[PATCH] nvmet: per-host namespaces masking

[Sagi Grimberg]

> Implement per-host namespaces masking, allowing to expose
> different namespaces to different hosts connecting to the
> same subsystem.
> The original method of adding host nqns to the subsystem
> takes precedence over the per-host namespace masking.

What is preventing a host to go ahead and issue an I/O command
to a ns that exists in the subsystem but is not exposed to
it in the nslist?

I don't think this can be even considered without a protection
against it. And given that you don't want to take the global
semaphore on each and every I/O, I'd say its a bit more involving
than what you have here...