+--------------+
                                 |   Identity   |
                                 |   Provider   |
                                 |    (IdP)     |
                                 +-^----------^-+
                                   * EAP      o RADIUS/
                                   *          o Diameter
                                 --v----------v--
                              ///                \\\
                            //                      \\     
                           |        Federation        | 
                           |        Substrate         | 
                            \\                      //  
                              \\\                ///
                                 --^----------^--
                                   * EAP      o RADIUS/
                                   *          o Diameter
+-------------+                  +-v----------v--+
|             |<---------------->|               |
| Client      |  EAP/EAP Method  | Relying Party |
| Application |<****************>|     (RP)      |
|             |  GSS-API         |               |
|             |<---------------->|               |
|             |  Application     |               |
|             |  Protocol        |               |
|             |<================>|               |
+-------------+                  +---------------+

Legend:

 <****>: Client-to-IdP Exchange
 <---->: Client-to-RP Exchange
 <oooo>: RP-to-IdP Exchange 
 <====>: Protocol through which GSS-API/GS2 exchanges are tunneled
 

Figure 2: ABFAB Protocol Instantiation

2.1. Relying Party to Identity Provider

Communications between the Relying Party and the Identity Provider is done by the federation substrate. This communication channel is responsible for:

• Establishing the trust relationship between the RP and the IdP.
• Determining the rules governing the relationship.
• Conveying EAP packets between the RP and IdP.

The ABFAB working group has chosen the AAA framework for the messages transported between the RP and IdP. This allows for the current AAA protocols to be used to establish the trust relationship between the RP and the IdP. Future protocols that support the same framework but do different routing may be used in the future. There is currently an effort to setup a framework that creates a trusted point-to-point channel on the fly. The ABFAB protocol itself details the method of establishing the trust relationship between the RP and the client.

2.1.1. AAA, RADIUS and Diameter

Interestingly, for network access authentication the usage of the AAA framework with RADIUS [RFC2865] and Diameter [RFC3588] was quite successful from a deployment point of view. To map the terminology used in Figure 1 to the AAA framework the IdP corresponds to the AAA server, the RP corresponds to the AAA client, and the technical building blocks of a federation are AAA proxies, relays and redirect agents (particularly if they are operated by third parties, such as AAA brokers and clearing houses). The front-end, i.e. the end host to AAA client communication, is in case of network access authentication offered by link layer protocols that forward authentication protocol exchanges back-and-forth. An example of a large scale RADIUS-based federation is EDUROAM.

By using the AAA framework, ABFAB gets a lot of mileage as many of the federation agreements already exist and merely need to be expanded to cover the ABFAB additions. The AAA framework has already addressed some of the problems outlined above. For example,

• It already has a method for routing requests based on a domain.
• It already has an extensible architecture allowing for new attributes to be defined and transported.
• Pre-existing relationships can be re-used.

The astute reader will notice that RADIUS and Diameter have substantially similar characteristics. Why not pick one? RADIUS and Diameter are deployed in different environments. RADIUS can often be found in enterprise and university networks, and is also in use by fixed network operators. Diameter, on the other hand, is deployed by mobile operators. Another key difference is that today RADIUS is largely transported upon UDP. We leave as a deployment decision, which protocol will be appropriate. The protocol defines all the necessary new AAA attributes as RADIUS attributes. A future document would defined the same AAA attributes for a Diameter environment. We also note that there exist proxies which convert from RADIUS to Diameter and back. This makes it possible for both to be deployed in a single federation substrate.

Through the integrity protection mechanisms in the AAA framework, the identity provider can establish technical trust that messages are being sent by the appropriate relying party. Any given interaction will be associated with one federation at the policy level. The legal or business relationship defines what statements the identity provider is trusted to make and how these statements are interpreted by the relying party. The AAA framework also permits the relying party or elements between the relying party and identity provider to make statements about the relying party.

The AAA framework provides transport for attributes. Statements made about the subject by the identity provider, statements made about the relying party and other information are transported as attributes.

One demand that the AAA substrate makes of the upper layers is that they must properly identify the end points of the communication. It must be possible for the AAA client at the RP to determine where to send each RADIUS or Diameter message. Without this requirement, it would be the RP's responsibility to determine the identity of the client on its own, without the assistance of an IdP. This architecture makes use of the Network Access Identifier (NAI), where the IdP is indicated by the realm component [RFC4282]. The NAI is represented and consumed by the GSS-API layer as GSS_C_NT_USER_NAME as specified in [RFC2743]. The GSS-API EAP mechanism includes the NAI in the EAP Response/Identity message.

2.1.2. Discovery and Rules Determination

While we are using the AAA protocols to communicate with the IdP, the RP may have multiple federation substrates to select from. The RP has a number of criteria that it will use in selecting which of the different federations to use:

• The federation selected must be able to communicate with the IdP.
• The federation selected must match the business rules and technical policies required for the RP security requirements.

The RP needs to discover which federation will be used to contact the IdP. The first selection criteria in discovery is going to be the name of the IdP to be contacted. The second selection criteria in discovery is going to be the set of business rules and technical policies governing the relationship; this is called rules determination. The RP also needs to establish technical trust in the communications with the IdP.

