libev/ev_kqueue.c

/*
 * Copyright 2007      Marc Alexander Lehmann <libev@schmorp.de>
 * Copyright 2000-2002 Niels Provos <provos@citi.umich.edu>
 * All rights reserved.
 *
 * 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 <sys/types.h>
#include <sys/time.h>
#include <sys/queue.h>
#include <sys/event.h>
#include <string.h>
#include <errno.h>

static void
kqueue_change (EV_P_ int fd, int filter, int flags, int fflags)
{
  struct kevent *ke;

  ++kqueue_changecnt;
  array_needsize (struct kevent, kqueue_changes, kqueue_changemax, kqueue_changecnt, EMPTY2);

  ke = &kqueue_changes [kqueue_changecnt - 1];
  memset (ke, 0, sizeof (struct kevent));
  ke->ident  = fd;
  ke->filter = filter;
  ke->flags  = flags;
  ke->fflags = fflags;
}

#ifndef NOTE_EOF
# define NOTE_EOF 0
#endif

static void
kqueue_modify (EV_P_ int fd, int oev, int nev)
{
  if (oev != nev)
    {
      if (oev & EV_READ)
        kqueue_change (EV_A_ fd, EVFILT_READ , EV_DELETE, 0);

      if (oev & EV_WRITE)
        kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_DELETE, 0);
    }

  /* to detect close/reopen reliably, we have to re-add */
  /* event requests even when oev == nev */

  if (nev & EV_READ)
    kqueue_change (EV_A_ fd, EVFILT_READ , EV_ADD, NOTE_EOF);

  if (nev & EV_WRITE)
    kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_ADD, NOTE_EOF);
}

static void
kqueue_poll (EV_P_ ev_tstamp timeout)
{
  int res, i;
  struct timespec ts;

  /* need to resize so there is enough space for errors */
  if (kqueue_changecnt > kqueue_eventmax)
    {
      ev_free (kqueue_events);
      kqueue_eventmax = array_roundsize (struct kevent, kqueue_changecnt);
      kqueue_events = (struct kevent *)ev_malloc (sizeof (struct kevent) * kqueue_eventmax);
    }

  ts.tv_sec  = (time_t)timeout;
  ts.tv_nsec = (long)((timeout - (ev_tstamp)ts.tv_sec) * 1e9);
  res = kevent (kqueue_fd, kqueue_changes, kqueue_changecnt, kqueue_events, kqueue_eventmax, &ts);
  kqueue_changecnt = 0;

  if (res < 0)
    { 
      if (errno != EINTR)
        syserr ("(libev) kevent");

      return;
    } 

  for (i = 0; i < res; ++i)
    {
      int fd = kqueue_events [i].ident;

      if (kqueue_events [i].flags & EV_ERROR)
        {
          int err = kqueue_events [i].data;

          /* 
           * errors that may happen
           *   EBADF happens when the file discriptor has been
           *   closed,
           *   ENOENT when the file descriptor was closed and
           *   then reopened.
           *   EINVAL for some reasons not understood; EINVAL
           *   should not be returned ever; but FreeBSD does :-\
           */

          /* we are only interested in errors for fds that we are interested in :) */
          if (anfds [fd].events)
            {
              if (err == ENOENT) /* resubmit changes on ENOENT */
                kqueue_modify (EV_A_ fd, 0, anfds [fd].events);
              else if (err == EBADF) /* on EBADF, we re-check the fd */
                {
                  if (fd_valid (fd))
                    kqueue_modify (EV_A_ fd, 0, anfds [fd].events);
                  else
                    fd_kill (EV_A_ fd);
                }
              else /* on all other errors, we error out on the fd */
                fd_kill (EV_A_ fd);
            }
        }
      else
        fd_event (
          EV_A_
          fd,
          kqueue_events [i].filter == EVFILT_READ ? EV_READ
          : kqueue_events [i].filter == EVFILT_WRITE ? EV_WRITE
          : 0
        );
    }

  if (expect_false (res == kqueue_eventmax))
    {
      ev_free (kqueue_events);
      kqueue_eventmax = array_roundsize (struct kevent, kqueue_eventmax << 1);
      kqueue_events = (struct kevent *)ev_malloc (sizeof (struct kevent) * kqueue_eventmax);
    }
}

static int
kqueue_init (EV_P_ int flags)
{
  struct kevent ch, ev;

  /* Initalize the kernel queue */
  if ((kqueue_fd = kqueue ()) < 0)
    return 0;

  fcntl (kqueue_fd, F_SETFD, FD_CLOEXEC); /* not sure if necessary, hopefully doesn't hurt */

  /* Check for Mac OS X kqueue bug. */
  ch.ident  = -1;
  ch.filter = EVFILT_READ;
  ch.flags  = EV_ADD;

