bufev, options); } /** Internal: Add the event 'ev' with timeout tv, unless tv is set to 0, in * which case add ev with no timeout. */ EVENT2_EXPORT_SYMBOL int bufferevent_add_event_(struct event *ev, const struct timeval *tv); /* ========= * These next functions implement timeouts for bufferevents that aren't doing * anything else with ev_read and ev_write, to handle timeouts. * ========= */ /** Internal use: Set up the ev_read and ev_write callbacks so that * the other "generic_timeout" functions will work on it. Call this from * the constructor function. */ EVENT2_EXPORT_SYMBOL void bufferevent_init_generic_timeout_cbs_(struct bufferevent *bev); /** Internal use: Add or delete the generic timeout events as appropriate. * (If an event is enabled and a timeout is set, we add the event. Otherwise * we delete it.) Call this from anything that changes the timeout values, * that enabled EV_READ or EV_WRITE, or that disables EV_READ or EV_WRITE. */ EVENT2_EXPORT_SYMBOL int bufferevent_generic_adj_timeouts_(struct bufferevent *bev); EVENT2_EXPORT_SYMBOL int bufferevent_generic_adj_existing_timeouts_(struct bufferevent *bev); EVENT2_EXPORT_SYMBOL enum bufferevent_options bufferevent_get_options_(struct bufferevent *bev); EVENT2_EXPORT_SYMBOL const struct sockaddr* bufferevent_socket_get_conn_address_(struct bufferevent *bev); EVENT2_EXPORT_SYMBOL void bufferevent_socket_set_conn_address_fd_(struct bufferevent *bev, evutil_socket_t fd); EVENT2_EXPORT_SYMBOL void bufferevent_socket_set_conn_address_(struct bufferevent *bev, struct sockaddr *addr, size_t addrlen); /** Internal use: We have just successfully read data into an inbuf, so * reset the read timeout (if any). */ #define BEV_RESET_GENERIC_READ_TIMEOUT(bev) \ do { \ if (evutil_timerisset(&(bev)->timeout_read)) \ event_add(&(bev)->ev_read, &(bev)->timeout_read); \ } while (0) /** Internal use: We have just successfully written data from an inbuf, so * reset the read timeout (if any). */ #define BEV_RESET_GENERIC_WRITE_TIMEOUT(bev) \ do { \ if (evutil_timerisset(&(bev)->timeout_write)) \ event_add(&(bev)->ev_write, &(bev)->timeout_write); \ } while (0) #define BEV_DEL_GENERIC_READ_TIMEOUT(bev) \ event_del(&(bev)->ev_read) #define BEV_DEL_GENERIC_WRITE_TIMEOUT(bev) \ event_del(&(bev)->ev_write) /** Internal: Given a bufferevent, return its corresponding * bufferevent_private. */ #define BEV_UPCAST(b) EVUTIL_UPCAST((b), struct bufferevent_private, bev) #ifdef EVENT__DISABLE_THREAD_SUPPORT #define BEV_LOCK(b) EVUTIL_NIL_STMT_ #define BEV_UNLOCK(b) EVUTIL_NIL_STMT_ #else /** Internal: Grab the lock (if any) on a bufferevent */ #define BEV_LOCK(b) do { \ struct bufferevent_private *locking = BEV_UPCAST(b); \ EVLOCK_LOCK(locking->lock, 0); \ } while (0) /** Internal: Release the lock (if any) on a bufferevent */ #define BEV_UNLOCK(b) do { \ struct bufferevent_private *locking = BEV_UPCAST(b); \ EVLOCK_UNLOCK(locking->lock, 0); \ } while (0) #endif /* ==== For rate-limiting. */ EVENT2_EXPORT_SYMBOL int bufferevent_decrement_write_buckets_(struct bufferevent_private *bev, ev_ssize_t bytes); EVENT2_EXPORT_SYMBOL int bufferevent_decrement_read_buckets_(struct bufferevent_private *bev, ev_ssize_t bytes); EVENT2_EXPORT_SYMBOL ev_ssize_t bufferevent_get_read_max_(struct bufferevent_private *bev); EVENT2_EXPORT_SYMBOL ev_ssize_t bufferevent_get_write_max_(struct bufferevent_private *bev); int bufferevent_ratelim_init_(struct bufferevent_private *bev); #ifdef __cplusplus } #endif #endif /* BUFFEREVENT_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef CHANGELIST_INTERNAL_H_INCLUDED_ #define CHANGELIST_INTERNAL_H_INCLUDED_ /* A "changelist" is a list of all the fd status changes that should be made between calls to the backend's dispatch function. There are a few reasons that a backend would want to queue changes like this rather than processing them immediately. 1) Sometimes applications will add and delete the same event more than once between calls to dispatch. Processing these changes immediately is needless, and potentially expensive (especially if we're on a system that makes one syscall per changed event). 2) Sometimes we can coalesce multiple changes on the same fd into a single syscall if we know about them in advance. For example, epoll can do an add and a delete at the same time, but only if we have found out about both of them before we tell epoll. 3) Sometimes adding an event that we immediately delete can cause unintended consequences: in kqueue, this makes pending events get reported spuriously. */ #include "event2/util.h" /** Represents a */ struct event_change { /** The fd or signal whose events are to be changed */ evutil_socket_t fd; /* The events that were enabled on the fd before any of these changes were made. May include EV_READ or EV_WRITE. */ short old_events; /* The changes that we want to make in reading and writing on this fd. * If this is a signal, then read_change has EV_CHANGE_SIGNAL set, * and write_change is unused. */ ev_uint8_t read_change; ev_uint8_t write_change; ev_uint8_t close_change; }; /* Flags for read_change and write_change. */ /* If set, add the event. */ #define EV_CHANGE_ADD 0x01 /* If set, delete the event. Exclusive with EV_CHANGE_ADD */ #define EV_CHANGE_DEL 0x02 /* If set, this event refers a signal, not an fd. */ #define EV_CHANGE_SIGNAL EV_SIGNAL /* Set for persistent events. Currently not used. */ #define EV_CHANGE_PERSIST EV_PERSIST /* Set for adding edge-triggered events. */ #define EV_CHANGE_ET EV_ET /* The value of fdinfo_size that a backend should use if it is letting * changelist handle its add and delete functions. */ #define EVENT_CHANGELIST_FDINFO_SIZE sizeof(int) /** Set up the data fields in a changelist. */ void event_changelist_init_(struct event_changelist *changelist); /** Remove every change in the changelist, and make corresponding changes * in the event maps in the base. This function is generally used right * after making all the changes in the changelist. */ void event_changelist_remove_all_(struct event_changelist *changelist, struct event_base *base); /** Free all memory held in a changelist. */ void event_changelist_freemem_(struct event_changelist *changelist); /** Implementation of eventop_add that queues the event in a changelist. */ int event_changelist_add_(struct event_base *base, evutil_socket_t fd, short old, short events, void *p); /** Implementation of eventop_del that queues the event in a changelist. */ int event_changelist_del_(struct event_base *base, evutil_socket_t fd, short old, short events, void *p); #endif /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef DEFER_INTERNAL_H_INCLUDED_ #define DEFER_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "evconfig-private.h" #include struct event_callback; typedef void (*deferred_cb_fn)(struct event_callback *, void *); /** Initialize an empty, non-pending event_callback. @param deferred The struct event_callback structure to initialize. @param priority The priority that the callback should run at. @param cb The function to run when the struct event_callback executes. @param arg The function's second argument. */ EVENT2_EXPORT_SYMBOL void event_deferred_cb_init_(struct event_callback *, ev_uint8_t, deferred_cb_fn, void *); /** Change the priority of a non-pending event_callback. */ void event_deferred_cb_set_priority_(struct event_callback *, ev_uint8_t); /** Cancel a struct event_callback if it is currently scheduled in an event_base. */ EVENT2_EXPORT_SYMBOL void event_deferred_cb_cancel_(struct event_base *, struct event_callback *); /** Activate a struct event_callback if it is not currently scheduled in an event_base. Return true if it was not previously scheduled. */ EVENT2_EXPORT_SYMBOL int event_deferred_cb_schedule_(struct event_base *, struct event_callback *); #ifdef __cplusplus } #endif #endif /* EVENT_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EPOLLTABLE_INTERNAL_H_INCLUDED_ #define EPOLLTABLE_INTERNAL_H_INCLUDED_ /* Here are the values we're masking off to decide what operations to do. Note that since EV_READ|EV_WRITE. Note also that this table is a little sparse, since ADD+DEL is nonsensical ("xxx" in the list below.) Note also that we are shifting old_events by only 5 bits, since EV_READ is 2 and EV_WRITE is 4. The table was auto-generated with a python script, according to this pseudocode:[*0] If either the read or the write change is add+del: This is impossible; Set op==-1, events=0. Else, if either the read or the write change is add: Set events to 0. If the read change is add, or (the read change is not del, and ev_read is in old_events): Add EPOLLIN to events. If the write change is add, or (the write change is not del, and ev_write is in old_events): Add EPOLLOUT to events. If old_events is set: Set op to EPOLL_CTL_MOD [*1,*2] Else: Set op to EPOLL_CTL_ADD [*3] Else, if the read or the write change is del: Set op to EPOLL_CTL_DEL. If the read change is del: If the write change is del: Set events to EPOLLIN|EPOLLOUT Else if ev_write is in old_events: Set events to EPOLLOUT Set op to EPOLL_CTL_MOD Else Set events to EPOLLIN Else: {The write change is del.} If ev_read is in old_events: Set events to EPOLLIN Set op to EPOLL_CTL_MOD Else: Set the events to EPOLLOUT Else: There is no read or write change; set op to 0 and events to 0. The logic is a little tricky, since we had no events set on the fd before, we need to set op="ADD" and set events=the events we want to add. If we had any events set on the fd before, and we want any events to remain on the fd, we need to say op="MOD" and set events=the events we want to remain. But if we want to delete the last event, we say op="DEL" and set events=(any non-null pointer). [*0] Actually, the Python script has gotten a bit more complicated, to support EPOLLRDHUP. [*1] This MOD is only a guess. MOD might fail with ENOENT if the file was closed and a new file was opened with the same fd. If so, we'll retry with ADD. [*2] We can't replace this with a no-op even if old_events is the same as the new events: if the file was closed and reopened, we need to retry with an ADD. (We do a MOD in this case since "no change" is more common than "close and reopen", so we'll usually wind up doing 1 syscalls instead of 2.) [*3] This ADD is only a guess. There is a fun Linux kernel issue where if you have two fds for the same file (via dup) and you ADD one to an epfd, then close it, then re-create it with the same fd (via dup2 or an unlucky dup), then try to ADD it again, you'll get an EEXIST, since the struct epitem is not actually removed from the struct eventpoll until the file itself is closed. EV_CHANGE_ADD==1 EV_CHANGE_DEL==2 EV_READ ==2 EV_WRITE ==4 EV_CLOSED ==0x80 Bit 0: close change is add Bit 1: close change is del Bit 2: read change is add Bit 3: read change is del Bit 4: write change is add Bit 5: write change is del Bit 6: old events had EV_READ Bit 7: old events had EV_WRITE Bit 8: old events had EV_CLOSED */ #define EPOLL_OP_TABLE_INDEX(c) \ ( (((c)->close_change&(EV_CHANGE_ADD|EV_CHANGE_DEL))) | \ (((c)->read_change&(EV_CHANGE_ADD|EV_CHANGE_DEL)) << 2) | \ (((c)->write_change&(EV_CHANGE_ADD|EV_CHANGE_DEL)) << 4) | \ (((c)->old_events&(EV_READ|EV_WRITE)) << 5) | \ (((c)->old_events&(EV_CLOSED)) << 1) \ ) #if EV_READ != 2 || EV_WRITE != 4 || EV_CLOSED != 0x80 || EV_CHANGE_ADD != 1 || EV_CHANGE_DEL != 2 #error "Libevent's internals changed! Regenerate the op_table in epolltable-internal.h" #endif static const struct operation { int events; int op; } epoll_op_table[] = { /* old= 0, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= 0, write: 0, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write: 0, read: 0, close:del */ { EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= 0, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= 0, write: 0, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_ADD }, /* old= 0, write: 0, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write: 0, read:add, close:del */ { EPOLLIN, EPOLL_CTL_ADD }, /* old= 0, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= 0, write: 0, read:del, close: 0 */ { EPOLLIN, EPOLL_CTL_DEL }, /* old= 0, write: 0, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write: 0, read:del, close:del */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= 0, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= 0, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= 0, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= 0, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= 0, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= 0, write:add, read: 0, close: 0 */ { EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:add, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= 0, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read:add, close:xxx */ { 0, 255 }, /* old= 0, write:add, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_ADD }, /* old= 0, write:add, read:del, close:xxx */ { 0, 255 }, /* old= 0, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= 0, write:add, read:xxx, close:add */ { 0, 255 }, /* old= 0, write:add, read:xxx, close:del */ { 0, 255 }, /* old= 0, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= 0, write:del, read: 0, close: 0 */ { EPOLLOUT, EPOLL_CTL_DEL }, /* old= 0, write:del, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:del, read: 0, close:del */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= 0, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= 0, write:del, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_ADD }, /* old= 0, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_ADD }, /* old= 0, write:del, read:add, close:xxx */ { 0, 255 }, /* old= 0, write:del, read:del, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_DEL }, /* old= 0, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_ADD }, /* old= 0, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= 0, write:del, read:del, close:xxx */ { 0, 255 }, /* old= 0, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= 0, write:del, read:xxx, close:add */ { 0, 255 }, /* old= 0, write:del, read:xxx, close:del */ { 0, 255 }, /* old= 0, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= 0, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= 0, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= 0, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= 0, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= 0, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= 0, write:xxx, read:add, close:add */ { 0, 255 }, /* old= 0, write:xxx, read:add, close:del */ { 0, 255 }, /* old= 0, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= 0, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= 0, write:xxx, read:del, close:add */ { 0, 255 }, /* old= 0, write:xxx, read:del, close:del */ { 0, 255 }, /* old= 0, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= 0, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= 0, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= 0, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= 0, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= r, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= r, write: 0, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write: 0, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= r, write: 0, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write: 0, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write: 0, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= r, write: 0, read:del, close: 0 */ { EPOLLIN, EPOLL_CTL_DEL }, /* old= r, write: 0, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write: 0, read:del, close:del */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= r, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= r, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= r, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= r, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= r, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= r, write:add, read: 0, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:add, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= r, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read:add, close:xxx */ { 0, 255 }, /* old= r, write:add, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= r, write:add, read:del, close:xxx */ { 0, 255 }, /* old= r, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= r, write:add, read:xxx, close:add */ { 0, 255 }, /* old= r, write:add, read:xxx, close:del */ { 0, 255 }, /* old= r, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= r, write:del, read: 0, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write:del, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:del, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= r, write:del, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= r, write:del, read:add, close:xxx */ { 0, 255 }, /* old= r, write:del, read:del, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_DEL }, /* old= r, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= r, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= r, write:del, read:del, close:xxx */ { 0, 255 }, /* old= r, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= r, write:del, read:xxx, close:add */ { 0, 255 }, /* old= r, write:del, read:xxx, close:del */ { 0, 255 }, /* old= r, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= r, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= r, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= r, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= r, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= r, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= r, write:xxx, read:add, close:add */ { 0, 255 }, /* old= r, write:xxx, read:add, close:del */ { 0, 255 }, /* old= r, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= r, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= r, write:xxx, read:del, close:add */ { 0, 255 }, /* old= r, write:xxx, read:del, close:del */ { 0, 255 }, /* old= r, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= r, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= r, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= r, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= r, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= w, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= w, write: 0, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write: 0, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= w, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write: 0, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write: 0, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= w, write: 0, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write: 0, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write: 0, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= w, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= w, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= w, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= w, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= w, write:add, read: 0, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:add, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= w, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read:add, close:xxx */ { 0, 255 }, /* old= w, write:add, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= w, write:add, read:del, close:xxx */ { 0, 255 }, /* old= w, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= w, write:add, read:xxx, close:add */ { 0, 255 }, /* old= w, write:add, read:xxx, close:del */ { 0, 255 }, /* old= w, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= w, write:del, read: 0, close: 0 */ { EPOLLOUT, EPOLL_CTL_DEL }, /* old= w, write:del, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:del, read: 0, close:del */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= w, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= w, write:del, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= w, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= w, write:del, read:add, close:xxx */ { 0, 255 }, /* old= w, write:del, read:del, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_DEL }, /* old= w, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= w, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= w, write:del, read:del, close:xxx */ { 0, 255 }, /* old= w, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= w, write:del, read:xxx, close:add */ { 0, 255 }, /* old= w, write:del, read:xxx, close:del */ { 0, 255 }, /* old= w, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= w, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= w, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= w, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= w, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= w, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= w, write:xxx, read:add, close:add */ { 0, 255 }, /* old= w, write:xxx, read:add, close:del */ { 0, 255 }, /* old= w, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= w, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= w, write:xxx, read:del, close:add */ { 0, 255 }, /* old= w, write:xxx, read:del, close:del */ { 0, 255 }, /* old= w, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= w, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= w, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= w, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= w, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= rw, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= rw, write: 0, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write: 0, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= rw, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= rw, write: 0, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= rw, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= rw, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= rw, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= rw, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= rw, write:add, read: 0, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:add, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= rw, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read:add, close:xxx */ { 0, 255 }, /* old= rw, write:add, read:del, close: 0 */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= rw, write:add, read:del, close:xxx */ { 0, 255 }, /* old= rw, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= rw, write:add, read:xxx, close:add */ { 0, 255 }, /* old= rw, write:add, read:xxx, close:del */ { 0, 255 }, /* old= rw, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= rw, write:del, read: 0, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= rw, write:del, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:del, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= rw, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= rw, write:del, read:add, close: 0 */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= rw, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= rw, write:del, read:add, close:xxx */ { 0, 255 }, /* old= rw, write:del, read:del, close: 0 */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_DEL }, /* old= rw, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= rw, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= rw, write:del, read:del, close:xxx */ { 0, 255 }, /* old= rw, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= rw, write:del, read:xxx, close:add */ { 0, 255 }, /* old= rw, write:del, read:xxx, close:del */ { 0, 255 }, /* old= rw, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= rw, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= rw, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= rw, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= rw, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= rw, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= rw, write:xxx, read:add, close:add */ { 0, 255 }, /* old= rw, write:xxx, read:add, close:del */ { 0, 255 }, /* old= rw, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= rw, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= rw, write:xxx, read:del, close:add */ { 0, 255 }, /* old= rw, write:xxx, read:del, close:del */ { 0, 255 }, /* old= rw, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= rw, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= rw, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= rw, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= rw, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= c, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= c, write: 0, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write: 0, read: 0, close:del */ { EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= c, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= c, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write: 0, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write: 0, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= c, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= c, write: 0, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write: 0, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write: 0, read:del, close:del */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= c, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= c, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= c, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= c, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= c, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= c, write:add, read: 0, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= c, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= c, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= c, write:add, read:add, close:xxx */ { 0, 255 }, /* old= c, write:add, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= c, write:add, read:del, close:xxx */ { 0, 255 }, /* old= c, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= c, write:add, read:xxx, close:add */ { 0, 255 }, /* old= c, write:add, read:xxx, close:del */ { 0, 255 }, /* old= c, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= c, write:del, read: 0, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read: 0, close:del */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= c, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= c, write:del, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= c, write:del, read:add, close:xxx */ { 0, 255 }, /* old= c, write:del, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= c, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= c, write:del, read:del, close:xxx */ { 0, 255 }, /* old= c, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= c, write:del, read:xxx, close:add */ { 0, 255 }, /* old= c, write:del, read:xxx, close:del */ { 0, 255 }, /* old= c, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= c, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= c, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= c, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= c, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= c, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= c, write:xxx, read:add, close:add */ { 0, 255 }, /* old= c, write:xxx, read:add, close:del */ { 0, 255 }, /* old= c, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= c, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= c, write:xxx, read:del, close:add */ { 0, 255 }, /* old= c, write:xxx, read:del, close:del */ { 0, 255 }, /* old= c, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= c, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= c, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= c, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= c, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= cr, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= cr, write: 0, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write: 0, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= cr, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= cr, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write: 0, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write: 0, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= cr, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= cr, write: 0, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write: 0, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write: 0, read:del, close:del */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= cr, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= cr, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= cr, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= cr, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= cr, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= cr, write:add, read: 0, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= cr, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= cr, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= cr, write:add, read:add, close:xxx */ { 0, 255 }, /* old= cr, write:add, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= cr, write:add, read:del, close:xxx */ { 0, 255 }, /* old= cr, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= cr, write:add, read:xxx, close:add */ { 0, 255 }, /* old= cr, write:add, read:xxx, close:del */ { 0, 255 }, /* old= cr, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= cr, write:del, read: 0, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= cr, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= cr, write:del, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= cr, write:del, read:add, close:xxx */ { 0, 255 }, /* old= cr, write:del, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cr, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= cr, write:del, read:del, close:xxx */ { 0, 255 }, /* old= cr, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= cr, write:del, read:xxx, close:add */ { 0, 255 }, /* old= cr, write:del, read:xxx, close:del */ { 0, 255 }, /* old= cr, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= cr, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= cr, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= cr, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= cr, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= cr, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= cr, write:xxx, read:add, close:add */ { 0, 255 }, /* old= cr, write:xxx, read:add, close:del */ { 0, 255 }, /* old= cr, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= cr, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= cr, write:xxx, read:del, close:add */ { 0, 255 }, /* old= cr, write:xxx, read:del, close:del */ { 0, 255 }, /* old= cr, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= cr, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= cr, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= cr, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= cr, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old= cw, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old= cw, write: 0, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write: 0, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old= cw, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:add, close:xxx */ { 0, 255 }, /* old= cw, write: 0, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write: 0, read:del, close:xxx */ { 0, 255 }, /* old= cw, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old= cw, write: 0, read:xxx, close:add */ { 0, 255 }, /* old= cw, write: 0, read:xxx, close:del */ { 0, 255 }, /* old= cw, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old= cw, write:add, read: 0, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read: 0, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read: 0, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write:add, read: 0, close:xxx */ { 0, 255 }, /* old= cw, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write:add, read:add, close:xxx */ { 0, 255 }, /* old= cw, write:add, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old= cw, write:add, read:del, close:xxx */ { 0, 255 }, /* old= cw, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old= cw, write:add, read:xxx, close:add */ { 0, 255 }, /* old= cw, write:add, read:xxx, close:del */ { 0, 255 }, /* old= cw, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old= cw, write:del, read: 0, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read: 0, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read: 0, close:del */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= cw, write:del, read: 0, close:xxx */ { 0, 255 }, /* old= cw, write:del, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old= cw, write:del, read:add, close:xxx */ { 0, 255 }, /* old= cw, write:del, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old= cw, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old= cw, write:del, read:del, close:xxx */ { 0, 255 }, /* old= cw, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old= cw, write:del, read:xxx, close:add */ { 0, 255 }, /* old= cw, write:del, read:xxx, close:del */ { 0, 255 }, /* old= cw, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old= cw, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old= cw, write:xxx, read: 0, close:add */ { 0, 255 }, /* old= cw, write:xxx, read: 0, close:del */ { 0, 255 }, /* old= cw, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old= cw, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old= cw, write:xxx, read:add, close:add */ { 0, 255 }, /* old= cw, write:xxx, read:add, close:del */ { 0, 255 }, /* old= cw, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old= cw, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old= cw, write:xxx, read:del, close:add */ { 0, 255 }, /* old= cw, write:xxx, read:del, close:del */ { 0, 255 }, /* old= cw, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old= cw, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old= cw, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old= cw, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old= cw, write:xxx, read:xxx, close:xxx */ { 0, 255 }, /* old=crw, write: 0, read: 0, close: 0 */ { 0, 0 }, /* old=crw, write: 0, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write: 0, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write: 0, read: 0, close:xxx */ { 0, 255 }, /* old=crw, write: 0, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:add, close:xxx */ { 0, 255 }, /* old=crw, write: 0, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write: 0, read:del, close:xxx */ { 0, 255 }, /* old=crw, write: 0, read:xxx, close: 0 */ { 0, 255 }, /* old=crw, write: 0, read:xxx, close:add */ { 0, 255 }, /* old=crw, write: 0, read:xxx, close:del */ { 0, 255 }, /* old=crw, write: 0, read:xxx, close:xxx */ { 0, 255 }, /* old=crw, write:add, read: 0, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read: 0, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read: 0, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write:add, read: 0, close:xxx */ { 0, 255 }, /* old=crw, write:add, read:add, close: 0 */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read:add, close:add */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read:add, close:del */ { EPOLLIN|EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write:add, read:add, close:xxx */ { 0, 255 }, /* old=crw, write:add, read:del, close: 0 */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read:del, close:add */ { EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:add, read:del, close:del */ { EPOLLOUT, EPOLL_CTL_MOD }, /* old=crw, write:add, read:del, close:xxx */ { 0, 255 }, /* old=crw, write:add, read:xxx, close: 0 */ { 0, 255 }, /* old=crw, write:add, read:xxx, close:add */ { 0, 255 }, /* old=crw, write:add, read:xxx, close:del */ { 0, 255 }, /* old=crw, write:add, read:xxx, close:xxx */ { 0, 255 }, /* old=crw, write:del, read: 0, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read: 0, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read: 0, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old=crw, write:del, read: 0, close:xxx */ { 0, 255 }, /* old=crw, write:del, read:add, close: 0 */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read:add, close:add */ { EPOLLIN|EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read:add, close:del */ { EPOLLIN, EPOLL_CTL_MOD }, /* old=crw, write:del, read:add, close:xxx */ { 0, 255 }, /* old=crw, write:del, read:del, close: 0 */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read:del, close:add */ { EPOLLRDHUP, EPOLL_CTL_MOD }, /* old=crw, write:del, read:del, close:del */ { EPOLLIN|EPOLLOUT|EPOLLRDHUP, EPOLL_CTL_DEL }, /* old=crw, write:del, read:del, close:xxx */ { 0, 255 }, /* old=crw, write:del, read:xxx, close: 0 */ { 0, 255 }, /* old=crw, write:del, read:xxx, close:add */ { 0, 255 }, /* old=crw, write:del, read:xxx, close:del */ { 0, 255 }, /* old=crw, write:del, read:xxx, close:xxx */ { 0, 255 }, /* old=crw, write:xxx, read: 0, close: 0 */ { 0, 255 }, /* old=crw, write:xxx, read: 0, close:add */ { 0, 255 }, /* old=crw, write:xxx, read: 0, close:del */ { 0, 255 }, /* old=crw, write:xxx, read: 0, close:xxx */ { 0, 255 }, /* old=crw, write:xxx, read:add, close: 0 */ { 0, 255 }, /* old=crw, write:xxx, read:add, close:add */ { 0, 255 }, /* old=crw, write:xxx, read:add, close:del */ { 0, 255 }, /* old=crw, write:xxx, read:add, close:xxx */ { 0, 255 }, /* old=crw, write:xxx, read:del, close: 0 */ { 0, 255 }, /* old=crw, write:xxx, read:del, close:add */ { 0, 255 }, /* old=crw, write:xxx, read:del, close:del */ { 0, 255 }, /* old=crw, write:xxx, read:del, close:xxx */ { 0, 255 }, /* old=crw, write:xxx, read:xxx, close: 0 */ { 0, 255 }, /* old=crw, write:xxx, read:xxx, close:add */ { 0, 255 }, /* old=crw, write:xxx, read:xxx, close:del */ { 0, 255 }, /* old=crw, write:xxx, read:xxx, close:xxx */ { 0, 255 }, }; #endif /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVBUFFER_INTERNAL_H_INCLUDED_ #define EVBUFFER_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "evconfig-private.h" #include "event2/util.h" #include "event2/event_struct.h" #include "util-internal.h" #include "defer-internal.h" /* Experimental cb flag: "never deferred." Implementation note: * these callbacks may get an inaccurate view of n_del/n_added in their * arguments. */ #define EVBUFFER_CB_NODEFER 2 #ifdef _WIN32 #include #endif #include /* Minimum allocation for a chain. We define this so that we're burning no * more than 5% of each allocation on overhead. It would be nice to lose even * less space, though. */ #if EVENT__SIZEOF_VOID_P < 8 #define MIN_BUFFER_SIZE 512 #else #define MIN_BUFFER_SIZE 1024 #endif /** A single evbuffer callback for an evbuffer. This function will be invoked * when bytes are added to or removed from the evbuffer. */ struct evbuffer_cb_entry { /** Structures to implement a doubly-linked queue of callbacks */ LIST_ENTRY(evbuffer_cb_entry) next; /** The callback function to invoke when this callback is called. If EVBUFFER_CB_OBSOLETE is set in flags, the cb_obsolete field is valid; otherwise, cb_func is valid. */ union { evbuffer_cb_func cb_func; evbuffer_cb cb_obsolete; } cb; /** Argument to pass to cb. */ void *cbarg; /** Currently set flags on this callback. */ ev_uint32_t flags; }; struct bufferevent; struct evbuffer_chain; struct evbuffer { /** The first chain in this buffer's linked list of chains. */ struct evbuffer_chain *first; /** The last chain in this buffer's linked list of chains. */ struct evbuffer_chain *last; /** Pointer to the next pointer pointing at the 'last_with_data' chain. * * To unpack: * * The last_with_data chain is the last chain that has any data in it. * If all chains in the buffer are empty, it is the first chain. * If the buffer has no chains, it is NULL. * * The last_with_datap pointer points at _whatever 'next' pointer_ * pointing at the last_with_data chain. If the last_with_data chain * is the first chain, or it is NULL, then the last_with_datap pointer * is &buf->first. */ struct evbuffer_chain **last_with_datap; /** Total amount of bytes stored in all chains.*/ size_t total_len; /** Number of bytes we have added to the buffer since we last tried to * invoke callbacks. */ size_t n_add_for_cb; /** Number of bytes we have removed from the buffer since we last * tried to invoke callbacks. */ size_t n_del_for_cb; #ifndef EVENT__DISABLE_THREAD_SUPPORT /** A lock used to mediate access to this buffer. */ void *lock; #endif /** True iff we should free the lock field when we free this * evbuffer. */ unsigned own_lock : 1; /** True iff we should not allow changes to the front of the buffer * (drains or prepends). */ unsigned freeze_start : 1; /** True iff we should not allow changes to the end of the buffer * (appends) */ unsigned freeze_end : 1; /** True iff this evbuffer's callbacks are not invoked immediately * upon a change in the buffer, but instead are deferred to be invoked * from the event_base's loop. Useful for preventing enormous stack * overflows when we have mutually recursive callbacks, and for * serializing callbacks in a single thread. */ unsigned deferred_cbs : 1; #ifdef _WIN32 /** True iff this buffer is set up for overlapped IO. */ unsigned is_overlapped : 1; #endif /** Zero or more EVBUFFER_FLAG_* bits */ ev_uint32_t flags; /** Used to implement deferred callbacks. */ struct event_base *cb_queue; /** A reference count on this evbuffer. When the reference count * reaches 0, the buffer is destroyed. Manipulated with * evbuffer_incref and evbuffer_decref_and_unlock and * evbuffer_free. */ int refcnt; /** A struct event_callback handle to make all of this buffer's callbacks * invoked from the event loop. */ struct event_callback deferred; /** A doubly-linked-list of callback functions */ LIST_HEAD(evbuffer_cb_queue, evbuffer_cb_entry) callbacks; /** The parent bufferevent object this evbuffer belongs to. * NULL if the evbuffer stands alone. */ struct bufferevent *parent; }; #if EVENT__SIZEOF_OFF_T < EVENT__SIZEOF_SIZE_T typedef ev_ssize_t ev_misalign_t; #define EVBUFFER_CHAIN_MAX ((size_t)EV_SSIZE_MAX) #else typedef ev_off_t ev_misalign_t; #if EVENT__SIZEOF_OFF_T > EVENT__SIZEOF_SIZE_T #define EVBUFFER_CHAIN_MAX EV_SIZE_MAX #else #define EVBUFFER_CHAIN_MAX ((size_t)EV_SSIZE_MAX) #endif #endif /** A single item in an evbuffer. */ struct evbuffer_chain { /** points to next buffer in the chain */ struct evbuffer_chain *next; /** total allocation available in the buffer field. */ size_t buffer_len; /** unused space at the beginning of buffer or an offset into a * file for sendfile buffers. */ ev_misalign_t misalign; /** Offset into buffer + misalign at which to start writing. * In other words, the total number of bytes actually stored * in buffer. */ size_t off; /** Set if special handling is required for this chain */ unsigned flags; #define EVBUFFER_FILESEGMENT 0x0001 /**< A chain used for a file segment */ #define EVBUFFER_SENDFILE 0x0002 /**< a chain used with sendfile */ #define EVBUFFER_REFERENCE 0x0004 /**< a chain with a mem reference */ #define EVBUFFER_IMMUTABLE 0x0008 /**< read-only chain */ /** a chain that mustn't be reallocated or freed, or have its contents * memmoved, until the chain is un-pinned. */ #define EVBUFFER_MEM_PINNED_R 0x0010 #define EVBUFFER_MEM_PINNED_W 0x0020 #define EVBUFFER_MEM_PINNED_ANY (EVBUFFER_MEM_PINNED_R|EVBUFFER_MEM_PINNED_W) /** a chain that should be freed, but can't be freed until it is * un-pinned. */ #define EVBUFFER_DANGLING 0x0040 /** a chain that is a referenced copy of another chain */ #define EVBUFFER_MULTICAST 0x0080 /** number of references to this chain */ int refcnt; /** Usually points to the read-write memory belonging to this * buffer allocated as part of the evbuffer_chain allocation. * For mmap, this can be a read-only buffer and * EVBUFFER_IMMUTABLE will be set in flags. For sendfile, it * may point to NULL. */ unsigned char *buffer; }; /** callback for a reference chain; lets us know what to do with it when * we're done with it. Lives at the end of an evbuffer_chain with the * EVBUFFER_REFERENCE flag set */ struct evbuffer_chain_reference { evbuffer_ref_cleanup_cb cleanupfn; void *extra; }; /** File segment for a file-segment chain. Lives at the end of an * evbuffer_chain with the EVBUFFER_FILESEGMENT flag set. */ struct evbuffer_chain_file_segment { struct evbuffer_file_segment *segment; #ifdef _WIN32 /** If we're using CreateFileMapping, this is the handle to the view. */ HANDLE view_handle; #endif }; /* Declared in event2/buffer.h; defined here. */ struct evbuffer_file_segment { void *lock; /**< lock prevent concurrent access to refcnt */ int refcnt; /**< Reference count for this file segment */ unsigned flags; /**< combination of EVBUF_FS_* flags */ /** What kind of file segment is this? */ unsigned can_sendfile : 1; unsigned is_mapping : 1; /** The fd that we read the data from. */ int fd; /** If we're using mmap, this is the raw mapped memory. */ void *mapping; #ifdef _WIN32 /** If we're using CreateFileMapping, this is the mapping */ HANDLE mapping_handle; #endif /** If we're using mmap or IO, this is the content of the file * segment. */ char *contents; /** Position of this segment within the file. */ ev_off_t file_offset; /** If we're using mmap, this is the offset within 'mapping' where * this data segment begins. */ ev_off_t mmap_offset; /** The length of this segment. */ ev_off_t length; /** Cleanup callback function */ evbuffer_file_segment_cleanup_cb cleanup_cb; /** Argument to be pass to cleanup callback function */ void *cleanup_cb_arg; }; /** Information about the multicast parent of a chain. Lives at the * end of an evbuffer_chain with the EVBUFFER_MULTICAST flag set. */ struct evbuffer_multicast_parent { /** source buffer the multicast parent belongs to */ struct evbuffer *source; /** multicast parent for this chain */ struct evbuffer_chain *parent; }; #define EVBUFFER_CHAIN_SIZE sizeof(struct evbuffer_chain) /** Return a pointer to extra data allocated along with an evbuffer. */ #define EVBUFFER_CHAIN_EXTRA(t, c) (t *)((struct evbuffer_chain *)(c) + 1) /** Assert that we are holding the lock on an evbuffer */ #define ASSERT_EVBUFFER_LOCKED(buffer) \ EVLOCK_ASSERT_LOCKED((buffer)->lock) #define EVBUFFER_LOCK(buffer) \ do { \ EVLOCK_LOCK((buffer)->lock, 0); \ } while (0) #define EVBUFFER_UNLOCK(buffer) \ do { \ EVLOCK_UNLOCK((buffer)->lock, 0); \ } while (0) #define EVBUFFER_LOCK2(buffer1, buffer2) \ do { \ EVLOCK_LOCK2((buffer1)->lock, (buffer2)->lock, 0, 0); \ } while (0) #define EVBUFFER_UNLOCK2(buffer1, buffer2) \ do { \ EVLOCK_UNLOCK2((buffer1)->lock, (buffer2)->lock, 0, 0); \ } while (0) /** Increase the reference count of buf by one. */ void evbuffer_incref_(struct evbuffer *buf); /** Increase the reference count of buf by one and acquire the lock. */ void evbuffer_incref_and_lock_(struct evbuffer *buf); /** Pin a single buffer chain using a given flag. A pinned chunk may not be * moved or freed until it is unpinned. */ void evbuffer_chain_pin_(struct evbuffer_chain *chain, unsigned flag); /** Unpin a single buffer chain using a given flag. */ void evbuffer_chain_unpin_(struct evbuffer_chain *chain, unsigned flag); /** As evbuffer_free, but requires that we hold a lock on the buffer, and * releases the lock before freeing it and the buffer. */ void evbuffer_decref_and_unlock_(struct evbuffer *buffer); /** As evbuffer_expand, but does not guarantee that the newly allocated memory * is contiguous. Instead, it may be split across two or more chunks. */ int evbuffer_expand_fast_(struct evbuffer *, size_t, int); /** Helper: prepares for a readv/WSARecv call by expanding the buffer to * hold enough memory to read 'howmuch' bytes in possibly noncontiguous memory. * Sets up the one or two iovecs in 'vecs' to point to the free memory and its * extent, and *chainp to point to the first chain that we'll try to read into. * Returns the number of vecs used. */ int evbuffer_read_setup_vecs_(struct evbuffer *buf, ev_ssize_t howmuch, struct evbuffer_iovec *vecs, int n_vecs, struct evbuffer_chain ***chainp, int exact); /* Helper macro: copies an evbuffer_iovec in ei to a win32 WSABUF in i. */ #define WSABUF_FROM_EVBUFFER_IOV(i,ei) do { \ (i)->buf = (ei)->iov_base; \ (i)->len = (unsigned long)(ei)->iov_len; \ } while (0) /* XXXX the cast above is safe for now, but not if we allow mmaps on win64. * See note in buffer_iocp's launch_write function */ /** Set the parent bufferevent object for buf to bev */ void evbuffer_set_parent_(struct evbuffer *buf, struct bufferevent *bev); void evbuffer_invoke_callbacks_(struct evbuffer *buf); int evbuffer_get_callbacks_(struct evbuffer *buffer, struct event_callback **cbs, int max_cbs); #ifdef __cplusplus } #endif #endif /* EVBUFFER_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVENT_INTERNAL_H_INCLUDED_ #define EVENT_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "evconfig-private.h" #include #include #include "event2/event_struct.h" #include "minheap-internal.h" #include "evsignal-internal.h" #include "mm-internal.h" #include "defer-internal.h" /* map union members back */ /* mutually exclusive */ #define ev_signal_next ev_.ev_signal.ev_signal_next #define ev_io_next ev_.ev_io.ev_io_next #define ev_io_timeout ev_.ev_io.ev_timeout /* used only by signals */ #define ev_ncalls ev_.ev_signal.ev_ncalls #define ev_pncalls ev_.ev_signal.ev_pncalls #define ev_pri ev_evcallback.evcb_pri #define ev_flags ev_evcallback.evcb_flags #define ev_closure ev_evcallback.evcb_closure #define ev_callback ev_evcallback.evcb_cb_union.evcb_callback #define ev_arg ev_evcallback.evcb_arg /** @name Event closure codes Possible values for evcb_closure in struct event_callback @{ */ /** A regular event. Uses the evcb_callback callback */ #define EV_CLOSURE_EVENT 0 /** A signal event. Uses the evcb_callback callback */ #define EV_CLOSURE_EVENT_SIGNAL 1 /** A persistent non-signal event. Uses the evcb_callback callback */ #define EV_CLOSURE_EVENT_PERSIST 2 /** A simple callback. Uses the evcb_selfcb callback. */ #define EV_CLOSURE_CB_SELF 3 /** A finalizing callback. Uses the evcb_cbfinalize callback. */ #define EV_CLOSURE_CB_FINALIZE 4 /** A finalizing event. Uses the evcb_evfinalize callback. */ #define EV_CLOSURE_EVENT_FINALIZE 5 /** A finalizing event that should get freed after. Uses the evcb_evfinalize * callback. */ #define EV_CLOSURE_EVENT_FINALIZE_FREE 6 /** @} */ /** Structure to define the backend of a given event_base. */ struct eventop { /** The name of this backend. */ const char *name; /** Function to set up an event_base to use this backend. It should * create a new structure holding whatever information is needed to * run the backend, and return it. The returned pointer will get * stored by event_init into the event_base.evbase field. On failure, * this function should return NULL. */ void *(*init)(struct event_base *); /** Enable reading/writing on a given fd or signal. 'events' will be * the events that we're trying to enable: one or more of EV_READ, * EV_WRITE, EV_SIGNAL, and EV_ET. 'old' will be those events that * were enabled on this fd previously. 'fdinfo' will be a structure * associated with the fd by the evmap; its size is defined by the * fdinfo field below. It will be set to 0 the first time the fd is * added. The function should return 0 on success and -1 on error. */ int (*add)(struct event_base *, evutil_socket_t fd, short old, short events, void *fdinfo); /** As "add", except 'events' contains the events we mean to disable. */ int (*del)(struct event_base *, evutil_socket_t fd, short old, short events, void *fdinfo); /** Function to implement the core of an event loop. It must see which added events are ready, and cause event_active to be called for each active event (usually via event_io_active or such). It should return 0 on success and -1 on error. */ int (*dispatch)(struct event_base *, struct timeval *); /** Function to clean up and free our data from the event_base. */ void (*dealloc)(struct event_base *); /** Flag: set if we need to reinitialize the event base after we fork. */ int need_reinit; /** Bit-array of supported event_method_features that this backend can * provide. */ enum event_method_feature features; /** Length of the extra information we should record for each fd that has one or more active events. This information is recorded as part of the evmap entry for each fd, and passed as an argument to the add and del functions above. */ size_t fdinfo_len; }; #ifdef _WIN32 /* If we're on win32, then file descriptors are not nice low densely packed integers. Instead, they are pointer-like windows handles, and we want to use a hashtable instead of an array to map fds to events. */ #define EVMAP_USE_HT #endif /* #define HT_CACHE_HASH_VALS */ #ifdef EVMAP_USE_HT #define HT_NO_CACHE_HASH_VALUES #include "ht-internal.h" struct event_map_entry; HT_HEAD(event_io_map, event_map_entry); #else #define event_io_map event_signal_map #endif /* Used to map signal numbers to a list of events. If EVMAP_USE_HT is not defined, this structure is also used as event_io_map, which maps fds to a list of events. */ struct event_signal_map { /* An array of evmap_io * or of evmap_signal *; empty entries are * set to NULL. */ void **entries; /* The number of entries available in entries */ int nentries; }; /* A list of events waiting on a given 'common' timeout value. Ordinarily, * events waiting for a timeout wait on a minheap. Sometimes, however, a * queue can be faster. **/ struct common_timeout_list { /* List of events currently waiting in the queue. */ struct event_list events; /* 'magic' timeval used to indicate the duration of events in this * queue. */ struct timeval duration; /* Event that triggers whenever one of the events in the queue is * ready to activate */ struct event timeout_event; /* The event_base that this timeout list is part of */ struct event_base *base; }; /** Mask used to get the real tv_usec value from a common timeout. */ #define COMMON_TIMEOUT_MICROSECONDS_MASK 0x000fffff struct event_change; /* List of 'changes' since the last call to eventop.dispatch. Only maintained * if the backend is using changesets. */ struct event_changelist { struct event_change *changes; int n_changes; int changes_size; }; #ifndef EVENT__DISABLE_DEBUG_MODE /* Global internal flag: set to one if debug mode is on. */ extern int event_debug_mode_on_; #define EVENT_DEBUG_MODE_IS_ON() (event_debug_mode_on_) #else #define EVENT_DEBUG_MODE_IS_ON() (0) #endif TAILQ_HEAD(evcallback_list, event_callback); /* Sets up an event for processing once */ struct event_once { LIST_ENTRY(event_once) next_once; struct event ev; void (*cb)(evutil_socket_t, short, void *); void *arg; }; struct event_base { /** Function pointers and other data to describe this event_base's * backend. */ const struct eventop *evsel; /** Pointer to backend-specific data. */ void *evbase; /** List of changes to tell backend about at next dispatch. Only used * by the O(1) backends. */ struct event_changelist changelist; /** Function pointers used to describe the backend that this event_base * uses for signals */ const struct eventop *evsigsel; /** Data to implement the common signal handler code. */ struct evsig_info sig; /** Number of virtual events */ int virtual_event_count; /** Maximum number of virtual events active */ int virtual_event_count_max; /** Number of total events added to this event_base */ int event_count; /** Maximum number of total events added to this event_base */ int event_count_max; /** Number of total events active in this event_base */ int event_count_active; /** Maximum number of total events active in this event_base */ int event_count_active_max; /** Set if we should terminate the loop once we're done processing * events. */ int event_gotterm; /** Set if we should terminate the loop immediately */ int event_break; /** Set if we should start a new instance of the loop immediately. */ int event_continue; /** The currently running priority of events */ int event_running_priority; /** Set if we're running the event_base_loop function, to prevent * reentrant invocation. */ int running_loop; /** Set to the number of deferred_cbs we've made 'active' in the * loop. This is a hack to prevent starvation; it would be smarter * to just use event_config_set_max_dispatch_interval's max_callbacks * feature */ int n_deferreds_queued; /* Active event management. */ /** An array of nactivequeues queues for active event_callbacks (ones * that have triggered, and whose callbacks need to be called). Low * priority numbers are more important, and stall higher ones. */ struct evcallback_list *activequeues; /** The length of the activequeues array */ int nactivequeues; /** A list of event_callbacks that should become active the next time * we process events, but not this time. */ struct evcallback_list active_later_queue; /* common timeout logic */ /** An array of common_timeout_list* for all of the common timeout * values we know. */ struct common_timeout_list **common_timeout_queues; /** The number of entries used in common_timeout_queues */ int n_common_timeouts; /** The total size of common_timeout_queues. */ int n_common_timeouts_allocated; /** Mapping from file descriptors to enabled (added) events */ struct event_io_map io; /** Mapping from signal numbers to enabled (added) events. */ struct event_signal_map sigmap; /** Priority queue of events with timeouts. */ struct min_heap timeheap; /** Stored timeval: used to avoid calling gettimeofday/clock_gettime * too often. */ struct timeval tv_cache; struct evutil_monotonic_timer monotonic_timer; /** Difference between internal time (maybe from clock_gettime) and * gettimeofday. */ struct timeval tv_clock_diff; /** Second in which we last updated tv_clock_diff, in monotonic time. */ time_t last_updated_clock_diff; #ifndef EVENT__DISABLE_THREAD_SUPPORT /* threading support */ /** The thread currently running the event_loop for this base */ unsigned long th_owner_id; /** A lock to prevent conflicting accesses to this event_base */ void *th_base_lock; /** A condition that gets signalled when we're done processing an * event with waiters on it. */ void *current_event_cond; /** Number of threads blocking on current_event_cond. */ int current_event_waiters; #endif /** The event whose callback is executing right now */ struct event_callback *current_event; #ifdef _WIN32 /** IOCP support structure, if IOCP is enabled. */ struct event_iocp_port *iocp; #endif /** Flags that this base was configured with */ enum event_base_config_flag flags; struct timeval max_dispatch_time; int max_dispatch_callbacks; int limit_callbacks_after_prio; /* Notify main thread to wake up break, etc. */ /** True if the base already has a pending notify, and we don't need * to add any more. */ int is_notify_pending; /** A socketpair used by some th_notify functions to wake up the main * thread. */ evutil_socket_t th_notify_fd[2]; /** An event used by some th_notify functions to wake up the main * thread. */ struct event th_notify; /** A function used to wake up the main thread from another thread. */ int (*th_notify_fn)(struct event_base *base); /** Saved seed for weak random number generator. Some backends use * this to produce fairness among sockets. Protected by th_base_lock. */ struct evutil_weakrand_state weakrand_seed; /** List of event_onces that have not yet fired. */ LIST_HEAD(once_event_list, event_once) once_events; }; struct event_config_entry { TAILQ_ENTRY(event_config_entry) next; const char *avoid_method; }; /** Internal structure: describes the configuration we want for an event_base * that we're about to allocate. */ struct event_config { TAILQ_HEAD(event_configq, event_config_entry) entries; int n_cpus_hint; struct timeval max_dispatch_interval; int max_dispatch_callbacks; int limit_callbacks_after_prio; enum event_method_feature require_features; enum event_base_config_flag flags; }; /* Internal use only: Functions that might be missing from */ #ifndef LIST_END #define LIST_END(head) NULL #endif #ifndef TAILQ_FIRST #define TAILQ_FIRST(head) ((head)->tqh_first) #endif #ifndef TAILQ_END #define TAILQ_END(head) NULL #endif #ifndef TAILQ_NEXT #define TAILQ_NEXT(elm, field) ((elm)->field.tqe_next) #endif #ifndef TAILQ_FOREACH #define TAILQ_FOREACH(var, head, field) \ for ((var) = TAILQ_FIRST(head); \ (var) != TAILQ_END(head); \ (var) = TAILQ_NEXT(var, field)) #endif #ifndef TAILQ_INSERT_BEFORE #define TAILQ_INSERT_BEFORE(listelm, elm, field) do { \ (elm)->field.tqe_prev = (listelm)->field.tqe_prev; \ (elm)->field.tqe_next = (listelm); \ *(listelm)->field.tqe_prev = (elm); \ (listelm)->field.tqe_prev = &(elm)->field.tqe_next; \ } while (0) #endif #define N_ACTIVE_CALLBACKS(base) \ ((base)->event_count_active) int evsig_set_handler_(struct event_base *base, int evsignal, void (*fn)(int)); int evsig_restore_handler_(struct event_base *base, int evsignal); int event_add_nolock_(struct event *ev, const struct timeval *tv, int tv_is_absolute); /** Argument for event_del_nolock_. Tells event_del not to block on the event * if it's running in another thread. */ #define EVENT_DEL_NOBLOCK 0 /** Argument for event_del_nolock_. Tells event_del to block on the event * if it's running in another thread, regardless of its value for EV_FINALIZE */ #define EVENT_DEL_BLOCK 1 /** Argument for event_del_nolock_. Tells event_del to block on the event * if it is running in another thread and it doesn't have EV_FINALIZE set. */ #define EVENT_DEL_AUTOBLOCK 2 /** Argument for event_del_nolock_. Tells event_del to proceed even if the * event is set up for finalization rather for regular use.*/ #define EVENT_DEL_EVEN_IF_FINALIZING 3 int event_del_nolock_(struct event *ev, int blocking); int event_remove_timer_nolock_(struct event *ev); void event_active_nolock_(struct event *ev, int res, short count); EVENT2_EXPORT_SYMBOL int event_callback_activate_(struct event_base *, struct event_callback *); int event_callback_activate_nolock_(struct event_base *, struct event_callback *); int event_callback_cancel_(struct event_base *base, struct event_callback *evcb); void event_callback_finalize_nolock_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *)); EVENT2_EXPORT_SYMBOL void event_callback_finalize_(struct event_base *base, unsigned flags, struct event_callback *evcb, void (*cb)(struct event_callback *, void *)); int event_callback_finalize_many_(struct event_base *base, int n_cbs, struct event_callback **evcb, void (*cb)(struct event_callback *, void *)); EVENT2_EXPORT_SYMBOL void event_active_later_(struct event *ev, int res); void event_active_later_nolock_(struct event *ev, int res); int event_callback_activate_later_nolock_(struct event_base *base, struct event_callback *evcb); int event_callback_cancel_nolock_(struct event_base *base, struct event_callback *evcb, int even_if_finalizing); void event_callback_init_(struct event_base *base, struct event_callback *cb); /* FIXME document. */ EVENT2_EXPORT_SYMBOL void event_base_add_virtual_(struct event_base *base); void event_base_del_virtual_(struct event_base *base); /** For debugging: unless assertions are disabled, verify the referential integrity of the internal data structures of 'base'. This operation can be expensive. Returns on success; aborts on failure. */ EVENT2_EXPORT_SYMBOL void event_base_assert_ok_(struct event_base *base); void event_base_assert_ok_nolock_(struct event_base *base); /* Helper function: Call 'fn' exactly once every inserted or active event in * the event_base 'base'. * * If fn returns 0, continue on to the next event. Otherwise, return the same * value that fn returned. * * Requires that 'base' be locked. */ int event_base_foreach_event_nolock_(struct event_base *base, event_base_foreach_event_cb cb, void *arg); /* Cleanup function to reset debug mode during shutdown. * * Calling this function doesn't mean it'll be possible to re-enable * debug mode if any events were added. */ void event_disable_debug_mode(void); #ifdef __cplusplus } #endif #endif /* EVENT_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVMAP_INTERNAL_H_INCLUDED_ #define EVMAP_INTERNAL_H_INCLUDED_ /** @file evmap-internal.h * * An event_map is a utility structure to map each fd or signal to zero or * more events. Functions to manipulate event_maps should only be used from * inside libevent. They generally need to hold the lock on the corresponding * event_base. **/ struct event_base; struct event; /** Initialize an event_map for use. */ void evmap_io_initmap_(struct event_io_map* ctx); void evmap_signal_initmap_(struct event_signal_map* ctx); /** Remove all entries from an event_map. @param ctx the map to clear. */ void evmap_io_clear_(struct event_io_map* ctx); void evmap_signal_clear_(struct event_signal_map* ctx); /** Add an IO event (some combination of EV_READ or EV_WRITE) to an event_base's list of events on a given file descriptor, and tell the underlying eventops about the fd if its state has changed. Requires that ev is not already added. @param base the event_base to operate on. @param fd the file descriptor corresponding to ev. @param ev the event to add. */ int evmap_io_add_(struct event_base *base, evutil_socket_t fd, struct event *ev); /** Remove an IO event (some combination of EV_READ or EV_WRITE) to an event_base's list of events on a given file descriptor, and tell the underlying eventops about the fd if its state has changed. @param base the event_base to operate on. @param fd the file descriptor corresponding to ev. @param ev the event to remove. */ int evmap_io_del_(struct event_base *base, evutil_socket_t fd, struct event *ev); /** Active the set of events waiting on an event_base for a given fd. @param base the event_base to operate on. @param fd the file descriptor that has become active. @param events a bitmask of EV_READ|EV_WRITE|EV_ET. */ void evmap_io_active_(struct event_base *base, evutil_socket_t fd, short events); /* These functions behave in the same way as evmap_io_*, except they work on * signals rather than fds. signals use a linear map everywhere; fds use * either a linear map or a hashtable. */ int evmap_signal_add_(struct event_base *base, int signum, struct event *ev); int evmap_signal_del_(struct event_base *base, int signum, struct event *ev); void evmap_signal_active_(struct event_base *base, evutil_socket_t signum, int ncalls); /* Return the fdinfo object associated with a given fd. If the fd has no * events associated with it, the result may be NULL. */ void *evmap_io_get_fdinfo_(struct event_io_map *ctx, evutil_socket_t fd); /* Helper for event_reinit(): Tell the backend to re-add every fd and signal * for which we have a pending event. */ int evmap_reinit_(struct event_base *base); /* Helper for event_base_free(): Call event_del() on every pending fd and * signal event. */ void evmap_delete_all_(struct event_base *base); /* Helper for event_base_assert_ok_(): Check referential integrity of the * evmaps. */ void evmap_check_integrity_(struct event_base *base); /* Helper: Call fn on every fd or signal event, passing as its arguments the * provided event_base, the event, and arg. If fn returns 0, process the next * event. If it returns any other value, return that value and process no * more events. */ int evmap_foreach_event_(struct event_base *base, event_base_foreach_event_cb fn, void *arg); #endif /* EVMAP_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2006-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVRPC_INTERNAL_H_INCLUDED_ #define EVRPC_INTERNAL_H_INCLUDED_ #include "event2/http.h" #include "http-internal.h" struct evrpc; struct evrpc_request_wrapper; #define EVRPC_URI_PREFIX "/.rpc." struct evrpc_hook { TAILQ_ENTRY(evrpc_hook) next; /* returns EVRPC_TERMINATE; if the rpc should be aborted. * a hook is is allowed to rewrite the evbuffer */ int (*process)(void *, struct evhttp_request *, struct evbuffer *, void *); void *process_arg; }; TAILQ_HEAD(evrpc_hook_list, evrpc_hook); /* * this is shared between the base and the pool, so that we can reuse * the hook adding functions; we alias both evrpc_pool and evrpc_base * to this common structure. */ struct evrpc_hook_ctx; TAILQ_HEAD(evrpc_pause_list, evrpc_hook_ctx); struct evrpc_hooks_ { /* hooks for processing outbound and inbound rpcs */ struct evrpc_hook_list in_hooks; struct evrpc_hook_list out_hooks; struct evrpc_pause_list pause_requests; }; #define input_hooks common.in_hooks #define output_hooks common.out_hooks #define paused_requests common.pause_requests struct evrpc_base { struct evrpc_hooks_ common; /* the HTTP server under which we register our RPC calls */ struct evhttp* http_server; /* a list of all RPCs registered with us */ TAILQ_HEAD(evrpc_list, evrpc) registered_rpcs; }; struct evrpc_req_generic; void evrpc_reqstate_free_(struct evrpc_req_generic* rpc_state); /* A pool for holding evhttp_connection objects */ struct evrpc_pool { struct evrpc_hooks_ common; struct event_base *base; struct evconq connections; int timeout; TAILQ_HEAD(evrpc_requestq, evrpc_request_wrapper) (requests); }; struct evrpc_hook_ctx { TAILQ_ENTRY(evrpc_hook_ctx) next; void *ctx; void (*cb)(void *, enum EVRPC_HOOK_RESULT); }; struct evrpc_meta { TAILQ_ENTRY(evrpc_meta) next; char *key; void *data; size_t data_size; }; TAILQ_HEAD(evrpc_meta_list, evrpc_meta); struct evrpc_hook_meta { struct evrpc_meta_list meta_data; struct evhttp_connection *evcon; }; /* allows association of meta data with a request */ static void evrpc_hook_associate_meta_(struct evrpc_hook_meta **pctx, struct evhttp_connection *evcon); /* creates a new meta data store */ static struct evrpc_hook_meta *evrpc_hook_meta_new_(void); /* frees the meta data associated with a request */ static void evrpc_hook_context_free_(struct evrpc_hook_meta *ctx); /* the server side of an rpc */ /* We alias the RPC specific structs to this voided one */ struct evrpc_req_generic { /* * allows association of meta data via hooks - needs to be * synchronized with evrpc_request_wrapper */ struct evrpc_hook_meta *hook_meta; /* the unmarshaled request object */ void *request; /* the empty reply object that needs to be filled in */ void *reply; /* * the static structure for this rpc; that can be used to * automatically unmarshal and marshal the http buffers. */ struct evrpc *rpc; /* * the http request structure on which we need to answer. */ struct evhttp_request* http_req; /* * Temporary data store for marshaled data */ struct evbuffer* rpc_data; }; /* the client side of an rpc request */ struct evrpc_request_wrapper { /* * allows association of meta data via hooks - needs to be * synchronized with evrpc_req_generic. */ struct evrpc_hook_meta *hook_meta; TAILQ_ENTRY(evrpc_request_wrapper) next; /* pool on which this rpc request is being made */ struct evrpc_pool *pool; /* connection on which the request is being sent */ struct evhttp_connection *evcon; /* the actual request */ struct evhttp_request *req; /* event for implementing request timeouts */ struct event ev_timeout; /* the name of the rpc */ char *name; /* callback */ void (*cb)(struct evrpc_status*, void *request, void *reply, void *arg); void *cb_arg; void *request; void *reply; /* unmarshals the buffer into the proper request structure */ void (*request_marshal)(struct evbuffer *, void *); /* removes all stored state in the reply */ void (*reply_clear)(void *); /* marshals the reply into a buffer */ int (*reply_unmarshal)(void *, struct evbuffer*); }; #endif /* EVRPC_INTERNAL_H_INCLUDED_ */ /* * Copyright 2000-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVSIGNAL_INTERNAL_H_INCLUDED_ #define EVSIGNAL_INTERNAL_H_INCLUDED_ #ifndef evutil_socket_t #include "event2/util.h" #endif #include typedef void (*ev_sighandler_t)(int); /* Data structure for the default signal-handling implementation in signal.c */ struct evsig_info { /* Event watching ev_signal_pair[1] */ struct event ev_signal; /* Socketpair used to send notifications from the signal handler */ evutil_socket_t ev_signal_pair[2]; /* True iff we've added the ev_signal event yet. */ int ev_signal_added; /* Count of the number of signals we're currently watching. */ int ev_n_signals_added; /* Array of previous signal handler objects before Libevent started * messing with them. Used to restore old signal handlers. */ #ifdef EVENT__HAVE_SIGACTION struct sigaction **sh_old; #else ev_sighandler_t **sh_old; #endif /* Size of sh_old. */ int sh_old_max; }; int evsig_init_(struct event_base *); void evsig_dealloc_(struct event_base *); void evsig_set_base_(struct event_base *base); void evsig_free_globals_(void); #endif /* EVSIGNAL_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2008-2012 Niels Provos, Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef EVTHREAD_INTERNAL_H_INCLUDED_ #define EVTHREAD_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "evconfig-private.h" #include "event2/thread.h" #include "util-internal.h" struct event_base; #if !defined(_WIN32) && !defined(__CYGWIN__) /* On Windows, the way we currently make DLLs, it's not allowed for us to * have shared global structures. Thus, we only do the direct-call-to-function * code path if we know that the local shared library system supports it. */ #define EVTHREAD_EXPOSE_STRUCTS #endif #if ! defined(EVENT__DISABLE_THREAD_SUPPORT) && defined(EVTHREAD_EXPOSE_STRUCTS) /* Global function pointers to lock-related functions. NULL if locking isn't enabled. */ EVENT2_EXPORT_SYMBOL extern struct evthread_lock_callbacks evthread_lock_fns_; EVENT2_EXPORT_SYMBOL extern struct evthread_condition_callbacks evthread_cond_fns_; extern unsigned long (*evthread_id_fn_)(void); EVENT2_EXPORT_SYMBOL extern int evthread_lock_debugging_enabled_; /** Return the ID of the current thread, or 1 if threading isn't enabled. */ #define EVTHREAD_GET_ID() \ (evthread_id_fn_ ? evthread_id_fn_() : 1) /** Return true iff we're in the thread that is currently (or most recently) * running a given event_base's loop. Requires lock. */ #define EVBASE_IN_THREAD(base) \ (evthread_id_fn_ == NULL || \ (base)->th_owner_id == evthread_id_fn_()) /** Return true iff we need to notify the base's main thread about changes to * its state, because it's currently running the main loop in another * thread. Requires lock. */ #define EVBASE_NEED_NOTIFY(base) \ (evthread_id_fn_ != NULL && \ (base)->running_loop && \ (base)->th_owner_id != evthread_id_fn_()) /** Allocate a new lock, and store it in lockvar, a void*. Sets lockvar to NULL if locking is not enabled. */ #define EVTHREAD_ALLOC_LOCK(lockvar, locktype) \ ((lockvar) = evthread_lock_fns_.alloc ? \ evthread_lock_fns_.alloc(locktype) : NULL) /** Free a given lock, if it is present and locking is enabled. */ #define EVTHREAD_FREE_LOCK(lockvar, locktype) \ do { \ void *lock_tmp_ = (lockvar); \ if (lock_tmp_ && evthread_lock_fns_.free) \ evthread_lock_fns_.free(lock_tmp_, (locktype)); \ } while (0) /** Acquire a lock. */ #define EVLOCK_LOCK(lockvar,mode) \ do { \ if (lockvar) \ evthread_lock_fns_.lock(mode, lockvar); \ } while (0) /** Release a lock */ #define EVLOCK_UNLOCK(lockvar,mode) \ do { \ if (lockvar) \ evthread_lock_fns_.unlock(mode, lockvar); \ } while (0) /** Helper: put lockvar1 and lockvar2 into pointerwise ascending order. */ #define EVLOCK_SORTLOCKS_(lockvar1, lockvar2) \ do { \ if (lockvar1 && lockvar2 && lockvar1 > lockvar2) { \ void *tmp = lockvar1; \ lockvar1 = lockvar2; \ lockvar2 = tmp; \ } \ } while (0) /** Lock an event_base, if it is set up for locking. Acquires the lock in the base structure whose field is named 'lockvar'. */ #define EVBASE_ACQUIRE_LOCK(base, lockvar) do { \ EVLOCK_LOCK((base)->lockvar, 0); \ } while (0) /** Unlock an event_base, if it is set up for locking. */ #define EVBASE_RELEASE_LOCK(base, lockvar) do { \ EVLOCK_UNLOCK((base)->lockvar, 0); \ } while (0) /** If lock debugging is enabled, and lock is non-null, assert that 'lock' is * locked and held by us. */ #define EVLOCK_ASSERT_LOCKED(lock) \ do { \ if ((lock) && evthread_lock_debugging_enabled_) { \ EVUTIL_ASSERT(evthread_is_debug_lock_held_(lock)); \ } \ } while (0) /** Try to grab the lock for 'lockvar' without blocking, and return 1 if we * manage to get it. */ static inline int EVLOCK_TRY_LOCK_(void *lock); static inline int EVLOCK_TRY_LOCK_(void *lock) { if (lock && evthread_lock_fns_.lock) { int r = evthread_lock_fns_.lock(EVTHREAD_TRY, lock); return !r; } else { /* Locking is disabled either globally or for this thing; * of course we count as having the lock. */ return 1; } } /** Allocate a new condition variable and store it in the void *, condvar */ #define EVTHREAD_ALLOC_COND(condvar) \ do { \ (condvar) = evthread_cond_fns_.alloc_condition ? \ evthread_cond_fns_.alloc_condition(0) : NULL; \ } while (0) /** Deallocate and free a condition variable in condvar */ #define EVTHREAD_FREE_COND(cond) \ do { \ if (cond) \ evthread_cond_fns_.free_condition((cond)); \ } while (0) /** Signal one thread waiting on cond */ #define EVTHREAD_COND_SIGNAL(cond) \ ( (cond) ? evthread_cond_fns_.signal_condition((cond), 0) : 0 ) /** Signal all threads waiting on cond */ #define EVTHREAD_COND_BROADCAST(cond) \ ( (cond) ? evthread_cond_fns_.signal_condition((cond), 1) : 0 ) /** Wait until the condition 'cond' is signalled. Must be called while * holding 'lock'. The lock will be released until the condition is * signalled, at which point it will be acquired again. Returns 0 for * success, -1 for failure. */ #define EVTHREAD_COND_WAIT(cond, lock) \ ( (cond) ? evthread_cond_fns_.wait_condition((cond), (lock), NULL) : 0 ) /** As EVTHREAD_COND_WAIT, but gives up after 'tv' has elapsed. Returns 1 * on timeout. */ #define EVTHREAD_COND_WAIT_TIMED(cond, lock, tv) \ ( (cond) ? evthread_cond_fns_.wait_condition((cond), (lock), (tv)) : 0 ) /** True iff locking functions have been configured. */ #define EVTHREAD_LOCKING_ENABLED() \ (evthread_lock_fns_.lock != NULL) #elif ! defined(EVENT__DISABLE_THREAD_SUPPORT) unsigned long evthreadimpl_get_id_(void); EVENT2_EXPORT_SYMBOL int evthreadimpl_is_lock_debugging_enabled_(void); EVENT2_EXPORT_SYMBOL void *evthreadimpl_lock_alloc_(unsigned locktype); EVENT2_EXPORT_SYMBOL void evthreadimpl_lock_free_(void *lock, unsigned locktype); EVENT2_EXPORT_SYMBOL int evthreadimpl_lock_lock_(unsigned mode, void *lock); EVENT2_EXPORT_SYMBOL int evthreadimpl_lock_unlock_(unsigned mode, void *lock); EVENT2_EXPORT_SYMBOL void *evthreadimpl_cond_alloc_(unsigned condtype); EVENT2_EXPORT_SYMBOL void evthreadimpl_cond_free_(void *cond); EVENT2_EXPORT_SYMBOL int evthreadimpl_cond_signal_(void *cond, int broadcast); EVENT2_EXPORT_SYMBOL int evthreadimpl_cond_wait_(void *cond, void *lock, const struct timeval *tv); int evthreadimpl_locking_enabled_(void); #define EVTHREAD_GET_ID() evthreadimpl_get_id_() #define EVBASE_IN_THREAD(base) \ ((base)->th_owner_id == evthreadimpl_get_id_()) #define EVBASE_NEED_NOTIFY(base) \ ((base)->running_loop && \ ((base)->th_owner_id != evthreadimpl_get_id_())) #define EVTHREAD_ALLOC_LOCK(lockvar, locktype) \ ((lockvar) = evthreadimpl_lock_alloc_(locktype)) #define EVTHREAD_FREE_LOCK(lockvar, locktype) \ do { \ void *lock_tmp_ = (lockvar); \ if (lock_tmp_) \ evthreadimpl_lock_free_(lock_tmp_, (locktype)); \ } while (0) /** Acquire a lock. */ #define EVLOCK_LOCK(lockvar,mode) \ do { \ if (lockvar) \ evthreadimpl_lock_lock_(mode, lockvar); \ } while (0) /** Release a lock */ #define EVLOCK_UNLOCK(lockvar,mode) \ do { \ if (lockvar) \ evthreadimpl_lock_unlock_(mode, lockvar); \ } while (0) /** Lock an event_base, if it is set up for locking. Acquires the lock in the base structure whose field is named 'lockvar'. */ #define EVBASE_ACQUIRE_LOCK(base, lockvar) do { \ EVLOCK_LOCK((base)->lockvar, 0); \ } while (0) /** Unlock an event_base, if it is set up for locking. */ #define EVBASE_RELEASE_LOCK(base, lockvar) do { \ EVLOCK_UNLOCK((base)->lockvar, 0); \ } while (0) /** If lock debugging is enabled, and lock is non-null, assert that 'lock' is * locked and held by us. */ #define EVLOCK_ASSERT_LOCKED(lock) \ do { \ if ((lock) && evthreadimpl_is_lock_debugging_enabled_()) { \ EVUTIL_ASSERT(evthread_is_debug_lock_held_(lock)); \ } \ } while (0) /** Try to grab the lock for 'lockvar' without blocking, and return 1 if we * manage to get it. */ static inline int EVLOCK_TRY_LOCK_(void *lock); static inline int EVLOCK_TRY_LOCK_(void *lock) { if (lock) { int r = evthreadimpl_lock_lock_(EVTHREAD_TRY, lock); return !r; } else { /* Locking is disabled either globally or for this thing; * of course we count as having the lock. */ return 1; } } /** Allocate a new condition variable and store it in the void *, condvar */ #define EVTHREAD_ALLOC_COND(condvar) \ do { \ (condvar) = evthreadimpl_cond_alloc_(0); \ } while (0) /** Deallocate and free a condition variable in condvar */ #define EVTHREAD_FREE_COND(cond) \ do { \ if (cond) \ evthreadimpl_cond_free_((cond)); \ } while (0) /** Signal one thread waiting on cond */ #define EVTHREAD_COND_SIGNAL(cond) \ ( (cond) ? evthreadimpl_cond_signal_((cond), 0) : 0 ) /** Signal all threads waiting on cond */ #define EVTHREAD_COND_BROADCAST(cond) \ ( (cond) ? evthreadimpl_cond_signal_((cond), 1) : 0 ) /** Wait until the condition 'cond' is signalled. Must be called while * holding 'lock'. The lock will be released until the condition is * signalled, at which point it will be acquired again. Returns 0 for * success, -1 for failure. */ #define EVTHREAD_COND_WAIT(cond, lock) \ ( (cond) ? evthreadimpl_cond_wait_((cond), (lock), NULL) : 0 ) /** As EVTHREAD_COND_WAIT, but gives up after 'tv' has elapsed. Returns 1 * on timeout. */ #define EVTHREAD_COND_WAIT_TIMED(cond, lock, tv) \ ( (cond) ? evthreadimpl_cond_wait_((cond), (lock), (tv)) : 0 ) #define EVTHREAD_LOCKING_ENABLED() \ (evthreadimpl_locking_enabled_()) #else /* EVENT__DISABLE_THREAD_SUPPORT */ #define EVTHREAD_GET_ID() 1 #define EVTHREAD_ALLOC_LOCK(lockvar, locktype) EVUTIL_NIL_STMT_ #define EVTHREAD_FREE_LOCK(lockvar, locktype) EVUTIL_NIL_STMT_ #define EVLOCK_LOCK(lockvar, mode) EVUTIL_NIL_STMT_ #define EVLOCK_UNLOCK(lockvar, mode) EVUTIL_NIL_STMT_ #define EVLOCK_LOCK2(lock1,lock2,mode1,mode2) EVUTIL_NIL_STMT_ #define EVLOCK_UNLOCK2(lock1,lock2,mode1,mode2) EVUTIL_NIL_STMT_ #define EVBASE_IN_THREAD(base) 1 #define EVBASE_NEED_NOTIFY(base) 0 #define EVBASE_ACQUIRE_LOCK(base, lock) EVUTIL_NIL_STMT_ #define EVBASE_RELEASE_LOCK(base, lock) EVUTIL_NIL_STMT_ #define EVLOCK_ASSERT_LOCKED(lock) EVUTIL_NIL_STMT_ #define EVLOCK_TRY_LOCK_(lock) 1 #define EVTHREAD_ALLOC_COND(condvar) EVUTIL_NIL_STMT_ #define EVTHREAD_FREE_COND(cond) EVUTIL_NIL_STMT_ #define EVTHREAD_COND_SIGNAL(cond) EVUTIL_NIL_STMT_ #define EVTHREAD_COND_BROADCAST(cond) EVUTIL_NIL_STMT_ #define EVTHREAD_COND_WAIT(cond, lock) EVUTIL_NIL_STMT_ #define EVTHREAD_COND_WAIT_TIMED(cond, lock, howlong) EVUTIL_NIL_STMT_ #define EVTHREAD_LOCKING_ENABLED() 0 #endif /* This code is shared between both lock impls */ #if ! defined(EVENT__DISABLE_THREAD_SUPPORT) /** Helper: put lockvar1 and lockvar2 into pointerwise ascending order. */ #define EVLOCK_SORTLOCKS_(lockvar1, lockvar2) \ do { \ if (lockvar1 && lockvar2 && lockvar1 > lockvar2) { \ void *tmp = lockvar1; \ lockvar1 = lockvar2; \ lockvar2 = tmp; \ } \ } while (0) /** Acquire both lock1 and lock2. Always allocates locks in the same order, * so that two threads locking two locks with LOCK2 will not deadlock. */ #define EVLOCK_LOCK2(lock1,lock2,mode1,mode2) \ do { \ void *lock1_tmplock_ = (lock1); \ void *lock2_tmplock_ = (lock2); \ EVLOCK_SORTLOCKS_(lock1_tmplock_,lock2_tmplock_); \ EVLOCK_LOCK(lock1_tmplock_,mode1); \ if (lock2_tmplock_ != lock1_tmplock_) \ EVLOCK_LOCK(lock2_tmplock_,mode2); \ } while (0) /** Release both lock1 and lock2. */ #define EVLOCK_UNLOCK2(lock1,lock2,mode1,mode2) \ do { \ void *lock1_tmplock_ = (lock1); \ void *lock2_tmplock_ = (lock2); \ EVLOCK_SORTLOCKS_(lock1_tmplock_,lock2_tmplock_); \ if (lock2_tmplock_ != lock1_tmplock_) \ EVLOCK_UNLOCK(lock2_tmplock_,mode2); \ EVLOCK_UNLOCK(lock1_tmplock_,mode1); \ } while (0) EVENT2_EXPORT_SYMBOL int evthread_is_debug_lock_held_(void *lock); void *evthread_debug_get_real_lock_(void *lock); void *evthread_setup_global_lock_(void *lock_, unsigned locktype, int enable_locks); #define EVTHREAD_SETUP_GLOBAL_LOCK(lockvar, locktype) \ do { \ lockvar = evthread_setup_global_lock_(lockvar, \ (locktype), enable_locks); \ if (!lockvar) { \ event_warn("Couldn't allocate %s", #lockvar); \ return -1; \ } \ } while (0); int event_global_setup_locks_(const int enable_locks); int evsig_global_setup_locks_(const int enable_locks); int evutil_global_setup_locks_(const int enable_locks); int evutil_secure_rng_global_setup_locks_(const int enable_locks); /** Return current evthread_lock_callbacks */ EVENT2_EXPORT_SYMBOL struct evthread_lock_callbacks *evthread_get_lock_callbacks(void); /** Return current evthread_condition_callbacks */ struct evthread_condition_callbacks *evthread_get_condition_callbacks(void); /** Disable locking for internal usage (like global shutdown) */ void evthreadimpl_disable_lock_debugging_(void); #endif #ifdef __cplusplus } #endif #endif /* EVTHREAD_INTERNAL_H_INCLUDED_ */ /* Copyright 2002 Christopher Clark */ /* Copyright 2005-2012 Nick Mathewson */ /* Copyright 2009-2012 Niels Provos and Nick Mathewson */ /* See license at end. */ /* Based on ideas by Christopher Clark and interfaces from Niels Provos. */ #ifndef HT_INTERNAL_H_INCLUDED_ #define HT_INTERNAL_H_INCLUDED_ #define HT_HEAD(name, type) \ struct name { \ /* The hash table itself. */ \ struct type **hth_table; \ /* How long is the hash table? */ \ unsigned hth_table_length; \ /* How many elements does the table contain? */ \ unsigned hth_n_entries; \ /* How many elements will we allow in the table before resizing it? */ \ unsigned hth_load_limit; \ /* Position of hth_table_length in the primes table. */ \ int hth_prime_idx; \ } #define HT_INITIALIZER() \ { NULL, 0, 0, 0, -1 } #ifdef HT_NO_CACHE_HASH_VALUES #define HT_ENTRY(type) \ struct { \ struct type *hte_next; \ } #else #define HT_ENTRY(type) \ struct { \ struct type *hte_next; \ unsigned hte_hash; \ } #endif #define HT_EMPTY(head) \ ((head)->hth_n_entries == 0) /* How many elements in 'head'? */ #define HT_SIZE(head) \ ((head)->hth_n_entries) /* Return memory usage for a hashtable (not counting the entries themselves) */ #define HT_MEM_USAGE(head) \ (sizeof(*head) + (head)->hth_table_length * sizeof(void*)) #define HT_FIND(name, head, elm) name##_HT_FIND((head), (elm)) #define HT_INSERT(name, head, elm) name##_HT_INSERT((head), (elm)) #define HT_REPLACE(name, head, elm) name##_HT_REPLACE((head), (elm)) #define HT_REMOVE(name, head, elm) name##_HT_REMOVE((head), (elm)) #define HT_START(name, head) name##_HT_START(head) #define HT_NEXT(name, head, elm) name##_HT_NEXT((head), (elm)) #define HT_NEXT_RMV(name, head, elm) name##_HT_NEXT_RMV((head), (elm)) #define HT_CLEAR(name, head) name##_HT_CLEAR(head) #define HT_INIT(name, head) name##_HT_INIT(head) /* Helper: */ static inline unsigned ht_improve_hash_(unsigned h) { /* Aim to protect against poor hash functions by adding logic here * - logic taken from java 1.4 hashtable source */ h += ~(h << 9); h ^= ((h >> 14) | (h << 18)); /* >>> */ h += (h << 4); h ^= ((h >> 10) | (h << 22)); /* >>> */ return h; } #if 0 /** Basic string hash function, from Java standard String.hashCode(). */ static inline unsigned ht_string_hash_(const char *s) { unsigned h = 0; int m = 1; while (*s) { h += ((signed char)*s++)*m; m = (m<<5)-1; /* m *= 31 */ } return h; } #endif /** Basic string hash function, from Python's str.__hash__() */ static inline unsigned ht_string_hash_(const char *s) { unsigned h; const unsigned char *cp = (const unsigned char *)s; h = *cp << 7; while (*cp) { h = (1000003*h) ^ *cp++; } /* This conversion truncates the length of the string, but that's ok. */ h ^= (unsigned)(cp-(const unsigned char*)s); return h; } #ifndef HT_NO_CACHE_HASH_VALUES #define HT_SET_HASH_(elm, field, hashfn) \ do { (elm)->field.hte_hash = hashfn(elm); } while (0) #define HT_SET_HASHVAL_(elm, field, val) \ do { (elm)->field.hte_hash = (val); } while (0) #define HT_ELT_HASH_(elm, field, hashfn) \ ((elm)->field.hte_hash) #else #define HT_SET_HASH_(elm, field, hashfn) \ ((void)0) #define HT_ELT_HASH_(elm, field, hashfn) \ (hashfn(elm)) #define HT_SET_HASHVAL_(elm, field, val) \ ((void)0) #endif /* Helper: alias for the bucket containing 'elm'. */ #define HT_BUCKET_(head, field, elm, hashfn) \ ((head)->hth_table[HT_ELT_HASH_(elm,field,hashfn) % head->hth_table_length]) #define HT_FOREACH(x, name, head) \ for ((x) = HT_START(name, head); \ (x) != NULL; \ (x) = HT_NEXT(name, head, x)) #define HT_PROTOTYPE(name, type, field, hashfn, eqfn) \ int name##_HT_GROW(struct name *ht, unsigned min_capacity); \ void name##_HT_CLEAR(struct name *ht); \ int name##_HT_REP_IS_BAD_(const struct name *ht); \ static inline void \ name##_HT_INIT(struct name *head) { \ head->hth_table_length = 0; \ head->hth_table = NULL; \ head->hth_n_entries = 0; \ head->hth_load_limit = 0; \ head->hth_prime_idx = -1; \ } \ /* Helper: returns a pointer to the right location in the table \ * 'head' to find or insert the element 'elm'. */ \ static inline struct type ** \ name##_HT_FIND_P_(struct name *head, struct type *elm) \ { \ struct type **p; \ if (!head->hth_table) \ return NULL; \ p = &HT_BUCKET_(head, field, elm, hashfn); \ while (*p) { \ if (eqfn(*p, elm)) \ return p; \ p = &(*p)->field.hte_next; \ } \ return p; \ } \ /* Return a pointer to the element in the table 'head' matching 'elm', \ * or NULL if no such element exists */ \ static inline struct type * \ name##_HT_FIND(const struct name *head, struct type *elm) \ { \ struct type **p; \ struct name *h = (struct name *) head; \ HT_SET_HASH_(elm, field, hashfn); \ p = name##_HT_FIND_P_(h, elm); \ return p ? *p : NULL; \ } \ /* Insert the element 'elm' into the table 'head'. Do not call this \ * function if the table might already contain a matching element. */ \ static inline void \ name##_HT_INSERT(struct name *head, struct type *elm) \ { \ struct type **p; \ if (!head->hth_table || head->hth_n_entries >= head->hth_load_limit) \ name##_HT_GROW(head, head->hth_n_entries+1); \ ++head->hth_n_entries; \ HT_SET_HASH_(elm, field, hashfn); \ p = &HT_BUCKET_(head, field, elm, hashfn); \ elm->field.hte_next = *p; \ *p = elm; \ } \ /* Insert the element 'elm' into the table 'head'. If there already \ * a matching element in the table, replace that element and return \ * it. */ \ static inline struct type * \ name##_HT_REPLACE(struct name *head, struct type *elm) \ { \ struct type **p, *r; \ if (!head->hth_table || head->hth_n_entries >= head->hth_load_limit) \ name##_HT_GROW(head, head->hth_n_entries+1); \ HT_SET_HASH_(elm, field, hashfn); \ p = name##_HT_FIND_P_(head, elm); \ r = *p; \ *p = elm; \ if (r && (r!=elm)) { \ elm->field.hte_next = r->field.hte_next; \ r->field.hte_next = NULL; \ return r; \ } else { \ ++head->hth_n_entries; \ return NULL; \ } \ } \ /* Remove any element matching 'elm' from the table 'head'. If such \ * an element is found, return it; otherwise return NULL. */ \ static inline struct type * \ name##_HT_REMOVE(struct name *head, struct type *elm) \ { \ struct type **p, *r; \ HT_SET_HASH_(elm, field, hashfn); \ p = name##_HT_FIND_P_(head,elm); \ if (!p || !*p) \ return NULL; \ r = *p; \ *p = r->field.hte_next; \ r->field.hte_next = NULL; \ --head->hth_n_entries; \ return r; \ } \ /* Invoke the function 'fn' on every element of the table 'head', \ * using 'data' as its second argument. If the function returns \ * nonzero, remove the most recently examined element before invoking \ * the function again. */ \ static inline void \ name##_HT_FOREACH_FN(struct name *head, \ int (*fn)(struct type *, void *), \ void *data) \ { \ unsigned idx; \ struct type **p, **nextp, *next; \ if (!head->hth_table) \ return; \ for (idx=0; idx < head->hth_table_length; ++idx) { \ p = &head->hth_table[idx]; \ while (*p) { \ nextp = &(*p)->field.hte_next; \ next = *nextp; \ if (fn(*p, data)) { \ --head->hth_n_entries; \ *p = next; \ } else { \ p = nextp; \ } \ } \ } \ } \ /* Return a pointer to the first element in the table 'head', under \ * an arbitrary order. This order is stable under remove operations, \ * but not under others. If the table is empty, return NULL. */ \ static inline struct type ** \ name##_HT_START(struct name *head) \ { \ unsigned b = 0; \ while (b < head->hth_table_length) { \ if (head->hth_table[b]) \ return &head->hth_table[b]; \ ++b; \ } \ return NULL; \ } \ /* Return the next element in 'head' after 'elm', under the arbitrary \ * order used by HT_START. If there are no more elements, return \ * NULL. If 'elm' is to be removed from the table, you must call \ * this function for the next value before you remove it. \ */ \ static inline struct type ** \ name##_HT_NEXT(struct name *head, struct type **elm) \ { \ if ((*elm)->field.hte_next) { \ return &(*elm)->field.hte_next; \ } else { \ unsigned b = (HT_ELT_HASH_(*elm, field, hashfn) % head->hth_table_length)+1; \ while (b < head->hth_table_length) { \ if (head->hth_table[b]) \ return &head->hth_table[b]; \ ++b; \ } \ return NULL; \ } \ } \ static inline struct type ** \ name##_HT_NEXT_RMV(struct name *head, struct type **elm) \ { \ unsigned h = HT_ELT_HASH_(*elm, field, hashfn); \ *elm = (*elm)->field.hte_next; \ --head->hth_n_entries; \ if (*elm) { \ return elm; \ } else { \ unsigned b = (h % head->hth_table_length)+1; \ while (b < head->hth_table_length) { \ if (head->hth_table[b]) \ return &head->hth_table[b]; \ ++b; \ } \ return NULL; \ } \ } #define HT_GENERATE(name, type, field, hashfn, eqfn, load, mallocfn, \ reallocfn, freefn) \ static unsigned name##_PRIMES[] = { \ 53, 97, 193, 389, \ 769, 1543, 3079, 6151, \ 12289, 24593, 49157, 98317, \ 196613, 393241, 786433, 1572869, \ 3145739, 6291469, 12582917, 25165843, \ 50331653, 100663319, 201326611, 402653189, \ 805306457, 1610612741 \ }; \ static unsigned name##_N_PRIMES = \ (unsigned)(sizeof(name##_PRIMES)/sizeof(name##_PRIMES[0])); \ /* Expand the internal table of 'head' until it is large enough to \ * hold 'size' elements. Return 0 on success, -1 on allocation \ * failure. */ \ int \ name##_HT_GROW(struct name *head, unsigned size) \ { \ unsigned new_len, new_load_limit; \ int prime_idx; \ struct type **new_table; \ if (head->hth_prime_idx == (int)name##_N_PRIMES - 1) \ return 0; \ if (head->hth_load_limit > size) \ return 0; \ prime_idx = head->hth_prime_idx; \ do { \ new_len = name##_PRIMES[++prime_idx]; \ new_load_limit = (unsigned)(load*new_len); \ } while (new_load_limit <= size && \ prime_idx < (int)name##_N_PRIMES); \ if ((new_table = mallocfn(new_len*sizeof(struct type*)))) { \ unsigned b; \ memset(new_table, 0, new_len*sizeof(struct type*)); \ for (b = 0; b < head->hth_table_length; ++b) { \ struct type *elm, *next; \ unsigned b2; \ elm = head->hth_table[b]; \ while (elm) { \ next = elm->field.hte_next; \ b2 = HT_ELT_HASH_(elm, field, hashfn) % new_len; \ elm->field.hte_next = new_table[b2]; \ new_table[b2] = elm; \ elm = next; \ } \ } \ if (head->hth_table) \ freefn(head->hth_table); \ head->hth_table = new_table; \ } else { \ unsigned b, b2; \ new_table = reallocfn(head->hth_table, new_len*sizeof(struct type*)); \ if (!new_table) return -1; \ memset(new_table + head->hth_table_length, 0, \ (new_len - head->hth_table_length)*sizeof(struct type*)); \ for (b=0; b < head->hth_table_length; ++b) { \ struct type *e, **pE; \ for (pE = &new_table[b], e = *pE; e != NULL; e = *pE) { \ b2 = HT_ELT_HASH_(e, field, hashfn) % new_len; \ if (b2 == b) { \ pE = &e->field.hte_next; \ } else { \ *pE = e->field.hte_next; \ e->field.hte_next = new_table[b2]; \ new_table[b2] = e; \ } \ } \ } \ head->hth_table = new_table; \ } \ head->hth_table_length = new_len; \ head->hth_prime_idx = prime_idx; \ head->hth_load_limit = new_load_limit; \ return 0; \ } \ /* Free all storage held by 'head'. Does not free 'head' itself, or \ * individual elements. */ \ void \ name##_HT_CLEAR(struct name *head) \ { \ if (head->hth_table) \ freefn(head->hth_table); \ name##_HT_INIT(head); \ } \ /* Debugging helper: return false iff the representation of 'head' is \ * internally consistent. */ \ int \ name##_HT_REP_IS_BAD_(const struct name *head) \ { \ unsigned n, i; \ struct type *elm; \ if (!head->hth_table_length) { \ if (!head->hth_table && !head->hth_n_entries && \ !head->hth_load_limit && head->hth_prime_idx == -1) \ return 0; \ else \ return 1; \ } \ if (!head->hth_table || head->hth_prime_idx < 0 || \ !head->hth_load_limit) \ return 2; \ if (head->hth_n_entries > head->hth_load_limit) \ return 3; \ if (head->hth_table_length != name##_PRIMES[head->hth_prime_idx]) \ return 4; \ if (head->hth_load_limit != (unsigned)(load*head->hth_table_length)) \ return 5; \ for (n = i = 0; i < head->hth_table_length; ++i) { \ for (elm = head->hth_table[i]; elm; elm = elm->field.hte_next) { \ if (HT_ELT_HASH_(elm, field, hashfn) != hashfn(elm)) \ return 1000 + i; \ if ((HT_ELT_HASH_(elm, field, hashfn) % head->hth_table_length) != i) \ return 10000 + i; \ ++n; \ } \ } \ if (n != head->hth_n_entries) \ return 6; \ return 0; \ } /** Implements an over-optimized "find and insert if absent" block; * not meant for direct usage by typical code, or usage outside the critical * path.*/ #define HT_FIND_OR_INSERT_(name, field, hashfn, head, eltype, elm, var, y, n) \ { \ struct name *var##_head_ = head; \ struct eltype **var; \ if (!var##_head_->hth_table || \ var##_head_->hth_n_entries >= var##_head_->hth_load_limit) \ name##_HT_GROW(var##_head_, var##_head_->hth_n_entries+1); \ HT_SET_HASH_((elm), field, hashfn); \ var = name##_HT_FIND_P_(var##_head_, (elm)); \ if (*var) { \ y; \ } else { \ n; \ } \ } #define HT_FOI_INSERT_(field, head, elm, newent, var) \ { \ HT_SET_HASHVAL_(newent, field, (elm)->field.hte_hash); \ newent->field.hte_next = NULL; \ *var = newent; \ ++((head)->hth_n_entries); \ } /* * Copyright 2005, Nick Mathewson. Implementation logic is adapted from code * by Christopher Clark, retrofit to allow drop-in memory management, and to * use the same interface as Niels Provos's tree.h. This is probably still * a derived work, so the original license below still applies. * * Copyright (c) 2002, Christopher Clark * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * * Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * * Neither the name of the original author; nor the names of any contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER * OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #endif /* * Copyright 2001-2007 Niels Provos * Copyright 2007-2012 Niels Provos and Nick Mathewson * * This header file contains definitions for dealing with HTTP requests * that are internal to libevent. As user of the library, you should not * need to know about these. */ #ifndef HTTP_INTERNAL_H_INCLUDED_ #define HTTP_INTERNAL_H_INCLUDED_ #include "event2/event_struct.h" #include "util-internal.h" #include "defer-internal.h" #define HTTP_CONNECT_TIMEOUT 45 #define HTTP_WRITE_TIMEOUT 50 #define HTTP_READ_TIMEOUT 50 enum message_read_status { ALL_DATA_READ = 1, MORE_DATA_EXPECTED = 0, DATA_CORRUPTED = -1, REQUEST_CANCELED = -2, DATA_TOO_LONG = -3 }; struct evbuffer; struct addrinfo; struct evhttp_request; /* Indicates an unknown request method. */ #define EVHTTP_REQ_UNKNOWN_ (1<<15) enum evhttp_connection_state { EVCON_DISCONNECTED, /**< not currently connected not trying either*/ EVCON_CONNECTING, /**< tries to currently connect */ EVCON_IDLE, /**< connection is established */ EVCON_READING_FIRSTLINE,/**< reading Request-Line (incoming conn) or **< Status-Line (outgoing conn) */ EVCON_READING_HEADERS, /**< reading request/response headers */ EVCON_READING_BODY, /**< reading request/response body */ EVCON_READING_TRAILER, /**< reading request/response chunked trailer */ EVCON_WRITING /**< writing request/response headers/body */ }; struct event_base; /* A client or server connection. */ struct evhttp_connection { /* we use this tailq only if this connection was created for an http * server */ TAILQ_ENTRY(evhttp_connection) next; evutil_socket_t fd; struct bufferevent *bufev; struct event retry_ev; /* for retrying connects */ char *bind_address; /* address to use for binding the src */ ev_uint16_t bind_port; /* local port for binding the src */ char *address; /* address to connect to */ ev_uint16_t port; size_t max_headers_size; ev_uint64_t max_body_size; int flags; #define EVHTTP_CON_INCOMING 0x0001 /* only one request on it ever */ #define EVHTTP_CON_OUTGOING 0x0002 /* multiple requests possible */ #define EVHTTP_CON_CLOSEDETECT 0x0004 /* detecting if persistent close */ /* set when we want to auto free the connection */ #define EVHTTP_CON_AUTOFREE EVHTTP_CON_PUBLIC_FLAGS_END /* Installed when attempt to read HTTP error after write failed, see * EVHTTP_CON_READ_ON_WRITE_ERROR */ #define EVHTTP_CON_READING_ERROR (EVHTTP_CON_AUTOFREE << 1) struct timeval timeout; /* timeout for events */ int retry_cnt; /* retry count */ int retry_max; /* maximum number of retries */ struct timeval initial_retry_timeout; /* Timeout for low long to wait * after first failing attempt * before retry */ enum evhttp_connection_state state; /* for server connections, the http server they are connected with */ struct evhttp *http_server; TAILQ_HEAD(evcon_requestq, evhttp_request) requests; void (*cb)(struct evhttp_connection *, void *); void *cb_arg; void (*closecb)(struct evhttp_connection *, void *); void *closecb_arg; struct event_callback read_more_deferred_cb; struct event_base *base; struct evdns_base *dns_base; int ai_family; }; /* A callback for an http server */ struct evhttp_cb { TAILQ_ENTRY(evhttp_cb) next; char *what; void (*cb)(struct evhttp_request *req, void *); void *cbarg; }; /* both the http server as well as the rpc system need to queue connections */ TAILQ_HEAD(evconq, evhttp_connection); /* each bound socket is stored in one of these */ struct evhttp_bound_socket { TAILQ_ENTRY(evhttp_bound_socket) next; struct evconnlistener *listener; }; /* server alias list item. */ struct evhttp_server_alias { TAILQ_ENTRY(evhttp_server_alias) next; char *alias; /* the server alias. */ }; struct evhttp { /* Next vhost, if this is a vhost. */ TAILQ_ENTRY(evhttp) next_vhost; /* All listeners for this host */ TAILQ_HEAD(boundq, evhttp_bound_socket) sockets; TAILQ_HEAD(httpcbq, evhttp_cb) callbacks; /* All live connections on this host. */ struct evconq connections; TAILQ_HEAD(vhostsq, evhttp) virtualhosts; TAILQ_HEAD(aliasq, evhttp_server_alias) aliases; /* NULL if this server is not a vhost */ char *vhost_pattern; struct timeval timeout; size_t default_max_headers_size; ev_uint64_t default_max_body_size; int flags; const char *default_content_type; /* Bitmask of all HTTP methods that we accept and pass to user * callbacks. */ ev_uint16_t allowed_methods; /* Fallback callback if all the other callbacks for this connection don't match. */ void (*gencb)(struct evhttp_request *req, void *); void *gencbarg; struct bufferevent* (*bevcb)(struct event_base *, void *); void *bevcbarg; struct event_base *base; }; /* XXX most of these functions could be static. */ /* resets the connection; can be reused for more requests */ void evhttp_connection_reset_(struct evhttp_connection *); /* connects if necessary */ int evhttp_connection_connect_(struct evhttp_connection *); enum evhttp_request_error; /* notifies the current request that it failed; resets connection */ EVENT2_EXPORT_SYMBOL void evhttp_connection_fail_(struct evhttp_connection *, enum evhttp_request_error error); enum message_read_status; EVENT2_EXPORT_SYMBOL enum message_read_status evhttp_parse_firstline_(struct evhttp_request *, struct evbuffer*); EVENT2_EXPORT_SYMBOL enum message_read_status evhttp_parse_headers_(struct evhttp_request *, struct evbuffer*); void evhttp_start_read_(struct evhttp_connection *); void evhttp_start_write_(struct evhttp_connection *); /* response sending HTML the data in the buffer */ void evhttp_response_code_(struct evhttp_request *, int, const char *); void evhttp_send_page_(struct evhttp_request *, struct evbuffer *); EVENT2_EXPORT_SYMBOL int evhttp_decode_uri_internal(const char *uri, size_t length, char *ret, int decode_plus); #endif /* HTTP_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef IOCP_INTERNAL_H_INCLUDED_ #define IOCP_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif struct event_overlapped; struct event_iocp_port; struct evbuffer; typedef void (*iocp_callback)(struct event_overlapped *, ev_uintptr_t, ev_ssize_t, int success); /* This whole file is actually win32 only. We wrap the structures in a win32 * ifdef so that we can test-compile code that uses these interfaces on * non-win32 platforms. */ #ifdef _WIN32 /** Internal use only. Wraps an OVERLAPPED that we're using for libevent functionality. Whenever an event_iocp_port gets an event for a given OVERLAPPED*, it upcasts the pointer to an event_overlapped, and calls the iocp_callback function with the event_overlapped, the iocp key, and the number of bytes transferred as arguments. */ struct event_overlapped { OVERLAPPED overlapped; iocp_callback cb; }; /* Mingw's headers don't define LPFN_ACCEPTEX. */ typedef BOOL (WINAPI *AcceptExPtr)(SOCKET, SOCKET, PVOID, DWORD, DWORD, DWORD, LPDWORD, LPOVERLAPPED); typedef BOOL (WINAPI *ConnectExPtr)(SOCKET, const struct sockaddr *, int, PVOID, DWORD, LPDWORD, LPOVERLAPPED); typedef void (WINAPI *GetAcceptExSockaddrsPtr)(PVOID, DWORD, DWORD, DWORD, LPSOCKADDR *, LPINT, LPSOCKADDR *, LPINT); /** Internal use only. Holds pointers to functions that only some versions of Windows provide. */ struct win32_extension_fns { AcceptExPtr AcceptEx; ConnectExPtr ConnectEx; GetAcceptExSockaddrsPtr GetAcceptExSockaddrs; }; /** Internal use only. Stores a Windows IO Completion port, along with related data. */ struct event_iocp_port { /** The port itself */ HANDLE port; /* A lock to cover internal structures. */ CRITICAL_SECTION lock; /** Number of threads ever open on the port. */ short n_threads; /** True iff we're shutting down all the threads on this port */ short shutdown; /** How often the threads on this port check for shutdown and other * conditions */ long ms; /* The threads that are waiting for events. */ HANDLE *threads; /** Number of threads currently open on this port. */ short n_live_threads; /** A semaphore to signal when we are done shutting down. */ HANDLE *shutdownSemaphore; }; EVENT2_EXPORT_SYMBOL const struct win32_extension_fns *event_get_win32_extension_fns_(void); #else /* Dummy definition so we can test-compile more things on unix. */ struct event_overlapped { iocp_callback cb; }; #endif /** Initialize the fields in an event_overlapped. @param overlapped The struct event_overlapped to initialize @param cb The callback that should be invoked once the IO operation has finished. */ EVENT2_EXPORT_SYMBOL void event_overlapped_init_(struct event_overlapped *, iocp_callback cb); /** Allocate and return a new evbuffer that supports overlapped IO on a given socket. The socket must be associated with an IO completion port using event_iocp_port_associate_. */ EVENT2_EXPORT_SYMBOL