yabridge/src/wine-host/bridges/group.h

// yabridge: a Wine VST bridge
// Copyright (C) 2020-2021 Robbert van der Helm
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program.  If not, see <https://www.gnu.org/licenses/>.

#pragma once

#include <atomic>
#include <thread>

#include "../boost-fix.h"

#include <boost/asio/local/stream_protocol.hpp>

#include "../common/logging/common.h"
#include "../utils.h"
#include "common.h"

/**
 * Encapsulate capturing the STDOUT or STDERR stream by opening a pipe and
 * reopening the passed file descriptor as one of the ends of the newly opened
 * pipe. This allows all output sent to be read from that pipe. This is needed
 * to capture all (debug) output from Wine and the hosted plugins so we can
 * prefix it with a timestamp and a group identifier and potentially write it to
 * a log file. Since the host application is run independently of the yabridge
 * instance that spawned it, this can't simply be done by the caller like we're
 * doing for Wine output in individually hosted plugins.
 */
class StdIoCapture {
   public:
    /**
     * Redirect all output sent to a file descriptor (e.g. `STDOUT_FILENO` or
     * `STDERR_FILENO`) to a pipe. `StdIoCapture::pipe` can be used to read from
     * this pipe.
     *
     * @param io_context The IO context to create the captured pipe stream on.
     * @param file_descriptor The file descriptor to remap.
     *
     * @throw std::system_error If the pipe could not be created.
     */
    StdIoCapture(boost::asio::io_context& io_context, int file_descriptor);

    StdIoCapture(const StdIoCapture&) = delete;
    StdIoCapture& operator=(const StdIoCapture&) = delete;

    /**
     * On cleanup, close the outgoing file descriptor from the pipe and restore
     * the original file descriptor for the captured stream.
     */
    ~StdIoCapture();

    /**
     * The pipe endpoint where all output from the original file descriptor gets
     * redirected to. This can be read from like any other `Boost.Asio` stream.
     */
    boost::asio::posix::stream_descriptor pipe;

   private:
    /**
     * The file descriptor of the stream we're capturing.
     */
    const int target_fd;

    /**
     * A copy of the original file descriptor. Will be used to undo
     * the capture when this object gets destroyed.
     */
    const int original_fd_copy;

    /**
     * The two file descriptors generated by the `pipe()` function call.
     * `pipe_fd[1]` is used to reopen/capture the passed file descriptor, and
     * `pipe_fd[0]` can be used to read the captured output from.
     */
    int pipe_fd[2];
};

/**
 * A 'plugin group' that listens on a _group socket_ for plugins to host in this
 * process. Once the plugin gets loaded into a new thread the actual bridging
 * process is identical to individually hosted plugins.
 *
 * An important detail worth mentioning here is that while this plugin group can
 * throw in the constructor when another process is already listening on the
 * socket, this should not be treated as an error. When using plugins groups,
 * yabridge will try to connect to the group socket on initialization and it
 * will launch a new group host process if it can't. If this is done for
 * multiple yabridge instances at the same time, then multiple group host
 * processes will be launched. Instead of using complicated inter-process
 * synchronization, we'll simply allow the processes to fail when another
 * process is already listening on the socket.
 */
class GroupBridge {
   public:
    /**
     * Create a plugin group by listening on the provided socket for incoming
     * plugin host requests.
     *
     * @param gruop_socket_path The path to the group socket endpoint. This path
     *   should be in the form of
     *   `/tmp/yabridge-group-<group_name>-<wine_prefix_id>-<architecture>.sock`
     *   where `<wine_prefix_id>` is a numerical hash as explained in the
     *   `create_logger_prefix()` function in `./group.cpp`.
     *
     * @throw boost::system::system_error If we can't listen on the socket.
     * @throw std::system_error If the pipe could not be created.
     *
     * @note Creating an `GroupBridge` instance has the side effect that the
     *   STDOUT and STDERR streams of the current process will be redirected to
     *   a pipe so they can be properly written to a log file.
     */
    explicit GroupBridge(boost::filesystem::path group_socket_path);

    ~GroupBridge();

    GroupBridge(const GroupBridge&) = delete;
    GroupBridge& operator=(const GroupBridge&) = delete;

    /**
     * If this returns `true`, then the group host's event loop should
     * temporarily be disabled. This simply calls
     * `HostBridge::inhibits_event_loop()` for all plugins hosted in this group
     * process.
     */
    bool is_event_loop_inhibited();

