It was previously generating TLS keys, which seems to have been an oversight.
Using EdDSA also has a slight performance edge, as there's some mutex contention when ECDSA keys are used.
It turns out the JDK implementation (`SunEC` provider) of Ed25519 signature verification is quite slow, slower than the abandoned library (i2p) it replaced. This has been replaced by Bouncy Castle, whereby the `EDDSA_ED25519_SHA512` signature scheme uses it. `SunEC` still remains the default implementation. `Crypto.toSupportedPublicKey` (and `toSupportedPrivateKey`) were tweaked to make sure any `SunEC` keys are converted to Bouncy Castle. The presence of two different `EdECPublicKey` implementations for the same key causes cache misses in `BasicHSMKeyManagementService`, resulting in another performance degradation.
The various crypto tests that were previously ignored have been re-enabled.
The abandoned i2p EdDSA library has been replaced with native support that was added in Java 15.
Java 17 (via the `SunEC` provider) does not support the secp256k1 curve (one of the two ECDSA curves supported in Corda). This would not normally have been an issue as secp256k1 is already taken care of by Bouncy Castle. However, this only works if the `Crypto` API is used or if `”BC”` is explicitly specified as the provider (e.g. `Signature.getInstance(“SHA256withECDSA”, “BC”)`). If no provider is specified, which is what is more common, and actually what the Java docs recommend, then this doesn’t work as the `SunEC` provider is selected. To resolve this, a custom provider was created, installed just in front of `SunEC`, which “augments” `SunEC` by delegating to Bouncy Castle if keys or parameters for secp256k1 are encountered.
`X509Utilities.createCertificate` now calls `X509Certificate.verify()` to verify the created certificate, rather than using the Bouncy Castle API. This is more representative of how certificates will be verified (e.g. during SSL handshake) and weeds out other issues (such as unsupported curve error for secp256k1).
`BCCryptoService` has been renamed to `DefaultCryptoService` as it no longer explicitly uses Bouncy Castle but rather uses the installed security providers. This was done to fix a failing test. Further, `BCCryptoService` was already relying on the installed providers in some places.
The hack to get Corda `SecureRandom` working was also resolved. Also, as an added bonus, tests which ignored `SPHINCS256_SHA256` have been reinstated.
Note, there is a slightly inconsistency between how EdDSA and ECDSA keys are handled (and also RSA). For the later, Bouncy Castle is preferred, and methods such as `toSupportedKey*` will convert any JDK class to Bouncy Castle. For EdDSA the preference is the JDK (`SunEC`). However, this is simply a continuation of the previous preference of the i2p library over Bouncy Castle.
"WARNING: sun.reflect.Reflection.getCallerClass is not supported. This will impact performance." warning was being caused by log4j. Latest version fixes this issue.
This is code refactoring and cleanup that is required to add a new WireTransaction component group for 4.12+ attachments, and for supporting legacy (4.11 or older) contract CorDapps in the node.
