Re: ext3 vs resiserfs vs xfs

[James A Sutherland]

On Wednesday 07 November 2001 9:11 pm, Andreas Dilger wrote:
> On Nov 07, 2001  20:44 +0000, James A Sutherland wrote:
> > On Wednesday 07 November 2001 7:38 pm, Ville Herva wrote:
> > > On Wed, Nov 07, 2001 at 04:31:24PM +0000, you [James A Sutherland] 
claimed:
> > > > Hm.. after a decidedly unclean shutdown, I decided to force an fsck
> > > > here and my ext3 partition DID have two inode errors on fsck...
> > > > (Having said that, the last entry in syslog was from the SCSI driver,
> > > > and ext3's journalling probably doesn't help much when the disk it's
> > > > on goes AWOL...)
> > >
> > > A stupid question: does ext3 replay the journal before fsck? If not,
> > > the inode errors would be expected...
> >
> > Yes, it does: this was AFTER the journal replay. And yes, it was ext3 not
> > ext2 mounting it (well, either that or ext2 has learned to do journal
> > replays...).
>
> Actuall, e2fsck can also do the journal replay, so depending on whether
> this is the root fs or not, it may be that you get a journal replay and
> still mount it as ext2...

The journal replay occurred on mount, well before fsck was invoked.

> > So, AFTER a journal replay, there were still two damaged inodes
> > - which sounds like Anton's problem. Maybe ext3 just hates Cambridge? :-)
>
> Well, if you had a SCSI error, then it may be that the fs marked an error
> in the superblock, which would force a full fsck also.
>
> Note also, that it is often normal to have "orphaned inodes" cleaned up
> when the journal is cleaned up.  This is not an error.  I normally have
> these on my system because of PCMCIA cardmgr creating device inodes in /tmp
> and then unlinking them immediately after opening them.

They were not orphaned inodes, they were inodes with incorrect size & block 
values...

> If you have an open but unlinked file, then ext3 will delete this file at
> mount/fsck time (unlike reiserfs which leaves it around wasting space).
> Did you actually get files in lost+found, or only the orphaned inode
> message?

Nothing in l&f, just the familiar (from ext2!) scenario of automatic fsck 
finding errors, then dropping me to a single-user login to run fsck manually.

James.