libev/ev_epoll.c

/*
 * libev epoll fd activity backend
 *
 * Copyright (c) 2007,2008 Marc Alexander Lehmann <libev@schmorp.de>
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without modifica-
 * tion, 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.
 * 
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
 * CIAL, 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 OTH-
 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
 * OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * Alternatively, the contents of this file may be used under the terms of
 * the GNU General Public License ("GPL") version 2 or any later version,
 * in which case the provisions of the GPL are applicable instead of
 * the above. If you wish to allow the use of your version of this file
 * only under the terms of the GPL and not to allow others to use your
 * version of this file under the BSD license, indicate your decision
 * by deleting the provisions above and replace them with the notice
 * and other provisions required by the GPL. If you do not delete the
 * provisions above, a recipient may use your version of this file under
 * either the BSD or the GPL.
 */

/*
 * general notes about epoll:
 *
 * a) epoll silently removes fds from the fd set. as nothing tells us
 *    that an fd has been removed otherwise, we have to continually
 *    "rearm" fds that we suspect *might* have changed (same
 *    problem with kqueue, but much less costly there).
 * b) the fact that ADD != MOD creates a lot of extra syscalls due to a)
 *    and seems not to have any advantage.
 * c) the inability to handle fork or file descriptors (think dup)
 *    limits the applicability over poll, so this is not a generic
 *    poll replacement.
 *
 * lots of "weird code" and complication handling in this file is due
 * to these design problems with epoll, as we try very hard to avoid
 * epoll_ctl syscalls for common usage patterns.
 */

#include <sys/epoll.h>

void inline_size
unsigned_char_init (unsigned char *base, int count)
{
  /* memset might be overkill */
  while (count--)
    *base++ = 0;
}

static void
epoll_modify (EV_P_ int fd, int oev, int nev)
{
  struct epoll_event ev;
  unsigned char oldmask;

  /*
   * we handle EPOLL_CTL_DEL by ignoring it here
   * on the assumption that the fd is gone anyways
   * if that is wrong, we have to handle the spurious
   * event in epoll_poll.
   * the fd is later added, we try to ADD it, and, if that
   * fails, we assume it still has the same eventmask.
   */
  if (!nev)
    return;

  oldmask = anfds [fd].emask;
  anfds [fd].emask = nev;

  /* store the generation counter in the upper 32 bits */
  ev.data.u64 = fd | ((uint64_t)++anfds [fd].egen << 32);
  ev.events   = (nev & EV_READ  ? EPOLLIN  : 0)
              | (nev & EV_WRITE ? EPOLLOUT : 0);

  if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
    return;

  if (expect_true (errno == ENOENT))
    {
      /* if ENOENT then the fd went away, so try to do the right thing */
      if (!nev)
        goto dec_egen;

      if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
        return;
    }
  else if (expect_true (errno == EEXIST))
    {
      /* EEXIST means we ignored a previous DEL, but the fd is still active */
      /* if the kernel mask is the same as the new mask, we assume it hasn't changed */
      if (oldmask == nev)
        goto dec_egen;

      if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
        return;
    }

  fd_kill (EV_A_ fd);

dec_egen:
  /* we didn't successfully call epoll_ctl, so decrement the generation counter again */
  --anfds [fd].egen;
}

static void
epoll_poll (EV_P_ ev_tstamp timeout)
{
  int i;
  int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));

  if (expect_false (eventcnt < 0))
    {
      if (errno != EINTR)
        syserr ("(libev) epoll_wait");

      return;
    }

  for (i = 0; i < eventcnt; ++i)
    {
      struct epoll_event *ev = epoll_events + i;

      int fd   = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
      int got  = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
               | (ev->events & (EPOLLIN  | EPOLLERR | EPOLLHUP) ? EV_READ  : 0);
      int want = anfds [fd].events;

      if (anfds [fd].egen != (unsigned char)(ev->data.u64 >> 32))
        /*fprintf (stderr, "spurious notification fd %d, %d vs %d\n", fd, (int)(ev->data.u64 >> 32), anfds [fd].egen);*/
        continue;

      if (expect_false (got & ~want))
        {
          anfds [fd].emask = want;

          /* we received an event but are not interested in it, try mod or del */
          /* I don't think we ever need MOD, but let's handle it anyways */
          ev->events = (want & EV_READ  ? EPOLLIN  : 0)
                     | (want & EV_WRITE ? EPOLLOUT : 0);

          epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev);
        }

      fd_event (EV_A_ fd, got);
    }

  /* if the receive array was full, increase its size */
  if (expect_false (eventcnt == epoll_eventmax))
    {
      ev_free (epoll_events);
      epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
      epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
    }
}

int inline_size
epoll_init (EV_P_ int flags)
{
  backend_fd = epoll_create (256);

  if (backend_fd < 0)
    return 0;

  fcntl (backend_fd, F_SETFD, FD_CLOEXEC);

  backend_fudge  = 0.; /* kernel sources seem to indicate this to be zero */
  backend_modify = epoll_modify;
  backend_poll   = epoll_poll;

  epoll_eventmax = 64; /* intiial number of events receivable per poll */
  epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);

  return EVBACKEND_EPOLL;
}

void inline_size
epoll_destroy (EV_P)
{
  ev_free (epoll_events);
}

void inline_size
epoll_fork (EV_P)
{
  close (backend_fd);

  while ((backend_fd = epoll_create (256)) < 0)
    syserr ("(libev) epoll_create");

  fcntl (backend_fd, F_SETFD, FD_CLOEXEC);

