bug in os_unix.c with TCP/IP. deadlock read after accept on 'dead' connection

The problem described in the comments
/**********************************************************************
  * This works around a problem on Linux 2.0.x and SCO Unixware (maybe
  * others?).  When a connect() is made to a Unix Domain socket, but its
  * not accept()ed before the web server gets impatient and close()s, an
  * accept() results in a valid file descriptor, but no data to read.
  * This causes a block on the first read() - which never returns!
  *
....

happening with TCP/IP sockets as well on Linux 2.4.17 and 2.6.10

sanity check performed by
static int is_af_unix_keeper
must be applied to AF_INET sockets as well in OS_Accept

a patch attached




E.Pirogov
Team Leader
OpenSRS Development
+1.416-535-0123x1392

___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/

--- /u/blogware_src/fcgi-2.4.0/libfcgi/os_unix.c	Mon Feb 21 11:26:05 2005
+++ os_unix.c	Mon Aug 15 10:04:40 2005
@@ -1205,7 +1205,7 @@
         if (ReleaseLock(listen_sock))
             return (-1);
 
-        if (sa.in.sin_family != AF_UNIX || is_af_unix_keeper(socket))
+        if (is_af_unix_keeper(socket))
             break;
 
         close(socket);