[cifs-protocol] SMB1 -- proper client behavior when it does not hold an oplock
jlayton at samba.org
Fri Apr 27 12:34:20 MDT 2012
-----BEGIN PGP SIGNED MESSAGE-----
Sorry for the duplicate emails, but I sent this to the wrong dochelp
address before. Let me try again...
I'm hoping you can help clarify some points about proper SMB1 (and
maybe SMB2?) client behavior when it does not hold an oplock (at least
one that allows write caching).
My understanding has always been that when a client does not have an
oplock that allows write caching, then it should not cache any writes
- -- full stop. If an application does a write then the kernel should not
return until it has been sent to the server and the reply has come
back. That behavior is at least suggested in MS-CIFS, though it does
not come out and state that explicitly.
OTOH, Steve French suggested that that's not required by the protocol
and that clients are allowed to buffer up writes "briefly" in order to
allow the redirector to batch up small writes into a single request as
long as it flushes them out in a timely fashion. That seems a little
crazy to me, but I guess it's not the craziest thing in SMB1 if so...
So I guess my questions are:
1) What does the protocol actually mandate? Are you allowed to briefly
buffer up writes before returning to an application when the client
holds no oplock?
2) What does Windows actually do in this regard? If you are not allowed
to do that by the protocol, then does it follow this strictly or does it
do as Steve suggests and batch up small writes until it can fill a
Thanks for any info you can provide!
Jeff Layton <jlayton at samba.org>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.18 (GNU/Linux)
-----END PGP SIGNATURE-----
More information about the cifs-protocol