sbom-operator

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

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.

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

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

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

kubectl apply -f deploy/standard/

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

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.

The --format-version flag (env: SBOM_FORMAT_VERSION, YAML: formatVersion) pins the CycloneDX or SPDX specification version used by the embedded Syft encoder. It only applies when --format selects a version-aware family; for non-version-aware formats (json, syftjson, text, table, github) the flag is silently accepted and Syft defaults are used.

Exact supported revisions and defaults come from the Syft version bundled in each operator release. Run sbom-operator --help to see the authoritative values for the installed binary.

CLI flag (explicit CycloneDX 1.3 JSON):

sbom-operator --format=cyclonedxjson --format-version=1.3

Environment variable (SPDX 2.3):

export SBOM_FORMAT_VERSION=2.3
sbom-operator --format=spdxjson

Helm YAML:

args:
  format: cyclonedxxml
  formatVersion: "1.5"

Starting the operator with a version not supported by the selected family produces a hard Fatal error before any Kubernetes runtime initialization. The error includes the requested family, rejected version, and the versions supported by the bundled Syft release.

An unknown format name (not in the alias matrix) with an omitted or whitespace-only format-version delegates to Syft's default JSON encoder, preserving backward compatibility with custom format strings:

sbom-operator --format=custom-format --format-version=""

When an unknown format is paired with a version-like string (e.g. --format-version=1.2), the version is resolved to its family (CycloneDX 1.2 → cyclonedxjson) and the operator logs the concrete effective version at startup.

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"

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.

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.

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.

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).

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.

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

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

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

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
    ...
...

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

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

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.

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 the configmap-key sbom.

If you don't want to use Syft to analyze your images, you can give the Job-Image feature a try. The operator creates a Kubernetes-Job which does the analysis with any possible tool inside. There's no target-handling done by the operator, the tool from the job has to process the SBOMs on its own. Currently there are two possible integrations:

This feature is built as generic approach. Any image which follows these specs can be used as job-image.

e.g. Manifest (deploy/job-image):

--job-image=ghcr.io/ckotzbauer/sbom-operator/vcn:<TAG>

e.g. Helm:

jobImageMode: true

envVars:
  - name: SBOM_JOB_VCN_LC_API_KEY
    value: "<KEY>"

All operator-environment variables prefixed with SBOM_JOB_ are passed to the Kubernetes job.

The docker-image is based on a scratch-image to reduce the attack-surface and keep the image small. Furthermore the image and release-artifacts are signed with cosign and attested with provenance-files. The release-process satisfies SLSA Level 2. All of those "metadata files" are also stored in a dedicated repository ghcr.io/ckotzbauer/sbom-operator-metadata. When discovering security issues please refer to the Security process.

COSIGN_EXPERIMENTAL=1 COSIGN_REPOSITORY=ghcr.io/ckotzbauer/sbom-operator-metadata cosign verify ghcr.io/ckotzbauer/sbom-operator:<tag-to-verify> --certificate-github-workflow-name create-release --certificate-github-workflow-repository ckotzbauer/sbom-operator

COSIGN_EXPERIMENTAL=1 COSIGN_REPOSITORY=ghcr.io/ckotzbauer/sbom-operator-metadata cosign verify-attestation ghcr.io/ckotzbauer/sbom-operator:<tag-to-verify> --certificate-github-workflow-name create-release --certificate-github-workflow-repository ckotzbauer/sbom-operator

COSIGN_REPOSITORY=ghcr.io/ckotzbauer/sbom-operator-metadata cosign download attestation ghcr.io/ckotzbauer/sbom-operator:<tag-to-verify> | jq -r '.payload' | base64 -d

COSIGN_REPOSITORY=ghcr.io/ckotzbauer/sbom-operator-metadata cosign download sbom ghcr.io/ckotzbauer/sbom-operator:<tag-to-verify> | jq -r '.payload' | base64 -d

Please refer to the Contribution guildelines.

Please refer to the Conduct guildelines.