Re: [babel] Solving HMAC replay once and for all

[Juliusz Chroboczek <[e-mail]>]

> The normal packet reception flow just:
> -  computes pseudoheader, hashes that + what was received, and compares it with the trailer HMAC. 
> - If success, compare counter, 
>   - if too old (modulo replay window if any), drop, otherwise process normally.
> - If failure, maybe challenge (see above).

Hmm... so now both challenges and challenge replies are unprotected by
HMAC (since they need to be acted upon even in case of HMAC failure),
right?  Therefore, we put them in the packet trailer, so that we can
retain the property that we don't parse the packet body before HMAC
validation.

Implementation complexity is comparable, I think.  However, now the
Neighbour Table entry is created before HMAC validation (to keep the
challenge state), so an evil peer is able to create unbounded amounts of
state (by spoofing multiple source addresses).  I guess you could avoid
that by encoding a cookie in your nonce, but that in turn increases
implementation complexity.

Am I missing something?

-- Juliusz "I'll sleep when I retire" Chroboczek