Re: [babel] Babel-MAC: Blake2s is 128-bits by default

[Juliusz Chroboczek <[e-mail]>]

After reading your replies, and thinking it over carefully, I disagree
with the apparent consensus.  I think Blake2s in Babel-MAC should use
128-bit keys, as in the initial implementation.  Please hear me out.

Toke and Dave originally suggested adding Blake2s to Babel-HMAC for two
reasons:

  - it is more computationally lightweight on platforms that don't
    implement SHA-256 in hardware;
  - it may serve as a fallback if SHA2 is ever broken.

While I am still not quite convinced by these arguments, I've agreed to
add the SHOULD Blake2s and to implement Blake2s for the following reasons:

  - it serves as a good example of adding a new algorithm to Babel-MAC;
  - with 128-bit signatures, it is useful on low-rate networks, which
    Babel is commonly used with.

For these reasons, we implemented Blake2s with 128-bit signatures.
Unfortunately, some confusion ensued, Toke implemented Blake2s with
256-bit signatures, Antonin changed our implementation to interoperate
with Toke's (without discussing the issue with me beforehand), hence the
current situation.

If Blake2s uses 256-bit signatures, then I see little convincing reason to
keep it around in addition to SHA-256.  For this reason, I recommend the
following course of action:

  1. Toke agrees to switch to 128-bit Blake2s signatures;
  2. we discard Antonin's branch (which is just a private branch), and do
     128-bit Blake2s signatures instead;
  3. during AUTH48, I make the clarification that the SHOULD implement
     applies to 128-bit signatures.

Two more points.  First, under no circumstances will I go against WG
consensus on this point -- if the WG strongly believes that the SHOULD
applies to 256-bit signatures, I will naturally follow the consensus.
Second, Babel-HMAC is an extensible protocol, so even if the RFC ends up
recommending n-bit signatures, nothing prevents a given implementation
from implementing m-bit signatures in addition to the RFC-recommended
n-bit signatures.

Sorry again for the confusion.

-- Juliusz