| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This fixes a bug that was causing the ephemeral keystore to retrieve
certs in a format that couldn't be handled by the `KeyMgr`. This caused
all certificate retrievals from `EphemeralKeystore` done via the
`KeyMgr` to fail with an internal error.
For context, the only supported cert type is `TorEd25519Cert`, which is
a pre-encoded certificate (i.e. a type wrapper over a `Vec<u8>`).
These certificates are stored as-is by the Arti native keystore (the
bytes are written to a file on disk). When retrieving a
`TorEd25519Cert`, the Arti keystore uses `parse_certificate_erased()` to
parse the cert into a `ParsedEd25519Cert` before returning it as a
type-erased `ErasedKey`. This works as intended with the `KeyMgr`
retrieval and downcasting logic, which expects the certificate to be
returned in the `ParsedCert` format specified in the `ToEncodableCert`
implementation.
Before this change, the ephemeral keystore, on the other hand, did not
play well with the `KeyMgr` when it came to cert retrieval: it would
incorrectly store the `KeystoreItem` as-is, and retrieve it as an
`ErasedKey` using the `ErasedKey::into_erased()` implementation. This
would then cause the `KeyMgr` to fail to downcast the `ErasedKey` to the
correct type (because the returned erased item was of a different type
than `ParsedCert`).
This commit also removes `KeystoreItem::into_erased()`, which was a
footgun (because certificates are not actually supposed to be retrieved
in the format returned by `CertData::into_erased()`).
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
We want to require that whenever we generate a key that's persistent
(stored in KeyMgr), it's going to be made from a stronger-than-usual
Rng. This trait helps us enforce that.
We also add a FakeEntropicRng struct to use for testing.
Note that this turned up a case that we'd missed, which required
an internal change in tor-hsservice.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
| |
This is necessary because `ParsedCert`s will not be `EncodableItem`s.
This is because we cannot (and don't want to) write certificates that
have not yet been validated to the keystore. They do need to be
retrievable from the keystore though, so we also change `ErasedKey`
to be `Box<dyn ItemType>` instead.
|
| | |
|
|
|
We can't possibly know the `ErrorKind` of such an error, unless we know
where the unsupported key came from. Since we can't know this, we should
let a higher level crate (like `tor-keymgr`) decide the `ErrorKind`
instead.
|
