一、 突破 1024 文件描述符限制
二、epoll
epoll是Linux下多路复用IO接口select/poll的增强版本，它能显著提高程序在大量并发连接中只有少量活跃的情况下的系统CPU利用率，都连接但不发送数据
1. epoll_create函数
2. epoll_ctl函数
3. epoll_wait函数
epoll实现多路IO转接思路：
代码实现：
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <arpa/inet.h>
#include <sys/epoll.h>
#include <errno.h>
#include <ctype.h>
#include "wrap.h"
#define MAXLINE 8192
#define SERV_PORT 8000
#define OPEN_MAX 5000
int main(int argc, char *argv[])
{
int i, listenfd, connfd, sockfd;
int n, num = 0;
ssize_t nready, efd, res;
char buf[MAXLINE], str[INET_ADDRSTRLEN];
socklen_t clilen;
struct sockaddr_in cliaddr, servaddr;
listenfd = Socket(AF_INET, SOCK_STREAM, 0);
int opt = 1;
setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)); //端口复用
bzero(&servaddr, sizeof(servaddr));
servaddr.sin_family = AF_INET;
servaddr.sin_addr.s_addr = htonl(INADDR_ANY);
servaddr.sin_port = htons(SERV_PORT);
Bind(listenfd, (struct sockaddr *) &servaddr, sizeof(servaddr));
Listen(listenfd, 20);
efd = epoll_create(OPEN_MAX); //创建epoll模型, efd指向红黑树根节点
if (efd == -1)
perr_exit("epoll_create error");
struct epoll_event tep, ep[OPEN_MAX]; //tep: epoll_ctl参数 ep[] : epoll_wait参数
tep.events = EPOLLIN;
tep.data.fd = listenfd; //指定lfd的监听时间为"读"
res = epoll_ctl(efd, EPOLL_CTL_ADD, listenfd, &tep); //将lfd及对应的结构体设置到树上,efd可找到该树
if (res == -1)
perr_exit("epoll_ctl error");
for ( ; ; ) {
/*epoll为server阻塞监听事件, ep为struct epoll_event类型数组, OPEN_MAX为数组容量, -1表永久阻塞*/
nready = epoll_wait(efd, ep, OPEN_MAX, -1);
if (nready == -1)
perr_exit("epoll_wait error");
for (i = 0; i < nready; i++) {
if (!(ep[i].events & EPOLLIN)) //如果不是"读"事件, 继续循环
continue;
if (ep[i].data.fd == listenfd) { //判断满足事件的fd是不是lfd
clilen = sizeof(cliaddr);
connfd = Accept(listenfd, (struct sockaddr *)&cliaddr, &clilen); //接受链接
printf("received from %s at PORT %d\n",
inet_ntop(AF_INET, &cliaddr.sin_addr, str, sizeof(str)),
ntohs(cliaddr.sin_port));
printf("cfd %d---client %d\n", connfd, ++num);
tep.events = EPOLLIN; tep.data.fd = connfd;
res = epoll_ctl(efd, EPOLL_CTL_ADD, connfd, &tep); //加入红黑树
if (res == -1)
perr_exit("epoll_ctl error");
} else { //不是lfd,
sockfd = ep[i].data.fd;
n = Read(sockfd, buf, MAXLINE);
if (n == 0) { //读到0,说明客户端关闭链接
res = epoll_ctl(efd, EPOLL_CTL_DEL, sockfd, NULL); //将该文件描述符从红黑树摘除
if (res == -1)
perr_exit("epoll_ctl error");
Close(sockfd); //关闭与该客户端的链接
printf("client[%d] closed connection\n", sockfd);
} else if (n < 0) { //出错
perror("read n < 0 error: ");
res = epoll_ctl(efd, EPOLL_CTL_DEL, sockfd, NULL); //摘除节点
Close(sockfd);
} else { //实际读到了字节数
for (i = 0; i < n; i++)
buf[i] = toupper(buf[i]); //转大写,写回给客户端
Write(STDOUT_FILENO, buf, n);
Writen(sockfd, buf, n);
}
}
}
}
Close(listenfd);
Close(efd);
return 0;
}
三、epoll 事件模型
1. epoll事件有两种模型：
基于管道epoll ET/LT触发模式
#include <stdio.h>
#include <stdlib.h>
#include <sys/epoll.h>
#include <errno.h>
#include <unistd.h>
#define MAXLINE 10
int main(int argc, char *argv[])
{
int efd, i;
int pfd[2];
pid_t pid;
char buf[MAXLINE], ch = 'a';
pipe(pfd);
pid = fork();
if (pid == 0) { //子 写
close(pfd[0]);
while (1) {
//aaaa\n
for (i = 0; i < MAXLINE/2; i++)
buf[i] = ch;
buf[i-1] = '\n';
ch++;
//bbbb\n
for (; i < MAXLINE; i++)
buf[i] = ch;
buf[i-1] = '\n';
ch++;
//aaaa\nbbbb\n
write(pfd[1], buf, sizeof(buf));
sleep(5);
}
close(pfd[1]);
} else if (pid > 0) { //父 读
struct epoll_event event;
struct epoll_event resevent[10]; //epoll_wait就绪返回event
int res, len;
close(pfd[1]);
efd = epoll_create(10);
event.events = EPOLLIN | EPOLLET; // ET 边沿触发
// event.events = EPOLLIN; // LT 水平触发 (默认)
event.data.fd = pfd[0];
epoll_ctl(efd, EPOLL_CTL_ADD, pfd[0], &event);
while (1) {
res = epoll_wait(efd, resevent, 10, -1);
printf("res %d\n", res);
if (resevent[0].data.fd == pfd[0]) {
len = read(pfd[0], buf, MAXLINE/2);
write(STDOUT_FILENO, buf, len);
}
}
close(pfd[0]);
close(efd);
} else {
perror("fork");
exit(-1);
}
return 0;
}
2. ET的非阻塞模式
epoll 的 ET模式为高效模式，但是只支持非阻塞模式。 --- 忙轮询。
基于网络C/S非阻塞模型的epoll ET触发模式：
#include <stdio.h>
#include <string.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <sys/wait.h>
#include <sys/types.h>
#include <sys/epoll.h>
#include <unistd.h>
#include <fcntl.h>
#define MAXLINE 10
#define SERV_PORT 8000
int main(void)
{
struct sockaddr_in servaddr, cliaddr;
socklen_t cliaddr_len;
int listenfd, connfd;
char buf[MAXLINE];
char str[INET_ADDRSTRLEN];
int efd, flag;
listenfd = socket(AF_INET, SOCK_STREAM, 0);
bzero(&servaddr, sizeof(servaddr));
servaddr.sin_family = AF_INET;
servaddr.sin_addr.s_addr = htonl(INADDR_ANY);
servaddr.sin_port = htons(SERV_PORT);
bind(listenfd, (struct sockaddr *)&servaddr, sizeof(servaddr));
listen(listenfd, 20);
///
struct epoll_event event;
struct epoll_event res_event[10];
int res, len;
efd = epoll_create(10);
event.events = EPOLLIN | EPOLLET; /* ET 边沿触发，默认是水平触发 */
//event.events = EPOLLIN;
printf("Accepting connections ...\n");
cliaddr_len = sizeof(cliaddr);
connfd = accept(listenfd, (struct sockaddr *)&cliaddr, &cliaddr_len);
printf("received from %s at PORT %d\n",
inet_ntop(AF_INET, &cliaddr.sin_addr, str, sizeof(str)),
ntohs(cliaddr.sin_port));
flag = fcntl(connfd, F_GETFL); /* 修改connfd为非阻塞读 */
flag |= O_NONBLOCK;
fcntl(connfd, F_SETFL, flag);
event.data.fd = connfd;
epoll_ctl(efd, EPOLL_CTL_ADD, connfd, &event); //将connfd加入监听红黑树
while (1) {
printf("epoll_wait begin\n");
res = epoll_wait(efd, res_event, 10, -1); //最多10个, 阻塞监听
printf("epoll_wait end res %d\n", res);
if (res_event[0].data.fd == connfd) {
while ((len = read(connfd, buf, MAXLINE/2)) >0 ) //非阻塞读, 轮询
write(STDOUT_FILENO, buf, len);
}
}
return 0;
}
3. epoll优缺点
四、epoll反应堆模型
核心：epoll ET模式+非阻塞+void *ptr
反应堆的理解：加入IO转接之后，有了事件，server才去处理，这里反应堆也是这样，由于网络环境复杂，服务器处理数据之后，可能并不能直接写回去，比如遇到网络繁忙或者对方缓冲区已经满了这种情况，就不能直接写回给客户端。反应堆就是在处理数据之后，监听写事件，能写回客户端了，才去做写回操作。写回之后，在改回监听读事件，以此循环
源码实现：
/*
*epoll基于非阻塞I/O事件驱动
*/
#include <stdio.h>
#include <sys/socket.h>
#include <sys/epoll.h>
#include <arpa/inet.h>
#include <fcntl.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <stdlib.h>
#include <time.h>
#define MAX_EVENTS 1024 //监听上限数
#define BUFLEN 4096
#define SERV_PORT 8080
void recvdata(int fd, int events, void *arg);
