(was slashes, now NFS devices)

[Robert Thurlow]

In <1991Mar9.025601.18479 at panix.uucp> zink at panix.uucp (David Zink) writes:

>thurlow at convex.com (Robert Thurlow) Chips in:
>>The locking protocol is separate from the NFS protocol.  The only thing

>But nobody in the previous conversation said 'protocol'.  We were all
>talking about NFS the method of remote mounting filesystems.  If NFS
>the stateless protocol needs locking to emulate real fileystems effectively
>then locking is a part of NFS.

Are you really this uninformed, or do you have someone helping you?

BSD 4.3 has the flock(2) system call for locking entire files, period.
On Suns, this works only on local filesystems.  Until Sun gave their
lock manager to the world, there was no way to do /usr/group style
record locking via lockf(3)/fcntl(2) at all.  As well, the lockf()
and flock() calls on current Sun-compatible systems do not interact in
any way; two local processes can blithely acquire each lock without
being aware of each other.  lockf() and the lock manager are "enhanced
functionality" (of questionable value as currently implemented);  they
do not in any way "complete" the semantics of the NFS file system, as
there is no precedent for lockf() working on local file systems.

>Trying ordering NFS from a vendor.  See whether they send you a protocol
>or an implementation.

Of course vendors ship implementations.  I contend that there are a
lot of good implementations of NFS out there, based on testing at
Connectathon.  If you ever actually *use* NFS sometime, you might
come to the same conclusion.

Rob T
--
Rob Thurlow, thurlow at convex.com
An employee and not a spokesman for Convex Computer Corp., Dallas, TX