path: root/doc/rfc/rfc3627.txt

Network Working Group                                          P. Savola
Request for Comments: 3627                                     CSC/FUNET
Category: Informational                                   September 2003


      Use of /127 Prefix Length Between Routers Considered Harmful

Status of this Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2003).  All Rights Reserved.

Abstract

   In some cases, the operational decision may be to use IPv6 /127
   prefix lengths, especially on point-to-point links between routers.
   Under certain situations, this may lead to one router claiming both
   addresses due to subnet-router anycast being implemented.  This
   document discusses the issue and offers a couple of solutions to the
   problem; nevertheless, /127 should be avoided between two routers.

1.  Introduction

   [ADDRARCH] defines Subnet-router anycast address: in a subnet prefix
   of n bits, the last 128-n bits are all zero.  It is meant to be in
   use of any one router in the subnet.

   Even though having prefix length longer than /64 is forbidden by
   [ADDRARCH] section 2.4 for non-000/3 unicast prefixes, using /127
   prefix length has gained a lot of operational popularity; it seems
   like that these prefix lengths are being used heavily in point-to-
   point links.  The operational practise has often been to use the
   least amount of address space especially in the presence of a large
   number of point-to-point links; it may be unlikely that all of these
   links would start to use /64's.  Using /127 has also other
   operational benefits: you always know which address the other end
   uses, and there is no "ping-pong" [PINGPONG] problem with older ICMP
   implementations (fixed now in [ICMPv3]).








Savola                       Informational                      [Page 1]
^L
RFC 3627         /127 Prefix Length Considered Harmful    September 2003


2.  Scope of this Memo

   This memo does not advocate the use of long prefixes, but brings up
   problems for those that do want to use them, for one reason or
   another.

   Detailed discussion on what is the "right" solution is out of the
   scope; it is not the goal of this memo to try to find the "best"
   addressing solution for everyone.

3.  Problem with /127 and Two Routers

   Note that this problem does not exist between a router and a host,
   assuming the PREFIX::0/127 address is assigned to the router.

   Using /127 can be especially harmful on a point-to-point link when
   Subnet-router anycast address is implemented.  Consider the following
   sequence of events:

   1. Router A and Router B are connected by a point-to-point link.

   2. Neither has anything configured or set up on this link.

   3. 3ffe:ffff::1/127 address is added to Router A; now it performs
      Duplicate Address Detection (DAD) [NDISC] for 3ffe:ffff::1.
      Router A also adds the Subnet-router anycast address
      3ffe:ffff::0/127.  (DAD is not performed for anycast addresses.)

   4. Now Router B has been planned and configured to use
      3ffe:ffff::0/127 as its unicast IPv6 address, but adding it will
      fail DAD, and Router B does not have any address.

   Similar scenarios also happen during router reboots, crashes and
   such.

   The usability of subnet-router anycast address between two routers on
   a point-to-point link is very questionable, but it is still a
   mandated feature of [ADDRARCH].  Workarounds for this are presented
   in the next section.

   As of yet, this kind of unexpected behavior hasn't been seen at large
   perhaps because the Subnet-router anycast address hasn't been
   implemented or too widely used.








Savola                       Informational                      [Page 2]
^L
RFC 3627         /127 Prefix Length Considered Harmful    September 2003


4.  Solutions

   1. One could use /64 for subnets, including point-to-point links.

   2. One could use only link-local addresses, but that may make network
      maintenance and debugging impractical at least in bigger networks;
      for example, "traceroute" can only return a list of nodes on the
      path, not the links which would have been used.

   3. Failing that, /126 does not have this problem, and it can be used
      safely on a point-to-point link (e.g., using the 2nd and the 3rd
      address for unicast).  This is analogous to using /30 for IPv4.
      Using two /128 addresses is also one, though often cumbersome,
      approach.  Naturally, not much would be lost if even a shorter
      prefix was used, e.g., /112 or /120.

      The author feels that if /64 cannot be used, /112, reserving the
      last 16 bits for node identifiers, has probably the least amount
      of drawbacks (also see section 3).

   4. [ADDRARCH] could be revised to state that Subnet-router anycast
      address should not be used if the prefix length of the link is not
      /64 (or even longer than /120).  This does not seem like a good
      approach, as we should avoid making assumptions about prefix
      lengths in the specifications, to maintain future flexibility.
      Also, in some cases, it might be usable to have a Subnet-router
      anycast address in some networks with a longer prefix length.

      A more conservative (implementation) approach would be not using
      Subnet-router anycast addresses in subnets with a prefix length of
      /127 if there are only two routers on the link: this can be
      noticed with [NDISC] 'Router' bit in Neighbor Advertisement
      messages.  However, this seems to overload the functionality of
      'R' bit, so it does not look like a good approach in the long run.

