Re: Cryptosplit note

[Adam Shostack <adam@bwh.harvard.edu>]

|    It uses rand() when it needs random numbers for the
| coefficients of the polynomial. I don't know what kind of
| security risk that poses, but it really should be using something
| better.  Where can I get Blum-Blum-Shub source or documentation on the
| algorithm?

rand() produces really bad random numbers.  Dose anyone have code for
Mac/dos/unix that figures out how to use the 'better' PRNG that the
vendor ships with ifdefs & stuff?  (On Unix, I use random(3) for bad
random numbers, on the Mac I use the toolbox Random().  I dont code on
pcs.

Adam


-- 
Adam Shostack 				       adam@bwh.harvard.edu

Politics.  From the greek "poly," meaning many, and ticks, a small,
annoying bloodsucker.


to do is to choose
a Blum modulus N = P*Q where P and Q are both equal to 3 mod 4, and of about
the same size.  Choose a random initial seed S and set X0 = S*S mod N.
Then repeatedly iterate X(i+1) = Xi * Xi mod N.  Use the low-order
log2 ( log2 ( N ) ) bits of Xi as the output of the PRNG; for N of 1000 bits
this means you get 10 bits per iteration.

For the cryptosplit application (nice program, BTW) you could use a fixed
pre-computed suitable N.  Then the only hard part is to seed X0.  Maybe you
could use a combination of a hash of the input file and the time of day;
that should be pretty safe although it might be subject to a known-plaintext
attack (where they think they know what you've split up, and they just want
to verify it).  You could add a switch for the user to throw in a random
string as additional seeding material.

The only other problem then is adding an MP package.  A lot of Unix systems
come with libmp, or you could use Gnu or even pgptools.

Hal