nmdanny/softauth

A software FIDO2/WebAuthn authenticator using CTAP2 and virtual HID

softauth's Issues

UI

• Choice of GUI framework:

  • Tauri, an electron-like framework. It would be nice for refreshing my front-end skills, and might be worthwhile if I ever require a more complex UI (e.g, credential syncing, more configuration options like storage backends, import/export)

  • iced - a cross platform UI library for Rust (made from scratch) that follows the Elm architecture.
    A lot simpler than Tauri, though it is a 'toy' framework and might be limited in some aspects.

  • relm - GTK based, also follows the Elm architecture.

  • Don't use a GUI and instead do something crude like starting an interactive terminal when asking for UV/UP.

  • Maybe a TUI library?

• Features:

  • User presence
  • User verification
  • Key management (CRUD operations - list, delete, add via makeCredential)
    • Credential management, import/export, sync?

Refactor CBOR (de)-serialization

• Find a more generic way of deserializing Keymappable objects that does not rely on intermediate conversion to a CBOR value
  (similar to how the Serializer is implemented)
• Find a nicer way of sorting CBOR maps without converting to intermediate CBOR values
• Alternatively, find an alternative CBOR library that already does some of this.
  minicbor seemed almost fitting, except that it forces you to you use integer keys for all structs,
  and I need string keys for structs that aren't top level request/response objects.

Implementing timeouts

• Timeouts on the CTAP-HID level - HID server should go idle after a while without receiving extra continuation packets (or if
  CTAP2 message processing timed out)
• Timeouts on CTAP2 message processing. This might be relatively easy with tower middleware
  • UI integration, once there is a UI for user presence/verification

UHID OutputEvent data comes with an extra byte at the front

Current workaround is to simply delete that byte

softauth/src/hid/linux/uhid_transport.rs

Lines 33 to 36 in 16f0b29

	Ok(OutputEvent::Output { mut data }) => {
	// TODO: BUG: why do UHID output event come with an extra byte in the front?
	data.remove(0);
	if let Err(e) = send_read.send(Ok(data)) {

I suspect that uhid-virt's C bindings (uhidrs-sys) might be outdated - might be missing a field

Build recommendations

Did you ever get this to work? Love the idea. Kept on running into build issues. What is the recommended build environment? I'll probably try a docker build.

Implement CTAP-HID keepalives

CTAP1/U2F

Long term goals - interop with U2F clients, processing CTAPHID_MSG messages

Secret storage

Where to store secrets:

• File based, e.g, SQLite, JSON file. Easiest in terms of OS portability, easy and familiar from a programmatic standpoint.
  Would require implementing encryption, thus adding complexity (need to define a UI for setting the private key) and increases
  the attack surface.

• Keyring: Linux secret store, Windows credential management, etc..
  Look into the keyring library - though its API is a bit limited (e.g, no ability to enumerate
  all secrets under the service, requires utf-8 strings), might be able to simply store all keys in a single JSON string.

• TPM: The most secure option, but also the most complex.

CTAP-HID abort and cancellation

• Both CTAPHID_CANCEL and CTAPHID_INIT(when busy) messages are treated the same
  

  softauth/src/hid/packet_processing.rs

  Lines 202 to 207 in b0b8677

  	if [Ok(CommandType::Init), Ok(CommandType::Cancel)]
  	.contains(&init.get_command_type())
  	{
  	// TODO: difference between abort and init
  	self.abort_transaction();
  	Ok(PacketProcessingResult::Aborted)

  Need to figure out what's the difference(if any)

• Propagate cancellation to CTAP2 message processing

  • Propagate to UI once it is implemented

Graceful shutdown

• Add SIGINT, SIGTERM handlers that allow graceful shutdown
• When a Tokio task/thread panics, the entire daemon should shutdown and not hang.

Windows support

• Since Windows doesn't have something analogous to Linux userspace HID, Windows support would require implementing a driver that
  emulates a HID device.

  This can be a nice exercise in driver development, but currently a low priority

• Cross platform UI toolkit, or re-implement the UI for Windows

• Cross platform storage