[SCSI] lpfc 8.2.4 : Rework misplaced reference taking on node structure
authorJames Smart <James.Smart@Emulex.Com>
Fri, 11 Jan 2008 06:53:27 +0000 (01:53 -0500)
committerJames Bottomley <James.Bottomley@HansenPartnership.com>
Wed, 23 Jan 2008 17:29:24 +0000 (11:29 -0600)
Rework misplaced reference taking on node structure

Signed-off-by: James Smart <James.Smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
drivers/scsi/lpfc/lpfc_ct.c
drivers/scsi/lpfc/lpfc_els.c
drivers/scsi/lpfc/lpfc_hbadisc.c
drivers/scsi/lpfc/lpfc_nportdisc.c
drivers/scsi/lpfc/lpfc_sli.c

index 3759ae1dc5e740b4be49bd3fe70c9f88ff45729b..92441ce610ede4601dd113ee9175f7ef5ebc81e8 100644 (file)
@@ -561,7 +561,6 @@ lpfc_cmpl_ct_cmd_gid_ft(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
        if (vport->load_flag & FC_UNLOADING)
                goto out;
 
-
        if (lpfc_els_chk_latt(vport) || lpfc_error_lost_link(irsp)) {
                lpfc_printf_vlog(vport, KERN_INFO, LOG_DISCOVERY,
                                 "0216 Link event during NS query\n");
index 8da6e8be9d83e24465c42dd00d5489e9f200bc49..c6b739dc6bc32c695e9721e32a5e4df79daa7900 100644 (file)
@@ -120,12 +120,8 @@ lpfc_prep_els_iocb(struct lpfc_vport *vport, uint8_t expectRsp,
        pcmd = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
        if (pcmd)
                pcmd->virt = lpfc_mbuf_alloc(phba, MEM_PRI, &pcmd->phys);
-       if (!pcmd || !pcmd->virt) {
-               kfree(pcmd);
-
-               lpfc_sli_release_iocbq(phba, elsiocb);
-               return NULL;
-       }
+       if (!pcmd || !pcmd->virt)
+               goto els_iocb_free_pcmb_exit;
 
        INIT_LIST_HEAD(&pcmd->list);
 
@@ -135,13 +131,8 @@ lpfc_prep_els_iocb(struct lpfc_vport *vport, uint8_t expectRsp,
                if (prsp)
                        prsp->virt = lpfc_mbuf_alloc(phba, MEM_PRI,
                                                     &prsp->phys);
-               if (!prsp || !prsp->virt) {
-                       kfree(prsp);
-                       lpfc_mbuf_free(phba, pcmd->virt, pcmd->phys);
-                       kfree(pcmd);
-                       lpfc_sli_release_iocbq(phba, elsiocb);
-                       return NULL;
-               }
+               if (!prsp || !prsp->virt)
+                       goto els_iocb_free_prsp_exit;
                INIT_LIST_HEAD(&prsp->list);
        } else {
                prsp = NULL;
@@ -152,15 +143,8 @@ lpfc_prep_els_iocb(struct lpfc_vport *vport, uint8_t expectRsp,
        if (pbuflist)
                pbuflist->virt = lpfc_mbuf_alloc(phba, MEM_PRI,
                                                 &pbuflist->phys);
-       if (!pbuflist || !pbuflist->virt) {
-               lpfc_sli_release_iocbq(phba, elsiocb);
-               lpfc_mbuf_free(phba, pcmd->virt, pcmd->phys);
-               lpfc_mbuf_free(phba, prsp->virt, prsp->phys);
-               kfree(pcmd);
-               kfree(prsp);
-               kfree(pbuflist);
-               return NULL;
-       }
+       if (!pbuflist || !pbuflist->virt)
+               goto els_iocb_free_pbuf_exit;
 
        INIT_LIST_HEAD(&pbuflist->list);
 
@@ -205,7 +189,10 @@ lpfc_prep_els_iocb(struct lpfc_vport *vport, uint8_t expectRsp,
                bpl->tus.w = le32_to_cpu(bpl->tus.w);
        }
 
+       /* prevent preparing iocb with NULL ndlp reference */
        elsiocb->context1 = lpfc_nlp_get(ndlp);
+       if (!elsiocb->context1)
+               goto els_iocb_free_pbuf_exit;
        elsiocb->context2 = pcmd;
        elsiocb->context3 = pbuflist;
        elsiocb->retry = retry;
@@ -231,8 +218,20 @@ lpfc_prep_els_iocb(struct lpfc_vport *vport, uint8_t expectRsp,
                                 cmdSize);
        }
        return elsiocb;
-}
 
+els_iocb_free_pbuf_exit:
+       lpfc_mbuf_free(phba, prsp->virt, prsp->phys);
+       kfree(pbuflist);
+
+els_iocb_free_prsp_exit:
+       lpfc_mbuf_free(phba, pcmd->virt, pcmd->phys);
+       kfree(prsp);
+
+els_iocb_free_pcmb_exit:
+       kfree(pcmd);
+       lpfc_sli_release_iocbq(phba, elsiocb);
+       return NULL;
+}
 
 static int
 lpfc_issue_fabric_reglogin(struct lpfc_vport *vport)
@@ -513,6 +512,9 @@ lpfc_cmpl_els_flogi(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
 
        /* Check to see if link went down during discovery */
        if (lpfc_els_chk_latt(vport)) {
+               /* One additional decrement on node reference count to
+                * trigger the release of the node
+                */
                lpfc_nlp_put(ndlp);
                goto out;
        }
@@ -731,6 +733,9 @@ lpfc_initial_flogi(struct lpfc_vport *vport)
        }
 
        if (lpfc_issue_els_flogi(vport, ndlp, 0)) {
+               /* This decrement of reference count to node shall kick off
+                * the release of the node.
+                */
                lpfc_nlp_put(ndlp);
        }
        return 1;
@@ -754,7 +759,11 @@ lpfc_initial_fdisc(struct lpfc_vport *vport)
                lpfc_dequeue_node(vport, ndlp);
        }
        if (lpfc_issue_els_fdisc(vport, ndlp, 0)) {
+               /* decrement node reference count to trigger the release of
+                * the node.
+                */
                lpfc_nlp_put(ndlp);
+               return 0;
        }
        return 1;
 }
@@ -1557,6 +1566,9 @@ lpfc_issue_els_scr(struct lpfc_vport *vport, uint32_t nportid, uint8_t retry)
                                     ndlp->nlp_DID, ELS_CMD_SCR);
 
        if (!elsiocb) {
+               /* This will trigger the release of the node just
+                * allocated
+                */
                lpfc_nlp_put(ndlp);
                return 1;
        }
@@ -1578,10 +1590,17 @@ lpfc_issue_els_scr(struct lpfc_vport *vport, uint32_t nportid, uint8_t retry)
        phba->fc_stat.elsXmitSCR++;
        elsiocb->iocb_cmpl = lpfc_cmpl_els_cmd;
        if (lpfc_sli_issue_iocb(phba, pring, elsiocb, 0) == IOCB_ERROR) {
+               /* The additional lpfc_nlp_put will cause the following
+                * lpfc_els_free_iocb routine to trigger the rlease of
+                * the node.
+                */
                lpfc_nlp_put(ndlp);
                lpfc_els_free_iocb(phba, elsiocb);
                return 1;
        }
+       /* This will cause the callback-function lpfc_cmpl_els_cmd to
+        * trigger the release of node.
+        */
        lpfc_nlp_put(ndlp);
        return 0;
 }
@@ -1613,6 +1632,9 @@ lpfc_issue_els_farpr(struct lpfc_vport *vport, uint32_t nportid, uint8_t retry)
        elsiocb = lpfc_prep_els_iocb(vport, 1, cmdsize, retry, ndlp,
                                     ndlp->nlp_DID, ELS_CMD_RNID);
        if (!elsiocb) {
+               /* This will trigger the release of the node just
+                * allocated
+                */
                lpfc_nlp_put(ndlp);
                return 1;
        }
@@ -1649,10 +1671,17 @@ lpfc_issue_els_farpr(struct lpfc_vport *vport, uint32_t nportid, uint8_t retry)
        phba->fc_stat.elsXmitFARPR++;
        elsiocb->iocb_cmpl = lpfc_cmpl_els_cmd;
        if (lpfc_sli_issue_iocb(phba, pring, elsiocb, 0) == IOCB_ERROR) {
+               /* The additional lpfc_nlp_put will cause the following
+                * lpfc_els_free_iocb routine to trigger the release of
+                * the node.
+                */
                lpfc_nlp_put(ndlp);
                lpfc_els_free_iocb(phba, elsiocb);
                return 1;
        }
+       /* This will cause the callback-function lpfc_cmpl_els_cmd to
+        * trigger the release of the node.
+        */
        lpfc_nlp_put(ndlp);
        return 0;
 }
@@ -1712,7 +1741,10 @@ lpfc_els_retry_delay(unsigned long ptr)
                return;
        }
 
