passenger 3.0.21 → 3.9.1.beta

data/ext/libev/ev_epoll.c

@@ -1,19 +1,19 @@
 /*
  * libev epoll fd activity backend
  *
- * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libev@schmorp.de>
+ * 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
@@ -49,6 +49,12 @@
  * 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
@@ -59,6 +65,8 @@
 
 #include <sys/epoll.h>
 
+#define EV_EMASK_EPERM 0x80
+
 static void
 epoll_modify (EV_P_ int fd, int oev, int nev)
 {
@@ -85,7 +93,7 @@ epoll_modify (EV_P_ int fd, int oev, int nev)
   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))
@@ -107,6 +115,21 @@ epoll_modify (EV_P_ int fd, int oev, int nev)
       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);
 
@@ -120,11 +143,14 @@ epoll_poll (EV_P_ ev_tstamp timeout)
 {
   int i;
   int eventcnt;
-  
+
+  if (expect_false (epoll_epermcnt))
+    timeout = 0.;
+
   /* 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.));
+  eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3);
   EV_ACQUIRE_CB;
 
   if (expect_false (eventcnt < 0))
@@ -144,7 +170,12 @@ epoll_poll (EV_P_ ev_tstamp timeout)
       int got  = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
                | (ev->events & (EPOLLIN  | EPOLLERR | EPOLLHUP) ? EV_READ  : 0);
 
-      /* check for spurious notification */
+      /*
+       * check for spurious notification.
+       * this only finds spurious notifications on egen updates
+       * other spurious notifications will be found by epoll_ctl, below
+       * 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 */
@@ -156,8 +187,15 @@ epoll_poll (EV_P_ ev_tstamp timeout)
         {
           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 */
+          /*
+           * we received an event but are not interested in it, try mod or del
+           * this often happens because we optimistically do not unregister fds
+           * when we are no longer interested in them, but also when we get spurious
+           * notifications for fds from another process. this is partially handled
+           * above with the gencounter check (== our fd is not the event fd), and
+           * partially here, when epoll_ctl returns an error (== a child has the fd
+           * but we closed it).
+           */
           ev->events = (want & EV_READ  ? EPOLLIN  : 0)
                      | (want & EV_WRITE ? EPOLLOUT : 0);
 
@@ -165,7 +203,7 @@ epoll_poll (EV_P_ ev_tstamp timeout)
           /* 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 */
+              postfork = 1; /* an error occurred, recreate kernel state */
               continue;
             }
         }
@@ -180,6 +218,18 @@ epoll_poll (EV_P_ ev_tstamp timeout)
       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
@@ -188,7 +238,7 @@ epoll_init (EV_P_ int flags)
 #ifdef EPOLL_CLOEXEC
   backend_fd = epoll_create1 (EPOLL_CLOEXEC);
 
-  if (backend_fd <= 0)
+  if (backend_fd < 0)
 #endif
     backend_fd = epoll_create (256);
 
@@ -197,9 +247,9 @@ epoll_init (EV_P_ int flags)
 
   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;
+  backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */
+  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);
@@ -211,6 +261,7 @@ void inline_size
 epoll_destroy (EV_P)
 {
   ev_free (epoll_events);
+  array_free (epoll_eperm, EMPTY);
 }
 
 void inline_size

data/ext/libev/ev_kqueue.c

@@ -1,19 +1,19 @@
 /*
  * libev kqueue backend
  *
- * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libev@schmorp.de>
+ * 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
@@ -39,7 +39,6 @@
 
 #include <sys/types.h>
 #include <sys/time.h>
-#include <sys/queue.h>
 #include <sys/event.h>
 #include <string.h>
 #include <errno.h>
@@ -53,6 +52,10 @@ kqueue_change (EV_P_ int fd, int filter, int flags, int fflags)
   EV_SET (&kqueue_changes [kqueue_changecnt - 1], fd, filter, flags, fflags, 0, 0);
 }
 
+/* OS X at least needs this */
+#ifndef EV_ENABLE
+# define EV_ENABLE 0
+#endif
 #ifndef NOTE_EOF
 # define NOTE_EOF 0
 #endif
@@ -73,10 +76,10 @@ kqueue_modify (EV_P_ int fd, int oev, int nev)
   /* event requests even when oev == nev */
 
   if (nev & EV_READ)
-    kqueue_change (EV_A_ fd, EVFILT_READ , EV_ADD, NOTE_EOF);
+    kqueue_change (EV_A_ fd, EVFILT_READ , EV_ADD | EV_ENABLE, NOTE_EOF);
 
   if (nev & EV_WRITE)
-    kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_ADD, NOTE_EOF);
+    kqueue_change (EV_A_ fd, EVFILT_WRITE, EV_ADD | EV_ENABLE, NOTE_EOF);
 }
 
 static void
@@ -94,19 +97,18 @@ kqueue_poll (EV_P_ ev_tstamp timeout)
     }
 
   EV_RELEASE_CB;
-  ts.tv_sec  = (time_t)timeout;
-  ts.tv_nsec = (long)((timeout - (ev_tstamp)ts.tv_sec) * 1e9);
+  EV_TS_SET (ts, timeout);
   res = kevent (backend_fd, kqueue_changes, kqueue_changecnt, kqueue_events, kqueue_eventmax, &ts);
   EV_ACQUIRE_CB;
   kqueue_changecnt = 0;
 
   if (expect_false (res < 0))
-    { 
+    {
       if (errno != EINTR)
         ev_syserr ("(libev) kevent");
 
       return;
-    } 
+    }
 
   for (i = 0; i < res; ++i)
     {
@@ -114,11 +116,11 @@ kqueue_poll (EV_P_ ev_tstamp timeout)
 
       if (expect_false (kqueue_events [i].flags & EV_ERROR))
         {
-	  int err = kqueue_events [i].data;
+          int err = kqueue_events [i].data;
 
           /* 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 */
@@ -130,7 +132,7 @@ kqueue_poll (EV_P_ ev_tstamp timeout)
                 }
               else /* on all other errors, we error out on the fd */
                 fd_kill (EV_A_ fd);
-	    }
+            }
         }
       else
         fd_event (
@@ -153,15 +155,15 @@ kqueue_poll (EV_P_ ev_tstamp timeout)
 int inline_size
 kqueue_init (EV_P_ int flags)
 {
-  /* Initalize the kernel queue */
+  /* Initialize the kernel queue */
   if ((backend_fd = kqueue ()) < 0)
     return 0;
 
   fcntl (backend_fd, F_SETFD, FD_CLOEXEC); /* not sure if necessary, hopefully doesn't hurt */
 
-  backend_fudge  = 0.;
-  backend_modify = kqueue_modify;
-  backend_poll   = kqueue_poll;
+  backend_mintime = 1e-9; /* apparently, they did the right thing in freebsd */
+  backend_modify  = kqueue_modify;
+  backend_poll    = kqueue_poll;
 
   kqueue_eventmax = 64; /* initial number of events receivable per poll */
   kqueue_events = (struct kevent *)ev_malloc (sizeof (struct kevent) * kqueue_eventmax);

data/ext/libev/ev_poll.c

@@ -1,19 +1,19 @@
 /*
  * libev poll fd activity backend
  *
- * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libev@schmorp.de>
+ * 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
@@ -42,7 +42,7 @@
 void inline_size
 pollidx_init (int *base, int count)
 {
-  /* consider using memset (.., -1, ...), which is pratically guarenteed
+  /* consider using memset (.., -1, ...), which is practically guaranteed
    * to work on all systems implementing poll */
   while (count--)
     *base++ = -1;
@@ -92,7 +92,7 @@ poll_poll (EV_P_ ev_tstamp timeout)
   int res;
   
   EV_RELEASE_CB;
-  res = poll (polls, pollcnt, (int)ceil (timeout * 1000.));
+  res = poll (polls, pollcnt, timeout * 1e3);
   EV_ACQUIRE_CB;
 
   if (expect_false (res < 0))
@@ -106,28 +106,32 @@ poll_poll (EV_P_ ev_tstamp timeout)
     }
   else
     for (p = polls; res; ++p)
-      if (expect_false (p->revents)) /* this expect is debatable */
-        {
-          --res;
-
-          if (expect_false (p->revents & POLLNVAL))
-            fd_kill (EV_A_ p->fd);
-          else
-            fd_event (
-              EV_A_
-              p->fd,
-              (p->revents & (POLLOUT | POLLERR | POLLHUP) ? EV_WRITE : 0)
-              | (p->revents & (POLLIN | POLLERR | POLLHUP) ? EV_READ : 0)
-            );
-        }
+      {
+        assert (("libev: poll() returned illegal result, broken BSD kernel?", p < polls + pollcnt));
+
+        if (expect_false (p->revents)) /* this expect is debatable */
+          {
+            --res;
+
+            if (expect_false (p->revents & POLLNVAL))
+              fd_kill (EV_A_ p->fd);
+            else
+              fd_event (
+                EV_A_
+                p->fd,
+                (p->revents & (POLLOUT | POLLERR | POLLHUP) ? EV_WRITE : 0)
+                | (p->revents & (POLLIN | POLLERR | POLLHUP) ? EV_READ : 0)
+              );
+          }
+      }
 }
 
 int inline_size
 poll_init (EV_P_ int flags)
 {
-  backend_fudge  = 0.; /* posix says this is zero */
-  backend_modify = poll_modify;
-  backend_poll   = poll_poll;
+  backend_mintime = 1e-3;
+  backend_modify  = poll_modify;
+  backend_poll    = poll_poll;
 
   pollidxs = 0; pollidxmax = 0;
   polls    = 0; pollmax    = 0; pollcnt = 0;

data/ext/libev/ev_port.c

@@ -1,19 +1,19 @@
 /*
  * libev solaris event port backend
  *
- * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libev@schmorp.de>
+ * 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
@@ -37,6 +37,17 @@
  * either the BSD or the GPL.
  */
 
