之前的Android热更新实践 里面使用替换默认ClassLoader的方式实现了热修复,但偶然发现这种方式在加载某些系统so库的时候出现了问题。
背景是我们的某个功能依赖了我司开发的XXX.so这个系统so,这个XXX.so依赖了libandroid_runtime.so,而libandroid_runtime.so又依赖了libandroidicu.so:
然后在应用层调用System.loadLibrary去加载XXX.so的时候报了下面的异常:
1 2 3 4 03-26 02:27:07.385 3671 3695 E Demo: err : java.lang.UnsatisfiedLinkError: dlopen failed: library "libandroidicu.so" not found: needed by /system/lib64/libandroid_runtime.so in namespace classloader-namespace 03-26 02:27:07.385 3671 3695 E Demo: at java.lang.Runtime.loadLibrary0(Runtime.java:1077) 03-26 02:27:07.385 3671 3695 E Demo: at java.lang.Runtime.loadLibrary0(Runtime.java:998) 03-26 02:27:07.385 3671 3695 E Demo: at java.lang.System.loadLibrary(System.java:1661)
so加载 从报错来看是libandroidicu.so在classloader-namespace这个命名空间里面不可访问。原生库的命名空间 是安卓7.0引入的东西,目的在于限制native层私有api的访问。
从文档上并不能直接定位到我们的问题,但是从触发异常的条件”默认的ClassLoader去加载这个so是没有问题的,用我们热修复的ClassLoader去加载就会出现异常”来看,
问题原因应该就是出在我们热修复的ClassLoader的对应的命名空间没有权限去加载libandroidicu.so了。
我们先用find命令看看libandroidicu.so在系统的什么目录:
1 2 3 4 5 console:/ # find . -name libandroidicu.so 2> /dev/null ./apex/com.android.i18n/lib/libandroidicu.so ./apex/com.android.i18n/lib64/libandroidicu.so ./system/apex/com.android.i18n/lib/libandroidicu.so ./system/apex/com.android.i18n/lib64/libandroidicu.so
的确不在/system/lib64下面,应用加载不到它是合理的。但我好奇的是默认的ClassLoader又是怎么加载到的?
从错误堆栈的Runtime.loadLibrary0一路往jni追踪可以发现so实际并不是由ClassLoader去加载的,而是通过ClassLoader找到了对应的NativeLoaderNamespace,然后用NativeLoaderNamespace::Load去加载的:
具体的代码调用如下:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 public class Runtime { ... public void loadLibrary (String libname) { loadLibrary0(Reflection.getCallerClass(), libname); } void loadLibrary0 (Class<?> fromClass, String libname) { ClassLoader classLoader = ClassLoader.getClassLoader(fromClass); loadLibrary0(classLoader, fromClass, libname); } private synchronized void loadLibrary0 (ClassLoader loader, Class<?> callerClass, String libname) { ... nativeLoad(filename, loader); ... } private static String nativeLoad (String filename, ClassLoader loader) { return nativeLoad(filename, loader, null ); } private static native String nativeLoad (String filename, ClassLoader loader, Class<?> caller) ; ... }
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 JNIEXPORT jstring JNICALL Runtime_nativeLoad (JNIEnv* env, jclass ignored, jstring javaFilename, jobject javaLoader, jclass caller) { return JVM_NativeLoad (env, javaFilename, javaLoader, caller); } static JNINativeMethod gMethods[] = { FAST_NATIVE_METHOD (Runtime, freeMemory, "()J" ), FAST_NATIVE_METHOD (Runtime, totalMemory, "()J" ), FAST_NATIVE_METHOD (Runtime, maxMemory, "()J" ), NATIVE_METHOD (Runtime, nativeGc, "()V" ), NATIVE_METHOD (Runtime, nativeExit, "(I)V" ), NATIVE_METHOD (Runtime, nativeLoad, "(Ljava/lang/String;Ljava/lang/ClassLoader;Ljava/lang/Class;)" "Ljava/lang/String;" ), }; JNIEXPORT jstring JVM_NativeLoad (JNIEnv* env, jstring javaFilename, jobject javaLoader, jclass caller) { ... bool success = vm->LoadNativeLibrary (env, filename.c_str (), javaLoader, caller, &error_msg); ... } bool JavaVMExt::LoadNativeLibrary (JNIEnv* env, const std::string& path, jobject class_loader, jclass caller_class, std::string* error_msg) { ... void * handle = android::OpenNativeLibrary ( env, runtime_->GetTargetSdkVersion (), path_str, class_loader, (caller_location.empty () ? nullptr : caller_location.c_str ()), library_path.get (), &needs_native_bridge, &nativeloader_error_msg); ... } LibraryNamespaces* g_namespaces = new LibraryNamespaces; void * OpenNativeLibrary (JNIEnv* env, int32_t target_sdk_version, const char * path, jobject class_loader, const char * caller_location, jstring library_path, bool * needs_native_bridge, char ** error_msg) { ... std::lock_guard<std::mutex> guard (g_namespaces_mutex) ; NativeLoaderNamespace* ns; if ((ns = g_namespaces->FindNamespaceByClassLoader (env, class_loader)) == nullptr ) { Result<NativeLoaderNamespace*> isolated_ns = CreateClassLoaderNamespaceLocked (env, target_sdk_version, class_loader, false , nullptr , library_path, nullptr , nullptr ); if (!isolated_ns.ok ()) { *error_msg = strdup (isolated_ns.error ().message ().c_str ()); return nullptr ; } else { ns = *isolated_ns; } } return OpenNativeLibraryInNamespace (ns, path, needs_native_bridge, error_msg); ... } void * OpenNativeLibraryInNamespace (NativeLoaderNamespace* ns, const char * path, bool * needs_native_bridge, char ** error_msg) { auto handle = ns->Load (path); ... return handle.ok () ? *handle : nullptr ; }
从ld.config.txt可以看到不同的namespace有不同的search paths:
1 2 3 4 5 6 additional.namespaces = com_android_adbd,com_android_art,com_android_conscrypt,com_android_cronet,com_android_i18n,com_android_media,com_android_neuralnetworks,com_android_os_statsd,com_android_resolv,com_android_runtime,com_product_service1,product,rs,sphal,vndk,vndk_product ... namespace.default.search.paths = /system/${LIB} namespace.default.search.paths += /system_ext/${LIB} ... namespace.com_android_i18n.search.paths = /apex/com.android.i18n/${LIB}
public libs g_namespaces负责NativeLoaderNamespace的创建和缓存,NativeLoaderNamespace也是通过g_namespaces去Create出来的:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Result<NativeLoaderNamespace*> CreateClassLoaderNamespaceLocked (JNIEnv* env, int32_t target_sdk_version, jobject class_loader, bool is_shared, jstring dex_path, jstring library_path, jstring permitted_path, jstring uses_library_list) REQUIRES (g_namespaces_mutex) { Result<NativeLoaderNamespace*> ns = g_namespaces->Create (env, target_sdk_version, class_loader, is_shared, dex_path, library_path, permitted_path, uses_library_list); ... return ns; }
我们看看g_namespaces的Create方法创建NativeLoaderNamespace干了些啥:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Result<NativeLoaderNamespace*> LibraryNamespaces::Create (JNIEnv* env, uint32_t target_sdk_version, jobject class_loader, bool is_shared, jstring dex_path_j, jstring java_library_path, jstring java_permitted_path, jstring uses_library_list) { ... auto app_ns = NativeLoaderNamespace::Create ( namespace_name, library_path, permitted_path, parent_ns, is_shared, target_sdk_version < 24 , also_used_as_anonymous); ... for (const auto &[apex_ns_name, public_libs] : apex_public_libraries ()) { auto ns = NativeLoaderNamespace::GetExportedNamespace (apex_ns_name, is_bridged); if (ns.ok ()) { linked = app_ns->Link (&ns.value (), public_libs); if (!linked.ok ()) { return linked.error (); } } } ... }
可以看到它在创建出app_ns之后会遍历apex_public_libraries去链接apex里面的公共库:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 constexpr const char * kApexLibrariesConfigFile = "/linkerconfig/apex.libraries.config.txt" ;const std::map<std::string, std::string>& apex_public_libraries () { static std::map<std::string, std::string> public_libraries = InitApexLibraries ("public" ); return public_libraries; } static std::map<std::string, std::string> InitApexLibraries (const std::string& tag) { std::string file_content; if (!base::ReadFileToString (kApexLibrariesConfigFile, &file_content)) { return {}; } auto config = ParseApexLibrariesConfig (file_content, tag); if (!config.ok ()) { LOG_ALWAYS_FATAL ("%s: %s" , kApexLibrariesConfigFile, config.error ().message ().c_str ()); return {}; } return *config; }
在实机的/linkerconfig/apex.libraries.config.txt下可以看到:
1 2 3 4 5 6 7 8 9 jni com_android_appsearch libicing.so public com_android_art libnativehelper.so jni com_android_btservices libbluetooth_jni.so jni com_android_conscrypt libjavacrypto.so public com_android_i18n libicui18n.so:libicuuc.so:libicu.so public com_android_neuralnetworks libneuralnetworks.so jni com_android_os_statsd libstats_jni.so jni com_android_tethering libframework-connectivity-jni.so:libframework-connectivity-tiramisu-jni.so:libandroid_net_connectivity_com_android_net_module_util_jni.so:libservice-connectivity.so jni com_android_uwb libuwb_uci_jni_rust.so
com_android_i18n的libicui18n.so、libicuuc.so、libicu.so是公共的可以直接访问会被链接到app_ns允许访问。而报错的libandroidicu.so的确没有在public里面,所以不会被链接,于是无法访问。
shared libs 实际上除了这个public libs配置之外,还有个shared libs的配置可以用于配置NativeLoaderNamespace对哪些NativeLoaderNamespace暴露哪些so。
详细的规则可以参考链接器命名空间 的文档
具体的原理我们可以继续往下追一层看看NativeLoaderNamespace::Create是怎么创建app_ns的:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 Result<NativeLoaderNamespace> NativeLoaderNamespace::Create ( const std::string& name, const std::string& search_paths, const std::string& permitted_paths, const NativeLoaderNamespace* parent, bool is_shared, bool is_exempt_list_enabled, bool also_used_as_anonymous) { ... uint64_t type = ANDROID_NAMESPACE_TYPE_ISOLATED; ... if (is_shared) { type |= ANDROID_NAMESPACE_TYPE_SHARED; } ... android_namespace_t * raw = android_create_namespace (name.c_str (), nullptr , search_paths.c_str (), type, permitted_paths.c_str (), effective_parent.ToRawAndroidNamespace ()); if (raw != nullptr ) { return NativeLoaderNamespace (name, raw); } ... }
可以看到is_shared为true的时候会给type添加一个ANDROID_NAMESPACE_TYPE_SHARED的flag,这个flag最终在create_namespace会判断:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 static android_namespace_t * g_anonymous_namespace = &g_default_namespace;... android_namespace_t * create_namespace (const void * caller_addr, const char * name, const char * ld_library_path, const char * default_library_path, uint64_t type, const char * permitted_when_isolated_path, android_namespace_t * parent_namespace) { if (parent_namespace == nullptr ) { soinfo* caller_soinfo = find_containing_library (caller_addr); parent_namespace = caller_soinfo != nullptr ? caller_soinfo->get_primary_namespace () : g_anonymous_namespace; } ProtectedDataGuard guard; std::vector<std::string> ld_library_paths; std::vector<std::string> default_library_paths; std::vector<std::string> permitted_paths; ... android_namespace_t * ns = new (g_namespace_allocator.alloc ()) android_namespace_t (); ... if ((type & ANDROID_NAMESPACE_TYPE_SHARED) != 0 ) { std::copy (parent_namespace->get_ld_library_paths ().begin (), parent_namespace->get_ld_library_paths ().end (), back_inserter (ld_library_paths)); std::copy (parent_namespace->get_default_library_paths ().begin (), parent_namespace->get_default_library_paths ().end (), back_inserter (default_library_paths)); std::copy (parent_namespace->get_permitted_paths ().begin (), parent_namespace->get_permitted_paths ().end (), back_inserter (permitted_paths)); add_soinfos_to_namespace (parent_namespace->soinfo_list (), ns); for (auto & link : parent_namespace->linked_namespaces ()) { ns->add_linked_namespace (link.linked_namespace (), link.shared_lib_sonames (), link.allow_all_shared_libs ()); } } else { add_soinfos_to_namespace (parent_namespace->get_shared_group (), ns); } ns->set_ld_library_paths (std::move (ld_library_paths)); ns->set_default_library_paths (std::move (default_library_paths)); ns->set_permitted_paths (std::move (permitted_paths)); ... }
可以看到添加了ANDROID_NAMESPACE_TYPE_SHARED这个flag的话会添加父namespace的所有so和链接父namespace连接过的namespace。而如果为false则只会添加父namespace的shared so。
系统默认创建的Classloader对应的namespace的parent_namespace为null,于是会把parent_namespace设置成默认的”default”这个namespace。
而从/linkerconfig/ld.config.txt里面可以看到”default”这个namespace是链接了com_android_i18n这个namespace,可以从里面访问libandroidicu.so:
1 2 3 4 5 6 7 8 9 10 ... namespace.default.links = com_android_adbd,com_android_i18n,default,com_android_art,com_android_resolv,com_android_tethering,com_android_neural namespace.default.link.com_android_adbd.shared_libs = libadb_pairing_auth.so namespace.default.link.com_android_adbd.shared_libs += libadb_pairing_connection.so namespace.default.link.com_android_adbd.shared_libs += libadb_pairing_server.so namespace.default.link.com_android_i18n.shared_libs = libandroidicu.so namespace.default.link.com_android_i18n.shared_libs += libicu.so namespace.default.link.com_android_i18n.shared_libs += libicui18n.so namespace.default.link.com_android_i18n.shared_libs += libicuuc.so ...
所以默认的ClassLoader创建NativeLoaderNamespace的时候is_shared是true可以加载到libandroidicu.so
而我们自定义的ClassLoader创建NativeLoaderNamespace的时候is_shared是false没有继承”default”这个parent_namepsace的links配置无法加载libandroidicu.so
另外也可以看到默认的Classloader对应的namespace会连接com_android_i18n这个命名空间两次
第一次在create\_namespace函数里由于ANDROID_NAMESPACE_TYPE_SHARED继承了defalut这个parent_namespace的links配置能访问/linkerconfig/ld.config.txt里的namespace.default.link.com_android_i18n.shared_libs配置的shared so,
第二次则是在LibraryNamespaces::Create里面赌钱/linkerconfig/apex.libraries.config.txt里public com_android_i18n配置链接com_android_i18n的public so。
链接器命名空间 这个文档里也有提到:
此属性与 public.libraries.txt 文件在底层实现上是相同的。这两种机制都通过使用库名称过滤器指定链接的方式来控制导入的共享库。
is shared 从前面CreateClassLoaderNamespaceLocked的传参来看我们的自定义Classloader创建namespace的时候is_shared的确为false:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 LibraryNamespaces* g_namespaces = new LibraryNamespaces; void * OpenNativeLibrary (JNIEnv* env, int32_t target_sdk_version, const char * path, jobject class_loader, const char * caller_location, jstring library_path, bool * needs_native_bridge, char ** error_msg) { ... std::lock_guard<std::mutex> guard (g_namespaces_mutex) ; NativeLoaderNamespace* ns; if ((ns = g_namespaces->FindNamespaceByClassLoader (env, class_loader)) == nullptr ) { Result<NativeLoaderNamespace*> isolated_ns = CreateClassLoaderNamespaceLocked (env, target_sdk_version, class_loader, false , nullptr , library_path, nullptr , nullptr ); if (!isolated_ns.ok ()) { *error_msg = strdup (isolated_ns.error ().message ().c_str ()); return nullptr ; } else { ns = *isolated_ns; } } return OpenNativeLibraryInNamespace (ns, path, needs_native_bridge, error_msg); ... }
那默认的ClassLoader的is_shared是怎么设置成true的呢?从LoadedApk代码里面可以看到系统app最终就会调用ClassLoaderFactory.createClassLoader在里面创建Classloader的同时调用createClassloaderNamespace创建namespace,传入的is_shared为isBundledApp即true:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 public ClassLoader getClassLoader () { synchronized (mLock) { if (mClassLoader == null ) { createOrUpdateClassLoaderLocked(null ); } return mClassLoader; } } private void createOrUpdateClassLoaderLocked (List<String> addedPaths) { ... boolean isBundledApp = mApplicationInfo.isSystemApp() && !mApplicationInfo.isUpdatedSystemApp(); ... mDefaultClassLoader = ApplicationLoaders.getDefault().getClassLoaderWithSharedLibraries( zip, mApplicationInfo.targetSdkVersion, isBundledApp, librarySearchPath, libraryPermittedPath, mBaseClassLoader, mApplicationInfo.classLoaderName, sharedLibraries.first, nativeSharedLibraries, sharedLibraries.second); mAppComponentFactory = createAppFactory(mApplicationInfo, mDefaultClassLoader); ... } ClassLoader getClassLoaderWithSharedLibraries ( String zip, int targetSdkVersion, boolean isBundled, String librarySearchPath, String libraryPermittedPath, ClassLoader parent, String classLoaderName, List<ClassLoader> sharedLibraries, List<String> nativeSharedLibraries, List<ClassLoader> sharedLibrariesLoadedAfterApp) { return getClassLoader(zip, targetSdkVersion, isBundled, librarySearchPath, libraryPermittedPath, parent, zip, classLoaderName, sharedLibraries, nativeSharedLibraries, sharedLibrariesLoadedAfterApp); } ClassLoader getClassLoaderWithSharedLibraries ( String zip, int targetSdkVersion, boolean isBundled, String librarySearchPath, String libraryPermittedPath, ClassLoader parent, String classLoaderName, List<ClassLoader> sharedLibraries, List<String> nativeSharedLibraries, List<ClassLoader> sharedLibrariesLoadedAfterApp) { return getClassLoader(zip, targetSdkVersion, isBundled, librarySearchPath, libraryPermittedPath, parent, zip, classLoaderName, sharedLibraries, nativeSharedLibraries, sharedLibrariesLoadedAfterApp); } rivate ClassLoader getClassLoader (String zip, int targetSdkVersion, boolean isBundled, String librarySearchPath, String libraryPermittedPath, ClassLoader parent, String cacheKey, String classLoaderName, List<ClassLoader> sharedLibraries, List<String> nativeSharedLibraries, List<ClassLoader> sharedLibrariesLoadedAfterApp) { ... ClassLoader classloader = ClassLoaderFactory.createClassLoader( zip, librarySearchPath, libraryPermittedPath, parent, targetSdkVersion, isBundled, classLoaderName, sharedLibraries, nativeSharedLibraries, sharedLibrariesLoadedAfterApp); ... return loader; ... } public static ClassLoader createClassLoader (String dexPath, String librarySearchPath, String libraryPermittedPath, ClassLoader parent, int targetSdkVersion, boolean isNamespaceShared, String classLoaderName, List<ClassLoader> sharedLibraries, List<String> nativeSharedLibraries, List<ClassLoader> sharedLibrariesAfter) {final ClassLoader classLoader = createClassLoader(dexPath, librarySearchPath, parent, classLoaderName, sharedLibraries, sharedLibrariesAfter); ... String errorMessage = createClassloaderNamespace(classLoader, targetSdkVersion, librarySearchPath, libraryPermittedPath, isNamespaceShared, dexPath, sonameList); ... return classLoader; }
解决方案 方法1 由于ClassLoaderFactory.createClassloaderNamespace是private的不能在外部调用,所以解决自定义classloader找不到libandroidicu.so的方法就是不要自己直接new ClassLoader,而是调用ClassLoaderFactory.createClassLoader去创建,传入isNamespaceShared为true。
方法2 由于系统默认的classloader对应的namespace已经加载了libandroid_runtime.so,如果将我们自定义的classloader的父classloader设置成系统默认的classloder,则自定义classloader对应的namespace的parent_namespace也会指向默认classloader的namespace。
然后这个namespace已经加载了libandroid_runtime.so,于是在后面的add_soinfos_to_namespace(parent_namespace->get_shared_group(), ns);里面就能直接使用已经加载好的shared so libandroid_runtime.so:
使用setprop debug.ld.all dlopen,dlerror命令打开全部linker打印也可以看到libandroid_runtime.so Already loaded的日志:
03-29 04:36:10.024 6046 6069 D linker : find_library_internal(ns=classloader-namespace, task=libandroid_runtime.so): Already loaded (by sona
me): /system/lib64/libandroid_runtime.so`
但是当父classloader设置成系统默认的classloader之后由于双亲委托的机制,会先从父classloader去加载class,达不到热修复的需求。
于是我们需要修改自定义classloader的loadClass打破双亲委托机制,先自己去加载class,加载不到再让父classloder去加载:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 @Override protected Class<?> loadClass(String name, boolean resolve) throws ClassNotFoundException { Class<?> c = null ; try { c = findClass(name); } catch (ClassNotFoundException e) { } if (c == null ) { c = getParent().loadClass(name); } return c; }