-       evtp->evt_arg1  = ndlp;
+       /* We need to hold the node by incrementing the reference
+        * count until the queued work is done
+        */
+       evtp->evt_arg1  = lpfc_nlp_get(ndlp);
        evtp->evt       = LPFC_EVT_ELS_RETRY;
        list_add_tail(&evtp->evt_listp, &phba->work_list);
        if (phba->work_wait)
@@ -2190,6 +2222,11 @@ lpfc_cmpl_els_logo_acc(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
                         * thread, just unregister the RPI.
                         */
                        lpfc_unreg_rpi(vport, ndlp);
+               } else {
+                       /* Indicate the node has already released, should
+                        * not reference to it from within lpfc_els_free_iocb.
+                        */
+                       cmdiocb->context1 = NULL;
                }
        }
        lpfc_els_free_iocb(phba, cmdiocb);
@@ -2208,7 +2245,6 @@ lpfc_mbx_cmpl_dflt_rpi(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
        mempool_free(pmb, phba->mbox_mem_pool);
        if (ndlp) {
                lpfc_nlp_put(ndlp);
-
                /* This is the end of the default RPI cleanup logic for this
                 * ndlp. If no other discovery threads are using this ndlp.
                 * we should free all resources associated with it.
@@ -2236,11 +2272,13 @@ lpfc_cmpl_els_rsp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
        if (cmdiocb->context_un.mbox)
                mbox = cmdiocb->context_un.mbox;
 
-       /* First determine if this is a LS_RJT cmpl */
+       /* First determine if this is a LS_RJT cmpl. Note, this callback
+        * function can have cmdiocb->contest1 (ndlp) field set to NULL.
+        */
        pcmd = (uint8_t *) (((struct lpfc_dmabuf *) cmdiocb->context2)->virt);
-       if (*((uint32_t *) (pcmd)) == ELS_CMD_LS_RJT) {
-               /* A LS_RJT associated with Default RPI cleanup
-                * has its own seperate code path.
+       if (ndlp && (*((uint32_t *) (pcmd)) == ELS_CMD_LS_RJT)) {
+               /* A LS_RJT associated with Default RPI cleanup has its own
+                * seperate code path.
                 */
                if (!(ndlp->nlp_flag & NLP_RM_DFLT_RPI))
                        ls_rjt = 1;
@@ -2257,8 +2295,14 @@ lpfc_cmpl_els_rsp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
                        mempool_free(mbox, phba->mbox_mem_pool);
                }
                if (ndlp && (ndlp->nlp_flag & NLP_RM_DFLT_RPI))
-                       if (lpfc_nlp_not_used(ndlp))
+                       if (lpfc_nlp_not_used(ndlp)) {
                                ndlp = NULL;
+                               /* Indicate the node has already released,
+                                * should not reference to it from within
+                                * the routine lpfc_els_free_iocb.
+                                */
+                               cmdiocb->context1 = NULL;
+                       }
                goto out;
        }
 