void senddata(int fd, int events, void *arg);
/* 描述就绪文件描述符相关信息 */
struct myevent_s {
int fd; //要监听的文件描述符
int events; //对应的监听事件
void *arg; //泛型参数
void (*call_back)(int fd, int events, void *arg); //回调函数
int status; //是否在监听:1->在红黑树上(监听), 0->不在(不监听)
char buf[BUFLEN];
int len;
long last_active; //记录每次加入红黑树 g_efd 的时间值
};
int g_efd; //全局变量, 保存epoll_create返回的文件描述符
struct myevent_s g_events[MAX_EVENTS+1]; //自定义结构体类型数组. +1-->listen fd
/*将结构体 myevent_s 成员变量 初始化*/
void eventset(struct myevent_s *ev, int fd, void (*call_back)(int, int, void *), void *arg)
{
ev->fd = fd;
ev->call_back = call_back;
ev->events = 0;
ev->arg = arg;
ev->status = 0;
memset(ev->buf, 0, sizeof(ev->buf));
ev->len = 0;
ev->last_active = time(NULL); //调用eventset函数的时间
return;
}
/* 向 epoll监听的红黑树 添加一个 文件描述符 */
//eventadd(efd, EPOLLIN, &g_events[MAX_EVENTS]);
void eventadd(int efd, int events, struct myevent_s *ev)
{
struct epoll_event epv = {0, {0}};
int op;
epv.data.ptr = ev;
epv.events = ev->events = events; //EPOLLIN 或 EPOLLOUT
if (ev->status == 0) { //已经在红黑树 g_efd 里
op = EPOLL_CTL_ADD; //将其加入红黑树 g_efd, 并将status置1
ev->status = 1;
}
if (epoll_ctl(efd, op, ev->fd, &epv) < 0) //实际添加/修改
printf("event add failed [fd=%d], events[%d]\n", ev->fd, events);
else
printf("event add OK [fd=%d], op=%d, events[%0X]\n", ev->fd, op, events);
return ;
}
/* 从epoll 监听的 红黑树中删除一个 文件描述符*/
void eventdel(int efd, struct myevent_s *ev)
{
struct epoll_event epv = {0, {0}};
if (ev->status != 1) //不在红黑树上
return ;
//epv.data.ptr = ev;
epv.data.ptr = NULL;
ev->status = 0; //修改状态
epoll_ctl(efd, EPOLL_CTL_DEL, ev->fd, &epv); //从红黑树 efd 上将 ev->fd 摘除
return ;
}
/* 当有文件描述符就绪, epoll返回, 调用该函数 与客户端建立链接 */
void acceptconn(int lfd, int events, void *arg)
{
struct sockaddr_in cin;
socklen_t len = sizeof(cin);
int cfd, i;
if ((cfd = accept(lfd, (struct sockaddr *)&cin, &len)) == -1) {
if (errno != EAGAIN && errno != EINTR) {
/* 暂时不做出错处理 */
}
printf("%s: accept, %s\n", __func__, strerror(errno));
return ;
}
do {
for (i = 0; i < MAX_EVENTS; i++) //从全局数组g_events中找一个空闲元素
if (g_events[i].status == 0) //类似于select中找值为-1的元素
break; //跳出 for
if (i == MAX_EVENTS) {
printf("%s: max connect limit[%d]\n", __func__, MAX_EVENTS);
break; //跳出do while(0) 不执行后续代码
}
int flag = 0;
if ((flag = fcntl(cfd, F_SETFL, O_NONBLOCK)) < 0) { //将cfd也设置为非阻塞
printf("%s: fcntl nonblocking failed, %s\n", __func__, strerror(errno));
break;
}
/* 给cfd设置一个 myevent_s 结构体, 回调函数 设置为 recvdata */
eventset(&g_events[i], cfd, recvdata, &g_events[i]);
eventadd(g_efd, EPOLLIN, &g_events[i]); //将cfd添加到红黑树g_efd中,监听读事件
} while(0);
printf("new connect [%s:%d][time:%ld], pos[%d]\n",
inet_ntoa(cin.sin_addr), ntohs(cin.sin_port), g_events[i].last_active, i);
return ;
}
void recvdata(int fd, int events, void *arg)
{
struct myevent_s *ev = (struct myevent_s *)arg;
int len;
len = recv(fd, ev->buf, sizeof(ev->buf), 0); //读文件描述符, 数据存入myevent_s成员buf中
eventdel(g_efd, ev); //将该节点从红黑树上摘除
if (len > 0) {
ev->len = len;
ev->buf[len] = '要求：能看懂epoll反应堆模型实现源码即可'; //手动添加字符串结束标记
printf("C[%d]:%s\n", fd, ev->buf);
eventset(ev, fd, senddata, ev); //设置该 fd 对应的回调函数为 senddata
eventadd(g_efd, EPOLLOUT, ev); //将fd加入红黑树g_efd中,监听其写事件
} else if (len == 0) {
close(ev->fd);
/* ev-g_events 地址相减得到偏移元素位置 */
printf("[fd=%d] pos[%ld], closed\n", fd, ev-g_events);
} else {
close(ev->fd);
printf("recv[fd=%d] error[%d]:%s\n", fd, errno, strerror(errno));
}
return;
}
void senddata(int fd, int events, void *arg)
{
struct myevent_s *ev = (struct myevent_s *)arg;
int len;
len = send(fd, ev->buf, ev->len, 0); //直接将数据 回写给客户端。未作处理
eventdel(g_efd, ev); //从红黑树g_efd中移除
if (len > 0) {
printf("send[fd=%d], [%d]%s\n", fd, len, ev->buf);
eventset(ev, fd, recvdata, ev); //将该fd的 回调函数改为 recvdata
eventadd(g_efd, EPOLLIN, ev); //从新添加到红黑树上， 设为监听读事件
} else {
close(ev->fd); //关闭链接
printf("send[fd=%d] error %s\n", fd, strerror(errno));
}
return ;
}
/*创建 socket, 初始化lfd */
void initlistensocket(int efd, short port)
{
struct sockaddr_in sin;
int lfd = socket(AF_INET, SOCK_STREAM, 0);
fcntl(lfd, F_SETFL, O_NONBLOCK); //将socket设为非阻塞
memset(&sin, 0, sizeof(sin)); //bzero(&sin, sizeof(sin))
sin.sin_family = AF_INET;
sin.sin_addr.s_addr = INADDR_ANY;
sin.sin_port = htons(port);
bind(lfd, (struct sockaddr *)&sin, sizeof(sin));
listen(lfd, 20);
/* void eventset(struct myevent_s *ev, int fd, void (*call_back)(int, int, void *), void *arg); */
eventset(&g_events[MAX_EVENTS], lfd, acceptconn, &g_events[MAX_EVENTS]);
/* void eventadd(int efd, int events, struct myevent_s *ev) */
eventadd(efd, EPOLLIN, &g_events[MAX_EVENTS]);
return ;
}
int main(int argc, char *argv[])
{
unsigned short port = SERV_PORT;
if (argc == 2)
port = atoi(argv[1]); //使用用户指定端口.如未指定,用默认端口
g_efd = epoll_create(MAX_EVENTS+1); //创建红黑树,返回给全局 g_efd
if (g_efd <= 0)
printf("create efd in %s err %s\n", __func__, strerror(errno));
initlistensocket(g_efd, port); //初始化监听socket
struct epoll_event events[MAX_EVENTS+1]; //保存已经满足就绪事件的文件描述符数组
printf("server running:port[%d]\n", port);
int checkpos = 0, i;
while (1) {
/* 超时验证，每次测试100个链接，不测试listenfd 当客户端60秒内没有和服务器通信，则关闭此客户端链接 */
long now = time(NULL); //当前时间
for (i = 0; i < 100; i++, checkpos++) { //一次循环检测100个。 使用checkpos控制检测对象
if (checkpos == MAX_EVENTS)
checkpos = 0;
if (g_events[checkpos].status != 1) //不在红黑树 g_efd 上
continue;
long duration = now - g_events[checkpos].last_active; //客户端不活跃的世间
if (duration >= 60) {
close(g_events[checkpos].fd); //关闭与该客户端链接
printf("[fd=%d] timeout\n", g_events[checkpos].fd);
eventdel(g_efd, &g_events[checkpos]); //将该客户端 从红黑树 g_efd移除
}
}
/*监听红黑树g_efd, 将满足的事件的文件描述符加至events数组中, 1秒没有事件满足, 返回 0*/
int nfd = epoll_wait(g_efd, events, MAX_EVENTS+1, 1000);
if (nfd < 0) {
printf("epoll_wait error, exit\n");
break;
}
for (i = 0; i < nfd; i++) {
/*使用自定义结构体myevent_s类型指针, 接收 联合体data的void *ptr成员*/
struct myevent_s *ev = (struct myevent_s *)events[i].data.ptr;
if ((events[i].events & EPOLLIN) && (ev->events & EPOLLIN)) { //读就绪事件
ev->call_back(ev->fd, events[i].events, ev->arg);
//lfd EPOLLIN
}
if ((events[i].events & EPOLLOUT) && (ev->events & EPOLLOUT)) { //写就绪事件
ev->call_back(ev->fd, events[i].events, ev->arg);
}
}
}
/* 退出前释放所有资源 */
return 0;
}