| RFC 0000 | Enhanced FP-uRPF | October 2019 |
| Sriram, et al. | Best Current Practice | [Page] |
This document identifies a need for and proposes improvement of the unicast Reverse Path Forwarding (uRPF) techniques (see RFC 3704) for detection and mitigation of source address spoofing (see BCP 38). The strict uRPF is inflexible about directionality, the loose uRPF is oblivious to directionality, and the current feasible-path uRPF attempts to strike a balance between the two (see RFC 3704). However, as shown in this document, the existing feasible-path uRPF still has shortcomings. This document describes enhanced feasible-path uRPF (EFP-uRPF) techniques, which are more flexible (in a meaningful way) about directionality than the feasible-path uRPF (RFC 3704). The proposed EFP-uRPF methods aim to significantly reduce false positives regarding invalid detection in source address validation (SAV). Hence they can potentially alleviate ISPs' concerns about the possibility of disrupting service for their customers, and encourage greater deployment of uRPF techniques. This document updates RFC 3704.¶
This memo documents an Internet Best Current Practice.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on BCPs is available in Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc0000.¶
Copyright (c) 2019 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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
Source Address Validation (SAV) refers to the detection and mitigation of source address (SA) spoofing [RFC2827]. This document identifies a need for and proposes improvement of improvement of the unicast Reverse Path Forwarding (uRPF) techniques [RFC3704] for SAV. The strict uRPF is inflexible about directionality (see [RFC3704] for definitions), the loose uRPF is oblivious to directionality, and the current feasible-path uRPF attempts to strike a balance between the two [RFC3704]. However, as shown in this document, the existing feasible-path uRPF still has shortcomings. Even with the feasible-path uRPF, ISPs are often apprehensive that they may be dropping customers' data packets with legitimate source addresses.¶
This document describes an enhanced feasible-path uRPF (EFP-uRPF) technique, which aims to be more flexible (in a meaningful way) about directionality than the feasible-path uRPF. It is based on the principle that if BGP updates for multiple prefixes with the same origin AS were received on different interfaces (at border routers), then incoming data packets with source addresses in any of those prefixes should be accepted on any of those interfaces (presented in Section 3). For some challenging ISP-customer scenarios (see Section 3.3), this document also describes a more relaxed version of the enhanced feasible-path uRPF technique (presented in Section 3.4). Implementation and operations considerations are discussed in Section 3.6.¶
Throughout this document, the routes under consideration are assumed to have been vetted based on prefix filtering [RFC7454] and possibly origin validation [RFC6811].¶
The EFP-uRPF methods aim to significantly reduce false positives regarding invalid detection in SAV. They are expected to add greater operational robustness and efficacy to uRPF, while minimizing ISPs' concerns about accidental service disruption for their customers. It is expected that this will encourage more deployment of uRPF to help realize its DDoS prevention benefits network wide.¶
Reverse Path Forwarding (RPF) list: The list of permissible source-address prefixes for incoming data packets on a given interface.¶
Peering relationships considered in this document are provider-to-customer (P2C), customer-to-provider (C2P), and peer-to-peer (p2p). Provider here refers to transit provider. The first two are transit relationships. A peer connected via a p2p link is known as a lateral peer (non-transit).¶
Customer Cone: AS A's customer cone is A plus all the ASes that can be reached from A following only P2C links [Luckie].¶
A stub AS is an AS that does not have any customers or lateral peers. In this document, a single-homed stub AS is one that has a single transit provider and a multi-homed stub AS is one that has multiple (two or more) transit providers.¶
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.¶
