Segment Routing Prefix Segment Identifier Extensions for BGP

Introduction The Segment Routing (SR) architecture leverages the source-routing paradigm. A segment represents either a topological instruction, such as "go to prefix P following shortest path", or a service instruction. Other types of segments may be defined in the future. A segment is identified through a Segment Identifier (SID). An "SR domain" is defined as a single administrative domain for global SID assignment. It may be comprised of a single Autonomous System (AS) or multiple ASes under consolidated global SID administration. Typically, the ingress node of the SR domain prepends an SR header containing SIDs to an incoming packet. As described in , when SR is applied to the MPLS data plane (), the SID consists of a label. also describes how Segment Routing can be applied to an IPv6 data plane (SRv6) using an IPv6 routing header containing a stack of SR SIDs encoded as IPv6 addresses . The applicability and support for Segment Routing over IPv6 is beyond the scope of this document. A BGP Prefix Segment is a BGP prefix with a Prefix-SID attached. A BGP Prefix-SID is always a global SID () within the SR domain and identifies an instruction to forward the packet over the Equal-Cost Multipath (ECMP) best path computed by BGP to the related prefix. The BGP Prefix-SID is the identifier of the BGP Prefix Segment. In this document, we always refer to the BGP Prefix Segment by the BGP Prefix-SID. This document describes the BGP extensions to signal the BGP Prefix-SID. Specifically, this document defines a BGP attribute known as the "BGP Prefix-SID attribute" and specifies the rules to originate, receive, and handle error conditions for the attribute. The BGP Prefix-SID attribute defined in this document can be attached to prefixes from Multiprotocol BGP IPv4/IPv6 Labeled Unicast ( ). Usage of the BGP Prefix-SID attribute for other Address Family Identifier (AFI) / Subsequent Address Family Identifier (SAFI) combinations is not defined herein but may be specified in future specifications. describes example use cases where the BGP Prefix-SID is used for the above AFI/SAFI combinations. It should be noted that:

A BGP Prefix-SID will be global across ASes when the interconnected ASes are part of the same SR domain. Alternatively, when interconnecting ASes, the ASBRs of each domain will have to handle the advertisement of unique SIDs. The mechanisms for such interconnection are outside the scope of the protocol extensions defined in this document.
A BGP Prefix-SID MAY be attached to a BGP prefix. This implies that each prefix is advertised individually, reducing the ability to pack BGP advertisements (when sharing common attributes).

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.

MPLS BGP Prefix-SID The BGP Prefix-SID is realized on the MPLS data plane () in the following way:

The operator assigns a globally unique label index, L_I, to a locally originated prefix of a BGP speaker N, which is advertised to all other BGP speakers in the SR domain.
According to , each BGP speaker is configured with a label block called the Segment Routing Global Block (SRGB). While recommends using the same SRGB across all the nodes within the SR domain, the SRGB of a node is a local property and could be different on different speakers. The drawbacks of the use case where BGP speakers have different SRGBs are documented in and .
If traffic engineering within the SR domain is required, each node may also be required to advertise topological information and Peer SIDs for each of its links and peers. This information is required to perform the explicit path computation and to express an explicit path as a list of SIDs. The advertisement of topological information and peer segments (Peer SIDs) is done through .
If a prefix segment is to be included in an MPLS label stack, e.g., for traffic-engineering purposes, knowledge of the prefix originator's SRGB is required in order to compute the local label used by the originator.
This document assumes that Border Gateway Protocol - Link State (BGP-LS) is the preferred method for a collecting both peer segments (Peer SIDs) and SRGB information through , , and . However, as an optional alternative for the advertisement of the local SRGB without the topology or the peer SIDs and, therefore, without applicability for TE, the Originator SRGB TLV of the BGP Prefix-SID attribute is specified in of this document.
A BGP speaker will derive its local MPLS label L from the label index L_I and its local SRGB as described in . The BGP speaker then programs the MPLS label L in its MPLS data plane as its incoming/local label for the prefix. See for more details.
The outgoing label for the prefix is found in the Network Layer Reachability Information (NLRI) of the Multiprotocol BGP IPv4/IPv6 Labeled Unicast prefix advertisement as defined in . The label index L_I is only used as a hint to derive the local/incoming label.
of this document specifies the Label-Index TLV of the BGP Prefix-SID attribute; this TLV can be used to advertise the label index for a given prefix.

BGP Prefix-SID Attribute The BGP Prefix-SID attribute is an optional, transitive BGP path attribute. The attribute type code 40 has been assigned by IANA (see ). The BGP Prefix-SID attribute is defined here to be a set of elements encoded as "Type/Length/Value" tuples (i.e., a set of TLVs). All BGP Prefix-SID attribute TLVs will start with a 1-octet type and a 2-octet length. The following TLVs are defined in this document:

Label-Index TLV
Originator SRGB TLV

The Label-Index and Originator SRGB TLVs are used only when SR is applied to the MPLS data plane. For future extensibility, unknown TLVs MUST be ignored and propagated unmodified.

