2.6.0-test{1,2,3} kernels + software raid 0 + xfs broken?

2.6.0-test{1,2,3} kernels + software raid 0 + xfs broken?

[Florian Schanda]

Hi all,

I am running a 2.4.21 kernel + xfs + raid0 (software) on 2 scsi disks. It 
works without flaws.

Trying the same setup in 2.6.0-test{1,2,3} (haven't tested 4 yet) _will_ give 
data corruption if I write to the array.

2 Things that happen are:
* Creating a directory and then being unable to remove it: (rm -r foo says foo 
is not directory, rmdir foo says foo is not a directory, rm foo says foo is a 
directory, ls -l foo says it is a directory)
* writing files to the array, unmount array (works fine up here), mounting 
again utterly fails. (bad superblock, etc, etc) Doing an "xfs_repair -L" will 
kill a lot of data (the files modifies / written) but fix it. Rebooting into 
2.4 will fix it too (It doesn't even complain when mounting the thing.)

Anybody has an idea?

Thanks,

	- Florian

Re: 2.6.0-test{1,2,3} kernels + software raid 0 + xfs broken?

[Florian Schanda]

On Saturday 23 August 2003 19:16, Florian Schanda wrote:
> I am running a 2.4.21 kernel + xfs + raid0 (software) on 2 scsi disks. It
> works without flaws.
>
> Trying the same setup in 2.6.0-test{1,2,3} (haven't tested 4 yet) _will_
> give data corruption if I write to the array.

Looking at: 
ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/must-fix/must-fix-6.txt

> o RAID0 dies on strangely aligned BIOs

Is that my problem? I'm not sure I fully undestand this -- what is strangely 
aligned here?

	Florian Schanda

Re: 2.6.0-test{1,2,3,4} kernels + software raid 0 + xfs broken?

[Florian Schanda]

[-- Attachment #1: Type: text/plain, Size: 1937 bytes --]

On Sunday 24 August 2003 08:56, Neil Brown wrote:
> On Saturday August 23, ma1flfs@bath.ac.uk wrote:
> > Hi all,
> >
> > I am running a 2.4.21 kernel + xfs + raid0 (software) on 2 scsi disks. It
> > works without flaws.
> >
> > Trying the same setup in 2.6.0-test{1,2,3} (haven't tested 4 yet) _will_
> > give data corruption if I write to the array.
> >
> Can you test without raid0 and confirm that it works with xfs on a
> single drive?

I have now also tested with -test4.
It works, in all of 2.6.0{1,2,3,4} and 2.4.21-xfs.

> Are there any kernel messages?

Only this: (from /var/log/kern.log)

Aug 27 17:57:02 omoikane kernel: XFS mounting filesystem md0
Aug 27 17:57:03 omoikane kernel: Ending clean XFS mount for filesystem: md0
Aug 27 17:58:22 omoikane kernel: XFS mounting filesystem md0
Aug 27 17:58:22 omoikane kernel: Starting XFS recovery on filesystem: md0 
(dev: 9/0)
Aug 27 17:58:22 omoikane kernel: Ending XFS recovery on filesystem: md0 (dev: 
9/0)
Aug 27 17:59:27 omoikane kernel: XFS mounting filesystem md0
Aug 27 17:59:27 omoikane kernel: Ending clean XFS mount for filesystem: md0
Aug 27 18:00:17 omoikane kernel: XFS mounting filesystem md0
Aug 27 18:00:17 omoikane kernel: Starting XFS recovery on filesystem: md0 
(dev: 9/0)
Aug 27 18:00:18 omoikane kernel: Ending XFS recovery on filesystem: md0 (dev: 
9/0)

What I did was (4 times):

1) mount /dev/md0 on /data
2) make a dir called testdir
3) copy some random files in there (about 20-500) MB each
4) umount /data
5) mount /data

In this particular case it was able to recover (but it really shouldn't... 
since it was cleanly unmounted), but If I do lots of stuff it usually can't 
recover. Usually the only way out is xfs_repair -L /dev/md0

Note that the very first initial mount (mounting the array as it was left by 
2.4.21) did not generate any messages and succeded flawlessly.

I've attached my 2.4.6 config below - just in case.

Thanks,

	Florian

[-- Attachment #2: config.bz2 --]
[-- Type: application/x-bzip2, Size: 5181 bytes --]

Re: 2.6.0-test{1,2,3} kernels + software raid 0 + xfs broken?

[Neil Brown]

On Wednesday August 27, ma1flfs@bath.ac.uk wrote:
> On Saturday 23 August 2003 19:16, Florian Schanda wrote:
> > I am running a 2.4.21 kernel + xfs + raid0 (software) on 2 scsi disks. It
> > works without flaws.
> >
> > Trying the same setup in 2.6.0-test{1,2,3} (haven't tested 4 yet) _will_
> > give data corruption if I write to the array.
> 
> Looking at: 
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/must-fix/must-fix-6.txt
> 
> > o RAID0 dies on strangely aligned BIOs
> 
> Is that my problem? I'm not sure I fully undestand this -- what is strangely 
> aligned here?

This has been fixed, though possibly it was in test4.
In any case, if this were your problem you would be seeing raid0
related messages in your kernel logs.

NeilBrown