On 2024/05/21 20:30, jrmu wrote: > Greetings, > > > > I also don't control the entire /48. > > > > > > Here is the information I was given: > > > > > > My IPv6 Address Subnet: 2602:fccf:400:41::/64 > > > Hypervisor' IPv6 Gateway: 2602:fccf:400::1 > > > > > > I was only given a /64. > > > > So you should use a /64 prefix length not the /48 which you have. > > > > See EXAMPLES in route(8) for how to set the gateway. > > Please excuse my ignorance here, as I am unfamiliar with networking. Can > you explain why /64 is the correct prefix length?
Because that is the information they gave you: "Here is the information I was given: My IPv6 Address Subnet: 2602:fccf:400:41::/64" > I am confused because it seems not analogous to IPv4. Your provider has decided to use a different config method for v6 compared to v4. They probably have a route for the whole /64 to your MAC address to avoid having to do neighbour discovery (NDP) for addresses in your subnet. If they did NDP, they have to try to find the MAC address to send packets for that individual address. So if that address isn't in the (limited size) NDP cache their router would need to buffer the packet, try to resolve the address, if that address is not configured anywhere they'd need to wait for a timeout before possibly generating a host-unreachable icmp6 message and discarding the packet. These are all slow operations using cpu resources on a router where those resources are usually quite limited. Now consider the number of addresses in the subnet and that someone on the internet can send packets to any address. There are similar issues for v4 (using ARP rather than NDP to find MAC addresses) but the scale is vastly different - and most addresses will be in use anyway so most of the time a randomly addressed packet will already have the MAC address in the ARP cache. There are other ways to handle this (e.g. add a small 'link net' between the router and your host) but config for that is a bit more hassle to do on the provider's side - typically with that setup you'd have a separate vlan per customer too, as well as the route table entry across the provider's network for the link net, using more resources on routers/switches. > In the IPv4 example, my address is 104.167.241.211, the gateway is > 104.167.241.193, and the subnet mask 255.255.255.192. The network length > then is /26. I don't control the entire /26 subnet, only one single IPv4 > address within it, but my network would have a prefix length of /26. All of the /26 is probably directly reachable (using ARP to lookup the MAC address). And vice-versa, other addresses in the /26 will be expecting to be able to send packets to you directly rather than going via the gateway. > Isn't using a prefix length of /48 the same in the case of IPv6? I don't > control the entire /48, but the gateway 2602:fccf:400::1 shares the > first 48 network bits with my IPv6 address 2602:fccf:400:41:: You almost certainly can't reach the rest of the /48 without going via the gateway. > If I were to set the routing prefix length to 64, then I could manually > add an extra route to the IPv6 gateway. But then, wouldn't I want to set > my IPv4 address with a subnet mask of 255.255.255.255, so that the > network length would be 32 rather than 26, and also add a manual route > there? Some providers do do that for v4, but if they had they'd be telling you to use the /32. There's a lot less reason to do it for v4 though.
