RE: recovery from NFS4ERR_FHEXPIRED

New Message Reply About this list Date view Thread view Subject view Author view Attachment view

From: Noveck, Dave (dave.noveck@netapp.com)
Date: 09/27/99-09:14:47 AM Z


Message-ID: <7F608EC0BDE6D111B53A00805FA7F7DA05D10C51@TAHOE>
From: "Noveck, Dave" <dave.noveck@netapp.com>
Subject: RE: recovery from NFS4ERR_FHEXPIRED
Date: Mon, 27 Sep 1999 07:14:47 -0700

 
> This is a good answer. It follows the client-side
> implementations need to be prepared for NFS4ERR_GRACE
> from all ops, not just OPEN or LOCK.

I'm not sure about *all* ops.  Should it be valid
to return NFS4ERR_GRACE for PUTFH or SETCLIENTID,
for example?  The list should be definitely expanded.
It's not clear whether we just have a small set
for which NFS4ERR_GRACE is allowed (OPEN, LOCK,
REMOVE, RENAME) or allow it for everything except
some other set of exceptions.

> 
> Is it safe to say the pathname is the only
> possible recovery for volatile file handles?
> Only repeated LOOKUPs and OPENs can recover
> expired handles? I believe so. Comments?
> 
> The NFSv4 spec should at least *RECOMMEND*
> that servers which return volatile file handles
> defend the file name space and also use
> the trans-reboot grace period as Carl
> described. The spec should also *REQUIRE*
> client implementations to handle NFS4ERR_GRACE
> for all operations.

The client should certainly be required to
handle anything that we recommend that servers
do. 

> A set of *RECOMMENDED*
> heuristics for the client-side to defend
> file content would also be a good topic
> for the spec, but since file identity is
> problematic, such heuristics should not be
> required of either the client nor server.

If a server is going to take an open file and
all of a sudden connect it to a different object
then I'm not sure what heuristics a client could
use to defend it's files other than to not use
such a server.

> 
> The whole mechanism begs the question of a lost
> CLOSE. There are similar issues for client-side
> caching, and I would urge that the same
> mechanism be used in this case.

I'm confused.  What mechanism begs what question
regarding a lost CLOSE? What are the similar
issues for client-side caching and what mechanism
needs to be used in that case?

> 
> OBTW, SMB ain't got a solution for the
> client/client trans-reboot race condition.
> Chalk one up for NFSv4. I love it when
> that happens. Nice going, Carl.
> 
 


New Message Reply About this list Date view Thread view Subject view Author view Attachment view

This archive was generated by hypermail 2.1.2 : 03/04/05-01:47:39 AM Z CST