Resources, scopes, permissions and policies in Keycloak

Question

I want to create a fairly simple role-based access control system using Keycloak's authorization system. The system Keycloak is replacing allows us to create a "user", who is a member of one or more "groups". In this legacy system, a user is given "permission" to access each of about 250 "capabilities" either through group membership (where groups are assigned permissions) or a direct grant of a permission to the user.

I would like to map the legacy system to Keycloak authorizations.

It should be simple for me to map each "capability" in the existing system to a Keycloak resource and a set of Keycloak scopes. For example, a "viewAccount" capability would obviously map to an "account" resource and a "view" scope; and "viewTransaction" maps to a "transaction" resource... but is it best practice to create just one "view" scope, and use it across multiple resources (account, transaction, etc.)? Or should I create a "viewAccount" scope, a "viewTransaction" scope, etc.?

Similarly, I'm a little confused about permissions. For each practical combination of resource and scope, is it usual practice to create a permission? If there are multiple permissions matching a given resource/scope, what does Keycloak do? I'm guessing that the intention of Keycloak is to allow me to configure a matrix of permissions against resources and scopes, so for example I could have permission to access "accounts" and permission for "view" scope, so therefore I would have permission to view accounts?

I ask because the result of all this seems to be that my old "viewAccount" capability ends up creating an "Account" resource, with "View" scope, and a "viewAccount" permission, which seems to get me back where I was. Which is fine, if it's correct.

Finally, obviously I need a set of policies that determine if viewAccount should be applied. But am I right that this means I need a policy for each of the legacy groups that a user could belong to? For example, if I have a "help desk" role, then I need a "help desk membership" policy, which I could then add to the "viewAccount" permission. Is this correct?

Answer 1

Full transparency—I am by no means a Keycloak, OAuth, or OIDC expert and what I know is mostly from reading the docs, books, good ol' YouTube and playing around with the tool.

This post will be comprised of two parts:

1. I'll attempt to answer all your questions to the best of my ability
2. I'll show you all how you can play around with policies/scopes/permissions in Keycloak without needing to deploy a separate app in order to better understand some of the core concepts in this thread. Do note though that this is mostly meant to get you all started. I'm using Keycloak 8.0.0.

Part I

Some terminology before we get started:

• In Keycloak, you can create two types of permissions: Resource-Based and Scope-Based.
• Simply put, for Resource-Based permissions, you apply it directly to your resource
• For Scoped-Based permission, you apply it to your scope(s) or scope(s) and resource.

is it best practice to create just one "view" scope, and use it across multiple resources (account, transaction, etc.)? Or should I create a "viewAccount" scope, a "viewTransaction" scope, etc.?

Scopes represent a set of rights at a protected resource. In your case, you have two resources: account and transaction, so I would lean towards the second approach.

In the long run, having a global view scope associated with all your resources (e.g., account, transaction, customer, settlement...) makes authorization difficult to both manage and adapt to security requirement changes.

Here are a few examples that you can check out to get a feel for design

• Slack API
• Box API
• Stripe

Do note though - I am not claiming that you shouldn't share scopes across resources. Matter of fact, Keycloak allows this for resources with the same type. You could for instance need both viewAccount and viewTransaction scope to read a transaction under a given account (after all you might need access to the account to view transactions). Your requirements and standards will heavily influence your design.

For each practical combination of resource and scope, is it usual practice to create a permission?

Apologies, I don't fully understand the question so I'll be a bit broad. In order to grant/deny access to a resource, you need to:

• Define your policies
• Define your permissions
• Apply your policies to your permissions
• Associate your permissions to a scope or resource (or both)

for policy enforcement to take effect. See Authorization Process.

How you go about setting all this up is entirely up to you. You could for instance:

• Define individual policies, and tie each policy under the appropriate permission.

• Better yet, define individual policies, then group all your related policies under an aggregated policy (a policy of policies) and then associate that aggregated policy with the scope-based permission. You could have that scoped-based permission apply to both the resource and all its associated scope.

• Or, you could further break apart your permissions by leveraging the two separate types. You could create permissions solely for your resources via the resource-based permission type, and separately associate other permissions solely with a scope via the scope-based permission type.

You have options.

If there are multiple permissions matching a given resource/scope, what does Keycloak do?

This depends on

1. The resource server's Decision Strategy
2. Each permission's Decision Strategy
3. Each policy's Logic value.

The Logic value is similar with Java's ! operator. It can either be Positive or Negative. When the Logic is Positive, the policy's final evaluation remains unchanged. When its Negative, the final result is negated (e.g. if a policy evaluates to false and its Logic is Negative, then it will be true). To keep things simple, let's assume that the Logic is always set to Positive.

