Linux select/poll源码剖析

linux内核版本：2.6.34

在读select、poll源码前，需要先了解的知识点：

1. 等待队列
2. 文件系统（主要是进程的打开文件描述符表以及struct file）
3. poll机制
   • 资源注册监听poll() -> poll_wait(struct file *, wait_queue_t *, poll_table *pt) -> pt->qproc(struct file *, wait_queue_t *, poll_table *)
   • 资源就绪通知callback_function(wait_queue_t *, unsigned mode, int sync, void *key)
4. select/poll主要数据结构
   • 一个select()/poll()调用对应一个struct poll_wqueues
   • 一个监听事件对应一个struct poll_table_entry

Common

poll机制是所有多路转接的共性；调用控制块(struct poll_wqueues)与监听事件项(struct poll_table_entry)是select()与poll()的共性；事件描述集(fdset)与事件描述符(struct pollfd)是select()与poll()的特性。

/*
 * Structures and helpers for sys_poll/sys_poll
 */
struct poll_wqueues {
    poll_table pt;
    struct poll_table_page *table;
    /* 
     * 每次select()都会初始化一个poll_wqueues结构与这个
     * 调用相对应。结构中的polling_task字段指向调用进程
     * （也就是current进程）的task_struct
     *
     * 资源等待队列节点wait_queue_t中的private字段指向
     * 对应的poll_wqueues对象。因此资源就绪时，通过获得
     * poll_wqueues对象，然后访问其中的polling_task字段
     * 能够得到调用进程的PCB，然后对其进行唤醒...
     */
    struct task_struct *polling_task;
    /* 已触发标记 */
    int triggered;
    int error;
    int inline_index;
    /* poll_table_entry结构缓存数组 */
    struct poll_table_entry inline_entries[N_INLINE_POLL_ENTRIES];
};

typedef struct poll_table_struct {
    poll_queue_proc qproc;
    unsigned long key;
} poll_table;

struct poll_table_page {
    struct poll_table_page * next;
    /* 指向下一可用的poll_table_entry结构 */
    struct poll_table_entry * entry;
    struct poll_table_entry entries[0];
};

/* 真正被挂载到资源等待队列中的结构 */
struct poll_table_entry {
    /* 链接到资源文件 */
    struct file *filp;
    /* 存储监听的事件 */
    unsigned long key;
    /* 
     * 挂载到资源等待队列的节点，其中包含了唤醒回调函数
     *
     * 节点的private成员关联poll_table_entry所属的poll_wqueues，
     */
    wait_queue_t wait;
    /* 指向资源等待队列队列头 */
    wait_queue_head_t *wait_address;
};

void poll_initwait(struct poll_wqueues *pwq)
{
    init_poll_funcptr(&pwq->pt, __pollwait);
    pwq->polling_task = current;
    pwq->triggered = 0;
    pwq->error = 0;
    pwq->table = NULL;
    pwq->inline_index = 0;
}

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

/* Add a new entry */
/**
 * __pollwait - 将poll_table_entry挂载到资源文件的监听队列
 * @file: 被监听的资源文件
 * @wait_address: 被监听的资源文件的等待队列头
 * @p: 在poll_initwait()中设置的poll_tbale
 */
static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
                poll_table *p)
{
    /* 获取poll_wqueues */
    struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
    /* 从poll_wqueues中取得一个poll_table_entry结构 */
    struct poll_table_entry *entry = poll_get_entry(pwq);
    if (!entry)
        return;
    /* 增加资源文件引用计数 */
    get_file(filp);
    /* 关联资源文件 */
    entry->filp = filp;
    /* 保存资源文件监听队列队列头 */
    entry->wait_address = wait_address;
    /* 设置想要监听事件 */
    entry->key = p->key;
    /* 
     * 初始化一个等待队列节点，其中唤醒函数设置为pollwake
     *
     * 重点！！！：
     * 唤醒函数为pollwake
     */
    init_waitqueue_func_entry(&entry->wait, pollwake);
    /*
     * 来看一下为什么等待队列节点的private要这样设计：
     * 1. 实际linux内核设计：
     *    每个wait_queue_t的private字段指向同一个poll_wqueues，然后
     *    共用的poll_wqueues中保存了指向调用进程PCB的指针，这样总共
     *    需要n + 1个指针...
     * 2. 假想设计：
     *    每个wait_queue_t的private字段指向调用进程PCB，对应的事件
     *    结构poll_table_entry中每个都保存了指向同一个poll_wqueues
     *    的指针，这样总共需要n + n个指针...
     */
    entry->wait.private = pwq;
    /* 将poll_table_entry挂载到资源文件的监听队列中 */
    add_wait_queue(wait_address, &entry->wait);
}

static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
{
    struct poll_table_page *table = p->table;

