IPv6 DHCP Routing
dhcp1 at thehobsons.co.uk
Tue Nov 17 19:13:10 UTC 2015
Brice Abba <briceabba at gmail.com> wrote:
> This seems to be a normal behaviors, because we dont have possibility to specify the prefix length in DHCPv6, so every simgle machine will have /128, then even if they are all on the same link we'll have routing, to avoid this,
> You can ask your router sending RA, to set the "L" flag to tell nodes to consider that they are on the same link, this way you could avoid all trafic to pass through the GW. And set manegement flag as well
I think something is misconfigured then. The RAs should contain prefix(es)/information on what's considered on-link. If properly configured, each node should know what prefixes are on-link and should never need to send "local" traffic via a router.
There was "extensive 'discussion'"* a while back on the IETF DHC WG list about whether a DCHP server should include a prefix length with an address, and whether addres & prefix should form an "inseparable pair". I learned a lot from that, and specifically :
Each router should be advertising both the routes it can handle, and the list of prefixes on the link. Part of the information is whether a prefix should be considered "on-link" - which is a very different concept between IPv4 and IPv6. With IPv4 we've got used to "same subnet = on-link". With IPv6 we can have multiple prefixes which are all on-link, and it's supported to have an Ipv6 address which is not on-link with another address in the same prefix.
So an IPv6 is just an address, it doesn't automatically have a prefix with it (in the same way that IPv4 addresses have a subnet mask with them) - though many implementations don't support having one without the other. "on-link" is determined by reference to a list of prefixes that the router(s) tell the node are local to it - ie if the destination address is in an on-link prefix then it's on-link regardless of the source address.
* For 'discussion', read 'at times heated debate' ;-)
More information about the dhcp-users