[linux-cifs-client] CIFS Mount and the old PCFS Mount (was re: CIFS Unix Extensions UIDs and client permission)

[Steven French]

> I'm not sure the semantics of the CIFS mount
> are all that sane...

I am not so sure either, but now is a reasonable time to make the mount
options make more sense to Linux users.    I would like the mount semantics
to mimic local linux filesystem behavior better (and more closely mimic
posix semantics where the tradeoffs are reasonable) rather than simply
follow nfs or smbfs behavior - although much of this has been debated in
nfs before.

Some of the decisions on behavior and mount defaults have been around since
smbfs (or earlier) and predate the CIFS Unix extensions.

Suggestions would be welcome.

If I can come up with a decent name for the alternative mount behaviors I
can code them easily enough and the defaults can be debated.

Two underlying issues are challenges to implement (presumably tough for NFS
too but for slightly different reasons):
1) redundant permission checks, once on client and once on the server (in a
perfect world they should only be checked at the server since the mode or
acls could change at any time and the client check is at best redundant)
2) smb header uids don't trivially map to unix uids which don't easily map
to rids, and it is not clear whether it would work in general to use the
same tid but with different uids sent on path based ops (to access the same
share with different users without resending tconx).

Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at-sign us dot ibm dot com