On Thu, Feb 22, 2024 at 07:54:18AM -0800, Luis Chamberlain wrote: > > I should have been more precise here, I was trying to say supporting > > real fabrics targets. blktests already has some logic for PCI targets > > with $TEST_DEV but I haven't really looked into this part yet. > > Do fabric targets have a symlink which remains static? A pretty typical nvme fabric test is: setup phase target side: - create backing device (file/block) - create loopback device - create nvme subsystem setup phase host side: - discover - connect to the target test phase do something like reading/writing from '/dev/nvmeX' or 'nvme id-ctrl /dev/nvmeX', etc. cleanup phase host side: - disconnect from the target cleanup phase target side: - remove nvme subsystem - remove loopback device - remove backing device I'd like to make the setup and cleanup target side more flexible. The host side will not be affected at all by exchanging the current soft target side (aka nvmet) with something else. This means it's not about any device links in /dev. Hope this makes it a bit clearer.
