Re: [TLS] RNG vs. PRNG

[Dean Anderson <[e-mail]>]

On Wed, 5 May 2010, Nicolas Williams wrote:

> I'm not sure what the point of this sub-thread is.  

The point of the draft is to extend the random bits.  So a question 
related to that is: "Is anything to be gained by having more 'random' 
bits?"  This sub-thread, in part, answers that question.

The real random value is pre_master_secret. The purpose of the 
*Hello.random additions is to ensure a unique master_secret on 
additional channels. These have no contribution to the security of the 
random set chosen for pre_master_secret:

master_secret = PRF(pre_master_secret, "master secret",
                          ClientHello.random + ServerHello.random)
                          [0..47];

The master_secret has 48 bytes, the pre_master_secret has 48 bytes. One 
cannot add "more randomness".  Hence the conclusion is this:

1) One cannot add more "randomness" into the calculation. Either 
pre_master_secret is a true random number or it isn't. 

2) Since the *Hello.random bit set is publicly exposed, using the same 
random generator for these as for pre_master_secret would potentially 
expose the [P]RNG to analysis and thereby expose pre_master_secret as 
well as *Hello.random. If all the inputs to the PRF are known, then 
master_secret is known.


> TLS can't really mandate that implementations use real RNGs

I agree. So we have to assume that some [many] implementers will use 
PRNGs and so exposing these bits to analysis is a very bad thing(tm).

> > True RNG "itself" can't be attacked - though its implementations could
> > (depending on many circumstances). Attacks against PRNG could be both
> > cryptanalytic and implementation-directed.
> 
> A true RNG depends on physical processes.  Therefore it can be attacked
> by physical means.  

Perhaps in principle, the RNG could be tampered with so future bits
aren't random. But the past bits would remain secure. So having a log of
messages encrypted with a true RNG remain unbreakable, even if you
subseqently tamper with the RNG.  This would not be true of a PRNG. Once
you gain the seed, you can compute any output.


> The simplest thing to do is to couple an RNG to a PRNG.

No. A PRNG only generates a subset of the input: equal in size to the
size of the seed. PRNGs reduce entropy to the size of the seed. 

Design Rule: You can't get better than a real RNG. Once you have that,
you can only make it worse. (I suspect that stems from thermodynamics)

		--Dean


-- 
Av8 Internet   Prepared to pay a premium for better service?
www.av8.net         faster, more reliable, better service
617 256 5494