Re: [Discuss-gnuradio] VT receives NSF grant for SDR security

[David P. Reed]

Greg - I think the concept of "software defined radio" being explored by 
the VT folks is a concept I persoally refer to as "crippled software radio".

It is based on a discredited theory of "security" that was called a 
"secure kernel" when I was a student 30 years ago.  In other words - 
that there is a small, well-defined portion of a system that can be 
certified separately from the rest of the system, which has the 
essential property that its *correct* operation *guarantees* that the 
entire system will be secure according to *all possible interpretations* 
of the word secure.

I worked on a project of this sort, and am currently ashamed that I 
helped perpetuate that charade.   I can only say that many others helped 
- it funded lots of work on "proving programs correct" - on the theory 
that it was feasible to prove small programs correct, and thus whole 
systems "secure".

The big lie, of course, is that the researchers essentially redefined 
the word "secure" to mean the trivial notion of security that you 
couldn't compromise the "kernel".   Of course today we stare the 
fraudulence of that idea in the face: phishing, XSS, and other very 
dangerous attacks do not depend one whit on a failure to secure a 
"kernel" of the operating system, or even the "kernel" of a router.

Yet the idea that incorrectness is the same thing as insecurity persists 
in such ideas as the idea that you need "hardware inegrity" to prevent 
attacks on radio systems.

I suggest that it is impossible to carry on a dialog with folks like the 
VT researchers, because they must necessarily buy into the 
"certification of correctness" notion of security.    If they were 
concerned with "correctness" that would be fine - we could carry out a 
meaningful discussion about the difficulty of determining correctness in 
a system that is inherently focusing on getting reliable communications 
through unreliable channels (information theory).   But since they play 
to the gods of deterministic correctness - unreliability doesn't fit in 
their notion of "security" - they cannot even consider the idea that 
there is no "kernel" that can be certified to reduce risk.