libev/ev_epoll.c

/*
 * libev epoll fd activity backend
 *
 * Copyright (c) 2007,2008,2009,2010,2011 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.
 * d) epoll doesn't work the same as select with many file descriptors
 *    (such as files). while not critical, no other advanced interface
 *    seems to share this (rather non-unixy) limitation.
 * e) epoll claims to be embeddable, but in practise you never get
 *    a ready event for the epoll fd (broken: <=2.6.26, working: >=2.6.32).
 * f) epoll_ctl returning EPERM means the fd is always ready.
 *
 * 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 and handle the breakage
 * ensuing from receiving events for closed and otherwise long gone
 * file descriptors.
 */

#include <sys/epoll.h>

#define EV_EMASK_EPERM 0x80

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.
   * if the fd is added again, 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, the fd in the lower 32 bits */
  ev.data.u64 = (uint64_t)(uint32_t)fd
              | ((uint64_t)(uint32_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 && oldmask != nev ? 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;
    }
  else if (expect_true (errno == EPERM))
    {
      /* EPERM means the fd is always ready, but epoll is too snobbish */
      /* to handle it, unlike select or poll. */
      anfds [fd].emask = EV_EMASK_EPERM;

      /* add fd to epoll_eperms, if not already inside */
      if (!(oldmask & EV_EMASK_EPERM))
        {
          array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2);
          epoll_eperms [epoll_epermcnt++] = fd;
        }

      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 times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */
  /* the default libev max wait time, however. */
  EV_RELEASE_CB;
  eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
  EV_ACQUIRE_CB;

  if (expect_false (eventcnt < 0))
    {
      if (errno != EINTR)
        ev_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 want = anfds [fd].events;
      int got  = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
               | (ev->events & (EPOLLIN  | EPOLLERR | EPOLLHUP) ? EV_READ  : 0);

      /* check for spurious notification */
      /* we assume that fd is always in range, as we never shrink the anfds array */
      if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
        {
          /* recreate kernel state */
          postfork = 1;
          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);

          /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */
          /* which is fortunately easy to do for us. */
          if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
            {
              postfork = 1; /* an error occurred, recreate kernel state */
              continue;
            }
        }

      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);
    }

  /* now synthesize events for all fds where epoll fails, while select works... */
  for (i = epoll_epermcnt; i--; )
    {
      int fd = epoll_eperms [i];
      unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE);

      if (anfds [fd].emask & EV_EMASK_EPERM && events)
        fd_event (EV_A_ fd, events);
      else
        epoll_eperms [i] = epoll_eperms [--epoll_epermcnt];
    }
}

int inline_size
epoll_init (EV_P_ int flags)
{
#ifdef EPOLL_CLOEXEC
  backend_fd = epoll_create1 (EPOLL_CLOEXEC);

  if (backend_fd <= 0)
#endif
    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; /* initial 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);
  array_free (epoll_eperm, EMPTY);
}

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

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

  fcntl (backend_fd, F_SETFD, FD_CLOEXEC);

