Linux Audio

Check our new training course

Loading...
v5.9
   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#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  31#define RPCDBG_FACILITY		RPCDBG_SCHED
  32#endif
  33
  34#define CREATE_TRACE_POINTS
  35#include <trace/events/sunrpc.h>
  36
  37/*
  38 * RPC slabs and memory pools
  39 */
  40#define RPC_BUFFER_MAXSIZE	(2048)
  41#define RPC_BUFFER_POOLSIZE	(8)
  42#define RPC_TASK_POOLSIZE	(8)
  43static struct kmem_cache	*rpc_task_slabp __read_mostly;
  44static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
  45static mempool_t	*rpc_task_mempool __read_mostly;
  46static mempool_t	*rpc_buffer_mempool __read_mostly;
  47
  48static void			rpc_async_schedule(struct work_struct *);
  49static void			 rpc_release_task(struct rpc_task *task);
  50static void __rpc_queue_timer_fn(struct work_struct *);
  51
  52/*
  53 * RPC tasks sit here while waiting for conditions to improve.
  54 */
  55static struct rpc_wait_queue delay_queue;
  56
  57/*
  58 * rpciod-related stuff
  59 */
  60struct workqueue_struct *rpciod_workqueue __read_mostly;
  61struct workqueue_struct *xprtiod_workqueue __read_mostly;
  62EXPORT_SYMBOL_GPL(xprtiod_workqueue);
  63
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  64unsigned long
  65rpc_task_timeout(const struct rpc_task *task)
  66{
  67	unsigned long timeout = READ_ONCE(task->tk_timeout);
  68
  69	if (timeout != 0) {
  70		unsigned long now = jiffies;
  71		if (time_before(now, timeout))
  72			return timeout - now;
  73	}
  74	return 0;
  75}
  76EXPORT_SYMBOL_GPL(rpc_task_timeout);
  77
  78/*
  79 * Disable the timer for a given RPC task. Should be called with
  80 * queue->lock and bh_disabled in order to avoid races within
  81 * rpc_run_timer().
  82 */
  83static void
  84__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  85{
  86	if (list_empty(&task->u.tk_wait.timer_list))
  87		return;
  88	dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  89	task->tk_timeout = 0;
  90	list_del(&task->u.tk_wait.timer_list);
  91	if (list_empty(&queue->timer_list.list))
  92		cancel_delayed_work(&queue->timer_list.dwork);
  93}
  94
  95static void
  96rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  97{
  98	unsigned long now = jiffies;
  99	queue->timer_list.expires = expires;
 100	if (time_before_eq(expires, now))
 101		expires = 0;
 102	else
 103		expires -= now;
 104	mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
 105}
 106
 107/*
 108 * Set up a timer for the current task.
 109 */
 110static void
 111__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
 112		unsigned long timeout)
 113{
 114	dprintk("RPC: %5u setting alarm for %u ms\n",
 115		task->tk_pid, jiffies_to_msecs(timeout - jiffies));
 116
 117	task->tk_timeout = timeout;
 118	if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
 119		rpc_set_queue_timer(queue, timeout);
 120	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
 121}
 122
 123static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 124{
 125	if (queue->priority != priority) {
 126		queue->priority = priority;
 127		queue->nr = 1U << priority;
 128	}
 129}
 130
 131static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 132{
 133	rpc_set_waitqueue_priority(queue, queue->maxpriority);
 134}
 135
 136/*
 137 * Add a request to a queue list
 138 */
 139static void
 140__rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
 141{
 142	struct rpc_task *t;
 143
 144	list_for_each_entry(t, q, u.tk_wait.list) {
 145		if (t->tk_owner == task->tk_owner) {
 146			list_add_tail(&task->u.tk_wait.links,
 147					&t->u.tk_wait.links);
 148			/* Cache the queue head in task->u.tk_wait.list */
 149			task->u.tk_wait.list.next = q;
 150			task->u.tk_wait.list.prev = NULL;
 151			return;
 152		}
 153	}
 154	INIT_LIST_HEAD(&task->u.tk_wait.links);
 155	list_add_tail(&task->u.tk_wait.list, q);
 156}
 157
 158/*
 159 * Remove request from a queue list
 160 */
 161static void
 162__rpc_list_dequeue_task(struct rpc_task *task)
 163{
 164	struct list_head *q;
 165	struct rpc_task *t;
 166
 167	if (task->u.tk_wait.list.prev == NULL) {
 168		list_del(&task->u.tk_wait.links);
 169		return;
 170	}
 171	if (!list_empty(&task->u.tk_wait.links)) {
 172		t = list_first_entry(&task->u.tk_wait.links,
 173				struct rpc_task,
 174				u.tk_wait.links);
 175		/* Assume __rpc_list_enqueue_task() cached the queue head */
 176		q = t->u.tk_wait.list.next;
 177		list_add_tail(&t->u.tk_wait.list, q);
 178		list_del(&task->u.tk_wait.links);
 179	}
 180	list_del(&task->u.tk_wait.list);
 181}
 182
 183/*
 184 * Add new request to a priority queue.
 185 */
 186static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
 187		struct rpc_task *task,
 188		unsigned char queue_priority)
 189{
 190	if (unlikely(queue_priority > queue->maxpriority))
 191		queue_priority = queue->maxpriority;
 192	__rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
 193}
 194
 195/*
 196 * Add new request to wait queue.
 197 *
 198 * Swapper tasks always get inserted at the head of the queue.
 199 * This should avoid many nasty memory deadlocks and hopefully
 200 * improve overall performance.
 201 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 202 */
 203static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
 204		struct rpc_task *task,
 205		unsigned char queue_priority)
 206{
 207	INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
 208	if (RPC_IS_PRIORITY(queue))
 209		__rpc_add_wait_queue_priority(queue, task, queue_priority);
 210	else if (RPC_IS_SWAPPER(task))
 211		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 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	dprintk("RPC: %5u added to queue %p \"%s\"\n",
 221			task->tk_pid, queue, rpc_qname(queue));
 222}
 223
 224/*
 225 * Remove request from a priority queue.
 226 */
 227static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 228{
 229	__rpc_list_dequeue_task(task);
 230}
 231
 232/*
 233 * Remove request from queue.
 234 * Note: must be called with spin lock held.
 235 */
 236static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 237{
 238	__rpc_disable_timer(queue, task);
 239	if (RPC_IS_PRIORITY(queue))
 240		__rpc_remove_wait_queue_priority(task);
 241	else
 242		list_del(&task->u.tk_wait.list);
 243	queue->qlen--;
 244	dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 245			task->tk_pid, queue, rpc_qname(queue));
 246}
 247
 248static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 249{
 250	int i;
 251
 252	spin_lock_init(&queue->lock);
 253	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 254		INIT_LIST_HEAD(&queue->tasks[i]);
 255	queue->maxpriority = nr_queues - 1;
 256	rpc_reset_waitqueue_priority(queue);
 257	queue->qlen = 0;
 258	queue->timer_list.expires = 0;
 259	INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
 260	INIT_LIST_HEAD(&queue->timer_list.list);
 261	rpc_assign_waitqueue_name(queue, qname);
 262}
 263
 264void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 265{
 266	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 267}
 268EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 269
 270void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 271{
 272	__rpc_init_priority_wait_queue(queue, qname, 1);
 273}
 274EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 275
 276void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 277{
 278	cancel_delayed_work_sync(&queue->timer_list.dwork);
 279}
 280EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 281
 282static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
 283{
 284	freezable_schedule_unsafe();
 285	if (signal_pending_state(mode, current))
 286		return -ERESTARTSYS;
 287	return 0;
 288}
 289
 290#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
 291static void rpc_task_set_debuginfo(struct rpc_task *task)
 292{
 293	static atomic_t rpc_pid;
 