struct evbuffer *evbuffer_overlapped_new_(evutil_socket_t fd); /** XXXX Document (nickm) */ evutil_socket_t evbuffer_overlapped_get_fd_(struct evbuffer *buf); void evbuffer_overlapped_set_fd_(struct evbuffer *buf, evutil_socket_t fd); /** Start reading data onto the end of an overlapped evbuffer. An evbuffer can only have one read pending at a time. While the read is in progress, no other data may be added to the end of the buffer. The buffer must be created with event_overlapped_init_(). evbuffer_commit_read_() must be called in the completion callback. @param buf The buffer to read onto @param n The number of bytes to try to read. @param ol Overlapped object with associated completion callback. @return 0 on success, -1 on error. */ EVENT2_EXPORT_SYMBOL int evbuffer_launch_read_(struct evbuffer *buf, size_t n, struct event_overlapped *ol); /** Start writing data from the start of an evbuffer. An evbuffer can only have one write pending at a time. While the write is in progress, no other data may be removed from the front of the buffer. The buffer must be created with event_overlapped_init_(). evbuffer_commit_write_() must be called in the completion callback. @param buf The buffer to read onto @param n The number of bytes to try to read. @param ol Overlapped object with associated completion callback. @return 0 on success, -1 on error. */ EVENT2_EXPORT_SYMBOL int evbuffer_launch_write_(struct evbuffer *buf, ev_ssize_t n, struct event_overlapped *ol); /** XXX document */ EVENT2_EXPORT_SYMBOL void evbuffer_commit_read_(struct evbuffer *, ev_ssize_t); EVENT2_EXPORT_SYMBOL void evbuffer_commit_write_(struct evbuffer *, ev_ssize_t); /** Create an IOCP, and launch its worker threads. Internal use only. This interface is unstable, and will change. */ EVENT2_EXPORT_SYMBOL struct event_iocp_port *event_iocp_port_launch_(int n_cpus); /** Associate a file descriptor with an iocp, such that overlapped IO on the fd will happen on one of the iocp's worker threads. */ EVENT2_EXPORT_SYMBOL int event_iocp_port_associate_(struct event_iocp_port *port, evutil_socket_t fd, ev_uintptr_t key); /** Tell all threads serving an iocp to stop. Wait for up to waitMsec for all the threads to finish whatever they're doing. If waitMsec is -1, wait as long as required. If all the threads are done, free the port and return 0. Otherwise, return -1. If you get a -1 return value, it is safe to call this function again. */ EVENT2_EXPORT_SYMBOL int event_iocp_shutdown_(struct event_iocp_port *port, long waitMsec); /* FIXME document. */ EVENT2_EXPORT_SYMBOL int event_iocp_activate_overlapped_(struct event_iocp_port *port, struct event_overlapped *o, ev_uintptr_t key, ev_uint32_t n_bytes); struct event_base; /* FIXME document. */ EVENT2_EXPORT_SYMBOL struct event_iocp_port *event_base_get_iocp_(struct event_base *base); /* FIXME document. */ EVENT2_EXPORT_SYMBOL int event_base_start_iocp_(struct event_base *base, int n_cpus); void event_base_stop_iocp_(struct event_base *base); /* FIXME document. */ EVENT2_EXPORT_SYMBOL struct bufferevent *bufferevent_async_new_(struct event_base *base, evutil_socket_t fd, int options); /* FIXME document. */ void bufferevent_async_set_connected_(struct bufferevent *bev); int bufferevent_async_can_connect_(struct bufferevent *bev); int bufferevent_async_connect_(struct bufferevent *bev, evutil_socket_t fd, const struct sockaddr *sa, int socklen); #ifdef __cplusplus } #endif #endif /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* Internal use only: Fake IPv6 structures and values on platforms that * do not have them */ #ifndef IPV6_INTERNAL_H_INCLUDED_ #define IPV6_INTERNAL_H_INCLUDED_ #include "event2/event-config.h" #include "evconfig-private.h" #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #include "event2/util.h" #ifdef __cplusplus extern "C" { #endif /** @file ipv6-internal.h * * Replacement types and functions for platforms that don't support ipv6 * properly. */ #ifndef EVENT__HAVE_STRUCT_IN6_ADDR struct in6_addr { ev_uint8_t s6_addr[16]; }; #endif #ifndef EVENT__HAVE_SA_FAMILY_T typedef int sa_family_t; #endif #ifndef EVENT__HAVE_STRUCT_SOCKADDR_IN6 struct sockaddr_in6 { /* This will fail if we find a struct sockaddr that doesn't have * sa_family as the first element. */ sa_family_t sin6_family; ev_uint16_t sin6_port; struct in6_addr sin6_addr; }; #endif #ifndef AF_INET6 #define AF_INET6 3333 #endif #ifndef PF_INET6 #define PF_INET6 AF_INET6 #endif #ifdef __cplusplus } #endif #endif /* * Copyright (c) 2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef KQUEUE_INTERNAL_H_INCLUDED_ #define KQUEUE_INTERNAL_H_INCLUDED_ /** Notification function, used to tell an event base to wake up from another * thread. Only works when event_kq_add_notify_event_() has previously been * called successfully on that base. */ int event_kq_notify_base_(struct event_base *base); /** Prepare a kqueue-using event base to receive notifications via an internal * EVFILT_USER event. Return 0 on sucess, -1 on failure. */ int event_kq_add_notify_event_(struct event_base *base); #endif /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef LOG_INTERNAL_H_INCLUDED_ #define LOG_INTERNAL_H_INCLUDED_ #include "event2/util.h" #ifdef __cplusplus extern "C" { #endif #ifdef __GNUC__ #define EV_CHECK_FMT(a,b) __attribute__((format(printf, a, b))) #define EV_NORETURN __attribute__((noreturn)) #else #define EV_CHECK_FMT(a,b) #define EV_NORETURN #endif #define EVENT_ERR_ABORT_ ((int)0xdeaddead) #if !defined(EVENT__DISABLE_DEBUG_MODE) || defined(USE_DEBUG) #define EVENT_DEBUG_LOGGING_ENABLED #endif #ifdef EVENT_DEBUG_LOGGING_ENABLED EVENT2_CORE_EXPORT_SYMBOL extern ev_uint32_t event_debug_logging_mask_; #define event_debug_get_logging_mask_() (event_debug_logging_mask_) #else #define event_debug_get_logging_mask_() (0) #endif EVENT2_EXPORT_SYMBOL void event_err(int eval, const char *fmt, ...) EV_CHECK_FMT(2,3) EV_NORETURN; EVENT2_EXPORT_SYMBOL void event_warn(const char *fmt, ...) EV_CHECK_FMT(1,2); EVENT2_EXPORT_SYMBOL void event_sock_err(int eval, evutil_socket_t sock, const char *fmt, ...) EV_CHECK_FMT(3,4) EV_NORETURN; EVENT2_EXPORT_SYMBOL void event_sock_warn(evutil_socket_t sock, const char *fmt, ...) EV_CHECK_FMT(2,3); EVENT2_EXPORT_SYMBOL void event_errx(int eval, const char *fmt, ...) EV_CHECK_FMT(2,3) EV_NORETURN; EVENT2_EXPORT_SYMBOL void event_warnx(const char *fmt, ...) EV_CHECK_FMT(1,2); EVENT2_EXPORT_SYMBOL void event_msgx(const char *fmt, ...) EV_CHECK_FMT(1,2); EVENT2_EXPORT_SYMBOL void event_debugx_(const char *fmt, ...) EV_CHECK_FMT(1,2); EVENT2_EXPORT_SYMBOL void event_logv_(int severity, const char *errstr, const char *fmt, va_list ap) EV_CHECK_FMT(3,0); #ifdef EVENT_DEBUG_LOGGING_ENABLED #define event_debug(x) do { \ if (event_debug_get_logging_mask_()) { \ event_debugx_ x; \ } \ } while (0) #else #define event_debug(x) ((void)0) #endif #undef EV_CHECK_FMT #ifdef __cplusplus } #endif #endif /* LOG_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Copyright (c) 2006 Maxim Yegorushkin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef MINHEAP_INTERNAL_H_INCLUDED_ #define MINHEAP_INTERNAL_H_INCLUDED_ #include "event2/event-config.h" #include "evconfig-private.h" #include "event2/event.h" #include "event2/event_struct.h" #include "event2/util.h" #include "util-internal.h" #include "mm-internal.h" typedef struct min_heap { struct event** p; unsigned n, a; } min_heap_t; static inline void min_heap_ctor_(min_heap_t* s); static inline void min_heap_dtor_(min_heap_t* s); static inline void min_heap_elem_init_(struct event* e); static inline int min_heap_elt_is_top_(const struct event *e); static inline int min_heap_empty_(min_heap_t* s); static inline unsigned min_heap_size_(min_heap_t* s); static inline struct event* min_heap_top_(min_heap_t* s); static inline int min_heap_reserve_(min_heap_t* s, unsigned n); static inline int min_heap_push_(min_heap_t* s, struct event* e); static inline struct event* min_heap_pop_(min_heap_t* s); static inline int min_heap_adjust_(min_heap_t *s, struct event* e); static inline int min_heap_erase_(min_heap_t* s, struct event* e); static inline void min_heap_shift_up_(min_heap_t* s, unsigned hole_index, struct event* e); static inline void min_heap_shift_up_unconditional_(min_heap_t* s, unsigned hole_index, struct event* e); static inline void min_heap_shift_down_(min_heap_t* s, unsigned hole_index, struct event* e); #define min_heap_elem_greater(a, b) \ (evutil_timercmp(&(a)->ev_timeout, &(b)->ev_timeout, >)) void min_heap_ctor_(min_heap_t* s) { s->p = 0; s->n = 0; s->a = 0; } void min_heap_dtor_(min_heap_t* s) { if (s->p) mm_free(s->p); } void min_heap_elem_init_(struct event* e) { e->ev_timeout_pos.min_heap_idx = -1; } int min_heap_empty_(min_heap_t* s) { return 0u == s->n; } unsigned min_heap_size_(min_heap_t* s) { return s->n; } struct event* min_heap_top_(min_heap_t* s) { return s->n ? *s->p : 0; } int min_heap_push_(min_heap_t* s, struct event* e) { if (s->n == UINT32_MAX || min_heap_reserve_(s, s->n + 1)) return -1; min_heap_shift_up_(s, s->n++, e); return 0; } struct event* min_heap_pop_(min_heap_t* s) { if (s->n) { struct event* e = *s->p; min_heap_shift_down_(s, 0u, s->p[--s->n]); e->ev_timeout_pos.min_heap_idx = -1; return e; } return 0; } int min_heap_elt_is_top_(const struct event *e) { return e->ev_timeout_pos.min_heap_idx == 0; } int min_heap_erase_(min_heap_t* s, struct event* e) { if (-1 != e->ev_timeout_pos.min_heap_idx) { struct event *last = s->p[--s->n]; unsigned parent = (e->ev_timeout_pos.min_heap_idx - 1) / 2; /* we replace e with the last element in the heap. We might need to shift it upward if it is less than its parent, or downward if it is greater than one or both its children. Since the children are known to be less than the parent, it can't need to shift both up and down. */ if (e->ev_timeout_pos.min_heap_idx > 0 && min_heap_elem_greater(s->p[parent], last)) min_heap_shift_up_unconditional_(s, e->ev_timeout_pos.min_heap_idx, last); else min_heap_shift_down_(s, e->ev_timeout_pos.min_heap_idx, last); e->ev_timeout_pos.min_heap_idx = -1; return 0; } return -1; } int min_heap_adjust_(min_heap_t *s, struct event *e) { if (-1 == e->ev_timeout_pos.min_heap_idx) { return min_heap_push_(s, e); } else { unsigned parent = (e->ev_timeout_pos.min_heap_idx - 1) / 2; /* The position of e has changed; we shift it up or down * as needed. We can't need to do both. */ if (e->ev_timeout_pos.min_heap_idx > 0 && min_heap_elem_greater(s->p[parent], e)) min_heap_shift_up_unconditional_(s, e->ev_timeout_pos.min_heap_idx, e); else min_heap_shift_down_(s, e->ev_timeout_pos.min_heap_idx, e); return 0; } } int min_heap_reserve_(min_heap_t* s, unsigned n) { if (s->a < n) { struct event** p; unsigned a = s->a ? s->a * 2 : 8; if (a < n) a = n; #if (SIZE_MAX == UINT32_MAX) if (a > SIZE_MAX / sizeof *p) return -1; #endif if (!(p = (struct event**)mm_realloc(s->p, a * sizeof *p))) return -1; s->p = p; s->a = a; } return 0; } void min_heap_shift_up_unconditional_(min_heap_t* s, unsigned hole_index, struct event* e) { unsigned parent = (hole_index - 1) / 2; do { (s->p[hole_index] = s->p[parent])->ev_timeout_pos.min_heap_idx = hole_index; hole_index = parent; parent = (hole_index - 1) / 2; } while (hole_index && min_heap_elem_greater(s->p[parent], e)); (s->p[hole_index] = e)->ev_timeout_pos.min_heap_idx = hole_index; } void min_heap_shift_up_(min_heap_t* s, unsigned hole_index, struct event* e) { unsigned parent = (hole_index - 1) / 2; while (hole_index && min_heap_elem_greater(s->p[parent], e)) { (s->p[hole_index] = s->p[parent])->ev_timeout_pos.min_heap_idx = hole_index; hole_index = parent; parent = (hole_index - 1) / 2; } (s->p[hole_index] = e)->ev_timeout_pos.min_heap_idx = hole_index; } void min_heap_shift_down_(min_heap_t* s, unsigned hole_index, struct event* e) { unsigned min_child = 2 * (hole_index + 1); while (min_child <= s->n) { min_child -= min_child == s->n || min_heap_elem_greater(s->p[min_child], s->p[min_child - 1]); if (!(min_heap_elem_greater(e, s->p[min_child]))) break; (s->p[hole_index] = s->p[min_child])->ev_timeout_pos.min_heap_idx = hole_index; hole_index = min_child; min_child = 2 * (hole_index + 1); } (s->p[hole_index] = e)->ev_timeout_pos.min_heap_idx = hole_index; } #endif /* MINHEAP_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef MM_INTERNAL_H_INCLUDED_ #define MM_INTERNAL_H_INCLUDED_ #include #ifdef __cplusplus extern "C" { #endif #ifndef EVENT__DISABLE_MM_REPLACEMENT /* Internal use only: Memory allocation functions. We give them nice short * mm_names for our own use, but make sure that the symbols have longer names * so they don't conflict with other libraries (like, say, libmm). */ /** Allocate uninitialized memory. * * @return On success, return a pointer to sz newly allocated bytes. * On failure, set errno to ENOMEM and return NULL. * If the argument sz is 0, simply return NULL. */ EVENT2_EXPORT_SYMBOL void *event_mm_malloc_(size_t sz); /** Allocate memory initialized to zero. * * @return On success, return a pointer to (count * size) newly allocated * bytes, initialized to zero. * On failure, or if the product would result in an integer overflow, * set errno to ENOMEM and return NULL. * If either arguments are 0, simply return NULL. */ EVENT2_EXPORT_SYMBOL void *event_mm_calloc_(size_t count, size_t size); /** Duplicate a string. * * @return On success, return a pointer to a newly allocated duplicate * of a string. * Set errno to ENOMEM and return NULL if a memory allocation error * occurs (or would occur) in the process. * If the argument str is NULL, set errno to EINVAL and return NULL. */ EVENT2_EXPORT_SYMBOL char *event_mm_strdup_(const char *str); EVENT2_EXPORT_SYMBOL void *event_mm_realloc_(void *p, size_t sz); EVENT2_EXPORT_SYMBOL void event_mm_free_(void *p); #define mm_malloc(sz) event_mm_malloc_(sz) #define mm_calloc(count, size) event_mm_calloc_((count), (size)) #define mm_strdup(s) event_mm_strdup_(s) #define mm_realloc(p, sz) event_mm_realloc_((p), (sz)) #define mm_free(p) event_mm_free_(p) #else #define mm_malloc(sz) malloc(sz) #define mm_calloc(n, sz) calloc((n), (sz)) #define mm_strdup(s) strdup(s) #define mm_realloc(p, sz) realloc((p), (sz)) #define mm_free(p) free(p) #endif #ifdef __cplusplus } #endif #endif #ifndef OPENSSL_COMPAT_H #define OPENSSL_COMPAT_H #include #include "util-internal.h" #if (OPENSSL_VERSION_NUMBER < 0x10100000L) || \ (defined(LIBRESSL_VERSION_NUMBER) && LIBRESSL_VERSION_NUMBER < 0x20700000L) static inline BIO_METHOD *BIO_meth_new(int type, const char *name) { BIO_METHOD *biom = calloc(1, sizeof(BIO_METHOD)); if (biom != NULL) { biom->type = type; biom->name = name; } return biom; } #define BIO_meth_set_write(b, f) (b)->bwrite = (f) #define BIO_meth_set_read(b, f) (b)->bread = (f) #define BIO_meth_set_puts(b, f) (b)->bputs = (f) #define BIO_meth_set_ctrl(b, f) (b)->ctrl = (f) #define BIO_meth_set_create(b, f) (b)->create = (f) #define BIO_meth_set_destroy(b, f) (b)->destroy = (f) #define BIO_set_init(b, val) (b)->init = (val) #define BIO_set_data(b, val) (b)->ptr = (val) #define BIO_set_shutdown(b, val) (b)->shutdown = (val) #define BIO_get_init(b) (b)->init #define BIO_get_data(b) (b)->ptr #define BIO_get_shutdown(b) (b)->shutdown #define TLS_method SSLv23_method #define X509_getm_notBefore X509_get_notBefore #define X509_getm_notAfter X509_get_notAfter #endif /* (OPENSSL_VERSION_NUMBER < 0x10100000L) || \ (defined(LIBRESSL_VERSION_NUMBER) && LIBRESSL_VERSION_NUMBER < 0x20700000L) */ #if defined(LIBRESSL_VERSION_NUMBER) && LIBRESSL_VERSION_NUMBER >= 0x20700000L #define BIO_get_init(b) (b)->init #endif #endif /* OPENSSL_COMPAT_H */ /* * Copyright (c) 2009-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef RATELIM_INTERNAL_H_INCLUDED_ #define RATELIM_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/util.h" /** A token bucket is an internal structure that tracks how many bytes we are * currently willing to read or write on a given bufferevent or group of * bufferevents */ struct ev_token_bucket { /** How many bytes are we willing to read or write right now? These * values are signed so that we can do "defecit spending" */ ev_ssize_t read_limit, write_limit; /** When was this bucket last updated? Measured in abstract 'ticks' * relative to the token bucket configuration. */ ev_uint32_t last_updated; }; /** Configuration info for a token bucket or set of token buckets. */ struct ev_token_bucket_cfg { /** How many bytes are we willing to read on average per tick? */ size_t read_rate; /** How many bytes are we willing to read at most in any one tick? */ size_t read_maximum; /** How many bytes are we willing to write on average per tick? */ size_t write_rate; /** How many bytes are we willing to write at most in any one tick? */ size_t write_maximum; /* How long is a tick? Note that fractions of a millisecond are * ignored. */ struct timeval tick_timeout; /* How long is a tick, in milliseconds? Derived from tick_timeout. */ unsigned msec_per_tick; }; /** The current tick is 'current_tick': add bytes to 'bucket' as specified in * 'cfg'. */ int ev_token_bucket_update_(struct ev_token_bucket *bucket, const struct ev_token_bucket_cfg *cfg, ev_uint32_t current_tick); /** In which tick does 'tv' fall according to 'cfg'? Note that ticks can * overflow easily; your code needs to handle this. */ ev_uint32_t ev_token_bucket_get_tick_(const struct timeval *tv, const struct ev_token_bucket_cfg *cfg); /** Adjust 'bucket' to respect 'cfg', and note that it was last updated in * 'current_tick'. If 'reinitialize' is true, we are changing the * configuration of 'bucket'; otherwise, we are setting it up for the first * time. */ int ev_token_bucket_init_(struct ev_token_bucket *bucket, const struct ev_token_bucket_cfg *cfg, ev_uint32_t current_tick, int reinitialize); int bufferevent_remove_from_rate_limit_group_internal_(struct bufferevent *bev, int unsuspend); /** Decrease the read limit of 'b' by 'n' bytes */ #define ev_token_bucket_decrement_read(b,n) \ do { \ (b)->read_limit -= (n); \ } while (0) /** Decrease the write limit of 'b' by 'n' bytes */ #define ev_token_bucket_decrement_write(b,n) \ do { \ (b)->write_limit -= (n); \ } while (0) #ifdef __cplusplus } #endif #endif #ifndef STRLCPY_INTERNAL_H_INCLUDED_ #define STRLCPY_INTERNAL_H_INCLUDED_ #ifdef __cplusplus extern "C" { #endif #include "event2/event-config.h" #include "event2/visibility.h" #include "evconfig-private.h" #ifndef EVENT__HAVE_STRLCPY #include EVENT2_EXPORT_SYMBOL size_t event_strlcpy_(char *dst, const char *src, size_t siz); #define strlcpy event_strlcpy_ #endif #ifdef __cplusplus } #endif #endif /* * Copyright (c) 2000-2007 Niels Provos * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef TIME_INTERNAL_H_INCLUDED_ #define TIME_INTERNAL_H_INCLUDED_ #include "event2/event-config.h" #include "evconfig-private.h" #ifdef EVENT__HAVE_MACH_MACH_TIME_H /* For mach_timebase_info */ #include #endif #include #include "event2/util.h" #ifdef __cplusplus extern "C" { #endif #if defined(EVENT__HAVE_CLOCK_GETTIME) && defined(CLOCK_MONOTONIC) #define HAVE_POSIX_MONOTONIC #elif defined(EVENT__HAVE_MACH_ABSOLUTE_TIME) #define HAVE_MACH_MONOTONIC #elif defined(_WIN32) #define HAVE_WIN32_MONOTONIC #else #define HAVE_FALLBACK_MONOTONIC #endif long evutil_tv_to_msec_(const struct timeval *tv); EVENT2_EXPORT_SYMBOL void evutil_usleep_(const struct timeval *tv); #ifdef _WIN32 typedef ULONGLONG (WINAPI *ev_GetTickCount_func)(void); #endif struct evutil_monotonic_timer { #ifdef HAVE_MACH_MONOTONIC struct mach_timebase_info mach_timebase_units; #endif #ifdef HAVE_POSIX_MONOTONIC int monotonic_clock; #endif #ifdef HAVE_WIN32_MONOTONIC ev_GetTickCount_func GetTickCount64_fn; ev_GetTickCount_func GetTickCount_fn; ev_uint64_t last_tick_count; ev_uint64_t adjust_tick_count; ev_uint64_t first_tick; ev_uint64_t first_counter; double usec_per_count; int use_performance_counter; #endif struct timeval adjust_monotonic_clock; struct timeval last_time; }; EVENT2_EXPORT_SYMBOL int evutil_configure_monotonic_time_(struct evutil_monotonic_timer *mt, int flags); EVENT2_EXPORT_SYMBOL int evutil_gettime_monotonic_(struct evutil_monotonic_timer *mt, struct timeval *tv); #ifdef __cplusplus } #endif #endif /* EVENT_INTERNAL_H_INCLUDED_ */ /* * Copyright (c) 2007-2012 Niels Provos and Nick Mathewson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #ifndef UTIL_INTERNAL_H_INCLUDED_ #define UTIL_INTERNAL_H_INCLUDED_ #include "event2/event-config.h" #include "evconfig-private.h" #include /* For EVUTIL_ASSERT */ #include "log-internal.h" #include #include #ifdef EVENT__HAVE_SYS_SOCKET_H #include #endif #ifdef EVENT__HAVE_SYS_EVENTFD_H #include #endif #include "event2/util.h" #include "time-internal.h" #include "ipv6-internal.h" #ifdef __cplusplus extern "C" { #endif /* __has_attribute() wrapper */ #ifdef __has_attribute # define EVUTIL_HAS_ATTRIBUTE __has_attribute #endif /** clang 3 __has_attribute misbehaves in some versions */ #if defined(__clang__) && __clang__ == 1 # if defined(__apple_build_version__) # if __clang_major__ <= 6 # undef EVUTIL_HAS_ATTRIBUTE # endif # else /* !__apple_build_version__ */ # if __clang_major__ == 3 && __clang_minor__ >= 2 && __clang_minor__ <= 5 # undef EVUTIL_HAS_ATTRIBUTE # endif # endif /* __apple_build_version__ */ #endif /*\ defined(__clang__) && __clang__ == 1 */ #ifndef EVUTIL_HAS_ATTRIBUTE # define EVUTIL_HAS_ATTRIBUTE(x) 0 #endif /* If we need magic to say "inline", get it for free internally. */ #ifdef EVENT__inline #define inline EVENT__inline #endif /* Define to appropriate substitute if compiler doesnt have __func__ */ #if defined(EVENT__HAVE___func__) # ifndef __func__ # define __func__ __func__ # endif #elif defined(EVENT__HAVE___FUNCTION__) # define __func__ __FUNCTION__ #else # define __func__ __FILE__ #endif /* A good no-op to use in macro definitions. */ #define EVUTIL_NIL_STMT_ ((void)0) /* A no-op that tricks the compiler into thinking a condition is used while * definitely not making any code for it. Used to compile out asserts while * avoiding "unused variable" warnings. The "!" forces the compiler to * do the sizeof() on an int, in case "condition" is a bitfield value. */ #define EVUTIL_NIL_CONDITION_(condition) do { \ (void)sizeof(!(condition)); \ } while(0) /* Internal use only: macros to match patterns of error codes in a cross-platform way. We need these macros because of two historical reasons: first, nonblocking IO functions are generally written to give an error on the "blocked now, try later" case, so sometimes an error from a read, write, connect, or accept means "no error; just wait for more data," and we need to look at the error code. Second, Windows defines a different set of error codes for sockets. */ #ifndef _WIN32 #if EAGAIN == EWOULDBLOCK #define EVUTIL_ERR_IS_EAGAIN(e) \ ((e) == EAGAIN) #else #define EVUTIL_ERR_IS_EAGAIN(e) \ ((e) == EAGAIN || (e) == EWOULDBLOCK) #endif /* True iff e is an error that means a read/write operation can be retried. */ #define EVUTIL_ERR_RW_RETRIABLE(e) \ ((e) == EINTR || EVUTIL_ERR_IS_EAGAIN(e)) /* True iff e is an error that means an connect can be retried. */ #define EVUTIL_ERR_CONNECT_RETRIABLE(e) \ ((e) == EINTR || (e) == EINPROGRESS) /* True iff e is an error that means a accept can be retried. */ #define EVUTIL_ERR_ACCEPT_RETRIABLE(e) \ ((e) == EINTR || EVUTIL_ERR_IS_EAGAIN(e) || (e) == ECONNABORTED) /* True iff e is an error that means the connection was refused */ #define EVUTIL_ERR_CONNECT_REFUSED(e) \ ((e) == ECONNREFUSED) #else /* Win32 */ #define EVUTIL_ERR_IS_EAGAIN(e) \ ((e) == WSAEWOULDBLOCK || (e) == EAGAIN) #define EVUTIL_ERR_RW_RETRIABLE(e) \ ((e) == WSAEWOULDBLOCK || \ (e) == WSAEINTR) #define EVUTIL_ERR_CONNECT_RETRIABLE(e) \ ((e) == WSAEWOULDBLOCK || \ (e) == WSAEINTR || \ (e) == WSAEINPROGRESS || \ (e) == WSAEINVAL) #define EVUTIL_ERR_ACCEPT_RETRIABLE(e) \ EVUTIL_ERR_RW_RETRIABLE(e) #define EVUTIL_ERR_CONNECT_REFUSED(e) \ ((e) == WSAECONNREFUSED) #endif /* Arguments for shutdown() */ #ifdef SHUT_RD #define EVUTIL_SHUT_RD SHUT_RD #else #define EVUTIL_SHUT_RD 0 #endif #ifdef SHUT_WR #define EVUTIL_SHUT_WR SHUT_WR #else #define EVUTIL_SHUT_WR 1 /* SD_SEND */ #endif #ifdef SHUT_BOTH #define EVUTIL_SHUT_BOTH SHUT_BOTH #else #define EVUTIL_SHUT_BOTH 2 #endif /* Helper: Verify that all the elements in 'dlist' are internally consistent. * Checks for circular lists and bad prev/next pointers. * * Example usage: * EVUTIL_ASSERT_LIST_OK(eventlist, event, ev_next); */ #define EVUTIL_ASSERT_LIST_OK(dlist, type, field) do { \ struct type *elm1, *elm2, **nextp; \ if (LIST_EMPTY((dlist))) \ break; \ \ /* Check list for circularity using Floyd's */ \ /* 'Tortoise and Hare' algorithm */ \ elm1 = LIST_FIRST((dlist)); \ elm2 = LIST_NEXT(elm1, field); \ while (elm1 && elm2) { \ EVUTIL_ASSERT(elm1 != elm2); \ elm1 = LIST_NEXT(elm1, field); \ elm2 = LIST_NEXT(elm2, field); \ if (!elm2) \ break; \ EVUTIL_ASSERT(elm1 != elm2); \ elm2 = LIST_NEXT(elm2, field); \ } \ \ /* Now check next and prev pointers for consistency. */ \ nextp = &LIST_FIRST((dlist)); \ elm1 = LIST_FIRST((dlist)); \ while (elm1) { \ EVUTIL_ASSERT(*nextp == elm1); \ EVUTIL_ASSERT(nextp == elm1->field.le_prev); \ nextp = &LIST_NEXT(elm1, field); \ elm1 = *nextp; \ } \ } while (0) /* Helper: Verify that all the elements in a TAILQ are internally consistent. * Checks for circular lists and bad prev/next pointers. * * Example usage: * EVUTIL_ASSERT_TAILQ_OK(activelist, event, ev_active_next); */ #define EVUTIL_ASSERT_TAILQ_OK(tailq, type, field) do { \ struct type *elm1, *elm2, **nextp; \ if (TAILQ_EMPTY((tailq))) \ break; \ \ /* Check list for circularity using Floyd's */ \ /* 'Tortoise and Hare' algorithm */ \ elm1 = TAILQ_FIRST((tailq)); \ elm2 = TAILQ_NEXT(elm1, field); \ while (elm1 && elm2) { \ EVUTIL_ASSERT(elm1 != elm2); \ elm1 = TAILQ_NEXT(elm1, field); \ elm2 = TAILQ_NEXT(elm2, field); \ if (!elm2) \ break; \ EVUTIL_ASSERT(elm1 != elm2); \ elm2 = TAILQ_NEXT(elm2, field); \ } \ \ /* Now check next and prev pointers for consistency. */ \ nextp = &TAILQ_FIRST((tailq)); \ elm1 = TAILQ_FIRST((tailq)); \ while (elm1) { \ EVUTIL_ASSERT(*nextp == elm1); \ EVUTIL_ASSERT(nextp == elm1->field.tqe_prev); \ nextp = &TAILQ_NEXT(elm1, field); \ elm1 = *nextp; \ } \ EVUTIL_ASSERT(nextp == (tailq)->tqh_last); \ } while (0) /* Locale-independent replacements for some ctypes functions. Use these * when you care about ASCII's notion of character types, because you are about * to send those types onto the wire. */ EVENT2_EXPORT_SYMBOL int EVUTIL_ISALPHA_(char c); EVENT2_EXPORT_SYMBOL int EVUTIL_ISALNUM_(char c); int EVUTIL_ISSPACE_(char c); EVENT2_EXPORT_SYMBOL int EVUTIL_ISDIGIT_(char c); EVENT2_EXPORT_SYMBOL int EVUTIL_ISXDIGIT_(char c); int EVUTIL_ISPRINT_(char c); int EVUTIL_ISLOWER_(char c); int EVUTIL_ISUPPER_(char c); EVENT2_EXPORT_SYMBOL char EVUTIL_TOUPPER_(char c); EVENT2_EXPORT_SYMBOL char EVUTIL_TOLOWER_(char c); /** Remove all trailing horizontal whitespace (space or tab) from the end of a * string */ EVENT2_EXPORT_SYMBOL void evutil_rtrim_lws_(char *); /** Helper macro. If we know that a given pointer points to a field in a structure, return a pointer to the structure itself. Used to implement our half-baked C OO. Example: struct subtype { int x; struct supertype common; int y; }; ... void fn(struct supertype *super) { struct subtype *sub = EVUTIL_UPCAST(super, struct subtype, common); ... } */ #define EVUTIL_UPCAST(ptr, type, field) \ ((type *)(((char*)(ptr)) - evutil_offsetof(type, field))) /* As open(pathname, flags, mode), except that the file is always opened with * the close-on-exec flag set. (And the mode argument is mandatory.) */ int evutil_open_closeonexec_(const char *pathname, int flags, unsigned mode); EVENT2_EXPORT_SYMBOL int evutil_read_file_(const char *filename, char **content_out, size_t *len_out, int is_binary); EVENT2_EXPORT_SYMBOL int evutil_socket_connect_(evutil_socket_t *fd_ptr, const struct sockaddr *sa, int socklen); int evutil_socket_finished_connecting_(evutil_socket_t fd); EVENT2_EXPORT_SYMBOL int evutil_ersatz_socketpair_(int, int , int, evutil_socket_t[]); int evutil_resolve_(int family, const char *hostname, struct sockaddr *sa, ev_socklen_t *socklen, int port); const char *evutil_getenv_(const char *name); /* Structure to hold the state of our weak random number generator. */ struct evutil_weakrand_state { ev_uint32_t seed; }; #define EVUTIL_WEAKRAND_MAX EV_INT32_MAX /* Initialize the state of a week random number generator based on 'seed'. If * the seed is 0, construct a new seed based on not-very-strong platform * entropy, like the PID and the time of day. * * This function, and the other evutil_weakrand* functions, are meant for * speed, not security or statistical strength. If you need a RNG which an * attacker can't predict, or which passes strong statistical tests, use the * evutil_secure_rng* functions instead. */ EVENT2_EXPORT_SYMBOL ev_uint32_t evutil_weakrand_seed_(struct evutil_weakrand_state *state, ev_uint32_t seed); /* Return a pseudorandom value between 0 and EVUTIL_WEAKRAND_MAX inclusive. * Updates the state in 'seed' as needed -- this value must be protected by a * lock. */ EVENT2_EXPORT_SYMBOL ev_int32_t evutil_weakrand_(struct evutil_weakrand_state *seed); /* Return a pseudorandom value x such that 0 <= x < top. top must be no more * than EVUTIL_WEAKRAND_MAX. Updates the state in 'seed' as needed -- this * value must be proteced by a lock */ EVENT2_EXPORT_SYMBOL ev_int32_t evutil_weakrand_range_(struct evutil_weakrand_state *seed, ev_int32_t top); /* Evaluates to the same boolean value as 'p', and hints to the compiler that * we expect this value to be false. */ #if defined(__GNUC__) && __GNUC__ >= 3 /* gcc 3.0 or later */ #define EVUTIL_UNLIKELY(p) __builtin_expect(!!(p),0) #else #define EVUTIL_UNLIKELY(p) (p) #endif #if EVUTIL_HAS_ATTRIBUTE(fallthrough) #define EVUTIL_FALLTHROUGH __attribute__((fallthrough)) #else #define EVUTIL_FALLTHROUGH /* fallthrough */ #endif /* Replacement for assert() that calls event_errx on failure. */ #ifdef NDEBUG #define EVUTIL_ASSERT(cond) EVUTIL_NIL_CONDITION_(cond) #define EVUTIL_FAILURE_CHECK(cond) 0 #else #define EVUTIL_ASSERT(cond) \ do { \ if (EVUTIL_UNLIKELY(!(cond))) { \ event_errx(EVENT_ERR_ABORT_, \ "%s:%d: Assertion %s failed in %s", \ __FILE__,__LINE__,#cond,__func__); \ /* In case a user-supplied handler tries to */ \ /* return control to us, log and abort here. */ \ (void)fprintf(stderr, \ "%s:%d: Assertion %s failed in %s", \ __FILE__,__LINE__,#cond,__func__); \ abort(); \ } \ } while (0) #define EVUTIL_FAILURE_CHECK(cond) EVUTIL_UNLIKELY(cond) #endif #ifndef EVENT__HAVE_STRUCT_SOCKADDR_STORAGE /* Replacement for sockaddr storage that we can use internally on platforms * that lack it. It is not space-efficient, but neither is sockaddr_storage. */ struct sockaddr_storage { union { struct sockaddr ss_sa; struct sockaddr_in ss_sin; struct sockaddr_in6 ss_sin6; char ss_padding[128]; } ss_union; }; #define ss_family ss_union.ss_sa.sa_family #endif /* Internal addrinfo error code. This one is returned from only from * evutil_getaddrinfo_common_, when we are sure that we'll have to hit a DNS * server. */ #define EVUTIL_EAI_NEED_RESOLVE -90002 struct evdns_base; struct evdns_getaddrinfo_request; typedef struct evdns_getaddrinfo_request* (*evdns_getaddrinfo_fn)( struct evdns_base *base, const char *nodename, const char *servname, const struct evutil_addrinfo *hints_in, void (*cb)(int, struct evutil_addrinfo *, void *), void *arg); EVENT2_EXPORT_SYMBOL void evutil_set_evdns_getaddrinfo_fn_(evdns_getaddrinfo_fn fn); typedef void (*evdns_getaddrinfo_cancel_fn)( struct evdns_getaddrinfo_request *req); EVENT2_EXPORT_SYMBOL void evutil_set_evdns_getaddrinfo_cancel_fn_(evdns_getaddrinfo_cancel_fn fn); EVENT2_EXPORT_SYMBOL struct evutil_addrinfo *evutil_new_addrinfo_(struct sockaddr *sa, ev_socklen_t socklen, const struct evutil_addrinfo *hints); EVENT2_EXPORT_SYMBOL struct evutil_addrinfo *evutil_addrinfo_append_(struct evutil_addrinfo *first, struct evutil_addrinfo *append); EVENT2_EXPORT_SYMBOL void evutil_adjust_hints_for_addrconfig_(struct evutil_addrinfo *hints); EVENT2_EXPORT_SYMBOL int evutil_getaddrinfo_common_(const char *nodename, const char *servname, struct evutil_addrinfo *hints, struct evutil_addrinfo **res, int *portnum); struct evdns_getaddrinfo_request *evutil_getaddrinfo_async_( struct evdns_base *dns_base, const char *nodename, const char *servname, const struct evutil_addrinfo *hints_in, void (*cb)(int, struct evutil_addrinfo *, void *), void *arg); void evutil_getaddrinfo_cancel_async_(struct evdns_getaddrinfo_request *data); /** Return true iff sa is a looback address. (That is, it is 127.0.0.1/8, or * ::1). */ EVENT2_EXPORT_SYMBOL int evutil_sockaddr_is_loopback_(const struct sockaddr *sa); /** Formats a sockaddr sa into a string buffer of size outlen stored in out. Returns a pointer to out. Always writes something into out, so it's safe to use the output of this function without checking it for NULL. */ EVENT2_EXPORT_SYMBOL const char *evutil_format_sockaddr_port_(const struct sockaddr *sa, char *out, size_t outlen); int evutil_hex_char_to_int_(char c); void evutil_free_secure_rng_globals_(void); void evutil_free_globals_(void); #ifdef _WIN32 EVENT2_EXPORT_SYMBOL HMODULE evutil_load_windows_system_library_(const TCHAR *library_name); #endif #ifndef EV_SIZE_FMT #if defined(_MSC_VER) || defined(__MINGW32__) || defined(__MINGW64__) #define EV_U64_FMT "%I64u" #define EV_I64_FMT "%I64d" #define EV_I64_ARG(x) ((__int64)(x)) #define EV_U64_ARG(x) ((unsigned __int64)(x)) #else #define EV_U64_FMT "%llu" #define EV_I64_FMT "%lld" #define EV_I64_ARG(x) ((long long)(x)) #define EV_U64_ARG(x) ((unsigned long long)(x)) #endif #endif #ifdef _WIN32 #define EV_SOCK_FMT EV_I64_FMT #define EV_SOCK_ARG(x) EV_I64_ARG((x)) #else #define EV_SOCK_FMT "%d" #define EV_SOCK_ARG(x) (x) #endif #if defined(__STDC__) && defined(__STDC_VERSION__) && !defined(__MINGW64_VERSION_MAJOR) #if (__STDC_VERSION__ >= 199901L) #define EV_SIZE_FMT "%zu" #define EV_SSIZE_FMT "%zd" #define EV_SIZE_ARG(x) (x) #define EV_SSIZE_ARG(x) (x) #endif #endif #ifndef EV_SIZE_FMT #if (EVENT__SIZEOF_SIZE_T <= EVENT__SIZEOF_LONG) #define EV_SIZE_FMT "%lu" #define EV_SSIZE_FMT "%ld" #define EV_SIZE_ARG(x) ((unsigned long)(x)) #define EV_SSIZE_ARG(x) ((long)(x)) #else #define EV_SIZE_FMT EV_U64_FMT #define EV_SSIZE_FMT EV_I64_FMT #define EV_SIZE_ARG(x) EV_U64_ARG(x) #define EV_SSIZE_ARG(x) EV_I64_ARG(x) #endif #endif EVENT2_EXPORT_SYMBOL evutil_socket_t evutil_socket_(int domain, int type, int protocol); evutil_socket_t evutil_accept4_(evutil_socket_t sockfd, struct sockaddr *addr, ev_socklen_t *addrlen, int flags); /* used by one of the test programs.. */ EVENT2_EXPORT_SYMBOL int evutil_make_internal_pipe_(evutil_socket_t fd[2]); evutil_socket_t evutil_eventfd_(unsigned initval, int flags); #ifdef SOCK_NONBLOCK #define EVUTIL_SOCK_NONBLOCK SOCK_NONBLOCK #else #define EVUTIL_SOCK_NONBLOCK 0x4000000 #endif #ifdef SOCK_CLOEXEC #define EVUTIL_SOCK_CLOEXEC SOCK_CLOEXEC #else #define EVUTIL_SOCK_CLOEXEC 0x80000000 #endif #ifdef EFD_NONBLOCK #define EVUTIL_EFD_NONBLOCK EFD_NONBLOCK #else #define EVUTIL_EFD_NONBLOCK 0x4000 #endif #ifdef EFD_CLOEXEC #define EVUTIL_EFD_CLOEXEC EFD_CLOEXEC #else #define EVUTIL_EFD_CLOEXEC 0x8000 #endif void evutil_memclear_(void *mem, size_t len); struct in_addr; struct in6_addr; /* This is a any, loopback, link-local, multicast */ EVENT2_EXPORT_SYMBOL int evutil_v4addr_is_local_(const struct in_addr *in); /* This is a reserved, ipv4compat, ipv4map, loopback, * link-local, multicast, or unspecified address. */ EVENT2_EXPORT_SYMBOL int evutil_v6addr_is_local_(const struct in6_addr *in); #ifdef __cplusplus } #endif #endif