Android ServiceConnectionLeaked异常的调查分析

本文的调查基于Android P的原生源代码。

本周在处理某一款应用的问题时,遇到一处名为ServiceConnectionLeaked的运行时异常,异常信息如下:

Service com.xxx.xxx.alexaservice.MetroAlexaAudioProviderService has leaked ServiceConnection com.xxx.xxx.connection.MetroConnection@32dc748 that was originally bound here
备注:因涉及三方应用,所以包名以com.xxx.xxx替换

详细信息:

10-29 15:15:19.422  1000  1190  1209 I ActivityManager: Start proc 2308:com.xxx.xxx/u0a165 for activity com.xxx.xxx/com.xxx.xxx.AlexaSettingsLauncherActivity caller=com.android.settings
......
10-29 15:15:20.360 10165  2308  2308 E ActivityThread: Service com.xxx.xxx.alexaservice.MetroAlexaAudioProviderService has leaked ServiceConnection com.xxx.xxx.connection.MetroConnection@32dc748 that was originally bound here
10-29 15:15:20.360 10165  2308  2308 E ActivityThread: android.app.ServiceConnectionLeaked: Service com.xxx.xxx.alexaservice.MetroAlexaAudioProviderService has leaked ServiceConnection com.xxx.xxx.connection.MetroConnection@32dc748 that was originally bound here
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.LoadedApk$ServiceDispatcher.(LoadedApk.java:1618)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.LoadedApk.getServiceDispatcher(LoadedApk.java:1510)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ContextImpl.bindServiceCommon(ContextImpl.java:1670)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ContextImpl.bindService(ContextImpl.java:1623)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.content.ContextWrapper.bindService(ContextWrapper.java:708)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at com.xxx.xxx.alexaservice.MetroAlexaAudioProviderService.connectToMetroService(MetroAlexaAudioProviderService.java:168)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at com.xxx.xxx.alexaservice.MetroAlexaAudioProviderService.doBind(MetroAlexaAudioProviderService.java:68)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at com.xxx.xxx.AlexaAudioProviderService.onBind(Unknown Source:0)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ActivityThread.handleBindService(ActivityThread.java:3626)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ActivityThread.access$1600(ActivityThread.java:207)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1722)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.os.Handler.dispatchMessage(Handler.java:106)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.os.Looper.loop(Looper.java:201)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at android.app.ActivityThread.main(ActivityThread.java:6831)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at java.lang.reflect.Method.invoke(Native Method)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:547)
10-29 15:15:20.360 10165  2308  2308 E ActivityThread:  at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:927)

从异常信息中,粗略看到Service对象MetroAlexaAudioProviderService泄漏了其成员变量MetroConnection@32dc748,这一变量是ServiceConnection类型。有些疑惑。

ServiceConnectionLeaked是继承自AndroidRuntimeException类的子类。



先解释下该应用(包名:com.xxx.xxx)的这段异常堆栈吧,通过反编译apk源代码发现,原来是:

  1. com.xxx.xxx进程启动后,有调用端通过bindService来绑定服务MetroAlexaAudioProviderService(父类:com.xxx.xxx.AlexaAudioProviderService),MetroAlexaAudioProviderService的doBind方法执行中,会调用connectToMetroService方法绑定另外一个新的服务

  2. 绑定新服务过程:ContextWrapper.bindService → ContextImpl.bindService → ContextImpl.bindServiceCommon → android.app.LoadedApk.getServiceDispatcher → android.app.LoadedApk$ServiceDispatcher.最后突然抛出ServiceConnectionLeaked异常,也就是我们看到的上面Java异常栈。

查看栈顶的调用at android.app.LoadedApk$ServiceDispatcher.(LoadedApk.java:1618),这里的代码是:

// android/app/LoadedApk.java
ServiceDispatcher(ServiceConnection conn,
        Context context, Handler activityThread, int flags) {
    mIServiceConnection = new InnerConnection(this);
    mConnection = conn;
    mContext = context;
    mActivityThread = activityThread;
    mLocation = new ServiceConnectionLeaked(null);
    mLocation.fillInStackTrace();              // LoadedApk.java:1618
    mFlags = flags;
}

第1618行是mLocation.fillInStackTrace方法。熟悉此方法的同学了解,这里相当于埋雷,预先收集了执行到此函数的Java栈调用,像是保存了相关快照,但异常并非在这里抛出!

