RFC Errata


Errata Search

 
Source of RFC  
Summary Table Full Records

RFC 4291, "IP Version 6 Addressing Architecture", February 2006

Note: This RFC has been updated by RFC 5952, RFC 6052, RFC 7136, RFC 7346, RFC 7371, RFC 8064

Source of RFC: ipv6 (int)

Errata ID: 5716
Status: Reported
Type: Technical
Publication Format(s) : TEXT

Reported By: Dennis Ferguson
Date Reported: 2019-05-01

Section 2.5 says:

   A slightly sophisticated host (but still rather simple) may
   additionally be aware of subnet prefix(es) for the link(s) it is
   attached to, where different addresses may have different values for
   n:

It should say:

   For Link-Local unicast addresses, which comprise a flat address
   space that is reused on every link, the consideration above is all
   that is relevant. For Global Unicast addresses, however, a slightly
   sophisticated host (but still rather simple) may additionally be
   aware of subnet prefix(es) for the link(s) it is attached to, where
   different addresses may have different values for n:

Notes:

Text in several sections of 4291 seem to be logically inconsistent with respect to Link-Local unicast addresses. The inconsistency spans Sections 2.1, 2.5 and 2.5.6.

Section 2.5 says that unicast addresses, including Link-Local unicast addresses, on links a host is attached to have the following internal structure, providing a definition for "subnet prefix":

| n bits | 128-n bits |
+-------------------------------+---------------------------------+
| subnet prefix | interface ID |
+-------------------------------+---------------------------------+

Section 2.5.6 specifies that Link-Local unicast addresses have the following format:

| 10 |
| bits | 54 bits | 64 bits |
+----------+-------------------------+----------------------------+
|1111111010| 0 | interface ID |
+----------+-------------------------+----------------------------+

Applying the Section 2.5 definition, Link-Local unicast addresses hence have a subnet prefix of fe80::/64 on every link.

Section 2.1 says this about the addressing model:

Currently, IPv6 continues the IPv4 model in that a subnet prefix is
associated with one link.

Clearly the use of a same Link-Local unicast subnet prefix on every link is inconsistent with an addressing model that requires a subnet prefix to be unique to one link. The plain reading of this text is that it disagrees with itself, so either something needs to change to bring it into agreement or some explanation of how this is in fact consistent needs to be made. Since I don't know what the intent of this was I don't really know what needs fixing.

The proposed text hence reflects my personal biases; while I understand subnet prefix uniqueness to be foundational to the forwarding behaviour of Global Unicasts I have no idea what a "subnet prefix" even means in the context of Link-Local unicasts. The proposed text hence modifies Section 2.5 to exclude Link-Local addresses from the part of that section defining a "subnet prefix" and instead be defines them to provide a flat address space that is reused on every link. That is, Link-Local addresses are defined without a subnet prefix.

This leaves the Section 2.1 statement about subnet prefixes, and the forwarding behaviour implied by that, applicable to Global Unicast addresses but inapplicable to Link-Local addresses. Given this Section 2.5.6 can be read to be solely a constraint on Link-Local address construction, i.e. requiring SLAAC and other methods of Link-Local configuration to only produce addresses from that subset, without any implication for forwarding behaviour that "subnet prefix" might suggest.

That this general topic seems to produce periodic flurries of concern might be an indication that greater clarity in specification here might be useful.

Report New Errata



Advanced Search