ipv4: Set skb->protocol properly for local output
authorEli Cooper <elicooper@gmx.com>
Thu, 1 Dec 2016 02:05:10 +0000 (10:05 +0800)
committerDanny Wood <danwood76@gmail.com>
Tue, 29 Jan 2019 13:15:24 +0000 (13:15 +0000)
commit f4180439109aa720774baafdd798b3234ab1a0d2 upstream.

When xfrm is applied to TSO/GSO packets, it follows this path:

    xfrm_output() -> xfrm_output_gso() -> skb_gso_segment()

where skb_gso_segment() relies on skb->protocol to function properly.

This patch sets skb->protocol to ETH_P_IP before dst_output() is called,
fixing a bug where GSO packets sent through a sit tunnel are dropped
when xfrm is involved.

Signed-off-by: Eli Cooper <elicooper@gmx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Willy Tarreau <w@1wt.eu>
net/ipv4/ip_output.c

index 47d236156f0f0b2d806ea188a69de18a07e35845..650873c3240c78abb3b3dc488367cfc7f8739b87 100644 (file)
@@ -97,6 +97,9 @@ int __ip_local_out(struct sk_buff *skb)
 
        iph->tot_len = htons(skb->len);
        ip_send_check(iph);
+
+       skb->protocol = htons(ETH_P_IP);
+
        return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
                       skb_dst(skb)->dev, dst_output);
 }