  fd_rearm_all (EV_A);
}

Revision:	1.36
Committed:	Mon Oct 27 11:08:29 2008 UTC (15 years, 6 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.35:	+16 -4 lines
Log Message:	work around epoll spurious notifications
#	Content
1	/*
2	* libev epoll fd activity backend
3	*
4	* Copyright (c) 2007,2008 Marc Alexander Lehmann <libev@schmorp.de>
5	* All rights reserved.
6	*
7	* Redistribution and use in source and binary forms, with or without modifica-
8	* tion, are permitted provided that the following conditions are met:
9	*
10	* 1. Redistributions of source code must retain the above copyright notice,
11	* this list of conditions and the following disclaimer.
12	*
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.
16	*
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.
27	*
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.
38	*/
39
40	/*
41	* general notes about epoll:
42	*
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
51	* poll replacement.
52	*
53	* lots of "weird code" and complication handling in this file is due
54	* to these design problems with epoll, as we try very hard to avoid
55	* epoll_ctl syscalls for common usage patterns.
56	*/
57
58	#include <sys/epoll.h>
59
60	void inline_size
61	unsigned_char_init (unsigned char *base, int count)
62	{
63	/* memset might be overkill */
64	while (count--)
65	*base++ = 0;
66	}
67
68	static void
69	epoll_modify (EV_P_ int fd, int oev, int nev)
70	{
71	struct epoll_event ev;
72	unsigned char oldmask;
73
74	/*
75	* we handle EPOLL_CTL_DEL by ignoring it here
76	* on the assumption that the fd is gone anyways
77	* if that is wrong, we have to handle the spurious
78	* event in epoll_poll.
79	* the fd is later added, we try to ADD it, and, if that
80	* fails, we assume it still has the same eventmask.
81	*/
82	if (!nev)
83	return;
84
85	oldmask = anfds [fd].emask;
86	anfds [fd].emask = nev;
87
88	/* store the generation counter in the upper 32 bits */
89	ev.data.u64 = fd \| ((uint64_t)++anfds [fd].egen << 32);
90	ev.events = (nev & EV_READ ? EPOLLIN : 0)
91	\| (nev & EV_WRITE ? EPOLLOUT : 0);
92
93	if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
94	return;
95
96	if (expect_true (errno == ENOENT))
97	{
98	/* if ENOENT then the fd went away, so try to do the right thing */
99	if (!nev)
100	goto dec_egen;
101
102	if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
103	return;
104	}
105	else if (expect_true (errno == EEXIST))
106	{
107	/* EEXIST means we ignored a previous DEL, but the fd is still active */
108	/* if the kernel mask is the same as the new mask, we assume it hasn't changed */
109	if (oldmask == nev)
110	goto dec_egen;
111
112	if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
113	return;
114	}
115
116	fd_kill (EV_A_ fd);
117
118	dec_egen:
119	/* we didn't successfully call epoll_ctl, so decrement the generation counter again */
120	--anfds [fd].egen;
121	}
122
123	static void
124	epoll_poll (EV_P_ ev_tstamp timeout)
125	{
126	int i;
127	int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
128
129	if (expect_false (eventcnt < 0))
130	{
131	if (errno != EINTR)
132	syserr ("(libev) epoll_wait");
133
134	return;
135	}
136
137	for (i = 0; i < eventcnt; ++i)
138	{
139	struct epoll_event *ev = epoll_events + i;
140
141	int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
142	int got = (ev->events & (EPOLLOUT \| EPOLLERR \| EPOLLHUP) ? EV_WRITE : 0)
143	\| (ev->events & (EPOLLIN \| EPOLLERR \| EPOLLHUP) ? EV_READ : 0);
144	int want = anfds [fd].events;
145
146	if (anfds [fd].egen != (unsigned char)(ev->data.u64 >> 32))
147	/fprintf (stderr, "spurious notification fd %d, %d vs %d\n", fd, (int)(ev->data.u64 >> 32), anfds [fd].egen);/
148	continue;
149
150	if (expect_false (got & ~want))
151	{
152	anfds [fd].emask = want;
153
154	/* we received an event but are not interested in it, try mod or del */
155	/* I don't think we ever need MOD, but let's handle it anyways */
156	ev->events = (want & EV_READ ? EPOLLIN : 0)
157	\| (want & EV_WRITE ? EPOLLOUT : 0);
158
159	epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev);
160	}
161
162	fd_event (EV_A_ fd, got);
163	}
164
165	/* if the receive array was full, increase its size */
166	if (expect_false (eventcnt == epoll_eventmax))
167	{
168	ev_free (epoll_events);
169	epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
170	epoll_events = (struct epoll_event )ev_malloc (sizeof (struct epoll_event) epoll_eventmax);
171	}
172	}
173
174	int inline_size
175	epoll_init (EV_P_ int flags)
176	{
177	backend_fd = epoll_create (256);
178
179	if (backend_fd < 0)
180	return 0;
181
182	fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
183
184	backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */
185	backend_modify = epoll_modify;
186	backend_poll = epoll_poll;
187
188	epoll_eventmax = 64; /* intiial number of events receivable per poll */
189	epoll_events = (struct epoll_event )ev_malloc (sizeof (struct epoll_event) epoll_eventmax);
190
191	return EVBACKEND_EPOLL;
192	}
193
194	void inline_size
195	epoll_destroy (EV_P)
196	{
197	ev_free (epoll_events);
198	}
199
200	void inline_size
201	epoll_fork (EV_P)
202	{
203	close (backend_fd);
204
205	while ((backend_fd = epoll_create (256)) < 0)
206	syserr ("(libev) epoll_create");
207
208	fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
209
210	fd_rearm_all (EV_A);
211	}
212