Library for ed25519 signatures

This library is implemented purely in Elixir, of course it will be slow. If you want to have faster and built-in implementation, then compile your Erlang with modern OpenSSL (1.1+ IIRC) and then you can use :crypto module directly.

I have tried generating key using :crypto.generate_key(:ecdh, :ed25519), but Edwards curves are still not implemented in :crypto.ec_curves in order to use it (Erlang/OTP 22).

What is output of info_lib/0? Because if you have compiled with OpenSSL 1.0 then it will not have Edward’s curves as these are implemented in OpenSSL 1.1 and later. In my case:

iex> :crypto.info_lib()
[{"OpenSSL", 269488207, "OpenSSL 1.1.1d  10 Sep 2019"}]
iex> :ed25519 in :crypto.ec_curves()
true

So there is support for such.

EDIT:

Also if you check edwards_curve_dh/0 type then you will see that you need to use :x25519 instead of :ed25519 as a curve name.

The output from info_lib/0 is the same. Creating keys worked without any issues with :x25519, but signing still fails:

:crypto.sign(:ecdsa, :sha256, "msg", [pk, :x25519])
** (ArgumentError) argument error
    (crypto) :crypto.pkey_sign_nif(:ecdsa, :sha256, "msg", {{:evp, :x25519}, <<127, 106, 130, 168, 185, 171, 210, 115, 88, 46, 158, 57, 39, 101, 193, 114, 130, 214, 69, 220, 117, 184, 188, 107, 199, 80, 49, 234, 39, 74, 176, 96>>}, [])
    (crypto) crypto.erl:1610: :crypto.sign/5

I have tried using binary instead of string for the message, passing only the key, still no luck…

Here is a discussion from 6 years back for incomplete elliptic curve implementation in Erlang, I suppose it still continues to be that way

Any opinion on the libsalty library - GitHub - ArteMisc/libsalty: Elixir bindings for libsodium (NIF) · GitHub?

Yes, now this fails, because when signing you need to use :ed25519 (don’t ask me why this works this way). You need to check documentation of crypto:sign/4.

I haven’t used libsalty, but if NaCL is what you need, or you just want to get crypto right without all that hassle, then it is good choice.

why does it work this way?

Yes, with :crypto.sign(:eddsa, :sha256, msg, [priv_key, :ed25519]) worked! I get between 1.5e-4 and 2.3e-4 in time which is great.

However, :crypto.verify(:eddsa, :sha256, msg, signature, [pub_key, :ed25519]) always returns false…

EDIT:

I have tested the Salty library and works impressively good - in terms of performance it is even faster than the Erlang :crypto. For now I will stick with it.

In :crypto.generate_key(:ecdh, :ed25519) ecdh refers to all kinds of eliptic curves and then we use x25519. While in :crypto.sign(:eddsa, :sha256, msg, [priv_key, :ed25519]) we have special function for Edwards curves, in which we use ed25519.
I hope it makes a bit more sense.

I believe :crypto.generate_key(:eddsa, :ed25519) should be used for EdDSA signatures. The curve x25519 is only used for key exchange algorithms (like ecdh). Anyway, the :eddsa, :ed25519 parameter combination seems to be supported only in Erlang 23.1.
Regarding libsalty, the repo has been archived, so I am not confident in using it.

Given this, any ideas? Perhaps the best option will be to depend on Erlang 23.1…

I think this was already solved… but I’ve been playing with the :crypto library recently and it looks like you needed to use a different parameter.

This is an easy mistake to do because of the formatting of the documenation. So, it looks like this:

edwards_curve_dh() = x25519 | x448
edwards_curve_ed() = ed25519 | ed448

But, the params for type :ecdh are the following: ecdh_params() = ec_named_curve() | edwards_curve_dh() | ec_explicit_curve()

If you switch out :ed25519 (which is edwards_curve_ed()) for :x25519, then it’ll work (assuming you’re now using erlang 23+). I’m currently on 23.2.1 and it’s working.