Bug#775169: work-around: Bug #775169 pvmove segfaults

[Michael Paoli]

I was encountering same, or highly similar issue,
notably pvmove segfaults.  Currently on Debian oldstable.
I also found the relatively informative similar/related thread:
https://lists.debian.org/debian-user/2011/06/msg01947.html
https://lists.debian.org/debian-user/2011/06/msg02076.html

My situation "the same", or highly similar:
pvmove segfaults
lvs -a
and
lvdisplay -a
show a [pvmove0] volume
attempts at:
pvmove --abort
lvremove [-f] ...pvmove0
lvchange ... pvmove0
lvmremove [-f] ... pvmove0
all failed
The threat mentioned above had suggestion to remove the dm device,
however, using blkid on all dm devices, I found no DM device with
the UUID shown by lvdisplay -a
nor any devices/links with or containing name pvmove0, and pvmove0
(in lvdisplay -a, lvs -a)
also showed it as size 0, and everything LVM related reported it
as locked and inactive.  Not sure what created this situation, but ..

workaround for the above:
I ran vgcfgbackup twice (so I'd have both "current" config saved,
and one slightly older archive saved copy of same).
I then edited the most recent saved copy, removing the pvmove0 section.
I then used vgcfgrestore.
After that, all was fine and good.  :-)  No more complaints or signs
about pvmove0, and pvmove worked again fine and as expected.

Suggestion - see if something can be added to the lvm code to recognize this
situation and self-correct (or even prevent).  It may also be quite feasible
to reproduce the problem, by injecting such data into a file created by
vgcfgbackup, then using vgcfgrestore (I've not attempted that).  But,
in case that may be useful, here's the section I'd removed:
# pwd -P; sed -ne '/pvmove0 {/,/}/p' tigger_00754-383252791.vg
/etc/lvm/archive
                 pvmove0 {
                         id = "23mVuC-uEaK-gB3Z-9WU0-nYbM-91ls-kQ9KXf"
                         status = ["READ", "WRITE", "PVMOVE", "LOCKED"]
                         flags = []
                         allocation_policy = "contiguous"
                         segment_count = 0

                 }
#