Using Pod Security Policies with Litmus

While working in environments (clusters) that have restrictive security policies, the default litmuschaos experiment execution procedure may be inhibited. This is mainly due to the fact that the experiment pods running the chaos injection tasks in privileged mode. This, in turn, is necessitated due to the mounting of container runtime-specific socket files from the Kubernetes nodes in order to invoke runtime APIs. While this is not needed for all experiments (a considerable number of them use purely the K8s API), those involving injection of chaos processes into the network/process namespaces of other containers have this requirement (ex: netem, stress).

The restrictive policies are often enforced via pod security policies (PSP) today, with organizations opting for the default "restricted" policy.

Applying Pod Security Policies to Litmus Chaos Pods

• To run the litmus pods with operating characteristics described above, first create a custom PodSecurityPolicy that allows the same:

  apiVersion: policy/v1beta1
  kind: PodSecurityPolicy
  metadata:
  name: litmus
  annotations:
      seccomp.security.alpha.kubernetes.io/allowedProfileNames: '*'
  spec:
  privileged: true
  # Required to prevent escalations to root.
  allowPrivilegeEscalation: true
  # Allow core volume types.
  volumes:
      # To mount script files/templates like ssm-docs in experiment
      - 'configMap'
      # Used for chaos injection like io chaos
      - 'emptyDir'
      - 'projected'
      # To authenticate with different cloud providers
      - 'secret'
      # To derive the experiment pod name in the experimemnt
      - 'downwardAPI'
      # Assume that persistentVolumes set up by the cluster admin are safe to use.
      - 'persistentVolumeClaim'
      # To mount the socket path directory used to perform container runtime operations
      - 'hostPath'
  
  allowedHostPaths:
      # substitutes this path with an appropriate socket path
      # ex: '/run/containerd/containerd.sock', '/run/containerd/containerd.sock', '/run/crio/crio.sock'
      - pathPrefix: "/run/containerd/containerd.sock"
      # substitutes this path with an appropriate container path
      # ex: '/var/lib/docker/containers', '/var/lib/containerd/io.containerd.runtime.v1.linux/k8s.io', '/var/lib/containers/storage/overlay/'
      - pathPrefix: "/var/lib/containerd/io.containerd.runtime.v1.linux/k8s.io"
  
  allowedCapabilities:
      # NET_ADMIN & SYS_ADMIN: used in network chaos experiments to perform
      # network operations (running tc command in network ns of target container). 
      - "NET_ADMIN"
      # SYS_ADMIN: used in stress chaos experiment to perform cgroup operations.
      - "SYS_ADMIN"
  hostNetwork: false
  hostIPC: false
      # To run fault injection on a target container using pid namespace.
      # It is used in stress, network, dns and http experiments. 
  hostPID: true
  seLinux:
      # This policy assumes the nodes are using AppArmor rather than SELinux.
      rule: 'RunAsAny'
  supplementalGroups:
      rule: 'MustRunAs'
      ranges:
      # Forbid adding the root group.
      - min: 1
        max: 65535
  fsGroup:
      rule: 'MustRunAs'
      ranges:
      # Forbid adding the root group.
      - min: 1
        max: 65535
  readOnlyRootFilesystem: false
  

  Note: This PodSecurityPolicy is a sample configuration which works for a majority of the usecases. It is left to the user's discretion to modify it based on the environment. For example, if the experiment doesn't need the socket file to be mounted, allowedHostPaths can be excluded from the psp spec. On the other hand, in case of CRI-O runtime, network-chaos tests need the chaos pods executed in privileged mode. It is also possible that different PSP configs are used in different namespaces based on ChaosExperiments installed/executed in them.

• Subscribe to the created PSP in the experiment RBAC (or in the admin-mode rbac, as applicable). For example, the pod-delete experiment rbac instrumented with the PSP is shown below:

  ---
  apiVersion: v1
  kind: ServiceAccount
  metadata:
  name: pod-delete-sa
  namespace: default
  labels:
      name: pod-delete-sa
      app.kubernetes.io/part-of: litmus
  ---
  apiVersion: rbac.authorization.k8s.io/v1
  kind: Role
  metadata:
  name: pod-delete-sa
  namespace: default
  labels:
      name: pod-delete-sa
      app.kubernetes.io/part-of: litmus
  rules:
  - apiGroups: [""]
  resources: ["pods","events"]
  verbs: ["create","list","get","patch","update","delete","deletecollection"]
  - apiGroups: [""]
  resources: ["pods/exec","pods/log","replicationcontrollers"]
  verbs: ["create","list","get"]
  - apiGroups: ["batch"]
  resources: ["jobs"]
  verbs: ["create","list","get","delete","deletecollection"]
  - apiGroups: ["apps"]
  resources: ["deployments","statefulsets","daemonsets","replicasets"]
  verbs: ["list","get"]
  - apiGroups: ["apps.openshift.io"]
  resources: ["deploymentconfigs"]
  verbs: ["list","get"]
  - apiGroups: ["argoproj.io"]
  resources: ["rollouts"]
  verbs: ["list","get"]
  - apiGroups: ["litmuschaos.io"]
  resources: ["chaosengines","chaosexperiments","chaosresults"]
  verbs: ["create","list","get","patch","update"]
  - apiGroups: ["policy"]
  resources: ["podsecuritypolicies"]
  verbs: ["use"]
  resourceNames: ["litmus"] 
  ---
  apiVersion: rbac.authorization.k8s.io/v1
  kind: RoleBinding
  metadata:
  name: pod-delete-sa
  namespace: default
  labels:
      name: pod-delete-sa
      app.kubernetes.io/part-of: litmus
  roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: pod-delete-sa
  subjects:
  - kind: ServiceAccount
  name: pod-delete-sa
  namespace: default
  

• Execute the ChaosEngine and verify that the litmus experiment pods are created successfully.