GeezerVison

Motivation

Add support for the KMS Decrypt Recipient field. This is used by Nitro Enclaves to decrypt sensitive without their host being able to read it, by KMS re-encrypting the decrypted data to the provided RSA key before it is returned to the caller. This flow is described here.

Changes

I've added basic support to the KMS Decrypt command for the Recipient field by looking for the attestation, parsing it to extract the recipient public key, and returning CiphertextForRecipient instead of Plaintext. It's not clear to me how to simulate verifying the attestation (which involves a signature from AWS that the PCR values are correct) without a significant effort to add proper support for Nitro to Localstack, and I think this PR as-is provides valuable functionality by allowing systems that use Nitro and KMS to be tested against Localstack without having to special-case their calls to Localstack's KMS implementation. So, I don't attempt to do any validation on the provided attestation document beyond what is required to extract the recipient public key.

I've unfortunately had to do a significant amount of manual crypto operations here with asn1crypto and PyCryptoDome because cryptography.hazmat doesn't appear to support the exact algorithms that AWS uses; I apologize for the verbosity and am happy to change it to cryptography.hazmat if there is a way to use it.

I've taken a compromise approach to ensuring that this change is backwards-compatible; if callers pass in a nonsense value to Recipient it will be logged and ignored and Plaintext will be returned, but if an attestation with a valid public key is passed in then CiphertextForRecipient instead of Plaintext will be returned. It is more to spec to error if a nonsense value is passed in for Recipient but that feels a bit more breaking. I'm happy to change to erroring if the maintainers would prefer that now, or it could be done as part of the next major release or similar.

Tests

I've added basic tests for the functionality. I've had to mark them as only_localstack because it is quite complex to call AWS to set up an actual Nitro Enclave, and until Localstack has broader Nitro support I think this will suffice to test. They do test that the flow works end to end. In addition, I have manually tested that this implementation produces a PKCS7 envelope similar to what actual AWS produces for an actual Nitro Enclave, and I have manually tested that my application that works on actual Nitro and KMS works against this PR.

Related

Addresses #12901.