1 files changed, 302 insertions, 139 deletions

diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 243fc09b164..c0365c14b85 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -18,6 +18,7 @@
 #include <linux/smp.h>
 #include <linux/spinlock.h>
 #include <linux/mutex.h>
+#include <linux/freezer.h>
 
 #include <linux/sunrpc/clnt.h>
 
@@ -27,6 +28,9 @@
 #define RPCDBG_FACILITY		RPCDBG_SCHED
 #endif
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/sunrpc.h>
+
 /*
  * RPC slabs and memory pools
  */
@@ -94,18 +98,55 @@ __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
 	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
 }
 
+static void rpc_rotate_queue_owner(struct rpc_wait_queue *queue)
+{
+	struct list_head *q = &queue->tasks[queue->priority];
+	struct rpc_task *task;
+
+	if (!list_empty(q)) {
+		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
+		if (task->tk_owner == queue->owner)
+			list_move_tail(&task->u.tk_wait.list, q);
+	}
+}
+
+static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
+{
+	if (queue->priority != priority) {
+		/* Fairness: rotate the list when changing priority */
+		rpc_rotate_queue_owner(queue);
+		queue->priority = priority;
+	}
+}
+
+static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
+{
+	queue->owner = pid;
+	queue->nr = RPC_BATCH_COUNT;
+}
+
+static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
+{
+	rpc_set_waitqueue_priority(queue, queue->maxpriority);
+	rpc_set_waitqueue_owner(queue, 0);
+}
+
 /*
  * Add new request to a priority queue.
  */
-static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
+static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
+		struct rpc_task *task,
+		unsigned char queue_priority)
 {
 	struct list_head *q;
 	struct rpc_task *t;
 
 	INIT_LIST_HEAD(&task->u.tk_wait.links);
-	q = &queue->tasks[task->tk_priority];
-	if (unlikely(task->tk_priority > queue->maxpriority))
-		q = &queue->tasks[queue->maxpriority];
+	if (unlikely(queue_priority > queue->maxpriority))
+		queue_priority = queue->maxpriority;
+	if (queue_priority > queue->priority)
+		rpc_set_waitqueue_priority(queue, queue_priority);
+	q = &queue->tasks[queue_priority];
 	list_for_each_entry(t, q, u.tk_wait.list) {
 		if (t->tk_owner == task->tk_owner) {
 			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
@@ -123,18 +164,24 @@ static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct r
  * improve overall performance.
  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
  */
-static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
+static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
+		struct rpc_task *task,
+		unsigned char queue_priority)
 {
-	BUG_ON (RPC_IS_QUEUED(task));
+	WARN_ON_ONCE(RPC_IS_QUEUED(task));
+	if (RPC_IS_QUEUED(task))
+		return;
 
 	if (RPC_IS_PRIORITY(queue))
-		__rpc_add_wait_queue_priority(queue, task);
+		__rpc_add_wait_queue_priority(queue, task, queue_priority);
 	else if (RPC_IS_SWAPPER(task))
 		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 	else
 		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 	task->tk_waitqueue = queue;
 	queue->qlen++;
+	/* barrier matches the read in rpc_wake_up_task_queue_locked() */
+	smp_wmb();
 	rpc_set_queued(task);
 
 	dprintk("RPC: %5u added to queue %p \"%s\"\n",
@@ -170,24 +217,6 @@ static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_tas
 			task->tk_pid, queue, rpc_qname(queue));
 }
 
-static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
-{
-	queue->priority = priority;
-	queue->count = 1 << (priority * 2);
-}
-
-static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
-{
-	queue->owner = pid;
-	queue->nr = RPC_BATCH_COUNT;
-}
-
-static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
-{
-	rpc_set_waitqueue_priority(queue, queue->maxpriority);
-	rpc_set_waitqueue_owner(queue, 0);
-}
-
 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 {
 	int i;
@@ -200,9 +229,7 @@ static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const c
 	queue->qlen = 0;
 	setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 	INIT_LIST_HEAD(&queue->timer_list.list);
-#ifdef RPC_DEBUG
-	queue->name = qname;
-#endif
+	rpc_assign_waitqueue_name(queue, qname);
 }
 