There are various existing techniques for mitigation against DDoS attacks with spoofed addresses [RFC2827] [RFC3704]. Source address validation (SAV) is performed in network edge devices such as border routers, Cable Modem Termination Systems (CMTS) [RFC4036], and Packet Data Network gateways (PDN-GW) in mobile networks [Firmin]. Ingress Access Control List (ACL) and unicast Reverse Path Forwarding (uRPF) are techniques employed for implementing SAV [RFC2827] [RFC3704] [ISOC].¶
Ingress/egress Access Control Lists (ACLs) are maintained to list acceptable (or alternatively, unacceptable) prefixes for the source addresses in the incoming/outgoing Internet Protocol (IP) packets. Any packet with a source address that fails the filtering criteria is dropped. The ACLs for the ingress/egress filters need to be maintained to keep them up to date. Updating the ACLs is an operator-driven manual process, and hence operationally difficult or infeasible.¶
Typically, the egress ACLs in access aggregation devices (e.g., CMTS, PDN-GW) permit source addresses only from the address spaces (prefixes) that are associated with the interface on which the customer network is connected. Ingress ACLs are typically deployed on border routers, and drop ingress packets when the source address is spoofed (e.g., belongs to obviously disallowed prefix blocks, IANA special-purpose prefixes [SPAR-v4][SPAR-v6], provider's own prefixes, etc.).¶
Note: In the figures (scenarios) in this section and the subsequent sections, the following terminology is used: "fails" means drops packets with legitimate source addresses; "works (but not desirable)" means passes all packets with legitimate source addresses but is oblivious to directionality; "works best" means passes all packets with legitimate source addresses with no (or minimal) compromise of directionality. Further, the notation Pi[ASn ASm ...] denotes a BGP update with prefix Pi and an AS_PATH as shown in the square brackets.¶
In the strict unicast Reverse Path Forwarding (uRPF) method, an ingress packet at a border router is accepted only if the Forwarding Information Base (FIB) contains a prefix that encompasses the source address, and forwarding information for that prefix points back to the interface over which the packet was received. In other words, the reverse path for routing to the source address (if it were used as a destination address) should use the same interface over which the packet was received. It is well known that this method has limitations when networks are multi-homed, routes are not symmetrically announced to all transit providers, and there is asymmetric routing of data packets. Asymmetric routing occurs (see Figure 1) when a customer AS announces one prefix (P1) to one transit provider (ISP-a) and a different prefix (P2) to another transit provider (ISP-b), but routes data packets with source addresses in the second prefix (P2) to the first transit provider (ISP-a) or vice versa. Then data packets with source address in prefix P2 that are received directly from AS1 will get dropped. Further, data packets with source address in prefix P1 that originate from AS1 and traverse via AS3 to AS2 will also get dropped at AS2.¶
+------------+ ---- P1[AS2 AS1] ---> +------------+
| AS2(ISP-a) | <----P2[AS3 AS1] ---- | AS3(ISP-b)|
+------------+ +------------+
/\ /\
\ /
\ /
\ /
P1[AS1]\ /P2[AS1]
\ /
+-----------------------+
| AS1(customer) |
+-----------------------+
P1, P2 (prefixes originated)
Consider data packets received at AS2
(1) from AS1 with source address (SA) in P2, or
(2) from AS3 that originated from AS1 with SA in P1:
* Strict uRPF fails
* Feasible-path uRPF fails
* Loose uRPF works (but not desirable)
* Enhanced Feasible-path uRPF works best
The feasible-path uRPF technique helps partially overcome the problem identified with the strict uRPF in the multi-homing case. The feasible-path uRPF is similar to the strict uRPF, but in addition to inserting the best-path prefix, additional prefixes from alternative announced routes are also included in the RPF list. This method relies on either (a) announcements for the same prefixes (albeit some may be prepended to effect lower preference) propagating to all transit providers performing feasible-path uRPF checks, or (b) announcement of an aggregate less specific prefix to all transit providers while announcing more specific prefixes (covered by the less specific