Policy

A Policy enables developers to configure fine-grained access control for Things and other entities easily.

Authorization concept

A specific policy provides someone (called subject), permission to read and/or write a given resource.

Please note, that in most cases it makes sense to grant read permission in addition to write permission, because write does not imply read.

Model specification

Subjects

Subjects in a policy define who gets permissions granted/revoked on the resources of a policy entry.
Each subject ID contains a prefix defining the subject “issuer” (so which party issued the authentication) and an actual subject, separated with a colon:

<subject-issuer>:<subject>

The subject can be one of the following ones:

• nginx:<nginx-username> - when using nginx as pre-authentication provider - by default enabled in the Ditto installation’s nginx
• <other-pre-auth-provider>:<username> - when using another custom provider as pre-authentication provider which sets the x-ditto-pre-authenticated HTTP header

• google:<google-user-id> - in general different JWT - the currently supported are listed in the table:

  Prefix	Type	Description
  google	jwt	A JWT issued by Google

• <custom-openid-connect-provider>:<jwt-sub-claim> - custom OpenID Connect compliant providers - supported providers are listed at OpenID Connect - Certified OpenID Provider Servers and Services - can be configured in Ditto defining the prefix in Ditto’s config file.
  The sub claim from the JWT and the configured provider name are used in the form <provider>:<jwt-sub-claim>.

Expiring Policy subjects

When a Policy subject contains an "expiry" timestamp (formatted as ISO-8601 string), this subject will be automatically deleted once this timestamp was reached.

When providing an "expiry" for a Policy subject, this timestamp is rounded up:

• by default to the next full hour
• this is configurable via the environment variable POLICY_SUBJECT_EXPIRY_GRANULARITY of the policies service which takes a HOCON duration, e.g.:
  • configured to “1s”: a received “expiry” is rounded up to the next full second
  • configured to “30s”: a received “expiry” is rounded up to the next half minute
  • configured to “1h”: a received “expiry” is rounded up to the next full hour (default)
  • configured to “12h”: a received “expiry” is rounded up to the next half day
  • configured to “1d”: a received “expiry” is rounded up to the next full day
  • configured to “15d”: a received “expiry” is rounded up to the next half month

Once an expired subject is deleted, it will immediately no longer have access to the resources protected by the policy it was deleted from.

Subject deletion announcements

To get notified when a subject is deleted, the "announcement" object can be configured in the respective subject section.

{
  "type": "my-subject",
  "expiry": "2099-12-31T23:59:59Z",
  "announcement": {
    "beforeExpiry": "1h",
    "whenDeleted": true,
    "requestedAcks": {
      "labels": ["my-connection:my-issued-acknowledgement"],
      "timeout": "10s"
    }
  }
}

Here are the meanings of the fields of "announcement":

• "beforeExpiry": The duration before expiration of the subject when a subject deletion announcement should be published if no previous subject deletion announcement was acknowledged.
  Supported unit suffixes:
  • "ms": for milliseconds
  • "s": for seconds
  • "m": for minutes
  • "h": for hours
• "whenDeleted": Boolean value to describe whether a subject deletion announcement should be published whenever a subject is manually deleted (e.g. via overwrite of a policy entry) from a policy, if no previous subject deletion announcement was acknowledged.
• "requestedAcks": Settings for at-least-once delivery of announcements via acknowledgements:
  • "labels": Array of Requested acknowledgement labels of the websocket or connectivity channel from which the issued acknowledgement is expected.
  • "timeout": Time in minutes (1m), seconds (60s), or milliseconds (600ms) how long to wait for acknowledgements before retrying to publish a timed out announcement.

The subject deletion announcements are published to any websocket or connection that has subscribed for policy announcements and was authenticated with the relevant subject ID.

Actions

Policy actions are available via Ditto’s HTTP API and can be invoked for certain policy entries or for complete policies.

They require neither READ nor WRITE permission, but instead a granted EXECUTE permission on the specific action name, e.g. for a single policy entry:

• policy:/entries/{label}/actions/{actionName}

Action activateTokenIntegration

When authenticated using OpenID Connect, it is possible to inject a subject into policies that expires when the JWT expires. The form of the injected subject (the token integration subject) is configurable globally in the Ditto installation.

