Starting esx8.03, vmware code is trying to use "nvme extended copy" for doing ds migrations. We do not support this option. In theory vmware should understand that during the connection negotiation phase, and avoid calling nvme extended copy. But we are not behaving 100% correctly which is causing the esx to believe that we do support it.
Affected esx version are 8.0u3 and beyond.
The affected ceph versions are 7.1 and beyond.
The fix is simple, we need to change the spdk a little bit to return error on one of the commands, if the "reserved" bits are not zeroed.