Join our Newsletter — 33% off our NHI Course
Home› Guides› Source Address Validation and Anti-Spoofing Guide: Deploying BCP…
Guide Network Security

Source Address Validation and Anti-Spoofing Guide: Deploying BCP 38, BCP 84 and uRPF Safely

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 27 October 2026 14 min read
On this page

Reflection and amplification attacks only work because some networks still forward packets with forged source addresses, and every network that skips source address validation lends its customers' bandwidth to the next flood. Source address validation (SAV) is the practice of dropping packets whose source address could not legitimately come from the interface they arrived on. It has been an IETF Best Current Practice since BCP 38 (RFC 2827, 2000), was extended for multihomed networks by BCP 84 (RFC 3704 and RFC 8704), and is one of the four MANRS actions for network operators. This guide is for network engineers at ISPs, hosting and cloud providers, data centres and enterprises with their own address space: which method to apply on which interface, how to roll it out without dropping legitimate traffic, and how to prove it is working. Defending against floods once they arrive is covered in the DDoS Protection Guide; this page is about not being a source.

Key takeaways

  • SAV protects other people first. RFC 2827 is explicit that ingress filtering does nothing against floods from legitimately announced prefixes; its value is that spoofed traffic cannot leave your network and attacks become traceable.
  • Match the method to the interface. RFC 8704 recommends strict uRPF or ACLs for directly connected single-homed customers, enhanced feasible-path uRPF (Algorithm B) for customer interfaces generally, and loose uRPF towards transit providers.
  • Loose uRPF is not anti-spoofing. RFC 3704 notes that it loses directionality and that a default route can make it pass everything; use it only to drop unrouted and martian sources.
  • Do not forget the edges inside your edge. MANRS asks for SAV on single-homed stub customers, your own end users and your infrastructure; SAVI (RFC 7039) extends validation to the first-hop switch where hosts share a link.
  • Measure from outside. Router drop counters show your filters fire, but only an external test such as the CAIDA Spoofer client shows whether spoofed packets can actually leave your network.

Why spoofing still matters

A spoofed packet carries a source address the sender does not own. In a reflection attack (MITRE ATT&CK T1498.002 Reflection Amplification) the attacker sends small requests to open services, such as DNS resolvers, with the victim's address as the source; the services send their larger replies to the victim. RFC 5358 (BCP 140) on recursive resolvers identifies ingress filtering as the only real long-term fix for this class of attack, alongside closing open recursion. RFC 2827 lists the older forms that SAV also blunts: SYN floods with forged sources, UDP floods that connect chargen and echo services on different sites, and smurf floods against directed broadcasts.

Attack patternRole of spoofingWhat SAV changesOther controls still needed
Reflection and amplification (T1498.002)Victim's address forged as source of requests to reflectorsAttackers inside your network cannot launch it from thereClose open resolvers and other reflectors (RFC 5358), upstream scrubbing
Direct floods with random sources (T1498.001)Random sources defeat simple source blocking and hide originFloods from your customers carry real addresses and can be tracedRate limiting, DDoS mitigation service
TCP SYN floods with forged sourcesExhaust state; backscatter hits bystandersRemoves forged-source variants from your networkSYN cookies, stateful edge protection
Spoofing your own prefixes from outsidePackets claiming to be internal arrive at your edgeInbound filter drops your own space arriving from the internetInternal segmentation
Same-subnet host impersonationHost on a shared LAN uses a neighbour's addressNot addressed by edge SAVSAVI at the access switch (RFC 7039)

The SAV methods compared

RFC 3704 defines the methods and their trade-offs; RFC 8704 adds enhanced feasible-path uRPF (EFP-uRPF), written by authors from NIST and Juniper, which reduces the false positives that make operators switch filtering off.

MethodHow it decidesStrengthWeaknessTypical place
Ingress ACLStatic list of the customer's prefixesMost deterministicGoes stale when the customer adds space or multihomesDirectly connected subnets, stable customers, extra layer
Strict uRPFSource must be reachable via the arrival interface (best path in the FIB)Simple, dynamic, cheapDrops legitimate traffic under asymmetric routingSingle-homed stub customers, access networks
Feasible-path uRPFSource must be reachable via the arrival interface on any feasible path in the RIBHandles multihoming and asymmetry betterFails if a route is not advertised on every path (NO_EXPORT, policy)Multihomed customers
EFP-uRPF Algorithm APrefixes sharing an origin AS seen on customer interfaces are accepted on those interfacesFewer false drops than feasible-pathLess flexible than Algorithm BOption for lateral peers
EFP-uRPF Algorithm BCustomer-cone origin ASes define acceptable prefixes, including routes learned from peers and transitsRecommended for customer interfaces in RFC 8704Needs vendor support; depends on BGP data qualityCustomer interfaces of transit and large access providers
Loose uRPFSource must exist anywhere in the FIBSafe on any interfaceNo directionality; a default route defeats itTransit provider and lateral peer interfaces, martian dropping
Loose uRPF ignoring defaultAs loose, but default route does not countWorks on asymmetric inter-ISP linksStill no directionalityAsymmetric inter-provider links

RFC 8704 also allows the acceptable list to be augmented with prefixes from RPKI ROAs, and from IRR data only where that data is reliable. That links SAV quality directly to routing data hygiene; see the RPKI and Route Origin Validation Deployment Guide.

