Skip to content
|
search
Plugin Hub
github-edit-pageEdit this page
report-issueReport an issue
header icon

OpenID Connect

By Kong Inc.

OpenID Connect (1.0) plugin allows for integration with a third party identity provider (IdP) in a standardized way. This plugin can be used to implement Kong as a (proxying) OAuth 2.0 resource server (RS) and/or as an OpenID Connect relying party (RP) between the client, and the upstream service.

The plugin supports several types of credentials and grants:

  • Signed JWT access tokens (JWS)
  • Opaque access tokens
  • Refresh tokens
  • Authorization code with client secret or PKCE
  • Username and password
  • Client credentials
  • Session cookies

The plugin has been tested with several OpenID Connect providers:

As long as your provider supports OpenID Connect standards, the plugin should work, even if it is not specifically tested against it. Let Kong know if you want your provider to be tested and added to the list.

Once applied, any user with a valid credential can access the Service.

This plugin can be used for authentication in conjunction with the Application Registration plugin.

Important Configuration Parameters

This plugin includes many configuration parameters that allow finely grained customization. The following steps will help you get started setting up the plugin:

  1. Configure: config.issuer.

    This parameter tells the plugin where to find discovery information, and it is the only required parameter. You should set the value realm or iss on this parameter if you don’t have a discovery endpoint.

    Note: This does not have to match the URL of the iss claim in the access tokens being validated. To set URLs supported in the iss claim, use config.issuers_allowed.

  2. Decide what authentication grants to use with this plugin and configure the config.auth_methods field accordingly.

    In order to restrict the scope of potential attacks, the parameter should only contain the grants that you want to use.

  3. In many cases, you also need to specify config.client_id, and if your identity provider requires authentication, such as on a token endpoint, you will need to specify the client authentication credentials too, for example config.client_secret.

  4. If you are using a public identity provider, such as Google, you should limit the audience with config.audience_required to contain only your config.client_id. You may also need to adjust config.audience_claim in case your identity provider uses a non-standard claim (other than aud as specified in JWT standard). This is important because some identity providers, such as Google, share public keys with different clients.

  5. If you are using Kong in DB-less mode with a declarative configuration and session cookie authentication, you should set config.session_secret. Leaving this parameter unset will result in every Nginx worker across your nodes encrypting and signing the cookies with their own secrets.

In summary, start with the following parameters:

  1. config.issuer
  2. config.auth_methods
  3. config.client_id (and in many cases the client authentication credentials)
  4. config.audience_required (if using a public identity provider)
  5. config.session_secret (if using Kong in DB-less mode)

Records

In the above parameter list, two configuration settings used an array of records as a data type:

  • config.client_jwk: array of JWK records (one for each client)
  • config.session_redis_cluster_nodes: array of host records, either as IP addresses or hostnames, and their ports.

Below are descriptions of the record types.

JWK Record

The JSON Web Key (JWK) record is specified in RFC7517. This record is used with the config.client_jwk when using private_key_jwk client authentication.

Here is an example of JWK record generated by the plugin itself (see: JSON Web Key Set):

{
    "kid": "B2FxBJ8G_e61tnZEfaYpaMLjswjNO3dbVEQhR7-i_9s",
    "kty": "RSA",
    "alg": "RS256",
    "use": "sig"
    "e": "AQAB",
    "n": "…",
    "d": "…",
    "p": "…",
    "q": "…",
    "dp": "…",
    "dq": "…",
    "qi": "…"
}

The JWK private fields (k, d, p, q, dp, dq, qi, oth, r, t) are referenceable, which means they can be securely stored as a secret in a vault. References must follow a specific format.

Host Record

The Host record used with the config.session_redis_cluster_nodes is simple. It contains ip or host, and the port where the port defaults to 6379.

Here is an example of Host the record:

{
    "ip": "127.0.0.1"
    "port": 6379
}

Admin API

The OpenID Connect plugin extends the Kong Admin API with a few endpoints.

Discovery Cache

When configuring the plugin using config.issuer, the plugin will store the fetched discovery information to the Kong database, or in the worker memory with DB-less. The discovery cache does not have an expiry or TTL, and so must be cleared manually using the DELETE endpoints listed below.

The discovery cache will attempt to be refreshed when a token is presented with required discovery information that is not already available, based on the config.issuer value. Once a rediscovery attempt has been made, a new attempt will not occur until the number of seconds defined in rediscovery_lifetime has elapsed - this avoids excessive discovery requests to the identity provider.

If a JWT cannot be validated due to missing discovery information and an invalid status code is received from the rediscovery request (for example, non-2xx), the plugin will attempt to validate the JWT by falling back to any sufficient discovery information that is still in the discovery cache.

Discovery Cache Object
{
    "id": "<uuid>",
    "issuer": "<config.issuer>"
    "created_at": <timestamp>,
    "configuration": {
        <discovery>
    },
    "keys": [
        <keys>
    ]
}

List All Discovery Cache Objects

/openid-connect/issuers
Response
HTTP 200 OK

{
    "data": [{
        "id": "<uuid>",
        "issuer": "<config.issuer>"
        "created_at": <timestamp>,
        "configuration": {
            <discovery>
        },
        "keys": [
            <keys>
        ]
    }],
    "next": null
}

Retrieve Discovery Cache Object

/openid-connect/issuers/{issuer or id}
Attributes Description
issuer or id
required 		The unique identifier or the value of config.issuer
Response
HTTP 200 OK

{
    "id": "<uuid>",
    "issuer": "<config.issuer>"
    "created_at": <timestamp>,
    "configuration": {
        <discovery>
    },
    "keys": [
        <keys>
    ]
}

Delete All Discovery Cache Objects

/openid-connect/issuers
Response
HTTP 204 No Content

Note: The automatically generated session secret (that can be overridden with the config.session_secret) is stored with the discovery cache objects. Deleting discovery cache objects will invalidate all the sessions created with the associated secret.

Delete Discovery Cache Object

/openid-connect/issuers/{issuer or id}
Attributes Description
issuer or id
required 		The unique identifier or the value of config.issuer
Response
HTTP 204 No Content

JSON Web Key Set

When the OpenID Connect client (the plugin) is set to communicate with the identity provider endpoints using private_key_jwt, the plugin needs to use public key cryptography. Thus, the plugin needs to generate the needed keys. Identity provider on the other hand has to verify that the assertions used for the client authentication.

The plugin will automatically generate the key pairs for the different algorithms. It will also publish the public keys with the admin api where the identity provider could fetch them.

{
    "keys": [{
        <keys>
    }]
}

Retrieve JWKS

/openid-connect/jwks

This endpoint will return a standard JWK Set document with the private keys stripped out.

Response
HTTP 200 OK

{
    "keys": [{
        <keys>
    }]
}

Rotate JWKS

/openid-connect/jwks

Deleting JWKS will also cause auto-generation of a new JWK set, so DELETE will actually cause a key rotation.

Response
HTTP 204 No Content

Preparations

The OpenID Connect plugin relies in most cases on a third party identity provider. In this section, we explain configuration of Keycloak and Kong.

All the *.test domains in the following examples point to the localhost (127.0.0.1 and/or ::1).

Keycloak Configuration

We use Keycloak as the identity provider in the following examples, but the steps will be similar in other standard identity providers. If you encounter difficulties during this phase, please refer to the Keycloak documentation.

  1. Create a confidential client kong with private_key_jwt authentication and configure Keycloak to download the public keys from the OpenID Connect Plugin JWKS endpoint:



  2. Create another confidential client service with client_secret_basic authentication. For this client, Keycloak will auto-generate a secret similar to the following: cf4c655a-0622-4ce6-a0de-d3353ef0b714. Enable the client credentials grant for the client:



  3. Create a verified user with the name: john and the non-temporary password: doe that can be used with the password grant:

Alternatively you can download the exported Keycloak configuration, and use it to configure the Keycloak. Please refer to Keycloak import documentation for more information.

You need to modify Keycloak standalone.xml configuration file, and change the socket binding from:

<socket-binding name="https" port="${jboss.https.port:8443}"/>

to

<socket-binding name="https" port="${jboss.https.port:8440}"/>

The Keycloak default https port conflicts with the default Kong TLS proxy port, and that can be a problem if both are started on the same host.

Kong Configuration

Create a Service

http -f put :8001/services/openid-connect url=http://httpbin.org/anything

HTTP/1.1 200 OK

{
    "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd",
    "name": "openid-connect",
    "protocol": "http",
    "host": "httpbin.org",
    "port": 80,
    "path": "/anything"
}

Create a Route

http -f put :8001/services/openid-connect/routes/openid-connect paths=/

HTTP/1.1 200 OK

{
    "id": "ac1e86bd-4bce-4544-9b30-746667aaa74a",
    "name": "openid-connect",
    "paths": [ "/" ]
}

Create a Plugin

You may execute this before patching the plugin (as seen on following examples) to reset the plugin configuration.

http -f put :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  name=openid-connect                                          \
  service.name=openid-connect                                  \
  config.issuer=http://keycloak.test:8080/auth/realms/master   \
  config.client_id=kong                                        \
  config.client_auth=private_key_jwt

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "issuer": "http://keycloak.test:8080/auth/realms/master",
        "client_id": [ "kong" ],
        "client_auth": [ "private_key_jwt" ]
    }
}

Also, check the discovery cache: http :8001/openid-connect/issuers. It should contain Keycloak OpenID Connect discovery document, and the keys.

Summary

At this point we have:

  1. Created a Service
  2. Routed traffic to the service
  3. Enabled OpenID Connect plugin on the service

The following sections will guide you through the process of enabling the OpenID Connect plugin for specific grants or flows.

Authentication

Before you proceed, check that you have done the preparations.

We use HTTPie to execute the examples. The output is stripped for a better readability. httpbin.org is used as an upstream service.

Using Admin API is convenient when testing the plugin, but similar configs can be done in declarative format as well.

When this plugin is configured with multiple grants/flows there is a hard-coded search order for the credentials:

  1. Session Authentication
  2. JWT Access Token Authentication
  3. Kong OAuth Token Authentication
  4. Introspection Authentication
  5. User Info Authentication
  6. Refresh Token Grant
  7. Password Grant
  8. Client Credentials Grant
  9. Authorization Code Flow

Once credentials are found, the plugin will stop searching further. Multiple grants may share the same credentials. For example, both the password and client credentials grants can use basic access authentication through the Authorization header.

The choices made in the examples below are solely aimed at simplicity. Because httpbin.org is used as an upstream service, it is highly recommended that you do not run these usage examples with a production identity provider as there is a great chance of leaking information. Also the examples below use the plain HTTP protocol that you should never use in production.

Authorization Code Flow

The authorization code flow is the three-legged OAuth/OpenID Connect flow. The sequence diagram below describes the participants and their interactions for this usage scenario, including the use of session cookies:

Authorization code flow diagram

If using PKCE, the identity provider must contain the code_challenge_methods_supported object in the /.well-known/openid-configuration issuer discovery endpoint response, as required by RFC 8414. If it is not included, the PKCE code_challenge query parameter will not be sent.

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the authorization code flow and the session authentication.
  2. We want to set the response mode to form_post so that authorization codes won’t get logged to the access logs.
  3. We want to preserve the original request query arguments over the authorization code flow redirection.
  4. We want to redirect the client to original request url after the authorization code flow so that the POST request (because of form_post) is turned to the GET request, and the browser address bar is updated with the original request query arguments.
  5. We don’t want to include any tokens in the browser address bar.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=authorization_code                         \
  config.auth_methods=session                                    \
  config.response_mode=form_post                                 \
  config.preserve_query_args=true                                \
  config.login_action=redirect                                   \
  config.login_tokens=

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "authorization_code",
            "session"
        ],
        "login_action": "redirect",
        "preserve_query_args": true,
        "login_tokens": null
    }
}

Test the Authorization Code Flow

  1. Open the Service Page with some query arguments: 
    open http://service.test:8000/?hello=world

  2. See that the browser is redirected to the Keycloak login page:


    You may examine the query arguments passed to Keycloak with the browser developer tools.

  3. And finally you will be presented a response from httpbin.org:

  4. Done.

It looks rather simple from the user point of view, but what really happened is described in the diagram above.

Password Grant

Password grant is a legacy authentication grant. This is a less secure way of authenticating end users than the authorization code flow, because, for example, the passwords are shared with third parties. The image below illustrates the grant:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the password grant.
  2. We want to search credentials for password grant from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password                                   \
  config.password_param_type=header

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "password_param_type": [ "header" ]
    }
}

Test the Password Grant

  1. Request the service with basic authentication credentials created in the Keycloak configuration step: 
    http -v -a john:doe :8000

    GET / HTTP/1.1
Authorization: Basic BEkg3bHT0ERXFmKr1qelBQYrLBeHb5Hr

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Done.

If you make another request using the same credentials, you should see that Kong adds less latency to the request as it has cached the token endpoint call to Keycloak.

Client Credentials Grant

The client credentials grant is very similar to the password grant. The most important difference in the Kong OpenID Connect plugin is that the plugin itself does not try to authenticate. It just forwards the credentials passed by the client to the identity server’s token endpoint. The client credentials grant is visualized below:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the client credentials grant.
  2. We want to search credentials for client credentials from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=client_credentials                         \
  config.client_credentials_param_type=header

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "client_credentials" ],
        "client_credentials_param_type": [ "header" ]
    }
}

Test the Client Credentials Grant

  1. Request the service with client credentials created in the Keycloak configuration step: 
    http -v -a service:cf4c655a-0622-4ce6-a0de-d3353ef0b714 :8000

    GET / HTTP/1.1
Authorization: Basic c2VydmljZTpjZjRjNjU1YS0wNjIyLTRjZTYtYTBkZS1kMzM1M2VmMGI3MTQ=

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Done.

If you make another request using the same credentials, you should see that Kong adds less latency to the request as it has cached the token endpoint call to Keycloak.

Refresh Token Grant

The refresh token grant can be used when the client has a refresh token available. There is a caveat with this: identity providers in general only allow refresh token grant to be executed with the same client that originally got the refresh token, and if there is a mismatch, it may not work. The mismatch is likely when Kong OpenID Connect is configured to use one client, and the refresh token is retrieved with another. The grant itself is very similar to password grant and client credentials grant:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the refresh token grant, but we also enable the password grant for demoing purposes
  2. We want to search the refresh token for the refresh token grant from the headers only.
  3. We want to pass refresh token from the client in Refresh-Token header.
  4. We want to pass refresh token to upstream in Refresh-Token header.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.refresh_token_param_name=refresh_token                  \
  config.refresh_token_param_type=header                         \
  config.auth_methods=refresh_token                              \
  config.auth_methods=password                                   \
  config.upstream_refresh_token_header=refresh_token

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "refresh_token",
            "password"
        ],
        "refresh_token_param_name": "refresh_token",
        "refresh_token_param_type": [ "header" ],
        "upstream_refresh_token_header": "refresh_token"
    }
}

The config.auth_methods and config.upstream_refresh_token_header are only enabled for demoing purposes so that we can get a refresh token with:

http -a john:doe :8000 | jq -r '.headers."Refresh-Token"'

Output:

<refresh-token>

We can use the output in Refresh-Token header.

Test the Refresh Token Grant

  1. Request the service with a bearer token: 
    http -v :8000 Refresh-Token:$(http -a john:doe :8000 | \
  jq -r '.headers."Refresh-Token"')

    or

    http -v :8000 Refresh-Token:"<refresh-token>"

    GET / HTTP/1.1
Refresh-Token: <refresh-token>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>",
        "Refresh-Token": "<refresh-token>"
    },
    "method": "GET"
}
  2. Done.

JWT Access Token Authentication

For legacy reasons, the stateless JWT Access Token authentication is named bearer with the Kong OpenID Connect plugin (see: config.auth_methods). Stateless authentication basically means the signature verification using the identity provider published public keys and the standard claims’ verification (such as exp (or expiry)). The client may have received the token directly from the identity provider or by other means. It is simple:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the bearer authentication, but we also enable the password grant for demoing purposes
  2. We want to search the bearer token for the bearer authentication from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.bearer_token_param_type=header                          \
  config.auth_methods=bearer                                     \
  config.auth_methods=password # only enabled for demoing purposes

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "bearer",
            "password"
        ],
        "bearer_token_param_type": [ "header" ]
    }
}

The password grant is enabled so that we can get a JWT access token that we can use to show how the JWT access token authentication works. That is: we need a token. One way to get a JWT access token is to issue the following call (we use jq to filter the response):

http -a john:doe :8000 | jq -r .headers.Authorization

Output:

Bearer <access-token>

We can use the output in Authorization header.

Test the JWT Access Token Authentication

  1. Request the service with a bearer token: 
    http -v :8000 Authorization:"$(http -a john:doe :8000 | \
  jq -r .headers.Authorization)"

    or

    http -v :8000 Authorization:"Bearer <access-token>"

    GET / HTTP/1.1
Authorization: Bearer <access-token>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Done.

Introspection Authentication

As with JWT Access Token Authentication), the introspection authentication relies on a bearer token that the client has already gotten from somewhere. The difference to stateless JWT authentication is that the plugin needs to call the introspection endpoint of the identity provider to find out whether the token is valid and active. This makes it possible to issue opaque tokens to the clients.

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the introspection authentication, but we also enable the password grant for demoing purposes
  2. We want to search the bearer token for the introspection from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.bearer_token_param_type=header                          \
  config.auth_methods=introspection                              \
  config.auth_methods=password # only enabled for demoing purposes

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "introspection",
            "password"
        ],
        "bearer_token_param_type": [ "header" ]
    }
}

Test the Introspection Authentication

  1. Request the service with a bearer token: 
    http -v :8000 Authorization:"$(http -a john:doe :8000 | \
  jq -r .headers.Authorization)"

    or

    http -v :8000 Authorization:"Bearer <access-token>"

    GET / HTTP/1.1
Authorization: Bearer <access-token>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Done.

User Info Authentication

The user info authentication uses OpenID Connect standard user info endpoint to verify the access token. In most cases it is preferable to use Introspection Authentication as that is meant for retrieving information from the token itself, whereas the user info endpoint is meant for retrieving information about the user for whom the token was given. The sequence diagram below looks almost identical to introspection authentication:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the user info authentication, but we also enable the password grant for demoing purposes
  2. We want to search the bearer token for the user info from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.bearer_token_param_type=header                          \
  config.auth_methods=userinfo                                   \
  config.auth_methods=password # only enabled for demoing purposes

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "userinfo",
            "password"
        ],
        "bearer_token_param_type": [ "header" ]
    }
}

Test the User Info Authentication

  1. Request the service with a bearer token: 
    http -v :8000 Authorization:"$(http -a john:doe :8000 | \
  jq -r .headers.Authorization)"

    or

    http -v :8000 Authorization:"Bearer <access-token>"

    GET / HTTP/1.1
Authorization: Bearer <access-token>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Done.

Kong OAuth Token Authentication

The OpenID Connect plugin can also verify the tokens issued by Kong OAuth 2.0 Plugin. This is very similar to third party identity provider issued JWT access token authentication or introspection authentication:

Prepare Kong OAuth Application

  1. Create a Consumer: 
    http -f put :8001/consumers/jane
  2. Create Kong OAuth Application for the consumer: 
    http -f put :8001/consumers/jane/oauth2/client \
  name=demo                                    \
  client_secret=secret                         \
  hash_secret=true
  3. Create a Route: 
    http -f put :8001/routes/auth paths=/auth
  4. Apply OAuth plugin to the Route: 
    http -f put :8001/plugins/7cdeaa2d-5faf-416d-8df5-533d1e4cd2c4 \
  name=oauth2                                                  \
  route.name=auth                                              \
  config.global_credentials=true                               \
  config.enable_client_credentials=true
  5. Test the token endpoint: 
    https -f --verify no post :8443/auth/oauth2/token \
  client_id=client                                \
  client_secret=secret                            \
  grant_type=client_credentials

    HTTP/1.1 200 OK

    {
    "access_token": "<access-token>",
    "expires_in": 7200,
    "token_type": "bearer"
}

At this point we should be able to retrieve a new access token with:

https -f --verify no post :8443/auth/oauth2/token \
  client_id=client                                \
  client_secret=secret                            \
  grant_type=client_credentials |                 \
  jq -r .access_token

Output:

<access-token>

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the Kong OAuth authentication
  2. We want to search the bearer token for the Kong OAuth authentication from the headers only.

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=kong_oauth2                                \
  config.bearer_token_param_type=header

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "kong_oauth2" ],
        "bearer_token_param_type": [ "header" ]
    }
}

Test the Kong OAuth Token Authentication

  1. Request the service with Kong OAuth token: 
    http -v :8000 Authorization:"Bearer $(https -f --verify no \
  post :8443/auth/oauth2/token                             \
  client_id=client                                         \
  client_secret=secret                                     \
  grant_type=client_credentials |                          \
  jq -r .access_token)"

    or

    http -v :8000 Authorization:"Bearer <access-token>"

    GET / HTTP/1.1
Authorization: Bearer <access-token>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>",
        "X-Consumer-Id": "<consumer-id>",
        "X-Consumer-Username": "jane"
    },
    "method": "GET"
}
  2. Done.

Session Authentication

Kong OpenID Connect plugin can issue a session cookie that can be used for further session authentication. To make OpenID Connect issue a session cookie, you need to first authenticate with one of the other grant / flows described above. In authorization code flow we already demonstrated session authentication when we used the redirect login action. The session authentication is described below:

Patch the Plugin

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the session authentication, but we also enable the password grant for demoing purposes

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=session                                    \
  config.auth_methods=password # only enabled for demoing purposes

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [
            "session",
            "password"
        ]
    }
}

Test the Session Authentication

  1. Request the service with basic authentication credentials (created in the Keycloak configuration step), and store the session: 
    http -v -a john:doe --session=john :8000

    GET / HTTP/1.1
Authorization: Basic BEkg3bHT0ERXFmKr1qelBQYrLBeHb5Hr

    HTTP/1.1 200 OK
Set-Cookie: session=<session-cookie>; Path=/; SameSite=Lax; HttpOnly

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  2. Make request with a session cookie (stored above): 
    http -v --session=john :8000

    GET / HTTP/1.1
Cookie: session=<session-cookie>

    HTTP/1.1 200 OK

    {
    "headers": {
        "Authorization": "Bearer <access-token>"
    },
    "method": "GET"
}
  3. Done.

If you want to disable session creation with some grants, you can use the config.disable_session.

Authorization

Before you proceed, check that you have completed the preparations.

The OpenID Connect plugin has several features to do coarse grained authorization:

  1. Claims based authorization
  2. ACL plugin authorization
  3. Consumer authorization

Claims Based Authorization

The following options can be configured to manage claims verification during authorization:

  1. config.scopes_claim and config.scopes_required
  2. config.audience_claim and config.audience_required
  3. config.groups_claim and config.groups_required
  4. config.roles_claim and config.roles_required

For example, the first configuration option, config.scopes_claim, points to a source, from which the value is retrieved and checked against the value of the second configuration option: config.scopes_required.

Let’s take a look at a JWT access token:

Reset the plugin configuration before patching.

  1. Patch the plugin to enable the password grant: 
    http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password
  2. Retrieve the content of an access token: 
    http -a john:doe :8000 | jq -r .headers.Authorization

    HTTP/1.1 200 OK

    Bearer <access-token>
  3. The signed JWT <access-token> (JWS) comes with three parts separated with a dot .: <header>.<payload>.<signature> (a JWS compact serialization format)
  4. We are interested with the <payload>, and you should have something similar to: 
    eyJleHAiOjE2MjI1NTY3MTMsImF1ZCI6ImFjY291bnQiLCJ0eXAiOiJCZWFyZXIiLC
JzY29wZSI6Im9wZW5pZCBlbWFpbCBwcm9maWxlIiwicHJlZmVycmVkX3VzZXJuYW1l
Ijoiam9obiIsImdpdmVuX25hbWUiOiJKb2huIiwiZmFtaWx5X25hbWUiOiJEb2UifQ

    That can be base64 url decoded to the following JSON:

    {
    "exp": 1622556713,
    "aud": "account",
    "typ": "Bearer",
    "scope": "openid email profile",
    "preferred_username": "john",
    "given_name": "John",
    "family_name": "Doe"
}

    This payload may contain arbitrary claims, such as user roles and groups, but as we didn’t configure them in Keycloak, let’s just use the claims that we have. In this case we want to authorize against the values in scope claim.

Let’s patch the plugin that we created in the Kong configuration step:

  1. We want to only use the password grant for demonstration purposes.
  2. We require the value openid and email to be present in scope claim of the access token.
http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password                                   \
  config.scopes_claim=scope                                      \
  config.scopes_required="openid email"

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "scopes_claim": [ "scope" ],
        "scopes_required": [ "openid email" ]
    }
}

Now let’s see if we can still access the service:

http -v -a john:doe :8000

GET / HTTP/1.1
Authorization: Basic BEkg3bHT0ERXFmKr1qelBQYrLBeHb5Hr

HTTP/1.1 200 OK

{
   "headers": {
       "Authorization": "Bearer <access-token>"
   },
   "method": "GET"
}

Works as expected, but let’s try to add another authorization:

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.audience_claim=aud                                      \
  config.audience_required=httpbin

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "audience_claim": [ "scope" ],
        "audience_required": [ "httpbin" ]
    }
}

As we know, the access token has "aud": "account", and that does not match with "httpbin", so the request should now be forbidden:

http -v -a john:doe :8000

HTTP/1.1 403 Forbidden

{
    "message": "Forbidden"
}

A few words about config.scopes_claim and config.scopes_required (and the similar configuration options). You may have noticed that config.scopes_claim is an array of string elements. Why? It is used to traverse the JSON when looking up a claim, take for example this imaginary payload:

{
    "user": {
        "name": "john",
        "groups": [
            "employee",
            "marketing"
        ]
    }
}

In this case you would probably want to use config.groups_claim to point to groups claim, but that claim is not a top-level claim, so you need to traverse there:

  1. Find the user claim.
  2. Inside the user claim, find the groups claim, and read its value:
{
    "config": {
        "groups_claim": [
            "user",
            "groups"
        ]
    }
}

or

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.groups_claim=user                                       \
  config.groups_claim=groups

The value of a claim can be the following:

  • a space separated string (such as scope claim usually is)
  • an JSON array of strings (such as the imaginary groups claim above)
  • a simple value, such as a string

What about the config.groups_required then? That is also an array?

That is correct, the required checks are arrays to allow logical and/or type of checks:

  1. and: use a space separated values
  2. or: specify value(s) in separate array indices

For example:

{
    "config": {
        "groups_required": [
            "employee marketing",
            "super-admins"
        ]
    }
}

or

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.groups_required="employee marketing"                    \
  config.groups_required="super-admins"

The above means that a claim has to have:

  1. employee and marketing values in it, OR
  2. super-admins value in it

ACL Plugin Authorization

The plugin can also be integrated with Kong ACL Plugin that provides access control functionality in form of allow and deny lists.

Please read the claims verification section for a basic information, as a lot of that applies here too.

Let’s first configure the OpenID Connect plugin for integration with the ACL plugin (please remove other authorization if enabled):

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password                                   \
  config.authenticated_groups_claim=scope

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "authorized_groups_claim": [ "scope" ]
    }
}

Before we apply the ACL plugin, let’s try it once:

http -v -a john:doe :8000

HTTP/1.1 200 OK

{
    "headers": {
        "X-Authenticated-Groups": "openid, email, profile",
    }
}

Interesting, the X-Authenticated-Groups header was injected in a request. This means that we are all good to add the ACL plugin:

http -f put :8001/plugins/b238b64a-8520-4bbb-b5ff-2972165cf3a2 \
  name=acl                                                     \
  service.name=openid-connect                                  \
  config.allow=openid

Let’s test it again:

http -v -a john:doe :8000

HTTP/1.1 200 OK

Let’s make it forbidden by changing it to a deny-list:

http -f patch :8001/plugins/b238b64a-8520-4bbb-b5ff-2972165cf3a2 \
  config.allow=                                                  \
  config.deny=profile

And try again:

http -v -a john:doe :8000

HTTP/1.1 403 Forbidden

{
    "message": "You cannot consume this service"
}

Consumer Authorization

The third option for authorization is to use Kong consumers and dynamically map from a claim value to a Kong consumer. This means that we restrict the access to only those that do have a matching Kong consumer. Kong consumers can have ACL groups attached to them and be further authorized with the Kong ACL Plugin.

As a remainder our token payload looks like this:

{
   "exp": 1622556713,
   "aud": "account",
   "typ": "Bearer",
   "scope": "openid email profile",
   "preferred_username": "john",
   "given_name": "John",
   "family_name": "Doe"
}

Out of these the preferred_username claim looks promising for consumer mapping. Let’s patch the plugin:

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password                                   \
  config.consumer_claim=preferred_username                       \
  config.consumer_by=username

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "consumer_claim": [ "preferred_username" ],
        "consumer_by": [ "username" ]
    }
}

Before we proceed, let’s make sure we don’t have consumer john:

http delete :8001/consumers/john

HTTP/1.1 204 No Content

Let’s try to access the service without a matching consumer:

http -a john:doe :8000

HTTP/1.1 403 Forbidden

{
    "message": "Forbidden"
}

Now, let’s add the consumer:

http -f put :8001/consumers/john

HTTP/1.1 200 OK


{
    "id": "<consumer-id>",
    "username": "john"
}

Let’s try to access the service again:

http -a john:doe :8000

HTTP/1.1 200 OK

{
    "headers": {
        "Authorization": "Bearer <access-token>",
        "X-Consumer-Id": "<consumer-id>",
        "X-Consumer-Username": "john"
    },
    "method": "GET"
}

Nice, as you can see the plugin even added the X-Consumer-Id and X-Consumer-Username as request headers.

It is possible to make consumer mapping optional and non-authorizing by setting the config.consumer_optional=true.

Headers

Before you proceed, check that you have completed the preparations.

The OpenID Connect plugin can pass claim values, tokens, JWKs, and the session identifier to the upstream service in request headers, and to the downstream client in response headers. By default, the plugin passes an access token in Authorization: Bearer <access-token> header to the upstream service (this can be controlled with config.upstream_access_token_header). The claim values can be taken from:

  • an access token,
  • an id token,
  • an introspection response, or
  • a user info response

Let’s take a look for an access token payload:

{
   "exp": 1622556713,
   "aud": "account",
   "typ": "Bearer",
   "scope": "openid email profile",
   "preferred_username": "john",
   "given_name": "John",
   "family_name": "Doe"
}

To pass the preferred_username claim’s value john to the upstream with an Authenticated-User header, we need to patch our plugin:

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=password                                   \
  config.upstream_headers_claims=preferred_username              \
  config.upstream_headers_names=authenticated_user

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password" ],
        "upstream_headers_claims": [ "preferred_username" ],
        "upstream_headers_names": [ "authenticated_user" ]
    }
}

Let’s see if it had any effect:

http -a john:doe :8000

HTTP/1.1 200 OK

{
    "headers": {
        "Authorization": "Bearer <access-token>",
        "Authenticated-User": "john"
    },
    "method": "GET"
}

See the configuration parameters for other options.

Logout

The logout functionality is mostly useful together with session authentication that is mostly useful with authorization code flow.

As part of the logout, the OpenID Connect plugin implements several features:

  • session invalidation
  • token revocation
  • relying party (RP) initiated logout

Let’s patch the OpenID Connect plugin to provide the logout functionality:

Reset the plugin configuration before patching.

http -f patch :8001/plugins/5f35b796-ced6-4c00-9b2a-90eef745f4f9 \
  config.auth_methods=session                                    \
  config.auth_methods=password                                   \
  config.logout_uri_suffix=/logout                               \
  config.logout_methods=POST                                     \
  config.logout_revoke=true

HTTP/1.1 200 OK

{
    "id": "5f35b796-ced6-4c00-9b2a-90eef745f4f9",
    "name": "openid-connect",
    "service": {
        "id": "5fa9e468-0007-4d7e-9aeb-49ca9edd6ccd"
    },
    "config": {
        "auth_methods": [ "password", "session" ],
        "logout_uri_suffix": "/logout",
        "logout_methods": [ "POST" ],
        "logout_revoke": true
    }
}

Login and establish a session:

http -a john:doe --session=john :8000

HTTP/1.1 200 OK

Test that session authentication works:

http --session=john :8000

HTTP/1.1 200 OK

Logout, and follow the redirect:

http --session=john --follow -a john: post :8000/logout

HTTP/1.1 200 OK

We needed to pass -a john: as there seems to be a feature with HTTPie that makes it to store the original basic authentication credentials in a session - not just the session cookies.

At this point the client has logged out from both Kong and the identity provider (Keycloak).

Check that the session is really gone:

http --session=john :8000

HTTP/1.1 401 Unauthorized

Debugging

The OpenID Connect plugin is complex, integrating with third-party identity providers can present challenges. If you have issues with the plugin or integration, try the following:

  1. Set Kong log level to debug, and check the Kong error.log (you can filter it with openid-connect) 
    KONG_LOG_LEVEL=debug kong restart
  2. Set the Kong OpenID Connect plugin to display errors: 
    {
    "config": {
        "display_errors": true
    }
}
  3. Disable the Kong OpenID Connect plugin verifications and see if you get further, just for debugging purposes: 
    {
    "config": {
        "verify_nonce": false,
        "verify_claims": false,
        "verify_signature": false
    }
}
  4. See what kind of tokens the Kong OpenID Connect plugin gets: 
    {
    "config": {
        "login_action": "response",
        "login_tokens": [ "tokens" ],
        "login_methods": [
            "password",
            "client_credentials",
            "authorization_code",
            "bearer",
            "introspection",
            "userinfo",
            "kong_oauth2",
            "refresh_token",
            "session"
        ]
    }
}

With session related issues, you can try to store the session data in Redis or memcache as that will make the session cookie much smaller. It is rather common that big cookies do cause issues. You can also enable session compression.

Also try to eliminate indirection as that makes it easier to find out where the problem is. By indirection, we mean other gateways, load balancers, NATs, and such in front of Kong. If there is such, you may look at using: