The following reply was made to PR user/6509; it has been noted by GNATS.
From: Reyk Floeter <[email protected]>
To: Stuart Henderson <[email protected]>
Cc: [email protected], Jonathan Gray <[email protected]>,
Reyk Floeter <[email protected]>
Subject: Re: user/6509: relayctl show sessions make relayd crash
Date: Thu, 19 May 2011 12:33:10 +0200
I still get the problem with your diff applied on -current.
fatal: wrong message for session
hce exiting, pid 28577
lost child: pfe exited abnormally
lost child: hce exited okay
relay exiting, pid 12860
relay exiting, pid 17479
relay exiting, pid 15617
relay exiting, pid 4932
relay exiting, pid 11902
parent terminating, pid 30803
I accelerated the test a little bit by running two fast loops at the
same time. It seems that it happens less often with the diff but it
is still there.
# relayd -dvvf relayd-make-me-crash.conf
$ while true; do relayctl show sessions; done
$ while true; do relayctl show sessions; done
I'm thinking about another solution to make it fully async. I
generally dislike this while(!done) loop in show_sessions() which is
causing all kinds of difficulties. I try to come up with a diff.
reyk
On Wed, May 18, 2011 at 03:30:58PM +0100, Stuart Henderson wrote:
> jsg@ wrote on 2010-11-18:
> > The problem is we have multiple handlers and get
> > IMSG_STATISTICS when in the secondary handler.
> >
> > Try the following diff from reyk:
>
> This still occurs with up-to-date relayd; here is an updated
> version of this diff (applies on top of the current reload diff)a
> which fixes the problem for me.
>
> Index: relay.c
> ===================================================================
> RCS file: /cvs/src/usr.sbin/relayd/relay.c,v
> retrieving revision 1.136
> diff -u -p -r1.136 relay.c
> --- relay.c 9 May 2011 12:08:47 -0000 1.136
> +++ relay.c 18 May 2011 14:26:45 -0000
> @@ -450,10 +450,12 @@ relay_statistics(int fd, short events, v
> if (resetday)
> cur->last_day = 0;
>
> - crs.id = rlay->rl_conf.id;
> - crs.proc = proc_id;
> - proc_compose_imsg(env->sc_ps, PROC_PFE, -1, IMSG_STATISTICS, -1,
> - &crs, sizeof(crs));
> + if (!env->sc_statstop) {
> + crs.id = rlay->rl_conf.id;
> + crs.proc = proc_id;
> + proc_compose_imsg(env->sc_ps, PROC_PFE, -1,
> + IMSG_STATISTICS, -1, &crs, sizeof(crs));
> + }
>
> for (con = SPLAY_ROOT(&rlay->rl_sessions);
> con != NULL; con = next_con) {
> @@ -2489,6 +2491,7 @@ relay_dispatch_pfe(int fd, struct privse
> evtimer_add(&con->se_ev, &tv);
> break;
> case IMSG_CTL_SESSION:
> + env->sc_statstop = 1;
> TAILQ_FOREACH(rlay, env->sc_relays, rl_entry) {
> SPLAY_FOREACH(con, session_tree,
> &rlay->rl_sessions) {
> @@ -2499,6 +2502,7 @@ relay_dispatch_pfe(int fd, struct privse
> }
> proc_compose_imsg(env->sc_ps, p->p_id, -1, IMSG_CTL_END,
> -1, NULL, 0);
> + env->sc_statstop = 0;
> break;
> default:
> return (-1);
> Index: relayd.h
> ===================================================================
> RCS file: /cvs/src/usr.sbin/relayd/relayd.h,v
> retrieving revision 1.146
> diff -u -p -r1.146 relayd.h
> --- relayd.h 9 May 2011 12:08:47 -0000 1.146
> +++ relayd.h 18 May 2011 14:26:45 -0000
> @@ -829,6 +829,7 @@ struct relayd {
>
> struct event sc_statev;
> struct timeval sc_statinterval;
> + int sc_statstop;
>
> int sc_snmp;
> struct event sc_snmpto;
