Traffic Steering using BGP FlowSpec with SR Policy

Segment Routing (SR) leverages the source routing paradigm for both SR-MPLS and SRv6 . An SR Policy defines an explicit ordering of segments to steer traffic through a network, which may be instantiated locally or signaled via BGP SR Policy or PCEP for SR Policy . BGP Flow Specification (FlowSpec) enables the dissemination of traffic filtering rules and associated actions using MP_REACH_NLRI / MP_UNREACH_NLRI and Extended Community attributes . This document specifies BGP FlowSpec mechanisms to steer matching traffic flows into an SR Policy . It defines two operational modes: Transport Steering Only (Mode 1): Steers traffic into a transport SR Policy toward a target Endpoint. Steering with Egress Service Action (Mode 2): Steers traffic into an SRv6 Policy while conveying an Egress Service SID for egress service action processing (e.g., VRF decapsulation). This document does not modify the base Segment Routing architecture or SR Policy constructs , but establishes the BGP FlowSpec interworking semantics for SR Policy steering.

BGP FS: BGP Flow Specification EC: Extended Community LPM: Longest Prefix Match Redirect-to-IP: The FlowSpec Redirect-to-IP Extended Community defined in Mode 1 (Transport Steering Only): Steering traffic into an SR Policy (SR-MPLS or SRv6) toward a target Endpoint solely for transport path selection. Mode 2 (Steering with Egress Service Action): Steering traffic into an SRv6 Policy toward a target Endpoint while executing an Egress Service SID function (e.g., End.DT4/DT6 ) at the egress PE. SID: Segment Identifier USD: Ultimate Segment Decapsulation (Flavor defined in )

BGP FlowSpec NLRI can be combined with BGP Extended Communities and, where applicable, the BGP Prefix-SID Attribute to steer matching traffic into a Segment Routing (SR) Policy in either SR-MPLS or SRv6 data planes. This specification relies on a two-tier control plane architecture: Transport Steering: The Color Extended Community is combined with the FlowSpec Redirect-to-IP Extended Community to form an explicit (Color, Endpoint) tuple. This tuple binds the matching FlowSpec traffic to an SR Policy (SR-MPLS or SRv6) at the head-end router (Ingress PE) . Egress Service Action (SRv6-specific): For SRv6 scenarios requiring a specific egress service action (e.g., decapsulation and table/VRF lookup) at the tail-end router (Egress PE), the BGP Prefix-SID Attribute may be attached to convey an SRv6 Service SID. The procedures in this document apply strictly to IPv4 and IPv6 FlowSpec address families in the global routing context (AFI=1/2, SAFI=133).

Traffic steering specified in this document leverages BGP FlowSpec filtering rules to steer flows into SR Policies . FlowSpec-based steering corresponds to Policy-Based Routing (PBR) steering per Section 8.7 of and does not perform per- destination steering (Section 8.4 of ). Implementations MUST NOT apply per-destination route resolution, BGP multipath path selection, or On-Demand BSID recursion to FlowSpec routes. A BGP FlowSpec sender MUST construct the BGP UPDATE message using one of two operational modes: Mode 1 (Transport Steering Only) or Mode 2 (Steering with Egress Service Action).

When matching traffic requires steering into an SR Policy (SR-MPLS or SRv6) toward an Endpoint without a specific egress service action, Mode 1 steering applies: Sender Behavior: The sender MUST attach a Redirect-to-IP Extended Community and a Color Extended Community , and MUST NOT attach the BGP Prefix-SID Attribute (see Section 4 for attribute validation). Headend Steering: Upon resolving the (Endpoint, Color) tuple per Section 4, the headend steers matching traffic into the resolved SR Policy by encapsulating it solely with the policy's segment list (MPLS label stack or SRH). Egress Behavior: Decapsulation or label popping at the egress PE relies on the final segment behavior of the target SR Policy (e.g., Implicit Null for SR-MPLS or End with USD flavor for SRv6).

