Re: [PATCH] Fix signal race during process exit

[Andrew Morton <akpm@osdl.org>]

Roland McGrath <roland@redhat.com> wrote:
>
> Is there a reproducer case around so we can test fixes for this problem?

Jeremy would have the details.

> It seems to me that signals sent to an already dying task might as well
> just be discarded anyway.  All they ever do now (except for trip bugs) is
> change what pending signals you see in the /proc/pid/status entry for a
> zombie.  What's wrong with this:
> 
> Index: linux-2.6/kernel/signal.c
> ===================================================================
> RCS file: /home/roland/redhat/bkcvs/linux-2.5/kernel/signal.c,v
> retrieving revision 1.120
> diff -u -b -p -r1.120 signal.c
> --- linux-2.6/kernel/signal.c 10 May 2004 20:28:20 -0000 1.120
> +++ linux-2.6/kernel/signal.c 4 Jun 2004 01:16:31 -0000
> @@ -161,6 +161,9 @@ static int sig_ignored(struct task_struc
>  {
>  	void * handler;
>  
> +	if (t->flags & PF_DEAD)
> +		return 1;
> +

I'm not sure about the locking here.  What happens if the task starts
exitting immediately after the above test?

Plus the above adds code to the delivery fastpath, rather than exit().

If we're going to do the above for other reasons (what are they?) then it
would be neater to add a new PF_NO_SIGNALS rather than overloading PF_DEAD.