Label-Index TLV The Label-Index TLV MUST be present in the BGP Prefix-SID attribute attached to IPv4/IPv6 Labeled Unicast prefixes (). It MUST be ignored when received for other BGP AFI/SAFI combinations. The Label-Index TLV has the following format: where:

Type:

1

Length:
7, the total length in octets of the value portion of the TLV.

RESERVED:
8-bit field. It MUST be clear on transmission and MUST be ignored on reception.

Flags:
16 bits of flags. None are defined by this document. The Flags field MUST be clear on transmission and MUST be ignored on reception.

Label Index:
32-bit value representing the index value in the SRGB space.

Originator SRGB TLV The Originator SRGB TLV is an optional TLV and has the following format: where:

Type:
3

Length:
The total length in octets of the value portion of the TLV: 2 + (non-zero multiple of 6).

Flags:
16 bits of flags. None are defined in this document. The Flags field MUST be clear on transmission and MUST be ignored on reception.

SRGB:
3 octets specifying the first label in the range followed by 3 octets specifying the number of labels in the range. Note that the SRGB field MAY appear multiple times. If the SRGB field appears multiple times, the SRGB consists of multiple ranges that are concatenated.

The Originator SRGB TLV contains the SRGB of the node originating the prefix to which the BGP Prefix-SID is attached. The Originator SRGB TLV MUST NOT be changed during the propagation of the BGP update. It is used to build SR policies when different SRGBs are used in the fabric, for example, . Examples of how the receiving routers concatenate the ranges and build their neighbor's Segment Routing Global Block (SRGB) are included in . The Originator SRGB TLV may only appear in a BGP Prefix-SID attribute attached to IPv4/IPv6 Labeled Unicast prefixes (). It MUST be ignored when received for other BGP AFI/SAFI combinations. Since the Label-Index TLV is required for IPv4/IPv6 prefix applicability, the Originator SRGB TLV will be ignored if it is not specified in a manner consistent with . If a BGP speaker receives a node's SRGB as an attribute of the BGP-LS Node NLRI and the BGP speaker also receives the same node's SRGB in a BGP Prefix-SID attribute, then the received values should be the same. If the values are different, the values advertised in the BGP-LS NLRI SHOULD be preferred, and an error should be logged.

Receiving BGP Prefix-SID Attribute A BGP speaker receiving a BGP Prefix-SID attribute from an External BGP (EBGP) neighbor residing outside the boundaries of the SR domain MUST discard the attribute unless it is configured to accept the attribute from the EBGP neighbor. A BGP speaker SHOULD log an error for further analysis when discarding an attribute.

MPLS Data Plane: Labeled Unicast A BGP session supporting the Multiprotocol BGP Labeled IPv4 or IPv6 Unicast () AFI/SAFI is required. When the BGP Prefix-SID attribute is attached to a BGP Labeled IPv4 or IPv6 Unicast AFI/SAFI, it MUST contain the Label-Index TLV and MAY contain the Originator SRGB TLV. A BGP Prefix-SID attribute received without a Label-Index TLV MUST be considered to be "invalid" by the receiving speaker. The label index provides guidance to the receiving BGP speaker as to the incoming label that SHOULD be allocated to the prefix. A BGP speaker may be locally configured with an SRGB=[SRGB_Start, SRGB_End]. The preferred method for deriving the SRGB is a matter of local node configuration. The mechanisms through which a given label-index value is assigned to a given prefix are outside the scope of this document. Given a label index L_I, we refer to (L = L_I + SRGB_Start) as the derived label. A BGP Prefix-SID attribute is designated "conflicting" for a speaker M if the derived label value L lies outside the SRGB configured on M. Otherwise, the Label-Index TLV is designated "acceptable" to speaker M. If multiple different prefixes are received with the same label index, all of the different prefixes MUST have their BGP Prefix-SID attribute considered to be "conflicting". If multiple valid paths for the same prefix are received from multiple BGP speakers or, in the case of , from the same BGP speaker, and the BGP Prefix-SID attributes do not contain the same label index, then the label index from the best path BGP Prefix-SID attribute SHOULD be chosen with a notable exception being when is being used to dampen route changes. When a BGP speaker receives a path from a neighbor with an "acceptable" BGP Prefix-SID attribute and that path is selected as the best path, it SHOULD program the derived label as the label for the prefix in its local MPLS data plane. When a BGP speaker receives a path from a neighbor with an "invalid" or "conflicting" BGP Prefix-SID attribute, or when a BGP speaker receives a path from a neighbor with a BGP Prefix-SID attribute but is unable to process it (e.g., local policy disables the functionality), it MUST ignore the BGP Prefix-SID attribute. For the purposes of label allocation, a BGP speaker MUST assign a local (also called dynamic) label (non-SRGB) for such a prefix as per classic Multiprotocol BGP IPv4/IPv6 Labeled Unicast () operation. In the case of an "invalid" BGP Prefix-SID attribute, a BGP speaker MUST follow the error-handling rules specified in . A BGP speaker SHOULD log an error for further analysis. In the case of a "conflicting" BGP Prefix-SID attribute, a BGP speaker SHOULD NOT treat it as an error and SHOULD propagate the attribute unchanged. A BGP speaker SHOULD log a warning for further analysis, i.e., in the case the conflict is not due to a label-index transition. When a BGP Prefix-SID attribute changes and transitions from "conflicting" to "acceptable", the BGP Prefix-SID attributes for other prefixes may also transition to "acceptable" as well. Implementations SHOULD ensure all impacted prefixes revert to using the label indices corresponding to these newly "acceptable" BGP Prefix-SID attributes. The outgoing label is always programmed as per classic Multiprotocol BGP IPv4/IPv6 Labeled Unicast () operation. Specifically, a BGP speaker receiving a prefix with a BGP Prefix-SID attribute and a label NLRI field of Implicit NULL from a neighbor MUST adhere to standard behavior and program its MPLS data plane to pop the top label when forwarding traffic to the prefix. The label NLRI defines the outbound label that MUST be used by the receiving node.

Advertising BGP Prefix-SID Attribute The BGP Prefix-SID attribute MAY be attached to BGP IPv4/IPv6 Labeled Unicast prefixes . In order to prevent distribution of the BGP Prefix-SID attribute beyond its intended scope of applicability, attribute filtering SHOULD be deployed to remove the BGP Prefix-SID attribute at the administrative boundary of the SR domain. A BGP speaker that advertises a path received from one of its neighbors SHOULD advertise the BGP Prefix-SID received with the path without modification as long as the BGP Prefix-SID was acceptable. If the path did not come with a BGP Prefix-SID attribute, the speaker MAY attach a BGP Prefix-SID to the path if configured to do so. The content of the TLVs present in the BGP Prefix-SID is determined by the configuration.

MPLS Data Plane: Labeled Unicast A BGP speaker that originates a prefix attaches the BGP Prefix-SID attribute when it advertises the prefix to its neighbors via Multiprotocol BGP IPv4/IPv6 Labeled Unicast (). The value of the label index in the Label-Index TLV is determined by configuration. A BGP speaker that originates a BGP Prefix-SID attribute MAY optionally announce the Originator SRGB TLV along with the mandatory Label-Index TLV. The content of the Originator SRGB TLV is determined by configuration. Since the label-index value must be unique within an SR domain, by default an implementation SHOULD NOT advertise the BGP Prefix-SID attribute outside an AS unless it is explicitly configured to do so. In all cases, the Label field of the advertised NLRI ( ) MUST be set to the local/incoming label programmed in the MPLS data plane for the given advertised prefix. If the prefix is associated with one of the BGP speaker's interfaces, this is the usual MPLS label (such as the Implicit or Explicit NULL label ).

Error Handling of BGP Prefix-SID Attribute When a BGP speaker receives a BGP UPDATE message containing a malformed or invalid BGP Prefix-SID attribute attached to an IPv4/IPv6 Labeled Unicast prefix (), it MUST ignore the received BGP Prefix-SID attribute and not advertise it to other BGP peers. In this context, a malformed BGP Prefix-SID attribute is one that cannot be parsed due to not meeting the minimum attribute length requirement, containing a TLV length that doesn't conform to the length constraints for the TLV, or containing a TLV length that would extend beyond the end of the attribute (as defined by the attribute length). This is equivalent to the "Attribute discard" action specified in . When discarding an attribute, a BGP speaker SHOULD log an error for further analysis. As per , if the BGP Prefix-SID attribute appears more than once in an UPDATE message, all the occurrences of the attribute other than the first one SHALL be discarded and the UPDATE message will continue to be processed. Similarly, if a recognized TLV appears more than once in a BGP Prefix-SID attribute while the specification only allows for a single occurrence, then all the occurrences of the TLV other than the first one SHALL be discarded and the Prefix-SID attribute will continue to be processed. For future extensibility, unknown TLVs MUST be ignored and propagated unmodified.

