[linux-lvm] LVM on Raid-5 woes...

[linux-lvm] LVM on Raid-5 woes...

[Fabian Wenzel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello there!

Currently I am trying to set up LVM on top of RAID-5. I am using Kernel
2.6.12.2, device-mapper 1.01.04-r2 and lvm2 2.01.15-r1 on an NSLU2 (big
endian ARM) with three USB hard disks connected (two via a hub).

My problem is that I cannot create any volume groups; even physical
volumes cause trouble: After typing

root@platon:/etc# pvcreate /dev/md0

the messages of "pvscan" randomly change without doing anything in
between - sometimes two error messages appear, sometimes there are three
(see attachment 1).

Similar things happen when trying to set up a volume group with

root@platon:/etc# vgcreate -s 64M data /dev/md0
  Incorrect metadata area header checksum
  Incorrect metadata area header checksum
  Volume group "data" successfully created


When scanning for volume groups using "vgscan", error messages change,
but the volume group is never found (see attachment 2).


I set up the device-filter in lvm.conf correctly such that only
md-devices are detected: When increasing the loglevel, it can be seen
that the errors are indeed related to the md-device (see attachment 3).

The weird thing is the fact that errors change from call to call. So
this must be a timing problem (I don't have another explanation for
it)?! However, I could write an ext3-file system to /dev/md0 and mount
it afterwards without any problems.

If anyone could help me, that'd be great!

	Fabian


- -----------------------------------------------------------------------
Attachment 1:

root@platon:/etc# pvscan /dev/md0
  Incorrect metadata area header checksum
  Incorrect metadata area header checksum
  Incorrect metadata area header checksum
  PV /dev/md0         lvm2 [372.28 GB]
  Total: 1 [372.28 GB] / in use: 0 [0   ] / in no VG: 1 [372.28 GB]
root@platon:/etc# pvscan /dev/md0
  Incorrect metadata area header checksum
  Incorrect metadata area header checksum
  PV /dev/md0         lvm2 [372.28 GB]
  Total: 1 [372.28 GB] / in use: 0 [0   ] / in no VG: 1 [372.28 GB]

- -----------------------------------------------------------------------
Attachment 2:

root@platon:/etc# vgscan
  Reading all physical volumes.  This may take a while...
  Incorrect metadata area header checksum
root@platon:/etc# vgscan
  Reading all physical volumes.  This may take a while...
  No volume groups found
root@platon:/etc# vgscan
  Reading all physical volumes.  This may take a while...
  Incorrect metadata area header checksum

- -----------------------------------------------------------------------
Attachment 2:

#label/label.c:168       /dev/md0: lvm2 label detected
#device/dev-io.c:484         Closed /dev/md0
#device/dev-io.c:438         Opened /dev/md0 RO O_DIRECT
#device/dev-io.c:134         /dev/md0: block size is 4096 bytes
#format_text/format-text.c:136   Incorrect metadata area header checksum
#format_text/format-text.c:838         <backtrace>
#device/dev-io.c:484         Closed /dev/md0
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFD0MZsbJzYO4mghQIRAhYlAJ0RdWYmBfv0XTP0bJVrItwYwtoGfACfejc9
x0CRv9U3gNtZANXEQXP5yVs=
=Z1NP
-----END PGP SIGNATURE-----

Re: [linux-lvm] LVM on Raid-5 woes...

[Michael Loftis]

--On January 20, 2006 12:15:56 PM +0100 Fabian Wenzel <f.wenzel@gmx.net> 
wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hello there!
>
> Currently I am trying to set up LVM on top of RAID-5. I am using Kernel
> 2.6.12.2, device-mapper 1.01.04-r2 and lvm2 2.01.15-r1 on an NSLU2 (big
> endian ARM) with three USB hard disks connected (two via a hub).
>
> My problem is that I cannot create any volume groups; even physical
> volumes cause trouble: After typing

<...>

cat /proc/mdstat ?  what state is MD in during all of this?  sounds more 
like an MD RAID problem than a LVM problem...IE MD isn't able to operate 
reliably.  Verify the proper operation of MD first.

Re: [linux-lvm] LVM on Raid-5 woes...

[Michael Loftis]

--On January 20, 2006 12:15:56 PM +0100 Fabian Wenzel <f.wenzel@gmx.net> 
wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hello there!
>
> Currently I am trying to set up LVM on top of RAID-5. I am using Kernel
> 2.6.12.2, device-mapper 1.01.04-r2 and lvm2 2.01.15-r1 on an NSLU2 (big
> endian ARM) with three USB hard disks connected (two via a hub).
>
> My problem is that I cannot create any volume groups; even physical
> volumes cause trouble: After typing

Oh also, do the drives work fine by themselves?  IE remove MD RAID fromt eh 
picture completely temporarily, can you pvcreate directly on the devices?

Re: [linux-lvm] LVM on Raid-5 woes...

[Fabian Wenzel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

> Oh also, do the drives work fine by themselves?  IE remove MD RAID fromt
> eh picture completely temporarily, can you pvcreate directly on the
> devices?

Will try tomorrow if the solution below is wrong. The raid is running
fine, all devices are up.

Meanwhile, I googled over and over again and found an two important bug
reports related to the same issue (Debian bug report #247896 and, more
important, #267157 - can be found on bugs.debian.org)

I think the problem is that I have to rebuild lvm2 using
"--disable-o_direct". I don't know why there is some trouble on an ARM,
and (of course) it would be good if "o_direct" can eventually be used
(whatever it is), but I'm positive that this will do the trick.

Otherwise, I will send another message to this thread.

Thanks anyway,

       Fabian





-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFD0QA2bJzYO4mghQIRAjrzAJ0YYElKKVpjiXJ/d0/5ZZah+pbYMQCfTDOY
QJNta7QiLT+HIaph9kk3s00=
=aXQE
-----END PGP SIGNATURE-----