On Wed, Jul 05, 2006 at 11:21:35AM -0700, Jason Thorpe wrote:
> 
> On Jul 5, 2006, at 4:23 AM, Manuel Bouyer wrote:
> 
> >No, I think this should be in fsck. When the system boots fsck is run
> >before filesystems are mounted read/write. There shouldn't be anything
> >in the kernel writing to the filesystem before we tell the system  
> >to mount it
> >read/write, even for recovering the journal. And if for whatever  
> >reason the
> >journal can't be recovered we certainly want fsck to check this  
> >filesystem
> >(or just fail, and stop the boot process).
> 
> No, it should not be in fsck.  What if you want to mount a file  
> system read-only?

If I mount a filesystem read-only I don't want anything to write to
disk, dammit ! not even journal recovery.

> You need a way to replay in-core with an on-disk  
> journal and on-disk-inconsistent metadata...

I'm not sure, as long as we keep the property that metadata writes
are done in an ordered way.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--