Alan Bateman wrote:
Christopher Hegarty -Sun Microsystems Ireland wrote:
Hi Pavel, Michael,
6887364: SetOutgoingIf.java fails if run on multihomed machine without
PIv6 on all interfaces [more details below]
Webrev:
http://cr.openjdk.java.net/~chegar/6887364/webrev/
I just realized why this test was failing on one of our lab machines.
The reason is that the test tries to join both IPv4 and IPv6 multicast
groups on network interfaces that are not loopback, are Up, and
support multicast. On our machine we have one interface that has only
an IPv4 address assigned to it, while another has both IPv4 and Ipv6
addresses.
I've changed the test to only try joining the appropriate multicast
group if the network interface supports that IP protocol.
Also, I noticed that the IPv4 mapped IPv6 address the test runs with
actually gets converted to a regular Inet4Address because the
InetAddress factory method is smarter than us! It never creates
Inet6Address instances that are IPv4 mapped IPv6 addresses, it just
creates Inet4Address. So (controversially) I manufactured an
Inet6Address multicast instance manually. It seems to work fine.
--------------
Details of failure and machine configuration:
Fails with:
Exception in thread "main" java.net.SocketException: No such device
or address
at java.net.PlainDatagramSocketImpl.join(Native Method)
at
java.net.AbstractPlainDatagramSocketImpl.joinGroup(AbstractPlainDatagramSocketImpl.java:184)
at java.net.MulticastSocket.joinGroup(MulticastSocket.java:382)
at SetOutgoingIf.main(SetOutgoingIf.java:116)
Machine configuration [IP address information removed for security
reasons]:
:> ifconfig -a
lo0: flags=2001000849<UP,LOOPBACK,RUNNING,MULTICAST,IPv4,VIRTUAL> mtu
8232 index 1
inet 127.0.0.1 netmask ff000000
bge0: flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index 2
inet XXX.XXX.XXX.160 netmask ffffff00 broadcast XXX.XXX.XXX.255
bge3: flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index 5
inet XXX.XXX.YYY.160 netmask ffffff00 broadcast XXX.XXX.YYY.255
bge0: flags=2000841<UP,RUNNING,MULTICAST,IPv6> mtu 1500 index 2
inet6 fe80::214:4fff:fe7a:b734/10
lo0: flags=2002000849<UP,LOOPBACK,RUNNING,MULTICAST,IPv6,VIRTUAL> mtu
8252 index 1
inet6 ::1/128
---------------
-Chris.
I don't know if you have a reviewer for this one yet. I went through the
changes to the test and the overall approach seems reasonable. The one
thing that doesn't seem right is attempting to do IPv4 multicasting when
an interface is only plumbed with IPv6 (I'm talking about the case where
it adds an IPv4-mapped address to the list of groups). I don't think the
RFC dealing with migration to IPv6 covers that scenario and I'm pretty
sure that many kernels don't support it either. I would suggest dropping
that code so that you simple test IPv4 multicast on interfaces that have
IPv4 address, and test IPv6 multicasting on interfaces that have IPv6
addresses.
I was coming around to this conclusion also. I'll just drop the IPv4
mapped IPv6 address and update the webrev.
Thanks for the review.
-Chris.
-Alan.
