Open ashleysommer opened 1 year ago
So it seems like there are two issues here, that are potentially solvable in python-fido2.
1) Not all devices that advertise uv=True
necessarily support clientPin
. There should be an additional call to ClientPin.is_supported()
guarding this line .
2) When "hmac-secret" extension is advertised, but "clientPin" is not, enter a fallback mode where the client can attempt to create a minimal ClientPin
instance without throwing the "Not Supported" exception, just for the purposes of hmac-secret
(it needs the connection's shared secret).
I agree with @ashleysommer's assessment. The assumption that uv==True
in the authenticatorGetInfo
response implies clientPin==True
seems incorrect. uv
refers to built-in user verification and clientPin
refers to the capability of accepting a PIN from the client platform, e.g. from the browser. In my opinion these two are completely independent. For example the CTAP spec explicitly states that "ClientPIN is not considered a built-in user verification method".
The Trezor T supports PIN entry directly on the device's touch screen, which is superior in terms of security to ClientPin, because the PIN cannot be intercepted by a keylogger or other potential malware on the user's machine. Supporting ClientPin on Trezor would reduce the security of the device, because it would encourage users to enter their PIN on their machine rather than on Trezor's built-in touchscreen.
I created two pull requests (#190 and #191) with fixes for these two issues.
Just to give a quick response to this: At first glance this looks fine, as do the two PRs. I will need to scrutinize the spec a bit before merging and unfortunately I don't have time to do that right now. Apologies for the delay!
I have an alternative proposal for changes that I believe should address this, and would love some feedback on it: #193.
As I don't have a Trezor T I haven't been able to test this myself, but looking at the proposed other PRs I believe these changes should address both issues. One difference is that in my PR the get_uv_token
method is preferred instead of internal UV if supported and additional permissions are required. I'm assuming that this is supported by the Trezor, but I am not sure. If not then this may need additional changes.
"UV token" sounds like something from CTAP 2.1. Trezor currently supports only CTAP 2.0.
"UV token" sounds like something from CTAP 2.1. Trezor currently supports only CTAP 2.0.
I believe that should still be fine, the code should fall back to pinAuth with internal_uv.
Version 1.1.2 is now released and should hopefully resolve this.
Creating a FIDO2 credential using the FIDO2 python module reference hmac-secret example fails with "Authenticator does not support ClientPin".
This is the corresponding Issue I created for the Trezor Firmware: https://github.com/trezor/trezor-firmware/issues/3068
To Reproduce Steps to reproduce the behavior:
The error is pretty easy to track down by stepping through the python-fido2 sourcecode. When creating a new credential the client library checks if it needs to ask for Pin Auth from the device. To determine this, it checks for if 'uv' is in ctap info options, and if
'uv==True'
. The Trezor model T does advertise "uv=True". On the very next line the fido2 library attempts to initialize the ClientPin verification method then throws an exception because "clientPin" is not in info.options.Seems like when it comes to 'uv' and 'clientPin', python-fido2 needs both advertised or neither advertised.
If I patch the python-fido2 library to ignore the advertised 'uv==True' and assume no UV, it works normally, asks for my pin on the screen, then creates and registers the credential correctly on the device.
But it fails again further on. When sending the hmac-secret salt for attestation the library needs access to the pin-protocol's shared secret (to encrypt the salt going over the wire). This causes the fido2 library to attempt to initialize the ClientPin module again that leads to the same "Authenticator does not support ClientPin" message.
Patching to ignore the "uv==true" info in this case does not work, because the
hmac-secret
requires clientPin support regardless.Note, a lot of this is mentioned already under the "User Verification" heading in the Trezor's webauthn app README. It explains that Trezor does support a very minimal implementation of
clientPin
, just to the extent that is required forhmac-secret
to work.Additional context For transparency: I was sent a Trezor Model T free of charge from SatoshiLabs for the purposes of testing and implementing FIDO2 hmac-secret support in a popular opensource application.