BGP-4 Path Attribute Filtering Capability

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶

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."¶

This Internet-Draft will expire on 21 February 2026.¶

Copyright Notice

Copyright (c) 2025 IETF Trust and the persons identified as the document authors. All rights reserved.¶

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶

Table of Contents

1. Introduction

A BGP Route (Section 1.1 of [RFC4271]) is a tuple consisting of a set of Path Attributes (Section 5 of [RFC4271]) and sets of network reachability carried as Network Layer Reachability Information (NLRI). Some of these Path Attributes are defined as part of the core BGP-4 protocol. Path Attributes are the main extension mechanism defined by BGP, and may be scoped as "transitive" or "non- transitive."¶

Non-Transitive Path Attributes require the BGP speaker to understand the attribute in order to determine if it will be locally used and perhaps later propagated to additional BGP speakers. Unrecognized non-transitive Path Attributes are discarded by the receiving BGP speaker.¶

Transitive Path Attributes, when not understood by the receiving BGP speaker, are required to be propagated to other BGP speakers.¶

Some Path Attributes defined by BGP extensions are intended to be used in limited scopes, such as a single BGP Autonomous System (AS). When such attributes are distributed beyond the expected scope, this is called an "Attribute Escape" [I-D.haas-idr-bgp-attribute-escape].¶

Such attribute escapes may lead to improper BGP protocol behavior when received outside of their expected scope, and may lead to incorrect forwarding, or be a serious security consideration.¶

This document defines a mechanism exchanged through BGP Capabilities [RFC5492] where BGP speakers can more appropriately scope both Path Attributes to prevent escape, and to limit the distribution of routes that carry escaped Path Attributes.¶

2. BGP Path Attribute Filtering Capability

The BGP Path Attribute Filtering Capability is encoded as follows:¶

• Capability Code of (TBD).¶

• Capability Length of 0..32 octets.¶

• Capability Value contains a bit-string padded to an octet boundary where a bit is set if this BGP speaker considers the corresponding BGP Path Attribute from the remote BGP speaker to be unwanted. Bit number 0 is the most significant bit of the first octet, bit number 1 is the second most significant bit of the first octet, and so on.¶

Bits 1, 2, 3, 6, 7, 14, 15, 17, 18 MUST be clear (value 0), because support for [RFC4271], [RFC4760], and [RFC6793] procedures are required when this specification is in use.¶

Any bit not explicitly represented (e.g., bits 24 and beyond in the above example) is deemed to be clear (value 0). That is, the default is to accept any path attribute not explicitly unwanted.¶

3. Operation

A clear (value 0) bit in the Path Attribute Filtering capability indicates that the BGP speaker advertising it is willing to accept the corresponding Path Attribute and will process it according to the normal rules of the BGP protocol and the attribute in question.¶

A set (value 1) bit in the Path Attribute Filtering capability indicates that the BGP speaker advertising it is not willing to accept the corresponding Path Attribute. We refer to such Path Attributes as "unwanted Path Attributes".¶

A BGP speaker MUST NOT send an unwanted Path Attribute to its peer. (Of course, this expectation will be met only by BGP speakers that support this specification; therefore a BGP speaker that implements this specification SHOULD be prepared for the possibility it will receive unwanted Path Attributes; this is discussed below.)¶

One strategy to accomplish the above requirement is for the BGP speaker to not advertise BGP routes containing the unwanted Path Attribute in question. This might require a withdraw to be sent instead. This is similar to treat-as-withdraw as defined in [RFC7606].¶

Another strategy that could be used, when appropriate for the procedure covering a given BGP Path Attribute, is for the BGP speaker to remove the unwanted Path Attributes when it distributes the route to the remote BGP speaker. This is similar to the Attribute Discard behavior defined in [RFC7606].¶

Receiving BGP speakers SHOULD filter routes or discard unwanted Path Attributes if they are incorrectly sent by the remote BGP speaker. Minimally, a receiving BGP speaker receiving an unwanted Path Attribute SHOULD use treat-as-withdraw procedures. Receiving BGP speakers MAY accept the route and discard the unwanted Path Attribute if permitted to by local configuration.¶

4. Selecting supported Path Attributes

Implementations MUST, as described in Figure 1, clear the bits covering required core eBGP Path Attributes.¶

Common BGP features that are defined for Internet use SHOULD be clear by default between two BGP speakers. These include:¶

• Communities (8)¶

• Extended Communities (16)¶

• Large BGP Communities (32)¶

• BGPsec_Path (33)¶

• Only To Customer (OTC) (35)¶

BGP features required to support a given AFI/SAFI MUST also be clear when that address family is configured. An example of this is the BGP-LS attribute (29) when the BGP-LS feature is in use.¶

