[jboss-jira] [JBoss JIRA] (WFLY-9876) ActiveMQ Artemis with IPv6 and Scope fails

[Jens Popp (JIRA)]

     [ https://issues.jboss.org/browse/WFLY-9876?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jens Popp updated WFLY-9876:
----------------------------
    Description: 
Somehow Java behaves differently on Linux and Windows regarding the IPv6 Network Interfaces. With a small programm you can test this:
{code}
    Enumeration<NetworkInterface> interfaces = NetworkInterface.getNetworkInterfaces();
    while (interfaces.hasMoreElements()) {
        NetworkInterface netInterface = interfaces.nextElement();
        Enumeration<InetAddress> address = netInterface.getInetAddresses();
        while (address.hasMoreElements()) {
            InetAddress addr = address.nextElement();
            if (addr instanceof Inet6Address) {
                Inet6Address inet6addr = (Inet6Address) addr;
                if (!inet6addr.isLinkLocalAddress() && !inet6addr.isLoopbackAddress())
                    System.out.println(netInterface.getName() +":"+ inet6addr.isLinkLocalAddress() +":"+ inet6addr +" - "+ inet6addr.getScopedInterface());

            }
        }
    }
{code}
In windows the scope will only be set for Link Local, in CentOS also for ULA addresses.

If I lookup the JMS from a ULA address (without scope) e.g. with:
{code}
        String JMS_CONNECTION_FACTORY_JNDI = "jms/RemoteConnectionFactory";
        Context context = getApplicationManager().getConnectionHandler().getContext();
        TopicConnectionFactory cf = (TopicConnectionFactory) context.lookup(JMS_CONNECTION_FACTORY_JNDI);
{code}

The TopicConnectionFactory is ActiveMQJMSConnectionFactory. This factory receives a TransportConfiguration from the server that includes the server scope (IPv6 Address with NetworkInterface after % e.g. [fd00::d7dc:b4cc:2e2a:ea1%enp0s3]). Trying to create a topic connection with the TopicConnectionFactory will fail, since the client does not know the scope (enp0s3).

There is a bug fix in NettyConnector (HORNETQ-907) to strip this "%enp0s3" scope, but this BugFix does not work. The fix uses the standard InetAddress methods to parse the host String, which fails, if the host String contains a server scope (network interface) not available on the client. This will only cause an exception, if server and client have different names for the NetworkInterfaces! Testing e.g. on two identical Linux machines will not uncover the problem! 

I think the underlying problem is, that the lookup distributes the server scope in the IPv6 Address. The scope should only be distributed for Link-local Addresses, for all other addresses the scope should be stripped already on the server and not be distributed during lookup.

  was:
Somehow Java behaves differently on Linux and Windows regarding the IPv6 Network Interfaces. With a small programm you can test this:

    Enumeration<NetworkInterface> interfaces = NetworkInterface.getNetworkInterfaces();
    while (interfaces.hasMoreElements()) {
        NetworkInterface netInterface = interfaces.nextElement();
        Enumeration<InetAddress> address = netInterface.getInetAddresses();
        while (address.hasMoreElements()) {
            InetAddress addr = address.nextElement();
            if (addr instanceof Inet6Address) {
                Inet6Address inet6addr = (Inet6Address) addr;
                if (!inet6addr.isLinkLocalAddress() && !inet6addr.isLoopbackAddress())
                    System.out.println(netInterface.getName() +":"+ inet6addr.isLinkLocalAddress() +":"+ inet6addr +" - "+ inet6addr.getScopedInterface());

            }
        }
    }

In windows the scope will only be set for Link Local, in CentOS also for ULA addresses.

If I lookup the JMS from a ULA address (without scope) e.g. with:

        String JMS_CONNECTION_FACTORY_JNDI = "jms/RemoteConnectionFactory";
        Context context = getApplicationManager().getConnectionHandler().getContext();
        TopicConnectionFactory cf = (TopicConnectionFactory) context.lookup(JMS_CONNECTION_FACTORY_JNDI);

The TopicConnectionFactory is ActiveMQJMSConnectionFactory. This factory receives a TransportConfiguration from the server that includes the server scope (IPv6 Address with NetworkInterface after % e.g. [fd00::d7dc:b4cc:2e2a:ea1%enp0s3]). Trying to create a topic connection with the TopicConnectionFactory will fail, since the client does not know the scope (enp0s3).

There is a bug fix in NettyConnector (HORNETQ-907) to strip this "%enp0s3" scope, but this BugFix does not work. The fix uses the standard InetAddress methods to parse the host String, which fails, if the host String contains a server scope (network interface) not available on the client. This will only cause an exception, if server and client have different names for the NetworkInterfaces! Testing e.g. on two identical Linux machines will not uncover the problem! 

I think the underlying problem is, that the lookup distributes the server scope in the IPv6 Address. The scope should only be distributed for Link-local Addresses, for all other addresses the scope should be stripped already on the server and not be distributed during lookup.



> ActiveMQ Artemis with IPv6 and Scope fails
> ------------------------------------------
>
>                 Key: WFLY-9876
>                 URL: https://issues.jboss.org/browse/WFLY-9876
>             Project: WildFly
>          Issue Type: Bug
>          Components: JMS
>    Affects Versions: 10.1.0.Final, 11.0.0.Final
>         Environment: CentOS 7 with IPv6 ULA Address as server for Wildfly 10.1.0Final
> Windows 10 as client
> Both with Java 8u161 (64bit)
>            Reporter: Jens Popp
>            Assignee: Jeff Mesnil
>
> Somehow Java behaves differently on Linux and Windows regarding the IPv6 Network Interfaces. With a small programm you can test this:
> {code}
>     Enumeration<NetworkInterface> interfaces = NetworkInterface.getNetworkInterfaces();
>     while (interfaces.hasMoreElements()) {
>         NetworkInterface netInterface = interfaces.nextElement();
>         Enumeration<InetAddress> address = netInterface.getInetAddresses();
>         while (address.hasMoreElements()) {
>             InetAddress addr = address.nextElement();
>             if (addr instanceof Inet6Address) {
>                 Inet6Address inet6addr = (Inet6Address) addr;
>                 if (!inet6addr.isLinkLocalAddress() && !inet6addr.isLoopbackAddress())
>                     System.out.println(netInterface.getName() +":"+ inet6addr.isLinkLocalAddress() +":"+ inet6addr +" - "+ inet6addr.getScopedInterface());
>             }
>         }
>     }
> {code}
> In windows the scope will only be set for Link Local, in CentOS also for ULA addresses.
> If I lookup the JMS from a ULA address (without scope) e.g. with:
> {code}
>         String JMS_CONNECTION_FACTORY_JNDI = "jms/RemoteConnectionFactory";
>         Context context = getApplicationManager().getConnectionHandler().getContext();
>         TopicConnectionFactory cf = (TopicConnectionFactory) context.lookup(JMS_CONNECTION_FACTORY_JNDI);
> {code}
> The TopicConnectionFactory is ActiveMQJMSConnectionFactory. This factory receives a TransportConfiguration from the server that includes the server scope (IPv6 Address with NetworkInterface after % e.g. [fd00::d7dc:b4cc:2e2a:ea1%enp0s3]). Trying to create a topic connection with the TopicConnectionFactory will fail, since the client does not know the scope (enp0s3).
> There is a bug fix in NettyConnector (HORNETQ-907) to strip this "%enp0s3" scope, but this BugFix does not work. The fix uses the standard InetAddress methods to parse the host String, which fails, if the host String contains a server scope (network interface) not available on the client. This will only cause an exception, if server and client have different names for the NetworkInterfaces! Testing e.g. on two identical Linux machines will not uncover the problem! 
> I think the underlying problem is, that the lookup distributes the server scope in the IPv6 Address. The scope should only be distributed for Link-local Addresses, for all other addresses the scope should be stripped already on the server and not be distributed during lookup.



--
This message was sent by Atlassian JIRA
(v7.5.0#75005)