The current fc transport add_port routine validates that there is a
matching port to the target port config. It then takes a reference
on the targetport. The del_port removes the reference.
Unfortunately, if the LLDD undergoes a hw reset or driver unload and
wants to unreg the targetport, due to the reference, the targetport
effectively can't be removed. It requires the admin to remove the
port from the nvmet config first, which calls the del_port.
Note: it appears nvmetcli clear skips over the del_port call (I'm
not attempting to change that).
There's no real reason to take the reference. With FC, there is nothing
to enable or disable as the presence of the FC targetport implicitly
means its enabled, and removal of the targtport means its disabled.
Change add_port to simply validate and change remove_port to a noop.
No references are taken on the targetport.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
list_for_each_entry(tgtport, &nvmet_fc_target_list, tgt_list) {
if ((tgtport->fc_target_port.node_name == traddr.nn) &&
(tgtport->fc_target_port.port_name == traddr.pn)) {
- /* a FC port can only be 1 nvmet port id */
- if (!tgtport->port) {
- tgtport->port = port;
- port->priv = tgtport;
- nvmet_fc_tgtport_get(tgtport);
- ret = 0;
- } else
- ret = -EALREADY;
+ tgtport->port = port;
+ ret = 0;
break;
}
}
static void
nvmet_fc_remove_port(struct nvmet_port *port)
{
- struct nvmet_fc_tgtport *tgtport = port->priv;
- unsigned long flags;
- bool matched = false;
-
- spin_lock_irqsave(&nvmet_fc_tgtlock, flags);
- if (tgtport->port == port) {
- matched = true;
- tgtport->port = NULL;
- }
- spin_unlock_irqrestore(&nvmet_fc_tgtlock, flags);
-
- if (matched)
- nvmet_fc_tgtport_put(tgtport);
+ /* nothing to do */
}
static struct nvmet_fabrics_ops nvmet_fc_tgt_fcp_ops = {