Re: FreeBSD rnd bug

[Jean-Yves Migeon <jym%NetBSD.org@localhost>]

Le 19/02/2015 09:47, Fredrik Pettai a écrit :
> Did you see this:
>
> https://lists.freebsd.org/pipermail/freebsd-current/2015-February/054580.html
>
> I know that the rnd code in NetBSD has been reworked several times lately,
> is the NetBSD design very different from FreeBSDs? or are they fairly equal?

They are completely different. There are two PRNG:
- a "fast" one (chacha20) which is consumed in places where plain 
randomization is required in performance critical paths (ASLR, TCP 
sequence, anon ports...);
- a "strong" one (NIST CTR DRBG with AES 128 as block cipher) for crypto 
related operations.

The seed for the fast RNG is obtained from the strong RNG, and the 
strong PRNG is initted through rndsink(9). Of course from there, quality 
depends on the entropy sources, but this is just to show that the call 
paths and APIs have nothing in common.

--
Jean-Yves Migeon