Re: kernel oops: trying to mount a corrupted xfs partition (2.6.16-rc3)

[Sonny Rao <sonny@burdell.org>]

On Sun, Feb 19, 2006 at 04:52:09PM -0500, Dave Jones wrote:
<snip> 
> Just for kicks, I just hacked this up..
> 
> #!/bin/bash
> wget http://www.digitaldwarf.be/products/mangle.c
> gcc mangle.c -o mangle
> 
> dd if=/dev/zero of=data.img count=70000
> 
> while [ 1 ];
> do
>         mkfs.xfs -f data.img >/dev/null
> 		./mangle data.img $RANDOM
>         sudo mount -t xfs data.img mntpt -o loop
>         sudo ls -R mntpt
>         sudo umount mntpt
> done

Cool script, you might want to multiply $RANDOM by some factor (I used
8) to catch some more stuff, I know JFS, for example, doesn't put
anything in the first 32k, so the first time I ran it on JFS it did
nothing ;-) 


Reiserfs folks, 

I also found an infinte loop in Reiserfs on 2.6.15, if the Reiser
folks are interested, I've gziped the fs and put it here:

http://burdell.org/~sonny/data.img.breaks.reiserfs.gz

The fs is only 52k when zipped, so its not too bad to download.

This is under stock 2.6.15, sorry I can't post dmesg output because I
end up having to reboot when it happens and don't have time to debug
right now.  It looks like it's in the journal replay code where it
keeps trying to grab some block with a ridiculously large offset. 


> 
> xfs wins the award for 'noisiest fs in the face of corruption' :-)
> After a few dozen backtraces from xfs_corruption_error,
> this fell out...
> 
> divide error: 0000 [1] SMP
<snip trace>
 
> (The kernel is based on 2.6.16rc4)

I see a similar breakage (divide error) on x86 using 2.6.15

Sonny