Internet Engineering Task Force Curtis Villamizar INTERNET-DRAFT UUNET draft-ietf-rps-auth-02 Cengiz Alaettinoglu ISI David M. Meyer Cisco Sandy Murphy TIS February 23, 1999 Routing Policy System Security Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. 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 mate- rial or to cite them other than as ``work in progress.'' The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Copyright (C) The Internet Society (February 23, 1999). All Rights Reserved. Abstract The RIPE database specifications [2] and RPSL language [1] define lan- guages used as the basis for representing information in a routing policy system. A repository for routing policy system information is known as a routing registry. A routing registry provides a means of INTERNET-DRAFT Routing Policy System Security February 23, 1999 exchanging information needed to address many issues on importance to the operation of the Internet. The implementation and deployment of a routing policy system must maintain some degree of integrity to be of any operational use. This document addresses the need to assure integrity of the data by providing an authentication and authorization model. 1 Overview The Internet Routing Registry (IRR) has evolved to meet a need for Internet-wide coordination. This need was described in RFC-1787, an informational RFC prepared on behalf of the IAB [17]. The following summary appears in Section 7 of RFC-1787. While ensuring Internet-wide coordination may be more and more difficult, as the Internet continues to grow, stability and con- sistency of the Internet-wide routing could significantly benefit if the information about routing requirements of various organi- zations could be shared across organizational boundaries. Such information could be used in a wide variety of situations ranging from troubleshooting to detecting and eliminating conflicting routing requirements. The scale of the Internet implies that the information should be distributed. Work is currently underway to establish depositories of this information (Routing Registries), as well as to develop tools that analyze, as well as utilize this information. A routing registry must maintain some degree of integrity to be of any use. The degree of integrity required depends on the usage of the routing policy system. An initial intended usage of routing policy systems such as the RIPE database had been in an advisory capacity, documenting the intended routing policies for the purpose of debugging. In this role a very weak form of authentication was deemed sufficient. The IRR is increasingly used for purposes that have a stronger re- quirement for data integrity and security. This document addresses issues of data integrity and security that is consistent with the usage of the IRR and which avoids compromising data integrity and se- curity even if the IRR is distributed among less trusted repositories. Villamizar, et. al. Expires August 23, 1999 [Page 2] INTERNET-DRAFT Routing Policy System Security February 23, 1999 2 Background An early routing policy system used in the T3--NSFNET, the policy routing database (PRDB), provided a means of determining who was au- thorized to announce specific prefixes to the NSFNET backbone. The need for a policy database was recognized as far back as 1989 [6, 4]. By 1991 the database was in place [5]. Authentication was accom- plished by requiring confirmation and was a manually intensive pro- cess. This solved the problem for the NSFNET, but was oriented toward holding the routing policy of a single organization. The problem since has become more difficult. New requirements have emerged. 1. There is a need to represent the routing policies of many organiza- tions. 2. CIDR and overlapping prefixes and the increasing complexity of routing policies and the needs of aggregation have introduced new requirements. 3. There is a need to assure integrity of the data and delegate au- thority for the data representing specifically allocated resources to multiple persons or organizations. 4. There is a need to assure integrity of the data and distribute the storage of data subsets to multiple repositories. The RIPE effort specificly focused on the first issue [2]. Its prede- cessor, the PRDB, addressed the needs of a single organization, while the RIPE database was initially intended to address the needs of the European Internet community. The RIPE database formats as described in [2] were the basis of the original IRR. Routing protocols themselves provide no assurance that the origination of a route is legitimate and can actually reach the stated destina- tion. The nature of CIDR allows more specific prefixes to override less specific prefixes [9, 18, 8]. Even with signed route origina- tion, there is no way to determine if a more specific prefix is legit- imate and should override a less specific route announcement without a means of determining who is authorized to announce specific prefixes. Failing to do so places no assurance of integrity of global routing information and leaves an opportunity for a very effective form of denial of service attack. The Routing Policy System Language (RPSL) [1, 13] was a fairly sub- stantial evolutionary step in the data representation which was largely targeted at addressing the second group of needs. The PRDB Villamizar, et. al. Expires August 23, 1999 [Page 3] INTERNET-DRAFT Routing Policy System Security February 23, 1999 accommodated CIDR in 1993 [12] and the RIPE database accommodated the entry of CIDR prefixes from inception, but RPSL provides many needed improvements including explicit support for aggregation. This document addresses the third group of needs identified above. While the current implementation supporting weak authentication doesn't guarantee integrity of the data, it does provide extensive mechanisms to make sure that all involved parties get notified when a change is made to the database, whether the change was malicious or intended. This provides inadequate protection against additions. Since the software is increasingly used to configure the major parts of the Internet infrastructure, it is not considered to be adequate anymore to know about and have the ability roll back unintended changes. Therefore, more active security mechanism need to be de- veloped to prevent such problems before they happen. A separate document addresses the fourth group of needs [14]. 3 Implicit Policy Assumptions The authorization model encodes certain policies for allocation of address numbers, AS numbers, and for the announcement of routes. Im- plicit to the authorization model are a very limited number of policy assumptions. 1. Address numbers are allocated hierarchically. The IANA delegates portions of the address space to the regional registries (currently ARIN, APNIC and RIPE), which in turn delegate address space to their members, who can assign addresses to their customers. 2. AS numbers are allocated either singly or in small blocks by reg- istries. Registries are allocated blocks of AS numbers, thereby making the allocation hierarchical. 3. Routes should only be announced with the consent of the holder of the origin AS number of the announcement and with the consent of the holder of the address space. 4. AS numbers and IP address registries may be different entities from routing registries. For subsets of any of these three allocation spaces, network ad- dresses, AS numbers, and routes, these restrictions may be loosened or disabled by specifying a very weak authorization method or an authentication method of ``none''. However, even when no authenti- cation mechanism is used, all involved parties can be notified about Villamizar, et. al. Expires August 23, 1999 [Page 4] INTERNET-DRAFT Routing Policy System Security February 23, 1999 the changes that occurred through use of the existing ``notify'' at- tribute. 4 Organization of this Document Familiarity with RIPE-181 [2] and RPSL [1] is assumed throughout this document. Goals are described in Section 5. Section 6 through Sec- tion 8 provide descriptions of the changes and discussion. Section 9 provides a concise summary of data formats and semantics. Appendix A through Appendix C provide additional technical discussion, examples, and deployment considerations. Goals and Requirements Section 5 provides a more detailed descrip- tion of the issues and identifies specific problems that need to be solved, some of which require a degree of cooperation in the Internet community. Data Representation Section 6 provides some characteristics of RPSL and formats for external representations of information. Authentication Model Section 7 describes current practice, proposes additional authentication methods, and describes the extension mechanism if additional methods are needed in the future. Authorization Model Section 8 describes the means of determining whether a transaction contains the authorization needed to add, modify, or delete specific data objects, based on stated authenti- cation requirements in related data objects. Data Format Summaries Section 9 provides a concise reference to the data formats and steps in transaction processing. Technical Discussion Section A contains some discussion of techni- cal tradeoffs. Common Operational Cases Section B provides some examples drawn from past operational experience with the IRR. Deployment Considerations Section C describes some deployment is- sues and discusses possible means of resolution. 5 Goals and Requirements The Internet is an open network. This openness and the large scale of the Internet can present operational problems. Technical weaknesses Villamizar, et. al. Expires August 23, 1999 [Page 5] INTERNET-DRAFT Routing Policy System Security February 23, 1999 that allow misconfiguration or errant operation in part of the network to propagate globally or which provide potentials for simple denial of service attacks should be eliminated to the extent that it is prac- tical. The integrity of routing information is critical in assuring that traffic goes where it is supposed to. An accidental misconfiguration can direct traffic toward routers that cannot reach a destination for which they are advertising reachabil- ity. This is commonly caused by misconfigured static routes though there are numerous other potential causes. Static routes are often used to provide constant apparent reachability to single homed desti- nations. Some of the largest ISPs literally have thousands of static routes in their networks. These are often entered manually by op- erators. Mistyping can divert traffic from a completely unrelated destination to a router with no actual reachability to the advertised destination. This can happen and does happen somewhat regularly. In addition, implementation bugs or severe misconfigurations that result in the loss of BGP AS path information or alteration of prefix length can result in the advertisement of large sets of routes. Though con- siderably more rare, on a few occasions where this has occurred the results were catastrophic. Where there is the potential for an accidental misconfiguration in a remote part of the Internet affecting the global Internet there is also the potential for malice. For example, it has been demonstrated by accident that multiple hour outages at a major institution can be caused by a laptop and a dial account if proper precautions are not taken. The dial account need not be with the same provider used by the major institution. The potential for error is increased by the CIDR preference for more specific routes [8]. If an institution advertises a single route of a given length and a distant router advertises a more specific router covering critical hosts, the more specific route, if accepted at all, is preferred regardless of administrative weighting or any routing protocol attributes. There is a need to provide some form of checks on whether a route ad- vertisement is valid. Today checks are typically made against the border AS advertising the route. This prevents accepting routes from the set of border AS that could not be legitimately advertise the route. Theses checks rely on the use of information registered in the IRR to generate lists of prefixes that could be advertised by a specific border AS. Checks can also be made against the origin AS. If policy information were sufficiently populated, checks could be made against the entire AS path, but this is not yet feasible. The use of a routing registry can also make it more difficult for pre- fixes to be used without authorization such as unallocated prefixes or prefixes allocated to another party. Villamizar, et. al. Expires August 23, 1999 [Page 6] INTERNET-DRAFT Routing Policy System Security February 23, 1999 In summary, some of the problems being addressed are: o Localizing the impact of accidental misconfiguration made by Inter- net Providers to that provider's networks only. o Eliminating the potential for an Internet provider's customer to use malicious misconfiguration of routing as a denial of service attack if the provider router filters their customers and local- izing the denial of service to that Internet provider only if the immediate Internet service provider does not route filter their customers but other providers route filter the route exchange at the inter-provider peering. o Eliminating the unauthorized use of address space. If the data within a routing registry is critical, then the ability to change the data must be controlled. Centralized authorities can provide control but centralization can lead to scaling problems (and is politically distasteful). Address allocation and name allocation is already delegated. Since delegation can be to outside registries it is at least somewhat dis- tributed [11]. Autonomous System (AS) numbers are allocated by the same authorities. It makes sense to delegate the routing number space in a manner similar to the address allocation and AS number alloca- tion. The need for this delegation of authority to numerous reg- istries increases the difficulty of maintaining the integrity of the body of information as a whole. As a first step, the database can be somewhat centrally administered with authority granted to many parties to change the information. This is the case with the current IRR. There are a very small number of well trusted repositories and a very large number of parties au- thorized to make changes. Control must be exercised over who can make changes and what changes they can make. The distinction of who vs what separates authentication from authorization. o Authentication is the means to determine who is attempting to make a change. o Authorization is the determination of whether a transaction pass- ing a specific authentication check is allowed to perform a given operation. Different portions of the database will require different methods of authentication. Some applications will require authentication based Villamizar, et. al. Expires August 23, 1999 [Page 7] INTERNET-DRAFT Routing Policy System Security February 23, 1999 on strong encryption. In other cases software supporting strong en- cryption may not be necessary or may not be legally available. For this reason multiple authentication methods must be supported, se- lected on a per object basis. The authentication methods may range from very weak data integrity checks to cryptographicly strong sig- natures. The authorization model must insure that the use of weak integrity checks in parts of the database does not compromise the overall integrity of the database. Additional requirements are placed on the authorization model if the database is widely distributed with delegations made to parties that may not be trustworthy or whose security practices may be lacking. This problem must be addressed in the authorization model in order to enable later evolution to a more distributed routing registry. Autonomous system numbers can be delegated in blocks and subdelegated as needed and then individual AS numbers assigned. Address alloca- tion is a simple numeric hierarchy. Route allocation is somewhat more complicated. The key attributes in a route object (key with re- gard to making it unique) contains both an address prefix and an AS number, known as the origin AS. The addition of a route object must be validated against both the authorization criteria for the AS and the address prefix. Route objects may exist for the same prefix with multiple origin AS values due to a common multihoming practice that does not require a unique origin AS. There is often no correlation be- tween the origin AS of a prefix and the origin AS of overlapping more specific prefixes. There are numerous operational cases that must be accommodated. Some of the more common are listed below. These are explored in greater detail in Appendix B with discussion of technical tradeoffs in Ap- pendix A. o simple hierarchical address allocation and route allocation o aggregation and multihomed more specific routes o provider independent addresses and multiple origin AS o changing Internet service providers o renumbering grace periods The authorization model must accommodate a variety of policies regard- ing the allocation of address space and cannot mandate the use of any one model. There is no standardization of address allocation policies though guidelines do exist [11, 19]. Whether authorization allows the recovery of address space must be selectable on a per object basis and may differ in parts of the database. This issue is discussed further in Appendix A. Villamizar, et. al. Expires August 23, 1999 [Page 8] INTERNET-DRAFT Routing Policy System Security February 23, 1999 6 Data Representation RPSL provides a complete description of the contents of a routing repository [1]. Many RPSL data objects remain unchanged from the RIPE and RPSL references the RIPE-181 specification as recorded in RFC-1786 [2]. RPSL provides external data representation. Data may be stored differently internal to a routing registry. Some database object types or database attributes must be added to RPSL to record the delegation of authority and to improve the authen- tication and authorization mechanisms. These additions are very few and are described in Section 7 and Section 8. Some form of encapsulation must be used to exchange data. The de- facto encapsulation has been the one which the RIPE tools accept, a plain text file or plain text in the body of an RFC-822 formatted mail message with information needed for authentication derived from the mail headers or the body of the message. Merit has slightly modified this using the PGP signed portion of a plain text file or PGP signed portion of the body of a mail message. These very simple forms of encapsulation are suitable for the initial submission of a database transaction. The encapsulation of registry transaction submissions, registry queries and registry responses and exchanges between registries is outside the scope of this document. The encapsulation of registry transaction submissions and exchanges between registries is covered in [14]. 7 Authentication Model The maintainer objects serve as a container to hold authentication filters. A reference to a maintainer within another object defines authorization to perform operations on the object or on a set of re- lated objects. The maintainer is typically referenced by name in mnt- by attributes of objects. Further details on the use of maintainers are provided in Section 8.1. The maintainer contains one or more ``auth'' attributes. Each ``auth'' attribute begins with a keyword identifying the authenti- cation method followed by the authentication information needed to enforce that method. The PGPKEY method is slightly syntactically different in that the method PGPKEY is a substring. Authentication methods currently supported include the following. Note that pgp-from is being replaced by the pgpkey (see Section 9 and [22]). Villamizar, et. al. Expires August 23, 1999 [Page 9] INTERNET-DRAFT Routing Policy System Security February 23, 1999 mail-from This is a very weak authentication check and is discour- aged. The authentication information is a regular expression over ASCII characters. The maintainer is authenticated if the from or reply-to fields in RFC-822 mail headers are matched by this regular expression. Since mail forgery is quite easy, this is a very weak form of authentication. crypt-pw This is another weak form of authentication. The authenti- cation information is a fixed encrypted password in UNIX crypt for- mat. The maintainer is authenticated if the transaction contains the clear text password of the maintainer. Since the password is in clear text in transactions, it can be captured by snooping. Since the encrypted form of the password is exposed, it is subject to password guessing attacks. pgp-from This format is being replaced by the ``pgpkey'' so that the public key certificate will be available to remote repositories. This is Merit's PGP extension. The authentication information is a signature identity pointing to an external public key ring. The maintainer is authenticated if the transaction (currently PGP signed portion of a mail message) is signed by the corresponding private key. pgpkey This keyword takes the form ``PGPKEY-hhhh'', where ``hhhh'' is the hex representation of the four bytes id of the PGP public key used for authentication. The public key certificate is stored in a separate object as described in [22]. Repositories may elect to disallow the addition of ``auth'' attributes specifying weaker forms of authentication and/or disallow their use in local transaction submissions. Repositories are encouraged to dis- allow the addition of ``auth'' attributes with the deprecated ``pgp- from'' method. Any digital signature technique can be used for authentication. Transactions should be signed using multiple digital signature tech- niques to allow repositories or mirrors that only use a subset of the techniques to verify at least one of the signatures. Any digital sig- nature techniques would be applicable. One that may be supported in the in the future is DSA [15, 16]. Numerous digital signature algo- rithms are described in [21]. 8 Authorization Model The authorization model must accommodate the requirements outlined in Section 5. A key feature of the authorization model is the recog- nition that authorization for the addition of certain types of data objects must be derived from related data objects. Villamizar, et. al. Expires August 23, 1999 [Page 10] INTERNET-DRAFT Routing Policy System Security February 23, 1999 With multiple repositories, objects not found in RPSL are needed to control AS delegations and new attributes are needed in existing ob- jects to control subdelegation. Objects are also needed to provide query information for other repositories. 8.1 Maintainer Objects The maintainer objects serve as a container to hold authentication filters. The authentication methods are described in Section 7. The maintainer can be referenced by name in other objects, most notably in the mnt-by attributes of those objects. Maintainers themselves contain mnt-by attributes. In some cases the mnt-by in a maintainer will reference the maintainer itself. In this case, authorization to modify the maintainer is provided to a (usu- ally very limited) set of identities. A good practice is to create a maintainer containing a long list of identities authorized to make specific types of changes but have the maintainer's mnt-by attribute reference a far more restrictive maintainer more tightly controlling changes to the maintainer object itself. The mnt-by attribute is mandatory in all objects. Some data already exists without mnt-by attributes. A missing mnt-by attribute is in- terpreted as the absence of any control over changes. This is highly inadvisable and most repositories will no longer allow this. The ``mnt-routes'' attribute are needed to reference maintainer ob- jects to provide specific permissions related to the object. This is an extensions to RPSL and RIPE-181 proposed in this document and are described in detail in Section 9. A mnt-routes attribute in an aut-num object allows addition of route objects with that AS number as the origin to the maintainers listed. A mnt-routes attribute in an inetnum object allows addition of route objects with exact matching or more specific prefixes. A mnt-routes attribute in a route object allows addition of route objects with ex- act matching or more specific prefixes. A mnt-routes attribute does not allow changes to the aut-num, inetnum, or route object where they appear. A mnt-routes may optionally be constrained to only apply to a subset of more specific routes. 8.2 as-block and aut-num objects An ``as-block'' object is needed to delegate a range of AS numbers to a given repository. This is needed for authorization and it is needed to avoid having to make an exhaustive search of all repositories to find a specific AS. This search would not be a problem now but would Villamizar, et. al. Expires August 23, 1999 [Page 11] INTERNET-DRAFT Routing Policy System Security February 23, 1999 be if a more distributed routing repository is used [14]. The ``as-block'' object also makes it possible to separate AS number allocation from registration of AS routing policy. as-block: AS1321 - AS1335 ... The ``aut-num'' describes the routing policy for an AS and is criti- cal for router configuration of that AS and for analysis performed by another AS. For the purpose of this document it is sufficient to con- sider the aut-num solely as a place holder identifying the existence of an AS and providing a means to associate authorization with that AS for the purpose of adding ``route'' objects. The ``as-block'' object is proposed here solely as a means of record- ing the delegation of blocks of AS numbers to alternate registries and in doing so providing a means to direct queries and a means to support hierarchical authorization across multiple repositories. 8.3 inetnum objects A delegation attribute is needed in the inetnum and route object. This will accommodate the delegation of address space from IANA to regional IP registries. When the routing registry becomes more widely distributed a delegation attribute is needed to support any subdelega- tions to more localized registries or delegations to Internet provider operated registries or organizations who may prefer to run their own routing registry. The delegation attribute for an inetnum or a route object can be multi-valued and refers to all registries in which more specific route objects can be found. inetnum: 193.0.0.0 - 193.0.0.255 ... source: IANA The ``inetnum'' exists to support address allocation. For external number registries, such as those using ``[r]whoisd[++]'' the ``inet- num'' can serve as a secondary record that is added when an address allocation is made in the authoritative database. Such records could be added by a address registry such as ARIN as a courtesy to the cor- Villamizar, et. al. Expires August 23, 1999 [Page 12] INTERNET-DRAFT Routing Policy System Security February 23, 1999 responding routing registry. 8.4 route objects Currently there are a quite few route objects in more than one reg- istry. Quite a few are registered with origin AS for which they have never been announced. There is a legitimate reason to be in more than one origin AS. The ``route'' object is used to record routes which may appear in the global routing table. Explicit support for aggregation is provided. Route objects exist both for the configuration of routing information filters used to contain incidents of erroneous route announcements (Section 5) and to support network problem diagnosis. 8.5 reclaim and no-reclaim attributes A reclaim attribute is needed in as-block, inetnum and route objects. The reclaim attribute allows a control to be retained over more spe- cific AS, address or route space by allowing modify and delete privi- leges regardless of the mnt-by in the object itself. The reclaim and no-reclaim attributes contain contain lists of ob- jects subject to the reclaim and no-reclaim. See Section 9 for a full description of the reclaim and no-reclaim attributes. The reclaim attribute provides the means to enforce address lending. It allows cleanup in cases where entities cease to exist or as a last resort means to correct errors such as parties locking themselves out of access to their own objects. To allow finer control a set of prefixes can be specified. A no-reclaim attribute can be used to provide explicit exceptions. A reclaim attribute can only be added to an existing object if the ad- dition of the reclaim attribute does not remove autonomy of existing more specific objects that are covered by the new reclaim attribute. 1. A reclaim attribute can be added to an existing object if there are no existing exact matches or more specific objects overlapped by the new reclaim attribute, or 2. if the submitter is listed in the maintainer pointed to be the mnt- by of the objects which are overlapped, or 3. if any overlapped object is listed in a no-reclaim attribute in the object where the reclaim is being added. Villamizar, et. al. Expires August 23, 1999 [Page 13] INTERNET-DRAFT Routing Policy System Security February 23, 1999 Similarly a no-reclaim attribute cannot be deleted unless there are no overlapped objects for which the submitter is not listed in the maintainer pointed to be the mnt-by of the overlapped object. If neither a reclaim or no-reclaim attribute is present, then more specific objects of a given object cannot be modified by the main- tainer of the less specified object unless the maintainer is also listed as a maintainer in the more specific object. 8.6 Other Objects Many of the RPSL ancillary objects have no natural hierarchy the way AS numbers, Internet addresses and routes have a numeric hierarchy. Some examples are ``maintainers'', ``people'' and ``role'' objects. For these objects, lack of any hierarchy leads to two problems. 1. There is no hierarchy that can be exploited to direct queries to alternate registries. At some point the query strategy of search- ing all known registries becomes impractical. 2. There is no hierarchy on which authorizations of additions can be based. The first problem can be addressed by considering the name space for each of the ancillary objects to be unique only within the lo- cal database and to use explicit references to an external repository where needed. The object key is preceded by the name of the reposi- tory and the delimiter ``::''. For example a NIC handle may take the form ``RIPE::CO19''. Currently there is a desire to keep NIC handles unique so the naming convention of appending a dash and the reposi- tory name is used. Prepending the repository name provides the unique name space since an object in the RIPE database referencing ``CO19'' would be interpreted as ``RIPE::CO19'' by default, but it would still be possible to query or reference ``IANA::CO19''. There is no pos- sibility of accidentally forgetting to adhere to the conventions when making an addition and the existing objects are accommodated, includ- ing cases where name conflicts have already occurred. The second problem can be partially addressed by using a referral system for the addition of maintainers and requiring that any other object be submitted by a registered maintainer. The referral system would allow any existing maintainer to add another maintainer. This can be used in parallel with the addition of other object types to support the maintenance of those objects. For example, when adding a subdomain to the ``domain'' hierarchy (in the RIPE repository where domains are also handled), even when adding a new domain to a rela- tively flat domain such as ``com'', there is already a maintainer for Villamizar, et. al. Expires August 23, 1999 [Page 14] INTERNET-DRAFT Routing Policy System Security February 23, 1999 the existing domain. The existing maintainer can add the maintainer that will be needed for the new domain in addition to adding the new domain and giving the new maintainer the right to modify it. An organization gaining a presence on the Internet for the first time would be given a maintainer. This maintainer may list a small number of very trusted employees that are authorized to modify the maintainer itself. The organization itself can then add another maintainer list- ing a larger set of employees but listing the more restrictive main- tainer in the mnt-by attributes of the maintainers themselves. The organization can then add people and role objects as needed and any other objects as needed and as authorization permits. 8.7 Objects with AS Hierarchical Names Many RPSL objects do not have a natural hierarchy of their own but al- low hierarchical names. Some examples are the object types ``as-set'' and ``route-set''. An as-set may have a name corresponding to no nam- ing hierarchy such as ``AS-Foo'' or it may have a hierarchical name of the form ``AS1:AS-Bar''. When a hierarchical name is not used, authorization for objects such as ``as-set'' and ``route-set'' correspond to the rules for objects with no hierarchy described in Section 8.6. If hierarchical names are used, then the addition of an object must be authorized by the aut-num for the AS in the name of the object. The authentication must be listed in a maintainer referenced by the mnt-lower attribute of the aut-num if present, or if absent, in a maintainer referenced by the mnt-by attribute for the aut-num. 8.8 Query Processing A query may have to span multiple repositories. All queries should be directed toward a local repository which may mirror the root repos- itory and others. Currently each IRR repository mirrors all others repositories. In this way, the query may be answered by the local repository but draw data from others. For object types that have a natural hierarchy, such as aut-num, inet- num, and route, the search begins at the root database and follows the hierarchy. For objects types that have no natural hierarchy, such as maintainers, people, and roles, the search is confined to a default database unless a database is specified. The default database is the same database as an object from which a reference is made if the query is launched through the need to follow a reference. Otherwise the default is generally the local database or a default set by the Villamizar, et. al. Expires August 23, 1999 [Page 15] INTERNET-DRAFT Routing Policy System Security February 23, 1999 repository. The default can be specified in the query itself. In searching for an AS, the AS blocks can be consulted, moving the search to data from other repositories. Eventually the AS is either found or the search fails. The search for an inetnum is similar. Less specific inetnums may refer the search to other databases. Eventually the most specific inetnum is found and its status can be determined and assignment de- termined if it is assigned. The search for a route is similar except the search may branch to more than one repository. The most specific route in one repository may be more specific than the most specific in another. In looking for a route object it makes sense to return the most specific route that is not more specific than the query requests regardless of which repository that route is in rather than return one route from each repository that contains a less specific overlap. 8.9 Adding to the Database The root repository [14] must be initially populated at some epoch with a few entries. An initial maintainer is needed to add more main- tainers. The referral-by attribute can be set to refer to itself in this special case (Section 9 describes the referral-by). When adding an inetnum or a route object an existing exact match or a less spe- cific overlap must exist. A route object may be added based on an exact match or a less specific inetnum. The root repository must be initially populated with the allocation of an inetnum covering the prefix 0/0, indicating that some address allocation authority exists. Similarly an initial as-block is needed covering the full AS number range. When adding an object with no natural hierarchy, the search for an existing object follows the procedure outlined in Section 8.8. When adding an aut-num (an AS), the same procedure used in a query is used to determine the appropriate repository for the addition and to determine which maintainer applies. The sequence of AS-block objects and repository delegations is followed. If the aut-num does not ex- ist, then the submission must match the authentication specified in the maintainer for the most specific AS-block in order to be added. The procedure for adding an inetnum is similar. The sequence of inet- num blocks is followed until the most specific is found. The submis- sion must match the authentication specified in the maintainer for the most specific inetnum overlapping the addition. Adding a route object is somewhat more complicated. The route object Villamizar, et. al. Expires August 23, 1999 [Page 16] INTERNET-DRAFT Routing Policy System Security February 23, 1999 submission must satisfy two authentication criteria. It must match the authentication specified in the aut-num and the authentication specified in either a route object or if no applicable route object is found, then an inetnum. An addition is submitted with an AS number and prefix as its key. If the object already exists, then the submission is treated as a modify (see Section 8.10). If the aut-num does not exist on a route add, then the addition is rejected (see Section A for further discussion of tradeoffs). If the aut-num exists then the submission is checked against the applicable maintainer. A search is then done for the prefix first looking for an exact match. If the search for an exact match fails, a search is made for the longest prefix match that is less specific than the prefix specified. If this search succeeds it will return one or more route objects. The submission must match an applicable maintainer in at least one of these route objects for the addition to succeed. If the search for a route object fails, then a search is performed for an inetnum that exactly matches the prefix or for the most specific inetnum that is less specific than the route object submission. The search for an inetnum should never fail but it may return an unallocated or reserved range. The inetnum status must be ``allocated'' and the submission must match the maintainer. Having found the AS and either a route object or inetnum, the autho- rization is taken from these two objects. The applicable maintainer object is any referenced by the mnt-routes attributes. If one or more mnt-routes attributes are present in an object, the mnt-by attributes are not considered. In the absence of a mnt-routes attribute in a given object, the mnt-by attributes are used for that object. The authentication must match one of the authorizations in each of the two objects. If the addition of a route object or inetnum contains a reclaim at- tribute, then any more specific objects of the same type must be ex- amined. The reclaim attribute can only be added if there are no more specific overlaps or if the authentication on the addition is present in the authorization of a less specific object that already has a re- claim attribute covering the prefix range, or if the authentication on the addition is authorized for the modification of all existing more specific prefixes covered by the addition. 8.10 Modifying or Deleting Database Objects When modifying or deleting any existing object a search for the object is performed as described in Section 8.8. If the submission matches an applicable maintainer for the object, then the operation can pro- ceed. An applicable maintainer for a modification is any maintainer referenced by the mnt-by attribute in the object. For route and inet- num objects an applicable maintainer may be listed in a less specific Villamizar, et. al. Expires August 23, 1999 [Page 17] INTERNET-DRAFT Routing Policy System Security February 23, 1999 object with a reclaim attribute. If the submission is for a route object, a search is done for all less specific route objects and inetnums. If the submission is for an inetnum, a search is done for all less specific inetnums. If the submission fails the authorization in the object itself but matches the reclaim attribute in any of the less specific objects, then the operation can proceed. Section A contains discussion of the rationale behind the use of the reclaim attribute. A modification to an inetnum object that adds a reclaim attribute or removes a no-reclaim attribute must be checked against all exist- ing inetnums that are more specific. The same check of the reclaim attribute that is made during addition must be made when a reclaim attribute is added by a modification (see Section 8.9). A deletion is considered a special case of the modify operation. The deleted object may remain in the database with a ``deleted'' at- tribute in which case the mnt-by can still be consulted to remove the ``deleted'' attribute. 9 Data Format Summaries RIPE-181 [2] and RPSL [1] data is represented externally as ASCII text. Objects consist of a set of attributes. Attributes are name value pairs. A single attribute is represented as a single line with the name followed by a colon followed by whitespace characters (space, tab, or line continuation) and followed by the value. Within a value all whitespace is equivalent to a single space. Line continuation is supported by a backslash at the end of a line or the following line beginning with whitespace. When transferred, externally attributes are generally broken into shorter lines using line continuation though this is not a requirement. An object is externally represented as a series of attributes. Objects are separated by blank lines. There are about 80 attribute types in the current RIPE schema and about 15 object types. Some of the attributes are mandatory in cer- tain objects. Some attributes may appear multiple times. One or more attributes may form a key. Some attributes or sets of attributes may be required to be unique across all repositories. Some of the attributes may reference a key field in an object type and may be required to be a valid reference. Some attributes may be used in inverse lookups. A review of the entire RIPE or RPSL schema would be too lengthy to include here. Only the differences in the schema are described. Villamizar, et. al. Expires August 23, 1999 [Page 18] INTERNET-DRAFT Routing Policy System Security February 23, 1999 9.1 Changes to the RIPE/RPSL Schema One object type is added to the RIPE/RPSL schema. A few attributes are added to existing object types. There are significant changes to the rules which determine if the addition of an object is authorized. Additional keywords representing new authentication types are added to the semantics of the existing ``auth'' attribute. The new object type is listed below. The first attribute listed is the key attribute and also serves as the name of the object type. as-block key mandatory single unique descr optional multiple remarks optional multiple admin-c mandatory multiple tech-c mandatory multiple notify optional multiple mnt-by mandatory multiple changed mandatory multiple source mandatory single In the above object type only the key attribute ``as-block'' is new: as-block This attribute provides the AS number range for an ``as- block'' object. The format is two AS numbers including the sub- string ``AS'' separated by a ``-'' delimiter and optional whites- pace before and after the delimiter. In order to support stronger authentication, the following keywords are added to the ``auth'' attribute: pgp-from The remainder of the attribute gives the string identify- ing a PGP identity whose public key is held in an external keyring. The use of this method is deprecated in favor of the ``pgpkey'' method. pgpkey See [22]. In order to disable authentication and give permission to anyone, the authentication method ``none'' is added. It has no arguments. Villamizar, et. al. Expires August 23, 1999 [Page 19] INTERNET-DRAFT Routing Policy System Security February 23, 1999 An additional change is the ``auth'' attribute is allowed to exist in a ``person'' or ``role'' object. The ``auth'' method ``role'' or ``person'' can be used to refer to a role or person object and take the ``auth'' fields from those objects. Care must be taken in imple- mentations to detect circular references and terminate expansion or the references already visited. A few attributes are added to the schema. These are: mnt-routes The mnt-routes attribute may appear in an aut-num, inet- num, or route object. When used with a aut-num, inetnum, or route object, this attribute references a maintainer object which is used in determining authorization for the addition of route objects. When placed in a less specific inetnum it is used in determining authorization for the addition of more specific inetnum objects. After the reference to the maintainer, an optional list of prefix ranges (as defined in RPSL) inside of curly braces or the keyword ``ANY'' may follow. If a mnt-routes attribute is absent, the mnt- lower or mnt-by mnt-by attribute is used for this purpose. The mnt-routes attribute is optional and multiple. mnt-lower The mnt-lower attribute may appear in an inetnum, route, as-block or aut-num object. This attribute references a maintainer object. When used in an inetnum or route object the effect is the same as a ``mnt-routes'' but applies only to prefixes more specific than the prefix of the object in which it is contained. In an as- block object, mnt-lower allows addition of more specific as-block objects or aut-num objects. In an aut-num object the mnt-lower at- tribute specifies a maintainer that can be used to add objects with hierarchical names as described in Section 8.7. If a mnt-lower attribute is absent, the mnt-by attribute is used. reclaim The reclaim attribute may appear in as-block, aut-num, inet- num, or route objects. Any object of the same type below in the hierarchy may be modified or deleted by the maintainer of the ob- ject containing a reclaim attribute. The value of the attribute is a set or range of objects of the same type where the syntax of the set or range is as defined in RPSL. See Section 8.5 for restric- tions on adding reclaim attributes. no-reclaim The no-reclaim attribute is used with the reclaim at- tribute. The no-reclaim attribute negates any reclaim attribute it overlaps. See Section 8.5 for restrictions on deleting no-reclaim attributes. referral-by This attribute is required in the maintainer object. It may never be altered after the addition of the maintainer. This attribute refers to the maintainer that created this maintainer. It may be multiple if more than one signature appeared on the transaction creating the object. Villamizar, et. al. Expires August 23, 1999 [Page 20] INTERNET-DRAFT Routing Policy System Security February 23, 1999 auth-override An auth-override attribute can be added, deleted, or changed by a transaction submitted by maintainer listed in the referral-by. An auth-override can only be added to a maintainer if that maintainer has been inactive for the prior 60 days. The auth-override attribute itself contains only the date when the at- tribute will go into effect which must be at least 60 days from the current date unless there is already authorization to modify the maintainer. After the date in the auth-override is reached, those identified by the maintainer in the referral-by have authoriza- tion to modify the maintainer. This attribute exists as a means to clean up should the holder of a maintainer become unresponsive and can only take effect if that maintainer does not remove the auth- override in response to the automatic notification that occurs on changes. Each repository must identify itself with a ``repository'' object. The repository must also contain a special ``repository'' whose key is ``ROOT''. The root repository is where all non-local queries are directed, including where hierarchical object queries start. The query methods listed for the root repository may actually be a subset of those offered by that repository if efficiency considerations and topologic distance make some methods less useful. The root repository must contain a copy of the repository objects in any repository considered valid. The repository objects will be essential when the routing registry becomes more widely distributed. The existing ``mnt-by'' attribute references the ``maintainer'' ob- ject type. The ``mnt-by'' attribute is now mandatory in all object types. A new maintainer may be added by any existing maintainer. The ``referral-by'' attribute is now mandatory in the ``maintainer'' ob- ject to keep a record of which maintainer made the addition and can never be changed. Maintainers cannot be deleted as long as they are referenced by a ``referral-by'' attribute elsewhere. A Technical Discussion A few design tradeoffs exist. Some of these tradeoffs, the selected solution, and the alternatives are discussed here. Some of the issues are listed below. 1. Whether to error on the side of permissiveness and weaken autho- rization controls or risk the possibility of erecting barriers to registering information. 2. Whether to support enforcible address lending or provide the Villamizar, et. al. Expires August 23, 1999 [Page 21] INTERNET-DRAFT Routing Policy System Security February 23, 1999 smaller or end user with ultimate control over the registration of the prefixes they are using. 3. What to do with older objects that either don't conform to newer requirements regarding minimum authorization, authentication, and accountability, or are of questionable validity. A.1 Relaxing requirements for ease of registry If the requirement that an aut-num exists is relaxed, then it is pos- sible for anyone to make use of an unassigned AS number or make use of an assigned AS number for which the aut-num has not been entered. Placing requirements on the entry of aut-num presumes cooperation of the Internet address allocation authority (if separate from the rout- ing registry). The address allocation authority must be willing to field requests to populate skeleton aut-nums from the party for which the allocation has been made. These aut-num must include a reference to a maintainer. A request to the address allocation authority must therefore include a reference to an existing maintainer. The ability to add route objects is also tied to the existence of less specific route objects or inetnums. The Internet address allocation authority (if separate from the routing registry) must also be will- ing to field requests to add inetnum records for the party already allocated the address space. The Internet address allocation authority should also add inetnums and aut-nums for new allocations. In order to do so, a maintainer must exist. If a party is going to connect to the Internet, they can get a maintainer by making a request to the Internet service provider they will be connecting to. Once they have a maintainer they can make a request for address space or an AS number. The maintainer can con- tain a public key for a cryptographicly strong authorization method or could contain a ``crypt-key'' or ``mail-to'' authorization check if that is considered adequate by the registering party. Furthermore an address allocation authority should verify that the request for an AS number or for address space matches the authorization criteria in the maintainer. Currently only the registry themselves may add maintainers. This be- comes a problem for the registry particularly verifying public keys. This requirement is relaxed by allowing existing maintainers to add maintainers. Unfortunately the accountability trail does not exist for existing maintainers. The requirement then should be relaxed such that existing maintainers may remain but only existing maintain- ers that have a ``referral-by'' attribute can add maintainers. The ``referral-by'' cannot be modified. This requirement can be relax slightly so that a ``referral-by'' can be added to a maintainer by an existing maintainer with a ``referral-by''. This will allow the Villamizar, et. al. Expires August 23, 1999 [Page 22] INTERNET-DRAFT Routing Policy System Security February 23, 1999 accountability trail to be added to existing maintainers and these maintainers can then add new maintainers. Verifying that a party is who they claim to be on initial addition, is one of the problems that currently falls upon the AS number and address registry. This problem is reduced by allowing existing main- tainers to add maintainers. This may actually make it easier to get maintainers and therefore easier to register. The number authority still must verify that the AS or address space is actually needed by the party making a request. Authorization checks made during the addition of route objects that refer to AS objects and inetnum strongly rely on the cooperation of the Internet address allocation authorities. The number authorities must register as-blocks, aut-nums, or inetnums as AS numbers or ad- dress space is allocated. If only a subset of the number authorities cooperate, then either an inetnum or as-block can be created cover- ing the space that registry allocates and essentially requiring null allocation (for example a ``crypt-pw'' authentication where the pass- word is given in the remarks in the object or its maintainer) or those obtaining addresses from that number authority will have trouble reg- istering in the routing registry. The authorization model supports either option, though it would be preferable if the number authorities cooperated and the issue never surfaced in practice. The maintainer requirements can be relaxed slightly for existing main- tainers making it easier to register. Relaxing requirements on other objects may defeat the authorization model, hence is not an option. A.2 The address lending issue The issue of whether lending contracts should be enforcible is an issue of who should ultimately be able to exercise control over al- locations of address space. The routing registry would be wise to stay as neutral as possible with regard to disputes between third par- ties. The ``reclaim'' and ``no-reclaim'' are designed to allow either outcome to the decision as to whether the holder of a less specific inetnum or route object can exercise control over suballocations in the registry. The routing registry itself must decide whether to re- tain control themselves and if so, should very clearly state under what conditions the registry would intervene. A registry could even go to the extreme of stating that they will intervene in such a dis- pute only after the dispute has been resolved in court and a court order has been issued. When an allocation is made by a registry, the registry should keep a ``reclaim'' attribute in the less specific object and make a strong policy statement that the reclaim privilege will not be used except under very clearly defined special circumstances (which at the very Villamizar, et. al. Expires August 23, 1999 [Page 23] INTERNET-DRAFT Routing Policy System Security February 23, 1999 minimum would include a court order). If the allocation is further subdivided the party subdividing the allocation and the party accept- ing the suballocation must decide whether a ``reclaim'' can be kept by the holder of the less specific allocation or whether a ``no-reclaim'' must be added transferring control to the holder of the more specific. The registry is not involved in that decision. Different pairs of third parties may do different decisions regarding the ``reclaim'' and any contractual restrictions on its use that may be expressed out- side of the registry in the form of a legal contract and ultimately resolved by the courts in the event of a bitter dispute. By retaining ``reclaim'' rights the registry retains the ability to abide by a court order. This may only truly become an issue in a dis- tributed registry environment where registries will be rechecking the authorization of transactions made elsewhere and may fail to process the attempt of another registry to abide by a court order by overrid- ing normal authorization to change the registry contents if a reclaim is not present. A.3 Dealing with non-conformant or questionable older data Some of the newer requirements include requiring that all objects reference a maintainer object responsible for the integrity of the object and requiring accountability for the creation of maintainers to be recorded in the maintainer objects so that accountability can be traced back from an unresponsive maintainer. In the event that contact information is absent or incorrect from objects and there is any question regarding the validity of the objects, the maintainer can be contacted. If the maintainer is unresponsive, the maintainer that authorized the addition of that maintainer can be contacted to either update the contact information on the maintainer or confirm that the entity no longer exists or is no longer actively using the Internet or the registry. Many route objects exist for which there are no maintainers and for which inetnum and AS objects do not exist. Some contain the now obso- leted guardian attribute rather than a mnt-by. It is not practical to unconditionally purge old data that does not have maintainers or does not conform to the authorization hierarchy. New additions must be required to conform to the new requirements (otherwise the requirements are meaningless). New requirements can be phased in by requiring modifications to conform to the new require- ments. A great deal of questionable data exists in the current registry. The requirement that all objects have maintainers and the requirements for improved accountability in the maintainers themselves may make it easier to determine contact information even where the objects are not Villamizar, et. al. Expires August 23, 1999 [Page 24] INTERNET-DRAFT Routing Policy System Security February 23, 1999 updated to reflect contact information changes. It is not unreasonable to require valid contact information on exist- ing data. A great deal of data appears to be unused, such as route objects for which no announcement has been seen in many months or years. An attempt should be made to contact the listed contacts in the object, in the maintainer if there is one, then up the maintainer referral-by chain if there is one, and using the number registry or origin AS contact information if there is no maintainer accountability trail to follow. Experience so far indicates that the vast majority of deletions identified by comparing registered prefixes against route dumps will be positively confirmed (allowing the deletion) or there will be no response due to invalid contact information (in many cases the IRR contact information points to nsfnet-admin@merit.edu). By allowing the registry to modify (or delete) any objects which are disconnected from the maintainer accountability trail, cleanup can be made possible (though mail header forging could in many cases have the same effect it is preferable to record the fact that the registry itself made the cleanup). Similarly, a mechanism may be needed in the future to allow the maintainer in the referral-by to override maintainer privileges in a referred maintainer if all contacts have become unresponsive for a maintainer. The referral-by maintainer is allowed to add an ``auth-override'' attribute which becomes usable as an ``auth'' within 60 days from the time of addition. The main- tainer themselves would be notified of the change and could remove the ``auth-override'' attribute before it becomes effective and inquire as to why it was added and correct whatever problem existed. This can be supported immediately or added later if needed. B Common Operational Cases In principle address allocation and route allocation should be hierar- chical with the hierarchy corresponding to the physical topology. In practice this is often not the case for numerous reasons. The pri- mary reasons are the topology is not strictly tree structured and the topology can change. More specificly: 1. The Internet topology is not strictly tree structured. o At the top level the network more closely resembles a moderately dense mesh. o Near the bottom level many attachments to the Internet are multi- homed to more than one Internet provider. 2. The Internet topology can and does change. Villamizar, et. al. Expires August 23, 1999 [Page 25] INTERNET-DRAFT Routing Policy System Security February 23, 1999 o Many attachments switch providers to obtain better service or terms. o Service providers may modify adjacencies to obtain better transit service or terms. o Service providers may disappear completely scattering attachments or merge. Renumbering is viewed as a practical means to maintain a strict nu- meric hierarchy [19]. It is also acknowledged that renumbering IPv4 networks can be difficult [19, 3, 20]. We examine first the simple case where hierarchy still exists. We then examine the operational cases where either initial topology is not tree structured or cases where topology changes. B.1 simple hierarchical address allocation and route allocation This is the simplest case. Large ranges of inetnums are assigned to address registries. These registries in turn assign smaller ranges for direct use or to topologically large entities where allocations according to topology can reduce the amount of routing information needed (promote better route aggregation). AS objects are allocated as topology dictates the need for additional AS [10]. Route objects can be registered by those with authoriza- tion given by the AS and by the address owner. This is never an issue where the maintainer of the AS and the inetnum are the same. Where they differ, either the provider can give permission to add route ob- jects for their AS, or the party allocated the address space can give the provider permission to add router objects for their address space, or both parties can sign the transaction. Permission is provided by adding to maintainer attributes. B.2 aggregation and multihomed more specific routes Aggregation is normally not a problem if a provider is aggregating ad- dress space allocated to the provider and then suballocated internally and/or to customers. In fact, the provider would be expected to do so. This is not a problem even if the route object for the aggrega- tion is added after the more specific route objects since only less specific objects are considered. Aggregation is potentially a problem if a provider or a set of providers plan to aggregate address space that was never explicitly allocated as a block to those providers but rather remains the alloca- tion of a address registry. These large aggregations can be expected Villamizar, et. al. Expires August 23, 1999 [Page 26] INTERNET-DRAFT Routing Policy System Security February 23, 1999 to be uncommon, but relatively easily dealt with. Superaggregates of this type will generally be formed by topologically close entities who have also managed to draw adjacent address allocations. In effect, the registry must give permission to form such as superaggregate by either giving permission to do so in the mnt-routes of an inetnum or by signing the submission along with the other parties. B.3 provider independent addresses and multiple origin AS Provider independent addresses and multihoming arrangement using mul- tiple origin AS present a similar problem to multihoming. The main- tainer of the address space and the maintainer of the AS is not the same. Permission can be granted using mnt-routes or multiple signa- tures can appear on the submission. B.4 change in Internet service provider A change in Internet service providers is similar to multihoming. A minor difference is that the AS for the more specific route will be the AS of the new provider rather than the AS of the multihomed customer. Permission can be granted using mnt-routes or multiple signatures can appear on the submission. B.5 renumbering grace periods Renumbering grace periods allow a provider who wants to keep an ad- dress allocation intact to allow a customer who has chosen to go to another provider to renumber their network gradually and then re- turn the address space after renumbering is completed. The issue of whether to require immediate renumbering or offer renumbering grace periods and how long they should be or whether they should be in- definite has been topic of bitter disputes. The authorization model can support no renumbering grace period, a finite renumbering grace period, or an indefinite renumbering grace period. The ``reclaim'' attribute described in Section 8.1 provides a means to end the grace period. C Deployment Considerations This section describes deployment considerations. The intention is to raise issues and discuss approaches rather than to provide a deploy- ment plan. Villamizar, et. al. Expires August 23, 1999 [Page 27] INTERNET-DRAFT Routing Policy System Security February 23, 1999 The use of routing registries is not yet universally accepted. There still remain Internet providers who see no reason to provide the added assurance of accurate routing information described in Section 5. More accurately, these benefits are viewed as being insufficient to justify the cost. This has been largely caused an inability of a very major router vendor up until recently to handle prefix lists of the size needed to specify routing policy on a per prefix basis. Another reason cited is that filtering on a prefix basis in an environment where routing registry is incomplete or inaccurate can interfere with connectivity. There clearly is a critical mass issue with regard to the use of rout- ing registries. A minority of providers use the existing IRR to filter on a per prefix basis. Another minority of providers do not support the IRR and generally fail to register prefixes until con- nectivity problems are reported. The majority of providers register prefixes but do not implement strict prefix filtering. Deploying new authentication mechanisms has no adverse consequences. This has been proven with Merit's deployment of PGP. In deploying new authorization mechanisms, a major issue is dealing with existing data of very questionable origin. A very large num- ber of route objects refer to prefixes that have not been announced for many years. Other route objects refer to prefixes that are no longer announced with the origin AS that they are register with (some were incorrectly registered to start with). There are many causes for this. 1. During the transition from the NSFNET PRDB to the RADB a large number of prefixes were registered with an origin AS correspond- ing to the border AS at which the NSFNET had once heard the route announcements. The PRDB did not support origin AS, so border AS was used. Many of these routes were no longer in use at the time and are now routed with a submitter listed as ``nsfnet- admin@merit.edu''. 2. As CIDR was deployed, aggregates replaced previously separately announced more specific prefixes. The route objects for the more specific prefixes were never withdrawn from the routing registries. 3. Some prefixes are simply no longer in use. Some networks have been renumbered. Some network no longer exist. Often the routing reg- istry information is not withdrawn. 4. As provider AS adjacencies changed and as end customers switched providers often the actual origin AS changed. This was often not reflected by a change in the routing registry. Inaccuracies will continue to occur due to the reasons above, except Villamizar, et. al. Expires August 23, 1999 [Page 28] INTERNET-DRAFT Routing Policy System Security February 23, 1999 the first. The hierarchical authorization provides greater account- ability. In the event that the contacts for specific objects become unresponsive traversal up the authorization hierarchy should help identify the parties having previous provided authorization. These contacts may still have sufficient authorization to perform the neces- sary cleanup. This issue is discussed in Section A. A great deal of information is currently missing in the IRR. Quite a few AS have no aut-num. Quite a lot of data has no maintainer and the vast majority of maintainers use only the weakest of authentication methods. Very little can be done by the registries to correct this. The defaults in the cases of missing objects needed for authorization has to be to make no authentication checks at all. The transition can be staged as follows: 1. Add and make use of stronger authorization models. 2. Make schema modifications necessary to support delegations. 3. Add delegation objects needed for query traversal. 4. Base query traversal on delegations rather than search of all known registries. 5. Obtain the cooperation of the address registries for the purpose of populating the ``inetnum'' entries on an ongoing basis. 6. Add hierarchical authorization support for critical object types, ``aut-num'', ``inetnum'' and ``route''. 7. Add the requirement that database object either be in use or have valid contact information and if queries are made by the registry a response from a contact indicating that the object serves a purpose if it is not clear what its use is. 8. Begin to purge data which is clearly not in use and for which there is no valid contact information or no response from the contacts. Deployment of hierarchical authorization requires cooperation among the existing routing registries. New code will have to be deployed. In some cases very little development resources are available and sub- stantial inertia exists due to the reliance on the current repository and the need to avoid disruption. If hierarchical authorization of route objects depends on the exis- tence of address registration information, minimal cooperation of the currently separate address registries is required. The extent of the cooperation amounts to sending cryptographically signed trans- actions from the address registry to the number registry as address Villamizar, et. al. Expires August 23, 1999 [Page 29] INTERNET-DRAFT Routing Policy System Security February 23, 1999 allocations are made or providing equivalent access to new address allocations. Currently most registries returns query results from all of the known repositories using their mirrored copies. Cross registry authoriza- tions are not yet implemented. Minimal schema changes have to be made to support the ability to delegate objects for which there is an au- thorization hierarchy and the support queries and references to other repositories. In the case of AS delegations, ``as-block'' need to be created solely for the purpose of traversal. Acknowledgments This document draws ideas from numerous discussions and contributions of the IETF Routing Policy System Work Group and RIPE Routing Work Group. Earlier drafts of this document listed Carol Orange as a co- author. Carol Orange made contributions to this document while at RIPE. References [1] C. Alaettinoglu, T. Bates, E. Gerich, D. Karrenberg, D. Meyer, M. Terpstra, and C. Villamizar. Routing policy specification language (rpsl). Technical Report RFC 2280, Internet Engineering Task Force, 1998. ftp://ftp.isi.edu/in-notes/rfc2280.txt. [2] T. Bates, E. Gerich, L. Joncheray, J-M. Jouanigot, D. Kar- renberg, M. Terpstra, and J. Yu. Representation of ip rout- ing policies in a routing registry (ripe-81++). Techni- cal Report RFC 1786, Internet Engineering Task Force, 1995. ftp://ftp.isi.edu/in-notes/rfc1786.txt. [3] H. Berkowitz. Router renumbering guide. Technical Re- port RFC 2072, Internet Engineering Task Force, 1997. ftp://ftp.isi.edu/in-notes/rfc2072.txt. [4] H.W. Braun. Models of policy based routing. Technical Report RFC 1104, Internet Engineering Task Force, 1989. ftp://ftp.isi.edu/in-notes/rfc1104.txt. [5] H.W. Braun and Y. Rekhter. Advancing the nsfnet routing archi- tecture. Technical Report RFC 1222, Internet Engineering Task Force, 1991. ftp://ftp.isi.edu/in-notes/rfc1222.txt. [6] D.D. Clark. Policy routing in internet protocols. Techni- cal Report RFC 1102, Internet Engineering Task Force, 1989. ftp://ftp.isi.edu/in-notes/rfc1102.txt. Villamizar, et. al. Expires August 23, 1999 [Page 30] INTERNET-DRAFT Routing Policy System Security February 23, 1999 [7] D. Crocker. Standard for the format of arpa internet text mes- sages. Technical Report RFC 822, Internet Engineering Task Force, 1982. ftp://ftp.isi.edu/in-notes/rfc822.txt. [8] V. Fuller, T. Li, J. Yu, and K. Varadhan. Classless inter-domain routing (cidr): an address assignment and aggregation strat- egy. Technical Report RFC 1519, Internet Engineering Task Force, 1993. ftp://ftp.isi.edu/in-notes/rfc1519.txt. [9] Internet Engineering Steering Group and R. Hinden. Applicability statement for the implementation of classless inter-domain rout- ing (cidr). Technical Report RFC 1517, Internet Engineering Task Force, 1993. ftp://ftp.isi.edu/in-notes/rfc1517.txt. [10] J. Hawkinson and T. Bates. Guidelines for creation, selec- tion, and registration of an autonomous system (as). Techni- cal Report RFC 1930, Internet Engineering Task Force, 1996. ftp://ftp.isi.edu/in-notes/rfc1930.txt. [11] K. Hubbard, M. Kosters, D. Conrad, D. Karrenberg, and J. Pos- tel. Internet registry ip allocation guidelines. Techni- cal Report RFC 2050, Internet Engineering Task Force, 1996. ftp://ftp.isi.edu/in-notes/rfc2050.txt. [12] Mark Knopper and Steven J. Richardson. Aggregation sup- port in the nsfnet policy-based routing database. Techni- cal Report RFC 1482, Internet Engineering Task Force, 1993. ftp://ftp.isi.edu/in-notes/rfc1482.txt. [13] David Meyer, Cengiz Alaettinoglu, J. Schmitz, and Carol Orange. Using rpsl in practice. Internet Draft (Work in Progress) draft- ietf-rps-appl-rpsl-03, Internet Engineering Task Force, 11 1998. ftp://ftp.isi.edu/internet-drafts/draft-ietf-rps-appl-rpsl- 03.txt. [14] David Meyer, Curtis Villamizar, Cengiz Alaettinoglu, and R. Govindan. Distributed routing policy system. Internet Draft (Work in Progress) draft-ietf-rps-dist-00, Internet Engineering Task Force, 9 1998. ftp://ftp.isi.edu/internet-drafts/draft- ietf-rps-dist-00.txt. [15] National Institute of Standards and Technology. The digital sig- nature standard, proposal and discussion. Communications of the ACM, 35(7):36--54, July 1992. [16] National Institute of Standards and Technology (NIST). Fips pub- lication 186: Digital signature standard (dss). Technical re- port, Gaithersburg, MD, May 1994. [17] Y. Rekhter. Routing in a multi-provider internet. Techni- cal Report RFC 1787, Internet Engineering Task Force, 1995. ftp://ftp.isi.edu/in-notes/rfc1787.txt. [18] Y. Rekhter and T. Li. An architecture for ip address allocation with cidr. Technical Report RFC 1518, Internet Engineering Task Force, 1993. ftp://ftp.isi.edu/in-notes/rfc1518.txt. Villamizar, et. al. Expires August 23, 1999 [Page 31] INTERNET-DRAFT Routing Policy System Security February 23, 1999 [19] Y. Rekhter and T. Li. Implications of various address alloca- tion policies for internet routing. Technical Report RFC 2008, Internet Engineering Task Force, 1996. ftp://ftp.isi.edu/in- notes/rfc2008.txt. [20] Y. Rekhter, P. Lothberg, R. Hinden, S. Deering, and J. Pos- tel. An ipv6 provider-based unicast address format. Techni- cal Report RFC 2073, Internet Engineering Task Force, 1997. ftp://ftp.isi.edu/in-notes/rfc2073.txt. [21] Bruce Schneier. Applied Cryptography. Wiley, New York, 1996. [22] Janos Zsako. Pgp authentication for ripe database updates. Internet Draft (Work in Progress) draft-ietf-rps-dbsec- pgp-authent-00, Internet Engineering Task Force, 11 1998. ftp://ftp.isi.edu/internet-drafts/draft-ietf-rps-dbsec-pgp- authent-00.txt. Security Considerations @@ This entire document is about security but at some point we need to figure out what belongs here in this case. Maybe the Area Directors can provide some advice. Author's Addresses Curtis Villamizar UUNET Network Architecture Group Cengiz Alaettinoglu ISI David M. Meyer Cisco Sandy Murphy Trusted Information Systems Villamizar, et. al. Expires August 23, 1999 [Page 32] INTERNET-DRAFT Routing Policy System Security February 23, 1999 Full Copyright Statement Copyright (C) The Internet Society (February 23, 1999). All Rights Reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implmentation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are in- cluded on all such copies and derivative works. However, this doc- ument itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other In- ternet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an ``AS IS'' basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEER- ING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUD- ING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MER- CHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Villamizar, et. al. Expires August 23, 1999 [Page 33]