On Thu, Aug 6, 2015 at 10:08 AM, David Woodhouse <dwmw2 at infradead.org> wrote: >> With this mail I bring up a discussion made on the irc channel here. >> As it is now ocserv, for IPv6 provides an IP address from the >> configured pool using a dummy prefix length, the same way as we do for >> IPv4. However, since in IPv6 the number of addresses are pretty much >> unlimited it makes sense to provide a real subnet to the client. That >> will not have much impact on the clients which handle the provided >> address as a point-to-point one, but will allow future clients to use >> multiple addresses from the VPN. Do you see any issues with that >> approach, or have an idea to improve it? > No recollection of this. Did I respond? I'm trawling my list archive > because I *know* someone sent me a patch to fix the Juniper multiple > DNS search domain issue. > I'd take a look at IPv6 Prefix Delegation as handled in PPP and DHCPv6. > It hands out subnets to the client and the client can then run RA on > its *other* interfaces (and hand them out further, perhaps). > I think it's normally done as a *separate* configuration item to the > main IPv6 address. Yes, there are plenty but there's no reason to be > entirely profligate with them. You can have a /127 for the point-to > -point link, and multiple /64s for the subnets you want to route to. I'm currently giving a 127 mask to the clients. That's the safest bet now, and can be easily extended if we decide for openconnect to make use of larger blocks. However, I am wondering whether the mask received by anyconnect servers has any meaning at all, i.e., whether it is a real mask with addresses the client can use, or is some mask that has meaning for the server only. In IPv4 the latter is the case. regards, Nikos