[Cryptography] Dieharder & symmetric cryptosystems

[Ray Dillinger]

On 3/10/22 14:00, Michel Arboi wrote:
> Hello,
>
> Sorry for awakening an old thread but it appears that miscellaneous
> changes to /dev/random or /dev/urandom during the last couple of years
> did not fundamentally fix what I noticed in 2019, supposing that this
> apparent weakness needed fixing.
> (In summary, some WEAK results appeared very often on some specific
> tests with the /dev/urandom generator, while they never appeared with
> /dev/random, or very rarely)
> The main change is that /dev/random is now "as weak" as /dev/urandom,
> probably because of this:
> https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.6-Random-Rework
> <https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.6-Random-Rework>
>
> Considering that the old /dev/random or other PRNG do not exhibit this
> behaviour, I suspect some kind of weakness here. I still could not
> find any explanation.
>

Okay, let's be absolutely sure what you're saying here.

Are there diehard(er) tests that /dev/random is now CONSISTENTLY
failing?  If so which specific tests?

It's normal to see diehard(er) rate something WEAK occasionally; It
happens usually in inverse proportion to the (log of) the number of
elements in the tested sample.  This is just statistics noting that the
particular sample exhibits some bias or some identifiably patterned
sequences. It happens by chance.  And usually we see such ratings less,
when we use a larger sample size.

What we need to worry about, and investigate, is if we're getting WEAK
results on a particular test, consistently.

If WEAK results on a particular test are persistent across several
different runs and regardless of whether we increase the sample size,
then that's something that has to be fixed.

                    Bear