Re: [PROPOSAL/PATCH] Fortuna PRNG in /dev/random

[Jean-Luc Cooke <jlcooke@certainkey.com>]

On Mon, Sep 27, 2004 at 10:23:52AM -0400, Theodore Ts'o wrote:
> On Mon, Sep 27, 2004 at 12:50:33AM -0000, linux@horizon.com wrote:
> > > And the ring-buffer system which delays the expensive mixing stages untill a
> > > a sort interrupt does a great job (current and my fortuna-patch).  Difference
> > > being, fortuna-patch appears to be 2x faster.
> > 
> > Ooh, cool!  Must play with to steal the speed benefits.  Thank you!
> 
> This is somewhat fundamental to the philosophical question of whether
> you store a large amount of entropy, taking advantage of the fact that
> the kernel has easy access to hardware-generated entropy, or use tiny
> pools and put a greater faith in crypto primitives.

Tiny in that at most you can only pull out 256bits of entropy from one pool,
you are correct.  SHA-256 buffers 64 bytes at time.  The transform requires
512 bytes for its mixing.  The mixing of the 512 byte W[] array is done
serially.

random_state->pool is POOLBYTES in size.  Which is poolwords*4, which is
DEFAULT'd to 512 bytes.  The "5 tap" LFSR reaches all over that 512byte
memory for its mixing.

If page sizes get big enough and we page-align the pool[] member, the
standard RNG will get faster.

JLC