on Sat, Apr 27, 2002 at 01:58:08AM -0400, Roland Dowdeswell wrote:
> If we had a capability model, then xdm or login could just assign the
> capabilities to the user as they logged in.  But, that's a very different
> security architecture and has some different setbacks Such as if I ssh
> in to a box on which I am logged in on the console, I can't play audio
> unless I initiated screen(1) from the console and I steal the session---not
> to mention that I can start screen(1) and then log out, leaving me floppy
> access which I can modify when the next user logs in before they mount
> it.  So that'd have to be worked out.  (Please note that I am not
> advocating a capability model.)

On a fairly old linux systems in /etc/login.defs, you can have:
CONSOLE_GROUPS         floppy:audio:cdrom

which sometimes comes in handy, but isn't much better than just adding the
user to the relevant groups permanently if they can log in at the console.
Much like certain userland interactions cause the groups to be assigned,
another interaction would have to cause the kernel to revoke those groups,
but from what..