[TCP]: Reordered ACK's (old) SACKs not included to discarded MIB
authorIlpo Järvinen <ilpo.jarvinen@helsinki.fi>
Wed, 26 Sep 2007 05:46:50 +0000 (22:46 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Wed, 10 Oct 2007 23:52:38 +0000 (16:52 -0700)
In case of ACK reordering, the SACK block might be valid in it's
time but is already obsoleted since we've received another kind
of confirmation about arrival of the segments through snd_una
advancement of an earlier packet.

I didn't bother to build distinguishing of valid and invalid
SACK blocks but simply made reordered SACK blocks that are too
old always not counted regardless of their "real" validity which
could be determined by using the ack field of the reordered
packet (won't be significant IMHO).

DSACKs can very well be considered useful even in this situation,
so won't do any of this for them.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_input.c

index 4c10d9cad20f594e2bd6a3a15b0c6d45554ece38..4866e75e98e01adcb7d920de89c17cdc65f7aa6b 100644 (file)
@@ -1247,8 +1247,13 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_
                                        NET_INC_STATS_BH(LINUX_MIB_TCPDSACKIGNOREDNOUNDO);
                                else
                                        NET_INC_STATS_BH(LINUX_MIB_TCPDSACKIGNOREDOLD);
-                       } else
+                       } else {
+                               /* Don't count olds caused by ACK reordering */
+                               if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) &&
+                                   !after(end_seq, tp->snd_una))
+                                       continue;
                                NET_INC_STATS_BH(LINUX_MIB_TCPSACKDISCARD);
+                       }
                        continue;
                }