[linux-lvm] Help! snapshot apparenly trashed LVM

[linux-lvm] Help! snapshot apparenly trashed LVM

[ross]

I did some expanding and shrinking of my volumes and then attempted to
take a snapshot.  I got an error message that started "Invalid LV in
extent map".  I now see that message when I do lvscan or pvscan as well.

This looks very bad, and seems to be the same situation as described here:
http://weblog.pell.portland.or.us/~orc/2008/02/20/000/index.html

Despite these warnings, I *seem* to be able to use my volumes.  However, I
have stopped doing anything on the system to try to prevent corruption.

First, is it safe to continue to use the system?

Second, what can I do to diagnose or recover?

I'm running a stock Debian 2.6.24 kernel with a single SCSI disk.  Some of
its partitions are dedicated to an LVM group, out of which most of my
partitions are carved.  The LVM groups were originally created under EVMS
with a 2.4 kernel.  I converted them to straight LVM months ago.

Thanks.
Ross Boylan

[linux-lvm] Re: Help! snapshot apparenly trashed LVM [recovered? but still Q's]

[ross]

> I did some expanding and shrinking of my volumes and then attempted to
> take a snapshot.  I got an error message that started "Invalid LV in
> extent map".  I now see that message when I do lvscan or pvscan as well.
>
> This looks very bad, and seems to be the same situation as described here:
> http://weblog.pell.portland.or.us/~orc/2008/02/20/000/index.html
>
> Despite these warnings, I *seem* to be able to use my volumes.  However, I
> have stopped doing anything on the system to try to prevent corruption.
>
> First, is it safe to continue to use the system?
>
> Second, what can I do to diagnose or recover?
>
> I'm running a stock Debian 2.6.24 kernel with a single SCSI disk.  Some of
> its partitions are dedicated to an LVM group, out of which most of my
> partitions are carved.  The LVM groups were originally created under EVMS
> with a 2.4 kernel.  I converted them to straight LVM months ago.
>
Following some suggestions in the archive, I used vgcfgrestore -M1 with
the last metadata before the backup, and things seem to be back.  At least
lvscan runs with out errors.

So, 2 new questions:
1) are my data OK? (I'm also curious about what state the system was in
before the recovery--was it safe to use?)

2) how can I take a snapshot without running into this problem?

Thanks.
Ross

Re: [linux-lvm] Re: Help! snapshot apparenly trashed LVM [recovered? but still Q's]

[Milan Broz]

ross@biostat.ucsf.edu wrote:
>> I did some expanding and shrinking of my volumes and then attempted to
>> take a snapshot.  I got an error message that started "Invalid LV in
>> extent map".  I now see that message when I do lvscan or pvscan as well.
>>
>> This looks very bad, and seems to be the same situation as described here:
>> http://weblog.pell.portland.or.us/~orc/2008/02/20/000/index.html

And where did you send bugreport for that problem?

ok, it seems that we have already one

https://bugzilla.redhat.com/show_bug.cgi?id=209927 

Should be fixed in LVM2 2.02.36.

Milan
--
mbroz@redhat.com

Re: [linux-lvm] Re: Help! snapshot apparenly trashed LVM [recovered? but still Q's]

[Ross Boylan]

On Fri, 2008-06-06 at 08:29 +0200, Milan Broz wrote:
> ross@biostat.ucsf.edu wrote:
> >> I did some expanding and shrinking of my volumes and then attempted to
> >> take a snapshot.  I got an error message that started "Invalid LV in
> >> extent map".  I now see that message when I do lvscan or pvscan as well.
> >>
> >> This looks very bad, and seems to be the same situation as described here:
> >> http://weblog.pell.portland.or.us/~orc/2008/02/20/000/index.html
> 
> And where did you send bugreport for that problem?
> 
> ok, it seems that we have already one
> 
> https://bugzilla.redhat.com/show_bug.cgi?id=209927 
> 
> Should be fixed in LVM2 2.02.36.
> 
> Milan
> --
> mbroz@redhat.com

Excellent!

Is that a kernel level or userspace fix?  How do I know if I've got it?
Currently lvm version says
  LVM version:     2.02.35 (2008-04-15)
  Library version: 1.02.24 (2007-12-20)
  Driver version:  4.12.0


The bug indicates that converting to LVM2 will fix things.  I have no
reason to stick with LVM1, so that may be the best way to go.  Are there
any gotcha's?  Can vgconvert be run while the volumes and group are in
use?

Thanks.
Ross