IPv6 router advertisement

[Paul Selkirk]

Mihani -

I've confirmed the RA behavior.  radvd is perfectly happy to advertise
a /96 prefix with the A bit set, but the client won't auto-configure
an address.  RFC 4862 doesn't put a limit on the prefix length, but
dmesg reports "IPv6 autoconf: prefix with wrong length 96" (Linux
kernel 2.6.25.17).  Do you feel like hacking the kernel?

I guess it goes back to the question of why you want to do this.

				paul

> Date: Tue, 05 May 2009 19:53:25 +0200
> From: Mihai Moldovan <ionic at ionic.de>
> 
> Sorry for not responding so long, I've been putting v6 aside for a while
> because of compilation and configuration problems and thought it might
> be cool to get it finally working, until...
> 
> 
> * On 20.04.2009 19:37, David W. Hankins wrote:
> > On Sun, Apr 19, 2009 at 09:38:54AM +0200, Mihai Moldovan wrote:
> >  =20
> >> radvd is due to the small subnet size not working of course, so not
> >> employable.
> >>    =20
> >
> > I thought radvd can announce the prefix with the SLAAC bit cleared?
> >
> > The currently (deployed) IPv6 model is that you MUST have an RA that
> > delivers the default router and locally attached prefix information.
> >
> > So working out how to get radvd to emit what you want is the key.
> >  =20
> 
> Currently, my radvd is set up this way: (information acquired by
> radvdump on one machine which is of course not the server/router itself)
> 
> #
> # radvd configuration generated by radvdump 1.1
> # based on Router Advertisement from fe80::21b:21ff:fe2a:c4fe
> # received by interface eth0
> #
> 
> interface eth0
> {
>     AdvSendAdvert on;
>     # Note: {Min,Max}RtrAdvInterval cannot be obtained with radvdump
>     AdvManagedFlag off;
>     AdvOtherConfigFlag off;
>     AdvReachableTime 0;
>     AdvRetransTimer 0;
>     AdvCurHopLimit 64;
>     AdvDefaultLifetime 30;
>     AdvHomeAgentFlag off;
>     AdvDefaultPreference low;
>     AdvSourceLLAddress on;
> 
>     prefix 2001:41d0:1:678e:abcd::/96
>     {
>         AdvValidLifetime 2592000;
>         AdvPreferredLifetime 604800;
>         AdvOnLink on;
>         AdvAutonomous on;
>         AdvRouterAddr off;
>     }; # End of prefix definition
> 
> }; # End of interface definition
> 
> 
> Also, the DHCP server is set up to give out one fixed address out of
> this very /96 pool. (ISC DHCPd of course! :))
> 
> subnet6 2001:41D0:1:678E:ABCD::/96 {
>   default-lease-time        3600;
>   max-lease-time        21600;
> }
> 
>   host sui-wlan {
>     host-identifier option dhcp6.client-id
> 00:01:00:01:4a:00:49:71:01:72:e6:b7:6c:21; # < seems like
> https://fedorahosted.org/dhcpv6/ is generating some other DUID than ISC
> dhclient-4?! > 00:01:00:01:11:92:FB:91:00:90:F5:48:96:D0; # default MAC
> > 00:90:F5:48:96:D0;
>     fixed-address6 2001:41D0:1:678E:ABCD::2;
>   }
> 
> Now to come to some terms...
> 
> > Note however that at least dhclient, and one or two others, DHCPv6
> > software will presume a /64 prefix for any offered addresses, in order
> > to work around a bug-in-the-protocol where an RA failure leads to the
> > addressed clients being unable to speak to each other.
> >  =20
> Exactly this is what is happening. I've tried all to me known DHCPv6
> clients, that is ISC dhclient and dhcpv6 (see URL above) - dhcpcd yet
> has no v6 support, though it is to be included soon according the
> project owner.
> 
> This behavior is of course causing a lot of trouble around - although
> the address is given out correctly, other hosts in the same /64 subnet
> cannot be accessed, but you already now this. :)
> 
> Another problem is that RA totally refused to work. Although I can see
> advertisements via radvdump, nothing happens automagically. I don't know
> how to track this, I even have got absolutely no evidence of ra in my
> kernel log ring buffer output (Linux 2.6.28.7 for instance), that is, in
> dmesg.
> 
> I have even tried removing all IPv6 addresses and routes and adding the
> IP manually via ip -6 addr a 2001... dev eth0 and waiting some minutes
> to see if ra is working then, but nothing happened. The only way seems
> to be adding the route by hand which... sucks.
> 
> > There's currently a default gateway and prefix option draft before the
> > DHC WG, which will help us resolve the above bug-in-the-protocol and
> > eliminate the strict requirement for RA
> I've stumpled upon an RFC implementing the routers option in DHCPv6, but
> can't remember the URL anymore - is this the thing you refer to? :)
> 
> It would be perfectly cool to have a working setup which is independant
> from autoconfig - either my current one is wrong (both dhcp and ra wise)
> or I'm doing some things completely wrong.
> 
> Best regards,
> 
> 
> Mihai Moldovan