Abstract 

This document contains non-normative release notes produced by the User-Managed Access Work Group explaining how new versions of the UMA specifications differ from previous ones.

Status

This document includes release notes for all versions of UMA.

Editor

• Eve Maler

Intellectual Property Notice

The User-Managed Access Work Group operates under Kantara IPR Policy - Option Patent & Copyright: Reciprocal Royalty Free with Opt-Out to Reasonable And Non discriminatory (RAND) (HTML version) and the publication of this document is governed by the policies outlined in this option.

The content of this document is copyright of Kantara Initiative. © 2018 Kantara Initiative

Table of Contents 

Introduction

This document contains non-normative release notes produced by the User-Managed Access Work Group explaining how new versions of the UMA specifications differ from previous ones.

NOTE: Reading the release notes is not a substitute for reading the specifications carefully. In each specification release, much work is typically done to improve clarity and applicability for implementers and others. See the UMA Implementer's Guide for additional commentary.

The UMA specifications use Semantic Versioning:

1. MAJOR version when you make incompatible API changes,

2. MINOR version when you add functionality in a backwards-compatible manner, and

3. PATCH version when you make backwards-compatible bug fixes.

The following shorthand terms and abbreviations are used in this document (see also the terminology, including abbreviations, defined in the specifications):

• AS: authorization server

• RS: resource server

• Core: UMA Core specification (applies to versions 1.0 and 1.0.1)

• RSR: OAuth Resource Set Registration specification (applies to versions 1.0 and 1.0.1)

• Grant: UMA Grant for OAuth Authorization (applies to version 2.0)

• FedAuthz: Federated Authorization for UMA (applies to version 2.0)

• I-D: IETF Internet-Draft specification

• Sec: section

Where a change relates to a GitHub issue, the linked issue number is provided.

From UMA1 to UMA V2.0

The UMA V2.0 Recommendations are User-Managed Access (UMA) 2.0 Grant for OAuth 2.0 Authorization (known as "Grant") and Federated Authorization for User-Managed Access (UMA) 2.0 (known as "FedAuthz"). The official versions are downloadable from the Kantara Reports & Recommendations page; this document links to specific sections within the HTML versions.

Differences and changes noted are between V2.0 and V1.0.n generally; note that internal revision differences between UMA2 revisions are not tracked here. (You may find it helpful to refer to the Disposition of Comments document, a record of specification changes during the Public Comment periods late in their final review cycle, and the GitHub repository where the specifications are managed.) Where the distinction between V1.0 and V1.0.1 is important, it will be noted; otherwise the label "UMA1" is used.

The following sequence diagrams may be of assistance as brief summaries of changes made:

• Sequence diagram for Grant, highlighting key changes from UMA1

• Sequence diagram for FedAuthz, highlighting key changes from UMA1

Version Themes

The major themes of this version, as determined by the Work Group's 2016 roadmap planning process, were (along with constantly improving security) to:

• Increase OAuth 2.0 alignment

• Improve Internet of Things readiness

• Improve readiness for "wide ecosystems", where the requesting party and the resource owner's AS have no pre-established relationship

Specification Reorganization and Conformance Levels

The two specifications were divided differently until late April 2017. Core and RSR were recombined into Grant and FedAuthz, as follows:

• All communications of the client and requesting party with the AS appear in Grant. This specification formally defines an extension OAuth grant.

• All communications of the resource owner and resource server with the AS appear in FedAuthz. This includes:

  • Policy setting (outside the scope of UMA)

  • PAT definition and issuance

  • Protection API

    • Resource registration (previously, RSR specified only this endpoint/API and Core specified everything else)

    • The RS's permission requests at the AS

    • The RS's token introspection at the AS

• The formal profiles for API extensibility URIs https://docs.kantarainitiative.org/uma/profiles/prot-ext-1.0, https://docs.kantarainitiative.org/uma/profiles/authz-ext-1.0, and https://docs.kantarainitiative.org/uma/profiles/rsrc-ext-1.0 were removed and replaced with recommendations (Grant Sec 4 and FedAuthz Sec 1.3) to define profiles as needed and to use uma_profiles_supported metadata (Grant Sec 2) to declare them.

