Re: Regression in Autofs, 2.6.15-git

[Ian Kent <raven@themaw.net>]

On Sat, 2006-01-14 at 17:25 +0200, P. Christeas wrote:
> On Saturday 14 January 2006 5:11 pm, P. Christeas wrote:
> > Doesn't that mean that mnt==0x0004 ? Clearly wrong. I can also see from
> > Christian's patch that mnt wasn't previously used, so it makes perfect
> > sense for that commit to introduce the oops.
> >
> > I guess the problem lies in autofs4_revalidate (fs/autofs4/root.c:420), the
> > nd->mnt value..
> >
> > I will add a silly validator (mnt>0xff) instead of (mnt) and see..
> Confirmed: ((u32)mnt>0xff) discards the invalid 'mnt' value and the oops 
> disappears.
> That is, the autofs4 code needs some debugging :( .

Yes. It's me again.

Could you try this patch please.

--- linux-2.6.15/fs/autofs4/root.c.dumb-nameidata	2006-01-15 01:01:26.000000000 +0800
+++ linux-2.6.15/fs/autofs4/root.c	2006-01-15 01:02:12.000000000 +0800
@@ -193,6 +193,8 @@ static int autofs4_dir_open(struct inode
 		if (!empty)
 			d_invalidate(dentry);
 
+		nd.dentry = dentry;
+		nd.mnt = mnt;
 		nd.flags = LOOKUP_DIRECTORY;
 		status = (dentry->d_op->d_revalidate)(dentry, &nd);