ph10 2005/05/11 10:26:55 BST
Modified files:
exim-src/src acl.c
Log:
Added a long comment to acl.c about the problem of detecting a client
going away during a "delay" wait.
Revision Changes Path
1.30 +16 -0 exim/exim-src/src/acl.c
Index: acl.c
===================================================================
RCS file: /home/cvs/exim/exim-src/src/acl.c,v
retrieving revision 1.29
retrieving revision 1.30
diff -u -r1.29 -r1.30
--- acl.c 10 May 2005 10:19:11 -0000 1.29
+++ acl.c 11 May 2005 09:26:55 -0000 1.30
@@ -1,4 +1,4 @@
-/* $Cambridge: exim/exim-src/src/acl.c,v 1.29 2005/05/10 10:19:11 ph10 Exp $ */
+/* $Cambridge: exim/exim-src/src/acl.c,v 1.30 2005/05/11 09:26:55 ph10 Exp $ */
/*************************************************
* Exim - an Internet mail transport agent *
@@ -2178,6 +2178,22 @@
HDEBUG(D_acl)
debug_printf("delay skipped in -bh checking mode\n");
}
+
+ /* It appears to be impossible to detect that a TCP/IP connection has
+ gone away without reading from it. This means that we cannot shorten
+ the delay below if the client goes away, because we cannot discover
+ that the client has closed its end of the connection. (The connection
+ is actually in a half-closed state, waiting for the server to close its
+ end.) It would be nice to be able to detect this state, so that the
+ Exim process is not held up unnecessarily. However, it seems that we
+ can't. The poll() function does not do the right thing, and in any case
+ it is not always available.
+
+ NOTE: If ever this state of affairs changes, remember that we may be
+ dealing with stdin/stdout here, in addition to TCP/IP connections.
+ Whatever is done must work in both cases. To detected the stdin/stdout
+ case, check for smtp_in or smtp_out being NULL. */
+
else
{
while (delay > 0) delay = sleep(delay);