  /* 
   * If kqueue works, then kevent will succeed, and it will
   * stick an error in ev. If kqueue is broken, then
   * kevent will fail.
   */
  if (kevent (kqueue_fd, &ch, 1, &ev, 1, 0) != 1
      || ev.ident != -1
      || ev.flags != EV_ERROR)
    {
      /* detected broken kqueue */
      close (kqueue_fd);
      return 0;
    }

  backend_fudge  = 1e-3; /* needed to compensate for kevent returning early */
  backend_modify = kqueue_modify;
  backend_poll   = kqueue_poll;

  kqueue_eventmax = 64; /* intiial number of events receivable per poll */
  kqueue_events = (struct kevent *)ev_malloc (sizeof (struct kevent) * kqueue_eventmax);

  kqueue_changes   = 0;
  kqueue_changemax = 0;
  kqueue_changecnt = 0;

  return EVBACKEND_KQUEUE;
}

static void
kqueue_destroy (EV_P)
{
  close (kqueue_fd);

  ev_free (kqueue_events);
  ev_free (kqueue_changes);
}

static void
kqueue_fork (EV_P)
{
  close (kqueue_fd);

  while ((kqueue_fd = kqueue ()) < 0)
    syserr ("(libev) kqueue");

  fcntl (kqueue_fd, F_SETFD, FD_CLOEXEC);

