pthread detach和joinable的资源释放

记录一下最近看android bionic pthread相关源码的理解

这里以android kikat bionic为例

先看一下pthread_create函数源码

int pthread_create(pthread_t* thread_out, pthread_attr_t const* attr,
                   void* (*start_routine)(void*), void* arg) {
  ErrnoRestorer errno_restorer;

  // Inform the rest of the C library that at least one thread
  // was created. This will enforce certain functions to acquire/release
  // locks (e.g. atexit()) to protect shared global structures.
  // This works because pthread_create() is not called by the C library
  // initialization routine that sets up the main thread's data structures.
  __isthreaded = 1;

  pthread_internal_t* thread = reinterpret_cast<pthread_internal_t*>(calloc(sizeof(*thread), 1));          使用calloc创建thread结构体，初始化为0
  if (thread == NULL) {
    __libc_format_log(ANDROID_LOG_WARN, "libc", "pthread_create failed: couldn't allocate thread");
    return EAGAIN;
  }
  thread->allocated_on_heap = true;    标记thread结构体是在heap上创建的，也提示了之后不使用的时候要用free去销毁

  if (attr == NULL) {
    pthread_attr_init(&thread->attr);  如果创建线程没有指定attr则使用default的attr
  } else {
    thread->attr = *attr;              如果有指定attr，则使用指定的attr
    attr = NULL; // Prevent misuse below.
  }

  // Make sure the stack size and guard size are multiples of PAGE_SIZE.
  thread->attr.stack_size = (thread->attr.stack_size + (PAGE_SIZE-1)) & ~(PAGE_SIZE-1);
  thread->attr.guard_size = (thread->attr.guard_size + (PAGE_SIZE-1)) & ~(PAGE_SIZE-1);

  if (thread->attr.stack_base == NULL) {
    // The caller didn't provide a stack, so allocate one.
    thread->attr.stack_base = __create_thread_stack(thread);  如果没有指定attr的stack_base，那么自己创建一个
    if (thread->attr.stack_base == NULL) {
      free(thread);
      return EAGAIN;
    }
  } else {
    // The caller did provide a stack, so remember we're not supposed to free it.
    thread->attr.flags |= PTHREAD_ATTR_FLAG_USER_STACK;  指定一个flag，之后由user自己释放这块地址
  }

  // Make room for the TLS area.
  // The child stack is the same address, just growing in the opposite direction.
  // At offsets >= 0, we have the TLS slots.
  // At offsets < 0, we have the child stack.
  void** tls = (void**)((uint8_t*)(thread->attr.stack_base) + thread->attr.stack_size - BIONIC_TLS_SLOTS * sizeof(void*));
  void* child_stack = tls;

  // Create a mutex for the thread in TLS_SLOT_SELF to wait on once it starts so we can keep
  // it from doing anything until after we notify the debugger about it
  //
  // This also provides the memory barrier we need to ensure that all
  // memory accesses previously performed by this thread are visible to
  // the new thread.
  pthread_mutex_t* start_mutex = (pthread_mutex_t*) &tls[TLS_SLOT_SELF];
  pthread_mutex_init(start_mutex, NULL);
  ScopedPthreadMutexLocker start_locker(start_mutex);

  tls[TLS_SLOT_THREAD_ID] = thread;        TLS数组的TLS_SLOT_THREAD_ID index保存thread结构体，后面其他线程函数可以通过从TLS拿到当前线程thread结构体

  int flags = CLONE_FILES | CLONE_FS | CLONE_VM | CLONE_SIGHAND | CLONE_THREAD | CLONE_SYSVSEM;

  int tid = __pthread_clone(start_routine, child_stack, flags, arg);  使用clone系统调用，指定stack的起始地址为TLS之后
  if (tid < 0) {
    int clone_errno = errno;
    if ((thread->attr.flags & PTHREAD_ATTR_FLAG_USER_STACK) == 0) {
      munmap(thread->attr.stack_base, thread->attr.stack_size);
    }
    free(thread);
    __libc_format_log(ANDROID_LOG_WARN, "libc", "pthread_create failed: clone failed: %s", strerror(errno));
    return clone_errno;
  }

  thread->tid = tid;

  int init_errno = _init_thread(thread, true);
  if (init_errno != 0) {
    // Mark the thread detached and let its __thread_entry run to
    // completion. (It'll just exit immediately, cleaning up its resources.)
    thread->internal_flags |= kPthreadInitFailed;
    thread->attr.flags |= PTHREAD_ATTR_FLAG_DETACHED;
    return init_errno;
  }

  // Notify any debuggers about the new thread.
  {
    ScopedPthreadMutexLocker debugger_locker(&gDebuggerNotificationLock);
    _thread_created_hook(thread->tid);
  }

  // Publish the pthread_t and let the thread run.
  *thread_out = (pthread_t) thread;

  return 0;
}

看一下pthread_attr_init函数指定的default值

#define DEFAULT_THREAD_STACK_SIZE ((1 * 1024 * 1024) - SIGSTKSZ)

int pthread_attr_init(pthread_attr_t* attr) {
  attr->flags = 0;
  attr->stack_base = NULL;
  attr->stack_size = DEFAULT_THREAD_STACK_SIZE;
  attr->guard_size = PAGE_SIZE;
  attr->sched_policy = SCHED_NORMAL;
  attr->sched_priority = 0;
  return 0;
}

再看一下__create_thread_stack函数实现

static void* __create_thread_stack(pthread_internal_t* thread) {
  ScopedPthreadMutexLocker lock(&gPthreadStackCreationLock);

  // Create a new private anonymous map.
  int prot = PROT_READ | PROT_WRITE;
  int flags = MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE;
  void* stack = mmap(NULL, thread->attr.stack_size, prot, flags, -1, 0);  使用mmap映射一块大小为stack_size的虚拟地址，这块内存是可读可写的，他是匿名映射，也就是不对应具体的文件，他是私有的，也就是每个进程/线程独有的
  if (stack == MAP_FAILED) {
    __libc_format_log(ANDROID_LOG_WARN,
                      "libc",
                      "pthread_create failed: couldn't allocate %zd-byte stack: %s",
                      thread->attr.stack_size, strerror(errno));
    return NULL;
  }

  // Set the guard region at the end of the stack to PROT_NONE.
  if (mprotect(stack, thread->attr.guard_size, PROT_NONE) == -1) {
    __libc_format_log(ANDROID_LOG_WARN, "libc",
                      "pthread_create failed: couldn't mprotect PROT_NONE %zd-byte stack guard region: %s",
                      thread->attr.guard_size, strerror(errno));
    munmap(stack, thread->attr.stack_size);
    return NULL;
  }

  return stack;
}

最后看一下pthread_exit函数

void pthread_exit(void * retval)
{
    pthread_internal_t*  thread     = __get_thread();  通过TLS的TLS_SLOT_THREAD_ID index获取当前线程的thread结构，大概思路是之前pthread_create mmap匿名创建那一块内存，栈顶之上一部分是TLS区域，所以可以很容易通过栈底的寄存器拿到TLS的首地址
    void*                stack_base = thread->attr.stack_base;
    int                  stack_size = thread->attr.stack_size;
    int                  user_stack = (thread->attr.flags & PTHREAD_ATTR_FLAG_USER_STACK) != 0;
    sigset_t mask;

    // call the cleanup handlers first
    while (thread->cleanup_stack) {       如果user有指定cleanup_stack的方法则先调用它
        __pthread_cleanup_t*  c = thread->cleanup_stack;
        thread->cleanup_stack   = c->__cleanup_prev;
        c->__cleanup_routine(c->__cleanup_arg);
    }

    // call the TLS destructors, it is important to do that before removing this
    // thread from the global list. this will ensure that if someone else deletes
    // a TLS key, the corresponding value will be set to NULL in this thread's TLS
    // space (see pthread_key_delete)
    pthread_key_clean_all();

    if (thread->alternate_signal_stack != NULL) {
      // Tell the kernel to stop using the alternate signal stack.
      stack_t ss;
      ss.ss_sp = NULL;
      ss.ss_flags = SS_DISABLE;
      sigaltstack(&ss, NULL);

      // Free it.
      munmap(thread->alternate_signal_stack, SIGSTKSZ);
      thread->alternate_signal_stack = NULL;
    }

    // if the thread is detached, destroy the pthread_internal_t
    // otherwise, keep it in memory and signal any joiners.
    pthread_mutex_lock(&gThreadListLock);
    if (thread->attr.flags & PTHREAD_ATTR_FLAG_DETACHED) {  如果是detached，会将本线程创建在heap上的thread结构体free
        _pthread_internal_remove_locked(thread);
    } else {
       /* make sure that the thread struct doesn't have stale pointers to a stack that
        * will be unmapped after the exit call below.
        */
        if (!user_stack) {
            thread->attr.stack_base = NULL;
            thread->attr.stack_size = 0;
            thread->tls = NULL;
        }

       /* Indicate that the thread has exited for joining threads. */
        thread->attr.flags |= PTHREAD_ATTR_FLAG_ZOMBIE;
        thread->return_value = retval;

       /* Signal the joining thread if present. */
        if (thread->attr.flags & PTHREAD_ATTR_FLAG_JOINED) {  如果是joined，会通过pthread_cond_signal唤醒监听当前thread的pthread_join函数
            pthread_cond_signal(&thread->join_cond);          会通过futex_wake系统调用唤醒睡眠的程序
        }
    }
    pthread_mutex_unlock(&gThreadListLock);

    sigfillset(&mask);
    sigdelset(&mask, SIGSEGV);
    (void)sigprocmask(SIG_SETMASK, &mask, (sigset_t *)NULL);

    // destroy the thread stack
    if (user_stack)           如果是user自己指定的stack_base，则user自己管理
        _exit_thread((int)retval);
    else
        _exit_with_stack_teardown(stack_base, stack_size, (int)retval);   如果是自己创建的stack，那么调用munmap系统调用取消这块内存的匿名映射
}

那么如何指定detached呢？

int pthread_detach(pthread_t t) {
  pthread_accessor thread(t);
  if (thread.get() == NULL) {
      return ESRCH;
  }

  if (thread->attr.flags & PTHREAD_ATTR_FLAG_DETACHED) {
    return EINVAL; // Already detached.
  }

  if (thread->attr.flags & PTHREAD_ATTR_FLAG_JOINED) {
    return 0; // Already being joined; silently do nothing, like glibc.
  }

  thread->attr.flags |= PTHREAD_ATTR_FLAG_DETACHED;    指定为detached
  return 0;
}

那么如何指定joined呢?

int pthread_join(pthread_t t, void** ret_val) {
  if (t == pthread_self()) {
    return EDEADLK;
  }

  pthread_accessor thread(t);
  if (thread.get() == NULL) {
      return ESRCH;
  }

  if (thread->attr.flags & PTHREAD_ATTR_FLAG_DETACHED) {
    return EINVAL;
  }

  if (thread->attr.flags & PTHREAD_ATTR_FLAG_JOINED) {
    return EINVAL;
  }

  // Signal our intention to join, and wait for the thread to exit.
  thread->attr.flags |= PTHREAD_ATTR_FLAG_JOINED;             指定为joined
  while ((thread->attr.flags & PTHREAD_ATTR_FLAG_ZOMBIE) == 0) {
    pthread_cond_wait(&thread->join_cond, &gThreadListLock);  会通过fetux_wait进入睡眠
  }
  if (ret_val) {
    *ret_val = thread->return_value;
  }

  _pthread_internal_remove_locked(thread.get());              销毁当前thread的thread结构体
  return 0;
}