We were needlessly revoking cache state (sending an "oplock break") in a case where a new client open had access rights that only allowed attribute changes. We're not supposed to revoke cache delegations in such a case.
There was also some poorly ordered logic in smb_open_subr() which we fixed while looking at this.

The general order of things should be:  (a) access checks, (b) oplock breaking, (c) share modes, (d) actually create etc.
Author: Alek Pinchuk

Reviewers:

  Gordon Ross

  Kevin Crowe

Samba "smbtorture" oplock tests.

Field tested for quite some time.