Android development—a brief introduction to hot-fixing Tinker source code

0. Preface

## Hot repair technology has basically become more important for Android projects module. Mainly because after a project goes online, it is inevitable that there will be various problems, and it is too expensive to rely on releases to fix problems.

The current hot repair technology basically includes Alibaba’s AndFix and QZone plan, the ideological plan proposed by Meituan, and Tencent’s Tinker, etc.

Among themAndFixpossibly access is the simplest one (andTinkercommand The row access method is similar), but AndFix has certain compatibility issues, the QZone solution has a negative impact on performance There will be a certain impact, and in the Art mode there will be memory confusion problems. The idea proposed by Meituan is mainly based on The principle of Instant Run is not yet open source and has good compatibility.

It seems that if you choose an open source solution, Tinker is currently the best choice. Let’s take a look## A rough principle analysis of #Tinker.

1. Principle overview

Tinker

willold.apk and new.apk did diff and got patch.dex and then make it with apk#classes.dex Merge, generate a new classes.dex, reflect the merged during runtime dexFile is placed in front of the loaded dexElements array. The principle of runtime substitution is actually similar to the solution of

Qzone

, which is to reflect and modify dexElements. The difference between the two is: Qzone directly inserts patch.dex into the front of the array; while Tinker is the full amount of merged dex inserted in front of the array . Because there is a problem with the CLASS_ISPREVERIFIED solution mentioned in the Qzone solution.

Android's ClassLoader Generally used to load classes in the system are PathClassLoader and DexClassLoader , everyone just needs to understand that Android uses PathClassLoader as its class loader,DexClassLoadercan be loaded internally from files of type .jar and .apk classes.dexfile will be fine. For loading a class, just give a classname, and then go to findClass, PathClassLoader and DexClassLoader both inherit from BaseDexClassLoader. There is the following source code in BaseDexClassLoader:

#BaseDexClassLoader
@Override
protected Class<?> findClass(String name) throws ClassNotFoundException {
    Class clazz = pathList.findClass(name);

    if (clazz == null) {
        throw new ClassNotFoundException(name);
    }

    return clazz;
}

#DexPathList
public Class findClass(String name) {
    for (Element element : dexElements) {
        DexFile dex = element.dexFile;

        if (dex != null) {
            Class clazz = dex.loadClassBinaryName(name, definingContext);
            if (clazz != null) {
                return clazz;
            }
        }
    }

    return null;
}

#DexFile
public Class loadClassBinaryName(String name, ClassLoader loader) {
    return defineClass(name, loader, mCookie);
}
private native static Class defineClass(String name, ClassLoader loader, int cookie);

##You can see There is a pathList object in BaseDexClassLoader, and pathList contains a collection of DexFile dexElements, and for class loading is to traverse this collection and find through DexFile.

In layman’s terms, a ClassLoader can contain multiple dex files, each A dex file is an Element, and multiple dex files are arranged into one Ordered array dexElements, when looking for a class, it will traverse the dex files in order, and then start from the currently traversed ## Find the class in the #dex file. If the class is found, it will be returned. If it is not found, continue searching from the next dex file. In this case, we canplace our patch.jar in the first element of this array, which contains the repaired class. In this case, when When traversing findClass, the class we repaired will be found, thus replacing the buggy class.

Two lines of Tinker source code analysis in this film article:

（

1） When the application starts, load the merged classes.dex

(

2) from the default directory patchAfter delivery, synthesize classes.dex to the target directory

2． Source code analysis

2.1  加载patch

加载的代码实际上在生成的Application中调用的，其父类为TinkerApplication，在其attachBaseContext中辗转会调用到loadTinker()方法，在该方法内部，反射调用了TinkerLoader的tryLoad方法。继而调用了tryLoadPatchFilesInternal方法。

@Override
public Intent tryLoad(TinkerApplication app, int tinkerFlag, boolean tinkerLoadVerifyFlag) {
    Intent resultIntent = new Intent();

    long begin = SystemClock.elapsedRealtime();
    tryLoadPatchFilesInternal(app, tinkerFlag, tinkerLoadVerifyFlag, resultIntent);
    long cost = SystemClock.elapsedRealtime() - begin;
    ShareIntentUtil.setIntentPatchCostTime(resultIntent, cost);
    return resultIntent;
}

private void tryLoadPatchFilesInternal(TinkerApplication app, int tinkerFlag, boolean tinkerLoadVerifyFlag, Intent resultIntent) {
    // 省略大量安全性校验代码

    if (isEnabledForDex) {
        //tinker/patch.info/patch-641e634c/dex
        boolean dexCheck = TinkerDexLoader.checkComplete(patchVersionDirectory, securityCheck, resultIntent);
        if (!dexCheck) {
            //file not found, do not load patch
            Log.w(TAG, "tryLoadPatchFiles:dex check fail");
            return;
        }
    }

    //now we can load patch jar
    if (isEnabledForDex) {
        boolean loadTinkerJars = TinkerDexLoader.loadTinkerJars(app, tinkerLoadVerifyFlag, patchVersionDirectory, resultIntent, isSystemOTA);
        if (!loadTinkerJars) {
            Log.w(TAG, "tryLoadPatchFiles:onPatchLoadDexesFail");
            return;
        }
    }
}

其中TinkerDexLoader.checkComplete主要是用于检查下发的meta文件中记录的dex信息（meta文件，可以查看生成patch的产物，在assets/dex-meta.txt），检查meta文件中记录的dex文件信息对应的dex文件是否存在，并把值存在TinkerDexLoader的静态变量dexList中。

TinkerDexLoader.loadTinkerJars传入四个参数，分别为application，tinkerLoadVerifyFlag（注解上声明的值，传入为false），patchVersionDirectory当前version的patch文件夹，intent，当前patch是否仅适用于art。

@TargetApi(Build.VERSION_CODES.ICE_CREAM_SANDWICH)
public static boolean loadTinkerJars(Application application, boolean tinkerLoadVerifyFlag, 
    String directory, Intent intentResult, boolean isSystemOTA) {
        PathClassLoader classLoader = (PathClassLoader) TinkerDexLoader.class.getClassLoader();

        String dexPath = directory + "/" + DEX_PATH + "/";
        File optimizeDir = new File(directory + "/" + DEX_OPTIMIZE_PATH);

        ArrayList<File> legalFiles = new ArrayList<>();

        final boolean isArtPlatForm = ShareTinkerInternals.isVmArt();
        for (ShareDexDiffPatchInfo info : dexList) {
            //for dalvik, ignore art support dex
            if (isJustArtSupportDex(info)) {
                continue;
            }
            String path = dexPath + info.realName;
            File file = new File(path);

            legalFiles.add(file);
        }
        // just for art
        if (isSystemOTA) {
            parallelOTAResult = true;
            parallelOTAThrowable = null;
            Log.w(TAG, "systemOTA, try parallel oat dexes!!!!!");

            TinkerParallelDexOptimizer.optimizeAll(
                legalFiles, optimizeDir,
                new TinkerParallelDexOptimizer.ResultCallback() {
                }
            );

        SystemClassLoaderAdder.installDexes(application, classLoader, optimizeDir, legalFiles);
        return true;

找出仅支持art的dex，且当前patch是否仅适用于art时，并行去loadDex。关键是最后的installDexes：

@SuppressLint("NewApi")
public static void installDexes(Application application, PathClassLoader loader, File dexOptDir, List<File> files)
    throws Throwable {

    if (!files.isEmpty()) {
        ClassLoader classLoader = loader;
        if (Build.VERSION.SDK_INT >= 24) {
            classLoader = AndroidNClassLoader.inject(loader, application);
        }
        //because in dalvik, if inner class is not the same classloader with it wrapper class.
        //it won&#39;t fail at dex2opt
        if (Build.VERSION.SDK_INT >= 23) {
            V23.install(classLoader, files, dexOptDir);
        } else if (Build.VERSION.SDK_INT >= 19) {
            V19.install(classLoader, files, dexOptDir);
        } else if (Build.VERSION.SDK_INT >= 14) {
            V14.install(classLoader, files, dexOptDir);
        } else {
            V4.install(classLoader, files, dexOptDir);
        }
        //install done
        sPatchDexCount = files.size();
        Log.i(TAG, "after loaded classloader: " + classLoader + ", dex size:" + sPatchDexCount);

        if (!checkDexInstall(classLoader)) {
            //reset patch dex
            SystemClassLoaderAdder.uninstallPatchDex(classLoader);
            throw new TinkerRuntimeException(ShareConstants.CHECK_DEX_INSTALL_FAIL);
        }
    }
}

这里实际上就是根据不同的系统版本，去反射处理dexElements。我们看一下V19的实现：

private static final class V19 {
    private static void install(ClassLoader loader, List<File> additionalClassPathEntries,
                                File optimizedDirectory)
        throws IllegalArgumentException, IllegalAccessException,
        NoSuchFieldException, InvocationTargetException, NoSuchMethodException, IOException {

        Field pathListField = ShareReflectUtil.findField(loader, "pathList");
        Object dexPathList = pathListField.get(loader);
        ArrayList<IOException> suppressedExceptions = new ArrayList<IOException>();
        ShareReflectUtil.expandFieldArray(dexPathList, "dexElements", makeDexElements(dexPathList,
            new ArrayList<File>(additionalClassPathEntries), optimizedDirectory,
            suppressedExceptions));
        if (suppressedExceptions.size() > 0) {
            for (IOException e : suppressedExceptions) {
                Log.w(TAG, "Exception in makeDexElement", e);
                throw e;
            }
        }
    }
}

（1）找到PathClassLoader（BaseDexClassLoader）对象中的pathList对象

（2）根据pathList对象找到其中的makeDexElements方法，传入patch相关的对应的实参，返回Element[]对象

（3）拿到pathList对象中原本的dexElements方法

（4）步骤2与步骤3中的Element[]数组进行合并，将patch相关的dex放在数组的前面

（5）最后将合并后的数组，设置给pathList

2.2 合成patch

入口为onReceiveUpgradePatch()方法：

TinkerInstaller.onReceiveUpgradePatch(getApplicationContext(),
                Environment.getExternalStorageDirectory().getAbsolutePath() + "/patch_signed.apk");

上述代码会调用DefaultPatchListener中的onPatchReceived方法：

# DefaultPatchListener
@Override
public int onPatchReceived(String path) {

    int returnCode = patchCheck(path);

    if (returnCode == ShareConstants.ERROR_PATCH_OK) {
        TinkerPatchService.runPatchService(context, path);
    } else {
        Tinker.with(context).getLoadReporter().onLoadPatchListenerReceiveFail(new File(path), returnCode);
    }
    return returnCode;
}

首先对Tinker的相关配置（isEnable）以及patch的合法性进行检测，如果合法，则调用TinkerPatchService.runPatchService(context, path)。

public static void runPatchService(Context context, String path) {
    try {
        Intent intent = new Intent(context, TinkerPatchService.class);
        intent.putExtra(PATCH_PATH_EXTRA, path);
        intent.putExtra(RESULT_CLASS_EXTRA, resultServiceClass.getName());
        context.startService(intent);
    } catch (Throwable throwable) {
        TinkerLog.e(TAG, "start patch service fail, exception:" + throwable);
    }
}

TinkerPatchService是IntentService的子类，这里通过intent设置了两个参数，一个是patch的路径，一个是resultServiceClass，该值是调用Tinker.install的时候设置的，默认为DefaultTinkerResultService.class。由于是IntentService，直接看onHandleIntent即可。

@Override
protected void onHandleIntent(Intent intent) {
    final Context context = getApplicationContext();
    Tinker tinker = Tinker.with(context);
    String path = getPatchPathExtra(intent);
    File patchFile = new File(path);
    boolean result;
    increasingPriority();
    PatchResult patchResult = new PatchResult();
    result = upgradePatchProcessor.tryPatch(context, path, patchResult);
    patchResult.isSuccess = result;
    patchResult.rawPatchFilePath = path;
    patchResult.costTime = cost;
    patchResult.e = e;
    AbstractResultService.runResultService(context, patchResult, getPatchResultExtra(intent));
}

比较清晰，主要关注upgradePatchProcessor.tryPatch方法，调用的是UpgradePatch.tryPatch。这里有个有意思的地方increasingPriority()，其内部实现为：

private void increasingPriority() {
    TinkerLog.i(TAG, "try to increase patch process priority");
    try {
        Notification notification = new Notification();
        if (Build.VERSION.SDK_INT < 18) {
            startForeground(notificationId, notification);
        } else {
            startForeground(notificationId, notification);
            // start InnerService
            startService(new Intent(this, InnerService.class));
        }
    } catch (Throwable e) {
        TinkerLog.i(TAG, "try to increase patch process priority error:" + e);
    }
}

如果你对“保活”这个话题比较关注，那么对这段代码一定不陌生，主要是利用系统的一个漏洞来启动一个前台Service。下面继续回到tryPatch方法：

# UpgradePatch
@Override
public boolean tryPatch(Context context, String tempPatchPath, PatchResult patchResult) {
    Tinker manager = Tinker.with(context);

    final File patchFile = new File(tempPatchPath);

    //it is a new patch, so we should not find a exist
    SharePatchInfo oldInfo = manager.getTinkerLoadResultIfPresent().patchInfo;
    String patchMd5 = SharePatchFileUtil.getMD5(patchFile);

    //use md5 as version
    patchResult.patchVersion = patchMd5;
    SharePatchInfo newInfo;

    //already have patch
    if (oldInfo != null) {
        newInfo = new SharePatchInfo(oldInfo.oldVersion, patchMd5, Build.FINGERPRINT);
    } else {
        newInfo = new SharePatchInfo("", patchMd5, Build.FINGERPRINT);
    }

    //check ok, we can real recover a new patch
    final String patchDirectory = manager.getPatchDirectory().getAbsolutePath();
    final String patchName = SharePatchFileUtil.getPatchVersionDirectory(patchMd5);
    final String patchVersionDirectory = patchDirectory + "/" + patchName;

