gro: fix aggregation for skb using frag_list
authorEric Dumazet <edumazet@google.com>
Mon, 29 Sep 2014 17:34:29 +0000 (10:34 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 29 Sep 2014 19:17:59 +0000 (15:17 -0400)
In commit 8a29111c7ca6 ("net: gro: allow to build full sized skb")
I added a regression for linear skb that traditionally force GRO
to use the frag_list fallback.

Erez Shitrit found that at most two segments were aggregated and
the "if (skb_gro_len(p) != pinfo->gso_size)" test was failing.

This is because pinfo at this spot still points to the last skb in the
chain, instead of the first one, where we find the correct gso_size
information.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Fixes: 8a29111c7ca6 ("net: gro: allow to build full sized skb")
Reported-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/skbuff.c

index da1378a3e2c72e0f23e9e21581dfd2ece5491d09..8d289697cc7adcb2ce28c9e793e49c25fbbba5b5 100644 (file)
@@ -3152,6 +3152,9 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
                NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
                goto done;
        }
+       /* switch back to head shinfo */
+       pinfo = skb_shinfo(p);
+
        if (pinfo->frag_list)
                goto merge;
        if (skb_gro_len(p) != pinfo->gso_size)