[take two!]
Tom,
This specific area of code was changed recently due to CR 7035556 [1],
changeset [2], and this issue was discussed during the code review [3].
Essentially, bind() already closes the impl internally before throwing
the exception. Does this resolve the issue for you? Or do you still see
potential to leak?
-Chris
[1] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7035556
[2] http://hg.openjdk.java.net/jdk8/tl/jdk/rev/07a12583d4ea
[3] http://mail.openjdk.java.net/pipermail/net-dev/2011-July/003318.html
On 08/29/11 03:27 PM, Salter, Thomas A wrote:
There appears to be a socket leak in both DatagramSocket and
MulticastSocket constructors. Both classes have constructors that create
a socket and then attempt to bind. The bind can fail with a variety of
exceptions none of which are caught by the constructor. Thus, the actual
system socket that was allocated by impl.create() is never closed.
My fix was to wrap a try-finally around the bind call and call close()
if isBound is false.
public DatagramSocket(SocketAddress bindaddr) throws SocketException {
// create a datagram socket.
createImpl();
if (bindaddr != null) {
try {
bind(bindaddr);
} finally {
if( !isBound() )
close();
}
}
}
Tom Salter
Unisys Corporation
