Skip site navigation (1)Skip section navigation (2) 
Date:      Sat, 11 Feb 2017 17:05:08 +0000 (UTC)
From:      Ryan Stone <rstone@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r313646 - head/sys/netinet
Message-ID:  <201702111705.v1BH58OE071548@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help 
Author: rstone
Date: Sat Feb 11 17:05:08 2017
New Revision: 313646
URL: https://svnweb.freebsd.org/changeset/base/313646

Log:
  Don't zero out srtt after excess retransmits
  
  If the TCP stack has retransmitted more than 1/4 of the total
  number of retransmits before a connection drop, it decides that
  its current RTT estimate is hopelessly out of date and decides
  to recalculate it from scratch starting with the next ACK.
  
  Unfortunately, it implements this by zeroing out the current RTT
  estimate.  Drop this hack entirely, as it makes it significantly more
  difficult to debug connection issues.  Instead check for excessive
  retransmits at the point where srtt is updated from an ACK being
  received.  If we've exceeded 1/4 of the maximum retransmits,
  discard the previous srtt estimate and replace it with the latest
  rtt measurement.
  
  Differential Revision:	https://reviews.freebsd.org/D9519
  Reviewed by:	gnn
  Sponsored by:	Dell EMC Isilon

Modified:
  head/sys/netinet/tcp_input.c
  head/sys/netinet/tcp_timer.c
  head/sys/netinet/tcp_timer.h

Modified: head/sys/netinet/tcp_input.c
==============================================================================
--- head/sys/netinet/tcp_input.c	Sat Feb 11 15:25:49 2017	(r313645)
+++ head/sys/netinet/tcp_input.c	Sat Feb 11 17:05:08 2017	(r313646)
@@ -3494,7 +3494,7 @@ tcp_xmit_timer(struct tcpcb *tp, int rtt
 
 	TCPSTAT_INC(tcps_rttupdated);
 	tp->t_rttupdated++;
-	if (tp->t_srtt != 0) {
+	if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
 		/*
 		 * srtt is stored as fixed point with 5 bits after the
 		 * binary point (i.e., scaled by 8).  The following magic

Modified: head/sys/netinet/tcp_timer.c
==============================================================================
--- head/sys/netinet/tcp_timer.c	Sat Feb 11 15:25:49 2017	(r313645)
+++ head/sys/netinet/tcp_timer.c	Sat Feb 11 17:05:08 2017	(r313646)
@@ -845,20 +845,16 @@ tcp_timer_rexmt(void * xtp)
 	    (tp->t_rxtshift == 3))
 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
 	/*
-	 * If we backed off this far, our srtt estimate is probably bogus.
-	 * Clobber it so we'll take the next rtt measurement as our srtt;
-	 * move the current srtt into rttvar to keep the current
-	 * retransmit times until then.
+	 * If we backed off this far, notify the L3 protocol that we're having
+	 * connection problems.
 	 */
-	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
+	if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
 #ifdef INET6
 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
 			in6_losing(tp->t_inpcb);
 		else
 #endif
 			in_losing(tp->t_inpcb);
-		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
-		tp->t_srtt = 0;
 	}
 	tp->snd_nxt = tp->snd_una;
 	tp->snd_recover = tp->snd_max;

Modified: head/sys/netinet/tcp_timer.h
==============================================================================
--- head/sys/netinet/tcp_timer.h	Sat Feb 11 15:25:49 2017	(r313645)
+++ head/sys/netinet/tcp_timer.h	Sat Feb 11 17:05:08 2017	(r313646)
@@ -119,6 +119,13 @@
 
 #define	TCPTV_DELACK	( hz/10 )		/* 100ms timeout */
 
+/*
+ * If we exceed this number of retransmits for a single segment, we'll consider
+ * the current srtt measurement no longer valid and will recalculate from
+ * scratch starting with the next ACK.
+ */
+#define TCP_RTT_INVALIDATE (TCP_MAXRXTSHIFT / 4)
+
 #ifdef	TCPTIMERS
 static const char *tcptimers[] =
     { "REXMT", "PERSIST", "KEEP", "2MSL", "DELACK" };

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201702111705.v1BH58OE071548>