Skip to content

network library based on io_uring and C++20 coroutine

License

Notifications You must be signed in to change notification settings

xuanyi-fu/xynet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

xynet

Experimental C++ network library based on io_uring and coroutine. This project is developed for understanding and practice C++20 new features(ranges, coroutine) and new Linux I/O API io_uring.

Features

  • Header only
  • Based on io_uring, the new Linux asynchronous I/O API
  • Based on C++20 coroutine
  • file_descriptor is designed in the CRTP Mixin pattern, all operations(recv, send, listen...) are modularized and thus optional.
  • buffer_sequence is designed to be compatible with C++20 std::span
  • like C++17 <filesystem>, support error handling in both std::error_code and exception
  • coroutine part is based on Lewis Baker's cppcoro, modified so that it could compile using gcc10.
  • Efficient:
    • Few dynamic allocation: most of the allocations are for the coroutine stack itself.
    • Few system calls compared with epoll based libraries: All asynchronous operations are submitted and reaped through io_uring by a single system call.

Requirements

  • Compiler: GCC10.0.1 (or later)
  • Linux Kernel: 5.2 (or later)
  • Liburing: 0.5(or later)
  • OpenSSL: for the SHA-1 in websocket(will be removed later)

Example

Usage

  • file_descriptor

    template <template <typename...> typename ... Modules>
    class file_descriptor<detail::module_list<Modules ...>>
      : public file_descriptor_base
      , public Modules<file_descriptor<detail::module_list<Modules ...>>>...
    {};

    file_descriptor is a handler that has unique ownership of file descriptor, like std::unique_ptr.

    • move-only,
    • RAII. ::close() is called in its destructor if the underlying file descriptor is valid.
    • Modularized: Modules are interfaces that are mixin'ed into the base class. Modules access file_descriptor by static polymorphism, i.e. cast this into the base class type.
    • default constructor will initialize the underlying file descriptor to be an invalid one.
    • Mixin'ed modules use set() and get() to work with the file_descriptor.
  • socket_t

    using socket_t = file_descriptor
    <
      detail::module_list
      <
        socket_init,
        address,
        operation_shutdown,
        operation_set_options,
        operation_bind,
        operation_listen,
        operation_accept,
        operation_connect,
        operation_send,
        operation_recv,
        operation_close
      >
    >;

    convenient type alias that has all modules for socket. You can define you own type alias. For example, an acceptor does not need to do send or recv.

    using acceptor_t = file_descriptor
    <
      detail::module_list
      <
        socket_init,
        operation_set_options,
        operation_bind,
        operation_listen,
        operation_accept
      >
    >;
  • buffer_sequence and const_buffer_sequence

    buffer_sequence takes a sequences of Containers which satisfy the concept std::ranges::contiguous_range then transform them into iovec.

    buffer_sequence could be constructed using:

    template<typename... Containers>
    buffer_sequence(Containers&&... containers)

    where Containers could be

    • std::span<std::byte, size or std::dynamic_extent>
    • std::span<Ts, size or std::dynamic_extent>
    • contiguous ranges that are writable

    or

    template<typename BufferRange>
    requires std::ranges::viewable_range<BufferRange&>
             && std::ranges::contiguous_range<std::ranges::range_value_t<BufferRange>>
    buffer_sequence(BufferRange& buffer_range)

    where BufferRange is a range of contiguous ranges. Notice that, different from the former one, iovecs are stored in a std::vector where dynamic allocation is inevitable.

  • Modules

    All modules support error handling by

    • std::error_code: by passing an lvalue reference of std::error_code (like C++17 <filesystem>), should there be an error, the lvalue reference passed will assign with a new error_code, otherwise, it will be clear().

    • exception. An std::system_error constructed by the reason encapsulated in a std::error_code will be throwed if there is an error.

    • socket_init

      • init(), init(std::erro_code& error) create a IPv4/TCP socket and set the file_descriptor with the file_descriptor returned form ::socket().
    • address

      the size of file_descriptor will increase if this module is used.

      Getters and setters are provided: set_local_address(const socket_address& address), set_peer_address(const socket_address& address), get_local_address(), get_peer_address(). Moudles like operation_accept, operation_connect , will set the address on success.

    synchronous operation modules

    • operation_shutdown

      shutdown(int how = SHUTWR), shutdown(int how, std::error_code& error), shutdown(std::error_code& error)

    • operation_bind

      bind the socket to a given address. If module address is used, set the local address on success. If bind with socket_address{0}, then a random port will be assigned and the local address will also be set on success.

      bind(const socket_address& address), bind(const socket_address& address, std::error_code& error)

    • operation_set_options

      call ::setsockopt.

      void setsockopt(int level, int optname, const void* optval, socklen_t optlen, std::error_code& error), void setsockopt(int level, int optname, const void* optval, socklen_t optlen), void reuse_address(std::error_code& error), void reuse_address()

    • operation_listen

      put the socket into listen state listen(int backlog = SOMAXCONN), listen(int backlog, std::error_code& error), listen(std::error_code& error).

    asynchronous operation modules

    all asynchronous operations provides an optional argument duration to impose a timeout.

    template<typename F2, typename... Args> [[nodiscard]]
    decltype(auto) accept(F2& peer_socket, Args&&... args) noexcept
    • peer_socket The socket into which the new connection will be accpeted.

      1. Args is void: The awaiter returned by the function will throw exception to report the error. A std::system_error constructed with the corresponding std::error_code will be throwed if the accept operation is failed after being co_await'ed.

      2. Args is a Duration, i.e. std::chrono::duration<Rep, Period>. This duration will be treated as the timeout for the operation. If the operation does not finish within the given duration, the operation will be canneled and an error_code (std::errc::operation_canceled) will be returned.

      3. Args is an lvalue reference of a std::error_code The awaiter returned by the function will use std::error_code to report the error. Should there be an error in the operation, the std::error_code passed by lvalue reference will be reset. Otherwise, it will be clear.

      4. Args are first a Duration, second an lvalue reference of a std::error_code The operation will have the features described in 2 and 3.

    • operation_connect

    template<typename... Args> [[nodiscard]]
    decltype(auto) connect(const socket_address& address, Args&&... args) noexcept

    address: the address with which the connection will be established. args: same as args desribed in operation_accept

    This operation will read from the socket until it reads a 0(eof). Then it will call close(2) on the socket.

    template<typename... Args>
    [[nodiscard]]
    decltype(auto) close(Args&&... args) noexcept 

    args: same as args desribed in operation_accept

    • operation_recv

      recv([std::error_code& error], [Duration&& duration], Args&&... args) where args will be forwarded to the constructor of buffer_sequence.

      • The buffers will be filled with the same order as the order they were in the lvalue reference of a viewable range or the order they were in args.
      • The operation will finish if there is an error(including the socket reads EOF) or all the buffers are filled. That is, it may use recvmsg(2) more than once.
      • After the operation is co_await'ed and then finishes, it will return the bytes transferred.

      recv_some([std::error_code& error], [Duration&& duration], Args&&... args) where args will be forwarded to the constructor of buffer_sequence.

      this awaiter will resume the coroutine after the first recv operation finished, regardless of whether the buffers are filled up or not.

    • operation_send

      recv([std::error_code& error], [Duration&& duration], Args&&... args) where args will be forwarded to the constructor of const_buffer_sequence

About

network library based on io_uring and C++20 coroutine

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published