RE: read() & close()

From: Filipau, Ihar (ifilipau@sussdd.de)
Date: Fri Mar 21 2003 - 08:07:34 EST


> From: David Schwartz Sent: Freitag, 21. Marz 2003 13:36
> On Thu, 20 Mar 2003 15:14:52 +0100, Filipau, Ihar wrote:
>
> >I have/had a simple issue with multi-threaded programs:
> >
> >one thread is doing blocking read(fd) or poll({fd}) on file/socket.
> >
> >another thread is doing close(fd).
> >
> >I expected first thread will unblock with some kind
> >of error - but nope! It is blocked!
> >
> >Is it expected behaviour?
>
> It is impossible to make this work reliably, so
> *please* don't do
> that. For example, how can you possibly assure that the first thread
> is actually in 'poll' when call 'close'? There is no atomic 'release
> mutex and poll' function.
>
> So what happens if the system pre-empts the thread
> right before it
> calls 'poll'. Then you call 'close'. Perhaps next a thread started by
> some library function calls 'socket' and gets the file descriptor you
> just 'close'd. Now your call to 'poll' polls on the *wrong* socket!
>
> You simply must accept the fact that you cannot free a
> resource in one thread while another thread is or might be using it.
>

    In my case I was actually trying to use pipe/socket handle
    as a synchronisation point. to signal threads to wake-up at
    programme shut-down.

    But yes - you are right. This approach is prone to errors.

    It looks like in this kind of situations /dev/epoll should be used -
    when set of fds is defined before. and all operations on this set
    can be easily synchronized.
    Sure if /dev/epoll can handle closing of fd correctly - and will
    remove it from set.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Mar 23 2003 - 22:00:36 EST