    /**
     * Run a plugin's dispatcher and message loop, processing all events on the
     * main IO context. The plugin will have already been created in
     * `accept_requests` since it has to be initiated inside of the IO context's
     * thread.
     *
     * Once the plugin has exited, this thread will then be joined to the main
     * thread and removed from the `active_plugins` from the main IO context. If
     * this causes the vector to become empty, we will terminate this process.
     * This check is delayed by a few seconds to prevent having to constantly
     * restart the group process during plugin scanning.
     *
     * @param plugin_id The ID of this plugin in the `active_plugins` map. Used
     *   to unload the plugin and join this thread again after the plugin exits.
     *
     * @note In the case that the process starts but no plugin gets initiated,
     *   then the process will never exit on its own. This should not happen
     *   though.
     */
    void handle_plugin_run(size_t plugin_id, HostBridge* bridge);

    /**
     * Listen for new requests to spawn plugins within this process and handle
     * them accordingly. Will terminate once all plugins have exited.
     */
    void handle_incoming_connections();

   private:
    /**
     * Listen on the group socket for incoming requests to host a new plugin
     * within this group process. This will read a `GoupRequest` object
     * containing information about the plugin, reply with this process's PID so
     * the yabridge instance can tell if the plugin crashed during
     * initialization, and it will then try to initialize the plugin. After
     * intialization the plugin handling will be handed over to a new thread
     * running `handle_plugin_run()`. Because of the way the Win32 API works,
     * all plugins have to be initialized from the same thread, and all event
     * handling and message loop interaction also has to be done from that
     * thread, which is why we initialize the plugin here and use the
     * `handle_dispatch()` function to run events within the same
     * `main_context`.
     *
     * @see handle_plugin_run
     */
    void accept_requests();

    /**
     * Handle both Win32 messages and X11 events on a timer within the IO
     * context for all plugins.
     */
    void async_handle_events();

    /**
     * The logging facility used for this group host process. Since we can't
     * identify which plugin is generating (debug) output, every line will only
     * be prefixed with the name of the group.
     */
    Logger logger;

    /**
     * The IO context that connections will be accepted on, and that any plugin
     * operations that may involve the Win32 mesasge loop (e.g. initialization
     * and most `AEffect::dispatcher()` calls) should be run on.
     */
    MainContext main_context;
    /**
     * A seperate IO context that handles the STDIO redirect through
     * `StdIoCapture`. This is separated from the `main_context` above so that
     * STDIO capture does not get blocked by GUI operations. Since every GUI
     * related operation should be run from the same thread, we can't just add
     * another thread to the main IO context.
     */
    boost::asio::io_context stdio_context;

    boost::asio::streambuf stdout_buffer;
    boost::asio::streambuf stderr_buffer;
    /**
     * Contains a pipe used for capturing this process's STDOUT stream. Needed
     * to be able to process the output generated by Wine and plugins and to be
     * able write it write it to an external log file.
     */
    StdIoCapture stdout_redirect;
    /**
     * Contains a pipe used for capturing this process's STDERR stream. Needed
     * to be able to process the output generated by Wine and plugins and to be
     * able write it write it to an external log file.
     */
    StdIoCapture stderr_redirect;
    /**
     * A thread that runs the `stdio_context` loop.
     */
    Win32Thread stdio_handler;

    boost::asio::local::stream_protocol::endpoint group_socket_endpoint;
    /**
     * The UNIX domain socket acceptor that will be used to listen for incoming
     * connections to spawn new plugins within this process.
     */
    boost::asio::local::stream_protocol::acceptor group_socket_acceptor;

    /**
     * A map of threads that are currently hosting a plugin within this process
     * along with their plugin instance. After a plugin has exited or its
     * initialization has failed, the thread handling it will remove itself from
     * this map. This is to keep track of the amount of plugins currently
     * running with their associated thread handles. The key that identifies the
     * thread and plugin is a unique plugin ID obtained by doing a fetch-and-add
     * on `next_plugin_id`.
     */
    std::unordered_map<size_t,
                       std::pair<Win32Thread, std::unique_ptr<HostBridge>>>
        active_plugins;
    /**
     * A counter for the next unique plugin ID. When hosting a new plugin we'll
     * do a fetch-and-add to ensure that every thread gets its own unique
     * identifier.
     */
    std::atomic_size_t next_plugin_id;
    /**
     * A mutex to prevent two threads from simultaneously accessing the plugins
     * map, and also to prevent `handle_plugin_run()` from terminating the
     * process because it thinks there are no active plugins left just as a new
     * plugin is being spawned.
     */
    std::mutex active_plugins_mutex;

    /**
     * A timer to defer shutting down the process, allowing for fast plugin
     * scanning without having to start a new group host process for each
     * plugin.
     *
     * @see handle_plugin_run
     */
    boost::asio::steady_timer shutdown_timer;
    /**
     * A mutex to prevent two threads from simultaneously modifying the shutdown
     * timer when multiple plugins exit at the same time.
     */
    std::mutex shutdown_timer_mutex;
};