Comment 4 for bug 1573475

The bug breaks any reasonable use of ndppd if the initial description is correct, it certainly breaks the setup I'm currently evaluating. I'm using a VM to test a networking setup for an unrelated project, thus it's more complicated than it'd be on real hardware.

My configuration uses ndppd to establish multiple IPv6 subnets in place of the usual /64 ISP assignment for LXD containers, while the ISP doesn't actually delegate the network to my host.

The VirtualBox VM environment is set up as follows:
- 1 ethernet adapter configured as a bridge with my physical nic, there's a DHCP server on the same network
- 1 ethernet adapter on a host only network, ip fc00:1234::ffff:ffff:ffff:ffff/64

The VM itself is configured as:
- Ubuntu 16.04 server x64
- The bridged nic (enp0s3) is configured to use dhcp and practically only used for apt
- The host only nic (enp0s8) is simulating the actual network of interest
- The nic veth1234 is connecting to the LXD container (p2p, set up by LXD)
- net.ipv6.conf.all.forwarding=1 in /etc/sysctl.conf
- LXD, bridging disabled
- ndppd installed
/etc/network/interfaces:
auto lo
iface lo inet loopback

auto enp0s3
iface enp0s3 inet dhcp

auto enp0s8
iface enp0s8 inet6 static
        address fc00:1234::1/127

        post-up /sbin/ip -6 route add fc00:1234::ffff:ffff:ffff:ffff dev enp0s8
        pre-down /sbin/ip -6 route del fc00:1234::ffff:ffff:ffff:ffff dev enp0s8

allow-hotplug veth1234
iface veth1234 inet6 static
        address fc00:1234::5:ffff/112
/etc/ndppd.conf:
route-ttl 30000

proxy enp0s8 {
 router yes
 timeout 500
 ttl 30000

 rule fc00:1234::1/64 {
  auto
 }
}

The LXD container is configured as:
- name test
- Image source images:ubuntu/xenial/amd64
- default nic device removed
- new nic device: lxc config device add test veth0 nic nictype=p2p name=veth0 host_name=veth1234
/etc/network/interfaces:
auto lo
iface lo inet loopback

auto veth0
iface veth0 inet6 static
 address fc00:1234::5:1/112
 gateway fc00:1234::5:ffff
 dns-nameserver 2001:4860:4860::8888 2001:4860:4860::8844

With Xenial's ndppd 0.2.4 it won't be possible to ping the LXD container from the host, only from the VM or LXD container):
$ ping6 fc00:1234::5:1
PING fc00:1234::5:1(fc00:1234::5:1) 56 data bytes
From fc00:1234::ffff:ffff:ffff:ffff icmp_seq=1 Destination unreachable: Address unreachable

Installing the ndppd 0.2.5 package from the Yakkety repo on the VM fixes it:
$ ping6 fc00:1234::5:1
PING fc00:1234::5:1(fc00:1234::5:1) 56 data bytes
64 bytes from fc00:1234::5:1: icmp_seq=1 ttl=63 time=0.815 ms

By observing the traffic in Wireshark it can be seen that the VM host only nic receives neighbor solicitation requests, but nothing responds - ndppd doesn't pick them up.

The regression risk should be zero as the 16.04 package appears to be unusable in its current form.