rebuilt HW RAID60 array; XFS filesystem looks bad now

rebuilt HW RAID60 array; XFS filesystem looks bad now

[Paul Brunk]

Hi:

Short version: XFS filesystem on HW RAID60 array.  Array has been
multiply rebuilt due to drive insertions.  XFS filesystem damaged and
trying to salvage what I can, and I want to make sure I have no option
other than "xfs_repair -L".  Details follow.

# uname -a
Linux rccstor7.local 2.6.32-431.5.1.el6.x86_64 #1 SMP Wed Feb 12 
00:41:43 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

xfs_repair version 3.1.1.  The box has one 4-core Opteron CPU and 8
GB of RAM.

I have a 32TB HW RAID60 volume (Areca 1680 HW RAID) made of two RAID6
raid sets.

This volume is a PV in Linux LVM, with a single LV defined in it.  The
LV had an XFS filesystem created on it (no external log).

I can't do xfs_info on it because I can't mount the filesystem.

I had multiple drive removals and insertions (due to timeout error
with non-TLER drives in the RAID array, an unfortunate setup I
inherited), which triggered multiple HW RAID rebuilds.  This caused
the RAID volume to end up defined twice in the controller, with each
of the two constituent RAID sets being defined twice.  At Areca's
direction, I did a "raid set rescue" in the Areca controller.  That
succeeded in reducing the number of volumes from two to one, and the
RAID volume is now "normal" in the RAID controller instead of
"failed".

The logical volume is visible to the OS now, unlike when the RAID
status was "failed".

   # lvdisplay
   --- Logical volume ---
   LV Path                /dev/vg0/lv0
   LV Name                lv0
   VG Name                vg0
   LV UUID                YMlFWe-PTGe-5kHx-V3uo-31Vp-grXR-9ZBt3R
   LV Write Access        read/write
   LV Creation host, time ,
   LV Status              available
   # open                 0
   LV Size                32.74 TiB
   Current LE             8582595
   Segments               1
   Allocation             inherit
   Read ahead sectors     auto
   - currently set to     256
   Block device           253:2

That's good, but now I think the XFS filesystem is in bad shape.

  # grep /media/shares /etc/fstab
  UUID="9cba4e90-1d8f-4a98-8701-df10a28556da" /media/shares xfs pquota 0 0

That UUID entry in /dev/disk/by-uuid is a link to /dev/dm-2.

"dm-2" is the RAID volume.  Here it is in /proc/partitions:
  major minor  #blocks     name
   253     2   35154309120 dm-2

When I try to mount the XFS filesystem:

  # mount /media/shares
  mount: wrong fs type, bad option, bad superblock on /dev/mapper/vg0-lv0,
         missing codepage or helper program, or other error
         In some cases useful info is found in syslog - try
         dmesg | tail  or so

  # dmesg|tail
  XFS (dm-2): Mounting Filesystem
  XFS (dm-2): Log inconsistent or not a log (last==0, first!=1)
  XFS (dm-2): empty log check failed
  XFS (dm-2): log mount/recovery failed: error 22
  XFS (dm-2): log mount failed

  # xfs_check /dev/dm-2
  xfs_check: cannot init perag data (117)
  XFS: Log inconsistent or not a log (last==0, first!=1)
  XFS: empty log check failed

  # xfs_repair -n /dev/dm-2
  produced at least 7863 lines of output.   It begins

  Phase 1 - find and verify superblock...
  Phase 2 - using internal log
          - scan filesystem freespace and inode maps...
  bad magic # 0xa04850d in btbno block 0/108
  expected level 0 got 10510 in btbno block 0/108
  bad btree nrecs (144, min=255, max=510) in btbno block 0/108
  block (0,80-80) multiply claimed by bno space tree, state - 2
  block (0,108-108) multiply claimed by bno space tree, state - 7

  # egrep -c "invalid start block" xfsrepair.out
  2061
  # egrep -c "multiply claimed by bno" xfsrepair.out
  4753

  Included in the output are 381 occurrences of this pair of messages:

  bad starting inode # (0 (0x0 0x0)) in ino rec, skipping rec
  badly aligned inode rec (starting inode = 0)

Is there anything I should try prior to xfs_repair -L?

I'm just trying to salvage whatever I can from this FS.  I'm aware it
could be all gone.  Thanks.

-- 
Paul Brunk, system administrator
Georgia Advanced Computing Resource Center (GACRC)
Enterprise IT Svcs, the University of Georgia

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: rebuilt HW RAID60 array; XFS filesystem looks bad now

[Eric Sandeen]

Hi Paul -

On 3/3/14, 3:05 PM, Paul Brunk wrote:
> Hi:

<snip story about very unhappy raid>

It sounds like things are badly scrambled.

