- djm@cvs.openbsd.org 2012/04/11 13:16:19

     [channels.c channels.h clientloop.c serverloop.c]
     don't spin in accept() when out of fds (ENFILE/ENFILE) - back off for a
     while; ok deraadt@ markus@

5 files changed, 68 insertions, 16 deletions

diff --git a/ChangeLog b/ChangeLog
index 1b5e78a42..d19da34ae 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -17,6 +17,10 @@
      [channels.c channels.h servconf.c]
      Add PermitOpen none option based on patch from Loganaden Velvindron
      (bz #1949).  ok djm@
+   - djm@cvs.openbsd.org 2012/04/11 13:16:19
+     [channels.c channels.h clientloop.c serverloop.c]
+     don't spin in accept() when out of fds (ENFILE/ENFILE) - back off for a
+     while; ok deraadt@ markus@
 
 20120420
  - (djm) [contrib/caldera/openssh.spec contrib/redhat/openssh.spec]
diff --git a/channels.c b/channels.c
index e5783b197..cacd2fe55 100644
--- a/channels.c
+++ b/channels.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: channels.c,v 1.316 2012/03/29 23:54:36 dtucker Exp $ */
+/* $OpenBSD: channels.c,v 1.317 2012/04/11 13:16:19 djm Exp $ */
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
  * Copyright (c) 1995 Tatu Ylonen <ylo@cs.hut.fi>, Espoo, Finland
@@ -311,6 +311,7 @@ channel_new(char *ctype, int type, int rfd, int wfd, int efd,
         c->istate = CHAN_INPUT_OPEN;
         c->flags = 0;
         channel_register_fds(c, rfd, wfd, efd, extusage, nonblock, 0);
+        c->notbefore = 0;
         c->self = found;
         c->type = type;
         c->ctype = ctype;
@@ -1339,6 +1340,8 @@ channel_post_x11_listener(Channel *c, fd_set *readset, fd_set *writeset)
                 }
                 if (newsock < 0) {
                         error("accept: %.100s", strerror(errno));
+                        if (errno == EMFILE || errno == ENFILE)
+                                c->notbefore = time(NULL) + 1;
                         return;
                 }
                 set_nodelay(newsock);
@@ -1482,6 +1485,8 @@ channel_post_port_listener(Channel *c, fd_set *readset, fd_set *writeset)
                 newsock = accept(c->sock, (struct sockaddr *)&addr, &addrlen);
                 if (newsock < 0) {
                         error("accept: %.100s", strerror(errno));
+                        if (errno == EMFILE || errno == ENFILE)
+                                c->notbefore = time(NULL) + 1;
                         return;
                 }
                 set_nodelay(newsock);
@@ -1514,7 +1519,10 @@ channel_post_auth_listener(Channel *c, fd_set *readset, fd_set *writeset)
                 addrlen = sizeof(addr);
                 newsock = accept(c->sock, (struct sockaddr *)&addr, &addrlen);
                 if (newsock < 0) {
-                        error("accept from auth socket: %.100s", strerror(errno));
+                        error("accept from auth socket: %.100s",
+                            strerror(errno));
+                        if (errno == EMFILE || errno == ENFILE)
+                                c->notbefore = time(NULL) + 1;
                         return;
                 }
                 nc = channel_new("accepted auth socket",
@@ -1917,6 +1925,8 @@ channel_post_mux_listener(Channel *c, fd_set *readset, fd_set *writeset)
         if ((newsock = accept(c->sock, (struct sockaddr*)&addr,
             &addrlen)) == -1) {
                 error("%s accept: %s", __func__, strerror(errno));
+                if (errno == EMFILE || errno == ENFILE)
+                        c->notbefore = time(NULL) + 1;
                 return;
         }
 
@@ -2067,16 +2077,21 @@ channel_garbage_collect(Channel *c)
 }
 
 static void
-channel_handler(chan_fn *ftab[], fd_set *readset, fd_set *writeset)
+channel_handler(chan_fn *ftab[], fd_set *readset, fd_set *writeset,
+    time_t *unpause_secs)
 {
         static int did_init = 0;
         u_int i, oalloc;
         Channel *c;
+        time_t now;
 
         if (!did_init) {
                 channel_handler_init();
                 did_init = 1;
         }
+        now = time(NULL);
+        if (unpause_secs != NULL)
+                *unpause_secs = 0;
         for (i = 0, oalloc = channels_alloc; i < oalloc; i++) {
                 c = channels[i];
                 if (c == NULL)
@@ -2087,10 +2102,30 @@ channel_handler(chan_fn *ftab[], fd_set *readset, fd_set *writeset)
                         else
                                 continue;
                 }
-                if (ftab[c->type] != NULL)
-                        (*ftab[c->type])(c, readset, writeset);
+                if (ftab[c->type] != NULL) {
+                        /*
+                         * Run handlers that are not paused.
+                         */
+                        if (c->notbefore <= now)
+                                (*ftab[c->type])(c, readset, writeset);
+                        else if (unpause_secs != NULL) {
+                                /*
+                                 * Collect the time that the earliest
+                                 * channel comes off pause.
+                                 */
+                                debug3("%s: chan %d: skip for %d more seconds",
+                                    __func__, c->self,
+                                    (int)(c->notbefore - now));
+                                if (*unpause_secs == 0 ||
+                                    (c->notbefore - now) < *unpause_secs)
+                                        *unpause_secs = c->notbefore - now;
+                        }
+                }
                 channel_garbage_collect(c);
         }
+        if (unpause_secs != NULL && *unpause_secs != 0)
+                debug3("%s: first channel unpauses in %d seconds",
+                    __func__, (int)*unpause_secs);
 }
 
