Configuration

Helm Values

kubefence is configured via Helm values. The chart renders the TOML plugin configuration automatically into a ConfigMap — you do not edit the TOML file directly when using Helm.

Core settings

Value	Default	Description
image.repository	ghcr.io/kubefence/nono-nri-plugin	Plugin container image repository
image.tag	""	Plugin container image tag. Empty means the chart's appVersion, i.e. the release the chart came from — so --version 0.8.0 pins the image to 0.8.0. Set explicitly for local builds or to pin a digest
image.pullPolicy	IfNotPresent	Kubernetes image pull policy
namespace	kube-system	Namespace for all kubefence resources

NRI plugin configuration

These values are rendered into the TOML config file loaded by the plugin.

Value	Default	Description
config.runtimeClasses	[nono-runc]	List of RuntimeClass handler names to intercept. Pods whose RuntimeClass handler matches are sandboxed; all others are skipped
config.defaultProfile	"default"	nono profile used when a pod has no nono.sh/profile annotation
config.nonoBinPath	"/opt/nono-nri/nono"	Absolute host path to the nono binary. Copied there by the node-setup DaemonSet init container
config.socketPath	""	NRI socket path. Empty string uses the runtime default (/var/run/nri/nri.sock)
config.seccompProfile	"restricted"	Seccomp policy injected into every sandboxed container. "restricted" blocks io_uring, ptrace, the seccomp syscall, and pidfd_getfd on top of RuntimeDefault. "runtime-default" applies the Docker RuntimeDefault allowlist. "" disables injection

RuntimeClass creation

Value	Default	Description
runtimeClasses.nonoRunc.enabled	true	Create the nono-runc RuntimeClass (handler: nono-runc)
runtimeClasses.kataNono.enabled	false	Create the kata-nono-sandbox RuntimeClass
runtimeClasses.kataNono.handler	"kata-qemu-runtime-rs"	Handler name for the Kata RuntimeClass. Must match a handler registered by kata-deploy

Kata Containers provisioning

Value	Default	Description
kata.enabled	false	Enable the kata-setup DaemonSet. Requires kata-deploy to be installed first
kata.extensionImage	""	OCI image carrying the nono guest extension (hardened kata-agent policy). Empty means ghcr.io/kubefence/kata-nono-extension:<appVersion>, so a pinned chart pins the extension too. Set explicitly for another registry or to pin a digest
kata.shareDir	/opt/kata/share/kata-containers	Directory where kata-deploy installs kata share files on each node
kata.qemuConfigPath	/opt/kata/share/defaults/kata-containers/runtime-rs/runtimes/qemu-runtime-rs/configuration-qemu-runtime-rs.toml	Path to the kata QEMU configuration file written by kata-deploy. runtime-rs configs sit under a runtime-rs/ prefix
kata.qemu.machineAccelerators	""	Additional QEMU machine accelerators. Set to "kernel_irqchip=split" for nested-KVM environments (e.g. Kind clusters)
kata.qemu.seccompSandbox	"on,obsolete=deny,spawn=deny,resourcecontrol=deny"	QEMU process-level seccomp sandbox (host kernel). Restricts syscalls available to the QEMU hypervisor process. spawn=deny prevents QEMU from exec'ing host binaries after a VM escape. Written to seccomp_sandbox — runtime-rs's spelling, not the Go runtime's seccompsandbox. Set to "" to disable
kata.qemu.disableGuestSeccomp	false	Maps to disable_guest_seccomp in the kata QEMU config. false enables the kata-agent to apply the container's OCI seccomp profile (written by the NRI plugin) inside the guest VM

Node setup

Value	Default	Description
nodeSetup.enabled	true	Enable the node-setup DaemonSet that patches containerd to enable NRI and register runtime handlers
nodeSetup.nri.socketPath	/var/run/nri/nri.sock	NRI socket path to configure in containerd
nodeSetup.nri.pluginPath	/opt/nri/plugins	NRI plugin path to configure in containerd
nodeSetup.nri.configPath	/etc/nri/conf.d	NRI config directory to configure in containerd
containerdConfigPath	/etc/containerd/config.toml	Host path to containerd's config file. Shared by the node-setup and kata-setup DaemonSets. They write drop-ins into its conf.d sibling and only edit this file if its imports array does not already cover them

Host paths

Value	Default	Description
hostPaths.nriSocket	/var/run/nri	Directory holding the NRI socket, mounted into the plugin
hostPaths.nonoHostDir	/opt/nono-nri	Host directory the nono binary and interpreter wrappers are installed into, and the source of the /nono bind-mount
hostPaths.nriPlugins	/opt/nri/plugins	NRI plugin directory on the host
hostPaths.readyDir	/run/kubefence	Where the setup DaemonSets record that they have finished with the node. The plugin waits for those markers before connecting — see Startup ordering. On tmpfs by design, so a reboot forces the setup DaemonSets to re-assert them

Resources

Value	Default	Description
resources.requests.cpu	50m	CPU request for the plugin container
resources.requests.memory	32Mi	Memory request for the plugin container
resources.limits.cpu	200m	CPU limit for the plugin container. Prevents burst container-creation events from exhausting node memory
resources.limits.memory	128Mi	Memory limit for the plugin container

Helper images

Value	Default	Description
helperImages.alpine	alpine:3.20	Alpine image used in privileged init containers. Pin to an immutable digest in production
helperImages.busybox	busybox:1.37.0-uclibc	Busybox image used in privileged init containers. Pin to an immutable digest in production

---

TOML Configuration

When using Helm, the chart renders the TOML configuration automatically into a ConfigMap and mounts it at /etc/nri/conf.d/10-nono-nri.toml inside the plugin container. You configure the plugin through Helm values — not by editing the TOML file directly.

If you deploy kubefence without Helm, create the TOML file manually:

# RuntimeClass handler names to intercept (matches pod.GetRuntimeHandler())
runtime_classes = ["nono-runc"]

# nono profile when pod has no nono.sh/profile annotation
default_profile = "default"

# Host path to the nono binary (copied there by the DaemonSet init container)
nono_bin_path = "/opt/nono-nri/nono"

# NRI socket (empty = use runtime default: /var/run/nri/nri.sock)
socket_path = ""

Field reference

Field	Required	Description
runtime_classes	Yes	List of RuntimeClass handler names. The plugin sandboxes pods whose handler matches this list; all others are skipped with zero overhead
default_profile	Yes	nono profile used when a pod has no nono.sh/profile annotation
nono_bin_path	Yes	Absolute path to the nono binary on the host. The plugin checks this path at startup and refuses to start if the file is absent
socket_path	No	NRI socket path. Defaults to /var/run/nri/nri.sock when empty
seccomp_profile	No	Seccomp policy injected into every sandboxed container. See below

The plugin validates runtime_classes and nono_bin_path at startup. Missing or empty values cause an immediate exit with an error message.

seccomp_profile values

Value	Description
"restricted"	Docker RuntimeDefault allowlist minus io_uring_setup, io_uring_enter, io_uring_register, ptrace, seccomp, and pidfd_getfd. Default for AI workloads. Blocks io_uring (CVE-2022-2639, CVE-2023-2598), cross-process inspection, and self-filter removal
"runtime-default"	Docker/moby RuntimeDefault allowlist verbatim. Equivalent to seccompProfile.type: RuntimeDefault in the pod security context
"" (empty)	Disabled. The plugin injects no seccomp policy. Pods may still set their own policy via securityContext.seccompProfile

Both profiles use SCMP_ACT_ERRNO as the default action (deny-by-default) and declare only SCMP_ARCH_X86_64. Syscalls already blocked by RuntimeDefault — including bpf, mount, kexec_load, init_module, setns, unshare, process_vm_readv, and userfaultfd — remain blocked in both profiles.

For Kata Containers, the NRI-injected policy takes effect inside the guest VM only when disable_guest_seccomp = false in the kata QEMU config. The kubefence kata-setup DaemonSet sets this automatically when kata.qemu.disableGuestSeccomp: false (the default).