[linux-lvm] dmeventd fails to release V_vg lock

[Jaco Kroon <jaco@uls.co.za>]

Hi Guys,

It looks like under some circumstances dmeventd doesn't properly release
a lock it's holding on /var/lock/lvm/V_${vg} ... I suspect it might be
related to snapshots (also see an email from a few hours back). 
However, dmeventd is trying to do *something* with the lock held, or
sleeping, please see below.  Firstly, at the time of testing, dmeventd
is the only process that has the lock file open:

[root@hostjmdb2 proc]# ls -la */fd/* 2>/dev/null | grep -E 'V_vg'
lrwx------ 1 root root 64 Feb 21 11:52 5519/fd/1046 ->
/var/lock/lvm/V_vg_hostdb02
[root@hostjmdb2 proc]#

At this point if I run "lvs" (and then press ^C) I get this:

[root@hostname proc]# lvs
  /dev/vg_hostdb02/snap-lib_mysql-2013-02-20.18: read failed after 0 of
4096 at 536870846464: Input/output error
... bunch of other IO failures on snapshots that has presumably become
invalid
  /dev/vg_hostdb02/snap-lib_mysql-2013-02-21.06: read failed after 0 of
4096 at 4096: Input/output error
^C  CTRL-c detected: giving up waiting for lock
  /var/lock/lvm/V_vg_hostdb02: flock failed: Interrupted system call
  Can't get lock for vg_hostdb02
  Skipping volume group vg_hostdb02
[root@hostname proc]#

So the only logical conclusion is that dmeventd has the lock held.  And
according to wchan in /proc/5519 dmeventd is blocking in
poll_schedule_timeout ... so my suspicion is that under some error
condition dmeventd doesn't release it's lock on V_ ... perhaps an
attempt to extend the snapshot, which then becomes invalid before the
extend is issued or something.

Specific distro in question is Centos 6.5, and lvm2-2.02.95-10.el6.x86_64
-- 
Kind Regards,
Jaco Kroon