On 10/07/2024 17:45, Chuck Lever III wrote:
On Jul 10, 2024, at 3:11 AM, Sagi Grimberg <sagi@xxxxxxxxxxx> wrote:
Yes, as an NFSD co-maintainer, I would like to see the
READ stateid issue addressed. We just got distracted
by other things in the meantime.
OK, so reading the correspondence from the last time, it seems that
the breakage was the usage of anon stateid on a read. The spec says that
the client should use a stateid associated with a open/deleg to avoid
self-recall, but allowed to use the anon stateid.
I think that Dai's patch is a good starting point but needs to add handling of
the anon stateid case. The server should check if the client holds a delegation,
if so simply allow, if another client holds a deleg, it should recall?
For an anon stateid, NFSD might just always recall if
there is a delegation on that file. The use of anon is
kind of a legacy behavior, IIUC, so no need to go to a
lot of trouble to make it optimal.
(This is my starting position; I'm open to be convinced
NFSD should take more pain for this use case).
Hey Chuck, didn't forget about this.
I'll look into this when I find some time (which I lack these days).
