Re: [RFC] frandom - fast random generator module

[Sandy Harris <sandy@storm.ca>]

H. Peter Anvin wrote:

> No, I mean that putting a piece of code in the kernel "so it can be
> accessed from shell scripts" is idiotic.  Make a binary of it and put
> it in the filesystem.

I posted one of those here during a previous discussion.
http://www.geocrawler.com/archives/3/35/2000/8/0/4192943/
The version I posted was first draft code, quite likely buggy,
but the general idea  was sound.

This was a while back, before the /dev/random code was
rewritten into a two-stage generator. Since my code
was to add a second stage to old /dev/random, I doubt it
is now a good idea.

If the problem is that /dev/urandom is too slow, then
we need to look at speeding it up, not adding a PRNG,
let alone one in the kernel.

Would a block cipher second stage as in Yarrow or my
example be faster than the hashing 2nd stage Ted used?
Can we use a block cipher without legal hassles? Is
there some third choice? A faster hash?