> Is there anything I should try prior to xfs_repair -L?

The log failure is probably just the first error xfs encounters;
I imagine that after you zap the log, you'll find that everything
else is in equally bad shape.

You could _try_ an 

# xfs_metadump -o /dev/whatever fs.metadump
# xfs_mdrestore fs.metadump fs.img

and then try your various metadata salvage techniques on fs.img
(recreating as necessary with another xfs_mdrestore after failure)
as practice runs, but I'm guessing that xfs_metadump will fail
in equally spectacular ways.

Unless you can get your storage somehow reassembled properly, I doubt
that there's much xfs tools can do to help you.

-Eric

> I'm just trying to salvage whatever I can from this FS.  I'm aware it
> could be all gone.  Thanks.
> 

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: rebuilt HW RAID60 array; XFS filesystem looks bad now

[Dave Chinner]

On Mon, Mar 03, 2014 at 04:05:27PM -0500, Paul Brunk wrote:
> Hi:
> 
> Short version: XFS filesystem on HW RAID60 array.  Array has been
> multiply rebuilt due to drive insertions.  XFS filesystem damaged and
> trying to salvage what I can, and I want to make sure I have no option
> other than "xfs_repair -L".  Details follow.
> 
> # uname -a
> Linux rccstor7.local 2.6.32-431.5.1.el6.x86_64 #1 SMP Wed Feb 12
> 00:41:43 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux
> 
> xfs_repair version 3.1.1.  The box has one 4-core Opteron CPU and 8
> GB of RAM.
> 
> I have a 32TB HW RAID60 volume (Areca 1680 HW RAID) made of two RAID6
> raid sets.

Hmmm - yet another horror story from someone using an Areca HW RAID
controller. I'm starting to wonder if we should be putting an entry
in the FAQ saying "don't use Areca RAID controllers if you value
your data"...

[snip]

>  # mount /media/shares
>  mount: wrong fs type, bad option, bad superblock on /dev/mapper/vg0-lv0,
>         missing codepage or helper program, or other error
>         In some cases useful info is found in syslog - try
>         dmesg | tail  or so
> 
>  # dmesg|tail
>  XFS (dm-2): Mounting Filesystem
>  XFS (dm-2): Log inconsistent or not a log (last==0, first!=1)
>  XFS (dm-2): empty log check failed
>  XFS (dm-2): log mount/recovery failed: error 22
>  XFS (dm-2): log mount failed

That's bad. The log does not contain a valid header in it's first
block.

>  # xfs_repair -n /dev/dm-2
>  produced at least 7863 lines of output.   It begins
> 
>  Phase 1 - find and verify superblock...
>  Phase 2 - using internal log
>          - scan filesystem freespace and inode maps...
>  bad magic # 0xa04850d in btbno block 0/108
>  expected level 0 got 10510 in btbno block 0/108
>  bad btree nrecs (144, min=255, max=510) in btbno block 0/108

Corrupted freespace btree blocks.

>  block (0,80-80) multiply claimed by bno space tree, state - 2
>  block (0,108-108) multiply claimed by bno space tree, state - 7

with duplicate entries in them. That's not a good sign...

> 
>  # egrep -c "invalid start block" xfsrepair.out
>  2061
>  # egrep -c "multiply claimed by bno" xfsrepair.out
>  4753
> 
>  Included in the output are 381 occurrences of this pair of messages:
> 
>  bad starting inode # (0 (0x0 0x0)) in ino rec, skipping rec
>  badly aligned inode rec (starting inode = 0)

Ok, so the inode btree is also full of corrupt blocks.

> Is there anything I should try prior to xfs_repair -L?

Pray? Basically, then primary metadata in the filesystem that tracks
allocated space and inodes looks to be badly corrupted. If the
metadata is corrupted like this from the rebuild, then the rest of
the block device is likely to be busted up just as badly. So you
might be able to recover some of the filesysetm structure with XFS
repair, but all your data is going to be just as corrupted.

I'd be using metadump like Eric suggested to create a test image to
see what filesystem structure you'll end up with after running
repair. But with the corrupt AG btrees, there's a good chance even
metadump won't be able to run successfully on the filesystem. And
even that won't tell you how badly damaged the data is, just what
data you will have access to after running repair.

> I'm just trying to salvage whatever I can from this FS.  I'm aware it
> could be all gone.  Thanks.

Good luck :/

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs