Re-exports types from the toplevel instead of placing them in individual
modules (often which only contain one type).
This makes the API easier for users to navigate, while still retaining
the same module structure internally.
Additionally, this commit uses the `uuid` crate for modeling UUIDs.
Adds a `yubikey::Result` alias with `yubikey::Error` as the error type.
Since we only have one `Error` type, this simplifies the return types
where a `Result` is returned.
* Add MgmKey::set_default method
This wipes any metadata related to derived and PIN-protected management
keys, returning the management key to its default state.
* Transaction::set_mgm_key: Take touch requirement as bool
The Option<u8> was inherited from the original C code's usage of an
unsigned char. We don't need that flexibility, because only two cases
are supported.
* Replace MgmKey::set with MgmKey::set_manual
MgmKey::set_default is now implemented as a wrapper around
MgmKey::set_manual, as they both require clearing metadata related to
derived and PIN-protected management keys.
MgmKey::set_protected and YubiKey::set_pin_last_changed both contained
bugs resulting from the conversion of C pointer logic (incorrect buffer
management). The new Metadata struct holds its own buffer, avoiding the
problem.
Also adds a protected management key integration test.
This factors the junk drawer of constants into the relevant files.
There are still a few "global" ones left but they can be addressed in a
followup commit.
YubiKey NEOs are legacy YubiKey devices, most of which contain
unpatchable security vulnerabilities.
They have smaller buffer sizes than YK4 and YK5, which necessitates a
whole bunch of conditional gating and buffer size calculations.
Getting rid of them simplifies this logic and allows us to assume
consistent buffer sizes everywhere.
We never tested on NEOs anyway, and looking at the deleted code it seems
it may have been miscalculating the NEO's buffer size!
If someone *really* wants to support NEOs, it shouldn't be that hard to
restore, but the codebase is definitely cleaner without it.
This commit contains a "big bang" refactor/rewrite which does the
following:
- Replaces all `SCard*` FFI calls with the `pcsc` crate, which provides
a safe, portable PC/SC API across Windows, macOS, and Linux
- Refactors the `util` module into modules representing the various
device functions and concepts, e.g. `certificate`, `key`, `mgm`
- Replaces all usage of `libc` with `std` functionality, and in many
places rewriting functionality to use safe code.
- Removes `ykpiv_` from all function names, and `Piv*` from type names.
In 20/20 hindsight I wish I had done this commit more incrementally so
as to make it easier to review. Que sera sera.
However, realistically we need to test all functionality on the device
to ensure that it actually works. Going forward I would like to put
pretty much all of the current code behind an `untested` cargo feature,
and then remove it for each bit of functionality we test.
Tony Arcieri (iqlusion)