When matching traffic requires steering into an SRv6 Policy and a specific service context or decapsulation behavior at the egress PE, Mode 2 steering applies: Sender Behavior: The sender MUST attach a Redirect-to-IP Extended Community, a Color Extended Community , and the BGP Prefix-SID Attribute carrying the Egress Service SID (encoded per Section 3.6). Headend Steering: The headend resolves the target SRv6 Policy using the (Endpoint, Color) tuple per Section 4, and synthesizes the active segment list in the SRH by combining the SR Policy segment list with the Egress Service SID per Section 3.3. Egress Behavior: Upon packet arrival, the egress PE decapsulates the SRH and executes the specific function bound to the Egress Service SID (e.g., End.DT4 or End.DT6 VRF lookup ).

A BGP speaker or controller originating or re-advertising a FlowSpec route to steer traffic into an SR Policy MUST perform the following: Construct the explicit (Color, Endpoint) tuple by attaching the Color Extended Community and setting the target IP address in the Redirect-to-IP Extended Community . For Mode 2 steering (Section 3.1.2), attach the BGP Prefix-SID Attribute carrying the SRv6 Service SID constructed per Section 3.6. For Mode 1 steering (Section 3.1.1), the BGP Prefix-SID Attribute MUST NOT be attached. A BGP speaker re-advertising a FlowSpec route MUST preserve these steering attributes unless local policy explicitly modifies them.

The procedures specified in this section apply exclusively to BGP FlowSpec receivers acting as SR Headends. Intermediate BGP speakers (e.g., Route Reflectors) that do not perform data-plane steering MUST transparently reflect and propagate FlowSpec routes and their attached attributes per without attempting SR Policy binding. Upon receiving a FlowSpec route, the headend performs evaluation and forwarding programming strictly per the sequential logic in Section 4.1: Validation and Mode Determination: Syntactic validation follows and . Co-existing steering attributes dictate whether the route is Mode 1 or Mode 2 per Section 4.1. SR Policy Binding & Synthesis: The headend uses the extracted (Color, Endpoint) tuple to resolve an active SR Policy Candidate Path per Section 8.8.1 of . For Mode 1, traffic is encapsulated using the active Segment List. For Mode 2, the headend synthesizes the Segment List with the Egress Service SID per and Section 3.6. Default Longest Prefix Match (LPM) redirection MUST NOT be applied when an SR Policy is successfully resolved. Non-steering FlowSpec actions in the same route remain applicable. Control Plane Retention & Decoupling: If an active SR Policy cannot be resolved or forwarding installation fails, the headend MUST NOT withdraw or mark the BGP route invalid, provided it passes BGP syntactic checks. The route MUST be retained in Loc-RIB, eligible for best-path selection, and propagated to peers. Data-plane fallback for matching traffic MUST follow Section 3.7.

When a FlowSpec route carries both a Color Extended Community and a Redirect-to-IP Extended Community , the target IP address MUST be interpreted as the target SR Policy Endpoint to form the explicit (Color, Endpoint) tuple . In accordance with Section 2.2.2 of , SR Policy steering defined herein takes precedence over default LPM IP redirection. LPM lookup toward the target IP address MUST NOT be performed when an SR Policy is resolved. If multiple Color Extended Communities are present, color evaluation and fallback follow Section 8.4.1 of .

Steering procedures defined in this document strictly require an explicit (Color, Endpoint) tuple. Color-Only (CO) steering per Section 8.8.1 of is outside the scope of this specification. If a received route contains non-zero CO bits in the Color Extended Community, the receiver MUST treat CO bits as zero and attempt explicit (Color, Endpoint) matching. If no matching SR Policy exists, steering failure procedures (Section 3.7) apply. VPN FlowSpec address families (SAFI=134) and the Redirect-to-VRF Extended Community are outside the scope of this document and MUST NOT be processed under these procedures.

The BGP Prefix-SID Attribute is used strictly for conveying SRv6 Service SIDs in SRv6 Mode 2 (Section 3.1.2) and MUST NOT be attached for SR-MPLS. Encoding rules per and apply: SRv6 Service TLVs: The SRv6 L3 Service TLV (Type 5) MUST be used, containing an SRv6 SID Information Sub-TLV (Type 1). The SRv6 SID Structure Sub-Sub-TLV (Type 1) MAY be included. Endpoint Behaviors: Supported codepoints are restricted to Layer 3 behaviors (End.DT4, End.DT6, End.DT46, End.DX4, End.DX6) . Layer 2 and transit behaviors MUST NOT be used. Disallowance of Transposition: Transposition Schemes MUST NOT be used because FlowSpec NLRIs lack an MPLS Label field. The full SRv6 Service SID MUST be encoded entirely within the SRv6 SID Information Sub-TLV.