    /* 缓存数组够用就从缓存数组中分配... */
    if (p->inline_index < N_INLINE_POLL_ENTRIES)
        return p->inline_entries + p->inline_index++;

    /* 动态分配的内存为空或者已用完... */
    if (!table || POLL_TABLE_FULL(table)) {
        struct poll_table_page *new_table;

        /* 分配一页的内存给poll_table_page使用 */
        new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
        if (!new_table) {
            p->error = -ENOMEM;
            return NULL;
        }
        new_table->entry = new_table->entries;
        new_table->next = table;
        p->table = new_table;
        table = new_table;
    }

    return table->entry++;
}

void poll_freewait(struct poll_wqueues *pwq)
{
    struct poll_table_page * p = pwq->table;
    int i;
    /* 对缓存数组中的poll_table_entry进行卸载 */
    for (i = 0; i < pwq->inline_index; i++)
        free_poll_entry(pwq->inline_entries + i);
    /* 对动态内存中的poll_table_entry进行卸载 */
    while (p) {
        struct poll_table_entry * entry;
        struct poll_table_page *old;

        entry = p->entry;
        do {
            /* 一个poll_table_page对象中至少分配了一个
             * poll_table_entry，所以entry--是安全的 */
            entry--;
            free_poll_entry(entry);
        } while (entry > p->entries);
        old = p;
        p = p->next;
        free_page((unsigned long) old);
    }
}

static void free_poll_entry(struct poll_table_entry *entry)
{
    remove_wait_queue(entry->wait_address, &entry->wait);
    fput(entry->filp);
}

/**
 * pollwake - 唤醒回调函数，这个函数验证资源当前状态中是否有我们所关心的
 *            事件，如果没有，就忽略这次唤醒；如果有，就转调用__pollwake...
 * @wait: poll_table_entry.wait
 * @mode:
 * @key: 携带资源当前状态
 */
static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
{
    struct poll_table_entry *entry;

    /* 通过poll_wqueues中的wait获取poll_table_entry */
    entry = container_of(wait, struct poll_table_entry, wait);
    /* 如果资源的已就绪的状态中没有我们所关心的events的话，直接忽略返回 */
    if (key && !((unsigned long)key & entry->key))
        return 0;
    /* 有我们所关心的events，那就转调用__pollwake去处理吧... */
    return __pollwake(wait, mode, sync, key);
}

/* 资源就绪时真正调用的唤醒回调函数 */
static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
{
    /* 见__pollwait()中关于等待队列节点private字段的注释 */
    struct poll_wqueues *pwq = wait->private;
    /* 构造一个有效的等待队列节点，private字段指向调用进程的PCB */
    DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);

    /*
     * Although this function is called under waitqueue lock, LOCK
     * doesn't imply write barrier and the users expect write
     * barrier semantics on wakeup functions.  The following
     * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
     * and is paired with set_mb() in poll_schedule_timeout.
     */
    smp_wmb();
    /* 设置已触发状态标志 */
    pwq->triggered = 1;

    /*
     * Perform the default wake up operation using a dummy
     * waitqueue.
     *
     * TODO: This is hacky but there currently is no interface to
     * pass in @sync.  @sync is scheduled to be removed and once
     * that happens, wake_up_process() can be used directly.
     */
    /* 唤醒select()的调用进程 */
    return default_wake_function(&dummy_wait, mode, sync, key);
}

select

调用链：

sys_select() -> core_sys_select() -> do_select() -> 1. poll_initwait()
                                                 -> 2. f_op->poll() [ -> poll_wait() -> __pollwait() -> poll_get_entry() ]
                                                 -> 3. [ block -> goto 2 ]
                                                 -> 4. poll_freewait() -> free_poll_entry()

pollwake() -> __pollwake()

typedef __kernel_fd_set     fd_set;

#define __NFDBITS   (8 * sizeof(unsigned long))
#define __FD_SETSIZE    1024
#define __FDSET_LONGS   (__FD_SETSIZE/__NFDBITS)

