net/802/mrp: fix possible race condition when calling mrp_pdu_queue()

(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: NDavid Ward <david.ward@ll.mit.edu>
Acked-by: NCong Wang <amwang@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

net/802/mrp.c

@@ -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);