xref: /openbmc/linux/net/sunrpc/sched.c (revision 2ae1beb3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/net/sunrpc/sched.c
4  *
5  * Scheduling for synchronous and asynchronous RPC requests.
6  *
7  * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
8  *
9  * TCP NFS related read + write fixes
10  * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/sched.h>
16 #include <linux/interrupt.h>
17 #include <linux/slab.h>
18 #include <linux/mempool.h>
19 #include <linux/smp.h>
20 #include <linux/spinlock.h>
21 #include <linux/mutex.h>
22 #include <linux/freezer.h>
23 #include <linux/sched/mm.h>
24 
25 #include <linux/sunrpc/clnt.h>
26 #include <linux/sunrpc/metrics.h>
27 
28 #include "sunrpc.h"
29 
30 #define CREATE_TRACE_POINTS
31 #include <trace/events/sunrpc.h>
32 
33 /*
34  * RPC slabs and memory pools
35  */
36 #define RPC_BUFFER_MAXSIZE	(2048)
37 #define RPC_BUFFER_POOLSIZE	(8)
38 #define RPC_TASK_POOLSIZE	(8)
39 static struct kmem_cache	*rpc_task_slabp __read_mostly;
40 static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
41 static mempool_t	*rpc_task_mempool __read_mostly;
42 static mempool_t	*rpc_buffer_mempool __read_mostly;
43 
44 static void			rpc_async_schedule(struct work_struct *);
45 static void			 rpc_release_task(struct rpc_task *task);
46 static void __rpc_queue_timer_fn(struct work_struct *);
47 
48 /*
49  * RPC tasks sit here while waiting for conditions to improve.
50  */
51 static struct rpc_wait_queue delay_queue;
52 
53 /*
54  * rpciod-related stuff
55  */
56 struct workqueue_struct *rpciod_workqueue __read_mostly;
57 struct workqueue_struct *xprtiod_workqueue __read_mostly;
58 EXPORT_SYMBOL_GPL(xprtiod_workqueue);
59 
60 gfp_t rpc_task_gfp_mask(void)
61 {
62 	if (current->flags & PF_WQ_WORKER)
63 		return GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
64 	return GFP_KERNEL;
65 }
66 EXPORT_SYMBOL_GPL(rpc_task_gfp_mask);
67 
68 bool rpc_task_set_rpc_status(struct rpc_task *task, int rpc_status)
69 {
70 	if (cmpxchg(&task->tk_rpc_status, 0, rpc_status) == 0)
71 		return true;
72 	return false;
73 }
74 
75 unsigned long
76 rpc_task_timeout(const struct rpc_task *task)
77 {
78 	unsigned long timeout = READ_ONCE(task->tk_timeout);
79 
80 	if (timeout != 0) {
81 		unsigned long now = jiffies;
82 		if (time_before(now, timeout))
83 			return timeout - now;
84 	}
85 	return 0;
86 }
87 EXPORT_SYMBOL_GPL(rpc_task_timeout);
88 
89 /*
90  * Disable the timer for a given RPC task. Should be called with
91  * queue->lock and bh_disabled in order to avoid races within
92  * rpc_run_timer().
93  */
94 static void
95 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
96 {
97 	if (list_empty(&task->u.tk_wait.timer_list))
98 		return;
99 	task->tk_timeout = 0;
100 	list_del(&task->u.tk_wait.timer_list);
101 	if (list_empty(&queue->timer_list.list))
102 		cancel_delayed_work(&queue->timer_list.dwork);
103 }
104 
105 static void
106 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
107 {
108 	unsigned long now = jiffies;
109 	queue->timer_list.expires = expires;
110 	if (time_before_eq(expires, now))
111 		expires = 0;
112 	else
113 		expires -= now;
114 	mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
115 }
116 
117 /*
118  * Set up a timer for the current task.
119  */
120 static void
121 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
122 		unsigned long timeout)
123 {
124 	task->tk_timeout = timeout;
125 	if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
126 		rpc_set_queue_timer(queue, timeout);
127 	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
128 }
129 
130 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
131 {
132 	if (queue->priority != priority) {
133 		queue->priority = priority;
134 		queue->nr = 1U << priority;
135 	}
136 }
137 
138 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
139 {
140 	rpc_set_waitqueue_priority(queue, queue->maxpriority);
141 }
142 
143 /*
144  * Add a request to a queue list
145  */
146 static void
147 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
148 {
149 	struct rpc_task *t;
150 
151 	list_for_each_entry(t, q, u.tk_wait.list) {
152 		if (t->tk_owner == task->tk_owner) {
153 			list_add_tail(&task->u.tk_wait.links,
154 					&t->u.tk_wait.links);
155 			/* Cache the queue head in task->u.tk_wait.list */
156 			task->u.tk_wait.list.next = q;
157 			task->u.tk_wait.list.prev = NULL;
158 			return;
159 		}
160 	}
161 	INIT_LIST_HEAD(&task->u.tk_wait.links);
162 	list_add_tail(&task->u.tk_wait.list, q);
163 }
164 
165 /*
166  * Remove request from a queue list
167  */
168 static void
169 __rpc_list_dequeue_task(struct rpc_task *task)
170 {
171 	struct list_head *q;
172 	struct rpc_task *t;
173 
174 	if (task->u.tk_wait.list.prev == NULL) {
175 		list_del(&task->u.tk_wait.links);
176 		return;
177 	}
178 	if (!list_empty(&task->u.tk_wait.links)) {
179 		t = list_first_entry(&task->u.tk_wait.links,
180 				struct rpc_task,
181 				u.tk_wait.links);
182 		/* Assume __rpc_list_enqueue_task() cached the queue head */
183 		q = t->u.tk_wait.list.next;
184 		list_add_tail(&t->u.tk_wait.list, q);
185 		list_del(&task->u.tk_wait.links);
186 	}
187 	list_del(&task->u.tk_wait.list);
188 }
189 
190 /*
191  * Add new request to a priority queue.
