epoll后端机制的实现代码在epoll.c文件中。
/*
* Copyright 2000-2007 Niels Provos <[email protected]>
* 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 <stdint.h>
#include <sys/types.h>
#include <sys/resource.h>
#ifdef EVENT__HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#include <sys/queue.h>
#include <sys/epoll.h>
#include <signal.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <errno.h>
#ifdef EVENT__HAVE_FCNTL_H
#include <fcntl.h>
#endif
#ifdef EVENT__HAVE_SYS_TIMERFD_H
#include <sys/timerfd.h>
#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,
, /* 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,
, /* need reinit */
EV_FEATURE_ET|EV_FEATURE_O1|EV_FEATURE_EARLY_CLOSE, }; #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 = -;
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 == -) {
/* Initialize the kernel queue using the old interface. (The
size field is ignored since 2.6.8.) */
if ((epfd = epoll_create()) == -) {
if (errno != ENOSYS)
event_warn("epoll_create");
return (NULL);
}
evutil_make_socket_closeonexec(epfd);
} if (!(epollop = mm_calloc(, 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) != ||
((base->flags & EVENT_BASE_FLAG_IGNORE_ENV) == &&
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 >= ) {
struct epoll_event epev;
memset(&epev, , sizeof(epev));
epev.data.fd = epollop->timerfd;
epev.events = EPOLLIN;
if (epoll_ctl(epollop->epfd, EPOLL_CTL_ADD, fd, &epev) < ) {
event_warn("epoll_ctl(timerfd)");
close(fd);
epollop->timerfd = -;
}
} 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 = -;
}
} else {
epollop->timerfd = -;
}
#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 == ) {
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 = ;
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 == );
return ;
} if ((ch->read_change|ch->write_change) & EV_CHANGE_ET)
events |= EPOLLET; memset(&epev, , sizeof(epev));
epev.data.fd = ch->fd;
epev.events = events;
if (epoll_ctl(epollop->epfd, op, ch->fd, &epev) == ) {
event_debug((PRINT_CHANGES(op, epev.events, ch, "okay")));
return ;
} 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) == -) {
event_warn("Epoll MOD(%d) on %d retried as ADD; that failed too",
(int)epev.events, ch->fd);
return -;
} else {
event_debug(("Epoll MOD(%d) on %d retried as ADD; succeeded.",
(int)epev.events,
ch->fd));
return ;
}
}
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) == -) {
event_warn("Epoll ADD(%d) on %d retried as MOD; that failed too",
(int)epev.events, ch->fd);
return -;
} else {
event_debug(("Epoll ADD(%d) on %d retried as MOD; succeeded.",
(int)epev.events,
ch->fd));
return ;
}
}
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 ;
}
break;
default:
break;
} event_warn(PRINT_CHANGES(op, epev.events, ch, "failed"));
return -;
} 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 = ;
int i; for (i = ; i < changelist->n_changes; ++i) {
ch = &changelist->changes[i];
if (epoll_apply_one_change(base, epollop, ch) < )
r = -;
} 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 = ;
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 = ;
if (events & EV_WRITE)
ch.write_change = EV_CHANGE_DEL;
if (events & EV_READ)
ch.read_change = EV_CHANGE_DEL;
if (events & EV_CLOSED)
ch.close_change = EV_CHANGE_DEL; 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 = -; #ifdef USING_TIMERFD
if (epollop->timerfd >= ) {
struct itimerspec is;
is.it_interval.tv_sec = ;
is.it_interval.tv_nsec = ;
if (tv == NULL) {
/* No timeout; disarm the timer. */
is.it_value.tv_sec = ;
is.it_value.tv_nsec = ;
} else {
if (tv->tv_sec == && tv->tv_usec == ) {
/* we need to exit immediately; timerfd can't
* do that. */
timeout = ;
}
is.it_value.tv_sec = tv->tv_sec;
is.it_value.tv_nsec = tv->tv_usec * ;
}
/* 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, , &is, NULL) < ) {
event_warn("timerfd_settime");
}
} else
#endif
if (tv != NULL) {
timeout = evutil_tv_to_msec_(tv);
if (timeout < || 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 == -) {
if (errno != EINTR) {
event_warn("epoll_wait");
return (-);
} return ();
} event_debug(("%s: epoll_wait reports %d", __func__, res));
EVUTIL_ASSERT(res <= epollop->nevents); for (i = ; i < res; i++) {
int what = events[i].events;
short ev = ;
#ifdef USING_TIMERFD
if (events[i].data.fd == epollop->timerfd)
continue;
#endif if (what & (EPOLLHUP|EPOLLERR)) {
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 * ;
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 ();
} 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 >= )
close(epollop->epfd);
#ifdef USING_TIMERFD
if (epollop->timerfd >= )
close(epollop->timerfd);
#endif memset(epollop, , sizeof(struct epollop));
mm_free(epollop);
} #endif /* EVENT__HAVE_EPOLL */
(1)第117行-127行定义的epollops对应了这篇文章里说的epoll后端机制的定义。
(2)该文件中定义了epoll_init函数用于初始化、epoll_add函数用于添加一个事件、epoll_del函数用于删除一个事件、epoll_dispatch用于事件循环。
1、epoll_init函数
1）调用epoll_create创建epfd。
2）在堆上分配一个struct epollop结构epollop。
3）把epollop的成员epfd赋值为刚才创建的epfd。
4）初始化成员events，调用mm_malloc函数来分配。
5）初始化成员nevents为INITIAL_NEVENT。
6）如果定义了USING_TIMERFD宏，就初始化成员timerfd。
7）调用svsig_init_函数。
8）返回epollop。
2、epoll_nochanglist_add函数
1）判断read、write、close是否有改变。
2）调用epoll_apply_one_change函数，在该函数中首先调用epool_ctl修改事件，然后处理各种异常情况，比如：ENOENT、EEXIST等等。
3、epoll_nochangelist_del函数
1）判断read、write、close是否有删除。
2）调用函数epoll_apply_one_change函数。
4、epoll_dispatch函数
1）通过event_base结构的evbase获取epollop指针，然后获取到初始化时传入的events指针并保存在events中。
2）获取timeout。
3）调用epoll_wait函数。
4）在一个for循环中处理激活事件，在每一次循环中，先把epoll事件转换为libevent定义的事件，EPOLLIN->EV_READ，EPOLLOUT->EV_WRITE，EPOLLRDHUP->EV_CLOSED，然后调用evmap_io_active_函数。
5）判断如果用完了所有事件，则为下一次准备更多的事件，扩展为原来的2倍，第一次默认是32。