| Internet-Draft | Link-Local Next Hop Capability for BGP | December 2021 |
| White & Abraitis | Expires 10 June 2022 | [Page] |
BGP, described in [RFC4271], was originally designed to provide reachability between domains and between the edges of a domain. As such, BGP assumes the next hop towards any reachable destination may not reside on the advertising speaker, but rather may either be through a router connected to the same subnet as the speaker, or through a router only reachable by traversing multiple hops through the network. Because of this, BGP does not recognize the use of IPv6 link-local addresses, as described in [RFC4291], as a valid next hop for forwarding purposes.¶
However, BGP speakers are now often deployed on point-to-point links in networks where multihop reachability of any kind is not assumed or desired (all next hops are assumed to be the speaker reachable through a directly connected point-to-point link). This is common, for instance, in data center fabrics. In these situations, a global IPv6 address is not required for the advertisement of reachability information; in fact, providing global IPv6 addresses in these kinds of networks can be detrimental to Zero Touch Provisioning (ZTP).¶
This draft standardizes the operation of BGP over a point-to-point link using link-local IPv6 addressing only.¶
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 4 June 2022.¶
Copyright (c) 2021 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.¶
BGP, described in [RFC4271], was originally designed to provide reachability between domains and between the edges of a domain. As such, BGP assumes the next hop towards any reachable destination may not reside on the advertising speaker, but rather may either be through a router connected to the same subnet as the speaker, or through a router only reachable by traversing multiple hops through the network. Because of this, BGP does not recognize the use of IPv6 link-local addresses, as described in [RFC4271], as a valid next hop for forwarding purposes.¶
However, BGP speakers are now often deployed on point-to-point links in networks where multihop reachability of any kind is not assumed or desired (all next hops are assumed to be the speaker reachable through a directly connected point-to-point link). This is common, for instance, in data center fabrics. In these situations, a global IPv6 address is not required for the advertisement of reachability information; in fact, providing global IPv6 addresses in these kinds of networks can be detrimental to Zero Touch Provisioning (ZTP).¶
Such BGP deployment models require BGP to run on each link, and any ease or simplification of BGP configuration can result in simplifying orchestration and configuration management. This proposal is a step in that direction.¶
With the requirement of any global interface address being removed by this new capability, BGP neighbor configuration can be further simplified by making it (look) address-family independent. E.g. BGP can just take the interface name for the peer config and link-local IPv6 address of the peer can be learned via a discovery protocol running on the link or by an out-of-band tool. In essence, link-local next hop in combination with [RFC5549] makes it possible to achieve an unnumbered interface-like solution [RFC5309] in BGP.¶
The Link-Local Next Hop capability is a new BGP capability. A BGP speaker that supports capabilities advertisement [RFC5492] in an OPEN message should send this capability only when:¶
The presence of this capability does not affect the support of global IPv6 only (16 bytes next hop) and global IPv6 combined with link-local IPv6 (32 bytes next hop), which should continue to be supported as before. The Capability Code for this capability is TBA (based on the procedure described in the IANA Considerations section of this document). The Capability Length field of this capability is 0.¶
The advantage of using this capability is that it can let two conforming implementations interoperate [correctly] without additional configuration, in contrast to the current situation. Existing implementations of using a BGP next hop over an IPv6 link-local address are [inconsistent], and can't readily change their behavior without negative side effects.¶
A BGP speaker that is willing to use (send and receive) only link-local addresses as next hops with a peer SHOULD advertise the Link-Local Next Hop Capability to the peer using BGP Capabilities advertisement.¶
The peers have the flexibility to include both link-local and global next hops or link-local only next hop.¶
[RFC2545], section 2, notes link-local IPv6 addresses are not generally suitable for use in the Next Hop field of the MP_REACH_NLRI. In order to support the many uses of link-local addresses, however, [RFC2545] constructs the Next Hop field in IPv6 route advertisements by setting the length of the field to 32, and including both a link-local and global IPv6 address in the resulting enlarged field. In this way, the receiving BGP speaker can use the global IPv6 address to build local forwarding information, and the link-local address for ICMPv6 redirects, etc. [RFC2545] does not, however, provide an explanation for situations where there is only a link-local IPv6 address in the Next Hop field of the MP_REACH_NLRI. The result is each implementation that supports link-local peering along with forwarding to a link-local address has implemented the construction of the Next Hop field in the MP_REACH_NLRI when there is only a link-local address available in slightly different ways.¶
If an implementation intends to send a single link-local forwarding address in the Next Hop field of the MP_REACH_NLRI, it MUST set the length of the Next Hop field to 16 and include only the IPv6 link-local address in the Next Hop field.¶
