Android Art Hook 技术方案

0x1 开始

Anddroid上的ART从5.0之后变成默认的选择,可见ART的重要性,目前关于Dalvik Hook方面研究的文章很多,但我在网上却找不到关于ART Hook相关的文章,甚至连鼎鼎大名的XPosed和Cydia Substrate到目前为止也不支持ART的Hook。当然我相信,技术方案他们肯定是的,估计卡在机型适配上的了。

既然网上找不到相关的资料,于是我决定自己花些时间去研究一下,终于黃天不负有心人,我找到了一个切实可行的方法,即本文所介绍的方法。

应该说明的是本文所介绍的方法肯定不是最好的,但大家看完本文之后,如果能启发大家找到更好的ART Hook方法,那我抛砖引玉的目的就达到了。废话不多说,我们开始吧。

  • 运行环境: 4.4.2 ART模式的模拟器
  • 开发环境: Mac OS X 10.10.3

0x2 ART类方法加载及执行

在ART中类方法的执行要比在Dalvik中要复杂得多,Dalvik如果除去JIT部分,可以理解为是一个解析执行的虚拟机,而ART则同时包含本地指令执行和解析执行两种模式,同时所生成的oat文件也包含两种类型,分别是portable和quick。portable和quick的主要区别是对于方法的加载机制不相同,quick大量使用了Lazy Load机制,因此应用的启动速度更快,但加载流程更复杂。其中quick是作为默认选项,因此本文所涉及的技术分析都是基于quick类型的。

由于ART存在本地指令执行和解析执行两种模式,因此类方法之间并不是能直接跳转的,而是通过一些预先定义的bridge函数进行状态和上下文的切换,这里引用一下老罗博客中的示意图: 

执行示意图

当执行某个方法时,如果当前是本地指令执行模式,则会执行ArtMethod::GetEntryPointFromCompiledCode()指向的函数,否则则执行ArtMethod::GetEntryPointFromInterpreter()指向的函数。因此每个方法,都有两个入口点,分别保存在ArtMethod::entry_point_from_compiled_code_ArtMethod::entry_point_from_interpreter_。了解这一点非常重要,后面我们主要就是在这两个入口做文章。

在讲述原理之前,需要先把以下两个流程了解清楚,这里的内容要展开是非常庞大的,我针对Hook的关键点,简明扼要的描述一下,但还是强烈建议大家去老罗的博客里细读一下其中关于ART的几篇文章。

  • ArtMethod加载流程

这个过程发生在oat被装载进内存并进行类方法链接的时候,类方法链接的代码在art/runtime/class_linker.cc中的LinkCode,如下所示:

static void LinkCode(SirtRef<mirror::ArtMethod>& method, const OatFile::OatClass* oat_class, uint32_t method_index)
    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {

  // Method shouldn't have already been linked.
  DCHECK(method->GetEntryPointFromCompiledCode() == NULL);
  // Every kind of method should at least get an invoke stub from the oat_method.
  // non-abstract methods also get their code pointers.
  const OatFile::OatMethod oat_method = oat_class->GetOatMethod(method_index);

  // 这里默认会把method::entry_point_from_compiled_code_设置oatmethod的code
  oat_method.LinkMethod(method.get());

  // Install entry point from interpreter.
  Runtime* runtime = Runtime::Current();
  bool enter_interpreter = NeedsInterpreter(method.get(), method->GetEntryPointFromCompiledCode()); //判断方法是否需要解析执行

  // 设置解析执行的入口点
  if (enter_interpreter) {
    method->SetEntryPointFromInterpreter(interpreter::artInterpreterToInterpreterBridge);
  } else {
    method->SetEntryPointFromInterpreter(artInterpreterToCompiledCodeBridge);
  }

  // 下面是设置本地指令执行的入口点
  if (method->IsAbstract()) {
    method->SetEntryPointFromCompiledCode(GetCompiledCodeToInterpreterBridge());
    return;
  }

  // 这里比较难理解,如果是静态方法,但不是clinit,但需要把entry_point_from_compiled_code_设置为GetResolutionTrampoline的返回值
  if (method->IsStatic() && !method->IsConstructor()) {
    // For static methods excluding the class initializer, install the trampoline.
    // It will be replaced by the proper entry point by ClassLinker::FixupStaticTrampolines
    // after initializing class (see ClassLinker::InitializeClass method).
    method->SetEntryPointFromCompiledCode(GetResolutionTrampoline(runtime->GetClassLinker()));
  } else if (enter_interpreter) {
    // Set entry point from compiled code if there's no code or in interpreter only mode.
    method->SetEntryPointFromCompiledCode(GetCompiledCodeToInterpreterBridge());
  }

  if (method->IsNative()) {
    // Unregistering restores the dlsym lookup stub.
    method->UnregisterNative(Thread::Current());
  }

  // Allow instrumentation its chance to hijack code.
  runtime->GetInstrumentation()->UpdateMethodsCode(method.get(),method->GetEntryPointFromCompiledCode());
}
  • 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

通过上面的代码我们可以得到,一个ArtMethod的入口主要有以下几种:

  1. Interpreter2Interpreter对应artInterpreterToInterpreterBridge(art/runtime/interpreter/interpreter.cc);
  2. Interpreter2CompledCode对应artInterpreterToCompiledCodeBridge(/art/runtime/entrypoints/interpreter/interpreter_entrypoints.cc);
  3. CompliedCode2Interpreter对应art_quick_to_interpreter_bridge(art/runtime/arch/arm/quick_entrypoints_arm.S);
  4. CompliedCode2ResolutionTrampoline对应art_quick_resolution_trampoline(art/runtime/arch/arm/quick_entrypoints_arm.S);
  5. CompliedCode2CompliedCode这个入口是直接指向oat中的指令,详细可见OatMethod::LinkMethod;

其中调用约定主要有两种,分别是:

  1. typedef void (EntryPointFromInterpreter)(Thread* self, MethodHelper& mh, const DexFile::CodeItem* code_item, ShadowFrame* shadow_frame, JValue* result), 这种对应上述1,3两种入口;
  2. 剩下的2,4,5三种入口对应的是CompledCode的入口,代码中并没有直接给出,但我们通过分析ArtMethod::Invoke的方法调用,就可以知道其调用约定了。Invoke过程中会调用art_quick_invoke_stub(/art/runtime/arch/arm/quick_entrypoints_arm.S),代码如下所示:

     /*
     * Quick invocation stub.
     * On entry:
     *   r0 = method pointer
     *   r1 = argument array or NULL for no argument methods
     *   r2 = size of argument array in bytes
     *   r3 = (managed) thread pointer
     *   [sp] = JValue* result
     *   [sp + 4] = result type char
     */
    ENTRY art_quick_invoke_stub
    push   {r0, r4, r5, r9, r11, lr}       @ spill regs
    .save  {r0, r4, r5, r9, r11, lr}
    .pad #24
    .cfi_adjust_cfa_offset 24
    .cfi_rel_offset r0, 0
    .cfi_rel_offset r4, 4
    .cfi_rel_offset r5, 8
    .cfi_rel_offset r9, 12
    .cfi_rel_offset r11, 16
    .cfi_rel_offset lr, 20
    mov    r11, sp                         @ save the stack pointer
    .cfi_def_cfa_register r11
    mov    r9, r3                          @ move managed thread pointer into r9
    mov    r4, #SUSPEND_CHECK_INTERVAL     @ reset r4 to suspend check interval
    add    r5, r2, #16                     @ create space for method pointer in frame
    and    r5, #0xFFFFFFF0                 @ align frame size to 16 bytes
    sub    sp, r5                          @ reserve stack space for argument array
    add    r0, sp, #4                      @ pass stack pointer + method ptr as dest for memcpy
    bl     memcpy                          @ memcpy (dest, src, bytes)
    ldr    r0, [r11]                       @ restore method*
    ldr    r1, [sp, #4]                    @ copy arg value for r1
    ldr    r2, [sp, #8]                    @ copy arg value for r2
    ldr    r3, [sp, #12]                   @ copy arg value for r3
    mov    ip, #0                          @ set ip to 0
    str    ip, [sp]                        @ store NULL for method* at bottom of frame
    ldr    ip, [r0, #METHOD_CODE_OFFSET]   @ get pointer to the code
    blx    ip                              @ call the method
    mov    sp, r11                         @ restore the stack pointer
    ldr    ip, [sp, #24]                   @ load the result pointer
    strd   r0, [ip]                        @ store r0/r1 into result pointer
    pop    {r0, r4, r5, r9, r11, lr}       @ restore spill regs
    .cfi_adjust_cfa_offset -24
    bx     lr
    END art_quick_invoke_stub
    • 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

