Security models (was: [lug] KDE ...)

[D. Stimits]

rm at fabula.de wrote:
> 
> On Tue, Jan 08, 2002 at 03:53:30PM -0700, D. Stimits wrote:
> >
> > Some filesystems support permissions beyond the usual user/group/other.
> > The XFS filesystem supports more advanced Access Control Lists (ACL's
> > for short) that go far beyond this course granularity. Check out:
> > http://oss.sgi.com/projects/xfs/features.html
> >
> > The only thing is that XFS is not supported without getting an XFS
> > kernel. But if you do this and mount a data and non-system bin
> > partition, you can do extraordinary things. You *must* be certain that
> > the version you get is considered "good", there are patches for various
> > kernel numbers, even for RH kernels to install by, but you want a solid
> > version for your running system; if you do not need to run your root
> > partition on XFS, this is trivial. And XFS is *very* good performance
> > and meta journalling.
> 
> Yes, i'mi aware of this. And there is capability support in the kernel as
> well. The problem as i see it: there is no unified support for these
> features. Programs like 'chmod' etc. need to support these features.
> There's no central place (or central tool) to mangage user permissions/capa-
> bilities. You need to use a plethora of tools to do something like:
> "This user is a normal user _but_ is allowed to open /dev/dsp, set
> realtime priority on programs that access /dev/dsp but can't use
> more than x% of the CPU".
> Also, most applications would need to be rewritten to take advantage
> of these features. No more 'setuid(2)' just to open port 53 ....

I think acl's might be useful to allow or deny any program, but setting
them would be a specialized program. It would be nice if chmod was
expanded to use acl's. As for standards, acl's do have a POSIX standard,
but this was invented long after UNIX and normal permissions. Sometimes
people complain about bloat, and suggest a series of smaller tools; next
day it is a problem that too many tools to do the job. Then there is the
problem of each distribution using it's own management tool (which I
usually don't like them anyway). I would hope acl's become more
integrated one of these days.

D. Stimits, stimits at idcomm.com

> 
>  Ralf Mattes
> 
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug