1. Documentation
×

Kernel Module Management Operator

Learn about the Kernel Module Management (KMM) Operator and how you can use it to deploy out-of-tree kernel modules and device plugins on OKD clusters.

About the Kernel Module Management Operator

The Kernel Module Management (KMM) Operator manages, builds, signs, and deploys out-of-tree kernel modules and device plugins on OKD clusters.

KMM adds a new Module CRD which describes an out-of-tree kernel module and its associated device plugin. You can use Module resources to configure how to load the module, define ModuleLoader images for kernel versions, and include instructions for building and signing modules for specific kernel versions.

KMM is designed to accommodate multiple kernel versions at once for any kernel module, allowing for seamless node upgrades and reduced application downtime.

Installing the Kernel Module Management Operator

As a cluster administrator, you can install the Kernel Module Management (KMM) Operator by using the OpenShift CLI or the web console.

The KMM Operator is supported on OKD 4.12 and later. Installing KMM on version 4.11 does not require specific additional steps. For details on installing KMM on version 4.10 and earlier, see the section "Installing the Kernel Module Management Operator on earlier versions of OKD".

Installing the Kernel Module Management Operator using the web console

As a cluster administrator, you can install the Kernel Module Management (KMM) Operator using the OKD web console.

Procedure

  1. Log in to the OKD web console.

  2. Install the Kernel Module Management Operator:

    1. In the OKD web console, click Operators → OperatorHub.

    2. Select Kernel Module Management Operator from the list of available Operators, and then click Install.

    3. From the Installed Namespace list, select the openshift-kmm namespace.

    4. Click Install.

Verification

To verify that KMM Operator installed successfully:

  1. Navigate to the Operators → Installed Operators page.

  2. Ensure that Kernel Module Management Operator is listed in the openshift-kmm project with a Status of InstallSucceeded.

    During installation, an Operator might display a Failed status. If the installation later succeeds with an InstallSucceeded message, you can ignore the Failed message.
Troubleshooting

  1. To troubleshoot issues with Operator installation:

    1. Navigate to the Operators → Installed Operators page and inspect the Operator Subscriptions and Install Plans tabs for any failure or errors under Status.

    2. Navigate to the Workloads → Pods page and check the logs for pods in the openshift-kmm project.

Installing the Kernel Module Management Operator by using the CLI

As a cluster administrator, you can install the Kernel Module Management (KMM) Operator by using the OpenShift CLI.

Prerequisites

  • You have a running OKD cluster.

  • You installed the OpenShift CLI (oc).

  • You are logged into the OpenShift CLI as a user with cluster-admin privileges.

Procedure

  1. Install KMM in the openshift-kmm namespace:

    1. Create the following Namespace CR and save the YAML file, for example, kmm-namespace.yaml:

      apiVersion: v1
kind: Namespace
metadata:
  name: openshift-kmm

    2. Create the following OperatorGroup CR and save the YAML file, for example, kmm-op-group.yaml:

      apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: kernel-module-management
  namespace: openshift-kmm

    3. Create the following Subscription CR and save the YAML file, for example, kmm-sub.yaml:

      apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: kernel-module-management
  namespace: openshift-kmm
spec:
  channel: release-1.0
  installPlanApproval: Automatic
  name: kernel-module-management
  source: redhat-operators
  sourceNamespace: openshift-marketplace
  startingCSV: kernel-module-management.v1.0.0

    4. Create the subscription object by running the following command:

      $ oc create -f kmm-sub.yaml
Verification

  • To verify that the Operator deployment is successful, run the following command:

    $ oc get -n openshift-kmm deployments.apps kmm-operator-controller
    Example output
    NAME                              READY UP-TO-DATE  AVAILABLE AGE
kmm-operator-controller           1/1   1           1         97s

    The Operator is available.

Installing the Kernel Module Management Operator on earlier versions of OKD

The KMM Operator is supported on OKD 4.12 and later. For version 4.10 and earlier, you must create a new SecurityContextConstraint object and bind it to the Operator’s ServiceAccount. As a cluster administrator, you can install the Kernel Module Management (KMM) Operator by using the OpenShift CLI.

Prerequisites

  • You have a running OKD cluster.

  • You installed the OpenShift CLI (oc).

  • You are logged into the OpenShift CLI as a user with cluster-admin privileges.

Procedure

  1. Install KMM in the openshift-kmm namespace:

    1. Create the following Namespace CR and save the YAML file, for example, kmm-namespace.yaml file:

      apiVersion: v1
kind: Namespace
metadata:
  name: openshift-kmm

    2. Create the following SecurityContextConstraint object and save the YAML file, for example, kmm-security-constraint.yaml:

      allowHostDirVolumePlugin: false
allowHostIPC: false
allowHostNetwork: false
allowHostPID: false
allowHostPorts: false
allowPrivilegeEscalation: false
allowPrivilegedContainer: false
allowedCapabilities:
  - NET_BIND_SERVICE
apiVersion: security.openshift.io/v1
defaultAddCapabilities: null
fsGroup:
  type: MustRunAs
groups: []
kind: SecurityContextConstraints
metadata:
  name: restricted-v2
priority: null
readOnlyRootFilesystem: false
requiredDropCapabilities:
  - ALL
runAsUser:
  type: MustRunAsRange
seLinuxContext:
  type: MustRunAs
seccompProfiles:
  - runtime/default
supplementalGroups:
  type: RunAsAny
users: []
volumes:
  - configMap
  - downwardAPI
  - emptyDir
  - persistentVolumeClaim
  - projected
  - secret

    3. Bind the SecurityContextConstraint object to the Operator’s ServiceAccount by running the following commands:

      $ oc apply -f kmm-security-constraint.yaml
      $ oc adm policy add-scc-to-user kmm-security-constraint -z kmm-operator-controller -n openshift-kmm

    4. Create the following OperatorGroup CR and save the YAML file, for example, kmm-op-group.yaml:

      apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: kernel-module-management
  namespace: openshift-kmm

    5. Create the following Subscription CR and save the YAML file, for example, kmm-sub.yaml:

      apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: kernel-module-management
  namespace: openshift-kmm
spec:
  channel: release-1.0
  installPlanApproval: Automatic
  name: kernel-module-management
  source: redhat-operators
  sourceNamespace: openshift-marketplace
  startingCSV: kernel-module-management.v1.0.0

    6. Create the subscription object by running the following command:

      $ oc create -f kmm-sub.yaml
Verification

  • To verify that the Operator deployment is successful, run the following command:

    $ oc get -n openshift-kmm deployments.apps kmm-operator-controller
    Example output
    NAME                              READY UP-TO-DATE  AVAILABLE AGE
kmm-operator-controller           1/1   1           1         97s

    The Operator is available.

Configuring the Kernel Module Management Operator

In most cases, the default configuration for the Kernel Module Management (KMM) Operator does not need to be modified. However, you can modify the Operator settings to suit your environment using the following procedure.

The Operator configuration is set in the kmm-operator-manager-config ConfigMap in the Operator namespace.

Procedure

  1. To modify the settings, edit the ConfigMap data by entering the following command:

    $ oc edit configmap -n "$namespace" kmm-operator-manager-config
    Example output
    healthProbeBindAddress: :8081
job:
  gcDelay: 1h
leaderElection:
  enabled: true
  resourceID: kmm.sigs.x-k8s.io
webhook:
  disableHTTP2: true  # CVE-2023-44487
  port: 9443
metrics:
  enableAuthnAuthz: true
  disableHTTP2: true  # CVE-2023-44487
  bindAddress: 0.0.0.0:8443
  secureServing: true
worker:
  runAsUser: 0
  seLinuxType: spc_t
  setFirmwareClassPath: /var/lib/firmware
    Table 1. Operator configuration parameters
    Parameter Description

    healthProbeBindAddress

    Defines the address on which the Operator monitors for kubelet health probes. The recommended value is :8081.

    job.gcDelay

    Defines the duration that successful build pods should be preserved for before they are deleted. There is no recommended value for this setting. For information about the valid values for this setting, see ParseDuration.

    leaderElection.enabled

    Determines whether leader election is used to ensure that only one replica of the KMM Operator is running at any time. For more information, see Leases. The recommended value is true.

    leaderElection.resourceID

    Determines the name of the resource that leader election uses for holding the leader lock. The recommended value is kmm.sigs.x-k8s.io.

    webhook.disableHTTP2

    If true, disables HTTP/2 for the webhook server, as a mitigation for cve-2023-44487. The recommended value is true.

    webhook.port

    Defines the port on which the Operator monitors webhook requests. The recommended value is 9443.

    metrics.enableAuthnAuthz

    Determines if metrics are authenticated using TokenReviews and authorized using SubjectAccessReviews with the kube-apiserver.

    For authentication and authorization, the controller needs a ClusterRole with the following rules:

    • apiGroups: authentication.k8s.io, resources: tokenreviews, verbs: create

    • apiGroups: authorization.k8s.io, resources: subjectaccessreviews, verbs: create

    To scrape metrics, for example, using Prometheus, the client needs a ClusterRole with the following rule:

    • nonResourceURLs: "/metrics", verbs: get

    The recommended value is true.

    metrics.disableHTTP2

    If true, disables HTTP/2 for the metrics server as a mitigation for CVE-2023-44487. The recommended value is true.

    metrics.bindAddress

    Determines the bind address for the metrics server. If unspecified, the default is :8080. To disable the metrics server, set to 0. The recommended value is 0.0.0.0:8443.

    metrics.secureServing

    Determines whether the metrics are served over HTTPS instead of HTTP. The recommended value is true.

    worker.runAsUser

    Determines the value of the runAsUser field of the worker container’s security context. For more information, see SecurityContext. The recommended value is 9443.

    worker.seLinuxType

    Determines the value of the seLinuxOptions.type field of the worker container’s security context. For more information, see SecurityContext. The recommended value is spc_t.

    worker.setFirmwareClassPath

    Sets the kernel’s firmware search path into the /sys/module/firmware_class/parameters/path file on the node. The recommended value is /var/lib/firmware if you need to set that value through the worker app. Otherwise, unset.

  2. After modifying the settings, restart the controller with the following command:

    $ oc delete pod -n "<namespace>" -l app.kubernetes.io/component=kmm

    The value of <namespace> depends on your original installation method.
Additional resources

Unloading the kernel module

You must unload the kernel modules when moving to a newer version or if they introduce some undesirable side effect on the node.

Procedure

  • To unload a module loaded with KMM from nodes, delete the corresponding Module resource. KMM then creates worker pods, where required, to run modprobe -r and unload the kernel module from the nodes.

    When unloading worker pods, KMM needs all the resources it uses when loading the kernel module. This includes the ServiceAccount referenced in the Module as well as any RBAC defined to allow privileged KMM worker Pods to run. It also includes any pull secret referenced in .spec.imageRepoSecret.

    To avoid situations where KMM is unable to unload the kernel module from nodes, make sure those resources are not deleted while the Module resource is still present in the cluster in any state, including Terminating. KMM includes a validating admission webhook that rejects the deletion of namespaces that contain at least one Module resource.

Setting the kernel firmware search path

The Linux kernel accepts the firmware_class.path parameter as a search path for firmware, as explained in Firmware search paths.

KMM worker pods can set this value on nodes by writing to sysfs before attempting to load kmods.

Procedure

  • To define a firmware search path, set worker.setFirmwareClassPath to /var/lib/firmware in the Operator configuration.

Additional resources

Uninstalling the Kernel Module Management Operator

Use one of the following procedures to uninstall the Kernel Module Management (KMM) Operator, depending on how the KMM Operator was installed.

Uninstalling a Red Hat catalog installation

Use this procedure if KMM was installed from the Red Hat catalog.

Procedure

Use the following method to uninstall the KMM Operator:

  • Use the OpenShift console under Operators -→ Installed Operators to locate and uninstall the Operator.

Alternatively, you can delete the Subscription resource in the KMM namespace.

Uninstalling a CLI installation

Use this command if the KMM Operator was installed using the OpenShift CLI.

Procedure

  • Run the following command to uninstall the KMM Operator:

    $ oc delete -k https://github.com/rh-ecosystem-edge/kernel-module-management/config/default

    Using this command deletes the Module CRD and all Module instances in the cluster.

Kernel module deployment

Kernel Module Management (KMM) monitors Node and Module resources in the cluster to determine if a kernel module should be loaded on or unloaded from a node.

To be eligible for a module, a node must contain the following:

  • Labels that match the module’s .spec.selector field.

  • A kernel version matching one of the items in the module’s .spec.moduleLoader.container.kernelMappings field.

  • If ordered upgrade (ordered_upgrade.md) is configured in the module, a label that matches its .spec.moduleLoader.container.version field.

When KMM reconciles nodes with the desired state as configured in the Module resource, it creates worker pods on the target nodes to run the necessary action. The KMM Operator monitors the outcome of the pods and records the information. The Operator uses this information to label the Node objects when the module is successfully loaded, and to run the device plugin, if configured.

Worker pods run the KMM worker binary that performs the following tasks:

  • Pulls the kmod image configured in the Module resource. Kmod images are standard OCI images that contain .ko files.

  • Extracts the image in the pod’s filesystem.

  • Runs modprobe with the specified arguments to perform the necessary action.

The Module custom resource definition

The Module custom resource definition (CRD) represents a kernel module that can be loaded on all or select nodes in the cluster, through a kmod image. A Module custom resource (CR) specifies one or more kernel versions with which it is compatible, and a node selector.

The compatible versions for a Module resource are listed under .spec.moduleLoader.container.kernelMappings. A kernel mapping can either match a literal version, or use regexp to match many of them at the same time.

The reconciliation loop for the Module resource runs the following steps:

  1. List all nodes matching .spec.selector.

  2. Build a set of all kernel versions running on those nodes.

  3. For each kernel version:

    1. Go through .spec.moduleLoader.container.kernelMappings and find the appropriate container image name. If the kernel mapping has build or sign defined and the container image does not already exist, run the build, the signing pod, or both, as needed.

    2. Create a worker pod to pull the container image determined in the previous step and run modprobe.

    3. If .spec.devicePlugin is defined, create a device plugin daemon set using the configuration specified under .spec.devicePlugin.container.

  4. Run garbage-collect on:

    1. Obsolete device plugin DaemonSets that do not target any node.

    2. Successful build pods.

    3. Successful signing pods.

Set soft dependencies between kernel modules

Some configurations require that several kernel modules be loaded in a specific order to work properly, even though the modules do not directly depend on each other through symbols. These are called soft dependencies. depmod is usually not aware of these dependencies, and they do not appear in the files it produces. For example, if mod_a has a soft dependency on mod_b, modprobe mod_a will not load mod_b.

You can resolve these situations by declaring soft dependencies in the Module custom resource definition (CRD) using the modulesLoadingOrder field.

# ...
spec:
  moduleLoader:
    container:
      modprobe:
        moduleName: mod_a
        dirName: /opt
        firmwarePath: /firmware
        parameters:
          - param=1
        modulesLoadingOrder:
          - mod_a
          - mod_b

In the configuration above, the worker pod will first try to unload the in-tree mod_b before loading mod_a from the kmod image. When the worker pod is terminated and mod_a is unloaded, mod_b will not be loaded again.

The first value in the list, to be loaded last, must be equivalent to the moduleName.

Security and permissions

Loading kernel modules is a highly sensitive operation. After they are loaded, kernel modules have all possible permissions to do any kind of operation on the node.

ServiceAccounts and SecurityContextConstraints

Kernel Module Management (KMM) creates a privileged workload to load the kernel modules on nodes. That workload needs ServiceAccounts allowed to use the privileged SecurityContextConstraint (SCC) resource.

The authorization model for that workload depends on the namespace of the Module resource, as well as its spec.

  • If the .spec.moduleLoader.serviceAccountName or .spec.devicePlugin.serviceAccountName fields are set, they are always used.

  • If those fields are not set, then:

    • If the Module resource is created in the Operator’s namespace (openshift-kmm by default), then KMM uses its default, powerful ServiceAccounts to run the worker and device plugin pods.

    • If the Module resource is created in any other namespace, then KMM runs the pods with the namespace’s default ServiceAccount. The Module resource cannot run a privileged workload unless you manually enable it to use the privileged SCC.

openshift-kmm is a trusted namespace.

When setting up RBAC permissions, remember that any user or ServiceAccount creating a Module resource in the openshift-kmm namespace results in KMM automatically running privileged workloads on potentially all nodes in the cluster.

To allow any ServiceAccount to use the privileged SCC and run worker or device plugin pods, you can use the oc adm policy command, as in the following example:

$ oc adm policy add-scc-to-user privileged -z "${serviceAccountName}" [ -n "${namespace}" ]

Pod security standards

OpenShift runs a synchronization mechanism that sets the namespace Pod Security level automatically based on the security contexts in use. No action is needed.

Additional resources

Replacing in-tree modules with out-of-tree modules

You can use Kernel Module Management (KMM) to build kernel modules that can be loaded or unloaded into the kernel on demand. These modules extend the functionality of the kernel without the need to reboot the system. Modules can be configured as built-in or dynamically loaded.

Dynamically loaded modules include in-tree modules and out-of-tree (OOT) modules. In-tree modules are internal to the Linux kernel tree, that is, they are already part of the kernel. Out-of-tree modules are external to the Linux kernel tree. They are generally written for development and testing purposes, such as testing the new version of a kernel module that is shipped in-tree, or to deal with incompatibilities.

Some modules that are loaded by KMM could replace in-tree modules that are already loaded on the node. To unload in-tree modules before loading your module, set the value of the .spec.moduleLoader.container.inTreeModulesToRemove field to the modules that you want to unload. The following example demonstrates module replacement for all kernel mappings:

# ...
spec:
  moduleLoader:
    container:
      modprobe:
        moduleName: mod_a

      inTreeModulesToRemove: [mod_a, mod_b]

In this example, the moduleLoader pod uses inTreeModulesToRemove to unload the in-tree mod_a and mod_b before loading mod_a from the moduleLoader image. When the moduleLoader`pod is terminated and `mod_a is unloaded, mod_b is not loaded again.

The following is an example for module replacement for specific kernel mappings:

# ...
spec:
  moduleLoader:
    container:
      kernelMappings:
        - literal: 6.0.15-300.fc37.x86_64
          containerImage: "some.registry/org/my-kmod:${KERNEL_FULL_VERSION}"
          inTreeModulesToRemove: [<module_name>, <module_name>]
Additional resources

Example Module CR

The following is an annotated Module example:

apiVersion: kmm.sigs.x-k8s.io/v1beta1
kind: Module
metadata:
  name: <my_kmod>
spec:
  moduleLoader:
    container:
      modprobe:
        moduleName: <my_kmod> (1)
        dirName: /opt (2)
        firmwarePath: /firmware (3)
        parameters:  (4)
          - param=1
      kernelMappings:  (5)
        - literal: 6.0.15-300.fc37.x86_64
          containerImage: some.registry/org/my-kmod:6.0.15-300.fc37.x86_64
        - regexp: '^.+\fc37\.x86_64$' (6)
          containerImage: "some.other.registry/org/<my_kmod>:${KERNEL_FULL_VERSION}"
        - regexp: '^.+$' (7)
          containerImage: "some.registry/org/<my_kmod>:${KERNEL_FULL_VERSION}"  (8)
          build:
            buildArgs:  (9)
              - name: ARG_NAME
                value: <some_value>
            secrets:
              - name: <some_kubernetes_secret>  (10)
            baseImageRegistryTLS: (11)
              insecure: false
              insecureSkipTLSVerify: false  (12)
            dockerfileConfigMap:  (13)
              name: <my_kmod_dockerfile>
          sign:
            certSecret:
              name: <cert_secret>  (14)
            keySecret:
              name: <key_secret>  (15)
            filesToSign:
              - /opt/lib/modules/${KERNEL_FULL_VERSION}/<my_kmod>.ko
          registryTLS: (16)
            insecure: false (17)
            insecureSkipTLSVerify: false
    serviceAccountName: <sa_module_loader>  (18)
  devicePlugin:  (19)
    container:
      image: some.registry/org/device-plugin:latest  (20)
      env:
        - name: MY_DEVICE_PLUGIN_ENV_VAR
          value: SOME_VALUE
      volumeMounts:  (21)
        - mountPath: /some/mountPath
          name: <device_plugin_volume>
    volumes:  (22)
      - name: <device_plugin_volume>
        configMap:
          name: <some_configmap>
    serviceAccountName: <sa_device_plugin> (23)
  imageRepoSecret:  (24)
    name: <secret_name>
  selector:
    node-role.kubernetes.io/worker: ""
1 Required.
2 Optional.
3 Optional: Copies the contents of this path into the path specified in worker.setFirmwareClassPath (which is preset to /var/lib/firmware) of the kmm-operator-manager-config config map. This action occurs before modprobe is called to insert the kernel module.
4 Optional.
5 At least one kernel item is required.
6 For each node running a kernel matching the regular expression, KMM checks if you have included a tag or a digest. If you have not specified a tag or digest in the container image, then the validation webhook returns an error and does not apply the module.
7 For any other kernel, build the image using the Dockerfile in the my-kmod ConfigMap.
8 The container image that holds the customer’s kmods. This container should contain the cp binary.
9 Optional.
10 Optional: A value for some-kubernetes-secret can be obtained from the build environment at /run/secrets/some-kubernetes-secret.
11 This field has no effect. When building kmod images or signing kmods within a kmod image, you might sometimes need to pull base images from a registry that serves a certificate signed by an untrusted Certificate Authority (CA). In order for KMM to trust that CA, it must also trust the new CA by replacing the cluster’s CA bundle.

See "Additional resources" to learn how to replace the cluster’s CA bundle.
12 Optional: Avoid using this parameter. If set to true, the build skips any TLS server certificate validation when pulling the image in the Dockerfile FROM instruction using plain HTTP.
13 Required.
14 Required: A secret holding the public secureboot key with the key 'cert'.
15 Required: A secret holding the private secureboot key with the key 'key'.
16 Optional: Avoid using this parameter. If set to true, KMM is allowed to check if the container image already exists using plain HTTP.
17 Optional: Avoid using this parameter. If set to true, KMM skips any TLS server certificate validation when checking if the container image already exists.
18 Optional.
19 Optional.
20 Required: If the device plugin section is present.
21 Optional.
22 Optional.
23 Optional.
24 Optional: Used to pull module loader and device plugin images.
Additional resources

Some kernel modules depend on other kernel modules that are shipped with the node’s operating system. To avoid copying those dependencies into the kmod image, Kernel Module Management (KMM) mounts /usr/lib/modules into both the build and the worker pod’s filesystems.

By creating a symlink from /opt/usr/lib/modules/<kernel_version>/<symlink_name> to /usr/lib/modules/<kernel_version>, depmod can use the in-tree kmods on the building node’s filesystem to resolve dependencies.

At runtime, the worker pod extracts the entire image, including the <symlink_name> symbolic link. That symbolic link points to /usr/lib/modules/<kernel_version> in the worker pod, which is mounted from the node’s filesystem. modprobe can then follow that link and load the in-tree dependencies as needed.

In the following example, host is the symbolic link name under /opt/usr/lib/modules/<kernel_version>:

ARG DTK_AUTO

FROM ${DTK_AUTO} as builder

#
# Build steps
#

FROM ubi9/ubi

ARG KERNEL_FULL_VERSION

RUN dnf update && dnf install -y kmod

COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_a.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/
COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_b.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/

# Create the symbolic link
RUN ln -s /lib/modules/${KERNEL_FULL_VERSION} /opt/lib/modules/${KERNEL_FULL_VERSION}/host

RUN depmod -b /opt ${KERNEL_FULL_VERSION}

depmod generates dependency files based on the kernel modules present on the node that runs the kmod image build.

On the node on which KMM loads the kernel modules, modprobe expects the files to be present under /usr/lib/modules/<kernel_version>, and the same filesystem layout. It is highly recommended that the build and the target nodes share the same operating system and release.

Creating a kmod image

Kernel Module Management (KMM) works with purpose-built kmod images, which are standard OCI images that contain .ko files. The location of the .ko files must match the following pattern: <prefix>/lib/modules/[kernel-version]/.

Keep the following in mind when working with the .ko files:

  • In most cases, <prefix> should be equal to /opt. This is the Module CRD’s default value.

  • kernel-version must not be empty and must be equal to the kernel version the kernel modules were built for.

Running depmod

It is recommended to run depmod at the end of the build process to generate modules.dep and .map files. This is especially useful if your kmod image contains several kernel modules and if one of the modules depends on another module.

You must have a Red Hat subscription to download the kernel-devel package.
Procedure

  • Generate modules.dep and .map files for a specific kernel version by running the following command:

    $ depmod -b /opt ${KERNEL_FULL_VERSION}+`.

Example Dockerfile

If you are building your image on OKD, consider using the Driver Tool Kit (DTK).

For further information, see using an entitled build.

apiVersion: v1
kind: ConfigMap
metadata:
  name: kmm-ci-dockerfile
data:
  dockerfile: |
    ARG DTK_AUTO
    FROM ${DTK_AUTO} as builder
    ARG KERNEL_FULL_VERSION
    WORKDIR /usr/src
    RUN ["git", "clone", "https://github.com/rh-ecosystem-edge/kernel-module-management.git"]
    WORKDIR /usr/src/kernel-module-management/ci/kmm-kmod
    RUN KERNEL_SRC_DIR=/lib/modules/${KERNEL_FULL_VERSION}/build make all
    FROM registry.redhat.io/ubi9/ubi-minimal
    ARG KERNEL_FULL_VERSION
    RUN microdnf install kmod
    COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_a.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/
    COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_b.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/
    RUN depmod -b /opt ${KERNEL_FULL_VERSION}
Additional resources

Building in the cluster

KMM can build kmod images in the cluster. Follow these guidelines:

  • Provide build instructions using the build section of a kernel mapping.

  • Copy the Dockerfile for your container image into a ConfigMap resource, under the dockerfile key.

  • Ensure that the ConfigMap is located in the same namespace as the Module.

KMM checks if the image name specified in the containerImage field exists. If it does, the build is skipped.

Otherwise, KMM creates a Build resource to build your image. After the image is built, KMM proceeds with the Module reconciliation. See the following example.

# ...
- regexp: '^.+$'
  containerImage: "some.registry/org/<my_kmod>:${KERNEL_FULL_VERSION}"
  build:
    buildArgs:  (1)
      - name: ARG_NAME
        value: <some_value>
    secrets: (2)
      - name: <some_kubernetes_secret> (3)
    baseImageRegistryTLS:
      insecure: false (4)
      insecureSkipTLSVerify: false (5)
    dockerfileConfigMap:  (6)
      name: <my_kmod_dockerfile>
  registryTLS:
    insecure: false (7)
    insecureSkipTLSVerify: false (8)
1 Optional.
2 Optional.
3 Will be mounted in the build pod as /run/secrets/some-kubernetes-secret.
4 Optional: Avoid using this parameter. If set to true, the build will be allowed to pull the image in the Dockerfile FROM instruction using plain HTTP.
5 Optional: Avoid using this parameter. If set to true, the build will skip any TLS server certificate validation when pulling the image in the Dockerfile FROM instruction using plain HTTP.
6 Required.
7 Optional: Avoid using this parameter. If set to true, KMM will be allowed to check if the container image already exists using plain HTTP.
8 Optional: Avoid using this parameter. If set to true, KMM will skip any TLS server certificate validation when checking if the container image already exists.

Successful build pods are garbage collected immediately, unless the job.gcDelay parameter is set in the Operator configuration. Failed build pods are always preserved and must be deleted manually by the administrator for the build to be restarted.

Additional resources

Using the Driver Toolkit

The Driver Toolkit (DTK) is a convenient base image for building build kmod loader images. It contains tools and libraries for the OpenShift version currently running in the cluster.

Procedure

Use DTK as the first stage of a multi-stage Dockerfile.

  1. Build the kernel modules.

  2. Copy the .ko files into a smaller end-user image such as ubi-minimal.

  3. To leverage DTK in your in-cluster build, use the DTK_AUTO build argument. The value is automatically set by KMM when creating the Build resource. See the following example.

    ARG DTK_AUTO
FROM ${DTK_AUTO} as builder
ARG KERNEL_FULL_VERSION
WORKDIR /usr/src
RUN ["git", "clone", "https://github.com/rh-ecosystem-edge/kernel-module-management.git"]
WORKDIR /usr/src/kernel-module-management/ci/kmm-kmod
RUN KERNEL_SRC_DIR=/lib/modules/${KERNEL_FULL_VERSION}/build make all
FROM ubi9/ubi-minimal
ARG KERNEL_FULL_VERSION
RUN microdnf install kmod
COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_a.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/
COPY --from=builder /usr/src/kernel-module-management/ci/kmm-kmod/kmm_ci_b.ko /opt/lib/modules/${KERNEL_FULL_VERSION}/
RUN depmod -b /opt ${KERNEL_FULL_VERSION}
Additional resources

Using signing with Kernel Module Management (KMM)

On a Secure Boot enabled system, all kernel modules (kmods) must be signed with a public/private key-pair enrolled into the Machine Owner’s Key (MOK) database. Drivers distributed as part of a distribution should already be signed by the distribution’s private key, but for kernel modules build out-of-tree, KMM supports signing kernel modules using the sign section of the kernel mapping.

For more details on using Secure Boot, see Generating a public and private key pair

Prerequisites

  • A public private key pair in the correct (DER) format.

  • At least one secure-boot enabled node with the public key enrolled in its MOK database.

  • Either a pre-built driver container image, or the source code and Dockerfile needed to build one in-cluster.

Adding the keys for secureboot

To use KMM Kernel Module Management (KMM) to sign kernel modules, a certificate and private key are required. For details on how to create these, see Generating a public and private key pair.

For details on how to extract the public and private key pair, see Signing kernel modules with the private key. Use steps 1 through 4 to extract the keys into files.

Procedure

  1. Create the sb_cert.cer file that contains the certificate and the sb_cert.priv file that contains the private key:

    $ openssl req -x509 -new -nodes -utf8 -sha256 -days 36500 -batch -config configuration_file.config -outform DER -out my_signing_key_pub.der -keyout my_signing_key.priv

  2. Add the files by using one of the following methods:

    • Add the files as secrets directly:

      $ oc create secret generic my-signing-key --from-file=key=<my_signing_key.priv>
      $ oc create secret generic my-signing-key-pub --from-file=cert=<my_signing_key_pub.der>

    • Add the files by base64 encoding them:

      $ cat sb_cert.priv | base64 -w 0 > my_signing_key2.base64
      $ cat sb_cert.cer | base64 -w 0 > my_signing_key_pub.base64

  3. Add the encoded text to a YAML file:

    apiVersion: v1
kind: Secret
metadata:
  name: my-signing-key-pub
  namespace: default (1)
type: Opaque
data:
  cert: <base64_encoded_secureboot_public_key>

---
apiVersion: v1
kind: Secret
metadata:
  name: my-signing-key
  namespace: default (1)
type: Opaque
data:
  key: <base64_encoded_secureboot_private_key>
    1 namespace - Replace default with a valid namespace.

  4. Apply the YAML file:

    $ oc apply -f <yaml_filename>

Checking the keys

After you have added the keys, you must check them to ensure they are set correctly.

Procedure

  1. Check to ensure the public key secret is set correctly:

    $ oc get secret -o yaml <certificate secret name> | awk '/cert/{print $2; exit}' | base64 -d  | openssl x509 -inform der -text

    This should display a certificate with a Serial Number, Issuer, Subject, and more.

  2. Check to ensure the private key secret is set correctly:

    $ oc get secret -o yaml <private key secret name> | awk '/key/{print $2; exit}' | base64 -d

    This should display the key enclosed in the -----BEGIN PRIVATE KEY----- and -----END PRIVATE KEY----- lines.

Signing kmods in a pre-built image

Use this procedure if you have a pre-built image, such as an image either distributed by a hardware vendor or built elsewhere.

The following YAML file adds the public/private key-pair as secrets with the required key names - key for the private key, cert for the public key. The cluster then pulls down the unsignedImage image, opens it, signs the kernel modules listed in filesToSign, adds them back, and pushes the resulting image as containerImage.

KMM then loads the signed kmods onto all the nodes with that match the selector. The kmods are successfully loaded on any nodes that have the public key in their MOK database, and any nodes that are not secure-boot enabled, which will ignore the signature.

Prerequisites

  • The keySecret and certSecret secrets have been created in the same namespace as the rest of the resources.

Procedure

  • Apply the YAML file:

    ---
apiVersion: kmm.sigs.x-k8s.io/v1beta1
kind: Module
metadata:
  name: example-module
spec:
  moduleLoader:
    serviceAccountName: default
    container:
      modprobe: (1)
        moduleName: '<module_name>'
      kernelMappings:
        # the kmods will be deployed on all nodes in the cluster with a kernel that matches the regexp
        - regexp: '^.*\.x86_64$'
          # the container to produce containing the signed kmods
          containerImage: <image_name> (2)
          sign:
            # the image containing the unsigned kmods (we need this because we are not building the kmods within the cluster)
            unsignedImage: <image_name> (3)
            keySecret: # a secret holding the private secureboot key with the key 'key'
              name: <private_key_secret_name>
            certSecret: # a secret holding the public secureboot key with the key 'cert'
              name: <certificate_secret_name>
            filesToSign: # full path within the unsignedImage container to the kmod(s) to sign
              - /opt/lib/modules/4.18.0-348.2.1.el8_5.x86_64/kmm_ci_a.ko
  imageRepoSecret:
    # the name of a secret containing credentials to pull unsignedImage and push containerImage to the registry
    name: repo-pull-secret
  selector:
    kubernetes.io/arch: amd64
1 The name of the kmod to load.
2 The name of the container image. For example, quay.io/myuser/my-driver:<kernelversion.
3 The name of the unsigned image. For example, quay.io/myuser/my-driver:<kernelversion.

Building and signing a kmod image

Use this procedure if you have source code and must build your image first.

The following YAML file builds a new container image using the source code from the repository. The image produced is saved back in the registry with a temporary name, and this temporary image is then signed using the parameters in the sign section.

The temporary image name is based on the final image name and is set to be <containerImage>:<tag>-<namespace>_<module name>_kmm_unsigned.