Rules determination covers a broad range of decisions about the exchange. One of these is whether the given RP is permitted to talk to the IdP using a given federation at all, so rules determination encompasses the basic authorization decision. Other factors are included, such as what policies govern release of information about the principal to the RP and what policies govern the RP's use of this information. While rules determination is ultimately a business function, it has significant impact on the technical exchanges. The protocols need to communicate the result of authorization. When multiple sets of rules are possible, the protocol must disambiguate which set of rules are in play. Some rules have technical enforcement mechanisms; for example in some federations intermediaries validate information that is being communicated within the federation.

ABFAB has not formally defined any part of discovery at this point. The process of specifying and evaluating the business rules and technical policies is too complex to provide a simple framework. There is not currently a way to know if a AAA proxy is able to communicate with a specific IdP, although this may change with some of the routing protocols that are being considered. At the present time, the discovery process is going to be a manual configuration process.

2.1.3. Routing and Technical Trust

Several approaches to having messages routed through the federation substrate are possible. These routing methods can most easily be classified based on the mechanism for technical trust that is used. The choice of technical trust mechanism constrains how rules determination is implemented. Regardless of what deployment strategy is chosen, it is important that the technical trust mechanism be able to validate the names of both parties to the exchange. The trust mechanism must to ensure that the entity acting as IdP for a given NAI is permitted to be the IdP for that realm, and that any service name claimed by the RP is permitted to be claimed by that entity. Here are the categories of technical trust determination:

AAA Proxy:

The simplest model is that an RP supports a request directly to an AAA proxy. The hop-by-hop integrity protection of the AAA fabric provides technical trust. An RP can submit a request directly to a federation. Alternatively, a federation disambiguation fabric can be used. Such a fabric takes information about what federations the RP is part of and what federations the IdP is part of and routes a message to the appropriate federation. The routing of messages across the fabric plus attributes added to requests and responses provides rules determination. For example, when a disambiguation fabric routes a message to a given federation, that federation's rules are chosen. Name validation is enforced as messages travel across the fabric. The entities near the RP confirm its identity and validate names it claims. The fabric routes the message towards the appropriate IdP, validating the IdP's name in the process. The routing can be statically configured. Alternatively a routing protocol could be developed to exchange reachability information about given IdPs and to apply policy across the AAA fabric. Such a routing protocol could flood naming constraints to the appropriate points in the fabric.

Trust Broker:

Instead of routing messages through AAA proxies, some trust broker could establish keys between entities near the RP and entities near the IdP. The advantage of this approach is efficiency of message handling. Fewer entities are needed to be involved for each message. Security may be improved by sending individual messages over fewer hops. Rules determination involves decisions made by trust brokers about what keys to grant. Also, associated with each credential is context about rules and about other aspects of technical trust including names that may be claimed. A routing protocol similar to the one for AAA proxies is likely to be useful to trust brokers in flooding rules and naming constraints.

Global Credential:

A global credential such as a public key and certificate in a public key infrastructure can be used to establish technical trust. A directory or distributed database such as the Domain Name System is used by the RP to discover the endpoint to contact for a given NAI. Either the database or certificates can provide a place to store information about rules determination and naming constraints. Provided that no intermediates are required (or appear to be required) and that the RP and IdP are sufficient to enforce and determine rules, rules determination is reasonably simple. However applying certain rules is likely to be quite complex. For example if multiple sets of rules are possible between an IdP and RP, confirming the correct set is used may be difficult. This is particularly true if intermediates are involved in making the decision. Also, to the extent that directory information needs to be trusted, rules determination may be more complex.

Real-world deployments are likely to be mixtures of these basic approaches. For example, it will be quite common for an RP to route traffic to a AAA proxy within an organization. That proxy could then use any of the three methods to get closer to the IdP. It is also likely that rather than being directly reachable, the IdP may have a proxy on the edge of its organization. Federations will likely provide a traditional AAA proxy interface even if they also provide another mechanism for increased efficiency or security.

2.1.4. SAML Assertions

For the traditional use of AAA frameworks, network access, the only requirement that was necessary to grant access was an affirmative response from the IdP. In the ABFAB world, the RP may need to get additional information about the client before granting access. ABFAB therefore has a requirement that it can transport an arbitrary set of attributes about the client from the IdP to the RP.

Security Assertions Markup Language (SAML) [OASIS.saml-core-2.0-os] was designed in order to carry an extensible set of attributes about a subject. Since SAML is extensible in the attribute space, ABFAB has no immediate needs to update the core SAML specifications for our work. It will be necessary to update IdPs that need to return SAML assertions to IdPs and for both the IdP and the RP to implement a new SAML profile designed to carry SAML assertions in AAA. The new profile can be found in RFCXXXX [I-D.ietf-abfab-aaa-saml].

There are two issues that need to be highlighted:

• The security of SAML assertions.
• Namespaces and mapping of SAML attributes.

SAML assertions have an optional signature that can be used to protect and provide origination of the assertion. These signatures are normally based on asymmetric key operations and require that the verifier be able to check not only the cryptographic operation, but also the binding of the originators name and the public key. In a federated environment it will not always be possible for the RP to validate the binding, for this reason the technical trust established in the federation is used as an alternate method of validating the origination and integrity of the SAML Assertion.

Attributes placed in SAML assertions can have different namespaces assigned to the same name. In many, but not all, cases the federation agreements will determine what attributes can be used in a SAML statement. This means that the RP needs to map from the federation names, types and semantics into the ones that the policies of the RP are written in. In other cases the federation substrate may modify the SAML assertions in transit to do the necessary namespace, naming and semantic mappings as the assertion crosses the different boundaries in the federation. If the proxies are modifying the SAML Assertion, then will obviously remove any signatures on the SAML assertions as they would no longer validate. In this case the technical trust is the required mechanism for validating the integrity of the assertion. Finally, the attributes may still be in the namespace of the originating IdP. When this occurs the RP will need to get the required mapping operations from the federation agreements and do the appropriate mappings itself.

2.2. Client To Identity Provider

Looking at the communications between the client and the IdP, the following items need to be dealt with:

• The client and the IdP need to mutually authenticate each other.
• The client and the IdP need to mutually agree on the identity of the RP.

ABFAB selected EAP for the purposes of mutual authentication and assisted in creating some new EAP channel binding documents for dealing with determining the identity of the RP. A framework for the channel binding mechanism has been defined in RFC 6677 [RFC6677] that allows the IdP to check the identity of the RP provided by the AAA framework with that provided by the client.

2.2.1. Extensible Authentication Protocol (EAP)

Traditional web federation does not describe how a subject interacts with an identity provider for authentication. As a result, this communication is not standardized. There are several disadvantages to this approach. Since the communication is not standardized, it is difficult for machines to correctly enter their credentials with different authentications, where Individuals can correctly identify the entyr mechanism on the fly. The use of browsers for authentication restricts the deployment of more secure forms of authentication beyond plaintext username and password known by the server. In a number of cases the authentication interface may be presented before the subject has adequately validated they are talking to the intended server. By giving control of the authentication interface to a potential attacker, then the security of the system may be reduced and phishing opportunities introduced.

As a result, it is desirable to choose some standardized approach for communication between the subject's end-host and the identity provider. There are a number of requirements this approach must meet.

Experience has taught us one key security and scalability requirement: it is important that the relying party not get possession of the long-term secret of the client. Aside from a valuable secret being exposed, a synchronization problem can develop when the client changes keys with the IdP.

Since there is no single authentication mechanism that will be used everywhere there is another associated requirement: The authentication framework must allow for the flexible integration of authentication mechanisms. For instance, some IdPs require hardware tokens while others use passwords. A service provider wants to provide support for both authentication methods, and other methods from IdPs not yet seen.

Fortunately, these requirements can be met by utilizing standardized and successfully deployed technology, namely by the Extensible Authentication Protocol (EAP) framework [RFC3748]. Figure 2 illustrates the integration graphically.

EAP is an end-to-end framework; it provides for two-way communication between a peer (i.e,service client or principal) through the authenticator (i.e., service provider) to the back-end (i.e., identity provider). Conveniently, this is precisely the communication path that is needed for federated identity. Although EAP support is already integrated in AAA systems (see [RFC3579] and [RFC4072]) several challenges remain:

• The first is how to carry EAP payloads from the end host to the relying party.
• Another is to verify statements the relying party has made to the subject, confirm these statements are consistent with statements made to the identity provider and confirm all the above are consistent with the federation and any federation-specific policy or configuration.
• Another challenge is choosing which identity provider to use for which service.

The EAP method used for ABFAB needs to meet the following requirements: [I-D.ietf-emu-eap-tunnel-method] either alone (if client certificates are used) or with an inner EAP method that does mutual authentication.

• It needs to provide mutual authentication of the client and IdP.
• It needs to support channel binding.

As of this writing, the only EAP method that meets these criteria is TEAP

2.2.2. EAP Channel Binding

EAP channel binding is easily confused with a facility in GSS-API also called channel binding. GSS-API channel binding provides protection against man-in-the-middle attacks when GSS-API is used as authentication inside some tunnel; it is similar to a facility called cryptographic binding in EAP. See [RFC5056] for a discussion of the differences between these two facilities and Section 6.1 for how GSS-API channel binding is handled in this mechanism.

The client knows, in theory, the name of the RP that it attempted to connect to, however in the event that an attacker has intercepted the protocol, the client and the IdP need to be able to detect this situation. A general overview of the problem along with a recommended way to deal with the channel binding issues can be found in RFC 6677 [RFC6677].

Since that document was published, a number of possible attacks were found and methods to address these attacks have been outlined in [I-D.hartman-emu-mutual-crypto-bind].

2.3. Client to Relying Party

The final set of interactions between parties to consider are those between the client and the RP. In some ways this is the most complex set since at least part of it is outside the scope of the ABFAB work. The interactions between these parties include:

• Running the protocol that implements the service that is provided by the RP and desired by the client.
• Authenticating the client to the RP and the RP to the client.
• Providing the necessary security services to the service protocol that it needs beyond authentication.

2.3.1. GSS-API

One of the remaining layers is responsible for integration of federated authentication into the application. There are a number of approaches that applications have adopted for security. So, there may need to be multiple strategies for integration of federated authentication into applications. However, we have started with a strategy that provides integration to a large number of application protocols.

Many applications such as SSH [RFC4462], NFS [RFC2203], DNS [RFC3645] and several non-IETF applications support the Generic Security Services Application Programming Interface [RFC2743]. Many applications such as IMAP, SMTP, XMPP and LDAP support the Simple Authentication and Security Layer (SASL) [RFC4422] framework. These two approaches work together nicely: by creating a GSS-API mechanism, SASL integration is also addressed. In effect, using a GSS-API mechanism with SASL simply requires placing some headers on the front of the mechanism and constraining certain GSS-API options.

GSS-API is specified in terms of an abstract set of operations which can be mapped into a programming language to form an API. When people are first introduced to GSS-API, they focus on it as an API. However, from the prospective of authentication for non-web applications, GSS-API should be thought of as a protocol not an API. It consists of some abstract operations such as the initial context exchange, which includes two sub-operations (gss_init_sec_context and gss_accept_sec_context). An application defines which abstract operations it is going to use and where messages produced by these operations fit into the application architecture. A GSS-API mechanism will define what actual protocol messages result from that abstract message for a given abstract operation. So, since this work is focusing on a particular GSS-API mechanism, we generally focus on protocol elements rather than the API view of GSS-API.

The API view has significant value. Since the abstract operations are well defined, the set of information that a mechanism gets from the application is well defined. Also, the set of assumptions the application is permitted to make is generally well defined. As a result, an application protocol that supports GSS-API or SASL is very likely to be usable with a new approach to authentication including this one with no required modifications. In some cases, support for a new authentication mechanism has been added using plugin interfaces to applications without the application being modified at all. Even when modifications are required, they can often be limited to supporting a new naming and authorization model. For example, this work focuses on privacy; an application that assumes it will always obtain an identifier for the principal will need to be modified to support anonymity, unlinkability or pseudonymity.

So, we use GSS-API and SASL because a number of the application protocols we wish to federate support these strategies for security integration. What does this mean from a protocol standpoint and how does this relate to other layers? This means we need to design a concrete GSS-API mechanism. We have chosen to use a GSS-API mechanism that encapsulates EAP authentication. So, GSS-API (and SASL) encapsulate EAP between the end-host and the service. The AAA framework encapsulates EAP between the relying party and the identity provider. The GSS-API mechanism includes rules about how principals and services are named as well as per-message security and other facilities required by the applications we wish to support.

2.3.2. Protocol Transport

The transport of data between the client and the relying party is not provided by GSS-API. GSS-API creates and consumes messages, but it does not provide the transport itself, instead the protocol using GSS-API needs to provide the transport. In many cases HTTP or HTTPS is used for this transport, but other transports are perfectly acceptable. The core GSS-API document [RFC2743] provides some details on what requirements exist.

In addition we highlight the following:

• The transport does not need to provide either privacy or integrity. After GSS-EAP has finished negotiation, GSS-API can be used to provide both services. If the negotiation process itself needs protection from eavesdroppers then the transport would need to provide the necessary services.
• The transport needs to provide reliable transport of the messages.
• The transport needs to ensure that tokens are delivered in order during the negotiation process.
• GSS-API messages need to be delivered atomically. If the transport breaks up a message it must also reassemble the message before delivery.

3. Application Security Services

One of the key goals is to integrate federated authentication into existing application protocols and where possible, existing implementations of these protocols. Another goal is to perform this integration while meeting the best security practices of the technologies used to perform the integration. This section describes security services and properties required by the EAP GSS-API mechanism in order to meet these goals. This information could be viewed as specific to that mechanism. However, other future application integration strategies are very likely to need similar services. So, it is likely that these services will be expanded across application integration strategies if new application integration strategies are adopted.

3.1. Authentication

GSS-API provides an optional security service called mutual authentication. This service means that in addition to the initiator providing (potentially anonymous or pseudonymous) identity to the acceptor, the acceptor confirms its identity to the initiator. Especially for the ABFAB context, this service is confusingly named. We still say that mutual authentication is provided when the identity of an acceptor is strongly authenticated to an anonymous initiator.

RFC 2743, unfortunately, does not explicitly talk about what mutual authentication means. Within this document we therefore define it as:

• If a target name is supplied to the initiator, then the initiator trusts that the supplied target name describes the acceptor. This implies both that appropriate cryptographic exchanges took place for the initiator to make such a trust decision, and that after evaluating the results of these exchanges, the initiator's policy trusts that the target name is accurate.
• If no target name is supplied to the initiator, then the initiator trusts that the acceptor name, supplied by the acceptor, correctly names the entity it is communicating with.
• Both the initiator and acceptor have the same key material for per-message keys and both parties have confirmed they actually have the key material. In EAP terms, there is a protected indication of success.

Mutual authentication is an important defense against certain aspects of phishing. Intuitively, users would like to assume that if some party asks for their credentials as part of authentication, successfully gaining access to the resource means that they are talking to the expected party. Without mutual authentication, the acceptor could "grant access" regardless of what credentials are supplied. Mutual authentication better matches this user intuition.

It is important, therefore, that the GSS-EAP mechanism implement mutual authentication. That is, an initiator needs to be able to request mutual authentication. When mutual authentication is requested, only EAP methods capabale of providing the necessary service can be used, and appropriate steps need to be taken to provide mutual authentication. A broader set of EAP methods could be supported when a particular application does not request mutual authentication. It is an open question whether the mechanism will permit this.

The AAA infrastructure MAY hide the peer's identity from the GSS-API acceptor, providing anonymity between the peer and initiator. At this time, whether the identity is disclosed is determined by EAP server policy rather than by an indication from the peer. Also, peers are unlikely to be able to determine whether anonymous communication will be provided. For this reason, peers are unlikely to set the anonymous return flag from GSS_Init_Sec_context.

3.2. GSS-API Channel Binding

[RFC5056] defines a concept of channel binding to prevent man-in-the-middle attacks. It is common to provide SASL and GSS-API with another layer to provide transport security; Transport Layer Security (TLS) is the most common such layer. TLS provides its own server authentication. However there are a variety of situations where this authentication is not checked for policy or usability reasons. Even when it is checked, if the trust infrastructure behind the TLS authentication is different from the trust infrastructure behind the GSS-API mutual authentication then confirming the end-points using both trust infrastructures is likely to enhance security. If the endpoints of the GSS-API authentication are different than the endpoints of the lower layer, this is a strong indication of a problem such as a man-in-the-middle attack. Channel binding provides a facility to determine whether these endpoints are the same.

The GSS-EAP mechanism needs to support channel binding. When an application provides channel binding data, the mechanism needs to confirm this is the same on both sides consistent with the GSS-API specification.

Typically when considering channel binding, people think of channel binding in combination with mutual authentication. This is sufficiently common that without additional qualification channel binding should be assumed to imply mutual authentication. Without mutual authentication, only one party knows that the endpoints are correct. That's sometimes useful. Consider for example a user who wishes to access a protected resource from a shared whiteboard in a conference room. The whiteboard is the initiator; it does not need to actually authenticate that it is talking to the correct resource because the user will be able to recognize whether the displayed content is correct. If channel binding were used without mutual authentication, it would in effect be a request to only disclose the resource in the context of a particular channel. Such an authentication would be similar in concept to a holder-of-key SAML assertion. However, also note that while it is not happening in the protocol, mutual authentication is happening in the overall system: the user is able to visually authenticate the content. This is consistent with all uses of channel binding without protocol level mutual authentication found so far.

RFC 5056 channel binding (also called GSS-API channel binding when GSS-API is involved) is not the same thing as EAP channel binding. EAP channel binding is also used in the ABFAB context in order to implement acceptor naming and mutual authentication. Details are discussed in the mechanisms specification [I-D.ietf-abfab-gss-eap].

3.3. Host-Based Service Names

IETF security mechanisms typically take the name of a service entered by a user and make some trust decision about whether the remote party in an interaction is the intended party. GSS-API has a relatively flexible naming architecture. However most of the IETF applications that use GSS-API, including SSH, NFS, IMAP, LDAP and XMPP, have chosen to use host-based service names when they use GSS-API. In this model, the initiator names an acceptor based on a service such as "imap" or "host" (for login services such as SSH) and a host name.

Using host-based service names leads to a challenging trust delegation problem. Who is allowed to decide whether a particular hostname maps to an entity. The public-key infrastructure (PKI) used by the web has chosen to have a number of trust anchors (root certificate authorities) each of which can map any name to a public key. A number of GSS-API mechanisms, such as Kerberos [RFC1964], split the problem into two parts. A new concept called a realm is introduced. Then the mechanism decides what realm is responsible for a given name. That realm is responsible for deciding if the acceptor entity is allowed to claim the name. ABFAB needs to adopt this approach.

Host-based service names do not work ideally when different instances of a service are running on different ports. Also, these do not work ideally when SRV record or other insecure referrals are used.

The GSS-EAP mechanism needs to support host-based service names in order to work with existing IETF protocols.

3.4. Per-Message Tokens

GSS-API provides per-message security services that can provide confidentiality and integrity. Some IETF protocols such as NFS and SSH take advantage of these services. As a result GSS-EAP needs to support these services. As with mutual authentication, per-message services will limit the set of EAP methods that are available. Any EAP method that produces a Master Session Key (MSK) is able to support per-message security services described in [X].

GSS-API provides a pseudo-random function. While the pseudo-random function does not involve sending data over the wire, it provides an algorithm that both the initiator and acceptor can run in order to arrive at the same key value. This is useful for designs where a successful authentication is used to key some other function. This is similar in concept to the TLS extractor. No current IETF protocols require this. However GSS-EAP supports this service because it is valuable for the future and easy to do given per-message services. Non-IETF protocols are expected to take advantage of this in the near future.

4. Future Work: Attribute Providers

This architecture provides for a federated authentication and authorization framework between IdPs, RPs, principals, and subjects. It does not at this time provide for a means to retrieve attributes from 3rd parties. However, it envisions such a possibility. We note that in any extension to the model, an attribute provider must be authorized to release specific attributes to a specific RP for a specific principal. In addition, we note that it is an open question beyond this architecture as to how the RP should know to trust a particular attribute provider.

There are a number of possible technical means to provide attribute provider capabilities. One possible approach is for the IdP to provide a signed attribute request to RP that it in turn will provide to the attribute authority. Another approach would be for the IdP to provide a URI to the RP that contains a token of some form. The form of communications between the IdP and attribute provider as well as other considerations are left for the future. One thing we can say now is that the IdP would merely be asserting who the attribute authority is, and not the contents of what the attribute authority would return. (Otherwise, the IdP might as well make the query to the attribute authority and then resign it.)

5. Privacy Considerations

ABFAB, as an architecture designed to enable federated authentication and allow for the secure transmission of identity information between entities, obviously requires careful consideration around privacy and the potential for privacy violations.

This section examines the privacy related information presented in this document, summarising the entities that are involved in ABFAB communications and what exposure they have to identity information. In discussing these privacy considerations in this section, we use terminology and ideas from [I-D.iab-privacy-considerations].

Note that the ABFAB architecture uses at its core several existing technologies and protocols; detailed privacy discussion around these is not examined. This section instead focuses on privacy considerations specifically related to overall architecture and usage of ABFAB.

5.1. Entities and their roles

In an ABFAB environment, there are four distinct types of entities involved in communication paths. Figure 2 shows the ABFAB architecture with these entity types. We have:

• The client application: usually a piece of software running on a user's device. This communicates with a service (the Relying Party) that the user wishes to interact with.
• The Identity Provider: The home AAA server for the user.
• The Relying Party: The service the user wishes to connect to.
• The federation substrate: A set of entities through which messages pass on their path between RP and AAA server.

As described in detail earlier in this document, when a user wishes to access a Relying Party, a secure tunnel is set up between their client application and their Identity Provider (via the Relying Party and the federation substrate) through which credentials are exchanged. An indication of success or failure, alongside a set of AAA attributes about a principal is then passed from the Identity Provider to the Relying Party (usually in the form of a SAML assertion).

5.2. Relationship between user and entities

• Between User and Identity Provider - the identity Provider is an entity the user will have a direct relationship with, created when the organisation that operates the entity provisioned and exchanged the user's credentials. Privacy and data protection guarantees may form a part of this relationship.
• Between User and Relying Party - the Relying Party is an entity the user may or may not have a direct relationship with, depending on the service in question. Some services may only be offered to those users where such a direct relationship exists (for particularly sensitive services, for example), while some may not require this and would instead be satisfied with basic federation trust guarantees between themselves and the Identity Provider). This may well include the option that the user stays anonymous with respect to the Relying Party (though obviously not to the Identity Provider). If attempting to preserve privacy through the mitigation of data minimisation, then the only attribute information about individuals exposed to the Relying Party should be that which is strictly necessary for the operation of the service.
• Between User and Federation substrate - the user is highly likely to have no knowledge of, or relationship with, any entities involved with the federation substrate (not that the Identity Provider and/or Relying Party may, however). Knowledge of attribute information about individuals for these entities is not necessary, and thus such information should be protected in such a way as to prevent access to this information from being possible.

5.3. Data and Identifiers in use

In the ABFAB architecture, there are a few different types of data and identifiers in use.

5.3.1. NAI

In order for the Relying Party to be able to route messages to enable an EAP transaction to occur between client application and the correct identity Provider, it is necessary for the client application to provide enough information to the Relying Party to enable the identification of the correct Identity Provider. This takes the form of an Network Access Identifier (NAI) (as specified in [RFC4282]). Note that an NAI can have inner and outer forms in a AAA architecture.

• The outer part of NAI is exposed to the Relying Party; this can simply contain realm information. Doing so (i.e. not including user identification details such as a username) minimises the data given to the Relying Part to that which is purely necessary to support the necessary routing decision.
• The inner part of NAI is sent through the secure tunnel as established by the EAP protocol; this form of the NAI will contain credentials for the user suitable for authenticating them successfully (e.g. a username and password). Since the entire purpose of the secure tunnel is to protect communications between client application (EAP client) and Identity Provider (EAP server), then it is considered secure from eavesdroppers or malicious intermediaries and no further privacy discussion is necessary.

5.3.2. Identity Information

As a part of the ABFAB process, after a successful authentication has occurred between client application and Identity Provider, an indication of this success is sent to the Relying Party. Alongside this message, information about the user may be returned through AAA attributes, usually in form of a SAML assertion. This information is arbitrary and may include either only attributes that prevent an individual from being identified by the Relying Party (thus enabling anonymous or pseudonymous access) or attributes that contain personally identifiable information.

Depending on the method used, this information carried through AAA attributes may or may not be accessible to intermediaries involved in communications - e.g. in the case of RADIUS and unencrypted SAML, these headers are plain text and could be seen by any observer, whereas if using RADSEC or encrypted SAML, these headers are protected from observers. Obviously, where the protection of the privacy of an individual is required then this information needs to be protected by some appropriate means.

5.3.3. Accounting Information

Alongside the core authentication and authorization that occurs in AAA communications, accounting information about resource consumption may be delivered as part of the accounting exchange during the lifetime of the granted application session.

5.3.4. Collection and retention of data and identifiers

In cases where Relying Parties do not require to identify a particular individual when an individual wishes to make use of their service, the ABFAB architecture enable anonymous or pseudonymous access. Thus data and identifiers other than pseudonyms and unlinkable attribute information need not be stored and retained.

However, in cases where Relying Parties require the ability to identify a particular individual (e.g. so they can link this identity information to a particular account in their service, or where identity information is required for audit purposes), the service will need to collect and store such information, and to retain it for as long as they require. Deprovisioning of such accounts and information is out of scope for ABFAB, but obviously for privacy protection any identifiers collected should be deleted when they are no longer needed.

5.4. User Participation

In the ABFAB architecture, by its very nature users are active participants in the sharing of their identifiers as they initiate the communications exchange every time they wish to access a server. They are, however, not involved in control of the set of information related to them that transmitted from Identity Provider to Relying Party for authorisation purposes.

6. Deployment Considerations

6.1. EAP Channel Binding

Discuss the implications of needing EAP channel binding.

6.2. AAA Proxy Behavior

Discuss deployment implications of our proxy requirements.

7. Security Considerations

This document describes the architecture for Application Bridging for Federated Access Beyond Web (ABFAB) and security is therefore the main focus. This section highlights the main communication channels and their security properties:

Client-to-RP Channel:

The channel binding material is provided by any certificates and the final message (i.e., a cryptographic token for the channel). Authentication may be provided by the RP to the client but a deployment without authentication at the TLS layer is possible as well. In addition, there is a channel between the GSS requestor and the GSS acceptor, but the keying material is provided by a "third party" to both entities. The client can derive keying material locally, but the RP gets the material from the IdP. In the absence of a transport that provides encryption and/or integrity, the channel between the client and the RP has no ability to have any cryptographic protection until the EAP authentication has been completed and the MSK is transfered from the IdP to the RP.

RP-to-IdP Channel:

The security of this communication channel is mainly provided by the functionality offered via RADIUS and Diameter. At the time of writing there are no end-to-end security mechanisms standardized and thereby the architecture has to rely on hop-by-hop security with trusted AAA entities or, as an alternative but possible deployment variant, direct communication between the AAA client to the AAA server. Note that the authorization result the IdP provides to the RP in the form of a SAML assertion may, however, be protected such that the SAML related components are secured end-to-end.

The MSK is transported from the IdP to the RP over this channel. As no end-to-end security is provided by AAA, all AAA entities on the path between the RP and IdP have the ability to eavesdrop if no additional security measures are taken. One such measure is to use a transport between the client and the IdP that provides confidentiality.

Client-to-IdP Channel:

This communication interaction is accomplished with the help of EAP and EAP methods. The offered security protection will depend on the EAP method that is chosen but a minimum requirement is to offer mutual authentication, and key derivation. The IdP is responsible during this process to determine that the RP that is communication to the client over the RP-to-IdP channel is the same one talking to the IdP. This is accomplished via the EAP channel binding.

Partial list of issues to be addressed in this section: Privacy, SAML, Trust Anchors, EAP Algorithm Selection, Diameter/RADIUS/AAA Issues, Naming of Entities, Protection of passwords, Channel Binding, End-point-connections (TLS), Proxy problems

When a psuedonym is generated as a unique long term identifier for a Subject by an IdP, care MUST be taken in the algorithm that it cannot easily be reverse engineered by the service provider. If it can be reversed then the service provider can consult an oracle to determine if a given unique long term identifier is associated with a different known identifier.

8. IANA Considerations

This document does not require actions by IANA.

9. Acknowledgments

We would like to thank Mayutan Arumaithurai and Klaas Wierenga for their feedback. Additionally, we would like to thank Eve Maler, Nicolas Williams, Bob Morgan, Scott Cantor, Jim Fenton, Paul Leach, and Luke Howard for their feedback on the federation terminology question.

Furthermore, we would like to thank Klaas Wierenga for his review of the pre-00 draft version.

10. References

10.1. Normative References

[RFC2743]	Linn, J., "Generic Security Service Application Program Interface Version 2, Update 1", RFC 2743, January 2000.
[RFC2865]	Rigney, C., Willens, S., Rubens, A. and W. Simpson, "Remote Authentication Dial In User Service (RADIUS)", RFC 2865, June 2000.
[RFC3588]	Calhoun, P., Loughney, J., Guttman, E., Zorn, G. and J. Arkko, "Diameter Base Protocol", RFC 3588, September 2003.
[RFC3748]	Aboba, B., Blunk, L., Vollbrecht, J., Carlson, J. and H. Levkowetz, "Extensible Authentication Protocol (EAP)", RFC 3748, June 2004.
[RFC3579]	Aboba, B. and P. Calhoun, "RADIUS (Remote Authentication Dial In User Service) Support For Extensible Authentication Protocol (EAP)", RFC 3579, September 2003.
[RFC4072]	Eronen, P., Hiller, T. and G. Zorn, "Diameter Extensible Authentication Protocol (EAP) Application", RFC 4072, August 2005.
[RFC4282]	Aboba, B., Beadles, M., Arkko, J. and P. Eronen, "The Network Access Identifier", RFC 4282, December 2005.
[I-D.ietf-abfab-gss-eap]	Howlett, J and S Hartman, "A GSS-API Mechanism for the Extensible Authentication Protocol", Internet-Draft draft-ietf-abfab-gss-eap-04, October 2011.
[I-D.ietf-abfab-aaa-saml]	Howlett, J and S Hartman, "A RADIUS Attribute, Binding and Profiles for SAML", Internet-Draft draft-ietf-abfab-aaa-saml-03, March 2012.
[RFC6677]	Hartman, S., Clancy, T. and K. Hoeper, "Channel-Binding Support for Extensible Authentication Protocol (EAP) Methods", RFC 6677, July 2012.

10.2. Informative References

[RFC2903]	de Laat, C., Gross, G., Gommans, L., Vollbrecht, J. and D. Spence, "Generic AAA Architecture", RFC 2903, August 2000.
[I-D.nir-tls-eap]	Nir, Y, Sheffer, Y, Tschofenig, H and P Gutmann, "A Flexible Authentication Framework for the Transport Layer Security (TLS) Protocol using the Extensible Authentication Protocol (EAP)", Internet-Draft draft-nir-tls-eap-13, December 2011.
[I-D.ietf-oauth-v2]	Hammer-Lahav, E, Recordon, D and D Hardt, "The OAuth 2.0 Authorization Protocol", Internet-Draft draft-ietf-oauth-v2-25, March 2012.
[I-D.iab-privacy-considerations]	Cooper, A, Tschofenig, H, Aboba, B, Peterson, J and J Morris, "Privacy Considerations for Internet Protocols", Internet-Draft draft-iab-privacy-considerations-01, October 2011.
[RFC4017]	Stanley, D., Walker, J. and B. Aboba, "Extensible Authentication Protocol (EAP) Method Requirements for Wireless LANs", RFC 4017, March 2005.
[RFC5106]	Tschofenig, H., Kroeselberg, D., Pashalidis, A., Ohba, Y. and F. Bersani, "The Extensible Authentication Protocol-Internet Key Exchange Protocol version 2 (EAP-IKEv2) Method", RFC 5106, February 2008.
[RFC1964]	Linn, J., "The Kerberos Version 5 GSS-API Mechanism", RFC 1964, June 1996.
[RFC2203]	Eisler, M., Chiu, A. and L. Ling, "RPCSEC_GSS Protocol Specification", RFC 2203, September 1997.
[RFC3645]	Kwan, S., Garg, P., Gilroy, J., Esibov, L., Westhead, J. and R. Hall, "Generic Security Service Algorithm for Secret Key Transaction Authentication for DNS (GSS-TSIG)", RFC 3645, October 2003.
[RFC2138]	Rigney, C., Rigney, C., Rubens, A.C., Simpson, W.A. and S. Willens, "Remote Authentication Dial In User Service (RADIUS)", RFC 2138, April 1997.
[RFC4462]	Hutzelman, J., Salowey, J., Galbraith, J. and V. Welch, "Generic Security Service Application Program Interface (GSS-API) Authentication and Key Exchange for the Secure Shell (SSH) Protocol", RFC 4462, May 2006.
[RFC4422]	Melnikov, A. and K. Zeilenga, "Simple Authentication and Security Layer (SASL)", RFC 4422, June 2006.
[RFC5056]	Williams, N., "On the Use of Channel Bindings to Secure Channels", RFC 5056, November 2007.
[RFC5801]	Josefsson, S. and N. Williams, "Using Generic Security Service Application Program Interface (GSS-API) Mechanisms in Simple Authentication and Security Layer (SASL): The GS2 Mechanism Family", RFC 5801, July 2010.
[RFC5849]	Hammer-Lahav, E., "The OAuth 1.0 Protocol", RFC 5849, April 2010.
[OASIS.saml-core-2.0-os]	Cantor, S., Kemp, J., Philpott, R. and E. Maler, "Assertions and Protocol for the OASIS Security Assertion Markup Language (SAML) V2.0", OASIS Standard saml-core-2.0-os, March 2005.
[RFC2904]	Vollbrecht, J., Calhoun, P., Farrell, S., Gommans, L., Gross, G., de Bruijn, B., de Laat, C., Holdrege, M. and D. Spence, "AAA Authorization Framework", RFC 2904, August 2000.
[I-D.hartman-emu-mutual-crypto-bind]	Hartman, S, Wasserman, M and D Zhang, "EAP Mutual Cryptographic Binding", Internet-Draft draft-hartman-emu-mutual-crypto-bind-00, March 2012.
[I-D.ietf-emu-eap-tunnel-method]	Zhou, H, Cam-Winget, N, Salowey, J and S Hanna, "Tunnel EAP Method (TEAP) Version 1", Internet-Draft draft-ietf-emu-eap-tunnel-method-01, October 2011.
[WS-TRUST]	Lawrence, K, Kaler, C, Nadalin, A, Goodner, M, Gudgin, M, Barbir, A and H Granqvist, "WS-Trust 1.4", OASIS Standard ws-trust-200902, February 2009.

Authors' Addresses