192  */
193 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
194 		struct rpc_task *task,
195 		unsigned char queue_priority)
196 {
197 	if (unlikely(queue_priority > queue->maxpriority))
198 		queue_priority = queue->maxpriority;
199 	__rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
200 }
201 
202 /*
203  * Add new request to wait queue.
204  */
205 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
206 		struct rpc_task *task,
207 		unsigned char queue_priority)
208 {
209 	INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
210 	if (RPC_IS_PRIORITY(queue))
211 		__rpc_add_wait_queue_priority(queue, task, queue_priority);
212 	else
213 		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
214 	task->tk_waitqueue = queue;
215 	queue->qlen++;
216 	/* barrier matches the read in rpc_wake_up_task_queue_locked() */
217 	smp_wmb();
218 	rpc_set_queued(task);
219 }
220 
221 /*
222  * Remove request from a priority queue.
223  */
224 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
225 {
226 	__rpc_list_dequeue_task(task);
227 }
228 
229 /*
230  * Remove request from queue.
231  * Note: must be called with spin lock held.
232  */
233 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
234 {
235 	__rpc_disable_timer(queue, task);
236 	if (RPC_IS_PRIORITY(queue))
237 		__rpc_remove_wait_queue_priority(task);
238 	else
239 		list_del(&task->u.tk_wait.list);
240 	queue->qlen--;
241 }
242 
243 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
244 {
245 	int i;
246 
247 	spin_lock_init(&queue->lock);
248 	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
249 		INIT_LIST_HEAD(&queue->tasks[i]);
250 	queue->maxpriority = nr_queues - 1;
251 	rpc_reset_waitqueue_priority(queue);
252 	queue->qlen = 0;
253 	queue->timer_list.expires = 0;
254 	INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
255 	INIT_LIST_HEAD(&queue->timer_list.list);
256 	rpc_assign_waitqueue_name(queue, qname);
257 }
258 
259 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
260 {
261 	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
262 }
263 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
264 
265 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
266 {
267 	__rpc_init_priority_wait_queue(queue, qname, 1);
268 }
269 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
270 
271 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
272 {
273 	cancel_delayed_work_sync(&queue->timer_list.dwork);
274 }
275 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
276 
277 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
278 {
279 	schedule();
280 	if (signal_pending_state(mode, current))
281 		return -ERESTARTSYS;
282 	return 0;
283 }
284 
285 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
286 static void rpc_task_set_debuginfo(struct rpc_task *task)
287 {
288 	struct rpc_clnt *clnt = task->tk_client;
289 
290 	/* Might be a task carrying a reverse-direction operation */
291 	if (!clnt) {
292 		static atomic_t rpc_pid;
293 
294 		task->tk_pid = atomic_inc_return(&rpc_pid);
295 		return;
296 	}
297 
298 	task->tk_pid = atomic_inc_return(&clnt->cl_pid);
299 }
300 #else
301 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
302 {
303 }
304 #endif
305 
306 static void rpc_set_active(struct rpc_task *task)
307 {
308 	rpc_task_set_debuginfo(task);
309 	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
310 	trace_rpc_task_begin(task, NULL);
311 }
312 
313 /*
314  * Mark an RPC call as having completed by clearing the 'active' bit
315  * and then waking up all tasks that were sleeping.
316  */
317 static int rpc_complete_task(struct rpc_task *task)
318 {
319 	void *m = &task->tk_runstate;
320 	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
321 	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
322 	unsigned long flags;
323 	int ret;
324 
325 	trace_rpc_task_complete(task, NULL);
326 
327 	spin_lock_irqsave(&wq->lock, flags);
328 	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
329 	ret = atomic_dec_and_test(&task->tk_count);
330 	if (waitqueue_active(wq))
331 		__wake_up_locked_key(wq, TASK_NORMAL, &k);
332 	spin_unlock_irqrestore(&wq->lock, flags);
333 	return ret;
334 }
335 
336 /*
337  * Allow callers to wait for completion of an RPC call
338  *
339  * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
340  * to enforce taking of the wq->lock and hence avoid races with
341  * rpc_complete_task().
342  */
343 int rpc_wait_for_completion_task(struct rpc_task *task)
344 {
345 	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
346 			rpc_wait_bit_killable, TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
347 }
348 EXPORT_SYMBOL_GPL(rpc_wait_for_completion_task);
349 
350 /*
351  * Make an RPC task runnable.
352  *
353  * Note: If the task is ASYNC, and is being made runnable after sitting on an
354  * rpc_wait_queue, this must be called with the queue spinlock held to protect
355  * the wait queue operation.
