epoll触发机制分析

[taikulawo]

目标：

分析 kernel 驱动收到后skb

1. 如何唤醒epoll callback？
2. accpet syscall 如何从listener socket spawn 子 socket？

一个极简epoll server程序（无法编译，只说明流程）

// 创建listener socket
int listener_fd = socket();
bind(listener_fd);
listen(listener_fd);
set_socket_non_blocking(listener_fd);

// 创建socket
ep = epoll_create1(0);
struct epoll_event event;
event.data.fd = listener_fd;
event.events = EPOLLIN | EPOLLET;
// 将listener socket注册进epoll
s = epoll_ctl(efd, EPOLL_CTL_ADD, sfd, &event);
for ( ;; ) {
    // eventloop
    int n = epoll_wait(ep, &events, 1024, -1);
    for (itn i = 0; i < n ; i ++) {
        if(events[i].data.fd == listener_fd) {
            // new incoming connection
            int child = accept(listener_fd, &addr, &addr_len);
            // 将子连接socket加入epoll eventloop
            make_socket_non_blocking(infd);
            struct epoll_event;
            event.data.fd = infd;
            event.events = EPOLLIN | EPOLLET;
            s = epoll_ctl(ep, EPOLL_CTL_ADD, child, &event);
        }
    }
}

epoll 注册 wake callback

直接看追踪的call stack

1. epoll_ctl(ep, EPOLL_CTL_ADD, child, &event): called from epoll_ctl syscall
2. ep_insert:

   // ep_insert
   epq.epi = epi;
   init_poll_funcptr(&epq.pt, ep_ptable_queue_proc);

把 ep_ptable_queue_proc 放入 _qproc，在接下来的 poll_wait 调用

static inline void init_poll_funcptr(poll_table *pt, poll_queue_proc qproc)
{
    pt->_qproc = qproc;
    pt->_key   = ~(__poll_t)0; /* all events enabled */
}

3. ep_item_poll
   1. 如果fd不是file epoll，那么通过vfs定向给具体实现
4. vfs_poll: virtual fs poll，call file->f_op->poll(file, pt);
   1. sock poll: 文件底层对应的是sock
   2. unix_poll: address family unix poll
5. sock_poll: socket 文件类型的poll
6. tcp_poll: tcp socket poll 实现
7. sock_poll_wait
8. poll_wait
9. ep_ptable_queue_proc
   1. ep_poll_callback 添加到 wait queue

struct file 转换成 struct sock

struct sock 有最核心弄明白的点在于skb buff 如何转换成 struct socket 结构

static __poll_t sock_poll(struct file *file, poll_table *wait)
{
    // 关键点
    struct socket *sock = file->private_data;
    const struct proto_ops *ops = READ_ONCE(sock->ops);
    __poll_t events = poll_requested_events(wait), flag = 0;

    if (!ops->poll)
        return 0;

    if (sk_can_busy_loop(sock->sk)) {
        /* poll once if requested by the syscall */
        if (events & POLL_BUSY_LOOP)
            sk_busy_loop(sock->sk, 1);

        /* if this socket can poll_ll, tell the system call */
        flag = POLL_BUSY_LOOP;
    }

    return ops->poll(file, sock, wait) | flag;
}

sock_poll将 file->private_data 转换成 struct sock，从而调用到epoll的socket实现

Why does the Linux kernel have `struct sock` and `struct socket`? - Stack Overflow

sock wake epoll callback

唤醒ep->wqep_poll_callback

[iugo]

是在尝试实现一个 epoll 吗?

[taikulawo]

是在尝试实现一个 epoll 吗?

调试kernel协议栈时分析的。这是分析TCP/IP skb rcv最后一环（wake waiter）的文章