Hi Bastian, TL;DR: I think it's actually https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1072526
On Sun, Aug 09, 2026 at 08:34:35AM +0200, Bastian Blank wrote: > On Sun, Aug 09, 2026 at 04:52:11AM +0000, Andy Smith wrote: > > I'm trying to install Debian 13 on a system that (initially) only has > > IPv6 connectivity which it has to get from DHCP. There is no SLAAC nor > > Router Advertisements. > > DHCPv6 is not expected to work without RA. An RA tells systems to use > DHCPv6, via the M or O flags, I don't think that is the case. There isn't any MUSTs for RAs in any of the DHCPv6 RFCs and in RFC 9915 section 18 signalling by flag M or O is only one of the scenarios in which DHCPv6 might be initiated. The client deciding to do so is a valid configuration and in fact d-i does do this. > and provides the router address. Gateway of course can't come from DHCPv6 but it can come from router solicitation; d-i does currently do an rdisc6 to get my correct gateway of fe80::1 before it attempts DHCPv6. d-i is in fact working with DHCPv6 past the point where the M flag and gateway would be an issue, which is why I observe this to work up until host name and domain name questions are answered. Then netcfg kills the dhcp6c client process with SIGTERM, causing its signal handler to do a DHCP RELEASE and d-i never asks for a lease again. Still, we did wonder if it might make a difference so we enabled RAs with M flag in them. It did not make a difference. I'll follow up in #1072526 but essentially I can reproduce the same issue there and the patch or manual workarounds in that bug do resolve this problem for me, so I'm fairly confident it's that bug. Thanks, A Andy