“ldr ip, [r0, #METHOD_CODE_OFFSET]”其实就是把ArtMethod::entry_point_from_compiled_code_赋值给ip,然后通过blx直接调用。通过这段小小的汇编代码,我们得出如下堆栈的布局:

   -(low)
   | caller(Method *)   | <- sp 
   | arg1               | <- r1
   | arg2               | <- r2
   | arg3               | <- r3
   | ...                | 
   | argN               |
   | callee(Method *)   | <- r0
   +(high)
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9

这种调用约定并不是平时我们所见的调用约定,主要体现在参数当超过4时,并不是从sp开始保存,而是从sp + 20这个位置开始存储,所以这就是为什么在代码里entry_point_from_compiled_code_的类型是void *的原因了,因为无法用代码表示。

理解好这个调用约定对我们方案的实现至头重要

  • ArtMethod执行流程

上面详细讲述了类方法加载和链接的过程,但在实际执行的过程中,其实还不是直接调用ArtMethod的entry_point(解析执行和本地指令执行的入口),为了加快执行速度,ART为oat文件中的每个dex创建了一个DexCache(art/runtime/mirror/dex_cache.h)结构,这个结构会按dex的结构生成一系列的数组,这里我们只分析它里面的methods字段。 DexCache初始化的方法是Init,实现如下:

void DexCache::Init(const DexFile* dex_file,
                    String* location,
                    ObjectArray* strings,
                    ObjectArray* resolved_types,
                    ObjectArray* resolved_methods,
                    ObjectArray* resolved_fields,
                    ObjectArray* initialized_static_storage) {
  //...
  //...
  Runtime* runtime = Runtime::Current();
  if (runtime->HasResolutionMethod()) {
    // Initialize the resolve methods array to contain trampolines for resolution.
    ArtMethod* trampoline = runtime->GetResolutionMethod();
    size_t length = resolved_methods->GetLength();
    for (size_t i = 0; i < length; i++) {
      resolved_methods->SetWithoutChecks(i, trampoline);
    }
  }
}
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19

根据dex方法的个数,产生相应长度resolved_methods数组,然后每一个都用Runtime::GetResolutionMethod()返回的结果进行填充,这个方法是由Runtime::CreateResolutionMethod产生的,代码如下:

mirror::ArtMethod* Runtime::CreateResolutionMethod() {
  mirror::Class* method_class = mirror::ArtMethod::GetJavaLangReflectArtMethod();
  Thread* self = Thread::Current();
  SirtRef<mirror::ArtMethod>
      method(self, down_cast<mirror::ArtMethod*>(method_class->AllocObject(self)));
  method->SetDeclaringClass(method_class);
  // TODO: use a special method for resolution method saves
  method->SetDexMethodIndex(DexFile::kDexNoIndex);
  // When compiling, the code pointer will get set later when the image is loaded.
  Runtime* r = Runtime::Current();
  ClassLinker* cl = r->GetClassLinker();
  method->SetEntryPointFromCompiledCode(r->IsCompiler() ? NULL : GetResolutionTrampoline(cl));
  return method.get();
}
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14

从method->SetDexMethodIndex(DexFile::kDexNoIndex)这句得知,所有的ResolutionMethod的methodIndexDexFile::kDexNoIndex。而ResolutionMethod的entrypoint就是我们上面入口分析中的第4种情况,GetResolutionTrampoline最终返回的入口为art_quick_resolution_trampoline(art/runtime/arch/arm/quick_entrypoints_arm.S)。我们看一下其实现代码:

    .extern artQuickResolutionTrampoline
ENTRY art_quick_resolution_trampoline
    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME
    mov     r2, r9                 @ pass Thread::Current
    mov     r3, sp                 @ pass SP
    blx     artQuickResolutionTrampoline  @ (Method* called, receiver, Thread*, SP)
    cbz     r0, 1f                 @ is code pointer null? goto exception
    mov     r12, r0
    ldr  r0, [sp, #0]              @ load resolved method in r0
    ldr  r1, [sp, #8]              @ restore non-callee save r1
    ldrd r2, [sp, #12]             @ restore non-callee saves r2-r3
    ldr  lr, [sp, #44]             @ restore lr
    add  sp, #48                   @ rewind sp
    .cfi_adjust_cfa_offset -48
    bx      r12                    @ tail-call into actual code
1:
    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
    DELIVER_PENDING_EXCEPTION
END art_quick_resolution_trampoline
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20

调整好寄存器后,直接跳转至artQuickResolutionTrampoline(art/runtime/entrypoints/quick/quick_trampoline_entrypoints.cc),接下来我们分析这个方法的实现(大家不要晕了。。。,我会把无关紧要的代码去掉):

// Lazily resolve a method for quick. Called by stub code.
extern "C" const void* artQuickResolutionTrampoline(mirror::ArtMethod* called,
                                                    mirror::Object* receiver,
                                                    Thread* thread, mirror::ArtMethod** sp)
    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
  FinishCalleeSaveFrameSetup(thread, sp, Runtime::kRefsAndArgs);
  // Start new JNI local reference state
  JNIEnvExt* env = thread->GetJniEnv();
  ScopedObjectAccessUnchecked soa(env);
  ScopedJniEnvLocalRefState env_state(env);
  const char* old_cause = thread->StartAssertNoThreadSuspension("Quick method resolution set up");

  // Compute details about the called method (avoid GCs)
  ClassLinker* linker = Runtime::Current()->GetClassLinker();
  mirror::ArtMethod* caller = QuickArgumentVisitor::GetCallingMethod(sp);
  InvokeType invoke_type;
  const DexFile* dex_file;
  uint32_t dex_method_idx;
  if (called->IsRuntimeMethod()) {
    //...
    //...
  } else {
    invoke_type = kStatic;
    dex_file = &MethodHelper(called).GetDexFile();
    dex_method_idx = called->GetDexMethodIndex();
  }

  //...

  // Resolve method filling in dex cache.
  if (called->IsRuntimeMethod()) {
    called = linker->ResolveMethod(dex_method_idx, caller, invoke_type);
  }

  const void* code = NULL;
  if (LIKELY(!thread->IsExceptionPending())) {
    //...

    linker->EnsureInitialized(called_class, true, true);

    //...
  }
  // ...
  return code;
}
  • 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
inline bool ArtMethod::IsRuntimeMethod() const {
  return GetDexMethodIndex() == DexFile::kDexNoIndex;
}
  • 1
  • 2
  • 3

called->IsRuntimeMethod()用于判断当前方法是否为ResolutionMethod。如果是,那么就走ClassLinker::ResolveMethod流程去获取真正的方法,见代码:

mirror::ArtMethod* ClassLinker::ResolveMethod(const DexFile& dex_file,
                                                   uint32_t method_idx,
                                                   mirror::DexCache* dex_cache,
                                                   mirror::ClassLoader* class_loader,
                                                   const mirror::ArtMethod* referrer,
                                                   InvokeType type) {
  DCHECK(dex_cache != NULL);
  // Check for hit in the dex cache.
  mirror::ArtMethod* resolved = dex_cache->GetResolvedMethod(method_idx);
  if (resolved != NULL) {
    return resolved;
  }
  // Fail, get the declaring class.
  const DexFile::MethodId& method_id = dex_file.GetMethodId(method_idx);
  mirror::Class* klass = ResolveType(dex_file, method_id.class_idx_, dex_cache, class_loader);

  if (klass == NULL) {
    DCHECK(Thread::Current()->IsExceptionPending());
    return NULL;
  }

  // Scan using method_idx, this saves string compares but will only hit for matching dex
  // caches/files.
  switch (type) {
    case kDirect:  // Fall-through.
    case kStatic:
      resolved = klass->FindDirectMethod(dex_cache, method_idx);
      break;
    case kInterface:
      resolved = klass->FindInterfaceMethod(dex_cache, method_idx);
      DCHECK(resolved == NULL || resolved->GetDeclaringClass()->IsInterface());
      break;
    case kSuper:  // Fall-through.
    case kVirtual:
      resolved = klass->FindVirtualMethod(dex_cache, method_idx);
      break;
    default:
      LOG(FATAL) << "Unreachable - invocation type: " << type;
  }

  if (resolved == NULL) {
    // Search by name, which works across dex files.
    const char* name = dex_file.StringDataByIdx(method_id.name_idx_);
    std::string signature(dex_file.CreateMethodSignature(method_id.proto_idx_, NULL));
    switch (type) {
      case kDirect:  // Fall-through.
      case kStatic:
        resolved = klass->FindDirectMethod(name, signature);
        break;
      case kInterface:
        resolved = klass->FindInterfaceMethod(name, signature);
        DCHECK(resolved == NULL || resolved->GetDeclaringClass()->IsInterface());
        break;
      case kSuper:  // Fall-through.
      case kVirtual:
        resolved = klass->FindVirtualMethod(name, signature);
        break;
    }
  }


  if (resolved != NULL) {
    // Be a good citizen and update the dex cache to speed subsequent calls.
    dex_cache->SetResolvedMethod(method_idx, resolved);
    return resolved;
  } else {
    // ...
    }
}
  • 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

其实这里发生了“连锁反应”,ClassLinker::ResolveType走的流程,跟ResolveMethod是非常类似的,有兴趣的朋友可以跟一下。 
找到解析后的klass,再经过一轮疯狂的搜索,把找到的resolved通过DexCache::SetResolvedMethod覆盖掉之前的“替身”。当再下次再通过ResolveMethod解析方法时,就可以直接把该方法返回,不需要再解析了。

我们回过头来再重新“复现”一下这个过程,当我们首次调用某个类方法,其过程如下所示:

  1. 调用ResolutionMethod的entrypoint,进入art_quick_resolution_trampoline;
  2. art_quick_resolution_trampoline跳转到artQuickResolutionTrampoline;
  3. artQuickResolutionTrampoline调用ClassLinker::ResolveMethod解析类方法;
  4. ClassLinker::ResolveMethod调用ClassLinkder::ResolveType解析类,再从解析好的类寻找真正的方法;
  5. 调用DexCache::SetResolvedMethod,用真正的方法覆盖掉“替身”方法;
  6. 调用真正方法的entrypoint代码;

也许你会问,为什么要把过程搞得这么绕? 一切都是为了延迟加载,提高启动速度,这个过程跟ELF Linker的PLT/GOT符号重定向的过程是何其相似啊,所以技术都是想通的,一通百明。

0x3 Hook ArtMethod

通过上述ArtMethod加载和执行两个流程的分析,对于如何Hook ArtMethod,我想到了两个方案,分别

  1. 修改DexCach里的methods,把里面的entrypoint修改为自己的,做一个中转处理;
  2. 直接修改加载后的ArtMethod的entrypoint,同样做一个中转处理;

上面两个方法都是可行的,但由于我希望整个项目可以在NDK环境(而不是在源码下)下编译,因为就采用了方案2,因为通过JNI的接口就可以直接获取解析之后的ArtMethod,可以减少很多文件依赖。

回到前面的调用约定,每个ArtMethod都有两个约定,按道理我们应该准备两个中转函数的,但这里我们不考虑强制解析模式执行,所以只要处理好entry_point_from_compiled_code的中转即可。

首先,我们找到对应的方法,先保存其entrypoint,然后再把我们的中转函数art_quick_dispatcher覆盖,代码如下所示:

extern int __attribute__ ((visibility ("hidden"))) art_java_method_hook(JNIEnv* env, HookInfo *info) {
    const char* classDesc = info->classDesc;
    const char* methodName = info->methodName;
    const char* methodSig = info->methodSig;
    const bool isStaticMethod = info->isStaticMethod;

    // TODO we can find class by special classloader what do just like dvm
    jclass claxx = env->FindClass(classDesc);
    if(claxx == NULL){
        LOGE("[-] %s class not found", classDesc);
        return -1;
    }

    jmethodID methid = isStaticMethod ?
            env->GetStaticMethodID(claxx, methodName, methodSig) :
            env->GetMethodID(claxx, methodName, methodSig);

    if(methid == NULL){
        LOGE("[-] %s->%s method not found", classDesc, methodName);
        return -1;
    }

    ArtMethod *artmeth = reinterpret_cast(methid);

    if(art_quick_dispatcher != artmeth->GetEntryPointFromCompiledCode()){
        uint64_t (*entrypoint)(ArtMethod* method, Object *thiz, u4 *arg1, u4 *arg2);
        entrypoint = (uint64_t (*)(ArtMethod*, Object *, u4 *, u4 *))artmeth->GetEntryPointFromCompiledCode();

        info->entrypoint = (const void *)entrypoint;
        info->nativecode = artmeth->GetNativeMethod();

        artmeth->SetEntryPointFromCompiledCode((const void *)art_quick_dispatcher);

        // save info to nativecode :)
        artmeth->SetNativeMethod((const void *)info);

        LOGI("[+] %s->%s was hooked\n", classDesc, methodName);
    }else{
        LOGW("[*] %s->%s method had been hooked", classDesc, methodName);
    }

    return 0;
}
  • 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

我们关键的信息通过ArtMethod::SetNativeMethod保存起来了。

考虑到ART特殊的调用约定,art_quick_dispatcher只能用汇编实现了,把寄存器适当的调整一下,再跳转到另一个函数artQuickToDispatcher,这样就可以很方便用c/c++访问参数了。

先看一下art_quick_dispatcher函数的实现如下:

/*
 * Art Quick Dispatcher.
 * On entry:
 *   r0 = method pointer
 *   r1 = arg1
 *   r2 = arg2
 *   r3 = arg3
 *   [sp] = method pointer
 *   [sp + 4] = addr of thiz
 *   [sp + 8] = addr of arg1
 *   [sp + 12] = addr of arg2
 *   [sp + 16] = addr of arg3
 * and so on
 */
    .extern artQuickToDispatcher
ENTRY art_quick_dispatcher
    push    {r4, r5, lr}           @ sp - 12
    mov     r0, r0                 @ pass r0 to method
    str     r1, [sp, #(12 + 4)]
    str     r2, [sp, #(12 + 8)]
    str     r3, [sp, #(12 + 12)]
    mov     r1, r9                 @ pass r1 to thread
    add     r2, sp, #(12 + 4)      @ pass r2 to args array
    add     r3, sp, #12            @ pass r3 to old SP
    blx     artQuickToDispatcher   @ (Method* method, Thread*, u4 **, u4 **)
    pop     {r4, r5, pc}           @ return on success, r0 and r1 hold the result
END art_quick_dispatcher
  • 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

我把r2指向参数数组,这样就我们就可以非常方便的访问所有参数了。另外,我用r3保存了旧的sp地址,这样是为后面调用原来的entrypoint做准备的。我们先看看artQuickToDispatcher的实现:

extern "C" uint64_t artQuickToDispatcher(ArtMethod* method, Thread *self, u4 **args, u4 **old_sp){
    HookInfo *info = (HookInfo *)method->GetNativeMethod();
    LOGI("[+] entry ArtHandler %s->%s", info->classDesc, info->methodName);

    // If it not is static method, then args[0] was pointing to this
    if(!info->isStaticMethod){
        Object *thiz = reinterpret_cast(args[0]);
        if(thiz != NULL){
            char *bytes = get_chars_from_utf16(thiz->GetClass()->GetName());
            LOGI("[+] thiz class is %s", bytes);
            delete bytes;
        }
    }

    const void *entrypoint = info->entrypoint;
    method->SetNativeMethod(info->nativecode); //restore nativecode for JNI method
    uint64_t res = art_quick_call_entrypoint(method, self, args, old_sp, entrypoint);

    JValue* result = (JValue* )&res;
    if(result != NULL){
        Object *obj = result->l;
        char *raw_class_name = get_chars_from_utf16(obj->GetClass()->GetName());

        if(strcmp(raw_class_name, "java.lang.String") == 0){
            char *raw_string_value = get_chars_from_utf16((String *)obj);
            LOGI("result-class %s, result-value \"%s\"", raw_class_name, raw_string_value);
            free(raw_string_value);
        }else{
            LOGI("result-class %s", raw_class_name);
        }

        free(raw_class_name);
    }

    // entrypoid may be replaced by trampoline, only once.
//  if(method->IsStatic() && !method->IsConstructor()){

    entrypoint = method->GetEntryPointFromCompiledCode();
    if(entrypoint != (const void *)art_quick_dispatcher){
        LOGW("[*] entrypoint was replaced. %s->%s", info->classDesc, info->methodName);

        method->SetEntryPointFromCompiledCode((const void *)art_quick_dispatcher);
        info->entrypoint = entrypoint;
        info->nativecode = method->GetNativeMethod();
    }

    method->SetNativeMethod((const void *)info);

//  }

    return res;
}
  • 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

这里参数解析就不详细说了,接下来是最棘手的问题——如何重新调回原来的entrypoint。

这里的关键点是要还原之前的堆栈布局,art_quick_call_entrypoint就是负责完成这个工作的,其实现如下所示:

/*
 *
 * Art Quick Call Entrypoint
 * On entry:
 *  r0 = method pointer
 *  r1 = thread pointer
 *  r2 = args arrays pointer
 *  r3 = old_sp
 *  [sp] = entrypoint
 */
ENTRY art_quick_call_entrypoint
    push    {r4, r5, lr}           @ sp - 12
    sub     sp, #(40 + 20)         @ sp - 40 - 20
    str     r0, [sp, #(40 + 0)]    @ var_40_0 = method_pointer
    str     r1, [sp, #(40 + 4)]    @ var_40_4 = thread_pointer
    str     r2, [sp, #(40 + 8)]    @ var_40_8 = args_array
    str     r3, [sp, #(40 + 12)]   @ var_40_12 = old_sp
    mov     r0, sp
    mov     r1, r3
    ldr     r2, =40
    blx     memcpy                 @ memcpy(dest, src, size_of_byte)
    ldr     r0, [sp, #(40 + 0)]    @ restore method to r0
    ldr     r1, [sp, #(40 + 4)]
    mov     r9, r1                 @ restore thread to r9
    ldr     r5, [sp, #(40 + 8)]    @ pass r5 to args_array
    ldr     r1, [r5]               @ restore arg1
    ldr     r2, [r5, #4]           @ restore arg2
    ldr     r3, [r5, #8]           @ restore arg3
    ldr     r5, [sp, #(40 + 20 + 12)] @ pass ip to entrypoint
    blx     r5
    add     sp, #(40 + 20)
    pop     {r4, r5, pc}           @ return on success, r0 and r1 hold the result
END art_quick_call_entrypoint
  • 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

这里我偷懒了,直接申请了10个参数的空间,再使用之前传进入来的old_sp进行恢复,使用memcpy直接复制40字节。之后就是还原r0, r1, r2, r3, r9的值了。调用entrypoint完后,结果保存在r0和r1,再返回给artQuickToDispatcher。

至此,整个ART Hook就分析完毕了。

0x4 4.4与5.X上实现的区别

我的整个方案都是在4.4上测试的,主要是因为我只有4.4的源码,而且硬盘空间不足,实在装不下5.x的源码了。但整个思路,是完全可以套用用5.X上。另外,5.X的实现代码比4.4上复杂了很多,否能像我这样在NDK下编译完成就不知道了。

正常的4.4模拟器是以dalvik启动的,要到设置里改为art,这里会要求进行重启,但一般无效,我们手动关闭再重新打开就OK了,但需要等上一段时间才可以。

0x5 结束

虽然这篇文章只是介绍了Art Hook的技术方案,但其中的技术原理,对于如何在ART上进行代码加固、动态代码原理等等也是很有启发性。

老样子,整个项目的代码,我已经提交到https://github.com/boyliang/AllHookInOne,大家遇到什么问题,欢迎提问,有问题记得反馈。

对了,请用https://github.com/boyliang/ndk-patch给你的NDK打一下patch。

你可能感兴趣的:(Android Art Hook 技术方案)