net: xdp: pull ethernet header off packet after computing skb->protocol
authorJason A. Donenfeld <Jason@zx2c4.com>
Sat, 15 Aug 2020 07:29:30 +0000 (09:29 +0200)
committerDavid S. Miller <davem@davemloft.net>
Sun, 16 Aug 2020 22:28:18 +0000 (15:28 -0700)
When an XDP program changes the ethernet header protocol field,
eth_type_trans is used to recalculate skb->protocol. In order for
eth_type_trans to work correctly, the ethernet header must actually be
part of the skb data segment, so the code first pushes that onto the
head of the skb. However, it subsequently forgets to pull it back off,
making the behavior of the passed-on packet inconsistent between the
protocol modifying case and the static protocol case. This patch fixes
the issue by simply pulling the ethernet header back off of the skb
head.

Fixes: 297249569932 ("net: fix generic XDP to handle if eth header was mangled")
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/dev.c

index b5d1129..7592d7d 100644 (file)
@@ -4676,6 +4676,7 @@ static u32 netif_receive_generic_xdp(struct sk_buff *skb,
            (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
                __skb_push(skb, ETH_HLEN);
                skb->protocol = eth_type_trans(skb, skb->dev);
+               __skb_pull(skb, ETH_HLEN);
        }
 
        switch (act) {