   5. It's also possible to improve implementations: if /127 is used on
      a point-to-point link, never claim two addresses.  This has the
      drawback that even if the router using the combined unicast and
      anycast address is down, the packets to subnet-router anycast
      address will be lost as the other cannot claim the address.  This
      approach might lead to unpredictability which would be hard to
      trace when debugging problems.  However, this would normally be an
      issue only when the Subnet-router anycast address is used from
      outside of the link; usually, this cannot be done reliably as the
      prefix length or EUI64 u/g bits cannot be known for certain.
      There are other problems with an address being anycast and unicast





Savola                       Informational                      [Page 3]
^L
RFC 3627         /127 Prefix Length Considered Harmful    September 2003


      too: use of it as a source address, whether to use unicast or
      anycast semantics in [NDISC], and others: allowing this behavior
      would seem to only add a lot of complexity to the implementations.

   1) is definitely the best solution, wherever it is possible.  2) may
   be usable in some scenarios, but in larger networks (where the most
   often the desire would be to use longer prefix length) it may be
   deemed very impractical.  There are some situations where one of
   these may not be an option; then an operational work-around for this
   operational problem, that is 3), appears to be the best course of
   action.  This is because it may be very difficult to know whether all
   implementations implement some checks, like ones described in 4) or
   5).

5.  Other Problems with Long Prefixes

   These issues are not specific to /127.

   One should note that [ADDRARCH] specifies universal/local bits (u/g),
   which are the 70th and 71st bits in any address from non-000/3 range.
   When assigning prefixes longer than 64 bits, these should be taken
   into consideration; in almost every case, u should be 0, as the last
   64 bits of a long prefix is very rarely unique.  'G' is still
   unspecified, but defaults to zero.  Thus, all prefixes with u or g=1
   should be avoided.

   [MIPV6] specifies "Mobile IPv6 Home-Agents" anycast address which is
   used for Home Agent Discovery.  In consequence, 7 last bits of have
   been reserved in [ANYCAST] of every non-000/3 non-multicast address,
   similar to [ADDRARCH].  Thus, at least /120 would seem to make sense.
   However, as the sender must know the destination's prefix length,
   this "reserved anycast addresses" mechanism is only applicable when
   the sender knows about the link and expects that there is a service
   it needs there.  In the case of e.g., /126 between routers, the only
   to node to be found on this link would be the other router, so the
   mechanism does not seem useful.  At least, Mobile IPv6 Home Agent
   Discovery should not be performed if the prefix length is longer than
   /120.

6.  References

6.1.  Normative References

   [ADDRARCH]  Hinden, R. and S. Deering, "IP Version 6 (IPv6)
               Addressing Architecture", RFC 3513, April 2003.

   [ANYCAST]   Johnson, D. and S. Deering, "Reserved IPv6 Subnet Anycast
               Addresses", RFC 2526, March 1999.



Savola                       Informational                      [Page 4]
^L
RFC 3627         /127 Prefix Length Considered Harmful    September 2003


6.2.  Informative References

   [NDISC]     Narten, T., Nordmark, E. and W. Simpson, "Neighbor
               Discovery for IP Version 6 (IPv6)", RFC 2461, December
               1998.

   [MIPV6]     Johnson, D., Perkins, C., Arkko, J., "Mobility Support in
               IPv6", Work in Progress.

   [ICMPv3]    Conta, A., Deering, S., "Internet Control Message
               Protocol (ICMPv6)", Work in Progress.

   [PINGPONG]  Hagino, J., Jinmei, T., Zill, B., "Avoiding ping-pong
               packets on point-to-point links", Work in Progress.

7.  Security Considerations

   Beyond those already existing in other specifications, solution 4)
   might lead to denial of service in the case that one router is down:
   the packet to subnet-router anycast address would be lost.

8.  Acknowledgements

   Thanks to Robert Elz and many others on the IPv6 Working Group for
   discussion, and Alain Durand for pointing out [ADDRARCH] requirements
   for prefix lengths.  Charles Perkins pointed out MIPv6 HA
   requirements.  Randy Bush and Ole Troan commented on the document
   extensively, and Erik Nordmark pointed out issues with u-bit.

9.  Author's Address

   Pekka Savola
   CSC/FUNET
   Espoo, Finland

   EMail: psavola@funet.fi















Savola                       Informational                      [Page 5]
^L
RFC 3627         /127 Prefix Length Considered Harmful    September 2003


10.  Full Copyright Statement

   Copyright (C) The Internet Society (2003).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assignees.

   This document and the information contained herein is provided on an
   "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.



















Savola                       Informational                      [Page 6]
^L