Steering failure scenarios are classified into two distinct operational cases: Attribute Incompleteness (Control Plane) and Policy Unavailability (Forwarding Plane).

If a received FlowSpec route contains a Redirect-to-IP Extended Community but lacks a valid Color Extended Community, the steering intent for SR Policy selection is incomplete. In this scenario: The headend MUST NOT attempt to resolve or steer traffic into any default, fallback, or "null-color" SR Policy. The headend MUST fall back to standard BGP FlowSpec Redirect-to-IP forwarding per [I-D.ietf-idr-flowspec-redirect-ip] (i.e., traffic is redirected to the Target IP address using standard IP routing). Any attached BGP Prefix-SID Attribute MUST be ignored for the purpose of SR Policy steering.

If a FlowSpec route contains both a valid Redirect-to-IP Extended Community and a Color Extended Community (expressing explicit SR steering intent), but the corresponding (Endpoint, Color) SR Policy is Down, unresolvable, or not programmed in the forwarding plane: To prevent silent SLA violations and unexpected path leakage, the headend MUST NOT silently fall back to shortest-path IP forwarding toward the Target IP address. Matching traffic MUST be processed according to local administrative failure policy. The default local action SHOULD be to drop the matching traffic (Drop-on-Failure). Other non-steering FlowSpec actions present in the same route (e.g., Traffic-Rate, DSCP remarking) MUST continue to be enforced.

The Redirect-to-IP Extended Community [I-D.ietf-idr-flowspec-redirect-ip] includes a C-bit (Copy Bit) in its Local Administrator field: C-bit = 0 (Direct Redirection): Matching packets MUST be encapsulated and steered into the resolved SR Policy, bypassing normal IP routing. C-bit = 1 (Traffic Copying): The headend MUST duplicate matching packets. The original packet continues along normal IP routing without encapsulation, while the copy is encapsulated and steered into the resolved SR Policy per Section 3.1. Setting C-bit = 1 does not alter SR Policy resolution logic; (Endpoint, Color) resolution and BGP Prefix-SID Attribute processing remain unchanged for the copied stream.

When matching traffic is steered into an SR Policy with multiple active Segment Lists , the headend MUST distribute traffic across active Segment Lists according to their defined weights (ECMP/UCMP). FlowSpec filtering actions direct traffic into the policy as a single aggregate stream and MUST NOT alter Segment List selection. For both SR-MPLS and SRv6, multipath hashing MUST use inner packet fields (e.g., 5-tuple). In SRv6 scenarios, the outer IPv6 Flow Label MUST NOT be overwritten by FlowSpec rules and SHOULD be computed from inner packet fields per Section 5.5 of to prevent hash polarization. If the headend lacks line-rate capabilities to compute the IPv6 Flow Label from inner packet fields, the headend MAY set the Flow Label to zero or a fixed value, provided that entropy-based load balancing is handled by downstream SRv6 transit nodes or alternative path hashing mechanisms.

A BGP speaker supporting base FlowSpec and Redirect-to-IP, but not this specification, processes updates under standard protocol rules: Attribute Pass-Through: As an optional transitive attribute, an unrecognized BGP Prefix-SID Attribute is propagated unmodified to peers per and . Forwarding Fallback: A non-supporting headend ignores unrecognized BGP Prefix-SID Attributes and falls back to default Redirect-to-IP behavior . To prevent unintended forwarding paths due to fallback, implementations MUST support administrative controls to enable or disable advertisement of SRv6 service steering attributes per neighbor and per service . Operators SHOULD apply attribute filtering at domain boundaries per Section 3.2.1 of . An exception to this SHOULD requirement occurs when multiple BGP domains (e.g., multiple ASes) reside within a single administrative control domain where end-to-end SRv6 policy propagation across domain boundaries is explicitly intended. In such controlled deployments, boundary filtering MAY be relaxed, provided that strict ingress filtering remains enforced at the untrusted edge of the single administrative domain.

When a router receives a BGP FlowSpec route carrying a combination of the Redirect-to-IP Extended Community, the Color Extended Community, and the BGP Prefix-SID Attribute, it MUST evaluate these attributes in a sequential pipeline to resolve the steering action.

