Re: [PATCH] lockd: don't allow locking on reexported NFSv2/3

[Jeff Layton <jlayton@kernel.org>]

On Fri, 2025-10-24 at 08:49 +1100, NeilBrown wrote:
> On Fri, 24 Oct 2025, Jeff Layton wrote:
> > Since commit 9254c8ae9b81 ("nfsd: disallow file locking and delegations
> > for NFSv4 reexport"), file locking when reexporting an NFS mount via
> > NFSv4 is expressly prohibited by nfsd. Do the same in lockd:
> > 
> > Add a new  nlmsvc_file_cannot_lock() helper that will test whether file
> > locking is allowed for a given file, and return nlm_lck_denied_nolocks
> > if it isn't.
> > 
> > Signed-off-by: Jeff Layton <jlayton@kernel.org>
> > ---
> > Regardless of how we fix the bug that Olga found recently, I think we
> > need to do this as well. We don't allow locking when reexporting via v4,
> > and I don't think we want to allow it when reexporting via v2/3 either.
> 
> I would like to see more justification.  The two locking protocols have
> substantial differences so it is not obvious that reasoning which
> applies to one also applies to the other.
> 
> What is the reason that we disable locking for v4?  If we could state
> that and justify that the same reasoning applies to v3, then certainly
> this would be an appropriate fix.
> 
> My guess is that it relates to handling restart of the forwarding server
> from the perspectives of both the ultimate client and the ultimate
> server.  Restart handling is quite different in the two protocols, but
> maybe they are equally unable to handle this situation?
> 

Yes, that's basically it. If the reexporting server crashes and
reboots, there is no mechanism to force the backend server into grace
and no coordination of their grace periods. That's true for all
versions of NFS.

Documentation/filesystems/nfs/reexport.rst has a section that talks
about reboot recovery which documents this, but we can certainly quote
that in the changelog.
-- 
Jeff Layton <jlayton@kernel.org>