The Open Policy Agent (OPA) is an open source, general-purpose policy engine that enables unified, context-aware policy enforcement across the entire stack.

OPA is hosted by the Cloud Native Computing Foundation (CNCF) as an incubating-level project. If you are an organization that wants to help shape the evolution of technologies that are container-packaged, dynamically-scheduled and microservices-oriented, consider joining the CNCF. For details read the CNCF announcement.

• See openpolicyagent.org to get started with documentation and tutorials.
• See blog.openpolicyagent.org for blog posts about OPA and policy.
• See ADOPTERS.md for a list of production OPA adopters and use cases.
• See the Roadmap slides for a snapshot of high-level OPA features in-progress and planned.
• Try play.openpolicyagent.org to experiment with OPA policies.
• Join the conversation on Slack.

• See Docker Hub for Docker images.
• See GitHub releases for binary releases and changelogs.

• See to integrate OPA with services written in Go.
• See REST API to integrate OPA with services written in other languages.

• See DEVELOPMENT.md to build and test OPA itself.
• See CONTRIBUTING.md to get started.
• Use GitHub Issues to request features or file bugs.
• Join bi-weekly meetings every other Tuesday at 10:00 (Pacific Timezone):
  • Meeting Notes
  • Google Hangouts
  • Calendar Invite

OPA gives you a high-level declarative language to author and enforce policies across your stack.

With OPA, you define rules that govern how your system should behave. These rules exist to answer questions like:

• Can user X call operation Y on resource Z?
• What clusters should workload W be deployed to?
• What tags must be set on resource R before it's created?

You integrate services with OPA so that these kinds of policy decisions do not have to be hardcoded in your service. Services integrate with OPA by executing queries when policy decisions are needed.

When you query OPA for a policy decision, OPA evaluates the rules and data (which you give it) to produce an answer. The policy decision is sent back as the result of the query.

For example, in a simple API authorization use case:

• You write rules that allow (or deny) access to your service APIs.
• Your service queries OPA when it receives API requests.
• OPA returns allow (or deny) decisions to your service.
• Your service enforces the decisions by accepting or rejecting requests accordingly.

The examples below show different kinds of policies you can define with OPA as well as different kinds of queries your system can execute against OPA. The example queries are executed inside OPA's REPL which was built to make it easy to develop and test policies.

For concrete examples of how to integrate OPA with systems like Kubernetes, Terraform, Docker, SSH, and more, see openpolicyagent.org.

This example shows how you can enforce access controls over salary information served by a simple HTTP API. In this example, users are allowed to access their own salary as well as the salary of anyone who reports to them.

The management chain is represented in JSON and stored in a file (data.json):

{
    "management_chain": {
        "bob": [
            "ken",
            "janet"
        ],
        "alice": [
            "janet"
        ]
    }
}

Start OPA and load the data.json file:

opa run data.json

Inside the REPL you can define rules and execute queries. Paste the following rules into the REPL.

default allow = false

allow {
    input.method = "GET"
    input.path = ["salary", id]
    input.user_id = id
}

allow {
    input.method = "GET"
    input.path = ["salary", id]
    managers = data.management_chain[id]
    input.user_id = managers[_]
}

Is someone allowed to access their own salary?

> input := {"method": "GET", "path": ["salary", "bob"], "user_id": "bob"}
> allow
true

Display the management chain for Bob:

> data.management_chain["bob"]
[
    "ken",
    "janet"
]

Is Alice allowed to access Bob's salary?

> input := {"method": "GET", "path": ["salary", "bob"], "user_id": "alice"}
> allow
false

Is Janet allowed to access Bob's salary?

> input := {"method": "GET", "path": ["salary", "bob"], "user_id": "janet"}
> allow
true

This example shows how you can enforce where apps are deployed inside a simple orchestrator. In this example, apps must be deployed onto clusters that satisfy PCI and jurisdiction requirements.

app_placement[cluster_id] {
    cluster = data.clusters[cluster_id]
    satisfies_jurisdiction(input.app, cluster)
    satisfies_pci(input.app, cluster)
}

satisfies_jurisdiction(app, cluster) {
    not app.tags["requires-eu"]
}

satisfies_jurisdiction(app, cluster) {
    app.tags["requires-eu"]
    startswith(cluster.region, "eu-")
}

satisfies_pci(app, cluster) {
    not app.tags["requires-pci-level"]
}

satisfies_pci(app, cluster) {
    level = to_number(app.tags["requires-pci-level"])
    level >= cluster.tags["pci-level"]
}

Where will this app be deployed?

> input := {"app": {"tags": {"requires-pci-level": "3", "requires-eu": "true"}}}
> app_placement
[
    "prod-eu"
]

Display clusters in EU region:

> startswith(data.clusters[cluster_id].region, "eu-")
+------------+
| cluster_id |
+------------+
| "prod-eu"  |
| "test-eu"  |
+------------+

Display all clusters:

> data.clusters[cluster_id]
+------------+------------------------------------------------+
| cluster_id |           data.clusters[cluster_id]            |
+------------+------------------------------------------------+
| "prod-eu"  | {"region":"eu-central","tags":{"pci-level":2}} |
| "prod-us"  | {"region":"us-east"}                           |
| "test-eu"  | {"region":"eu-west","tags":{"pci-level":4}}    |
| "test-us"  | {"region":"us-west"}                           |
+------------+------------------------------------------------+

This example shows how you can audit who has SSH access to hosts within different clusters. We will assume that SSH access is granted via group access in LDAP.

import data.ldap
import data.clusters

ssh_access[[cluster_name, host_id, user_id]] {
    host_id = clusters[cluster_name].hosts[_]
    group_id = ldap.users[user_id].groups[_]
    group_id = clusters[cluster_name].groups[_]
}

prod_users = {user_id | ssh_access[["prod", _, user_id]]}

Who can access production hosts?

> prod_users
[
  "alice",
  "bob"
]

Display all LDAP users:

> data.ldap.users[user_id]
+-------------------------------+---------+
|   data.ldap.users[user_id]    | user_id |
+-------------------------------+---------+
| {"groups":["dev","platform"]} | "alice" |
| {"groups":["dev","ops"]}      | "bob"   |
| {"groups":["dev"]}            | "janet" |
+-------------------------------+---------+

Display all cluster/group pairs:

> data.clusters[cluster_id].groups[_] = group_id
+------------+------------+
| cluster_id |  group_id  |
+------------+------------+
| "test"     | "dev"      |
| "test"     | "ops"      |
| "prod"     | "ops"      |
| "prod"     | "platform" |
+------------+------------+

Does Janet have access to the test cluster?

> ssh_access[["test", _, "janet"]]
true

What are the addresses of the hosts in the test cluster that Janet can access?

> ssh_access[["test", host_id, "janet"]]; addr = data.hosts[host_id].addr
+------------+------------+
|    addr    |  host_id   |
+------------+------------+
| "10.0.0.1" | "host-abc" |
| "10.0.0.2" | "host-cde" |
| "10.0.0.3" | "host-efg" |
+------------+------------+

• Open Policy Agent Introduction @ CloudNativeCon EU 2018: video, slides
• Rego Deep Dive @ CloudNativeCon EU 2018: video, slides
• How Netflix Is Solving Authorization Across Their Cloud @ CloudNativeCon US 2017: video, slides.
• Policy-based Resource Placement in Kubernetes Federation @ LinuxCon Beijing 2017: slides, screencast.
• Enforcing Bespoke Policies In Kubernetes @ KubeCon US 2017: video, slides.
• Istio's Mixer: Policy Enforcement with Custom Adapters @ CloudNativeCon US 2017: video, slides.

A third party security audit was performed by Cure53, you can see the full report here

Please report vulnerabilities by email to open-policy-agent-security. We will send a confirmation message to acknowledge that we have received the report and then we will send additional messages to follow up once the issue has been investigated.