If an implementation intends to send both a link-local and global IPv6 forwarding address in the Next Hop field of the MP_REACH_NLRI, it MUST set the length of the Next Hop field to 32 and include both the IPv6 link-local and global IPv6 forwarding addresses in the Next Hop field. If both link local and global IPv6 forwarding addresses are carried in the Next Hop Field, the speaker SHOULD provide a local configuration option to determine which address is preferred for forwarding.¶
For internal BGP peers configured as a route-reflector, when route-reflector isn't configured to be in the data-path, the proposed link-local (only) next hops MUST not be reflected.¶
A single (only) link-local next hop address needs to always be reset as next hop self when passed to another link.¶
On receiving an MP_REACH_NLRI with a Next Hop length of 16, implementations SHOULD form the forwarding information using the IPv6 next hop contained in the Next Hop field, regardless of whether it is a link-local or globally reachable IPv6 address.¶
Implementations MAY check the validity of any IPv6 link-local address used to calculate forwarding information by insuring the address is in the local neighbor table for the interface on which the BGP update was received (or through which the BGP speaker from which the update was received is reachable). There MUST be a configuration option to enable/disable this check.¶
Note: It is possible that checking the IPv6 neighbor table for the existence or validity of a link-local next hop may make instances where a link is being overwhelmed through some form of Denial of Service (DoS) attack worse than they would otherwise be. If the IPv6 neighbor cache is overrun in a way that causes the link-local address being used for BGP peering to be removed from the table, which is possible through an on-link DoS attack, any fresh BGP update will cause the entire peering session to fail if the implementation is checking the validity of link-local next hops as described above. Operators should carefully assess the use of validation against the local IPv6 neighbor table to determine if it is appropriate for any particular peering session.¶
A BGP speaker receiving an MP_REACH_NLRI with the length of the Next Hop Field set to 32, where the update contains anything other than a link-local IPv6 address and a global IPv6 address, SHOULD consider this a malformed UPDATE message, and proceed as described in the following paragraphs. In order to support backward compatibility with existing implementations, an implementation MAY ignore a second link-local IPv6 address or 0::0/0 included with an IPv6 link-local address when the length of the Next Hop Field is set to 32; in this case, the implementation SHOULD report the existence of this additional information so the operator can correct the sending BGP implementation.¶
If the Next Hop field is malformed, the implementation MUST handle the malformed UPDATE message using the approach of "treat-as-withdraw", as described in section 7.3 of [RFC7606]. It MAY send a NOTIFICATION message as described in section 4 of [RFC4271], using the UPDATE error message code (8 - Invalid NEXT_HOP Attribute) indicating there is an invalid NEXT_HOP field¶
If the Next Hop field is properly formed, but the link-local next hop is not reachable (as determined by an examination of the IPv6 neighbor table), the implementation MAY handle the malformed UPDATE message using the approach of "treat-as-withdraw", as described in section 7.3 of [RFC7606] (see the note above on checking the local neighbor table for the correctness of the next hop). The implementation MAY send a NOTIFICATION message as described in section 4 of [RFC4271] using the UPDATE error message code (TBA), indicating a link-local address was included in the MP_REACH_NLRI, but the link-local address included cannot be reached. As this could indicate a security breach of some type (see the security considerations section below), the operator SHOULD have a local configuration option to terminate the peering session until manual intervention is initiated.¶
The authors would like to thank Vipin Kumar, Dinesh Dutt, Jeff Haas, and for their contributions to this draft.¶
This memo requests IANA assign a number from the "Error Subcodes" registry defined in the IANA Considerations section in [RFC4271]. This allocation will be for a new UPDATE error subcode, code (TBA), with a value of "Unreachable Link-Local Address."¶
Also, IANA is requested to assign a capability number to the same.¶
The mechanism described in this draft can be used as a component of ZTP for building BGP adjacencies across point-to-point links. This method, then, can be used by an attacker to form a peering session with a BGP speaker, ultimately advertising incorrect routing information into a routing domain in order to misdirect traffic or cause a denial of service. By using link-local IPv6 addresses, the attacker would be able to forego the use of a valid IPv6 address within the domain, making such an attack easier.¶
Operators SHOULD carefully consider security when deploying link-local addresses for BGP peering. Operators SHOULD filter traffic on links where BGP peering is not intended to occur to prevent speakers from accepting BGP session requests, as well as other mechanisms described in [RFC7454].¶
Operators MAY also use some form of cryptographic validation on links within the network to prevent unauthorized devices from forming BGP peering sessions. Authentication, such as the TCP authentication described in [RFC5925], may provide some relief if it is present and correctly configured. However, the distribution and management of keys in an environment where global addresses are not present on BGP speakers may be challenging.¶
Operators also MAY instruct a BGP peer which has received an UPDATE with an unreachable NEXT_HOP to disable the peering session over which the invalid NEXT_HOP was received pending manual intervention.¶