[Dnsmasq-discuss] DHCPv6 + dhcp-relay + dhcpcd

[Geert Stappers]

On Mon, Sep 01, 2025 at 10:15:13PM +0100, Simon Kelley wrote:
> On 8/26/25 16:03, Petr Menšík via Dnsmasq-discuss wrote:
> > Hello!
> > 
> > I were just debugging one of our old tests, which tested whether dnsmasq
> > --dhcp-relay worked. Both for IPv6 and IPv4 addresses. But that test
> > were written with dhclient originally and worked fined with that.
> > 
> > But you might know ISC dhcp client is deprecated and unsupported. There
> > is a nice alternative still supported, dhcpcd. But it requires on
> > difference and that is presence of Router Advertisement Managed flag
> > set.
> > 
> > dnsmasq has support for --enable-ra to send exactly such advertisements,
> > which will then dhcpcd use to query DHCP server. The problem I see is it
> > requires configured --dhcp-range for the network.
> > 
> > But using --dhcp-relay and --dhcp-range do not make sense on the same
> > network, right?
> 
> Right.
> 
> > Unfortunately dhcpcd just won't even try to send DHCP request to
> > multicast address, until it has indication from the network it should
> > try it.
> > 
> > I think it would make sense to extend --enable-ra to send RA also from
> > local addresses specified by --dhcp-relay address.
> 
> Agreed.
> 
> > Or at least have it explicitly enabled by --ra-param=<device>. The
> > similar problem seems to have --dhcp-range=::,static. That won't be used
> > by dhcpcd also, because it did not match correct network.
> 
> The --dhcp-range=::,static problem, I don't understand. More explanation
> would be good.
> 
> > 
> > Would be patches implementing this desired? Do you know any different
> > option, how to test dnsmasq --dhcp-relay with dhcpcd?
> 
> Definitely will look at patches. I just looked at what needs to be done, and
> came up with a five-line addition that looked good, but then realised that
> all the timer infrastructure supporting unsolicited RAs is predicated on
> having a struct dhcp_context.
> 
> The simplest way to fix that might be to add a struct dhcp_relay * to struct
> dhcp_context and create a context for each relay address with a flag set to
> avoid it's being used anywhere else than in the RA code.
> 
> It's a bit messy. I'll keep working on it unless you'd like to have a go.
> 
> If you're working on testing dhcp relay, maybe have a look at the new
> --dhcp-split-relay option that's in the bleeding edge code. That's IPv4
> only.




This post aims to revive the above discussion.



-- 
Silence is hard to parse