[ipoib] Fix a race when chain-loading undionly.kpxe in IPoIB

The Infiniband link status change callback ipoib_link_state_changed()
may be called while the IPoIB device is closed, in which case there
will not be an IPoIB queue pair to be joined to the IPv4 broadcast
group.  This leads to NULL pointer dereferences in ib_mcast_attach()
and ib_mcast_detach().

Fix by not attempting to join (or leave) the broadcast group unless we
actually have an IPoIB queue pair.

Signed-off-by: Wissam Shoukair <wissams@mellanox.com>
Modified-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>

src/drivers/net/ipoib.c

@@ -791,7 +791,8 @@ static void ipoib_link_state_changed ( struct ib_device *ibdev ) {
 	int rc;
 
 	/* Leave existing broadcast group */
-	ipoib_leave_broadcast_group ( ipoib );
+	if ( ipoib->qp )
+		ipoib_leave_broadcast_group ( ipoib );
 
 	/* Update MAC address based on potentially-new GID prefix */
 	memcpy ( &ipoib->mac.gid.s.prefix, &ibdev->gid.s.prefix,
@@ -806,7 +807,7 @@ static void ipoib_link_state_changed ( struct ib_device *ibdev ) {
 	netdev_link_err ( netdev, ( rc ? rc : -EINPROGRESS_JOINING ) );
 
 	/* Join new broadcast group */
-	if ( ib_is_open ( ibdev ) && ib_link_ok ( ibdev ) &&
+	if ( ib_is_open ( ibdev ) && ib_link_ok ( ibdev ) && ipoib->qp &&
 	     ( ( rc = ipoib_join_broadcast_group ( ipoib ) ) != 0 ) ) {
 		DBGC ( ipoib, "IPoIB %p could not rejoin broadcast group: "
 		       "%s\n", ipoib, strerror ( rc ) );
@@ -894,7 +895,9 @@ static void ipoib_close ( struct net_device *netdev ) {
 
 	/* Tear down the queues */
 	ib_destroy_qp ( ibdev, ipoib->qp );
+	ipoib->qp = NULL;
 	ib_destroy_cq ( ibdev, ipoib->cq );
+	ipoib->cq = NULL;
 
 	/* Close IB device */
 	ib_close ( ibdev );

src/net/infiniband.c

@@ -718,6 +718,9 @@ int ib_mcast_attach ( struct ib_device *ibdev, struct ib_queue_pair *qp,
 	struct ib_multicast_gid *mgid;
 	int rc;
 
+	/* Sanity check */
+	assert ( qp != NULL );
+
 	/* Add to software multicast GID list */
 	mgid = zalloc ( sizeof ( *mgid ) );
 	if ( ! mgid ) {
@@ -751,6 +754,9 @@ void ib_mcast_detach ( struct ib_device *ibdev, struct ib_queue_pair *qp,
 		       union ib_gid *gid ) {
 	struct ib_multicast_gid *mgid;
 
+	/* Sanity check */
+	assert ( qp != NULL );
+
 	/* Remove from hardware multicast GID list */
 	ibdev->op->mcast_detach ( ibdev, qp, gid );
 

src/net/infiniband/ib_mcast.c

@@ -150,6 +150,9 @@ int ib_mcast_join ( struct ib_device *ibdev, struct ib_queue_pair *qp,
 	DBGC ( ibdev, "IBDEV %p QPN %lx joining " IB_GID_FMT "\n",
 	       ibdev, qp->qpn, IB_GID_ARGS ( gid ) );
 
+	/* Sanity check */
+	assert ( qp != NULL );
+
 	/* Initialise structure */
 	membership->qp = qp;
 	memcpy ( &membership->gid, gid, sizeof ( membership->gid ) );
@@ -199,6 +202,9 @@ void ib_mcast_leave ( struct ib_device *ibdev, struct ib_queue_pair *qp,
 	DBGC ( ibdev, "IBDEV %p QPN %lx leaving " IB_GID_FMT "\n",
 	       ibdev, qp->qpn, IB_GID_ARGS ( gid ) );
 
+	/* Sanity check */
+	assert ( qp != NULL );
+
 	/* Detach from multicast GID */
 	ib_mcast_detach ( ibdev, qp, &membership->gid );