Commissioning of IPv6 machine results in incorrect /128 subnets

Hi Stephane,

can you share the DHCP configuration with us?

MAAS is operating the DHCPv6 server. It's just a VLAN with MAAS providing DHCPv4 and DHCPv6.
Attached a screenshot of the DHCP config page not that it shows a whole bunch.
Also grabbed the DHCPv6 config from the MAAS server.

i can't seem to reproduce this. is this real hardware, vm? any additional info you can give me? i seem to always get /64 on bolla (which is same rev level as your screenshot).

Affects real hardware and virtual machines.

Router is configured to advertise the subnet (/64) with the managed-config bit set which then correctly sends the machines to perform DHCPv6. MAAS' dhcp server gives them a dynamic address which then for some reason shows up as a /128.

Stephane, could you please provide the network config for one of the machines that are getting the /128 address?

The bad data is added during commissioning so not as a result of a user entered network config in MAAS.

The most recent example I have is of a Raspberry Pi 4 (using a UEFI firmware on it) that has the most simple gigabit network card you can find. It's connected to a regular switch with a single VLAN which goes to a MAAS managed network.

That network uses 2602:fc62:b:1016::/64 on IPv6 and 172.17.16.0/24 on IPv4, both configured to have MAAS operated DHCP without anything fancy in the config. On the IPv6 side, the gateway provides a router advertisement to announce the default gateway and has the "managed" bit set to tell the clients to request an IPv6 address from MAAS' DHCP.

I've attached the wireshark output for a router advertisement. It looks correct to me, it announces the /64 and shows the managed flag as expected.

Also attached is a bunch of screenshots of the MAAS config.

Could you please attach the output for the 50-maas-01-commissioning script for that machine?

Here you go

Any chance you could grab a tcpdump of dhcp traffic during commissioning?

I can't seem to reproduce the issue locally and AFAICS the dhcp6 config looks correct.
I'm not sure why the machine would get a /128 address.

Also could you add a screenshot of the VLANs tab for the MAAS controller?

I finally managed to reproduce the issue locally.

So, after further investigation it seems that the behavior WRT network configuration is actually correct.

MAAS is configured to act as a stateful DHCPv6, and looking at the DHCPv6 exchange, dhcpd6 is correctly offering an address from the dynamic range, which according to the RFC [1] must be applied with a /128 netmask from the host.

We could (should?) probably handle this on the MAAS side and fold those /128 addresses in the corresponding /64.
This already happens (see fix_link_addresses()) for /32 and /128 addresses when the node also reports addresses on the wider subnet, but it doesn't in this case because the machine only reports the /128 address.

[1] https://datatracker.ietf.org/doc/html/rfc8415

hi bill wear,
We are installing maas in ipv6 machine.
MAAS version: 3.2.6

we have provisioned ip address configured in eth0 like below:
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 1c:98:ec:13:17:10 brd ff:ff:ff:ff:ff:ff
    inet6 fd17:22:1::1/64 scope global
       valid_lft forever preferred_lft forever
    inet6 fe80::1e98:ecff:fe13:1710/64 scope link
       valid_lft forever preferred_lft forever

And the subnet (fd17:22:1::1/64) is enlisted in maas console (screenshots attached), it is ipv6 we may not need dhcp but since we do pxe boot it requires a dhcp server. so we enabled dhcp server in maas and setup the ranges.

And also we create radvd.conf : (contents are attached)

When we boot the idrac by using pxe boot, it assigned /128 ip address instead of /64 ip address and that causes the boot process fail (screenshot attached)

Please find all the screenshots as zip file