prevent rx peer timeout from reaching 0.0 seconds

The rx peer timeout is computed from the round trip time
calculation.  It traditionally has had a lowerbound of 350ms.
The computation in rxi_ComputeRoundTripTime() was incorrect
and instead used 350ms as an upperbound.

rxi_ComputeRoundTripTime() had a second problem wherein if
the actually RTT is shorter than the resolution of the clock
then the RTT would quickly approach 0.0 seconds.  Enforce
a lowerbound of 1ms if the RTT for a given packet appears
to be 0.0 seconds.

LICENSE BSD

Change-Id: I83d3a3d64127d4c023df2fce5e45a4df3eb35679
Reviewed-on: http://gerrit.openafs.org/696
Tested-by: Jeffrey Altman <jaltman@openafs.org>
Reviewed-by: Jeffrey Altman <jaltman@openafs.org>
Tested-by: Derrick Brashear <shadow@dementia.org>
Reviewed-by: Derrick Brashear <shadow@dementia.org>
Reviewed-on: http://gerrit.openafs.org/974

diff --git a/src/rx/rx.c b/src/rx/rx.c
index 812eeab1a6711a93d6d8df18ef2d47b2698c6edb..0fdb8ab5a78e2aa8afbb99ac7da6af3693fa45eb 100644 (file)
--- a/src/rx/rx.c
+++ b/src/rx/rx.c
@@ -5708,6 +5708,15 @@ rxi_ComputeRoundTripTime(register struct rx_packet *p,
        return;                 /* somebody set the clock back, don't count this time. */
     }
     clock_Sub(rttp, sentp);
+    if (rttp->sec == 0 && rttp->usec == 0) {
+        /*
+         * The actual round trip time is shorter than the
+         * clock_GetTime resolution.  It is most likely 1ms or 100ns.
+         * Since we can't tell which at the moment we will assume 1ms.
+         */
+        rttp->usec = 1000;
+    }
+
     MUTEX_ENTER(&rx_stats_mutex);
     if (clock_Lt(rttp, &rx_stats.minRtt))
        rx_stats.minRtt = *rttp;
@@ -5781,7 +5790,7 @@ rxi_ComputeRoundTripTime(register struct rx_packet *p,
      * be switched and/or swapped out.  So on fast, reliable networks, the
      * timeout would otherwise be too short.  
      */
-    rtt_timeout = MIN((peer->rtt >> 3) + peer->rtt_dev, 350);
+    rtt_timeout = MAX(((peer->rtt >> 3) + peer->rtt_dev), 350);
     clock_Zero(&(peer->timeout));
     clock_Addmsec(&(peer->timeout), rtt_timeout);