OpenLDAP backend broken by the new ldb.tdb event context work, would appreciate advice on how to fix

[Andrew Bartlett]

On Wed, 2017-07-19 at 17:00 +0300, Nadezhda Ivanova wrote:
> Hi team and Andrew,
> I'm trying to start using the newest Samba for my OpenLDAp backend work 
> - after the long hiatus, I was using code that was 2 years old, so I 
> decided to adapt to the newest master.

It is so good to see you back at this.  It has been a long time!

> However, we are no longer able to provision an OpenLDAP backend - one of 
> the issues I have a workaround for, by adding "client ldap sasl wrapping 
> = plain" to my smb.conf, until I'm able to fix it properly, but the 
> other I need some advice on.
> 
> While trying to create the domain naming context, provisioning hangs here:
> 
> 
> #0  0x00007ffff78f1673 in __epoll_wait_nocancel () at 
> ../sysdeps/unix/syscall-template.S:81
> #1  0x00007ffff5b0a074 in epoll_event_loop (epoll_ev=0x1ec7360, 
> tvalp=0x7fffffffc640) at ../lib/tevent/tevent_epoll.c:642
> #2  0x00007ffff5b0a98c in epoll_event_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent_epoll.c:930
> #3  0x00007ffff5b07693 in std_event_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent_standard.c:114
> #4  0x00007ffff5b0050f in _tevent_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent.c:721
> #5  0x00007ffff64a71bc in ldb_wait (handle=0x15a1fb0, type=LDB_WAIT_ALL) 
> at ../lib/ldb/common/ldb.c:637
> #6  0x00007fffda0d3d7e in new_partition_set_replicated_metadata 
> (ldb=0x1c52460, module=0x1b77090, last_req=0x1d5a070, data=0x15f8bd0, 
> partition=0x1c75bd0)
>      at ../source4/dsdb/samdb/ldb_modules/partition_init.c:601
> #7  0x00007fffda0d46da in partition_create (module=0x1b77090, 
> req=0x174c980) at ../source4/dsdb/samdb/ldb_modules/partition_init.c:827
> #8  0x00007fffda0d21e2 in partition_extended (module=0x1b77090, 
> req=0x174c980) at ../source4/dsdb/samdb/ldb_modules/partition.c:1172
> #9  0x00007ffff6488c42 in ldb_next_request (module=0x1b77090, 
> request=0x174c980) at ../lib/ldb/common/ldb_modules.c:543
> #10 0x00007fffdaefe427 in np_part_search_callback (req=0x2162590, 
> ares=0x15a47a0) at ../source4/dsdb/samdb/ldb_modules/new_partition.c:143
> #11 0x00007fffd6fd7275 in ltdb_request_done (ctx=0x1c9b4c0, error=32) at 
> ../lib/ldb/ldb_tdb/ldb_tdb.c:1314
> #12 0x00007fffd6fd7556 in ltdb_callback (ev=0x1ec70e0, te=0x2303ba0, 
> t=..., private_data=0x1c9b4c0) at ../lib/ldb/ldb_tdb/ldb_tdb.c:1424
> #13 0x00007ffff5b08602 in tevent_common_loop_timer_delay (ev=0x1ec70e0) 
> at ../lib/tevent/tevent_timed.c:369
> #14 0x00007ffff5b0a90c in epoll_event_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent_epoll.c:915
> #15 0x00007ffff5b07693 in std_event_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent_standard.c:114
> #16 0x00007ffff5b0050f in _tevent_loop_once (ev=0x1ec70e0, 
> location=0x7ffff64b0edd "../lib/ldb/common/ldb.c:637") at 
> ../lib/tevent/tevent.c:721
> #17 0x00007ffff64a71bc in ldb_wait (handle=0x15af320, type=LDB_WAIT_ALL) 
> at ../lib/ldb/common/ldb.c:637
> #18 0x00007ffff68c4af2 in py_ldb_add (self=0x7fffde1f1eb0, 
> args=0x7fffd647d908, kwargs=0x0) at ../lib/ldb/pyldb.c:1415
> #19 0x0000000000523f6d in PyEval_EvalFrameEx ()
> #20 0x0000000000555551 in PyEval_EvalCodeEx ()
> 
> 
> This seems to start happening after this commit:
> 
> commit e67d3568e5853387d0675ed13d51df21e6274770
> Author: Andrew Bartlett <abartlet at samba.org>
> Date:   Fri May 12 02:30:01 2017 +0200
> 
>      ldb: Use the private event context in ldb_tdb and ldb_wait()
> 
>      This enables the previous commits, and ensures that ldb_tdb is safe 
> from operations while locks
>      are held
> 
>      Signed-off-by: Andrew Bartlett <abartlet at samba.org>
>      Reviewed-by: Stefan Metzmacher <metze at samba.org>
>      Reviewed-by: Garming Sam <garming at catalyst.net.nz>
> 
> 
> 
> Could you advise me on why this happens, and possible steps to fix?

ldb_ildap.c:ildb_connect()
	ildb->event_ctx = ldb_get_event_context(ldb);

This uses one of those evil nested event loops, which are essentially
banned in Samba and now also ldb, for causing too much mischief.  

Instead, we now create a new event context for every request, so for
every request, you need to find the per-request event context and wait
on that instead.  

Eliminating long-term variables with the event context in it, and
instead using ldb_handle_get_event_context(req->handle) would make a
big difference.

Andrew Bartlett

-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba