I ran across this article today:

http://www.acsa-admin.org/2005/abstracts/47.html

Apparently this is a paper that was just presented at a computer
security conference. I read Ken Thompson's "trusting trust" paper way
back in the day and was pretty impressed but completely forgot about it
because it's not worth worrying about when there is nothing you can do
about it. Now this this paper comes along and claims to detect the attack.

But it makes a pretty huge assumption: towards the end of section 5 it
says "Instead, during DDC we only only use a trusted compilation process
T, code generated by T, and other programs in environements which we
trust do not trigger on the compilation of Sa." And the first line of
section 6 is "DDC requires a trusted compiler T and trusted
environment(s) where there is a high degree of confidence that any
triggers against Sa that may be in compiler A will not also be present."

I thought it was that kind of trust that we couldn't trust in the first
place? Have we really accomplished anything towards countering this
attack? It seems this counter to the "trusting trust" attack is all
about hoping you have enough diversity not to activate the triggers
while never really knowing for sure.

Section 8 talks about tcc which I had never heard of. Apparently this is
a completely fresh ANSI C compiler implementation which is capable of
compiling the Linux kernel.

In section 8.1 they use gcc to check for malicious behavior in tcc. Why
didn't they do it the other way around?! I don't give a care if tcc is
backdoored or not but I know a lot of people would be interested in
knowing if gcc is! Instead this is left as "Future potential work".

-- 
Tracy R Reed
http://copilotconsulting.com
1-877-MY-COPILOT


-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to