News for hackers. Stuff that matters.
Re:The Linux Security Circus: On GUI isolation by schoen on Thursday April 17, 2014 @08:16PM (#7607320)
I agree with the concern and appreciate the solution that Qubes offers.

It might be worth pointing out that some of the "hippies" also became concerned with this and implemented their own partial solution within the X framework decades ago -- the "secure keyboard" feature. You can see it in action: run an xterm, Ctrl-Left Click in the window, and choose "Secure Keyboard" from the menu. Now all X11 keyboard events are sent exclusively to that xterm, not to any other X application (even if the other application has focus). Caution: for these purposes the window manager is also an "application", so you won't even be able to Alt-Tab to switch applications while the secure keyboard mode is active!

The X11 developers assumed that you would use the Secure Keyboard feature while typing important passwords. There's a pretty elaborate discussion of this in the man page for xterm(1).

There, the threat that they're most focused on is remote display access from applications on other hosts, rather than malicious software potentially running on the local host. For example, the authors of xterm fear that you'll be using machines foo and bar in the same computer lab, and run "xhost foo" on bar, and then an attacker will rsh into foo and log in to their account there and then run "DISPLAY=bar:0 keyboard_sniffer" from their account on foo, whereupon the X server on bar will conclude that the remote client request from foo is perfectly legitimate because you told it to accept all X11 connections originating from foo.

Of course the secure keyboard feature only partially mitigates one aspect of this threat, while Qubes offers a much more thoroughgoing and useful mitigation to a larger range of things. (I might also mention ptrace, where recent Linux distributions have activated a kernel policy which forbids, by default, having a non-root process start to ptrace another running process that isn't its child. I believe this was in response to malware grabbing private keys from programs like ssh-agent via the ptrace interface. Oops. It does also mean that you can't strace -p or gdb attach something that's already running, unless you become root and change the policy.)

In general, the idea of protecting against software that's running locally as the same user is something of a novelty on most desktop OSes.

https://en.wikipedia.org/wiki/Discretionary_access_control

Re:The Linux Security Circus: On GUI isolation by Someone on Friday April 18, 2014 @05:48AM (#7608595)
We also have http://www.cs.dartmouth.edu/~doug/IX/, in particular "Multilevel Windows on a Single-Level Terminal" (http://www.cs.dartmouth.edu/~doug/IX/163f.ps):

"The workings of mux, a windowed-terminal handler, that can run differently classified sessions in different windows."

[google search tip for historians/computer archeologists who don't want to find SE-Linux when searching for Unix history: enter "before 1/1/{years ago} as a custom time range]