CDRIVER-5634: SCRAM-SHA-256 FIPS Compliance

[Julia-Garland]

Summary

(Mostly) ensures the driver implementation of SCRAM-SHA-256 (as well as SCRAM-SHA-1) delegates implementation of key derivation is delegated to a cryptographic provider that already is or can be configured to be FIPS-compliant.

What changed?

Previously the driver implemented the PBKDF2 (Password-based-Key-Derivative-Function) itself in _mongoc_salt_password().

• When MONGOC_ENABLE_CRYPTO=LIBCRYPTO is enabled, the driver uses PKCS5_PBKDF2_HMAC() with EVP_sha256() (or EVP_sha1()) from the OpenSSL EVP library.
• When MONGOC_ENABLE_CRYPTO=COMMON_CRYPTO is enabled, the driver uses CCKeyDerivationPBKDF() with kCCPRFHmacAlgSHA256 (or kCCPRFHmacAlgSHA1) from the CommonCrypto KeyDerivation library.
• When MONGOC_ENABLE_CRYPTO=CNG is enabled, the driver uses BCryptDeriveKeyPBKDF2() with a handle to HMAC-SHA-256 from the Bcrypt library. If the symbol is not supported, the driver reverts to the original key derivation function.

MinGW-w64

BCryptDeriveKeyPBKDF2 was added in MinGW-w64 6.0.0, but some hosts for Evergreen tests use older versions. I filed CDRIVER-5649 to “Require MinGW 6.0.0 or higher”, but until then only users with more modern MinGW versions get FIPS compliance on Windows.

[kevinAlbs]

Looks good overall. Suggested changing returns from int to bool and a possible fix to check for BCryptDeriveKeyPBKDF2.

Nice spot of the duplicate call to BCryptCloseAlgorithmProvider (&_sha256_hmac_algo, 0);.

[kevinAlbs]

LGTM with another check.