dovecot - May 2012 - Dovecot auth process delays exiting if LDAPS passdb used

Hi,

I'm using Dovecot 2.1.6 and LDAP server as a backend for
Postfix SMTP-Auth (SASL) on Debian GNU/Linux (wheezy),
Solaris 10 and AIX 6.1/7.1.

If Dovecot passdb is configured with LDAP (no TLS/SSL),
it is no problem. But if Dovecot passdb is configured with
LDAPS (or LDAP+TLS), Dovecot auth process has a problem
that Dovecot auth delays exiting about between 20 and
60 seconds when Dovecot dovecot (master) process is already
terminated by an administrator.

Is this known problem?

On AIX, this problem prevents restart Dovecot service by
the following error:

May 24 00:42:03 build-aix6 mail:warn|warning dovecot: master: Warning: Killed
with signal 15 (by pid=11337890 uid=0 code=kill)
May 24 00:42:10 build-aix6 mail:err|error dovecot: master: Error: service(auth):
Socket already exists: /opt/osstech/var/run/dovecot/auth-login
May 24 00:42:11 build-aix6 mail:crit dovecot: master: Fatal: Failed to start
listeners

My Dovecot 2.1.6 platforms are:

  * Debian GNU/Linux wheezy + OpenLDAP 2.4.28 (linked with GNU TLS 2.12.18)
  * Solaris 10 + OpenLDAP 2.4.26 (linked with OpenSSL 0.9.7)
  * AIX 6.1 + OpenLDAP 2.4.31 (linked with OpenSSL 1.0.0)
  * AIX 7.1 + OpenLDAP 2.4.28 (linked with OpenSSL 1.0.0)

Because I see the problem on the all of above, there is a
bug in Dovecot 2.1.6, I think.

Thanks.

-- 
-- Name: SATOH Fumiyasu (fumiyas @ osstech co jp)
-- Business Home: http://www.OSSTech.co.jp/
-- Personal Home: http://www.SFO.jp/blog/

Hi,

At Thu, 24 May 2012 01:01:25 +0900,
SATOH Fumiyasu wrote:
> If Dovecot passdb is configured with LDAP (no TLS/SSL),
> it is no problem. But if Dovecot passdb is configured with
> LDAPS (or LDAP+TLS), Dovecot auth process has a problem
> that Dovecot auth delays exiting about between 20 and
> 60 seconds when Dovecot dovecot (master) process is already
> terminated by an administrator.
> 
> Is this known problem?
GDB backtrace in Dovecot auth process on Debian GNU/Linux:

(gdb) bt full
#0  0x00007fa725c59f43 in __epoll_wait_nocancel ()
    at ../sysdeps/unix/syscall-template.S:82
No locals.
#1  0x00007fa7267a6db7 in io_loop_handler_run (ioloop=0x17686d0) at
ioloop-epoll.c:181
        ctx = 0x1770350
        events = 0xfffffffffffffffc
        event = 0x7530
        list = <optimized out>
        io = <optimized out>
        tv = {tv_sec = 29, tv_usec = 999374}
        msecs = 30000
        ret = <optimized out>
        i = <optimized out>
        j = <optimized out>
        call = <optimized out>
#2  0x00007fa7267a5cc9 in io_loop_run (ioloop=0x17686d0) at ioloop.c:398
No locals.
#3  0x00007fa7267938ff in master_service_run (service=0x1768580,
callback=<optimized out>)
    at master-service.c:544
No locals.
#4  0x0000000000418517 in main (argc=1, argv=0x1768370) at main.c:373
        c = <optimized out>

DBX backtrace in Dovecot auth process on Solaris 10:

(dbx) where
  [1] __pollsys(0x80bc3c8, 0x4, 0x8047cd8, 0x0), at 0xfed75a15
  [2] _pollsys(0x80bc3c8, 0x4, 0x8047cd8, 0x0), at 0xfed69fc8
  [3] _poll(0x80bc3c8, 0x4, 0x7530), at 0xfed1e95a
=>[4] io_loop_handler_run(ioloop = ???) (optimized), at 0xfef59294 (line
~166) in "ioloop-poll.c"
  [5] io_loop_run(ioloop = ???) (optimized), at 0xfef586cf (line ~398) in
"ioloop.c"
  [6] master_service_run(service = ???, callback = ???) (optimized), at
0xfef42e03 (line ~544) in "master-service.c"
  [7] main(argc = ???, argv = ???) (optimized), at 0x8077d9b (line ~374) in
"main.c"

Thanks.

-- 
-- Name: SATOH Fumiyasu (fumiyas @ osstech co jp)
-- Business Home: http://www.OSSTech.co.jp/
-- Personal Home: http://www.SFO.jp/blog/
> On AIX, this problem prevents restart Dovecot service by
> the following error:
> 
> May 24 00:42:03 build-aix6 mail:warn|warning dovecot: master: Warning:
Killed with signal 15 (by pid=11337890 uid=0 code=kill)
> May 24 00:42:10 build-aix6 mail:err|error dovecot: master: Error:
service(auth): Socket already exists: /opt/osstech/var/run/dovecot/auth-login
> May 24 00:42:11 build-aix6 mail:crit dovecot: master: Fatal: Failed to
start listeners
> 
> My Dovecot 2.1.6 platforms are:
> 
>   * Debian GNU/Linux wheezy + OpenLDAP 2.4.28 (linked with GNU TLS 2.12.18)
>   * Solaris 10 + OpenLDAP 2.4.26 (linked with OpenSSL 0.9.7)
>   * AIX 6.1 + OpenLDAP 2.4.31 (linked with OpenSSL 1.0.0)
>   * AIX 7.1 + OpenLDAP 2.4.28 (linked with OpenSSL 1.0.0)
> 
> Because I see the problem on the all of above, there is a
> bug in Dovecot 2.1.6, I think.

At Thu, 24 May 2012 01:01:25 +0900,
SATOH Fumiyasu wrote:
> If Dovecot passdb is configured with LDAP (no TLS/SSL),
> it is no problem. But if Dovecot passdb is configured with
> LDAPS (or LDAP+TLS), Dovecot auth process has a problem
> that Dovecot auth delays exiting about between 20 and
> 60 seconds when Dovecot dovecot (master) process is already
> terminated by an administrator.
I can reproduce this problem with LDAP (no TLS/SSL) passdb.

Sorry...

-- 
-- Name: SATOH Fumiyasu (fumiyas @ osstech co jp)
-- Business Home: http://www.OSSTech.co.jp/
-- Personal Home: http://www.SFO.jp/blog/