RPKI route origin validation and IRR based prefix filtering are often treated as rival answers to BGP hijacks, but they check different things and each leaves gaps the other partly closes. Origin validation asks whether the origin AS of a route is authorised by a signed object; a prefix filter asks whether this particular neighbour should be sending this prefix at all. Neither validates the AS path, so route leaks and forged-origin hijacks slip past both unless you add path controls. This guide compares the two on threat model, data source, failure modes and operating cost, then shows how to run them together. It is for network engineers, peering coordinators and security architects at ISPs, hosting providers and multihomed enterprises. For deployment detail on ROAs and validators see the RPKI and Route Origin Validation Deployment Guide; for the wider BGP picture see the BGP and Routing Security Guide.
Key takeaways
- They answer different questions. ROV checks the origin AS against signed ROAs on every route from every neighbour; prefix filters check that a neighbour only sends what it is entitled to send, which matters most on customer sessions.
- Neither validates the path. RFC 6811 says origin validation "does not protect against "AS-in-the-middle attacks" or provide any path validation", and a prefix list built from a customer's registered space accepts a leaked route with the right prefix and origin.
- Data quality is the real difference. ROAs are cryptographically tied to the address holder; IRR objects vary by registry in how they are authorised, so filters inherit whatever stale or unchecked objects the AS-SET expands to.
- Layer, do not choose. Drop RPKI Invalid on all eBGP sessions, apply strict prefix filters and max-prefix on customers, and add AS path checks and RFC 9234 roles for leaks.
- Build filters from both sources. Using RPKI data to prune or supplement IRR derived prefix lists catches registry objects that contradict signed ROAs before they reach your edge.
The short answer
If you run eBGP with anyone other than a single upstream, you need both. Route origin validation is a global, low maintenance check that stops origin mistakes and simple exact or sub-prefix hijacks from any direction, provided the victim has published ROAs. Prefix filtering is a per-neighbour allow list that stops a customer from announcing space it does not hold, including space with no ROA, and it is the control that keeps one customer's error from becoming a global incident. MANRS Action 1 requires network operators to have a system that stops incorrect routing announcements, names explicit prefix-level filters or equivalent mechanisms as the main way to do it, and states that filtering customer announcements by AS-path filters alone is insufficient. RFC 7454 (BCP 194) recommends both: discard RPKI Invalid and build customer filters from registry data.
Enterprises that only originate their own space have a smaller job: publish ROAs, apply a strict outbound prefix list so they never leak upstream routes, and ask their providers to do the inbound work.
How the two controls compare
| Dimension | RPKI route origin validation | IRR based prefix filtering |
|---|---|---|
| Question answered | Is this origin AS authorised for this prefix and length? | Should this neighbour be sending this prefix? |
| Data source | ROAs signed under RIR resource certificates, validated by relying party software | route and route6 objects plus AS-SETs in one or more Internet Routing Registries |
| Trust basis | Cryptographic chain to the RIR trust anchors | Registry authorisation rules, which differ between registries |
| Where applied | All eBGP sessions: transit, peers, customers, route servers | Mainly customer sessions; strict mode on peers where the registry is not fully trusted |
| Delivery to router | Validated ROA payloads over the RPKI to Router protocol (RFC 8210) | Generated prefix lists pushed by configuration automation |
| Update model | Continuous, as validators refresh | Batch; RFC 7454 says refreshing filters daily SHOULD be considered |
| Coverage gap | Prefixes with no ROA are NotFound and pass | Customers with no or wrong registry objects need manual entries |
| Path awareness | None | None, unless paired with AS path filters |
| Main failure mode | Wrong or stale ROA makes legitimate routes Invalid; validator outage silently stops enforcement | Stale or rogue IRR objects let bad routes in; filter generation failure blocks a customer's new prefix |
| Config footprint | Small: one policy per session type | Large: RFC 7454 notes filters can run to tens of thousands of lines for large peers |
RPKI origin validation: threat model and limits
What it stops: an AS originating a prefix that a ROA authorises for a different AS, and a more-specific announcement longer than the ROA's maxLength. RFC 6811 labels each route Valid, Invalid or NotFound, and RFC 7454 says an Invalid route SHOULD be discarded. Because the check uses global data, it works on routes from a peer you have never filtered, which prefix lists cannot do.
