xref: /openbmc/linux/net/sunrpc/sched.c (revision e8e0929d)
1 /*
2  * linux/net/sunrpc/sched.c
3  *
4  * Scheduling for synchronous and asynchronous RPC requests.
5  *
6  * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
7  *
8  * TCP NFS related read + write fixes
9  * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  */
11 
12 #include <linux/module.h>
13 
14 #include <linux/sched.h>
15 #include <linux/interrupt.h>
16 #include <linux/slab.h>
17 #include <linux/mempool.h>
18 #include <linux/smp.h>
19 #include <linux/spinlock.h>
20 #include <linux/mutex.h>
21 
22 #include <linux/sunrpc/clnt.h>
23 
24 #include "sunrpc.h"
25 
26 #ifdef RPC_DEBUG
27 #define RPCDBG_FACILITY		RPCDBG_SCHED
28 #define RPC_TASK_MAGIC_ID	0xf00baa
29 #endif
30 
31 /*
32  * RPC slabs and memory pools
33  */
34 #define RPC_BUFFER_MAXSIZE	(2048)
35 #define RPC_BUFFER_POOLSIZE	(8)
36 #define RPC_TASK_POOLSIZE	(8)
37 static struct kmem_cache	*rpc_task_slabp __read_mostly;
38 static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
39 static mempool_t	*rpc_task_mempool __read_mostly;
40 static mempool_t	*rpc_buffer_mempool __read_mostly;
41 
42 static void			rpc_async_schedule(struct work_struct *);
43 static void			 rpc_release_task(struct rpc_task *task);
44 static void __rpc_queue_timer_fn(unsigned long ptr);
45 
46 /*
47  * RPC tasks sit here while waiting for conditions to improve.
48  */
49 static struct rpc_wait_queue delay_queue;
50 
51 /*
52  * rpciod-related stuff
53  */
54 struct workqueue_struct *rpciod_workqueue;
55 
56 /*
57  * Disable the timer for a given RPC task. Should be called with
58  * queue->lock and bh_disabled in order to avoid races within
59  * rpc_run_timer().
60  */
61 static void
62 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
63 {
64 	if (task->tk_timeout == 0)
65 		return;
66 	dprintk("RPC: %5u disabling timer\n", task->tk_pid);
67 	task->tk_timeout = 0;
68 	list_del(&task->u.tk_wait.timer_list);
69 	if (list_empty(&queue->timer_list.list))
70 		del_timer(&queue->timer_list.timer);
71 }
72 
73 static void
74 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
75 {
76 	queue->timer_list.expires = expires;
77 	mod_timer(&queue->timer_list.timer, expires);
78 }
79 
80 /*
81  * Set up a timer for the current task.
82  */
83 static void
84 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
85 {
86 	if (!task->tk_timeout)
87 		return;
88 
89 	dprintk("RPC: %5u setting alarm for %lu ms\n",
90 			task->tk_pid, task->tk_timeout * 1000 / HZ);
91 
92 	task->u.tk_wait.expires = jiffies + task->tk_timeout;
93 	if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
94 		rpc_set_queue_timer(queue, task->u.tk_wait.expires);
95 	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
96 }
97 
98 /*
99  * Add new request to a priority queue.
100  */
101 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
102 {
103 	struct list_head *q;
104 	struct rpc_task *t;
105 
106 	INIT_LIST_HEAD(&task->u.tk_wait.links);
107 	q = &queue->tasks[task->tk_priority];
108 	if (unlikely(task->tk_priority > queue->maxpriority))
109 		q = &queue->tasks[queue->maxpriority];
110 	list_for_each_entry(t, q, u.tk_wait.list) {
111 		if (t->tk_owner == task->tk_owner) {
112 			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
113 			return;
114 		}
115 	}
116 	list_add_tail(&task->u.tk_wait.list, q);
117 }
118 
119 /*
120  * Add new request to wait queue.
121  *
122  * Swapper tasks always get inserted at the head of the queue.
123  * This should avoid many nasty memory deadlocks and hopefully
124  * improve overall performance.
125  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
126  */
127 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
128 {
129 	BUG_ON (RPC_IS_QUEUED(task));
130 
131 	if (RPC_IS_PRIORITY(queue))
132 		__rpc_add_wait_queue_priority(queue, task);
133 	else if (RPC_IS_SWAPPER(task))
134 		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
135 	else
136 		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
137 	task->tk_waitqueue = queue;
138 	queue->qlen++;
139 	rpc_set_queued(task);
140 
141 	dprintk("RPC: %5u added to queue %p \"%s\"\n",
142 			task->tk_pid, queue, rpc_qname(queue));
143 }
144 
145 /*
146  * Remove request from a priority queue.
147  */
148 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
149 {
150 	struct rpc_task *t;
151 
152 	if (!list_empty(&task->u.tk_wait.links)) {
153 		t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
154 		list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
155 		list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
156 	}
157 }
158 
159 /*
160  * Remove request from queue.
161  * Note: must be called with spin lock held.
162  */
163 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
164 {
165 	__rpc_disable_timer(queue, task);
166 	if (RPC_IS_PRIORITY(queue))
167 		__rpc_remove_wait_queue_priority(task);
168 	list_del(&task->u.tk_wait.list);
169 	queue->qlen--;
170 	dprintk("RPC: %5u removed from queue %p \"%s\"\n",
171 			task->tk_pid, queue, rpc_qname(queue));
172 }
173 
174 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
175 {
176 	queue->priority = priority;
177 	queue->count = 1 << (priority * 2);
178 }
179 
180 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
181 {
182 	queue->owner = pid;
183 	queue->nr = RPC_BATCH_COUNT;
184 }
185 
186 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
187 {
188 	rpc_set_waitqueue_priority(queue, queue->maxpriority);
189 	rpc_set_waitqueue_owner(queue, 0);
190 }
191 
192 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
193 {
194 	int i;
195 
196 	spin_lock_init(&queue->lock);
197 	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
198 		INIT_LIST_HEAD(&queue->tasks[i]);
199 	queue->maxpriority = nr_queues - 1;
200 	rpc_reset_waitqueue_priority(queue);
201 	queue->qlen = 0;
202 	setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
203 	INIT_LIST_HEAD(&queue->timer_list.list);
204 #ifdef RPC_DEBUG
205 	queue->name = qname;
206 #endif
207 }
208 
209 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
210 {
211 	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
212 }
213 
214 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
215 {
216 	__rpc_init_priority_wait_queue(queue, qname, 1);
217 }
218 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
219 
220 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
221 {
222 	del_timer_sync(&queue->timer_list.timer);
223 }
224 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
225 
226 static int rpc_wait_bit_killable(void *word)
227 {
228 	if (fatal_signal_pending(current))
229 		return -ERESTARTSYS;
230 	schedule();
231 	return 0;
232 }
233 
234 #ifdef RPC_DEBUG
235 static void rpc_task_set_debuginfo(struct rpc_task *task)
236 {
237 	static atomic_t rpc_pid;
238 
239 	task->tk_magic = RPC_TASK_MAGIC_ID;
240 	task->tk_pid = atomic_inc_return(&rpc_pid);
241 }
242 #else
243 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
244 {
245 }
246 #endif
247 
248 static void rpc_set_active(struct rpc_task *task)
249 {
250 	struct rpc_clnt *clnt;
251 	if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
252 		return;
253 	rpc_task_set_debuginfo(task);
254 	/* Add to global list of all tasks */
255 	clnt = task->tk_client;
256 	if (clnt != NULL) {
257 		spin_lock(&clnt->cl_lock);
258 		list_add_tail(&task->tk_task, &clnt->cl_tasks);
259 		spin_unlock(&clnt->cl_lock);
260 	}
261 }
262 
263 /*
264  * Mark an RPC call as having completed by clearing the 'active' bit
265  */
266 static void rpc_mark_complete_task(struct rpc_task *task)
267 {
268 	smp_mb__before_clear_bit();
269 	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
270 	smp_mb__after_clear_bit();
271 	wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
272 }
273 
274 /*
275  * Allow callers to wait for completion of an RPC call
276  */
277 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
278 {
279 	if (action == NULL)
280 		action = rpc_wait_bit_killable;
281 	return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
282 			action, TASK_KILLABLE);
283 }
284 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
285 
286 /*
287  * Make an RPC task runnable.
288  *
289  * Note: If the task is ASYNC, this must be called with
290  * the spinlock held to protect the wait queue operation.
291  */
292 static void rpc_make_runnable(struct rpc_task *task)
293 {
294 	rpc_clear_queued(task);
295 	if (rpc_test_and_set_running(task))
296 		return;
297 	if (RPC_IS_ASYNC(task)) {
298 		int status;
299 
300 		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
301 		status = queue_work(rpciod_workqueue, &task->u.tk_work);
302 		if (status < 0) {
303 			printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
304 			task->tk_status = status;
305 			return;
306 		}
307 	} else
308 		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
309 }
310 
311 /*
312  * Prepare for sleeping on a wait queue.
313  * By always appending tasks to the list we ensure FIFO behavior.
314  * NB: An RPC task will only receive interrupt-driven events as long
315  * as it's on a wait queue.
316  */
317 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
318 			rpc_action action)
319 {
320 	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
321 			task->tk_pid, rpc_qname(q), jiffies);
322 
323 	if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
324 		printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
325 		return;
326 	}
327 
328 	__rpc_add_wait_queue(q, task);
329 
330 	BUG_ON(task->tk_callback != NULL);
331 	task->tk_callback = action;
332 	__rpc_add_timer(q, task);
333 }
334 
335 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
336 				rpc_action action)
337 {
338 	/* Mark the task as being activated if so needed */
339 	rpc_set_active(task);
340 
341 	/*
342 	 * Protect the queue operations.
343 	 */
344 	spin_lock_bh(&q->lock);
345 	__rpc_sleep_on(q, task, action);
346 	spin_unlock_bh(&q->lock);
347 }
348 EXPORT_SYMBOL_GPL(rpc_sleep_on);
349 
350 /**
351  * __rpc_do_wake_up_task - wake up a single rpc_task
352  * @queue: wait queue
353  * @task: task to be woken up
354  *
355  * Caller must hold queue->lock, and have cleared the task queued flag.
356  */
357 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
358 {
359 	dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
360 			task->tk_pid, jiffies);
361 
362 #ifdef RPC_DEBUG
363 	BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
364 #endif
365 	/* Has the task been executed yet? If not, we cannot wake it up! */
366 	if (!RPC_IS_ACTIVATED(task)) {
367 		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
368 		return;
369 	}
370 
371 	__rpc_remove_wait_queue(queue, task);
372 
373 	rpc_make_runnable(task);
374 
375 	dprintk("RPC:       __rpc_wake_up_task done\n");
376 }
377 
378 /*
379  * Wake up a queued task while the queue lock is being held
380  */
381 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
382 {
383 	if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
384 		__rpc_do_wake_up_task(queue, task);
385 }
386 
387 /*
388  * Wake up a task on a specific queue
389  */
390 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
391 {
392 	spin_lock_bh(&queue->lock);
393 	rpc_wake_up_task_queue_locked(queue, task);
394 	spin_unlock_bh(&queue->lock);
395 }
396 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
397 
398 /*
399  * Wake up the specified task
400  */
401 static void rpc_wake_up_task(struct rpc_task *task)
402 {
403 	rpc_wake_up_queued_task(task->tk_waitqueue, task);
404 }
405 
406 /*
407  * Wake up the next task on a priority queue.
408  */
409 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
410 {
411 	struct list_head *q;
412 	struct rpc_task *task;
413 
414 	/*
415 	 * Service a batch of tasks from a single owner.
416 	 */
417 	q = &queue->tasks[queue->priority];
418 	if (!list_empty(q)) {
419 		task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
420 		if (queue->owner == task->tk_owner) {
421 			if (--queue->nr)
422 				goto out;
423 			list_move_tail(&task->u.tk_wait.list, q);
424 		}
425 		/*
426 		 * Check if we need to switch queues.
427 		 */
428 		if (--queue->count)
429 			goto new_owner;
430 	}
431 
432 	/*
433 	 * Service the next queue.
434 	 */
435 	do {
436 		if (q == &queue->tasks[0])
437 			q = &queue->tasks[queue->maxpriority];
438 		else
439 			q = q - 1;
440 		if (!list_empty(q)) {
441 			task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
442 			goto new_queue;
443 		}
444 	} while (q != &queue->tasks[queue->priority]);
445 
446 	rpc_reset_waitqueue_priority(queue);
447 	return NULL;
448 
449 new_queue:
450 	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
451 new_owner:
452 	rpc_set_waitqueue_owner(queue, task->tk_owner);
453 out:
454 	rpc_wake_up_task_queue_locked(queue, task);
455 	return task;
456 }
457 
458 /*
459  * Wake up the next task on the wait queue.
460  */
461 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
462 {
463 	struct rpc_task	*task = NULL;
464 
465 	dprintk("RPC:       wake_up_next(%p \"%s\")\n",
466 			queue, rpc_qname(queue));
467 	spin_lock_bh(&queue->lock);
468 	if (RPC_IS_PRIORITY(queue))
469 		task = __rpc_wake_up_next_priority(queue);
470 	else {
471 		task_for_first(task, &queue->tasks[0])
472 			rpc_wake_up_task_queue_locked(queue, task);
473 	}
474 	spin_unlock_bh(&queue->lock);
475 
476 	return task;
477 }
478 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
479 
480 /**
481  * rpc_wake_up - wake up all rpc_tasks
482  * @queue: rpc_wait_queue on which the tasks are sleeping
483  *
484  * Grabs queue->lock
485  */
486 void rpc_wake_up(struct rpc_wait_queue *queue)
487 {
488 	struct rpc_task *task, *next;
489 	struct list_head *head;
490 
491 	spin_lock_bh(&queue->lock);
492 	head = &queue->tasks[queue->maxpriority];
493 	for (;;) {
494 		list_for_each_entry_safe(task, next, head, u.tk_wait.list)
495 			rpc_wake_up_task_queue_locked(queue, task);
496 		if (head == &queue->tasks[0])
497 			break;
498 		head--;
499 	}
500 	spin_unlock_bh(&queue->lock);
501 }
502 EXPORT_SYMBOL_GPL(rpc_wake_up);
503 
504 /**
505  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
506  * @queue: rpc_wait_queue on which the tasks are sleeping
507  * @status: status value to set
508  *
509  * Grabs queue->lock
510  */
511 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
512 {
513 	struct rpc_task *task, *next;
514 	struct list_head *head;
515 
516 	spin_lock_bh(&queue->lock);
517 	head = &queue->tasks[queue->maxpriority];
518 	for (;;) {
519 		list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
520 			task->tk_status = status;
521 			rpc_wake_up_task_queue_locked(queue, task);
522 		}
523 		if (head == &queue->tasks[0])
524 			break;
525 		head--;
526 	}
527 	spin_unlock_bh(&queue->lock);
528 }
529 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
530 
531 static void __rpc_queue_timer_fn(unsigned long ptr)
532 {
533 	struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
534 	struct rpc_task *task, *n;
535 	unsigned long expires, now, timeo;
536 
537 	spin_lock(&queue->lock);
538 	expires = now = jiffies;
539 	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
540 		timeo = task->u.tk_wait.expires;
541 		if (time_after_eq(now, timeo)) {
542 			dprintk("RPC: %5u timeout\n", task->tk_pid);
543 			task->tk_status = -ETIMEDOUT;
544 			rpc_wake_up_task_queue_locked(queue, task);
545 			continue;
546 		}
547 		if (expires == now || time_after(expires, timeo))
548 			expires = timeo;
549 	}
550 	if (!list_empty(&queue->timer_list.list))
551 		rpc_set_queue_timer(queue, expires);
552 	spin_unlock(&queue->lock);
553 }
554 
555 static void __rpc_atrun(struct rpc_task *task)
556 {
557 	task->tk_status = 0;
558 }
559 
560 /*
561  * Run a task at a later time
562  */
563 void rpc_delay(struct rpc_task *task, unsigned long delay)
564 {
565 	task->tk_timeout = delay;
566 	rpc_sleep_on(&delay_queue, task, __rpc_atrun);
567 }
568 EXPORT_SYMBOL_GPL(rpc_delay);
569 
570 /*
571  * Helper to call task->tk_ops->rpc_call_prepare
572  */
573 void rpc_prepare_task(struct rpc_task *task)
574 {
575 	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
576 }
577 
578 /*
579  * Helper that calls task->tk_ops->rpc_call_done if it exists
580  */
581 void rpc_exit_task(struct rpc_task *task)
582 {
583 	task->tk_action = NULL;
584 	if (task->tk_ops->rpc_call_done != NULL) {
585 		task->tk_ops->rpc_call_done(task, task->tk_calldata);
586 		if (task->tk_action != NULL) {
587 			WARN_ON(RPC_ASSASSINATED(task));
588 			/* Always release the RPC slot and buffer memory */
589 			xprt_release(task);
590 		}
591 	}
592 }
593 EXPORT_SYMBOL_GPL(rpc_exit_task);
594 
595 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
596 {
597 	if (ops->rpc_release != NULL)
598 		ops->rpc_release(calldata);
599 }
600 
601 /*
602  * This is the RPC `scheduler' (or rather, the finite state machine).
603  */
604 static void __rpc_execute(struct rpc_task *task)
605 {
606 	struct rpc_wait_queue *queue;
607 	int task_is_async = RPC_IS_ASYNC(task);
608 	int status = 0;
609 
610 	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
611 			task->tk_pid, task->tk_flags);
612 
613 	BUG_ON(RPC_IS_QUEUED(task));
614 
615 	for (;;) {
616 
617 		/*
618 		 * Execute any pending callback.
619 		 */
620 		if (task->tk_callback) {
621 			void (*save_callback)(struct rpc_task *);
622 
623 			/*
624 			 * We set tk_callback to NULL before calling it,
625 			 * in case it sets the tk_callback field itself:
626 			 */
627 			save_callback = task->tk_callback;
628 			task->tk_callback = NULL;
629 			save_callback(task);
630 		}
631 
632 		/*
633 		 * Perform the next FSM step.
634 		 * tk_action may be NULL when the task has been killed
635 		 * by someone else.
636 		 */
637 		if (!RPC_IS_QUEUED(task)) {
638 			if (task->tk_action == NULL)
639 				break;
640 			task->tk_action(task);
641 		}
642 
643 		/*
644 		 * Lockless check for whether task is sleeping or not.
645 		 */
646 		if (!RPC_IS_QUEUED(task))
647 			continue;
648 		/*
649 		 * The queue->lock protects against races with
650 		 * rpc_make_runnable().
651 		 *
652 		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
653 		 * rpc_task, rpc_make_runnable() can assign it to a
654 		 * different workqueue. We therefore cannot assume that the
655 		 * rpc_task pointer may still be dereferenced.
656 		 */
657 		queue = task->tk_waitqueue;
658 		spin_lock_bh(&queue->lock);
659 		if (!RPC_IS_QUEUED(task)) {
660 			spin_unlock_bh(&queue->lock);
661 			continue;
662 		}
663 		rpc_clear_running(task);
664 		spin_unlock_bh(&queue->lock);
665 		if (task_is_async)
666 			return;
667 
668 		/* sync task: sleep here */
669 		dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
670 		status = out_of_line_wait_on_bit(&task->tk_runstate,
671 				RPC_TASK_QUEUED, rpc_wait_bit_killable,
672 				TASK_KILLABLE);
673 		if (status == -ERESTARTSYS) {
674 			/*
675 			 * When a sync task receives a signal, it exits with
676 			 * -ERESTARTSYS. In order to catch any callbacks that
677 			 * clean up after sleeping on some queue, we don't
678 			 * break the loop here, but go around once more.
