Broken reiser4 FS

Broken reiser4 FS

[Ari Sovijärvi]

Hi.

I tested the reiser4 shipped with mm-patch, and I managed to corrupt the 
filesystem after the first unmount. See the URL for details:
http://apz.fi/apz/reiser4.txt

Oh, and I'm not on the list, so I'd appreciate if you'd cc your replies. Thanks.

-- 
Ari Sovijärvi
http://apz.fi/

Re: Broken reiser4 FS

[Christophe Saout]

[-- Attachment #1: Type: text/plain, Size: 1205 bytes --]

Am Mittwoch, den 25.08.2004, 14:21 +0300 schrieb Ari Sovijärvi:

> I tested the reiser4 shipped with mm-patch, and I managed to corrupt the 
> filesystem after the first unmount. See the URL for details:
> http://apz.fi/apz/reiser4.txt
>
> CHECKING SEMANTiC TREE
> FSCK: Directory [2a:6d703300000000:10001] (dir40), node [2281606], item [0], 
> unit [4]: entry has wrong offset 
> [10001:0(NAME):1446f6e2048756f:6e6f74202d204879:d5b58f25a1408e04]. Should be 
> [10001:0(NAME):1446f6e2048756f:6e6f74202d204879:6e433d19ec834164]. 

Just FYI:

I've also seen a bunch of these on my notebook. I'm ruling out memory
problems, my hardware is reliable.

fsck then removed the directories. I'm not exactly sure which directory
was damaged, but it seems the directoriy were a temporary one (?), I
found some library files in lost+found but the ones in /usr/lib were
still there.

My notebook is sometimes low on memory and swaps a lot while I'm doing
some disk I/O. This is perhaps increasing the chance of getting this
corruption, I'm not sure.

I panicked and didn't create a metadata dump or something, I justed
wanted it fixed before the damage starts to spread. Sorry for that.


[-- Attachment #2: Dies ist ein digital signierter Nachrichtenteil --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Broken reiser4 FS

[mjt]

On Wed, Aug 25, 2004 at 02:22:31PM +0200, Christophe Saout wrote:
>
>I've also seen a bunch of these on my notebook. I'm ruling out memory
>problems, my hardware is reliable.

Wasn't there some discussion that there may be some issues with Reiser4
still committing journals during umount which leads to data loss in -mm4?

Or was it so that this was fixed in -mm4?

That should not destroy the directory plugin type, though.

>My notebook is sometimes low on memory and swaps a lot while I'm doing
>some disk I/O. This is perhaps increasing the chance of getting this
>corruption, I'm not sure.

That might be consistent with something about files being in swap during
umount and thus not commited?

I don't know if this is a shot in the dark or clutching at straws ;)

Any Namesys guys around? :)

-- 
mjt