@@ -2302,14 +2346,27 @@ lpfc_cmpl_els_rsp(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
                                ndlp->nlp_DID, ndlp->nlp_flag, ndlp->nlp_state,
                                ndlp->nlp_rpi);
 
-                       if (lpfc_nlp_not_used(ndlp))
+                       if (lpfc_nlp_not_used(ndlp)) {
                                ndlp = NULL;
+                               /* Indicate node has already been released,
+                                * should not reference to it from within
+                                * the routine lpfc_els_free_iocb.
+                                */
+                               cmdiocb->context1 = NULL;
+                       }
                } else {
                        /* Do not drop node for lpfc_els_abort'ed ELS cmds */
                        if (!lpfc_error_lost_link(irsp) &&
                            ndlp->nlp_flag & NLP_ACC_REGLOGIN) {
-                               if (lpfc_nlp_not_used(ndlp))
+                               if (lpfc_nlp_not_used(ndlp)) {
                                        ndlp = NULL;
+                                       /* Indicate node has already been
+                                        * released, should not reference
+                                        * to it from within the routine
+                                        * lpfc_els_free_iocb.
+                                        */
+                                       cmdiocb->context1 = NULL;
+                               }
                        }
                }
                mp = (struct lpfc_dmabuf *) mbox->context1;
@@ -2331,7 +2388,12 @@ out:
                 * resources.
                 */
                if (ls_rjt)
-                       lpfc_nlp_not_used(ndlp);
+                       if (lpfc_nlp_not_used(ndlp))
+                               /* Indicate node has already been released,
+                                * should not reference to it from within
+                                * the routine lpfc_els_free_iocb.
+                                */
+                               cmdiocb->context1 = NULL;
        }
 
        lpfc_els_free_iocb(phba, cmdiocb);
@@ -3292,7 +3354,10 @@ lpfc_els_rsp_rps_acc(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
        elsiocb = lpfc_prep_els_iocb(phba->pport, 0, cmdsize,
                                     lpfc_max_els_tries, ndlp,
                                     ndlp->nlp_DID, ELS_CMD_ACC);
+
+       /* Decrement the ndlp reference count from previous mbox command */
        lpfc_nlp_put(ndlp);
+
        if (!elsiocb)
                return;
 
@@ -3375,11 +3440,13 @@ lpfc_els_rcv_rps(struct lpfc_vport *vport, struct lpfc_iocbq *cmdiocb,
                        mbox->context2 = lpfc_nlp_get(ndlp);
                        mbox->vport = vport;
                        mbox->mbox_cmpl = lpfc_els_rsp_rps_acc;
-                       if (lpfc_sli_issue_mbox (phba, mbox, MBX_NOWAIT)
+                       if (lpfc_sli_issue_mbox(phba, mbox, MBX_NOWAIT)
                                != MBX_NOT_FINISHED)
                                /* Mbox completion will send ELS Response */
                                return 0;
-
+                       /* Decrement reference count used for the failed mbox
+                        * command.
+                        */
                        lpfc_nlp_put(ndlp);
                        mempool_free(mbox, phba->mbox_mem_pool);
                }
@@ -4284,7 +4351,6 @@ lpfc_cmpl_reg_new_vport(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
        spin_lock_irq(shost->host_lock);
        vport->fc_flag &= ~FC_VPORT_NEEDS_REG_VPI;
        spin_unlock_irq(shost->host_lock);
-       lpfc_nlp_put(ndlp);
 
        if (mb->mbxStatus) {
                lpfc_printf_vlog(vport, KERN_ERR, LOG_MBOX,
@@ -4317,6 +4383,12 @@ lpfc_cmpl_reg_new_vport(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
                else
                        lpfc_do_scr_ns_plogi(phba, vport);
        }
+
+       /* Now, we decrement the ndlp reference count held for this
+        * callback function
+        */
+       lpfc_nlp_put(ndlp);
+
        mempool_free(pmb, phba->mbox_mem_pool);
        return;
 }
@@ -4336,26 +4408,29 @@ lpfc_register_new_vport(struct lpfc_hba *phba, struct lpfc_vport *vport,
                mbox->mbox_cmpl = lpfc_cmpl_reg_new_vport;
                if (lpfc_sli_issue_mbox(phba, mbox, MBX_NOWAIT)
                    == MBX_NOT_FINISHED) {
+                       /* mailbox command not success, decrement ndlp
+                        * reference count for this command
+                        */
+                       lpfc_nlp_put(ndlp);
                        mempool_free(mbox, phba->mbox_mem_pool);
-                       spin_lock_irq(shost->host_lock);
-                       vport->fc_flag &= ~FC_VPORT_NEEDS_REG_VPI;
-                       spin_unlock_irq(shost->host_lock);
 
-                       lpfc_vport_set_state(vport, FC_VPORT_FAILED);
                        lpfc_printf_vlog(vport, KERN_ERR, LOG_MBOX,
                                "0253 Register VPI: Can't send mbox\n");
+                       goto mbox_err_exit;
                }
        } else {
-               lpfc_vport_set_state(vport, FC_VPORT_FAILED);
-
                lpfc_printf_vlog(vport, KERN_ERR, LOG_MBOX,
                                 "0254 Register VPI: no memory\n");
-
-               spin_lock_irq(shost->host_lock);
-               vport->fc_flag &= ~FC_VPORT_NEEDS_REG_VPI;
-               spin_unlock_irq(shost->host_lock);
-               lpfc_nlp_put(ndlp);
+               goto mbox_err_exit;
        }
+       return;
+
+mbox_err_exit:
+       lpfc_vport_set_state(vport, FC_VPORT_FAILED);
+       spin_lock_irq(shost->host_lock);
+       vport->fc_flag &= ~FC_VPORT_NEEDS_REG_VPI;
+       spin_unlock_irq(shost->host_lock);
+       return;
 }
 
 static void
@@ -4436,7 +4511,8 @@ lpfc_cmpl_els_fdisc(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
                else
                        lpfc_do_scr_ns_plogi(phba, vport);
 
-               lpfc_nlp_put(ndlp); /* Free Fabric ndlp for vports */
+               /* Unconditionaly kick off releasing fabric node for vports */
+               lpfc_nlp_put(ndlp);
        }
 
 out:
index 644d96012d565330ff870abe1895242d3bf9ba0c..dc042bd97baa62e3ebb2fcd09b322f85f921df2e 100644 (file)
@@ -149,7 +149,10 @@ lpfc_dev_loss_tmo_callbk(struct fc_rport *rport)
                return;
 
        spin_lock_irq(&phba->hbalock);
-       evtp->evt_arg1  = ndlp;
+       /* We need to hold the node by incrementing the reference
+        * count until this queued work is done
+        */
+       evtp->evt_arg1  = lpfc_nlp_get(ndlp);
        evtp->evt       = LPFC_EVT_DEV_LOSS;
        list_add_tail(&evtp->evt_listp, &phba->work_list);
        if (phba->work_wait)
@@ -300,12 +303,18 @@ lpfc_work_list_done(struct lpfc_hba *phba)
                        ndlp = (struct lpfc_nodelist *) (evtp->evt_arg1);
                        lpfc_els_retry_delay_handler(ndlp);
                        free_evt = 0; /* evt is part of ndlp */
+                       /* decrement the node reference count held
+                        * for this queued work
+                        */
+                       lpfc_nlp_put(ndlp);
                        break;
                case LPFC_EVT_DEV_LOSS:
                        ndlp = (struct lpfc_nodelist *)(evtp->evt_arg1);
-                       lpfc_nlp_get(ndlp);
                        lpfc_dev_loss_tmo_handler(ndlp);
                        free_evt = 0;
+                       /* decrement the node reference count held for
+                        * this queued work
+                        */
                        lpfc_nlp_put(ndlp);
                        break;
                case LPFC_EVT_ONLINE:
@@ -1176,6 +1185,9 @@ lpfc_mbx_cmpl_reg_login(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
        lpfc_mbuf_free(phba, mp->virt, mp->phys);
        kfree(mp);
        mempool_free(pmb, phba->mbox_mem_pool);
+       /* decrement the node reference count held for this callback
+        * function.
+        */
        lpfc_nlp_put(ndlp);
 
