RDMA/cxgb4: Fix endpoint timeout race condition

The endpoint timeout logic had a race that could cause an endpoint
object to be freed while it was still on the timedout list.  This
can happen if the timer is stopped after it had fired, but before
the timedout thread processed the endpoint timeout.
Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>

drivers/infiniband/hw/cxgb4/cm.c

@@ -159,9 +159,11 @@ static void start_ep_timer(struct c4iw_ep *ep)
 {
 	PDBG("%s ep %p\n", __func__, ep);
 	if (timer_pending(&ep->timer)) {
-		PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
-		del_timer_sync(&ep->timer);
-	} else
+		pr_err("%s timer already started! ep %p\n",
+		       __func__, ep);
+		return;
+	}
+	clear_bit(TIMEOUT, &ep->com.flags);
 	c4iw_get_ep(&ep->com);
 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
 	ep->timer.data = (unsigned long)ep;
@@ -171,13 +173,9 @@ static void start_ep_timer(struct c4iw_ep *ep)
 
 static void stop_ep_timer(struct c4iw_ep *ep)
 {
-	PDBG("%s ep %p\n", __func__, ep);
-	if (!timer_pending(&ep->timer)) {
-		WARN(1, "%s timer stopped when its not running! "
-		       "ep %p state %u\n", __func__, ep, ep->com.state);
-		return;
-	}
+	PDBG("%s ep %p stopping\n", __func__, ep);
 	del_timer_sync(&ep->timer);
+	if (!test_and_set_bit(TIMEOUT, &ep->com.flags))
 		c4iw_put_ep(&ep->com);
 }
 
@@ -3191,10 +3189,15 @@ static DECLARE_WORK(skb_work, process_work);
 static void ep_timeout(unsigned long arg)
 {
 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
+	int kickit = 0;
 
 	spin_lock(&timeout_lock);
+	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
 		list_add_tail(&ep->entry, &timeout_list);
+		kickit = 1;
+	}
 	spin_unlock(&timeout_lock);
+	if (kickit)
 		queue_work(workq, &skb_work);
 }
 

drivers/infiniband/hw/cxgb4/iw_cxgb4.h

@@ -716,6 +716,7 @@ enum c4iw_ep_flags {
 	ABORT_REQ_IN_PROGRESS	= 1,
 	RELEASE_RESOURCES	= 2,
 	CLOSE_SENT		= 3,
+	TIMEOUT                 = 4,
 	QP_REFERENCED           = 5,
 };