Re: [nfsv4] Should client issue close in stale fh?

[Trond Myklebust <[e-mail]>]

On Thu, 2018-08-23 at 16:38 +0200, Mkrtchyan, Tigran wrote:
> 
> Hi NFSv4 gurus,
> 
> we have a following situation. A file, opened by
> client-A get delete by client-B before LAYOUTGET
> is called:
> 
> 
> client-A -> server : OPEN(file)
> server -> client-A : fh
> 
> client-B -> server : REMOVE(file)
> server -> client-A : OK
> 
> client-A -> server : LAYOUTGET(fh)
> server -> client-A : ERR_STALE
> 
> 
> Client-A never sends close. The spec is not
> clear about it. Well, I understand, that probably
> if file is opened, then file handle must still be
> valid, but (a) this is hard to implement for distributed
> systems and (b) NFS4ERR_STALE is valid reply to
> LAYOUTGET.

It is also a valid and indeed expected reply to PUTFH+CLOSE in that
situation, so why should the client not be allowed to optimise it away?

If the server doesn't want that to happen, then it has the option of
not actually deleting the file when it is unlinked above, and instead
waiting until the client chooses to close the file.

Cheers
  Trond