FIDO2 screenlock authentication fixes

[TheMartinizer]

When registering a FIDO2 key using the ScreenLockCredentialStore, the process currently throws an exception if the Android Keystore does not have a default attestation key to sign the newly generated key pair with (as is the case for both phones I've tested on). I've added code to catch this exception and try generating a key pair again without any attestation. Also added code to use Strongbox if available, although this hasn't been tested since none of my phones have Strongbox.

Also added code that allows signing in with screenlock even if the requesting party has not supplied an allowed credential list. Currently, the code looks through the keystore for all keys that match the requesting party ID, and returns the first one it finds. In the future, the user should probably be allowed to choose a credential, as specified in the Web Authentication standard: https://www.w3.org/TR/webauthn-2/#sctn-op-get-assertion

[ale5000-git]

Quote from https://developer.android.com/privacy-and-security/keystore:

If the StrongBox Keymaster isn't available for the given algorithm and key size associated with a key, the framework throws a StrongBoxUnavailableException. If you get this exception, try using TEE for your key storage as a fallback option.

It seems that it may be available but still fail so the StrongBoxUnavailableException need to be handled.

[TheMartinizer]

You are right! It really doesn't hurt to do some extra error checking. I think that according to the documentation, all StrongBoxes should be able to handle the key parameters that that the code currently chooses for the screenlock key, but that doesn't mean that all phones will have implemented it correctly. I don't have a phone with StrongBox to test on myself, but I have tested the error handling, and it seems to work.

[mar-v-in]

Also added code that allows signing in with screenlock even if the requesting party has not supplied an allowed credential list. Currently, the code looks through the keystore for all keys that match the requesting party ID, and returns the first one it finds. In the future, the user should probably be allowed to choose a credential, as specified in the Web Authentication standard: https://www.w3.org/TR/webauthn-2/#sctn-op-get-assertion

Not only that. To support requests without an allowed credential list, we'd need to support storing of user entities (including handle) on the device when registering, as the result must include the corresponding user handle. Returning the user handle is optional when the allowed credential list is provided (as it is assumed that the user handle is already known as the authenticating entity already knows which credentials are valid), but is needed as soon as we do support resident/discoverable credentials. It is also required to have the user entity (which includes name, icon and display name) to make the credential chooser any reasonable (we can't ask users to select from random IDs or public keys).

Another complexity comes from the question if and how we want to support backups of keys used for WebAuthN/Passkeys. Google now stores a backup of all passkeys with the Google account instead of storing it exclusively in the secure element of the device. This is so that user's don't lose access to their account when losing their phone or moving to another phone. A backup functionality certainly would be desirable for microG's Passkeys implementation as well.

Independent of all of this, your work is a good start and thanks for that.

[ale5000-git]

@TheMartinizer
Since the fix for the exception is mostly ready but the support for requests without an allowed credential list will likely need more time why not create a separate PR for the first so maybe will get merged in the less time?

I suggest also to log warnings to make things clearer in the logcat, like this:

Log.w(TAG, "Failed with StrongBox, retrying without it...")
Log.w(TAG, "Failed with attestation challenge, retrying without it...")

[ale5000-git]

If you want to reuse, I have done the separate changes in my fork here: micro5k@6b334ef

Patch: https://github.com/micro5k/GmsCoreMod/commit/6b334efc89866098337ba39b543044579ab70fca.patch

[TheMartinizer]

Thank you! I used your patch, and now the pull request only contains the exception fixes. I might start to look into credential selection, but that might take a while, and will be in a different pull request!

[ale5000-git]

Thanks.
Time doesn't matter, any help is really appreciated :-)