服务报价 | 域名主机 | 网络营销 | 软件工具| [加入收藏]
 热线电话: #

理解 Android Crash处理流程

时间:2016-10-05 06:29来源:未知 作者:最模板 点击:
基于Android 6.0的源码剖析, 分析Android应用Crash是如何处理的。 /frameworks/base/core/java/com/android/internal/os/RuntimeInit.java/frameworks/base/core/java/android/app/ActivityManagerNative.java (含内部类AMP)/framewo

基于Android 6.0的源码剖析, 分析Android应用Crash是如何处理的。

/frameworks/base/core/java/com/android/internal/os/RuntimeInit.java
/frameworks/base/core/java/android/app/ActivityManagerNative.java (含内部类AMP)
/frameworks/base/core/java/android/app/ApplicationErrorReport.java
 
/frameworks/base/services/core/java/com/android/server/
    - am/ActivityManagerService.java
    - am/ProcessRecord.java
    - am/ActivityRecord.java
    - am/ActivityStackSupervisor.java
    - am/ActivityStack.java
    - am/ActivityRecord.java
    - am/BroadcastQueue.java
    - wm/WindowManagerService.java
 
/libcore/libart/src/main/java/java/lang/Thread.java

一、概述

App crash(全称Application crash), 对于Crash可分为native crash和framework crash(包含app crash在内),对于crash相信很多app开发者都会遇到,那么上层什么时候会出现crash呢,系统又是如何处理crash的呢。例如,在app大家经常使用 try...catch语句,那么如果没有有效catch exception,就是导致应用crash,发生没有catch exception,系统便会来进行捕获,并进入crash流程。如果你是从事Android系统开发或者架构相关工作,或者遇到需要解系统性的疑难杂症,那么很有必要了解系统Crash处理流程,知其然还需知其所以然;如果你仅仅是App初级开发,可能本文并非很适合阅读,整个系统流程错中复杂。

在Android系统启动系列文章,已讲述过上层应用都是由Zygote fork孵化而来,分为system_server系统进程和各种应用进程,在这些进程创建之初会设置未捕获异常的处理器,当系统抛出未捕获的异常时,最终都交给异常处理器。

  • 对于system_server进程:文章 Android系统启动-SystemServer上篇 ,system_server启动过程中由RuntimeInit.java的 commonInit 方法设置UncaughtHandler,用于处理未捕获异常;
  • 对于普通应用进程:文章 理解Android进程创建流程 ,进程创建过程中,同样会调用RuntimeInit.java的 commonInit 方法设置UncaughtHandler。

1.1 crash调用链

crash流程的方法调用关系来结尾:

AMP.handleApplicationCrash
    AMS.handleApplicationCrash
        AMS.findAppProcess
        AMS.handleApplicationCrashInner
            AMS.addErrorToDropBox
            AMS.crashApplication
                AMS.makeAppCrashingLocked
                    AMS.startAppProblemLocked
                    ProcessRecord.stopFreezingAllLocked
                        ActivityRecord.stopFreezingScreenLocked
                            WMS.stopFreezingScreenLocked
                                WMS.stopFreezingDisplayLocked
                    AMS.handleAppCrashLocked
                mUiHandler.sendMessage(SHOW_ERROR_MSG)
 
Process.killProcess(Process.myPid());
System.exit(10);

接下来说说这个过程。

二、Crash处理流程

那么接下来以commonInit()方法为起点来展开说明。

1. RuntimeInit.commonInit

public class RuntimeInit {
    ...
    private static final void commonInit() {
        //设置默认的未捕获异常处理器,UncaughtHandler实例化过程【见小节2】
        Thread.setDefaultUncaughtExceptionHandler(new UncaughtHandler());
        ...
    }
}

setDefaultUncaughtExceptionHandler()只是将异常处理器handler对象赋给Thread成员变量,即 Thread.defaultUncaughtHandler = new UncaughtHandler() 。接下来看看UncaughtHandler对象实例化过程。

2. UncaughtHandler

[–>RuntimeInit.java]

private static class UncaughtHandlerimplements Thread.UncaughtExceptionHandler {
    //覆写接口方法
    public void uncaughtException(Thread t, Throwable e) {
        try {
            //保证crash处理过程不会重入
            if (mCrashing) return;
            mCrashing = true;
 
            if (mApplicationObject == null) {
                //system_server进程
                Clog_e(TAG, "*** FATAL EXCEPTION IN SYSTEM PROCESS: " + t.getName(), e);
            } else {
                //普通应用进程
                StringBuildermessage = new StringBuilder();
                message.append("FATAL EXCEPTION: ").append(t.getName()).append("\n");
                final String processName = ActivityThread.currentProcessName();
                if (processName != null) {
                    message.append("Process: ").append(processName).append(", ");
                }
                message.append("PID: ").append(Process.myPid());
                Clog_e(TAG, message.toString(), e);
            }
 
            //启动crash对话框,等待处理完成 【见小节2.1和3】
            ActivityManagerNative.getDefault().handleApplicationCrash(
                    mApplicationObject, new ApplicationErrorReport.CrashInfo(e));
        } catch (Throwablet2) {
            ...
        } finally {
            //确保当前进程彻底杀掉【见小节11】
            Process.killProcess(Process.myPid());
            System.exit(10);
        }
    }
}
  1. 当system进程crash的信息:
    • 开头 *** FATAL EXCEPTION IN SYSTEM PROCESS [线程名] ;
    • 接着输出发生crash时的调用栈信息;
  2. 当app进程crash时的信息:
    • 开头 FATAL EXCEPTION: [线程名] ;
    • 紧接着 Process: [进程名], PID: [进程id] ;
    • 最后输出发生crash时的调用栈信息。

看到这里,你就会发现要从log中搜索crash信息,只需要搜索关键词 FATAL EXCEPTION ;如果需要进一步筛选只搜索系统crash信息,则可以搜索的关键词可以有多样,比如 *** FATAL EXCEPTION 。

当输出完crash信息到logcat里面,这只是crash流程的刚开始阶段,接下来弹出 crash对话框 ,ActivityManagerNative.getDefault()返回的是ActivityManagerProxy(简称AMP ), AMP 经过binder调用最终交给ActivityManagerService(简称 AMS )中相应的方法去处理,故接下来调用的是AMS.handleApplicationCrash()。

2.1 CrashInfo

[-> ApplicationErrorReport.java]

public class ApplicationErrorReport implements Parcelable {
    ...
    public static class CrashInfo {
        public CrashInfo(Throwabletr) {
            StringWritersw = new StringWriter();
            PrintWriterpw = new FastPrintWriter(sw, false, 256);
            tr.printStackTrace(pw); //输出栈trace
            pw.flush();
            stackTrace = sw.toString();
            exceptionMessage = tr.getMessage();
 
            ThrowablerootTr = tr;
            while (tr.getCause() != null) {
                tr = tr.getCause();
                if (tr.getStackTrace() != null && tr.getStackTrace().length > 0) {
                    rootTr = tr;
                }
                String msg = tr.getMessage();
                if (msg != null && msg.length() > 0) {
                    exceptionMessage = msg;
                }
            }
 
            exceptionClassName = rootTr.getClass().getName();
            if (rootTr.getStackTrace().length > 0) {
                StackTraceElementtrace = rootTr.getStackTrace()[0];
                throwFileName = trace.getFileName();
                throwClassName = trace.getClassName();
                throwMethodName = trace.getMethodName();
                throwLineNumber = trace.getLineNumber();
            } else {
                throwFileName = "unknown";
                throwClassName = "unknown";
                throwMethodName = "unknown";
                throwLineNumber = 0;
            }
        }
        ...
    }
}

将crash信息 文件名 , 类名 , 方法名 , 对应行号 以及 异常信息 都封装到CrashInfo对象。

3. handleApplicationCrash

[–>ActivityManagerService.java]

public void handleApplicationCrash(IBinderapp, ApplicationErrorReport.CrashInfocrashInfo) {
    //获取进程record对象【见小节3.1】
    ProcessRecord r = findAppProcess(app, "Crash");
    final String processName = app == null ? "system_server"
            : (r == null ? "unknown" : r.processName);
    //【见小节4】
    handleApplicationCrashInner("crash", r, processName, crashInfo);
}

关于进程名(processName):

  1. 当远程IBinder对象为空时,则进程名为 system_server ;
  2. 当远程IBinder对象不为空,且ProcessRecord为空时,则进程名为 unknown ;
  3. 当远程IBinder对象不为空,且ProcessRecord不为空时,则进程名为ProcessRecord对象中相应进程名。

3.1 findAppProcess

[–>ActivityManagerService.java]

private ProcessRecordfindAppProcess(IBinderapp, String reason) {
    if (app == null) {
        return null;
    }
 
    synchronized (this) {
        final int NP = mProcessNames.getMap().size();
        for (int ip=0; ip<NP; ip++) {
            SparseArray<ProcessRecord> apps = mProcessNames.getMap().valueAt(ip);
            final int NA = apps.size();
            for (int ia=0; ia<NA; ia++) {
                ProcessRecord p = apps.valueAt(ia);
                //当找到目标进程则返回
                if (p.thread != null && p.thread.asBinder() == app) {
                    return p;
                }
            }
        }
        //如果代码执行到这里,表明无法找到应用所在的进程
        return null;
    }
}

其中 mProcessNames = new ProcessMap<ProcessRecord>(); 对于代码 mProcessNames.getMap() 返回的是 mMap ,而 mMap= new ArrayMap<String, SparseArray<ProcessRecord>>() ;

知识延伸: SparseArray 和 ArrayMap 是Android专门针对内存优化而设计的取代Java API中的 HashMap 的数据结构。对于key是int类型则使用 SparseArray ,可避免自动装箱过程;对于key为其他类型则使用 ArrayMap 。 HashMap 的查找和插入时间复杂度为O(1)的代价是牺牲大量的内存来实现的,而 SparseArray 和 ArrayMap 性能略逊于 HashMap ,但更节省内存。

再回到 mMap ,这是以进程name为key,再以(uid为key,以ProcessRecord为Value的)结构体作为value。下面看看其get()和put()方法

//获取mMap中(name,uid)所对应的ProcessRecord
public ProcessRecordget(String name, int uid) {};
//将(name,uid, value)添加到mMap
public ProcessRecordput(String name, int uid, ProcessRecordvalue) {};

findAppProcess()根据app(IBinder类型)来查询相应的目标对象ProcessRecord。

有了进程记录对象ProcessRecord和进程名processName,则进入执行Crash处理方法,继续往下看。

4. handleApplicationCrashInner

[–>ActivityManagerService.java]

void handleApplicationCrashInner(String eventType, ProcessRecord r, String processName,
        ApplicationErrorReport.CrashInfocrashInfo) {
    //将Crash信息写入到Event log
    EventLog.writeEvent(EventLogTags.AM_CRASH,...);
    //将错误信息添加到DropBox
    addErrorToDropBox(eventType, r, processName, null, null, null, null, null, crashInfo);
    //【见小节5】
    crashApplication(r, crashInfo);
}

其中 addErrorToDropBox 是将crash的信息输出到目录 /data/system/dropbox 。例如system_server的dropbox文件名为system_server_crash@xxx.txt (xxx代表的是时间戳)

5. crashApplication

[–>ActivityManagerService.java]

private void crashApplication(ProcessRecord r, ApplicationErrorReport.CrashInfocrashInfo) {
    long timeMillis = System.currentTimeMillis();
    String shortMsg = crashInfo.exceptionClassName;
    String longMsg = crashInfo.exceptionMessage;
    String stackTrace = crashInfo.stackTrace;
    if (shortMsg != null && longMsg != null) {
        longMsg = shortMsg + ": " + longMsg;
    } else if (shortMsg != null) {
        longMsg = shortMsg;
    }
 
    AppErrorResultresult = new AppErrorResult();
    synchronized (this) {
        //清除远程调用者uid和pid信息,并保存到origId
        final long origId = Binder.clearCallingIdentity();
        ...
 
        //【见小节6】
        if (r == null || !makeAppCrashingLocked(r, shortMsg, longMsg, stackTrace)) {
            Binder.restoreCallingIdentity(origId);
            return;
        }
 
        Messagemsg = Message.obtain();
        msg.what = SHOW_ERROR_MSG;
        HashMapdata = new HashMap();
        data.put("result", result);
        data.put("app", r);
        msg.obj = data;
        //发送消息SHOW_ERROR_MSG,弹出提示crash的对话框,等待用户选择【见小节10】
        mUiHandler.sendMessage(msg);
        //恢复远程调用者uid和pid
        Binder.restoreCallingIdentity(origId);
    }
 
    //进入阻塞等待,直到用户选择crash对话框"退出"或者"退出并报告"
    int res = result.get();
 
    IntentappErrorIntent = null;
    synchronized (this) {
        if (r != null && !r.isolated) {
            // 将崩溃的进程信息保存到mProcessCrashTimes
            mProcessCrashTimes.put(r.info.processName, r.uid,
                    SystemClock.uptimeMillis());
        }
        if (res == AppErrorDialog.FORCE_QUIT_AND_REPORT) {
            //创建action="android.intent.action.APP_ERROR",组件为r.errorReportReceiver的Intent
            appErrorIntent = createAppErrorIntentLocked(r, timeMillis, crashInfo);
        }
    }
 
    if (appErrorIntent != null) {
        try {
            //启动Intent为appErrorIntent的Activity
            mContext.startActivityAsUser(appErrorIntent, new UserHandle(r.userId));
        } catch (ActivityNotFoundException e) {
            Slog.w(TAG, "bug report receiver dissappeared", e);
        }
    }
}

该方法主要做的两件事:

  • 调用 makeAppCrashingLocked ,继续处理crash流程;
  • 发送消息SHOW_ERROR_MSG,弹出提示crash的对话框,等待用户选择;

6. makeAppCrashingLocked

[–>ActivityManagerService.java]

private boolean makeAppCrashingLocked(ProcessRecordapp,
        String shortMsg, String longMsg, String stackTrace) {
    app.crashing = true;
    //封装crash信息到crashingReport对象
    app.crashingReport = generateProcessError(app,
            ActivityManager.ProcessErrorStateInfo.CRASHED, null, shortMsg, longMsg, stackTrace);
    //【见小节7】
    startAppProblemLocked(app);
    //停止屏幕冻结【见小节8】
    app.stopFreezingAllLocked();
    //【见小节9】
    return handleAppCrashLocked(app, "force-crash", shortMsg, longMsg, stackTrace);
}

7. startAppProblemLocked

[–>ActivityManagerService.java]

void startAppProblemLocked(ProcessRecordapp) {
    app.errorReportReceiver = null;
 
    for (int userId : mCurrentProfileIds) {
        if (app.userId == userId) {
            //获取当前用户下的crash应用的error receiver【见小节7.1】
            app.errorReportReceiver = ApplicationErrorReport.getErrorReportReceiver(
                    mContext, app.info.packageName, app.info.flags);
        }
    }
    //忽略当前app的广播接收【见小节7.2】
    skipCurrentReceiverLocked(app);
}

该方法主要功能:

  • 获取当前用户下的crash应用的error receiver;
  • 忽略当前app的广播接收;

7.1 getErrorReportReceiver

[-> ApplicationErrorReport.java]

public static ComponentNamegetErrorReportReceiver(Contextcontext,
        String packageName, int appFlags) {
    //检查Settings中的"send_action_app_error"是否使能错误报告的功能
    int enabled = Settings.Global.getInt(context.getContentResolver(),
            Settings.Global.SEND_ACTION_APP_ERROR, 0);
    if (enabled == 0) {
        //1.当未使能时,则直接返回
        return null;
    }
 
    PackageManagerpm = context.getPackageManager();
 
    String candidate = null;
    ComponentNameresult = null;
    try {
        //获取该crash应用的安装器的包名
        candidate = pm.getInstallerPackageName(packageName);
    } catch (IllegalArgumentException e) {
    }
 
    if (candidate != null) {
        result = getErrorReportReceiver(pm, packageName, candidate);//【见下文】
        if (result != null) {
            //2.当找到该crash应用的安装器,则返回;
            return result;
        }
    }
 
    if ((appFlags&ApplicationInfo.FLAG_SYSTEM) != 0) {
        //该系统属性名为"ro.error.receiver.system.apps"
        candidate = SystemProperties.get(SYSTEM_APPS_ERROR_RECEIVER_PROPERTY);
        result = getErrorReportReceiver(pm, packageName, candidate);//【见下文】
        if (result != null) {
            //3.当crash应用是系统应用时,且系统属性指定error receiver时,则返回;
            return result;
        }
    }
 
    //该默认属性名为"ro.error.receiver.default"
    candidate = SystemProperties.get(DEFAULT_ERROR_RECEIVER_PROPERTY);
    //4.当默认属性值指定error receiver时,则返回;
    return getErrorReportReceiver(pm, packageName, candidate); //【见下文】
}

getErrorReportReceiver:这是同名不同输入参数的另一个方法:

static ComponentNamegetErrorReportReceiver(PackageManagerpm, String errorPackage,
        String receiverPackage) {
    if (receiverPackage == null || receiverPackage.length() == 0) {
        return null;
    }
 
    //当安装应用程序的安装器Crash,则直接返回
    if (receiverPackage.equals(errorPackage)) {
        return null;
    }
 
    //ACTION_APP_ERROR值为"android.intent.action.APP_ERROR"
    Intentintent = new Intent(Intent.ACTION_APP_ERROR);
    intent.setPackage(receiverPackage);
    ResolveInfoinfo = pm.resolveActivity(intent, 0);
    if (info == null || info.activityInfo == null) {
        return null;
    }
    //创建包名为receiverPackage的组件
    return new ComponentName(receiverPackage, info.activityInfo.name);
}

7.2 skipCurrentReceiverLocked

[–>ActivityManagerService.java]

void skipCurrentReceiverLocked(ProcessRecordapp) {
    for (BroadcastQueuequeue : mBroadcastQueues) {
        queue.skipCurrentReceiverLocked(app);  //【见小节7.2.1】
    }
}

7.2.1 skipCurrentReceiverLocked

[-> BroadcastQueue.java]

public void skipCurrentReceiverLocked(ProcessRecordapp) {
    BroadcastRecord r = null;
    //查看app进程中的广播
    if (mOrderedBroadcasts.size() > 0) {
        BroadcastRecordbr = mOrderedBroadcasts.get(0);
        if (br.curApp == app) {
            r = br;
        }
    }
    if (r == null && mPendingBroadcast != null && mPendingBroadcast.curApp == app) {
        r = mPendingBroadcast;
    }
 
    if (r != null) {
        //结束app进程的广播结束
        finishReceiverLocked(r, r.resultCode, r.resultData,
                r.resultExtras, r.resultAbort, false);
        //广播调度
        scheduleBroadcastsLocked();
    }
}

8. PR.stopFreezingAllLocked

[-> ProcessRecord.java]

public void stopFreezingAllLocked() {
    int i = activities.size();
    while (i > 0) {
        i--;
        activities.get(i).stopFreezingScreenLocked(true); //【见小节8.1】
    }
}

其中activities类型为ArrayList <ActivityRecord >,停止进程里所有的Activity

8.1. AR.stopFreezingScreenLocked

[-> ActivityRecord.java]

public void stopFreezingScreenLocked(boolean force) {
    if (force || frozenBeforeDestroy) {
        frozenBeforeDestroy = false;
        //mWindowManager类型为WMS //【见小节8.1.1】
        service.mWindowManager.stopAppFreezingScreen(appToken, force);
    }
}

其中appToken是IApplication.Stub类型,即WindowManager的token。

8.1.1 WMS.stopFreezingScreenLocked

[-> WindowManagerService.java]

@Override
public void stopFreezingScreen() {
    //权限检查
    if (!checkCallingPermission(android.Manifest.permission.FREEZE_SCREEN,
            "stopFreezingScreen()")) {
        throw new SecurityException("Requires FREEZE_SCREEN permission");
    }
 
    synchronized(mWindowMap) {
        if (mClientFreezingScreen) {
            mClientFreezingScreen = false;
            mLastFinishedFreezeSource = "client";
            final long origId = Binder.clearCallingIdentity();
            try {
                stopFreezingDisplayLocked(); //【见流程8.1.1.1】
            } finally {
                Binder.restoreCallingIdentity(origId);
            }
        }
    }
}

8.1.1.1 WMS.stopFreezingDisplayLocked

[-> WindowManagerService.java]

private void stopFreezingDisplayLocked() {
    if (!mDisplayFrozen) {
        return; //显示没有冻结,则直接返回
    }
 
    //往往跟屏幕旋转相关
    ...
 
    mDisplayFrozen = false;
    //从上次冻屏到现在的总时长
    mLastDisplayFreezeDuration = (int)(SystemClock.elapsedRealtime() - mDisplayFreezeTime);
 
    //移除冻屏的超时消息
    mH.removeMessages(H.APP_FREEZE_TIMEOUT);
    mH.removeMessages(H.CLIENT_FREEZE_TIMEOUT);
 
    boolean updateRotation = false;
    //获取默认的DisplayContent
    final DisplayContentdisplayContent = getDefaultDisplayContentLocked();
    final int displayId = displayContent.getDisplayId();
    ScreenRotationAnimationscreenRotationAnimation =
            mAnimator.getScreenRotationAnimationLocked(displayId);
 
    //屏幕旋转动画的相关操作
    if (CUSTOM_SCREEN_ROTATION && screenRotationAnimation != null
            && screenRotationAnimation.hasScreenshot()) {
        DisplayInfodisplayInfo = displayContent.getDisplayInfo();
        boolean isDimming = displayContent.isDimming();
        if (!mPolicy.validateRotationAnimationLw(mExitAnimId, mEnterAnimId, isDimming)) {
            mExitAnimId = mEnterAnimId = 0;
        }
        //加载动画最大时长为10s
        if (screenRotationAnimation.dismiss(mFxSession, MAX_ANIMATION_DURATION,
                getTransitionAnimationScaleLocked(), displayInfo.logicalWidth,
                    displayInfo.logicalHeight, mExitAnimId, mEnterAnimId)) {
            scheduleAnimationLocked();
        } else {
            screenRotationAnimation.kill();
            mAnimator.setScreenRotationAnimationLocked(displayId, null);
            updateRotation = true;
        }
    } else {
        if (screenRotationAnimation != null) {
            screenRotationAnimation.kill();
            mAnimator.setScreenRotationAnimationLocked(displayId, null);
        }
        updateRotation = true;
    }
    //经过层层调用到InputManagerService服务,IMS服务使能输入事件分发功能
    mInputMonitor.thawInputDispatchingLw();
 
    boolean configChanged;
    //当display被冻结时不再计算屏幕方向,以避免不连续的状态。
    configChanged = updateOrientationFromAppTokensLocked(false);
 
    //display冻结时,执行gc操作
    mH.removeMessages(H.FORCE_GC);
    mH.sendEmptyMessageDelayed(H.FORCE_GC, 2000);
 
    //mScreenFrozenLock的类型为PowerManager.WakeLock,即释放屏幕冻结的锁
    mScreenFrozenLock.release();
 
    if (updateRotation) {
        //更新当前的屏幕方向
        configChanged |= updateRotationUncheckedLocked(false);
    }
 
    if (configChanged) {
        //向mH发送configuraion改变的消息
        mH.sendEmptyMessage(H.SEND_NEW_CONFIGURATION);
    }
}

该方法主要功能:

  1. 处理屏幕旋转相关逻辑;
  2. 移除冻屏的超时消息;
  3. 屏幕旋转动画的相关操作;
  4. 使能输入事件分发功能;
  5. display冻结时,执行gc操作;
  6. 更新当前的屏幕方向;
  7. 向mH发送configuraion改变的消息。

9.AMS.handleAppCrashLocked

[-> ActivityManagerService.java]

private boolean handleAppCrashLocked(ProcessRecordapp, String reason,
        String shortMsg, String longMsg, String stackTrace) {
    long now = SystemClock.uptimeMillis();
 
    Long crashTime;
    if (!app.isolated) {
        crashTime = mProcessCrashTimes.get(app.info.processName, app.uid);
    } else {
        crashTime = null;
    }
    //当同一个进程,连续两次crash的时间间隔小于1分钟时,则认为crash太过于频繁
    if (crashTime != null && now < crashTime+ProcessList.MIN_CRASH_INTERVAL) {
        EventLog.writeEvent(EventLogTags.AM_PROCESS_CRASHED_TOO_MUCH,
                app.userId, app.info.processName, app.uid);
        //【见小节9.1】
        mStackSupervisor.handleAppCrashLocked(app);
        if (!app.persistent) {
            //不再重启非persistent进程,除非用户显式地调用
            EventLog.writeEvent(EventLogTags.AM_PROC_BAD, app.userId, app.uid,
                    app.info.processName);
            if (!app.isolated) {
                //将当前app加入到mBadProcesses
                mBadProcesses.put(app.info.processName, app.uid,
                        new BadProcessInfo(now, shortMsg, longMsg, stackTrace));
                mProcessCrashTimes.remove(app.info.processName, app.uid);
            }
            app.bad = true;
            app.removed = true;
            //移除进程的所有服务,保证不再重启【见小节9.2】
            removeProcessLocked(app, false, false, "crash");
            //恢复最顶部的Activity【见小节9.3】
            mStackSupervisor.resumeTopActivitiesLocked();
            return false;
        }
        mStackSupervisor.resumeTopActivitiesLocked();
    } else {
        //此处reason="force-crash"【见小节9.4】
        mStackSupervisor.finishTopRunningActivityLocked(app, reason);
    }
 
    //运行在当前进程中的所有服务的crash次数执行加1操作
    for (int i=app.services.size()-1; i>=0; i--) {
        ServiceRecordsr = app.services.valueAt(i);
        sr.crashCount++;
    }
 
    //当桌面应用crash,并且被三方app所取代,那么需要清空桌面应用的偏爱选项。
    final ArrayList<ActivityRecord> activities = app.activities;
    if (app == mHomeProcess && activities.size() > 0
                && (mHomeProcess.info.flags & ApplicationInfo.FLAG_SYSTEM) == 0) {
        for (int activityNdx = activities.size() - 1; activityNdx >= 0; --activityNdx) {
            final ActivityRecord r = activities.get(activityNdx);
            if (r.isHomeActivity()) {
                //清空偏爱应用
                ActivityThread.getPackageManager()
                        .clearPackagePreferredActivities(r.packageName);
            }
        }
    }
 
    if (!app.isolated) {
        //无法记录孤立进程的crash时间点,由于他们并没有一个固定身份
        mProcessCrashTimes.put(app.info.processName, app.uid, now);
    }
    //当app存在crash的handler,那么交给其处理
    if (app.crashHandler != null) mHandler.post(app.crashHandler);
    return true;
}
  1. 当同一进程在时间间隔小于1分钟时连续两次crash,则执行的情况下:
    • 对于非persistent进程:
      • [9.1] mStackSupervisor.handleAppCrashLocked(app);
      • [9.2] removeProcessLocked(app, false, false, “crash”);
      • [9.3] mStackSupervisor.resumeTopActivitiesLocked();
    • 对于persistent进程,则只执行
      • [9.3] mStackSupervisor.resumeTopActivitiesLocked();
  2. 否则执行
    • [9.4] mStackSupervisor.finishTopRunningActivityLocked(app, reason);

