[olug] share a folder rw, but not deletable?

[Matthew G. Marsh]

On Thu, 27 Mar 2003, William E. Kempf wrote:

>
> Jay Hannah said:
> >
> > Brian Wiese wrote:
> >> |A user can delete a directory only if they have write permissions to
> >> the |directory above the directory in question.
> >>
> >> Great, this... 'will work' and I can make it happen, thanks!  I'm
> >> really starting to realize just how limited a filesystems is without
> >> mandatory and discretionary access control lists.  It's just an odd
> >> thing to keep in mind (of several I've discovered) difference between
> >> Unix and Windows NTFS file systems... where you can give a user
> >> "delete" permissions.
> >
> > Limited? How so? What can't you do? Once you spend a half hour learning
> > it, Unixy file permissioning is easy. I've never bothered to care about
> > WinX file permissioning. (I am not an MS admin.)
>
> Apply more advanced security.  For example, I administer multiple CVS
> repositories for a few development teams.  We would like to have a
> repository secured such that: Team A has full access; Other Teams have
> read only access; Everyone else has no access.

Linux (or any UNIX) security is not relevant here. Instead go look for the
ACL version of CVS that allows you to specify ACLs within the context of
the CVS. CVS was not designed for multiuser security however there are
many excellent modifications adding both ACLs as well as SSL connection
structures to CVS. BTW - the ACL version of CVS does not require that the
client be aware of the ACL...

>  But Linux security
> (without ACLs) doesn't allow this granularity of security, since only a
> single user and a single group is allowed access.  (I do know work
> arounds, after doing a lot of research, so I'm not asking for help on this
> particular security case.  Just pointing out that the standard file
> permission scheme used in Linux isn't up to the task.)

Never rely on a filesystem to secure a program ;-} Just ask your friendly
local NTFS guru what the DATA fork means...

Seriously - any network capable program that can be remotely used for file
system access should (personally I lean to must) provide security
structures (ACL or whatever) to limit and structure user access. The file
system cannot know who the users are in many cases (consider non-local
users - ie not in the passwd/group type files) and thus should never be
required to enforce permissions on such users.

> Yep... Linux security is pretty easy to deal with.  But with out ACLs it's
> not very flexible.

User/Group/Permission = ACL (filesystem)  - sorry for the hard fact ;-}

An ACL merely is a permission list that specifies what a user may access.
That is exactly what CHMOD/CHOWN/GHGRP and the related utilities were
designed to do for the set of local users the box has control over. BTW -
standard CVS does not pay much attention to this.

This is a very good discussion - pity I just started reading through it
(from Florida...)

> --
> William E. Kempf

--------------------------------------------------
Matthew G. Marsh,  President
Paktronix Systems LLC
1506 North 59th Street
Omaha  NE  68104
Phone: (402) 553-2288
Email: mgm at midwestlinux.com
WWW:  http://www.midwestlinux.com
--------------------------------------------------