Re: [nfsv4] [FedFS] Meeting Minutes, 9/30/2010

[James Lentini <[e-mail]>]

On Fri, 1 Oct 2010, Trond Myklebust wrote:

> On Fri, 2010-10-01 at 09:50 -0400, James Lentini wrote:
> > 
> > On Thu, 30 Sep 2010, Trond Myklebust wrote:
> > 
> > > On Thu, 2010-09-30 at 17:33 -0400, James Lentini wrote:
> > > > 
> > > > On Thu, 30 Sep 2010, Chuck Lever wrote:
> > > > > Some convention for the NFS server's reply in this case should be 
> > > > > agreed upon outside of the ongoing FedFS discussion.
> > > > 
> > > > I think another way to phrase this question is: How does an NFS server 
> > > > indicate that a file system has moved to an unknown location? One 
> > > > obvious possibility is to return a zero element locations array in the 
> > > > fs_locations structure.
> > > 
> > > NFS4ERR_STALE is the usual response. If you don't know where the 
> > > filesystem was migrated to, then what you have is not a migration 
> > > event. It is an unexport of the partition.
> > 
> > Suppose the server waits to resolve the junction until a GETATTR that 
> > includes fs_locations[_info]. If the server has already responded with 
> > an NFS4ERR_MOVED, will an NFS4ERR_STALE return for the GETATTR be a 
> > reasonable response to the client?
> 
> I'm not sure I understand. If the location of the filesystem is unknown,
> then why would the server reply with NFS4ERR_MOVED in the first place?

I think there are a few scenarios where this might happen. 

Suppose the client issues the operations in Section 11.8.1 of RFC 
5661. The server will respond to the GETFH with an NFS4ERR_MOVED. A 
server could delay resolving the junction at /this/is/the/path until 
the client sends a GETATTR for /this/is/the/path that includes 
fs_locations[_info] (at which point the resolution might fail) or the 
server might have a cached value for the junction that expires between 
receipt of the GETFH and GETATTR (and the new resolution might fail).

> That said, I believe that the answer to your question is 'yes'. See 
> for instance the discussion in section 4.2.2 in RFC3530.

Thanks. I think we have our answer. The server should return 
NFS4ERR_STALE in these instances.