Re: tcp_sack_output: Computed sack hole not the same as cached value

This patch (from Noritoshi Demizu) should fix this.
Andrey Chernov reported this as well and the patch fixes 
the problem for him. This will be committed to -current 
shortly.

mohan

Index: netinet/tcp_input.c
===================================================================
RCS file: /home/cvsup/FreeBSD/ncvs/src/sys/netinet/tcp_input.c,v
retrieving revision 1.277
diff -u -r1.277 tcp_input.c
--- netinet/tcp_input.c 27 Jun 2005 22:27:42 -0000      1.277
+++ netinet/tcp_input.c 29 Jun 2005 09:42:35 -0000
_at__at_ -1147,7 +1147,8 _at__at_
                            ((!tcp_do_newreno && !tp->sack_enable &&
                              tp->t_dupacks < tcprexmtthresh) ||
                             ((tcp_do_newreno || tp->sack_enable) &&
-                             !IN_FASTRECOVERY(tp) && to.to_nsacks == 0))) {
+                             !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
+                             TAILQ_EMPTY(&tp->snd_holes)))) {
                                KASSERT(headlocked, ("headlocked"));
                                INP_INFO_WUNLOCK(&tcbinfo);
                                headlocked = 0;
_at__at_ -1822,7 +1823,8 _at__at_
                        tcpstat.tcps_rcvacktoomuch++;
                        goto dropafterack;
                }
-               if (tp->sack_enable)
+               if (tp->sack_enable &&
+                   (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
                        tcp_sack_doack(tp, &to, th->th_ack);
                if (SEQ_LEQ(th->th_ack, tp->snd_una)) {  
                        if (tlen == 0 && tiwin == tp->snd_wnd) {
Index: netinet/tcp_subr.c
===================================================================
RCS file: /home/cvsup/FreeBSD/ncvs/src/sys/netinet/tcp_subr.c,v
retrieving revision 1.227
diff -u -r1.227 tcp_subr.c
--- netinet/tcp_subr.c  1 Jun 2005 12:06:07 -0000       1.227
+++ netinet/tcp_subr.c  29 Jun 2005 09:42:35 -0000
_at__at_ -605,8 +605,7 _at__at_
        if (tcp_do_rfc1323)
                tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
        tp->sack_enable = tcp_do_sack;
-       if (tp->sack_enable)
-               TAILQ_INIT(&tp->snd_holes);
+       TAILQ_INIT(&tp->snd_holes);
        tp->t_inpcb = inp;      /* XXX */
        /*
         * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no

--- Daniel Eriksson <daniel_k_eriksson_at_telia.com> wrote:

> 
> With latest CURRENT (2005.06.30.12.00.00) I get a ton of these messages
> on the console. 50-100 per minute and that's when load is light. With
> the old kernel (2005.06.23.00.30.00) no such messages show up.
> 
> tcp_sack_output: Computed sack hole not the same as cached value
> 
> This is on an SMP machine. On my UP server I saw no messages, but I only
> had the two machines (one UP and one SMP) running the new kernel for a
> few minutes.
> 
> /Daniel Eriksson
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
>