IANA Considerations This document defines a BGP path attribute known as the BGP Prefix-SID attribute. IANA has assigned attribute code type 40 to the BGP Prefix-SID attribute from the "BGP Path Attributes" registry. This document defines two TLVs for the BGP Prefix-SID attribute. These TLVs have been registered with IANA. IANA has created a registry for BGP Prefix-SID Attribute TLVs as follows: Under the "Border Gateway Protocol (BGP) Parameters" registry, the new registry titled "BGP Prefix-SID TLV Types" has been created and points to this document as the reference. Registration Procedure(s):

Values 1-254, Expert Review as defined in
Values 0 and 255, Reserved

BGP Prefix-SID TLV Types

Value	Type	Reference
0	Reserved	This document
1	Label-Index	This document
2	Deprecated	This document
3	Originator SRGB	This document
4-254	Unassigned
255	Reserved	This document

The value 2 previously corresponded to the IPv6 SID TLV, which was specified in previous versions of this document. It was removed, and use of the BGP Prefix-SID for Segment Routing over the IPv6 data plane has been deferred to future specifications. IANA has also created the "BGP Prefix-SID Label-Index TLV Flags" registry under the "Border Gateway Protocol (BGP) Parameters" registry, with a reference to this document. Initially, this 16-bit flags registry is empty. The registration policy for flag bits is Expert Review , consistent with the "BGP Prefix-SID TLV Types" registry. Finally, IANA has created the "BGP Prefix-SID Originator SRGB TLV Flags" registry under the "Border Gateway Protocol (BGP) Parameters" registry, with a reference to this document. Initially, this 16-bit flags registry is empty. The registration policy for flag bits is Expert Review consistent with the BGP Prefix-SID TLV Types registry. The designated experts must be good and faithful stewards of the above registries, ensuring that each request is legitimate and corresponds to a viable use case. Given the limited number of bits in the flags registries and the applicability to a single TLV, additional scrutiny should be afforded to requests for flag-bit allocation. In general, no single use case should require more than one flag bit and, should the use case require more, alternate encodings using new TLVs should be considered.

Manageability Considerations This document defines a BGP attribute to address use cases such as the one described in . It is assumed that advertisement of the BGP Prefix-SID attribute is controlled by the operator in order to:

Prevent undesired origination/advertisement of the BGP Prefix-SID attribute. By default, a BGP Prefix-SID attribute SHOULD NOT be attached to a prefix and advertised. Hence, BGP Prefix-SID Advertisement SHOULD require explicit enablement.
Prevent any undesired propagation of the BGP Prefix-SID attribute. By default, the BGP Prefix-SID is not advertised outside the boundary of a single SR/administrative domain that may include one or more ASes. The propagation to other ASes MUST be explicitly configured.

The deployment model described in assumes multiple ASes under a common administrative domain. For this use case, the BGP Prefix-SID Advertisement is applicable to the inter-AS context, i.e., EBGP, while it is confined to a single administrative domain.

Security Considerations This document introduces a BGP attribute (BGP Prefix-SID), which inherits the security considerations expressed in: , , and . When advertised using BGPsec as described in , the BGP Prefix-SID attribute doesn't impose any unique security considerations. It should be noted that the BGP Prefix-SID attribute is not protected by the BGPsec signatures. It should be noted that, as described in , this document refers to a deployment model where all nodes are under the single administrative domain. In this context, we assume that the operator doesn't want to leak any information related to internal prefixes and topology outside of the administrative domain. The internal information includes the BGP Prefix-SID. In order to prevent such leaking, the common BGP mechanisms (filters) are applied at the boundary of the SR/administrative domain. Local BGP-attribute-filtering policies and mechanisms are not standardized and, consequently, are beyond the scope of this document. To prevent a Denial-of-Service (DoS) or Distributed-Denial-of-Service (DDoS) attack due to excessive BGP updates with an invalid or conflicting BGP Prefix-SID attribute, error log message rate limiting as well as suppression of duplicate error log messages SHOULD be deployed. Since BGP-LS is the preferred method for advertising SRGB information, the BGP speaker SHOULD log an error if a BGP Prefix-SID attribute is received with SRGB information different from that received as an attribute of the same node's BGP-LS Node NLRI.

References Normative References Segment Routing with the MPLS Data Plane Informative References BGP Prefix Segment in Large-Scale Data Centers

Acknowledgements The authors would like to thank Satya Mohanty for his contribution to this document. The authors would like to thank Alvaro Retana for substantive comments as part of the Routing AD review. The authors would like to thank Bruno Decraene for substantive comments and suggested text as part of the Routing Directorate review. The authors would like to thank Shyam Sethuram for comments and discussion of TLV processing and validation. The authors would like to thank Robert Raszuk for comments and suggestions regarding the MPLS data-plane behavior. The authors would like to thank Krishna Deevi, Juan Alcaide, Howard Yang, and Jakob Heitz for discussions on conflicting BGP Prefix-SID label indices and BGP add paths. The authors would like to thank Peter Yee, Tony Przygienda, Mirja Kuhlewind, Alexey Melnikov, Eric Rescorla, Suresh Krishnan, Warren Kumari, Ben Campbell Sue Hares, and Martin Vigoureux for IDR Working Group last call, IETF Last Call, directorate, and IESG reviews.

Contributors