| Network Working Group | J. Richer, Ed. |
| Internet-Draft | Bespoke Engineering |
| Intended status: Standards Track | L. Johansson |
| Expires: September 19, 2018 | Swedish University Network |
| March 18, 2018 |
Vectors of Trust
draft-richer-vectors-of-trust-08
This document defines a mechanism for describing and signaling several aspects that are used to calculate trust placed in a digital identity transaction.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 RFC 2119 RFC 8174 when, and only when, they appear in all capitals, as shown here.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on September 19, 2018.
Copyright (c) 2018 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Methods for measuring trust in digital identity transactions have historically fallen into two main categories: either all measurements are combined into a single scalar value, or trust decisions are calculated locally based on a detailed set of attribute metadata. This document defines a method of conveying trust information that is more expressive than a single value but less complex than comprehensive attribute metadata.
Prior to the third edition published in 2017, NIST Special Publication 800-63 used a single scalar measurement of trust called a Level of Assurance (LoA). An LoA can be used to compare different transactions within a system at a coarse level. For instance, an LoA4 transaction is generally considered more trusted (across all measured categories) than an LoA2 transaction. The LoA for a given transaction is computed by the identity provider (IdP) and is consumed by a relying party (RP). Since the trust measurement is a simple numeric value, it's trivial for RPs to process and compare. However, since each LoA encompasses many different aspects of a transaction, it can't express many real-world situations. For instance, an anonymous user account might have a very strong credential, such as would be common of a whistle-blower or political dissident. Despite the strong credential, the lack of identity proofing would make any transactions conducted by the account to fall into a low LoA. Furthermore, different use cases and domains require subtly different definitions for their LoA categories, and one group's LoA2 is not equivalent or even comparable to another group's LoA2.
Attribute based access control (ABAC) systems used by RPs may need to know details about a user's attributes, such as how recently the attribute data was verified and by whom. Attribute metadata systems are capable of expressing extremely fine-grained detail about the transaction. However, this approach requires the IdP to collect, store, and transmit all of this attribute data for the RP's consumption. The RP must process this data, which may be prohibitive for trivial security decisions.
Vectors of Trust (VoT) seeks a balance between these two alternatives by allowing expression of multiple aspects of an identity transaction (including but not limited to identity proofing, credential strength, credential management, and assertion strength), without requiring full attribute metadata descriptions. This method of measurement gives more actionable data and expressiveness than an LoA, but is still relatively easy for the RP to process. It is anticipated that VoT can be used alongside more detailed attribute metadata systems as has that proposed by NISITIR 8112. The RP can use the vector value for most basic decisions but be able to query the IdP for additional attribute metadata where needed. Furthermore, it is anticipated that some trust frameworks will provide a simple mapping between certain sets of vector values to LoAs, for RPs that do not have a need for the vector's more fine-grained detail. In such systems, an RP is given a choice of how much detail it needs in order to process a given request.
This document defines a data model for these vectors and an on-the-wire format for conveying them between parties, anchored in a trust definition. This document also provides guidance for defining values for use in conveying this information, including four component categories and guidance on defining values within those categories. Additionally, this document defines a general-purpose set of component values in an appendix for use cases that do not need something more specific.
This document assumes the following model for identity based on identity federation technologies:
The identity subject (also known as the user) is associated with an identity provider which acts as a trusted third party on behalf of the user with regard to a relying party by making identity assertions about the user to the relying party.
The real-world person represented by the identity subject is in possession of a primary credential bound to the identity subject by the identity provider (or an agent thereof) in such a way that the binding between the credential and the real-world user is a representation of the identity proofing process performed by the identity provider (or an agent thereof) to verify the identity of the real-world person. This is all carried by an identity assertion across the network to the relying party during the authentication transaction.
The term Vectors of Trust is based on the mathematical construct of a vector, which is defined as an item composed of multiple independent values.
An important goal for this work is to balance the need for simplicity (particularly on the part of the relying party) with the need for expressiveness. As such, this vector construct is designed to be composable and extensible.
All components of the vector construct MUST be orthogonal such that no aspect of a component overlaps an aspect of another component, as much as is possible.
This specification defines four orthogonal components: identity proofing, primary credential usage, primary credential management, and assertion presentation. These dimensions MUST be evaluated by the RP in the context of a trust framework and SHOULD be combined with other information when making a trust and authorization decision.
This specification also defines values for each component to be used in the absence of a more specific trust framework in Appendix A. It is expected that trust frameworks will provide context, semantics, and mapping to legal statutes and business rules for each value in each component. Consequently, a particular vector value can only be compared with vectors defined in the same context. The RP MUST understand and take into account the trust framework context in which a vector is being expressed in order for to process a vector securely.
Each component is identified by a demarcator consisting of a single uppercase ASCII letter in the range [A-Z]. The demarcator SHOULD reflect the category with which it is associated in a natural manner. Demarcators for components MUST be registered as described in Section 9. It is anticipated that trust framework definitions will use this registry to define specialized components, though it is RECOMMENDED that trust frameworks re-use existing components wherever possible.
The value for a given component within a vector of trust is defined by its demarcator character followed by a single digit or lowercase ASCII letter in the range [0-9a-z]. Categories which have a natural ordering SHOULD use digits, with 0 as the lowest value. Categories which do not have a natural ordering, or which can have an ambiguous ordering, SHOULD use letters. Categories MAY use both letter style and number style value indicators simultaneously. For example, a category could define 0 as a special "empty" value while using letters such as a, b, c for normal values can to differentiate between these types of options. Another system could have a base category with a numeric value with additonal details provided by letter values.
Regardless of the type of value indicator used, the values assigned to each component of a vector MUST NOT be assumed always to have inherent ordinal properties when compared to the same or other components in the vector space. In other words, 1 is different from 2, but it is dangerous to assume that 2 is always better than 1 in a given transaction.
Each component MAY be repeated with multiple different values within a single vector. The same component and value combination MUST NOT be repeated within a single vector.
The Identity Proofing dimension defines, overall, how strongly the set of identity attributes have been verified and vetted. In other words, this dimension describes how likely it is that a given digital identity transaction corresponds to a particular (real-world) identity subject.
This dimension SHALL be represented by the P demarcator and a single-character level value, such as P0, P1, etc. Most definitions of identity proofing will have a natural ordering, as more or less stringent proofing can be applied to an individual. In such cases it is RECOMMENDED that a digit style value be used for this component and that only a single value be allowed to be communicated in a transaction.
The primary credential usage dimension defines how strongly the primary credential can be verified by the IdP. In other words, how easily that credential could be spoofed or stolen.
This dimension SHALL be represented by the C demarcator and a single-character level value, such as Ca, Cb, etc. Most definitions of credential usage will not have an overall natural ordering, as there may be several equivalent classes described within a trust framework. In such cases it is RECOMMENDED that a letter style value be used for this component and that multiple distinct credential usage factors be allowed to be communicated simultaneously, such as when Multi-Factor Authentication is used.
The primary credential management dimension conveys information about the expected lifecycle of the primary credential in use, including its binding, rotation, and revocation. In other words, the use and strength of policies, practices, and security controls used in managing the credential at the IdP and its binding to the intended individual.
This dimension SHALL be represented by the M demarcator and a single-character level value, such as Ma, Mb, etc. Most definitions of credential management will not have an overall natural ordering, though there can be preference and comparison between values in some circumstances. In such cases it is RECOMMENDED that a letter style value be used for this component and that multiple distinct values be allowed to be communicated simultaneously.
The Assertion Presentation dimension defines how well the given digital identity can be communicated across the network without information leaking to unintended parties, and without spoofing. In other words, this dimension describes how likely it is that a given digital identity was actually asserted by a given identity provider for a given transaction. While this information is largely already known by the RP as a side effect of processing an identity assertion, this dimension is still very useful when the RP requests a login and when describing the capabilities of an IdP.
This dimension SHALL be represented by the A demarcator and a level value, such as Aa, Ab, etc. Most definitions of assertion presentation will not have an overall natural ordering. In such cases, it is RECOMMENDED that a letter style value be used for this component and that multiple values be allowed to be communicated simultaneously.
A vector of trust is designed to be used in the context of an identity and authentication transaction, providing information about the context of a federated credential. The vector therefore needs to be able to be communicated in the context of the federated credential in a way that is strongly bound to the assertion representing the federated credential.
This vector has several requirements for use.
These requirements lead us to defining a simple string-based representation of the vector that can be incorporated within a number of different locations and protocols without further encoding.
The vector MUST be represented as a period-separated ('.') list of vector components, with no specific order. A vector component type MAY occur multiple times within a single vector, with each component separated by periods. Multiple values for a component are considered a logical AND of the values. A specific value of a vector component MUST NOT occur more than once in a single vector. That is, while Cc.Cd is a valid vector, Cc.Cc is not.
Vector components MAY be omitted from a vector. No holding space is left for an omitted vector component. If a vector component is omitted, the vector is making no claim for that component. This MAY be distinct from a specific component value stating that a component was not used.
Vector values MUST be communicated along side of a trustmark definition to give the components context. The trustmark MUST be cryptographically bound to the vector value, such as in a signed assertion. A vector value without context is unprocessable, and vectors defined in different contexts are not directly comparable as whole values. Different trustmarks MAY re-use component definitions (including their values), allowing comparison of individual components across contexts without requiring complete understanding of all aspects of a context. The proper processing of such cross-context values is outside the scope of this specification.
For example, the vector value P1.Cc.Ab translates to "pseudonymous, proof of shared key, signed browser-passed verified assertion, and no claim made toward credential management" in the context of this specification's definitions. The vector value of Cb.Mc.Cd.Ac translates to "known device, full proofing require for issuance and rotation, cryptographic proof of possession of a shared key, signed back-channel verified assertion, and no claim made toward identity proofing" in the same context.
In OpenID Connect, the IdP MUST send the vector as a string within the vot (vector of trust) claim in the ID token. The trustmark that applies to this vector MUST be sent as an HTTPS URL in the vtm (vector trust mark) claim to provide context to the vector.
For example, the body of an ID token claiming "pseudonymous, proof of shared key, signed back-channel verified token, and no claim made toward credential management" could look like this JSON object payload of the ID token.
{
"iss": "https://idp.example.com/",
"sub": "jondoe1234",
"vot": "P1.Cc.Ac",
"vtm": "https://trustmark.example.org/trustmark/idp.example.com"
}
The body of the ID token is signed and optionally encrypted using JOSE, as per the OpenID Connect specification. By putting the vot and vtm values inside the ID token, the vector and its context are strongly bound to the federated credential represented by the ID token.
In SAML, a vector is communicated as an AuthenticationContextDeclRef. A vector is represented by prefixing it with the urn urn:ietf:param:[TBD] to form a full URN. The AuthenticationContextDeclaration corresponding to a given vector is a AuthenticationContextDeclaration element containing an Extension element with components of the vector represented by the following XML schema:
<?xml version="1.0" encoding="UTF-8"?>
<xs:schema
targetNamespace="urn:ietf:param:[TBD]:schema"
xmlns:xs="http://www.w3.org/2001/XMLSchema"
<xs:element name="Vector">
<xs:annotation>
<xs:documentation>This represents a set of
vector components.</xs:documentation>
</xs:annotation>
<xs:simpleType>
<xs:restriction base="xsd:token">
<xs:pattern value="([A-Z][a-z0-9])(\.[A-Z][a-z0-9])*"/>
</xs:restriction>
</xs:simpleType>
</xs:element>
</xs:schema>
For instance the vector P1.Cc.Ac is represented by the AuthenticationContextDeclRef URN urn:ietf:param:[TBD]:P1.Cc.Ac (or urn:ietf:param:[TBD]:Cc.P1.Ac or ...) which corresponds to the following AuthenticationContextDeclaration:
<?xml version="1.0" encoding="UTF-8"?>
<AuthenticationContextDeclaration
xmlns="urn:oasis:names:tc:SAML:2.0:ac">
<Extension>
<vot:Vector>P1.Cc.Ac</vot:Vector>
</Extension>
</AuthenticationContextDeclaration>
In some identity protocols, the RP can request that particular vector components be applied to a given identity transaction. Using the same syntax as defined in Section 3.1, an RP can indicate that it desires particular aspects be present in the authentication. Processing and fulfillment of these requests are in the purview of the IdP and details are outside the scope of this specification.
In OpenID Connect, the client MAY request a set of acceptable VoT values with the vtr (vector of trust request) claim request as part of the Request Object. The value of this field is an array of JSON strings, each string identifying an acceptable set of vector components. The component values within each vector are ANDed together while the separate vectors are ORed together. For example, a list of vectors in the form ["P1.Cb.Cc.Ab", "Ce.Ab"] is stating that either the full set of "P1 AND Cb AND Cc AND Ab" simultaneously OR the full set of "Ce AND Ab" simultaneously are acceptable to this RP for this transaction.
Vector request values MAY omit components, indicating that any value is acceptable for that component category, including omission of that component in the response vector.
The mechanism by which the IdP processes the vtr and maps that to the authentication transaction are out of scope of this specification.
In SAML the client can request a set of acceptable VoT values by including the corresponding AuthenticationContextDeclRef URIs together with an AuthenticationContextClassRef corresponding to the trust mark (cf below). The processing rules in Section 3.3 apply.
When an RP receives a specific vector from an IdP, it needs to make a decision to trust the vector within a specific context. A trust framework can provide such a context, allowing legal and business rules to give weight to an IdP's claims. A trustmark is a verifiable claim to conform to a specific component of a trust framework, such as a verified identity provider. The trustmark conveys the root of trustworthiness about the claims and assertions made by the IdP, including the vector itself.
The trustmark MUST be available from an HTTPS URL served by the trust framework provider. The contents of this URL are a JSON document with the following fields:
Vector component values offered by this IdP are be listed in a using their demarcator. For the four component categories defined in this specification:
For example, the following trustmark provided by the trustmark.example.org organization applies to the idp.example.org identity provider:
{
"idp": "https://idp.example.org/",
"trustmark_provider": "https://trustmark.example.org/",
"P": ["P0", "P1"],
"C": ["C0", "Ca", "Cb"],
"M": ["Mb"],
"A": ["Ab", "Ac"]
}
An RP wishing to check the claims made by an IdP can fetch the information from the trustmark provider about what claims the IdP is allowed to make in the first place and process them accordingly. The RP MAY cache the information returned from the trustmark URL.
The operational aspects of the IdP MAY be included in the trustmark definition. For example, if a trustmark can indicate that an IdP uses multiple redundant hosts, encrypts all data at rest, or other operational security mechanisms that affect the trustworthiness of assertions made by the IdP. The definition of these additional aspects is outside the scope of this specfication.
The means by which the RP decides which trustmark providers it trusts is out of scope for this specification and is generally configured out of band.
Though most trust frameworks will provide a third-party independent verification service for components, an IdP MAY host its own trustmark. For example, a self-hosted trustmark would look like:
{
"idp": "https://idp.example.org/",
"trustmark_provider": "https://idp.example.org/",
"P": ["P0", "P1"],
"C": ["C0", "Ca", "Cb"],
"M": ["Mb"],
"A": ["Ab", "Ac"]
}
The IdP MAY list all of its available trustmarks as part of its discovery document, such as the OpenID Connect Discovery server configuration document. In this context, trustmarks are listed in the trustmarks element which contains a single JSON object. The keys of this JSON object are trustmark provider issuer URLs and the values of this object are the corresponding trustmark URLs for this IdP.
{
"iss": "https://idp.example.org/",
"trustmarks": {
"https://trustmark.example.org/":
"https://trustmark.example.org/trustmark/idp.example.org",
"https://trust.example.net/":
"https://trust.example.net/trustmark/idp.example.org"
}
}
Vectors of Trust is meant to be a flexible and reusable framework for communicating authentication data between networked parties in an identity federation protocol. However, the exact nature of the information needed is reliant on the parties requiring the information and the relationship between them. While this document does define a usable default set of values in Appendix A, it is anticipated that many situations will require an extension of this specification for their own use.
Components categories such as those defined in Section 2 are intended to be general purpose and reusable in a variety of circumstances. Extension specifications SHOULD re-use existing category definitions where possible. Extensions MAY create additional categories where needed by using the registry defined in Section 9. The registry encourages re-use and discovery of existing categories across different implementations. In other words, the P category in another framework SHOULD be used for identity proofing and related information.
The values of components such as those defined in Appendix A are intended to be contextual to the defining trust document. While this specification's component values are intended to be general-purpose and extensions MAY re-use the values and their definitions, extensions MUST define all allowable values. As these values are always interpreted in the context of a trustmark, these values are not recorded in a central registry. Consequently, a P1 value from one framework and a P1 value from another framework could have very different interpretations depending on their contextual trustmark documents.
Extensions to this specification SHOULD choose either a numerical ordering or a group category approach to component values as described in Section 2, though combinations of both types MAY be used. Extensions to this specification MUST specify whether multiple values are allowed for each category, and while any component category is generally allowed to have multiple distinct values, a specific definition of a set of values in an extension MAY limit a given component category to a single value per transaction.
The authors would like to thank the members of the Vectors of Trust mailing list in the IETF for discussion and feedback on the concept and document, and the members of the ISOC Trust and Identity team for their support.
This specification creates one registry and registers several values into existing registries.
This specification establishes the Vectors of Trust Components Registry.
Component demarcators are registered by Specification Required after a two-week review period on the vot@ietf.org mailing list, on the advice of one or more Designated Experts.
Criteria that should be applied by the Designated Experts includes determining whether the proposed registration duplicates existing functionality, whether it is likely to be of general applicability or whether it is useful only for a single application, and whether the registration description is clear.
Registration requests sent to the mailing list for review should use an appropriate subject (e.g., "Request to register Vector of Trust Component name: example"). Within the review period, the Designated Expert(s) will either approve or deny the registration request, communicating this decision to the review list and IANA. Denials should include an explanation and, if applicable, suggestions as to how to make the request successful. IANA must only accept registry updates from the Designated Expert(s) and should direct all requests for registration to the review mailing list.
The Vector of Trust Components Registry contains the definitions of vector components and their associated demarcators.
This specification adds the following values to the OAuth Parameters Registry established by [RFC6749].
This specification adds the following values to the JSON Web Token Claims Registry established by [RFC7519].
The vector of trust value MUST be cryptographically protected in transit, using TLS as described in [BCP195]. The vector of trust value must be associated with a trustmark marker, and the two must be carried together in a cryptographically bound mechanism such as a signed identity assertion. A signed OpenID Connect ID Token and a signed SAML assertion both fulfil this requirement.
The vector value is always associated with a trustmark and needs to be interpreted by the RP in the context of that trustmark. Different trust frameworks can apply different interpretations to the same component value, much as was the case with LoA. Therefore, an RP interpreting a component value in the the wrong context could mistakenly accept or reject a request. In order to avoid this mistake, RPs need to reject vectors that are defined in trust frameworks that they do not understand how to interpret properly.
The VoT framework provides a mechanism for describing and conveying trust information. It does not define any policies for asserting the values of the vector, nor does it define any policies for applying the values of a vector to an RP's security decision process. These policies must be agreed upon by the IdP and RP, and they should be expressed in detail in an associated human-readable trust framework document.
By design, vector of trust values contain information about the user's authentication and associations that can be made thereto. Therefore, all aspects of a vector of trust contain potentially privacy-sensitive information and must be guarded as such. Even in the absence of specific attributes about a user, knowledge that the user has been highly proofed or issued a strong token could provide more information about the user than was intended. It is recommended that systems in general use the minimum vectors applicable to their use case in order to prevent inadvertent information disclosure.
| [OpenID] | Sakimura, N., Bradley, J. and M. Jones, "OpenID Connect Core 1.0", November 2014. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC6749] | Hardt, D., "The OAuth 2.0 Authorization Framework", RFC 6749, DOI 10.17487/RFC6749, October 2012. |
| [RFC7519] | Jones, M., Bradley, J. and N. Sakimura, "JSON Web Token (JWT)", RFC 7519, DOI 10.17487/RFC7519, May 2015. |
| [RFC8126] | Cotton, M., Leiba, B. and T. Narten, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 8126, DOI 10.17487/RFC8126, June 2017. |
| [RFC8174] | Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017. |
| [RFC8259] | Bray, T., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, December 2017. |
| [BCP195] | Sheffer, Y., Holz, R. and P. Saint-Andre, "Recommendations for Secure Use of Transport Layer Security (TLS) and Datagram Transport Layer Security (DTLS)", BCP 195, RFC 7525, DOI 10.17487/RFC7525, May 2015. |
| [NISTIR-8112] | National Institute of Standards and Technology, U.S. Department of Commerce, "A Proposed Schema for Evaluating Federated Attributes", NIST NISTIR 8112, January 2018. |
| [SP-800-63-2] | National Institute of Standards and Technology, U.S. Department of Commerce, "Electronic Authentication Guideline", NIST SP 800-63-2, DOI 10.6028/NIST.SP.800-63-2, August 2013. |
| [SP-800-63-3] | National Institute of Standards and Technology, U.S. Department of Commerce, "Digital Identity Guideline", NIST SP 800-63-3, DOI 10.6028/NIST.SP.800-63-3, June 2017. |
The following general-purpose component definitions MAY be used when a more specific set is unavailable. These component values are referenced in a trustmark definition defined by [[ this document URL ]].
Extensions of this specification SHOULD define their own component values as described in Section 7. Where possible, extensions MAY re-use specific values here.
The identity proofing component of this vector definition represents increasing scrutiny during the proofing process. Higher levels are largely subsumptive of lower levels, such that P2 fulfills requirements for P1, etc. Mutltiple distinct values from this category MUST NOT be used in a single transaction.
The primary credential usage component of this vector definition represents distinct categories of primary credential that MAY be used together in a single transaction. Multiple distinct values from this category MAY be used in a single transaction.
The primary credential management component of this vector definition represents distinct categories of management that MAY be considered separately or together in a single transaction. Many trust framework deployments MAY use a single value for this component as a baseline for all transactions and thereby omit it. Multiple distinct values from this category MAY be used in a single transaction.
The assertion presentation component of this vector definition represents distinct categories of assertion which are RECOMMENDED to be used in a subsumptive manner but MAY be used together. Multiple distinct values from this category MAY be used in a single transaction.
-08
-07
-06
-05
-04
-03
-02
- 01
- 00