Re: [PATCH] NFSD: TEST_STATEID should not return NFS4ERR_STALE_STATEID

["J. Bruce Fields" <bfields@fieldses.org>]

On Sun, May 27, 2012 at 04:30:24PM -0400, Chuck Lever wrote:
> Hi-
> 
> On May 25, 2012, at 7:30 PM, Chuck Lever wrote:
> 
> > On May 25, 2012, at 9:46 AM, J. Bruce Fields wrote:
> > 
> >> If you think it's worth checking for those buggy client cases, we could
> >> instaed check that stateid->si_opaque.so_clid and cl->clientid agree.
> >> That'd cover the special-stateid checks too.
> > 
> > That's interesting.  I can redrive this patch with this change.
> 
> Doesn't a positive find_stateid(cl, stateid) result guarantee that so_clid and clientid are the same?

No, it only looks at the si_opaque.so_id, and assumes this stateid
belongs to the given client.  The passed-in client was actually
extracted from the sequence at the start of the compound:

	nfsd4_sequence(...)
	{
		...
		session = find_in_sessionid_hashtbl(&seq->sessionid);
		...
		cstate->session = session;
	}

	nfsd4_free_stateid(...)
	{
		struct nfs4_client *cl = cstate->session->se_client;
		...
			... nfs4_validate_stateid(cl, &stateid->ts_id_stateid);
	}

So in theory a confused client sent a free_stateid referencing a stateid
that belonged to a client other than the one that the session belonged
to, and then we'd see this inconsistency.

--b.