Using audit seccomp

The audit seccomp gadget provides a stream of events with syscalls that had their seccomp filters generating an audit log. An audit log can be generated in one of these two conditions:

• The Seccomp profile has the flag SECCOMP_FILTER_FLAG_LOG (currently unsupported by runc) and returns any action other than SECCOMP_RET_ALLOW.
• The Seccomp profile does not have the flag SECCOMP_FILTER_FLAG_LOG but returns SCMP_ACT_LOG or SCMP_ACT_KILL*.

On Kubernetes

• Install the Seccomp Operator.

• Install a SeccompProfile that log the mkdir and unshare syscalls.

apiVersion: security-profiles-operator.x-k8s.io/v1beta1
kind: SeccompProfile
metadata:
  name: log
  annotations:
    description: "Log some syscalls"
spec:
  defaultAction: SCMP_ACT_ALLOW
  syscalls:
  - action: SCMP_ACT_KILL
    names:
    - unshare
  - action: SCMP_ACT_LOG
    names:
    - mkdir

• Start a pod with that SeccompProfile.

apiVersion: v1
kind: Pod
metadata:
  name: mypod
spec:
  securityContext:
    seccompProfile:
      type: Localhost
      localhostProfile: operator/default/log.json
  restartPolicy: Never
  containers:
  - name: container1
    image: busybox
    command: ["sh"]
    args: ["-c", "sleep infinity"]

• Start the audit-seccomp gadget.

$ kubectl gadget audit seccomp -o columns=k8s.namespace,k8s.podname,syscall,code
K8S.NAMESPACE    K8S.PODNAME      SYSCALL          CODE

• In another terminal, execute the aforementioned syscalls in the pod.

$ kubectl exec -ti  mypod -- /bin/sh
/ # mkdir /tmp/dir42 ; unshare -i
Bad system call (core dumped)

• Observe the syscalls logged by seccomp in the first terminal.

K8S.NAMESPACE    K8S.PODNAME      SYSCALL          CODE
default          mypod            mkdir            log
default          mypod            unshare          kill_thread

With ig

• Prepare a Seccomp Profile.

{
  "defaultAction": "SCMP_ACT_ALLOW",
  "syscalls": [
    {
      "action": "SCMP_ACT_KILL",
      "names": [
        "unshare"
      ]
    }
  ]
}

• Start the audit-seccomp gadget.

$ sudo ig audit seccomp -r docker
RUNTIME.CONTAINERNAME                              PID        COMM             SYSCALL     CODE

• In another terminal, start a container and run unshare:

$ docker run -ti --rm --security-opt seccomp=profile.json ubuntu
# unshare -i
Bad system call (core dumped)

• Observe the syscalls logged by seccomp in the first terminal.

$ sudo ig audit seccomp -r docker
RUNTIME.CONTAINERNAME                              PID        COMM             SYSCALL     CODE
eager_mclean                                       231712     unshare          unshare     kill_thread