Horizon Policy Enforcement (RBAC: Role Based Access Control)


Horizon’s policy enforcement builds on the oslo_policy engine. The basis of which is openstack_auth/policy.py. Services in OpenStack use the oslo policy engine to define policy rules to limit access to APIs based primarily on role grants and resource ownership.

The Keystone v3 API provides an interface for creating/reading/updating policy files in the keystone database. However, at this time services do not load the policy files into Keystone. Thus, the implementation in Horizon is based on copies of policy.json files found in the service’s source code. The long-term goal is to read/utilize/update these policy files in Horizon.

The service rules files are loaded into the policy engine to determine access rights to actions and service APIs.

Horizon Settings

There are a few settings that must be in place for the Horizon policy engine to work.


Default: os.path.join(ROOT_PATH, "conf")

Specifies where service based policy files are located. These are used to define the policy rules actions are verified against. This value must contain the files listed in POLICY_FILES or all policy checks will pass.


The path to deployment specific policy files can be specified in local_settings.py to override the default location.


Default: {'identity': 'keystone_policy.json', 'compute': 'nova_policy.json'}

This should essentially be the mapping of the contents of POLICY_FILES_PATH to service types. When policy.json files are added to the directory POLICY_FILES_PATH, they should be included here too. Without this mapping, there is no way to map service types with policy rules, thus two policy.json files containing a “default” rule would be ambiguous.


Deployment specific policy files can be specified in local_settings.py to override the default policy files. It is imperative that these policy files match those deployed in the target OpenStack installation. Otherwise, the displayed actions and the allowed action will not match.


Default: policy.check

This value should not be changed, although removing it would be a means to bypass all policy checks.

How user’s roles are determined

Each policy check uses information about the user stored on the request to determine the user’s roles. This information was extracted from the scoped token received from Keystone when authenticating.

Entity ownership is also a valid role. To verify access to specific entities like a project, the target must be specified. See the section rule targets later in this document.

How to Utilize RBAC

The primary way to add role based access control checks to panels is in the definition of table actions. When implementing a derived action class, setting the policy_rules attribute to valid policy rules will force a policy check before the horizon.tables.Action.allowed() method is called on the action. These rules are defined in the policy files pointed to by POLICY_PATH and POLICY_FILES. The rules are role based, where entity owner is also a role. The format for the policy_rules is a list of two item tuples. The first component of the tuple is the scope of the policy rule, this is the service type. This informs the policy engine which policy file to reference. The second component is the rule to enforce from the policy file specified by the scope. An example tuple is:

("identity", "identity:get_user")

x tuples can be added to enforce x rules.


If a rule specified is not found in the policy file, the policy check will return False and the action will not be allowed.

The secondary way to add a role based check is to directly use the check() method. The method takes a list of actions, same format as the policy_rules attribute detailed above; the current request object; and a dictionary of action targets. This is the method that horizon.tables.Action class utilizes. Examples look like:

from openstack_dashboard import policy

allowed = policy.check((("identity", "identity:get_user"),
                       ("identity", "identity:get_project"),), request)

can_see = policy.check((("identity", "identity:get_user"),), request,
                       target={"domain_id": domainId})


Any time multiple rules are specified in a single policy.check method call, the result is the logical and of each rule check. So, if any rule fails verification, the result is False.

The third way to add a role based check is in javascript files. Use the method ‘ifAllowed()’ in file ‘openstack_dashboard.static.app.core.policy.service.js’. The method takes a list of actions, similar format with the policy_rules attribute detailed above. An Example looks like:

.controller('identityUsersTableController', identityUsersTableController);

identityUsersTableController.$inject = [

function identityUsersTableController(toast, gettext, policy, keystone) {
  var rules = [['identity', 'identity:list_users']];
  policy.ifAllowed({ rules: rules }).then(policySuccess, policyFailed);

The fourth way to add a role based check is in html files. Use angular directive ‘hz-if-policies’ in file ‘openstack_dashboard.static.app.core.cloud-services.hz-if-policies-directive.js’. Assume you have the following policy defined in your angular controller:

ctrl.policy = { rules: [["identity", "identity:update_user"]] }

Then in your HTML, use it like so:

<div hz-if-policies='ctrl.policy'>
  <span>I am visible if the policy is allowed!</span>

Rule Targets

Some rules allow access if the user owns the entity. Policy check targets specify particular entities to check for user ownership. The target parameter to the check() method is a simple dictionary. For instance, the target for checking access a project looks like:

{"project_id": "0905760626534a74979afd3f4a9d67f1"}

If the value matches the project_id to which the user’s token is scoped, then access is allowed.

When deriving the horizon.tables.Action class for use in a table, if a policy check is desired for a particular target, the implementer should override the horizon.tables.Action.get_policy_target() method. This allows a programmatic way to specify the target based on the current datum. The value returned should be the target dictionary.