static ngx_int_t ngx_epoll_add_event(ngx_event_t *ev, ngx_int_t event, ngx_uint_t flags) { int op; uint32_t events, prev; ngx_event_t *e; ngx_connection_t *c; struct epoll_event ee; c = ev->data; events = (uint32_t) event; if (event == NGX_READ_EVENT) { e = c->write; prev = EPOLLOUT; } else { e = c->read; prev = EPOLLIN; } if (e->active) { op = EPOLL_CTL_MOD; events |= prev; } else { op = EPOLL_CTL_ADD; } ee.events = events | (uint32_t) flags; ee.data.ptr = (void *) ((uintptr_t) c | ev->instance); if (epoll_ctl(ep, op, c->fd, &ee) == -1) { return NGX_ERROR; } ev->active = 1; return NGX_OK; }
这就是那个在群里被问了无数遍的函数,在讨论它之前,可以先来看下这个http://blog.csdn.net/dingyujie/article/details/8570764,在这篇文章中特别强调了两个重要变量(即active和ready)的用途。看过之后,再来看这个函数。大家的问题主要是围绕在函数的开头:
为什么明明处理的是NGX_READ_EVENT,即所谓的读事件,为什么要去管c->write?即代码:
if (event == NGX_READ_EVENT) { e = c->write; prev = EPOLLOUT; } else { e = c->read; prev = EPOLLIN; }
其实答案就跟在后面:
if (e->active) { op = EPOLL_CTL_MOD; events |= prev; } else { op = EPOLL_CTL_ADD; }
结合这两段代码,意图就很明显了(明显吗?既然明显,为什么还有不少人在这里被绊住了呢?)。在epoll中监控读写事件时,主要通过接口epoll_ctl来处理。注意使用时有两种操作add和mod,当一个fd第一次注册到epoll时,使用add方式。而如果之前已经添加过对该fd读写事件的监控,就要通过mod来修改原来的监控方式,告知epoll我们的需求。如果一个fd被重复执行add会报错。
man epoll: Q: What happens if you register the same file descriptor on an epoll instance twice? A: You will probably get EEXIST. However, it is possible to add a duplicate (dup(2), dup2(2), fcntl(2) F_DUPFD) descriptor to the same epoll instance. This can be a useful technique for filtering events, if the duplicate file descriptors are registered with dif- ferent events masks.
所以nginx这里就是为了避免这种情况,当要在epoll中加入对一个fd读事件(即NGX_READ_EVENT)的监听时,nginx先看一下与这个fd相关的写事件的状态,即e=c->write,如果此时e->active为1,说明该fd之前已经以NGX_WRITE_EVENT方式被加到epoll中了,此时只需要使用mod方式,将我们的需求加进去,否则才使用add方式,将该fd注册到epoll中。反之处理NGX_WRITE_EVENT时道理是一样的。