5. Error Handling

If the received Capability Length for the Path Attributes Filtering Capability is greater than 32, the filtering capability MUST be ignored and treated as if not received by the BGP speaker.¶

To support core BGP features, Section 2 requires that bits 1, 2, 3, 6, 7, 14, 15, 17, 18 be clear (value 0). A BGP speaker receiving a Path Attribute Filtering Capability with these bits set (value 1) MAY reject the BGP session. If it does so, it uses a NOTIFICATION message with an error subcode of "Unsupported Capability".¶

6. Operational Considerations

7. IANA Considerations

This document requests a new BGP Capability Code to be allocated from the First Come First Served range of the Capability Codes registry. The description should be "Path Attribute Filtering", and the reference should be this document.¶

This document requests that the "BGP Path Attributes" registry be updated to add two new columns, called "Should Filter By Default" and "Filtering Profile". These columns are to be seeded with the values shown in Table 1. This document should be added as a reference for the registry. Authors seeking guidance for how to populate these columns should refer to Section 6.3 and Section 10.¶

8. Security Considerations

The motivation for this feature is to attempt to address the numerous BGP security implications where BGP Path Attributes propagate beyond their intended scope.¶

The definition of a feature that limits the distribution of BGP Path Attributes unfortunately moves BGP's default behavior away from "distribute unknown things easily" and thus hampers incremental deployment of new features. However, operators have already begun indiscriminate filtering of Path Attributes they do not themselves require. This feature attempts to provide a more flexible negotiated mode to permit such filtering while at the same time not completely precluding incremental deployment of new features.¶

9. Recommendations for Future BGP Extensions

When a new BGP extension defines a new BGP Path Attribute, the specifying document MUST define its expected filtering profile for routes containing the new Path Attribute. The following are suggested propagation policies for new features:¶

• Default deny: Without explicit configuration from the sending BGP speaker and the receiving BGP speaker, routes containing the new Path Attribute are considered unwanted. Implementations supporting the Path Attribute Filtering Capability SHOULD set the bit for this Path Attribute to 1 (unwanted) by default. The sending BGP speaker MUST NOT advertise routes containing such a Path Attribute to the remote BGP speaker. BGP speakers receiving routes containing the unwanted Path Attribute SHOULD NOT accept the route into its Adj-Rib-In for that peer (treat-as-withdraw).¶

• Default discard: Without explicit configuration from the sending BGP speaker and the receiving BGP speaker, routes containing the new Path Attribute are considered unwanted. Implementations supporting the Path Attribute Filtering Capability SHOULD set the bit of this Path Attribute to 1 (unwanted) by default.¶

  The sending BGP speaker MUST use one of the following two options to avoid sending the unwanted Path Attribute: it can either send the route to the remote speaker after discarding the offending Path Attribute (attribute discard) or it can avoid sending the route to such a peer.¶

  The receiving BGP speaker SHOULD NOT accept routes with the offending Path Attribute into its Adj-Rib-In for that peer (treat-as-withdraw) or MAY discard the offending Path Attribute (attribute-discard). The choice of the desired behavior to discard or to not propagate will depend on the operational expectations for the feature.¶

• AFI/SAFI conditional: When the specified AFI/SAFI is configured for the BGP session and the Path Attribute Filtering Capability is supported, the filtering capability bit SHOULD be set to 0 (wanted) by default. When that AFI/SAFI is not configured, the defining specification should detail whether the default filtering policy is wanted and the filtering capability bit for that Path Attribute MUST be set accordingly.¶

• Default permit: Without explicit configuration from the sending BGP speaker and the receiving BGP speaker, routes containing the new Path Attribute are considered wanted. Implementations supporting the Path Attribute Filtering Capability SHOULD set the bit of this Path Attribute to 0 (wanted) by default.¶

10. Recommendations for Filtering Path Attributes

The table below provides suggested default filtering behaviors for many BGP Path Attributes. In addition to "yes" and "no", which indicate that an implementation SHOULD either filter, or not filter, the given attribute by default, "never" indicates that an implementation MUST NOT filter the given attribute under any circumstances, and "-" indicates that no recommendation is made.¶

The table also recommends filtering profiles, as defined in Section 9. Where no filtering recommendation is made, no profile is suggested either.¶

It is not recommended that BGP Path Attributes that are deprecated should be filtered by default. This permits their long term reassignment and re-use. Operators with a strong filtering policy may consider filtering these to block routes from older implementations of these features that may still be deployed.¶

Upon publication of this document as an RFC, this table will become historic and the authoritative source of recommendations will be the relevant IANA registry (Section 7).¶

Acknowledgements

The authors would like to thank the following individuals whose comments contributed to the refinement of this document: Donatas Abraitis, Ignas Bagdonas, Bruno Decraene, David Lamparter, Robert Raszuk.¶

References

Authors' Addresses