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

Linas Vepstas wrote:
> 
> Hi,
> 
> I've been experimenting with automatic bus error recovery in the
> 2.6.11 kernel.  During one of my failed experiments, I tripped over
> a Reiserfs bug, below.  Basically, my error recovery failed, which
> means a SCSI disk went permanently offline, which, admitedly,
> is pretty catastrophic, but shouldn't be a kernel panic.  It seems
> that reiser hits a 'BUG_ON' in this case.
> 
> FWIW, in my limited experience with ext3 in the same exact situation, 
> it seems that ext3 handles this gracefully, returning -EIO to all 
> affected apps accessing the disk.
> 
> Unfortunately, I don't know how to tell you how to reproduce this :)

Hi Linas -

Finally getting a chance to look into this one a little bit more. What
were your test cases? I've seen this bug before, and in doing a quick
trace, it may be possible to hit this if you attempt to write to a file
with a hole and an I/O error occurs while flushing that buffer. When
map_block_for_writepage calls journal_end and it fails, it can still
call reiserfs_get_block for the hole even though the journal has been
aborted. If the buffer is !uptodate (due to the i/o error failure),
you'll hit that BUG.

I'll continue to track this one down, but any more info you can provide
on your test environment would be helpful.

Thanks.

- -Jeff

- --
Jeff Mahoney
SuSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (GNU/Linux)

iD8DBQFCXW7ELPWxlyuTD7IRAh4jAJ4zB4eMUxoZjhnaOkoSDZ/yDHMtDACggkXi
y1ESZm40aGqJ0S2SfGLgBBQ=
=yYCv
-----END PGP SIGNATURE-----

Reply via email to