Re: [PATCH] fix handling of SIGCHILD from reaped child

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Tue, 20 Feb 2007 20:20:49 +0300
Oleg Nesterov <oleg@tv-sign.ru> wrote:
> > > > +		clear_stale_sigchild(current, retval);
> > > 
> > > But we are not checking that SIGCHLD is blocked?
> > > 
> > I'm sorry if I don't read SUSv3 correctly. SUSv3 doesn't define how we should
> > do if SIGCHLD is not blocked.(so I don't check not-blocked case.)
> 
> Probably it is me who misunderstands SUSv3. Could you point me the reference
> to authoritative document? My understanding: if blocked AND wait() succeeds.
> 
I read this:
http://www.opengroup.org/onlinepubs/009695399/


> > IMHO, user's sig-child-handler is tend to call wait()/waitpid() and expects
> > successful return. So removing stale signal here may be good.
> 
> Yes. But sig-child-handler should do
> 
> 	while (wait() >= 0)
> 		....
> 
> anyway, because SIGCHLD is not a realtime signal.
> 

Ah...it looks I should explation why I found this more.

A user (who is migrated from Solaris) met following situation.
==(single threaded program.)

pid1 = fork();
if (!pid1) {
	do_something_forever();
}
....set SIGCHLD handler here to catch pid1's error-exit....

if (!pid2) {
	do_something_in_shortterm();
}
Block SIGCHLD
ret = waitpid(pid2, hoge, hoge); // wait for pid2
UNBlock SIGCHLD
==

And SIGCHLD handler didn't use WNOHANG. 
I asked him to fix his program. He agreed.(So, no problem now.)

While our problem was fixed, it seems Linux doesn't meet spec.(SUSv3)
So I posted. 
But this is rare situation and this fix makes codes ugly....


Thanks,
-Kame