2.6.19: File system corruption "stuck" until device is replugged

2.6.19: File system corruption "stuck" until device is replugged

[Andrey Borzenkov]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I had USB stick (fat32) that reported file system corruption on mount and 
hence was mounted read-only. No amount of umount/dosfsck/mount could make it 
rw again. dosfsck reported device as clean but it still would mount ro and I 
continued to see directory that had been deleted by the very first dosfsck 
run! I unplugged it, looked under Win2k - it was OK - and only then did I 
notice that directory claimed as corrupted did not even exist. Replugging 
it - mounted OK.

I am not sure if this is a bug or "work as designed". May be this is specific 
fat32 problem; still it does not look right?

TIA

- -andrey
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)

iD8DBQFFi2YfR6LMutpd94wRAquCAKC3n8DjRGRqDYdfP6tNGvlg5sG0MQCfQRNJ
89HQuNaAWuLzJkkKayVrLks=
=m0rH
-----END PGP SIGNATURE-----

Re: 2.6.19: File system corruption "stuck" until device is replugged

[Sven-Haegar Koch]

On Fri, 22 Dec 2006, Andrey Borzenkov wrote:

> I had USB stick (fat32) that reported file system corruption on mount and
> hence was mounted read-only. No amount of umount/dosfsck/mount could make it
> rw again. dosfsck reported device as clean but it still would mount ro and I
> continued to see directory that had been deleted by the very first dosfsck
> run! I unplugged it, looked under Win2k - it was OK - and only then did I
> notice that directory claimed as corrupted did not even exist. Replugging
> it - mounted OK.
>
> I am not sure if this is a bug or "work as designed". May be this is specific
> fat32 problem; still it does not look right?

I think at the first mount time there was a read error for an unknown 
reason, and this turned the partition (not the filesystem) into read-only 
mode. All further mount attempts only found a read-only medium, and thus 
where only able to mount read-only. I've been bitten by this more than 
once, too - there seems to be no way to avoid a reboot/replug when one of 
your data disks has thrown a read error, taking (for me) the whole server 
down, when just one unimportant data-disk thinks it has problems again (as 
happens with my home-server about once every one or two weeks).

c'ya
sven

-- 

The Internet treats censorship as a routing problem, and routes around it.
(John Gilmore on http://www.cygnus.com/~gnu/)