[PATCH] Fix coredump vs exec deadlock

[David Howells]

Fix a race between the coredump code and the exec dethreader that causes the
two to deadlock.

The problem is that if both events happen simultaneously (exec and death by
signal) in two different threads, both threads sit there waiting for the other
to die.

This patch makes exec inferior to death by signal, the dethreader for the
former aborting if it detects the latter.

I've also fixed a potential race in which de_thread() was doing a wait loop,
but with the setting of the task state _after_ the loop condition check.  I've
moved the task state set before the check.

Signed-Off-By: David Howells <dhowells@xxxxxxxxxx>
---
warthog>diffstat -p1 /tmp/dethread.diff 
 fs/exec.c |   16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 3a79d97..b66cc82 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -647,10 +647,13 @@ static int de_thread(struct task_struct 
 			hrtimer_restart(&sig->real_timer);
 		spin_lock_irq(lock);
 	}
-	while (atomic_read(&sig->count) > count) {
+	for (;;) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		if (atomic_read(&sig->count) <= count ||
+		    tsk->mm->core_startup_done)
+			break;
 		sig->group_exit_task = current;
 		sig->notify_count = count;
-		__set_current_state(TASK_UNINTERRUPTIBLE);
 		spin_unlock_irq(lock);
 		schedule();
 		spin_lock_irq(lock);
@@ -658,6 +661,13 @@ static int de_thread(struct task_struct 
 	sig->group_exit_task = NULL;
 	sig->notify_count = 0;
 	spin_unlock_irq(lock);
+	__set_current_state(TASK_RUNNING);
+
+	/* exec loses out to death-by-signal in some other thread */
+	if (tsk->mm->core_startup_done) {
+		kmem_cache_free(sighand_cachep, newsighand);
+		return -EINTR;
+	}
 
 	/*
 	 * At this point all other threads have exited, all we have to
@@ -1394,6 +1404,8 @@ static void zap_threads (struct mm_struc
 	do_each_thread(g,p)
 		if (mm == p->mm && p != tsk) {
 			force_sig_specific(SIGKILL, p);
+			/* make sure de_thread() gets the message */
+			wake_up_state(p, TASK_UNINTERRUPTIBLE);
 			mm->core_waiters++;
 			if (unlikely(p->ptrace) &&
 			    unlikely(p->parent->mm == mm))
-
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/