356  * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
357  * which is needed to ensure that __rpc_execute() doesn't loop (due to the
358  * lockless RPC_IS_QUEUED() test) before we've had a chance to test
359  * the RPC_TASK_RUNNING flag.
360  */
361 static void rpc_make_runnable(struct workqueue_struct *wq,
362 		struct rpc_task *task)
363 {
364 	bool need_wakeup = !rpc_test_and_set_running(task);
365 
366 	rpc_clear_queued(task);
367 	if (!need_wakeup)
368 		return;
369 	if (RPC_IS_ASYNC(task)) {
370 		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
371 		queue_work(wq, &task->u.tk_work);
372 	} else {
373 		smp_mb__after_atomic();
374 		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
375 	}
376 }
377 
378 /*
379  * Prepare for sleeping on a wait queue.
380  * By always appending tasks to the list we ensure FIFO behavior.
381  * NB: An RPC task will only receive interrupt-driven events as long
382  * as it's on a wait queue.
383  */
384 static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
385 		struct rpc_task *task,
386 		unsigned char queue_priority)
387 {
388 	trace_rpc_task_sleep(task, q);
389 
390 	__rpc_add_wait_queue(q, task, queue_priority);
391 }
392 
393 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
394 		struct rpc_task *task,
395 		unsigned char queue_priority)
396 {
397 	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
398 		return;
399 	__rpc_do_sleep_on_priority(q, task, queue_priority);
400 }
401 
402 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
403 		struct rpc_task *task, unsigned long timeout,
404 		unsigned char queue_priority)
405 {
406 	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
407 		return;
408 	if (time_is_after_jiffies(timeout)) {
409 		__rpc_do_sleep_on_priority(q, task, queue_priority);
410 		__rpc_add_timer(q, task, timeout);
411 	} else
412 		task->tk_status = -ETIMEDOUT;
413 }
414 
415 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
416 {
417 	if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
418 		task->tk_callback = action;
419 }
420 
421 static bool rpc_sleep_check_activated(struct rpc_task *task)
422 {
423 	/* We shouldn't ever put an inactive task to sleep */
424 	if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
425 		task->tk_status = -EIO;
426 		rpc_put_task_async(task);
427 		return false;
428 	}
429 	return true;
430 }
431 
432 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
433 				rpc_action action, unsigned long timeout)
434 {
435 	if (!rpc_sleep_check_activated(task))
436 		return;
437 
438 	rpc_set_tk_callback(task, action);
439 
440 	/*
441 	 * Protect the queue operations.
442 	 */
443 	spin_lock(&q->lock);
444 	__rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
445 	spin_unlock(&q->lock);
446 }
447 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
448 
449 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
450 				rpc_action action)
451 {
452 	if (!rpc_sleep_check_activated(task))
453 		return;
454 
455 	rpc_set_tk_callback(task, action);
456 
457 	WARN_ON_ONCE(task->tk_timeout != 0);
458 	/*
459 	 * Protect the queue operations.
460 	 */
461 	spin_lock(&q->lock);
462 	__rpc_sleep_on_priority(q, task, task->tk_priority);
463 	spin_unlock(&q->lock);
464 }
465 EXPORT_SYMBOL_GPL(rpc_sleep_on);
466 
467 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
468 		struct rpc_task *task, unsigned long timeout, int priority)
469 {
470 	if (!rpc_sleep_check_activated(task))
471 		return;
472 
473 	priority -= RPC_PRIORITY_LOW;
474 	/*
475 	 * Protect the queue operations.
476 	 */
477 	spin_lock(&q->lock);
478 	__rpc_sleep_on_priority_timeout(q, task, timeout, priority);
479 	spin_unlock(&q->lock);
480 }
481 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
482 
483 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
484 		int priority)
485 {
486 	if (!rpc_sleep_check_activated(task))
487 		return;
488 
489 	WARN_ON_ONCE(task->tk_timeout != 0);
490 	priority -= RPC_PRIORITY_LOW;
491 	/*
492 	 * Protect the queue operations.
493 	 */
494 	spin_lock(&q->lock);
495 	__rpc_sleep_on_priority(q, task, priority);
496 	spin_unlock(&q->lock);
497 }
498 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
499 
500 /**
501  * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
502  * @wq: workqueue on which to run task
503  * @queue: wait queue
504  * @task: task to be woken up
505  *
506  * Caller must hold queue->lock, and have cleared the task queued flag.
507  */
508 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
509 		struct rpc_wait_queue *queue,
510 		struct rpc_task *task)
511 {
512 	/* Has the task been executed yet? If not, we cannot wake it up! */
513 	if (!RPC_IS_ACTIVATED(task)) {
514 		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
515 		return;
516 	}
517 
518 	trace_rpc_task_wakeup(task, queue);
519 
520 	__rpc_remove_wait_queue(queue, task);
521 
522 	rpc_make_runnable(wq, task);
523 }
524 
525 /*
526  * Wake up a queued task while the queue lock is being held
527  */
528 static struct rpc_task *
529 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
530 		struct rpc_wait_queue *queue, struct rpc_task *task,
531 		bool (*action)(struct rpc_task *, void *), void *data)
532 {
533 	if (RPC_IS_QUEUED(task)) {
534 		smp_rmb();
535 		if (task->tk_waitqueue == queue) {
536 			if (action == NULL || action(task, data)) {
537 				__rpc_do_wake_up_task_on_wq(wq, queue, task);
538 				return task;
539 			}
540 		}
541 	}
542 	return NULL;
543 }
544 
545 /*
546  * Wake up a queued task while the queue lock is being held
547  */
548 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
549 					  struct rpc_task *task)
550 {
551 	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
552 						   task, NULL, NULL);
553 }
554 
555 /*
556  * Wake up a task on a specific queue
557  */
558 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
559 {
560 	if (!RPC_IS_QUEUED(task))
561 		return;
562 	spin_lock(&queue->lock);
563 	rpc_wake_up_task_queue_locked(queue, task);
564 	spin_unlock(&queue->lock);
565 }
566 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
567 
568 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
569 {
570 	task->tk_status = *(int *)status;
571 	return true;
572 }
573 
574 static void
575 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
576 		struct rpc_task *task, int status)
577 {
578 	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
579 			task, rpc_task_action_set_status, &status);
580 }
581 
582 /**
583  * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
584  * @queue: pointer to rpc_wait_queue
585  * @task: pointer to rpc_task
586  * @status: integer error value
587  *
588  * If @task is queued on @queue, then it is woken up, and @task->tk_status is
589  * set to the value of @status.
590  */
591 void
592 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
593 		struct rpc_task *task, int status)
594 {
595 	if (!RPC_IS_QUEUED(task))
596 		return;
597 	spin_lock(&queue->lock);
598 	rpc_wake_up_task_queue_set_status_locked(queue, task, status);
599 	spin_unlock(&queue->lock);
600 }
601 
602 /*
603  * Wake up the next task on a priority queue.
604  */
605 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
606 {
607 	struct list_head *q;
608 	struct rpc_task *task;
609 
610 	/*
611 	 * Service the privileged queue.
612 	 */
613 	q = &queue->tasks[RPC_NR_PRIORITY - 1];
614 	if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
615 		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
616 		goto out;
617 	}
618 
619 	/*
620 	 * Service a batch of tasks from a single owner.
621 	 */
622 	q = &queue->tasks[queue->priority];
623 	if (!list_empty(q) && queue->nr) {
624 		queue->nr--;
625 		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
626 		goto out;
627 	}
628 
629 	/*
630 	 * Service the next queue.
631 	 */
632 	do {
633 		if (q == &queue->tasks[0])
634 			q = &queue->tasks[queue->maxpriority];
635 		else
636 			q = q - 1;
637 		if (!list_empty(q)) {
638 			task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
639 			goto new_queue;
640 		}
641 	} while (q != &queue->tasks[queue->priority]);
642 
643 	rpc_reset_waitqueue_priority(queue);
644 	return NULL;
645 
646 new_queue:
647 	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
648 out:
649 	return task;
650 }
651 
652 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
653 {
654 	if (RPC_IS_PRIORITY(queue))
655 		return __rpc_find_next_queued_priority(queue);
656 	if (!list_empty(&queue->tasks[0]))
657 		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
658 	return NULL;
659 }
660 
661 /*
662  * Wake up the first task on the wait queue.
663  */
664 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
665 		struct rpc_wait_queue *queue,
666 		bool (*func)(struct rpc_task *, void *), void *data)
667 {
668 	struct rpc_task	*task = NULL;
669 
670 	spin_lock(&queue->lock);
671 	task = __rpc_find_next_queued(queue);
672 	if (task != NULL)
673 		task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
674 				task, func, data);
675 	spin_unlock(&queue->lock);
676 
677 	return task;
678 }
679 
680 /*
681  * Wake up the first task on the wait queue.
682  */
683 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
684 		bool (*func)(struct rpc_task *, void *), void *data)
685 {
686 	return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
687 }
688 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
689 
690 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
691 {
692 	return true;
693 }
694 
695 /*
696  * Wake up the next task on the wait queue.
