memcached探索之thread model(2)

6. 下面我们看看主线程dispatch_thread的事件处理设置。在memcached.c中的main函数中

/* create unix mode sockets after dropping privileges */
    if (settings.socketpath != NULL) {
        errno = 0;
        if (server_socket_unix(settings.socketpath,settings.access)) {
            vperror("failed to listen on UNIX socket: %s", settings.socketpath);
            exit(EX_OSERR);
        }
    }

    /* create the listening socket, bind it, and init */
    if (settings.socketpath == NULL) {
        int udp_port;

        const char *portnumber_filename = getenv("MEMCACHED_PORT_FILENAME");
        char temp_portnumber_filename[PATH_MAX];
        FILE *portnumber_file = NULL;

        if (portnumber_filename != NULL) {
            snprintf(temp_portnumber_filename,
                     sizeof(temp_portnumber_filename),
                     "%s.lck", portnumber_filename);

            portnumber_file = fopen(temp_portnumber_filename, "a");
            if (portnumber_file == NULL) {
                fprintf(stderr, "Failed to open \"%s\": %s\n",
                        temp_portnumber_filename, strerror(errno));
            }
        }

        errno = 0;
        if (settings.port && server_socket(settings.port, tcp_transport,
                                           portnumber_file)) {
            vperror("failed to listen on TCP port %d", settings.port);
            exit(EX_OSERR);
        }

        /*
         * initialization order: first create the listening sockets
         * (may need root on low ports), then drop root if needed,
         * then daemonise if needed, then init libevent (in some cases
         * descriptors created by libevent wouldn't survive forking).
         */
        udp_port = settings.udpport ? settings.udpport : settings.port;

        /* create the UDP listening socket and bind it */
        errno = 0;
        if (settings.udpport && server_socket(settings.udpport, udp_transport,
                                              portnumber_file)) {
            vperror("failed to listen on UDP port %d", settings.udpport);
            exit(EX_OSERR);
        }

        if (portnumber_file) {
            fclose(portnumber_file);
            rename(temp_portnumber_filename, portnumber_filename);
        }
    }
    /* Drop privileges no longer needed */
    drop_privileges();
    /* enter the event loop */
    event_base_loop(main_base, 0); //主线程(dispatcher_thread)的事件监听循环。。。




7. 继续跟踪server_socket函数（memcached.c中）

/**
 * Create a socket and bind it to a specific port number
 * @param port the port number to bind to
 * @param transport the transport protocol (TCP / UDP)
 * @param portnumber_file A filepointer to write the port numbers to
 * when they are successfully added to the list of ports we
 * listen on.
 */
static int server_socket(int port, enum network_transport transport,
                         FILE *portnumber_file) {
    int sfd;
    struct linger ling = {0, 0};
    struct addrinfo *ai;
    struct addrinfo *next;
    struct addrinfo hints = { .ai_flags = AI_PASSIVE,
                              .ai_family = AF_UNSPEC };
    char port_buf[NI_MAXSERV];
    int error;
    int success = 0;
    int flags =1;

    hints.ai_socktype = IS_UDP(transport) ? SOCK_DGRAM : SOCK_STREAM;

    if (port == -1) {
        port = 0;
    }
    snprintf(port_buf, sizeof(port_buf), "%d", port);
    error= getaddrinfo(settings.inter, port_buf, &hints, &ai);
    if (error != 0) {
        if (error != EAI_SYSTEM)
          fprintf(stderr, "getaddrinfo(): %s\n", gai_strerror(error));
        else
          perror("getaddrinfo()");
        return 1;
    }

    for (next= ai; next; next= next->ai_next) {
        conn *listen_conn_add;
        if ((sfd = new_socket(next)) == -1) { //creat socket
            /* getaddrinfo can return "junk" addresses,
             * we make sure at least one works before erroring.
             */
            continue;
        }

#ifdef IPV6_V6ONLY
        if (next->ai_family == AF_INET6) {
            error = setsockopt(sfd, IPPROTO_IPV6, IPV6_V6ONLY, (char *) &flags, sizeof(flags));
            if (error != 0) {
                perror("setsockopt");
                close(sfd);
                continue;
            }
        }
#endif
        //setsockopt
        setsockopt(sfd, SOL_SOCKET, SO_REUSEADDR, (void *)&flags, sizeof(flags));
        if (IS_UDP(transport)) {
            maximize_sndbuf(sfd);
        } else {
            error = setsockopt(sfd, SOL_SOCKET, SO_KEEPALIVE, (void *)&flags, sizeof(flags));
            if (error != 0)
                perror("setsockopt");

            error = setsockopt(sfd, SOL_SOCKET, SO_LINGER, (void *)&ling, sizeof(ling));
            if (error != 0)
                perror("setsockopt");

            error = setsockopt(sfd, IPPROTO_TCP, TCP_NODELAY, (void *)&flags, sizeof(flags));
            if (error != 0)
                perror("setsockopt");
        }
        //bind
        if (bind(sfd, next->ai_addr, next->ai_addrlen) == -1) {
            if (errno != EADDRINUSE) {
                perror("bind()");
                close(sfd);
                freeaddrinfo(ai);
                return 1;
            }
            close(sfd);
            continue;
        } else {
            success++;

