Network Working Group|
Request for Comments: 2904
Interlink Networks, Inc.
Sun Microsystems, Inc.
Enterasys Networks EMEA
B. de Bruijn
Interpay Nederland B.V.
C. de Laat
Interlink Networks, Inc.
This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.
Copyright © The Internet Society (2000). All Rights Reserved.
This memo serves as the base requirements for Authorization of Internet Resources and Services (AIRS). It presents an architectural framework for understanding the authorization of Internet resources and services and derives requirements for authorization protocols.
2. Authorization Entities and Trust Relationships
3. Message Sequences
3.1. Single Domain Case
3.1.1. The Agent Sequence
3.1.2. The Pull Sequence
3.1.3. The Push Sequence
3.3. Distributed Services
3.4. Combining Roaming and Distributed Services
4. Relationship of Authorization and Policy
4.1. Policy Retrieval
4.2. Policy Evaluation
4.3. Policy Enforcement
4.4. Distributed Policy
5. Use of Attribute Certificates
6. Resource Management
6.1. Session Management
6.2. The Resource Manager
7. AAA Message Forwarding and Delivery
8. End-to-End Security
9. Streamlined Authorization Process
10. Summary of the Authorization Framework
11. Security Considerations
Full Copyright Statement
This document is one of a series of three documents under consideration by the AAAarch RG dealing with the authorization requirements for AAA protocols. The three documents are:
AAA Authorization Framework (this document)
AAA Authorization Requirements 
AAA Authorization Application Examples 
There is a demonstrated need for a common scheme which covers all Internet services which offer Authorization. This common scheme will address various functional architectures which meet the requirements of basic services. We attempt to describe these architectures and functions as a basis for deriving requirements for an authorization protocol .
These architectures include Policy structures, Certificate Authorities, Resource Managers, Inter-Domain and Multi-Domain schemes, and Distributed Services. The requirements are for the expected use of Authorization services across these architectures.
A representative set of applications that may use this architecture to support their authorization needs is presented in . The examples in  show how this framework may be used to meet a wide variety of different authorization needs.
We expect that this work may be extended in the future to a more
comprehensive model and that the scheme described here will be
incorporated into a framework that includes authentication,
accounting and auditing. We have referenced a number of
authorization sources, but also recognize that there may be some that we have missed and that should be included. Please notify one of the authors of any such oversight so it can be corrected in a future revision.
In general, it is assumed that the parties who are participating in the authorization process have already gone through an authentication phase. The authentication method used by those parties is outside the scope of this document except to the extent that it influences the requirements found in a subsequent authorization process. Likewise, accounting requirements are outside the scope of this document other than recording accounting data or establishing trust relationships during an authorization that will facilitate a subsequent accounting phase.
The work for this memo was done by a group that originally was the Authorization subgroup of the AAA Working Group of the IETF. When the charter of the AAA working group was changed to focus on MobileIP and NAS requirements, the AAAarch Research Group was chartered within the IRTF to continue and expand the architectural work started by the Authorization subgroup. This memo is one of four which were created by the subgroup. This memo is a starting point for further work within the AAAarch Research Group. It is still a work in progress and is published so that the work will be available for the AAAarch subgroup and others working in this area, not as a definitive description of architecture or requirements.
This document uses the terms 'MUST', 'SHOULD' and 'MAY', and their negatives, in the way described in RFC 2119 .
The following framework is being presented in order to provide a framework for discussing authorization requirements for a large number of applications. The intent is to provide some common vocabulary for the discussion. Terminology is introduced for basic elements in the authorization transaction and for concepts that appear to be common to all (or at least many) authorization proposals.
Figure 1, below, identifies the basic conceptual entities that may be participants in an authorization:
+------+ +-------------------------+ | | | User Home Organization | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | +-------------------------+ | | | | | | | User | +-------------------------+ | | | Service Provider | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | | +-------------------+ | | | | | Service | | | | | | Equipment | | | | | +-------------------+ | | | | | +------+ +-------------------------+
Fig. 1 -- The Basic Authorization Entities
These entities will be referenced in the authorization requirements.
There may be bilateral agreements between pairs of organizations involved in an authorization transaction. Agreements between organizations may take the form of formal contracts or Service Level Agreements. Figure 2 uses double lines to show relationships that may exist between the User and the User Home Organization and between the User Home Organization and the Service Provider.
+------+ +-------------------------+ | | | User Home Organization | | |======| +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | +-------------------------+ | | || | | || | | || | User | +-------------------------+ | | | Service Provider | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | | +-------------------+ | | | | | Service | | | | | | Equipment | | | | | +-------------------+ | | | | | +------+ +-------------------------+
Fig. 2 -- Service Agreements
Authorization is based on these bilateral agreements between entities. Agreements may be chained, as shown in figure 2. The User has an agreement with the User Home Organization (e.g., the User may have access to the service between 9:00 a.m. and 11:00 a.m. daily). The User Home Organization has an agreement with the Service Provider (e.g., that all requests for access will be granted, except between 5:00 a.m. and 10:00 a.m. on Sunday). The fulfillment of the User's request depends on both agreements being honored.
Note that these agreements may be implemented by hand configuration or by evaluation of Policy data stored in a Policy database. The point is that there must be a set of known rules in place between entities in order for authorization transactions to be executed.