697 */
698 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
699 {
700 	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
701 }
702 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
703 
704 /**
705  * rpc_wake_up_locked - wake up all rpc_tasks
706  * @queue: rpc_wait_queue on which the tasks are sleeping
707  *
708  */
709 static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
710 {
711 	struct rpc_task *task;
712 
713 	for (;;) {
714 		task = __rpc_find_next_queued(queue);
715 		if (task == NULL)
716 			break;
717 		rpc_wake_up_task_queue_locked(queue, task);
718 	}
719 }
720 
721 /**
722  * rpc_wake_up - wake up all rpc_tasks
723  * @queue: rpc_wait_queue on which the tasks are sleeping
724  *
725  * Grabs queue->lock
726  */
727 void rpc_wake_up(struct rpc_wait_queue *queue)
728 {
729 	spin_lock(&queue->lock);
730 	rpc_wake_up_locked(queue);
731 	spin_unlock(&queue->lock);
732 }
733 EXPORT_SYMBOL_GPL(rpc_wake_up);
734 
735 /**
736  * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
737  * @queue: rpc_wait_queue on which the tasks are sleeping
738  * @status: status value to set
739  */
740 static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
741 {
742 	struct rpc_task *task;
743 
744 	for (;;) {
745 		task = __rpc_find_next_queued(queue);
746 		if (task == NULL)
747 			break;
748 		rpc_wake_up_task_queue_set_status_locked(queue, task, status);
749 	}
750 }
751 
752 /**
753  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
754  * @queue: rpc_wait_queue on which the tasks are sleeping
755  * @status: status value to set
756  *
757  * Grabs queue->lock
758  */
759 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
760 {
761 	spin_lock(&queue->lock);
762 	rpc_wake_up_status_locked(queue, status);
763 	spin_unlock(&queue->lock);
764 }
765 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
766 
767 static void __rpc_queue_timer_fn(struct work_struct *work)
768 {
769 	struct rpc_wait_queue *queue = container_of(work,
770 			struct rpc_wait_queue,
771 			timer_list.dwork.work);
772 	struct rpc_task *task, *n;
773 	unsigned long expires, now, timeo;
774 
775 	spin_lock(&queue->lock);
776 	expires = now = jiffies;
777 	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
778 		timeo = task->tk_timeout;
779 		if (time_after_eq(now, timeo)) {
780 			trace_rpc_task_timeout(task, task->tk_action);
781 			task->tk_status = -ETIMEDOUT;
782 			rpc_wake_up_task_queue_locked(queue, task);
783 			continue;
784 		}
785 		if (expires == now || time_after(expires, timeo))
786 			expires = timeo;
787 	}
788 	if (!list_empty(&queue->timer_list.list))
789 		rpc_set_queue_timer(queue, expires);
790 	spin_unlock(&queue->lock);
791 }
792 
793 static void __rpc_atrun(struct rpc_task *task)
794 {
795 	if (task->tk_status == -ETIMEDOUT)
796 		task->tk_status = 0;
797 }
798 
799 /*
800  * Run a task at a later time
801  */
802 void rpc_delay(struct rpc_task *task, unsigned long delay)
803 {
804 	rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
805 }
806 EXPORT_SYMBOL_GPL(rpc_delay);
807 
808 /*
809  * Helper to call task->tk_ops->rpc_call_prepare
810  */
811 void rpc_prepare_task(struct rpc_task *task)
812 {
813 	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
814 }
815 
816 static void
817 rpc_init_task_statistics(struct rpc_task *task)
818 {
819 	/* Initialize retry counters */
820 	task->tk_garb_retry = 2;
821 	task->tk_cred_retry = 2;
822 
823 	/* starting timestamp */
824 	task->tk_start = ktime_get();
825 }
826 
827 static void
828 rpc_reset_task_statistics(struct rpc_task *task)
829 {
830 	task->tk_timeouts = 0;
831 	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
832 	rpc_init_task_statistics(task);
833 }
834 
835 /*
836  * Helper that calls task->tk_ops->rpc_call_done if it exists
837  */
838 void rpc_exit_task(struct rpc_task *task)
839 {
840 	trace_rpc_task_end(task, task->tk_action);
841 	task->tk_action = NULL;
842 	if (task->tk_ops->rpc_count_stats)
843 		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
844 	else if (task->tk_client)
845 		rpc_count_iostats(task, task->tk_client->cl_metrics);
846 	if (task->tk_ops->rpc_call_done != NULL) {
847 		trace_rpc_task_call_done(task, task->tk_ops->rpc_call_done);
848 		task->tk_ops->rpc_call_done(task, task->tk_calldata);
849 		if (task->tk_action != NULL) {
850 			/* Always release the RPC slot and buffer memory */
851 			xprt_release(task);
852 			rpc_reset_task_statistics(task);
853 		}
854 	}
855 }
856 
857 void rpc_signal_task(struct rpc_task *task)
858 {
859 	struct rpc_wait_queue *queue;
860 
861 	if (!RPC_IS_ACTIVATED(task))
862 		return;
863 
864 	if (!rpc_task_set_rpc_status(task, -ERESTARTSYS))
865 		return;
866 	trace_rpc_task_signalled(task, task->tk_action);
867 	set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
868 	smp_mb__after_atomic();
869 	queue = READ_ONCE(task->tk_waitqueue);
870 	if (queue)
871 		rpc_wake_up_queued_task(queue, task);
872 }
873 
874 void rpc_task_try_cancel(struct rpc_task *task, int error)
875 {
876 	struct rpc_wait_queue *queue;
877 
878 	if (!rpc_task_set_rpc_status(task, error))
879 		return;
880 	queue = READ_ONCE(task->tk_waitqueue);
881 	if (queue)
882 		rpc_wake_up_queued_task(queue, task);
883 }
884 
885 void rpc_exit(struct rpc_task *task, int status)
886 {
887 	task->tk_status = status;
888 	task->tk_action = rpc_exit_task;
889 	rpc_wake_up_queued_task(task->tk_waitqueue, task);
890 }
891 EXPORT_SYMBOL_GPL(rpc_exit);
892 
893 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
894 {
895 	if (ops->rpc_release != NULL)
896 		ops->rpc_release(calldata);
897 }
898 
899 static bool xprt_needs_memalloc(struct rpc_xprt *xprt, struct rpc_task *tk)
900 {
901 	if (!xprt)
902 		return false;
903 	if (!atomic_read(&xprt->swapper))
904 		return false;
905 	return test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == tk;
906 }
907 
908 /*
909  * This is the RPC `scheduler' (or rather, the finite state machine).
