rsync - Mar 2004 - Issue with rsync 2.6.1-pre1 on AIX

Hi all,

I've recently tried rsync in daemon mode on AIX. Unfortunately, after
handling connections, the daemon segfaulted. I've located this problem
to the signal handler for SIGCHLD in socket.c - it seems reinstalling
the signal handler before doing waitpid caused the signal to be
redelivered, so the program recurses to the signal handler before it
runs out of stack space in the end and segfaults. Simply moving the
handler reinstaller to after the waitpid seems to work (and should,
AFAIK not break anything on other systems).

I can also report that IPv6 seems to work fine on AIX (by editing
config.h after configuration), the only thing that bothers me is that
I haven't managed to make the same daemon listen on both v4 and v6
(with the same socket), but that's not something I really need.

(The above goes for rsync 2.6.1-pre1 built with Gcc 3.3.2 on AIX 5.2
[oslevel -r says 5200-02]).

-------------- next part --------------
*** socket.c.orig       Mon Mar 29 15:29:10 2004
--- socket.c    Mon Mar 29 15:30:23 2004
***************
*** 431,440 ****
  
  static RETSIGTYPE sigchld_handler(UNUSED(int val))
  {
-       signal(SIGCHLD, sigchld_handler);
  #ifdef WNOHANG
        while (waitpid(-1, NULL, WNOHANG) > 0) {}
  #endif
  }
  
  
--- 431,440 ----
  
  static RETSIGTYPE sigchld_handler(UNUSED(int val))
  {
  #ifdef WNOHANG
        while (waitpid(-1, NULL, WNOHANG) > 0) {}
  #endif
+       signal(SIGCHLD, sigchld_handler);
  }
  
  
-------------- next part --------------

cheers
	/Pontus
-- 

Pontus Sk?ld, see <URL:http://soua.net/> for more information.

On Mon, Mar 29, 2004 at 04:42:09PM +0200, Pontus Skoeld
wrote:
> Simply moving the handler reinstaller to after the waitpid seems to
> work (and should, AFAIK not break anything on other systems).
Thanks -- I've checked-in your patch to CVS.

..wayne..

--On Monday, March 29, 2004 16:42:09 +0200 Pontus Skoeld 
<pont_rsynclistg@soua.net> wrote:
> I've recently tried rsync in daemon mode on AIX. Unfortunately, after
> handling connections, the daemon segfaulted. I've located this problem
> to the signal handler for SIGCHLD in socket.c - it seems reinstalling
> the signal handler before doing waitpid caused the signal to be
> redelivered, so the program recurses to the signal handler before it
> runs out of stack space in the end and segfaults. Simply moving the
> handler reinstaller to after the waitpid seems to work (and should,
> AFAIK not break anything on other systems).
Blech. Why is rsync not just using sigaction()? It's the _only_ way to get 
cross-platform reliable signals. SYSV/BSD signal() is just too broken on 
too many platforms.

If it would be accepted, I'd happily submit a patch fixing up signal 
handling to only use the nice, functional POSIX API.

-- 
Carson