|
|
最近有朋友在面試的時(shí)候被問(wèn)了select 和epoll效率差的原因，和一般人一樣，大部分都會(huì)回答select是輪詢(xún)、epoll是觸發(fā)式的，所以效率高。這個(gè)答案聽(tīng)上去很完美，大致也說(shuō)出了二者的主要區(qū)別。今天閑來(lái)無(wú)事，翻看了下內(nèi)核代碼，結(jié)合內(nèi)核代碼和大家分享下我的觀點(diǎn)。一、連接數(shù)我本人也曾經(jīng)在項(xiàng)目中用過(guò)select和epoll,對(duì)于select，感觸最深的是linux下select最大數(shù)目限制(windows 下似乎沒(méi)有限制)，每個(gè)進(jìn)程的select最多能處理FD_SETSIZE個(gè)FD(文件句柄)，如果要處理超過(guò)1024個(gè)句柄，只能采用多進(jìn)程了。常見(jiàn)的使用slect的多進(jìn)程模型是這樣的： 一個(gè)進(jìn)程專(zhuān)門(mén)accept，成功后將fd通過(guò)unix socket傳遞給子進(jìn)程處理，父進(jìn)程可以根據(jù)子進(jìn)程負(fù)載分派。曾經(jīng)用過(guò)1個(gè)父進(jìn)程+4個(gè)子進(jìn)程 承載了超過(guò)4000個(gè)的負(fù)載。這種模型在我們當(dāng)時(shí)的業(yè)務(wù)運(yùn)行的非常好。epoll在連接數(shù)方面沒(méi)有限制，當(dāng)然可能需要用戶(hù)調(diào)用API重現(xiàn)設(shè)置進(jìn)程的資源限制。二、IO差別1、select的實(shí)現(xiàn)這段可以結(jié)合linux內(nèi)核代碼描述了，我使用的是2.6.28，其他2.6的代碼應(yīng)該差不多吧。先看看select:select系統(tǒng)調(diào)用的代碼在fs/Select.c下，asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp){ struct timespec end_time, *to = NULL; struct timeval tv; int ret; if (tvp) { if (copy_from_user(&tv, tvp, sizeof(tv))) return -EFAULT; to = &end_time; if (poll_select_set_timeout(to, tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) return -EINVAL; } ret = core_sys_select(n, inp, outp, exp, to); ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); return ret;} 前面是從用戶(hù)控件拷貝各個(gè)fd_set到內(nèi)核空間，接下來(lái)的具體工作在core_sys_select中，core_sys_select->do_select,真正的核心內(nèi)容在do_select里：int do_select(int n, fd_set_bits *fds, struct timespec *end_time){ ktime_t expire, *to = NULL; struct poll_wqueues table; poll_table *wait; int retval, i, timed_out = 0; unsigned long slack = 0; rcu_read_lock(); retval = max_select_fd(n, fds); rcu_read_unlock(); if (retval < 0) return retval; n = retval; poll_initwait(&table); wait = &table.pt; if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { wait = NULL; timed_out = 1; } if (end_time && !timed_out) slack = estimate_accuracy(end_time); retval = 0; for (;;) { unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; set_current_state(TASK_INTERRUPTIBLE); inp = fds->in; outp = fds->out; exp = fds->ex; rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; for (i = 0; i < n; ++rinp, ++routp, ++rexp) { unsigned long in, out, ex, all_bits, bit = 1, mask, j; unsigned long res_in = 0, res_out = 0, res_ex = 0; const struct file_operations *f_op = NULL; struct file *file = NULL; in = *inp++; out = *outp++; ex = *exp++; all_bits = in | out | ex; if (all_bits == 0) { i += __NFDBITS; continue; } for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { int fput_needed; if (i >= n) break; if (!(bit & all_bits)) continue; file = fget_light(i, &fput_needed); if (file) { f_op = file->f_op; mask = DEFAULT_POLLMASK; if (f_op && f_op->poll) mask = (*f_op->poll)(file, retval ? NULL : wait); fput_light(file, fput_needed); if ((mask & POLLIN_SET) && (in & bit)) { res_in |= bit; retval++; } if ((mask & POLLOUT_SET) && (out & bit)) { res_out |= bit; retval++; } if ((mask & POLLEX_SET) && (ex & bit)) { res_ex |= bit; retval++; } } } if (res_in) *rinp = res_in; if (res_out) *routp = res_out; if (res_ex) *rexp = res_ex; cond_resched(); } wait = NULL; if (retval || timed_out || signal_pending(current)) break; if (table.error) { retval = table.error; break; } /* * If this is the first loop and we have a timeout * given, then we convert to ktime_t and set the to * pointer to the expiry value. */ if (end_time && !to) { expire = timespec_to_ktime(*end_time); to = &expire; } if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) timed_out = 1; } __set_current_state(TASK_RUNNING); poll_freewait(&table); return retval;} 上面的代碼很多，其實(shí)真正關(guān)鍵的代碼是這一句:mask = (*f_op->poll)(file, retval ? NULL : wait); 這個(gè)是調(diào)用文件系統(tǒng)的 poll函數(shù)，不同的文件系統(tǒng)poll函數(shù)自然不同，由于我們這里關(guān)注的是tcp連接，而socketfs的注冊(cè)在 net/Socket.c里。register_filesystem(&sock_fs_type); socket文件系統(tǒng)的函數(shù)也是在net/Socket.c里：static const struct file_operations socket_file_ops = { .owner = THIS_MODULE, .llseek = no_llseek, .aio_read = sock_aio_read, .aio_write = sock_aio_write, .poll = sock_poll, .unlocked_ioctl = sock_ioctl,#ifdef CONFIG_COMPAT .compat_ioctl = compat_sock_ioctl,#endif .mmap = sock_mmap, .open = sock_no_open, /* special open code to disallow open via /proc */ .release = sock_close, .fasync = sock_fasync, .sendpage = sock_sendpage, .splice_write = generic_splice_sendpage, .splice_read = sock_splice_read,};從sock_poll跟隨下去，最后可以到 net/ipv4/tcp.c的unsigned int tcp_poll(struct file *file, struct socket *sock, poll_table *wait) 這個(gè)是最終的查詢(xún)函數(shù)，也就是說(shuō)select 的核心功能是調(diào)用tcp文件系統(tǒng)的poll函數(shù)，不停的查詢(xún)，如果沒(méi)有想要的數(shù)據(jù)，主動(dòng)執(zhí)行一次調(diào)度（防止一直占用cpu），直到有一個(gè)連接有想要的消息為止。從這里可以看出select的執(zhí)行方式基本就是不同的調(diào)用poll,直到有需要的消息為止，如果select 處理的socket很多，這其實(shí)對(duì)整個(gè)機(jī)器的性能也是一個(gè)消耗。2、epoll的實(shí)現(xiàn)epoll的實(shí)現(xiàn)代碼在 fs/EventPoll.c下，由于epoll涉及到幾個(gè)系統(tǒng)調(diào)用，這里不逐個(gè)分析了，僅僅分析幾個(gè)關(guān)鍵點(diǎn)，第一個(gè)關(guān)鍵點(diǎn)在static int ep_insert(struct eventpoll *ep, struct epoll_event *event, struct file *tfile, int fd) 這是在我們調(diào)用sys_epoll_ctl 添加一個(gè)被管理socket的時(shí)候調(diào)用的函數(shù)，關(guān)鍵的幾行如下：epq.epi = epi; init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); /* * Attach the item to the poll hooks and get current event bits. * We can safely use the file* here because its usage count has * been increased by the caller of this function. Note that after * this operation completes, the poll callback can start hitting * the new item. */ revents = tfile->f_op->poll(tfile, &epq.pt); 這里也是調(diào)用文件系統(tǒng)的poll函數(shù)，不過(guò)這次初始化了一個(gè)結(jié)構(gòu)，這個(gè)結(jié)構(gòu)會(huì)帶有一個(gè)poll函數(shù)的callback函數(shù)：ep_ptable_queue_proc，在調(diào)用poll函數(shù)的時(shí)候，會(huì)執(zhí)行這個(gè)callback，這個(gè)callback的功能就是將當(dāng)前進(jìn)程添加到 socket的等待進(jìn)程上。static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, poll_table *pt){ struct epitem *epi = ep_item_from_epqueue(pt); struct eppoll_entry *pwq; if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); pwq->whead = whead; pwq->base = epi; add_wait_queue(whead, &pwq->wait); list_add_tail(&pwq->llink, &epi->pwqlist); epi->nwait++; } else { /* We have to signal that an error occurred */ epi->nwait = -1; }} 注意到參數(shù) whead 實(shí)際上是 sk->sleep，其實(shí)就是將當(dāng)前進(jìn)程添加到sk的等待隊(duì)列里，當(dāng)該socket收到數(shù)據(jù)或者其他事件觸發(fā)時(shí)，會(huì)調(diào)用sock_def_readable 或者sock_def_write_space 通知函數(shù)來(lái)喚醒等待進(jìn)程，這2個(gè)函數(shù)都是在socket創(chuàng)建的時(shí)候填充在sk結(jié)構(gòu)里的。從前面的分析來(lái)看，epoll確實(shí)是比select聰明的多、輕松的多，不用再苦哈哈的去輪詢(xún)了。
|