net/core: dev_forward_skb() should clear skb_iif

While investigating another bug, I found that the code on the incoming path
in __netif_receive_skb will only set skb->skb_iif if it is already 0.  When
dev_forward_skb() is used in the case of interfaces like veth, skb_iif may
already have been set.  Making dev_forward_skb() cause the packet to look
like a newly received packet would seem to the the correct behaviour here,
as otherwise the wrong incoming interface can be reported for such a packet.

Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/core/dev.c b/net/core/dev.c
index 452db7090d18b39995a302843d05eeed0042414d..723a4065a00ead5cbe5f75e7f2fa1b7444b4a64f 100644 (file)
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1597,6 +1597,7 @@ int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
                kfree_skb(skb);
                return NET_RX_DROP;
        }
+       skb->skb_iif = 0;
        skb_set_dev(skb, dev);
        skb->tstamp.tv64 = 0;
        skb->pkt_type = PACKET_HOST;