The S3PR reservation is "owned" by the host (ITN), which means even if a VM currently has exclusive access to the LUN, another VM that doesn't hold the reservation but is running on the same node ultimately has full access to the LUN.

If the other VM that doesn't have the reservation tries to write/mount the disk, it will likely succeed, which may cause corruption.

For this reason it is safer if we never let VMs sharing the same S3PR disk to run on the same node. This would never happen:

# oc get vm rhel-94 -o yaml | yq '.spec.template.spec.domain.devices.disks[3]'
lun:
  bus: scsi
  reservation: true
name: iscsi

# oc get vm rhel-96 -o yaml | yq '.spec.template.spec.domain.devices.disks[3]'
lun:
  bus: scsi
  reservation: true
name: iscsi 

# oc get vmi | grep rhel-9 
rhel-94      3m45s   Running   192.168.3.14   cyan.shift.home.arpa     True
rhel-96      4m22s   Running   192.168.3.15   cyan.shift.home.arpa     True

Please have CNV create some automatic hard anti-affinity rule to prevent this. Maybe a knob to enable/disable the automatic creation may be useful as well.