    //copy file
    File destPatchFile = new File(patchVersionDirectory + "/" + SharePatchFileUtil.getPatchVersionFile(patchMd5));
    // check md5 first
    if (!patchMd5.equals(SharePatchFileUtil.getMD5(destPatchFile))) {
        SharePatchFileUtil.copyFileUsingStream(patchFile, destPatchFile);
    }

 //we use destPatchFile instead of patchFile, because patchFile may be deleted during the patch process
    if (!DexDiffPatchInternal.tryRecoverDexFiles(manager, signatureCheck, context, patchVersionDirectory, 
                destPatchFile)) {
        TinkerLog.e(TAG, "UpgradePatch tryPatch:new patch recover, try patch dex failed");
        return false;
    }
    return true;
}

拷贝patch文件拷贝至私有目录，然后调用DexDiffPatchInternal.tryRecoverDexFiles：

protected static boolean tryRecoverDexFiles(Tinker manager, ShareSecurityCheck checker, Context context,
                                                
    String patchVersionDirectory, File patchFile) {
    String dexMeta = checker.getMetaContentMap().get(DEX_META_FILE);
    boolean result = patchDexExtractViaDexDiff(context, patchVersionDirectory, dexMeta, patchFile);
    return result;
}

直接看patchDexExtractViaDexDiff：

private static boolean patchDexExtractViaDexDiff(Context context, String patchVersionDirectory, String meta, final File patchFile) {
    String dir = patchVersionDirectory + "/" + DEX_PATH + "/";

    if (!extractDexDiffInternals(context, dir, meta, patchFile, TYPE_DEX)) {
        TinkerLog.w(TAG, "patch recover, extractDiffInternals fail");
        return false;
    }

    final Tinker manager = Tinker.with(context);

    File dexFiles = new File(dir);
    File[] files = dexFiles.listFiles();

    ...files遍历执行：DexFile.loadDex
     return true;
}

核心代码主要在extractDexDiffInternals中：

private static boolean extractDexDiffInternals(Context context, String dir, String meta, File patchFile, int type) {
    //parse meta
    ArrayList<ShareDexDiffPatchInfo> patchList = new ArrayList<>();
    ShareDexDiffPatchInfo.parseDexDiffPatchInfo(meta, patchList);

    File directory = new File(dir);
    //I think it is better to extract the raw files from apk
    Tinker manager = Tinker.with(context);
    ZipFile apk = null;
    ZipFile patch = null;

    ApplicationInfo applicationInfo = context.getApplicationInfo();

    String apkPath = applicationInfo.sourceDir; //base.apk
    apk = new ZipFile(apkPath);
    patch = new ZipFile(patchFile);

    for (ShareDexDiffPatchInfo info : patchList) {

        final String infoPath = info.path;
        String patchRealPath;
        if (infoPath.equals("")) {
            patchRealPath = info.rawName;
        } else {
            patchRealPath = info.path + "/" + info.rawName;
        }

        File extractedFile = new File(dir + info.realName);

        ZipEntry patchFileEntry = patch.getEntry(patchRealPath);
        ZipEntry rawApkFileEntry = apk.getEntry(patchRealPath);

        patchDexFile(apk, patch, rawApkFileEntry, patchFileEntry, info, extractedFile);
    }

    return true;
}

这里的代码比较关键了，可以看出首先解析了meta里面的信息，meta中包含了patch中每个dex的相关数据。然后通过Application拿到sourceDir，其实就是本机apk的路径以及patch文件；根据mate中的信息开始遍历，其实就是取出对应的dex文件，最后通过patchDexFile对两个dex文件做合并。

private static void patchDexFile(
            ZipFile baseApk, ZipFile patchPkg, ZipEntry oldDexEntry, ZipEntry patchFileEntry,
            ShareDexDiffPatchInfo patchInfo,  File patchedDexFile) throws IOException {
    InputStream oldDexStream = null;
    InputStream patchFileStream = null;

    oldDexStream = new BufferedInputStream(baseApk.getInputStream(oldDexEntry));
    patchFileStream = (patchFileEntry != null ? new BufferedInputStream(patchPkg.getInputStream(patchFileEntry)) : null);

    new DexPatchApplier(oldDexStream, patchFileStream).executeAndSaveTo(patchedDexFile);

}

Get the InputStream of its internal file through ZipFile, which is actually reading The local apk corresponding dex file, and patch corresponds to the dex file, and the executeAndSaveTo method is used for both Merge into patchedDexFile, which is the target private directory of patch. As for the mergealgorithm, this is actually the core of Tinker. If you are interested, you can refer to this article!

The above is the detailed content of Android development—a brief introduction to hot-fixing Tinker source code. For more information, please follow other related articles on the PHP Chinese website!