Validate Rules

Check resources configurations for policy compliance.

Validation rules are probably the most common and practical types of rules you will be working with, and the main use case for admission controllers such as Kyverno. In a typical validation rule, one defines the mandatory properties with which a given resource should be created. When a new resource is created by a user or process, the properties of that resource are checked by Kyverno against the validate rule. If those properties are validated, meaning there is agreement, the resource is allowed to be created. If those properties are different, the creation is blocked. The behavior of how Kyverno responds to a failed validation check is determined by the validationFailureAction field. It can either be blocked (Enforce) or allowed yet recorded in a policy report (Audit). Validation rules in Audit mode can also be used to get a report on matching resources which violate the rule(s), both upon initial creation and when Kyverno initiates periodic scans of Kubernetes resources. Resources in violation of an existing rule placed in Audit mode will also surface in an event on the resource in question.

To validate resource data, define a pattern in the validation rule. For more advanced processing using tripartite expressions (key-operator-value), define a deny element in the validation rule along with a set of conditions that control when to allow or deny the request.

Basic Validations

As a basic example, consider the below ClusterPolicy which validates that any new Namespace that is created has the label purpose with the value of production.

 1apiVersion: kyverno.io/v1
 2# The `ClusterPolicy` kind applies to the entire cluster.
 3kind: ClusterPolicy
 4metadata:
 5  name: require-ns-purpose-label
 6# The `spec` defines properties of the policy.
 7spec:
 8  # The `validationFailureAction` tells Kyverno if the resource being validated should be allowed but reported (`Audit`) or blocked (`Enforce`).
 9  validationFailureAction: Enforce
10  # The `rules` is one or more rules which must be true.
11  rules:
12  - name: require-ns-purpose-label
13    # The `match` statement sets the scope of what will be checked. In this case, it is any `Namespace` resource.
14    match:
15      any:
16      - resources:
17          kinds:
18          - Namespace
19    # The `validate` statement tries to positively check what is defined. If the statement, when compared with the requested resource, is true, it is allowed. If false, it is blocked.
20    validate:
21      # The `message` is what gets displayed to a user if this rule fails validation.
22      message: "You must have label `purpose` with value `production` set on all new namespaces."
23      # The `pattern` object defines what pattern will be checked in the resource. In this case, it is looking for `metadata.labels` with `purpose=production`.
24      pattern:
25        metadata:
26          labels:
27            purpose: production

If a new Namespace with the following definition is submitted to Kyverno, given the ClusterPolicy above, it will be allowed (validated). This is because it contains the label of purpose=production, which is the only pattern being validated in the rule.

1apiVersion: v1
2kind: Namespace
3metadata:
4  name: prod-bus-app1
5  labels:
6    purpose: production

By contrast, if a new Namespace with the below definition is submitted, given the ClusterPolicy above, it will be blocked (invalidated). As you can see, its value of the purpose label differs from that required in the policy. But this isn’t the only way a validation can fail. If, for example, the same Namespace is requested which has no labels defined whatsoever, it too will be blocked for the same reason.

1apiVersion: v1
2kind: Namespace
3metadata:
4  name: prod-bus-app1
5  labels:
6    purpose: development

Save the above manifest as ns.yaml and try to create it with your sample ClusterPolicy in place.

1$ kubectl create -f ns.yaml
2Error from server: error when creating "ns.yaml": admission webhook "validate.kyverno.svc" denied the request:
3
4resource Namespace//prod-bus-app1 was blocked due to the following policies
5
6require-ns-purpose-label:
7  require-ns-purpose-label: 'Validation error: You must have label `purpose` with value `production` set on all new namespaces.; Validation rule require-ns-purpose-label failed at path /metadata/labels/purpose/'

Change the development value to production and try again. Kyverno permits creation of your new Namespace resource.

Validation Failure Action

The validationFailureAction attribute controls admission control behaviors for resources that are not compliant with a policy. If the value is set to Enforce, resource creation or updates are blocked when the resource does not comply. When the value is set to Audit, a policy violation is logged in a PolicyReport or ClusterPolicyReport but the resource creation or update is allowed. For preexisting resources which violate a newly-created policy set to Enforce mode, Kyverno will allow subsequent updates to those resources which continue to violate the policy as a way to ensure no existing resources are impacted. However, should a subsequent update to the violating resource(s) make them compliant, any further updates which would produce a violation are blocked.

Validation Failure Action Overrides

Using validationFailureActionOverrides, you can specify which actions to apply per Namespace. This attribute is only available for ClusterPolicies.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-label-app
 5spec:
 6  validationFailureAction: Audit
 7  validationFailureActionOverrides:
 8    - action: Enforce     # Action to apply
 9      namespaces:       # List of affected namespaces
10        - default
11    - action: Audit
12      namespaces:
13        - test
14  rules:
15    - name: check-label-app
16      match:
17        any:
18        - resources:
19            kinds:
20            - Pod
21      validate:
22        message: "The label `app` is required."
23        pattern:
24          metadata:
25            labels:
26              app: "?*"

In the above policy, for Namespace default, validationFailureAction is set to Enforce and for Namespace test, it’s set to Audit. For all other Namespaces, the action defaults to the validationFailureAction field.

Patterns

A validation rule which checks resource data is defined as an overlay pattern that provides the desired configuration. Resource configurations must match fields and expressions defined in the pattern to pass the validation rule. The following rules are followed when processing the overlay pattern:

  1. Validation will fail if a field is defined in the pattern and if the field does not exist in the configuration.
  2. Undefined fields are treated as wildcards.
  3. A validation pattern field with the wildcard value ‘*’ will match zero or more alphanumeric characters. Empty values are matched. Missing fields are not matched.
  4. A validation pattern field with the wildcard value ‘?’ will match any single alphanumeric character. Empty or missing fields are not matched.
  5. A validation pattern field with the wildcard value ‘?*’ will match any alphanumeric characters and requires the field to be present with non-empty values.
  6. A validation pattern field with the value null or "" (empty string) requires that the field not be defined or has no value.
  7. The validation of siblings is performed only when one of the field values matches the value defined in the pattern. You can use the conditional anchor to explicitly specify a field value that must be matched. This allows writing rules like ‘if fieldA equals X, then fieldB must equal Y’.
  8. Validation of child values is only performed if the parent matches the pattern.

Wildcards

  1. * - matches zero or more alphanumeric characters
  2. ? - matches a single alphanumeric character

For a couple of examples on how wildcards work in rules, see the following.

This policy requires that all containers in all Pods have resource requests and limits defined (CPU limits intentionally omitted):

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: all-containers-need-requests-and-limits
 5spec:
 6  validationFailureAction: Enforce
 7  rules:
 8  - name: check-container-resources
 9    match:
10      any:
11      - resources:
12          kinds:
13          - Pod
14    validate:
15      message: "All containers must have CPU and memory resource requests and limits defined."
16      pattern:
17        spec:
18          containers:
19          # Select all containers in the pod. The `name` field here is not specifically required but serves
20          # as a visual aid for instructional purposes.
21          - name: "*"
22            resources:
23              limits:
24                # '?' requires 1 alphanumeric character and '*' means that
25                # there can be 0 or more characters. Using them together
26                # e.g. '?*' requires at least one character.
27                memory: "?*"
28              requests:
29                memory: "?*"
30                cpu: "?*"

The following validation rule checks for a label in Deployment, StatefulSet, and DaemonSet resources:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-label-app
 5spec:
 6  validationFailureAction: Enforce
 7  rules:
 8    - name: check-label-app
 9      match:
10        any:
11        - resources:
12            kinds:
13            - Deployment
14            - StatefulSet
15            - DaemonSet
16      validate:
17        message: "The label `app` is required."
18        pattern:
19          spec:
20            template:
21              metadata:
22                labels:
23                  app: "?*"

In order to treat special characters like wildcards as literals, see this section in the JMESPath page.

Operators

Operators in the following support list values in addition to scalar values. Many of these operators also support checking of durations (ex., 12h) and semver (ex., 1.4.1).

OperatorMeaning
>greater than
<less than
>=greater than or equals to
<=less than or equals to
!not equals
|logical or
&logical and
-within a range
!-outside a range

An example of using an operator in a pattern style validation rule is shown below.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: validate
 5spec:
 6  validationFailureAction: Enforce
 7  rules:
 8    - name: validate-replica-count
 9      match:
10        any:
11        - resources:
12            kinds:
13            - Deployment
14      validate:
15        message: "Replica count for a Deployment must be greater than or equal to 2."
16        pattern:
17          spec:
18            replicas: ">=2"

Anchors

Anchors allow conditional processing (i.e. “if-then-else”) and other logical checks in validation patterns. The following types of anchors are supported:

AnchorTagBehavior
Conditional()If tag with the given value (including child elements) is specified, then peer elements will be processed.
e.g. If image has tag latest then imagePullPolicy cannot be IfNotPresent.
    (image): “*:latest”
    imagePullPolicy: “!IfNotPresent”
Equality=()If tag is specified, then processing continues. For tags with scalar values, the value must match. For tags with child elements, the child element is further evaluated as a validation pattern.
e.g. If hostPath is defined then the path cannot be /var/lib
    =(hostPath):
        path: “!/var/lib”
Existence^()Works on the list/array type only. If at least one element in the list satisfies the pattern. In contrast, a conditional anchor would validate that all elements in the list match the pattern.
e.g. At least one container with image nginx:latest must exist.
    ^(containers):
    - image: nginx:latest
NegationX()The tag cannot be specified. The value of the tag is not evaluated (use exclamation point to negate a value). The value should ideally be set to "null" (quotes surrounding null).
e.g. Hostpath tag cannot be defined.
    X(hostPath): “null”
Global<()The content of this condition, if false, will cause the entire rule to be skipped. Valid for both validate and strategic merge patches.

Anchors and child elements: Conditional and Equality

Child elements are handled differently for conditional and equality anchors.

For conditional anchors, the child element is considered to be part of the “if” clause, and all peer elements are considered to be part of the “then” clause. For example, consider the following ClusterPolicy pattern statement:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: conditional-anchor-dockersock
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: conditional-anchor-dockersock
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      message: "If a hostPath volume exists and is set to `/var/run/docker.sock`, the label `allow-docker` must equal `true`."
17      pattern:
18        metadata:
19          labels:
20            allow-docker: "true"
21        (spec):
22          (volumes):
23          - (hostPath):
24              path: "/var/run/docker.sock"

This reads as “If a hostPath volume exists and the path equals /var/run/docker.sock, then a label “allow-docker” must be specified with a value of true.” In this case, the conditional checks the spec.volumes and spec.volumes.hostPath map. The child element of spec.volumes.hostPath is the path key and so the check ends the “If” evaluation at path. The entire metadata object is a peer element to the spec object because these reside at the same hierarchy within a Pod definition. Therefore, conditional anchors can not only compare peers when they are simple key/value, but also when peers are objects or YAML maps.

For equality anchors, a child element is considered to be part of the “then” clause. Now, consider the same ClusterPolicy as above but using equality anchors:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: equality-anchor-no-dockersock
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: equality-anchor-no-dockersock
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      message: "If a hostPath volume exists, it must not be set to `/var/run/docker.sock`."
17      pattern:
18        =(spec):
19          =(volumes):
20          - =(hostPath):
21              path: "!/var/run/docker.sock"

This is read as “If a hostPath volume exists, then the path must not be equal to /var/run/docker.sock”. In this sample, the object spec.volumes.hostPath is being checked, which is where the “If” evaluation ends. Similar to the conditional example above, the path key is a child to hostPath and therefore is the one being evaluated under the “then” check.

Existence anchor: At Least One

The existence anchor is used to check that, in a list of elements, at least one element exists that matches the pattern. This is done by using the ^() notation for the field. The existence anchor only works on array/list type data.

For example, this pattern will check that at least one container is using an image named nginx:latest:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: existence-anchor-at-least-one-nginx
 5spec:
 6  validationFailureAction: Enforce
 7  rules:
 8  - name: existence-anchor-at-least-one-nginx
 9    match:
10      any:
11      - resources:
12          kinds:
13          - Pod
14    validate:
15      message: "At least one container must use the image `nginx:latest`."
16      pattern:
17        spec:
18          ^(containers):
19            - image: nginx:latest

Contrast this existence anchor, which checks for at least one instance, with a wildcard which checks for every instance.

1      pattern:
2        spec:
3          containers:
4          - name: "*"
5            image: nginx:latest

This snippet above instead states that every entry in the array of containers, regardless of name, must have the image set to nginx:latest.

Global Anchor

The global anchor is a way to use a condition anywhere in a resource to base a decision. If the condition enclosed in the global anchor is true, the rest of the rule must apply. If the condition enclosed in the global anchor is false, the rule is skipped.

In this example, a container image coming from a registry called corp.reg.com is required to mount an imagePullSecret called my-registry-secret.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: sample
 5spec:
 6  validationFailureAction: Enforce
 7  rules:
 8  - name: check-container-image
 9    match:
10      any:
11      - resources:
12          kinds:
13          - Pod
14    validate:
15      message: Images coming from corp.reg.com must use the correct imagePullSecret.
16      pattern:
17        spec:
18          containers:
19          - name: "*"
20            <(image): "corp.reg.com/*"
21          imagePullSecrets:
22          - name: my-registry-secret

The below Pod has a single container which meets the global anchor’s specifications, but the rest of the pattern does not match. The Pod is therefore blocked.

 1apiVersion: v1
 2kind: Pod
 3metadata:
 4  name: static-web
 5  labels:
 6    role: myrole
 7spec:
 8  containers:
 9    - name: web
10      image: corp.reg.com/nginx
11      ports:
12        - name: web
13          containerPort: 80
14          protocol: TCP
15  imagePullSecrets:
16  - name: other-secret

anyPattern

In some cases, content can be defined at different levels. For example, a security context can be defined at the Pod or Container level. The validation rule should pass if either one of the conditions is met.

