Re: Crash...

[Ryusuke Konishi <ryusuke-sG5X7nlA6pw@public.gmane.org>]

On Fri, 24 Jul 2009 10:58:03 +0200, Reinoud Zandijk <reinoud-S783fYmB3Ccdnm+yROfE0A@public.gmane.org> wrote:
> On Thu, Jul 23, 2009 at 11:02:47PM +0200, Andrea Gelmini wrote:
> > 2009/7/23 Ryusuke Konishi <ryusuke-sG5X7nlA6pw@public.gmane.org>:
> > > It also suggests that an inconsistent state in page cache of B-tree
> > > nodes hit the function; the function found a dirty page, but the page
> > > didn't have buffer heads which was supposed to be impossible for the
> > > b-tree of nilfs.
> >    -> mkfs.nilfs2 -b 1024 /dev/mapper/VG-NilfHome (maybe the problem
> > is the 1K block size?)
> 
> 1024 is lower than the page size so that might explain a lot!

Yes, I suspect this in fact.  The small size block changes several
code paths.

> I think its a missing check in the mkfs.nilfs2 to never allow lower
> values than the page size for block size.... but Ryusuke can better
> answer that :-D
>
> With regards,
> Reinoud
 
We have implemented nilfs to allow smaller block sizes in order to
support some devices such like DVD/CD-ROM or flash based ones.

So, I count it a bug which we should fix.

Thanks,
Ryusuke Konishi