            //TCP listen
            if (!IS_UDP(transport) && listen(sfd, settings.backlog) == -1) {
                perror("listen()");
                close(sfd);
                freeaddrinfo(ai);
                return 1;
            }
            if (portnumber_file != NULL &&
                (next->ai_addr->sa_family == AF_INET ||
                 next->ai_addr->sa_family == AF_INET6)) {
                union {
                    struct sockaddr_in in;
                    struct sockaddr_in6 in6;
                } my_sockaddr;
                socklen_t len = sizeof(my_sockaddr);
                if (getsockname(sfd, (struct sockaddr*)&my_sockaddr, &len)==0) {
                    if (next->ai_addr->sa_family == AF_INET) {
                        fprintf(portnumber_file, "%s INET: %u\n",
                                IS_UDP(transport) ? "UDP" : "TCP",
                                ntohs(my_sockaddr.in.sin_port));
                    } else {
                        fprintf(portnumber_file, "%s INET6: %u\n",
                                IS_UDP(transport) ? "UDP" : "TCP",
                                ntohs(my_sockaddr.in6.sin6_port));
                    }
                }
            }
        }

        if (IS_UDP(transport)) {


            //UDP 的处理中不需要accept,所以直接派发connection到工作线程。
            int c;

            for (c = 0; c < settings.num_threads; c++) {
                /* this is guaranteed to hit all threads because we round-robin */
                dispatch_conn_new(sfd, conn_read, EV_READ | EV_PERSIST,
                                  UDP_READ_BUFFER_SIZE, transport);
            }
        } else { 


           //TCP的处理(注意，这里dispatcher_thread同样调用了conn_new来绑定conn_event到其main_base. 并且此时conn的初始状态为conn_listening, 事件为持久可读， 而在conn_new中注册了conn_event的回调函数为event_handler,所以，dispatche_thread在当前listen的socket可读时就会调用event_handler,进而调用driver_machine(c) 进入状态机。而在driver_machine中如果是主线程(dispatcher_thread)则会在accept socket后调用dispatch_new_conn函数来给各worker_thread派发connection...)
            if (!(listen_conn_add = conn_new(sfd, conn_listening,
                                             EV_READ | EV_PERSIST, 1,
                                             transport, main_base))) {
                fprintf(stderr, "failed to create listening connection\n");
                exit(EXIT_FAILURE);
            }
            listen_conn_add->next = listen_conn;
            listen_conn = listen_conn_add;
        }
    }

    freeaddrinfo(ai);

    /* Return zero iff we detected no errors in starting up connections */
    return success == 0;
}

8. 看看UDP和TCP模式下dispatcher_thread都会调用的dispatch_new_conn函数(在thread.c中)

/*
 * Dispatches a new connection to another thread. This is only ever called
 * from the main thread, either during initialization (for UDP) or because
 * of an incoming connection.
 */
void dispatch_conn_new(int sfd, enum conn_states init_state, int event_flags,
                       int read_buffer_size, enum network_transport transport) {
    CQ_ITEM *item = cqi_new();
    int tid = (last_thread + 1) % settings.num_threads; //轮询的方式找worker_thread

    LIBEVENT_THREAD *thread = threads + tid;

    last_thread = tid;

    item->sfd = sfd;
    item->init_state = init_state;
    item->event_flags = event_flags;
    item->read_buffer_size = read_buffer_size;
    item->transport = transport;
    //push conn到worker_thread的CQ中
    cq_push(thread->new_conn_queue, item);

    MEMCACHED_CONN_DISPATCH(sfd, thread->thread_id);
    if (write(thread->notify_send_fd, "", 1) != 1) {
        perror("Writing to thread notify pipe");
    }
}


9. 先看看unix domain socket模式下主线程的事件处理设置。（在上面的6中调用的 server_socket_unix函数）

static int server_socket_unix(const char *path, int access_mask) {
    int sfd;
    struct linger ling = {0, 0};
    struct sockaddr_un addr;
    struct stat tstat;
    int flags =1;
    int old_umask;

    if (!path) {
        return 1;
    }

    if ((sfd = new_socket_unix()) == -1) {
        return 1;
    }

    /*
     * Clean up a previous socket file if we left it around
     */
    if (lstat(path, &tstat) == 0) {
        if (S_ISSOCK(tstat.st_mode))
            unlink(path);
    }

    setsockopt(sfd, SOL_SOCKET, SO_REUSEADDR, (void *)&flags, sizeof(flags));
    setsockopt(sfd, SOL_SOCKET, SO_KEEPALIVE, (void *)&flags, sizeof(flags));
    setsockopt(sfd, SOL_SOCKET, SO_LINGER, (void *)&ling, sizeof(ling));

    /*
     * the memset call clears nonstandard fields in some impementations
     * that otherwise mess things up.
     */
    memset(&addr, 0, sizeof(addr));

    addr.sun_family = AF_UNIX;
    strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
    assert(strcmp(addr.sun_path, path) == 0);
    old_umask = umask( ~(access_mask&0777));
    if (bind(sfd, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
        perror("bind()");
        close(sfd);
        umask(old_umask);
        return 1;
    }
    umask(old_umask);
    if (listen(sfd, settings.backlog) == -1) {
        perror("listen()");
        close(sfd);
        return 1;
    }
    if (!(listen_conn = conn_new(sfd, conn_listening, //同样是调用conn_new
                                 EV_READ | EV_PERSIST, 1,
                                 local_transport, main_base))) {
        fprintf(stderr, "failed to create listening connection\n");
        exit(EXIT_FAILURE);
    }

    return 0;
}