Re: CAJ "AlreadyBoundException"

[Till Straumann <strauman@slac.stanford.edu>]

I recently ported JCA/CAJ to android and came across the fact that
BroadcastConnector does not bind the underlying DatagramSocket
to a local address.

According to the java.nio.channels.DatagramChannel docs a channel
is open()ed and then the underlying socket can be bound. Under android
it apparently must be. openjdk is more forgiving and binds the socket
automatically. However, there might be a problem:

For openjdk I came across the following source code

http://cr.openjdk.java.net/~dxu/8000404/webrev/src/share/classes/sun/nio/ch/DatagramChannelImpl.java.html

which happens to match the lines in your stack trace. In this class (I 
have no
other reason to believe it is the same implementation you are using than
two matching line numbers) the socket is bound to a local address 
automatically
(if the user has not done so already).

At first glance there seems to be a race condition which could explain what
you experience:

receive(ByteBuffer dst) [starting at line 317] contains the following code:

   synchronized( readLock ) {
      ...
      if ( localAddress() == null ) {
          bind(null);
      }
.     ...
   }

and bind(SocketAddress local) [starting at line 654] is coded:

  synchronized( readLock ) {
    synchronized ( writeLock ) {
        if ( localAddress != null ) {
           throw new AlreadyBoundException();
        }
        // do actual binding here
        localAddress = xxx;
    }
  }

However, in send(ByteBuffer src, SocketAddress target) [starting at line 
413] we find

  synchronized ( writeLock ) {
        ...
        if ( localAddress == null ) {
          localAddress = Net.localAddress( fd );
        }
  }


Hence, it could occur that  thread A takes the read lock, tests 
localAddress and
then proceeds to bind. It finds the writeLock held by thread B which - 
in the middle
of 'send' - sets 'localAddress'. When thread A resumes and obtains 
'writeLock'
then it finds 'localAddress' non-null and throws the exception.

I don't know how likely this is (do you see this often?) - there might 
be other scenarios
or you might not even use the implementation I looked at (I didn't spend 
more than 1h
on this analysis).

In any case it is IMHO worthwhile to fix CAJ so that it binds datagram 
channels
at creation time:

--- a/src/com/cosylab/epics/caj/impl/BroadcastConnector.java Fri Nov 08 
11:26:19 2013 -0800
+++ b/src/com/cosylab/epics/caj/impl/BroadcastConnector.java Fri Nov 08 
11:29:22 2013 -0800
@@ -58,6 +58,9 @@

             // use non-blocking channel (no need for soTimeout)
             socket.configureBlocking(false);
+
+            // bind to any local port
+            socket.socket().bind( null );

             // set SO_BROADCAST
             socket.socket().setBroadcast(true);


HTH
-- Till

On 12/04/2013 06:48 PM, Kasemir, Kay wrote:
> Hello:
>
> Does anybody else see these messages from CAJ, the pure java channel access client?
>
> Thanks,
> Kay
>
> 2013-12-04 14:44:27.667 SEVERE [Thread 25] com.cosylab.epics.caj.impl.reactor.Reactor (processInternal) -
> java.nio.channels.AlreadyBoundException
>          at sun.nio.ch.DatagramChannelImpl.bind(DatagramChannelImpl.java:661)
>          at sun.nio.ch.DatagramChannelImpl.receive(DatagramChannelImpl.java:327)
>          at com.cosylab.epics.caj.impl.BroadcastTransport.processRead(BroadcastTransport.java:178)
>          at com.cosylab.epics.caj.impl.BroadcastTransport.handleEvent(BroadcastTransport.java:155)
>          at com.cosylab.epics.caj.impl.reactor.lf.LeaderFollowersHandler.handleEvent(LeaderFollowersHandler.java:77)
>          at com.cosylab.epics.caj.impl.reactor.Reactor.processInternal(Reactor.java:400)
>          at com.cosylab.epics.caj.impl.reactor.Reactor.process(Reactor.java:284)
>          at com.cosylab.epics.caj.CAJContext$2.run(CAJContext.java:708)
>          at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
>          at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
>          at java.lang.Thread.run(Thread.java:722)