Atom feed of this document
Kilo -  Kilo -  Kilo -  Kilo -  Kilo -  Kilo -  Kilo -  Kilo - 

 Appendix A. The policy.json file

Each OpenStack service, Identity, Compute, Networking and so on, has its own role-based access policies. They determine which user can access which objects in which way, and are defined in the service's policy.json file.

Whenever an API call to an OpenStack service is made, the service's policy engine uses the appropriate policy definitions to determine if the call can be accepted. Any changes to policy.json are effective immediately, which allows new policies to be implemented while the service is running.

A policy.json file is a text file in JSON (Javascript Object Notation) format. Each policy is defined by a one-line statement in the form "<target>" : "<rule>".

The policy target, also named "action", represents an API call like "start an instance" or "attach a volume".

Action names are usually qualified. Example: OpenStack Compute features API calls to list instances, volumes and networks. In /etc/nova/policy.json, these APIs are represented by compute:get_all, volume:get_all and network:get_all, respectively.

The mapping between API calls and actions is not generally documented.

The policy rule determines under which circumstances the API call is permitted. Usually this involves the user who makes the call (hereafter named the "API user") and often the object on which the API call operates. A typical rule checks if the API user is the object's owner.

[Caution]Modifying the policy

While recipes for editing policy.json files are found on blogs, modifying the policy can have unexpected side effects and is not encouraged.


A simple rule might look like this:

"compute:get_all" : ""

The target is "compute:get_all", the "list all instances" API of the Compute service. The rule is an empty string meaning "always". This policy allows anybody to list instances.

You can also decline permission to use an API:

"compute:shelve": "!"

The exclamation mark stands for "never" or "nobody", which effectively disables the Compute API "shelve an instance".

Many APIs can only be called by admin users. This can be expressed by the rule "role:admin". The following policy ensures that only administrators can create new users in the Identity database:

"identity:create_user" : "role:admin"

You can limit APIs to any role. For example, the Orchestration service defines a role named heat_stack_user. Whoever has this role isn't allowed to create stacks:

"stacks:create": "not role:heat_stack_user"

This rule makes use of the boolean operator not. More complex rules can be built using operators and, or and parentheses.

You can define aliases for rules:

"deny_stack_user": "not role:heat_stack_user"

The policy engine understands that "deny_stack_user" is not an API and consequently interprets it as an alias. The stack creation policy above can then be written as:

"stacks:create": "rule:deny_stack_user"

This is taken verbatim from /etc/heat/policy.json.

Rules can compare API attributes to object attributes. For example:

"compute:start" : "user_id:%(user_id)s"

states that only the owner of an instance can start it up. The user_id string before the colon is an API attribute, namely the user ID of the API user. It is compared with the user ID of the object (in this case, an instance); more precisely, it is compared with the user_id field of that object in the database. If the two values are equal, permission is granted.

An admin user always has permission to call APIs. This is how /etc/keystone/policy.json makes this policy explicit:

"admin_required": "role:admin or is_admin:1",
"owner" : "user_id:%(user_id)s",
"admin_or_owner": "rule:admin_required or rule:owner",
"identity:change_password": "rule:admin_or_owner"

The first line defines an alias for "user is an admin user". The is_admin flag is only used when setting up the Identity service for the first time. It indicates that the user has admin privileges granted by the service token (--os-token parameter of the keystone command line client).

The second line creates an alias for "user owns the object" by comparing the API's user ID with the object's user ID.

Line 3 defines a third alias admin_or_owner, combining the two first aliases with the Boolean operator or.

Line 4 sets up the policy that a password can only be modified by its owner or an admin user.

As a final example, let's examine a more complex rule:

"identity:ec2_delete_credential": "rule:admin_required or
             (rule:owner and user_id:%(target.credential.user_id)s)"

This rule determines who can use the Identity API "delete EC2 credential". Here, boolean operators and parentheses combine three simpler rules. admin_required and owner are the same aliases as in the previous example. user_id:%(target.credential.user_id)s compares the API user with the user ID of the credential object associated with the target.


A policy.json file consists of policies and aliases of the form target:rule or alias:definition, separated by commas and enclosed in curly braces:

       "alias 1" : "definition 1",
       "alias 2" : "definition 2",
       "target 1" : "rule 1",
       "target 2" : "rule 2",

Targets are APIs and are written "service:API" or simply "API". For example, "compute:create" or "add_image".

Rules determine whether the API call is allowed.

Rules can be:

  • always true. The action is always permitted. This can be written as "" (empty string), [], or "@".

  • always false. The action is never permitted. Written as "!".

  • a special check

  • a comparison of two values

  • boolean expressions based on simpler rules

Special checks are

  • <role>:<role name>, a test whether the API credentials contain this role.

  • <rule>:<rule name>, the definition of an alias.

  • http:<target URL>, which delegates the check to a remote server. The API is authorized when the server returns True.

Developers can define additional special checks.

Two values are compared in the following way:

"value1 : value2"

Possible values are

  • constants: Strings, numbers, true, false

  • API attributes

  • target object attributes

  • the flag is_admin

API attributes can be project_id, user_id or domain_id.

Target object attributes are fields from the object description in the database. For example in the case of the "compute:start" API, the object is the instance to be started. The policy for starting instances could use the %(project_id)sattribute, that is the project that owns the instance. The trailing s indicates this is a string.

is_admin indicates that administrative privileges are granted via the admin token mechanism (the --os-token option of the keystone command). The admin token allows initialisation of the identity database before the admin role exists.

The alias construct exists for convenience. An alias is short name for a complex or hard to understand rule. It is defined in the same way as a policy:

alias name : alias definition

Once an alias is defined, use the rule keyword to use it in a policy rule.

 Older syntax

You may encounter older policy.json files that feature a different syntax, where JavaScript arrays are used instead of boolean operators. For example, the EC2 credentials rule above would have been written as follows:

"identity:ec2_delete_credential": [ [ "rule:admin_required ],
             [ "rule:owner", "user_id:%(target.credential.user_id)s)" ] ]

The rule is an array of arrays. The innermost arrays are or'ed together, whereas elements inside the innermost arrays are and'ed.

While the old syntax is still supported, we recommend using the newer, more intuitive syntax.

Questions? Discuss on
Found an error? Report a bug against this page

loading table of contents...