ET模式下，需要循环从缓存中读取，直到返回EAGAIN没有数据可读后，一个被通知的事件才算结束。如果还读取过程中，同一个连接又有新的事件到来，触发其他线程处理同一个socket，就乱了。EPOLL_ONESHOT就是用来避免这种情况发生的。将事件设置为EPOLL_ONESHOT后，每次事件触发后就会将这个FD从rbtree中删除，就不会再监听该事件了。
下面是我模仿别人写的一段代码：
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <fcntl.h>
#include <libgen.h>
#include <assert.h>
#include <stdbool.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 #define ONE_SHOT true
#define NOT_ONE_SHOT false struct data {
int epollfd;
int sockfd;
}; void setnonblockling(int fd)
{
int fd_option = fcntl(fd, F_GETFL);
fd_option |= O_NONBLOCK;
fcntl(fd, F_SETFL, fd_option);
} void add_fd(int epollfd, int fd, bool flag)
{
struct epoll_event event;
event.data.fd = fd;
event.events = EPOLLIN | EPOLLET; //监听可读事件,ET模式
if (flag)
event.events |= EPOLLONESHOT; epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &event);
setnonblockling(fd);
} void reset_oneshot(int epollfd, int fd)
{
struct epoll_event event;
event.data.fd = fd;
event.events = EPOLLIN | EPOLLET | EPOLLONESHOT;
epoll_ctl(epollfd, EPOLL_CTL_MOD, fd, &event);
} void *threadMain(void *instance)
{
struct data *data = (struct data *)instance;
int epollfd = data->epollfd;
int sockfd = data->sockfd;
pthread_t pid = pthread_self(); //pid or tid, confused printf("start new thread %u to recv data on fd: %d\n", pid, sockfd);
char buf[BUFFER_SIZE];
memset(buf, , sizeof(buf)); for(;;) {
int n = recv(sockfd, buf, BUFFER_SIZE-, );
if (n == ) {
close(sockfd);
printf("foreiner closed the connection\n");
break;
}
else if (n < ) {
if (errno == EAGAIN) {
reset_oneshot(epollfd, sockfd);
printf("EAGAIN. Read later\n");
break;
}
}
else {
buf[n] = '\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 ret;
int listenfd;
int port = atoi(argv[]);
struct 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);
printf("port = %d\n", port);
listenfd = socket(PF_INET, SOCK_STREAM, );
assert(listenfd >= ); ret = bind(listenfd, (struct sockaddr *)&address, sizeof(address));
assert(ret != -); ret = listen(listenfd, );
assert(ret != -); struct epoll_event events[MAX_EVENT_NUMBER];
int epollfd = epoll_create();
assert(epollfd != -); /* 需要反复监听listenfd */
add_fd(epollfd, listenfd, NOT_ONE_SHOT); while() {
int fd_number = epoll_wait(epollfd, events, MAX_EVENT_NUMBER, -); //设置为永久阻塞
if (fd_number < ) {
perror("epoll wait failed\n");
break;
} for (int i = ; i < fd_number; i++) {
int sockfd = events[i].data.fd;
/* listenfd上有新事件，则创建新的fd，并加入监听队列 */
if (sockfd == listenfd) {
struct sockaddr_in client_addr;
socklen_t client_socklen = sizeof(client_addr);
int connfd = accept(sockfd, (struct sockaddr*)&client_addr, &client_socklen);
add_fd(epollfd, connfd, ONE_SHOT);
}
else if (events[i].events & EPOLLIN) {
pthread_t tid;
struct data instance;
instance.epollfd = epollfd;
instance.sockfd = sockfd;
/* 一个连接对应一个线程，最好写成threadpool-BlockingQueue模式 */
pthread_create(&tid, NULL, threadMain, (void *)&instance);
}
else {
printf("something else happened\n");
}
}
} close(listenfd);
return ;
}