Erik Nordmark <erik.nordm...@oracle.com> writes: > While some implementations of RFC 4861/62 might make assumptions about > /64 being magic, there isn't anything in those RFCs that specify that it > is magic. RFC 4862 merely says that the sum of the length of the link > prefix and the interface identifier needs to be 128.
Right. That was by design, as I recall. :) > The IPv6-over-foo documents tend to say that the IID is 64 bits, and we > have the text in RFC4291. But stateless addr autoconf is actually more > general than that. I think "tend to say" is too mild. They all require it, in a MUST sense, even if they don't use the MUST keyword. Unfortunate, IMO, as the reality is this will be very hard to change now, given the code that has already been written. And there is no good reason (IMO) to require that they be exactly 64 bits long. (The original motivation for doing so was to leave the door open to approaches like GSE/8+8, but the reality is that this restriction doesn't seem to have bought us much, if anything, in practice.) RFC 2464 (IPv6 over Ethernet) says, for example: An IPv6 address prefix used for stateless autoconfiguration [ACONF] of an Ethernet interface must have a length of 64 bits. Thomas -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
