Skip to content

Custom Policies

Overview

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

trivy conf --policy /path/to/policy.rego --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/Snapshot)

This is useful for filtering inputs, as described below.

Rego format

A single package must contain only one policy.

Example

# METADATA
# title: Deployment not allowed
# description: Deployments are not allowed because of some reasons.
# schemas:
#   - input: schema["kubernetes"]
# custom:
#   id: ID001
#   severity: LOW
#   input:
#     selector: 
#     - type: kubernetes
package user.kubernetes.ID001

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

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

# METADATA (optional)
  • SHOULD be defined for clarity since these values will be displayed in the scan results
  • custom.input SHOULD be set to indicate the input type the policy should be applied to. See list of available types
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
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.

The annotation format is described in the OPA documentation.

Trivy supports extra fields in the custom section as described below.

Example

# METADATA
# title: Deployment not allowed
# description: Deployments are not allowed because of some reasons.
# custom:
#   id: ID001
#   severity: LOW
#   input:
#     selector:
#     - type: kubernetes

All fields are optional. The schemas field should be used to enable policy validation using a built-in schema. The schema that will be used is based on the input document type. It is recommended to use this to ensure your policies are correct and do not reference incorrect properties/values.

Field name Allowed values Default value In table In JSON
title Any characters N/A
description Any characters
schemas.input schema["kubernetes"], schema["dockerfile"], schema["cloud"] (applied to all input types)
custom.id Any characters N/A
custom.severity LOW, MEDIUM, HIGH, CRITICAL UNKNOWN
custom.recommended_actions Any characters
custom.input.selector.type Any item(s) in this list
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 the custom.input annotation.

Example

# METADATA
# custom:
#   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.

Possible values for input types are:

  • dockerfile (Dockerfile)
  • kubernetes (Kubernetes YAML/JSON)
  • rbac (Kubernetes RBAC YAML/JSON)
  • cloud (Cloud format, as defined by defsec - this is used for Terraform, CloudFormation, and Cloud/AWS scanning)
  • yaml (Generic YAML)
  • json (Generic JSON)
  • toml (Generic TOML)

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.

Schemas

See here for the detail.