net: keep original skb which only needs header checking during software GSO
authorJason Wang <jasowang@redhat.com>
Fri, 19 Sep 2014 08:04:38 +0000 (16:04 +0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 22 Sep 2014 18:57:08 +0000 (14:57 -0400)
Commit ce93718fb7cdbc064c3000ff59e4d3200bdfa744 ("net: Don't keep
around original SKB when we software segment GSO frames") frees the
original skb after software GSO even for dodgy gso skbs. This breaks
the stream throughput from untrusted sources, since only header
checking was done during software GSO instead of a true
segmentation. This patch fixes this by freeing the original gso skb
only when it was really segmented by software.

Fixes ce93718fb7cdbc064c3000ff59e4d3200bdfa744 ("net: Don't keep
around original SKB when we software segment GSO frames.")

Cc: David S. Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/dev.c

index e916ba8caccfc90c8c2128bbcac415ef74d37da3..52cd71a4a3439c131db57f8c301376af399995f1 100644 (file)
@@ -2694,10 +2694,12 @@ struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
                struct sk_buff *segs;
 
                segs = skb_gso_segment(skb, features);
-               kfree_skb(skb);
-               if (IS_ERR(segs))
+               if (IS_ERR(segs)) {
                        segs = NULL;
-               skb = segs;
+               } else if (segs) {
+                       consume_skb(skb);
+                       skb = segs;
+               }
        } else {
                if (skb_needs_linearize(skb, features) &&
                    __skb_linearize(skb))