Re: Refactoring HMAC in the core code

On Fri, Dec 18, 2020 at 08:41:01AM +0900, Michael Paquier wrote:
> Knowing that we are in a period of vacations for a lot of people, and
> that this is a sensitive area of the code that involves
> authentication, I think that it is better to let this thread brew
> longer and get more eyes to look at it.  As this also concerns
> external SSL libraries like libnss, making sure that the APIs have a
> shape flexible enough would be good.  Based on my own checks with
> OpenSSL and libnss, I think that's more than enough.  But let's be
> sure.

FWIW, I got my eyes on this stuff again today, and please find
attached a v2, where I have fixed a certain number of issues:
- Fixed a memory leak with the shrink buffer in the fallback
implementation.
- Fixed a couple of incorrect comments.
- The logic around the resowner was a bit busted with OpenSSL <=
1.0.2.  So I haev reorganized the code a bit.

This has been tested on Windows and Linux across all the versions of
OpenSSL we support on HEAD.  I am also attaching a small module called
hmacfuncs that I used as a way to validate this patch across all the
versions of OpenSSL and the fallback implementation.  As a reference,
this matches with the results from Wikipedia here:
https://en.wikipedia.org/wiki/HMAC#Examples
--
Michael