/* fd_set中能够表示的最大文件描述符为__FD_SETSIZE - 1 */
typedef struct {
    unsigned long fds_bits [__FDSET_LONGS];
} __kernel_fd_set;

/*
 * Scaleable version of the fd_set.
 */
typedef struct {
    unsigned long *in, *out, *ex;
    unsigned long *res_in, *res_out, *res_ex;
} fd_set_bits;

/*
 * How many longwords for "nr" bits?
 */
/* 一个long包含的bit数 */
#define FDS_BITPERLONG  (8*sizeof(long))
/* nr个bit需要多少long才能装下 */
#define FDS_LONGS(nr)   (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
/* nr个bit需要多大的long数组才能装下 */
#define FDS_BYTES(nr)   (FDS_LONGS(nr)*sizeof(long))

SYSCALL_DEFINE5(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;

    /* 
     * 1. 永久等待      tvp == NULL
     * 2. 根本不等待    tvp->tv_sec == 0 && tvp->tc_nsec == 0
     * 3. 等待指定时间  tvp->tv_sec != 0 || tvp->tc_nsec != 0
     */
    if (tvp) {
        /* 将超时时间从用户空间拷贝到内核空间 */
        if (copy_from_user(&tv, tvp, sizeof(tv)))
            return -EFAULT;

        to = &end_time;
        /* 处理超时时间 */
        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;
    }

    /*
     * 处理完超时时间之后：
     * 1. 永久等待      to == NULL
     * 2. 根本不等待    to->ts_sec == 0 && to->ts_nsec == 0
     * 3. 等待指定时间  to指向的timespec结构表示绝对超时时间
     */
    /* 主线（至此，我们只进行了超时时间处理） */
    ret = core_sys_select(n, inp, outp, exp, to);
    /* 将剩余超时时间通过tvp指向的timeval结构返回给用户空间 */
    ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);

    return ret;
}

int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
               fd_set __user *exp, struct timespec *end_time)
{
    fd_set_bits fds;
    void *bits;
    int ret, max_fds;
    unsigned int size;
    struct fdtable *fdt;
    /* Allocate small arguments on the stack to save memory and be faster */
    /*
     * 我们需要将用户空间传进来的inset、outset、exset拷贝到内核空间，并且
     * 需要等容量的空间来存储结果集，之后会将结果集的内容写回到用户空间。
     * 我们先在栈上分配一块缓冲区，用于缓存输入集以及结果集，如果缓存的
     * 空间大小不够，那么再使用kmalloc()动态分配，优先使用栈缓存而不用动态
     * 内存可以加快访问...
     */
    long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];

    ret = -EINVAL;
    if (n < 0)
        goto out_nofds;

    /* max_fds can increase, so grab it once to avoid race */
    /* rdlock加锁：保护struct files的访问 */
    rcu_read_lock();
    fdt = files_fdtable(current->files);
    max_fds = fdt->max_fds;
    rcu_read_unlock();
    /*
     * 根据当前文件描述符表能表示的最大文件描述符对输入参数n进行修正
     *
     * fdtable.maxfd的意义：
     * 在expand_fdtable() -> alloc_fdtable()中有以下代码：
     *  fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
     *  fdt->max_fds = nr;
     *  data = alloc_fdmem(nr * sizeof(struct file *));
     *  fdt->fd = (struct file **)data;
     */
    if (n > max_fds)
        n = max_fds;

    /*
     * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
     * since we used fdset we need to allocate memory in units of
     * long-words. 
     */
    /* 
     * n个bits至少需要size个long才能装下（之后我们使用long表示bits段）
     * 为了存储输入集与结果集，我们需要6*size个long的存储空间
     * 如果我们在栈上分配的那个缓冲区够用，那么就用它；而如果空间
     * 容纳不下的话，那么我们只好kmalloc()动态分配内存了...
     */
    size = FDS_BYTES(n);
    bits = stack_fds;
    if (size > sizeof(stack_fds) / 6) {
        /* Not enough space in on-stack array; must use kmalloc */
        ret = -ENOMEM;
        bits = kmalloc(6 * size, GFP_KERNEL);
        if (!bits)
            goto out_nofds;
    }
    /* 
     * 将数组划分为6份...
     * inset、outset、exset
     * res_inset、res_outset、res_exset
     */
    fds.in      = bits;
    fds.out     = bits +   size;
    fds.ex      = bits + 2*size;
    fds.res_in  = bits + 3*size;
    fds.res_out = bits + 4*size;
    fds.res_ex  = bits + 5*size;

    /* 从用户空间将输入集拷贝到内核空间 */
    if ((ret = get_fd_set(n, inp, fds.in)) ||
        (ret = get_fd_set(n, outp, fds.out)) ||
        (ret = get_fd_set(n, exp, fds.ex)))
        goto out;
    /* 将结果集清0 */
    zero_fd_set(n, fds.res_in);
    zero_fd_set(n, fds.res_out);
    zero_fd_set(n, fds.res_ex);

    /* 主线（至此，我们只进行了描述符集处理） */
    ret = do_select(n, &fds, end_time);

    if (ret < 0)
        goto out;
    if (!ret) {
        ret = -ERESTARTNOHAND;
        if (signal_pending(current))
            goto out;
        ret = 0;
    }

    /* 最后...将结果集写给用户空间 */
    if (set_fd_set(n, inp, fds.res_in) ||
        set_fd_set(n, outp, fds.res_out) ||
        set_fd_set(n, exp, fds.res_ex))
        ret = -EFAULT;

out:
    if (bits != stack_fds)
        kfree(bits);
out_nofds:
    return ret;
}

/**
 * do_select - select()核心函数
 * @n: select()的第一个参数
 * @fd_set_bits: core_sys_select()处理的描述符集
 * @end_time: 绝对超时时间
 */
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;
    /* timed_out指示是否已经超时，超时1，未超时0 */
    int retval, i, timed_out = 0;
    unsigned long slack = 0;

    /* 根据输入集以及当前文件描述符打开情况对n修正 */
    rcu_read_lock();
    retval = max_select_fd(n, fds);
    rcu_read_unlock();

    if (retval < 0)
        return retval;
    n = retval;

    /* ------------------------------------------------------------- */
    /* ---------------------------- main --------------------------- */
    /* ------------------------------------------------------------- */
    /* 
     * 注意：
     * poll_table被封装在了poll_wqueues结构体中，以便之后向资源
     * 注册监听的时候，能够用poll_table得到对应的poll_wqueues
     *
     * 初始化poll_wqueues
     * 1. 初始化poll_wqueues中的poll_table：
     *      * 设置监听注册函数为__pollwait
     *      * 设置想要监听的事件为所有事件（没必要，之后会修改）
     * 2. 设置polling_task指向当前进程PCB
     *
     * 重点：资源注册函数为__pollwait
     */
    poll_initwait(&table);
    wait = &table.pt;
    /* 根本不等待 */
    if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
        /* 注意：wait设为NULL了！！！ */
        wait = NULL;
        /* 还没开始就已经超时，这样就实现了根本不等待... */
        timed_out = 1;
    }

    /* 重新估算相对超时时间... */
    if (end_time && !timed_out)
        slack = estimate_accuracy(end_time);

    retval = 0;
    for (;;) {
        unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;

        inp = fds->in; outp = fds->out; exp = fds->ex;
        rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;

        /* 通过i遍历[0, n)范围内的文件描述符 */
        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;
            /* 跳过我们不关心的bits段 */
            if (all_bits == 0) {
                i += __NFDBITS;
                continue;
            }

            /* 遍历bits段中的每一个bit */
            for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
                int fput_needed;
                /* 超出了关心的文件描述符范围[0, n)，那么跳出... */
                if (i >= n)
                    break;
                
                /* 跳过我们不关心的bit */
                if (!(bit & all_bits))
                    continue;
                /* 通过current->files->fdt->fd[i]获得当前进程中描述符i
                 * 对应文件的struct file...同时，增加struct file的引用
                 * 计数，防止在获取struct file之后，它被异步删除... */
                file = fget_light(i, &fput_needed);
                /* 因为没有rdlock加锁，因此当前进程中描述符i对应的文件可能已经
                 * 被异步关闭。这就是为什么需要判断file是否为空的原因...
                 */
                if (file) {
                    f_op = file->f_op;
                    /* 注意：
                     * mask = POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM; */
                    mask = DEFAULT_POLLMASK;
                    /*
                     * 如果这个文件支持poll()，那么我们就向这个文件
                     * 注册监听函数；如果不支持，那么我们就忽略掉这
                     * 个文件描述符...
                     *
                     * 还有三种特殊情况：
                     *  1. 我们将超时时间设置为根本不等待
                     *  2. 已经发生了就绪事件
                     *  3. 已经注册了监听函数
                     *  这三种情况下，wait都为NULL，此时poll()只是
                     *  简单地返回资源文件当前状态...
                     */
                    if (f_op && f_op->poll) {
                        /* 设置poll_table中想要监听的事件 */
                        wait_key_set(wait, in, out, bit);
                        /* 对文件注册监听函数，并返回资源的当前状态 */
                        mask = (*f_op->poll)(file, wait);
                    }
                    fput_light(file, fput_needed);
                    /*
                     * 如果文件不支持poll()的话，那么mask总为DEFAULT_POLLMASK，
                     * 表示当前资源对于任何事件都就绪...
                     * 对于不支持poll()的文件，主要是ext2/ext3/ext4等块设备上
                     * 的文件，这些文件总是对任意事件就绪的，读写操作不会引起
                     * 阻塞，有数据就返回数据，没有数据就返回0表示end-of-file
                     */
                    /* events验证，其中retval表示就绪的资源数 */
                    if ((mask & POLLIN_SET) && (in & bit)) {
                        res_in |= bit;
                        retval++;
                        /* 发生了就绪事件后就不再对资源进行注册了...
                         * 因为我们遍历完这次就可以返回了！ */
                        wait = NULL;
                    }
                    if ((mask & POLLOUT_SET) && (out & bit)) {
                        res_out |= bit;
                        retval++;
                        wait = NULL;
                    }
                    if ((mask & POLLEX_SET) && (ex & bit)) {
                        res_ex |= bit;
                        retval++;
                        wait = NULL;
                    }
                }
            } // end of for (j = 0; j < __NFDBITS; ++j)

            /* 写出结果，注意写出的目的地是传进来的fd_set_bits */
            if (res_in)
                *rinp = res_in;
            if (res_out)
                *routp = res_out;
            if (res_ex)
                *rexp = res_ex;
            /* 提供一个调度时机，给资源就绪更多的准备时间... */
            cond_resched();
        } // end of for (i = 0; i < n; )
        /* 将wait设为NULL，表示我们不希望再进行监听注册... */
        wait = NULL;
        /* 
         * 1. 有事件发生了(retval)
         * 2. 超时了(timed_out)
         * 3. 发生了中断(signal_pending(current))
         * 任意事件发生了都跳出死循环...
         */
        /*
         * 感觉这里存在问题：
         * e.g.
         * 假设我们监听文件描述符集#1#2#3#4，poll(#1)与poll(#2)时
         * 向资源文件注册监听函数并返回0 events，poll(#3)时向资源
         * 文件注册监听函数并返回可读events，然后wait被置NULL，
         * poll(#4)时由于wait为NULL，所以只是返回#4的当前文件状态。
         * 遍历描述符集完毕，retval != 0，此时跳出循环，返回用户
         * 空间#3与#4的文件状态。但是在处理后续文件描述符的时候，
         * #1和#2可能异步就绪，但是我们没有返回它们...
         */
        if (retval || timed_out || signal_pending(current))
            break;
        /* 发生了错误，我们也跳出死循环... */
        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;
        }

        /* 能够到达这一步就说明没有发生就绪、中断以及超时 */
        /* 
         * 判断poll_wqueues是否已触发，如果还没有触发，那就设置
         * 当前运行状态为可中断阻塞并进行睡眠，等待被唤醒...
         * 被唤醒之后重新进行迭代，获取资源就绪情况...
         * 在向资源注册监听与判断poll_wqueues是否已触发这段时间
         * 内，可能资源异步就绪了，如果没有触发标志，那么可能就
         * 会丢失资源就绪这个事件，可能导致select()永久沉睡...
         * 这就是为什么需要poll_wqueues.triggered字段的原因...
         */
        if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
                       to, slack))
            timed_out = 1;
    } // end of for ( ; ; )

    /* 
     * 1. 卸载安装到资源监听队列上的poll_table_entry
     * 2. 释放poll_wqueues占用的资源
     */
    poll_freewait(&table);

    return retval;
}

#define FDS_IN(fds, n)      (fds->in + n)
#define FDS_OUT(fds, n)     (fds->out + n)
#define FDS_EX(fds, n)      (fds->ex + n)

#define BITS(fds, n)    (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))

static int max_select_fd(unsigned long n, fd_set_bits *fds)
{
    unsigned long *open_fds;
    unsigned long set;
    int max;
    struct fdtable *fdt;

    /* handle last in-complete long-word first */
    /* 
     * set = ~(1111....11111111 << (n % __NFDBITS))
     * e.g.
     * n = 16, set = (unsigned long)0000....00000000B
     * n = 18, set = (unsigned long)0000....00000011B
     * n = 23, set = (unsigned long)0000....01111111B
     */
    set = ~(~0UL << (n & (__NFDBITS-1)));
    n /= __NFDBITS;
    fdt = files_fdtable(current->files);
    open_fds = fdt->open_fds->fds_bits+n;
    max = 0;

    /* 处理最后一个不完整的bits段 */
    if (set) {
        set &= BITS(fds, n);
        if (set) {
            /* 有效性验证：
             * 输入集中不能含有未打开的文件描述符 */
            if (!(set & ~*open_fds))
                goto get_max;
            return -EBADF;
        }
    }
    /* 循环处理完整的bits段 */
    while (n) {
        open_fds--;
        n--;
        set = BITS(fds, n);
        if (!set)
            continue;
        /* 有效性验证：
         * 输入集中不能含有未打开的文件描述符 */
        if (set & ~*open_fds)
            return -EBADF;
        /* 判断max是否被设置而continue，而不是根据max是否被设置
         * 而break，是想对fd_set中设置的所有文件描述符都进行有效
         * 性验证 */
        if (max)
            continue;
get_max:
        /* max会是__NFDBITS的整数倍 */
        do {
            max++;
            set >>= 1;
        } while (set);
        max += n * __NFDBITS;
    }

    return max;
}

#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
#define POLLEX_SET (POLLPRI)

static inline void wait_key_set(poll_table *wait, unsigned long in,
                unsigned long out, unsigned long bit)
{
    if (wait) {
        wait->key = POLLEX_SET;
        if (in & bit)
            wait->key |= POLLIN_SET;
        if (out & bit)
            wait->key |= POLLOUT_SET;
    }
}

poll

调用链：

sys_poll() -> do_sys_poll() -> poll_initwait()
                            -> do_poll() -> 1. do_pollfd() -> f_op->poll() [ -> poll_wait() -> __pollwait() -> poll_get_entry() ]
                                         -> 2. [ block -> goto 1 ]
                            -> poll_freewait() -> free_poll_entry()

pollwake() -> __pollwake()

struct pollfd {
    /* 想要监听的文件描述符 */
    int fd;
    /* 关心的events */
    short events;
    /* 就绪的events */
    short revents;
};

struct poll_list {
    struct poll_list *next;
    int len;
    struct pollfd entries[0];
};

SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
        long, timeout_msecs)
{
    struct timespec end_time, *to = NULL;
    int ret;

    if (timeout_msecs >= 0) {
        to = &end_time;
        poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
            NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
    }

    /* 主线（至此，我们只进行了超时时间处理） */
    ret = do_sys_poll(ufds, nfds, to);

    /* ... */
    if (ret == -EINTR) {
        struct restart_block *restart_block;

        restart_block = &current_thread_info()->restart_block;
        restart_block->fn = do_restart_poll;
        restart_block->poll.ufds = ufds;
        restart_block->poll.nfds = nfds;

        if (timeout_msecs >= 0) {
            restart_block->poll.tv_sec = end_time.tv_sec;
            restart_block->poll.tv_nsec = end_time.tv_nsec;
            restart_block->poll.has_timeout = 1;
        } else
            restart_block->poll.has_timeout = 0;

        ret = -ERESTART_RESTARTBLOCK;
    }
    return ret;
}

int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
        struct timespec *end_time)
{
    struct poll_wqueues table;
    int err = -EFAULT, fdcount, len, size;
    /* Allocate small arguments on the stack to save memory and be
       faster - use long to make sure the buffer is aligned properly
       on 64 bit archs to avoid unaligned access */
    /* poll_list缓存数组 */
    long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
    struct poll_list *const head = (struct poll_list *)stack_pps;
    struct poll_list *walk = head;
    unsigned long todo = nfds;

    /* 输入验证 */
    if (nfds > rlimit(RLIMIT_NOFILE))
        return -EINVAL;

    /* 将用户空间中的pollfd拷贝到内核空间 */
    len = min_t(unsigned int, nfds, N_STACK_PPS);
    for (;;) {
        walk->next = NULL;
        walk->len = len;
        if (!len)
            break;

        if (copy_from_user(walk->entries, ufds + nfds-todo,
                    sizeof(struct pollfd) * walk->len))
            goto out_fds;

        todo -= walk->len;
        if (!todo)
            break;

        len = min(todo, POLLFD_PER_PAGE);
        size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
        walk = walk->next = kmalloc(size, GFP_KERNEL);
        if (!walk) {
            err = -ENOMEM;
            goto out_fds;
        }
    }

    poll_initwait(&table);
    /* 主线（至此，我们只进行了描述符集处理） */
    fdcount = do_poll(nfds, head, &table, end_time);
    poll_freewait(&table);

    /* 将结果写给用户空间 */
    for (walk = head; walk; walk = walk->next) {
        struct pollfd *fds = walk->entries;
        int j;

        for (j = 0; j < walk->len; j++, ufds++)
            if (__put_user(fds[j].revents, &ufds->revents))
                goto out_fds;
    }

    err = fdcount;
out_fds:
    walk = head->next;
    while (walk) {
        struct poll_list *pos = walk;
        walk = walk->next;
        kfree(pos);
    }

    return err;
}

static int do_poll(unsigned int nfds,  struct poll_list *list,
           struct poll_wqueues *wait, struct timespec *end_time)
{
    poll_table* pt = &wait->pt;
    ktime_t expire, *to = NULL;
    int timed_out = 0, count = 0;
    unsigned long slack = 0;

    /* Optimise the no-wait case */
    if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
        pt = NULL;
        timed_out = 1;
    }

    if (end_time && !timed_out)
        slack = estimate_accuracy(end_time);

    for (;;) {
        struct poll_list *walk;

        /* 遍历poll_list */
        for (walk = list; walk != NULL; walk = walk->next) {
            struct pollfd * pfd, * pfd_end;

            /* 遍历pollfd */
            pfd = walk->entries;
            pfd_end = pfd + walk->len;
            for (; pfd != pfd_end; pfd++) {
                /*
                 * Fish for events. If we found one, record it
                 * and kill the poll_table, so we don't
                 * needlessly register any other waiters after
                 * this. They'll get immediately deregistered
                 * when we break out and return.
                 */
                /* 
                 * 对每一个想要监听的pollfd调用do_pollfd()
                 * 返回非0表示监听的事件已发生...
                 */
                if (do_pollfd(pfd, pt)) {
                    count++;
                    pt = NULL;
                }
            }
        }
        /*
         * All waiters have already been registered, so don't provide
         * a poll_table to them on the next loop iteration.
         */
        pt = NULL;
        if (!count) {
            count = wait->error;
            if (signal_pending(current))
                count = -EINTR;
        }
        if (count || timed_out)
            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 (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
            timed_out = 1;
    }
    return count;
}

/*
 * Fish for pollable events on the pollfd->fd file descriptor. We're only
 * interested in events matching the pollfd->events mask, and the result
 * matching that mask is both recorded in pollfd->revents and returned. The
 * pwait poll_table will be used by the fd-provided poll handler for waiting,
 * if non-NULL.
 */
static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
{
    unsigned int mask;
    int fd;

    mask = 0;
    fd = pollfd->fd;
    if (fd >= 0) {
        int fput_needed;
        struct file * file;

        file = fget_light(fd, &fput_needed);
        mask = POLLNVAL;
        if (file != NULL) {
            mask = DEFAULT_POLLMASK;
            if (file->f_op && file->f_op->poll) {
                if (pwait)
                    /* poll()总是设置events中的POLLERR与POLLHUP */
                    pwait->key = pollfd->events |
                            POLLERR | POLLHUP;

                mask = file->f_op->poll(file, pwait);
            }
            /* Mask out unneeded events. */
            mask &= pollfd->events | POLLERR | POLLHUP;
            fput_light(file, fput_needed);
        }
    }
    pollfd->revents = mask;

    return mask;
}

关于惊群：select()惊群与poll()惊群目前无法等到解决，因为它无法像accept()那样使得每次只有一个调用者能挂载到资源的监听队列上...虽然epoll()已经较好的解决了惊群，但只限于同一个epoll实例的ET模式。