阅读需 10 分钟

初阶技能:应用运行异常如何丰富线索

在我们容器虚拟化产品开发过程中,时常会遇到某些应用无法启动或运行时异常崩溃的问题;让应用行为信息丰富,则能还原应用异常发生过程,对我们快速分析问题至关重要。

在这里需要用到以下几个开源工具:

01. 举个栗子

我们在做应用容器时,尝尝会遇到应用进程崩溃,通过adb抓取的日志,却只有很少的几行日志,没有其他任何信息,因此通过一些简单的方式让应用行为的痕迹还原,是我们第一步要做的事情。

9402  9402 D AndroidRuntime: Shutting down VM
9402 9402 I Process : Sending signal. PID: 9402 SIG: 9

02. 打印退出痕迹

通过拦截native函数,打印Java层的异常退出:

import android.os;

public class Process {

/**
* Returns the identifier of this process, which can be used with
* {@link #killProcess} and {@link #sendSignal}.
*/
public static final native void sendSignal(int pid, int signal);

/**
* @hide
* Private impl for avoiding a log message... DO NOT USE without doing
* your own log, or the Android Illuminati will find you some night and
* beat you up.
*/
@UnsupportedAppUsage(maxTargetSdk = Build.VERSION_CODES.P)
public static final native void sendSignalQuiet(int pid, int signal);

}

public class Runtime {
private static native void nativeExit(int code);
}

通过拦截以下函数,打印Native层的异常退出:

void exit(int status);
int kill(pid_t pid, int sig);

打印调用栈:

// 非 JNI 环境获取从当前线程获取 JNIEnv(AttachCurrentThread)
void jni_thread_dump(JNIEnv* env) {
jclass jcls = env->FindClass("java/lang/Thread");
if (!jcls) return;
jmethodID jm = env->GetStaticMethodID(jcls, "dumpStack", "()V");
if (!jm) return;
env->CallStaticVoidMethod(jcls, jm);
env->ExceptionClear();
}

输出结果:

5511  5511 W System.err: java.lang.Exception: Stack trace
5511 5511 W System.err: at java.lang.Thread.dumpStack(Thread.java:1527)
5511 5511 W System.err: at android.os.Process.sendSignal(Native Method)
5511 5511 W System.err: at android.os.Process.killProcess(Process.java:1295)
5511 5511 W System.err: at com.android.internal.os.RuntimeInit$KillApplicationHandler.uncaughtException(RuntimeInit.java:207)
5511 5511 W System.err: at java.lang.ThreadGroup.uncaughtException(ThreadGroup.java:1073)
5511 5511 W System.err: at java.lang.ThreadGroup.uncaughtException(ThreadGroup.java:1068)
5511 5511 W System.err: at java.lang.Thread.dispatchUncaughtException(Thread.java:2211)

03. 追踪Throwable异常调用栈

在我们常用的try ... catch语句中,异常时打印调用栈是常规操作,在这里可以拦截并打印栈日志信息。

import java.lang;

public class Throwable implements Serializable {

private static native StackTraceElement[] nativeGetStackTrace(Object stackState);

private static native Object nativeFillInStackTrace();

}

04. 打印VM异常调用栈

package dalvik.system;

/**
* Provides a limited interface to the Dalvik VM stack. This class is mostly
* used for implementing security checks.
*/
public final class VMStack {
/**
* Retrieves the stack trace from the specified thread.
*
* @param t
* thread of interest
* @return an array of stack trace elements, or null if the thread
* doesn't have a stack trace (e.g. because it exited)
*/
native public static StackTraceElement[] getThreadStackTrace(Thread t);

/**
* Retrieves an annotated stack trace from the specified thread.
*
* @param t
* thread of interest
* @return an array of annotated stack frames, or null if the thread
* doesn't have a stack trace (e.g. because it exited)
*/
native public static AnnotatedStackTraceElement[] getAnnotatedThreadStackTrace(Thread t);

/**
* Retrieves a partial stack trace from the specified thread into
* the provided array.
*
* @param t
* thread of interest
* @param stackTraceElements
* preallocated array for use when only the top of stack is
* desired. Unused elements will be filled with null values.
* @return the number of elements filled
*/
native public static int fillStackTraceElements(Thread t, StackTraceElement[] stackTraceElements);
}

输出:

9402  9402 W stack  : android.app.LoadedApk.makeApplication(LoadedApk.java:1554)
9402 9402 W stack : android.app.ActivityThread.handleBindApplication(ActivityThread.java:8522)
9402 9402 W stack : android.app.ActivityThread.access$2800(ActivityThread.java:311)
9402 9402 W stack : android.app.ActivityThread$H.handleMessage(ActivityThread.java:2889)
9402 9402 W stack : android.os.Handler.dispatchMessage(Handler.java:117)
9402 9402 W stack : android.os.Looper.loopOnce(Looper.java:205)
9402 9402 W stack : android.os.Looper.loop(Looper.java:293)
9402 9402 W stack : android.app.ActivityThread.loopProcess(ActivityThread.java:9934)
9402 9402 W stack : android.app.ActivityThread.main(ActivityThread.java:9923)
9402 9402 W stack : java.lang.reflect.Method.invoke(Native Method)
9402 9402 W stack : com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:586)
9402 9402 W stack : com.android.internal.os.ZygoteInit.main(ZygoteInit.java:1240)
9402 9402 W stack : ohos.abilityshell.HarmonyLoader.tryLoadHarmony(HarmonyLoader.java:130)
9402 9402 W stack : ohos.abilityshell.HarmonyApplication.tryLoadHarmony(HarmonyApplication.java:673)
9402 9402 W stack : ohos.abilityshell.HarmonyApplication.attachBaseContext(HarmonyApplication.java:168)
9402 9402 W stack : com.demo.app.DemoApp.attachBaseContext(SourceFile:1)
9402 9402 W stack : android.app.Application.attach(Application.java:338)
9402 9402 W stack : android.app.Instrumentation.newApplication(Instrumentation.java:1191)
9402 9402 W stack : android.app.LoadedApk.makeApplication(LoadedApk.java:1546)
9402 9402 W stack : android.app.ActivityThread.handleBindApplication(ActivityThread.java:8522)
9402 9402 W stack : android.app.ActivityThread.access$2800(ActivityThread.java:311)
9402 9402 W stack : android.app.ActivityThread$H.handleMessage(ActivityThread.java:2889)
9402 9402 W stack : android.os.Handler.dispatchMessage(Handler.java:117)
9402 9402 W stack : android.os.Looper.loopOnce(Looper.java:205)
9402 9402 W stack : android.os.Looper.loop(Looper.java:293)
9402 9402 W stack : android.app.ActivityThread.loopProcess(ActivityThread.java:9934)
9402 9402 W stack : android.app.ActivityThread.main(ActivityThread.java:9923)
9402 9402 W stack : java.lang.reflect.Method.invoke(Native Method)
9402 9402 W stack : com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:586)
9402 9402 W stack : com.android.internal.os.ZygoteInit.main(ZygoteInit.java:1240)

05. 打印异常信息

Android框架默认会为每个应用进程设置一个全局的异常Handler,我们可以替换掉输出打印更多内容。在这里要考虑应用自身也会设置的情况。

Thread.setDefaultUncaughtExceptionHandler(new Thread.UncaughtExceptionHandler() {
@Override
public void uncaughtException(Thread t, Throwable e) {
// TODO
}
});

06. 打印异常传送

AndroidAMS服务中增加了一个应用向服务提交异常的接口,我们通过android.reflect.Proxy方式代理IActivityManager的实例,打印输出。

package android.app;

// @source code: /frameworks/base/core/java/android/app/IActivityManager.aidl
interface IActivityManager {
void handleApplicationCrash(IBinder app, ApplicationErrorReport.ParcelableCrashInfo crashInfo);
}

// @source code: /frameworks/base/core/java/android/app/ApplicationErrorReport.java
public class ApplicationErrorReport implements Parcelable {
public static class CrashInfo {
/**
* Dump a CrashInfo instance to a Printer.
*/
public void dump(Printer pw, String prefix) {
pw.println(prefix + "exceptionHandlerClassName: " + exceptionHandlerClassName);
pw.println(prefix + "exceptionClassName: " + exceptionClassName);
pw.println(prefix + "exceptionMessage: " + exceptionMessage);
pw.println(prefix + "throwFileName: " + throwFileName);
pw.println(prefix + "throwClassName: " + throwClassName);
pw.println(prefix + "throwMethodName: " + throwMethodName);
pw.println(prefix + "throwLineNumber: " + throwLineNumber);
pw.println(prefix + "stackTrace: " + stackTrace);
}
}
public static class ParcelableCrashInfo extends CrashInfo implements Parcelable {

}
}

如输出结果为:

9402  9402 W Crash   : exceptionClassName: java.lang.IllegalStateException
9402 9402 W Crash : exceptionMessage: failed to attach Application, errorCode=30, errorInfo=bms service error, code is 8519969. 8519797
9402 9402 W Crash : throwFileName: HarmonyLoader.java
9402 9402 W Crash : throwClassName: ohos.abilityshell.HarmonyLoader
9402 9402 W Crash : throwMethodName: tryLoadHarmony
9402 9402 W Crash : throwLineNumber: 130
9402 9402 W Crash : stackTrace: java.lang.RuntimeException: Unable to instantiate application com.demo.app.DemoApp package com.demo.app: java.lang.IllegalStateException: failed to attach Application, errorCode=30, errorInfo=bms service error, code is 8519969.
9402 9402 W Crash : at android.app.LoadedApk.makeApplication(LoadedApk.java:1554)
9402 9402 W Crash : at android.app.ActivityThread.handleBindApplication(ActivityThread.java:8522)
9402 9402 W Crash : at android.app.ActivityThread.access$2800(ActivityThread.java:311)
9402 9402 W Crash : at android.app.ActivityThread$H.handleMessage(ActivityThread.java:2889)
9402 9402 W Crash : at android.os.Handler.dispatchMessage(Handler.java:117)
9402 9402 W Crash : at android.os.Looper.loopOnce(Looper.java:205)
9402 9402 W Crash : at android.os.Looper.loop(Looper.java:293)
9402 9402 W Crash : at android.app.ActivityThread.loopProcess(ActivityThread.java:9934)
9402 9402 W Crash : at android.app.ActivityThread.main(ActivityThread.java:9923)
9402 9402 W Crash : at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:586)
9402 9402 W Crash : at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:1240)
9402 9402 W Crash : Caused by: java.lang.IllegalStateException: fail

07. 打印Binder通信信息

可以打印应用访问了那些系统服务。可以通过Binderceptor项目来完成。

9402  9402 W Trace   : Binder::tgt: 0x3887b52, 3, : android.content.pm.IPackageManager
9402 9402 W Trace : Binder::tgt: 0xe45f2c2, 2, : android.os.IServiceManager
9402 9402 W Trace : Binder::tgt: 0x76c2822, 56, : com.huawei.android.view.IHwWindowManager
9402 9417 W Trace : Binder::tgt: 0x3887b52, 20, : android.content.pm.IPackageManager
9402 9402 W Trace : Binder::tgt: 0x7db76a2, 30, : android.net.IConnectivityManager
9402 9402 W Trace : Binder::tgt: 0x3887b52, 20, : android.content.pm.IPackageManager
9402 9402 W Trace : Binder::tgt: 0x3887b52, 9, : android.content.pm.IPackageManager
9402 9402 W Trace : Binder::tgt: 0x3887b52, 95, : android.content.pm.IPackageManager
9402 9402 W Trace : Binder::tgt: 0x3887b52, 3, : android.content.pm.IPackageManager
......

binder codeaidl接口函数的对应关系:

public static void printStub(String clsName) {
android.util.Log.w(TAG, ">>>>>> " + clsName);
try {
Class<?> nmClassStub = Class.findClass(clsName + "$Stub");
Method[] mm = nmClassStub.getMethods();

ArrayList<TRANSACTION_Item> list = new ArrayList<TRANSACTION_Item>();
for (Method m : mm) {
if (TextUtils.equals(m.getName(), "asBinder")) continue;
try {
funcName = m.toGenericString();
funcName = funcName.replace("abstract ", "");
funcName = funcName.replace(clsName + ".", "");

fieldName = "TRANSACTION_" + m.getName();
Field f = nmClassStub.getDeclaredField(fieldName);
f.setAccessible(true);
int val = f.getInt(null);
// android.util.Log.w(TAG, fieldName + " " + val);
list.add(new TRANSACTION_Item(val, funcName, fieldName));
} catch (Exception e) {
android.util.Log.e(TAG, e.toString());
}
}
Collections.sort(list, new TRANSACTION_Comparator());
} catch (Exception e) {
android.util.Log.e(TAG, e.toString());
}
android.util.Log.w(TAG, "<<<<<< " + clsName);
}

排序后可以得到输出的结果:

package android.content.pm;

interface IPackageManager {
public void checkPackageStartable(java.lang.String,int) throws android.os.RemoteException;// 1
public boolean isPackageAvailable(java.lang.String,int) throws android.os.RemoteException;// 2
public android.content.pm.PackageInfo getPackageInfo(java.lang.String,int,int) throws android.os.RemoteException;// 3
public android.content.pm.PackageInfo getPackageInfoVersioned(android.content.pm.VersionedPackage,int,int) throws android.os.RemoteException;// 4
}

08. 打印JNI调用轨迹

通过拦截JNIEnv层提供调用Java层方法的函数,打印调用信息。

env->FindClass(...);
env->GetMethodID(...);
env->GetStaticMethodID(...);
env->RegisterNatives(...);

09. 打印访问文件行为

通过拦截open系列函数(为什么是系列,因为libc.so经过多年迭代,衍生出来很多函数,所以说Android的兼容性让广大开发者开发Demo容易,做稳定则难上加难),打印文件访问记录。

int open(const char *fileName, int flags, ...);
int open2(const char *fileName, int flags, ...);
int _open(const char *fileName, int flags, ...);
int openat(int dirfd, const char *fileName, int flags, ...);
int _openat(int dirfd, const char *fileName, int flags, ...);
...

至此,通过以上基本的方法,可以快速的增加应用运行轨迹信息。

若还不够,则需要针对应用内部调试,动态库加载的进阶方法深一步剖析,后面会给大家整理进阶技能。