Secure Inter-Domain Routing X. Lee Internet-Draft X. Liu Intended status: Informational Z. Yan Expires: January 9, 2017 G. Geng Y. Fu CNNIC July 8, 2016 RPKI Deployment Considerations: Problem Analysis and Alternative Solutions draft-lee-sidr-rpki-deployment-02 Abstract With the global deployment of RPKI, a lot of concerns about technical problems have been and will be raised. In this draft, we collect and analyze the problems that have appeared or that seem likely to appear during the process of RPKI deployment, and suggest some solutions to address or mitigate these problems. Status of This Memo 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 http://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 January 9, 2017. Copyright Notice Copyright (c) 2016 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 (http://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 Lee, et al. Expires January 9, 2017 [Page 1] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 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. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1. RPKI Architecture . . . . . . . . . . . . . . . . . . . . 2 1.2. Status of RPKI Deployment . . . . . . . . . . . . . . . . 3 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 4 3. Considerations of RPKI Deployment . . . . . . . . . . . . . . 4 3.1. More than One TA . . . . . . . . . . . . . . . . . . . . 4 3.2. Problems of CAs . . . . . . . . . . . . . . . . . . . . . 5 3.2.1. Operational Errors . . . . . . . . . . . . . . . . . 5 3.2.2. Unilateral Resource Revocation . . . . . . . . . . . 5 3.3. Mirror World Attacks . . . . . . . . . . . . . . . . . . 5 3.4. Data Synchronization . . . . . . . . . . . . . . . . . . 6 3.5. Problems of Staged and Incomplete Deployment . . . . . . 6 3.6. Low Validation Coverage . . . . . . . . . . . . . . . . . 7 4. Alternative Solutions to RPKI Deployment Problems . . . . . . 8 4.1. Solutions to Multiple TAs . . . . . . . . . . . . . . . . 8 4.2. Solutions to Misbehaving CAs . . . . . . . . . . . . . . 9 4.3. Solutions to Data Synchronization . . . . . . . . . . . . 9 4.4. Solutions to Incomplete Deployment and Low Validation Coverage . . . . . . . . . . . . . . . . . . . . . . . . 10 5. Security Considerations . . . . . . . . . . . . . . . . . . . 10 6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 10 7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 10 8. References . . . . . . . . . . . . . . . . . . . . . . . . . 10 8.1. Normative References . . . . . . . . . . . . . . . . . . 10 8.2. Informative References . . . . . . . . . . . . . . . . . 11 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11 1. Introduction 1.1. RPKI Architecture In RPKI, CAs (Certification Authorities) are organized in a hierarchical structure which is aligned to the existing INR (Internet Number Resources) allocation hierarchy (including IP prefixes and AS numbers). Each INR allocation requires corresponding resource certificates to attest to it, for security. In RPKI, two types of resource certificates [RFC6480] are generated as adjuncts to this allocation process: CA certificates and EE (End-entity) certificates. CA certificates attest to the INR holdings; EE certificates are primarily used for ROAs (Route Origin Authorizations) [RFC6482] and Router Certificates. ROAs are used to bind IP prefixes to the ASes Lee, et al. Expires January 9, 2017 [Page 2] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 that is permitted to originate routes for these IP prefixes. Manifests [RFC6486] are also validated using EE certificates. Manifests are used to ensure the integrity of the RPKI repository system. The process of using the RPKI to verify the origin of a route is as follows. 1. CAs, including IANA (Internet Assigned Numbers Authority), five RIRs (Regional Internet Registries), NIRs (National Internet Registries) and ISPs (Internet Service Providers), publish authoritative objects (including resource certificates, ROAs, Manifest and so on) into their repositories. 2. RPs (Relying Parties) all over the world collect (using rsync or RRDP protocol [I-D.ietf-sidr-delta-protocol]) and verify (using rcynic or RPSTIR) the RPKI objects from these repositories, and provide the results of verification to BGP border routers or other routing practices such as RPSL-based. 3. Finally, BGP border routers can make use of these results to verify the route origin information in the BGP update messages they receive. This may be done by generating route filters from the validated RPKI data, or by using the RPKI-to-router protocol [RFC6810]. 1.2. Status of RPKI Deployment Each of the five RIRs has initiated the deployment of RPKI, and each now offers RPKI services to its members. A number of countries (Ecuador, Japan, Bangladesh, etc.) have also started to test and deploy RPKI internally. In order to promote the deployment of RPKI, ICANN (Internet Corporation for Assigned Names and Numbers), the five RIRs, many NIRs and companies have making continuous efforts to solve the existing problems and improve the corresponding policies and technical standards. However, RPKI is still in its early stages of global deployment. According to the data provided by RPKI Dashboard as of July 2016, the current routing table holds about 659,271 IP prefixes in total, and the RPKI validation state has been determined for 44983 IP prefixes, which means that only 6.82% of the prefixes in the routing table can be validated using the RPKI. Table 1 details of the RPKI "adoption rate" (the percentage of members deployed RPKI) in each of the RIRs. Lee, et al. Expires January 9, 2017 [Page 3] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 +---------------+---------+-------+-------+--------+----------+ | RIR | AFRINIC | APNIC | ARIN | LACNIC | RIPE NCC | +---------------+---------+-------+-------+--------+----------+ | Adoption Rate | 0% | 3.44% | 1.22% | 20.66% | 12.14% | +---------------+---------+-------+-------+--------+----------+ Table 1.Adoption rate of RPKI in 5 RIRs As we can see from Table 1, LACNIC has the highest adoption rate, which is about 20.66%. While the adoption rates in ARIN and AFRINIC are much lower, which are only 1.22% and 0% respectively. RIPE NCC provides some statistics regarding the number of resource certificates and ROAs in each RIR. From these statistics we find a good sign that the global deployment status of RPKI rises gradually, and with its further evolution, the global adoption rate of RPKI should achieve a faster growth. 2. Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119]. 3. Considerations of RPKI Deployment During the process of incremental deployment of RPKI, several technical problems have appeared and others may appear. In this section, we attempt to collect and analyze the problems that seem most critical. 3.1. More than One TA A TA (Trust Anchor) is an authoritative entity represented by a public key and its associated data [RFC5914]. The public key is used to verify digital signatures and the associated data describes the types of information and actions for which the TA is authoritative. There are multiple TAs in the RPKI architecture today, for example, the five RIRs are generally viewed as default TAs. With more than one TA, there is no technical mechanism to prevent two or more TAs from asserting control over the same set of INRs accidentally or maliciously, which means that certificates might be issued for allocations of the overlapping INRs. This, in turn, may lead to inconsistent and conflicting assertions about to whom the specific INRs have been allocated. This kind of problem obviously may cause resource conflicts on the Internet. Lee, et al. Expires January 9, 2017 [Page 4] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 3.2. Problems of CAs 3.2.1. Operational Errors Operatioanl errors by CAs are inevitable and may cause significant impact on Internet routing. Thus such errors by CAs in RPKI constitute a risk to widespread deployment. Operatioanl errors by CAs in the RPKI may lead to serious consequences similar to those caused by malicious attacks (black-hole routes, traffic interception, and denial-of-service attacks). For example, an error in using a ROA (such as adding a new erroneous ROA or whacking an existing ROA) may cause all routes covered by the (original) ROA to become invalid (or to assume an "unknown" security status [RFC6483]). Note that, if the old validating ROA still matches (not just covers) the announce prefix, the announcement will still be marked as valid. 3.2.2. Unilateral Resource Revocation In the RPKI architecture, there is a risk that CAs have the power to unilaterally revoke the INRs that have been allocated to their descendants, e.g., by revoking corresponding CA certificates [RFC6480]. This is a natural aspect of PKIs and it is a necessary capability for CAs as they manage re-allocation of resources within their domains. However, if revocation occurs accidentally, or because the CA has been compelled by authorities, the results can be significant. Specifically, all RPs will view the origin assertions by the CA (and its descendants) to be not found. This may cause ISPs to depreference routes to the affected prefixes. 3.3. Mirror World Attacks In mirror world attacks, a malicious CA presents one view of the RPKI repository (that it manages) to some RPs, and a different view to others. (Because repository data may be cached by ISPs, it may not be possible for a malicious CA to provide erroneous results to a narrowly targeted set of RPs.) Since a CA in the RPKI controls everything in its own repository, it may be easy for a malicious CA to perform such a attack. For example, a malicious CA presents the correct view of its repository to some RPs, but a forged view (e.g., the CA adds a specific, erroneous ROA) to the others. When these deceived RPs offer their validation results to BGP routers, the routers may abandon the Lee, et al. Expires January 9, 2017 [Page 5] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 legitimate routes that are considered to be invalid according to the (erroneous) validation results they have received. 3.4. Data Synchronization It is required in [RFC6480] that all repositories must be accessible via rsync protocol which is used by RPs to get the RPKI objects in the global distributed repositories. However, the rsync protocol is considered to be controversial with respect to the following disadvantages: 1. Lack of standards and non-modular implementation: Although rsync is widely adopted in backup, restore, and file transfer, it has not been standardized by IETF. And the rsync implementation is non-modular, making it difficult to use its source code. 2. Underlying overhead caused by repository updates during active data transmissions: During data transmissions between RPs and the repository, a new update to the repository may cause data inconsistency between them. In order to rectify this inconsistency, extra overhead costs (such as performing the synchronization once more) are required. 3. This is being solved by the new RRDP protocol, now in deployment. 3.5. Problems of Staged and Incomplete Deployment Since the global deployment of RPKI is an incremental and staged process, unexpected problems may appear during this process. Let's take an example to explain why the incomplete deployment of RPKI may cause legitimate routes to be misclassified into invalid. In Fig.1, we make the following assumptions: 1. CNNIC, ISP1 and ISP2 have deployed the RPKI, but ISP3 has not yet. ISP1 and ISP2 received allocations form CNNIC, and ISP3 received its allocation from ISP1. 2. CNNIC allocated IP prefix 218.241.104.0/22 to ISP1 and 218.241.108.0/22 to ISP2. 3. Three ROAs (ROA1, ROA2, ROA3) are issued respectively by CNNIC, ISP1 and ISP2. Lee, et al. Expires January 9, 2017 [Page 6] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 ------- -------------- |APNIC| | Resource | ------- |Certificates| | -------------- | .............. ................. ------- . ROA . . ROA1: . | | .............. .218.241.96.0/20.<--|CNNIC| . AS1 . | | ................. ------- / \ / \ .................. ------ ------ .................. . ROA2: . | | | | . ROA3: . .218.241.104.0/22.<--|ISP1| |ISP2|-->.218.241.108.0/22. . AS2 . | | | | . AS3 . .................. ------ ------ .................. | | ------ | | |ISP3| | | ------ Fig.1: An example of incomplete deployment Now ISP3 announces to be the origin of 218.241.106.0/23. When other entities receive this announcement, they can validate it with ROAs information. Since prefix 218.241.104.0/22 described in ROA2 encompasses prefix 218.241.106.0/23 and no matching ROA describes 218.241.106.0/23 could be found [RFC6483], the announcement for prefix 218.241.106.0/23 will be considered to be invalid. This example illustrates why careful coordination is needed when (non- leaf) ISPs incrementally deploy the RPKI. This example illustrates why careful coordination is needed when (non-leaf) ISPs incrementally deploy the RPKI. Therefore, if an ISP knows its customer is not creating a ROA, it is the ISP's responsibility to create that ROA, just as it is that ISP's responsibility to do 42 other things for their 'customer'. 3.6. Low Validation Coverage The route origin validation coverage refers to the percentage of valid routes attested to by the RPKI. i.e., Coverage = Lee, et al. Expires January 9, 2017 [Page 7] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 number_of_valid_routes / (number_of_valid_routes + number_of_invalid_routes). As we can see from Table 2, the coverage of route origin validation in the five RIRs differs a lot. LACNIC and RIPE NCC have the highest validation coverage and both of them are over 90%, while the coverage in APNIC is less than 70%. Many reasons may account for the low validation coverage, such as misconfigurations, low RPKI adoption rates, etc. +---------+-------+-------+---------+---------+----------+----------+ | RIR | Total | Valid | Invalid | Unknown | Accuracy | Adoption | | | | | | | | Rate | +---------+-------+-------+---------+---------+----------+----------+ | AFRI- | 14948 | 242 | 5 | 14701 | 97.98% | 1.65% | | NIC | | | | | | | | APNIC | 15802 | 3332 | 1564 | 153124 | 68.06% | 3.1% | | | 0 | | | | | | | ARIN | 21977 | 1911 | 337 | 217531 | 85.01% | 1.02% | | | 9 | | | | | | | LACNIC | 76841 | 13379 | 736 | 62726 | 94.79% | 18.37% | | RIPE | 15925 | 16771 | 1307 | 141178 | 92.77% | 11.35% | | NCC | 6 | | | | | | +---------+-------+-------+---------+---------+----------+----------+ Table 2. Route Origin Validation Accuracy in 5 RIRs 4. Alternative Solutions to RPKI Deployment Problems In this section, we propose and analyze the alternative solutions and strategies to solve or mitigate the problems mentioned in Section 3. 4.1. Solutions to Multiple TAs The RIRs say they are trying to continually evolve RPKI. ICANN (IANA) and RIRs have developed a technical testbed with an RPKI GTA. It's assumed that there must be a single root trust anchor eventually. With this single root trust anchor deployed, the risks of resource conflicts (at the level of RIR certificates) could be significantly reduced. However, this solution cedes more power to ICANN (IANA) and thus might exacerbate the risk of "Unilateral Resource Revocation" (power imbalance) mentioned in Section 3.2.2. Lee, et al. Expires January 9, 2017 [Page 8] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 4.2. Solutions to Misbehaving CAs S. Kent et al. put forward a collection of mechanisms named "Suspenders". "Suspenders" is designed to address the adverse effects on INR holders which were caused by CAs' accidental or deliberate misbehavior or attacks on CAs and repositories. This mechanism imports two new objects: an INRD (Internet Number Resource Declaration) file and a LOCK object. The INRD file is external to the RPKI repository, and it contains the most recent changes that were made by the INR holder. The LOCK object is published in the INR holder's repository. It contains a URL which points to the INRD file, and a public key used to verify the signature of INRD file. Whenever the RPs detect the inconsistencies between the actual changes and the INRD file, they can determine individually whether to accept these changes or not. (This proposal is being revised to address operational concerns, but it is anticipated that a subsequent version of Suspenders will preserve the primary features noted above.) 4.3. Solutions to Data Synchronization A number of alternative protocols have been presented to take the place of "rsync" protocol due to its shortcomings mentioned above. 1) RRDP T. Bruijnzeels et al. have proposed an alternative protocol (RRDP, RPKI Repository Delta Protocol) for RPs to keep their local caches in sync with the repository system [I-D.ietf-sidr-delta-protocol]. This new protocol is based on notification, snapshot and delta files. When RPs query a repository for updates, they will use delta files (and snapshot files as needed) to keep their local caches updated. Moreover, RRDP protocol can work with the existing rsync URIs. Compared with rsync protocol, RRDP is considered to be effective as a way to eliminate a number of consistency related issues, help to reduce the load on publication servers, and have improved scalability. RRDP is in current RIPE and DRL software. 2) Improved Rsync Protocol CNNIC also proposed an improved rsync mechanism which transfers the work of checksums calculation to RPs in order to reduce the computation load on the rsync server side. The mechanism also offered a NOTIFY method that send NOTIFY message to make some important RPs to actively fetch the updated RPKI objects in time. Lee, et al. Expires January 9, 2017 [Page 9] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 4.4. Solutions to Incomplete Deployment and Low Validation Coverage Both of the two problems (incomplete deployment and low validation accuracy) are caused by the partial deployment of RPKI. With the widely deployment of RPKI in the near future, these two problems ought to be mitigated. 5. Security Considerations TBD 6. IANA Considerations This draft does not request any IANA action. 7. Acknowledgements The authors would like to thanks the valuable comments made by Stephen Kent and other members of sidr WG. This document was produced using the xml2rfc tool [RFC2629]. 8. References 8.1. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC6480] Lepinski, M. and S. Kent, "An Infrastructure to Support Secure Internet Routing", RFC 6480, DOI 10.17487/RFC6480, February 2012, . [RFC6482] Lepinski, M., Kent, S., and D. Kong, "A Profile for Route Origin Authorizations (ROAs)", RFC 6482, DOI 10.17487/RFC6482, February 2012, . [RFC6483] Huston, G. and G. Michaelson, "Validation of Route Origination Using the Resource Certificate Public Key Infrastructure (PKI) and Route Origin Authorizations (ROAs)", RFC 6483, DOI 10.17487/RFC6483, February 2012, . Lee, et al. Expires January 9, 2017 [Page 10] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 [RFC6486] Austein, R., Huston, G., Kent, S., and M. Lepinski, "Manifests for the Resource Public Key Infrastructure (RPKI)", RFC 6486, DOI 10.17487/RFC6486, February 2012, . 8.2. Informative References [I-D.ietf-sidr-delta-protocol] Bruijnzeels, T., Muravskiy, O., Weber, B., and R. Austein, "RPKI Repository Delta Protocol", draft-ietf-sidr-delta- protocol-03 (work in progress), July 2016. [RFC2629] Rose, M., "Writing I-Ds and RFCs using XML", RFC 2629, DOI 10.17487/RFC2629, June 1999, . [RFC5914] Housley, R., Ashmore, S., and C. Wallace, "Trust Anchor Format", RFC 5914, DOI 10.17487/RFC5914, June 2010, . [RFC6810] Bush, R. and R. Austein, "The Resource Public Key Infrastructure (RPKI) to Router Protocol", RFC 6810, DOI 10.17487/RFC6810, January 2013, . Authors' Addresses Xiaodong Lee CNNIC No.4 South 4th Street, Zhongguancun Beijing, 100190 P.R. China Email: xl@cnnic.cn Xiaowei Liu CNNIC No.4 South 4th Street, Zhongguancun Beijing, 100190 P.R. China Email: liuxiaowei@cnnic.cn Lee, et al. Expires January 9, 2017 [Page 11] Internet-Draft draft-lee-sidr-pki-deployment-02 July 2016 Zhiwei Yan CNNIC No.4 South 4th Street, Zhongguancun Beijing, 100190 P.R. China Email: yanzhiwei@cnnic.cn Guanggang Geng CNNIC No.4 South 4th Street, Zhongguancun Beijing, 100190 P.R. China Email: gengguanggang@cnnic.cn Yu Fu CNNIC No.4 South 4th Street, Zhongguancun Beijing, 100190 P.R. China Email: fuyu@cnnic.cn Lee, et al. Expires January 9, 2017 [Page 12]