Boost Asio异步发送数据（async_write）崩溃问题记录

背景

---

服务端与客户端之间的网络通信（使用Boost Asio库异步编程模式实现），客户端会向服务端请求数据。

在刚开始的测试中，是没有出现问题的。后来有一次测试时，服务端查询完数据后，向客户端发送时总是崩溃。

通过gdb调试，可以发现是在调用到异步发送函数（boost::asio::async_write）后崩溃的。

打印的栈信息如下：

Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x000000000056625c in boost::asio::detail::reactive_socket_service_base::start_op (this=0x2e33313a30313a5f, impl=..., op_type=1, op=0x7fc5b4000af0, is_continuation=false, 
    is_non_blocking=true, noop=false) at /usr/local/include/boost/asio/detail/impl/reactive_socket_service_base.ipp:219
219	  reactor_.post_immediate_completion(op, is_continuation);
[Current thread is 1 (Thread 0x7fc5f7fff700 (LWP 128087))]
(gdb) bt
#0  0x000000000056625c in boost::asio::detail::reactive_socket_service_base::start_op (this=0x2e33313a30313a5f, impl=..., op_type=1, op=0x7fc5b4000af0, is_continuation=false, 
    is_non_blocking=true, noop=false) at /usr/local/include/boost/asio/detail/impl/reactive_socket_service_base.ipp:219
#1  0x000000000056bed5 in boost::asio::detail::reactive_socket_service_base::async_send<boost::asio::mutable_buffers_1, boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> > > (this=0x2e33313a30313a5f, impl=..., buffers=..., flags=0, handler=...)
    at /usr/local/include/boost/asio/detail/reactive_socket_service_base.hpp:216
#2  0x000000000056b20f in boost::asio::stream_socket_service<boost::asio::ip::tcp>::async_send<boost::asio::mutable_buffers_1, boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> > >(boost::asio::detail::reactive_socket_service<boost::asio::ip::tcp>::implementation_type&, boost::asio::mutable_buffers_1 const&, int, boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> >&&) (this=0x2e33313a30313a37, impl=..., buffers=..., flags=0, handler=<unknown type in /a.out, CU 0x5ed5a4, DIE 0x6317d1>)
    at /usr/local/include/boost/asio/stream_socket_service.hpp:334
#3  0x000000000056a708 in boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >::async_write_some<boost::asio::mutable_buffers_1, boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> > >(boost::asio::mutable_buffers_1 const&, boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> >&&) (
    this=0x2473c38, buffers=..., handler=<unknown type in /a.out, CU 0x5ed5a4, DIE 0x62fcbd>) at /usr/local/include/boost/asio/basic_stream_socket.hpp:732
#4  0x0000000000569831 in boost::asio::detail::write_op<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::transfer_all_t, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> >::operator() (this=0x7fc5f7ffde20, ec=..., bytes_transferred=0, start=1) at /usr/local/include/boost/asio/impl/write.hpp:258
#5  0x0000000000568263 in boost::asio::async_write<boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >, boost::asio::mutable_buffers_1, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running> >(boost::asio::basic_stream_socket<boost::asio::ip::tcp, boost::asio::stream_socket_service<boost::asio::ip::tcp> >&, boost::asio::mutable_buffers_1 const&, boost::asio::detail::wrapped_handler<boost::asio::io_service::strand, boost::_bi::bind_t<void, boost::_mfi::mf2<void, TcpSession, boost::system::error_code const&, unsigned long>, boost::_bi::list3<boost::_bi::value<boost::shared_ptr<TcpSession> >, boost::arg<1>, boost::arg<2> > >, boost::asio::detail::is_continuation_if_running>&&) (s=..., buffers=..., handler=<unknown type in /a.out, CU 0x5ed5a4, DIE 0x62aa66>)
    at /usr/local/include/boost/asio/impl/write.hpp:621
#6  0x000000000056306b in TcpSession::StartSend (this=0x246ec20, pData=0x7fc5a000f158 "\314F", nDataSize=18128) at SourceFiles/Common/ClientMgrBase.cpp:259

分析

---

1. 数据缓冲区有效性

在Boost Asio异步编程中，最常见的错误莫过于数据缓冲区的失效。这个在编程初期已经进行了严格的检查。

在服务端的实现中，使用了TcpSession类来管理socket和它相关的read/write缓冲，它继承自boost::enable_shared_from_this<TcpSession>，并在bind仿函数中使用智能指针，所以这块应该是没有问题的。

2. socket类非线程安全

另外一个，是socket的非线程安全性。对于同一个socket，不能在一个线程读的同时，在另一个线程中执行写操作。

其实，在数据量比较小时（比如小于65535），多线程操作socket一般也不会有什么问题。

3. 结合测试情况和debug

结合测试具体情况，在先前的测试中，并未出现问题，测试条件的改变在于读取的数据量变大。

而在boost::asio::async_write实现中，使用定长的数组作为发送数据缓冲区。在发送函数到来的时候，先把数据拷贝到这个缓冲区，再调用发送函数。

一个现实的情况是，当数据量太大时，会超过发送数据缓冲区，这样就会导致内存使用异常。

虽然在拷贝的时候没有崩溃，但在真正发送数据时，发现buff违例，导致崩溃。接收数据时也会存在类似情况。

所以最终定准为数据缓冲区的使用问题，最简单的方法是修改为string，这样就由string类的asign方法来承载待发送的数据，不会出现超出范围的问题（内存允许）。

小结

---

在实际测试中，当我修改客户端的recv函数的数据缓冲区小到不能容纳所接收的数据时，也会出现崩溃的现象，但打印的栈信息与上述不同。

可见，在数据缓冲区的处理方面，我的程序还有不少需要改进的地方。由于不能预知数据的最大大小，所以应该考虑动态分配内存的方式，比如std::vector类型。

总之，当Boost Asio库由于异常崩溃时，不要因为看不懂栈信息而着急上火，还是要结合相关信息和已有知识储备，由浅入深，逐步排查定位。

参考资料

---

boost::asio::socket thread safety