The anyPattern tag is a logical OR across multiple validation patterns and can be used to check if any one of the patterns in the list match.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: require-run-as-non-root
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: check-containers
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      message: >-
17        Running as root is not allowed. The fields spec.securityContext.runAsNonRoot,
18        spec.containers[*].securityContext.runAsNonRoot, and
19        spec.initContainers[*].securityContext.runAsNonRoot must be `true`.                
20      anyPattern:
21      # spec.securityContext.runAsNonRoot must be set to true. If containers and/or initContainers exist which declare a securityContext field, those must have runAsNonRoot also set to true.
22      - spec:
23          securityContext:
24            runAsNonRoot: true
25          containers:
26          - =(securityContext):
27              =(runAsNonRoot): true
28          =(initContainers):
29          - =(securityContext):
30              =(runAsNonRoot): true
31      # All containers and initContainers must define (not optional) runAsNonRoot=true.
32      - spec:
33          containers:
34          - securityContext:
35              runAsNonRoot: true
36          =(initContainers):
37          - securityContext:
38              runAsNonRoot: true

The anyPattern method is best suited for validation cases which do not use a negated condition. In the above example, only one of the spec contents must be valid. The same is true of negated conditions, however in the below example, this is slightly more difficult to reason about in that when negated, the anyPattern option allows any resource to pass so long as it doesn’t have at least one of the patterns.

1validate:
2  message: Cannot use Flux v1 annotation.
3  anyPattern:
4  - metadata:
5      =(annotations):
6        X(fluxcd.io/*): "*?"
7  - metadata:
8      =(annotations):
9        X(flux.weave.works/*): "*?"

If the desire is to state, “neither annotation named fluxcd.io/ nor flux.weave.works/ may be present”, then this would need two separate rules to express as including either one would mean the other is valid and therefore the resource is allowed.

Deny rules

In addition to applying patterns to check resources, a validate rule can deny a request based on a set of conditions written as expressions. A deny condition is an expression constructed of key, operator, value, and an optional message field. Unlike a pattern, when a deny condition evaluates to true it blocks a resource. Pattern expressions by contrast, when true, allow a resource.

Deny rules are more powerful and expressive than simple patterns but are also more complex to write. Use deny rules when:

An example of a deny rule is shown below. In deny rules, you write expressions similar to those in Kubernetes resources such as selectors. Deny rules, or “conditions”, must be nested under an any, all, or potentially both in order to control the decision-making logic. In this snippet, a resource will be denied if ANY of the following expressions are true.

  1. {{ request.object.data.team }} Equals eng
  2. {{ request.object.data.unit }} Equals green
 1validate:
 2  message: Main message is here.
 3  deny:
 4    conditions:
 5      any:
 6      - key: "{{ request.object.data.team }}"
 7        operator: Equals
 8        value: eng
 9        message: The expression team = eng failed.
10      - key: "{{ request.object.data.unit }}"
11        operator: Equals
12        value: green
13        message: The expression unit = green failed.

Placing these two conditions under an all block instead would require that both of them be true to produce the deny behavior.

Kyverno performs short-circuiting on deny conditions to abort processing when a decision can be reached. The first expression to evaluate to a true in an any block discontinues further evaluation. The first expression to evaluate to false in an all block does the same.

If the optional message field is included, it will be printed for a condition which evaluates to false keeping in mind how short-circuiting works.

See also Preconditions.

Deny DELETE requests based on labels

This policy denies delete requests for objects with the label app.kubernetes.io/managed-by: kyverno and for all users who do not have the cluster-admin role.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: deny-deletes
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: block-deletes-for-kyverno-resources
10    match:
11      any:
12      - resources:
13          selector:
14            matchLabels:
15              app.kubernetes.io/managed-by: kyverno
16    exclude:
17      any:
18      - clusterRoles:
19        - cluster-admin
20    validate:
21      message: "Deleting {{request.oldObject.kind}}/{{request.oldObject.metadata.name}} is not allowed"
22      deny:
23        conditions:
24          any:
25          - key: "{{request.operation}}"
26            operator: Equals
27            value: DELETE

Block changes to a custom resource

This policy denies admission review requests for updates or deletes to a custom resource, unless the request is from a specific ServiceAccount or matches specified Roles.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: block-updates-to-custom-resource
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: block-updates-to-custom-resource
10    match:
11      any:
12      - resources:
13          kinds:
14          - SomeCustomResource
15    exclude:
16      any:
17      - subjects:
18        - kind: ServiceAccount
19          name: custom-controller
20      - clusterRoles:
21        - custom-controller:*
22        - cluster-admin
23    validate:
24      message: "Modifying or deleting this custom resource is forbidden."
25      deny: {}

Prevent changing NetworkPolicy resources

This policy prevents users from changing NetworkPolicy resources with names that end with -default.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: deny-netpol-changes
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: deny-netpol-changes
10    match:
11      any:
12      - resources:
13          kinds:
14          - NetworkPolicy
15          names:
16          - "*-default"
17    exclude:
18      any:
19      - clusterRoles:
20        - cluster-admin
21    validate:
22      message: "Changing default network policies is not allowed."
23      deny: {}

foreach

The foreach declaration simplifies validation of sub-elements in resource declarations, for example containers in a Pod.

A foreach declaration can contain multiple entries to process different sub-elements e.g. one to process a list of containers and another to process the list of initContainers in a Pod.

Each foreach entry must contain a list attribute, written as a JMESPath expression without braces, that defines the sub-elements it processes. For example, iterating over the list of containers in a Pod is performed using this list declaration:

1list: request.object.spec.containers

When a foreach is processed, the Kyverno engine will evaluate list as a JMESPath expression to retrieve zero or more sub-elements for further processing. The value of the list field may also resolve to a simple array of strings, for example as defined in a context variable. The value of the list field should not be enclosed in braces even though it is a JMESPath expression.

A variable element is added to the processing context on each iteration. This allows referencing data in the element using element.<name> where name is the attribute name. For example, using the list request.object.spec.containers when the request.object is a Pod allows referencing the container image as element.image within a foreach.

The following child declarations are permitted in a foreach:

In addition, each foreach declaration can contain the following declarations:

  • Context: to add additional external data only available per loop iteration.
  • Preconditions: to control when a loop iteration is skipped
  • elementScope: controls whether to use the current list element as the scope for validation. Defaults to “true” if not specified.

Here is a complete example to enforce that all container images are from a trusted registry:

 1apiVersion : kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-images
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: check-registry
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    preconditions:
16      any:
17      - key: "{{request.operation}}"
18        operator: NotEquals
19        value: DELETE
20    validate:
21      message: "unknown registry"  
22      foreach:
23      - list: "request.object.spec.initContainers"
24        pattern:
25          image: "trusted-registry.io/*"      
26      - list: "request.object.spec.containers"
27        pattern:
28          image: "trusted-registry.io/*"

Note that the pattern is applied to the element and hence does not need to specify spec.containers and can directly reference the attributes of the element, which is a container in the example above.