910  */
911 static void __rpc_execute(struct rpc_task *task)
912 {
913 	struct rpc_wait_queue *queue;
914 	int task_is_async = RPC_IS_ASYNC(task);
915 	int status = 0;
916 	unsigned long pflags = current->flags;
917 
918 	WARN_ON_ONCE(RPC_IS_QUEUED(task));
919 	if (RPC_IS_QUEUED(task))
920 		return;
921 
922 	for (;;) {
923 		void (*do_action)(struct rpc_task *);
924 
925 		/*
926 		 * Perform the next FSM step or a pending callback.
927 		 *
928 		 * tk_action may be NULL if the task has been killed.
929 		 */
930 		do_action = task->tk_action;
931 		/* Tasks with an RPC error status should exit */
932 		if (do_action && do_action != rpc_exit_task &&
933 		    (status = READ_ONCE(task->tk_rpc_status)) != 0) {
934 			task->tk_status = status;
935 			do_action = rpc_exit_task;
936 		}
937 		/* Callbacks override all actions */
938 		if (task->tk_callback) {
939 			do_action = task->tk_callback;
940 			task->tk_callback = NULL;
941 		}
942 		if (!do_action)
943 			break;
944 		if (RPC_IS_SWAPPER(task) ||
945 		    xprt_needs_memalloc(task->tk_xprt, task))
946 			current->flags |= PF_MEMALLOC;
947 
948 		trace_rpc_task_run_action(task, do_action);
949 		do_action(task);
950 
951 		/*
952 		 * Lockless check for whether task is sleeping or not.
953 		 */
954 		if (!RPC_IS_QUEUED(task)) {
955 			cond_resched();
956 			continue;
957 		}
958 
959 		/*
960 		 * The queue->lock protects against races with
961 		 * rpc_make_runnable().
962 		 *
963 		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
964 		 * rpc_task, rpc_make_runnable() can assign it to a
965 		 * different workqueue. We therefore cannot assume that the
966 		 * rpc_task pointer may still be dereferenced.
967 		 */
968 		queue = task->tk_waitqueue;
969 		spin_lock(&queue->lock);
970 		if (!RPC_IS_QUEUED(task)) {
971 			spin_unlock(&queue->lock);
972 			continue;
973 		}
974 		/* Wake up any task that has an exit status */
975 		if (READ_ONCE(task->tk_rpc_status) != 0) {
976 			rpc_wake_up_task_queue_locked(queue, task);
977 			spin_unlock(&queue->lock);
978 			continue;
979 		}
980 		rpc_clear_running(task);
981 		spin_unlock(&queue->lock);
982 		if (task_is_async)
983 			goto out;
984 
985 		/* sync task: sleep here */
986 		trace_rpc_task_sync_sleep(task, task->tk_action);
987 		status = out_of_line_wait_on_bit(&task->tk_runstate,
988 				RPC_TASK_QUEUED, rpc_wait_bit_killable,
989 				TASK_KILLABLE|TASK_FREEZABLE);
990 		if (status < 0) {
991 			/*
992 			 * When a sync task receives a signal, it exits with
993 			 * -ERESTARTSYS. In order to catch any callbacks that
994 			 * clean up after sleeping on some queue, we don't
995 			 * break the loop here, but go around once more.
996 			 */
997 			rpc_signal_task(task);
998 		}
999 		trace_rpc_task_sync_wake(task, task->tk_action);
1000 	}
1001 
1002 	/* Release all resources associated with the task */
1003 	rpc_release_task(task);
1004 out:
1005 	current_restore_flags(pflags, PF_MEMALLOC);
1006 }
1007 
1008 /*
1009  * User-visible entry point to the scheduler.
1010  *
1011  * This may be called recursively if e.g. an async NFS task updates
1012  * the attributes and finds that dirty pages must be flushed.
1013  * NOTE: Upon exit of this function the task is guaranteed to be
1014  *	 released. In particular note that tk_release() will have
1015  *	 been called, so your task memory may have been freed.
1016  */
1017 void rpc_execute(struct rpc_task *task)
1018 {
1019 	bool is_async = RPC_IS_ASYNC(task);
1020 
1021 	rpc_set_active(task);
1022 	rpc_make_runnable(rpciod_workqueue, task);
1023 	if (!is_async) {
1024 		unsigned int pflags = memalloc_nofs_save();
1025 		__rpc_execute(task);
1026 		memalloc_nofs_restore(pflags);
1027 	}
1028 }
1029 
1030 static void rpc_async_schedule(struct work_struct *work)
1031 {
1032 	unsigned int pflags = memalloc_nofs_save();
1033 
1034 	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
1035 	memalloc_nofs_restore(pflags);
1036 }
1037 
1038 /**
1039  * rpc_malloc - allocate RPC buffer resources
1040  * @task: RPC task
1041  *
1042  * A single memory region is allocated, which is split between the
1043  * RPC call and RPC reply that this task is being used for. When
1044  * this RPC is retired, the memory is released by calling rpc_free.
1045  *
1046  * To prevent rpciod from hanging, this allocator never sleeps,
1047  * returning -ENOMEM and suppressing warning if the request cannot
1048  * be serviced immediately. The caller can arrange to sleep in a
1049  * way that is safe for rpciod.
1050  *
1051  * Most requests are 'small' (under 2KiB) and can be serviced from a
1052  * mempool, ensuring that NFS reads and writes can always proceed,
1053  * and that there is good locality of reference for these buffers.
1054  */
1055 int rpc_malloc(struct rpc_task *task)
1056 {
1057 	struct rpc_rqst *rqst = task->tk_rqstp;
1058 	size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1059 	struct rpc_buffer *buf;
1060 	gfp_t gfp = rpc_task_gfp_mask();
1061 
1062 	size += sizeof(struct rpc_buffer);
1063 	if (size <= RPC_BUFFER_MAXSIZE) {
1064 		buf = kmem_cache_alloc(rpc_buffer_slabp, gfp);
1065 		/* Reach for the mempool if dynamic allocation fails */
1066 		if (!buf && RPC_IS_ASYNC(task))
1067 			buf = mempool_alloc(rpc_buffer_mempool, GFP_NOWAIT);
1068 	} else
1069 		buf = kmalloc(size, gfp);
1070 
1071 	if (!buf)
1072 		return -ENOMEM;
1073 
1074 	buf->len = size;
1075 	rqst->rq_buffer = buf->data;
1076 	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1077 	return 0;
1078 }
1079 EXPORT_SYMBOL_GPL(rpc_malloc);
1080 
1081 /**
1082  * rpc_free - free RPC buffer resources allocated via rpc_malloc
1083  * @task: RPC task
1084  *
1085  */
1086 void rpc_free(struct rpc_task *task)
1087 {
1088 	void *buffer = task->tk_rqstp->rq_buffer;
1089 	size_t size;
1090 	struct rpc_buffer *buf;
1091 
1092 	buf = container_of(buffer, struct rpc_buffer, data);
1093 	size = buf->len;
1094 
1095 	if (size <= RPC_BUFFER_MAXSIZE)
1096 		mempool_free(buf, rpc_buffer_mempool);
1097 	else
1098 		kfree(buf);
1099 }
1100 EXPORT_SYMBOL_GPL(rpc_free);
1101 
1102 /*
1103  * Creation and deletion of RPC task structures
1104  */
1105 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1106 {
1107 	memset(task, 0, sizeof(*task));
1108 	atomic_set(&task->tk_count, 1);
1109 	task->tk_flags  = task_setup_data->flags;
1110 	task->tk_ops = task_setup_data->callback_ops;
1111 	task->tk_calldata = task_setup_data->callback_data;
1112 	INIT_LIST_HEAD(&task->tk_task);
1113 
1114 	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1115 	task->tk_owner = current->tgid;
1116 
1117 	/* Initialize workqueue for async tasks */
1118 	task->tk_workqueue = task_setup_data->workqueue;
1119 
1120 	task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1121 			xprt_get(task_setup_data->rpc_xprt));
1122 
1123 	task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1124 
1125 	if (task->tk_ops->rpc_call_prepare != NULL)
1126 		task->tk_action = rpc_prepare_task;
1127 
1128 	rpc_init_task_statistics(task);
1129 }
1130 
1131 static struct rpc_task *rpc_alloc_task(void)
1132 {
1133 	struct rpc_task *task;
1134 
1135 	task = kmem_cache_alloc(rpc_task_slabp, rpc_task_gfp_mask());
1136 	if (task)
1137 		return task;
1138 	return mempool_alloc(rpc_task_mempool, GFP_NOWAIT);
1139 }
1140 
1141 /*
1142  * Create a new task for the specified client.
1143  */
1144 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1145 {
1146 	struct rpc_task	*task = setup_data->task;
1147 	unsigned short flags = 0;
1148 
1149 	if (task == NULL) {
1150 		task = rpc_alloc_task();
1151 		if (task == NULL) {
1152 			rpc_release_calldata(setup_data->callback_ops,
1153 					     setup_data->callback_data);
1154 			return ERR_PTR(-ENOMEM);
1155 		}
1156 		flags = RPC_TASK_DYNAMIC;
1157 	}
1158 
1159 	rpc_init_task(task, setup_data);
1160 	task->tk_flags |= flags;
1161 	return task;
1162 }
1163 
1164 /*
1165  * rpc_free_task - release rpc task and perform cleanups
1166  *
1167  * Note that we free up the rpc_task _after_ rpc_release_calldata()
1168  * in order to work around a workqueue dependency issue.
1169  *
1170  * Tejun Heo states:
1171  * "Workqueue currently considers two work items to be the same if they're
1172  * on the same address and won't execute them concurrently - ie. it
1173  * makes a work item which is queued again while being executed wait
1174  * for the previous execution to complete.
1175  *
1176  * If a work function frees the work item, and then waits for an event
1177  * which should be performed by another work item and *that* work item
1178  * recycles the freed work item, it can create a false dependency loop.
1179  * There really is no reliable way to detect this short of verifying
1180  * every memory free."