  fd_rearm_all (EV_A);
}

Revision:	1.57
Committed:	Sat Jan 8 10:25:16 2011 UTC (13 years, 4 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.56:	+2 -0 lines
Log Message:	* empty log message *
#	User	Rev	Content
1	root	1.5	/*
2	root	1.9	* libev epoll fd activity backend
3			*
4	root	1.55	* Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de>
5	root	1.5	* All rights reserved.
6			*
7	root	1.33	* 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	root	1.5	*
28	root	1.33	* 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	root	1.5	*/
39
40	root	1.32	/*
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	root	1.50	* 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	root	1.55	* 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.
58	root	1.32	*
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	root	1.37	* epoll_ctl syscalls for common usage patterns and handle the breakage
62			* ensuing from receiving events for closed and otherwise long gone
63			* file descriptors.
64	root	1.32	*/
65
66	root	1.1	#include <sys/epoll.h>
67
68	root	1.55	#define EV_EMASK_EPERM 0x80
69
70	root	1.1	static void
71	root	1.13	epoll_modify (EV_P_ int fd, int oev, int nev)
72	root	1.1	{
73	root	1.32	struct epoll_event ev;
74	root	1.35	unsigned char oldmask;
75	root	1.32
76			/*
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	root	1.43	* if the fd is added again, we try to ADD it, and, if that
82	root	1.35	* fails, we assume it still has the same eventmask.
83	root	1.32	*/
84			if (!nev)
85			return;
86	root	1.1
87	root	1.35	oldmask = anfds [fd].emask;
88			anfds [fd].emask = nev;
89
90	root	1.42	/* 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	root	1.32	ev.events = (nev & EV_READ ? EPOLLIN : 0)
94			\| (nev & EV_WRITE ? EPOLLOUT : 0);
95
96	root	1.51	if (expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
97	root	1.32	return;
98
99			if (expect_true (errno == ENOENT))
100			{
101	root	1.35	/* if ENOENT then the fd went away, so try to do the right thing */
102	root	1.32	if (!nev)
103	root	1.36	goto dec_egen;
104	root	1.32
105			if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
106			return;
107			}
108			else if (expect_true (errno == EEXIST))
109			{
110	root	1.35	/* 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 */
112	root	1.36	if (oldmask == nev)
113			goto dec_egen;
114
115			if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
116	root	1.32	return;
117			}
118	root	1.55	else if (expect_true (errno == EPERM))
119			{
120	root	1.57	/* EPERM means the fd is always ready, but epoll is too snobbish */
121			/* to handle it, unlike select or poll. */
122	root	1.55	anfds [fd].emask = EV_EMASK_EPERM;
123
124			/* add fd to epoll_eperms, if not already inside */
125			if (!(oldmask & EV_EMASK_EPERM))
126			{
127			array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2);
128			epoll_eperms [epoll_epermcnt++] = fd;
129			}
130
131			return;
132			}
133	root	1.32
134			fd_kill (EV_A_ fd);
135	root	1.36
136			dec_egen:
137			/* we didn't successfully call epoll_ctl, so decrement the generation counter again */
138			--anfds [fd].egen;
139	root	1.1	}
140
141	root	1.6	static void
142	root	1.13	epoll_poll (EV_P_ ev_tstamp timeout)
143	root	1.1	{
144	root	1.20	int i;
145	root	1.44	int eventcnt;
146
147	root	1.48	/* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */
148			/* the default libev max wait time, however. */
149	root	1.46	EV_RELEASE_CB;
150	root	1.44	eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
151	root	1.46	EV_ACQUIRE_CB;
152	root	1.1
153	root	1.29	if (expect_false (eventcnt < 0))
154	root	1.20	{
155			if (errno != EINTR)
156	root	1.38	ev_syserr ("(libev) epoll_wait");
157	root	1.20
158			return;
159			}
160	root	1.1
161			for (i = 0; i < eventcnt; ++i)
162	root	1.32	{
163			struct epoll_event *ev = epoll_events + i;
164
165	root	1.37	int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
166			int want = anfds [fd].events;
167	root	1.32	int got = (ev->events & (EPOLLOUT \| EPOLLERR \| EPOLLHUP) ? EV_WRITE : 0)
168			\| (ev->events & (EPOLLIN \| EPOLLERR \| EPOLLHUP) ? EV_READ : 0);
169
170	root	1.37	/* check for spurious notification */
171	sf-exg	1.54	/* we assume that fd is always in range, as we never shrink the anfds array */
172	root	1.40	if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
173	root	1.37	{
174			/* recreate kernel state */
175			postfork = 1;
176			continue;
177			}
178	root	1.36
179	root	1.32	if (expect_false (got & ~want))
180			{
181	root	1.35	anfds [fd].emask = want;
182
183	root	1.32	/* we received an event but are not interested in it, try mod or del */
184	root	1.35	/* I don't think we ever need MOD, but let's handle it anyways */
185	root	1.32	ev->events = (want & EV_READ ? EPOLLIN : 0)
186			\| (want & EV_WRITE ? EPOLLOUT : 0);
187
188	root	1.49	/* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */
189			/* which is fortunately easy to do for us. */
190	root	1.37	if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
191			{
192	sf-exg	1.52	postfork = 1; /* an error occurred, recreate kernel state */
193	root	1.37	continue;
194			}
195	root	1.32	}
196
197			fd_event (EV_A_ fd, got);
198			}
199	root	1.1
200			/* if the receive array was full, increase its size */
201	root	1.14	if (expect_false (eventcnt == epoll_eventmax))
202	root	1.1	{
203	root	1.20	ev_free (epoll_events);
204	root	1.28	epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
205	root	1.23	epoll_events = (struct epoll_event )ev_malloc (sizeof (struct epoll_event) epoll_eventmax);
206	root	1.1	}
207	root	1.55
208	root	1.56	/* now synthesize events for all fds where epoll fails, while select works... */
209	root	1.55	for (i = epoll_epermcnt; i--; )
210			{
211			int fd = epoll_eperms [i];
212			unsigned char events = anfds [fd].events & (EV_READ \| EV_WRITE);
213
214			if (anfds [fd].emask & EV_EMASK_EPERM && events)
215			fd_event (EV_A_ fd, events);
216			else
217			epoll_eperms [i] = epoll_eperms [--epoll_epermcnt];
218			}
219	root	1.1	}
220
221	root	1.27	int inline_size
222	root	1.13	epoll_init (EV_P_ int flags)
223	root	1.1	{
224	root	1.47	#ifdef EPOLL_CLOEXEC
225			backend_fd = epoll_create1 (EPOLL_CLOEXEC);
226
227			if (backend_fd <= 0)
228			#endif
229			backend_fd = epoll_create (256);
230	root	1.1
231	root	1.26	if (backend_fd < 0)
232	root	1.13	return 0;
233	root	1.3
234	root	1.26	fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
235	root	1.1
236	root	1.31	backend_fudge = 0.; /* kernel sources seem to indicate this to be zero */
237	root	1.25	backend_modify = epoll_modify;
238			backend_poll = epoll_poll;
239	root	1.1
240	root	1.37	epoll_eventmax = 64; /* initial number of events receivable per poll */
241	root	1.23	epoll_events = (struct epoll_event )ev_malloc (sizeof (struct epoll_event) epoll_eventmax);
242	root	1.13
243	root	1.24	return EVBACKEND_EPOLL;
244	root	1.1	}
245	root	1.6
246	root	1.27	void inline_size
247	root	1.15	epoll_destroy (EV_P)
248			{
249	root	1.20	ev_free (epoll_events);
250	root	1.55	array_free (epoll_eperm, EMPTY);
251	root	1.15	}
252
253	root	1.27	void inline_size
254	root	1.15	epoll_fork (EV_P)
255			{
256	root	1.26	close (backend_fd);
257	root	1.21
258	root	1.26	while ((backend_fd = epoll_create (256)) < 0)
259	root	1.38	ev_syserr ("(libev) epoll_create");
260	root	1.21
261	root	1.26	fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
262	root	1.15
263	root	1.16	fd_rearm_all (EV_A);
264	root	1.15	}
265