sbom-operator

Catalogue all images of a Kubernetes cluster to multiple targets with Syft.

Overview

This operator maintains a central place to track all packages and software used in all those images in a Kubernetes cluster. For this a Software Bill of Materials (SBOM) is generated from each image with Syft. They are all stored in one or more targets. Currently Git, Dependency Track, OCI-Registry and ConfigMaps are supported. With this it is possible to do further analysis, vulnerability scans and much more in a single place. To prevent scans of images that have already been analyzed pods are annotated with the imageID of the already processed image.

Kubernetes Compatibility

The image contains versions of k8s.io/client-go. Kubernetes aims to provide forwards & backwards compatibility of one minor version between client and server:

sbom-operator	k8s.io/{api,apimachinery,client-go}	expected kubernetes compatibility
main	v0.34.1	1.33.x, 1.34.x, 1.35.x
0.38.0	v0.34.1	1.33.x, 1.34.x, 1.35.x
0.33.0	v0.30.0	1.29.x, 1.30.x, 1.31.x
0.31.0	v0.29.3	1.28.x, 1.29.x, 1.30.x
0.30.0	v0.28.4	1.27.x, 1.28.x, 1.29.x
0.28.0	v0.27.4	1.26.x, 1.27.x, 1.28.x
0.26.0	v0.26.3	1.25.x, 1.26.x, 1.27.x
0.22.0	v0.25.4	1.24.x, 1.25.x, 1.26.x
0.15.0	v0.24.4	1.23.x, 1.24.x, 1.25.x
0.10.0	v0.23.6	1.22.x, 1.23.x, 1.24.x

However, the operator will work with more versions of Kubernetes in general.

Container Registry Support

The operator relies on the syft-internal mechanism to download images from OCI-compliant registries.

Installation

Manifests

kubectl apply -f deploy/standard/

Helm-Chart

Create a YAML file first with the required configurations or use helm-flags instead.

helm repo add ckotzbauer https://ckotzbauer.github.io/helm-charts
helm install ckotzbauer/sbom-operator -f your-values.yaml

Configuration

All parameters are cli-flags. The flags can be configured as args or as environment-variables prefixed with SBOM_ to inject sensitive configs as secret values.

Common parameters

Parameter	Required	Default	Description
verbosity	false	info	Log-level (debug, info, warn, error, fatal, panic)
cron	false	""	Backround-Service interval (CRON). See Trigger for details.
ignore-annotations	false	false	Force analyzing of all images, including those from annotated pods.
format	false	json	SBOM-Format. (One of json, syftjson, cyclonedxjson, spdxjson, github, githubjson, cyclonedx, cyclone, cyclonedxxml, spdx, spdxtv, spdxtagvalue, text, table)
targets	false	git	Comma-delimited list of targets to sent the generated SBOMs to. Possible targets git, dtrack, oci, configmap. Ignored with a job-image
pod-label-selector	false	""	Kubernetes Label-Selector for pods.
namespace-label-selector	false	""	Kubernetes Label-Selector for namespaces.
fallback-pull-secret	false	""	Kubernetes Pull-Secret Name to load as a fallback when all others fail (must be in the same namespace as the sbom-operator)
registry-proxy	false	[]	Proxy-Registry-Hosts to use. Flag can be used multiple times. Value-Mapping e.g. docker.io=ghcr.io
delete-orphan-images	false	true	Delete orphan images automatically

Example Helm-Config

args:
  targets: git
  git-author-email: XXX
  git-author-name: XXX
  git-repository: https://github.com/XXX/XXX
  git-path: dev-cluster/sboms
  verbosity: debug
  cron: "0 30 * * * *"

envVars:
  - name: SBOM_GIT_ACCESS_TOKEN
    valueFrom:
      secretKeyRef:
        name: "sbom-operator"
        key: "accessToken"

Analysis-Trigger

Cron

With the cron flag set, the operator runs with a specified interval and checks for changed images in your cluster. All options from github.com/robfig/cron are allowed as cron-syntax.

Real-Time

When you omit the cron flag, the operator uses a Cache-Informer to process changed pods immediately. In this mode there's also a one-time analysis at startup to sync the targets with the actual cluster-state. If you configured a job-image there's no initial startup sync.

AWS ECR (IRSA / Pod Identity)

When images are hosted in private AWS Elastic Container Registry (*.dkr.ecr.*.amazonaws.com) and no pull secrets are configured on the pod, sbom-operator automatically obtains a registry token via the AWS SDK default credentials chain. IRSA, EKS Pod Identity, and EC2 instance roles are all supported transparently. The required IAM permission for the operator's role is ecr:GetAuthorizationToken. For successful image layer pulls, the role also needs ecr:BatchGetImage and ecr:GetDownloadUrlForLayer on the relevant repositories.

ECR Public (public.ecr.aws/*) and cross-account ECR via assume-role are not supported by this auto-detection. Use --fallback-pull-secret for those scenarios.

Targets

It is possible to store the generated SBOMs to different targets (even multple at once). All targets are using Syft as analyzer. If you want to use another tool to analyze your images, then have a look at the Job image section. Images which are not present in the cluster anymore are removed from the configured targets (except for the OCI-Target).

Dependency Track

Dependency Track Parameter

Parameter	Required	Default	Description
dtrack-base-url	true when dtrack target is used	""	Dependency-Track base URL, e.g. 'https://dtrack.example.com'
dtrack-api-key	true when dtrack target is used	""	Dependency-Track API key
dtrack-label-tag-matcher	false	""	Dependency-Track Pod-Label-Tag matcher regex
dtrack-ca-cert-file	false	""	CA-Certificate filepath when using mTLS to connect to dtrack
dtrack-client-cert-file	true when dtrack-ca-cert-file is provided	""	Client-Certificate filepath when using mTLS to connect to dtrack
dtrack-client-key-file	true when dtrack-ca-cert-file is provided	""	Client-Key filepath when using mTLS to connect to dtrack
dtrack-default-parent-project	false	""	The default parent project UUID that should be used. This is overwritten in case a parent project annotation is set and found
dtrack-use-short-name	false	""	Use short image name (without registry) for project names
dtrack-kubernetes-cluster-id-mode	false	"tag"	Mode for using the kubernetes-cluster-id. Options: tag (adds cluster ID as a project tag) or prefix (prefixes the project name with the cluster ID).
dtrack-parent-project-annotation-key	false	""	Kubernetes pod annotation key to set parent project automatically, e.g. "my.pod.annotation"
dtrack-project-name-annotation-key	false	""	Kubernetes pod annotation key to set custom dtrack project name automatically, e.g. "my.pod.annotation"
kubernetes-cluster-id	false	"default"	Kubernetes Cluster ID (to be used in Dependency-Track or Job-Images)

Each image in the cluster is created as project with the full-image name (registry and image-path without tag) and the image-tag as project-version. When there's no image-tag, but a digest, the digest is used as project-version. The autoCreate option of DT is used. You have to set the --format flag to cyclonedx with this target.

---

Custom dtrack project name:

The key at kubernetes has to be suffixed with the container name the project is for. e.g. my.project.name/my-nginx.

[!IMPORTANT] The suffix regarding container name must not be added to the config value and must not include /. e.g. my.project.name

The value for a custom project name in dtrack by annotation at the specific Pod is written in the format of project:version or just project where version defaults to latest. E.g. MyParentProject or MyParentProject:1.0

---

Setting parent project at Dependency Track automatically:

The key at kubernetes has to be suffixed with the container name the parent project is for. e.g. my.parent.project/my-nginx. The value for the parent project annotation at the specific Pod is written in the format of project:version or just project where version defaults to latest. E.g. MyParentProject or MyParentProject:1.0

[!IMPORTANT] The suffix regarding container name must not be added to the config value and must not include /. e.g. my.parent.project

---

Example Pod Annotation:

apiVersion: v1
kind: Pod
metadata:
  annotations:
    my.parent.project/my-nginx: MyParentProject
    my.project.name/my-nginx: MyNginxProject:1.0
    my.parent.project/my-sidecar: MyOtherParentProject
    my.project.name/my-sidecar: MySidecarProject:1.0.1
spec:
  containers:
    - image: nginx:latest
      name: my-nginx
    ...
    - image: some-other-image:latest
      name: my-sidecar
    ...
...

---

sbom-operator config:

--dtrack-parent-project-annotation-key=my.parent.project
--dtrack-project-name-annotation-key=my.project.name

---

Git

Git Parameter

Parameter	Required	Default	Description
git-workingtree	false	/work	Directory to place the git-repo.
git-repository	true when git target is used.	""	Git-Repository-URL (HTTPS).
git-branch	false	main	Git-Branch to checkout.
git-path	false	""	Folder-Path inside the Git-Repository.
git-author-name	true when git target is used.	""	Author name to use for Git-Commits.
git-author-email	true when git target is used.	""	Author email to use for Git-Commits.
git-access-token	false	""	Git-Personal-Access-Token with write-permissions.
git-username	false	""	Git-Username
git-password	false	""	Git-Password
github-app-id	false	""	GitHub App-ID.
github-app-installation-id	false	""	GitHub App-Installation-ID.

The operator will save all files with a specific folder structure as described below. When a git-path is configured, all folders above this path are not touched from the application. Assuming that git-path is set to dev-cluster/sboms. When no git-path is given, the structure below is directly in the repository-root. The structure is basically <git-path>/<registry-server>/<image-path>/<image-digest>/sbom.json (see example below). The file-extension may differ when another output-format is configured. You can use a token-based authentication (e.g. a PAT for GitHub) with --git-access-token, BasicAuth with username and password (--git-username, --git-password) or Github App Authentication (--github-app-id, --github-app-installation-id, env: SBOM_GITHUB_APP_PRIVATE_KEY) The private-key has to be Base64 encoded.

Note: It is required, that the specified branch of the repo is fully initialized. There's no logic which creates a non-existent branch. Just commit a README.md or something similar, to make things work.

dev-cluster
│
└───sboms
    │
    └───docker.io
    |   │
    |   └───library
    |       │
    |       └───busybox
    |           │
    |           └───sha256_ae39a6f5...
    |               │   sbom.json
    |
    └───ghcr.io
        │
        └───kyverno
            │
            └───kyverno
            |   │
            |   └───sha256_9e3f14e5...
            |       │   sbom.json
            |
            └───kyvernopre
                │
                └───sha256_e48f87fd...
                    │   sbom.json
            |
            └───policy-reporter
                │
                └───sha256_b70caa7a...
                    │   sbom.json

OCI-Registry

OCI-Registry Parameter

Parameter	Required	Default	Description
oci-registry	true when oci target is used	""	OCI-Registry
oci-user	true when oci target is used	""	OCI-User
oci-token	true when oci target is used	""	OCI-Token

In this mode the operator will generate a SBOM and store it into an OCI-Registry. The SBOM then can be processed by cosign, Kyverno or any other tool. E.g.:

COSIGN_REPOSITORY=<yourregistry> cosign download sbom <your full image digest>

The operator needs the Registry-URL, a user and a token as password to authenticate to the registry. Write-permissions are needed.

ConfigMap

This target stores the SBOM as Kubernetes-ConfigMap. They are placed in the same namespace as the corresponding pod and the name consists of the pod- and container-name. The configmap is labeled with ckotzbauer.sbom-operator.io=true and annotated with ckotzbauer.sbom-operator.io/image-id=<full-image-repo-with-digest>. The content is stored as brotli-compressed binary-data with