679 			 */
680 			dprintk("RPC: %5u got signal\n", task->tk_pid);
681 			task->tk_flags |= RPC_TASK_KILLED;
682 			rpc_exit(task, -ERESTARTSYS);
683 			rpc_wake_up_task(task);
684 		}
685 		rpc_set_running(task);
686 		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
687 	}
688 
689 	dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
690 			task->tk_status);
691 	/* Release all resources associated with the task */
692 	rpc_release_task(task);
693 }
694 
695 /*
696  * User-visible entry point to the scheduler.
697  *
698  * This may be called recursively if e.g. an async NFS task updates
699  * the attributes and finds that dirty pages must be flushed.
700  * NOTE: Upon exit of this function the task is guaranteed to be
701  *	 released. In particular note that tk_release() will have
702  *	 been called, so your task memory may have been freed.
703  */
704 void rpc_execute(struct rpc_task *task)
705 {
706 	rpc_set_active(task);
707 	rpc_set_running(task);
708 	__rpc_execute(task);
709 }
710 
711 static void rpc_async_schedule(struct work_struct *work)
712 {
713 	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
714 }
715 
716 /**
717  * rpc_malloc - allocate an RPC buffer
718  * @task: RPC task that will use this buffer
719  * @size: requested byte size
720  *
721  * To prevent rpciod from hanging, this allocator never sleeps,
722  * returning NULL if the request cannot be serviced immediately.
723  * The caller can arrange to sleep in a way that is safe for rpciod.
724  *
725  * Most requests are 'small' (under 2KiB) and can be serviced from a
726  * mempool, ensuring that NFS reads and writes can always proceed,
727  * and that there is good locality of reference for these buffers.
728  *
729  * In order to avoid memory starvation triggering more writebacks of
730  * NFS requests, we avoid using GFP_KERNEL.
731  */
732 void *rpc_malloc(struct rpc_task *task, size_t size)
733 {
734 	struct rpc_buffer *buf;
735 	gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
736 
737 	size += sizeof(struct rpc_buffer);
738 	if (size <= RPC_BUFFER_MAXSIZE)
739 		buf = mempool_alloc(rpc_buffer_mempool, gfp);
740 	else
741 		buf = kmalloc(size, gfp);
742 
743 	if (!buf)
744 		return NULL;
745 
746 	buf->len = size;
747 	dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
748 			task->tk_pid, size, buf);
749 	return &buf->data;
750 }
751 EXPORT_SYMBOL_GPL(rpc_malloc);
752 
753 /**
754  * rpc_free - free buffer allocated via rpc_malloc
755  * @buffer: buffer to free
756  *
757  */
758 void rpc_free(void *buffer)
759 {
760 	size_t size;
761 	struct rpc_buffer *buf;
762 
763 	if (!buffer)
764 		return;
765 
766 	buf = container_of(buffer, struct rpc_buffer, data);
767 	size = buf->len;
768 
769 	dprintk("RPC:       freeing buffer of size %zu at %p\n",
770 			size, buf);
771 
772 	if (size <= RPC_BUFFER_MAXSIZE)
773 		mempool_free(buf, rpc_buffer_mempool);
774 	else
775 		kfree(buf);
776 }
777 EXPORT_SYMBOL_GPL(rpc_free);
778 
779 /*
780  * Creation and deletion of RPC task structures
781  */
782 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
783 {
784 	memset(task, 0, sizeof(*task));
785 	atomic_set(&task->tk_count, 1);
786 	task->tk_flags  = task_setup_data->flags;
787 	task->tk_ops = task_setup_data->callback_ops;
788 	task->tk_calldata = task_setup_data->callback_data;
789 	INIT_LIST_HEAD(&task->tk_task);
790 
791 	/* Initialize retry counters */
792 	task->tk_garb_retry = 2;
793 	task->tk_cred_retry = 2;
794 
795 	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
796 	task->tk_owner = current->tgid;
