Re: [PATCH] fix send_sigqueue() vs thread exit race

[Thomas Gleixner]

Oleg,

On Sun, 2005-08-21 at 14:59 +0400, Oleg Nesterov wrote:
> CPU_0					CPU_1
> 
> release_task:				posix_timer_fn:
> 	write_lock(tasklist);			lock(timer->it_lock);
> 
> 	exit_timers:				send_sigxxx();
> 		lock(timer->it_lock)			read_lock(tasklist);
> 				
> Deadlock.
> 
> Dear cc-list, what do you think?

The patch below on top of your patch should solve this. We don't need
tasklist_lock to check p->flags. As you pointed out p cannot be invalid
in send_sigqueue as it's protected by get_task_struct() in
create_timer()

For send_group_sigqueue it's protected by exit_itimers() waiting for
k_itimer.it_lock.

It still does not solve the ugly dependency on tasklist_lock but at
least the race and the deadlock are fixed.

tglx

Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>


--- linux-2.6.13-rc6.signal/kernel/signal.oleg.c	2005-08-21 23:07:03.000000000 +0200
+++ linux-2.6.13-rc6.signal/kernel/signal.c	2005-08-21 23:09:07.000000000 +0200
@@ -1381,11 +1381,14 @@ send_sigqueue(int sig, struct sigqueue *
 	int ret = 0;
 
 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
-	read_lock(&tasklist_lock);
 
-	if (unlikely(p->flags & PF_EXITING)) {
-		ret = -1;
-		goto out_err;
+retry:
+	if (unlikely(p->flags & PF_EXITING))
+		return -1;
+
+	if ((unlikely(!read_trylock(&tasklist_lock))) {
+		cpu_relax();
+		goto retry;
 	}
 
 	spin_lock_irqsave(&p->sighand->siglock, flags);
@@ -1414,7 +1417,6 @@ send_sigqueue(int sig, struct sigqueue *
 
 out:
 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
-out_err:
 	read_unlock(&tasklist_lock);
 
 	return ret;
@@ -1427,7 +1429,14 @@ send_group_sigqueue(int sig, struct sigq
 	int ret = 0;
 
 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
-	read_lock(&tasklist_lock);
+retry:
+	if (unlikely(p->flags & PF_EXITING))
+		return -1;
+
+	if (unlikely(!read_trylock(&tasklist_lock))) {
+		cpu_relax();
+		goto retry;
+	}
 	spin_lock_irqsave(&p->sighand->siglock, flags);
 	handle_stop_signal(sig, p);
 
--- linux-2.6.13-rc6.signal/kernel/posix-timers.oleg.c	2005-08-21 23:09:58.000000000 +0200
+++ linux-2.6.13-rc6.signal/kernel/posix-timers.c	2005-08-21 23:19:42.000000000 +0200
@@ -501,7 +501,8 @@ static void posix_timer_fn(unsigned long
 			remove_from_abslist(timr);
 		}
 
-		if (posix_timer_event(timr, si_private))
+		/* Do not rearm the timer, when we are exiting */
+		if (posix_timer_event(timr, si_private) > 0)
 			/*
 			 * signal was not sent because of sig_ignor
 			 * we will not get a call back to restart it AND


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