Re: [Fwd: I-D Action: draft-carpenter-6man-why64-00.txt]

[Alexandru Petrescu <[e-mail]>]

Le 07/01/2014 12:11, Ray Hunter a écrit :
[...]
> Where exactly is the requirement that IPv6 has to use /64 prefix
> length really enshrined?

Simply put, I agree with Brian that there is no one single specification
place where this could be found enshrined.

But a huge devil lies in details I dare uncover just a bit here.

First, key places like the Addressing Architecture and SLAAC RFCs
generally read as avoiding setting rigid limits on the prefix length or
the Interface ID.  For example the subnet prefix is said to be of length
'n', and the Interface ID of length '128-n'.

But even there in RFC4291 there is a quirk saying:
> For all unicast addresses, except those that start with the binary
> value 000, Interface IDs are required to be 64 bits long and to be
> constructed in Modified EUI-64 format.

(not sure what that 000 means).

Then, an IP address being of fixed length, the prefix length could be
inferred by looking at the length of the Interface ID, or other
identifiers which look like it (e.g. an anycast IID).  If that length is
rigid, one may legitimately suppose strict limits on the prefix length
as well.

The IP-over-Ethernet RFC is precise in its limit of 64bit for the
Interface ID.  The success of its deployment made it so that others copy
on it, together with that rigid limit.  Today, not only Ethernet
Interface IDs are precisely 64bit, but also WiFi, and, more surprisingly
even 802.15.4 (surprisingly because even its short address on 16bit
leads to an Interface ID on 64bits - 3rd par, sec 6 RFC 4944).  Worse,
other links which are not even documented as IP-over-foo RFCs do
implement that 64bit limit although nobody requires them to.

As for implementation enshrinment places there are many.  One easy to
spot is dmesg on modern linux upon reception of an RA containing a
prefix whose length is 65, or 63 if one so wishes.  Decimal.

Alex