Re: [tlug] Permissions are like Rubik's cubes [SOLVED]

["Stephen J. Turnbull" <stephen@example.com>]

Dave M G writes:

 > I always thought the whole permissions deal was a matter of the 
 > operating system, not the disk file system.

Koan: so how are permissions preserved with removable media (across
mount points or even moving between hosts)?  There has to be data
regarding permissions associated with each file, and in fact it lives
in the file's directory.  It makes sense to keep that directory
information in specially marked (ie, the d bit that you see when doing
ls -l) files in the file system itself, you see?  (In most Unix file
systems, anyway, although even there the root directory needs special
treatment because it doesn't have a name until mounted!)

In the old days, many file system drivers didn't support UID/GID and
group/world permissions as mount options.  Instead, they took them
from the mount point.

Note that life once again gets complicated when you deal with
networked file systems.  NFS is designed to give as close to POSIX (==
Unixoid local file system) semantics as possible, but even it has
well-known problems.  AFS, Coda, Intermezzo, and others should be
considered suspect until you (or somebody you trust) determines what
works and what doesn't.  For example, Coda semantics allow operation
even though disconnected from the server, but cache consistency
requires that the whole file be in the cache.  This means that open(2)
can take many seconds or minutes for GB-size files like video.  This
can be a shock to naive programs and their users!