A user is authorized to inject the token integration subject when granted the EXECUTE permission on a policy entry.
The WRITE permission is not necessary. To activate or deactivate a token integration subject, send a POST request to the following HTTP routes:

• POST /api/2/policies/{policyId}/actions/activateTokenIntegration
  Injects a new subject into all matched policy entries calculated with information extracted from the authenticated JWT.
  • the authenticated token must be granted the EXECUTE permission to perform the activateTokenIntegration action
  • one of the subject IDs must be contained in the authenticated token
  • at least one READ permission to a thing:/ resource path must be granted
• POST /api/2/policies/{policyId}/actions/deactivateTokenIntegration
  Removes the calculated subject with information extracted from the authenticated JWT from all matched policy entries.
  • the authenticated token must be granted the EXECUTE permission to perform the deactivateTokenIntegration action
  • one of the subject IDs must be contained in the authenticated token
• POST /api/2/policies/{policyId}/entries/{label}/actions/activateTokenIntegration
  Injects the calculated subject into the policy entry calculated with information extracted from the authenticated JWT.
  • the authenticated token must be granted the EXECUTE permission to perform the activateTokenIntegration action
  • one of the subject IDs must be contained in the authenticated token
  • at least one READ permission to a thing:/ resource path must be granted
• POST /api/2/policies/{policyId}/entries/{label}/actions/deactivateTokenIntegration
  Removes the calculated subject with information extracted from the authenticated JWT from the policy entry.
  • the authenticated token must be granted the EXECUTE permission to perform the deactivateTokenIntegration action
  • one of the subject IDs must be contained in the authenticated token

The injected subject pattern is configurable in Ditto and is by default:

integration:{{policy-entry:label}}:{{jwt:aud}}

To configure the token integration subject, set the path

ditto.gateway.authentication.oauth.token-integration-subject

in gateway-extension.conf, or set the environment variable OAUTH_TOKEN_INTEGRATION_SUBJECT for Gateway Service.

ditto.gateway.authentication.oauth.token-integration-subject =
  "my-token-integration-issuer:{{policy-entry:label}}:{{jwt:sub}}"

ditto.gateway.authentication.oauth.token-integration-subject =
  ${?OAUTH_TOKEN_INTEGRATION_SUBJECT}

The placeholders below are usable as a part of the activateTokenIntegration configuration:

Which Resources can be controlled?

A Policy can contain access control definitions for several resources:

• Policy: Someone who was granted write permission at the policy root resource (i.e. policy:/) is allowed to manage the policy itself.
• Thing: The resource can be defined as fine-grained as necessary for the respective use case: e.g. thing:/ as top-level resource or on sub-resources such as thing:/features. At runtime, the permissions are propagated down to all Thing sub-entities.
  • In case you grant read permission on top-level and revoke it at a sub-entity, the subject can read the upper part only.
  • In case you omit a subject at top-level but grant permission at a sub-entity, the subject can access the lower part only (and the Thing ID).

Policy

The Policy resource (addressable as policy:/) defines the access control for the Policy itself.

The Things example at the end of the page also defines access control on the policy resource.

Thing

The Thing resource (addressable as thing:/) defines the access control for Things.

The access control definitions defined in a Policy’s Thing resource will be applied to all Things referencing this Policy.

Find a Things example at the end of the page.

Feature

Find a Things example at the end of the page.

Message

The Message resource (addressable as message:/) defines the access control for Messages.

The access control definitions defined in a Policy’s Message resource will be applied to all Messages sent to or from Things referencing this Policy.

• For sending messages to a Thing or its Features write permission is required
• For receiving messages from a Thing or its Features read permission is required.

Such permissions can be defined at resources of different granularity.

The Things example at the end of the page also defines access control on messages.

Grant and Revoke some Permission

Policy imports

With policy imports it is possible to import entries from other referenced policies. Which parts of the referenced policy are imported is controlled by two properties of a policy.

Firstly, the imported policy can define for each entry whether and how a policy entry is importable by others using JSON field importable. The field can have one of the following three values:

1. implicit (default): the policy entry is imported without being listed in the importing policy individually
2. explicit: the policy entry is only imported if it is listed in the importing policy
3. never: the policy entry is not imported, regardless of being listed in the importing policy

If the field is not specified, the default value is implicit.

Example of a policy specifying different types of importable entries:

{
  "entries": {
    "DEFAULT": {
      "subjects": { ... },
      "resources": { ... }
    },
    "IMPLICIT": {
      "subjects": { ... },
      "resources": { ... },
      "importable": "implicit"
    },
    "EXPLICIT": {
      "subjects": { ... },
      "resources": { ... },
      "importable": "explicit"
    },
    "NEVER": {
      "subjects": { ... },
      "resources": { ... },
      "importable": "never"
    }
  }
}

Secondly, the importing policy may define a set of entries (identified by their label) it wants to import in addition to those entries that are implicitly imported.

Example of a policy importing two other policies:

{
  "policyId": "ditto:importing-policy",
  "entries": {  ...  },
  "imports": {
    "ditto:imported-policy" : {
      // import the "EXPLICIT" entry and entries that are of importable type implicit      
      "entries": [ "EXPLICIT" ] 
    },
    "ditto:another-imported-policy" : { } // import only entries that are of importable type implicit
  }
}

A subject creating or modifying a policy with policy imports must have the following permissions:

• permission on the importing policy to WRITE the modified policy import or policy imports
• permission on the imported policy to READ entries that are implicitly or explicitly referenced in the policy imports

The entries of the importing policy and the entries of the imported policy are merged at runtime and evaluated as if the entries were defined in one single policy. The same rules for the evaluation of grant/revoke permissions apply to imported entries in the same way as for entries defined in the importing policy itself. This also means that changes to an imported policy are reflected in all policies that import it. For direct access to the entities protected by a policy (things, policies, …) via e.g. HTTP this happens instantaneously. When searching things it may take some time to sync the changes as the search index is eventually consistent. This is particularly the case if the changed policy is imported by a large number of other policies.

Limitations

When managing and using policy imports the following limitations apply:

• The maximum number of policy imports allowed per policy is 10.
• To avoid conflicts with imported entries, it is not allowed to use the prefix imported for the name of a policy entry label. Trying to do so will result in an error.

Tools for editing a Policy

The Policy can be edited with a text editor of your choice. Just make sure it is in valid JSON representation, and that at least one valid subject is granted write permission at the root resources.

In case of fine-grained access on Things, keep an eye on your actual Thing structure to make sure that all paths will be granted or revoked the permissions your use case is supposed to support.

Example

Given you need to support the following scenario:

• Owner: The Thing my.namespace:thing-0123 is owned by a user. Thus, she needs full access and admin rights for the complete Thing. In our example her ID is ditto
• Observer of changes at featureX and featureY:
  • Another application needs to be informed on each change at those features. In our example its ID is observer-client.
  • There is a group of users who are allowed to read both features. In our example the group ID is some-users.
• Privacy: The value of the “city” property at “featureY” is confidential and needs to be “hidden” from the group of users.

Your Policy then might look like the following:

The correct Policy JSON object notation would be as shown in the following code block.

{
  "policyId": "my.namespace:policy-a",
  "entries": {
    "owner": {
      "subjects": {
        "nginx:ditto": {
          "type": "nginx basic auth user"
        }
      },
      "resources": {
        "thing:/": {
          "grant": ["READ", "WRITE"],
          "revoke": []
        },
        "policy:/": {
          "grant": ["READ", "WRITE"],
          "revoke": []
        },
        "message:/": {
          "grant": ["READ", "WRITE"],
          "revoke": []
        }
      }
    },
    "observer": {
      "subjects": {
        "nginx:observer-client": {
          "type": "technical client"
        },
        "nginx:some-users": {
          "type": "a group of users"
        }
      },
      "resources": {
        "thing:/features/featureX": {
          "grant": ["READ"],
          "revoke": []
        },
        "thing:/features/featureY": {
          "grant": ["READ"],
          "revoke": []
        }
      }
    },
    "private": {
      "subjects": {
        "nginx:some-users": {
          "type": "a group of users"
        },
        "resources": {
          "thing:/features/featureX/properties/location/city": {
            "grant": [],
            "revoke": ["READ"]
          }
        }
      }
    }
  }
}

The Policy can be found:

• Via GET request at /api/2/policies/<policyId>, and
• Via GET request at /api/2/things/{thingId}/policyId
• At any Thing itself in its JSON representation. It is however not included by default, but can be retrieved by specifying the /api/2/things/<thingId>?fields=_policy query parameter.