spin_lock_irqsave(&ndlp->lock, iflags);
ndlp->nlp_flag |= NLP_IN_DEV_LOSS;
- ndlp->nlp_flag &= ~NLP_NPR_2B_DISC;
+
+ /* If there is a PLOGI in progress, and we are in a
+ * NLP_NPR_2B_DISC state, don't turn off the flag.
+ */
+ if (ndlp->nlp_state != NLP_STE_PLOGI_ISSUE)
+ ndlp->nlp_flag &= ~NLP_NPR_2B_DISC;
/*
* The backend does not expect any more calls associated with this
}
/**
+ * lpfc_check_nlp_post_devloss - Check to restore ndlp refcnt after devloss
+ * @vport: Pointer to vport object.
+ * @ndlp: Pointer to remote node object.
+ *
+ * If NLP_IN_RECOV_POST_DEV_LOSS flag was set due to outstanding recovery of
+ * node during dev_loss_tmo processing, then this function restores the nlp_put
+ * kref decrement from lpfc_dev_loss_tmo_handler.
+ **/
+void
+lpfc_check_nlp_post_devloss(struct lpfc_vport *vport,
+ struct lpfc_nodelist *ndlp)
+{
+ unsigned long iflags;
+
+ spin_lock_irqsave(&ndlp->lock, iflags);
+ if (ndlp->save_flags & NLP_IN_RECOV_POST_DEV_LOSS) {
+ ndlp->save_flags &= ~NLP_IN_RECOV_POST_DEV_LOSS;
+ spin_unlock_irqrestore(&ndlp->lock, iflags);
+ lpfc_nlp_get(ndlp);
+ lpfc_printf_vlog(vport, KERN_INFO, LOG_DISCOVERY | LOG_NODE,
+ "8438 Devloss timeout reversed on DID x%x "
+ "refcnt %d ndlp %p flag x%x "
+ "port_state = x%x\n",
+ ndlp->nlp_DID, kref_read(&ndlp->kref), ndlp,
+ ndlp->nlp_flag, vport->port_state);
+ spin_lock_irqsave(&ndlp->lock, iflags);
+ }
+ spin_unlock_irqrestore(&ndlp->lock, iflags);
+}
+
+/**
* lpfc_dev_loss_tmo_handler - Remote node devloss timeout handler
* @ndlp: Pointer to remote node object.
*
uint8_t *name;
int warn_on = 0;
int fcf_inuse = 0;
+ bool recovering = false;
+ struct fc_vport *fc_vport = NULL;
unsigned long iflags;
vport = ndlp->vport;
/* Fabric nodes are done. */
if (ndlp->nlp_type & NLP_FABRIC) {
+ spin_lock_irqsave(&ndlp->lock, iflags);
+ /* In massive vport configuration settings, it's possible
+ * dev_loss_tmo fired during node recovery. So, check if
+ * fabric nodes are in discovery states outstanding.
+ */
+ switch (ndlp->nlp_DID) {
+ case Fabric_DID:
+ fc_vport = vport->fc_vport;
+ if (fc_vport &&
+ fc_vport->vport_state == FC_VPORT_INITIALIZING)
+ recovering = true;
+ break;
+ case Fabric_Cntl_DID:
+ if (ndlp->nlp_flag & NLP_REG_LOGIN_SEND)
+ recovering = true;
+ break;
+ case FDMI_DID:
+ fallthrough;
+ case NameServer_DID:
+ if (ndlp->nlp_state >= NLP_STE_PLOGI_ISSUE &&
+ ndlp->nlp_state <= NLP_STE_REG_LOGIN_ISSUE)
+ recovering = true;
+ break;
+ }
+ spin_unlock_irqrestore(&ndlp->lock, iflags);
+
+ /* Mark an NLP_IN_RECOV_POST_DEV_LOSS flag to know if reversing
+ * the following lpfc_nlp_put is necessary after fabric node is
+ * recovered.
