1) Skip condition used to be wrong way around which made SACK
processing very broken, missed many blocks because of that.
2) Use highest_sack advancement only if some skbs are already
sacked because otherwise tcp_write_queue_next may move things
too far (occurs mainly with GSO). The other similar advancement
is not problem because highest_sack was previosly put to point
a sacked skb.
These problems were located because of problem report from Matt
Mathis <[EMAIL PROTECTED]>.
Signed-off-by: Ilpo Järvinen <[EMAIL PROTECTED]>
---
...I verified result from time-seq graphs now, they are right in my simple
test case.
In case there's a rebase of net-2.6.25, it might be good to just combine
173b7ae366ac3f69b3030a6ab308cf3fa3ab751a ([TCP]: Correct DSACK check
placing) and this one to the a1598ba3339ebeeb0a959141a122cbe69956e713
([TCP]: Rewrite SACK block processing & sack_recv_cache use), they just
fix its bugs.
net/ipv4/tcp_input.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 0cee3dc..cadde8b 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -1384,7 +1384,7 @@ static struct sk_buff *tcp_sacktag_skip(struct sk_buff
*skb, struct sock *sk,
if (skb == tcp_send_head(sk))
break;
- if (before(TCP_SKB_CB(skb)->end_seq, skip_to_seq))
+ if (!before(TCP_SKB_CB(skb)->end_seq, skip_to_seq))
break;
}
return skb;
@@ -1575,7 +1575,7 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff
*ack_skb, u32 prior_snd_
continue;
}
- if (!before(start_seq, tcp_highest_sack_seq(tp))) {
+ if (tp->sacked_out && after(start_seq,
tcp_highest_sack_seq(tp))) {
skb = tcp_write_queue_next(sk, tp->highest_sack);
fack_count = tp->fackets_out;
}
--
1.5.0.6
