Fix multiple connections problems to one same nqn in a single node (#23)

* fix: make multiple connections to one same nqn possible in a single node

Multiple connections to one same nqn in a single node return -EALREADY error.
So, we should use a hostnqn to make every connections seperate. It is appropriate
to use a target path from a volume request as a hostnqn.

* fix: make ro and rw mounts for one same nqn possible in a single node

ReadOnly and ReadWrite mounts for one same nqn in a single node return -EBUSY error.
So, we should remove readonly and readwrite options when mounting a block device in a
csi driver. A cri runtime makes a mount as readonly inside containers if needed.

* fix: add fallback supports for no hostnqn sysfs file supports

Directories per each nqn will be created in /run/nvmf and files per each hostnqn
will be created in /run/nvmf/{nqn}. If linux kernel has no hostnqn sysfs file
supports, we will disconnect all connections for a nqn at once when a directory for
the nqn is empty.

* refactor: rename sysfs_nqn_path to sysfs_subsysnqn_path

* refactor: use filepath.Join instead of strings.Join

* fix: add explicitly UnsupportedHostnqnError for fallback supports

If linux kernel has no hostnqn sysfs file supports, return UnsupportedHostnqnError
and switch to fallback mode which will disconnect all connections at once when
a last controller in a nqn is disconnected.

* fix: return directly after fallback mode

* fix: mount the host's /run/nvmf directory to csi-node-driver

pkg/nvmf/const.go

@@ -18,6 +18,7 @@ package nvmf
 const (
 	NVMF_NQN_SIZE = 223
 	SYS_NVMF      = "/sys/class/nvme"
+	RUN_NVMF      = "/run/nvmf"
 )
 
 // Here erron