Update tests for OpenSSL 3 and fix bugs found in test

[mamckee]

This PR updates existing tests for OpenSSL 3, and fixes any bugs found in testing

• Added test cases to be run with the OpenSSL evp_test utility
  • Test cases for both OpenSSL 1.1.1 and OpenSSL 3.0
  • Most test cases are unmodified from the OpenSSL source. The test cases that are modified are due to known implementation differences with SymCrypt or unsupported configurations (e.g. omitting L/separator in SP800-108 or enforcing RSA parameter null termination)
  • Added a readme with instructions on how to test the SymCrypt provider and engine with these test files
• Remove engine EVP interface for OpenSSL 3 to prevent intercepting calls that should be routed to the provider.
• Updated SSLPlay to pass in OpenSSL 3 with the engine and/or provider
  • Support variable-length IVs in the provider AES-GCM
  • A scenario exists in the engine that prevents this from working. EVP_CIPHER_CTX_iv_length calls into the engine, so the engine has no way of knowing the IV length for an IV passed to e_scossl_aes_gcm_init_key. The engine will continue to reject any IVs that aren't 12-bytes.
• Fixed an illegal cast i the engine RSA implementation
• Fixed expected behavior differences in the provider ECC key management, ECDSA, RSA sign and decrypt that caused test failures
• Fixed missing OPENSSL_strcasecmp preventing the provider from running with OpenSSL versions below 3.0.4, when this API was exported

[mlindgren]

  message(FATAL_ERROR "Provide symcrypt pkg-config or set SYMCRYPT_ROOT_DIR variable")

np: tabs #Resolved

---

Refers to: CMakeLists.txt:19 in 55d107f. [](commit_id = 55d107fb4626a97dcb778ffeba4846c8f3c54615, deletion_comment = False)

[mlindgren]

A scenario exists in the engine that prevents this from working. EVP_CIPHER_CTX_iv_length calls into the engine, so the engine has no way of knowing the IV length for an IV passed to e_scossl_aes_gcm_init_key. The engine will continue to reject any IVs that aren't 12-bytes.

I don't follow this comment. Can't the engine save the IV length and use that for subsequent calls?

[mamckee]

A scenario exists in the engine that prevents this from working. EVP_CIPHER_CTX_iv_length calls into the engine, so the engine has no way of knowing the IV length for an IV passed to e_scossl_aes_gcm_init_key. The engine will continue to reject any IVs that aren't 12-bytes.

I don't follow this comment. Can't the engine save the IV length and use that for subsequent calls?

It depends on EVP_CTRL_AEAD_SET_IVLEN being set before EVP_CTRL_INIT. We might be able to rework this (i.e. not setting the IV until e_scossl_aes_gcm_init_key) but I don't want to spend a ton of time on adding new behavior to the engine. I'm sure there was a reason the IV was set in EVP_CTRL_INIT so it's something I'm hesitant to modify unless we need to.