prefix) to different transit providers as needed for traffic engineering. As an example, in the multi-homing scenario (see Scenario 2 in Figure 2), if the customer AS announces routes for both prefixes (P1, P2) to both transit providers (with suitable prepends if needed for traffic engineering), then the feasible-path uRPF method works. It should be mentioned that the feasible-path uRPF works in this scenario only if customer routes are preferred at AS2 and AS3 over a shorter non-customer route. However, the feasible-path uRPF method has limitations as well. One form of limitation naturally occurs when the recommendation (a) or (b) mentioned above regarding propagation of prefixes is not followed. Another form of limitation can be described as follows. In Scenario 2 (described here, illustrated in Figure 2), it is possible that the second transit provider (ISP-b or AS3) does not propagate the prepended route for prefix P1 to the first transit provider (ISP-a or AS2). This is because AS3's decision policy permits giving priority to a shorter route to prefix P1 via a lateral peer (AS2) over a longer route learned directly from the customer (AS1). In such a scenario, AS3 would not send any route announcement for prefix P1 to AS2 (over the p2p link). Then a data packet with source address in prefix P1 that originates from AS1 and traverses via AS3 to AS2 will get dropped at AS2.¶
+------------+ routes for P1, P2 +-----------+
| AS2(ISP-a) |<-------------------->| AS3(ISP-b)|
+------------+ (p2p) +-----------+
/\ /\
\ /
P1[AS1]\ /P2[AS1]
\ /
P2[AS1 AS1 AS1]\ /P1[AS1 AS1 AS1]
\ /
+-----------------------+
| AS1(customer) |
+-----------------------+
P1, P2 (prefixes originated)
Consider data packets received at AS2 via AS3
that originated from AS1 and have source address in P1:
* Feasible-path uRPF works (if customer route to P1
is preferred at AS3 over shorter path)
* Feasible-path uRPF fails (if shorter path to P1
is preferred at AS3 over customer route)
* Loose uRPF works (but not desirable)
* Enhanced Feasible-path uRPF works best
In the loose unicast Reverse Path Forwarding (uRPF) method, an ingress packet at the border router is accepted only if the FIB has one or more prefixes that encompass the source address. That is, a packet is dropped if no route exists in the FIB for the source address. Loose uRPF sacrifices directionality. It only drops packets if the source address is unreachable in the current FIB (e.g., IANA special-purpose prefixes [SPAR-v4][SPAR-v6], unallocated, allocated but currently not routed).¶
The Virtual Routing and Forwarding (VRF) technology [RFC4364] [Juniper] allows a router to maintain multiple routing table instances separate from the global Routing Information Base (RIB). External BGP (eBGP) peering sessions send specific routes to be stored in a dedicated VRF table. The uRPF process queries the VRF table (instead of the FIB) for source address validation. A VRF table can be dedicated per eBGP peer and used for uRPF for only that peer, resulting in strict mode operation. For implementing loose uRPF on an interface, the corresponding VRF table would be global, i.e., contains the same routes as in the FIB.¶
Enhanced feasible-path uRPF (EFP-uRPF) method adds greater operational robustness and efficacy to existing uRPF methods discussed in Section 2. That is because it avoids dropping legitimate data packets and avoids compromising directionality. The method is based on the principle that if BGP updates for multiple prefixes with the same origin AS were received on different interfaces (at border routers), then incoming data packets with source addresses in any of those prefixes should be accepted on any of those interfaces. The EFP-uRPF method can be best explained with an example as follows:¶
Let us say, a border router of ISP-A has in its Adj-RIBs-In [RFC4271] the set of prefixes {Q1, Q2, Q3} each of which has AS-x as its origin and AS-x is in ISP-A's customer cone. In this set, the border router received the route for prefix Q1 over a customer facing interface, while it learned the routes for prefixes Q2 and Q3 from a lateral peer and an upstream transit provider, respectively. In this example scenario, the enhanced feasible-path uRPF method requires Q1, Q2, and Q3 be included in the RPF list for the customer interface under consideration.¶
Thus, the enhanced feasible-path uRPF (EFP-uRPF) method gathers feasible paths for customer interfaces in a more precise way (as compared to feasible-path uRPF) so that all legitimate packets are accepted while the directionality property is not compromised.¶
The above described EFP-uRPF method is recommended to be applied on customer interfaces. It can be extended to create the RPF lists for lateral peer interfaces also. That is, the EFP-uRPF method can be applied (and loose uRPF avoided) on lateral peer interfaces. That will help avoid compromise of directionality for lateral peer interfaces (which is inevitable with loose uRPF; see Section 2.4).¶
Looking back at Scenarios 1 and 2 (Figures 1 and 2), the enhanced feasible-path uRPF (EFP-uRPF) method works better than the other uRPF methods. Scenario 3 (Figure 3) further illustrates the enhanced feasible-path uRPF method with a more concrete example. In this scenario, the focus is on operation of the feasible-path uRPF at ISP4 (AS4). ISP4 learns a route for prefix P1 via a customer-to-provider (C2P) interface from customer ISP2 (AS2). This route for P1 has origin AS1. ISP4 also learns a route for P2 via another C2P interface from customer ISP3 (AS3). Additionally, AS4 learns a route for P3 via a lateral peer-to-peer (p2p) interface from ISP5 (AS5). Routes for all three prefixes have the same origin AS (i.e., AS1). Using the enhanced feasible-path uRPF scheme, given the commonality of the origin AS across the routes for P1, P2 and P3, AS4 includes all of these prefixes in the RPF list for the customer interfaces (from AS2 and AS3).¶
+----------+ P3[AS5 AS1] +------------+
| AS4(ISP4)|<---------------| AS5(ISP5) |
+----------+ (p2p) +------------+
/\ /\ /\
/ \ /
P1[AS2 AS1]/ \P2[AS3 AS1] /
(C2P)/ \(C2P) /
/ \ /
+----------+ +----------+ /
| AS2(ISP2)| | AS3(ISP3)| /
+----------+ +----------+ /
/\ /\ /
\ / /
P1[AS1]\ /P2[AS1] /P3[AS1]
(C2P)\ /(C2P) /(C2P)
\ / /
+----------------+ /
| AS1(customer) |/
+----------------+
P1, P2, P3 (prefixes originated)
Consider that data packets (sourced from AS1)
may be received at AS4 with source address
in P1, P2 or P3 via any of the neighbors (AS2, AS3, AS5):
* Feasible-path uRPF fails
* Loose uRPF works (but not desirable)
* Enhanced Feasible-path uRPF works best
The underlying algorithm in the solution method described above (Section 3.1) can be specified as follows (to be implemented in a transit AS):¶
The above algorithm can be extended to apply EFP-uRPF method to lateral peer interfaces also. However, it is left up to the operator to decide whether they should apply EFP-uRPF or loose uRPF method on lateral peer interfaces. The loose uRPF method is recommended to be applied on transit provider interfaces.¶
The following operational recommendations will make the operation of the enhanced feasible-path uRPF robust:¶
For multi-homed stub AS:¶
For non-stub AS:¶
It should be observed that in the absence of ASes adhering to above recommendations, the following example scenario may be constructed which poses a challenge for the enhanced feasible-path uRPF (as well as for traditional feasible-path uRPF). In the scenario illustrated in Figure 4, since routes for neither P1 nor P2 are propagated on the AS2-AS4 interface (due to the presence of NO_EXPORT Community), the enhanced feasible-path uRPF at AS4 will reject data packets received on that interface with source addresses in P1 or P2. (For a little more complex example scenario, see slide #10 in [Sriram-URPF].)¶
+----------+
| AS4(ISP4)|
+----------+
/\ /\
/ \ P1[AS3 AS1]
P1 and P2 not / \ P2[AS3 AS1]
propagated / \ (C2P)
(C2P) / \
+----------+ +----------+
| AS2(ISP2)| | AS3(ISP3)|
+----------+ +----------+
/\ /\
\ / P1[AS1]
P1[AS1] NO_EXPORT \ / P2[AS1]
P2[AS1] NO_EXPORT \ / (C2P)
(C2P) \ /
+----------------+
| AS1(customer) |
+----------------+
P1, P2 (prefixes originated)
Consider that data packets (sourced from AS1)
may be received at AS4 with source address
in P1 or P2 via AS2:
* Feasible-path uRPF fails
* Loose uRPF works (but not desirable)
* Enhanced Feasible-path uRPF with Algorithm A fails
* Enhanced Feasible-path uRPF with Algorithm B works best
Adding further flexibility to the enhanced feasible-path uRPF method can help address the potential limitation identified above using the scenario in Figure 4 (Section 3.3). In the following, "route" refers to a route currently existing in the Adj-RIB-in. Including the additional degree of flexibility, the modified algorithm called Algorithm B (implemented in a transit AS) can be described as follows:¶
When Algorithm B (which is more flexible than Algorithm A) is employed on customer interfaces, the type of limitation identified in Figure 4 (Section 3.3) is overcome and the method works. The directionality property is minimally compromised, but still the proposed EFP-uRPF method with Algorithm B is a much better choice (for the scenario under consideration) than applying the loose uRPF method which is oblivious to directionality.¶
So, applying EFP-uRPF method with Algorithm B is recommended on customer interfaces for the challenging scenarios such as those described in Section 3.3.¶
It is worth emphasizing that an indirect part of the proposal in this document is that RPF filters may be augmented from secondary sources. Hence, the construction of RPF lists using a method proposed in this document (Algorithm A or B) can be augmented with data from Route Origin Authorization (ROA) [RFC6482] as well as Internet Routing Registry (IRR) data. Special care should be exercised when using IRR data because it not always accurate or trusted. In the EFP-uRPF method with Algorithm A (see Section 3.1.1), if a ROA includes prefix Pi and ASj, then augment with Pi the RPF list of each customer interface on which at least one route with origin ASj was received. In the EFP-uRPF method with Algorithm B, if ASj belongs in set A (see Step #3 Section 3.4) and if a ROA includes prefix Pi and ASj, then augment with Pi the RPF list Z in Step 5 of Algorithm B. Similar procedures can be followed with reliable IRR data as well. This will help make the RPF lists more robust about source addresses that may be legitimately used by customers of the ISP.¶
The existing RPF checks in edge routers take advantage of existing line card implementations to perform the RPF functions. For implementation of the enhanced feasible-path uRPF, the general necessary feature would be to extend the line cards to take arbitrary RPF lists that are not necessarily the same as the existing FIB contents. In the algorithms (Section 3.1.1 and Section 3.4) described here, the RPF lists are constructed by applying a set of rules to all received BGP routes (not just those selected as best path and installed in the FIB). The concept of uRPF querying an RPF list (instead of the FIB) is similar to uRPF querying a VRF table (see (Section 2.5).¶
The techniques described in this document require that there should be additional memory (i.e., ternary content addressable memory (TCAM)) available to store the RPF lists in line cards. For an ISP's AS, the RPF list size for each line card will roughly equal the total number of originated prefixes from ASes in its customer cone (assuming Algorithm B in Section 3.4 is used). (Note: EFP-uRPF with Algorithm A (see Section 3.1.1) requires much less memory than EFP-uRPF with Algorithm B.)¶
The following table shows the measured customer cone sizes in number of prefixes originated (from all ASes in the customer cone) for various types of ISPs [Sriram-RIPE63]:¶
| Type of ISP | Measured Customer Cone Size in # Prefixes (in turn this is an estimate for RPF list size on the line card) |
|---|---|
| Very Large Global ISP #1 | 32393 |
| Very Large Global ISP #2 | 29528 |
| Large Global ISP | 20038 |
| Mid-size Global ISP | 8661 |
| Regional ISP (in Asia) | 1101 |
For some super large global ISPs that are at the core of the Internet, the customer cone size (# prefixes) can be as high as a few hundred thousand [CAIDA]. But uRPF is most effective when deployed at ASes at the edges of the Internet where the customer cone sizes are smaller as shown in Table 1.¶
A very large global ISP's router line card is likely to have a FIB size large enough to accommodate 2 million routes [Cisco1]. Similarly, the line cards in routers corresponding to a large global ISP, a mid-size global ISP, and a regional ISP are likely to have FIB sizes large enough to accommodate about 1 million, 0.5 million, and 100K routes, respectively [Cisco2]. Comparing these FIB size numbers with the corresponding RPF list size numbers in Table 1, it can be surmised that the conservatively estimated RPF list size is only a small fraction of the anticipated FIB memory size under relevant ISP scenarios. What is meant here by relevant ISP scenarios is that only smaller ISPs (and possibly some mid-size and regional ISPs) are expected to implement the proposed EFP-uRPF method since it is most effective closer to the edges of the Internet.¶
BGP routing announcements can exhibit transient behavior. Routes may be withdrawn temporarily and then re-announced due to transient conditions such as BGP session reset or link failure-recovery. To cope with this, hysteresis should be introduced in the maintenance of the RPF lists. Deleting entries from the RPF lists SHOULD be delayed by a pre-determined amount (the value based on operational experience) when responding to route withdrawals. This should help suppress the effects due to the transients in BGP.¶
Depending on the scenario, an ISP or enterprise AS operator should follow one of the following recommendations concerning uRPF/SAV:¶
For all other scenarios:¶
It is also recommended that prefixes from registered ROAs and IRR route objects that include ASes in an ISP's customer cone SHOULD be used to augment the pertaining RPF lists (see Section 3.5 for details).¶
EFP-uRPF method with Algorithm A is not mentioned in the above set of recommendations. It is an alternative to EFP-uRPF with Algorithm B and can be used in limited circumstances. The EFP-uRPF method with Algorithm A is expected to work fine if an ISP deploying it has only multi-homed stub customers. It is trivially equivalent to strict uRPF if an ISP deploys it for a single-homed stub customer. More generally, it is also expected to work fine when there is absence of limitations such as those described in Section 3.3. However, caution is required for use of EFP-uRPF with Algorithm A because even if the limitations are not expected at the time of deployment, the vulnerability to change in conditions exists. It may be difficult for an ISP to know or track the extent of use of NO_EXPORT (see Section 3.3) on routes within its customer cone. If an ISP decides to use EFP-uRPF with Algorithm A, it should make its direct customers aware of the operational recommendations in Section 3.2. This means that the ISP notifies direct customers that at least one prefix originated by each AS in the direct customer's customer cone must propagate to the ISP.¶
On a lateral peer interface, an ISP may choose to apply the EFP-uRPF method with Algorithm A (with appropriate modification of the algorithm). This is because stricter forms of uRPF (than the loose uRPF) may be considered applicable by some ISPs on interfaces with lateral peers.¶
The security considerations in BCP 38 [RFC2827] and BCP 84 [RFC3704] apply for this document as well. In addition, if considering using EFP-uRPF method with Algorithm A, an ISP or AS operator should be aware of the applicability considerations and potential vulnerabilities discussed in Section 3.7.1.¶
In augmenting RPF lists with ROA (and possibly reliable IRR) information (see Section 3.5), a trade-off is made in favor of reducing false positives (regarding invalid detection in SAV) at the expense of a slight other risk. The other risk being a malicious actor at another AS in the neighborhood within the customer cone might take advantage (of the augmented prefix) to some extent. This risk also exists even with normal announced prefixes (i.e., without ROA augmentation) for any uRPF method other than the strict. However, the risk is mitigated if the transit provider of the other AS in question is performing SAV.¶
Though not within the scope of this document, security hardening of routers and other supporting systems (e.g., Resource PKI (RPKI) and ROA management systems) against compromise is extremely important. The compromise of those systems can affect the operation and performance of the SAV methods described in this document.¶
This document does not request new capabilities or attributes. It does not create any new IANA registries.¶
The authors would like to thank Sandy Murphy, Alvaro Retana, Job Snijders, Marco Marzetti, Marco d'Itri, Nick Hilliard, Gert Doering, Fred Baker, Igor Gashinsky, Igor Lubashev, Andrei Robachevsky, Barry Greene, Amir Herzberg, Ruediger Volk, Jared Mauch, Oliver Borchert, Mehmet Adalier, and Joel Jaeggli for comments and suggestions. The comments and suggestions received from the IESG reviewers are also much appreciated.¶