Re: [PATCH v12 9/9] manual: Add documentation for arc4random functions

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

On 22/07/22 12:33, Zack Weinberg via Libc-alpha wrote:
> On 2022-07-22 8:21 AM, Adhemerval Zanella via Libc-alpha wrote:
>> +Although these functions provide higher random quality than ISO, BSD, and
>> +SVID functions, these still use a Pseudo-Random generator and should not
>> +be used in cryptographic contexts.
> 
> Huh? Isn't the whole point of arc4random that it _is_ cryptographically strong?
> 
> zw

Even OpenBSD manual is not clear about it [1].  This implementation is similar
to OpenBSD, the main differences is that glibc implements a per-thread buffer
instead of a global state (and although OpenBSD does think a global state is 
good qualitiy the are candid say they have no formal proof of these 
qualities [2]); and fork handling is done by glibc instead of relying on 
kernel (I do want to see if we can use a block allocator in this case).

The main problem is if we state this is a CSRNG we need to certify somehow,
which I don't have the background to do; and it might incur in extra 
certification depending of the intended glibc usage [2].

So I think it would be a more conservative approach to state our implementation
is not a CSRNG (although it does follow most of the principles layout as 
fast-key-erasure random-number generators [4]).

[1] https://sourceware.org/pipermail/libc-alpha/2022-July/140728.html
[2] https://www.youtube.com/watch?v=gp_90-3R0pE
[3] https://sourceware.org/pipermail/libc-alpha/2022-July/140782.html
[4] http://blog.cr.yp.to/20170723-random.html