TOC 
Internet Engineering Task ForceS. SUZUKI
Internet-DraftHitachi, Ltd.
Expires: August 2, 2003T. JINMEI
 Toshiba Corporation
 Feb 2003

PIM upstream detection among multiple addresses
draft-suz-pim-upstream-detection-00.txt

Status of this Memo

This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.

This Internet-Draft will expire on August 2, 2003.

Copyright Notice

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

Abstract

This memo describes a PIM routing problem owing to the reverse path forwarding (RPF) failure on routers having multiple addresses on a link. This memo also considers possible solutions with their pros and cons. The solutions include operational one which does not require any protocol change, and one which uses a new PIM Hello option. However, it is beyond the scope of this memo to discuss what is the most appropriate solution for this problem.



 TOC 

Table of Contents




 TOC 

1. Overview of the problem

The PIM protocol uses the reverse path forwarding (RPF) algorithm to detect an upstream PIM router with respect to the address of a packet sender or a rendezvous point (RP) router [1].

The RPF algorithm requires a look-up in a multicast routing information base (multicast RIB, which is normally derived from unicast routing table, but may be from other routing protocols such as MBGP [2]) to find the next hop address for a given address. If the next hop address matches with the address of a PIM neighbor then that PIM neighbor can be regarded as the upstream PIM router for the address. The address of the upstream neighbor is then used in the Upstream Neighbor Address field of Join/Prune or Graft messages.

The address of a PIM neighbor is given as the source address of PIM Hello messages from the neighbor. The PIM protocol specifies to use a link-local address for this purpose. (Note: it is not very clear what "link-local" means for IPv4, but link-local addresses are well-defined for IPv6.)

The entire procedure to detect the upstream address assumes the address of a PIM neighbor is always same as the address of the next hop router, as long as they refer to the same router. This is typically the case for IPv4, and so is for IPv6 when an Interior Gateway Protocol is used to build the multicast RIB. In general, however, it may not be the case when a router has multiple addresses on a link.

Suppose a router-A has address-A1 and address-A2 on a link and it establishes a PIM neighboring with a router-B using address-A1.

	          address-A1        PIM neighbor = address-A1
	          address-A2	    next-hop to S = address-A2
	S----router-A ------|------- router-B
 

When router-B tries to detect an upstream router for a source address S, router-B cannot detect an upstream PIM router even though its RPF calculation says the S's upstream is address-A2, because router-B does not know address-A2 and address-A1 refer to the same router. In this case, router-B does not create Join/Prune or Graft messages for S, and PIM does not work correctly.

There are two typical cases that lead to this situation for IPv6. The first case can occur when the multicast RIB is not built by an IPv6 Interior Gateway Protocol. This includes static routing and MBGP. The second case occurs when the address of RP shares a subnet prefix with down stream routers (note that the RP router's address has to be domain-wide and thus cannot be a link-local address).

The following figure depicts the second case.

	      fe80::1             PIM neighbor = fe80::1
	      2001:db8::1         RP(G) = 2001:db8::1
	RP router ---------|---------- router
		     2001:db8::/64
 

Since this issue can happen in some typical cases for IPv6, the issue has to be resolved in some manner, in order to deploy IPv6 multicasting (with PIM).



 TOC 

2. Possible solutions

There are four possible solutions to this problem. This section describes their details and their pros and cons.

2.1 Avoid such situation by operation

There is operational workaround for this problem.

When manually specifying an upstream router's address, it is always possible to use a link-local address. When using MBGP, a link-local next hop address can be specified if the BGP peering is single-hop and a separate non link-local prefix is shared on the peering link [3].

The problem occurring when a RP shares a subnet prefix with downstream routers can also be avoided by using a separate RP address. In this case, the RP address has to be picked up from some link whose prefix is not shared with any downstream routers. One practical way to implement this would be to choose a new prefix and assign an address derived by the prefix to a loopback interface.

Another possible approach is to advertise a host route for the upstream address in the routing protocol used to build multicast RIB. This can work unless static routing is used, and if the host route does not interfere with other intra-link information, such as IPv6 Neighbor Discovery (ND) or IPv4 Address Resolution Protocol (ARP).

All of the approaches will work without any protocol extensions. However, they impose additional costs in operation and/or restrict flexibility in operation.

pro =
Does not need to change anything in the PIM protocol.
con =
Imposes additional costs in operation, and/or restricts operational flexibility.

2.2 Use a new PIM Hello option

This is a solution to let the PIM protocol avoid such situation by adding a PIM hello message option including all the addresses on the interface where PIM hello message is advertised.

When a PIM router finds an upstream router for some address, the result of RPF calculation is compared with the addresses in this option, in addition to PIM neighbor's address itself. Since this option includes all the possible addresses of a PIM router on that link, it always includes the RPF calculation result if it refers to the PIM router supporting this option.

   0                   1                   2                   3
   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |             Type              |            Length             |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |                         Address-List                          |
  |                              ...                              |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 
Type :
TBD (All vendors supporting this option use 65001)
Length :
The Length of the Address-List field in byte.
Address-List:
The list of the IP addresses of the router on the link where this Hello message is advertised. Each IP address is encoded in Encoded-Unicast format [1].

This approach does not require a separate configuration to deal with the problem, because the address mapping is provided automatically. Thus, this approach reduces the operational costs comparing to the first solution.

Furthermore, since this is an optional attribute in PIM hello message, PIM implementations not supporting this option do not have any problem in PIM routing, unless the problem described here does not occur.

pro =
Does not increase operational cost. Keeps flexibility of configuration.
con =
Requires a change (an extension) in the PIM protocol, and increases implementation cost.

