Re: XFS v 1.02 800GB Filesystem problem

[Seth Mos <knuffie@xs4all.nl>]

On Sat, 4 May 2002, Kevin Giguere wrote:

> Hello Folks,
>
> I have a filesystem that was acting strangely and I have not been able to
> pin point the true cause.
> The system was a 1ghz P3, 1 gb ram with external 800GB ide to scsi raid.
> The scsi HBA is an adaptec 29160 using the aic78xx driver.
>
> The  filesystem (/dev/sdb1) at its best point would allow me to mount it and
> I could view files, however this is all that I could do.  Any attempt to
> write a file would freeze the mount and the machine would have to be reset
> to free the mount point.  Another raid (/dev/sdc1)in the scsi chain remained
> fully usable.

Do you have any idea if some changed in the machine configuration during
that time or if you noticed other strange behaviour. Did you have a
systemcrash?

> I was able to do an xfs_check /dev/sdb1 and it went straight back to prompt.
> No apparent problems.  Restart. Remount. Same behavior; couldn't write
> anything...
>
> So I made sure the fs was unmounted and ran xfs_repair -n /dev/sdb1 and a
> whole bunch of  errors came up, but it didn't do anything b/c -n option.
> Spoke to my Irix guy asking what some of them meant.  We decided to run
> xfs_repair and try to fix some of these things.  This produced a message
> saying that the log is messed up and I would have to mount to rerun the log.
> The new problem was that I could not mount it anymore.  Shutdown checked the
> cables, everything was ok.  Other Raid was still fine.  Doesn't look like a
> HW problem ( Though I am convinced it is---  SCSI bios looks good,
> /proc/scsi/scsi sees everything, /proc/partitions sees everything..but
> HMMM).

Have you saved some of the xfs_repair output? The SGI people would
probably be able to tell what was wrong with the filesystem.

> After wanting to break the thing, I tried again. Same as results as above.
> Can't mount if I wanted to, can't run xfs_repair or check.  The only option
> to run xfs_repair (and where I am going with this) is xfs_repair -L
> /dev/sdb1.  I understood that this is a potentially dangerous option.  But I
> didn't have another 800GB volume free to dd onto as b/u.  So I ran this. And
> it apparently did a lot.  After this was run, I ran xfs_repair again to
> rebuild the log and then mounted.  The great thing is I could mount, read
> and write.  The terrible thing is that I couldn't see any data.  All that
> was there was an empty lost and found.....,  but xfs_check finds a ton of
> mismatches and lost inodes.

It sounds like it is toast. Maybe a newer xfs_repair will find some data
left to repair although I think there is not much left.

> As a last ditch I hooked it up to a SGI to see if it could do anything.
> From what I can tell all Irix sees is a raw disk.  I didn't do any fx stuff
> fearing ruining any other chances for recovery.

The linux XFS port can currently only read disks with blocksize ==
pagesize == 4K on ia32. So I am afraid that won't work easily.
The FAQ has better description of the caveats, it is possible though.

> I think that there is very very small chance to get something back, Any one
> have any Ideas????

I'm afraid not, maybe one of the developers has an idea to get something
back. The odds are against you.

Cheers
Seth