Where to apply each method

RFC 3704 states that ingress filtering should always be applied between an ISP and a single-homed edge network, can usually be applied to multihomed edges with feasible or strict RPF, and is reasonable between ISPs where the prefix count is low. It also says both ISPs and edge networks should check that their own addresses are not used as sources from outside.

Interface or networkRecommended methodSource
Directly connected subnets (your own LANs, server VLANs)ACL permitting only the subnet as sourceRFC 8704
Single-homed stub customerStrict uRPF (or ACL)RFC 8704, RFC 3704
Multihomed customerEFP-uRPF Algorithm B, or feasible-path uRPF, or ACL built from the customer's registered prefixesRFC 8704
Lateral peerLoose uRPF, or EFP-uRPF Algorithm A where suitableRFC 8704
Transit providerLoose uRPF; inbound ACL dropping your own prefixes and special-purpose spaceRFC 8704, RFC 3704
Enterprise internet edge, outboundACL permitting only your assigned public prefixes (and NAT pool) as sourceRFC 2827 principle applied to the edge
Hosts on a shared linkSAVI binding of address to switch port or MAC at the access layerRFC 7039

Special-purpose and bogon sources

Whatever method is used on an interface, packets arriving from the internet with special-purpose source addresses should be dropped. The IANA special-purpose registry described in RFC 6890 includes, for IPv4: 0.0.0.0/8, 10.0.0.0/8, 100.64.0.0/10 (shared address space), 127.0.0.0/8, 169.254.0.0/16, 172.16.0.0/12, 192.0.0.0/24, 192.0.2.0/24, 192.88.99.0/24, 192.168.0.0/16, 198.18.0.0/15, 198.51.100.0/24, 203.0.113.0/24, 240.0.0.0/4 and 255.255.255.255/32. Check the live IANA registry when building the list, and treat 100.64.0.0/10 with care if a provider uses it for carrier-grade NAT on your side of the edge.

Implementation by environment

  • Access ISPs and broadband: the highest value deployment. Apply strict uRPF or per-subscriber source checks on the broadband network gateway or access router, so every end user is validated. This is the "own end-users" part of the MANRS action.
  • Transit and large providers: run EFP-uRPF (or feasible-path) on customer-facing interfaces, built from BGP data and customer cones; keep ACL generation from IRR and RPKI data as a cross-check for customers with few prefixes.
  • Hosting, data centres and cloud: tenant virtual machines are a common source of spoofed traffic. Validate at the hypervisor or top-of-rack switch so each tenant can only use its assigned addresses, and at the data centre edge so only your announced prefixes leave. Check the provider controls in your platform (many virtual networks enforce per-interface source checks by default; confirm it has not been disabled for routing or NAT appliances).
  • Enterprises with their own address space: outbound ACL at the internet edge permitting only your public prefixes as source, inbound ACL dropping your own prefixes and special-purpose space. Internally, ACLs on routed VLAN interfaces stop a compromised host forging addresses from other segments.
  • Linux routers and hosts: the kernel rp_filter sysctl implements RFC 3704 checks: 0 is no validation, 1 is strict, 2 is loose. The kernel documentation notes the default is 0 (some distributions enable it at start-up), that the higher of the all and per-interface values applies, and recommends loose mode where routing is asymmetric.

Vendor syntax differs, but most routing platforms support strict and loose uRPF per interface, often with options to allow a default route or to log dropped packets. Confirm in your platform's documentation whether uRPF runs in hardware at line rate, whether it supports IPv6 equally, and whether feasible-path or EFP modes are available.

Phased rollout and rollback

  1. Inventory interfaces: classify every external and customer-facing interface by the table above (stub, multihomed, peer, transit, internal subnet). Interfaces that cannot be classified are the risk.
  2. Start with the safe wins: outbound and inbound bogon and own-prefix ACLs at the edge, and strict uRPF or ACLs on directly connected single-homed customers and internal subnets. These rarely break legitimate traffic.
  3. Monitor before dropping where the platform allows: use counters or a log-only filter action on the interfaces with any asymmetry, for one to two weeks, to see what would be dropped and why.
  4. Move multihomed customers to feasible-path or EFP-uRPF: agree with each customer which prefixes they will source, and ask them to announce all of them on every link, which feasible-path relies on.
  5. Automate ACLs: generate customer ACLs from the provisioning system, IRR objects and ROAs rather than by hand, so new customer prefixes are added on the same day.
  6. Verify externally: run the CAIDA Spoofer client from representative customer and internal networks; it sends spoofed UDP probes to measure whether forged-source packets can leave and enter, and publishes results by autonomous system.
  7. Roll back narrowly: if legitimate traffic drops, move that interface from strict to feasible-path or add the missing prefix to the ACL; do not remove SAV from the whole router. Record every relaxation with an owner and review date.

Decision guidance: a small enterprise with one upstream needs only an edge ACL and should ask its ISP to apply SAV on the access link. A regional ISP should prioritise strict uRPF on every subscriber, because that is where most spoofing originates. A transit provider gets most value from EFP-uRPF on customer interfaces, with loose uRPF on peers and transits, accepting that loose mode is a martian filter rather than anti-spoofing.

Detection and measurement

  • Router counters: uRPF and ACL drop counters per interface, graphed over time. A sudden spike on a customer port can mean a compromised customer host; a steady rise after a routing change can mean a false positive.
  • Flow data: NetFlow or IPFIX at the edge showing outbound packets with source addresses outside your prefixes. If any appear, SAV is missing somewhere upstream of the collector.
  • Inbound own-prefix sources: any packet arriving from the internet that claims one of your own prefixes as its source is spoofed by definition.
  • External testing: CAIDA Spoofer results for your AS, and registration for notification when spoofed packets are detected from it. The project page notes that its funding has lapsed and it relies on community donations, so do not make it your only test.
  • Abuse reports: reports that your address space sourced reflection traffic should be rare; track them.

Metrics:

  • Percentage of customer-facing interfaces with SAV appropriate to their class (target 100 percent of single-homed stubs).
  • Percentage of end-user access ports or sessions with source validation.
  • Days from new customer prefix to ACL update (target same day, through automation).
  • Number of external Spoofer tests from your networks that show spoofed packets escaping (target zero).
  • Number of SAV exceptions or relaxations past their review date.

Where identity fits

Source addresses are a weak form of identity, and many systems still treat them as one: IP allowlists on APIs, management interfaces trusted by subnet, and partner integrations authenticated by source address. SAV makes those addresses harder to forge for connectionless protocols, but it does not make an address a credential. Any service that authenticates workloads or machines should use real credentials (mutual TLS, signed tokens, workload identity) rather than relying on network origin. See the Network Segmentation Guide for where address-based controls still fit.

Common failure patterns

  • Loose uRPF everywhere: operators enable loose mode on customer interfaces because strict mode caused complaints. With a default route, loose mode accepts everything, so the control exists on paper only.
  • Strict uRPF on multihomed customers: asymmetric return paths cause legitimate drops, the filter is removed under pressure, and nothing replaces it. Use feasible-path or EFP-uRPF instead.
  • Hand-maintained ACLs: customer prefix lists drift from reality, legitimate traffic is dropped after a customer grows, and engineers disable filtering rather than update it.
  • Edge only: the internet edge filters correctly but internal VLANs and tenant networks do not, so a compromised server can still spoof addresses from other internal segments or tenants.
  • IPv4 only: SAV is configured for IPv4 and forgotten for IPv6, which spoofing tests cover separately.
  • Never testing: configuration is assumed to work, but a single unfiltered path (a backup link, a new data centre) lets spoofed traffic out. Only an external test shows this.

Framework mapping

FrameworkReferenceRelevance
IETF BCPsBCP 38 (RFC 2827); BCP 84 (RFC 3704, RFC 8704); BCP 140 (RFC 5358)The technical definition of SAV and reflector prevention
NISTSP 800-189 Rev. 1 (Initial Public Draft, January 2025, "Border Gateway Protocol Security and Resilience")Covers SAV using ACLs and uRPF alongside ROA-based route origin validation; status was draft when checked on 27 October 2026
MANRSAction 2, Anti-spoofingEnable SAV for at least single-homed stub customers, own end users and infrastructure
NIST SP 800-53 Rev. 5SC-7 Boundary Protection; SC-5 Denial-of-service Protection; SI-4 System MonitoringEdge filtering, DoS resilience, drop and flow monitoring
NIST CSF 2.0PR.IR-01; DE.CM-01Protecting networks from unauthorised use; monitoring network services
MITRE ATT&CK v20T1498.002 Reflection Amplification; T1498.001 Direct Network FloodThe impact techniques SAV makes harder to launch

Practitioner checklist

  • Classify every external and customer-facing interface as subnet, stub, multihomed, peer or transit.
  • Apply ACL or strict uRPF on every directly connected subnet and single-homed customer.
  • Use feasible-path or EFP-uRPF (Algorithm B) for multihomed customers instead of strict mode.
  • Restrict loose uRPF to peer and transit interfaces, and understand that it is a martian filter only.
  • Drop special-purpose sources and your own prefixes arriving from the internet.
  • Permit only your own public prefixes as source on outbound edge traffic.
  • Configure SAV for IPv6 everywhere it exists for IPv4.
  • Generate customer ACLs automatically from provisioning, IRR and ROA data.
  • Graph SAV drop counters and run an external spoofing test from each network at least yearly.
  • Close open recursive resolvers and other reflectors you operate (RFC 5358).

Standards and references

Related NHI Mgmt Group resources: DDoS Protection Guide · MANRS Network Operator Actions Guide · BGP and Routing Security Guide · RPKI and Route Origin Validation Deployment Guide · DNS Security Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 27 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.