ceph: cancel delayed work when closing connection

This ensures that if/when we reopen the connection, we can requeue work on
the connection immediately, without waiting for an old timer to expire.
Queue new delayed work inside con->mutex to avoid any race.

This fixes problems with clients failing to reconnect to the MDS due to
the client_reconnect message arriving too late (due to waiting for an old
delayed work timeout to expire).

Signed-off-by: Sage Weil <[email protected]>

fs/ceph/messenger.c

@@ -344,6 +344,7 @@ void ceph_con_close(struct ceph_connection *con)
 	clear_bit(STANDBY, &con->state);  /* avoid connect_seq bump */
 	mutex_lock(&con->mutex);
 	reset_connection(con);
+	cancel_delayed_work(&con->work);
 	mutex_unlock(&con->mutex);
 	queue_con(con);
 }
@@ -1841,6 +1842,8 @@ static void ceph_fault(struct ceph_connection *con)
 	clear_bit(BUSY, &con->state);  /* to avoid an improbable race */
 
 	mutex_lock(&con->mutex);
+	if (test_bit(CLOSED, &con->state))
+		goto out_unlock;
 
 	con_close_socket(con);
 
@@ -1876,15 +1879,15 @@ static void ceph_fault(struct ceph_connection *con)
 	else if (con->delay < MAX_DELAY_INTERVAL)
 		con->delay *= 2;
 
-	mutex_unlock(&con->mutex);
-
 	/* explicitly schedule work to try to reconnect again later. */
 	dout("fault queueing %p delay %lu\n", con, con->delay);
 	con->ops->get(con);
 	if (queue_delayed_work(ceph_msgr_wq, &con->work,
 			       round_jiffies_relative(con->delay)) == 0)
 		con->ops->put(con);
 
+out_unlock:
+	mutex_unlock(&con->mutex);
 out:
 	if (con->ops->fault)
 		con->ops->fault(con);