scsi: lpfc: Fix NPIV Fabric Node reference counting

While testing initiator-side cable swaps with NPIV, oops occur.  The
reference counts for the Fabric nodes on the NPIV vports isn't balanced,
resulting in premature node removal.

The following fixes were made:

 - Removed the FC_LBIT check in lpfc_linkup_port. This removed the special
   case for vports that didn't have them clean up just like the physical
   port.

 - Removed the unreg_rpi call in lpfc_cleanup_node. In this section, the
   node is being removed in the context of a reference count release and a
   mailbox command can't be issued at this point.

 - Remove special case handling in the default mailbox completion handler
   that allowed the skipping of a node reference. Now, reference counting
   always requires the removal of the reference.

 - Move the location of the DEVICE_RM event is done during LOGO handling as
   the driver has additional work to do on the ndlp before puts/releases
   can be performed.

Link: https://lore.kernel.org/r/20201115192646.12977-10-james.smart@broadcom.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

1 files changed, 2 insertions, 4 deletions

diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c
index f0cbf98e7caa..86d9ab4bcebb 100644
--- a/drivers/scsi/lpfc/lpfc_init.c
+++ b/drivers/scsi/lpfc/lpfc_init.c
@@ -2851,10 +2851,8 @@ lpfc_cleanup(struct lpfc_vport *vport)
 			continue;
 		}
 
-		/* take care of nodes in unused state before the state
-		 * machine taking action.
-		 */
-		if (ndlp->nlp_state == NLP_STE_UNUSED_NODE) {
+		if (ndlp->nlp_DID == Fabric_Cntl_DID &&
+		    ndlp->nlp_state == NLP_STE_UNUSED_NODE) {
 			lpfc_nlp_put(ndlp);
 			continue;
 		}