+/* useful reading:
+ *
+ * http://bugs.opensolaris.org/view_bug.do?bug_id=6268715 (random results)
+ * http://bugs.opensolaris.org/view_bug.do?bug_id=6455223 (just totally broken)
+ * http://bugs.opensolaris.org/view_bug.do?bug_id=6873782 (manpage ETIME)
+ * http://bugs.opensolaris.org/view_bug.do?bug_id=6874410 (implementation ETIME)
+ * http://www.mail-archive.com/networking-discuss@opensolaris.org/msg11898.html ETIME vs. nget
+ * http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/lib/libc/port/gen/event_port.c (libc)
+ * http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/fs/portfs/port.c#1325 (kernel)
+ */
+
 #include <sys/types.h>
 #include <sys/time.h>
 #include <poll.h>
@@ -85,19 +96,20 @@ port_poll (EV_P_ ev_tstamp timeout)
   struct timespec ts;
   uint_t nget = 1;
 
+  /* we initialise this to something we will skip in the loop, as */
+  /* port_getn can return with nget unchanged, but no indication */
+  /* whether it was the original value or has been updated :/ */
+  port_events [0].portev_source = 0;
+
   EV_RELEASE_CB;
-  ts.tv_sec  = (time_t)timeout;
-  ts.tv_nsec = (long)(timeout - (ev_tstamp)ts.tv_sec) * 1e9;
+  EV_TS_SET (ts, timeout);
   res = port_getn (backend_fd, port_events, port_eventmax, &nget, &ts);
   EV_ACQUIRE_CB;
 
-  if (res == -1)
-    { 
-      if (errno != EINTR && errno != ETIME)
-        ev_syserr ("(libev) port_getn (see http://bugs.opensolaris.org/view_bug.do?bug_id=6268715, try LIBEV_FLAGS=3 env variable)");
-
-      return;
-    } 
+  /* port_getn may or may not set nget on error */
+  /* so we rely on port_events [0].portev_source not being updated */
+  if (res == -1 && errno != ETIME && errno != EINTR)
+    ev_syserr ("(libev) port_getn (see http://bugs.opensolaris.org/view_bug.do?bug_id=6268715, try LIBEV_FLAGS=3 env variable)");
 
   for (i = 0; i < nget; ++i)
     {
@@ -112,7 +124,7 @@ port_poll (EV_P_ ev_tstamp timeout)
             | (port_events [i].portev_events & (POLLIN | POLLERR | POLLHUP) ? EV_READ : 0)
           );
 
-          port_associate_and_check (EV_A_ fd, anfds [fd].events);
+          fd_change (EV_A_ fd, EV__IOFDSET);
         }
     }
 
@@ -127,17 +139,25 @@ port_poll (EV_P_ ev_tstamp timeout)
 int inline_size
 port_init (EV_P_ int flags)
 {
-  /* Initalize the kernel queue */
+  /* Initialize the kernel queue */
   if ((backend_fd = port_create ()) < 0)
     return 0;
 
+  assert (("libev: PORT_SOURCE_FD must not be zero", PORT_SOURCE_FD));
+
   fcntl (backend_fd, F_SETFD, FD_CLOEXEC); /* not sure if necessary, hopefully doesn't hurt */
 
-  backend_fudge  = 1e-3; /* needed to compensate for port_getn returning early */
-  backend_modify = port_modify;
-  backend_poll   = port_poll;
+  /* if my reading of the opensolaris kernel sources are correct, then
+   * opensolaris does something very stupid: it checks if the time has already
+   * elapsed and doesn't round up if that is the case,m otherwise it DOES round
+   * up. Since we can't know what the case is, we need to guess by using a
+   * "large enough" timeout. Normally, 1e-9 would be correct.
+   */
+  backend_mintime = 1e-3; /* needed to compensate for port_getn returning early */
+  backend_modify  = port_modify;
+  backend_poll    = port_poll;
 
-  port_eventmax = 64; /* intiial number of events receivable per poll */
+  port_eventmax = 64; /* initial number of events receivable per poll */
   port_events = (port_event_t *)ev_malloc (sizeof (port_event_t) * port_eventmax);
 
   return EVBACKEND_PORT;