With repeated port swaps between separate fabrics, there can be multiple
registrations for fabric well known address 0xfffffe. This can cause ndlp
reference confusion due to the usage of a single ndlp ptr that stores the
rport object in fc_rport struct private storage during transport
registration. Subsequent registrations update the ndlp->rport field with
the newer rport, so when transport layer triggers dev_loss_tmo for the
earlier registered rport the ndlp->rport private storage is referencing the
newer rport instead of the older rport in dev_loss_tmo callbk.
Because the older ndlp->rport object is already cleaned up elsewhere in
driver code during the time of fabric swap, check that the rport provided
in dev_loss_tmo callbk actually matches the rport stored in the LLDD's
ndlp->rport field. Otherwise, skip dev_loss_tmo work on a stale rport.
Signed-off-by: Justin Tee <justin.tee@xxxxxxxxxxxx>
---
drivers/scsi/lpfc/lpfc_hbadisc.c | 12 +++++++++---
1 file changed, 9 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_hbadisc.c b/drivers/scsi/lpfc/lpfc_hbadisc.c
index 36e66df36a18..2dfcf1db5395 100644
--- a/drivers/scsi/lpfc/lpfc_hbadisc.c
+++ b/drivers/scsi/lpfc/lpfc_hbadisc.c
@@ -228,10 +228,16 @@ lpfc_dev_loss_tmo_callbk(struct fc_rport *rport)
if (ndlp->nlp_state == NLP_STE_MAPPED_NODE)
return;
- /* check for recovered fabric node */
- if (ndlp->nlp_state == NLP_STE_UNMAPPED_NODE &&
- ndlp->nlp_DID == Fabric_DID)
+ /* Ignore callback for a mismatched (stale) rport */
+ if (ndlp->rport != rport) {
+ lpfc_vlog_msg(vport, KERN_WARNING, LOG_NODE,
+ "6788 fc rport mismatch: d_id x%06x ndlp x%px "
+ "fc rport x%px node rport x%px state x%x "
+ "refcnt %u\n",
+ ndlp->nlp_DID, ndlp, rport, ndlp->rport,
+ ndlp->nlp_state, kref_read(&ndlp->kref));
return;
+ }
if (rport->port_name != wwn_to_u64(ndlp->nlp_portname.u.wwn))
lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
--
2.38.0
