--- libev/ev_epoll.c	2008/10/29 07:09:37	1.40
+++ libev/ev_epoll.c	2010/04/21 13:13:18	1.51
@@ -1,7 +1,7 @@
 /*
  * 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-
@@ -49,6 +49,11 @@
  * 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
@@ -70,7 +75,7 @@
    * 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)
@@ -79,12 +84,13 @@
   oldmask = anfds [fd].emask;
   anfds [fd].emask = nev;
 
-  /* store the generation counter in the upper 32 bits */
-  ev.data.u64 = fd | ((uint64_t)(uint32_t)++anfds [fd].egen << 32);
+  /* 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 ? 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))
@@ -118,7 +124,13 @@
 epoll_poll (EV_P_ ev_tstamp timeout)
 {
   int i;
-  int eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, (int)ceil (timeout * 1000.));
+  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))
     {
@@ -154,6 +166,8 @@
           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 occured, recreate kernel state */
@@ -176,7 +190,12 @@
 int inline_size
 epoll_init (EV_P_ int flags)
 {
-  backend_fd = epoll_create (256);
+#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;