Re: XFS corruption on SoftRAID5

[Seth Mos <knuffie@xxxxxxxxx>]

At 10:29 29-6-2001 +0200, Simon Matter wrote:
> Seth Mos schrieb:
>
> I'm not complaining.

Your not. The coffee here must have been a bit on the strong side.

> > Maybe a bit harsh but the md author might just be listening on the
> > linux-kernel list.
>
> Until today, it seemed to be XFS related.

Oh. I thought you noticed it earlier.

> > The people here understand XFS all too well but they don't know the
> > complete kernel in and out (could be wrong though). Another problem is that
> > they unfortunately don't really have the time to fix all sorts of 
> kernel bugs.
> >
>
> You're right. But on this list we have all those people using big disks
> and raid volumes. So if the problem was somehow XFS/SoftRAID related,
> where could I ask.

True, but a lot of them are using hardware raid either IDE or scsi or fibre 
based.

> > If you can produce a testcase in which you can generate corruption on the
> > fs no matter what the fs is that would be helpful. Are you just seeing file
> > names being garbled or ar the files themeselves also corrupt. What does a
> > xfs_repair mention when you try to check it? Does it even report anything
> > on that matter at all or does it decide to core dump because it's checking
> > swiss cheese?
>
> It's the filnames and the files themselve. The hole blockdevice seems to
> be corrupted.
> Its not XFS,not SoftRAID.
> Its something in the IDE subsystem.

What IDE controller was it? A promise I believe? I unfortunately don't have 
experience with those controllers except for a Promise Ultra66 controller.
You don't happen to have another IDE controller to test it with do you :)

Do you also see a certain pattern in the fs corruption or is it just 
/dev/random ?

> > Can you check out the CVS tree and build a kernel with that to simulate it.
> > 2.4.5+ makes a big difference relative to 2.4.3. There have been some raid
> > fixes in the past time. And 2.4.6 is approaching in a rapid pace.
> >
> > I'm placing my bet on the next version being 2.4.6.
> >
> > If you build a new kernel with the CVS tree (currently at 2.4.6-pre6) and
> > can test if you see corruption again that would be helpful. Then we at
> > least now what issues remain for the 1.0.1 installer. Although shipping a
> > 2.4.5 in 1.0.1 might not be possible.
>
> Just tried rawhide 2.4.5-20010613 and it's exactly the same.

Crap, so much for my theory. Oh well.

Bye
--
Seth
Every program has two purposes one for which
it was written and another for which it wasn't
I use the last kind.