Using external authentication with Keystone

When Keystone is executed in a web server like Apache HTTPD it is possible to use external authentication methods different from the authentication provided by the identity store backend or the different authentication plugins. For example, this makes possible to use an SQL identity backend together with, X.509 authentication or Kerberos, for example, instead of using the username and password combination.

When a web server is in charge of authentication, it is normally possible to set the REMOTE_USER environment variable so that it can be used in the underlying application. Keystone can be configured to use that environment variable if set, so that the authentication is handled by the web server.

Configuration

In Identity API v2, there is no way to disable external authentication. In order to activate the external authentication mechanism for Identity API v3, the external method must be in the list of enabled authentication methods. By default it is enabled, so if you don’t want to use external authentication, remove it from the methods option in the auth section.

To configure the plugin that should be used set the external option again in the auth section. There are two external authentication method plugins provided by Keystone:

• DefaultDomain: This plugin won’t take into account the domain information that the external authentication method may pass down to Keystone and will always use the configured default domain. The REMOTE_USER variable is the username. This is the default if no plugin is given.
• Domain: This plugin expects that the REMOTE_DOMAIN variable contains the domain for the user. If this variable is not present, the configured default domain will be used. The REMOTE_USER variable is the username.

Using HTTPD authentication

Web servers like Apache HTTP support many methods of authentication. Keystone can profit from this feature and let the authentication be done in the web server, that will pass down the authenticated user to Keystone using the REMOTE_USER environment variable. This user must exist in advance in the identity backend to get a token from the controller.

To use this method, Keystone should be running on HTTPD.

X.509 example

The following snippet for the Apache conf will authenticate the user based on a valid X.509 certificate from a known CA:

<VirtualHost _default_:5000>
    SSLEngine on
    SSLCertificateFile    /etc/ssl/certs/ssl.cert
    SSLCertificateKeyFile /etc/ssl/private/ssl.key

    SSLCACertificatePath /etc/ssl/allowed_cas
    SSLCARevocationPath  /etc/ssl/allowed_cas
    SSLUserName          SSL_CLIENT_S_DN_CN
    SSLVerifyClient      require
    SSLVerifyDepth       10

    (...)
</VirtualHost>

Developing a WSGI middleware for authentication

In addition to the method described above, it is possible to implement other custom authentication mechanisms using the REMOTE_USER WSGI environment variable.

Attention

Please note that even if it is possible to develop a custom authentication module, it is preferable to use the modules in the HTTPD server. Such authentication modules in webservers like Apache have normally undergone years of development and use in production systems and are actively maintained upstream. Developing a custom authentication module that implements the same authentication as an existing Apache module likely introduces a higher security risk.

If you find you must implement a custom authentication mechanism, you will need to develop a custom WSGI middleware pipeline component. This middleware should set the environment variable REMOTE_USER to the authenticated username. Keystone then will assume that the user has been already authenticated upstream and will not try to authenticate it. However, as with HTTPD authentication, the user must exist in advance in the identity backend so that a proper token can be issued.

Your code should set the REMOTE_USER if the user is properly authenticated, following the semantics below:

from keystone.common import wsgi
from keystone import exception

class MyMiddlewareAuth(wsgi.Middleware):
    def __init__(self, *args, **kwargs):
        super(MyMiddlewareAuth, self).__init__(*args, **kwargs)

    def process_request(self, request):
        if request.environ.get('REMOTE_USER', None) is not None:
            # Assume that it is authenticated upstream
            return self.application

        if not self.is_auth_applicable(request):
            # Not applicable
            return self.application

        username = self.do_auth(request)
        if username is not None:
            # User is authenticated
            request.environ['REMOTE_USER'] = username
        else:
            # User is not authenticated, render exception
            raise exception.Unauthorized("Invalid user")

Pipeline configuration

Once you have your WSGI middleware component developed you have to add it to your pipeline. The first step is to add the middleware to your configuration file. Assuming that your middleware module is keystone.middleware.MyMiddlewareAuth, you can configure it in your keystone-paste.ini as:

[filter:my_auth]
paste.filter_factory = keystone.middleware.MyMiddlewareAuth.factory

The second step is to add your middleware to the pipeline. The exact place where you should place it will depend on your code (i.e. if you need for example that the request body is converted from JSON before perform the authentication you should place it after the json_body filter) but it should be set before the public_service (for the public_api pipeline) or admin_service (for the admin_api pipeline), since they consume authentication.

For example, if the original pipeline looks like this:

[pipeline:public_api]
pipeline = url_normalize token_auth admin_token_auth json_body debug ec2_extension user_crud_extension public_service

Your modified pipeline might then look like this:

[pipeline:public_api]
pipeline = url_normalize token_auth admin_token_auth json_body my_auth debug ec2_extension user_crud_extension public_service