[CentOS] CentOS 6.0 CR mdadm-3.2.2 breaks Intel BIOS RAID

Sat Oct 8 21:31:43 UTC 2011
William Warren <hescominsoon at emmanuelcomputerconsulting.com>

On 10/8/2011 3:14 PM, Trey Dockendorf wrote:
> On Sat, Oct 8, 2011 at 2:04 PM, Steve Brooks<steveb at mcs.st-and.ac.uk>wrote:
>
>> On Sat, 8 Oct 2011, Trey Dockendorf wrote:
>>
>>> I just upgraded my home KVM server to CentOS 6.0 CR to make use of the
>>> latest libvirt and now my RAID array with my VM storage is missing.  It
>>> seems that the upgrade to mdadm-3.2.2 is the culprit.
>>>
>>> This is the output from mdadm when scanning that array,
>>>
>>> # mdadm --detail --scan
>>> ARRAY /dev/md0 metadata=imsm UUID=734f79cf:22200a5a:73be2b52:3388006b
>>> ARRAY /dev/md126 metadata=imsm UUID=3d135942:f0fad0b0:33255f78:29c3f50a
>>> mdadm(IMSM): Unsupported attributes : 40000000
>>> mdadm: IMSM metadata loading not allowed due to attributes
>> incompatibility.
>>> mdadm(IMSM): Unsupported attributes : 40000000
>>> mdadm: IMSM metadata loading not allowed due to attributes
>> incompatibility.
>>> ARRAY /dev/md127 container=/dev/md0 member=0
>>> UUID=734f79cf:22200a5a:73be2b52:3388006b
>>>
>>> The error about IMSM shows up on google as something that happened to
>> Fedora
>>> users during a FC14->FC15 upgrade.
>>>
>>> The server itself isn't old, it's a Supermicro 2U with Dual Xeon 5400
>> family
>>> of CPU.  There are two RAIDs on this one controller...a RAID1 which still
>>> functions and a RAID5 which is the one that is unable to be seen.  I
>> don't
>>> know what IMSM is for, but the only thing strange about that array is it
>> is
>>> 2.7TB so the BIOS configured it as two separate arrays, one as 2TB and
>> one
>>> as 700GB, but it was showing up to CentOS as a single volume.
>>>
>>> I downgraded to 3.2.1 , ran mdadm again and bam...it works,
>>>
>>> # mdadm --detail --scan
>>> ARRAY /dev/md0 metadata=imsm UUID=734f79cf:22200a5a:73be2b52:3388006b
>>> ARRAY /dev/md126 metadata=imsm UUID=3d135942:f0fad0b0:33255f78:29c3f50a
>>> ARRAY /dev/md127 container=/dev/md0 member=0
>>> UUID=691f975d:6beecfd8:67b39886:b7ee7f6e
>>>
>>> Hopefully this can be fixed before this version makes it to 6.1, though
>> it's
>>> likely a problem for upstream RHEL as well.
>>>
>>> - Trey
>>> _______________________________________________
>>> CentOS mailing list
>>> CentOS at centos.org
>>> http://lists.centos.org/mailman/listinfo/centos
>>
>>
>> Hmm I recall seeing something like this on an "sl6" box. I think it needed
>> a "/etc/mdadm.conf" with som metadata id code.. I am pretty sure I fixed
>> it with
>>
>> mdadm --detail --scan>>  /etc/mdadm.conf
>>
>> and a reboot.
>>
>> Steve
>> _______________________________________________
>> CentOS mailing list
>> CentOS at centos.org
>> http://lists.centos.org/mailman/listinfo/centos
>>
> That's the problem, mdadm --detail --scan throws errors.  After updating to
> mdadm-3.2.2 , and rebooting, the entry /dev/md126 was removed, likely
> because when I run it and don't pipe the output, it shows this...
>
> # mdadm --detail --scan
> ARRAY /dev/md0 metadata=imsm UUID=734f79cf:22200a5a:73be2b52:3388006b
> ARRAY /dev/md126 metadata=imsm UUID=3d135942:f0fad0b0:33255f78:29c3f50a
> mdadm(IMSM): Unsupported attributes : 40000000
> mdadm: IMSM metadata loading not allowed due to attributes incompatibility.
> mdadm(IMSM): Unsupported attributes : 40000000
> mdadm: IMSM metadata loading not allowed due to attributes incompatibility.
> ARRAY /dev/md127 container=/dev/md0 member=0
> UUID=734f79cf:22200a5a:73be2b52:3388006b
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos
why bother with bios FAKERAID using PIO mode(effectively).  Make a 
backup and redo using MD raid and turn off the bios raid. Then things 
will work correctly.