 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
@@ -227,11 +254,11 @@ static int rpc_wait_bit_killable(void *word)
 {
 	if (fatal_signal_pending(current))
 		return -ERESTARTSYS;
-	schedule();
+	freezable_schedule_unsafe();
 	return 0;
 }
 
-#ifdef RPC_DEBUG
+#if defined(RPC_DEBUG) || defined(RPC_TRACEPOINTS)
 static void rpc_task_set_debuginfo(struct rpc_task *task)
 {
 	static atomic_t rpc_pid;
@@ -246,29 +273,47 @@ static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 
 static void rpc_set_active(struct rpc_task *task)
 {
+	trace_rpc_task_begin(task->tk_client, task, NULL);
+
 	rpc_task_set_debuginfo(task);
 	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 }
 
 /*
  * Mark an RPC call as having completed by clearing the 'active' bit
+ * and then waking up all tasks that were sleeping.
  */
-static void rpc_mark_complete_task(struct rpc_task *task)
+static int rpc_complete_task(struct rpc_task *task)
 {
-	smp_mb__before_clear_bit();
+	void *m = &task->tk_runstate;
+	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
+	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
+	unsigned long flags;
+	int ret;
+
+	trace_rpc_task_complete(task->tk_client, task, NULL);
+
+	spin_lock_irqsave(&wq->lock, flags);
 	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
-	smp_mb__after_clear_bit();
-	wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
+	ret = atomic_dec_and_test(&task->tk_count);
+	if (waitqueue_active(wq))
+		__wake_up_locked_key(wq, TASK_NORMAL, &k);
+	spin_unlock_irqrestore(&wq->lock, flags);
+	return ret;
 }
 
 /*
  * Allow callers to wait for completion of an RPC call
+ *
+ * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
+ * to enforce taking of the wq->lock and hence avoid races with
+ * rpc_complete_task().
  */
 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 {
 	if (action == NULL)
 		action = rpc_wait_bit_killable;
-	return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
+	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 			action, TASK_KILLABLE);
 }
 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
@@ -276,24 +321,24 @@ EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 /*
  * Make an RPC task runnable.
  *
- * Note: If the task is ASYNC, this must be called with
- * the spinlock held to protect the wait queue operation.
+ * Note: If the task is ASYNC, and is being made runnable after sitting on an
+ * rpc_wait_queue, this must be called with the queue spinlock held to protect
+ * the wait queue operation.
+ * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
+ * which is needed to ensure that __rpc_execute() doesn't loop (due to the
+ * lockless RPC_IS_QUEUED() test) before we've had a chance to test
+ * the RPC_TASK_RUNNING flag.
  */
 static void rpc_make_runnable(struct rpc_task *task)
 {
+	bool need_wakeup = !rpc_test_and_set_running(task);
+
 	rpc_clear_queued(task);
-	if (rpc_test_and_set_running(task))
+	if (!need_wakeup)
 		return;
 	if (RPC_IS_ASYNC(task)) {
-		int status;
-
 		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
-		status = queue_work(rpciod_workqueue, &task->u.tk_work);
-		if (status < 0) {
-			printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
-			task->tk_status = status;
-			return;
-		}
+		queue_work(rpciod_workqueue, &task->u.tk_work);
 	} else
 		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 }
@@ -304,15 +349,19 @@ static void rpc_make_runnable(struct rpc_task *task)
  * NB: An RPC task will only receive interrupt-driven events as long
  * as it's on a wait queue.
  */
-static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
-			rpc_action action)
+static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
+		struct rpc_task *task,
+		rpc_action action,
+		unsigned char queue_priority)
 {
 	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 			task->tk_pid, rpc_qname(q), jiffies);
 
-	__rpc_add_wait_queue(q, task);
+	trace_rpc_task_sleep(task->tk_client, task, q);
 
-	BUG_ON(task->tk_callback != NULL);
+	__rpc_add_wait_queue(q, task, queue_priority);
+
+	WARN_ON_ONCE(task->tk_callback != NULL);
 	task->tk_callback = action;
 	__rpc_add_timer(q, task);
 }
@@ -321,17 +370,42 @@ void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 				rpc_action action)
 {
 	/* We shouldn't ever put an inactive task to sleep */
-	BUG_ON(!RPC_IS_ACTIVATED(task));
+	WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
+	if (!RPC_IS_ACTIVATED(task)) {
+		task->tk_status = -EIO;
+		rpc_put_task_async(task);
+		return;
+	}
 
 	/*
 	 * Protect the queue operations.
 	 */
 	spin_lock_bh(&q->lock);
-	__rpc_sleep_on(q, task, action);
+	__rpc_sleep_on_priority(q, task, action, task->tk_priority);
 	spin_unlock_bh(&q->lock);
 }
 EXPORT_SYMBOL_GPL(rpc_sleep_on);
 
+void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
+		rpc_action action, int priority)
+{
+	/* We shouldn't ever put an inactive task to sleep */
+	WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
+	if (!RPC_IS_ACTIVATED(task)) {
+		task->tk_status = -EIO;
+		rpc_put_task_async(task);
+		return;
+	}
+
+	/*
+	 * Protect the queue operations.
+	 */
+	spin_lock_bh(&q->lock);
+	__rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
+	spin_unlock_bh(&q->lock);
+}
+EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
+
 /**
  * __rpc_do_wake_up_task - wake up a single rpc_task
  * @queue: wait queue
@@ -350,6 +424,8 @@ static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task
 		return;
 	}
 
+	trace_rpc_task_wakeup(task->tk_client, task, queue);
+
 	__rpc_remove_wait_queue(queue, task);
 
 	rpc_make_runnable(task);
@@ -362,23 +438,12 @@ static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task
  */
 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
-	if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
-		__rpc_do_wake_up_task(queue, task);
-}
-
-/*
- * Tests whether rpc queue is empty
- */
-int rpc_queue_empty(struct rpc_wait_queue *queue)
-{
-	int res;
-
-	spin_lock_bh(&queue->lock);
-	res = queue->qlen;
-	spin_unlock_bh(&queue->lock);
-	return res == 0;
+	if (RPC_IS_QUEUED(task)) {
+		smp_rmb();
+		if (task->tk_waitqueue == queue)
+			__rpc_do_wake_up_task(queue, task);
+	}
 }
-EXPORT_SYMBOL_GPL(rpc_queue_empty);
 
 /*
  * Wake up a task on a specific queue
@@ -394,7 +459,7 @@ EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 /*
  * Wake up the next task on a priority queue.
  */
-static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
+static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 {
 	struct list_head *q;
 	struct rpc_task *task;
@@ -413,8 +478,7 @@ static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queu
 		/*
 		 * Check if we need to switch queues.
 		 */
-		if (--queue->count)
-			goto new_owner;
+		goto new_owner;
 	}
 
 	/*
@@ -439,30 +503,54 @@ new_queue:
 new_owner:
 	rpc_set_waitqueue_owner(queue, task->tk_owner);
 out:
-	rpc_wake_up_task_queue_locked(queue, task);
 	return task;
 }
 
+static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
+{
+	if (RPC_IS_PRIORITY(queue))
+		return __rpc_find_next_queued_priority(queue);
+	if (!list_empty(&queue->tasks[0]))
+		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
+	return NULL;
+}
+
 /*
- * Wake up the next task on the wait queue.
+ * Wake up the first task on the wait queue.
  */
-struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
+struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
+		bool (*func)(struct rpc_task *, void *), void *data)
 {
 	struct rpc_task	*task = NULL;
 
-	dprintk("RPC:       wake_up_next(%p \"%s\")\n",
+	dprintk("RPC:       wake_up_first(%p \"%s\")\n",
 			queue, rpc_qname(queue));
 	spin_lock_bh(&queue->lock);
-	if (RPC_IS_PRIORITY(queue))
-		task = __rpc_wake_up_next_priority(queue);
-	else {
-		task_for_first(task, &queue->tasks[0])
+	task = __rpc_find_next_queued(queue);
+	if (task != NULL) {
+		if (func(task, data))
 			rpc_wake_up_task_queue_locked(queue, task);
+		else
+			task = NULL;
 	}
 	spin_unlock_bh(&queue->lock);
 
 	return task;
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_first);
+
+static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
+{
+	return true;
+}
+
+/*
+ * Wake up the next task on the wait queue.
+*/
+struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
+{
+	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
+}
 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 
 /**
@@ -473,14 +561,18 @@ EXPORT_SYMBOL_GPL(rpc_wake_up_next);
  */
 void rpc_wake_up(struct rpc_wait_queue *queue)
 {
-	struct rpc_task *task, *next;
 	struct list_head *head;
 
 	spin_lock_bh(&queue->lock);
 	head = &queue->tasks[queue->maxpriority];
 	for (;;) {
-		list_for_each_entry_safe(task, next, head, u.tk_wait.list)
+		while (!list_empty(head)) {
+			struct rpc_task *task;
+			task = list_first_entry(head,
+					struct rpc_task,
+					u.tk_wait.list);
 			rpc_wake_up_task_queue_locked(queue, task);
+		}
 		if (head == &queue->tasks[0])
 			break;
 		head--;
@@ -498,13 +590,16 @@ EXPORT_SYMBOL_GPL(rpc_wake_up);
  */
 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 {
-	struct rpc_task *task, *next;
 	struct list_head *head;
 
 	spin_lock_bh(&queue->lock);
 	head = &queue->tasks[queue->maxpriority];
 	for (;;) {
-		list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
+		while (!list_empty(head)) {
+			struct rpc_task *task;
+			task = list_first_entry(head,
+					struct rpc_task,
+					u.tk_wait.list);
 			task->tk_status = status;
 			rpc_wake_up_task_queue_locked(queue, task);
 		}
@@ -542,7 +637,8 @@ static void __rpc_queue_timer_fn(unsigned long ptr)
 
 static void __rpc_atrun(struct rpc_task *task)
 {
-	task->tk_status = 0;
+	if (task->tk_status == -ETIMEDOUT)
+		task->tk_status = 0;
 }
 
 /*
@@ -563,6 +659,27 @@ void rpc_prepare_task(struct rpc_task *task)
 	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 }
 
+static void
+rpc_init_task_statistics(struct rpc_task *task)
+{
+	/* Initialize retry counters */
+	task->tk_garb_retry = 2;
+	task->tk_cred_retry = 2;
+	task->tk_rebind_retry = 2;
+
+	/* starting timestamp */
+	task->tk_start = ktime_get();
+}
+
+static void
+rpc_reset_task_statistics(struct rpc_task *task)
+{
+	task->tk_timeouts = 0;
+	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
+
+	rpc_init_task_statistics(task);
+}
+
 /*
  * Helper that calls task->tk_ops->rpc_call_done if it exists
  */
@@ -575,6 +692,7 @@ void rpc_exit_task(struct rpc_task *task)
 			WARN_ON(RPC_ASSASSINATED(task));
 			/* Always release the RPC slot and buffer memory */
 			xprt_release(task);
+			rpc_reset_task_statistics(task);
 		}
 	}
 }
@@ -606,35 +724,31 @@ static void __rpc_execute(struct rpc_task *task)
 	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 			task->tk_pid, task->tk_flags);
 
-	BUG_ON(RPC_IS_QUEUED(task));
+	WARN_ON_ONCE(RPC_IS_QUEUED(task));
+	if (RPC_IS_QUEUED(task))
+		return;
 
 	for (;;) {
+		void (*do_action)(struct rpc_task *);
 
 		/*
-		 * Execute any pending callback.
+		 * Execute any pending callback first.
 		 */
-		if (task->tk_callback) {
-			void (*save_callback)(struct rpc_task *);
-
+		do_action = task->tk_callback;
+		task->tk_callback = NULL;
+		if (do_action == NULL) {
 			/*
-			 * We set tk_callback to NULL before calling it,
-			 * in case it sets the tk_callback field itself:
+			 * Perform the next FSM step.
+			 * tk_action may be NULL if the task has been killed.
+			 * In particular, note that rpc_killall_tasks may
+			 * do this at any time, so beware when dereferencing.
 			 */
-			save_callback = task->tk_callback;
-			task->tk_callback = NULL;
-			save_callback(task);
-		}
-
-		/*
-		 * Perform the next FSM step.
-		 * tk_action may be NULL when the task has been killed
-		 * by someone else.
-		 */
-		if (!RPC_IS_QUEUED(task)) {
-			if (task->tk_action == NULL)
+			do_action = task->tk_action;
+			if (do_action == NULL)
 				break;
-			task->tk_action(task);
 		}
+		trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
+		do_action(task);
 
 		/*
 		 * Lockless check for whether task is sleeping or not.
@@ -677,7 +791,6 @@ static void __rpc_execute(struct rpc_task *task)
 			task->tk_flags |= RPC_TASK_KILLED;
 			rpc_exit(task, -ERESTARTSYS);
 		}
-		rpc_set_running(task);
 		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 	}
 
@@ -698,15 +811,19 @@ static void __rpc_execute(struct rpc_task *task)
  */
 void rpc_execute(struct rpc_task *task)
 {
+	bool is_async = RPC_IS_ASYNC(task);
+
 	rpc_set_active(task);
 	rpc_make_runnable(task);
-	if (!RPC_IS_ASYNC(task))
+	if (!is_async)
 		__rpc_execute(task);
 }
 
 static void rpc_async_schedule(struct work_struct *work)
 {
+	current->flags |= PF_FSTRANS;
 	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
+	current->flags &= ~PF_FSTRANS;
 }
 
 /**
@@ -715,7 +832,8 @@ static void rpc_async_schedule(struct work_struct *work)
  * @size: requested byte size
  *
  * To prevent rpciod from hanging, this allocator never sleeps,
- * returning NULL if the request cannot be serviced immediately.
+ * returning NULL and suppressing warning if the request cannot be serviced
+ * immediately.
  * The caller can arrange to sleep in a way that is safe for rpciod.
  *
  * Most requests are 'small' (under 2KiB) and can be serviced from a
@@ -728,7 +846,10 @@ static void rpc_async_schedule(struct work_struct *work)
 void *rpc_malloc(struct rpc_task *task, size_t size)
 {
 	struct rpc_buffer *buf;
-	gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
+	gfp_t gfp = GFP_NOWAIT | __GFP_NOWARN;
+
+	if (RPC_IS_SWAPPER(task))
+		gfp |= __GFP_MEMALLOC;
 
 	size += sizeof(struct rpc_buffer);
 	if (size <= RPC_BUFFER_MAXSIZE)
@@ -784,10 +905,6 @@ static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *ta
 	task->tk_calldata = task_setup_data->callback_data;
 	INIT_LIST_HEAD(&task->tk_task);
 
-	/* Initialize retry counters */
-	task->tk_garb_retry = 2;
-	task->tk_cred_retry = 2;
-
 	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 	task->tk_owner = current->tgid;
 
@@ -797,8 +914,7 @@ static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *ta
 	if (task->tk_ops->rpc_call_prepare != NULL)
 		task->tk_action = rpc_prepare_task;
 
-	/* starting timestamp */
-	task->tk_start = ktime_get();
+	rpc_init_task_statistics(task);
 
 	dprintk("RPC:       new task initialized, procpid %u\n",
 				task_pid_nr(current));
@@ -807,7 +923,7 @@ static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *ta
 static struct rpc_task *
 rpc_alloc_task(void)
 {
-	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
+	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO);
 }
 
 /*
@@ -829,27 +945,40 @@ struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 	}
 
 	rpc_init_task(task, setup_data);
-	if (task->tk_status < 0) {
-		int err = task->tk_status;
-		rpc_put_task(task);
-		return ERR_PTR(err);
-	}
-
 	task->tk_flags |= flags;
 	dprintk("RPC:       allocated task %p\n", task);
 	return task;
 }
 
+/*
+ * rpc_free_task - release rpc task and perform cleanups
+ *
+ * Note that we free up the rpc_task _after_ rpc_release_calldata()
+ * in order to work around a workqueue dependency issue.
+ *
+ * Tejun Heo states:
+ * "Workqueue currently considers two work items to be the same if they're
+ * on the same address and won't execute them concurrently - ie. it
+ * makes a work item which is queued again while being executed wait
+ * for the previous execution to complete.
+ *
+ * If a work function frees the work item, and then waits for an event
+ * which should be performed by another work item and *that* work item
+ * recycles the freed work item, it can create a false dependency loop.
+ * There really is no reliable way to detect this short of verifying
+ * every memory free."
+ *
+ */
 static void rpc_free_task(struct rpc_task *task)
 {
-	const struct rpc_call_ops *tk_ops = task->tk_ops;
-	void *calldata = task->tk_calldata;
+	unsigned short tk_flags = task->tk_flags;
+
+	rpc_release_calldata(task->tk_ops, task->tk_calldata);
 
-	if (task->tk_flags & RPC_TASK_DYNAMIC) {
+	if (tk_flags & RPC_TASK_DYNAMIC) {
 		dprintk("RPC: %5u freeing task\n", task->tk_pid);
 		mempool_free(task, rpc_task_mempool);
 	}
-	rpc_release_calldata(tk_ops, calldata);
 }
 
 static void rpc_async_release(struct work_struct *work)
@@ -857,34 +986,68 @@ static void rpc_async_release(struct work_struct *work)
 	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 }
 
-void rpc_put_task(struct rpc_task *task)
+static void rpc_release_resources_task(struct rpc_task *task)
 {
-	if (!atomic_dec_and_test(&task->tk_count))
-		return;
-	/* Release resources */
-	if (task->tk_rqstp)
-		xprt_release(task);
-	if (task->tk_msg.rpc_cred)
+	xprt_release(task);
+	if (task->tk_msg.rpc_cred) {
 		put_rpccred(task->tk_msg.rpc_cred);
+		task->tk_msg.rpc_cred = NULL;
+	}
 	rpc_task_release_client(task);
-	if (task->tk_workqueue != NULL) {
+}
+
+static void rpc_final_put_task(struct rpc_task *task,
+		struct workqueue_struct *q)
+{
+	if (q != NULL) {
 		INIT_WORK(&task->u.tk_work, rpc_async_release);
-		queue_work(task->tk_workqueue, &task->u.tk_work);
+		queue_work(q, &task->u.tk_work);
 	} else
 		rpc_free_task(task);
 }
+
+static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
+{
+	if (atomic_dec_and_test(&task->tk_count)) {
+		rpc_release_resources_task(task);
+		rpc_final_put_task(task, q);
+	}
+}
+
+void rpc_put_task(struct rpc_task *task)
+{
+	rpc_do_put_task(task, NULL);
+}
 EXPORT_SYMBOL_GPL(rpc_put_task);
 
+void rpc_put_task_async(struct rpc_task *task)
+{
+	rpc_do_put_task(task, task->tk_workqueue);
+}
+EXPORT_SYMBOL_GPL(rpc_put_task_async);
+
 static void rpc_release_task(struct rpc_task *task)
 {
 	dprintk("RPC: %5u release task\n", task->tk_pid);
 
-	BUG_ON (RPC_IS_QUEUED(task));
+	WARN_ON_ONCE(RPC_IS_QUEUED(task));
 
-	/* Wake up anyone who is waiting for task completion */
-	rpc_mark_complete_task(task);
+	rpc_release_resources_task(task);
 
-	rpc_put_task(task);
+	/*
+	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
+	 * so it should be safe to use task->tk_count as a test for whether
+	 * or not any other processes still hold references to our rpc_task.
+	 */
+	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
+		/* Wake up anyone who may be waiting for task completion */
+		if (!rpc_complete_task(task))
+			return;
+	} else {
+		if (!atomic_dec_and_test(&task->tk_count))
+			return;
+	}
+	rpc_final_put_task(task, task->tk_workqueue);
 }
 
 int rpciod_up(void)
@@ -908,7 +1071,7 @@ static int rpciod_start(void)
 	 * Create the rpciod thread and wait for it to start.
 	 */
 	dprintk("RPC:       creating workqueue rpciod\n");
-	wq = alloc_workqueue("rpciod", WQ_RESCUER, 0);
+	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 1);
 	rpciod_workqueue = wq;
 	return rpciod_workqueue != NULL;
 }