target: Re-org of core_tmr_lun_reset
authorNicholas Bellinger <nab@linux-iscsi.org>
Thu, 29 Sep 2011 04:37:29 +0000 (21:37 -0700)
committerNicholas Bellinger <nab@linux-iscsi.org>
Mon, 24 Oct 2011 03:19:32 +0000 (03:19 +0000)
This patch is a re-orginzation of core_tmr_lun_reset() logic to properly
scan the active tmr_list, dev->state_task_list and qobj->qobj_list w/ the
relivent locks held, and performing a list_move_tail onto seperate local
scope lists before performing the full drain.

This involves breaking out the code into three seperate list specific
functions: core_tmr_drain_tmr_list(), core_tmr_drain_task_list() and
core_tmr_drain_cmd_list().

(nab: Include target: Remove non-active tasks from execute list during
      LUN_RESET patch to address original breakage)

Reported-by: Roland Dreier <roland@purestorage.com>
Cc: Roland Dreier <roland@purestorage.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: stable@kernel.org
Signed-off-by: Nicholas Bellinger <nab@risingtidesystems.com>
drivers/target/target_core_tmr.c

index 98b12a8..ed0b1ff 100644 (file)
@@ -66,15 +66,16 @@ void core_tmr_release_req(
        struct se_tmr_req *tmr)
 {
        struct se_device *dev = tmr->tmr_dev;
+       unsigned long flags;
 
        if (!dev) {
                kmem_cache_free(se_tmr_req_cache, tmr);
                return;
        }
 
-       spin_lock_irq(&dev->se_tmr_lock);
+       spin_lock_irqsave(&dev->se_tmr_lock, flags);
        list_del(&tmr->tmr_list);
-       spin_unlock_irq(&dev->se_tmr_lock);
+       spin_unlock_irqrestore(&dev->se_tmr_lock, flags);
 
        kmem_cache_free(se_tmr_req_cache, tmr);
 }
@@ -99,54 +100,20 @@ static void core_tmr_handle_tas_abort(
        transport_cmd_finish_abort(cmd, 0);
 }
 
-int core_tmr_lun_reset(
+static void core_tmr_drain_tmr_list(
        struct se_device *dev,
        struct se_tmr_req *tmr,
-       struct list_head *preempt_and_abort_list,
-       struct se_cmd *prout_cmd)
+       struct list_head *preempt_and_abort_list)
 {
-       struct se_cmd *cmd, *tcmd;
-       struct se_node_acl *tmr_nacl = NULL;
-       struct se_portal_group *tmr_tpg = NULL;
-       struct se_queue_obj *qobj = &dev->dev_queue_obj;
+       LIST_HEAD(drain_tmr_list);
        struct se_tmr_req *tmr_p, *tmr_pp;
-       struct se_task *task, *task_tmp;
+       struct se_cmd *cmd;
        unsigned long flags;
-       int fe_count, tas;
-       /*
-        * TASK_ABORTED status bit, this is configurable via ConfigFS
-        * struct se_device attributes.  spc4r17 section 7.4.6 Control mode page
-        *
-        * A task aborted status (TAS) bit set to zero specifies that aborted
-        * tasks shall be terminated by the device server without any response
-        * to the application client. A TAS bit set to one specifies that tasks
-        * aborted by the actions of an I_T nexus other than the I_T nexus on
-        * which the command was received shall be completed with TASK ABORTED
-        * status (see SAM-4).
-        */
-       tas = dev->se_sub_dev->se_dev_attrib.emulate_tas;
-       /*
-        * Determine if this se_tmr is coming from a $FABRIC_MOD
-        * or struct se_device passthrough..
-        */
-       if (tmr && tmr->task_cmd && tmr->task_cmd->se_sess) {
-               tmr_nacl = tmr->task_cmd->se_sess->se_node_acl;
-               tmr_tpg = tmr->task_cmd->se_sess->se_tpg;
-               if (tmr_nacl && tmr_tpg) {
-                       pr_debug("LUN_RESET: TMR caller fabric: %s"
-                               " initiator port %s\n",
-                               tmr_tpg->se_tpg_tfo->get_fabric_name(),
-                               tmr_nacl->initiatorname);
-               }
-       }
-       pr_debug("LUN_RESET: %s starting for [%s], tas: %d\n",
-               (preempt_and_abort_list) ? "Preempt" : "TMR",
-               dev->transport->name, tas);
        /*
         * Release all pending and outgoing TMRs aside from the received
         * LUN_RESET tmr..
         */
-       spin_lock_irq(&dev->se_tmr_lock);
+       spin_lock_irqsave(&dev->se_tmr_lock, flags);
        list_for_each_entry_safe(tmr_p, tmr_pp, &dev->dev_tmr_list, tmr_list) {
                /*
                 * Allow the received TMR to return with FUNCTION_COMPLETE.
@@ -168,29 +135,48 @@ int core_tmr_lun_reset(
                    (core_scsi3_check_cdb_abort_and_preempt(
                                        preempt_and_abort_list, cmd) != 0))
                        continue;
-               spin_unlock_irq(&dev->se_tmr_lock);
 
-               spin_lock_irqsave(&cmd->t_state_lock, flags);
+               spin_lock(&cmd->t_state_lock);
                if (!atomic_read(&cmd->t_transport_active)) {
-                       spin_unlock_irqrestore(&cmd->t_state_lock, flags);
-                       spin_lock_irq(&dev->se_tmr_lock);
+                       spin_unlock(&cmd->t_state_lock);
                        continue;
                }
                if (cmd->t_state == TRANSPORT_ISTATE_PROCESSING) {
-                       spin_unlock_irqrestore(&cmd->t_state_lock, flags);
-                       spin_lock_irq(&dev->se_tmr_lock);
+                       spin_unlock(&cmd->t_state_lock);
                        continue;
                }
+               spin_unlock(&cmd->t_state_lock);
+
+               list_move_tail(&tmr->tmr_list, &drain_tmr_list);
+       }
+       spin_unlock_irqrestore(&dev->se_tmr_lock, flags);
+
+       while (!list_empty(&drain_tmr_list)) {
+               tmr = list_entry(drain_tmr_list.next, struct se_tmr_req, tmr_list);
+               list_del(&tmr->tmr_list);
+               cmd = tmr_p->task_cmd;
+
                pr_debug("LUN_RESET: %s releasing TMR %p Function: 0x%02x,"
                        " Response: 0x%02x, t_state: %d\n",
-                       (preempt_and_abort_list) ? "Preempt" : "", tmr_p,
-                       tmr_p->function, tmr_p->response, cmd->t_state);
-               spin_unlock_irqrestore(&cmd->t_state_lock, flags);
+                       (preempt_and_abort_list) ? "Preempt" : "", tmr,
+                       tmr->function, tmr->response, cmd->t_state);
 
                transport_cmd_finish_abort_tmr(cmd);
-               spin_lock_irq(&dev->se_tmr_lock);
        }
-       spin_unlock_irq(&dev->se_tmr_lock);
+}
+
+static void core_tmr_drain_task_list(
+       struct se_device *dev,
+       struct se_cmd *prout_cmd,
+       struct se_node_acl *tmr_nacl,
+       int tas,
+       struct list_head *preempt_and_abort_list)
+{
+       LIST_HEAD(drain_task_list);
+       struct se_cmd *cmd;
+       struct se_task *task, *task_tmp;
+       unsigned long flags;
+       int fe_count;
        /*
         * Complete outstanding struct se_task CDBs with TASK_ABORTED SAM status.
         * This is following sam4r17, section 5.6 Aborting commands, Table 38
@@ -235,9 +221,23 @@ int core_tmr_lun_reset(
                if (prout_cmd == cmd)
                        continue;
 
-               list_del(&task->t_state_list);
+               list_move_tail(&task->t_state_list, &drain_task_list);
                atomic_set(&task->task_state_active, 0);
-               spin_unlock_irqrestore(&dev->execute_task_lock, flags);
+               /*
+                * Remove from task execute list before processing drain_task_list
+                */
+               if (atomic_read(&task->task_execute_queue) != 0) {
+                       list_del(&task->t_execute_list);
+                       atomic_set(&task->task_execute_queue, 0);
+                       atomic_dec(&dev->execute_tasks);
+               }
+       }
+       spin_unlock_irqrestore(&dev->execute_task_lock, flags);
+
+       while (!list_empty(&drain_task_list)) {
+               task = list_entry(drain_task_list.next, struct se_task, t_state_list);
+               list_del(&task->t_state_list);
+               cmd = task->task_se_cmd;
 
                spin_lock_irqsave(&cmd->t_state_lock, flags);
                pr_debug("LUN_RESET: %s cmd: %p task: %p"
@@ -274,20 +274,14 @@ int core_tmr_lun_reset(
 
                        atomic_set(&task->task_active, 0);
                        atomic_set(&task->task_stop, 0);
-               } else {
-                       if (atomic_read(&task->task_execute_queue) != 0)
-                               transport_remove_task_from_execute_queue(task, dev);
                }
                __transport_stop_task_timer(task, &flags);
 
                if (!atomic_dec_and_test(&cmd->t_task_cdbs_ex_left)) {
-                       spin_unlock_irqrestore(
-                                       &cmd->t_state_lock, flags);
+                       spin_unlock_irqrestore(&cmd->t_state_lock, flags);
                        pr_debug("LUN_RESET: Skipping task: %p, dev: %p for"
                                " t_task_cdbs_ex_left: %d\n", task, dev,
                                atomic_read(&cmd->t_task_cdbs_ex_left));
-
-                       spin_lock_irqsave(&dev->execute_task_lock, flags);
                        continue;
                }
                fe_count = atomic_read(&cmd->t_fe_count);
@@ -297,22 +291,31 @@ int core_tmr_lun_reset(
                                " task: %p, t_fe_count: %d dev: %p\n", task,
                                fe_count, dev);
                        atomic_set(&cmd->t_transport_aborted, 1);
-                       spin_unlock_irqrestore(&cmd->t_state_lock,
-                                               flags);
-                       core_tmr_handle_tas_abort(tmr_nacl, cmd, tas, fe_count);
+                       spin_unlock_irqrestore(&cmd->t_state_lock, flags);
 
-                       spin_lock_irqsave(&dev->execute_task_lock, flags);
+                       core_tmr_handle_tas_abort(tmr_nacl, cmd, tas, fe_count);
                        continue;
                }
                pr_debug("LUN_RESET: Got t_transport_active = 0 for task: %p,"
                        " t_fe_count: %d dev: %p\n", task, fe_count, dev);
                atomic_set(&cmd->t_transport_aborted, 1);
                spin_unlock_irqrestore(&cmd->t_state_lock, flags);
-               core_tmr_handle_tas_abort(tmr_nacl, cmd, tas, fe_count);
 
-               spin_lock_irqsave(&dev->execute_task_lock, flags);
+               core_tmr_handle_tas_abort(tmr_nacl, cmd, tas, fe_count);
        }
-       spin_unlock_irqrestore(&dev->execute_task_lock, flags);
+}
+
+static void core_tmr_drain_cmd_list(
+       struct se_device *dev,
+       struct se_cmd *prout_cmd,
+       struct se_node_acl *tmr_nacl,
+       int tas,
+       struct list_head *preempt_and_abort_list)
+{
+       LIST_HEAD(drain_cmd_list);
+       struct se_queue_obj *qobj = &dev->dev_queue_obj;
+       struct se_cmd *cmd, *tcmd;
+       unsigned long flags;
        /*
         * Release all commands remaining in the struct se_device cmd queue.
         *
@@ -337,10 +340,15 @@ int core_tmr_lun_reset(
                if (prout_cmd == cmd)
                        continue;
 
-               atomic_dec(&cmd->t_transport_queue_active);
+               atomic_set(&cmd->t_transport_queue_active, 0);
                atomic_dec(&qobj->queue_cnt);
+               list_move_tail(&cmd->se_queue_node, &drain_cmd_list);
+       }
+       spin_unlock_irqrestore(&qobj->cmd_queue_lock, flags);
+
+       while (!list_empty(&drain_cmd_list)) {
+               cmd = list_entry(drain_cmd_list.next, struct se_cmd, se_queue_node);
                list_del_init(&cmd->se_queue_node);
-               spin_unlock_irqrestore(&qobj->cmd_queue_lock, flags);
 
                pr_debug("LUN_RESET: %s from Device Queue: cmd: %p t_state:"
                        " %d t_fe_count: %d\n", (preempt_and_abort_list) ?
@@ -353,9 +361,53 @@ int core_tmr_lun_reset(
 
                core_tmr_handle_tas_abort(tmr_nacl, cmd, tas,
                                atomic_read(&cmd->t_fe_count));
-               spin_lock_irqsave(&qobj->cmd_queue_lock, flags);
        }
-       spin_unlock_irqrestore(&qobj->cmd_queue_lock, flags);
+}
+
+int core_tmr_lun_reset(
+        struct se_device *dev,
+        struct se_tmr_req *tmr,
+        struct list_head *preempt_and_abort_list,
+        struct se_cmd *prout_cmd)
+{
+       struct se_node_acl *tmr_nacl = NULL;
+       struct se_portal_group *tmr_tpg = NULL;
+       int tas;
+        /*
+        * TASK_ABORTED status bit, this is configurable via ConfigFS
+        * struct se_device attributes.  spc4r17 section 7.4.6 Control mode page
+        *
+        * A task aborted status (TAS) bit set to zero specifies that aborted
+        * tasks shall be terminated by the device server without any response
+        * to the application client. A TAS bit set to one specifies that tasks
+        * aborted by the actions of an I_T nexus other than the I_T nexus on
+        * which the command was received shall be completed with TASK ABORTED
+        * status (see SAM-4).
+        */
+       tas = dev->se_sub_dev->se_dev_attrib.emulate_tas;
+       /*
+        * Determine if this se_tmr is coming from a $FABRIC_MOD
+        * or struct se_device passthrough..
+        */
+       if (tmr && tmr->task_cmd && tmr->task_cmd->se_sess) {
+               tmr_nacl = tmr->task_cmd->se_sess->se_node_acl;
+               tmr_tpg = tmr->task_cmd->se_sess->se_tpg;
+               if (tmr_nacl && tmr_tpg) {
+                       pr_debug("LUN_RESET: TMR caller fabric: %s"
+                               " initiator port %s\n",
+                               tmr_tpg->se_tpg_tfo->get_fabric_name(),
+                               tmr_nacl->initiatorname);
+               }
+       }
+       pr_debug("LUN_RESET: %s starting for [%s], tas: %d\n",
+               (preempt_and_abort_list) ? "Preempt" : "TMR",
+               dev->transport->name, tas);
+
+       core_tmr_drain_tmr_list(dev, tmr, preempt_and_abort_list);
+       core_tmr_drain_task_list(dev, prout_cmd, tmr_nacl, tas,
+                               preempt_and_abort_list);
+       core_tmr_drain_cmd_list(dev, prout_cmd, tmr_nacl, tas,
+                               preempt_and_abort_list);
        /*
         * Clear any legacy SPC-2 reservation when called during
         * LOGICAL UNIT RESET
@@ -378,3 +430,4 @@ int core_tmr_lun_reset(
                        dev->transport->name);
        return 0;
 }
+