 /*
@@ -2099,7 +2134,7 @@ channel_handler(chan_fn *ftab[], fd_set *readset, fd_set *writeset)
  */
 void
 channel_prepare_select(fd_set **readsetp, fd_set **writesetp, int *maxfdp,
-    u_int *nallocp, int rekeying)
+    u_int *nallocp, int *minwait_secs, int rekeying)
 {
         u_int n, sz, nfdset;
 
@@ -2122,7 +2157,8 @@ channel_prepare_select(fd_set **readsetp, fd_set **writesetp, int *maxfdp,
         memset(*writesetp, 0, sz);
 
         if (!rekeying)
-                channel_handler(channel_pre, *readsetp, *writesetp);
+                channel_handler(channel_pre, *readsetp, *writesetp,
+                    minwait_secs);
 }
 
 /*
@@ -2132,7 +2168,7 @@ channel_prepare_select(fd_set **readsetp, fd_set **writesetp, int *maxfdp,
 void
 channel_after_select(fd_set *readset, fd_set *writeset)
 {
-        channel_handler(channel_post, readset, writeset);
+        channel_handler(channel_post, readset, writeset, NULL);
 }
 
 
diff --git a/channels.h b/channels.h
index 6ed1ce00c..d75b800f7 100644
--- a/channels.h
+++ b/channels.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: channels.h,v 1.110 2012/03/29 23:54:36 dtucker Exp $ */
+/* $OpenBSD: channels.h,v 1.111 2012/04/11 13:16:19 djm Exp $ */
 
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
@@ -105,6 +105,7 @@ struct Channel {
         int     wfd_isatty;     /* wfd is a tty */
         int     client_tty;     /* (client) TTY has been requested */
         int     force_drain;    /* force close on iEOF */
+        time_t  notbefore;      /* Pause IO until deadline (time_t) */
         int     delayed;        /* post-select handlers for newly created
                                  * channels are delayed until the first call
                                  * to a matching pre-select handler. 
@@ -238,7 +239,8 @@ void	 channel_input_status_confirm(int, u_int32_t, void *);
 
 /* file descriptor handling (read/write) */
 
-void     channel_prepare_select(fd_set **, fd_set **, int *, u_int*, int);
+void     channel_prepare_select(fd_set **, fd_set **, int *, u_int*,
+             time_t*, int);
 void     channel_after_select(fd_set *, fd_set *);
 void     channel_output_poll(void);
 
diff --git a/clientloop.c b/clientloop.c
index f69a9b025..58357cf39 100644
--- a/clientloop.c
+++ b/clientloop.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: clientloop.c,v 1.238 2012/01/18 21:46:43 dtucker Exp $ */
+/* $OpenBSD: clientloop.c,v 1.239 2012/04/11 13:16:19 djm Exp $ */
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
  * Copyright (c) 1995 Tatu Ylonen <ylo@cs.hut.fi>, Espoo, Finland
@@ -583,10 +583,12 @@ client_wait_until_can_do_something(fd_set **readsetp, fd_set **writesetp,
 {
         struct timeval tv, *tvp;
         int timeout_secs;
+        time_t minwait_secs;
         int ret;
 
         /* Add any selections by the channel mechanism. */
-        channel_prepare_select(readsetp, writesetp, maxfdp, nallocp, rekeying);
+        channel_prepare_select(readsetp, writesetp, maxfdp, nallocp,
+            &minwait_secs, rekeying);
 
         if (!compat20) {
                 /* Read from the connection, unless our buffers are full. */
@@ -639,6 +641,8 @@ client_wait_until_can_do_something(fd_set **readsetp, fd_set **writesetp,
                 if (timeout_secs < 0)
                         timeout_secs = 0;
         }
+        if (minwait_secs != 0)
+                timeout_secs = MIN(timeout_secs, (int)minwait_secs);
         if (timeout_secs == INT_MAX)
                 tvp = NULL;
         else {
diff --git a/serverloop.c b/serverloop.c
index 19b84ff27..50be16b7c 100644
--- a/serverloop.c
+++ b/serverloop.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: serverloop.c,v 1.160 2011/05/15 08:09:01 djm Exp $ */
+/* $OpenBSD: serverloop.c,v 1.161 2012/04/11 13:16:19 djm Exp $ */
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
  * Copyright (c) 1995 Tatu Ylonen <ylo@cs.hut.fi>, Espoo, Finland
@@ -281,9 +281,18 @@ wait_until_can_do_something(fd_set **readsetp, fd_set **writesetp, int *maxfdp,
 {
         struct timeval tv, *tvp;
         int ret;
+        time_t minwait_secs;
         int client_alive_scheduled = 0;
         int program_alive_scheduled = 0;
 
+        /* Allocate and update select() masks for channel descriptors. */
+        channel_prepare_select(readsetp, writesetp, maxfdp, nallocp,
+            &minwait_secs, 0);
+
+        if (minwait_secs != 0)
+                max_time_milliseconds = MIN(max_time_milliseconds,
+                    (u_int)minwait_secs * 1000);
+
         /*
          * if using client_alive, set the max timeout accordingly,
          * and indicate that this particular timeout was for client
@@ -298,9 +307,6 @@ wait_until_can_do_something(fd_set **readsetp, fd_set **writesetp, int *maxfdp,
                 max_time_milliseconds = options.client_alive_interval * 1000;
         }
 
-        /* Allocate and update select() masks for channel descriptors. */
-        channel_prepare_select(readsetp, writesetp, maxfdp, nallocp, 0);
-
         if (compat20) {
 #if 0
                 /* wrong: bad condition XXX */