Deprecation of AS

BCP 172 makes a recommendation for not using AS_SET (see ) and AS_CONFED_SET (see ) in the Border Gateway Protocol (BGP). This document advances the BCP recommendation to a standards requirement in BGP; it proscribes the use of the AS_SET types of path segments in the AS_PATH. The purpose is to simplify the design and implementation of BGP and to make the semantics of the originator of a route clearer. This will also simplify the design, implementation, and deployment of various BGP security mechanisms. In particular, the proscription of AS_SETs removes the possibility of ambiguity about origin AS in RPKI-based route origin validation (RPKI-ROV) . The AS_SET path segment in the AS_PATH attribute (Sections 4.3 and 5.1.2 of ) is created by a router that is performing route aggregation and contains an unordered set of Autonomous Systems (ASes) that contributing prefixes in the aggregate have traversed. By performing aggregation, a router is combining multiple BGP routes for more specific destinations into a new route for a less specific destination (, Section 9.1.2.2.). Aggregation may blur the semantics of the origin AS for the prefix being announced by producing an AS_SET. AS_SETs can cause operational issues, such as not being able to authenticate a route origin for the aggregate prefix in new BGP security technologies such as those that take advantage of X.509 extensions for IP addresses and AS identifiers (, , , , , ). This in turn could result in reachability problems for the aggregated prefix and its components; i.e., more specific prefixes. From analysis of historical Internet routing data, it is apparent that aggregation that involves AS_SETs is very seldom used in practice on the public Internet . When it is used, it is often used incorrectly; only a single AS in the AS_SET are by far the most common cases. Also, very often the same AS appears in the AS_SEQUENCE and the AS_SET in the BGP update. The occurrence of reserved AS numbers () is also somewhat frequent.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

BGP speakers conforming to this document (i.e., conformant BGP speakers) SHOULD NOT locally generate BGP UPDATE messages containing AS_SET. Conformant BGP speakers SHOULD NOT send BGP UPDATE messages containing AS_SETs. Upon receipt of such messages, conformant BGP speakers SHOULD use the "Treat-as-withdraw" error handling behavior as per . If a network operator wishes to consider BGP UPDATE messages with AS_SETs received from an external BGP peers, they MAY have a feature (knob) in their implementation to do so on a per-peer basis. The operator should understand the full implications of choosing this option. Network operators SHOULD NOT locally generate any new announcements containing AS_SETs. BGP security technologies (such as those that take advantage of X.509 extensions for IP addresses and AS identifiers (, , , ) might not support routes with AS_SETs or AS_CONFED_SETs in them. Routes with AS_SETs have no possibility of ever being considered RPKI-ROV valid .

This document deprecates the origination of BGP routes with AS_SET (type 1) AS_PATH segments. (, Section 4.3.) BGP speakers conforming to this document — i.e., conformant BGP speakers — SHOULD NOT originate BGP UPDATE messages containing AS_SETs. Upon receipt of BGP routes containing AS_SETs, conformant BGP speakers SHOULD use the "Treat-as-withdraw" error handling behavior as per .

, Sections 9.1.4 and 9.2.2.2, describes BGP aggregation procedures. , Appendix F.6 describes a generally unimplemented "Complex AS_PATH Aggregation" procedure. , Section 5.1.6 describing the ATOMIC_AGGREGATE Path Attribute notes that:

When a BGP speaker aggregates several routes for the purpose of advertisement to a particular peer, the AS_PATH of the aggregated route normally includes an AS_SET formed from the set of ASes from which the aggregate was formed. In many cases, the network administrator can determine if the aggregate can safely be advertised without the AS_SET, and without forming route loops. If an aggregate excludes at least some of the AS numbers present in the AS_PATH of the routes that are aggregated as a result of dropping the AS_SET, the aggregated route, when advertised to the peer, SHOULD include the ATOMIC_AGGREGATE attribute.

When BGP AS_PATH aggregation is done according to the Section 9.2.2.2 procedures and any resulting AS_SETs are discarded, this is typically referred to as "brief" aggreation in implementations. This results in an AS_PATH that has the property (from Section 9.2.2.2):

determine the longest leading sequence of tuples (as defined above) common to all the AS_PATH attributes of the routes to be aggregated. Make this sequence the leading sequence of the aggregated AS_PATH attribute.

The ATOMIC_AGGREGATE Path Attribute is subsequently attached to the BGP route, if AS_SETs are dropped.

While brief AS_PATH aggregation has the desirable property of not containing AS_SETs, the resulting aggregated AS_PATH may contain an unpredictable origin AS. Such an unpredictable origin ASes may result in RPKI-ROV validation issues:

Depending on the contributing routes to the aggregate route, the resulting origin AS may vary.
The presence of expected contributing routes may be unpredictable due to route availability from BGP neighbors.
In the presence of such varying origin ASes, it would be necessary for the resource holder to register Route Origin Authorizations (ROAs) for each potential origin AS that may result from the expected aggregated AS_PATHs.

In order to ensure a consistent BGP origin AS is announced for aggregate BGP routes for implementations of "brief" BGP aggregation, the implementation should be configured to truncate the AS_PATH after the right-most instance of the desired origin AS for the aggregate. If the resulting AS_PATH would be truncated from the otherwise expected result of BGP AS_PATH aggregation (an AS_SET would be generated, or ASes are removed from the "longest leading sequence" of ASes), the ATOMIC_AGGREGATE Path Attribute SHALL be attached. This is consistent with the intent of Section 5.1.6 of .

When aggregating prefixes, network operators MUST use brief aggregation. In brief aggregation, the AGGREGATOR attribute is included but the AS_SET attribute is not included. When doing the above, operators MUST form the aggregate at the border in the outbound BGP policy and omit any prefixes from the AS that the aggregate is being advertised to. In other words, an aggregate prefix MUST NOT be announced to the contributing ASes. Instead, more specific prefixes (from the aggregate) MUST be announced to each contributing AS, excluding any that were learned from the contributing AS in consideration. For illustration, if p1/24 (from AS1), p2/24 (from AS2), p3/24 (from AS3) and p4/24 (from AS4) are aggregated to p/22, then p/22 will not be announced to AS1, AS2, AS3, or AS4. Instead, as further illustration, p1/24, p2/24 and p4/24 are announced to AS3. Or, possibly q/23 (aggregate of p1/24 and p2/24) and p4/24 are announced to AS3. Operators MUST install egress filters to block data packets when the destination address belongs to an internal prefix. Similarly, any known single-homed customer prefix MUST also be included in the egress filters except on the interface for that customer. This mitigates looping in the data plane when connection to such an internal or customer prefix is lost. This mechanism effectively compensates for the lack of the additional loop detection capability accorded by AS_SETs (if they were allowed).

This document obsoletes the use of aggregation techniques that create AS_SETs. Obsoleting these path segment types from BGP and removal of the related code from implementations would potentially decrease the attack surface for BGP. Deployments of new BGP security technologies (, , ) benefit greatly if AS_SETs are not used in BGP.

This document requires no IANA actions.

The authors would like to thank John Heasley, Job Snijders, Jared Mauch, Jakob Heitz, Keyur Patel, Douglas Montgomery, Randy Bush, Susan Hares, John Scudder, Curtis Villamizar, Danny McPherson, Chris Morrow, Tom Petch, Ilya Varlashkin, Enke Chen, Tony Li, Florian Weimer, John Leslie, Paul Jakma, Rob Austein, Russ Housley, Sandra Murphy, Steve Bellovin, Steve Kent, Steve Padgett, Alfred Hoenes, and Alvaro Retana for comments and suggestions.