]> git.karo-electronics.de Git - linux-beck.git/commitdiff
net: allow vlan traffic to be received under bond
authorJohn Fastabend <john.r.fastabend@intel.com>
Mon, 10 Oct 2011 09:16:41 +0000 (09:16 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 19 Oct 2011 03:46:46 +0000 (23:46 -0400)
The following configuration used to work as I expected. At least
we could use the fcoe interfaces to do MPIO and the bond0 iface
to do load balancing or failover.

       ---eth2.228-fcoe
       |
eth2 -----|
          |
          |---- bond0
          |
eth3 -----|
       |
       ---eth3.228-fcoe

This worked because of a change we added to allow inactive slaves
to rx 'exact' matches. This functionality was kept intact with the
rx_handler mechanism. However now the vlan interface attached to the
active slave never receives traffic because the bonding rx_handler
updates the skb->dev and goto's another_round. Previously, the
vlan_do_receive() logic was called before the bonding rx_handler.

Now by the time vlan_do_receive calls vlan_find_dev() the
skb->dev is set to bond0 and it is clear no vlan is attached
to this iface. The vlan lookup fails.

This patch moves the VLAN check above the rx_handler. A VLAN
tagged frame is now routed to the eth2.228-fcoe iface in the
above schematic. Untagged frames continue to the bond0 as
normal. This case also remains intact,

eth2 --> bond0 --> vlan.228

Here the skb is VLAN tagged but the vlan lookup fails on eth2
causing the bonding rx_handler to be called. On the second
pass the vlan lookup is on the bond0 iface and completes as
expected.

Putting a VLAN.228 on both the bond0 and eth2 device will
result in eth2.228 receiving the skb. I don't think this is
completely unexpected and was the result prior to the rx_handler
result.

Note, the same setup is also used for other storage traffic that
MPIO is used with eg. iSCSI and similar setups can be contrived
without storage protocols.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jesse Gross <jesse@nicira.com>
Reviewed-by: Jiri Pirko <jpirko@redhat.com>
Tested-by: Hans Schillstrom <hams.schillstrom@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/dev.c

index 70ecb86439ca25f60b9166efb3669c1a3b3e793f..8b6118a16b87c9536a9b06b9e553a4637d98eede 100644 (file)
@@ -3231,6 +3231,17 @@ another_round:
 ncls:
 #endif
 
+       if (vlan_tx_tag_present(skb)) {
+               if (pt_prev) {
+                       ret = deliver_skb(skb, pt_prev, orig_dev);
+                       pt_prev = NULL;
+               }
+               if (vlan_do_receive(&skb))
+                       goto another_round;
+               else if (unlikely(!skb))
+                       goto out;
+       }
+
        rx_handler = rcu_dereference(skb->dev->rx_handler);
        if (rx_handler) {
                if (pt_prev) {
@@ -3251,17 +3262,6 @@ ncls:
                }
        }
 
-       if (vlan_tx_tag_present(skb)) {
-               if (pt_prev) {
-                       ret = deliver_skb(skb, pt_prev, orig_dev);
-                       pt_prev = NULL;
-               }
-               if (vlan_do_receive(&skb))
-                       goto another_round;
-               else if (unlikely(!skb))
-                       goto out;
-       }
-
        /* deliver only exact match when indicated */
        null_or_dev = deliver_exact ? skb->dev : NULL;