Re: [PATCH,RFC] random: make fast_mix() honor its name

["Jörn Engel" <joern@logfs.org>]

On Sun, 22 September 2013 17:27:52 -0400, Theodore Ts'o wrote:
> 
> The structure of the mixing functions in /dev/random has been well
> studied, and validatetd in a number of different academic papers.  So
> I prefer to stick with the basic architecture, even as it is scaled
> down for speed reasons and beause the pool is smaller.

And I want to keep that function.  Essentially the point of fast_mix()
is to ratelimit _mix_pool_bytes().  Naïve ratelimiting would simply
discard the input once the ratelimit has been reached.  My proposal is
to still use the input bits, but use a really cheap mixing function.

Your version of fast_mix() failed in the "really cheap" department.
As a result, it showed up in profiles and at least one idiot (me)
reverted to naïve ratelimiting.  It could have been worse, I was
explicitly asked twice to just remove the call to
add_interrupt_randomness().

So don't think of my patch as weakening the mixing, but as
strengthening the ratelimited mixing.  If we have few interrupts,
_mix_pool_bytes() will be run once a second, if we have many it will
be run once every 64 interrupts.  And in the latter case, the input
for _mix_pool_bytes() is much better than with naïve ratelimiting.

And you should do the same for add_timer_randomness(), where again you
have ratelimiting.  Once trickle_thresh is reached your code simply
discards most randomness.  Only once in 4096 call do you use all the
bits you get - most of which will be predictable.  Why not use a cheap
mixing function for the other 4095 calls and ensure we have many good
bits on call 4096?

Jörn

--
You can't tell where a program is going to spend its time. Bottlenecks
occur in surprising places, so don't try to second guess and put in a
speed hack until you've proven that's where the bottleneck is.
-- Rob Pike