Architecture in words: RIRs certify address holders; holders sign ROAs in a hosted portal or delegated CA; validators fetch and verify the objects and produce validated payloads; routers receive them over RTR and tag routes; your import policy drops Invalid. Every stage is a dependency, which is why the deployment guide insists on at least two validators.
What it misses:
- Unsigned space: a prefix with no covering ROA is NotFound and accepted, so ROV protects only holders who have signed.
- Forged origin: RFC 9319 describes an attacker who puts the legitimate origin AS at the end of a forged AS_PATH. With a loose maxLength the attacker can announce an unused more-specific and pass ROV; with minimal ROAs the attack is reduced to a forged-origin exact prefix hijack, which still passes.
- Route leaks: a leaked route keeps its real origin, so ROV marks it Valid.
- Tampered validation data: RFC 6811 warns that blocking Invalid creates a denial of service vector if an attacker can add or remove records in the validation database.
Prefix filtering: threat model and limits
What it stops: a customer or peer announcing anything outside its registered customer cone, whether or not the prefix has a ROA. That includes leaked full tables, internal more-specifics (RFC 7908 Type 6), and announcements of space the customer simply does not hold. RFC 7454 section 6.2.2.1 says inbound customer policy should accept only the customer's prefixes and discard the rest.
Architecture in words: the customer registers route objects and an AS-SET listing itself and its own customers; your tooling recursively expands the AS-SET, collects every route object whose origin is in that set, aggregates the result and pushes a per-session prefix list. Around it sit the generic filters RFC 7454 lists: special-purpose prefixes marked not globally routable by IANA, too-specific routes (IPv4 longer than /24, IPv6 longer than /48 as a common convention), your own prefixes, the IXP LAN and default route where not expected.
What it misses:
- Weak registry authorisation: in the RIPE Database a route object needs authorisation from the address holder, and RIPE does not check the origin AS holder. Other registries apply their own rules, and some accept objects for space they do not administer, so a stale or false object can put a prefix into your filter.
- AS-SET sprawl: a customer's AS-SET that includes a large transit network's AS-SET expands to a large share of the internet and turns the filter into an open door.
- Peers and transit: you rarely hold a meaningful filter for a transit provider, so prefix filtering offers little on those sessions beyond bogon and too-specific rules.
- Path manipulation: a correctly registered prefix with a forged path or a leak from the customer's other upstream (RFC 7908 Type 1) passes a prefix list.
The gaps neither control closes
| Event | ROV alone | Prefix filter alone | Extra control that helps |
|---|---|---|---|
| Customer originates someone else's signed prefix | Dropped (Invalid) | Dropped if not in cone | Either is enough |
| Customer originates unsigned space it does not hold | Accepted (NotFound) | Dropped | Prefix filter is the control |
| Peer announces signed prefix with wrong origin | Dropped | Usually accepted (no strict filter on peer) | ROV is the control |
| Hairpin leak: customer passes provider A routes to you | Accepted (Valid) | Dropped if filter is strict | Max-prefix, RFC 9234 roles and OTC |
| Lateral leak from a peer (RFC 7908 Type 2 to 4) | Accepted | Accepted | Roles and OTC, AS path filters, ASPA when available |
| Forged-origin exact prefix hijack | Accepted | Accepted unless from a filtered customer | Monitoring; ASPA may detect some cases |
| Forged-origin sub-prefix with loose maxLength | Accepted | Dropped only on a filtered customer | Minimal ROAs (RFC 9319) |
| Rogue IRR object for victim space, victim has ROA | Dropped | Accepted | Prune IRR filters with RPKI |
Path controls fill the leak rows. RFC 9234 (Standards Track, May 2022) adds a BGP Role capability so each eBGP session declares Provider, Customer, RS, RS-Client or Peer; mismatched roles reject the session, and the Only to Customer (OTC) attribute marks routes so they cannot be sent back up to providers or across to peers, with detection possible several hops away. ASPA lets an AS publish its providers in the RPKI so receivers can check the path; the verification document is still an Internet-Draft (draft-ietf-sidrops-aspa-verification-29, dated 16 October 2026 and with the IESG when checked on 14 November 2026), so treat it as something to publish and observe rather than enforce. RFC 7454 section 9 covers AS path sanity: reject your own AS, private ASNs from non-customers, and paths whose first AS is not the neighbour's except at transparent route servers. See the RTBH and BGP Flowspec DDoS Mitigation Guide for how blackhole routes interact with these filters.
Deploying them together
The two controls work best when filter generation uses RPKI as a cross-check and router policy applies them in a fixed order.
- Classify every eBGP session: customer, peer, transit, route server. Record the role in your source of truth; it drives filter choice, max-prefix values and the RFC 9234 role.
- Clean your own data first: publish minimal ROAs and matching route objects in your RIR's registry for everything you originate, and AS0 ROAs for unused space. Keep one AS-SET for your customer cone and remove anything you no longer serve.
- Generate customer filters from both sources: expand the customer's AS-SET, prefer RIR-operated registries, then drop any IRR derived prefix that is RPKI Invalid for the registered origin and add prefixes covered by ROAs for the customer's ASNs. Log every entry removed so the customer can fix its objects.
- Set import policy order: generic rejects (bogons, too-specific, own space, private ASNs, wrong first AS) first; then ROV drop Invalid; then the per-customer prefix list; then max-prefix. Tag accepted routes with community values for customer, peer and transit.
- Set export policy by relationship: send only own and customer routes to peers and transit, and enable RFC 9234 roles where both sides support them.
- Observe before dropping: run ROV in tag-and-log mode and run new customer filters in a compare mode against the routes actually received for two to four weeks; chase every route that would be dropped.
- Enforce in stages: ROV drop on peers and transit, then customers after notice; new strict filters on one customer group at a time.
- Prepare rollback: keep a pre-approved policy that de-preferences Invalid instead of dropping it for a validator or repository failure, and keep the last good prefix list per customer so a failed regeneration never pushes an empty or truncated filter.
Max-prefix belongs with both. RFC 7454 section 8 recommends a limit below the internet table size on peer sessions so a full-table leak tears the session down, and a limit above table size on full-feed upstreams to protect router memory, with regular review.
Decision guidance by network type
| Network | ROV | Prefix filtering | Priority |
|---|---|---|---|
| Single-homed enterprise, provider space | Provider's job | None to run | Contract for ROAs, ROV and customer filtering |
| Multihomed enterprise with own AS | Drop Invalid if you take full tables | Strict outbound list of your own prefixes only | Outbound filter first: it prevents you from becoming a hairpin leaker |
| Hosting or cloud provider with BYOIP customers | All external sessions | Inbound on customers; check customer ROAs authorise your AS before you announce | Both, plus BYOIP onboarding checks |
| Transit ISP | All eBGP sessions | Strict, automated, IRR plus RPKI on every customer | Both, plus roles, AS path filters and max-prefix |
| Internet exchange route server | On route server import | Per member from IRR and RPKI | Both; members benefit even if they do not validate |
Small networks: ROV costs little once two validators run, and a hand-maintained outbound list covers prefix filtering. Large networks: the cost is in filter automation, registry data quality and customer communication; put the effort into generation tooling with RPKI pruning, change review and alerting on large filter deltas.
Worked scenario: a regional ISP closing both gaps
A regional ISP has 300 BGP customers, two transit providers and an IXP presence. It already drops RPKI Invalid on transit sessions and believes it is protected. A customer misconfigures a new router and re-announces 40,000 routes learned from its other provider. Every route is Valid because origins are unchanged, so ROV does nothing. The ISP's customer session has max-prefix set at 1,000, so the session drops before the leak propagates. The review finds that 30 smaller customers have no prefix lists at all because their IRR objects were missing when they were onboarded.
The fix plan: generate prefix lists for every customer from IRR plus RPKI, with the 30 missing customers given a deadline to register route objects or publish ROAs; prune 12 stale IRR entries that RPKI marks Invalid; enable roles on the IXP route server sessions that support them; extend ROV drop to customer sessions; and add an alert when any generated filter changes by more than 20 percent. ROV now catches origin errors from anyone, prefix lists catch unsigned space from customers, max-prefix and roles catch leaks.
Detection and measurement
- External view: watch your prefixes in public route collectors for unexpected origins, more-specifics and unusual paths; internal routers cannot see a hijack of your own space.
- Dropped routes: log routes rejected by ROV and by prefix lists separately, with top originating ASNs, so you can tell customer registry errors from hijacks.
- Filter health: alert on generation failures, empty results and large deltas; track filter age per session.
- Validator health: data age and payload count per validator, and disagreement between implementations.
Metrics:
- Percentage of eBGP sessions dropping RPKI Invalid (target 100 percent).
- Percentage of customer sessions with an automated prefix list refreshed within 24 hours (target 100 percent).
- Percentage of customer sessions with max-prefix and RFC 9234 role configured.
- IRR derived entries pruned as RPKI Invalid per month, trending down as customers fix objects.
- Your own ROA and route object coverage for announced prefixes (target 100 percent).
Common failure patterns
- ROV treated as complete protection: the leak in the scenario above passes untouched; leaks need filters, max-prefix and roles.
- Customers without filters: onboarding accepts "any" until registry objects appear, and the temporary state becomes permanent.
- Over-broad AS-SETs: a customer includes an upstream's AS-SET by mistake and the generated filter permits most of the routing table.
- Failed regeneration pushes an empty list: the customer is blackholed, or worse, automation falls back to permit all.
- Stale IRR objects never pruned: space returned years ago still sits in a third-party registry and in your filters.
- Loose maxLength on your own ROAs: it makes forged-origin sub-prefix hijacks of your space pass every validating network, as RFC 9319 explains.
Where identity fits
Both controls depend on a few powerful accounts: the RIR portal account that creates ROAs and route objects, IRR maintainer credentials, and the automation service accounts that push prefix lists to routers. A stolen maintainer password can add a route object that opens your filters; a stolen RIR login can make your prefixes Invalid. Use named accounts with phishing-resistant MFA, store maintainer and API credentials in a vault, and alert on any object change outside a ticket. See the API Key Management Guide and the Privileged Access Management Guide.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-189 Rev. 1 (Initial Public Draft, 3 January 2025) | RPKI, ROAs, ROA based origin validation, prefix filtering | Recommends both controls; no final version listed when checked on 14 November 2026 |
| NIST SP 800-53 Rev. 5 | SC-7 Boundary Protection; SC-5 Denial-of-service Protection | Filtering routes at the network edge |
| NIST SP 800-53 Rev. 5 | CM-3 Configuration Change Control; SI-4 System Monitoring | Filter generation under change control; route and validator monitoring |
| NIST SP 800-53 Rev. 5 | SA-9 External System Services | Provider commitments on ROV and customer filtering |
| NIST CSF 2.0 | PR.IR-01; PR.PS-01; DE.CM-01; GV.SC-05 | Network protection, configuration management, monitoring, supplier requirements |
| MANRS for network operators | Action 1 (prevent propagation of incorrect routing information); Action 4 (IRR and RPKI) | Compulsory prefix-level customer filtering; registry documentation, ROAs recommended |
| ISO/IEC 27001:2022 | A.8.20 Networks security; A.8.21 Security of network services | Router policy and provider requirements |
Practitioner checklist
- Classify every eBGP session by relationship and record it in your source of truth.
- Publish minimal ROAs and matching route objects for everything you originate.
- Drop RPKI Invalid on all eBGP sessions after a tag-and-log period, with a de-preference rollback.
- Generate strict prefix lists for every customer from IRR, pruned and supplemented with RPKI data, refreshed daily.
- Reject bogons, too-specific routes, your own space, private ASNs from non-customers and wrong first AS.
- Set max-prefix on every session: below table size on peers and customers, above it on full-feed upstreams.
- Enable RFC 9234 roles where supported and publish ASPA objects as a monitoring step.
- Alert on filter generation failures, empty filters and large deltas, and keep the last good list.
- Protect RIR, IRR maintainer and automation credentials with MFA, vaulting and change alerts.
Standards and references
- IETF: RFC 7454 BGP Operations and Security (BCP 194)
- IETF: RFC 6811 BGP Prefix Origin Validation
- IETF: RFC 9319 The Use of maxLength in the RPKI
- IETF: RFC 9234 Route Leak Prevention and Detection Using Roles in UPDATE and OPEN Messages
- IETF: RFC 7908 Problem Definition and Classification of BGP Route Leaks
- IETF: BGP AS_PATH Verification Based on ASPA Objects (Internet-Draft)
- NIST: SP 800-189 Rev. 1 Border Gateway Protocol Security and Resilience (Initial Public Draft)
- MANRS: Actions for Network Operators
Related NHI Mgmt Group resources: RPKI and Route Origin Validation Deployment Guide · BGP and Routing Security Guide · DDoS Protection Guide · Telecom Network Security Guide