2.3 Use a separate protocol

This solution is similar to the previous one except for the protocol to be used; this solution assumes the existence of some other protocol to detect a list of addresses that point to the same node pointed by the given address.

One possible example is to use a protocol to resolve link-layer addresses. IPv6 ND or IPv4 ARP is available for this purpose. With this approach, the downstream router resolves the link-layer address both for the upstream router address and for the PIM neighbor address. If the link-layer addresses coincides, the corresponding layer 3 addresses are regarded to specify a same node.

IPv6 also has a dedicated layer 3 protocol that can be used for this purpose; ICMPv6 Node Information Query [4] using Node Addresses query type.

These protocols can provide a complete mapping among PIM neighbor addresses and upstream router addresses. Thus, they can solve the problem just like the solution described in the previous section.

It does not increase operational costs, just like the previous solution. This approach does not require any change in the PIM protocol, either. However, PIM implementations will need to be modified to solve the PIM-specific problem using the separate protocol.

Additionally, there is limitation on applicability of this approach. ND or ARP does not work to solve the problem on a link that does not need to resolve link-layer addresses (e.g. a point-to-point link). Also, in theory, different layer-3 addresses on one interface may be mapped to different link-layer addresses resolved by ND or ARP. Though this should be practically rare, ND or ARP does not work in this case, either.

There may be a different kind of clue for this mapping in such cases. For example, in case of point-to-point link, addresses not belonging to one side should belong to the other side of the link. The implementation can also use inverse ARP [5] or inverse ND [6] to deal with these cases. However, an exceptional rule has to be implemented in addition to ND or ARP, which may increase implementation costs.

ICMPv6 Node Information Query, particularly the Node Addresses query type, is not widely implemented. This means the protocol cannot be used as a solution today, and will increase implementation cost for router developers.

The use of a separate protocol also requires additional process to ensure consistency of the mapping. In the first place, there is no guarantee that the mapping is provided when it is necessary in the PIM protocol. Thus, implementation may have to invoke the other protocol after it noticed the need for the mapping. This can delay the detection procedure. Secondly, in order to follow address changes, it is necessary for the downstream router to poll the neighbor's address periodically.

pro =
Does not increase operational cost. Keeps flexibility of configuration.
con =
May require additional implementation cost.
con =
May require additional consideration (and implementation costs as a result) for corner cases or to ensure consistency.

2.4 Loosen the PIM protocol

The problem can be solved by loosening restrictions of the PIM protocol. For example, the loosened protocol would allow a router to specify a non-link-local address as the Upstream Neighbor Address of Join/Prune or Graft messages. If the upstream router does not check if the Upstream Neighbor Address is link-local, the corresponding part of the PIM protocol should still work.

However, this solution itself does not provide a mapping between the address of a PIM neighbor and the RPF upstream address corresponding to the neighbor. Thus, the downstream router cannot make fast retransmission of Join/Prune messages when the router detects recovery from failure of an upstream router by Generation ID.

This approach may require further changes in the PIM protocol in the future, if the protocol has a new header field where an upstream neighbor address should be specified.

pro =
Does not increase operational cost. Keeps flexibility of configuration.
con =
Requires a change in the PIM protocol
con =
Not a complete solution. It would hinder the optimization by Generation ID.



 TOC 

3. Security considerations

When the new PIM hello option described in Section Use a new PIM Hello option is used, a forged message can be used to hijack a multicast distribution path or to cause a denial of service attack. However, since the use of PIM hello messages is limited to a single link, such attacks cannot be made off-link. Additionally, these attacks within a single link can be done using existing PIM messages. The new PIM hello option thus does not increase the security threat of the existing PIM protocol.



 TOC 

4. Acknowledgements

We would like to express thanks to Pekka Savola, Dave Thaler, and Brian Haberman for their valuable comments.



 TOC 

Normative References

[1] Fenner, B., Handley, M., Holbrook, H. and I. Kouvelas, "Protocol Independent Multicast Sparse Mode (revised)", Internet Draft draft-ietf-pim-sm-v2-new-06.txt, December 2002.
[2] Bates, T., Rekhter, Y., Chandra, R. and D. Katz, "Multiprotocol Extensions for BGP-4", RFC 2858, June 2000.
[3] Marques, P. and F. Dupont, "Use of BGP-4 Multiprotocol Extensions for IPv6 Inter-Domain Routing", RFC 2545, March 1999.
[4] Crawford, M., "IPv6 Node Information Queries", Internet Draft draft-ietf-ipngwg-icmp-name-lookups-09.txt, May 2002.
[5] Bradley, T., Brown, C. and A. Malis, "Inverse Address Resolution Protocol", RFC 2390, August 1998.
[6] Conta, A., "Extensions to IPv6 Neighbor Discovery for Inverse Discovery Specification", RFC 3122, June 2001.


 TOC 

Informative References



 TOC 

Authors' Addresses

  SUZUKI Shinsuke
  Hitachi, Ltd.
  1-280 Higashi-Koigakubo
  Kokubunji-shi, Tokyo 185-8601
  Japan
EMail:  suz@crl.hitachi.co.jp
  
  JINMEI Tatuya
  Toshiba Corporation
  1 Komukai Toshiba-cho
  Kawasaki-shi, Kanagawa 212-8582
  Japan
EMail:  jinmei@isl.rdc.toshiba.co.jp


 TOC 

Full Copyright Statement

Acknowledgement