ipv6: gro: fix CHECKSUM_COMPLETE support

When GRE support was added in linux-3.14, CHECKSUM_COMPLETE handling
broke on GRE+IPv6 because we did not update/use the appropriate csum :

GRO layer is supposed to use/update NAPI_GRO_CB(skb)->csum instead of
skb->csum

Tested using a GRE tunnel and IPv6 traffic. GRO aggregation now happens
at the first level (ethernet device) instead of being done in gre
tunnel. Native IPv6+TCP is still properly aggregated.

Fixes: bf5a755f5e918 ("net-gre-gro: Add GRE support to the GRO stack")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jerry Chu <hkchu@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 2 insertions, 6 deletions

diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
index 59f95affceb0..b2f091566f88 100644
--- a/net/ipv6/ip6_offload.c
+++ b/net/ipv6/ip6_offload.c
@@ -196,7 +196,6 @@ static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
         unsigned int off;
         u16 flush = 1;
         int proto;
-        __wsum csum;
 
         off = skb_gro_offset(skb);
         hlen = off + sizeof(*iph);
@@ -264,13 +263,10 @@ static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 
         NAPI_GRO_CB(skb)->flush |= flush;
 
-        csum = skb->csum;
-        skb_postpull_rcsum(skb, iph, skb_network_header_len(skb));
+        skb_gro_postpull_rcsum(skb, iph, nlen);
 
         pp = ops->callbacks.gro_receive(head, skb);
 
-        skb->csum = csum;
-
 out_unlock:
         rcu_read_unlock();
 
diff --git a/net/ipv6/tcpv6_offload.c b/net/ipv6/tcpv6_offload.c
index 0d78132ff18a..8517d3cd1aed 100644
--- a/net/ipv6/tcpv6_offload.c
+++ b/net/ipv6/tcpv6_offload.c
@@ -42,7 +42,7 @@ static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
         if (NAPI_GRO_CB(skb)->flush)
                 goto skip_csum;
 
-        wsum = skb->csum;
+        wsum = NAPI_GRO_CB(skb)->csum;
 
         switch (skb->ip_summed) {
         case CHECKSUM_NONE: