Weird IPv6 behavior, running in Access-Point mode

Hi All
I'm running OpenWrt 25.12.4 (and before 25.12.2) on a Powerline Kit.
I don't use IPv6 yet internally, as my ISP doesn't provide IPv6. The Powerline is setup with one bridged LAN interface, over all ports, wired and wireless. There is no WAN interface. This is working fine, but I have one weird issue with IPv6 now in my internal network.
I've disabled on the bridge interface all IPv6 options. I went also through all other menus and disabled IPv6 where I found something.
My clients in my network now receive/find/use the IPv6 address of the Powerline adapter as a DNS server and have started to send DNS queries to it, which he faithfully answers with AAAA not found (as he has no valid IPv6 WAN configuration). The learned IPv6 address on my clients is a fe80:: one.
Is there somewhere a hidden IPv6 config that I might have overlooked?
Thanks
pato

Google for: "IPv6 address classes Global, LLA, ULA". You have IPv6 adresses even if your ISP doesn't provide globally routed IPv6 prefix. Those adresses for LAN use only. In "Global network options" tab of your OpenWrt router you can find (and change) ULA prefix which is automatically generated during first booting.

Look at what Timur said.

fe80::/10 is link-local address (LLA), it's locally generated on your device regardless of router settings. It does not hurt anything and I would argue it's one of nice additios of IPv6 -- you can talk to local devices without much of a setup point to point.

Then there are ULAs -- unique local addresses are from fc00::/7, router like OpenWrt generates random /48 from this range. These are local addresses akin to 10.0.0.0/8 or 192.168.0.0/16. Again hurt nothing usually. (EDIT: also this is probably enabled by default on OpenWrt and works regardless of ISPs setup).

Also why disable IPv6? Please don't.

Then the powerline must also advertise itself as DNS server, either via DHCPv6 or via RA.

Have you disabled dnsmasq/odhcpd/firewall on the AP?
You are trying to run OpenWrt as "dump AP", right?

My only guess is that odhcpd is running in server mode on your interface serving DHCPv6 and RA.

I'll post the full config later, but yes, it's running as a dumb AP.
I have disabled DHCPv6 and RA in the LUCI and also couldn't find any remains in the CLI, which is the part that confuses me. I don't understand why my LAN clients learn the fe80:: IP of the OpenWRT dumb AP as a DNS server.
Maybe there is some old stuff remaining when I upgraded from 17.x (initial version to switch to OpenWRT) to 25.x that wasn't cleanly replaced.

Try:

for i in firewall dnsmasq odhcpd; do
  if /etc/init.d/"$i" enabled; then
    /etc/init.d/"$i" disable
    /etc/init.d/"$i" stop
  fi
done

And then maybe try tcpdump/wireguard to look at dhcp/ra packets.
EDIT: something like:

tcpdump -i eth0 -n -vv "udp port 546 or udp port 547 or (icmp6 and (ip6[40] == 133 or ip6[40] == 134))"

EDIT2: or maybe replace -vv with -w ipv6_dhcp_ra.pcap so you can open it in wireshark later to analyze it better.

Thanks all!
It seems it was a remainder in the configuration and more importantly, me not realizing that I have to "ipconfig /release6" instead of "ipconfig /release" to actually get rid of the learned IPv6 config, which even survived a reboot of Windows 11.