pNFS consistency model

I have been trying to find out what the consistency model used by
pNFS is, and have been failing dismally - I can't even find any
description for NFSv4, despite Web references to it.

Note that I am NOT talking about the file system corruption issue,
but about what is and is not defined when you get race conditions
between two different clients. With NFSv3, it was easy to get the
metadata out of step with the contents, even for a single client,
but NFSv4 at least allows a client to avoid that. I can't see that
it requires clients to do anything sanely, and it specifically says
that COMPOUND doesn't imply atomic.

Now, this is a MAJOR matter for pNFS, because the direct route
from the client to the storage brings back the problems of NFSv3,
except in a much worse form. Note that locking is NOT a solution,
as the question is what pNFS specifies - not how to program round
its lack of specification.

Assuming that I have failed to find such a description, can anyone
point me in the right direction? Just the working document name
would be enough, but the section would be useful.


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:
Tel.: +44 1223 334761 Fax: +44 1223 334679