On Fri, Nov 16, 2007 at 01:39:09PM +0100, Otto Moerbeek wrote: > On Fri, Nov 16, 2007 at 01:13:42PM +0100, Otto Moerbeek wrote: > > > On Fri, Nov 16, 2007 at 12:40:29PM +0100, frantisek holop wrote: > > > hmm, on Fri, Nov 16, 2007 at 12:30:00PM +0100, Toni Mueller said that > > > > > could someone test this before i submit a bug report? > > > > > > > > I've removed the '-s' flag for this reason, although I would very much > > > > prefer to have it in place in the case that I have net access. I don't > > > > know whether it would be feasible for ntpd to see whether there's an > > > > appropriate route, and whether the relevant interfaces are up. > > > > > > it definitely worked in 4.1 even with "-s". > > > > It is this commit, which manages to keep the outging buffer full. > > > > http://www.openbsd.org/cgi-bin/cvsweb/src/usr.sbin/ntpd/client.c.diff?r1=1.75&r2=1.76&f=h > > > > Reverting it makes the problem go away. > > > > Still thinking how to solve this. > > > > -Otto > > This seems to work for me, > > -Otto
Grrrr, sent the wrong diff. This has a far batter chanche. Index: client.c =================================================================== RCS file: /cvs/src/usr.sbin/ntpd/client.c,v retrieving revision 1.76 diff -u -p -r1.76 client.c --- client.c 1 May 2007 07:40:45 -0000 1.76 +++ client.c 17 Nov 2007 16:34:07 -0000 @@ -123,7 +123,8 @@ client_query(struct ntp_peer *p) int tos = IPTOS_LOWDELAY; if (p->addr == NULL && client_nextaddr(p) == -1) { - set_next(p, scale_interval(INTERVAL_QUERY_AGGRESSIVE)); + set_next(p, MAX(SETTIME_TIMEOUT, + scale_interval(INTERVAL_QUERY_AGGRESSIVE))); return (0); } @@ -140,8 +141,8 @@ client_query(struct ntp_peer *p) if (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EHOSTUNREACH || errno == EADDRNOTAVAIL) { client_nextaddr(p); - set_next(p, - scale_interval(INTERVAL_QUERY_AGGRESSIVE)); + set_next(p, MAX(SETTIME_TIMEOUT, + scale_interval(INTERVAL_QUERY_AGGRESSIVE))); return (-1); } else fatal("client_query connect");