Local Block Volume should judge device mount point

General Discussions
1

Cluster information:

Kubernetes version: 1.22

Problem

When using Local Block Volume, we may mistakenly assign an used device. For example, when we use a local block volume as follow

apiVersion: v1
kind: PersistentVolume
metadata:
  name: support-es1-1-1-block-0
spec:
  accessModes:
  - ReadWriteOnce
  capacity:
    storage: 3500Gi
  local:
    path: /dev/nvme2n1
  nodeAffinity:
    required:
      nodeSelectorTerms:
      - matchExpressions:
        - key: kubernetes.io/hostname
          operator: In
          values:
          - 10.28.0.172
  persistentVolumeReclaimPolicy: Delete

If someone already used the device /dev/nvme2n1, made its filesystem and mounted this device with some path like /data. Then some POD uses a PVC which is bounded to this PV. The POD will directly write some data to this block device. This case will cause some damage to filesystem.

Solution

I propose some check should be added here. In Local Volume Setup code, now the code is shown as following.

// SetUpDevice prepares the volume to the node by the plugin specific way.
func (m *localVolumeMapper) SetUpDevice() (string, error) {
	return "", nil
}

We should add MountPoint check in this function as what we do in Filesystem Volume

// SetUpAt bind mounts the directory to the volume path and sets up volume ownership
func (m *localVolumeMounter) SetUpAt(dir string, mounterArgs volume.MounterArgs) error {
	...

	notMnt, err := mount.IsNotMountPoint(m.mounter, dir)
	klog.V(4).Infof("LocalVolume mount setup: PodDir(%s) VolDir(%s) Mounted(%t) Error(%v), ReadOnly(%t)", dir, m.globalPath, !notMnt, err, m.readOnly)
	if err != nil && !os.IsNotExist(err) {
		klog.Errorf("cannot validate mount point: %s %v", dir, err)
		return err
	}

	if !notMnt {
		return nil
	}
    ...