EPOLL---Linux Programmer's Manual

sparrow_lee

EPOLL(4)                   Linux Programmer's Manual                  EPOLL(4)
NAME
       epoll - I/O event notification facility
SYNOPSIS
       #include
DESCRIPTION
       epoll  is a variant of poll(2) that can be used either as Edge or Level
       Triggered interface and scales well to large numbers  of  watched  fds.
       Three  system  calls  are  provided to set up and control an epoll set:
       epoll_create(2), epoll_ctl(2), epoll_wait(2).
       An epoll set is connected to a file descriptor  created  by  epoll_cre-
       ate(2).   Interest  for certain file descriptors is then registered via
       epoll_ctl(2).  Finally, the actual wait is started by epoll_wait(2).
NOTES
       The epoll event distribution interface is able to behave both  as  Edge
       Triggered  ( ET ) and Level Triggered ( LT ). The difference between ET
       and LT event distribution mechanism can be described as  follows.  Sup-
       pose that this scenario happens :
       1      The file descriptor that represent the read side of a pipe ( RFD
              ) is added inside the epoll device.
       2      Pipe writer writes 2Kb of data on the write side of the pipe.
       3      A call to epoll_wait(2) is done that will return  RFD  as  ready
              file descriptor.
       4      The pipe reader reads 1Kb of data from RFD.
       5      A call to epoll_wait(2) is done.
       If  the RFD file descriptor has been added to the epoll interface using
       the EPOLLET flag, the call to epoll_wait(2) done in step 5 will  proba-
       bly  hang because of the available data still present in the file input
       buffers and the remote peer might be expecting a response based on  the
       data  it already sent. The reason for this is that Edge Triggered event
       distribution delivers events only when events happens on the  monitored
       file.  So, in step 5 the caller might end up waiting for some data that
       is already present inside the input buffer. In the  above  example,  an
       event on RFD will be generated because of the write done in 2 , and the
       event is consumed in 3.  Since the read operation done in  4  does  not
       consume the whole buffer data, the call to epoll_wait(2) done in step 5
       might lock indefinitely. The epoll interface, when used with the  EPOL-
       LET flag ( Edge Triggered ) should use non-blocking file descriptors to
       avoid having a blocking read or write starve the task that is  handling
       multiple  file  descriptors.  The suggested way to use epoll as an Edge
       Triggered ( EPOLLET ) interface is  below,  and  possible  pitfalls  to
       avoid follow.
       i      with non-blocking file descriptors
       ii     by  going  to  wait  for an event only after read(2) or write(2)
              return EAGAIN
       On the contrary, when used as a Level Triggered interface, epoll is  by
       all means a faster poll(2), and can be used wherever the latter is used
       since it shares the same semantics. Since even with the Edge  Triggered
       epoll  multiple  events  can  be  generated  up on receival of multiple
       chunks of data, the caller has the option to specify  the  EPOLLONESHOT
       flag, to tell epoll to disable the associated file descriptor after the
       receival of an event with epoll_wait(2).  When the EPOLLONESHOT flag is
       specified,  it  is  caller  responsibility to rearm the file descriptor
       using epoll_ctl(2) with EPOLL_CTL_MOD.
EXAMPLE FOR SUGGESTED USAGE
       While the usage of epoll when employed like a Level Triggered interface
       does  have  the  same  semantics  of  poll(2),  an Edge Triggered usage
       requires more clarifiction to avoid stalls  in  the  application  event
       loop.  In this example, listener is a non-blocking socket on which lis-
       ten(2) has been called. The function do_use_fd()  uses  the  new  ready
       file descriptor until EAGAIN is returned by either read(2) or write(2).
       An event driven state machine application should, after having received
       EAGAIN,  record  its  current  state  so  that  at  the  next  call  to
       do_use_fd() it will continue to  read(2)  or  write(2)  from  where  it
       stopped before.
       struct epoll_event ev, *events;
       for(;;) {
           nfds = epoll_wait(kdpfd, events, maxevents, -1);
           for(n = 0; n

本文来自ChinaUnix博客，如果查看原文请点：http://blog.chinaunix.net/u1/49071/showart_401448.html