Oops in NFSv4 server in 2.6.23.17

[J. Bruce Fields]

On Thu, Mar 13, 2008 at 03:36:31PM +0100, Lukas Hejtmanek wrote:
> Hello,
> 
> On Wed, Mar 12, 2008 at 12:00:07PM -0400, J. Bruce Fields wrote:
> > In this case it looks like something in the gss context cache is bad.
> > 
> > I'm not sure what's going on.
> 
> I discovered what triggers this bug. If I have default export with secure
> option and I try to connect from insecure port, it produces the oops.
> If I add insecure option to the export, no oops at all and mount succeeds.
> 
> Also, the bug is not triggered by the first attempt but by the second one.
> 
> So I guess that from the first attempt there is something inserted into gss
> context cache - error occurs (insecure attempt with the secure option), gss
> context is removed from the cache as invalid but not completely. And the
> second attempt does oops. Sounds reasonable for me. 
> 
> Hope this helps to find the bug.

That's an excellent clue, thanks.  I wonder it would be explainable by a
reference count imbalance of some kind in fh_verify?  I don't quite see
it yet, and unfortunately can't yet reproduce the bug here.

--b.