diff options
| author | 2022-06-28 19:24:19 +0000 | |
|---|---|---|
| committer | 2022-07-02 04:25:11 +0000 | |
| commit | b5a17e8e5c8187b428946ca5dfe007c854c4cbbb (patch) | |
| tree | 4745c9c99b400f0a3f632707e3de2f46c19d76ea /libs/androidfw/ResourceTimer.cpp | |
| parent | 716cce695e7e6b91cfd16d7b7a49ff8a32813822 (diff) | |
Fix leak of RemoteAccessibilityController through SurfacePackage.
When a child SurfacePackage is set on a SurfaceView, the view's
RemoteAccessibilityController calls linkToDeath on the SurfacePackage's
IAccessibilityEmbeddedConnection. The death recipient -- the controller
itself -- is unlinked when the Surface is destroyed -- being notified
through a SurfaceChangedCallback registered on the ViewRootImpl.
There are 2 issues there:
* When the SurfaceView is detached, the SurfaceChangedCallback is
unregistered, so the unlinkToDeath never happens.
* The Surface holding the SurfaceView may never be destroyed, for
example in the case of the IME.
SurfacePackages sent over ipc seem to be used in only 2 places
currently:
* For inlined autofill suggestions in the input method. This bug causes
a serious leak here, since the memory is held in 2 persistent
processes -- autofill Session data in the system server (referenced
from the SurfaceViews), and SurfaceViews in the IME.
* The Wallpaper app.
For now, this is fixed by using a WeakRef for the
RemoteAccessibilityController. The root cause will be fixed in a
followup change in a later release.
Fix: 183402294
Test: manual:
* atest CtsAutoFillServiceTestCases:InlineLoginActivityTest\
--iterations 5
* Force gc: adb shell kill -10 $(adb shell pgrep ext.services mockime\
system_server -d '\ ')
* Rerun that a couple of times so the binder objects get cleared.
* Check for leaks of autofill.Session in system_server and
SurfaceViews or RemoteAccessibilityEmbeddedConnection in
com.android.cts.mockime.
Change-Id: I5447b173313919507969872bdeb7ff3038152c23
Diffstat (limited to 'libs/androidfw/ResourceTimer.cpp')
0 files changed, 0 insertions, 0 deletions