Nested foreach

The foreach object also supports nesting multiple foreach declarations to form loops within loops. When using nested loops, the special variable {{elementIndex}} requires a loop number to identify which element to process. Preconditions are supported only at the top-level loop and not per inner loop.

This sample illustrates using nested foreach loops to validate that every hostname does not ends with new.com.

 1apiVersion : kyverno.io/v2beta1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-ingress
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9  - name: check-tls-secret-host
10    match:
11      any:
12      - resources:
13          kinds:
14          - Ingress
15    validate:
16      message: "All TLS hosts must use a domain of old.com."  
17      foreach:
18      - list: request.object.spec.tls[]
19        foreach:
20        - list: "element.hosts"
21          deny:
22            conditions:
23              all:
24              - key: "{{element}}"
25                operator: Equals
26                value: "*.new.com"

A sample Ingress which may get blocked by this look like the below.

 1apiVersion: networking.k8s.io/v1
 2kind: Ingress
 3metadata:
 4  name: kuard
 5  labels:
 6    app: kuard
 7spec:
 8  rules:
 9  - host: kuard.old.com
10    http:
11      paths:
12      - backend:
13          service: 
14            name: kuard
15            port: 
16              number: 8080
17        path: /
18        pathType: ImplementationSpecific
19  - host: hr.old.com
20    http:
21      paths:
22      - backend:
23          service: 
24            name: kuard
25            port: 
26              number: 8090
27        path: /myhr
28        pathType: ImplementationSpecific
29  tls:
30  - hosts:
31    - kuard.old.com
32    - kuard-foo.new.com
33    secretName: foosecret.old.com
34  - hosts:
35    - hr.old.com
36    secretName: hr.old.com

Nested foreach statements are also supported in mutate rules. See the documentation here for further details.

Manifest Validation

Kyverno has the ability to verify signed Kubernetes YAML manifests created with the Sigstore k8s-manifest-sigstore project. Using this capability, a Kubernetes YAML manifest is signed using one or multiple methods, which includes support for both keyed and keyless signing like in image verification, and through a policy definition Kyverno can validate these signatures prior to creation. This capability also includes support for field exclusions, multiple signatures, and dry-run mode.

Generate a key pair used to sign a manifest by using the cosign command-line tool.

1cosign generate-key-pair

Install the kubectl-sigstore command-line tool using one of the provided methods.

Sign the YAML manifest using the private key generated in the first step.

1$ kubectl-sigstore sign -f secret.yaml -k cosign.key --tarball no -o secret-signed.yaml
2Enter password for private key: 
3Using payload from: /tmp/kubectl-sigstore-temp-dir1572288324/tmp-blob-file
40D 7ѫO2�Ď��D)�I��!@t�0���X� Xmj���7���+u
5                                        ���_ڑ)ۆ�d�0�qHINFO[0004] signed manifest generated at secret-signed.yaml

The secret.yaml manifest provided as an input has been signed using your private key and the signed version is output at secret-signed.yaml.

 1apiVersion: v1
 2data:
 3  api_token: MDEyMzQ1Njc4OWFiY2RlZg==
 4kind: Secret
 5metadata:
 6  annotations:
 7    cosign.sigstore.dev/message: H4sIAAAAAAAA/zTMPQrCQBBA4X5OMVeIWA2kU7sYVFC0kXEzyJr9c3cirKcXlXSv+R4ne5RcbAyErwYGViZA5GSvGkcJhN1qXbv3rtk+zLI/bex5sXeXe9vCaMNAeBCTRcGL8owd38SVbyG6aFh/d5lyTAKIgb0Q+lr+UmsSwj7xcxL4BAAA//+dVuynjwAAAA==
 8    cosign.sigstore.dev/signature: MEQCIDfRq08y5MSOFo3iiEQUKdRJw9YhQHTjMAXwgO0eWO+hAiBYbR5qpa3wBjfN+d4rdQy5iNFf2pEp24aHZJgwyHEaSA==
 9  labels:
10    location: europe
11  name: mysecret
12type: Opaque

Create the Kyverno policy which matches on Secrets and will be used to validate the signatures.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: validate-secrets
 5spec:
 6  validationFailureAction: Enforce
 7  background: true
 8  rules:
 9    - name: validate-secrets
10      match:
11        any:
12        - resources:
13            kinds:
14              - Secret
15      validate:
16        manifests:
17          attestors:
18          - count: 1
19            entries:
20            - keys:
21                publicKeys: |-
22                  -----BEGIN PUBLIC KEY-----
23                  MFkwEwYHKoZIzj0CAQYIKoZIzj0DAQcDQgAEStoX3dPCFYFD2uPgTjZOf1I5UFTa
24                  1tIu7uoGoyTxJqqEq7K2aqU+vy+aK76uQ5mcllc+TymVtcLk10kcKvb3FQ==
25                  -----END PUBLIC KEY-----                  

To test the operation of this rule, modify the signed Secret to change some aspect of the manifest. For example, by changing even the value of the location label from europe to asia will cause the signed manifest to be invalid. Kyverno will reject the altered manifest because the signature was only valid for the original Secret manifest.

1$ kubectl apply -f secret-signed.yaml 
2Error from server: error when creating "secret-signed.yaml": admission webhook "validate.kyverno.svc-fail" denied the request: 
3
4policy Secret/default/mysecret for resource violation: 
5
6validate-secrets:
7  validate-secrets: 'manifest verification failed; verifiedCount 0; requiredCount
8    1; message .attestors[0].entries[0].keys: failed to verify signature. diff found;
9    {"items":[{"key":"metadata.labels.location","values":{"after":"asia","before":"europe"}}]}'

The difference between the signed manifest and supplied manifest is shown as part of the failure message.

Change the value of the location label back to europe and attempt to create the manifest once again.

1$ kubectl apply -f secret-signed.yaml 
2secret/mysecret created

The creation is allowed since the signature was validated according to the original contents.

In many cases, you may wish to secure a portion of a manifest while allowing alterations to other portions. For example, you may wish to sign manifests for Deployments which prevent tampering with any fields other than the replica count. Use the ignoreFields portion to define the object type and allowed fields which can differ from the signed original.

The below policy example shows how to match on Deployments and verify signed manifests while allowing changes to the spec.replicas field.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: validate-deployment
 5spec:
 6  validationFailureAction: Enforce
 7  background: true
 8  rules:
 9    - name: validate-deployment
10      match:
11        any:
12        - resources:
13            kinds:
14              - Deployment
15      validate:
16        manifests:
17          attestors:
18          - count: 1
19            entries:
20            - keys:
21                publicKeys: |-
22                  -----BEGIN PUBLIC KEY-----
23                  MFkwEwYHKoZIzj0CAQYIKoZIzj0DAQcDQgAEStoX3dPCFYFD2uPgTjZOf1I5UFTa
24                  1tIu7uoGoyTxJqqEq7K2aqU+vy+aK76uQ5mcllc+TymVtcLk10kcKvb3FQ==
25                  -----END PUBLIC KEY-----                  
26          ignoreFields:
27          - objects:
28            - kind: Deployment
29            fields:
30            - spec.replicas

Kyverno will permit the creation of a signed Deployment as long as the only difference between the signed original and the submitted manifest is the spec.replicas field. Modifications to any other field(s) will trigger a failure, for example if the spec.template.spec.containers[0].image field is changed from the default of busybox:1.28 to evilimage:1.28.

Rather than using ignoreFields to handle expected controller mutations, the dryRun object can be used to eliminate default changes by these and admission controllers. Set enable to true under the dryRun object as shown below and specify a Namespace in which the dry run will occur. Using other Namespaces or dry running with cluster-scoped or custom resources may entail giving additional privileges to the Kyverno ServiceAccount.

1validate:
2  manifests:
3    dryRun: 
4      enable: true
5      namespace: my-dryrun-ns

The manifest validation feature shares many of the same abilities as the verify images rule type. For a more thorough explanation of the available fields, use the kubectl explain clusterpolicy.spec.rules.validate.manifests command.

Pod Security

Starting in Kyverno 1.8, a new subrule type called podSecurity is available. This subrule type dramatically simplifies the process of writing and applying Pod Security Standards profiles and controls. By integrating the same libraries as used in Kubernetes’ Pod Security Admission, enabled by default in 1.23 and stable in 1.25, Kyverno is able to apply all or some of the controls and profiles in a single rule while providing capabilities not possible in Pod Security Admission. Standard match and exclude processing is available just like with other rules. This subrule type is enabled when a validate rule is written with a podSecurity object, detailed below.

The podSecurity feature has the following advantages over the Kubernetes built-in Pod Security Admission feature:

  1. Cluster-wide application of Pod Security Standards does not require an AdmissionConfiguration file nor any modifications to any control plane components.
  2. Namespace application of Pod Security Standards does not require assignment of a label.
  3. Specific controls may be exempted from a given profile.
  4. Container images may be exempted along with a control exemption.
  5. Enforcement of Pod controllers is automatic.
  6. Auditing of Pods in violation may be viewed in-cluster by examining a Policy Report Custom Resource.
  7. Testing of Pods and Pod controller manifests in a CI/CD pipeline is enabled via the Kyverno CLI.

For example, this policy enforces the latest version of the Pod Security Standards baseline profile in a single rule across the entire cluster.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: psa
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: baseline
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      podSecurity:
17        level: baseline
18        version: latest

The podSecurity.level field indicates the profile to be applied. Applying the baseline profile automatically includes all the controls outlined in the baseline profile.

The podSecurity.version field indicates which version of the Pod Security Standards should be applied. Use of the latest version indicates the latest version of the Pod Security Standards should be applied. This field allows prior versions, for example v1.24, to support the pinning to specific versions of the Pod Security Standards.

Attempting to apply a Pod which does not meet all of the controls included in the baseline profile will result in a blocking action.

1apiVersion: v1
2kind: Pod
3metadata:
4  name: badpod01
5spec:
6  hostIPC: true
7  containers:
8  - name: container01
9    image: dummyimagename

The failure message returned indicates which level, version, and specific control(s) were responsible for the failure.

1Error from server: error when creating "bad.yaml": admission webhook "validate.kyverno.svc-fail" denied the request: 
2
3policy Pod/default/badpod01 for resource violation: 
4
5psa:
6  baseline: |
7    Validation rule 'baseline' failed. It violates PodSecurity "baseline:latest": ({Allowed:false ForbiddenReason:host namespaces ForbiddenDetail:hostIPC=true})

Similarly, the restricted profile may be applied by changing the level field.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: psa
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: restricted
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      podSecurity:
17        level: restricted
18        version: latest

Applying the same Pod as above will now return additional information in the message about the cumulative violations.

Error from server: error when creating "bad.yaml": admission webhook "validate.kyverno.svc-fail" denied the request: 

policy Pod/default/badpod01 for resource violation: 

psa:
  baseline: |
    Validation rule 'baseline' failed. It violates PodSecurity "restricted:latest": ({Allowed:false ForbiddenReason:allowPrivilegeEscalation != false ForbiddenDetail:container "container01" must set securityContext.allowPrivilegeEscalation=false})
    ({Allowed:false ForbiddenReason:unrestricted capabilities ForbiddenDetail:container "container01" must set securityContext.capabilities.drop=["ALL"]})
    ({Allowed:false ForbiddenReason:host namespaces ForbiddenDetail:hostIPC=true})
    ({Allowed:false ForbiddenReason:runAsNonRoot != true ForbiddenDetail:pod or container "container01" must set securityContext.runAsNonRoot=true})
    ({Allowed:false ForbiddenReason:seccompProfile ForbiddenDetail:pod or container "container01" must set securityContext.seccompProfile.type to "RuntimeDefault" or "Localhost"})

Exemptions

When it is necessary to exempt specific controls within a profile while applying all others, the podSecurity.exclude[] object may be used. Controls which have restricted fields at the Pod spec level need only specify the controlName field, the value of which must be a valid name of a Pod Security Standard control. Controls which have restricted fields at the Pod containers[] level must additionally specify the images[] list. Wildcards are supported in the value of images[] allowing for flexible exemption. And controls which have restricted fields at both spec and containers[] levels must specify two objects in the exclude[] field: once with controlName and the other with both controlName and images[].

For example, the below policy applies the baseline profile across the entire cluster while exempting any Pod that violates the Host Namespaces control.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: psa
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: baseline
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      podSecurity:
17        level: baseline
18        version: latest
19        exclude:
20        - controlName: Host Namespaces

The following Pod violates the Host Namespaces control because it sets spec.hostIPC: true yet is allowed due to the exclusion.

1apiVersion: v1
2kind: Pod
3metadata:
4  name: badpod01
5spec:
6  hostIPC: true
7  containers:
8  - name: container01
9    image: dummyimagename

When a control exemption is requested where the control defines only container-level fields, the images[] list must be present with at least one entry. Wildcards (*) are supported both as the sole value and as a component of an image name.

For example, the below policy enforces the restricted profile but exempts containers running either the nginx or redis image from following the Capabilities control.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: psa
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: restricted
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      podSecurity:
17        level: restricted
18        version: latest
19        exclude:
20        - controlName: Capabilities
21          images:
22          - nginx*
23          - redis*

The following Pod, running the nginx:1.1.9 image, will be allowed although it violates the Capabilities control by virtue of it adding a forbidden capability.

 1apiVersion: v1
 2kind: Pod
 3metadata:
 4  name: goodpod01
 5spec:
 6  containers:
 7  - name: container01
 8    image: nginx:1.1.9
 9    securityContext:
10      allowPrivilegeEscalation: false
11      runAsNonRoot: true
12      seccompProfile:
13        type: RuntimeDefault
14      capabilities:
15        add:
16        - SYS_ADMIN
17        drop:
18        - ALL

The same policy would result in blocking a Pod in which a container running the busybox:1.28 image attempted the same thing.