  /* re-register interest in fds */
  fd_rearm_all (EV_A);
}

Revision:	1.22
Committed:	Fri Nov 23 05:28:17 2007 UTC (16 years, 5 months ago) by root
Content type:	text/plain
Branch:	MAIN
Changes since 1.21:	+10 -7 lines
Log Message:	minor kqueue optimisation
#	User	Rev	Content
1	root	1.1	/*
2			* Copyright 2007 Marc Alexander Lehmann <libev@schmorp.de>
3			* Copyright 2000-2002 Niels Provos <provos@citi.umich.edu>
4			* All rights reserved.
5			*
6			* Redistribution and use in source and binary forms, with or without
7			* modification, are permitted provided that the following conditions
8			* are met:
9			*
10			* 1. Redistributions of source code must retain the above copyright
11			* notice, this list of conditions and the following disclaimer.
12			* 2. Redistributions in binary form must reproduce the above copyright
13			* notice, this list of conditions and the following disclaimer in the
14			* documentation and/or other materials provided with the distribution.
15			* 3. The name of the author may not be used to endorse or promote products
16			* derived from this software without specific prior written permission.
17			*
18			* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19			* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20			* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21			* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22			* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23			* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24			* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25			* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26			* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27			* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28			*/
29
30			#include <sys/types.h>
31			#include <sys/time.h>
32			#include <sys/queue.h>
33			#include <sys/event.h>
34			#include <string.h>
35			#include <errno.h>
36
37			static void
38	root	1.6	kqueue_change (EV_P_ int fd, int filter, int flags, int fflags)
39	root	1.1	{
40			struct kevent *ke;
41
42	ayin	1.16	++kqueue_changecnt;
43	root	1.17	array_needsize (struct kevent, kqueue_changes, kqueue_changemax, kqueue_changecnt, EMPTY2);
44	root	1.1
45	root	1.6	ke = &kqueue_changes [kqueue_changecnt - 1];
46	root	1.1	memset (ke, 0, sizeof (struct kevent));
47			ke->ident = fd;
48			ke->filter = filter;
49			ke->flags = flags;
50			ke->fflags = fflags;
51			}
52
53	root	1.4	#ifndef NOTE_EOF
54			# define NOTE_EOF 0
55			#endif
56
57	root	1.1	static void
58	root	1.6	kqueue_modify (EV_P_ int fd, int oev, int nev)
59	root	1.1	{
60	root	1.22	if (oev != nev)
61			{
62			if (oev & EV_READ)
63			kqueue_change (EV_A_ fd, EVFILT_READ , EV_DELETE, 0);
64	root	1.14
65	root	1.22	if (oev & EV_WRITE)
66			kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_DELETE, 0);
67			}
68	root	1.14
69	root	1.22	/* to detect close/reopen reliably, we have to re-add */
70			/* event requests even when oev == nev */
71	root	1.14
72			if (nev & EV_READ)
73	root	1.22	kqueue_change (EV_A_ fd, EVFILT_READ , EV_ADD, NOTE_EOF);
74	root	1.1
75	root	1.14	if (nev & EV_WRITE)
76			kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_ADD, NOTE_EOF);
77	root	1.1	}
78
79			static void
80	root	1.6	kqueue_poll (EV_P_ ev_tstamp timeout)
81	root	1.1	{
82			int res, i;
83			struct timespec ts;
84
85	root	1.14	/* need to resize so there is enough space for errors */
86			if (kqueue_changecnt > kqueue_eventmax)
87			{
88			ev_free (kqueue_events);
89			kqueue_eventmax = array_roundsize (struct kevent, kqueue_changecnt);
90	root	1.15	kqueue_events = (struct kevent )ev_malloc (sizeof (struct kevent) kqueue_eventmax);
91	root	1.14	}
92
93	root	1.1	ts.tv_sec = (time_t)timeout;
94	root	1.18	ts.tv_nsec = (long)((timeout - (ev_tstamp)ts.tv_sec) * 1e9);
95	root	1.6	res = kevent (kqueue_fd, kqueue_changes, kqueue_changecnt, kqueue_events, kqueue_eventmax, &ts);
96			kqueue_changecnt = 0;
97	root	1.1
98			if (res < 0)
99	root	1.9	{
100			if (errno != EINTR)
101	root	1.10	syserr ("(libev) kevent");
102	root	1.9
103			return;
104			}
105	root	1.1
106			for (i = 0; i < res; ++i)
107			{
108	root	1.14	int fd = kqueue_events [i].ident;
109
110	root	1.6	if (kqueue_events [i].flags & EV_ERROR)
111	root	1.1	{
112	root	1.14	int err = kqueue_events [i].data;
113
114	root	1.1	/*
115	root	1.14	* errors that may happen
116	root	1.1	* EBADF happens when the file discriptor has been
117			* closed,
118	root	1.14	* ENOENT when the file descriptor was closed and
119	root	1.1	* then reopened.
120			* EINVAL for some reasons not understood; EINVAL
121			* should not be returned ever; but FreeBSD does :-\
122			*/
123	root	1.14
124			/* we are only interested in errors for fds that we are interested in :) */
125			if (anfds [fd].events)
126			{
127			if (err == ENOENT) /* resubmit changes on ENOENT */
128			kqueue_modify (EV_A_ fd, 0, anfds [fd].events);
129			else if (err == EBADF) /* on EBADF, we re-check the fd */
130			{
131			if (fd_valid (fd))
132			kqueue_modify (EV_A_ fd, 0, anfds [fd].events);
133			else
134			fd_kill (EV_A_ fd);
135			}
136			else /* on all other errors, we error out on the fd */
137			fd_kill (EV_A_ fd);
138			}
139	root	1.1	}
140			else
141	root	1.3	fd_event (
142	root	1.6	EV_A_
143	root	1.14	fd,
144	root	1.6	kqueue_events [i].filter == EVFILT_READ ? EV_READ
145			: kqueue_events [i].filter == EVFILT_WRITE ? EV_WRITE
146	root	1.1	: 0
147			);
148			}
149
150	root	1.6	if (expect_false (res == kqueue_eventmax))
151	root	1.1	{
152	root	1.9	ev_free (kqueue_events);
153	root	1.14	kqueue_eventmax = array_roundsize (struct kevent, kqueue_eventmax << 1);
154	root	1.15	kqueue_events = (struct kevent )ev_malloc (sizeof (struct kevent) kqueue_eventmax);
155	root	1.1	}
156			}
157
158	root	1.6	static int
159			kqueue_init (EV_P_ int flags)
160	root	1.1	{
161	root	1.2	struct kevent ch, ev;
162
163	root	1.1	/* Initalize the kernel queue */
164	root	1.6	if ((kqueue_fd = kqueue ()) < 0)
165			return 0;
166	root	1.1
167	root	1.7	fcntl (kqueue_fd, F_SETFD, FD_CLOEXEC); /* not sure if necessary, hopefully doesn't hurt */
168
169	root	1.2	/* Check for Mac OS X kqueue bug. */
170			ch.ident = -1;
171			ch.filter = EVFILT_READ;
172			ch.flags = EV_ADD;
173	root	1.1
174			/*
175			* If kqueue works, then kevent will succeed, and it will
176	root	1.4	* stick an error in ev. If kqueue is broken, then
177	root	1.1	* kevent will fail.
178			*/
179	root	1.6	if (kevent (kqueue_fd, &ch, 1, &ev, 1, 0) != 1
180	root	1.2	\|\| ev.ident != -1
181			\|\| ev.flags != EV_ERROR)
182	root	1.1	{
183			/* detected broken kqueue */
184	root	1.6	close (kqueue_fd);
185			return 0;
186	root	1.1	}
187
188	root	1.21	backend_fudge = 1e-3; /* needed to compensate for kevent returning early */
189			backend_modify = kqueue_modify;
190			backend_poll = kqueue_poll;
191	root	1.1
192	root	1.6	kqueue_eventmax = 64; /* intiial number of events receivable per poll */
193	root	1.15	kqueue_events = (struct kevent )ev_malloc (sizeof (struct kevent) kqueue_eventmax);
194	root	1.6
195	root	1.7	kqueue_changes = 0;
196			kqueue_changemax = 0;
197			kqueue_changecnt = 0;
198
199	root	1.20	return EVBACKEND_KQUEUE;
200	root	1.1	}
201
202	root	1.7	static void
203			kqueue_destroy (EV_P)
204			{
205			close (kqueue_fd);
206
207	root	1.9	ev_free (kqueue_events);
208			ev_free (kqueue_changes);
209	root	1.7	}
210
211			static void
212			kqueue_fork (EV_P)
213			{
214	root	1.11	close (kqueue_fd);
215	root	1.10
216	root	1.11	while ((kqueue_fd = kqueue ()) < 0)
217			syserr ("(libev) kqueue");
218	root	1.10
219	root	1.7	fcntl (kqueue_fd, F_SETFD, FD_CLOEXEC);
220
221			/* re-register interest in fds */
222	root	1.8	fd_rearm_all (EV_A);
223	root	1.7	}
224