[ViewVC] Diff of: cvs/libev/ev

Comparing libev/ev_epoll.c (file contents):
Revision 1.36 by root, Mon Oct 27 11:08:29 2008 UTC vs.
Revision 1.54 by sf-exg, Fri Oct 22 10:50:24 2010 UTC

 /*
  * libev epoll fd activity backend
  *
- * Copyright (c) 2007,2008 Marc Alexander Lehmann <libev@schmorp.de>
+ * Copyright (c) 2007,2008,2009,2010 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:
  *
  * 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.
  *
  * 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.
+ * 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>
-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;
   /*
    * 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
+   * 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 */
+  /* store the generation counter in the upper 32 bits, the fd in the lower 32 bits */
-  ev.data.u64 = fd | ((uint64_t)++anfds [fd].egen << 32);
+  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 ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
+  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 */
 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;
-  int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
+  eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
+  EV_ACQUIRE_CB;
   if (expect_false (eventcnt < 0))
     {
       if (errno != EINTR)
-        syserr ("(libev) epoll_wait");
+        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 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);
-      int want = anfds [fd].events;
-      if (anfds [fd].egen != (unsigned char)(ev->data.u64 >> 32))
+      /* check for spurious notification */
-        /*fprintf (stderr, "spurious notification fd %d, %d vs %d\n", fd, (int)(ev->data.u64 >> 32), anfds [fd].egen);*/
+      /* 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;
+          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. */
-          epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev);
+          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);
     }
 }
 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);
+    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_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;
 }
 epoll_fork (EV_P)
 {
   close (backend_fd);
   while ((backend_fd = epoll_create (256)) < 0)
-    syserr ("(libev) epoll_create");
+    ev_syserr ("(libev) epoll_create");
   fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
   fd_rearm_all (EV_A);
 }

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing libev/ev_epoll.c (file contents): Revision 1.36 by root, Mon Oct 27 11:08:29 2008 UTC vs. Revision 1.54 by sf-exg, Fri Oct 22 10:50:24 2010 UTC

Diff Legend

Comparing libev/ev_epoll.c (file contents):
Revision 1.36 by root, Mon Oct 27 11:08:29 2008 UTC vs.
Revision 1.54 by sf-exg, Fri Oct 22 10:50:24 2010 UTC