Case Study: Access Management 2.0 for the Enterprise

Introduction

Although UMA's primary use cases have centered on individual people (that is, the "users" who manage access to their own online resources), the UMA notion of authorization as a service also has relevance to modern enterprises that must secure APIs and other web resources in a developer-friendly way.

Problem Scenario

Where once web access management (WAM) and single sign-on (SSO) were sufficient for many purposes in the enterprise context, a new requirement has surfaced: managing access to an enterprise's web APIs, not just web apps. Today's systems for managing this type of access have a number of challenges.

Using current WAM solutions to provide API security can be unfriendly to developers, complex, expensive, and likely proprietary. Mobile clients struggle to deal with XML-based and SOAP-based security mechanisms. Enterprise IT struggles to deploy agents or proxies.

Since it's currently overly complex to centralize access authorization, we find too much authorization code in applications, which slows service delivery by forcing developers to redevelop authorization logic, as well as hindering effective auditing and policy administration.

WAM solutions are not fully agnostic as to authentication method. WAM solutions previously could make simplifying assumptions about how users authenticate (typically username and password into a web app). With mobile applications, game consoles, and other device types, and with strong authentication needs increasing, old assumptions are no longer viable.

APIs by their nature are subdivisions of functionality exposed at a single domain, and would map well to arbitrarily fine-grained policy, for example, at the method or even parameter level. However, outside the use of XACML, authorization policy granularity is coarse in traditional solutions: group filtering or URL regular expression matching at most.

Proposed Improvements

UMA makes the following solutions possible.

As a profile of OAuth 2.0 (IETF RFCs 6749 and 6750) that is complementary to OpenID Connect, UMA defines RESTful, JSON-based, standardized flows and constructs for coordinating the protection of any API or web resource in a way that will be familiar to any developer already acquainted with OAuth. Mobile developers accept technologies that use HTTP and JSON at their core.

UMA's notion of machine-readable resource set and scope descriptions creates an access control mechanism that enables control over specific API scopes (customizable buckets of API functionality), not just domains. With UMA, client app developers can handle authorization tasks by calling simple REST/JSON endpoints; administrators don't have to deploy a web server agent or reverse proxy to enable centralization.

UMA defines interfaces between authorization servers and resource servers that, by default, enable centralized policy decision-making for improved service delivery, auditing, policy administration, and accountability, even in a very loosely coupled "public API" environment. Custom profiles enable flexibility to move the decision-making line outward to distributed applications, to account for local preferences in API ecosystems. UMA does not standardize a policy expression language, enabling flexibility in policy expression and evaluation through XACML, other declarative policy languages, or procedural code as warranted by conditions.

UMA inherits authentication agnosticism from OAuth. It concentrates on authorization, not on authentication. It has been profiled to work with OpenID Connect to gather identity claims from whoever is attempting access, and enables true claims-based authorization (with simple group- or role-based policies a natural subset).

Solution Scenario

In UMA trust model terminology, this scenario is in the category non-person entity (NPE) to person sharing. An organization – say, BusinessCo -- is both the resource owner (technical term) and the Authorizing Party (contractual term), acting on its own behalf, protecting its own resources. A human "resource owner agent" acts in a IT administrator role.

BusinessCo runs a service that does whatever elements of the authorization job it has chosen to centralize; this is the UMA authorization server. Think of this as a policy decision point (PDP), though UMA's default profile gives less than 100% of the decision-making responsibility to it (the authorization server may in turn outsource actual decision-making to an XACML PDP or some other web service). This service would also expose policy administration point (PAP) functions to the IT administrator in some fashion. The service may itself be a policy information point (PIP), or may call out to one or more PIPs.

BusinessCo's protected resources would sit behind the enterprise apps and APIs it has chosen to expose, some of which may be run and hosted by third-party SaaS vendors; these apps and APIs represent the UMA resource servers. Think of these as policy enforcement points (PEPs), though UMA's default profile gives a bit of decision-making responsibility to them.

Client web or mobile applications, wielded by enterprise resource users such as employees and partners, are UMA clients. Those users are variously Requesting Parties or Requesting Party Agents.

Solution Flow

This scenario uses ordinary UMA flows, noting that it is a human "Authorizing Party Agent", not BusinessCo, that sets policy and possibly performs any policy-dictated manual intervention to enable access requests to go through. The company Gluu has produced a swimlane diagram to illustrate.

Solution Demo

Gluu gave a demo of an enterprise UMA scenario at an UMA WG meeting on Jan 10, 2013; see the meeting notes. Also see their series of demo videos on YouTube.