9.1 ASS.handleAppCrashLocked

[-> ActivityStackSupervisor.java]

void handleAppCrashLocked(ProcessRecordapp) {
    for (int displayNdx = mActivityDisplays.size() - 1; displayNdx >= 0; --displayNdx) {
        final ArrayList<ActivityStack> stacks = mActivityDisplays.valueAt(displayNdx).mStacks;
        int stackNdx = stacks.size() - 1;
        while (stackNdx >= 0) {
            //调用ActivityStack【见小节9.1.1】
            stacks.get(stackNdx).handleAppCrashLocked(app);
            stackNdx--;
        }
    }
}

9.1.1 AS.handleAppCrashLocked

[-> ActivityStack.java]

void handleAppCrashLocked(ProcessRecordapp) {
    for (int taskNdx = mTaskHistory.size() - 1; taskNdx >= 0; --taskNdx) {
        final ArrayList<ActivityRecord> activities = mTaskHistory.get(taskNdx).mActivities;
        for (int activityNdx = activities.size() - 1; activityNdx >= 0; --activityNdx) {
            final ActivityRecord r = activities.get(activityNdx);
            if (r.app == app) {
                r.app = null;
                //结束当前activity
                finishCurrentActivityLocked(r, FINISH_IMMEDIATELY, false);
            }
        }
    }
}

这里的 mTaskHistory 数据类型为ArrayList,记录着所有先前的后台activities。遍历所有activities,找到位于该ProcessRecord的所有ActivityRecord,并结束该Acitivity。

9.2 AMS.removeProcessLocked

[-> ActivityManagerService.java]

private final boolean removeProcessLocked(ProcessRecordapp,
        boolean callerWillRestart, boolean allowRestart, String reason) {
    final String name = app.processName;
    final int uid = app.uid;
    //从mProcessNames移除该进程
    removeProcessNameLocked(name, uid);
    ...
    if (app.pid > 0 && app.pid != MY_PID) {
        int pid = app.pid;
        synchronized (mPidsSelfLocked) {
            mPidsSelfLocked.remove(pid); //移除该pid
            mHandler.removeMessages(PROC_START_TIMEOUT_MSG, app);
        }
        ...
        boolean willRestart = false;
        //对于非孤立的persistent进程设置成可重启flags
        if (app.persistent && !app.isolated) {
            if (!callerWillRestart) {
                willRestart = true;
            } else {
                needRestart = true;
            }
        }
        // 杀进程【9.2.1】
        app.kill(reason, true);
        //移除进程并清空该进程相关联的activity/service等组件 【9.2.2】
        handleAppDiedLocked(app, willRestart, allowRestart);
 
        if (willRestart) {
            //此处willRestart=false,不进入该分支
            removeLruProcessLocked(app);
            addAppLocked(app.info, false, null /* ABI override */);
        }
    } else {
        mRemovedProcesses.add(app);
    }
    return needRestart;
}
  • mProcessNames 数据类型为ProcessMap,这是以进程名为key,记录着所有的ProcessRecord信息
  • mPidsSelfLocked 数据类型为SparseArray,这是以pid为key,记录着所有的ProcessRecord信息。该对象的同步保护是通过自身锁,而非全局ActivityManager锁。
  • (责任编辑:最模板)
顶一下
(0)
0%
踩一下
(0)
0%
------分隔线----------------------------
栏目列表
热点内容