The Decision Strategy is what we really want to tackle. The Decision Strategy can either be Unanimous or Affirmative. From the docs,

Decision Strategy

This configurations changes how the policy evaluation engine decides whether or not a resource or scope should be granted based on the outcome from all evaluated permissions. Affirmative means that at least one permission must evaluate to a positive decision in order grant access to a resource and its scopes. Unanimous means that all permissions must evaluate to a positive decision in order for the final decision to be also positive. As an example, if two permissions for a same resource or scope are in conflict (one of them is granting access and the other is denying access), the permission to the resource or scope will be granted if the chosen strategy is Affirmative. Otherwise, a single deny from any permission will also deny access to the resource or scope.

Let's use an example to better understand the above. Suppose you have a resource with two permissions and someone is trying to access that resource (remember, the Logic is Positive for all policies). Now:

1. Permission One has a Decision Strategy set to Affirmative. It also has 3 policies where they each evaluate to:

• true

• false

• false

  Since one of the policies is set to true, Permission One is set to true (Affirmative - only 1 needs to be true).

2. Permission Two has a Decision Strategy set to Unanimous with two policies:

• true

• false

  In this case Permission Two is false since one policy is false (Unanimous - they all need to be true).

3. Now comes the final evaluation. If the resource server's Decision Strategy is set to Affirmative, access to that resource would be granted because Permission One is true. If on the other hand, the resource server's Decision Strategy is set to Unanimous, access would be denied.

See:

• Resource Server Settings
• Managing Permissions

We'll keep revisiting this. I explain how to set the resource sever's Decision Strategy in Part II.

so for example I could have permission to access "accounts" and permission for "view" scope, so therefore I would have permission to view accounts?

The short answer is yes. Now, let's expand on this a bit :)

If you have the following scenario:

1. Resource server's Decision Strategy set to Unanimous or Affirmative
2. Permission to access the account/{id} resource is true
3. Permission to access the view scope is true

You will be granted access to view the account.

• true + true is equal to true under the Affirmative or Unanimous Decision Strategy.

Now if you have this

1. Resource server's Decision Strategy set to Affirmative
2. Permission to access the account/{id} resource is true
3. Permission to access the view scope is false

You will also be granted access to view the account.

• true + false is true under the Affirmative strategy.

The point here is that access to a given resource also depends on your setup so be careful as you may not want the second scenario.

But am I right that this means I need a policy for each of the legacy groups that a user could belong to?

I'm not sure how Keycloak behaved two years ago, but you can specify a Group-Based policy and simply add all your groups under that policy. You certainly do not need to create one policy per group.

For example, if I have a "help desk" role, then I need a "help desk membership" policy, which I could then add to the "viewAccount" permission. Is this correct?

Pretty much. There are many ways you can set this up. For instance, you can:

1. Create your resource (e.g. /account/{id}) and associate it with the account:view scope.
2. create a Role-Based Policy and add the helpdesk role under that policy
3. Create a Scope-Based permission called viewAccount and tie it with scope, resource and policy

We'll set up something similar in Part II.

Part II

Keycloak has a neat little tool which allows you test all your policies. Better yet, you actually do not need to spin up another application server and deploy a separate app for this to work.

Here's the scenario that we'll set up:

1. We'll create a new realm called stackoverflow-demo
2. We'll create a bank-api client under that realm
3. We will define a resource called /account/{id} for that client
4. The account/{id} will have the account:view scope
5. We'll create a user called bob under the new realm
6. We'll also create three roles: bank_teller, account_owner and user

• We will not associate bob with any roles. This is not needed right now.

7. We'll set up the following two Role-Based policies:
   • bank_teller and account_owner have access to the /account/{id} resource
   • account_owner has access to the account:view scope
   • user does not have access to the resource or scope
8. We'll play around with the Evaluate tool to see how access can be granted or denied.

Do forgive me, this example is unrealistic but I'm not familiar with the banking sector :)

Keycloak setup

Download and run Keycloak

cd tmp
wget https://downloads.jboss.org/keycloak/8.0.0/keycloak-8.0.0.zip
unzip keycloak-8.0.0.zip
cd keycloak-8.0.0/bin
./standalone.sh

Create initial admin user

1. Go to http://localhost:8080/auth
2. Click on the Administration Console link
3. Create the admin user and login

Visit Getting Started for more information. For our purposes, the above is enough.

Setting up the stage

Create a new realm

1. Hover your mouse around the master realm and click on the Add Realm button.
2. Enter stackoverflow-demo as the name.
3. Click on Create.
4. The top left should now say stackoverflow-demo instead of the master realm.

See Creating a New Realm

Create a new user

1. Click on the Users link on the left
2. Click on the Add User button
3. Enter the username (e.g., bob)
4. Ensure that User Enabled is turned on
5. Click Save

See Creating a New User

Create new roles

1. Click on the Roles link
2. Click on Add Role
3. Add the following roles: bank_teller, account_owner and user

Again, do not associate your user with the roles. For our purposes, this is not needed.

