[Cerowrt-devel] New random number generator for linux...

[Dave Taht]

On Thu, Oct 17, 2013 at 5:57 PM, Stephen Hemminger
<stephen at networkplumber.org> wrote:
> On Thu, 17 Oct 2013 17:40:37 -0700
> Dave Taht <dave.taht at gmail.com> wrote:
>
>> This new design for a rng actually appears to be pretty reasonable.
>>
>> http://thread.gmane.org/gmane.linux.kernel.cryptoapi/9878
>>
>> Full details, paper, and code patches here:
>>
>> http://www.chronox.de/
>>
>> I am (as usual) low on time to package it up for openwrt, or patch it
>> into the kernel, and try it. I will probably be able to get around to
>> it next weekend if nobody beats me to it, slamming it into my pie
>> build for x86 first.
>>
>> I'd be very interested if some folk could compare the standalone
>> daemon to rngd on their favorite platform? And if anyone here has the
>> chops to think hard about the algo, that would be nice, too...
>>
>> In other news , theodore ts'o's got a whole bunch more new patches for
>> the random driver on kernel.org that I HAVE patched in and am trying
>> to get into the next cero release. Fixes for get_cycles(), arrived
>> already in mainline, and the mips patch is around somewhere.
>>
>> While that stuff improves the quality of the entropy in the system,
>> the above rng would bring *more* entropy into the system, maybe get
>> rid of the existing daemon (or integrate with it) and/or hostapd
>> patches, etc, which I find exciting. That said, validating rng stuff
>> is *hard*.
>>
>
> I would not trust it until accepted upstream.

Hah. I wouldn't trust it even then! For half a decade, at least!

> It is better to mix multiple sources in rather than just relying on
> any single source. That way it makes it harder for organizations to
> poison the well.

That I agree with also. However it seemed like this could be used as
an additional source better than the existing rngd (from userspace)
and perhaps the kernelspace one makes sense.

Please note that the current situation in cerowrt (where we are
basically mixing back urandom into random via rngd, and get_cycles()
returns 0) is untenable.

Additional ideas for entropy welcomed....

Another patch in openwrt I'm kind of dubious about is this one...

https://github.com/dtaht/cerowrt-next/blob/master/package/network/services/hostapd/patches/500-random_pool_add_kernel.patch

-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html