2 * libev epoll fd activity backend
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de>
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
41 * general notes about epoll:
43 * a) epoll silently removes fds from the fd set. as nothing tells us
44 * that an fd has been removed otherwise, we have to continually
45 * "rearm" fds that we suspect *might* have changed (same
46 * problem with kqueue, but much less costly there).
47 * b) the fact that ADD != MOD creates a lot of extra syscalls due to a)
48 * and seems not to have any advantage.
49 * c) the inability to handle fork or file descriptors (think dup)
50 * limits the applicability over poll, so this is not a generic
52 * d) epoll doesn't work the same as select with many file descriptors
53 * (such as files). while not critical, no other advanced interface
54 * seems to share this (rather non-unixy) limitation.
55 * e) epoll claims to be embeddable, but in practise you never get
56 * a ready event for the epoll fd (broken: <=2.6.26, working: >=2.6.32).
57 * f) epoll_ctl returning EPERM means the fd is always ready.
59 * lots of "weird code" and complication handling in this file is due
60 * to these design problems with epoll, as we try very hard to avoid
61 * epoll_ctl syscalls for common usage patterns and handle the breakage
62 * ensuing from receiving events for closed and otherwise long gone
66 #include <sys/epoll.h>
68 #define EV_EMASK_EPERM 0x80
71 epoll_modify (EV_P_ int fd, int oev, int nev)
73 struct epoll_event ev;
74 unsigned char oldmask;
77 * we handle EPOLL_CTL_DEL by ignoring it here
78 * on the assumption that the fd is gone anyways
79 * if that is wrong, we have to handle the spurious
80 * event in epoll_poll.
81 * if the fd is added again, we try to ADD it, and, if that
82 * fails, we assume it still has the same eventmask.
87 oldmask = anfds [fd].emask;
88 anfds [fd].emask = nev;
90 /* store the generation counter in the upper 32 bits, the fd in the lower 32 bits */
91 ev.data.u64 = (uint64_t)(uint32_t)fd
92 | ((uint64_t)(uint32_t)++anfds [fd].egen << 32);
93 ev.events = (nev & EV_READ ? EPOLLIN : 0)
94 | (nev & EV_WRITE ? EPOLLOUT : 0);
96 if (expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
99 if (expect_true (errno == ENOENT))
101 /* if ENOENT then the fd went away, so try to do the right thing */
105 if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
108 else if (expect_true (errno == EEXIST))
110 /* EEXIST means we ignored a previous DEL, but the fd is still active */
111 /* if the kernel mask is the same as the new mask, we assume it hasn't changed */
115 if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
118 else if (expect_true (errno == EPERM))
120 /* EPERM means the fd is always ready, but epoll is too snobbish */
121 /* to handle it, unlike select or poll. */
122 anfds [fd].emask = EV_EMASK_EPERM;
124 /* add fd to epoll_eperms, if not already inside */
125 if (!(oldmask & EV_EMASK_EPERM))
127 array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2);
128 epoll_eperms [epoll_epermcnt++] = fd;
137 /* we didn't successfully call epoll_ctl, so decrement the generation counter again */
142 epoll_poll (EV_P_ ev_tstamp timeout)
147 if (expect_false (epoll_epermcnt))
150 /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */
151 /* the default libev max wait time, however. */
153 eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3);
156 if (expect_false (eventcnt < 0))
159 ev_syserr ("(libev) epoll_wait");
164 for (i = 0; i < eventcnt; ++i)
166 struct epoll_event *ev = epoll_events + i;
168 int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
169 int want = anfds [fd].events;
170 int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
171 | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
174 * check for spurious notification.
175 * this only finds spurious notifications on egen updates
176 * other spurious notifications will be found by epoll_ctl, below
177 * we assume that fd is always in range, as we never shrink the anfds array
179 if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
181 /* recreate kernel state */
186 if (expect_false (got & ~want))
188 anfds [fd].emask = want;
191 * we received an event but are not interested in it, try mod or del
192 * this often happens because we optimistically do not unregister fds
193 * when we are no longer interested in them, but also when we get spurious
194 * notifications for fds from another process. this is partially handled
195 * above with the gencounter check (== our fd is not the event fd), and
196 * partially here, when epoll_ctl returns an error (== a child has the fd
199 ev->events = (want & EV_READ ? EPOLLIN : 0)
200 | (want & EV_WRITE ? EPOLLOUT : 0);
202 /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */
203 /* which is fortunately easy to do for us. */
204 if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
206 postfork |= 2; /* an error occurred, recreate kernel state */
211 fd_event (EV_A_ fd, got);
214 /* if the receive array was full, increase its size */
215 if (expect_false (eventcnt == epoll_eventmax))
217 ev_free (epoll_events);
218 epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
219 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
222 /* now synthesize events for all fds where epoll fails, while select works... */
223 for (i = epoll_epermcnt; i--; )
225 int fd = epoll_eperms [i];
226 unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE);
228 if (anfds [fd].emask & EV_EMASK_EPERM && events)
229 fd_event (EV_A_ fd, events);
232 epoll_eperms [i] = epoll_eperms [--epoll_epermcnt];
233 anfds [fd].emask = 0;
239 epoll_init (EV_P_ int flags)
242 backend_fd = epoll_create1 (EPOLL_CLOEXEC);
244 if (backend_fd < 0 && (errno == EINVAL || errno == ENOSYS))
246 backend_fd = epoll_create (256);
251 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
253 backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */
254 backend_modify = epoll_modify;
255 backend_poll = epoll_poll;
257 epoll_eventmax = 64; /* initial number of events receivable per poll */
258 epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
260 return EVBACKEND_EPOLL;
266 ev_free (epoll_events);
267 array_free (epoll_eperm, EMPTY);
275 while ((backend_fd = epoll_create (256)) < 0)
276 ev_syserr ("(libev) epoll_create");
278 fcntl (backend_fd, F_SETFD, FD_CLOEXEC);