Custom Policies

Overview

You can write custom policies in Rego. Once you finish writing custom policies, you can pass the directory where those policies are stored with --policy option.

trivy conf --policy /path/to/custom_policies --namespaces user /path/to/config_dir

As for --namespaces option, the detail is described as below.

File formats

If a file name matches the following file patterns, Trivy will parse the file and pass it as input to your Rego policy.

File format	File pattern
JSON	*.json
YAML	*.yaml and *.yml
Dockerfile	Dockerfile, Dockerfile.*, and *.Dockerfile
Containerfile	Containerfile, Containerfile.*, and *.Containerfile
Terraform	*.tf and *.tf.json

Configuration languages

In the above general file formats, Trivy automatically identifies the following types of configuration files:

• CloudFormation (JSON/YAML)
• Kubernetes (JSON/YAML)
• Helm (YAML)
• Terraform Plan (JSON)

This is useful for filtering inputs, as described below.

Rego format

A single package must contain only one policy.

Example

package user.kubernetes.ID001

import lib.result

__rego_metadata__ := {
    "id": "ID001",
    "title": "Deployment not allowed",
    "severity": "LOW",
    "description": "Deployments are not allowed because of some reasons.",
}

__rego_input__ := {
    "selector": [
        {"type": "kubernetes"},
    ],
}

deny[res] {
    input.kind == "Deployment"
    msg := sprintf("Found deployment '%s' but deployments are not allowed", [input.metadata.name])
    res := result.new(msg, input)
}

In this example, ID001 "Deployment not allowed" is defined under user.kubernetes.ID001. If you add a new custom policy, it must be defined under a new package like user.kubernetes.ID002.

Policy structure

package (required)

• MUST follow the Rego's specification
• MUST be unique per policy
• SHOULD include policy id for uniqueness
• MAY include the group name such as kubernetes for clarity
  • Group name has no effect on policy evaluation

import data.lib.result (optional)

• MAY be defined if you would like to embellish your result(s) with line numbers and code highlighting

__rego_metadata__ (optional)

• SHOULD be defined for clarity since these values will be displayed in the scan results

__rego_input__ (optional)

• MAY be defined when you want to specify input format

deny (required)

• SHOULD be deny or start with deny_
  • Although warn, warn_*, violation, violation_ also work for compatibility, deny is recommended as severity can be defined in __rego_metadata__.
• SHOULD return ONE OF:
  • The result of a call to result.new(msg, cause). The msg is a string describing the issue occurrence, and the cause is the property/object where the issue occurred. Providing this allows Trivy to ascertain line numbers and highlight code in the output.
  • A string denoting the detected issue
    • Although object with msg field is accepted, other fields are dropped and string is recommended if result.new() is not utilised.
    • e.g. {"msg": "deny message", "details": "something"}

Package

A package name must be unique per policy.

Example

package user.kubernetes.ID001

By default, only builtin.* packages will be evaluated. If you define custom packages, you have to specify the package prefix via --namespaces option.

trivy conf --policy /path/to/custom_policies --namespaces user /path/to/config_dir

In this case, user.* will be evaluated. Any package prefixes such as main and user are allowed.

Metadata

Metadata helps enrich Trivy's scan results with useful information.

Example

__rego_metadata__ := {
    "id": "ID001",
    "title": "Deployment not allowed",
    "severity": "LOW",
    "description": "Deployments are not allowed because of some reasons.",
    "recommended_actions": "Remove Deployment",
    "url": "https://cloud.google.com/blog/products/containers-kubernetes/kubernetes-best-practices-resource-requests-and-limits",
}

All fields under __rego_metadata__ are optional.

Field name	Allowed values	Default value	In table	In JSON
id	Any characters	N/A
title	Any characters	N/A
severity	LOW, MEDIUM, HIGH, CRITICAL	UNKNOWN
description	Any characters
recommended_actions	Any characters
url	Any characters

Some fields are displayed in scan results.

k.yaml (kubernetes)
───────────────────

Tests: 32 (SUCCESSES: 31, FAILURES: 1, EXCEPTIONS: 0)
Failures: 1 (UNKNOWN: 0, LOW: 1, MEDIUM: 0, HIGH: 0, CRITICAL: 0)

LOW: Found deployment 'my-deployment' but deployments are not allowed
════════════════════════════════════════════════════════════════════════
Deployments are not allowed because of some reasons.
────────────────────────────────────────────────────────────────────────
 k.yaml:1-2
────────────────────────────────────────────────────────────────────────
   1 ┌ apiVersion: v1
   2 └ kind: Deployment
────────────────────────────────────────────────────────────────────────

Input

You can specify input format via __rego_input__. All fields under __rego_input are optional.

Example

__rego_input__ := {
    "combine": false,
    "selector": [
        {"type": "kubernetes"},
    ],
}

combine (boolean)

The details are here.

selector (array)

This option filters the input by file format or configuration language. In the above example, Trivy passes only Kubernetes files to this policy. Even if a Dockerfile exists in the specified directory, it will not be passed to the policy as input.

When configuration languages such as Kubernetes are not identified, file formats such as JSON will be used as type. When a configuration language is identified, it will overwrite type.

Example

pod.yaml including Kubernetes Pod will be handled as kubernetes, not yaml. type is overwritten by kubernetes from yaml.

type accepts kubernetes, dockerfile, cloudformation, terraform, terraformplan, json, or yaml.