xen/netfront: stop tx queues during live migration
authorDongli Zhang <dongli.zhang@oracle.com>
Fri, 22 Oct 2021 23:31:39 +0000 (16:31 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 25 Oct 2021 15:13:12 +0000 (16:13 +0100)
The tx queues are not stopped during the live migration. As a result, the
ndo_start_xmit() may access netfront_info->queues which is freed by
talk_to_netback()->xennet_destroy_queues().

This patch is to netif_device_detach() at the beginning of xen-netfront
resuming, and netif_device_attach() at the end of resuming.

     CPU A                                CPU B

 talk_to_netback()
 -> if (info->queues)
        xennet_destroy_queues(info);
    to free netfront_info->queues

                                        xennet_start_xmit()
                                        to access netfront_info->queues

  -> err = xennet_create_queues(info, &num_queues);

The idea is borrowed from virtio-net.

Cc: Joe Jin <joe.jin@oracle.com>
Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/xen-netfront.c

index e31b984..fc41ba9 100644 (file)
@@ -1730,6 +1730,10 @@ static int netfront_resume(struct xenbus_device *dev)
 
        dev_dbg(&dev->dev, "%s\n", dev->nodename);
 
+       netif_tx_lock_bh(info->netdev);
+       netif_device_detach(info->netdev);
+       netif_tx_unlock_bh(info->netdev);
+
        xennet_disconnect_backend(info);
        return 0;
 }
@@ -2349,6 +2353,10 @@ static int xennet_connect(struct net_device *dev)
         * domain a kick because we've probably just requeued some
         * packets.
         */
+       netif_tx_lock_bh(np->netdev);
+       netif_device_attach(np->netdev);
+       netif_tx_unlock_bh(np->netdev);
+
        netif_carrier_on(np->netdev);
        for (j = 0; j < num_queues; ++j) {
                queue = &np->queues[j];