@craigmcc/oauth-orchestrator

oauth-orchestrator

Basic implementation of OAuth 2.0 RFC 6749, and the corresponding usage of Bearer tokens for authorizing incoming requests RFC 6750. It provides back end functionality, suitable for integration with a Node-based web application.

It supports the Resource Owner Password Credentials Grant (Section 4.3) and the Refreshing an Access Token (Section 6) flows.

The primary focus is supporting use cases where (in RFC 6749 terminology) the Authorization Server (which validates passwords and gives out access and refresh tokens) and the Resource Server (which authorizes incoming requests and performs your application logic) are in the same web application. However, it would be straightforward to split these responsibilities into different servers with support for some back-channel communication between the two.

1. Installation

npm install @craigmcc/oauth-orchestrator

2. Features

• Supports password and refresh token grants for Authorization Server implementations.
• Supports an authorize() method for the Resource Server to verify the validity of an access token (is it a valid token, has it expired, does it possess the scope required for the application function being requested) on each request to a protected API endpoint.
• Orchestrator is agnostic about application-specific concerns, such as:
  • Where and how user information is stored.
  • Where and how access token and refresh token information is stored.
  • What HTTP framework might be in use.
• Instead, an application integrating this library must provide a small set (6) of handler functions to provide concrete integrations of necessary features.

Authenticating client applications (via client_id and client_secret properties) is not currently supported. This is a likely future addition, but to maintain backwards compatibility it will likely remain optional.

3. Technologies

In contrast to many of the OAuth packages currently available for Node-based web applications, Orchestrator strives to be minimalist in its requirements. Indeed, if you peruse the package.json file, defining it, you will note that there are zero runtime dependencies. It also presumes that a reasonably current set of technologies are available.

As such, the following technologies form the basis of this library:

• NodeJS. I use version 14.15 or later, although it may work with previous versions.
• Typescript. Coming out of a primarily Java-based software development career, the object orientation and error catching was very comfortable. I use version 4.1 or later, transpiling to target es6 by default. The library should (of course) be usable in pure Javascript environments as well.
• Mocha and Chai for testing. This was mostly personal preference, but I like the robust support for async/await based functions out of the box, as well as the cleaner output formats than some other testing libraries.

No web framework is defined as a dependency - it is up to the handler functions provided by your integration to adapt to whatever request/response support your web framework offers. I use Express for my own personal projects, where the middleware support makes integration for authorization calls very easy, but it should be possible to use other frameworks as well.

4. Integration Steps:

4.1 Developer Notes

For Typescript-based applications, all the type definitions mentioned below are exported by the library, so you can say things like this in your application:

import {
  AccessToken,
  RefreshToken,
  User } 
from "@craigmcc/oauth-orchestrator";

For non-Typescript-based applications, you cannot reference the type definitions in your code, but they are available to read in a liberally commented text file, which (after installation) will be at node_modules/@craigmcc/oauth-orchestrator/types.d.ts relative to your project directory. This will help you get the parameter types and return values right on your handler function implementations.

4.2 Create Persistent Storage Implementations

Your application is responsible for providing persistent storage (typically in a database, but that is up to you) for the following object types:

AccessToken - Access tokens that are created or retrieved by the Orchestrator.

export interface AccessToken {
    expires: Date;              // Timestamp when this token expires
    scope: string;              // Scope granted to this token
    token: string;              // Actual access token value
    userId: Identifier;         // User this token is associated with
}

The token field of an access token is an opaque string that is send on each request to the Resource Server. They expire at a certain time, and can be refreshed or revoked (which is effectively a logout operation).

To minimize security risks, you should generate reasonably long random character strings as token values.

RefreshToken - Refresh tokens that are created or retrieved by the Orchestrator.

export interface RefreshToken {
    accessToken: string;        // Access token value this refresh token is for
    expires: Date;              // Timestamp when this token expires
    token: string;              // Actual refresh token value
    userId: Identifier;         // User this token is associated with
}

If configured (and it is by default), a user who successfully authenticates will receive both an access token and a refresh token. The refresh token will generally have a much longer lifetime than an access token, and can (when the access token nears its expiration) be exchanged for a new access token and refresh token, without requiring the user to be authenticated again.

