netdev-owner@vger.kernel.org wrote on 09/08/2008 11:52:05 PM:

On Tue, 9 Sep 2008 02:38:53 +0200, Bernhard Schmidt 

[off-list ref]

wrote:

quoted

ff00::/8 dev teredo  metric 256  mtu 1280 advmss 1220 hoplimit 

4294967295

               ^^^^^^

Uho... that interface is not multicast-capable. Not sure how the kernel

handles the conflicting routes.

        Multicast programs shouldn't rely on the routing
table at all, IMAO. Unicast routing has nothing at all
to do with multicast routing, where a single address
means copying and forwarding to multiple different
segments, and the address has nothing at all to do with
the routing topology.
        That's why the scope_id should be set (and to
the proper index for eth0), but I think either strace
is lying to us here or they're using SO_BINDTODEVICE
or IP_MULTICAST_IF.
        But when it's breaking, I imagine it is ending
up on the wrong interface and thus the EINVAL... I
just don't see how some addresses still work while
others fail at the same time. :-)

                                                +-DLS