I posted a patch for bio_clone() back in May
I thought that you would add it into the main line kernel. I just
checked 2.6.13-rc3 and I did not see it. Could you please add the
attached patch. Thanks.
Please apply the attached patch to the 2.6 kernel. If you still run
into problem, please let us know.
On Wed, 2005-07-27 at 16:57, John Marco wrote:
> I probably ought to chime in here after reading the last couple
> of threads about snapshot corruption.
> About two months ago, I was building a system at home to run
> multiple UML jails using EVMS-managed filesystems. Anyhow, the
> root partition, /dev/evms/root, experienced severe corruption and
> meltdown when I tried to snapshot and back it up.
> After a short time copying the snapshotted filesystem, the 'cpio'
> process froze. Attempts to run any other commands resulted in
> segmentation faults or 'not found'. After a reboot, the system
> would not come back up due to severe damage to the root volume.
> I booted using the Gentoo LiveCD disc, and mounted the root
> volume. Most of the files were either missing or corrupted with
> binary gibberish. I don't have the exact details at this time,
> because I've since rebuilt my root volume, but at the time, I'm
> pretty sure I did a little postmortem snooping with evmsn, and
> saw that the root_snap shapshot object _appeared_ to have the
> "root" object as it's backing store, instead of the 'snap_root'
> region I had actually allocated for that purpose. If the original
> root was being used as backing store for a snapshot of itself,
> that would certainly explain the corruption. I assumed at the
> time that I had probably cut a brain fart and accidentally backed
> the snapshot on the root volume's region instead of snap_root,
> but I'm not normally the type to make that kind of mistake.
> Having seen the recent threads on snapshot corruption, I'm
> starting to think I may have experienced the same problem.
> So, the details of my system at the time were:
> Gentoo 2005.1
> Kernel 2.6.11-gentoo-something. (Exact release not known)
> EVMS 2.5.1
> 2 ATA100 disc drives.
> md0: 256MB Raid-1 for /dev/evms/boot, mirrored across equal 256M
> segments on each disc. Both active/bootable.
> md1: Rest of disc space Raid-1 for LVM2 container named 'raid1',
> Mirrored across segment#2 on each disc.
> /dev/evms/root: 5GB Reiserfs volume allocated from the LVM2
> container on md1.
> Various Reiserfs root filesystems for jailed UML instances.
> At work, I'm running two production systems on Gentoo 2.6.10-hardened-r1 with
> EVMS 2.5.1, and am not seeing any problems. I have automatic nightly backup
> scripts on these servers that snapshot the root and database partitions and
> back them up with no apparent problems.
> I suspect that something has crept into the devmapper or md code
> in 2.6.11+ that causes problems when mirroring over a md raid1 region.
> There's a thread on the Redhat dm mailing list that would seem to indicate
> that this is the case...
> This thread has been going on since March and resurfaced recently. It
> seems to describe exactly the kind of corruption I'm seeing.
> I'm still in the process of building my home system, and I'm tempted to
> see if I can reproduce this behavior with my current setup, which is
> still a gentoo 2.6.11 kernel with EVMS 2.5.1. (After copying my root FS
> in case I need to restore it this time.) At this point, I'm a bit paranoid
> about using snapshots until I can confirm that there are no corruption
> John Marco
> jmarco@... / Samba Holdings, Inc. / Albuquerque, NM
> SF.Net email is Sponsored by the Better Software Conference & EXPO September
> 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
> Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
> Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
> Evms-devel mailing list
> To subscribe/unsubscribe, please visit: