tcp: don't backtrack to sacked skbs

Backtracking to sacked skbs is a horrible performance killer
since the hint cannot be advanced successfully past them...
...And it's totally unnecessary too.

In theory this is 2.6.27..28 regression but I doubt anybody
can make .28 to have worse performance because of other TCP
improvements.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Ilpo Järvinen 2009-02-28 04:44:27 +00:00 committed by David S. Miller
parent 79cb5def8d
commit ac11ba753f
1 changed files with 1 additions and 1 deletions

View File

@ -2065,7 +2065,7 @@ begin_fwd:
goto begin_fwd; goto begin_fwd;
} else if (!(sacked & TCPCB_LOST)) { } else if (!(sacked & TCPCB_LOST)) {
if (hole == NULL && !(sacked & TCPCB_SACKED_RETRANS)) if (hole == NULL && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
hole = skb; hole = skb;
continue; continue;