| Age | Commit message (Collapse) | Author |
|---|
|
Introduce platform-neutral C library ("libeic") which can be used to
implement an Identity Credential Trusted Application/Applet in Secure
Hardware.
The libeic library is intentionally low-level, has no dependencies
(not even libc), uses very little run-time memory (less than 500 bytes
during a provisioning or presentation session), and doesn't
dynamically allocate any memory. Crypto routines are provided by the
library user through a simple crypto interface defined in EicOps.
Also provide an Android-side HAL implementation designed to
communicate with libeic running in Secure Hardware outside
Android. Abstract out communications between HAL and TA in a couple of
SecureHardwareProxy* classes which mimic libeic 1:1.
The default implementation of the HAL is a combination of the
aforementioned HAL using libeic in-process backed by BoringSSL for the
crypto bits.
Test: atest VtsHalIdentityTargetTest
Test: atest android.security.identity.cts
Bug: 170146643
Change-Id: I3bf43fa7fd9362f94023052591801f2094a04607
|
|
- The docs said that IdentityCredential.createEphemeralKey() returned
data encoded PKCS#8 which is wrong. It's supposed to be in DER format
which is also what the VTS tests and credstore expects.
- Clarify that createEphemeralKeyPair(), setReaderEphemeralPublicKey(),
and createAuthChallenge() are all optional.
- Avoid passing an invalid profile ID in the IdentityCredentialTests.
verifyOneProfileAndEntryPass test.
- Update requirements for which tags must be present in the attestation
for CredentialKey as well as the requirements on expiration date and
the issuer name. Update default implementation to satisfy these
requirements. Update VTS tests to carefully verify these requrements
are met.
- Clarify requirements for X.509 cert for AuthenticationKey. Add VTS
test to verify.
- Mandate that TAG_IDENTITY_CREDENTIAL_KEY must not be set for test
credentials. Add VTS test to verify this.
- Make default implementation pretend to be implemented in a trusted
environment and streamline VTS tests to not special-case for the
default implementation.
- Switch to using the attestation extension parser from the KM 4.1
support library instead of the one from system/keymaster. The latter
one did not support the latest attestation extension and thus would
fail for pretty much anything that wasn't the default HAL impl.
- Fix a couple of bugs in keymaster::V4_1::parse_attestation_record():
- Report root_of_trust.security_level
- Add support for Tag::IDENTITY_CREDENTIAL_KEY
- Fix how EMacKey is calculated.
- Add test vectors to verify how EMacKey and DeviceMac is calculated.
Test: atest VtsHalIdentityTargetTest
Test: atest android.security.identity.cts
Bug: 171745570
Change-Id: I2f8bd772de078556733f769cec2021918d1d7de6
|
|
These updates are based on input/experiences implementing this
HAL. There are no API changes.
- Specify that the validity for credentialKey certificate shall be
from current time and expire at the same time as the attestation
batch certificate.
- Require challenge passed to getAttestationCertificate() is
non-empty.
- Fix bug in VTS tests where the startPersonlization() result was not
checked.
- Remove verifyStartPersonalizationZero test since it cannot be
completed.
- Ensure secureUserId is non-zero if user authentication is needed.
- Specify format for signingKeyBlob in generateSigningKeyPair() same
way we do for credentialData in finishAddingEntries().
- Modify EndToEndTest to decrypt/unpack credentialData to obtain
credentialPrivKey and storageKey and do cross-checks on these.
- Modify EndToEndTest to decrypt/unpack signingKeyBlob to obtain
signingKeyPriv and check it matches the public key in the returned
certificate.
- Add new VTS tests for user and reader authentication.
- Relax unnecessary requirements about SessionTranscript structure -
just require it has X and Y of the ephemeral key created earlier.
- Allow calls in VTS tests to v2 HAL to fail - this should allow
these VTS tests to pass on a compliant v1 HAL.
Bug: 156911917
Bug: 158107945
Test: atest VtsHalIdentityTargetTest
Test: atest android.security.identity.cts
Change-Id: I11b79dbd57b1830609c70301fea9c99f9e5080cb
|
|
Without this extra information passed upfront it's not practical to
implement a HAL which incrementally builds up cryptographically
authenticated data.
Two new methods are added to facilitate this and the HAL version
number is bumped to 2.
Bug: 154631410
Test: atest VtsHalIdentityTargetTest
Test: atest android.security.identity.cts
Change-Id: Iff63dfa2c4485c8768e06e7f6d70e940cfc8f68e
|
|
In order to implement Identity Credential on resource-restricted
secure hardware, we need to limit the number of possible
AccessControlProfile in a credential. A limit of 32 means that such
hardware only need to devote four bytes of RAM for a bitmask with
information about which profiles are authorized.
Document this, add new VTS test, and update the default
implementation.
Bug: 155100967
Test: atest android.security.identity.cts
Test: atest VtsHalIdentityTargetTest
Change-Id: Ia4f2ee0013b330561df744e0595f298a0d156122
|
|
Fixed following bugs in WritableIdentityCredential.cpp
- Do not allow startPersonalization to be called more than once per
aidl.
- Do not preceed with beginAddEntry if addAccessControlProfile and
startPersonalization profile count mismatch.
- Verify access control profile ids are unique.
- Do not let empty name space to mess up beginAddEntry.
- Do not allow beginAddEntry to add entries interleaving namespace
groupings. Enforce all entries must be added in namespace "groups"
per aidl.
- Fix counting error that allowed one entries to be added per name
space than startPersonalization limit.
- Do not approve finishAddingEntries if there are more profiles or
entries to be added than startPersonalization set accounting.
- Add testing utilities library for identity credential.
- Refactored end to end tests.
Test: atest VtsHalIdentityTargetTest
Test: atest android.security.identity.cts
Change-Id: I51902681776c6230e49589fc75a8145e79d7d1a6
|
|
In native world, byte stream is typically represented in uint8_t[]
or vector<uint8_t>. C++ backend already generates that way. This
change involves NDK backend.
Now NDK backend also uses vector<uint8_t> just like C++ backend.
Bug: 144957764
Test: atest CtsNdkBinderTestCases
Merged-In: I8de348b57cf92dd99b3ee16252f56300ce5f4683
Change-Id: I8de348b57cf92dd99b3ee16252f56300ce5f4683
(cherry picked from commit 9070318462e5e73acf1509cf7e75ac260e51e43a)
Exempt-From-Owner-Approval: cp from internal
|
|
Bug: 149908474
Test: atest android.security.identity.cts.AttestationTest
Test: atest VtsHalIdentityCredentialTargetTest
Test: atest android.hardware.identity-support-lib-test
Change-Id: I18c5d05d806d4157c9dce42a398cc89421e26907
|
|
This includes add a partial types-only HAL for KeyMaster for
HardwareAuthToken.
Bug: 111446262
Test: atest android.security.identity.cts
Test: VtsHalIdentityTargetTest
Test: android.hardware.identity-support-lib-test
Change-Id: I7a6254d33200bfd62269aed1957cbb2a84b16272
|
