curve25519

[Aris Adamantiadis]

Hi,

The main advantage of your contribution is a speed increase. The 
disadvantage is that your implementation has not been reviewed for 
security by experts yet, and thus is not as reliable as the reference 
implementation.
I believe OpenSSH (and libssh from my pov) is not the right place to 
introduce experimental cryptographic code. The speed increase advantage 
is not very relevant to SSH, because the key exchange happens only once 
per session (on average), and we were using much slower algorithms till 
last year (DH and ECDH), that nobody ever complained about.

You should probably try to get that code to be part of OpenSSL. I 
Believe cryptographic implementations should go in crypto libs, and we 
should bundle/maintain as little crypto code as possible in crypto 
consuming projects.

Aris

Le 10/06/15 05:16, Mehdi Sotoodeh a écrit :
> I have developed a compact at the same time high performance library for
> curve25519/ed25519 and I have placed it in the public domain. It support DH
> key exchange as well as ed25519 keygen, sign and verify. The implementation
> is constant-time, supports blinding, bulk-verify and more.
>
> The library is available as portable-C as well as ASM for Intel-x64 CPUs.
> It outperforms curve25519-donna by a factor of 3.6 to 11 depending on the
> target.
>
> You may have a look at the source code hosted at:
> https://github.com/msotoodeh/curve25519.
>
> I was wondering if OpenSSH is a suitable home for this library?
>
>
> Thanks, Mehdi.
> _______________________________________________
> openssh-unix-dev mailing list
> openssh-unix-dev at mindrot.org
> https://lists.mindrot.org/mailman/listinfo/openssh-unix-dev
>