Bluetooth: Fix deadlock in l2cap_conn_del()

A deadlock occurs when PDU containing invalid SMP opcode is received on
Security Manager Channel over LE link and conn->pending_rx_work worker
has not run yet.

When LE link is created l2cap_conn_ready() is called and before
returning it schedules conn->pending_rx_work worker to hdev->workqueue.
Incoming data to SMP fixed channel is handled by l2cap_recv_frame()
which calls smp_sig_channel() to handle the SMP PDU. If
smp_sig_channel() indicates failure l2cap_conn_del() is called to delete
the connection. When deleting the connection, l2cap_conn_del() purges
the pending_rx queue and calls flush_work() to wait for the
pending_rx_work worker to complete.

Since incoming data is handled by a worker running from the same
workqueue as the pending_rx_work is being scheduled on, we will deadlock
on waiting for pending_rx_work to complete.

This patch fixes the deadlock by calling cancel_work_sync() instead of
flush_work().

Signed-off-by: Jukka Taimisto <jtt@codenomicon.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Cc: stable@vger.kernel.org

diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
index 6eabbe0..323f23c 100644 (file)
--- a/net/bluetooth/l2cap_core.c
+++ b/net/bluetooth/l2cap_core.c
@@ -1663,7 +1663,13 @@ static void l2cap_conn_del(struct hci_conn *hcon, int err)
        kfree_skb(conn->rx_skb);
 
        skb_queue_purge(&conn->pending_rx);
-       flush_work(&conn->pending_rx_work);
+
+       /* We can not call flush_work(&conn->pending_rx_work) here since we
+        * might block if we are running on a worker from the same workqueue
+        * pending_rx_work is waiting on.
+        */
+       if (work_pending(&conn->pending_rx_work))
+               cancel_work_sync(&conn->pending_rx_work);
 
        l2cap_unregister_all_users(conn);