Custodian policies for Infrastructure Code
This package allows cloud custodian to evaluate policies directly against infrastructure as code source assets.
It also provides a separate cli for better command line ux for source asset evaluation.
Install
We currently only support python > 3.10 on mac and linux, to run on windows we recommend using our docker images.
pip install c7n-left
We also provide signed docker images. These images are built on top of chainguard’s wolfi linux distribution which is designed to be minimal, auditable, and secure.
docker pull cloudcustodian/c7n-left:dev
Images signatures can be verified using cosign
export IMAGE=$(docker image inspect cloudcustodian/c7n-left:dev -f '{{index .RepoDigests 0}}')
cosign verify $IMAGE \
--certificate-identity 'https://github.com/cloud-custodian/cloud-custodian/.github/workflows/docker.yml@refs/heads/main' \
--certificate-oidc-issuer 'https://token.actions.githubusercontent.com'
Usage
❯ c7n-left run --help
Usage: c7n-left run [OPTIONS]
evaluate policies against IaC sources.
c7n-left -p policy_dir -d terraform_root --filters "severity=HIGH"
WARNING - CLI interface subject to change.
Options:
--format TEXT
--filters TEXT Filter policies or resources as k=v pairs
with globbing
--warn-on TEXT Select policies to log instead of fail on
via k=v pairs with globbing
-p, --policy-dir PATH Directory with policies
-d, --directory PATH IaC directory to evaluate
-o, --output [cli|github|json] Output format (default cli)
--output-file FILENAME Output file (default stdout)
--var-file FILE Load variables from the given file, can be
used more than once
--output-query TEXT Use a jmespath expression to filter json
output
--summary [policy|resource]
--help Show this message and exit.
We’ll create an empty directory with a policy in it
policies:
- name: test
resource: terraform.aws_s3_bucket
metadata:
severity: medium
filters:
- server_side_encryption_configuration: absent
And now we can use it to evaluate a terraform root module
❯ c7n-left run -p policies -d module
Running 1 policies on 1 resources
test - terraform.aws_s3_bucket
Failed
File: s3.tf:1-8
1 resource "aws_s3_bucket" "example" {
2 bucket = "my-custodian-test-bucket"
3 acl = "private"
4
5 tags = {
6 original-tag = "original-value"
7 }
8 }
Evaluation complete 0.00 seconds -> 1 Failures
Summary - By Policy
┏━━━━━━━━━━┳━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Severity ┃ Policy ┃ Result ┃
┡━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ medium │ test │ 1 failed 0 passed │
└──────────┴────────┴───────────────────┘
0 compliant of 1 total, 1 resource has 1 policy violation
For running in docker, you’ll need to use volume mounts to provide access to the policy directory and terraform root module.
docker run -ti --rm -v $(pwd)/policies:/policies -v $(pwd)/root-module:/module \
cloudcustodian/c7n-left:dev run -p /policies -d /module
If the terraform root module has other remote module dependencies, you’ll need to fetch those first using terraform before running c7n-left.
terraform get -update
CLI Filters
Which policies and which resources are evaluated can be controlled via
command line via --filters option.
Available filters
name- policy namecategory- policy categoryseverity- minimum policy severity (unknown, low, medium, high, critical)type- resource type, ie. aws_security_groupid- resource id ie. aws_vpc.example
Multiple values for a given filter can be specified as comma separate values, and all filters except severity support globbing.
Examples
# run all encryption policies on ebs volumes and sqs queues
c7n-left run -p policy_dir -d terraform --filters="category=encryption type=aws_ebs_volume,aws_sqs_queue"
# run all medium and higher level policies cost policies
c7n-left run -p policy_dir -d terraform --filters="severity=medium category=cost"
policy values for severity and category are specified in its metadata section. ie
policies:
- name: check-encryption
resource: [terraform.aws_ebs_volume, terraform.aws_sqs_queue]
metadata:
category: [encryption, security]
severity: high
filters:
- kms_master_key_id: absent
Outputs
if your using this in github actions, we have special output mode for
reporting annotations directly into pull requests with --output github
We also display a summary output after displaying resource matches,
there are two summary displays available, the default policy summary,
and a resource summary which can be enabled via --summary resource.
By default any policy matches cause a run to exit code 1 to mark failure,
this behavior can be controlled via the --warn-on cli flag. ie. given a policy
with
policies:
- name: check-encryption
resource: [terraform.aws_ebs_volume, terraform.aws_sqs_queue]
metadata:
category: [beta, security]
severity: high
running the policy with --warn-on category=beta will cause matches to be logged only instead
of causing an exit code 1.
Policy Language
Standard Custodian filters (value, list-item, and, or, not, reduce and event) are available
Policies for c7n-left support a few additional capabilities beyond what’s common for custodian policies.
Policies can be specified against multiple resource types either as an array or glob.
policies:
- name: check-encryption
resource: [aws_ebs_volume, aws_sqs_queue]
taggable filter
A taggable filter is available that allows filtering to only resources that support tagging.
In combination with resource wild card support, this allows using a single policy to enforce an organization’s tag standards.
policies:
- name: check-tag-policy
resource: "terraform.aws*"
filters:
- taggable
- or:
- tag:Env: absent
- tag:Owner: absent
- tag:App: absent
This filter supports resources from several terraform providers including aws, azure, gcp, oci, tencentcloud.
terraform providers that support default_tags have those values automatically available on the applicable resources.
traverse filter
A traverse filter is available that allows for multi-hop graph traversal from a resource
to any related resource.
ie, here’s a policy against an aws ec2 instance, that checks if any of the security groups attached to the instance, have a permission defined that allows access from 0.0.0.0/0
policies:
- name: check-security-group-open-cidr
resource: terraform.aws_instance
description: "EC2 should not be open to world on ssh"
filters:
- type: traverse
resources:
- aws_security_group
- aws_security_ingress_permission
attrs:
- Ipv4: 0.0.0.0/0
terraform data resources
Policies can also be written against data resources, note data
resources are prefixed with data..
policies:
- name: check-ami-data
resource: terraform.data.aws_ami
filters:
- type: value
key: owners
op: contains
value: "099720109477" # Canonical/ubuntu
and you can traverse from a resource to its data usage as well.
policies:
- name: check-owner-specified
resource: terraform.aws_instance
filters:
- type: traverse
resources: data.aws_ami
attrs:
- owners: present
environment variables
c7n-left is typically run in CI systems, which provide a wealth of information in environment variables. Policies can make use of these environment variables in two different ways.
They can be used to interpolate the content of a policy, where they will they will be substituted prior to the policy execution. Note this uses python’s format capabilties
policies:
- name: "check-{env[REPO]}-{env[PR_NUMBER]}"
resource: terraform.aws*
Additionally they can be evaluated by the policy using the event filter
policies:
- name: check-aws
resource: terraform.aws*
filters:
- type: event
key: env.REPO
value: "cloud-custodian/cloud-custodian"
jmespath
Sometimes you may have JSON encoded strings as part of your terraform. For
example if you’re working with an AWS ECS Task Definition the
container_defintions key will be a string and not a proper object. You
may want to have a policy that prevents environment variables like JWT_TOKEN
or DATABASE_PASSWORD from being passed in because you should be using the
secrets parameter instead.
You can do this with the following filter:
policies:
- name: ecs-task-definition-with-plaintext-password
description: >
It's not recommended to use plaintext environment variables for sensitive
information, such as credential data. Pass them through the `secrets`
parameter instead.
resource: terraform.aws_ecs_task_definition
metadata:
severity: High
category: Encryption
provider: aws
filters:
- container_definitions: not-null
- type: list-item
key: from_json(container_definitions)[].environment[]
attrs:
- type: value
key: name
op: regex
value: '(?:.|\n)*(password|secret|token|key)'
The key here is the from_json call to convert it from a string to an object.
Policy Testing
c7n-left supports writing and running tests for policies.
To create a test for a policy, create a tests directory next to your policy files.
Within that tests directory, create a sub directory with the policy name.
Next add terraform files to this sub directory. Typically you would add both terraform files that would match the policy and those that should not.
Finally you add assertions in a left.plan[.yaml|.json] file. The
format of the file is an array of dictionaries. The dictionaries are
used to match against the policy findings. The data its matching
against is what is found by using c7n-left run --output json. Each
key/value pair in the dictionary is matched against the finding.
So putting it all together, we’ve setup our tests as follows
❯ tree policy-dir-a/
policy-dir-a/
├── alb.yaml
└── tests
└── alb-deletion-protection-disabled
├── left.plan.yaml
├── negative1.tf
└── positive1.tf
3 directories, 4 files
❯ cat policy-dir-a/alb.yaml
policies:
- name: alb-deletion-protection-disabled
resource: [terraform.aws_lb, terraform.aws_alb]
description: |
Application Load Balancer should have deletion protection enabled
metadata:
severity: low
category: "Insecure Configurations"
filters:
- enable_deletion_protection: empty
❯ cat policy-dir-a/tests/alb-deletion-protection-disabled/left.plan.yaml
- "resource.__tfmeta.filename": "positive1.tf"
and now we can run a test
❯ c7n-left test -p policy-dir-a/
Discovered 1 Tests
Failure alb-deletion-protection-disabled [{'resource.__tfmeta.filename':
'positive1.tf'}] checks not used
1 Test Complete (0.05s) 1 Failure
A test fails if either an assertion in the plan file does not match one policy finding, or if a policy finding is not matched by an assertion.