1181  *
1182  */
1183 static void rpc_free_task(struct rpc_task *task)
1184 {
1185 	unsigned short tk_flags = task->tk_flags;
1186 
1187 	put_rpccred(task->tk_op_cred);
1188 	rpc_release_calldata(task->tk_ops, task->tk_calldata);
1189 
1190 	if (tk_flags & RPC_TASK_DYNAMIC)
1191 		mempool_free(task, rpc_task_mempool);
1192 }
1193 
1194 static void rpc_async_release(struct work_struct *work)
1195 {
1196 	unsigned int pflags = memalloc_nofs_save();
1197 
1198 	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1199 	memalloc_nofs_restore(pflags);
1200 }
1201 
1202 static void rpc_release_resources_task(struct rpc_task *task)
1203 {
1204 	xprt_release(task);
1205 	if (task->tk_msg.rpc_cred) {
1206 		if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1207 			put_cred(task->tk_msg.rpc_cred);
1208 		task->tk_msg.rpc_cred = NULL;
1209 	}
1210 	rpc_task_release_client(task);
1211 }
1212 
1213 static void rpc_final_put_task(struct rpc_task *task,
1214 		struct workqueue_struct *q)
1215 {
1216 	if (q != NULL) {
1217 		INIT_WORK(&task->u.tk_work, rpc_async_release);
1218 		queue_work(q, &task->u.tk_work);
1219 	} else
1220 		rpc_free_task(task);
1221 }
1222 
1223 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1224 {
1225 	if (atomic_dec_and_test(&task->tk_count)) {
1226 		rpc_release_resources_task(task);
1227 		rpc_final_put_task(task, q);
1228 	}
1229 }
1230 
1231 void rpc_put_task(struct rpc_task *task)
1232 {
1233 	rpc_do_put_task(task, NULL);
1234 }
1235 EXPORT_SYMBOL_GPL(rpc_put_task);
1236 
1237 void rpc_put_task_async(struct rpc_task *task)
1238 {
1239 	rpc_do_put_task(task, task->tk_workqueue);
1240 }
1241 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1242 
1243 static void rpc_release_task(struct rpc_task *task)
1244 {
1245 	WARN_ON_ONCE(RPC_IS_QUEUED(task));
1246 
1247 	rpc_release_resources_task(task);
1248 
1249 	/*
1250 	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1251 	 * so it should be safe to use task->tk_count as a test for whether
1252 	 * or not any other processes still hold references to our rpc_task.
1253 	 */
1254 	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1255 		/* Wake up anyone who may be waiting for task completion */
1256 		if (!rpc_complete_task(task))
1257 			return;
1258 	} else {
1259 		if (!atomic_dec_and_test(&task->tk_count))
1260 			return;
1261 	}
1262 	rpc_final_put_task(task, task->tk_workqueue);
1263 }
1264 
1265 int rpciod_up(void)
1266 {
1267 	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1268 }
1269 
1270 void rpciod_down(void)
1271 {
1272 	module_put(THIS_MODULE);
1273 }
1274 
1275 /*
1276  * Start up the rpciod workqueue.
1277  */
1278 static int rpciod_start(void)
1279 {
1280 	struct workqueue_struct *wq;
1281 
1282 	/*
1283 	 * Create the rpciod thread and wait for it to start.
1284 	 */
1285 	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1286 	if (!wq)
1287 		goto out_failed;
1288 	rpciod_workqueue = wq;
1289 	wq = alloc_workqueue("xprtiod", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
1290 	if (!wq)
1291 		goto free_rpciod;
1292 	xprtiod_workqueue = wq;
1293 	return 1;
1294 free_rpciod:
1295 	wq = rpciod_workqueue;
1296 	rpciod_workqueue = NULL;
1297 	destroy_workqueue(wq);
1298 out_failed:
1299 	return 0;
1300 }
1301 
1302 static void rpciod_stop(void)
1303 {
1304 	struct workqueue_struct *wq = NULL;
1305 
1306 	if (rpciod_workqueue == NULL)
1307 		return;
1308 
1309 	wq = rpciod_workqueue;
1310 	rpciod_workqueue = NULL;
1311 	destroy_workqueue(wq);
1312 	wq = xprtiod_workqueue;
1313 	xprtiod_workqueue = NULL;
1314 	destroy_workqueue(wq);
1315 }
1316 
1317 void
1318 rpc_destroy_mempool(void)
1319 {
1320 	rpciod_stop();
1321 	mempool_destroy(rpc_buffer_mempool);
1322 	mempool_destroy(rpc_task_mempool);
1323 	kmem_cache_destroy(rpc_task_slabp);
1324 	kmem_cache_destroy(rpc_buffer_slabp);
1325 	rpc_destroy_wait_queue(&delay_queue);
1326 }
1327 
1328 int
1329 rpc_init_mempool(void)
1330 {
1331 	/*
1332 	 * The following is not strictly a mempool initialisation,
1333 	 * but there is no harm in doing it here
1334 	 */
1335 	rpc_init_wait_queue(&delay_queue, "delayq");
1336 	if (!rpciod_start())
1337 		goto err_nomem;
1338 
1339 	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1340 					     sizeof(struct rpc_task),
1341 					     0, SLAB_HWCACHE_ALIGN,
1342 					     NULL);
1343 	if (!rpc_task_slabp)
1344 		goto err_nomem;
1345 	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1346 					     RPC_BUFFER_MAXSIZE,
1347 					     0, SLAB_HWCACHE_ALIGN,
1348 					     NULL);
1349 	if (!rpc_buffer_slabp)
1350 		goto err_nomem;
1351 	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1352 						    rpc_task_slabp);
1353 	if (!rpc_task_mempool)
1354 		goto err_nomem;
1355 	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1356 						      rpc_buffer_slabp);
1357 	if (!rpc_buffer_mempool)
1358 		goto err_nomem;
1359 	return 0;
1360 err_nomem:
1361 	rpc_destroy_mempool();
1362 	return -ENOMEM;
1363 }
1364