See Roles

Create a client

1. Click on the Clients link
2. Click on Create
3. Enter bank-api for the Client ID
4. For the Root URL enter http://127.0.0.1:8080/bank-api
5. Click on Save
6. Ensure that Client Protocol is openid-connect
7. Change the Access Type to confidential
8. Change Authorization Enabled to On
9. Scroll down and hit Save. A new Authorization tab should appear at the top.
10. Click on the Authorization tab and then Settings
11. Ensure that the Decision Strategy is set to Unanimous

• This is the resource server's Decision Strategy

See:

• Creating a Client Application
• Enabling Authorization Services

Create Custom Scopes

1. Click on the Authorization tab
2. Click on Authorization Scopes > Create to bring up Add Scope page
3. Enter account:view in the name and hit enter.

Create "View Account Resource"

1. Click on Authorization link above
2. Click on Resources
3. Click on Create
4. Enter View Account Resource for both the Name and Display name
5. Enter account/{id} for the URI
6. Enter account:view in the Scopes textbox
7. Click Save

See Creating Resources

Create your policies

1. Again under the Authorization tab, click on Policies
2. Select Role from the the Create Policy dropdown
3. In the Name section, type Only Bank Teller and Account Owner Policy
4. Under Realm Roles select both the bank_teller and account_owner role
5. Ensure that Logic is set to Positive
6. Click Save
7. Click on the Policies link
8. Select Role again from the Create Policy dropdown.
9. This time use Only Account Owner Policy for the Name
10. Under Realm Roles select account_owner
11. Ensure that Logic is set to Positive
12. Click Save
13. Click on the Policies link at the top, you should now see your newly created policies.

See Role-Based Policy

Do note that Keycloak has much more powerful policies. See Managing Policies

Create Resource-Based Permission

1. Again under the Authorization tab, click on Permissions
2. Select Resource-Based
3. Type View Account Resource Permission for the Name
4. Under Resources select View Account Resource
5. Under Apply Policy select Only Bank Teller and Account Owner Policy
6. Ensure that the Decision Strategy is set to Unanimous
7. Click Save

See Create Resource-Based Permissions

Phew...

Evaluating the Resource-Based permission

1. Again under the Authorization tab, select Evaluate

2. Under User enter bob

3. Under Roles select user

   • This is where we will associate our user with our created roles.

4. Under Resources select View Account Resource and click Add

5. Click on Evaluate.

6. Expand the View Account Resource with scopes [account:view] to see the results and you should see DENY.

   

7. This makes sense because we only allow two roles access to that resource via the Only Bank Teller and Account Owner Policy. Let's test this to make sure this is true!

8. Click on the Back link right above the evaluation result

9. Change bob's role to account_owner and click on Evaluate. You should now see the result as PERMIT. Same deal if you go back and change the role to bank_teller

See Evaluating and Testing Policies

Create Scope-Based Permission

1. Go back to the Permissions section
2. Select Scope-Based this time under the Create Permission dropdown.
3. Under Name, enter View Account Scope Permission
4. Under Scopes, enter account:view
5. Under Apply Policy, enter Only Account Owner Policy
6. Ensure that the Decision Strategy is set to Unanimous
7. Click Save

See Creating Scope-Based Permissions

Second test run

Evaluating our new changes

1. Go back to the Authorization section
2. Click on Evaluate
3. User should be bob
4. Roles should be bank_teller
5. Resources should be View Account Resource and click Add
6. Click on Evaluate and we should get DENY.
   • Again this should come as no surprise as the bank_teller has access to the resource but not the scope. Here one permission evaluates to true, and the other to false. Given that the resource server's Decision Strategy is set to Unanimous, the final decision is DENY.
7. Click on Settings under the Authorization tab, and change the Decision Strategy to Affirmative and go back to steps 1-6 again. This time, the final result should be PERMIT (one permission is true, so final decision is true).
8. For the sake of completeness, turn the resource server's Decision Strategy back to Unanimous. Again, go back to steps 1 through 6 but this time, set the role as account_owner. This time, the final result is again PERMIT which makes sense, given that the account_owner has access to both the resource and scope.

Answer 2

I know I am bit late to the party, but let me try to explain as much as I can.

In Keycloak, we have terms like:

Resource: object which users will be accessing or performing the action on

Auth scopes: Actions that users can perform on the specific object

Policies: Policy

Permission: Mapping actually occur here

If you don't want to follow manual way you can export this JSON and all the users, resources, permissions will be auto-set bu keycloak

JSON configuration file

Now let's see a scenario where:

Now we have few resources like:

1. Account
2. Bot
3. Report

We want to implement the scenario where only specific user can performe specific actions.

Setting up the Keycloak

Create a new realm

1. Click on the Add Realm button.

   

2. Enter test-v1 as the name.

3. Click on Create.

   

Create new roles

1. Click on the Roles

2. Click on Add Role

3. Create the roles "admin", "agent" & "super_admin"

   

Create a client

• Click on the Clients tab

• Enter app-client in Client ID textbox

• Click on Save

• Select and choose client again to configure other settings

• Verify the client Protocol is openid-connect

• Set Access Type to confidential

• Set Authorization Enabled to On

• Click on the Save button at last.

• A new Authorization tab will appear at the top.

• Select on the Authorization tab and then Settings

  Check that the Decision Strategy is set to Unanimous. This is the resource server strategy

  

Create Custom Auth Scopes

Go the Authorization tab Select Authorization Scopes → and click on Create Enter scopes:create & scopes:view in the text and save values.

Create Resource

• Go to Resources tab now → and click on Create

• Enter one by one and create the following resources res:account & res:bot & res:report

• For all resources in scope text, select both scopes that we created early scopes:create & scopes:view

• Click Save

  

Create policies

• Again inside the Authorization tab, select on Policies

• Click on the Create Policy dropdown and select the Role

• In the Name textbox, Admin

• In Realm Roles select role Admin

• Check Logic is set to Positive

• Click Save, and do the same for "Agent" & "Super_admin"

  

• Again inside the Authorization tab, select on Policies

• Click on the Create Policy dropdown and select the Aggregated

• In the Name textbox, Admin or Super_admin or Agent

• In Realm Roles select role Admin & Super_admin & Agent

• Check Logic is set to Affirmative

• Click Save

  

Create permission

• Again inside the Authorization tab, select on Permission

• Click on the Create Permissions dropdown and select the Scope-Based

• In the Name textbox, account-create

• In resource box, select the "resource res:account"

• In scopes select, scopes:create

• Apply policy Admin

• We have to set permission same way for all the resources as per requirement

  

Create user

• Inside the user tab create one test user We will not assign any roles, scopes, or group to it for testing

Let's Evaluate

• Again inside the Authorization tab, select on Evaluate

• Select the client we created, app-client

• In user select the created user, test

• In roles select the created user, admin

• Resource value, res:account

• Click on the Add button

• Click on the Evaluate button

  

  You will see grant is permitted as Admin role has access to do operations create and view on resource account.

Let's Evaluate Again

• Again inside the Authorization tab, select on Evaluate

• Select the client we created, app-client

• In user select the created user, test

• In roles select the created user, admin

• Resource value, res:report

• Scopes value, scopes:create

• Click on Add button

• Click on Evaluate button

  

You will see grant is Deny as only Super_Admin role has access to do operations create on resource report.

Answer 3

I was looking to enforce authorization via pure HTTP methods, without using the adapter as Lua did not have an adapter. Hope this answer helps people looking for non-adapter based method.

If you are looking for the adapter the quick start guide is the best place to start. Especially the spring boot authz example.

For pure HTTP based implementation:

Step 1:

Define the policies and permission in the Keycloak Admin UI

Step 2

Have an internal mapping of which HTTP paths belong to which resources and the required scopes for each path. This can be also saved in the configuration file. When a particular route is invoked, call the Keycloak token endpoint to validate the claims of the access token.

{
  "policy-enforcer": {
    "user-managed-access" : {},
    "enforcement-mode" : "ENFORCING"
    "paths": [
      {
        "path" : "/someUri/*",
        "methods" : [
          {
            "method": "GET",
            "scopes" : ["urn:app.com:scopes:view"]
          },
          {
            "method": "POST",
            "scopes" : ["urn:app.com:scopes:create"]
          }
        ]
      }
    ]
  }
}

If you are using an adapter and does not specify the path or resource, the adapter will internally search for the paths and resources from Keycloak.

Step 3:

Use the token endpoint to get or evaluate the permissions. You can use response_mode parameter to either obtain the final decision (whether to provide access) or retrieve the entire permissions.

curl -X POST \
  http://${host}:${port}/auth/realms/${realm}/protocol/openid-connect/token \
  -H "Authorization: Bearer ${access_token}" \
  --data "grant_type=urn:ietf:params:oauth:grant-type:uma-ticket" \
  --data "permission=Resource A#Scope A"

If the authorization request does not map to any permission, a 403 HTTP status code is returned instead.

Answer 4

A working solution with resources, scopes and permissions can be found here keycloak-nodejs-example

Just run already configured Keycloak using docker-compose using Quick Start guide.

Other useful examples in the project

• Custom login without using Keycloak login page.
• Stateless Node.js server without using a session. Keycloak token is stored using cookies.
• A centralized middleware to check permissions. Routes are not described explicitly can't be accessed.
• Configuration without keycloak.json. It can be used to having configuration for multiple environments. For example — DEV, QA.
• Examples of using Keycloak REST API to create users, roles and custom attributes.