Don't hold mutex until release cv in cv_wait

If a thread is holding mutex when doing cv_destroy, it might end up waiting a
thread in cv_wait. The waiter would wake up trying to aquire the same mutex
and cause deadlock.

We solve this by move the mutex_enter to the bottom of cv_wait, so that
the waiter will release the cv first, allowing cv_destroy to succeed and have
a chance to free the mutex.

This would create race condition on the cv_mutex. We use xchg to set and check
it to ensure we won't be harmed by the race. This would result in the cv_mutex
debugging becomes best-effort.

Signed-off-by: Chunwei Chen <tuxoko@gmail.com>

module/spl/spl-condvar.c

@@ -80,18 +80,20 @@ static void
 cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io)
 {
 	DEFINE_WAIT(wait);
+	kmutex_t *m;
 
 	ASSERT(cvp);
 	ASSERT(mp);
 	ASSERT(cvp->cv_magic == CV_MAGIC);
 	ASSERT(mutex_owned(mp));
 	atomic_inc(&cvp->cv_refs);
 
-	if (cvp->cv_mutex == NULL)
-		cvp->cv_mutex = mp;
-
+	m = ACCESS_ONCE(cvp->cv_mutex);
+	if (!m)
+		m = xchg(&cvp->cv_mutex, mp);
 	/* Ensure the same mutex is used by all callers */
-	ASSERT(cvp->cv_mutex == mp);
+	if (m)
+		ASSERT(m == mp);
 
 	prepare_to_wait_exclusive(&cvp->cv_event, &wait, state);
 	atomic_inc(&cvp->cv_waiters);
@@ -106,16 +108,23 @@ cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io)
 		io_schedule();
 	else
 		schedule();
-	mutex_enter(mp);
 
 	/* No more waiters a different mutex could be used */
 	if (atomic_dec_and_test(&cvp->cv_waiters)) {
+		/* This is set without any lock, so it's racy. But this is
+		 * just for debug anyway, so make it best-effort */
 		cvp->cv_mutex = NULL;
 		wake_up(&cvp->cv_destroy);
 	}
 
 	finish_wait(&cvp->cv_event, &wait);
 	atomic_dec(&cvp->cv_refs);
+
+	/*
+	 * Hold mutex after we release the cvp, otherwise we could dead lock
+	 * with a thread holding the mutex and call cv_destroy.
+	 */
+	mutex_enter(mp);
 }
 
 void
@@ -148,6 +157,7 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
     int state)
 {
 	DEFINE_WAIT(wait);
+	kmutex_t *m;
 	clock_t time_left;
 
 	ASSERT(cvp);
@@ -156,15 +166,17 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
 	ASSERT(mutex_owned(mp));
 	atomic_inc(&cvp->cv_refs);
 
-	if (cvp->cv_mutex == NULL)
-		cvp->cv_mutex = mp;
-
+	m = ACCESS_ONCE(cvp->cv_mutex);
+	if (!m)
+		m = xchg(&cvp->cv_mutex, mp);
 	/* Ensure the same mutex is used by all callers */
-	ASSERT(cvp->cv_mutex == mp);
+	if (m)
+		ASSERT(m == mp);
 
 	/* XXX - Does not handle jiffie wrap properly */
 	time_left = expire_time - jiffies;
 	if (time_left <= 0) {
+		/* XXX - doesn't reset cv_mutex */
 		atomic_dec(&cvp->cv_refs);
 		return (-1);
 	}
@@ -179,17 +191,23 @@ __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time,
 	 */
 	mutex_exit(mp);
 	time_left = schedule_timeout(time_left);
-	mutex_enter(mp);
 
 	/* No more waiters a different mutex could be used */
 	if (atomic_dec_and_test(&cvp->cv_waiters)) {
+		/* This is set without any lock, so it's racy. But this is
+		 * just for debug anyway, so make it best-effort */
 		cvp->cv_mutex = NULL;
 		wake_up(&cvp->cv_destroy);
 	}
 
 	finish_wait(&cvp->cv_event, &wait);
 	atomic_dec(&cvp->cv_refs);
 
+	/*
+	 * Hold mutex after we release the cvp, otherwise we could dead lock
+	 * with a thread holding the mutex and call cv_destroy.
+	 */
+	mutex_enter(mp);
 	return (time_left > 0 ? time_left : -1);
 }