1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/ipc/chromium/src/third_party/libevent/epoll.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,473 @@
1.4 +/*
1.5 + * Copyright 2000-2007 Niels Provos <provos@citi.umich.edu>
1.6 + * Copyright 2007-2012 Niels Provos, Nick Mathewson
1.7 + *
1.8 + * Redistribution and use in source and binary forms, with or without
1.9 + * modification, are permitted provided that the following conditions
1.10 + * are met:
1.11 + * 1. Redistributions of source code must retain the above copyright
1.12 + * notice, this list of conditions and the following disclaimer.
1.13 + * 2. Redistributions in binary form must reproduce the above copyright
1.14 + * notice, this list of conditions and the following disclaimer in the
1.15 + * documentation and/or other materials provided with the distribution.
1.16 + * 3. The name of the author may not be used to endorse or promote products
1.17 + * derived from this software without specific prior written permission.
1.18 + *
1.19 + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
1.20 + * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
1.21 + * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
1.22 + * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
1.23 + * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
1.24 + * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
1.25 + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
1.26 + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
1.27 + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
1.28 + * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
1.29 + */
1.30 +#include "event2/event-config.h"
1.31 +
1.32 +#include <stdint.h>
1.33 +#include <sys/types.h>
1.34 +#include <sys/resource.h>
1.35 +#ifdef _EVENT_HAVE_SYS_TIME_H
1.36 +#include <sys/time.h>
1.37 +#endif
1.38 +#include <sys/queue.h>
1.39 +#include <sys/epoll.h>
1.40 +#include <signal.h>
1.41 +#include <limits.h>
1.42 +#include <stdio.h>
1.43 +#include <stdlib.h>
1.44 +#include <string.h>
1.45 +#include <unistd.h>
1.46 +#include <errno.h>
1.47 +#ifdef _EVENT_HAVE_FCNTL_H
1.48 +#include <fcntl.h>
1.49 +#endif
1.50 +
1.51 +#include "event-internal.h"
1.52 +#include "evsignal-internal.h"
1.53 +#include "event2/thread.h"
1.54 +#include "evthread-internal.h"
1.55 +#include "log-internal.h"
1.56 +#include "evmap-internal.h"
1.57 +#include "changelist-internal.h"
1.58 +
1.59 +struct epollop {
1.60 + struct epoll_event *events;
1.61 + int nevents;
1.62 + int epfd;
1.63 +};
1.64 +
1.65 +static void *epoll_init(struct event_base *);
1.66 +static int epoll_dispatch(struct event_base *, struct timeval *);
1.67 +static void epoll_dealloc(struct event_base *);
1.68 +
1.69 +static const struct eventop epollops_changelist = {
1.70 + "epoll (with changelist)",
1.71 + epoll_init,
1.72 + event_changelist_add,
1.73 + event_changelist_del,
1.74 + epoll_dispatch,
1.75 + epoll_dealloc,
1.76 + 1, /* need reinit */
1.77 + EV_FEATURE_ET|EV_FEATURE_O1,
1.78 + EVENT_CHANGELIST_FDINFO_SIZE
1.79 +};
1.80 +
1.81 +
1.82 +static int epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd,
1.83 + short old, short events, void *p);
1.84 +static int epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd,
1.85 + short old, short events, void *p);
1.86 +
1.87 +const struct eventop epollops = {
1.88 + "epoll",
1.89 + epoll_init,
1.90 + epoll_nochangelist_add,
1.91 + epoll_nochangelist_del,
1.92 + epoll_dispatch,
1.93 + epoll_dealloc,
1.94 + 1, /* need reinit */
1.95 + EV_FEATURE_ET|EV_FEATURE_O1,
1.96 + 0
1.97 +};
1.98 +
1.99 +#define INITIAL_NEVENT 32
1.100 +#define MAX_NEVENT 4096
1.101 +
1.102 +/* On Linux kernels at least up to 2.6.24.4, epoll can't handle timeout
1.103 + * values bigger than (LONG_MAX - 999ULL)/HZ. HZ in the wild can be
1.104 + * as big as 1000, and LONG_MAX can be as small as (1<<31)-1, so the
1.105 + * largest number of msec we can support here is 2147482. Let's
1.106 + * round that down by 47 seconds.
1.107 + */
1.108 +#define MAX_EPOLL_TIMEOUT_MSEC (35*60*1000)
1.109 +
1.110 +static void *
1.111 +epoll_init(struct event_base *base)
1.112 +{
1.113 + int epfd;
1.114 + struct epollop *epollop;
1.115 +
1.116 + /* Initialize the kernel queue. (The size field is ignored since
1.117 + * 2.6.8.) */
1.118 + if ((epfd = epoll_create(32000)) == -1) {
1.119 + if (errno != ENOSYS)
1.120 + event_warn("epoll_create");
1.121 + return (NULL);
1.122 + }
1.123 +
1.124 + evutil_make_socket_closeonexec(epfd);
1.125 +
1.126 + if (!(epollop = mm_calloc(1, sizeof(struct epollop)))) {
1.127 + close(epfd);
1.128 + return (NULL);
1.129 + }
1.130 +
1.131 + epollop->epfd = epfd;
1.132 +
1.133 + /* Initialize fields */
1.134 + epollop->events = mm_calloc(INITIAL_NEVENT, sizeof(struct epoll_event));
1.135 + if (epollop->events == NULL) {
1.136 + mm_free(epollop);
1.137 + close(epfd);
1.138 + return (NULL);
1.139 + }
1.140 + epollop->nevents = INITIAL_NEVENT;
1.141 +
1.142 + if ((base->flags & EVENT_BASE_FLAG_EPOLL_USE_CHANGELIST) != 0 ||
1.143 + ((base->flags & EVENT_BASE_FLAG_IGNORE_ENV) == 0 &&
1.144 + evutil_getenv("EVENT_EPOLL_USE_CHANGELIST") != NULL))
1.145 + base->evsel = &epollops_changelist;
1.146 +
1.147 + evsig_init(base);
1.148 +
1.149 + return (epollop);
1.150 +}
1.151 +
1.152 +static const char *
1.153 +change_to_string(int change)
1.154 +{
1.155 + change &= (EV_CHANGE_ADD|EV_CHANGE_DEL);
1.156 + if (change == EV_CHANGE_ADD) {
1.157 + return "add";
1.158 + } else if (change == EV_CHANGE_DEL) {
1.159 + return "del";
1.160 + } else if (change == 0) {
1.161 + return "none";
1.162 + } else {
1.163 + return "???";
1.164 + }
1.165 +}
1.166 +
1.167 +static const char *
1.168 +epoll_op_to_string(int op)
1.169 +{
1.170 + return op == EPOLL_CTL_ADD?"ADD":
1.171 + op == EPOLL_CTL_DEL?"DEL":
1.172 + op == EPOLL_CTL_MOD?"MOD":
1.173 + "???";
1.174 +}
1.175 +
1.176 +static int
1.177 +epoll_apply_one_change(struct event_base *base,
1.178 + struct epollop *epollop,
1.179 + const struct event_change *ch)
1.180 +{
1.181 + struct epoll_event epev;
1.182 + int op, events = 0;
1.183 +
1.184 + if (1) {
1.185 + /* The logic here is a little tricky. If we had no events set
1.186 + on the fd before, we need to set op="ADD" and set
1.187 + events=the events we want to add. If we had any events set
1.188 + on the fd before, and we want any events to remain on the
1.189 + fd, we need to say op="MOD" and set events=the events we
1.190 + want to remain. But if we want to delete the last event,
1.191 + we say op="DEL" and set events=the remaining events. What
1.192 + fun!
1.193 + */
1.194 +
1.195 + /* TODO: Turn this into a switch or a table lookup. */
1.196 +
1.197 + if ((ch->read_change & EV_CHANGE_ADD) ||
1.198 + (ch->write_change & EV_CHANGE_ADD)) {
1.199 + /* If we are adding anything at all, we'll want to do
1.200 + * either an ADD or a MOD. */
1.201 + events = 0;
1.202 + op = EPOLL_CTL_ADD;
1.203 + if (ch->read_change & EV_CHANGE_ADD) {
1.204 + events |= EPOLLIN;
1.205 + } else if (ch->read_change & EV_CHANGE_DEL) {
1.206 + ;
1.207 + } else if (ch->old_events & EV_READ) {
1.208 + events |= EPOLLIN;
1.209 + }
1.210 + if (ch->write_change & EV_CHANGE_ADD) {
1.211 + events |= EPOLLOUT;
1.212 + } else if (ch->write_change & EV_CHANGE_DEL) {
1.213 + ;
1.214 + } else if (ch->old_events & EV_WRITE) {
1.215 + events |= EPOLLOUT;
1.216 + }
1.217 + if ((ch->read_change|ch->write_change) & EV_ET)
1.218 + events |= EPOLLET;
1.219 +
1.220 + if (ch->old_events) {
1.221 + /* If MOD fails, we retry as an ADD, and if
1.222 + * ADD fails we will retry as a MOD. So the
1.223 + * only hard part here is to guess which one
1.224 + * will work. As a heuristic, we'll try
1.225 + * MOD first if we think there were old
1.226 + * events and ADD if we think there were none.
1.227 + *
1.228 + * We can be wrong about the MOD if the file
1.229 + * has in fact been closed and re-opened.
1.230 + *
1.231 + * We can be wrong about the ADD if the
1.232 + * the fd has been re-created with a dup()
1.233 + * of the same file that it was before.
1.234 + */
1.235 + op = EPOLL_CTL_MOD;
1.236 + }
1.237 + } else if ((ch->read_change & EV_CHANGE_DEL) ||
1.238 + (ch->write_change & EV_CHANGE_DEL)) {
1.239 + /* If we're deleting anything, we'll want to do a MOD
1.240 + * or a DEL. */
1.241 + op = EPOLL_CTL_DEL;
1.242 +
1.243 + if (ch->read_change & EV_CHANGE_DEL) {
1.244 + if (ch->write_change & EV_CHANGE_DEL) {
1.245 + events = EPOLLIN|EPOLLOUT;
1.246 + } else if (ch->old_events & EV_WRITE) {
1.247 + events = EPOLLOUT;
1.248 + op = EPOLL_CTL_MOD;
1.249 + } else {
1.250 + events = EPOLLIN;
1.251 + }
1.252 + } else if (ch->write_change & EV_CHANGE_DEL) {
1.253 + if (ch->old_events & EV_READ) {
1.254 + events = EPOLLIN;
1.255 + op = EPOLL_CTL_MOD;
1.256 + } else {
1.257 + events = EPOLLOUT;
1.258 + }
1.259 + }
1.260 + }
1.261 +
1.262 + if (!events)
1.263 + return 0;
1.264 +
1.265 + memset(&epev, 0, sizeof(epev));
1.266 + epev.data.fd = ch->fd;
1.267 + epev.events = events;
1.268 + if (epoll_ctl(epollop->epfd, op, ch->fd, &epev) == -1) {
1.269 + if (op == EPOLL_CTL_MOD && errno == ENOENT) {
1.270 + /* If a MOD operation fails with ENOENT, the
1.271 + * fd was probably closed and re-opened. We
1.272 + * should retry the operation as an ADD.
1.273 + */
1.274 + if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, ch->fd, &epev) == -1) {
1.275 + event_warn("Epoll MOD(%d) on %d retried as ADD; that failed too",
1.276 + (int)epev.events, ch->fd);
1.277 + return -1;
1.278 + } else {
1.279 + event_debug(("Epoll MOD(%d) on %d retried as ADD; succeeded.",
1.280 + (int)epev.events,
1.281 + ch->fd));
1.282 + }
1.283 + } else if (op == EPOLL_CTL_ADD && errno == EEXIST) {
1.284 + /* If an ADD operation fails with EEXIST,
1.285 + * either the operation was redundant (as with a
1.286 + * precautionary add), or we ran into a fun
1.287 + * kernel bug where using dup*() to duplicate the
1.288 + * same file into the same fd gives you the same epitem
1.289 + * rather than a fresh one. For the second case,
1.290 + * we must retry with MOD. */
1.291 + if (epoll_ctl(epollop->epfd, EPOLL_CTL_MOD, ch->fd, &epev) == -1) {
1.292 + event_warn("Epoll ADD(%d) on %d retried as MOD; that failed too",
1.293 + (int)epev.events, ch->fd);
1.294 + return -1;
1.295 + } else {
1.296 + event_debug(("Epoll ADD(%d) on %d retried as MOD; succeeded.",
1.297 + (int)epev.events,
1.298 + ch->fd));
1.299 + }
1.300 + } else if (op == EPOLL_CTL_DEL &&
1.301 + (errno == ENOENT || errno == EBADF ||
1.302 + errno == EPERM)) {
1.303 + /* If a delete fails with one of these errors,
1.304 + * that's fine too: we closed the fd before we
1.305 + * got around to calling epoll_dispatch. */
1.306 + event_debug(("Epoll DEL(%d) on fd %d gave %s: DEL was unnecessary.",
1.307 + (int)epev.events,
1.308 + ch->fd,
1.309 + strerror(errno)));
1.310 + } else {
1.311 + event_warn("Epoll %s(%d) on fd %d failed. Old events were %d; read change was %d (%s); write change was %d (%s)",
1.312 + epoll_op_to_string(op),
1.313 + (int)epev.events,
1.314 + ch->fd,
1.315 + ch->old_events,
1.316 + ch->read_change,
1.317 + change_to_string(ch->read_change),
1.318 + ch->write_change,
1.319 + change_to_string(ch->write_change));
1.320 + return -1;
1.321 + }
1.322 + } else {
1.323 + event_debug(("Epoll %s(%d) on fd %d okay. [old events were %d; read change was %d; write change was %d]",
1.324 + epoll_op_to_string(op),
1.325 + (int)epev.events,
1.326 + (int)ch->fd,
1.327 + ch->old_events,
1.328 + ch->read_change,
1.329 + ch->write_change));
1.330 + }
1.331 + }
1.332 + return 0;
1.333 +}
1.334 +
1.335 +static int
1.336 +epoll_apply_changes(struct event_base *base)
1.337 +{
1.338 + struct event_changelist *changelist = &base->changelist;
1.339 + struct epollop *epollop = base->evbase;
1.340 + struct event_change *ch;
1.341 +
1.342 + int r = 0;
1.343 + int i;
1.344 +
1.345 + for (i = 0; i < changelist->n_changes; ++i) {
1.346 + ch = &changelist->changes[i];
1.347 + if (epoll_apply_one_change(base, epollop, ch) < 0)
1.348 + r = -1;
1.349 + }
1.350 +
1.351 + return (r);
1.352 +}
1.353 +
1.354 +static int
1.355 +epoll_nochangelist_add(struct event_base *base, evutil_socket_t fd,
1.356 + short old, short events, void *p)
1.357 +{
1.358 + struct event_change ch;
1.359 + ch.fd = fd;
1.360 + ch.old_events = old;
1.361 + ch.read_change = ch.write_change = 0;
1.362 + if (events & EV_WRITE)
1.363 + ch.write_change = EV_CHANGE_ADD |
1.364 + (events & EV_ET);
1.365 + if (events & EV_READ)
1.366 + ch.read_change = EV_CHANGE_ADD |
1.367 + (events & EV_ET);
1.368 +
1.369 + return epoll_apply_one_change(base, base->evbase, &ch);
1.370 +}
1.371 +
1.372 +static int
1.373 +epoll_nochangelist_del(struct event_base *base, evutil_socket_t fd,
1.374 + short old, short events, void *p)
1.375 +{
1.376 + struct event_change ch;
1.377 + ch.fd = fd;
1.378 + ch.old_events = old;
1.379 + ch.read_change = ch.write_change = 0;
1.380 + if (events & EV_WRITE)
1.381 + ch.write_change = EV_CHANGE_DEL;
1.382 + if (events & EV_READ)
1.383 + ch.read_change = EV_CHANGE_DEL;
1.384 +
1.385 + return epoll_apply_one_change(base, base->evbase, &ch);
1.386 +}
1.387 +
1.388 +static int
1.389 +epoll_dispatch(struct event_base *base, struct timeval *tv)
1.390 +{
1.391 + struct epollop *epollop = base->evbase;
1.392 + struct epoll_event *events = epollop->events;
1.393 + int i, res;
1.394 + long timeout = -1;
1.395 +
1.396 + if (tv != NULL) {
1.397 + timeout = evutil_tv_to_msec(tv);
1.398 + if (timeout < 0 || timeout > MAX_EPOLL_TIMEOUT_MSEC) {
1.399 + /* Linux kernels can wait forever if the timeout is
1.400 + * too big; see comment on MAX_EPOLL_TIMEOUT_MSEC. */
1.401 + timeout = MAX_EPOLL_TIMEOUT_MSEC;
1.402 + }
1.403 + }
1.404 +
1.405 + epoll_apply_changes(base);
1.406 + event_changelist_remove_all(&base->changelist, base);
1.407 +
1.408 + EVBASE_RELEASE_LOCK(base, th_base_lock);
1.409 +
1.410 + res = epoll_wait(epollop->epfd, events, epollop->nevents, timeout);
1.411 +
1.412 + EVBASE_ACQUIRE_LOCK(base, th_base_lock);
1.413 +
1.414 + if (res == -1) {
1.415 + if (errno != EINTR) {
1.416 + event_warn("epoll_wait");
1.417 + return (-1);
1.418 + }
1.419 +
1.420 + return (0);
1.421 + }
1.422 +
1.423 + event_debug(("%s: epoll_wait reports %d", __func__, res));
1.424 + EVUTIL_ASSERT(res <= epollop->nevents);
1.425 +
1.426 + for (i = 0; i < res; i++) {
1.427 + int what = events[i].events;
1.428 + short ev = 0;
1.429 +
1.430 + if (what & (EPOLLHUP|EPOLLERR)) {
1.431 + ev = EV_READ | EV_WRITE;
1.432 + } else {
1.433 + if (what & EPOLLIN)
1.434 + ev |= EV_READ;
1.435 + if (what & EPOLLOUT)
1.436 + ev |= EV_WRITE;
1.437 + }
1.438 +
1.439 + if (!ev)
1.440 + continue;
1.441 +
1.442 + evmap_io_active(base, events[i].data.fd, ev | EV_ET);
1.443 + }
1.444 +
1.445 + if (res == epollop->nevents && epollop->nevents < MAX_NEVENT) {
1.446 + /* We used all of the event space this time. We should
1.447 + be ready for more events next time. */
1.448 + int new_nevents = epollop->nevents * 2;
1.449 + struct epoll_event *new_events;
1.450 +
1.451 + new_events = mm_realloc(epollop->events,
1.452 + new_nevents * sizeof(struct epoll_event));
1.453 + if (new_events) {
1.454 + epollop->events = new_events;
1.455 + epollop->nevents = new_nevents;
1.456 + }
1.457 + }
1.458 +
1.459 + return (0);
1.460 +}
1.461 +
1.462 +
1.463 +static void
1.464 +epoll_dealloc(struct event_base *base)
1.465 +{
1.466 + struct epollop *epollop = base->evbase;
1.467 +
1.468 + evsig_dealloc(base);
1.469 + if (epollop->events)
1.470 + mm_free(epollop->events);
1.471 + if (epollop->epfd >= 0)
1.472 + close(epollop->epfd);
1.473 +
1.474 + memset(epollop, 0, sizeof(struct epollop));
1.475 + mm_free(epollop);
1.476 +}
