2009-04-27 19:53:19

by Oleg Nesterov

[permalink] [raw]
Subject: [PATCH] copy_process: remove the unneeded clear_tsk_thread_flag(TIF_SIGPENDING)

The forked child can have TIF_SIGPENDING if it was copied from parent's
ti->flags. But this is harmless and actually almost never happens, because
copy_process() can't succeed if signal_pending() == T.

Signed-off-by: Oleg Nesterov <[email protected]>

--- PTRACE/kernel/fork.c~FORK_SIGPENDING 2009-04-13 17:05:52.000000000 +0200
+++ PTRACE/kernel/fork.c 2009-04-27 21:38:57.000000000 +0200
@@ -1027,7 +1027,6 @@ static struct task_struct *copy_process(
p->vfork_done = NULL;
spin_lock_init(&p->alloc_lock);

- clear_tsk_thread_flag(p, TIF_SIGPENDING);
init_sigpending(&p->pending);

p->utime = cputime_zero;


2009-04-28 02:40:24

by Roland McGrath

[permalink] [raw]
Subject: Re: [PATCH] copy_process: remove the unneeded clear_tsk_thread_flag(TIF_SIGPENDING)

Acked-by: Roland McGrath <[email protected]>

> The forked child can have TIF_SIGPENDING if it was copied from parent's
> ti->flags. But this is harmless and actually almost never happens, because
> copy_process() can't succeed if signal_pending() == T.

When it does happen, it's actually improper to clear it. In a CLONE_THREAD
case, the pending signals might include shared_pending signals that the
child too should take. (Arguably there is no way to notice, since the
parent thread will be racing to dequeue the same signals.)


Thanks,
Roland

2009-04-28 12:52:27

by Oleg Nesterov

[permalink] [raw]
Subject: Re: [PATCH] copy_process: remove the unneeded clear_tsk_thread_flag(TIF_SIGPENDING)

On 04/27, Roland McGrath wrote:
>
> Acked-by: Roland McGrath <[email protected]>
>
> > The forked child can have TIF_SIGPENDING if it was copied from parent's
> > ti->flags. But this is harmless and actually almost never happens, because
> > copy_process() can't succeed if signal_pending() == T.
>
> When it does happen, it's actually improper to clear it. In a CLONE_THREAD
> case, the pending signals might include shared_pending signals that the
> child too should take. (Arguably there is no way to notice, since the
> parent thread will be racing to dequeue the same signals.)

Yes, sure. Now I see the changelog is not very clear.

I meant, it is possible that the parent has the false TIF_SIGPENDING which
is cleared later by recalc_sigpending(). In this case it is correct to do
clear_tsk_thread_flag(p, TIF_SIGPENDING), but this almost never happens.

Oleg.