epoll中et+多线程模式中很重要的EPOLL_ONESHOT实验

因为et模式需要循环读取，但是在读取过程中，如果有新的事件到达，很可能触发了其他线程来处理这个socket，那就乱了。

EPOLL_ONESHOT就是用来避免这种情况。注意在一个线程处理完一个socket的数据，也就是触发EAGAIN errno时候，就应该重置EPOLL_ONESHOT的flag，这时候，新到的事件，就可以重新进入触发流程了。

注：EPOLL_ONESHOT的原理其实是，每次触发事件之后，就将事件注册从fd上清除了，也就不会再被追踪到；下次需要用epoll_ctl的EPOLL_CTL_MOD来手动加上才行。

服务器代码如下：

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <fcntl.h>
#include <assert.h>

#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <sys/epoll.h>
#include <pthread.h>

#define MAX_EVENT_NUMBER 1024
#define BUFFER_SIZE 1024
struct fds {
        int epollfd;
        int sockfd;
};

int setnonblocking(int fd) {
        int old_option = fcntl(fd, F_GETFL);
        int new_option = old_option | O_NONBLOCK;
        fcntl(fd, F_SETFL, new_option);
        return old_option;
}

void addfd(int epollfd, int fd, bool oneshot) {
        epoll_event event;
        event.data.fd = fd;
        event.events = EPOLLIN | EPOLLET;
        if (oneshot) {
                event.events |= EPOLLONESHOT;
        }
        epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &event);
        setnonblocking(fd);
}

void reset_oneshot(int epollfd, int fd) {
        epoll_event event;
        event.data.fd = fd;
        event.events = EPOLLIN | EPOLLET | EPOLLONESHOT;
        epoll_ctl(epollfd, EPOLL_CTL_MOD, fd, &event);
}

void* worker(void *arg) {
        int sockfd = ((fds*)arg)->sockfd;
        int epollfd = ((fds*)arg)->epollfd;
        pthread_t pid = pthread_self();

        printf("start new thread %u to recv data on fd: %d\n", pid, sockfd);
        char buf[BUFFER_SIZE];
        memset(buf, '\0', BUFFER_SIZE);

        while() {
                int ret = recv(sockfd, buf, BUFFER_SIZE-, );
                if (ret == ) {
                        close(sockfd);
                        printf("foreiner closed the connection\n");
                        break;
                }
                else if (ret < ) {
                        if (errno == EAGAIN) {
                                reset_oneshot(epollfd, sockfd);
                                printf("EAGAIN read later\n");
                                break;
                        }
                }
                else {
                        buf[ret] = '\0';
                        printf("thread %u get content: %s\n", pid, buf);
                        printf("thread %u about to sleep\n", pid);
                        sleep();
                        printf("thread %u back from sleep\n", pid);
                }
        }
        //printf("end thread %u receiving data on fd: %d\n", pid, sockfd);

}

int main(int argc, char *argv[]) {
        if (argc <= ) {
                printf("usage: %s port_number [ip_address]\n", basename(argv[]));
                return ;
        }
        int port = atoi(argv[]);

        int ret = ;
        sockaddr_in address;
        bzero(&address, sizeof(address));
        address.sin_family = AF_INET;
        if (argc >= ) {
                const char *ip =argv[];
                inet_pton(AF_INET, ip, &address.sin_addr);
        }
        else {
                address.sin_addr.s_addr = INADDR_ANY;
        }
        address.sin_port = htons(port);
        int listenfd = socket(PF_INET, SOCK_STREAM, );
        assert(listenfd >= );

        ret = bind(listenfd, (sockaddr*)&address, sizeof(address));
        assert(ret != -);

        ret = listen(listenfd, );
        assert(ret != -);

        epoll_event events[MAX_EVENT_NUMBER];
        int epollfd = epoll_create();
        assert(epollfd != -);

        addfd(epollfd, listenfd, false);

        while() {
                int ret = epoll_wait(epollfd, events, MAX_EVENT_NUMBER, -);
                if (ret < ) {
                        printf("epoll failure\n");
                        break;
                }

                for (int i=; i<ret; i++) {
                        int sockfd = events[i].data.fd;
                        if (sockfd == listenfd) {
                                sockaddr_in client_address;
                                socklen_t client_addrlength = sizeof(client_address);
                                int connfd = accept(listenfd, (sockaddr*)&client_address,
                                                        &client_addrlength);

                                addfd(epollfd, connfd, true);
                                printf("new connection is added to epollfd\n");
                        }
                        else if (events[i].events & EPOLLIN) {
                                pthread_t thread;
                                fds fds_for_new_worker;
                                fds_for_new_worker.epollfd = epollfd;
                                fds_for_new_worker.sockfd = sockfd;
                                // new thread
                                pthread_create(&thread, NULL, worker,
                                                (void*)&fds_for_new_worker);
                        }
                        else {
                                printf("something else happened\n");
                        }
                }
        }

        close(listenfd);
        return ;

}

以下是使用telnet客户端发送的文本，匀速敲入代码：

$telnet 127.0.0.1
Trying 127.0.0.1...
Connected to 127.0.0.1.
Escape character is '^]'.
hi1
hi2
hi3
hi4
hi5
hi6
hi7
hi8
Connection closed by foreign host.

以下是服务器的运行和输出：

$./epoll_oneshot
new connection is added to epollfd
start new thread  to recv data on fd:
thread  get content: hi1

thread  about to sleep
thread  back from sleep
thread  get content: hi2
hi3

thread  about to sleep
thread  back from sleep
thread  get content: hi4
hi5
hi6

thread  about to sleep
thread  back from sleep
thread  get content: hi7

thread  about to sleep
thread  back from sleep
EAGAIN read later
start new thread  to recv data on fd:
thread  get content: hi8

thread  about to sleep
thread  back from sleep
EAGAIN read later
^C

最后用Ctrl+C来结束服务器。

可以看出，在hi7文本和hi8文本之间，服务器收到了EAGAIN，表示读取告一段落。而之后的线程id也换成了新线程id。在hi7之前，因为每次服务器sleep结束之后，都还有没有读完的数据，所以线程id始终没有变，始终是同一个线程处理数据。

另外，要注意的是，EPOLL_ONESHOT既可以在et下也可以在lt下设置。效果是一样的，都是同一个fd上面的相同事件只会触发一次。上面是et的例子，对于lt，如果设置了EPOLL_ONESHOT，也是需要把数据读完，然后重置event。而不能像原始lt编程方式那样依赖于事件通知来读取数据了。