UIDs and GIDs

[Dominic Dunlop]

From:  Dominic Dunlop <domo at tsa.co.uk>

In article <743 at longway.TIC.COM> Mark Brown (mbrown at osf.org) writes:
>In 1003.1, "User ID" is defined as a positive integer (so is GID)...
>
>Also, uid_t is defined as an arithmetic type (same for gid_t).
>
>How does one handle (or can one handle) certain networking conventions that
>use a "dummy" user ("nobody") and require a user id of -2 ?
>
>Do these conflict as they seem, or am I missing something (always possible..)

No, you're spotting something.  Yes, this is a known conflict between
``certain networking conventions'' and POSIX.1.  My guess is that it falls
to POSIX.8 (transparent file access) to unwind.  As POSIX.8 is now defining
two styles of remote file access -- full POSIX.1 semantics (namely better
than ``certain networking conventions''), and highly curtailed semantics
(considerably less than ``certain networking conventions''), one option at
its disposal is to let negative user id's fall down the crack (gulf?)
between the two styles.  An alternative is to weasel out of the conflict by
saying that accesses to remote files by unrecognised users map onto some
unique, unprivileged uid without actually admitting that the uid might be
negative.  Or that they map onto UID_MAX - 1 (except that POSIX.1 does not
have a UID_MAX because uid_t is allowed to be a magic cookie -- albeit a
magic cookie of arithmetic type).  (Incidentally, ISO's central secretariat
has, not ureasonably, asked us for a definition of ``magic cookie''.
Suggestions?)
-- 
Dominic Dunlop

Volume-Number: Volume 20, Number 62