Error from server: error when creating "temp.yaml": admission webhook "validate.kyverno.svc-fail" denied the request: 

policy Pod/default/badpod01 for resource violation: 

psa:
  restricted: |
    Validation rule 'restricted' failed. It violates PodSecurity "restricted:latest": ({Allowed:false ForbiddenReason:non-default capabilities ForbiddenDetail:container "container01" must not include "SYS_ADMIN" in securityContext.capabilities.add})
    ({Allowed:false ForbiddenReason:unrestricted capabilities ForbiddenDetail:container "container01" must not include "SYS_ADMIN" in securityContext.capabilities.add})

When a control is to be excluded which contains fields at both the spec and containers[] level, in order for that control to be fully excluded it must have exclusions for both. The controlName field assumes spec level while controlName plus images[] assumes the containers[] level.

For example, the Seccomp control in the restricted profile mandates that the securityContext.seccompProfile.type field be set to either RuntimeDefault or Localhost. The securityContext object may be defined at one or both the spec or container[] levels. The container[] fields may be undefined/nil if the Pod-level field is set appropriately. Conversely, the Pod-level field may be undefined/nil if all container- level fields are set. In order to completely exclude this control, two entries must exist in the podSecurity.exclude[] object. The below policy enforces the restricted profile across the entire cluster while fully exempting the Seccomp control from all images.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: psa
 5spec:
 6  background: true
 7  validationFailureAction: Enforce
 8  rules:
 9  - name: restricted
10    match:
11      any:
12      - resources:
13          kinds:
14          - Pod
15    validate:
16      podSecurity:
17        level: restricted
18        version: latest
19        exclude:
20        - controlName: Seccomp
21        - controlName: Seccomp
22          images:
23          - '*'

An example Pod which satisfies all controls in the restricted profile except the Seccomp control is therefore allowed.

 1apiVersion: v1
 2kind: Pod
 3metadata:
 4  name: goodpod01
 5spec:
 6  securityContext:
 7    seccompProfile:
 8      type: Unconfined
 9  containers:
10  - name: container01
11    image: busybox:1.28
12    securityContext:
13      allowPrivilegeEscalation: false
14      runAsNonRoot: true
15      # seccompProfile:
16      #   type: Unconfined
17      capabilities:
18        drop:
19        - ALL

Regardless of where the disallowed type: Unconfined field is specified, Kyverno allows the Pod.

Multiple control names may be excluded by listing them individually keeping in mind the previously-described rules. Refer to the Pod Security Standards documentation for a listing of all present controls, restricted fields, and allowed values.

PSA Interoperability

Kyverno’s podSecurity validate subrule type and Kubernetes’ Pod Security Admission (PSA) are compatible and may be used together in a single cluster with an understanding of where each begins and ends. These are a few of the most common strategies when employing both technologies.

  1. Use PSA to enforce the baseline profile cluster-wide and use Kyverno podSecurity subrule to enforce or audit the restricted profile with more granularity.

    Advantage: Reduces some of the processing on Kyverno by blocking non-compliant Pods at the source while allowing more flexible control on exclusions not possible with PSA.

  2. Use PSA to enforce either baseline or restricted on a per-Namespace basis and use Kyverno podSecurity cluster-wide or on different Namespaces.

    Advantage: Does not require configuring an AdmissionConfiguration file for PSA.

  3. Use PSA to enforce the baseline profile cluster-wide, relax certain Namespaces to the privileged profile, and use Kyverno podSecurity at the baseline or restricted profile.

    Advantage: Combines both AdmissionConfiguration with Namespace labeling while layering in Kyverno for granular control over baseline and restricted. A Kyverno mutate rule may also be separately employed here to handle the Namespace labeling as desired.

  4. Use both PSA and Kyverno to enforce the same profile at the same scope.

    Advantage: Provides a safety net in case either technology is inadvertently/maliciously disabled or becomes unavailable.

Common Expression Language (CEL)

Starting in Kyverno 1.11, a new subrule type called cel is available. This subrule type allows users to write CEL expressions for resource validation. CEL was initially introduced to Kubernetes for the validation rules of CustomResourceDefinitions and is now also utilized by Kubernetes ValidatingAdmissionPolicies. Standard match and exclude processing is available just like with other rules. This subrule type is enabled when a validate rule is written with a cel object, detailed below.

For example, this policy ensures that deployment replicas are less than 4.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-deployment-replicas
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9    - name: check-deployment-replicas
10      match:
11        any:
12        - resources:
13            kinds:
14              - Deployment
15      validate:
16        cel:
17          expressions:
18            - expression: "object.spec.replicas < 4"
19              message:  "Deployment spec.replicas must be less than 4."

The cel.expressions contains CEL expressions which use the Common Expression Language (CEL) to validate the request. If an expression evaluates to false, the validation check is enforced according to the spec.validationFailureAction field.

When trying to create a Deployment with replicas set not satisfying the validation expression, the creation of the Deployment will be blocked.

Error from server: error when creating "STDIN": admission webhook "validate.kyverno.svc-fail" denied the request: 

resource Deployment/default/nginx was blocked due to the following policies

check-deployment-replicas:
  check-deployment-replicas: Deployment spec.replicas must be less than 4.

The following policy ensures that any StatefulSet is created in the production namespace. The CEL expression access the namespace that the incoming object belongs to via namespaceObject.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-statefulset-namespace
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9    - name: statefulset-namespace
10      match:
11        any:
12        - resources:
13            kinds:
14              - StatefulSet
15      validate:
16        cel:
17          expressions:
18            - expression: "namespaceObject.metadata.name == 'production'"
19              message: "The StatefulSet must be created in the 'production' namespace."

When trying to create a StatefulSet in the default namespace, the creation of the StatefulSet will be blocked.

Error from server: error when creating "STDIN": admission webhook "validate.kyverno.svc-fail" denied the request: 

resource StatefulSet/default/bad-statefulset was blocked due to the following policies 

check-statefulset-namespace:
  statefulset-namespace: The StatefulSet must be created in the 'production' namespace.

CEL expressions have access to the contents of the Admission request/response, organized into CEL variables as well as some other useful variables:

  • object - The object from the incoming request. The value is null for DELETE requests.
  • oldObject - The existing object. The value is null for CREATE requests.
  • request - Attributes of the admission request.
  • params - Parameter resource referred to by cel.paramKind and cel.paramRef.
  • namespaceObject - The namespace, as a Kubernetes resource, that the incoming object belongs to. The value is null if the incoming object is cluster-scoped.
  • authorizer - It can be used to perform authorization checks.
  • authorizer.requestResource - A shortcut for an authorization check configured with the request resource (group, resource, (subresource), namespace, name).

Read Supported evaluation on CEL for more information about CEL rules.

validate.cel subrules also supports autogen rules for higher-level controllers that directly or indirectly manage Pods: Deployment, DaemonSet, StatefulSet, Job, and CronJob resources. Check the autogen section for more information.

 1status:
 2  autogen:
 3    rules:
 4    - exclude:
 5        resources: {}
 6      generate:
 7        clone: {}
 8        cloneList: {}
 9      match:
10        any:
11        - resources:
12            kinds:
13            - DaemonSet
14            - Deployment
15            - Job
16            - StatefulSet
17            - ReplicaSet
18            - ReplicationController
19        resources: {}
20      mutate: {}
21      name: autogen-disallow-latest-tag
22      validate:
23        cel:
24          expressions:
25          - expression: object.spec.template.spec.containers.all(container, !container.image.contains('latest'))
26            message: Using a mutable image tag e.g. 'latest' is not allowed.
27    - exclude:
28        resources: {}
29      generate:
30        clone: {}
31        cloneList: {}
32      match:
33        any:
34        - resources:
35            kinds:
36            - CronJob
37        resources: {}
38      mutate: {}
39      name: autogen-cronjob-disallow-latest-tag
40      validate:
41        cel:
42          expressions:
43          - expression: object.spec.jobTemplate.spec.template.spec.containers.all(container,
44              !container.image.contains('latest'))
45            message: Using a mutable image tag e.g. 'latest' is not allowed.

Parameter Resources

Parameter resources enable a policy configuration to be separated from its definition. A policy can define cel.paramKind, which outlines the GVK of the parameter resource, and then associate the policy with a specific parameter resource via cel.paramRef.

For example, the above policy can be modified to make it configurable:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: check-deployment-replicas
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9    - name: check-deployment-replicas
10      match:
11        any:
12        - resources:
13            kinds:
14              - Deployment
15      validate:
16        cel:
17          paramKind: 
18            apiVersion: rules.example.com/v1
19            kind: ReplicaLimit
20          paramRef:
21            name: "replica-limit"
22            parameterNotFoundAction: "Deny"
23          expressions:
24            - expression: "object.spec.replicas < params.maxReplicas"
25              messageExpression:  "'Deployment spec.replicas must be less than ' + string(params.maxReplicas)"

Here, cel.paramKind defines the resources used to configure the policy and the expression uses the params variable to access the parameter resource. The cel.paramRef is used to bind the policy to a specific resource.

The parameter resource could be as following:

1apiVersion: rules.example.com/v1
2kind: ReplicaLimit
3metadata:
4  name: "replica-limit"
5maxReplicas: 4

This policy parameter resource limits deployments to a max of 4 replicas.

Per-namespace Parameters

There are two types of parameter resources: cluster-wide parameters and per-namespace parameters. If you specify a namespace for the policy cel.paramRef, then Kyverno only searches for parameters in that namespace.

However, if namespace is not specified in cel.paramRef, then Kyverno can search for relevant parameters in the namespace that a request is against. For example, if you make a request to modify a ConfigMap in the default namespace and there is a relevant policy with no namespace set in cel.paramRef, then Kyverno looks for a parameter object in default.

Parameter selector

In addition to specify a parameter in cel.paramRef by name, you may choose instead to specify label selector, such that all resources of the policy’s paramKind, and the param’s namespace (if applicable) that match the label selector are selected for evaluation.

If multiple parameters are found to meet the condition, the policy’s rule is evaluated for each parameter found and the results will be ANDed together.

If cel.paramRef.namespace is provided, only objects of the paramKind in the provided namespace are eligible for selection. Otherwise, when namespace is empty and paramKind is namespace-scoped, the namespace used in the request being admitted will be used.

CEL Preconditions

CEL Preconditions allow for more fine-grained selection of resources than the options allowed by match and exclude statements. Preconditions consist of one or more CEL expressions which are evaluated after a resource has been successfully matched (and not excluded) by a rule. When preconditions are evaluated to an overall TRUE result, processing of the rule body begins.

For example, if you wished to apply policy to all Kubernetes Services which were of type NodePort, since neither the match/exclude blocks provide access to fields within a resource’s spec, a CEL precondition could be used. In the below rule, while all Services are initially selected by Kyverno, only the ones which have the field spec.type set to NodePort will go on to be processed to ensure the field spec.externalTrafficPolicy equals a value of Local.

 1rules:
 2  - name: validate-nodeport-trafficpolicy
 3    match:
 4      any:
 5      - resources:
 6          kinds:
 7            - Service
 8    celPreconditions:
 9        - name: check-service-type
10          expression: "object.spec.type.matches('NodePort')"
11    validate:
12      cel:
13        expressions:
14        - expression: "object.spec.externalTrafficPolicy.matches('Local')"
15          message: "All NodePort Services must use an externalTrafficPolicy of Local."

Attempting to apply a Service of type NodePort with externalTrafficPolicy set to Cluster will result in a blocking action.

 1apiVersion: v1
 2kind: Service
 3metadata:
 4  name: my-service
 5spec:
 6  type: "NodePort"
 7  selector:
 8    app.kubernetes.io/name: MyApp
 9  ports:
10    - port: 80
11      targetPort: 80
12  externalTrafficPolicy: "Cluster"

CEL Variables

If an expression grows too complicated, or part of the expression is reusable and computationally expensive to evaluate. We can extract some parts of the expressions into variables. A variable is a named expression that can be referred later as variables in other expressions.

The order of variables is important because a variable can refer to other variables defined before it. This ordering prevents circular references.

The below policy enforces that image repo names match the environment defined in its namespace. It enforces that all containers of deployment have the image repo match the environment label of its namespace except for “exempt” deployments or any containers that do not belong to the “example.com” organization (e.g., common sidecars). For example, if the namespace has a label of {“environment”: “staging”}, all container images must be either staging.example.com/* or do not contain “example.com” at all, unless the deployment has {“exempt”: “true”} label.

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: image-matches-namespace-environment.policy.example.com
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9    - name: image-matches-namespace-environment
10      match:
11        any:
12        - resources:
13            kinds:
14              - Deployment
15      validate:
16        cel:
17          variables:
18            - name: environment
19              expression: "'environment' in namespaceObject.metadata.labels ? namespaceObject.metadata.labels['environment'] : 'prod'"
20            - name: exempt
21              expression: "has(object.metadata.labels) && 'exempt' in object.metadata.labels && object.metadata.labels['exempt'] == 'true'"
22            - name: containers
23              expression: "object.spec.template.spec.containers"
24            - name: containersToCheck
25              expression: "variables.containers.filter(c, c.image.contains('example.com/'))"
26          expressions:
27            - expression: "variables.exempt || variables.containersToCheck.all(c, c.image.startsWith(variables.environment + '.'))"
28              messageExpression: "'only ' + variables.environment + ' images are allowed in namespace ' + namespaceObject.metadata.name"

Attempting to apply a deployment whose image is example.com/nginx in the staging-ns namespace will result in a blocking action.

 1apiVersion: apps/v1
 2kind: Deployment
 3metadata:
 4  name: deployment-fail
 5  namespace: staging-ns
 6spec:
 7  replicas: 1
 8  selector:
 9    matchLabels:
10      app: app
11  template:
12    metadata:
13      labels:
14        app: app
15    spec:
16      containers:
17      - name: container2
18        image: example.com/nginx

However, setting the deployment image as staging.example.com/nginx will allow it to be created.

Validating Admission Policies

A ValidatingAdmissionPolicy provides a declarative, in-process option for validating admission webhooks using the Common Expression Language (CEL) to perform resource validation checks directly in the API server.

Kubernetes ValidatingAdmissionPolicy was first introduced in 1.26, and it’s not fully enabled by default as of Kubernetes versions up to and including 1.28.

The ValidatingAdmissionPolicy is designed to perform basic validation checks for an admission request. In contrast, Kyverno is capable of performing complex validation checks, validation across resources with API lookups, mutation, generation, image verification, exception management, reporting, and off-cluster validation.

To unify the policy management, Kyverno policies can be used to generate and manage the lifecycle of Kubernetes ValidatingAdmissionPolicies. This allows the process of resource validation to take place directly in the API server, whenever possible, and extends Kyverno’s reporting and testing capabilities for ValidatingAdmissionPolicy resources.

When Kyverno manages ValidatingAdmissionPolicies and their bindings it is necessary to grant the Kyverno admission controller’s ServiceAccount additional permissions. To enable Kyverno to generate these types, see the section on customizing permissions. Kyverno will assist you in these situations by validating and informing you if the admission controller does not have the level of permissions required at the time the policy is installed.

To generate ValidatingAdmissionPolicies, make sure to:

  1. Enable ValidatingAdmissionPolicy feature gate.

  2. For 1.27, enable admissionregistration.k8s.io/v1alpha1 API, and for 1.28 enable both admissionregistration.k8s.io/v1alpha1 and admissionregistration.k8s.io/v1beta1 API.

    Here is the minikube command to enable ValidatingAdmissionPolicy:

    minikube start --extra-config=apiserver.runtime-config=admissionregistration.k8s.io/v1beta1,apiserver.runtime-config=admissionregistration.k8s.io/v1alpha1  --feature-gates='ValidatingAdmissionPolicy=true'
    
  3. Configure Kyverno to manage ValidatingAdmissionPolicies using the --generateValidatingAdmissionPolicy=true flag in the admission controller.

  4. Configure Kyverno to generate reports for ValidatingAdmissionPolicies using the --validatingAdmissionPolicyReports=true flag in the reports controller.

  5. Grant the admission controller’s ServiceAccount permissions to manage ValidatingAdmissionPolicies.

    Here is an aggregated cluster role you can apply:

     1apiVersion: rbac.authorization.k8s.io/v1
     2kind: ClusterRole
     3metadata:
     4  labels:
     5    app.kubernetes.io/component: admission-controller
     6    app.kubernetes.io/instance: kyverno
     7    app.kubernetes.io/part-of: kyverno
     8  name: kyverno:generate-validatingadmissionpolicy
     9rules:
    10- apiGroups:
    11  - admissionregistration.k8s.io
    12  resources:
    13  - validatingadmissionpolicies
    14  - validatingadmissionpolicybindings
    15  verbs:
    16  - create
    17  - update
    18  - delete
    19  - list
    

ValidatingAdmissionPolicies can only be generated from the validate.cel sub-rules in Kyverno policies. Refer to the CEL subrule section on the validate page for more information.

Below is an example of a Kyverno policy that can be used to generate a ValidatingAdmissionPolicy and its binding:

 1apiVersion: kyverno.io/v1
 2kind: ClusterPolicy
 3metadata:
 4  name: disallow-host-path
 5spec:
 6  validationFailureAction: Enforce
 7  background: false
 8  rules:
 9    - name: host-path
10      match:
11        any:
12        - resources:
13            kinds:
14              - Deployment
15      validate:
16        cel:
17          expressions:
18            - expression: "!has(object.spec.template.spec.volumes) || object.spec.template.spec.volumes.all(volume, !has(volume.hostPath))"
19              message: "HostPath volumes are forbidden. The field spec.template.spec.volumes[*].hostPath must be unset."

Once the policy is created, it is possible to check whether there is a corresponding ValidatingAdmissionPolicy was generated under the status object.

1status:
2  validatingadmissionpolicy:
3    generated: true
4    message: ""

The generated ValidatingAdmissionPolicy:

 1apiVersion: admissionregistration.k8s.io/v1beta1
 2kind: ValidatingAdmissionPolicy
 3metadata:
 4  labels:
 5    app.kubernetes.io/managed-by: kyverno
 6  name: disallow-host-path
 7  ownerReferences:
 8  - apiVersion: kyverno.io/v1
 9    kind: ClusterPolicy
10    name: disallow-host-path
11spec:
12  failurePolicy: Fail
13  matchConstraints:
14    matchPolicy: Equivalent
15    namespaceSelector: {}
16    objectSelector: {}
17    resourceRules:
18    - apiGroups:
19      - apps
20      apiVersions:
21      - v1
22      operations:
23      - CREATE
24      - UPDATE
25      resources:
26      - deployments
27      scope: '*'
28  validations:
29  - expression: '!has(object.spec.template.spec.volumes) || object.spec.template.spec.volumes.all(volume,
30      !has(volume.hostPath))'
31    message: HostPath volumes are forbidden. The field spec.template.spec.volumes[*].hostPath
32      must be unset.

The generated ValidatingAdmissionPolicyBinding:

 1apiVersion: admissionregistration.k8s.io/v1beta1
 2kind: ValidatingAdmissionPolicyBinding
 3metadata:
 4  labels:
 5    app.kubernetes.io/managed-by: kyverno
 6  name: disallow-host-path-binding
 7  ownerReferences:
 8  - apiVersion: kyverno.io/v1
 9    kind: ClusterPolicy
10    name: disallow-host-path
11spec:
12  policyName: disallow-host-path
13  validationActions:
14  - Deny

Both the ValidatingAdmissionPolicy and its binding have the same naming convention as the Kyverno policy they originate from, with the binding having a “-binding” suffix.

If there is a request to create the following deployment given the generated ValidatingAdmissionPolicy above, it will be denied by the API server.

 1apiVersion: apps/v1
 2kind: Deployment
 3metadata:
 4  name: nginx
 5spec:
 6  replicas: 2
 7  selector:
 8    matchLabels:
 9      app: nginx
10  template:
11    metadata:
12      labels:
13        app: nginx
14    spec:
15      containers:
16      - name: nginx-server
17        image: nginx
18        volumeMounts:
19          - name: udev
20            mountPath: /data
21      volumes:
22      - name: udev
23        hostPath:
24          path: /etc/udev

The response returned from the API server.

1The deployments "nginx" is invalid:  ValidatingAdmissionPolicy 'disallow-host-path' with binding 'disallow-host-path-binding' denied request: HostPath volumes are forbidden. The field spec.template.spec.volumes[*].hostPath must be unset.

The generated ValidatingAdmissionPolicy with its binding is totally managed by the Kyverno admission controller which means deleting/modifying these generated resources will be reverted. Any updates to Kyverno policy triggers synchronization in the corresponding ValidatingAdmissionPolicy.


Last modified January 29, 2024 at 5:13 PM PST: Update validate.md (cherry-pick #1120) (#1123) (f2f22fc)