[PATCH 5/7] nvme: Fix controller creation races with teardown flow

[Sagi Grimberg]

> FYI - I've hit a scenario with this patch, where if the device starts 
> rejecting the initial connections or they continuously hit a failure - 
> we're forced to wait ctrl_loss_tmo before it goes away. We can't 
> forcibly delete the controller via sysfs.  This shouldn't be possible.

The user should always be allowed to get rid of the controller. We
probably need to fix that.