tree: 625dacd96593a7d40ad8d3e7624b73b8c2d74676 [path history] [tgz]
  1. BUILD.gn
  2. DEPS
  3. OWNERS
  4. README.md
  5. navigator_wake_lock.cc
  6. navigator_wake_lock.h
  7. navigator_wake_lock.idl
  8. wake_lock.cc
  9. wake_lock.h
  10. wake_lock.idl
  11. wake_lock_manager.cc
  12. wake_lock_manager.h
  13. wake_lock_manager_test.cc
  14. wake_lock_sentinel.cc
  15. wake_lock_sentinel.h
  16. wake_lock_sentinel.idl
  17. wake_lock_sentinel_test.cc
  18. wake_lock_test.cc
  19. wake_lock_test_utils.cc
  20. wake_lock_test_utils.h
  21. wake_lock_type.cc
  22. wake_lock_type.h
  23. worker_navigator_wake_lock.cc
  24. worker_navigator_wake_lock.h
  25. worker_navigator_wake_lock.idl
third_party/blink/renderer/modules/wake_lock/README.md

modules/wake_lock

This directory contains an implementation of the Wake Lock specification, a Web API that allows script authors to prevent both the screen from turning off (“screen wake locks”) as well as the CPU from entering a deep power state (“system wake locks”). There are platform implementations for ChromeOS, Linux (X11), Mac, Android, and Windows.

At the time of writing (October 2019), system wake lock requests are always denied, as allowing them depends on a proper permission model for the requests being figured out first.

The code required to implement the Wake Lock API is spread across multiple Chromium subsystems: Blink, //content, //services and //chrome. This document focuses on the Blink part, and the other subsystems are mentioned when necessary but without much detail.

High level overview

Wake Lock's API surface is fairly small: Navigator and WorkerNavigator provide a wakeLock attribute that exposes the WakeLock, an interface with a single method to request a wake lock, and a WakeLockSentinel can be used to both release the requested lock and receive an event when it is released. All the parts that actually communicate with platform APIs are implemented elsewhere, so the Blink side only exposes the JavaScript API to script authors, validates API calls and manages state records.

Wake Lock usage in scripts looks like this:

let lock = await navigator.wakeLock.request("screen");
lock.addEventListener("release", (ev) => {
  console.log(`${ev.target.type} wake lock released`);
});
lock.release();

The three main Blink classes implementing the spec are:

  • WakeLock: per-Navigator/WorkerNavigator class that implements the bindings for the WakeLock IDL interface. Its responsibilities also include performing permission requests and Wake Lock management tasks that apply to documents and/or workers (e.g. page visibility handling). Lock acquisition calls are all forwarded to WakeLockManager. Its managers_ array contains per-wake lock type WakeLockManager instances, so all wake lock types are managed independently.
  • WakeLockManager: Owned by WakeLock. This is an implementation of the state records Wake Lock concept in a per-type fashion. Like in the spec, it keeps track of all active locks of a certain type, and it is also responsible for communicating with the //content and //services layers to request and cancel wake locks.
  • WakeLockSentinel: Owned by WakeLockManager. This is an implementation of the WakeLockSentinel IDL interface that is used to both release a lock requested by WakeLock.request() and receive a release event when it is released (either by WakeLockSentinel.release() or due to a platform event, such as a loss of context, or a page visibility change in the case of screen wake locks). This class is an event target, so it inherits from ActiveScriptWrappable to avoid being garbage-collected while there is pending activity.

Furthermore, wake_lock.mojom defines the Mojo interface implemented by //content's WakeLockServiceImpl that WakeLockManager uses to obtain a device::mojom::blink::WakeLock and request/cancel a wake lock.

The rest of the implementation is found in the following directories:

  • content/browser/wake_lock implements the WakeLockService Mojo interface defined in Blink. It is responsible for communicating with Blink and connecting Blink to services/device/wake_lock.
  • services/device/wake_lock contains the platform-specific parts of the implementation and implements the Wake Lock Mojo interfaces.
  • chrome/browser/wake_lock contains the Chrome-specific side of permission management for Wake Locks. When the Blink implementation needs to either query or request permission for wake locks, the request bubbles up to this directory, where the decision is made based on the wake lock type (for testing purposes, content_shell always grants screen wake locks and denies system wake locks in shell_permission_manager.cc).

Testing

Validation, exception types, feature policy integration and general IDL compliance are tested in web platform tests, while Chromium-specific implementation details (e.g. permission handling) are tested in web tests.

Larger parts of the Blink implementation are tested as browser and unit tests:

  • *_test.cc and wake_lock_test_utils.{cc,h} are built as part of the blink_unittests GN target, and attempt to have coverage over most of the code in this directory.
  • The unit tests in services/device/wake_lock test the service side of the API implementation.
  • chrome/browser/wake_lock has unit tests for WakeLockPermissionContext, and browser tests for end-to-end behavior testing.
  • content_shell implements its own permission logic that mimics what is done in //chrome in shell_permission_manager.cc.

Example workflows

Wake Lock acquisition

This section describes how the classes described above interact when the following excerpt is run in the browser:

const lock = await navigator.wakeLock.request("screen");
  1. WakeLock::request() performs all the validation steps described in the spec. If all checks have passed, it creates a ScriptPromiseResolver and calls WakeLock::DoRequest().
  2. WakeLock::DoRequest() simply forwards its arguments to WakeLock::ObtainPermission(). It exists as a separate method just to make writing unit tests easier, as we'd otherwise be unable to use our own ScriptPromiseResolvers in tests.
  3. WakeLock::ObtainPermission() connects to the permission service and asynchronously requests permission for a screen wake lock.
  4. In the browser process, the permission request bubbles up through //content and reaches //chrome's WakeLockPermissionContext, where WakeLockPermissionContext::GetPermissionStatusInternal() always grants CONTENT_SETTINGS_TYPE_WAKE_LOCK_SCREEN permission requests.
  5. Back in Blink, the permission request callback in this case is WakeLock::DidReceivePermissionResponse(). It performs some sanity checks such as verifying if the page visibility changed while waiting for the permission request to be processed. If any of the checks fail, or if the permission request was denied, the ScriptPromiseResolver instance created earlier by WakeLock::request() is rejected and we stop here. If everything went well, WakeLockManager::AcquireWakeLock() is called.
  6. If there are no existing screen wake locks, WakeLockManager::AcquireWakeLock() will connect to the WakeLockService Mojo interface, invoke its GetWakeLock() method to obtain a device::mojom::blink::WakeLock and call its RequestWakeLock() method.
  7. WakeLockManager::AcquireWakeLock() creates a new WakeLockSentinel instance, passing this as the WakeLockSentinel's WakeLockManager. This new WakeLockSentinel is added to its set of active locks.
  8. The ScriptPromiseResolver created by WakeLock::request() is resolved with the new WakeLockSentinel object.

Wake Lock cancellation

Given the excerpt below:

const lock = await navigator.wakeLock.request("screen");
await lock.release();

This section describes what happens when lock.release() is called.

  1. lock.release() results in a call to WakeLockSentinel::release().
  2. WakeLockSentinel::release() calls WakeLockSentinel::DoRelease() and returns a resolved promise. WakeLockSentinel::DoRelease() exists as a separate method because it is also called directly by WakeLockManager::ClearWakeLocks() when an event such as a page visibility change causes all screen wake locks to be released.
  3. WakeLockSentinel::DoRelease() aborts early if its manager_ member is not set. This can happen if WakeLock::release() has already been called before, or if WakeLockManager::ClearWakeLocks() has already released this WakeLockSentinel.
  4. WakeLockSentinel::DoRelease() calls WakeLockManager::UnregisterSentinel().
  5. WakeLockManager::UnregisterSentinel() implements the spec's release wake lock algorithm. If the given WakeLockSentinel is in WakeLockManager's wake_lock_sentinels_, it will be removed and, if the list is empty, WakeLockManager will communicate with its device::mojom::blink::WakeLock instance and call its CancelWakeLock() method.
  6. Back in WakeLockSentinel::DoRelease(), it then clears its manager_ member, and dispatches a release event with itself as a target.

Other Wake Lock usage in Chromium

Inside Blink

Video playback via the <video> tag currently uses a screen wake lock behind the scenes to prevent the screen from turning off while a video is being played. The implementation can be found in the VideoWakeLock class.

This is an implementation detail, but the code handling wake locks in VideoWakeLock is similar to WakeLockManager's, where Blink needs to talk to //content to connect to a WakeLockServiceImpl and use that to get to a device::mojom::blink::WakeLock.

Note: when writing new code that uses Wake Locks in Blink, it is recommended to follow the same pattern outlined above. That is, connect to WakeLockService via the //content layer and request a device::mojom::blink::WakeLock via WakeLockService::GetWakeLock(). Do not go through the classes in this module, and do not connect to the Wake Lock services directly. See the example below:

mojo::Remote<mojom::blink::WakeLockService> wake_lock_service;
mojo::Remote<device::mojom::blink::WakeLock> wake_lock;
execution_context->GetInterface(wake_lock_service.BindNewPipeAndPassReceiver());
wake_lock_service->GetWakeLock(..., wake_lock.BindNewPipeAndPassReceiver());
wake_lock_->RequestWakeLock();

Outside Blink

The Wake Lock service is also used by multiple parts of Chromium outside Blink. In other words, it is possible to use the Wake Lock service and prevent screen and CPU from entering a deep power state directly from the browser side.

In fact, this is why WakeLockProvider::GetWakeLockWithoutContext() exists in the first place. One consequence is that one needs to bear in mind these other usages when changing the public API exposed by the Wake Lock service.

Note: Avoid using WakeLockProvider::GetWakeLockWithoutContext() whenever possible. By design, it does not work on Android.

Example usage outside Blink includes:

Permission Model

The Wake Lock API spec checks for user activation in the context of wake lock permission requests, as a result of a call to WakeLock.request(). If a user agent is configured to prompt a user when a wake lock is requested, user activation is required, otherwise the request will be denied.

In the Chromium implementation, there currently is no “prompt” state, and no permission UI or settings: wake lock requests are either always granted or always denied:

  • Screen wake lock request are always granted without prompting or user activation checks. This is based on the existing precedent of the <video> tag's use of VideoWakeLocks: they are always requested and granted transparently, so even if the Wake Lock API implementation in Chromium started requiring stricter checks, malicious actors could still embed a <video> tag and prevent the screen from turning off without any user interaction.

  • System wake lock requests are always denied in chrome/browser/wake_lock/wake_lock_permission_context.cc. This means the entirety of the code is present and enabled in Blink, but all calls to WakeLock.request('system') currently return a promise that will be rejected with a NotAllowedError. Changing that requires figuring out a permission model for system wake lock requests, which, at the moment, is future work.