So that's interesting. But what I'm mostly worried about is how the nvmet kernel code allows a request without ->port to progress to the actual command handler. We should never allow a command to get that far if ->port is NULL, and should not allow to clear ->port while commands are still handled.