Trust is necessary to allow each entity to "know" that the policy it is authorizing is correct. This is a business issue as well as a protocol issue. Trust is often established through third party authentication servers (such as Kerberos), via a certificate authority, by configuring shared secrets or passwords, or by sharing a common facility (such as a connecting wire between processors). These "static" trust relationships are necessary for authorization
transactions to take place. Static trust relationships are used in an authorization sequence to establish a "dynamic" relationship between the User and the Service Equipment. Several possible authorization sequences are possible, each of which use the static trust "chain" to have the user first be approved by the User Home Organization, and then have the Service Provider accept the request based on its trust of the User Home Organization.
In general, the User Home Organization and the Service Provider are different entities or different "administrative domains". In the simplest case, however, the User Home Organization and the Service Provider may be combined as a single entity. This case will be used to describe three authorization sequences possible with the simple case.
In following sections these concepts will be applied to more complicated cases involving separate User Home Organization and Service Provider entities (as in roaming) and multiple Service Providers each with their own AAA Servers and Service Equipment (as in distributed services).
This case includes the User, the Service Provider's AAA Server, and the Service Provider's Service Equipment. Examples of this case include a NAS supported by a standalone RADIUS server, or a QoS Router supported by a local bandwidth broker.
The sequences considered in the following figures are the "agent", "pull", and "push" sequences for the single domain case.
In the agent sequence (see figure 3), the Service Provider AAA Server
functions as an agent between the User and the service itself. The
AAA Server receives a request from the User and forwards
authorization and possibly configuration information to the Service Equipment. In this model, the User sends a request to the Service Provider's AAA Server (1), which will apply a policy associated with the User and the particular service being requested. The AAA Server sends a request to the Service Equipment, and the Service Equipment sets up whatever is requested (2). The Service Equipment then responds to the AAA Server acknowledging that it has set up the Service for the user (3). The AAA Server replies to the User telling it that the Service is set up (4).
Depending on the nature of the service, further communication may take place between the User and the Service Equipment. For this to occur, there needs to be a binding between the User and the authorized service. This requires further study.
+-------------------------+ +------+ | Service Provider | | | 1 | +-------------------+ | | |------+->| AAA Server | | | |<-----+--| | | | | 4 | +-------------------+ | | User | | | /|\ | | | | |2 |3 | | | | \|/ | | | | | +-------------------+ | | | | | Service | | | | | | Equipment | | | | | +-------------------+ | +------+ | | +-------------------------+
Fig. 3 -- Agent Sequence
Example: A regular user may ask for 1 Mb/s bandwidth (1). The bandwidth broker (AAA Server) tells the router (Service Equipment) to set this user into the 1Mb/s "queue" (2). The router responds that it has done so (3), and the bandwidth broker tells the User the bandwidth is set up (4).
The pull sequence (figure 4) is what is typically used in the Dialin application, in the Mobile-IP proposal, and in some QoS proposals. The User sends a request to the Service Equipment (1), which forwards it to the Service Provider's AAA Server (2), which evaluates the request and returns an appropriate response to the Service Equipment (3), which sets up the service and tells the User it is ready (4).
+-------------------------+ +------+ | Service Provider | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | User | | /|\ | | | | | |2 |3 | | | | | \|/ | | | 1 | +-------------------+ | | |------+->| Service | | | |<-----+--| Equipment | | | | 4 | +-------------------+ | +------+ | | +-------------------------+
Fig. 4 -- Pull Sequence
The push sequence (figure 5) requires that the User get from the Service Provider's AAA Server a ticket or certificate verifying that it is o.k. for the User to have access to the service (1,2). The User includes the ticket in the request (3) to the Service Equipment. The Service Equipment uses the ticket to verify that the request is approved by the Service Provider's AAA Server. The Service Equipment then sends an o.k. to the User (4).
The ticket the user gets from the Service Provider's AAA Server will typically have some time limit on it. It may contain an indication of service location, and in some applications, it might be used for more than one request.
In the push sequence the communication between the AAA Server and the Service Equipment is relayed through the User rather than directly between themselves.
+-------------------------+ +------+ | Service Provider | | | 1 | +-------------------+ | | |------+->| AAA Server | | | |<-----+--| | | | | 2 | +-------------------+ | | User | | | | | | | | | | | | | 3 | +-------------------+ | | |------+->| Service | | | |<-----+--| Equipment | | | | 4 | +-------------------+ | +------+ | | +-------------------------+
Fig. 5 -- Push Sequence
In many interesting situations, the organization that authorizes and authenticates the User is different from the organization providing the service. This situation has been explored in the Roaming Operations (roamops) Working Group. For purposes of this discussion, any situation in which the User Home Organization is different from the Service Provider is considered to be roaming.
Examples of roaming include an ISP selling dialin ports to other organizations or a Mobile-IP provider allowing access to a user from another domain.
The same agent, pull and push sequences are possible with roaming. If the Service Provider's AAA Server and the Service Equipment are grouped as a logical entity for purposes of description, then the following figures illustrate these cases.
+------+ +-------------------------+ | | 1 | User Home Organization | | |----->| +-------------------+ | | | | | AAA Server | | | |<-----| | | | | | 4 | +-------------------+ | | | | | | | +-------------------------+ | | | /|\ | | |2 |3 | | \|/ | | User | +-------------------------+ | | | Service Provider | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | | +-------------------+ | | | | | Service | | | | | | Equipment | | | | | +-------------------+ | | | | | +------+ +-------------------------+
Fig. 6 -- Roaming Agent Sequence
+------+ +-------------------------+ | | | User Home Organization | | | | +-------------------+ | | | | | AAA Server | | | | | | | | | | | +-------------------+ | | | | | | | +-------------------------+ | | /|\ | | | |2 |3 | | | \|/ | | +-------------------------+ | | | Service Provider | | User | | +-------------------+ | | | | | AAA Server | | | | 1 | | | | | |----->| +-------------------+ | | | | | | |<-----| +-------------------+ | | | 4 | | Service | | | | | | Equipment | | | | | +-------------------+ | | | | | +------+ +-------------------------+
Fig. 7 -- Roaming Pull Sequence
+------+ +-------------------------+ | | 1 | User Home Organization | | |----->| +-------------------+ | | | | | AAA Server | | | |<-----| | | | | | 2 | +-------------------+ | | | | | | | +-------------------------+ | | | | | | | User | +-------------------------+ | | | Service Provider | | | | +-------------------+ | | | | | AAA Server | | | | 3 | | | | | |----->| +-------------------+ | | | | | | |<-----| +-------------------+ | | | 4 | | Service | | | | | | Equipment | | | | | +-------------------+ | | | | | +------+ +-------------------------+
Fig. 8 -- Roaming Push Sequence
To provide a complete service to a user, offerings from several service providers may need to be combined. An example would be a user who requires a QoS service for a session that crosses multiple ISPs. Any service that is provided by more than one Service Provider acting in concert is a distributed service. Figure 9 illustrates distributed services.
+-------------------+ +-------------------+ +------+ | Org1 | | Org2 | | | | +-------------+ | | +-------------+ | | | | | AAA Server | | | | AAA Server | | | | | | | | | | | | | | | +-------------+ | | +-------------+ | | User |======| |======| | | | | +-------------+ | | +-------------+ | | | | | Service | | | | Service | | | | | | Equipment | | | | Equipment | | | | | +-------------+ | | +-------------+ | +------+ | | | | +-------------------+ +-------------------+
Fig. 9 -- Distributed Services
The agreements between entities in figure 9 imply that the request from the User will be authenticated and authorized by the first organization, then forwarded to the second organization. Note that the sequence between User and Org1 may be different than between Org1 and Org2. The first might use a pull sequence, and the second might use an agent sequence. This example is illustrated in figure 10.
+-------------------+ +-------------------+ +------+ | Org1 | | Org2 | | | | +-------------+ | 3 | +-------------+ | | | | | AAA Server |--+------+->| AAA Server | | | | | | |<-+------+--| | | | | | +-------------+ | 6 | +-------------+ | | User | | /|\ | | | | /|\ | | | | |2 |7 | | |4 |5 | | | | | \|/ | | \|/ | | | | 1 | +-------------+ | | +-------------+ | | |------+->| Service | | | | Service | | | |<-----+--| Equipment | | | | Equipment | | | | 8 | +-------------+ | | +-------------+ | +------+ | | | | +-------------------+ +-------------------+
Fig. 10 -- A Possible Distributed Sequence
There are a number of other ways that authorization sequences for distributed services can be set up. For example, it is possible that, in order to reduce delay time in setting up a session, Org1 could send a response to the user before receiving a verification that Org2 has authorized the service. In that case Org1 would need to be able to revoke the authorization sent earlier if Org2 does not send the authorization in some amount of time.
Figure 11 shows a combination of Roaming and Distributed Services. Contract and trust relationships may be set up in number of ways, depending on a variety of factors, especially the business model.
+------+ +-------------------+ +-------------------+ | | | User Home Org | | SuperOrg | | | | +-------------+ | | +-------------+ | | | | | AAA Server | | | | AAA Server | | | | | | | | | | | | | | | +-------------+ | | +-------------+ | | | | | | | | | +-------------------+ +-------------------+ | | | | | | +-------------------+ +-------------------+ | User | | Org1 | | Org2 | | | | +-------------+ | | +-------------+ | | | | | AAA Server | | | | AAA Server | | | | | | | | | | | | | | | +-------------+ | | +-------------+ | | | | | | | | | | +-------------+ | | +-------------+ | | | | | Service | | | | Service | | | | | | Equipment | | | | Equipment | | | | | +-------------+ | | +-------------+ | | | | | | | +------+ +-------------------+ +-------------------+
Fig. 11 -- Roaming and Distributed Services
New entities that combine or add capabilities can be created to meet business needs. In figure 11, one such possibility, a SuperOrg entity is shown. The idea is that this entity would provide authentication and authorization for organizations that are providing services to end-users. It could be considered to be a wholesaler or broker. While not all authorization will require having a broker, authorization protocols should allow such entities to be created to meet legitimate requirements.
Having considered the basic players and how they interact, we will now consider different ways that authorization data may be stored in the network.
The Policy Framework (policy) Working Group is seeking to provide a framework to represent, manage, and share policies and policy information in a vendor-independent, interoperable, scalable manner. ,,. This section explores the relationship of policy and authorization and sets the stage for defining protocol requirements for supporting policy when included as part of multi-domain authorization. The work presented here builds on the policy framework, extending it to support policy across multiple domains.
One view of an authorization is that it is the result of evaluating policies of each organization that has an interest in the authorization decision. In this document the assumption is that each administration may have policies which may be indexed by user, by service, or by other attributes of the request. The policies of each administration are defined independently of other administrations.
Each independent policy must be 1) retrieved, 2) evaluated, and 3) enforced.
Policy definitions are maintained and stored in a policy repository  by (or on behalf of) the organization that requires them. The Policy Framework WG is working on a way to describe policy . Other implementations describe policy as a set of ACL lists. Policy definitions must be retrieved in order to be evaluated and enforced. Policy Definitions can be indexed by requester, by service attribute, or by some other key. The organization requiring the policy is also responsible for determining which policy is to be applied to a specific authorization request.
Policy retrieval is typically done by the administration that defines the policy or by an agent acting for that administration. Thus a policy defining the times of day that a particular User is allowed to connect to the network is maintained and retrieved by the User Organization. A policy defining a time that ports will be unusable because of maintenance is maintained and retrieved by the Service Provider.
Note that some implementation may choose to have the Service Provider retrieve a policy from the User Home Organization using a distributed directory access protocol. This may be appropriate in some cases, but is not a general solution. To understand why, suppose the remote administration and the home administration communicate via a broker which proxies their communications. In this case the remote and home
administrations have no prior relationship, and therefore the home administration directory is unlikely to be open for access by the remote administration and vice versa.
Evaluation of policy requires access to information referenced by the policy. Often the information required is not available in the administration where the policy is retrieved. For example, checking that a user is allowed to login at the current time can readily be done by the User Home Organization because the User Home Organization has access to current time. But authorizing a user requiring a 2Mb/s path with less than 4 hops requires information available at a Service Provider and not directly available to the UHO, so the UHO must either 1) have a way to query a remote administration for the needed information or 2) forward the policy to the remote administration and have the remote administration do the actual evaluation or 3) attempt somehow to "shadow" the authoritative source of the information (e.g by having the Service Provider send updates to the UHO).
Applications might support either 1) or 2), and a general authorization protocol must allow both. Case 3) is not considered further as shadowing seems too "expensive" to be supported by an AAA protocol.
An example of case 1 is when a Service Provider forwards a request to a UHO which includes a query for the clearance code of the User. The Service Provider policy includes reference to the clearance code and the information in the reply is used as input to that policy.
An example of case 2 is when the UHO approves an authorization conditional on the Service Provider confirming that there is currently a specific resource available for its use. The UHO includes the "policy" along with a conditional authorization to the Service Provider.
Policy Enforcement is typically done by the Service Provider on the Service Equipment. The Service Equipment is equivalent to the Policy Target described in the Policy Framework . Thus a NAS may enforce destination IP address limits via "filters" and a Router may enforce QoS restrictions on incoming packets. The protocol that sends the information between the Service Equipment and the Service Provider AAA Server may be specific to the Service Equipment, but it seems that the requirements are not different in kind from what is required between other AAA servers.
In particular, an AAA Server could send a "policy" to the Service Equipment stating what the equipment should do under various situations. The Service equipment should either set up to "enforce" the policy or reject the request.
The AAA Server could also send a query to the Service Equipment for information it requires to evaluate a policy.
A policy is retrieved by a Policy Retrieval Point (PRP) from a Policy Repository, evaluated at a Policy Decision Point (PDP) or Policy Consumer, and enforced at a Policy Enforcement Point (PEP) or Policy Target .
Generally, any of the AAA Servers involved in an authorization transaction may retrieve a policy or evaluate a policy, and any of the Service Equipment may enforce a policy. Policy Repositories may reside on any of the AAA Servers or be located elsewhere in the network.
Information against which policy conditions are evaluated (such as resource status, session state, or time of day) are accessible at Policy Information Points (PIPs) and might be accessed using Policy Information Blocks (PIBs). An interesting question in any authorization application that uses policy is where are the PDPs, PRPs, PIPs and PEPs?
Figure 12 shows which policy elements may be available at different points in the model. In distributed services, there may be multiple Service Providers involved in the authorization transaction, and each may act as the policy elements shown below.
Note that the User (or requester) may also be a PRP (e.g. use policy description to specify what service is being requested), a PIP (have information needed by other entities to evaluate their policy), and a PDP (decide if it will accept a service with specific parameters).
+------+ +------------------------------+ | | | User Home Organization | | | | +-------------------+ PRP | | | | | AAA Server | PIP | | | | | | PDP | | | | +-------------------+ | | | | | | | +------------------------------+ | | | | | | +------------------------------+ | User | | Service Provider | | | | +-------------------+ PRP | | PRP | | | AAA Server | PIP | | PIP | | | | PDP | | PDP | | +-------------------+ | | | | | | | | +-------------------+ | | | | | Service | PIP | | | | | Equipment | PEP | | | | +-------------------+ | | | | | +------+ +------------------------------+ PRP = Policy Retrieval Point PIP = Policy Information Point PDP = Policy Decision Point PEP = Policy Enforcement Point
Fig. 12 -- Where Different Policy Elements May be Located
An AAA protocol must be able to transport both policy definitions and the information needed to evaluate policies. It must also support queries for policy information.
This section outlines another mechanism that could be used for securely transporting the attributes on which an authorization decision is to be made. Work on X.509 Attribute Certificates is currently being undertaken in the Public Key Infrastructure (PKIX) Working Group . This proposal is largely based on that work.
When considering authorization using certificate-based mechanisms, one is often less interested in the identity of the entity than in some other attributes, (e.g. roles, account limits etc.), which should be used to make an authorization decision.
In many such cases, it is better to separate this information from the identity for management, security, interoperability or other reasons. However, this authorization information may also need to be protected in a fashion similar to a public key certificate. The name used here for such a structure is an Attribute Certificate (AC) which is a digitally signed (certified) set of attributes.
An AC is a structure that is similar to an X.509 public key certificate  with the main difference being that it contains no public key. The AC typically contains group membership, role, clearance and other access control information associated with the AC owner. A syntax for ACs is also defined in the X.509 standard.
When making an access decision based on an AC, an access decision function (in a PEP, PDP or elsewhere) may need to ensure that the appropriate AC owner is the entity that has requested access. The linkage between the request and the AC can be achieved if the AC has a "pointer" to a Public Key Certificate (PKC) for the requester and that the PKC has been used to authenticate the request. Other forms of linkage can be defined which work with other authentication schemes.
As there is often confusion about the difference between public key certificates (PKC's) and attribute certificates (ACs), an analogy may help. A PKC can be considered to be like a passport: it identifies the owner, it tends to be valid for a long period, it is difficult to forge, and it has a strong authentication process to establish the owner's identity. An AC is more like an entry visa in that it is typically issued by a different authority than the passport issuing authority, and it doesn't have as long a validity period as a passport. Acquiring an entry visa typically requires presenting a passport that authenticates that owner's identity. As a consequence, acquiring the entry visa becomes a simpler procedure. The entry visa will refer to the passport as a part of how that visa specifies the terms under which the passport owner is authorized to enter a country.
In conjunction with authentication services, ACs provide a means to transport authorization information securely to applications. However, there are a number of possible communication paths that an AC may take.
In some environments, it is suitable for a client to "push" an AC to a server. This means that no new connections between the client and server domains are required. It also means that no search burden is imposed on servers, which improves performance.
In other cases, it is more suitable for a client simply to authenticate to the server and for the server to request the client's AC from an AC issuer or a repository. A major benefit of the this model is that it can be implemented without changes to the client and client/server protocol. It is also more suitable for some inter- domain cases where the client's rights should be assigned within the server's domain, rather than within the client's "home" domain.
There are a number of possible exchanges that can occur, and there are three entities involved: client, server, and AC issuer. In addition the use of a directory service as a repository for AC retrieval may be supported.
Figure 13 shows an abstract view of the exchanges that may involve ACs. Note that the lines in the diagram represent protocols which must be defined, not data flows. The PKIX working group will define the required acquisition protocols. One candidate for the lookup protocols is LDAP (once an LDAP schema exists which states where an AC is to be found).
+--------------+ +---------------+ | AAA Server/ | | | | AC Issuer +----+ | Directory | | | | | | +--+-----------+ | Server +-------+-------+ | | Acquisition | |Client | |Server |Acquisition +----------------------+ |Lookup | | | +--+-----------+ +--+----+-------+ | | AC in application | Service | | User +------------------------+ Equipment/ | | | protocol | AAA Server | +--+-----------+ +---------------+ | | Client Lookup +--+-----------+ | | | Directory | | | +--------------+
Fig. 13 -- AC Exchanges
Figure 14 shows the data flows which may occur in one particular case, that termed "push" above (section 2.1.3).
+--------------+ | AAA Server/ | | AC Issuer | | | +--+-----------+ | |Client |Acquisition (1) | +--+-----------+ +---------------+ | | AC in application | Service | | User +------------------------+ Equipment/ | | | protocol (2) | AAA Server | +--------------+ +---------------+
Fig. 14 -- One example of an AC exchange
In the diagram, the user first contacts the AC Issuer and then incorporates the AC into the application protocol. The Service Equipment must then validate the AC and use it as the basis for the access decision (this functionality may be distributed between a PEP and PDP).
Authorization requests may be chained through a set of servers, as described in previous sections. Each of the servers may have a contractual relationship with servers on either side of it in the chain. In many of the applications being considered, the authorization results in establishing of an ongoing service which we call a session. Each of the servers involved in the authorization may also want to keep track of the state of the session, and be able to effect changes to the session if required. To make it simple to discuss this capability, we assume that each AAA Server MAY have a Resource Manager component. Resource Managers tracking the same session need to be able to initiate changes to the session, and to inform other Resource Managers when changes occur. Communication between Resource Managers creates requirements for an authorization protocol.
An example of the use of resource management might be a user which sets up a QoS path through two ISPs, and while this path is active, one of the ISPs gets a request for more bandwidth from a higher priority user. The ISP may need to take some bandwidth from a the lower priority user's previously allocated session and give it to the
new request. To do this, each of the administrations in the authorization path must be informed and agree to the change (this could be considered to be "authorizing the new value").
When an AAA Server grants authorization of some resource to an AAA requester (either a User or another AAA Server), the server may need to maintain session state information. This is used to make decisions about new sessions based on the state of current sessions, and to allow monitoring of sessions by all interested AAA Servers.
Each session is identified by a session identifier, which must be unique within each AAA Server. Communication between AAA Servers must include the session identifier. It is desirable that the session identifier is the same across all AAA servers, otherwise each server will have to map identifiers from other servers to its own identifiers. A single session identifier significantly simplifies auditing and session control functions.
Maintaining session state across AAA administrative boundaries increases the complexity of the problem, especially if each AAA Server in the trust chain must keep state as well. This can be viewed as an interdomain database replication problem. The protocol must include tools to help manage replicated state. Some of the problems to be addressed are:
a) Service Equipment must be able to notify its Resource Manager when a session terminates or changes state in some other way. The Resource Manager must inform other Resource Managers which keep state for this session.
b) The Resource Manager will need to set a time limit for each session which must be refreshed by having the Resource Manager query for authoritative status or by having the authoritative source send periodic keep alive messages that are forwarded to all Resource Managers in the authorization chain. Determining the appropriate session lifetime may be application specific and depends on the acceptable level of risk. If the service being offered is billed based on time, the session lifetime may need to be relatively small; if the service is billed on usage, the lifetime may be relatively large.
c) Any Resource Manager in the chain must have the ability to
terminate a session. This requires the Resource Manager to have
knowledge of at least the adjacent AAA Servers in the
An example of how resource management can be used is in the PPP dialin application. A home ISP may wish to restrict the number of concurrent sessions that a user can have at any given time. This is particularly important when service providers give all-you-can-eat Internet access. The possibility for fraud is quite large, since a user could provide his or her username/password to many people, causing a loss of revenue. Resource management would allow the home ISP AAA server to identify when a user is active and to reject any authorization request for the user until termination indication is received from the NAS or until the session expires.
This section describes the functions of the Resource Manager in more detail.
The Resource Manager is the component which tracks the state of sessions associated with an AAA Server or Service Equipment. It also may allocate resources to a session (e.g. IP addresses) and may track use of resources allocated by peer resource managers to a session (e.g. bandwidth in a foreign administrative domain). The resource manager also provides interfaces to allow the User to acquire or release authorized sessions.
The Resource Manager maintains all session specific AAA state information required by the AAA Server. That state information may include pointers to peer Resource Managers in other administrative domains that possess additional AAA state information that refers to the same session. The Resource Manager is the anchor point in the AAA Server from which a session can be controlled, monitored, and coordinated even if that session is consuming network resources or services across multiple Service Provider administrative domains.
The Resource Manager has several important functions:
a) It allows a Service Provider operations staff to inspect the status of any of the allocated resources and services including resources that span foreign Service Provider administrative boundaries. The peer Resource Managers will cooperatively share only the state information subset that is required to assist in diagnosing cross-domain trouble tickets. The network operator may also modify or altogether cancel one of the User's active authorizations.
b) It is the process contacted by other Resource Managers to inform the AAA Server that a specific session has been cancelled. This information is relayed to the other peer Resource Managers that also know about that session and hence must cancel it.
c) The Resource Manager conceals the identity and location of its private internal AAA components from other administrative domains and from the User, while at the same time facilitating cooperation between those domains.
d) The Resource Manager cooperates with "policy servers" or Policy Decision Points (PDPs). The Resource Manager maintains internal state information, possibly complex cross-administrative domain information, supported by dialogues with its peer Resource Managers. A policy server can use the state information when evaluating a particular policy.
e) The separation of the Resource Manager and the policy server into two distinct architectural components allows a single session to span multiple administrative domains, where each administrative domain has one or more policy server cooperating with its Resource Manager.
AAA resource managers will normally use the same trust relationships needed for authorization sequences. It is possible for independent relationships to be established, but that is discouraged.
An AAA Server is responsible for securely forwarding AAA messages to
the correct destination system or process in the AAA infrastructure.
Two well known examples are forwarding AAA messages for a roaming AAA
service, and forwarding AAA messages for a distributed AAA service.
The same principle can also be applied to intra-domain
communications. The message forwarding is done in one of two modes.
The first mode is when an AAA server needs to forward a message to a peer AAA server that has a known "logical destination address" that must be resolved by an application-specific procedure into its actual network address. Typically the forwarding procedure indexes into a database by an application-specific identifier to discover the peer's network address. For example, in the roaming dialin application, the application-specific identifier may be an NAI. A bandwidth brokerage application would use other search indices unique to its problem domain to select the addressed peer AAA server. After the address resolution procedure has completed successfully, then the AAA server transmits the message to its peer over the connection associated with that destination network address.
The second mode is when the AAA server already has an established session representing an authorization. The session's state contains the addressing and context used to direct the message to its destination peer AAA server, PDP, PEP, or User. The message is sent
over the AAA server's connection to that destination peer, multiplexed with other session's messages. The message must be qualified by a session identifier that is understood by both end points. The AAA message's destination may be either intra- administrative domain, or inter-administrative domain. In the former case, the destination process may reside on the same system as the AAA server.
In addition to the above message forwarding processing, the underlying message delivery service must meet the following requirements:
- Unicast capability -- An end system can send a message to any other end system with minimal latency of session setup/disconnect overhead messages, and no end system overhead of keeping state information about every potential peer. - Data integrity and error detection -- This data transport protocol assumes an underlying datagram network layer service that includes packet discard on error detection, and data integrity protection against third party modifications. - Reliable data transport assurance -- When an end system successfully receives a message marked receipt requested, it must acknowledge that message to the sending system by either piggybacking the acknowledgement on an application-specific reply message, or else as a standalone acknowledgement message. The sending system maintains a retry timer; when the timer expires, the sending system retransmits a copy of its original message. It gives up after a configurable number of unsuccessful retries. - Sequenced data delivery -- If multiple messages are sent between a pair of end systems, those messages are delivered to the addressed application in the same order as they were transmitted. Duplicates are silently suppressed. - Responsive to network congestion feedback -- When the network enters into congestion, the end systems must detect that condition, and they must back off their transmission rate until the congestion subsides. The back off and recovery algorithms must avoid oscillations.
When AAA servers communicate through intermediate AAA servers, such as brokers, it may be necessary that a part of the payload be secure between the originator and the target AAA server. The security requirement may consist of one or more of the following: end-to-end
message integrity, confidentiality, replay protection, and nonrepudiation. Furthermore, it is a requirement that intermediate AAA servers be able to append information such as local policy to a message before forwarding the message to its intended destination. It may also be required that an intermediate AAA Server sign such appended information.
This requirement has been clearly documented in , which describes many current weaknesses of the RADIUS protocol  in roaming networks since RADIUS does not provide such functionality. One well-known attack is the ability for the intermediate nodes to modify critical accounting information, such as a session time.
Most popular security protocols (e.g. IPSec, SSL, etc.) do not provide the ability to secure a portion of the payload. Therefore, it may be necessary for the AAA protocol to implement its own security extensions to provide end-to-end security.
The techniques described above allow for great flexibility in distributing the components required for authentication and authorization. However, working groups such as Roamops and MobileIP have identified requirements to minimize Internet traversals in order to reduce latency. To support these requirements, data fields necessary for both authentication and authorization SHOULD be able to be carried in a single message set. This is especially important when there are intermediate servers (such as Brokers) in the AAA chain.
Furthermore, it should be possible for the Brokers to allow end-to- end (direct) authentication and authorization. This can be done as follows. The User Home Organization generates a ticket which is signed using the UHO's private key. The ticket is carried in the accounting messages. The accounting messages must flow through the Broker since the Broker is acting as the settlement agent and requires this information. There are Brokers that will require to be in the authentication and authorization path as well since they will use this information to detect fraudulent activity, so the above should be optional.
In order for end-to-end authentication and authorization to occur, it may be necessary for the Broker to act as a certificate authority. All members of the roaming consortium would be able to trust each other (to an extent) using the certificates. A Service Provider's AAA server that sends a request to the Broker should be able to receive a redirect message which would allow the two peers (Service Provider and UHO) to interact directly. The redirect message from
the Broker should include the UHO's certificate, which eliminates the Service Provider from accessing the certificate archive. The request from the Service Provider could include its own certificate, and a token from the Broker's redirect message that is timestamped and guarantees that the Service Provider is in good standing with the Broker. This eliminates the home domain from accessing the Certificate Revocation List (CRL).
The above has introduced the basic players in an authorization transaction as User, User Home Organization, Service Provider's AAA Server, and Service Equipment. It has discussed relationships between entities based on agreements or contracts, and on "trust". Examples of authorization sequences have been given.
Concepts of roaming and distributed services have been briefly described. Combination of roaming and distributed services was also considered and the concept of a "wholesaler" or Broker was introduced. We have considered the use of policies and attribute certificates to store and transmit authorization data. We discussed the problem of managing the resources to which access has been authorized including the problem of tracking state information for session-oriented services, and we defined the Resource Manager component of a AAA Server. We considered the problem of forwarding AAA messages among servers in possibly different administrative domains. We considered the need for end-to-end security of portions of the payload of authorization messages that pass through intermediate AAA Servers. Finally we stressed the need for support of a streamlined authorization process that minimizes delay for latency-sensitive applications.
The intent is that this will provide support for discussing and understanding requirements of specific applications that need authorization services.
Authorization is itself a security mechanism. As such, it is important that authorization protocols cannot easily be abused to circumvent the protection they are intended to ensure. It is the responsibility of protocol designers to design their protocols to be resilient against well-known types of attacks. The following are some considerations that may guide protocol designers in the development of authorization protocols.
Authorization protocols must not be susceptible to replay attacks. If authentication data is carried with the authorization data, for example, the authentication protocol used must either be impervious to replay or else the confidentiality of the authentication data must be protected.
If proxying is required, the authorization protocol must not be susceptible to man-in-the-middle attacks.
If the push model is used, the confidentiality of the authorization data must be ensured so that it may not be hijacked by third parties and used to obtain a service fraudulently.
If the agent model is used, the binding between the authorization and the service itself must be protected to prevent service authorized to one party from being fraudulently received by another.
In addition to guarding against circumvention, authorization protocols designed according to this framework will have some intrinsic security requirements. These are included among the requirements in  and summarized briefly below.
Among the intrinsic security needs is the fact that authorization protocols may carry sensitive information. It is necessary to protect such information from disclosure to unauthorized parties including (as discussed in section 8) even certain parties involved in the authorization decision.
We have discussed the use of multi-party trust chains involving relaying of authorization data through brokers or other parties. In such cases, the integrity of the chain must be maintained. It may be necessary to protect the data exchanged between parties using such mechanisms as encryption and digital signatures.
Finally, because authorization will be necessary to gain access to many Internet services, a denial of service attack against an authorization server can be just as effective as a denial of service attack against the service equipment itself in preventing access to Internet services.
Attribute Certificate -- structure containing authorization
attributes which is digitally signed using public key
Contract Relationship -- a relation established between two or more business entities where terms and conditions determine the exchange of goods or services.
Distributed Service -- a service that is provided by more than one Service Provider acting in concert.
Dynamic Trust Relationship -- a secure relationship which is dynamically created between two entities who may never have had any prior relationship. This relationship can be created if the involved entities have a mutually trusted third party. Example: A merchant trusts a cardholder at the time of a payment transaction because they both are known by a credit card organization.
Policy Decision Point (PDP) -- The point where policy decisions are made.
Policy Enforcement Point (PEP) -- The point where the policy decisions are actually enforced.
Resource Manager -- the component of an AAA Server which tracks the state of sessions associated with the AAA Server or its associated Service Equipment and provides an anchor point from which a session can be controlled, monitored, and coordinated.
Roaming -- An authorization transaction in which the Service Provider and the User Home Organization are two different organizations. (Note that the dialin application is one for which roaming has been actively considered, but this definition encompasses other applications as well.)
Security Association -- a collection of security contexts, between a pair of nodes, which may be applied to protocol messages exchanged between them. Each context indicates an authentication algorithm and mode, a secret (a shared key, or appropriate public/private key pair), and a style of replay protection in use. 
Service Equipment -- the equipment which provides a service.
Service Provider -- an organization which provides a service.
Static Trust Relationship -- a pre-established secure relationship between two entities created by a trusted party. This relationship facilitates the exchange of AAA messages with a certain level of security and traceability. Example: A network operator (trusted party) who has access to the wiring closet
creates a connection between a user's wall outlet and a particular network port. The user is thereafter trusted -- to a certain level -- to be connected to this particular network port.
User -- the entity seeking authorization to use a resource or a service.
User Home Organization (UHO) -- An organization with whom the User has a contractual relationship which can authenticate the User and may be able to authorize access to resources or services.
 Bradner, S., "The Internet Standards Process -- Revision 3", BCP 9, RFC 2026, October 1996.
 Farrell, S., Vollbrecht, J., Calhoun, P., Gommans, L., Gross, G., de Bruijn, B., de Laat, C., Holdrege, M. and D. Spence, "AAA Authorization Requirements", RFC 2906, August 2000.
 Vollbrecht, J., Calhoun, P., Farrell, S., Gommans, L., Gross, G., de Bruijn, B., de Laat, C., Holdrege, M. and D. Spence, "AAA Authorization Application Examples", RFC 2905, August 2000.
 Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
 Stevens, M., "Policy Framework", Work in Progress.
 Strassner, John, Ed Ellesson, and Bob Moore, "Policy Core Information Model -- Version 1 Specification", Work in Progress.
 Strassner, John, et al, "Policy Framework LDAP Core Schema", Work in Progress.
 Farrell, Stephen and Russell Housley, "An Internet Attribute Certificate Profile for Authorization", Work in Progress.
 Housley, R., Ford, W., Polk, W. and D. Solo, "Internet X.509 Public Key Infrastructure -- Certificate and CRL Profile", RFC 2459, January 1999.
 Aboba, B. and J. Vollbrecht, "Proxy Chaining and Policy Implementation in Roaming", RFC 2607, June 1999.
 Rigney, C., Rubens, A., Simpson, W. and S. Willens, "Remote Authentication Dial In User Service (RADIUS)", RFC 2138, April 1997.
 Perkins, C., "IP Mobility Support", RFC 2002, October 1996.
 Yavatkar, R., Pendarakis, D. and R. Guerin, "A Framework for Policy-based Admission Control", RFC 2753, January 2000.
John R. Vollbrecht
Interlink Networks, Inc.
775 Technology Drive, Suite 200
Ann Arbor, MI 48108
Phone: +1 734 821 1205
Fax: +1 734 821 1235 Mail: firstname.lastname@example.org
Pat R. Calhoun
Network and Security Research Center, Sun Labs
Sun Microsystems, Inc.
15 Network Circle
Menlo Park, California, 94025
Phone: +1 650 786 7733
Fax: +1 650 786 6445 EMail: email@example.com
61 Fitzwilliam Lane
Phone: +353 1 647 7406
Fax: +353 1 647 7499 EMail: firstname.lastname@example.org
Enterasys Networks EMEA
2841 XM Moordrecht
Phone: +31 182 379279
or at University of Utrecht:
George M. Gross
184 Liberty Corner Road, m.s. LC2N-D13
Warren, NJ 07059
Phone: +1 908 580 4589
Fax: +1 908-580-4991 Email: email@example.com
Betty de Bruijn
Interpay Nederland B.V.
3526 LB Utrecht
Phone: +31 30 2835104
Cees T.A.M. de Laat
Physics and Astronomy dept.
Phone: +31 30 2534585
Phone: +31 30 2537555
223 Ximeno Ave.
Long Beach, CA 90803
David W. Spence
Interlink Networks, Inc.
775 Technology Drive, Suite 200
Ann Arbor, MI 48108
Phone: +1 734 821 1203
Fax: +1 734 821 1235 EMail: firstname.lastname@example.org
Copyright © The Internet Society (2000). 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 implementation 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 included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet 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 ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Funding for the RFC Editor function is currently provided by the Internet Society.