Heads up for login managers

[David Zeuthen]

On Mon, 2007-02-12 at 14:53 -0500, Alan Cox wrote:
> On Mon, Feb 12, 2007 at 02:43:13PM -0500, David Zeuthen wrote:
> > (and the other things I'm saying are 1) we really want revoke() to
> > ensure privacy of sessions on the same seat; and 2) it would be nice to
> > use a kernel mechanism rather than the rather ugly XDG_SESSION_COOKIE
> > mechanism.)
> 
> 1. Agreed (and SELinux)
> 
> 2. Your XDG_SESSION_COOKIE doesn't do anything anyway.

Of course it does. 

User alan is logged in on two seats A and B. 

Seat A gives access to Hardware X (say, a usb optical drive).

Seat B gives access to Hardware Y (say, another usb optical drive).

Login session for alan on seat A is active.

Login session for alan on seat B is inactive.

Hence, user alan should only have access to hardware Y.

Further, programs running for alan on seat B needs to be denied access
to certain D-Bus methods on system daemons like Bluez, HAL,
NetworkManager and other services. Programs running for alan on seat A
need not to be denied this access. The system daemons are smart enough
to know what seats the hardware belong to.

Bottom line: system-level daemons needs to know more than the uid of the
caller. They actually need to know what session they originate from.
Hence XDG_SESSION_COOKIE. I'm just asking for a nicer way to do this.
Thanks.

      David