vxlan: Don't call gro_cells_destroy() before device is unregistered

Commit ad6c9986bcb62 ("vxlan: Fix GRO cells race condition between
receive and link delete") fixed a race condition for the typical case a vxlan
device is dismantled from the current netns. But if a netns is dismantled,
vxlan_destroy_tunnels() is called to schedule a unregister_netdevice_queue()
of all the vxlan tunnels that are related to this netns.

In vxlan_destroy_tunnels(), gro_cells_destroy() is called and finished before
unregister_netdevice_queue(). This means that the gro_cells_destroy() call is
done too soon, for the same reasons explained in above commit.

So we need to fully respect the RCU rules, and thus must remove the
gro_cells_destroy() call or risk use after-free.

Fixes: 58ce31cca1ff ("vxlan: GRO support at tunnel layer")
Signed-off-by: Suanming.Mou <mousuanming@huawei.com>
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
index 077f1b9f27616d34e966fc3760b40fd1f6a5915a..d76dfed8d9bbef1d1ae8470686e417af2e531ac8 100644 (file)
--- a/drivers/net/vxlan.c
+++ b/drivers/net/vxlan.c
@@ -4335,10 +4335,8 @@ static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
                /* If vxlan->dev is in the same netns, it has already been added
                 * to the list by the previous loop.
                 */
-               if (!net_eq(dev_net(vxlan->dev), net)) {
-                       gro_cells_destroy(&vxlan->gro_cells);
+               if (!net_eq(dev_net(vxlan->dev), net))
                        unregister_netdevice_queue(vxlan->dev, head);
-               }
        }
 
        for (h = 0; h < PORT_HASH_SIZE; ++h)