net/802/mrp: fix possible race condition when calling mrp_pdu_queue()
David Ward [Thu, 11 Apr 2013 13:47:15 +0000 (13:47 +0000)]
(Adapted from a very similar change to net/802/garp.c by Cong Wang.)

mrp_pdu_queue() should ways be called with the applicant spin lock.
mrp_uninit_applicant() only holds the rtnl lock which is not enough;
a race is possible because mrp_rcv() is called in BH context:

mrp_rcv()
  |->mrp_pdu_parse_msg()
    |->mrp_pdu_parse_vecattr()
      |->mrp_pdu_parse_vecattr_event()
        |-> mrp_attr_event()
          |-> mrp_pdu_append_vecattr_event()
            |-> mrp_pdu_queue()

Cc: Cong Wang <amwang@redhat.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David Ward <david.ward@ll.mit.edu>
Acked-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/802/mrp.c

index a4cc322..e085bcc 100644 (file)
@@ -870,8 +870,12 @@ void mrp_uninit_applicant(struct net_device *dev, struct mrp_application *appl)
         * all pending messages before the applicant is gone.
         */
        del_timer_sync(&app->join_timer);
+
+       spin_lock(&app->lock);
        mrp_mad_event(app, MRP_EVENT_TX);
        mrp_pdu_queue(app);
+       spin_unlock(&app->lock);
+
        mrp_queue_xmit(app);
 
        dev_mc_del(dev, appl->group_address);