The headend evaluates received attribute containers step-by-step: Step 1: Redirect-to-IP Extended Community Absent: Procedures in this document do not apply. Process route per standard BGP FlowSpec . Present: Extract target IP as SR Policy Endpoint, process C-bit (Section 3.8), and proceed to Step 2. Step 2: Color Extended Community Absent: The steering intent for SR Policy is incomplete. The headend MUST fall back to standard Redirect-to-IP forwarding per Section 3.7 (i.e., steering into an SR Policy is bypassed, and any received BGP Prefix-SID Attribute is ignored for SR Policy resolution). Present: Extract Color value C to form the (Endpoint, Color) tuple, and proceed to Step 3. Step 3: BGP Prefix-SID Attribute (PSID) Note: This step applies only when the Endpoint is an IPv6 address and SRv6 steering is enabled. Absent (Mode 1): Resolve and steer traffic into the SR Policy corresponding to (Endpoint, Color) per Section 3.1.1. Present (Mode 2): Extract the Egress Service SID from the PSID, resolve the SR Policy via (Endpoint, Color), and synthesize the active segment list per Section 3.1.2 and Section 3.3.

Table 1 summarizes all possible attribute combinations and their corresponding resolution modes after applying the sequential logic above.

To prevent unintentional redirection and SLA violations, implementations MUST strictly enforce the following rules: Handling Invalid Steering Intent: If a route expresses SR steering intent (contains Color EC or Prefix-SID) but required steering attributes are invalid or incomplete, the receiver MUST NOT fall back to shortest-path IP forwarding toward the target IP address. Matching traffic MUST be dropped or processed per local failure policy (Section 3.7). Enforcement of Non-Steering Actions: In all failure cases, other valid non-steering actions present in the FlowSpec route (e.g., Traffic-Rate/Drop , DSCP remarking) MUST continue to be enforced in the forwarding plane. Multiple Redirect-to-IP Communities: When multiple valid Redirect-to-IP Extended Communities are attached alongside a Color Extended Community, load-sharing across the resulting target SR Policies MUST follow EC-level load-sharing defined in Section 2.2 of . Logging and Diagnostics: Implementations SHOULD generate diagnostic log entries whenever an SR Policy steering route contains corrupted or invalid steering attributes. An exception to this SHOULD requirement occurs during high-volume error events or flood conditions (e.g., malformed route injection attacks), where strict per-route logging could exhaust control plane CPU or storage resources. In such scenarios, implementations MAY rate-limit, aggregate, or temporarily suppress individual diagnostic log entries, provided that summary metrics or counter-based telemetry remain available to operators.

When a headend router fails to program a FlowSpec SR Policy steering entry into the forwarding plane (Section 3.3), the implementation SHOULD make relevant diagnostic information available to operators. This includes generating appropriate management notifications (e.g., Syslog messages, SNMP traps, or YANG notifications) detailing the uninstalled FlowSpec NLRI, the associated (Color, Endpoint) tuple, and the specific failure reason (e.g., resource exhaustion). An exception to this SHOULD requirement occurs during rapid sequence failures or resource exhaustion storms, where notification rate- limiting or aggregation MAY be applied to protect control plane stability. Implementations SHOULD monitor forwarding plane resource availability. When sufficient resources become available, the router SHOULD automatically re-attempt to program any previously uninstalled FlowSpec steering entries. An exception occurs if an administrative policy explicitly disables auto-recovery or if rapid retry thrashing is detected; in such cases, implementations MAY apply backoff timers or require manual operator intervention before retrying. Implementations SHOULD also provide hardware or software counter metrics (e.g., byte/packet drop counters) associated with uninstalled or failed FlowSpec steering entries for operational visibility. An exception occurs when underlying hardware forwarding platforms lack granular per-entry counter registers; in such cases, implementations MAY provide aggregated software-level counters instead.

The security considerations of BGP FlowSpec , Redirect-to-IP , BGP Prefix-SID , and core Segment Routing apply to this document. Combining FlowSpec redirection with SR Policy selection and SRv6 Service SIDs introduces specific threat vectors that MUST be mitigated.

The BGP Prefix-SID Attribute in a FlowSpec route conveys an SRv6 Service SID (e.g., End.DT4, End.DT6) that determines the egress VRF or interface at the egress PE. A malicious or compromised speaker could steer traffic into an arbitrary remote VRF, causing cross-VRF data leakage or traffic interception. To mitigate this, receiving PEs MUST validate that any SRv6 Service SID belongs to an authorized Locator range allocated for the target VRF or service context. BGP sessions with FlowSpec controllers MUST be secured using TCP-AO and restricted to trusted entities.

Standard BGP FlowSpec validation validates the NLRI and Redirect-to-IP Extended Community against the origin of the destination prefix, but does not define validation for the Color Extended Community or BGP Prefix-SID Attribute. These unvalidated attributes dictate path selection and service contexts. Headends MUST apply local policy checks to verify that the (Color, Endpoint) tuple resolves to an authorized SR Policy before binding traffic. Operators MUST strip or reject unknown or unauthorized Color Extended Communities and BGP Prefix-SID Attributes from untrusted peers.

SRv6 SIDs and BGP Prefix-SID Attributes are intended strictly for use within a trusted SR domain . Propagation across untrusted EBGP boundaries leaks internal SRv6 Locators and topology, exposing the domain to external steering attacks. Autonomous System Border Routers (ASBRs) MUST filter or strip the BGP Prefix-SID Attribute and SRv6-related Extended Communities from BGP FlowSpec routes traversing administrative boundaries, unless explicit inter-domain trust agreements and policies are enforced.

A single FlowSpec route can match high-volume distributed flows across multiple Ingress PEs and redirect them to a single destination . Binding redirected flows to an SR Policy allows an attacker to pin amplified traffic to a specific engineered path, potentially overloading transit nodes or egress interfaces. Ingress PEs SHOULD combine FlowSpec redirect actions with Traffic Rate Extended Communities to bound redirected bandwidth. An exception to this SHOULD requirement occurs when a FlowSpec route is deployed for legitimate traffic steering and rate limiting would cause unwanted packet drops, provided that adequate capacity planning guarantees sufficient bandwidth along the candidate path. Operators SHOULD monitor path utilization on SR Policies used for FlowSpec steering to detect and mitigate link congestion dynamically. An exception to this SHOULD requirement occurs in centralized control plane architectures where automated Controller-based TE or PCE monitoring dynamically adjusts policy pathing and prevents link congestion automatically.

[Note to the RFC Editor - remove this section before publication, as well as remove the reference to . This section records the status of known implementations of the protocol defined by this specification at the time of posting of this Internet-Draft, and is based on a proposal described in . The description of implementations in this section is intended to assist the IETF in its decision processes in progressing drafts to RFCs. Please note that the listing of any individual implementation here does not imply endorsement by the IETF. Furthermore, no effort has been spent to verify the information presented here that was supplied by IETF contributors. This is not intended as, and must not be construed to be, a catalog of available implementations or their features. Readers are advised to note that other implementations may exist. According to , "this will allow reviewers and working groups to assign due consideration to documents that have the benefit of running code, which may serve as evidence of valuable experimentation and feedback that have made the implemented protocols more mature. It is up to the individual working groups to use this information as they see fit".

The Traffic Steering using BGP FlowSpec with SR-MPLS / SRv6 Policy mechanism has been implemented on the following hardware devices, Network Operating System software, and SDN controllers. They have also successfully participated in a series of joint interoperability testing events hosted by China Mobile from July 2021 to October 2021. The following hardware devices and Network Operating System software passed the interoperability testing (in alphabetical order).

As of August 2022, this feature has been deployed on the IP backbone network of China Mobile. China Mobile has fully transitioned to the SRv6 architecture, with no scenarios where SRv6 and SR-MPLS coexist. As a result, when utilizing Color C and IPv6 redirect addresses, traffic steering is executed exclusively via SRv6 policies. In scenarios where an operator supports the coexistence of SRv6 and SR-MPLS, it is recommended to adhere to to govern policy selection for traffic steering based on Color C and IPv6 endpoints.

This document requires no IANA actions.