那异常是在哪里抛出的?根据"Service xxx has leaked ServiceConnection xxx that was originally bound here",我们可以发现,真正触发并抛出异常的代码位置是:android/app/LoadedApk.java中 removeContextRegistrations方法,

public void removeContextRegistrations(Context context,
        String who, String what) {
    // ...
    synchronized (mServices) {
        //Slog.i(TAG, "Receiver registrations: " + mReceivers);
        ArrayMap smap =
                mServices.remove(context);
        if (smap != null) {
            for (int i = 0; i < smap.size(); i++) {
                LoadedApk.ServiceDispatcher sd = smap.valueAt(i);
                ServiceConnectionLeaked leak = new ServiceConnectionLeaked(
                        what + " " + who + " has leaked ServiceConnection "
                        + sd.getServiceConnection() + " that was originally bound here");
                leak.setStackTrace(sd.getLocation().getStackTrace());
                Slog.e(ActivityThread.TAG, leak.getMessage(), leak);
                if (reportRegistrationLeaks) {
                    StrictMode.onServiceConnectionLeaked(leak);
                }
                try {
                    ActivityManager.getService().unbindService(
                            sd.getIServiceConnection());
                } catch (RemoteException e) {
                    throw e.rethrowFromSystemServer();
                }
                sd.doForget();
            }
        }
        // ...
    }
}

removeContextRegistrations方法最后触发了ServiceConnectionLeaked异常,那什么时候会调用removeContextRegistrations,并触发这样的异常呢?

通过查看frameworks/base/core/java/android/app/源码,以下是removeContextRegistrations的正常2种使用场景调用:

  1. ActivityThread#handleDestroyActivity → ContextImpl#scheduleFinalCleanup → ContextImpl#performFinalCleanup → LoadedApk#removeContextRegistrations

  2. ActivityThread#handleStopService → ContextImpl#scheduleFinalCleanup → ContextImpl#performFinalCleanup → LoadedApk#removeContextRegistrations

总结来说,应用的Acitivty和Service在销毁时,会最终调用到LoadedApk#removeContextRegistrations方法,这里有可能会抛出ServiceConnectionLeaked异常。



那下一个问题是,为何会抛出ServiceConnectionLeaked异常?它的作用是什么呢?
这里先要学习下Android中bindService、unbindService的代码实现知识:

  1. 首先我们知道,客户端在bindService时,需要生成一个ServiceConnection对象,该对象会封装为一个ServiceDispatcher对象,并将其内部类InnerConnection对象,通过ActivityManager#bindService,binder调用传递给AMS

  2. 注意:内部类InnerConnection对象是一个Binder对象,传递给AMS的目的是,bindService成功执行后,AMS可以通过该对象回调其connected方法,最终在应用端ServiceConnection对象的onServiceConnected方法被调用,一次完成的bindService成功完成!具体可参考:

    • 服务端:com/android/server/am/ActiveServices.java#bindServiceLocked
    • 客户端:android/app/LoadedApk.java#ServiceDispatcher#InnerConnection类的connected方法
  3. 客户端断开服务时,会通过调用unbindService方法,ContextImpl#unbindService执行时,会先调用 LoadedApk#forgetServiceDispatcher方法,该方法作用非常重要:其会将本地ServiceConnection对象与其ServiceDispatcher对象关系从ArrayMap中移除,并执行ServiceDispatcher的doForget方法(主要作用:解除针对服务端Service的死亡回调注册。因为这是正常的unbind,不再需要监听服务端的死亡回调)。

// android/app/ContextImpl.java
@Override
public void unbindService(ServiceConnection conn) {
    if (conn == null) {
        throw new IllegalArgumentException("connection is null");
    }
    if (mPackageInfo != null) {
        IServiceConnection sd = mPackageInfo.forgetServiceDispatcher(
                getOuterContext(), conn);
        try {
            ActivityManager.getService().unbindService(sd);
        } catch (RemoteException e) {
            throw e.rethrowFromSystemServer();
        }
    } else {
        throw new RuntimeException("Not supported in system context");
    }
}



