srtt and rtt_seq out of sync

From: Prasanna Meda
Date: Wed Aug 18 2004 - 21:09:55 EST



tcp_init_metrics initialises srtt from route cache.
tcp_rtt_estimator has an assumption that
whenerver srtt is set, rtt_seq is set.
Invariant assumed before(rtt_seq,snd_nxt) is
not true most of the time while initailising from
cache, so rttvar will not be calcualted for this
socket, rto will be wrong for long time.
The condition after(send_una, rtt_seq=0) is
correct 50% of the time of random send_una.
mdev check also resets the rttvar, but it will be
slower.

One fix is, initialising the rtt_seq in tcp_init_metrics
properly. Another fix is, doing the special check
like the below to test whether the invariant has
failed, i.e. initialised from the route cache.

--- net/ipv4/tcp_input.c Tue Jun 15 22:19:43 2004
+++ net/ipv4/tcp_input.c Wed Aug 18 17:36:18 2004
@@ -641,7 +641,7 @@
if (tp->mdev_max > tp->rttvar)
tp->rttvar = tp->mdev_max;
}
- if (after(tp->snd_una, tp->rtt_seq)) {
+ if (after(tp->snd_una, tp->rtt_seq) ||
before(tp->snd_nxt, tp->rtt_seq))
if (tp->mdev_max < tp->rttvar)
tp->rttvar -=
(tp->rttvar-tp->mdev_max)>>2;
tp->rtt_seq = tp->snd_nxt;

Fix from Dave miller:
===== net/ipv4/tcp_input.c 1.69 vs edited =====
--- 1.69/net/ipv4/tcp_input.c 2004-08-02 17:05:30 -07:00
+++ edited/net/ipv4/tcp_input.c 2004-08-18 17:53:54 -07:00
@@ -852,8 +852,10 @@
* to low value, and then abruptly stops to do it and starts to
delay
* ACKs, wait for troubles.
*/
- if (dst_metric(dst, RTAX_RTT) > tp->srtt)
+ if (dst_metric(dst, RTAX_RTT) > tp->srtt) {
tp->srtt = dst_metric(dst, RTAX_RTT);
+ tp->rtt_seq = tp->snd_nxt;
+ }
if (dst_metric(dst, RTAX_RTTVAR) > tp->mdev) {
tp->mdev = dst_metric(dst, RTAX_RTTVAR);
tp->mdev_max = tp->rttvar = max(tp->mdev, TCP_RTO_MIN);

Thanks,
Prasanna.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/