Coda File System

Re: Linux 2.6.x still suffering from NFS "stale" check in namei.c

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Thu, 15 Apr 2004 13:08:23 -0400
On Thu, Apr 15, 2004 at 06:25:34PM +0200, Ivan wrote:
> On Thu, Apr 15, 2004 at 09:59:51AM -0400, Jan Harkes wrote:
> > > in fs/namei.c and that removing that code eliminates some obscure problems
> > > with Coda.
> > 
> > That is very strange, because in 2.6 the ESTALE return only happens for
> > filesystems that have the FS_REVAL_DOT flag set (i.e. only NFS).
> 
> > I don't see how we can ever trip that code, because we don't set the
> > related flag.
> 
> - if (nd->dentry && nd->dentry->d_sb &&
> -                   (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
> 
> Hmm. You are right. Something is wrong. Possibly my interpretation :)

Maybe the ESTALE came from somewhere else.. What filesystem is used to
store the venus cache. I just did a quick grep in a 2.6.4 tree and found
that a couple of filesystems like to return their own ESTALE errors,

    afs, exportfs, fat, jfs, lockd, nfsd, nfs, reiserfs

(likely candidates here would probably be reiserfs or jfs)

> Sorry for pointing in a wrong direction, I'll keep you posted if I see
> the problem again.

That's ok, this just adds another datapoint to keep my eyes open for
possibly similar reports.

Jan
Received on 2004-04-15 13:09:50