crypto: use a one-step kdf for session keys, fixes #7953

[ThomasWaldmann]

No description provided.

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (6a511fd) 83.65% compared to head (74c34ba) 83.65%.
Report is 2 commits behind head on master.

Files	Patch %	Lines
src/borg/crypto/key.py	83.33%	0 Missing and 1 partial ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7955      +/-   ##
==========================================
- Coverage   83.65%   83.65%   -0.01%     
==========================================
  Files          66       66              
  Lines       11860    11861       +1     
  Branches     2149     2150       +1     
==========================================
  Hits         9922     9922              
- Misses       1363     1364       +1     
  Partials      575      575              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ThomasWaldmann]

Nice:

• faster key derivation (relevant for existing content)
• get slightly faster startup time (less self-tests)
• looks like we can get rid of quite some crypto code
• no "own crypto implementations" (was: HKDF-HMAC-SHA512) any more

[marti4d]

Just a thought, but I know I used the word flavour (spelled a certain way 🇨🇦 🇨🇦 🇨🇦) when I was talking about the string detailing what encryption suite you are using, but that's just kind-of a cheeky word I used for it. I think most organizations like FIPS would use the more standard term domain for it, since "domain separation" is the reason it's there.

[ThomasWaldmann]

@enkore can you review, please? You added the hkdf code 7y ago for the TAM stuff.

In master branch, i removed all the TAM stuff because we have typed repo objects now, so borg actually knows what's metadata and what's user data and verifies that it gets the type of object it requested from the repo.

So the session key computation was the last user of that hkdf and after discussion in #7953 a way simpler "one-step kdf" was found.