It is now optional to implement the features appearing in FedAuthz; thus, this specification effectively defines a conformance level. (Note: To receive the full benefits of "user-managed access", it is best to implement and use the features of both specifications.)

Terminology Changes

Note the following terminology changes made throughout the specifications. (256) See also Summary of API and Endpoint Changes below for naming changes made to some of the endpoints.

API and Endpoint Changes

These design changes include naming changes made to some of the endpoints.

Authorization Server Discovery Document and Metadata Changes

Discovery Document and Metadata Simplification

UMA1's endpoint and feature discovery mechanism was defined in total by its Core specification. UMA2 makes use of the OAuth Authorization Server Discovery mechanism instead (still in Internet-Draft form at the time of UMA2 publication), eliminating metadata fields already defined by the OAuth discovery or OpenID Connect specification. The Grant (Sec 2) and FedAuthz (Sec 2) specifications each define only the metadata fields they require. (59, 157, 159, 305)

Definition of OAuth Dynamic Client Registration Metadata Field

The new metadata field claims_redirect_uris enables the client to pre-register claims redirection URIs. (Grant Sec 2, Sec 3.3.2, Sec 7.3) (337 sub-issues c and d)

permissions Claim and Sub-Claims in Token Introspection Object Not Requested to Be IANA-Registered as JWT Claims

Previously, it was intended to make an IANA registration request of the claims inside the introspection object as independent JWT claims. This would enable them to be formally used in RPTs, such that an RS can validate the access token locally with these claims packed inside it. Because of potential security and privacy considerations, it was determined not to define this token format for now. (FedAuthz Sec 9) (334)

Changes to AS-Client, RS-Client, and AS-Requesting Party Interfaces (Now UMA Grant Specification)

Authorization Server Rotates Permission Ticket

After the AS initially generates the permission ticket and the RS conveys it to the client, whenever the client subsequently approaches the AS token endpoint or redirects the requesting party to the AS claims gathering endpoint, the AS is required to rotate the value of the permission ticket every time it hands a permission ticket value back to the client (Grant Sec 3.3.3, Sec 3.3.6). This action obsoletes the need for the UMA Claims-Gathering Extension for Enhanced Security specification (see this explanation of that specification for more information).

Token Endpoint Replaces RPT Endpoint; Client-Side Communications Defined as Extension Grant

The specialized RPT endpoint was removed in favor of using the standard OAuth token endpoint (Grant Sec 3.3.1). A formal extension OAuth grant was defined (same section), working with regular OAuth capabilities and OAuth error codes to the extent possible (Sec 3.3.6). This enabled reuse of large portions of the threat model and the client type model, along with the ability for the client to request scopes and to authenticate using its own client credentials at the token endpoint (see the next section for additional discussion). (153, 165)

AAT Removed in Favor of PCT

An end-user requesting party no longer needs to mediate issuance of an AAT at the AS, and the client no longer needs to use an AAT in order to request a token; it simply uses its own client credentials at the OAuth token endpoint as in a normal grant (see Token Endpoint Replaces RPT Endpoint and Client-Side Communications Defined as Grant). Thus, the first time the requesting party needs to interact with the AS, if at all, is to provide claims interactively when redirected by the client as part of claims collection. This is in contrast to UMA1, where an end-user requesting party would have been expected to engage in an interactive OAuth flow to log in and then authorize AAT issuance at the AS's authorization endpoint. In UMA1, the (required) AAT could have been used by the AS as a reminder of claims about the current requesting party. In UMA2, the (optional) PCT is available to serve in this capacity instead, without the OAuth mechanism being involved (Grant Sec 3.3.1). Note that UMA2 does not require the AS to involve the requesting party in an interactive flow authorizing PCT issuance (Grant Sec 3.3.3). (154, 264)

Deprecated Response-Body Permission Ticket Return Option By RS Removed

In UMA V1.0.1 the RS was able to return the initial permission ticket to the client in the response body for backwards compatibility with UMA V1.0, but this option was deprecated; now this option has been removed. (233)

Permission Ticket Return By AS With Redirect-User Hint No Longer Deprecated

In UMA V1.0.1 the AS was able to return the permission ticket to the client along with the redirect_user hint, but the client was not supposed to depend on ticket accuracy, and the supply of this ticket was deprecated. Now all permission tickets directly supplied by the AS are rotated and the value is safe for the client to depend on (Grant Sec 3.3.6). (233)

More Discretionary Permission Requests

The instruction for the RS to request permissions on the client's behalf (which can be a private interface or the standardized interface governed by FedAuthz) is now defined as a recommendation ("SHOULD") to be reasonable for the client's resource request, rather than being required to meet it ("minimally suffices"). The UMA Implementer's Guide has a section on Considerations Regarding Resource Server Permission Requests that explains how and why this level of discretion is more appropriate.

need_info Response Structured Flattened

The JSON nested object structure of the need_info error response from the AS has been flattened. Now it directly contains a permission ticket and either a required_claims or a redirect_user hint (or both) (Grant Sec 3.3.6). (237, 308)

not_authorized Error Renamed to request_denied

The UMA1 error not_authorized has been renamed to request_denied. Note that this error was re-added only in a later revision of UMA2. See the UMA Implementer's Guide section called Understanding Authorization Server Response Options From the Token Endpoint to understand AS error semantics. (Grant Sec 3.3.6) (340)

Added interval parameter to request_submitted Error

An optional interval parameter was added to the request_submitted error to enable the AS to inform the client about appropriate polling intervals. (Grant Sec 3.3.6) (341)

New Refresh Token Clarity

It has been clarified that the AS can issue a refresh token and the client can use the refresh token grant to attempt to get a new RPT with it (Grant Sec 3.3.5, Sec 3.6). (238, 284)

Authorization Assessment Gains Precision

Inputs to authorization assessment and results calculation are more normative and precise. It is also now possible for permissions with zero scopes to be granted (Grant Sec 3.3.4). (266, 310, 317)

Permission Ticket Ecosystem Rationalized

The permission ticket generation ecosystem has been rationalized. In UMA2, a permission ticket is always generated, and the value rotated, in cases of a redirect back from the claims interaction endpoint and in cases of need_info and request_submitted errors from token endpoint requests, and never in cases of other errors. An authorization process is still ongoing while the authorization server is still generating permission tickets. (275, 279, 298)

Only One Pushed Claim Token Now Allowed at a Time

In UMA1, the mechanism for claim token pushing was a JSON-encoded request message sent to the RPT endpoint, optionally including with a claim_tokens array each of whose objects had a format parameter and a token parameter. In UMA2 (Grant Sec 3.3.4), , due to increased alignment with OAuth, this structure was flattened and the request message – now sent to the token endpoint as application/x-www-form-urlencoded format – contains each of the inner parameters only once. (If it is desired to send multiple claim tokens in a single request message, a compound claim token format could be defined.)

RPT Upgrading Logic Improved

UMA2 includes more comprehensive and normative logic around RPT upgrading (Grant Sec 3.3.5, Sec 3.3.5.1). (281)

Token Revocation Clarifications

UMA2 includes more comprehensive and normative text around token revocation, and defines a token type hint for PCTs (Grant Sec 3.7). (295)

Refresh Token Grant and Downscoping Logic Clarifications

UMA2 ensures that the logic of downscoping during token refreshing is properly defined given that UMA scopes are bound to resources, and clarifies that the AS does not perform authorization assessment in this context (Grant Sec 3.6). (306)

Changes to AS-RS Interface/Protection API (Now Federated Authorization Specification)

Resource Registration Endpoint

Extraneous URL Parts Removed From Resource Registration API

The API available at the resource registration endpoint required the path to contain the string resource_set. This string has ben removed (FedAuthz Sec 3.2). (155)

Scope Description Documents No Longer Expected to Resolve at Run Time When Scopes Are URLs

The AS is no longer expected to resolve scope description details at resource registration time or at any other run-time requirement (FedAuthz Sec 3.1.1). (269)

Resource Descriptions Lose uri Parameter

The uri parameter in the resource description was removed due to potential security and privacy concerns. (FedAuthz Sec 3.1) (270)

Resource and Scope Description Documents Gain Description Parameters

Resource description documents and scope description documents each now have a new parameter, description, for a human-readable string describing the resource or scope (respectively) at length. (271, 272)

scopes Parameter in Resource Description Document Renamed to resource_scopes

The scopes parameter in the resource description document has been renamed to resource_scopes (FedAuthz Sec 3.1). (318)

New HTTP 400 and invalid_request Error

For a typical variety of malformed-request errors, a response of an HTTP 400 (Bad Request) status code and an optional invalid_request error code is now defined. (FedAuthz Sec 3.2) (354-1)

Permission Endpoint

Requesting Multiple Permissions and Permissions With Zero Scopes

It is now possible for the RS to request multiple permissions on the client's behalf, not just one; this enables the RS to request "packages" of multiple resources that are likely to need to be accessed together. It is also possible for the RS to supply zero scopes on a requested permission (FedAuthz Sec 4.1); this is because the client can request its own scopes directly from the AS (for more discussion see Token Endpoint Replaces RPT Endpoint; Client-Side Communications Defined as Extension Grant). (317)

Token Introspection Endpoint

scopes parameter renamed to resource_scopes in Introspection Response Object

The scopes parameter in the token introspection response object has been renamed to resource_scopes (FedAuthz Sec 5.1.1). (158)

Options Not to Use Token Introspection Explicitly Allowed

In UMA2, the RPT is explicitly a type of OAuth access token, and it has been clarified that the token can be self-contained and valided locally by the RS, or introspected at the AS at run time, or its cached value used as appropriate (FedAuthz Sec 5). (261)

permissions Claim in Token Introspection Object Must Be Used

If token introspection is used (see Options Not to Use Token Introspection Explicitly Allowed), the introspection object can no longer be extended to replace the permissions claim with an entirely different structure. (322)

permission Claim exp Sub-Claim's Meaning If Absent Removed

The statement about the permission claim's exp sub-claim not expiring if it is absent was removed for the multi-part rationale given in the linked issue. (337 sub-issue a)

From V1.0 to V1.0.1

The UMA V1.0 specifications (Core, RSR) were approved in March 2015. The UMA V1.0.1 specifications (Core, RSR) are were approved in an All-Member Ballot to be Kantara Recommendations and were published in December 2015.

The following release notes are catalogued according to their impact on software implementations (where impact on client software in addition to authorization server or resource server software is denoted with (+Client) in the section title). Links to relevant GitHub issues and specific section numbers are provided where possible, enabling old-to-new text comparisons and tracking of discussions and rationales.

The following themes animated the V1.0.1 release process:

• Account for V1.0 lessons learned out of the gate

• Achieve timeline predictability and minimization of disruption for V1.0 implementers

• Achieve efficiency, speed, and accuracy in specification revisions

• Achieve issue solution consistency with OAuth 2.0 and OpenID Connect where possible

• Within the allotted time, prioritize first blocking and critical bug fixes, then low-impact specification and implementation changes

Minor changes, such as changes that don't impact implementations or specification interpretations, are not discussed in this section. To see a full list of issues disposed of and specification commits related to V1.0.1, see the list of GitHub issues with the "V1.0.1" label and the commit histories for Core and RSR.

Changes Affecting Authorization Server (+Client) Implementations

Following are specification changes in V1.0.1 that affect authorization servers, and possibly clients that interact with them as well.

AS Now Has Unique Redirect URI Endpoint for Claims Gathering (+Client)

Previously, the client was instructed to present the ordinary OAuth redirect_uri endpoint to which the AS should redirect requesting parties back after claims gathering, but this was ambiguously specified and incorrect. Now the client has a unique endpoint, claims_redirect_uri, that it needs to register. (144)

Permission Ticket Lifecycle Management (+Client)

Previously, little guidance was offered on how to manage permission tickets. Now some implications are explored, particularly as they relate to client interaction. (172) (Core Sec 3.2.2)

Requested Permission and Permission Ticket Matching

Previously, the matching of the "extents of access" of the requested permission registered by the RS and the permission ticket issued by the AS was implicit. Now it is spelled out. (175) (Core Sec 3.2)

Permission Ticket on Redirect Back to Client (+Client)

Previously, the AS was required to repeat the client’s permission ticket back to it in a ticket property when offering a redirect_user hint in error_details. Now this is optional and the client is encouraged to ignore the property's value, preparatory to removing the property entirely in a future UMA version. The reason is that the value can't be guaranteed good; repeating the value was in order to save the client work; and having the client check the value would ultimately have caused both sides work for no gain. (205) (Core Sec 3.5.4.2)

PUT Means Complete Replacement

Previously, the requirement for an Update method in resource set registration to completely replace the previous resource set description was implicit. Now it is spelled out. (177) (RSR Sec 2.2.3)

Default-Deny for Authorization Data Issuance

Previously, a naive implementation could have resulted in accidental default-permit authorization data issuance in some cases. Now a default-deny authorization assessment model has been made explicit, with an example given of how implementations could get into trouble. (194) (Core Sec 3.5.2)

base64url-Encoded Claims (+Client)

Previously, the wording about base64url-encoding pushed claims was ambiguous about whether double-encoding was necessary in the case of claim formats that were already base64url-encoded. Now it has been clarified that double-encoding should not be performed. (206) (Core Sec 3.6.2)

Enhanced Security Considerations

Previously, the security considerations around accepting policy-setting context information from an incompletely trusted RS only covered "bad icon URIs". Now they cover all such policy-setting context information, following roughly the OAuth example. (151) (RSR Sec 4)

Previously, the security considerations around client-pushed claims were explored only in a very cursory fashion in the body of the text. Now they are treated at length in a new subsection. (160) (Core Sec 7.4.1)

Enhanced Privacy Considerations

Previously, little was said about privacy implications of requesting party claims being transmitted to the AS. Now this section has been greatly expanded. (211) (Core Sec 8.2)

Changes Affecting Resource Server (+Client) Implementations

Following are specification changes in V1.0.1 that affect resource servers, and possibly clients that interact with them as well.

Caveat About Resource Server API Constraint

Previously, the specification was missing an important caveat: Based on a client's initial RPT-free resource request, the RS needs to know the correct AS, PAT, and resource set ID to include in its follow-on call to the permission request endpoint at the AS. Thus, the API of the RS needs be structured so that it can derive this information from the client's request. Now this caveat appears in several locations. (161, 162, 225)

Adjustment of Other Resource Server API Constraints (+Client)

Previously, the specification wording was inconsistent and problematic regarding how the RS responds to a client request accompanied by no RPT or an RPT with insufficient authorization data (assuming permission request success). Now the ability not to respond at all is more fully acknowledged; all responses intended to be interpreted in an UMA fashion are required to be accompanied by a WWW-Authenticate: UMA header; the permission ticket is required to be returned in a new ticket parameter in that header; complete freedom is given regarding the RS's choice of HTTP status code; and only in the case of a 403 choice is a ticket in a JSON-encoded body suggested, preparatory to removing the body option in a future UMA version. The rationale for this somewhat dramatic set of changes is that the original prescription to return HTTP status code 403 was incorrect; the specification gave too little guidance about responses other than 403 responses to be useful for client interoperability; and its requirement to return the permission ticket in a JSON-encoded body regardless of expected content type was an issue. (163, 164, 168) (Core Sec 3.3.1)

Solution for Permission Registration Failure (+Client)

Previously, the specification gave no guidance on how the RS should respond to the client in case of permission registration failure at the AS. Now, if the RS responds at all, it is required to substitute a Warning: 199 - "UMA Authorization Server Unreachable" header for WWW-Authenticate: UMA. (176) (Core Sec 3.3.2)

Authorization Server URI to Return to Client (+Client)