User - Instances of all users allowed to be authenticated.

export type Identifier = string | number;

export interface User {
    scope: string;              // Space-separated scopes granted to this user.
    userId: Identifier;         // Unique user identifier.
}

For maximum flexibility, an Identifer is either a string or a number.

This is all that the Orchestrator needs to keep track of about users, once they have been authenticated.

• scope - A space delimited maximum list of permissions that this user will be granted when they authenticate. (They can ask for fewer permissions if they want, but that is optional).
• userId - An opaque identifier for this user (typically it is the primary key for your user table, but can be whatever you want as long as it is unique between users). This identifier is used to tie together the tokens that have been granted to this user.

If you have a much richer model of what a "user" is in your application, that is fine. Because Orchestrator does not care about these details, all you have to do is satisfy the object definition above in order to operate successfully with it.

Note in particular that there are no username or password fields included in this object. Orchestrator does not need to know or care about those details. If you use the Password Grant approach to OAuth, you'll be requiring the user to specify username and password to perform the authentication, but after that they no longer matter, and are not kept inside.

Indeed, even the mechanism that your application uses to perform this authentication is totally up to you as well.

We will get to how your Resource Server can leverage the authorize() capability to check for access on each request later, after we set up the Authorization Server integration.

4.3 Create Required OrchestratorHandlers Object and Handler Implementations

The definition of an OrchestratorHandler is pretty straightforward. It is merely a list that maps implementation-specific handlers to names that the Orchestrator knows, so that it can call out to your methods as needed. The Typescript-y version of this object is:

export interface OrchestratorHandlers {
    authenticateUser: AuthenticateUser;
    createAccessToken: CreateAccessToken;
    createRefreshToken: CreateRefreshToken;
    retrieveAccessToken: RetrieveAccessToken;
    retrieveRefreshToken: RetrieveRefreshToken;
    revokeAccessToken: RevokeAccessToken;
}

It is likely easiest to create this object in a separate file (MyOrchestratorHandlers.ts or whatever), which also includes the non-exported implementations of each handler function. For more complex scenarios, you might prefer to separate the handlers into their own individual files.

Note that all the handler functions return Promises, and are therefore expected to be async functions. Fortunately, pretty much any libraries you need for database access, password hashing, and your web framework will be very comfortable with this.

As a personal preference, I like using "await" over then/catch chains (with try/catch blocks to deal with errors as needed), but that style choice us up to you. There is no support for the older Javascript style of appending a callback function to the parameters.

The implementation requirements for each handler are described in the following sections (along with the Typescript signature, with definitions of the parameters and return type).

4.3.1 Authenticate User

export type AuthenticateUser
    = (username: string, password: string)
    => Promise<User>;

For the Password Grant flow in OAuth, it is presumed that your application will have some sort of login screen that asks for username and password. These are then submitted to the Orchestrator in order to ask the AuthenticateUser handler to actually authenticate these credentials.

The most common approach to this is to have your application provide an API endpoint (often /oauth/token but this is up to you) with incoming data that looks like a PasswordTokenRequest that triggers a call to the AuthenticateUser handler. Per the OAuth specification, this request should have property names matching those in PasswordTokenRequest, with content type application/x-www-form-urlencoded (i.e. the standard format for form submission).

If authentication is successful, your handler should return a User object, as described above. Orchestrator will then use that object to create an access token and (optional, but turned on by default) refresh token, which will be returned as a TokenResponse object (in JSON format).

If authentication fails (invalid username or password) an InvalidGrantError will be returned to you, with the underlying error included in the inner property. To avoid showing potentially sensitive information to client callers, this property should be suppressed in any response sent back to the calling application.

IMPLEMENTATION NOTE: It is strongly recommended that you do not store plaintext passwords in a database! Instead, encrypt them with a one-way hash function (I like bcrypt but the choice is yours), and implement your authentication handler to verify the submitted password against the hashed version retrieved from your database. Just be sure that you use the same algorithm for hashing and verifying.

4.3.2 Create Access Token

export type CreateAccessToken
    = (expires: Date, scope: string, userId: Identifier)
    => Promise<AccessToken>;

This handler will be called whenever the Orchestrator needs to create a new access token. This will happen in the following scenarios:

• A newly logging in user is successfully authenticated, and needs to receive an access token for use on all the subsequent API requests for that user.
• An existing logged in user has a valid access token, but it is approaching the end of its life. The client application recognizes this situation, and uses the refresh token it previously received to trigger creation of a new access token and (optional) refresh token. As a side effect, the old tokens will be revoked so that they are no longer valid.

The returned access token will be returned to the user, and then it's token value will be sent with each incoming API request, in order to prove the user's identity has been confirmed, and to validate whether the user is allowed to perform the operation being requested (by comparing the scope assigned to this token to that required by the requested operation).

If your application has problems storing or returning this token, simply throw an appropriate Error.

4.3.3 Create Refresh Token

export type CreateRefreshToken
    = (accessToken: string, expires: Date, userId: Identifier)
    => Promise<RefreshToken>;

This handler will be called if Orchestrator is configured to return refresh tokens (it is by default), and a refresh token will be created via this call, at the same time that a new access token was created.

If your application has problems storing or returning this token, simply throw an appropriate Error.

If your application turns off refresh token creation, it will not be possible to use such a token to receive new tokens. When the user's access token eventually expires, calls to the authorize() Orchestrator method will start failing (if you follow best practices, an HTTP 401 status will be returned to the user), after which they will need to reauthenticate to continue using the application.

4.3.4 Retrieve Access Token

export type RetrieveAccessToken
    = (token: string)
    => Promise<AccessToken>;

This handler will be called whenever an incoming request is authorized (via a call to the authorize() method of the Orchestrator). You do not need to worry about checking whether the token has expired or not - Orchestrator will take care of that detail.

If the requested token does not exist, or your internal systems have problems, throw an approprite Error.

PERFORMANCE NOTE: This handler will be called many many times - once per API call to a protected resource. If this creates performance issues, consider using some sort of in-memory cache to minimize the number of database calls needed. Just be sure that you purge the cache entries when the RevokeAccessToken handler is called, to avoid stale tokens from being used for additional requests.

4.3.5 Retrieve Refresh Token

export type RetrieveRefreshToken
    = (token: string)
    => Promise<RefreshToken>;

Existing refresh tokens are retrieved by this handler, whenever needed for the Refresh Token flow.

As usual, if your server environment has problems, or if the requested refresh token does not exist, throw an Error.

4.3.6 Revoke Access Token (and any related Refresh Tokens)

export type RevokeAccessToken
    = (token: string)
    => Promise<void>;

Conventionally, your application will offer an endpoint (I like to use DELETE /oauth/token, but this it is up to you) that effectively logs the user off by removing or deactivating the specified access token, along with any associated refresh tokens. The processing logic for this endpoint will call this handler. Afterwards, the access token that was used will no longer be valid, and the user will need to log back in again via the Password Grant flow.

Whether you offer such an endpoint or not, you will want to build some sort of scheduled job (daily or whatever) to prune expired access tokens (and their corresponding refresh tokens, if any).

4.4 Create Optional OrchestratorOptions To Override Defaults

If you want to modify some or all of the configuration options for the Orchestrator, you can create an options object that is passed to the Orchestrator at instantiation time. The type definitions for this object describe the options that can be modified, with default values in square brackets.

export interface OrchestratorOptions {
    accessTokenLifetime?: number;       // In seconds [86400, one day]
    issueRefreshToken?: boolean;        // Issue refresh token with
                                        // access token? [true]
    refreshTokenLifetime?: number;      // In seconds [604800, one week]
}

4.5 Instantiate and Configure an Orchestrator Singleton

In the top-level Javascript object that controls your application, include logic similar to this:

import { Orchestrator } from "@craigmcc/oauth-orchestrator";
import { MyOrchestratorHandlers } from "...wherever...";
export const OAuthOrchestrator: Orchestrator
  = new Orchestrator(MyOrchestratorHandlers);

This will make a configured OAuthOrchestrator instance available to any other part of your application that needs access to it.

If you want to override some of the configuration properties, pass a suitable OrchestratorOptions object as the second parameter to the constructor.

4.6 Integrate Authorization Calls In Your Resource Server

4.6.1 Requirements

For each request to an API that is protected by access restrictions, the goal is to implement a call to the authorize() method of the Orchestrator. Two pieces of information are required:

• The access token value (extracted from the Authorization header) for this request.
• The scope that is required to complete this request. Knowing this will require an understanding of the scope architecture, to understand what operations are allowed by what scopes.

The details of how this can be implemented will vary depending on the web framework you are using. Most such frameworks allow the configuration of an interceptor of some sort, before the normal processing of the request, that extracts the access token, and knows the scope requirements for that request (based on an understanding of the scope architecture).

4.6.2 Example Express Middleware

In a framework like Express, such a technique can be implemented by creating "middleware" functions that are inserted in the processing flow for a particular route (or all routes to a particular router, or the entire application). Using Express Typescript declarations, a middleware function that requires the "admin" scope might be implemented like this (with supporting utility functions shared across middleware implementations):

export const requireAdmin: RequestHandler =
    async (req: Request, res: Response, next: NextFunction) => {
        const token = extractToken(req);
        if (!token) {
            throw new Forbidden("No access token presented", "requireToken");
        }
        const required = "admin";
        await authorizeToken(token, required);
        res.locals.token = token;
        next();
}

const extractToken = (req: Request) : string | null => {
  const header: string | undefined = req.header("Authorization");
  if (!header) {
    return null;
  }
  const fields: string[] = header.split(" ");
  if (fields.length != 2) {
    return null;
  }
  if (fields[0] !== "Bearer") {
    return null;
  }
  return fields[1];
}

const authorizeToken = async (token: string, required: string): Promise<void> => {
  try {
    await OAuthOrchestrator.authorize(token, required);
  } catch (error) {
    throw error;
  }

}

Note that the "required" scope being specified is a minimum requirement. It is perfectly fine if the access token has been granted wider scope than what is required, but being granted less than the required scope will trigger a 403 response.

In more complex environments, the middleware might need to have access to which information is being requested, which will often be based on the value for a request parameter. In Express, the middleware has access to these parameters via req.params.{name}, which can then be used to influence the scope that is being defined as required for this operation.

4.6.3 Express Implementing Per-Route Restrictions

Imagine we have a library application, where any user is allowed to retrieve information about a particular library, but only an admin is allowed to change it. The route for the HTTP PUT that supports this function might be configured like this:

LibraryRouter.put("/:libraryId",
    requireAdmin,
    async (req: Request, res: Response) => {
        res.send(await LibraryServices.update(
            parseInt(req.params.libraryId, 10),
            req.body
        ));
    });

where requireAdmin is the middleware function we defined above. Because this is placed ahead of the actual update processing, it will be called first -- and any attempt by a non-admin to perform this operation will have received a 403 response, without the actual business logic of doing the update from having to care about the permissions check.

4.7 Client Application Responsibilities

4.7.1 Design Scope Architecture

One of the most interesting application design tasks is to define the architecture of what scope settings should be defined, and what operations they should allow.

As a quick review, a scope (in OAuth terms) is a space-delimited set of one or more "permissions" to do things (other approaches are possible, but this is likely to be very common). In an OAuth environment, an access token is associated with such a scope, and is typically either:

• A scope that is associated with the user, based on their role in this application. This will be used if the request to create an access token does not include a specified scope.
• A scope that is requested when requesting an access token.

A properly implemented Resource Server will deny any request to ask for more permissions than the user is allowed. For example, an "admin" scope might only be assigned to certain individuals, and then be used to limit functionality that a user might try to use.

Similarly, since a request for an access token might include a request for a particular scope that is not permitted for that user. In this case, the Authentication Server should disallow that request.

Scopes can range from simple (for instance, an "admin" user versus a "regular" user) to more complex. Consider an online SaaS app that supports multiple customers, each of which might have their own hierarchies of admin and regular users -- but none of them should be able to access any of the information for another customer. This requirement can be implemented in any number of ways. Using scope is an option, but will sometimes require a Resource Server to extract request parameters from the URL (for example, a customer id), or (worse from a performance viewpoint) from data retrieved from the database.

4.7.2 Implement Login Capabilities

How does a user request an access token in the first place? If the application uses the Password Grant flow, this will typically be handled by:

• Providing a login view that accepts username and password.
• Formulate a PasswordTokenRequest request, and send it to the "token" endpoint of the Authorization Server (often /oauth/token);
• Process the returned result, which will typically be:
  • HTTP Status 200, with a TokenResponse response body. Record the access_token and (optional) refresh token, as well as the scope assigned to this access token. You might also choose to record the expires_in so that you can warn the user about impending expiration, and/or proactively execute the Refresh Token flow to get a new access token without user intervention.
  • HTTP Status 401. Normally this will mean a problem in validating the username and password credentials. Ask the user to enter them again.
  • Other HTTP Status. Probably a server side issue.

In any of the non-200 cases, the response body might include a JSON structure such as an InvalidGrantError, which will include a message property with an explanatory message. However, any such response means that an access token was not received, so the login was not successful.

4.7.3 (Optional) Implement Logout Capabilities

A client application may choose to offer a "logout" or "sign out" option. Implementing this will typically include:

• Format a request to the appropriate endpoint that is provided by the server for this purpose.
• Make sure this request is authorized with the same access token that the user is sending on all normal requests (after all, we don't want the server invalidating access tokens that might actually belong to someone else).
• Dealing with responses as usual.

If logout has occurred, the client application should throw away any access token information related to this user, so that any subsequent request will trigger a 401 or 403 response that should then trigger having to login again.

4.7.4 Include Access Token On Every Protected API Request

When sending such a request, include an HTTP Authorization header, including the users's access token (if any), like this:

Authorization: Bearer {user-access-token}

Note that the OAuth specification supports a couple of alternative ways to include the Bearer token (as a query parameter, or as a field in a form submit), but these are problematic from a security (query parameters are often logged by the server), or accurate data modelling (polluting the request body with something that is not related to the actual data being transmitted) perspective. Therefore, the oauth-orchestrator implementation only accepts Bearer tokens.

4.7.5 Deal With Authentication or Authorization Responses

In addition to normal application responses, two particular HTTP status codes indicate issues with OAuth authentication and authorization issues:

• HTTP Status 401: Most likely scenario is that the previously usable access token has expired. Send the user back to the login mechanism, so they can request a new token. For extra credit, return them to where they were, perhaps resubmitting the request that failed.
• HTTP Status 403: Most likely scenario is that the user submitted an API request that is not allowed by the scope granted to the access token they are using. This type of request will not be successful unless the user logs in with credentials that allow the required scope.

In an ideal scenario, the client application design will seek to minimize the chances for a 403 response happening, by not even offering (through the application's UI) a way to invoke the API request that is going to fail. For example, if a user is not allowed to request or modify data of a particular type, do not even offer the UI that makes such a request possible.

Of course, the Authorization Server is going to enforce scope restrictions under all circumstances, because this application might not be the only one utilizing the server. Avoiding the possibility of triggering such issues in the first place will lead to a better user experience.

How do we do this? As part of the information returned when an access token is received, the scope assigned to that access token is also returned. If the client application understands the scope architecture, it can make decisions about what functions can be offered in the UI, avoiding those that will trigger 403 responses.