virtio_net: avoid using netif_tx_disable() for serializing tx routine
authorAke Koomsin <ake@igel.co.jp>
Wed, 17 Oct 2018 10:44:12 +0000 (19:44 +0900)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 4 Nov 2018 13:52:49 +0000 (14:52 +0100)
[ Upstream commit 05c998b738fdd3e5d6a257bcacc8f34b6284d795 ]

Commit 713a98d90c5e ("virtio-net: serialize tx routine during reset")
introduces netif_tx_disable() after netif_device_detach() in order to
avoid use-after-free of tx queues. However, there are two issues.

1) Its operation is redundant with netif_device_detach() in case the
   interface is running.
2) In case of the interface is not running before suspending and
   resuming, the tx does not get resumed by netif_device_attach().
   This results in losing network connectivity.

It is better to use netif_tx_lock_bh()/netif_tx_unlock_bh() instead for
serializing tx routine during reset. This also preserves the symmetry
of netif_device_detach() and netif_device_attach().

Fixes commit 713a98d90c5e ("virtio-net: serialize tx routine during reset")
Signed-off-by: Ake Koomsin <ake@igel.co.jp>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/virtio_net.c

index 910c46b4776999e9f43251801c0cc65a70d186b2..f528e9ac3413b8ef71d8a7ea419ca395f4a09c96 100644 (file)
@@ -1872,8 +1872,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev)
        /* Make sure no work handler is accessing the device */
        flush_work(&vi->config_work);
 
+       netif_tx_lock_bh(vi->dev);
        netif_device_detach(vi->dev);
-       netif_tx_disable(vi->dev);
+       netif_tx_unlock_bh(vi->dev);
        cancel_delayed_work_sync(&vi->refill);
 
        if (netif_running(vi->dev)) {
@@ -1909,7 +1910,9 @@ static int virtnet_restore_up(struct virtio_device *vdev)
                }
        }
 
+       netif_tx_lock_bh(vi->dev);
        netif_device_attach(vi->dev);
+       netif_tx_unlock_bh(vi->dev);
        return err;
 }