*** empty log message ***

[software/libev.git] / ev_epoll.c

diff --git a/ev_epoll.c b/ev_epoll.c
index da2b5bc0b3c4984feb1a91fdb59c8ec5f757a4d9..52f8337604cc699c84fc68c136331bd6c20b2f8f 100644 (file)
--- a/ev_epoll.c
+++ b/ev_epoll.c
@@ -1,75 +1,174 @@
-#include <sys/epoll.h>
+/*
+ * libev epoll fd activity backend
+ *
+ * Copyright (c) 2007 Marc Alexander Lehmann <libev@schmorp.de>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *
+ *     * 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 COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "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 COPYRIGHT
+ * OWNER OR CONTRIBUTORS 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.
+ */
+
+/*
+ * 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.
+ */

 
 

-static int epoll_fd = -1;
+#include <sys/epoll.h>

 
 static void
 
 static void

-epoll_modify (int fd, int oev, int nev)
+epoll_modify (EV_P_ int fd, int oev, int nev)

 {
 {

-  int mode = nev ? oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD : EPOLL_CTL_DEL;
-

   struct epoll_event ev;
   struct epoll_event ev;

-  ev.data.fd = fd;
-  ev.events =
-      (nev & EV_READ ? EPOLLIN : 0)
-      | (nev & EV_WRITE ? EPOLLOUT : 0);

 
 

-  epoll_ctl (epoll_fd, mode, fd, &ev);
-}
+  /*
+   * 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 (!nev)
+    return;

 
 

-void epoll_postfork_child (void)
-{
-  int fd;
+  ev.data.u64 = fd; /* use u64 to fully initialise the struct, for nicer strace etc. */
+  ev.events   = (nev & EV_READ  ? EPOLLIN  : 0)
+              | (nev & EV_WRITE ? EPOLLOUT : 0);

 
 

-  epoll_fd = epoll_create (256);
-  fcntl (epoll_fd, F_SETFD, FD_CLOEXEC);
+  if (expect_true (!epoll_ctl (backend_fd, oev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
+    return;

 
 

-  /* re-register interest in fds */
-  for (fd = 0; fd < anfdmax; ++fd)
-    if (anfds [fd].wev)
-      epoll_modify (fd, EV_NONE, anfds [fd].wev);
-}
+  if (expect_true (errno == ENOENT))
+    {
+      /* on ENOENT the fd went away, so try to do the right thing */
+      if (!nev)
+        return;

 
 

-static struct epoll_event *events;
-static int eventmax;
+      if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
+        return;
+    }
+  else if (expect_true (errno == EEXIST))
+    {
+      /* on EEXIST we ignored a previous DEL */
+      if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
+        return;
+    }

 
 

-static void epoll_poll (ev_tstamp timeout)
+  fd_kill (EV_A_ fd);
+}
+
+static void
+epoll_poll (EV_P_ ev_tstamp timeout)

 {
 {

-  int eventcnt = epoll_wait (epoll_fd, events, eventmax, ceil (timeout * 1000.));

   int i;
   int i;

+  int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));

 
 

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

 
   for (i = 0; i < eventcnt; ++i)
 
   for (i = 0; i < eventcnt; ++i)

-    fd_event (
-      events [i].data.fd,
-      (events [i].events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
-      | (events [i].events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0)
-    );
+    {
+      struct epoll_event *ev = epoll_events + i;
+
+      int fd   = ev->data.u64;
+      int got  = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
+               | (ev->events & (EPOLLIN  | EPOLLERR | EPOLLHUP) ? EV_READ  : 0);
+      int want = anfds [fd].events;
+
+      if (expect_false (got & ~want))
+        {
+          /* we received an event but are not interested in it, try mod or del */
+          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 the receive array was full, increase its size */

-  if (eventcnt == eventmax)
+  if (expect_false (eventcnt == epoll_eventmax))

     {
     {

-      free (events);
-      eventmax += eventmax >> 1;
-      events = malloc (sizeof (struct epoll_event) * 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);

     }
 }
 
     }
 }
 

-void epoll_init (int flags)
+int inline_size
+epoll_init (EV_P_ int flags)

 {
 {

-  epoll_fd = epoll_create (256);
+  backend_fd = epoll_create (256);

 
 

-  if (epoll_fd < 0)
-    return;
+  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;

 
 

-  fcntl (epoll_fd, F_SETFD, FD_CLOEXEC);
+  epoll_eventmax = 64; /* intiial number of events receivable per poll */
+  epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);

 
 

-  ev_method     = EVMETHOD_EPOLL;
-  method_fudge  = 1e-3; /* needed to compensate for epoll returning early */
-  method_modify = epoll_modify;
-  method_poll   = epoll_poll;
+  return EVBACKEND_EPOLL;
+}

 
 

-  eventmax = 64; /* intiial number of events receivable per poll */
-  events = malloc (sizeof (struct epoll_event) * eventmax);
+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);
+}
+