On devices with a fingerprint sensor, users can enroll one or more fingerprints and use those fingerprints to unlock the device and perform other tasks. Android uses the Fingerprint Hardware Abstraction Layer (HAL) to connect to a vendor-specific library and fingerprint hardware, e.g. a fingerprint sensor.
To implement the Fingerprint HAL, you must
implement the major functions of
/hardware/libhardware/include/hardware/fingerprint.h in a
The fingerprint sensor of a device is generally idle. However, in response to
a call to the
enroll function, the
fingerprint sensor listens for a touch (the screen might also wake when a user
touches the fingerprint sensor). The high-level flow of fingerprint matching
includes the following steps:
- User places a finger on the fingerprint sensor.
- The vendor-specific library determines if there is a fingerprint match in the current set of enrolled fingerprint templates.
- Matching results are passed to the Fingerprint HAL, which notifies
fingerprintd(the Fingerprint daemon) of a fingerprint authentication (or non-authentication).
This flow assumes a fingerprint has already been enrolled on the device, i.e. the vendor-specific library has enrolled a template for the fingerprint (for details, see Authentication).
The Fingerprint HAL interacts with the following components:
- Biometric API. For devices that launcher with Android 8.1 and lower,
FingerprintManagerinteracts directly with an app in an app process. Each app has an instance of
FingerprintManager, a wrapper that communicates with
Devices that ship with Android 9 and higher, should use the
BiometricPromptAPI instead of
FingerprintService. Singleton service that operates in the system process, which handles communication with
fingerprintd. C/C++ implementation of the binder interface from FingerprintService. The
fingerprintddaemon operates in its own process and wraps the Fingerprint HAL vendor-specific library.
- Fingerprint HAL vendor-specific library. Hardware vendor's implementation of the Fingerprint HAL. The vendor-specific library communicates with the device-specific hardware.
- Keystore API and Keymaster. Components that provide hardware-backed cryptography for secure key storage in a Trusted Execution Environment (TEE).
A vendor-specific HAL implementation must use the communication protocol required by a TEE. Thus, raw images and processed fingerprint features must not be passed in untrusted memory. All such biometric data needs to be secured within sensor hardware or trusted memory. (Memory inside the TEE is considered trusted; memory outside the TEE is considered untrusted.) Rooting must not compromise biometric data.
fingerprintd makes calls through the Fingerprint HAL to the
vendor-specific library to enroll fingerprints and perform other operations:
The following Fingerprint HAL guidelines are designed to ensure that fingerprint data is not leaked and is removed when a user is removed from a device:
- Raw fingerprint data or derivatives (e.g. templates) must never be accessible from outside the sensor driver or TEE. If the hardware supports it, hardware access must be limited to the TEE and protected by an SELinux policy. The Serial Peripheral Interface (SPI) channel must be accessible only to the TEE and there must be an explicit SELinux policy on all device files.
- Fingerprint acquisition, enrollment, and recognition must occur inside the TEE.
- Only the encrypted form of the fingerprint data can be stored on the file system, even if the file system itself is encrypted.
- Fingerprint templates must be signed with a private, device-specific key. For Advanced Encryption Standard (AES), at a minimum a template must be signed with the absolute file-system path, group, and finger ID such that template files are inoperable on another device or for anyone other than the user that enrolled them on the same device. For example, copying fingerprint data from a different user on the same device or from another device must not work.
- Implementations must either use the file system path provided by the
set_active_group()function or provide a way to erase all user template data when the user is removed. It is strongly recommended that fingerprint template files be stored as encrypted in the path provided. If this is infeasible due to TEE storage requirements, the implementer must add hooks to ensure removal of the data when the user is removed.
Fingerprint HAL functions
The Fingerprint HAL contains the following major functions in
enroll. Switches the HAL state machine to start the collection and storage of a fingerprint template. When enrollment is complete, or after a timeout, the HAL state machine returns to the idle state.
pre_enroll. Generates a unique token to indicate the start of a fingerprint enrollment. Provides a token to the
enrollfunction to ensure there was prior authentication, e.g. using a password. To prevent tampering, the token is wrapped (e.g. HMAC'd) after the device credential is confirmed. The token must be checked during enrollment to verify the token is still valid.
get_authenticator_id. Returns a token associated with the current fingerprint set.
cancel. Cancels pending enroll or authenticate operations. The HAL state machine is returned to the idle state.
enumerate. Synchronous call for enumerating all known fingerprint templates.
remove. Deletes a fingerprint template.
set_active_group. Restricts a HAL operation to a set of fingerprints that belong to a specified group, identified by a group identifier (GID).
authenticate. Authenticates a fingerprint-related operation (identified by an operation ID).
set_notify. Registers a user function that receives notifications from the HAL. If the HAL state machine is in a busy state, the function is blocked until the HAL leaves the busy state.
For details on these functions, refer to the comments in