]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
Bluetooth: L2CAP: cancel pending_rx_work before taking conn->lock
authorRunyu Xiao <runyu.xiao@seu.edu.cn>
Wed, 17 Jun 2026 15:36:13 +0000 (23:36 +0800)
committerLuiz Augusto von Dentz <luiz.von.dentz@intel.com>
Mon, 6 Jul 2026 14:46:57 +0000 (10:46 -0400)
l2cap_conn_del() takes conn->lock and then calls cancel_work_sync() for
pending_rx_work.  process_pending_rx() takes the same mutex, so teardown
can deadlock against the worker it is flushing.

This issue was found by our static analysis tool and then manually
reviewed against the current tree.

The grounded PoC kept the l2cap_conn_ready() -> queue_work(...,
&conn->pending_rx_work) submit path, the l2cap_conn_del() ->
cancel_work_sync(&conn->pending_rx_work) teardown path, and the
process_pending_rx() -> mutex_lock(&conn->lock) worker edge.  Lockdep
reported:

  WARNING: possible circular locking dependency detected
  process_pending_rx+0x21/0x2a [vuln_msv]
  l2cap_conn_del.constprop.0+0x3f/0x4e [vuln_msv]
  *** DEADLOCK ***

Cancel pending_rx_work before taking conn->lock, matching the existing
lock-before-drain ordering used for the two delayed works in the same
teardown path.  The pending_rx queue is still purged after the work has
been cancelled and conn->lock has been acquired.

Fixes: 7ab56c3a6ecc ("Bluetooth: Fix deadlock in l2cap_conn_del()")
Cc: stable@vger.kernel.org
Signed-off-by: Runyu Xiao <runyu.xiao@seu.edu.cn>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
net/bluetooth/l2cap_core.c

index 62133eef9d2fea14e7bf8d203fc951f4ed4499b4..036d887dec34c2ecdd62f30d9aa7fc81f03fdbea 100644 (file)
@@ -1775,19 +1775,13 @@ static void l2cap_conn_del(struct hci_conn *hcon, int err)
        disable_delayed_work_sync(&conn->info_timer);
        disable_delayed_work_sync(&conn->id_addr_timer);
 
+       cancel_work_sync(&conn->pending_rx_work);
+
        mutex_lock(&conn->lock);
 
        kfree_skb(conn->rx_skb);
 
        skb_queue_purge(&conn->pending_rx);
-
-       /* 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);
-
        ida_destroy(&conn->tx_ida);
 
        l2cap_unregister_all_users(conn);