        return;
@@ -1363,6 +1375,9 @@ lpfc_mbx_cmpl_ns_reg_login(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
 
        if (mb->mbxStatus) {
 out:
+               /* decrement the node reference count held for this
+                * callback function.
+                */
                lpfc_nlp_put(ndlp);
                lpfc_mbuf_free(phba, mp->virt, mp->phys);
                kfree(mp);
@@ -1414,6 +1429,9 @@ out:
                goto out;
        }
 
+       /* decrement the node reference count held for this
+        * callback function.
+        */
        lpfc_nlp_put(ndlp);
        lpfc_mbuf_free(phba, mp->virt, mp->phys);
        kfree(mp);
@@ -1661,13 +1679,14 @@ void
 lpfc_drop_node(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
 {
        /*
-        * Use of lpfc_drop_node and UNUSED list. lpfc_drop_node should
+        * Use of lpfc_drop_node and UNUSED list: lpfc_drop_node should
         * be used if we wish to issue the "last" lpfc_nlp_put() to remove
-        * the ndlp from the vport.  The ndlp resides on the UNUSED list
-        * until ALL other outstanding threads have completed. Thus, if a
-        * ndlp is on the UNUSED list already, we should never do another
-        * lpfc_drop_node() on it.
+        * the ndlp from the vport. The ndlp marked as UNUSED on the list
+        * until ALL other outstanding threads have completed. We check
+        * that the ndlp not already in the UNUSED state before we proceed.
         */
+       if (ndlp->nlp_state == NLP_STE_UNUSED_NODE)
+               return;
        lpfc_nlp_set_state(vport, ndlp, NLP_STE_UNUSED_NODE);
        lpfc_nlp_put(ndlp);
        return;
@@ -2767,7 +2786,9 @@ lpfc_mbx_cmpl_fdmi_reg_login(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
        else
                mod_timer(&vport->fc_fdmitmo, jiffies + HZ * 60);
 
-                               /* Mailbox took a reference to the node */
+       /* decrement the node reference count held for this callback
+        * function.
+        */
        lpfc_nlp_put(ndlp);
        lpfc_mbuf_free(phba, mp->virt, mp->phys);
        kfree(mp);
index 783659aa2102927d935d36877be3d754456d9f7e..4a0e3406e37ab787060c8713cf56a2581bdcb256 100644 (file)
@@ -922,6 +922,9 @@ lpfc_cmpl_plogi_plogi_issue(struct lpfc_vport *vport,
                                           NLP_STE_REG_LOGIN_ISSUE);
                        return ndlp->nlp_state;
                }
+               /* decrement node reference count to the failed mbox
+                * command
+                */
                lpfc_nlp_put(ndlp);
                mp = (struct lpfc_dmabuf *) mbox->context1;
                lpfc_mbuf_free(phba, mp->virt, mp->phys);
index 584c5451641fbb2f74be6103b3062811e9d39c52..fdd01e384e363d39b0ebf0114a1df68cbdf2dc56 100644 (file)
@@ -2605,7 +2605,7 @@ lpfc_sli_issue_mbox(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmbox, uint32_t flag)
                                        "1806 Mbox x%x failed. No vport\n",
                                        pmbox->mb.mbxCommand);
                        dump_stack();
-                       return MBXERR_ERROR;
+                       return MBX_NOT_FINISHED;
                }
        }