797 
798 	/* Initialize workqueue for async tasks */
799 	task->tk_workqueue = task_setup_data->workqueue;
800 
801 	task->tk_client = task_setup_data->rpc_client;
802 	if (task->tk_client != NULL) {
803 		kref_get(&task->tk_client->cl_kref);
804 		if (task->tk_client->cl_softrtry)
805 			task->tk_flags |= RPC_TASK_SOFT;
806 	}
807 
808 	if (task->tk_ops->rpc_call_prepare != NULL)
809 		task->tk_action = rpc_prepare_task;
810 
811 	if (task_setup_data->rpc_message != NULL) {
812 		task->tk_msg.rpc_proc = task_setup_data->rpc_message->rpc_proc;
813 		task->tk_msg.rpc_argp = task_setup_data->rpc_message->rpc_argp;
814 		task->tk_msg.rpc_resp = task_setup_data->rpc_message->rpc_resp;
815 		/* Bind the user cred */
816 		rpcauth_bindcred(task, task_setup_data->rpc_message->rpc_cred, task_setup_data->flags);
817 		if (task->tk_action == NULL)
818 			rpc_call_start(task);
819 	}
820 
821 	/* starting timestamp */
822 	task->tk_start = jiffies;
823 
824 	dprintk("RPC:       new task initialized, procpid %u\n",
825 				task_pid_nr(current));
826 }
827 
828 static struct rpc_task *
829 rpc_alloc_task(void)
830 {
831 	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
832 }
833 
834 /*
835  * Create a new task for the specified client.
836  */
837 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
838 {
839 	struct rpc_task	*task = setup_data->task;
840 	unsigned short flags = 0;
841 
842 	if (task == NULL) {
843 		task = rpc_alloc_task();
844 		if (task == NULL)
845 			goto out;
846 		flags = RPC_TASK_DYNAMIC;
847 	}
848 
849 	rpc_init_task(task, setup_data);
850 
851 	task->tk_flags |= flags;
852 	dprintk("RPC:       allocated task %p\n", task);
853 out:
854 	return task;
855 }
856 
857 static void rpc_free_task(struct rpc_task *task)
858 {
859 	const struct rpc_call_ops *tk_ops = task->tk_ops;
860 	void *calldata = task->tk_calldata;
861 
862 	if (task->tk_flags & RPC_TASK_DYNAMIC) {
863 		dprintk("RPC: %5u freeing task\n", task->tk_pid);
864 		mempool_free(task, rpc_task_mempool);
865 	}
866 	rpc_release_calldata(tk_ops, calldata);
867 }
868 
869 static void rpc_async_release(struct work_struct *work)
870 {
871 	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
872 }
873 
874 void rpc_put_task(struct rpc_task *task)
875 {
876 	if (!atomic_dec_and_test(&task->tk_count))
877 		return;
878 	/* Release resources */
879 	if (task->tk_rqstp)
880 		xprt_release(task);
881 	if (task->tk_msg.rpc_cred)
882 		rpcauth_unbindcred(task);
883 	if (task->tk_client) {
884 		rpc_release_client(task->tk_client);
885 		task->tk_client = NULL;
886 	}
887 	if (task->tk_workqueue != NULL) {
888 		INIT_WORK(&task->u.tk_work, rpc_async_release);
889 		queue_work(task->tk_workqueue, &task->u.tk_work);
890 	} else
891 		rpc_free_task(task);
892 }
893 EXPORT_SYMBOL_GPL(rpc_put_task);
894 
895 static void rpc_release_task(struct rpc_task *task)
896 {
897 #ifdef RPC_DEBUG
898 	BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
899 #endif
900 	dprintk("RPC: %5u release task\n", task->tk_pid);
901 
902 	if (!list_empty(&task->tk_task)) {
903 		struct rpc_clnt *clnt = task->tk_client;
904 		/* Remove from client task list */
905 		spin_lock(&clnt->cl_lock);
906 		list_del(&task->tk_task);
907 		spin_unlock(&clnt->cl_lock);
908 	}
909 	BUG_ON (RPC_IS_QUEUED(task));
910 
911 #ifdef RPC_DEBUG
912 	task->tk_magic = 0;
913 #endif
914 	/* Wake up anyone who is waiting for task completion */
915 	rpc_mark_complete_task(task);
916 
917 	rpc_put_task(task);
918 }
919 
920 /*
921  * Kill all tasks for the given client.
922  * XXX: kill their descendants as well?
923  */
924 void rpc_killall_tasks(struct rpc_clnt *clnt)
925 {
926 	struct rpc_task	*rovr;
927 
928 
929 	if (list_empty(&clnt->cl_tasks))
930 		return;
931 	dprintk("RPC:       killing all tasks for client %p\n", clnt);
932 	/*
933 	 * Spin lock all_tasks to prevent changes...
934 	 */
935 	spin_lock(&clnt->cl_lock);
936 	list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
937 		if (! RPC_IS_ACTIVATED(rovr))
938 			continue;
939 		if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
940 			rovr->tk_flags |= RPC_TASK_KILLED;
941 			rpc_exit(rovr, -EIO);
942 			rpc_wake_up_task(rovr);
943 		}
944 	}
945 	spin_unlock(&clnt->cl_lock);
946 }
947 EXPORT_SYMBOL_GPL(rpc_killall_tasks);
948 
949 int rpciod_up(void)
950 {
951 	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
952 }
953 
954 void rpciod_down(void)
955 {
956 	module_put(THIS_MODULE);
957 }
958 
959 /*
960  * Start up the rpciod workqueue.
961  */
962 static int rpciod_start(void)
963 {
964 	struct workqueue_struct *wq;
965 
966 	/*
967 	 * Create the rpciod thread and wait for it to start.
968 	 */
969 	dprintk("RPC:       creating workqueue rpciod\n");
970 	wq = create_workqueue("rpciod");
971 	rpciod_workqueue = wq;
972 	return rpciod_workqueue != NULL;
973 }
974 
975 static void rpciod_stop(void)
976 {
977 	struct workqueue_struct *wq = NULL;
978 
979 	if (rpciod_workqueue == NULL)
980 		return;
981 	dprintk("RPC:       destroying workqueue rpciod\n");
982 
983 	wq = rpciod_workqueue;
984 	rpciod_workqueue = NULL;
985 	destroy_workqueue(wq);
986 }
987 
988 void
989 rpc_destroy_mempool(void)
990 {
991 	rpciod_stop();
992 	if (rpc_buffer_mempool)
993 		mempool_destroy(rpc_buffer_mempool);
994 	if (rpc_task_mempool)
995 		mempool_destroy(rpc_task_mempool);
996 	if (rpc_task_slabp)
997 		kmem_cache_destroy(rpc_task_slabp);
998 	if (rpc_buffer_slabp)
999 		kmem_cache_destroy(rpc_buffer_slabp);
1000 	rpc_destroy_wait_queue(&delay_queue);
1001 }
1002 
1003 int
1004 rpc_init_mempool(void)
1005 {
1006 	/*
1007 	 * The following is not strictly a mempool initialisation,
1008 	 * but there is no harm in doing it here
1009 	 */
1010 	rpc_init_wait_queue(&delay_queue, "delayq");
1011 	if (!rpciod_start())
1012 		goto err_nomem;
1013 
1014 	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1015 					     sizeof(struct rpc_task),
1016 					     0, SLAB_HWCACHE_ALIGN,
1017 					     NULL);
1018 	if (!rpc_task_slabp)
1019 		goto err_nomem;
1020 	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1021 					     RPC_BUFFER_MAXSIZE,
1022 					     0, SLAB_HWCACHE_ALIGN,
1023 					     NULL);
1024 	if (!rpc_buffer_slabp)
1025 		goto err_nomem;
1026 	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1027 						    rpc_task_slabp);
1028 	if (!rpc_task_mempool)
1029 		goto err_nomem;
1030 	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1031 						      rpc_buffer_slabp);
1032 	if (!rpc_buffer_mempool)
1033 		goto err_nomem;
1034 	return 0;
1035 err_nomem:
1036 	rpc_destroy_mempool();
1037 	return -ENOMEM;
1038 }
1039