Internet-Draft | BAR-SAV | December 2022 |
Sriram, et al. | Expires 20 June 2023 | [Page] |
Designing an efficient source address validation (SAV) filter requires minimizing false positives (i.e., avoiding dropping legitimate traffic) while maintaining directionality (see RFC8704). This document advances the technology for SAV filter design through a method that makes use of BGP UPDATE messages, Autonomous System Provider Authorization (ASPA), and Route Origin Authorization (ROA). The proposed method's name is abbreviated as BAR-SAV. BAR-SAV can be used by network operators to derive more robust SAV filters and thus improve network resilience.¶
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 20 June 2023.¶
Copyright (c) 2022 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.¶
Spoofed source addresses are often used in Denial of Service (DoS) and Distributed DoS (DDoS) attacks. Source address validation (SAV) filtering is used to drop packets with spoofed source addresses (see BCP 84 [RFC3704] [RFC8704]). A detailed review of unicast Reverse Path Forwarding (uRPF) techniques for SAV is provided in [RFC8704]). Also, [RFC8704] describes enhanced feasible-path uRPF (EFP-uRPF) methods that aim to minimize false positives (i.e., avoid dropping legitimate traffic) while maintaining directionality (see definitions in [RFC3704]).¶
New technology for securing the Border Gateway Protocol (BGP) [RFC4271] using Resource Public Key Infrastructure (RPKI) [RFC6480] is seeing increasing adoption. Two of the currently existing or proposed types of signed objects in the RPKI can be leveraged for a more accurate SAV filter design as well. These are the Route Origin Authorization (ROA) and the Autonomous System Provider Authorizations (ASPA) objects. A ROA is a cryptographically signed attestation by an IP address-resource holder listing their prefixes that are authorized to be originated in BGP by a specific autonomous system (AS) [RFC6482]. ROAs are currently used for RPKI-based Route Origin Validation (RPKI-ROV) [RFC6811] [RFC9319]. An ASPA is a cryptographically signed attestation by an AS listing its transit provider AS numbers (ASNs) [I-D.ietf-sidrops-aspa-profile]. The ASPA data is designed to be used for a form of AS path validation that can detect and mitigate route leaks [I-D.ietf-sidrops-aspa-verification] [sriram1]. See [RFC7908] for the definition of route leaks.¶
This document advances the technology for SAV filter design using methods that make use of ASPA, ROA, and/or BGP UPDATE data. A method is presented in Section 3 that makes use of only ASPA and ROA data to design the SAV filter. This method is for use in the future when the adoption of ROA and ASPA is considered to be ubiquitous. However, for use in the period before that, another method for SAV is presented in Section 4 that makes complementary use of BGP UPDATE messages along with ASPA and ROA data. Accordingly, the latter method's name is abbreviated as BAR-SAV. It is hoped that just as the adoption of ROAs is growing at present [Monitor], the adoption of ASPA will also gain momentum in the near future. The BAR-SAV method additionally incorporates a refined version of Algorithm A of the EFP-uRPF technique (Section 3.1 of [RFC8704]). BAR-SAV can be used by network operators to derive more robust SAV filters and thus improve network resilience.¶
The focus of this document is on the design of ingress SAV filters for an interface facing a customer or lateral peer AS. The same procedure applies in both cases (Section 2).¶
Throughout this document, ROA and ASPA data mean the payload data in cryptographically valid ROA and ASPA objects (see Section 4 in [RFC6482] and Section 4 in [I-D.ietf-sidrops-aspa-profile]).¶
The reader is encouraged to be familiar with [RFC8704], [RFC6482], [RFC6811], [I-D.ietf-sidrops-aspa-profile], and [I-D.ietf-sidrops-aspa-verification].¶
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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
The same procedure applies for the construction of a permissible ingress SAV filter for a customer or lateral peer interface, because the data packets received from a customer or lateral peer should have source addresses belonging only to the prefixes in the customer cone (CC) of said customer or lateral peer. The focus, therefore, is only on the CC of the neighbor in each case. Note that the CC includes the AS belonging to the customer or lateral peer.¶
The procedure (called Procedure X) described in this section is for future scenarios when ASPA and ROA adoption is ubiquitous. In that scenario, robust SAV filters can be generated from the RPKI information (ASPA and ROA data) alone. The procedure is applicable for ingress SAV filter design for customer and lateral peer interfaces. An ISP may use Procedure X on a customer (or lateral peer) interface if it expects full adoption of ROAs and ASPAs in the CC of the neighbor.¶
A description of Procedure X (one that makes use of only ASPA and ROA data):¶
A detailed description of Procedure X is as follows:¶
SAV using BGP UPDATE Messages, ASPA, and ROA (BAR-SAV) is described in this section and is meant for the period when there is a partial deployment of ROAs and ASPAs. To compensate for incomplete RPKI information, BAR-SAV augments ASPA data with BGP UPDATE AS_PATH data for discovering CC ASes, and it augments ROA data with BGP UPDATE data for discovering all prefixes associated with ASes in the CC. The details of this procedure are described below.¶
BAR-SAV additionally incorporates a refined version of Algorithm A of EFP-uRPF (Section 3.1 of [RFC8704]). Algorithm A in [RFC8704] picked only the originating ASes from AS_PATHs received on the customer or lateral peer interface in consideration and included them for SAV filter computation. The variant of Algorithm A in [RFC8704] used here includes all ASes in the AS_PATHs for the SAV filter computation. Unless there is a route leak [RFC7908], each AS is a customer of the AS added next in AS_PATHs of BGP UPDATE messages received from a customer or lateral peer. Further customer-provider AS relations within the CC are discovered by examining all unique ASes in the AS_PATHs in BGP UPDATEs received on all interfaces (from transit providers, customers, lateral peers, and IBGP peers). This is described in the step-by-step procedure later in this section.¶
Note that if a multi-homed AS is present in an above-mentioned AS_PATH and did not originate any prefix in the CC in consideration but originated a prefix into an overlapping neighboring CC, then the AS and prefix will still be detected and included in the design of the SAV filter. This improves the accuracy of the SAV filter in the BAR-SAV method in comparison to Algorithm A in [RFC8704].¶
One should not compute a customer cone by separately processing ASPA data and AS_PATH data and then merging the two sets of ASes at the end. Doing so is likely to miss ASes from the customer cone. Instead, both ASPAs and AS_PATHs should be used to iteratively expand the discovered customer cone. When new ASes are discovered, both ASPA and AS_PATH data should be used to discover customers of those ASes. This process is repeated for newly discovered customer ASes until there are no new ASes to be found.¶
As a measure of security, validation of the AS_PATH data in Adj-RIBs-In [RFC4271] SHOULD be performed using the procedures in [I-D.ietf-sidrops-aspa-verification] and any Invalid AS_PATHs must be excluded from inputs to the BAR-SAV procedure. This ensures that BGP UPDATEs containing route leaks are not considered for BAR-SAV filter design. Please see additional discussion about route leaks in Section 8.¶
As a further measure of security, validation of BGP routes in Adj-RIBs-In MUST be performed by applying RPKI-ROV [RFC6811] and any Invalid routes must be excluded from inputs to the BAR-SAV procedure. Please see additional discussion about prefix/route filtering in Section 8.¶
A detailed description of the BAR-SAV procedure is as follows:¶
Network operators SHOULD implement the BAR-SAV method (Section 4) for computing the permissible ingress prefix list for SAV on interfaces facing customers and lateral peers. BAR-SAV offers immediate incremental benefits to early adopters.¶
The operational recommendations provided in Section 3.2 of [RFC8704] are applicable and helpful for BAR-SAV (Section 4). Since Procedure X (Section 3) and the BAR-SAV procedure (Section 4) benefit from the registration of ROAs, network operators are RECOMMENDED to register ROAs and enable RPKI-ROV in their ASes. When ASPA registration becomes available, network operators are also RECOMMENDED to register ASPAs at that time.¶
The registration of ROAs and ASPAs helps with the detection and inclusion of otherwise hidden prefixes in the permissible list for SAV. As mentioned earlier, prefixes hidden in other SAV techniques often arise from the use of multi-homing in conjunction with limited propagation of prefixes in a given CC (for example, by attaching NO_EXPORT to all prefixes announced from a customer AS to a transit provider AS). In these situations, the registration of ASPAs and ROAs helps improve the accuracy of SAV.¶
Direct Server Return (DSR) is a common asymmetric routing scenario that is not supported by existing BCP-84 uRPF [RFC3704] and EFP-uRPF [RFC8704] SAV methods. DSR is commonly used by Content Delivery Networks (CDNs) that wish to use anycast service addresses but deliver data from edge locations that do not announce anycast addresses.¶
For example, in Figure 1, the CDN announces an anycast prefix P3 (from AS3) from a well-connected location with CDN control infrastructure. When a User from prefix P1 (AS1) establishes a connection to the anycast address and requests an object, an Anycast Server at the CDN may determine that the best location to serve the object is an Edge Server in a location close to the User. The Edge Server is reachable only via prefix P2 (AS2). The Anycast Server can forward packets arriving from the User to the Edge Server (via IP-IP tunneling or similar means), but the bulk data transmission would need to happen directly from the Edge Server to the User with an anycast source address (a P3 address).¶
Existing SAV methods of [RFC3704] and EFP-uRPF [RFC8704] would not allow AS4 to include P3 as a legitimate SA prefix on the interface to AS2. However, if the CDN (owner of prefix P3) registers a ROA object authorizing AS2 to originate P3, and AS4 uses an SAV procedure specified in this document (Section 4), then AS4 will use that ROA object to include P3 as a valid source prefix for the AS2 customer interface. The CDN may never want to announce a route to P3 from AS2, but the existence of this ROA would result in the construction of an SAV filter that would permit AS2 to send data packets with source addresses belonging to P3.¶
The CDN example above is just one DSR scenario. There are other cloud-based DSR scenarios that include low-latency gaming, mobile roaming, corporate networks of global enterprises, and others.¶
Recommendation: In a DSR scenario, a network operator MUST register ROAs that bind the edge server ASes with the anycast service prefix. This is in addition to registering a ROA authorizing the anycast server AS to announce the anycast prefix.¶
This section highlights some important operations and management considerations and was motivated in part to address the comments received from the SIDROPS working group members.¶
A transit provider is a network that (a) offers its customers outbound (customer to Internet) data traffic connectivity and/or (b) further propagates in all directions (towards providers, lateral peers, and other customers) any BGP Updates that the customer may send [I-D.ietf-sidrops-aspa-profile]. In the latter case, it also provides transport for inbound data traffic. In all cases, the customer AS SHOULD follow the specification in [I-D.ietf-sidrops-aspa-profile] and include the transit provider AS in its ASPA. Registering an ASPA prevents forged-origin hijacks for the customer AS and its prefixes, prevents route leaks involving the customer AS, and facilitates BAR-SAV.¶
If a prefix is used for source addresses for hosts attached at an AS but not announced in BGP from that AS (e.g., the DSR scenario in Section 5.1), a ROA MUST be registered binding the prefix and the AS. This ROA registration assists in preventing hijacking of the prefix and helps facilitate BAR-SAV. The risk of this ROA registration enabling a forged-origin prefix hijack for the prefix is minimal since the ASPA-based path verification [I-D.ietf-sidrops-aspa-verification] prevents forged-origin attacks. It may be noted that a similar usage of ROA is made in the context of DDoS mitigation (see Section 5.1 in [RFC9319]), where hypothetically the prefix may never need to be originated by the AS of the DDoS mitigation provider.¶
BAR-SAV identifies all ASes in a customer's (or lateral peer's) customer cone (CC), and then it discovers all prefixes that could plausibly be used as source addresses in data traffic originated from the ASes in the CC. If ASPA and ROA have been adopted by all ASes and prefix owners, respectively, in the CC of interest, then the list of plausible source address prefixes will be complete with no improper drops (i.e., traffic with legitimate source addresses is not dropped). Further, deploying BAR-SAV by all ASes within the CC ensures no improper admits (i.e., traffic with spoofed source address is not admitted). Note that routing policies of ASes may be such that some of the discovered prefixes may never be used as source addresses on a given customer interface of interest, but this does not impact BAR-SAV's accuracy.¶
The discussion in Section 3.6.1 of [RFC8704] of the Forwarding Information Base (FIB) size estimates and the networks where SAV would be most effective are applicable to BAR-SAV as well. Smaller ISPs (and possibly some midsize and regional ISPs) are expected to implement the BAR-SAV method, since SAV in general is most effective closer to the edges of the Internet. For such networks, the conservatively estimated SAV filter list size is only a small fraction of the anticipated FIB memory size (see details in Section 3.6.1 of [RFC8704]).¶
SAV done manually, e.g., using ACLs, usually does not get much adoption because of operational costs, susceptibility to human errors, and tendency of SAV filters to get out of date due to the need for any changes by customers or peers to be coordinated with multiple parties (providers and peers). Automated uRPF technique, such as BAR-SAV, however, allow for easy, accurate, and cost effective deployments. The BAR-SAV method makes it possible to automate the construction of SAV filter lists aiming for no improper drops and a minimal probability of improper admit of data traffic. As ASPA adoption picks up alongside the ongoing ROA adoption, BAR-SAV's accuracy of discovering all possible source addresses (prefixes) for the customer cone of interest improves even further in complex scenarios.¶
When a SAV filter is used to police data traffic, and an incomplete SAV filter list could cause legitimate traffic to be dropped, the use of robust implementation practices for RPKI data retrieval and cache management practices become paramount. Some of such recommended practices are discussed in this section.¶
RPKI infrastructure does not guarantee continuous availability of RPKI repositories. Local caches of RPKI signed objects, manifest files (MFTs), and certificate revocation lists (CRLs) are already maintained for managing ROA objects and router certificates [RFC8210]. That is being extended to ASPA objects as well [I-D.ietf-sidrops-8210bis]. The cache refresh frequency currently used for RPKI data should be sufficient for BAR-SAV purposes as well. If an RPKI repository publication point is unavailable, or there is any other failure in fetching its objects, the latest cached version of the objects associated with the repository MUST continue to be used, as described in [RFC9286].¶
If the local cache of some repository objects required for BAR-SAV computation is unavailable (for example, due to a filesystem failure) and/or the RPKI data cannot be fetched from the repository publication point, the SAV system SHOULD "fail open" and downgrade the SAV function on a given interface to "loose uRPF" described in [RFC3704] and [RFC8704]. This downgrade is better than suspending SAV entirely since at least source addresses in unallocated and bogon space are rejected.¶
This document includes no request to IANA.¶
The security considerations described in [RFC8704], [RFC6811], [I-D.ietf-sidrops-aspa-profile], and [I-D.ietf-sidrops-aspa-verification] also apply to this document.¶
The security and robustness of BAR-SAV are strengthened by supporting mechanisms for detecting and dropping BGP routes that are misoriginations or leaks. Section 4 stated the requirement of validating BGP route origins using RPKI-ROV [RFC6811]. It further helps if route origin validation using trusted IRR route objects and prefix filtering are also deployed (see [RFC7454] [NIST-800-189]). It is also advised that one or more of the available methods to prevent, detect, and mitigate route leaks are deployed (e.g., [RFC9234] [I-D.ietf-grow-route-leak-detection-mitigation] [I-D.ietf-sidrops-aspa-verification] [sriram1]).¶
The authors would like to thank Oliver Borchert, Job Snijders, Ben Maddison, and Geoff Huston for comments and discussion.¶