Youpk 脱壳机脱壳原理分析

youpk,脱壳机,脱壳,原理,分析 · 浏览次数 : 131

小编点评

The context describes the process of generating and repairing a DEX file using the dextool package. Here's a summary: **1. Initializing DEX:** - `pointer_size` gets the size of the `ClassLinker` object. - `methods` gets a list of methods from the class. - `Unpacker::getMethodDumpPath(method)` finds the path to the method dump file. **2. Reading Method Dump:** - `dump_path` stores the path to the method dump file. - `fd` stores the file descriptor for the dump file. - `index` holds the index of the DexMethod in the class. - `name` stores the name of the method. - `code_item` holds the code item associated with the method. - `total_size` calculates the total size of the data to be written to the file. **3. Writing Method Dump:** - The data is written to the file in a specific format. - `nop_size` specifies the number of bytes to write from the `DexFile::CodeItem` struct. - `write()` function writes the data to the file. **4. Repairing DEX:** - `dexfixer.jar` is used to repair the DEX file. - `dump_path` is the path to the dump file. - `output_dir` is the path to the output directory. - `args` contains the command to repair the DEX file. **5. Usage:** 1. Generate the DEX file dump. 2. Run the `dexfixer.jar` repair tool with the `dump_path` and `output_dir` as arguments. **Key Takeaways:** - The process involves reading methods from the class, writing them to a file in a specific format, and repairing the DEX file using `dexfixer.jar`. - The generated dump file is stored in the `/data/data/` directory with a filename based on the class name. - The repair tool can fix corrupt DEX files and generate a new DEX file with the repaired data.

正文

Youpk 是一个针对整体加固和Dex抽取加固壳的脱壳机

主要是基于虚拟机的,也就是基于VA的脱壳机, 相对FART出来的更晚一些, 厂商针对少一些, 脱壳位置相对更底层一些,还提供了Dex修复的工具,简直棒棒

1. 先分析整体脱壳的原理

在ActivityThread 的 handleBindApplication 中增加了代码
image

也就是说,在应用的启动流程中,在makeApplication后,就开始干活

Unpacker.java -> unpack

 public static void unpack() {
        if (Unpacker.unpackerThread != null) {
            return;
        }

        if (!shouldUnpack()) {
            return;
        }
        
        //开启线程调用
        Unpacker.unpackerThread = new Thread() {
            @Override public void run() {
                while (true) {
                    try {
                        Thread.sleep(UNPACK_INTERVAL);
                    }
                    catch (InterruptedException e) {
                        e.printStackTrace();
                    }
                    Unpacker.unpackNative();
                }
            }
        };
        Unpacker.unpackerThread.start();
    }

这里开启一个线程,每一段时间就执行一下native的unpackNative
对应的是unpacker.cc

//注册native方法

static void Unpacker_unpackNative(JNIEnv*, jclass) {
  Unpacker::unpack();
}
....

void Unpacker::unpack() {
  ScopedObjectAccess soa(Thread::Current());
  ULOGI("%s", "unpack begin!");
  //1. 初始化
  init();
  //2. dump所有dex
  dumpAllDexes();
  //3. 主动调用所有方法
  invokeAllMethods();
  //4. 还原
  fini();
  ULOGI("%s", "unpack end!");
}

init() 主要是初始化工作,比如建立dump的目录,寻找需要dump的dex

void Unpacker::init() {
  Unpacker_fake_invoke_ = false;
  Unpacker_self_ = Thread::Current();
  Unpacker_dump_dir_ = getDumpDir();
  mkdir(Unpacker_dump_dir_.c_str(), 0777);
  Unpacker_dex_dir_ = getDumpDir() + "/dex";
  mkdir(Unpacker_dex_dir_.c_str(), 0777);
  Unpacker_method_dir_ = getDumpDir() + "/method";
  mkdir(Unpacker_method_dir_.c_str(), 0777);
  Unpacker_json_path_ = getDumpDir() + "/unpacker.json";
  Unpacker_json_fd_ = -1;
  Unpacker_json_fd_ = open(Unpacker_json_path_.c_str(), O_RDWR | O_CREAT, 0777);
  if (Unpacker_json_fd_ == -1) {
    ULOGE("open %s error: %s", Unpacker_json_path_.c_str(), strerror(errno));
  }
  Unpacker_json_ = parseJson();
  if (Unpacker_json_ == nullptr) {
    Unpacker_json_ = createJson();
  }
  CHECK(Unpacker_json_ != nullptr);

  Unpacker_dex_files_ = getDexFiles();
  Unpacker_class_loader_ = getAppClassLoader();
}

Unpacker_dex_files_ 在这里进行了寻找和赋值的操作

std::list<const DexFile*> Unpacker::getDexFiles() {
  std::list<const DexFile*> dex_files;
  Thread* const self = Thread::Current();
  ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
  ReaderMutexLock mu(self, *class_linker->DexLock());
  const std::list<ClassLinker::DexCacheData>& dex_caches = class_linker->GetDexCachesData();
  for (auto it = dex_caches.begin(); it != dex_caches.end(); ++it) {
    ClassLinker::DexCacheData data = *it;
    const DexFile* dex_file = data.dex_file;
    const std::string& dex_location = dex_file->GetLocation();
    if (dex_location.rfind("/system/", 0) == 0) {
      continue;
    }
    dex_files.push_back(dex_file);
  }
  return dex_files;
}

这里通过RunTime 拿到class_linker,然后通过classLinker来获得所有的Dex的指针(看得出作者对虚拟机有比较深的研究)

dumpAllDexes();就是我们整体dump的逻辑所在

void Unpacker::dumpAllDexes() {
  for (const DexFile* dex_file : Unpacker_dex_files_) {
    std::string dump_path = getDexDumpPath(dex_file);
    if (access(dump_path.c_str(), F_OK) != -1) {
      ULOGI("%s already dumped, ignored", dump_path.c_str());
      continue;
    }
    const uint8_t* begin = dex_file->Begin();
    size_t size = dex_file->Size();
    int fd = open(dump_path.c_str(), O_RDWR | O_CREAT, 0777);
    if (fd == -1) {
      ULOGE("open %s error: %s", dump_path.c_str(), strerror(errno));
      continue;
    }

    std::vector<uint8_t> data(size);
    memcpy(data.data(), "dex\n035", 8);
    memcpy(data.data() + 8, begin + 8, size - 8);

    size_t written_size = write(fd, data.data(), size);
    if (written_size < size) {
      ULOGW("fwrite %s %zu/%zu error: %s", dump_path.c_str(), written_size, size, strerror(errno));
    }
    close(fd);
    ULOGI("dump dex %s to %s successful!", dex_file->GetLocation().c_str(), dump_path.c_str());
  }
}

整体dump最终把数据写入到了.dex文件中(还做了一个dex文件前缀魔数修复)

2. 再看对抽取壳的处理

首先是构建主动调用链,来欺骗壳,使壳进行函数指令填充
对应的就是 unpack方法中的第三步
//3. 主动调用所有方法
invokeAllMethods();

注意标志的六种状态
  //dump类的六种status: 
  //Ready: 该类准备dump
  //Resolved: ResolveClass成功
  //ResolveClassFailed: ResolveClass失败
  //Inited: EnsureInitialized成功
  //EnsureInitializedFailed: EnsureInitialized失败
  //Dumped: dump所有method成功

整体来说分两步,
一: 往unpacker.json里写每个方法的关键元数据

...
    if (dex == nullptr) {
      dex = cJSON_CreateObject();
      cJSON_AddStringToObject(dex, "location", dex_file->GetLocation().c_str());
      cJSON_AddStringToObject(dex, "dump_path", getDexDumpPath(dex_file).c_str());
      cJSON_AddNumberToObject(dex, "class_size", dex_file->NumClassDefs());
      current = cJSON_AddObjectToObject(dex, "current");
      cJSON_AddNumberToObject(current, "index", class_idx);
      cJSON_AddStringToObject(current, "descriptor", dex_file->GetClassDescriptor(dex_file->GetClassDef(class_idx)));
      cJSON_AddStringToObject(current, "status", "Ready");
      failures = cJSON_AddArrayToObject(dex, "failures");
      cJSON_AddItemToArray(dexes, dex);
    }
...

记录着dex的位置,dex整体dump下来的位置,有多少个class,class的id等等数据.方便后续codeitem.bin和整体dump的dex进行融合的操作

二: 构造参数发起主动调用

std::string Unpacker::getMethodDumpPath(ArtMethod* method) {
  CHECK(method->GetDeclaringClass() != nullptr) << method;
  const DexFile& dex_file = method->GetDeclaringClass()->GetDexFile();
  std::string dex_location = dex_file.GetLocation();
  size_t size = dex_file.Size();
  //替换windows文件不支持的字符
  for (size_t i = 0; i < dex_location.length(); i++) {
    if (dex_location[i] == '/' || dex_location[i] == ':') {
      dex_location[i] = '_';
    }
  }
  std::string dump_path = Unpacker_method_dir_ + "/" + dex_location;
  dump_path += StringPrintf("_%zu_codeitem.bin", size);
  return dump_path;
}

从这里可以看出,函数的元数据写入到unpacker.json,而函数的codeItem(即指令数据),写入到了xxx_codeitem.bin的文件中,方便后续函数修复使用

三 获得 classDef后发起对class所有方法的主动调用()

// 前面还有一步主动初始化,
...
      size_t pointer_size = class_linker->GetImagePointerSize();
      auto methods = klass->GetDeclaredMethods(pointer_size);

      Unpacker::enableFakeInvoke();
      for (auto& m : methods) {
        ArtMethod* method = &m;
        if (!method->IsProxyMethod() && method->IsInvokable()) {
          uint32_t args_size = (uint32_t)ArtMethod::NumArgRegisters(method->GetShorty());
          if (!method->IsStatic()) {
            args_size += 1;
          }
          
          JValue result;
          std::vector<uint32_t> args(args_size, 0);
          if (!method->IsStatic()) {
            mirror::Object* thiz = klass->AllocObject(self);
            args[0] = StackReference<mirror::Object>::FromMirrorPtr(thiz).AsVRegValue();  
          }
		  // 重点这里
          method->Invoke(self, args.data(), args_size, &result, method->GetShorty());
        }
      }
      Unpacker::disableFakeInvoke();

      cJSON_ReplaceItemInObject(current, "status", cJSON_CreateString("Dumped"));
      writeJson();
...

四 发起invoke后,会走到java解释器中(youpk 强制走switch解释器), youpk修改了其中的一个宏
interpreter_switch_impl.cc

#define PREAMBLE()                                                                              \
  do {                                                                                          \
    inst_count++;                                                                               \
    bool dumped = Unpacker::beforeInstructionExecute(self, shadow_frame.GetMethod(),            \
                                                     dex_pc, inst_count);                       \
    if (dumped) {                                                                               \
      return JValue();                                                                          \
    }                                                                                           \
    if (UNLIKELY(instrumentation->HasDexPcListeners())) {                                       \
      instrumentation->DexPcMovedEvent(self, shadow_frame.GetThisObject(code_item->ins_size_),  \
                                       shadow_frame.GetMethod(), dex_pc);                       \
    }                                                                                           \
  } while (false)

这个宏在每个指令执行前都会调用,那么就一定会执行到 Unpacker::beforeInstructionExecute,在这里发起了对method的codeitem的dump操作

bool Unpacker::beforeInstructionExecute(Thread *self, ArtMethod *method, uint32_t dex_pc, int inst_count) {
  if (Unpacker::isFakeInvoke(self, method)) {
    const uint16_t* const insns = method->GetCodeItem()->insns_;
    const Instruction* inst = Instruction::At(insns + dex_pc);
    uint16_t inst_data = inst->Fetch16(0);
    Instruction::Code opcode = inst->Opcode(inst_data);

    //对于一般的方法抽取(非ijiami, najia), 直接在第一条指令处dump即可
    if (inst_count == 0 && opcode != Instruction::GOTO && opcode != Instruction::GOTO_16 && opcode != Instruction::GOTO_32) {
      Unpacker::dumpMethod(method);
      return true;
    }
    //ijiami, najia的特征为: goto: goto_decrypt; nop; ... ; return; const vx, n; invoke-static xxx; goto: goto_origin;
    else if (inst_count == 0 && opcode >= Instruction::GOTO && opcode <= Instruction::GOTO_32) {
      return false;
    } else if (inst_count == 1 && opcode >= Instruction::CONST_4 && opcode <= Instruction::CONST_WIDE_HIGH16) {
      return false;
    } else if (inst_count == 2 && (opcode == Instruction::INVOKE_STATIC || opcode == Instruction::INVOKE_STATIC_RANGE)) {
      //让这条指令真正的执行
      Unpacker::disableFakeInvoke();
      Unpacker::enableRealInvoke();
      return false;
    } else if (inst_count == 3) {
      if (opcode >= Instruction::GOTO && opcode <= Instruction::GOTO_32) {
        //写入时将第一条GOTO用nop填充
        const Instruction* inst_first = Instruction::At(insns);
        Instruction::Code first_opcode = inst_first->Opcode(inst->Fetch16(0));
        CHECK(first_opcode >= Instruction::GOTO && first_opcode <= Instruction::GOTO_32);
        ULOGD("found najia/ijiami %s", PrettyMethod(method).c_str());
        switch (first_opcode)
        {
        case Instruction::GOTO:
          Unpacker::dumpMethod(method, 2);
          break;
        case Instruction::GOTO_16:
          Unpacker::dumpMethod(method, 4);
          break;
        case Instruction::GOTO_32:
          Unpacker::dumpMethod(method, 8);
          break;
        default:
          break;
        }
      } else {
        Unpacker::dumpMethod(method);
      }
      return true;
    }
    Unpacker::dumpMethod(method);
    return true;
  }
  return false;
}

从这里可以看到,它即可一脱一般的抽取壳,也可以脱那种goto类型(ijiami, najia)的抽取壳,最终会走到
dumpMethod

void Unpacker::dumpMethod(ArtMethod *method, int nop_size) {
  std::string dump_path = Unpacker::getMethodDumpPath(method);
  int fd = -1;
  if (Unpacker_method_fds_.find(dump_path) != Unpacker_method_fds_.end()) {
    fd = Unpacker_method_fds_[dump_path];
  }
  else {
    fd = open(dump_path.c_str(), O_RDWR | O_CREAT | O_APPEND, 0777);
    if (fd == -1) {
      ULOGE("open %s error: %s", dump_path.c_str(), strerror(errno));
      return;
    }
    Unpacker_method_fds_[dump_path] = fd;
  }

  uint32_t index = method->GetDexMethodIndex();
  std::string str_name = PrettyMethod(method);
  const char* name = str_name.c_str();
  const DexFile::CodeItem* code_item = method->GetCodeItem();
  uint32_t code_item_size = (uint32_t)Unpacker::getCodeItemSize(method);

  size_t total_size = 4 + strlen(name) + 1 + 4 + code_item_size;
  std::vector<uint8_t> data(total_size);
  uint8_t* buf = data.data();
  memcpy(buf, &index, 4);
  buf += 4;
  memcpy(buf, name, strlen(name) + 1);
  buf += strlen(name) + 1;
  memcpy(buf, &code_item_size, 4);
  buf += 4;
  memcpy(buf, code_item, code_item_size);
  if (nop_size != 0) {
    memset(buf + offsetof(DexFile::CodeItem, insns_), 0, nop_size);
  }

  ssize_t written_size = write(fd, data.data(), total_size);
  if (written_size > (ssize_t)total_size) {
    ULOGW("write %s in %s %zd/%zu error: %s", PrettyMethod(method).c_str(), dump_path.c_str(), written_size, total_size, strerror(errno));
  }
}

这里就是把数据按照固定的格式把数据写入到.bin文件中
脱壳完成

3. dex修复

一 adb pull出dump文件, dump文件路径为 /data/data/包名/unpacker

adb pull /data/data/xxx.xxx.myxxxdemo/unpacker

二 调用修复工具 dexfixer.jar, 两个参数, 第一个为dump文件目录(必须为有效路径), 第二个为重组后的DEX目录(不存在将会创建)
youpk 比较爽的就是这里提供了修复的jar(还有源码),而fart的只是一个修复对比文件,未真正修复到dex中

java -jar dexfixer.jar /path/to/unpacker /path/to/output

完成dex的修复

与Youpk 脱壳机脱壳原理分析相似的内容: