KQUEUE(2) FreeBSD System Calls Manual KQUEUE(2)

 
NAME
     kqueue, kevent -- kernel event notification mechanism 内核事件通知机制
LIBRARY
     Standard C Library (libc, -lc)
SYNOPSIS
     #include <sys/types.h>
     #include <sys/event.h>
     #include <sys/time.h>

     int
     kqueue(void);

     int
     kevent(int kq, const struct kevent *changelist, int nchanges,
  struct kevent *eventlist, int nevents,
  const struct timespec *timeout);

     EV_SET(_kev, ident, filter, flags, fflags, data, udata);

DESCRIPTION
     The kqueue() system call provides a generic method of notifying the user
     when an event happens or a condition holds, based on the results of small
     pieces of kernel code termed filters.  A kevent is identified by the
     (ident, filter) pair; there may only be one unique kevent per kqueue.

     系统调用 kqueue() 函数 提供了一个 当内核事件发生或莫个条件被满足时通知用户
     的方法，它是基于一小片内核代码的条件过滤器的结果，一个事件（Kevent）被一对（标识，
     过滤条件）所定义；每一个队列（kqueue）的事件（Kevent）都是唯一的。
  

     The filter is executed upon the initial registration of a kevent in order
     to detect whether a preexisting condition is present, and is also exe-
     cuted whenever an event is passed to the filter for evaluation.  If the
     filter determines that the condition should be reported, then the kevent
     is placed on the kqueue for the user to retrieve.

     过滤器被执行在一个初始化并注册的事件之上，是为了发现一个预置的条件是否满足
     并且事件随时被赋值到过滤器上被执行，如果过滤器确定条件被满足，事件（kevent）
     将被放到队列中（kqueue ），用户通过队列（kqueue ）得到它

     The filter is also run when the user attempts to retrieve the kevent from
     the kqueue.  If the filter indicates that the condition that triggered
     the event no longer holds, the kevent is removed from the kqueue and is
     not returned.

     当用户尝试从队列（kqueue）返回事件（kevent）时，过滤器依然起作用，如果
     过滤器需要触发事件的条件不在被保持，事件将被从队列中移出并不在被放回

     Multiple events which trigger the filter do not result in multiple
     kevents being placed on the kqueue; instead, the filter will aggregate
     the events into a single struct kevent.  Calling close() on a file
     descriptor will remove any kevents that reference the descriptor.
     过滤器不支持多重事件被设置到队列中，替代的办法是，它将允许聚合起来的事件
     放到一个结构中，通过调用 文件描述符上的close函数 移出事件描述符。

     The kqueue() system call creates a new kernel event queue and returns a
     descriptor.  The queue is not inherited by a child created with fork(2).
     However, if rfork(2) is called without the RFFDG flag, then the descrip-
     tor table is shared, which will allow sharing of the kqueue between two
     processes.
     系统调用kqueue 函数创建个一个新的内核事件队列，并且返回一个描述符，队列
     不能够被fork函数创建的子进程继承，然而，如果用rfork函数，同时没有设置
     RFFDG标志，描述符表（因该指进程）被共享，将允许两个进程共享队列（kqueue ）

     The kevent() system call is used to register events with the queue, and
     return any pending events to the user.  The changelist argument is a
     pointer to an array of kevent structures, as defined in <sys/event.h>.
     All changes contained in the changelist are applied before any pending
     events are read from the queue.  The nchanges argument gives the size of
     changelist.  The eventlist argument is a pointer to an array of kevent
     structures.  The nevents argument determines the size of eventlist.  When
     nevents is zero, kevent() will return immediately even if there is a
     timeout specified unlike select(2).  If timeout is a non-NULL pointer, it
     specifies a maximum interval to wait for an event, which will be inter-
     preted as a struct timespec.  If timeout is a NULL pointer, kevent()
     waits indefinitely.  To effect a poll, the timeout argument should be
     non-NULL, pointing to a zero-valued timespec structure.  The same array
     may be used for the changelist and eventlist.
     系统调用（）kevent函数 用于注册事件到队列，并且返回任何挂起的事件给用户
     changelist参数是指向事件结构的队列，它的定义在<sys/event.h>。所有的变化
     包含在changelist被被应用在任何未决事件被从队列读出之前。nchanges参数是
     changelist的大小。eventlist参数是一个指向内核事件（kevent）结构的数组指
     针。nevents参数是eventlist的大小。当nevents它为零时，即使他们是不同的超时
     指定kevent函数立即返回。如果timeout为指定为非NULL，指定的是一个等待事件的
     最大的时间间隔，它将被定义为timespec结构。如果timeout参数设置为NULL，kevent
     的超时将不确定。影响poll，timeout参数将被设置成非NULL，指向一个零值的一个
     timespec，changelist和eventlist两个数组将被设置为一样。

     The EV_SET() macro is provided for ease of initializing a kevent struc-
     ture.
     EV_SET 宏是为了更容易的初始化内核世界结构而提供的

     The kevent structure is defined as: //内核事件结构的定义
     struct kevent {
      uintptr_t ident;      /* identifier for this event */ 
      short     filter;      /* filter for event */
      u_short   flags;      /* action flags for kqueue */
      u_int     fflags;      /* filter flag value */
      intptr_t  data;      /* filter data value */
      void      *udata;      /* opaque user data identifier */
     };

     The fields of struct kevent are:

     ident Value used to identify this event.  The exact interpretation
  is determined by the attached filter, but often is a file
  descriptor.
  ident的值用于标识这个事件。准确的解释是ident的值是通过附加的过滤器
  被确定的，但经常指的是一个文件描述符。

     filter Identifies the kernel filter used to process this event.  The
  pre-defined system filters are described below.
  filter定义了用于处理这个事件的内核过滤器。预定义系统的过滤器被下面描述。
     flags Actions to perform on the event.

     fflags Filter-specific flags.
     fflags 过滤器指定的标志  

     data Filter-specific data value.
     data 过滤器指定的数据值 

     udata Opaque user-defined value passed through the kernel unchanged.
     udata 用户定义的值，穿透整个过程，内核不去改变，对内核不透明。

     The flags field can contain the following values:
     flags可以包含一下的值。

     EV_ADD     Adds the event to the kqueue.  Re-adding an existing event
      will modify the parameters of the original event, and not
      result in a duplicate entry.  Adding an event automati-
      cally enables it, unless overridden by the EV_DISABLE
      flag.
      添加事件到队列。重复添加已存在的事件，将修改最初事件的参数，并且
      在重复入口里不返回。增加的函数被自动的启用，如果不设置EV_DISABLE
      标志
     

     EV_ENABLE     Permit kevent() to return the event if it is triggered.
                    如果事件被触发，kevent 函数将返回事件。

     EV_DISABLE     Disable the event so kevent() will not return it.  The
      filter itself is not disabled.
             将使得事件失效，kevent 函数将返回它，但过滤器不会失效。

     EV_DELETE     Removes the event from the kqueue. Events which are
      attached to file descriptors are automatically deleted on
      the last close of the descriptor.
      从内核队列（kqueue）中移出事件。在描述符关闭后被增加到文件描述
      符的事件将自动被删除。
     

     EV_ONESHOT     Causes the event to return only the first occurrence of
      the filter being triggered.  After the user retrieves the
      event from the kqueue, it is deleted.
      导致过滤器只返回过滤器被触发的第一个发生的事件。用户从队列（kqueue）
      得到事件后，事件将被删除。
     

     EV_CLEAR     After the event is retrieved by the user, its state is
      reset.  This is useful for filters which report state
      transitions instead of the current state.  Note that some
      filters may automatically set this flag internally.
      事件被用户取得后，它的状态将被重置。它被用于发布状态代替当前
      状态。注意：一些过滤器可能制动的被设置这个标志。

     EV_EOF     Filters may set this flag to indicate filter-specific EOF
      condition.
         过滤器可以设置这个标志去标识过滤器队列的结尾位置。

     EV_ERROR     See RETURN VALUES below.
                    请看下面的 返回值。

     The predefined system filters are listed below.  Arguments may be passed
     to and from the filter via the fflags and data fields in the kevent
     structure.
     系统预先定义的过滤器想在下面被列出来。参数可以经过过滤器在kevent结构里
     的fflags和data参数传进和传出。

     EVFILT_READ    Takes a descriptor as the identifier, and returns whenever
      there is data available to read.  The behavior of the fil-
      ter is slightly different depending on the descriptor
      type.

      Sockets
   Sockets which have previously been passed to listen()
   return when there is an incoming connection pending.
   data contains the size of the listen backlog.

   Other socket descriptors return when there is data to
   be read, subject to the SO_RCVLOWAT value of the
   socket buffer. This may be overridden with a per-fil-
   ter low water mark at the time the filter is added by
   setting the NOTE_LOWAT flag in fflags, and specifying
   the new low water mark in data.  On return, data con-
   tains the number of bytes of protocol data available
   to read.

   If the read direction of the socket has shutdown, then
   the filter also sets EV_EOF in flags, and returns the
   socket error (if any) in fflags.  It is possible for
   EOF to be returned (indicating the connection is gone)
   while there is still data pending in the socket
   buffer.

      Vnodes
   Returns when the file pointer is not at the end of
   file.  data contains the offset from current position
   to end of file, and may be negative.

      Fifos, Pipes
   Returns when the there is data to read; data contains
   the number of bytes available.

   When the last writer disconnects, the filter will set
   EV_EOF in flags.  This may be cleared by passing in
   EV_CLEAR, at which point the filter will resume wait-
   ing for data to become available before returning.

      BPF devices
   Returns when the BPF buffer is full, the BPF timeout
   has expired, or when the BPF has ``immediate mode''
   enabled and there is any data to read; data contains
   the number of bytes available.

     EVFILT_WRITE   Takes a descriptor as the identifier, and returns whenever
      it is possible to write to the descriptor. For sockets,
      pipes and fifos, data will contain the amount of space
      remaining in the write buffer.  The filter will set EV_EOF
      when the reader disconnects, and for the fifo case, this
      may be cleared by use of EV_CLEAR. Note that this filter
      is not supported for vnodes or BPF devices.

      For sockets, the low water mark and socket error handling
      is identical to the EVFILT_READ case.

     EVFILT_AIO     The sigevent portion of the AIO request is filled in, with
      sigev_notify_kqueue containing the descriptor of the
      kqueue that the event should be attached to, sigev_value
      containing the udata value, and sigev_notify set to
      SIGEV_KEVENT.  When the aio_*() system call is made, the
      event will be registered with the specified kqueue, and
      the ident argument set to the struct aiocb returned by the
      aio_*() system call.  The filter returns under the same
      conditions as aio_error.

     EVFILT_VNODE   Takes a file descriptor as the identifier and the events
      to watch for in fflags, and returns when one or more of
      the requested events occurs on the descriptor.  The events
      to monitor are:

      NOTE_DELETE    The unlink() system call was called on the
       file referenced by the descriptor.

      NOTE_WRITE    A write occurred on the file referenced by
       the descriptor.

      NOTE_EXTEND    The file referenced by the descriptor was
       extended.

      NOTE_ATTRIB    The file referenced by the descriptor had
       its attributes changed.

      NOTE_LINK    The link count on the file changed.

      NOTE_RENAME    The file referenced by the descriptor was
       renamed.

      NOTE_REVOKE    Access to the file was revoked via
       revoke(2) or the underlying file system was
       unmounted.

      On return, fflags contains the events which triggered the
      filter.

     EVFILT_PROC    Takes the process ID to monitor as the identifier and the
      events to watch for in fflags, and returns when the
      process performs one or more of the requested events.  If
      a process can normally see another process, it can attach
      an event to it.  The events to monitor are:

      NOTE_EXIT      The process has exited.

      NOTE_FORK      The process has called fork().

      NOTE_EXEC      The process has executed a new process
         via execve(2) or similar call.

      NOTE_TRACK      Follow a process across fork() calls.
         The parent process will return with
         NOTE_TRACK set in the fflags field, while
         the child process will return with
         NOTE_CHILD set in fflags and the parent
         PID in data.

      NOTE_TRACKERR    This flag is returned if the system was
         unable to attach an event to the child
         process, usually due to resource limita-
         tions.

      On return, fflags contains the events which triggered the
      filter.

     EVFILT_SIGNAL  Takes the signal number to monitor as the identifier and
      returns when the given signal is delivered to the process.
      This coexists with the signal() and sigaction() facili-
      ties, and has a lower precedence.  The filter will record
      all attempts to deliver a signal to a process, even if the
      signal has been marked as SIG_IGN. Event notification
      happens after normal signal delivery processing.  data
      returns the number of times the signal has occurred since
      the last call to kevent(). This filter automatically sets
      the EV_CLEAR flag internally.

     EVFILT_TIMER   Establishes an arbitrary timer identified by ident.  When
      adding a timer, data specifies the timeout period in mil-
      liseconds. The timer will be periodic unless EV_ONESHOT
      is specified.  On return, data contains the number of
      times the timeout has expired since the last call to
      kevent().  This filter automatically sets the EV_CLEAR
      flag internally.  There is a system wide limit on the num-
      ber of timers which is controlled by the
      kern.kq_calloutmax sysctl.

     EVFILT_NETDEV  Takes a descriptor to a network interface as the identi-
      fier, and the events to watch for in fflags.  It returns,
      when one or more of the requested events occur on the
      descriptor.  The events to monitor are:

      NOTE_LINKUP      The link is up.

      NOTE_LINKDOWN    The link is down.

      NOTE_LINKINV     The link state is invalid.

      On return, fflags contains the events which triggered the
      filter.

RETURN VALUES
     The kqueue() system call creates a new kernel event queue and returns a
     file descriptor.  If there was an error creating the kernel event queue,
     a value of -1 is returned and errno set.

     The kevent() system call returns the number of events placed in the
     eventlist, up to the value given by nevents.  If an error occurs while
     processing an element of the changelist and there is enough room in the
     eventlist, then the event will be placed in the eventlist with EV_ERROR
     set in flags and the system error in data.  Otherwise, -1 will be
     returned, and errno will be set to indicate the error condition.  If the
     time limit expires, then kevent() returns 0.

ERRORS
     The kqueue() system call fails if:

     [ENOMEM]  The kernel failed to allocate enough memory for the
   kernel queue.

     [EMFILE]  The per-process descriptor table is full.

     [ENFILE]  The system file table is full.

     The kevent() system call fails if:

     [EACCES]  The process does not have permission to register a
   filter.

     [EFAULT]  There was an error reading or writing the kevent
   structure.

     [EBADF]  The specified descriptor is invalid.

     [EINTR]  A signal was delivered before the timeout expired and
   before any events were placed on the kqueue for
   return.

     [EINVAL]  The specified time limit or filter is invalid.

     [ENOENT]  The event could not be found to be modified or
   deleted.

     [ENOMEM]  No memory was available to register the event or, in
   the special case of a timer, the maximum number of
   timers has been exceeded.  This maximum is config-
   urable via the kern.kq_calloutmax sysctl.

     [ESRCH]  The specified process to attach to does not exist.

SEE ALSO
     aio_error(2), aio_read(2), aio_return(2), poll(2), read(2), select(2),
     sigaction(2), write(2), signal(3)

HISTORY
     The kqueue() and kevent() system calls first appeared in FreeBSD 4.1.

AUTHORS
     The kqueue() system and this manual page were written by Jonathan Lemon
     <jlemon@FreeBSD.org>.

BUGS
     It is currently not possible to watch a vnode(9) that resides on anything
     but a UFS file system.

     The EVFILT_NETDEV filter is currently only implemented for devices that
     use the miibus(4) driver for LINKUP and LINKDOWN operations.  Therefore,
     it will not work with many non-ethernet devices.

     The timeout value is limited to 24 hours; longer timeouts will be
     silently reinterpreted as 24 hours.

FreeBSD 4.11         September 2, 2006    FreeBSD 4.11