+ */
+ if (recovering) {
+ lpfc_printf_vlog(vport, KERN_INFO,
+ LOG_DISCOVERY | LOG_NODE,
+ "8436 Devloss timeout marked on "
+ "DID x%x refcnt %d ndlp %p "
+ "flag x%x port_state = x%x\n",
+ ndlp->nlp_DID, kref_read(&ndlp->kref),
+ ndlp, ndlp->nlp_flag,
+ vport->port_state);
+ spin_lock_irqsave(&ndlp->lock, iflags);
+ ndlp->save_flags |= NLP_IN_RECOV_POST_DEV_LOSS;
+ spin_unlock_irqrestore(&ndlp->lock, iflags);
+ } else if (ndlp->nlp_state == NLP_STE_UNMAPPED_NODE) {
+ /* Fabric node fully recovered before this dev_loss_tmo
+ * queue work is processed. Thus, ignore the
+ * dev_loss_tmo event.
+ */
+ lpfc_printf_vlog(vport, KERN_INFO,
+ LOG_DISCOVERY | LOG_NODE,
+ "8437 Devloss timeout ignored on "
+ "DID x%x refcnt %d ndlp %p "
+ "flag x%x port_state = x%x\n",
+ ndlp->nlp_DID, kref_read(&ndlp->kref),
+ ndlp, ndlp->nlp_flag,
+ vport->port_state);
+ return fcf_inuse;
+ }
+
lpfc_nlp_put(ndlp);
return fcf_inuse;
}
ndlp->nlp_state, ndlp->nlp_rpi);
}
+ /* If we are devloss, but we are in the process of rediscovering the
+ * ndlp, don't issue a NLP_EVT_DEVICE_RM event.
+ */
+ if (ndlp->nlp_state >= NLP_STE_PLOGI_ISSUE &&
+ ndlp->nlp_state <= NLP_STE_PRLI_ISSUE) {
+ return fcf_inuse;
+ }
+
if (!(ndlp->fc4_xpt_flags & NVME_XPT_REGD))
lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM);
goto out;
}
+ lpfc_check_nlp_post_devloss(vport, ndlp);
+
if (phba->sli_rev < LPFC_SLI_REV4)
ndlp->nlp_rpi = mb->un.varWords[0];
void
lpfc_nlp_reg_node(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
{
-
unsigned long iflags;
+ lpfc_check_nlp_post_devloss(vport, ndlp);
+
spin_lock_irqsave(&ndlp->lock, iflags);
if (ndlp->fc4_xpt_flags & NLP_XPT_REGD) {
/* Already registered with backend, trigger rescan */
/* Issue LOGO, if no LOGO is outstanding */
spin_lock_irqsave(&pnode->lock, flags);
- if (!(pnode->upcall_flags & NLP_WAIT_FOR_LOGO) &&
+ if (!(pnode->save_flags & NLP_WAIT_FOR_LOGO) &&
!pnode->logo_waitq) {
pnode->logo_waitq = &waitq;
pnode->nlp_fcp_info &= ~NLP_FCP_2_DEVICE;
pnode->nlp_flag |= NLP_ISSUE_LOGO;
- pnode->upcall_flags |= NLP_WAIT_FOR_LOGO;
+ pnode->save_flags |= NLP_WAIT_FOR_LOGO;
spin_unlock_irqrestore(&pnode->lock, flags);
lpfc_unreg_rpi(vport, pnode);
wait_event_timeout(waitq,
- (!(pnode->upcall_flags &
+ (!(pnode->save_flags &
NLP_WAIT_FOR_LOGO)),
msecs_to_jiffies(dev_loss_tmo *
1000));
- if (pnode->upcall_flags & NLP_WAIT_FOR_LOGO) {
+ if (pnode->save_flags & NLP_WAIT_FOR_LOGO) {
lpfc_printf_vlog(vport, KERN_ERR, logit,
"0725 SCSI layer TGTRST "
"failed & LOGO TMO (%d, %llu) "
"return x%x\n",
tgt_id, lun_id, status);
spin_lock_irqsave(&pnode->lock, flags);
- pnode->upcall_flags &= ~NLP_WAIT_FOR_LOGO;
+ pnode->save_flags &= ~NLP_WAIT_FOR_LOGO;
} else {
spin_lock_irqsave(&pnode->lock, flags);
}