Iljitsch van Beijnum wrote:
This still breaks the notion that once you do duplicate address
detection for your MAC-derived link-local address you can assume that
any other MAC-derived addresses are also unique.
How does it break this notion?
Consider:
RFC 2462 says if DAD fails for link-local, the interface should be disabled.
I claim that if it succeeds for link-local, under either scheme, it will
succeed for all autoconf addresses on the same interface, under either
scheme.
(BTW, a corrolary is: it will even succeed when both schemes are in use
on /64 prefixes, since FFFE does not collide with any part of any
assigned OUI.)
This is clear enough, by working backwards:
If DAD would fail for some other address built on this new format for
II, <padding><oui24><24bitSerial>, then it would fail for
<oui24><24bitSerial>, which means DAD would fail for link-local.
In order for this to happen, however, DAD on the new address must be
attempted, which can only occur if link-local DAD had succeeded, or else
the interface would have been disabled.
Link-local cannot both succeed and fail.
Ergo, DAD on link-local will guarantee DAD success on all other
MAC-derived addresses under this scheme.
I.e. your assertion is incorrect, while mine is correct.
Brian
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------