Re: [PATCH] fix exit_itimers() vs posix_timer_event() AB-BAdeadlock

[Oleg Nesterov <oleg@tv-sign.ru>]

Andrew Morton wrote:
> 
> Oleg Nesterov <oleg@tv-sign.ru> wrote:
> >
> > +     /*
> >  +     * We are locking ->it_lock + tasklist_lock backwards
> >  +     * from release_task()->exit_itimers(), beware deadlock.
> >  +     */
> >  +    leader = timr->it_process->group_leader;
> >  +    while (unlikely(!read_trylock(&tasklist_lock))) {
> >  +            if (leader->flags & PF_EXITING) {
> >  +                    smp_rmb();
> >  +                    if (thread_group_empty(leader))
> >  +                            return 0;
> >  +            }
> >  +            cpu_relax();
> >  +    }
> 
> Oh dear.  Is there no way to fix this up by taking the locks in the correct
> order?  (Whatever that is).

Yes, this is ugly and that is why I hesitated so long before posting this patch.

But I don't see the simple and correct fix. The lock ordering is not the only
problem. In fact we don't need to take the ->it_lock in the itimer_delete() at
all. The problem is that itimer_delete() can't delete k_itimer.it.real.timer
while holding tasklist_lock, because this lock will prevent (without this patch)
the completition of posix_timer_fn(). It is unsafe to unlock tasklist in __exit_signal()
before calling exit_itimers(), and it is too late to call exit_itimers() after
unlock_irq(&tasklist_lock) in release_task(), at this moment we don't have
->signal/->sighand already.

I beleive the only sane fix would be to eliminate tasklist_lock from signal
sending path (see the patches from Paul and Thomas), but this is hard and
(I think) needs more working/testing.

Oleg.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/