The following people made significant contributions to this document:

The authors would like to express special thanks to: Nat Kao for suggesting the inclusion of SR-MPLS use cases and providing detailed recommendations regarding failure and fallback procedures. Ketan Talaulikar for his in-depth review and valuable input, which substantially enhanced the precision and overall quality of this document. Donald E. Eastlake, 3rd for his thorough review and insightful suggestions. The authors also gratefully acknowledge the review, comments, and contributions from Jeffrey Haas, Susan Hares, Keyur Patel, Weiqiang Cheng, Kaliraj Vairavakkalai, Robin Li, Acee Lindem, Gunter Van De Velde, John Scudder, Rainbow Wu, Linda Dunbar, Gang Yan, Feng Yang, Wim Henderickx, Robert Raszuk, Changwang Lin, Aijun Wang, Hao Li, Huaimo Chen, Sheng Fang, Yuanxiang Qiu, Ran Chen, Cheng Li, Zheng Zhang, Xuewei Wang, Yanrong Liang, Xuhui Cai, Haojie Wang, Lili Wang, Nan Geng, Stephane Litkowski, Zhenqiang Li, Jinming Li, Shengnan Yue, and Ziqing Cao.

This section provides illustrative examples for steering BGP FlowSpec traffic into SR Policies in SR-MPLS and SRv6 networks per Section 3 and Section 4.

illustrates steering FlowSpec traffic into an SR-MPLS Policy without egress service actions (Section 3.1.1).

) |Tailend| +-------+ (SR List) +-------+ '--( )--' ( ) '-----' ]]> Control Plane: The BGP FS Controller advertises a FlowSpec route with matching rules, Redirect-to-IP (Tailend Address), and Color C0. The BGP Prefix-SID Attribute is absent. Headend Resolution: The headend resolves the SR-MPLS Policy matching (Endpoint: Tailend, Color: C0). Encapsulation: Matching traffic is encapsulated into the policy's MPLS label stack:

In SRv6 networks, FlowSpec traffic steering operates under either Mode 2 (Transport Steering with Egress Service Action) or Mode 1 (Transport Steering Only).

illustrates signaling an explicit egress service action (Egress_Service_SID) alongside transport steering (Section 3.1.2).

) |Tailend| +-------+ (SR List) +-------+ '--( )--' SRv6 SID: Egress_Service_SID ( ) (e.g., End.DT4, End.DT6, etc.) '-----' ]]> Control Plane & Resolution: The route carries Color C1, Redirect-to-IPv6 (Tailend Address), and BGP Prefix-SID Egress_Service_SID. The headend resolves the SRv6 Policy matching (Endpoint: Tailend, Color: C1) yielding candidate list <S1, S2, S3>. Segment List Synthesis (Section 3.3): Scenario A (Locator Match / Replacement): If Egress_Service_SID shares the SRv6 Locator of S3 (e.g., S3 is an End/End.X SID allocated from the same Locator), S3 is replaced, yielding the synthesized segment list <S1, S2, Egress_Service_SID>. Scenario B (No Locator Match / Appending): Otherwise, Egress_Service_SID is appended, yielding the synthesized segment list <S1, S2, S3, Egress_Service_SID>. Packet Encapsulation Format: At the Headend device, matching traffic is encapsulated into an outer IPv6 header and SRH with the corresponding synthesized segment list (Scenario A or B), as shown below:

(A) | | OR: (B)| +-----------------------------------------------------+ | Inner Payload (Original IPv4/IPv6 Packet) | +-----------------------------------------------------+ ]]> Upon reaching the Tailend device, the packet is decapsulated and processed per the function denoted by Egress_Service_SID (e.g., VRF lookup for End.DT4).

When the final SID S3 instantiated on the Tailend is USD-flavored , an explicit Egress Service SID is omitted.

) |Tailend| +-------+ (SR List) +-------+ '--( )--' ( ) '-----' Note: S3 is an SRv6 SID with USD flavor on the Tailend. ]]> The controller advertises the route with Color C2 and Redirect-to-IPv6 without a BGP Prefix-SID. The headend encapsulates traffic using Segment List <S1, S2, S3>. Upon arrival, the Tailend executes USD behavior, pops the outer IPv6 header/SRH, and performs standard table lookup on the inner packet.

The mechanisms defined in this document are scoped strictly for deployment within a single SR Trusted Domain under unified administrative control. As detailed in Section 6.3, network operators should enforce BGP trust boundaries at domain ingress interfaces to filter unauthorized FlowSpec routes carrying SR-related Extended Communities or BGP Prefix-SID attributes, preventing traffic redirection or label/SID injection outside the trusted domain.