[TCP]: Fix two off-by-one errors in fackets_out adjusting logic
Ilpo Järvinen [Mon, 8 Oct 2007 06:36:41 +0000 (23:36 -0700)]
1) Passing wrong skb to tcp_adjust_fackets_out could corrupt
fastpath_cnt_hint as tcp_skb_pcount(next_skb) is not included
to it if hint points exactly to the next_skb (it's lagging
behind, see sacktag).

2) When fastpath_skb_hint is put backwards to avoid dangling
skb reference, the skb's pcount must also be removed from count
(not included like above).

Reported by Cedric Le Goater <legoater@free.fr>

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/tcp_output.c

index 6199abe..5329675 100644 (file)
@@ -1755,14 +1755,16 @@ static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *skb, int m
                if (tcp_is_reno(tp) && tp->sacked_out)
                        tcp_dec_pcount_approx(&tp->sacked_out, next_skb);
 
-               tcp_adjust_fackets_out(tp, skb, tcp_skb_pcount(next_skb));
+               tcp_adjust_fackets_out(tp, next_skb, tcp_skb_pcount(next_skb));
                tp->packets_out -= tcp_skb_pcount(next_skb);
 
                /* changed transmit queue under us so clear hints */
                tcp_clear_retrans_hints_partial(tp);
                /* manually tune sacktag skb hint */
-               if (tp->fastpath_skb_hint == next_skb)
+               if (tp->fastpath_skb_hint == next_skb) {
                        tp->fastpath_skb_hint = skb;
+                       tp->fastpath_cnt_hint -= tcp_skb_pcount(skb);
+               }
 
                sk_stream_free_skb(sk, next_skb);
        }