那什么情况下会抛出ServiceConnectionLeaked异常呢?我们看到正常情况下,通过unbindService断开服务时,本地保存的ArrayMap(ArrayMap)信息会被清理、删除。但是ServiceConnectionLeaked异常抛出时,也就是方法

removeContextRegistrations中的逻辑执行时,本地保存的ArrayMap(ArrayMap)信息仍然存在!这里可推断说明:

结论:
应用的Acitivty和Service在销毁前,如果其已绑定一个服务Service,那么应用需主动调用unbindService方法(例如Acitity或Service的onDestroy方法中)解绑。否则销毁走到removeContextRegistrations方法逻辑中时,就会抛出ServiceConnectionLeaked异常,并通过提前埋雷的方式,将相关栈信息打印出来,以便开发者查清原因。



最后,本地Demo复现此问题:
我们在Demo应用(包名:com.kevin.test)的Activity#onCreate中调用bindService来绑定服务,onDestroy方法中未调用unbindService,那么打开页面Activity后,进入最近任务,强制上滑杀掉应用,那么日志中就会发现ServiceConnectionLeaked异常信息了。示例如下:

public class MainActivity extends Activity {
 
    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_main);
 
        TestService service = new TestService(this);
        service.connectService(); //其中会调用mContext.bindService(intent, mConnection, mContext.BIND_AUTO_CREATE);
    }
 
    @Override
    protected void onDestroy() {
        super.onDestroy();
    }
}

抛出的异常信息:

11-04 14:39:20.624 11439 11439 E ActivityThread: Activity com.kevin.test.MainActivity has leaked ServiceConnection com.kevin.test.TestService$2@3201e07 that was originally bound here
11-04 14:39:20.624 11439 11439 E ActivityThread: android.app.ServiceConnectionLeaked: Activity com.kevin.test.MainActivity has leaked ServiceConnection com.kevin.test.TestService$2@3201e07 that was originally bound here
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.LoadedApk$ServiceDispatcher.(LoadedApk.java:1618)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.LoadedApk.getServiceDispatcher(LoadedApk.java:1510)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ContextImpl.bindServiceCommon(ContextImpl.java:1669)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ContextImpl.bindService(ContextImpl.java:1622)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.content.ContextWrapper.bindService(ContextWrapper.java:708)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at com.kevin.test.TestService.connectService(TestService.java:29)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at com.kevin.test.MainActivity.onCreate(MainActivity.java:20)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.Activity.performCreate(Activity.java:7224)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.Activity.performCreate(Activity.java:7213)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.Instrumentation.callActivityOnCreate(Instrumentation.java:1272)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:2926)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ActivityThread.handleLaunchActivity(ActivityThread.java:3081)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.servertransaction.LaunchActivityItem.execute(LaunchActivityItem.java:78)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.servertransaction.TransactionExecutor.executeCallbacks(TransactionExecutor.java:108)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.servertransaction.TransactionExecutor.execute(TransactionExecutor.java:68)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1831)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.os.Handler.dispatchMessage(Handler.java:106)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.os.Looper.loop(Looper.java:201)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at android.app.ActivityThread.main(ActivityThread.java:6806)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at java.lang.reflect.Method.invoke(Native Method)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:547)
11-04 14:39:20.624 11439 11439 E ActivityThread:    at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:873)

这样我们的分析,便得到验证!

最后说明下,此问题影响不大,更多的是暴露了开发者开发时的不规范,未考虑到bindService、unbindService的成对调用或正确时机调用。因为在最后Activity、Service销毁时removeContextRegistrations中会最后调用AMS#unbindService方法,ServiceDispatcher#doForget方法完成补救。但值得开发者注意!





作者:kevin song,2019.11.4于南京建邺区

你可能感兴趣的:(Android ServiceConnectionLeaked异常的调查分析)