BugTraq
RE: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 05:21PM
Rizwan Jiwan (Rizwan Jiwan KINGSTON Hummingbird com) (2 replies)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Aug 01 2003 06:45AM
Randy Kaelber (randall nimitz net)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 05:54PM
Alaric B Snell (alaric alaric-snell com) (1 replies)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 07:02PM
MightyE (trash mightye org) (2 replies)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 08:06PM
Barry Fitzgerald (bkfsec sdf lonestar org)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 07:53PM
David Riley (oscar the-rileys net) (1 replies)
Re: Another Mac OS X ScreenSaver Security Issue (after Security Update 2003-07-14) Jul 31 2003 08:17PM
MightyE (trash mightye org)

>a) If the screensaver engine is compromised (as it was earlier this month,
>though likely not in a command-execution sort of way), you don't want to
>be able to give the user root privileges. Presumably, xscreensaver has
>safeguards against that (or they assume it'll never be exploited). It
>would be pretty sad to have a root security hole through the screensaver.
>
>
Yes, this would certainly be pretty sad, but as I see it, the
screensaver is one of the most important local (non remote) security
devices, it protects against casual cracking attempts of using a
restricted application that was left running when someone was called
away from their desk. Although a root expoit on the screensaver would
be worse than merely crashing the screensaver, neither are tolerable if
you ask me, so may as well stress test the screensaver engine till you
can rely on it, then trust it with root access.

>b) Sometimes the screensaver does crash. Keep in mind that since the
>screensaver modules are executable code (as xscreensaver modules probably
>are as well, though I've never made one), that's the responsibility of the
>individual screensaver developer to fix. It's nice to be able to kill it
>when it does crash so that you can use the computer again.
>
A developer could add a --PERMIT_KILL option while testing their
screensaver which would tell Escape Pod that it's ok to kill this
process when the request came in. Also if they had a remote terminal
up, and root access, they could always kill it that way. As I see it, a
user really needs to be able to depend on their screensaver to protect
their workstation while they're not at it. A person doesn't necessarily
have the option of logging out of their workstation completely when they
go away from their desk, nor the time to take other security steps.

It's true that someone having physical access to your machine (and a
desire to use such access for evil) gives them a level of control that
can't be truly protected against with a screensaver of any calibre, but
it does protect the payroll manager from having one of his lackeys give
himself a pay raise while he's at a meeting.

-e

[ reply ]


 

Privacy Statement
Copyright 2010, SecurityFocus