For example, using the following YAML file, Kernel Module Management (KMM) builds an image named example.org/repository/minimal-driver:final-default_example-module_kmm_unsigned containing the build with unsigned kmods and pushes it to the registry. Then it creates a second image named example.org/repository/minimal-driver:final that contains the signed kmods. It is this second image that is pulled by the worker pods and contains the kmods to be loaded on the cluster nodes.

After it is signed, you can safely delete the temporary image from the registry. It will be rebuilt, if needed.

Prerequisites

  • The keySecret and certSecret secrets have been created in the same namespace as the rest of the resources.

Procedure

  • Apply the YAML file:

    ---
apiVersion: v1
kind: ConfigMap
metadata:
  name: example-module-dockerfile
  namespace: <namespace> (1)
data:
  Dockerfile: |
    ARG DTK_AUTO
    ARG KERNEL_VERSION
    FROM ${DTK_AUTO} as builder
    WORKDIR /build/
    RUN git clone -b main --single-branch https://github.com/rh-ecosystem-edge/kernel-module-management.git
    WORKDIR kernel-module-management/ci/kmm-kmod/
    RUN make
    FROM registry.access.redhat.com/ubi9/ubi:latest
    ARG KERNEL_VERSION
    RUN yum -y install kmod && yum clean all
    RUN mkdir -p /opt/lib/modules/${KERNEL_VERSION}
    COPY --from=builder /build/kernel-module-management/ci/kmm-kmod/*.ko /opt/lib/modules/${KERNEL_VERSION}/
    RUN /usr/sbin/depmod -b /opt
---
apiVersion: kmm.sigs.x-k8s.io/v1beta1
kind: Module
metadata:
  name: example-module
  namespace: <namespace> (1)
spec:
  moduleLoader:
    serviceAccountName: default (2)
    container:
      modprobe:
        moduleName: simple_kmod
      kernelMappings:
        - regexp: '^.*\.x86_64$'
          containerImage: <final_driver_container_name>
          build:
            dockerfileConfigMap:
              name: example-module-dockerfile
          sign:
            keySecret:
              name: <private_key_secret_name>
            certSecret:
              name: <certificate_secret_name>
            filesToSign:
              - /opt/lib/modules/4.18.0-348.2.1.el8_5.x86_64/kmm_ci_a.ko
  imageRepoSecret: (3)
    name: repo-pull-secret
  selector: # top-level selector
    kubernetes.io/arch: amd64
1 Replace default with a valid namespace.
2 The default serviceAccountName does not have the required permissions to run a module that is privileged. For information on creating a service account, see "Creating service accounts" in the "Additional resources" of this section.
3 Used as imagePullSecrets in the DaemonSet object and to pull and push for the build and sign features.
Additional resources

Using tolerations for kernel module scheduling

There are circumstances where you need to evacuate workloads on a node before upgrading a kernel module, which you can do through taints. You can use a taint to schedule only pods that contain a matching toleration on the node.

However, you also need to set tolerations to allow Kernel Module Management (KMM) pods that run housekeeping operations, such as kernel module upgrades. The tolerations must match the taint that is added to the nodes.

You can create user-defined tolerations to kernel modules to schedule selected KMM pods on a cordoned node. For example, during a device driver upgrade you cordon a node, at the same time, you can run housekeeping pods that perform the driver upgrades.

The ModuleSpec field is used to carry the tolerations to the KMM pods that are used during pod creation.

Applying tolerations to kernel module pods

Taints and tolerations consist of effect, key, and value parameters. Tolerations include additional operator and tolerationSeconds parameters.

effect

Indicates the taint effect to match. If left empty, all taint effects are matched. When you set effect, valid values are: NoSchedule, PreferNoSchedule, or NoExecute.

key

The taint key that the toleration applies to. If left empty, all taint keys are matched. If the key is empty, you must set the operator parameter to Exists. This combination matches all values and all keys.

value

The taint value the toleration matches to. If the operator parameter is Exists, the value must be empty, otherwise use a regular string.

operator

Represents a relationship of a key to the value. Valid operator parameters are Exists and Equal. The default value is Equal. Exists is equivalent to wildcard for value, so that a pod can tolerate all taints of a particular category.

tolerationSeconds

Represents the period of time the toleration (which must be of effect NoExecute, otherwise this field is ignored) tolerates the taint. By default, it is not set and the taint is tolerated forever without eviction. Zero and negative values are treated as 0 and immediately evicted by the system.

Example taint in a node specification
apiVersion: v1
kind: Node
metadata:
  name: <my_node>
#...
spec:
  taints:
  - effect: NoSchedule
    key: key1
    value: value1
#...
Example toleration in a module specification
apiVersion: kmm.sigs.x-k8s.io/v1beta1
kind: Module
metadata:
  name: <my_kmod>
spec:
  ...
  tolerations:
    effect: NoSchedule
    key: key1
    operator: Equal
    tolerationSeconds: 36000
    value: value1

Toleration values must match the taint that is added to the nodes. A toleration matches a taint:

  • If the operator parameter is set to Equal:

    • the key parameters are the same;

    • the value parameters are the same;

    • the effect parameters are the same.

  • If the operator parameter is set to Exists:

    • the key parameters are the same;

    • the effect parameters are the same.

Additional resources

KMM hub and spoke

In hub and spoke scenarios, many spoke clusters are connected to a central, powerful hub cluster. Kernel Module Management (KMM) depends on Red Hat Advanced Cluster Management (RHACM) to operate in hub and spoke environments.

KMM is compatible with hub and spoke environments through decoupling KMM features. A ManagedClusterModule custom resource definition (CRD) is provided to wrap the existing Module CRD and extend it to select Spoke clusters. Also provided is KMM-Hub, a new standalone controller that builds images and signs modules on the hub cluster.

In hub and spoke setups, spokes are focused, resource-constrained clusters that are centrally managed by a hub cluster. Spokes run the single-cluster edition of KMM, with those resource-intensive features disabled. To adapt KMM to this environment, you should reduce the workload running on the spokes to the minimum, while the hub takes care of the expensive tasks.

Building kernel module images and signing the .ko files, should run on the hub. The scheduling of the Module Loader and Device Plugin DaemonSets can only happen on the spokes.

Additional resources

KMM-Hub

The KMM project provides KMM-Hub, an edition of KMM dedicated to hub clusters. KMM-Hub monitors all kernel versions running on the spokes and determines the nodes on the cluster that should receive a kernel module.

KMM-Hub runs all compute-intensive tasks such as image builds and kmod signing, and prepares the trimmed-down Module to be transferred to the spokes through RHACM.

KMM-Hub cannot be used to load kernel modules on the hub cluster. Install the regular edition of KMM to load kernel modules.
Additional resources

Installing KMM-Hub

You can use one of the following methods to install KMM-Hub:

  • With the Operator Lifecycle Manager (OLM)

  • Creating KMM resources

Additional resources

Installing KMM-Hub using the Operator Lifecycle Manager

Use the Operators section of the OpenShift console to install KMM-Hub.

Installing KMM-Hub by creating KMM resources

Procedure

  • If you want to install KMM-Hub programmatically, you can use the following resources to create the Namespace, OperatorGroup and Subscription resources:

---
apiVersion: v1
kind: Namespace
metadata:
  name: openshift-kmm-hub
---
apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: kernel-module-management-hub
  namespace: openshift-kmm-hub
---
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: kernel-module-management-hub
  namespace: openshift-kmm-hub
spec:
  channel: stable
  installPlanApproval: Automatic
  name: kernel-module-management-hub
  source: redhat-operators
  sourceNamespace: openshift-marketplace

Using the ManagedClusterModule CRD

Use the ManagedClusterModule Custom Resource Definition (CRD) to configure the deployment of kernel modules on spoke clusters. This CRD is cluster-scoped, wraps a Module spec and adds the following additional fields:

apiVersion: hub.kmm.sigs.x-k8s.io/v1beta1
kind: ManagedClusterModule
metadata:
  name: <my-mcm>
  # No namespace, because this resource is cluster-scoped.
spec:
  moduleSpec: (1)
    selector: (2)
      node-wants-my-mcm: 'true'

  spokeNamespace: <some-namespace> (3)

  selector: (4)
    wants-my-mcm: 'true'
1 moduleSpec: Contains moduleLoader and devicePlugin sections, similar to a Module resource.
2 Selects nodes within the ManagedCluster.
3 Specifies in which namespace the Module should be created.
4 Selects ManagedCluster objects.

If build or signing instructions are present in .spec.moduleSpec, those pods are run on the hub cluster in the operator’s namespace.

When the .spec.selector matches one or more ManagedCluster resources, then KMM-Hub creates a ManifestWork resource in the corresponding namespace(s). ManifestWork contains a trimmed-down Module resource, with kernel mappings preserved but all build and sign subsections are removed. containerImage fields that contain image names ending with a tag are replaced with their digest equivalent.

Running KMM on the spoke

After installing Kernel Module Management (KMM) on the spoke, no further action is required. Create a ManagedClusterModule object from the hub to deploy kernel modules on spoke clusters.

Procedure

You can install KMM on the spokes cluster through a RHACM Policy object. In addition to installing KMM from the OperatorHub and running it in a lightweight spoke mode, the Policy configures additional RBAC required for the RHACM agent to be able to manage Module resources.

  • Use the following RHACM policy to install KMM on spoke clusters:

    ---
apiVersion: policy.open-cluster-management.io/v1
kind: Policy
metadata:
  name: install-kmm
spec:
  remediationAction: enforce
  disabled: false
  policy-templates:
    - objectDefinition:
        apiVersion: policy.open-cluster-management.io/v1
        kind: ConfigurationPolicy
        metadata:
          name: install-kmm
        spec:
          severity: high
          object-templates:
          - complianceType: mustonlyhave
            objectDefinition:
              apiVersion: v1
              kind: Namespace
              metadata:
                name: openshift-kmm
          - complianceType: mustonlyhave
            objectDefinition:
              apiVersion: operators.coreos.com/v1
              kind: OperatorGroup
              metadata:
                name: kmm
                namespace: openshift-kmm
              spec:
                upgradeStrategy: Default
          - complianceType: mustonlyhave
            objectDefinition:
              apiVersion: operators.coreos.com/v1alpha1
              kind: Subscription
              metadata:
                name: kernel-module-management
                namespace: openshift-kmm
              spec:
                channel: stable
                config:
                  env:
                    - name: KMM_MANAGED (1)
                      value: "1"
                installPlanApproval: Automatic
                name: kernel-module-management
                source: redhat-operators
                sourceNamespace: openshift-marketplace
          - complianceType: mustonlyhave
            objectDefinition:
              apiVersion: rbac.authorization.k8s.io/v1
              kind: ClusterRole
              metadata:
                name: kmm-module-manager
              rules:
                - apiGroups: [kmm.sigs.x-k8s.io]
                  resources: [modules]
                  verbs: [create, delete, get, list, patch, update, watch]
          - complianceType: mustonlyhave
            objectDefinition:
              apiVersion: rbac.authorization.k8s.io/v1
              kind: ClusterRoleBinding
              metadata:
                name: klusterlet-kmm
              subjects:
              - kind: ServiceAccount
                name: klusterlet-work-sa
                namespace: open-cluster-management-agent
              roleRef:
                kind: ClusterRole
                name: kmm-module-manager
                apiGroup: rbac.authorization.k8s.io
---
apiVersion: apps.open-cluster-management.io/v1
kind: PlacementRule
metadata:
  name: all-managed-clusters
spec:
  clusterSelector: (2)
    matchExpressions: []
---
apiVersion: policy.open-cluster-management.io/v1
kind: PlacementBinding
metadata:
  name: install-kmm
placementRef:
  apiGroup: apps.open-cluster-management.io
  kind: PlacementRule
  name: all-managed-clusters
subjects:
  - apiGroup: policy.open-cluster-management.io
    kind: Policy
    name: install-kmm
    1 This environment variable is required when running KMM on a spoke cluster.
    2 The spec.clusterSelector field can be customized to target select clusters only.

Customizing upgrades for kernel modules

Use this procedure to upgrade the kernel module while running maintenance operations on the node, including rebooting the node, if needed. To minimize the impact on the workloads running in the cluster, run the kernel upgrade process sequentially, one node at a time.

This procedure requires knowledge of the workload utilizing the kernel module and must be managed by the cluster administrator.
Prerequisites

  • Before upgrading, set the kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name>=$moduleVersion label on all the nodes that are used by the kernel module.

  • Terminate all user application workloads on the node or move them to another node.

  • Unload the currently loaded kernel module.

  • Ensure that the user workload (the application running in the cluster that is accessing kernel module) is not running on the node prior to kernel module unloading and that the workload is back running on the node after the new kernel module version has been loaded.

Procedure

  1. Ensure that the device plugin managed by KMM on the node is unloaded.

  2. Update the following fields in the Module custom resource (CR):

    • containerImage (to the appropriate kernel version)

    • version

      The update should be atomic; that is, both the containerImage and version fields must be updated simultaneously.

  3. Terminate any workload using the kernel module on the node being upgraded.

  4. Remove the kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name> label on the node. Run the following command to unload the kernel module from the node:

    $ oc label node/<node_name> kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name>-

  5. If required, as the cluster administrator, perform any additional maintenance required on the node for the kernel module upgrade.

    If no additional upgrading is needed, you can skip Steps 3 through 6 by updating the kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name> label value to the new $moduleVersion as set in the Module.

  6. Run the following command to add the kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name>=$moduleVersion label to the node. The $moduleVersion must be equal to the new value of the version field in the Module CR.

    $ oc label node/<node_name> kmm.node.kubernetes.io/version-module.<module_namespace>.<module_name>=<desired_version>

    Because of Kubernetes limitations in label names, the combined length of Module name and namespace must not exceed 39 characters.

  7. Restore any workload that leverages the kernel module on the node.

  8. Reload the device plugin managed by KMM on the node.

Day 1 kernel module loading

Kernel Module Management (KMM) is typically a Day 2 Operator. Kernel modules are loaded only after the complete initialization of a Linux (RHCOS) server. However, in some scenarios the kernel module must be loaded at an earlier stage. Day 1 functionality allows you to use the Machine Config Operator (MCO) to load kernel modules during the Linux systemd initialization stage.

Additional resources

Day 1 supported use cases

The Day 1 functionality supports a limited number of use cases. The main use case is to allow loading out-of-tree (OOT) kernel modules prior to NetworkManager service initialization. It does not support loading kernel module at the initramfs stage.

The following are the conditions needed for Day 1 functionality:

  • The kernel module is not loaded in the kernel.

  • The in-tree kernel module is loaded into the kernel, but can be unloaded and replaced by the OOT kernel module. This means that the in-tree module is not referenced by any other kernel modules.

  • In order for Day 1 functionlity to work, the node must have a functional network interface, that is, an in-tree kernel driver for that interface. The OOT kernel module can be a network driver that will replace the functional network driver.

OOT kernel module loading flow

The loading of the out-of-tree (OOT) kernel module leverages the Machine Config Operator (MCO). The flow sequence is as follows:

Procedure

  1. Apply a MachineConfig resource to the existing running cluster. In order to identify the necessary nodes that need to be updated, you must create an appropriate MachineConfigPool resource.

  2. MCO applies the reboots node by node. On any rebooted node, two new systemd services are deployed: pull service and load service.

  3. The load service is configured to run prior to the NetworkConfiguration service. The service tries to pull a predefined kernel module image and then, using that image, to unload an in-tree module and load an OOT kernel module.

  4. The pull service is configured to run after NetworkManager service. The service checks if the preconfigured kernel module image is located on the node’s filesystem. If it is, the service exists normally, and the server continues with the boot process. If not, it pulls the image onto the node and reboots the node afterwards.

The kernel module image

The Day 1 functionality uses the same DTK based image leveraged by Day 2 KMM builds. The out-of-tree kernel module should be located under /opt/lib/modules/${kernelVersion}.

Additional resources

In-tree module replacement

The Day 1 functionality always tries to replace the in-tree kernel module with the OOT version. If the in-tree kernel module is not loaded, the flow is not affected; the service proceeds and loads the OOT kernel module.

MCO yaml creation

KMM provides an API to create an MCO YAML manifest for the Day 1 functionality:

ProduceMachineConfig(machineConfigName, machineConfigPoolRef, kernelModuleImage, kernelModuleName string) (string, error)

The returned output is a string representation of the MCO YAML manifest to be applied. It is up to the customer to apply this YAML.

The parameters are:

machineConfigName

The name of the MCO YAML manifest. This parameter is set as the name parameter of the metadata of the MCO YAML manifest.

machineConfigPoolRef

The MachineConfigPool name used to identify the targeted nodes.

kernelModuleImage

The name of the container image that includes the OOT kernel module.

kernelModuleName

The name of the OOT kernel module. This parameter is used both to unload the in-tree kernel module (if loaded into the kernel) and to load the OOT kernel module.

The API is located under pkg/mcproducer package of the KMM source code. The KMM operator does not need to be running to use the Day 1 functionality. You only need to import the pkg/mcproducer package into their operator/utility code, call the API, and apply the produced MCO YAML to the cluster.

The MachineConfigPool

The MachineConfigPool identifies a collection of nodes that are affected by the applied MCO.

kind: MachineConfigPool
metadata:
  name: sfc
spec:
  machineConfigSelector: (1)
    matchExpressions:
      - {key: machineconfiguration.openshift.io/role, operator: In, values: [worker, sfc]}
  nodeSelector: (2)
    matchLabels:
      node-role.kubernetes.io/sfc: ""
  paused: false
  maxUnavailable: 1
1 Matches the labels in the MachineConfig.
2 Matches the labels on the node.

There are predefined MachineConfigPools in the OCP cluster:

  • worker: Targets all worker nodes in the cluster

  • master: Targets all master nodes in the cluster

Define the following MachineConfig to target the master MachineConfigPool:

metadata:
  labels:
    machineconfiguration.opensfhit.io/role: master

Define the following MachineConfig to target the worker MachineConfigPool:

metadata:
  labels:
    machineconfiguration.opensfhit.io/role: worker
Additional resources

Debugging and troubleshooting

If the kmods in your driver container are not signed or are signed with the wrong key, then the container can enter a PostStartHookError or CrashLoopBackOff status. You can verify by running the oc describe command on your container, which displays the following message in this scenario:

modprobe: ERROR: could not insert '<your_kmod_name>': Required key not available

KMM firmware support

Kernel modules sometimes need to load firmware files from the file system. KMM supports copying firmware files from the kmod image to the node’s file system.

The contents of .spec.moduleLoader.container.modprobe.firmwarePath are copied into the /var/lib/firmware path on the node before running the modprobe command to insert the kernel module.

All files and empty directories are removed from that location before running the modprobe -r command to unload the kernel module, when the pod is terminated.

Configuring the lookup path on nodes

On OKD nodes, the set of default lookup paths for firmwares does not include the /var/lib/firmware path.

Procedure

  1. Use the Machine Config Operator to create a MachineConfig custom resource (CR) that contains the /var/lib/firmware path:

    apiVersion: machineconfiguration.openshift.io/v1
kind: MachineConfig
metadata:
  labels:
    machineconfiguration.openshift.io/role: worker (1)
  name: 99-worker-kernel-args-firmware-path
spec:
  kernelArguments:
    - 'firmware_class.path=/var/lib/firmware'
    1 You can configure the label based on your needs. In the case of single-node OpenShift, use either control-pane or master objects.

  2. By applying the MachineConfig CR, the nodes are automatically rebooted.

Additional resources

Building a kmod image

Procedure

  • In addition to building the kernel module itself, include the binary firmware in the builder image:

    FROM registry.redhat.io/ubi9/ubi-minimal as builder

# Build the kmod

RUN ["mkdir", "/firmware"]
RUN ["curl", "-o", "/firmware/firmware.bin", "https://artifacts.example.com/firmware.bin"]

FROM registry.redhat.io/ubi9/ubi-minimal

# Copy the kmod, install modprobe, run depmod

COPY --from=builder /firmware /firmware

Tuning the Module resource

Procedure

  • Set .spec.moduleLoader.container.modprobe.firmwarePath in the Module custom resource (CR):

    apiVersion: kmm.sigs.x-k8s.io/v1beta1
kind: Module
metadata:
  name: my-kmod
spec:
  moduleLoader:
    container:
      modprobe:
        moduleName: my-kmod  # Required

        firmwarePath: /firmware (1)
    1 Optional: Copies /firmware/* into /var/lib/firmware/ on the node.

Day 0 through Day 2 kmod installation

You can install some kernel modules (kmods) during Day 0 through Day 2 operations without Kernel Module Management (KMM). This could assist in the transition of the kmods to KMM.

Use the following criteria to determine suitable kmod installations.

Day 0

The most basic kmods that are required for a node to become Ready in the cluster. Examples of these types of kmods include:

  • A storage driver that is required to mount the rootFS as part of the boot process

  • A network driver that is required for the machine to access machine-config-server on the bootstrap node to pull the ignition and join the cluster

Day 1

Kmods that are not required for a node to become Ready in the cluster but cannot be unloaded when the node is Ready.

An example of this type of kmod is an out-of-tree (OOT) network driver that replaces an outdated in-tree driver to exploit the full potential of the NIC while NetworkManager depends on it. When the node is Ready, you cannot unload the driver because of the NetworkManager dependency.

Day 2

Kmods that can be dynamically loaded to the kernel or removed from it without interfering with the cluster infrastructure, for example, connectivity.

Examples of these types of kmods include:

  • GPU operators

  • Secondary network adapters

  • field-programmable gate arrays (FPGAs)

Layering background

When a Day 0 kmod is installed in the cluster, layering is applied through the Machine Config Operator (MCO) and OKD upgrades do not trigger node upgrades.

You only need to recompile the driver if you add new features to it, because the node’s operating system will remain the same.

Lifecycle management

You can leverage KMM to manage the Day 0 through Day 2 lifecycle of kmods without a reboot when the driver allows it.

This will not work if the upgrade requires a node reboot, for example, when rebuilding initramfs files is needed.

Use one of the following options for lifecycle management.

Treat the kmod as an in-tree driver

Use this method when you want to upgrade the kmods. In this case, treat the kmod as an in-tree driver and create a Module in the cluster with the inTreeRemoval field to unload the old version of the driver.

Note the following characteristics of treating the kmod as an in-tree driver:

  • Downtime might occur as KMM tries to unload and load the kmod on all the selected nodes simultaneously.

  • This works if removing the driver makes the node lose connectivity because KMM uses a single pod to unload and load the driver.

Use ordered upgrade

You can use ordered upgrade (ordered_upgrade.md) to create a versioned Module in the cluster representing the kmods with no effect, because the kmods are already loaded.

Note the following characteristics of using ordered upgrade:

  • There is no cluster downtime because you control the pace of the upgrade and how many nodes are upgraded at the same time; therefore, an upgrade with no downtime is possible.

  • This method will not work if unloading the driver results in losing connection to the node, because KMM creates two different worker pods for unloading and another for loading. These pods will not be scheduled.

Troubleshooting KMM

When troubleshooting KMM installation issues, you can monitor logs to determine at which stage issues occur. Then, retrieve diagnostic data relevant to that stage.

Reading Operator logs

You can use the oc logs command to read Operator logs, as in the following examples.

Example command for KMM controller
$ oc logs -fn openshift-kmm deployments/kmm-operator-controller
Example command for KMM webhook server
$ oc logs -fn openshift-kmm deployments/kmm-operator-webhook-server
Example command for KMM-Hub controller
$ oc logs -fn openshift-kmm-hub deployments/kmm-operator-hub-controller
Example command for KMM-Hub webhook server
$ oc logs -fn openshift-kmm deployments/kmm-operator-hub-webhook-server

Observing events

Use the following methods to view KMM events.

Build & sign

KMM publishes events whenever it starts a kmod image build or observes its outcome. These events are attached to Module objects and are available at the end of the output of oc describe module command, as in the following example:

$ oc describe modules.kmm.sigs.x-k8s.io kmm-ci-a
[...]
Events:
  Type    Reason          Age                From  Message
  ----    ------          ----               ----  -------
  Normal  BuildCreated    2m29s              kmm   Build created for kernel 6.6.2-201.fc39.x86_64
  Normal  BuildSucceeded  63s                kmm   Build job succeeded for kernel 6.6.2-201.fc39.x86_64
  Normal  SignCreated     64s (x2 over 64s)  kmm   Sign created for kernel 6.6.2-201.fc39.x86_64
  Normal  SignSucceeded   57s                kmm   Sign job succeeded for kernel 6.6.2-201.fc39.x86_64

Module load or unload

KMM publishes events whenever it successfully loads or unloads a kernel module on a node. These events are attached to Node objects and are available at the end of the output of oc describe node command, as in the following example:

$ oc describe node my-node
[...]
Events:
  Type    Reason          Age    From  Message
  ----    ------          ----   ----  -------
[...]
  Normal  ModuleLoaded    4m17s  kmm   Module default/kmm-ci-a loaded into the kernel
  Normal  ModuleUnloaded  2s     kmm   Module default/kmm-ci-a unloaded from the kernel

Using the must-gather tool

The oc adm must-gather command is the preferred way to collect a support bundle and provide debugging information to Red Hat Support. Collect specific information by running the command with the appropriate arguments as described in the following sections.

Additional resources

Gathering data for KMM

Procedure

  1. Gather the data for the KMM Operator controller manager:

    1. Set the MUST_GATHER_IMAGE variable:

      $ export MUST_GATHER_IMAGE=$(oc get deployment -n openshift-kmm kmm-operator-controller -ojsonpath='{.spec.template.spec.containers[?(@.name=="manager")].env[?(@.name=="RELATED_IMAGE_MUST_GATHER")].value}')
$ oc adm must-gather --image="${MUST_GATHER_IMAGE}" -- /usr/bin/gather

      Use the -n <namespace> switch to specify a namespace if you installed KMM in a custom namespace.

    2. Run the must-gather tool:

      $ oc adm must-gather --image="${MUST_GATHER_IMAGE}" -- /usr/bin/gather

  2. View the Operator logs:

    $ oc logs -fn openshift-kmm deployments/kmm-operator-controller
    Example output 
    I0228 09:36:37.352405       1 request.go:682] Waited for 1.001998746s due to client-side throttling, not priority and fairness, request: GET:https://172.30.0.1:443/apis/machine.openshift.io/v1beta1?timeout=32s
I0228 09:36:40.767060       1 listener.go:44] kmm/controller-runtime/metrics "msg"="Metrics server is starting to listen" "addr"="127.0.0.1:8080"
I0228 09:36:40.769483       1 main.go:234] kmm/setup "msg"="starting manager"
I0228 09:36:40.769907       1 internal.go:366] kmm "msg"="Starting server" "addr"={"IP":"127.0.0.1","Port":8080,"Zone":""} "kind"="metrics" "path"="/metrics"
I0228 09:36:40.770025       1 internal.go:366] kmm "msg"="Starting server" "addr"={"IP":"::","Port":8081,"Zone":""} "kind"="health probe"
I0228 09:36:40.770128       1 leaderelection.go:248] attempting to acquire leader lease openshift-kmm/kmm.sigs.x-k8s.io...
I0228 09:36:40.784396       1 leaderelection.go:258] successfully acquired lease openshift-kmm/kmm.sigs.x-k8s.io
I0228 09:36:40.784876       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module" "source"="kind source: *v1beta1.Module"
I0228 09:36:40.784925       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module" "source"="kind source: *v1.DaemonSet"
I0228 09:36:40.784968       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module" "source"="kind source: *v1.Build"
I0228 09:36:40.785001       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module" "source"="kind source: *v1.Job"
I0228 09:36:40.785025       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module" "source"="kind source: *v1.Node"
I0228 09:36:40.785039       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="Module" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="Module"
I0228 09:36:40.785458       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="PodNodeModule" "controllerGroup"="" "controllerKind"="Pod" "source"="kind source: *v1.Pod"
I0228 09:36:40.786947       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="PreflightValidation" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidation" "source"="kind source: *v1beta1.PreflightValidation"
I0228 09:36:40.787406       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="PreflightValidation" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidation" "source"="kind source: *v1.Build"
I0228 09:36:40.787474       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="PreflightValidation" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidation" "source"="kind source: *v1.Job"
I0228 09:36:40.787488       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="PreflightValidation" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidation" "source"="kind source: *v1beta1.Module"
I0228 09:36:40.787603       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="NodeKernel" "controllerGroup"="" "controllerKind"="Node" "source"="kind source: *v1.Node"
I0228 09:36:40.787634       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="NodeKernel" "controllerGroup"="" "controllerKind"="Node"
I0228 09:36:40.787680       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="PreflightValidation" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidation"
I0228 09:36:40.785607       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream" "source"="kind source: *v1.ImageStream"
I0228 09:36:40.787822       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="preflightvalidationocp" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidationOCP" "source"="kind source: *v1beta1.PreflightValidationOCP"
I0228 09:36:40.787853       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream"
I0228 09:36:40.787879       1 controller.go:185] kmm "msg"="Starting EventSource" "controller"="preflightvalidationocp" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidationOCP" "source"="kind source: *v1beta1.PreflightValidation"
I0228 09:36:40.787905       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="preflightvalidationocp" "controllerGroup"="kmm.sigs.x-k8s.io" "controllerKind"="PreflightValidationOCP"
I0228 09:36:40.786489       1 controller.go:193] kmm "msg"="Starting Controller" "controller"="PodNodeModule" "controllerGroup"="" "controllerKind"="Pod"

Gathering data for KMM-Hub

Procedure

  1. Gather the data for the KMM Operator hub controller manager:

    1. Set the MUST_GATHER_IMAGE variable:

      $ export MUST_GATHER_IMAGE=$(oc get deployment -n openshift-kmm-hub kmm-operator-hub-controller -ojsonpath='{.spec.template.spec.containers[?(@.name=="manager")].env[?(@.name=="RELATED_IMAGE_MUST_GATHER")].value}')
$ oc adm must-gather --image="${MUST_GATHER_IMAGE}" -- /usr/bin/gather -u

      Use the -n <namespace> switch to specify a namespace if you installed KMM in a custom namespace.

    2. Run the must-gather tool:

      $ oc adm must-gather --image="${MUST_GATHER_IMAGE}" -- /usr/bin/gather -u

  2. View the Operator logs:

    $ oc logs -fn openshift-kmm-hub deployments/kmm-operator-hub-controller
    Example output 
    I0417 11:34:08.807472       1 request.go:682] Waited for 1.023403273s due to client-side throttling, not priority and fairness, request: GET:https://172.30.0.1:443/apis/tuned.openshift.io/v1?timeout=32s
I0417 11:34:12.373413       1 listener.go:44] kmm-hub/controller-runtime/metrics "msg"="Metrics server is starting to listen" "addr"="127.0.0.1:8080"
I0417 11:34:12.376253       1 main.go:150] kmm-hub/setup "msg"="Adding controller" "name"="ManagedClusterModule"
I0417 11:34:12.376621       1 main.go:186] kmm-hub/setup "msg"="starting manager"
I0417 11:34:12.377690       1 leaderelection.go:248] attempting to acquire leader lease openshift-kmm-hub/kmm-hub.sigs.x-k8s.io...
I0417 11:34:12.378078       1 internal.go:366] kmm-hub "msg"="Starting server" "addr"={"IP":"127.0.0.1","Port":8080,"Zone":""} "kind"="metrics" "path"="/metrics"
I0417 11:34:12.378222       1 internal.go:366] kmm-hub "msg"="Starting server" "addr"={"IP":"::","Port":8081,"Zone":""} "kind"="health probe"
I0417 11:34:12.395703       1 leaderelection.go:258] successfully acquired lease openshift-kmm-hub/kmm-hub.sigs.x-k8s.io
I0417 11:34:12.396334       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "source"="kind source: *v1beta1.ManagedClusterModule"
I0417 11:34:12.396403       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "source"="kind source: *v1.ManifestWork"
I0417 11:34:12.396430       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "source"="kind source: *v1.Build"
I0417 11:34:12.396469       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "source"="kind source: *v1.Job"
I0417 11:34:12.396522       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "source"="kind source: *v1.ManagedCluster"
I0417 11:34:12.396543       1 controller.go:193] kmm-hub "msg"="Starting Controller" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule"
I0417 11:34:12.397175       1 controller.go:185] kmm-hub "msg"="Starting EventSource" "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream" "source"="kind source: *v1.ImageStream"
I0417 11:34:12.397221       1 controller.go:193] kmm-hub "msg"="Starting Controller" "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream"
I0417 11:34:12.498335       1 filter.go:196] kmm-hub "msg"="Listing all ManagedClusterModules" "managedcluster"="local-cluster"
I0417 11:34:12.498570       1 filter.go:205] kmm-hub "msg"="Listed ManagedClusterModules" "count"=0 "managedcluster"="local-cluster"
I0417 11:34:12.498629       1 filter.go:238] kmm-hub "msg"="Adding reconciliation requests" "count"=0 "managedcluster"="local-cluster"
I0417 11:34:12.498687       1 filter.go:196] kmm-hub "msg"="Listing all ManagedClusterModules" "managedcluster"="sno1-0"
I0417 11:34:12.498750       1 filter.go:205] kmm-hub "msg"="Listed ManagedClusterModules" "count"=0 "managedcluster"="sno1-0"
I0417 11:34:12.498801       1 filter.go:238] kmm-hub "msg"="Adding reconciliation requests" "count"=0 "managedcluster"="sno1-0"
I0417 11:34:12.501947       1 controller.go:227] kmm-hub "msg"="Starting workers" "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream" "worker count"=1
I0417 11:34:12.501948       1 controller.go:227] kmm-hub "msg"="Starting workers" "controller"="ManagedClusterModule" "controllerGroup"="hub.kmm.sigs.x-k8s.io" "controllerKind"="ManagedClusterModule" "worker count"=1
I0417 11:34:12.502285       1 imagestream_reconciler.go:50] kmm-hub "msg"="registered imagestream info mapping" "ImageStream"={"name":"driver-toolkit","namespace":"openshift"} "controller"="imagestream" "controllerGroup"="image.openshift.io" "controllerKind"="ImageStream" "dtkImage"="quay.io/openshift-release-dev/ocp-v4.0-art-dev@sha256:df42b4785a7a662b30da53bdb0d206120cf4d24b45674227b16051ba4b7c3934" "name"="driver-toolkit" "namespace"="openshift" "osImageVersion"="412.86.202302211547-0" "reconcileID"="e709ff0a-5664-4007-8270-49b5dff8bae9"