More About Using OAuth 2.0¶
Review the CU-STD-INTR-001: OAuth 2.0 Scope Standard:
N.B. OAuth 2.0 is largely unrelated to OAuth 1.0; it’s not just an upgrade, but a totally new protocol. Many LTI APIs are based on OAuth 1.0. Different thing entirely.
The players in the world of OAuth 2.0 are:
User: The user is the human who logs in the usual way via a browser (e.g. using CAS in our case). The user login is generally initiated by the client making a request of the AS (below).
Client: The client app operates on behalf of the user. Since it’s frequently referred to as just the “client”, it is sometimes confused with the user, but it isn’t. It is the app that the user is authorizing to act on her behalf, sometimes even when the user is no longer present.
Resource Server (RS): The resource server is the “backend” (in our cast RESTful Django) server. It uses information passed by the client (described below) to see if the resources it manages are permitted to be manipulated using the HTTP methods.
Authorization Server (AS): This is the central control point. All clients and RS’s talk to the AS to get and validated a Token. Clients and RS’s are registered in advance with the AS.
More on these roles below.
Access Tokens are used for all authorization decisions. An Access Token is usually an opaque object that has no intrinsic meaning to the client app that received it from the AS. How access tokens get created and constrained is described below. Additional token types include Refresh and ID tokens.
Refresh Tokens, which can only be granted for an Authorization Code grant, are used to get a new Access Token when the old one expires. These are used for long-term authorization of 3rd-party apps to do something on behalf of the initial authorizing user without them having to be present at the time.
ID tokens contain information that identifies the user to the client app
and are only made available if using the Open ID Connect (OIDC) profile (with
openid scope, described below). In general, the client app does not know
who their user is, since the Access Token is an opaque object.
Scopes are requested by a registered Client app. They are granted
(and possibly down-scoped for so-called Enterprise Scopes) by the
OAuth 2.0 Authorization Server (AS). Scopes are then validated by the
Resource Server based on the
access_token provided in the Authorization
Bearer header of the HTTP request. The set of scopes “in” the
access_token must match the scopes required by the Resource Server.
Beyond the required scopes, additional scopes might be useful for
Our implementation of scopes includes several flavors:
Authentication Selector Scopes¶
Authentication selector scopes specify what method of user authentication the client app is requesting the AS to use. The two we are concerned with are:
auth-columbia: Perform a CAS user login. See Client is operating with an end user present, below.
auth-none: Do not perform a user login. This scope is used for backend server-to-server relationships. See Client is a trusted server, below.
Scope for Client to Learn the User’s Identity¶
In normal OAuth 2.0 operations, there may be no need for the client app to know who the end user is. If the client does need to know this, then use the openid scope. This results in an id_token being returned to the requesting client, in addition to the usual access_token – only if the client app has been registered with permission to request the openid scope. id_tokens only return the “sub” claim (email@example.com) unless additional scopes are added:
openid: Return an id_token to the Client app.
profile: Added to openid, includes user’s name, etc.
email: Added to openid, includes user’s email address
There’s a complementary userinfo endpoint that returns similar information, called claims. It is authorized using the access_token.
TODO: Update documentation for group claim.
End-user Generic Scopes¶
A series of generic scopes are generally used for end-user data. If requested, these scopes are always granted if requested, so they are not appropriate to limit access to enterprise-owned data that the end-user might need special permissions for (see Enterprise Scopes):
create: Permission to create new resources (POST) on my behalf.
read: Permission to read an existing resource (GET) of mine.
update: Permission to update an existing resource (PATCH) of mine.
delete: Permission to delete a resource (DELETE) of mine.
Enterprise scopes confirm (in near real-time) end-user membership in a Grouper-managed “OAUTH Group” and are only applicable when combined with auth-columbia scope. There are currently no real enterprise scopes defined. There is one named demo-netphone-admin for demonstration purposes. Also, auth-columbia is both a scope selector and an enterprise scope; one can be effectively blocked from access via Grouper removing a user’s membership from the auth-columbia OAUTH Group.
Enterprise Scopes may also be used with auth-none scope as a means
of defining access for trusted servers.
In this case, both the trusted server’s client credentials must be
configured to allow auth-none and the enterprise scope, for example,
What kind of client is it?¶
No matter the OAuth “style”, the Resource Server (RS) is always
presented with basically the same thing: an opaque Access Token
Authorization: Bearer <token> header.
The RS (backend Django app in this context) must
validate the Access Token to grant or deny appropriate permission
for a given HTTP request.
Client is operating with an end user present¶
This use case is similar to the typical front-end client app connecting to a back-end resource server in which the user must log in (via a browser redirect to Shibboleth/CAS/Duo). The result is, as above, an access token. In a fully browser-based client, the access token would be persisted in a cookie, analogously to the way a session cookie is used.
Client is a trusted server¶
This use case is the example of backend server-to-server trust that would traditionally use IP addresses or HTTP Basic Auth to determine whether one server should trust another. In the OAuth 2.0 world, this is implemented by the client using HTTP Basic Auth with the AS to get back an Access Token. Validation of (introspecting) the access token is done by the server using HTTP Basic Auth to identify itself and then introspecting the token. No user identity or permissions are provided because there is no user. In addition to using the scopes associated with the access token, the server app can always keep a list of client IDs if necessary. However, the cleanest design approach is to base the authorization decision entirely on the access token’s granted scopes.
Registering with the AS¶
Register Resource Server for Token Introspection¶
The Resource Server must be a registered client with the OAuth 2.0 Authorization Server before it is allowed to perform token introspection.
This process is currently manual via a service request to the IAM team. You need to provide:
the name of your Resource Server (e.g. cas-coursemgt-server)
that it should be allowed to perform introspection (is a resource server)
You’ll be given a client_id and client_secret. Configure these in settings.OAUTH2_PROVIDER using environment variables or some other method to protect the credentials (e.g. not in the source code!).
Register Client App(s)¶
Every client app must be a registered OAuth 2.0 client. Things to think about when registering a client are:
Client is operating on behalf of an end user¶
What scopes should the app be allowed to request on behalf of the user and what scopes are required by its Resource Server?
Should the end-user scope-approval page be presented or bypassed? In most cases, this will be bypassed.
Client is a server¶
In this case, the app should be registered with permission to request the auth-none and additional scopes.