 
 
 
 
 
 
 
 294
 295	task->tk_pid = atomic_inc_return(&rpc_pid);
 296}
 297#else
 298static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 299{
 300}
 301#endif
 302
 303static void rpc_set_active(struct rpc_task *task)
 304{
 305	rpc_task_set_debuginfo(task);
 306	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 307	trace_rpc_task_begin(task, NULL);
 308}
 309
 310/*
 311 * Mark an RPC call as having completed by clearing the 'active' bit
 312 * and then waking up all tasks that were sleeping.
 313 */
 314static int rpc_complete_task(struct rpc_task *task)
 315{
 316	void *m = &task->tk_runstate;
 317	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
 318	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
 319	unsigned long flags;
 320	int ret;
 321
 322	trace_rpc_task_complete(task, NULL);
 323
 324	spin_lock_irqsave(&wq->lock, flags);
 325	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 326	ret = atomic_dec_and_test(&task->tk_count);
 327	if (waitqueue_active(wq))
 328		__wake_up_locked_key(wq, TASK_NORMAL, &k);
 329	spin_unlock_irqrestore(&wq->lock, flags);
 330	return ret;
 331}
 332
 333/*
 334 * Allow callers to wait for completion of an RPC call
 335 *
 336 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
 337 * to enforce taking of the wq->lock and hence avoid races with
 338 * rpc_complete_task().
 339 */
 340int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
 341{
 342	if (action == NULL)
 343		action = rpc_wait_bit_killable;
 344	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 345			action, TASK_KILLABLE);
 346}
 347EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 348
 349/*
 350 * Make an RPC task runnable.
 351 *
 352 * Note: If the task is ASYNC, and is being made runnable after sitting on an
 353 * rpc_wait_queue, this must be called with the queue spinlock held to protect
 354 * the wait queue operation.
 355 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
 356 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
 357 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
 358 * the RPC_TASK_RUNNING flag.
 359 */
 360static void rpc_make_runnable(struct workqueue_struct *wq,
 361		struct rpc_task *task)
 362{
 363	bool need_wakeup = !rpc_test_and_set_running(task);
 364
 365	rpc_clear_queued(task);
 366	if (!need_wakeup)
 367		return;
 368	if (RPC_IS_ASYNC(task)) {
 369		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 370		queue_work(wq, &task->u.tk_work);
 371	} else
 372		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 373}
 374
 375/*
 376 * Prepare for sleeping on a wait queue.
 377 * By always appending tasks to the list we ensure FIFO behavior.
 378 * NB: An RPC task will only receive interrupt-driven events as long
 379 * as it's on a wait queue.
 380 */
 381static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
 382		struct rpc_task *task,
 383		unsigned char queue_priority)
 384{
 385	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 386			task->tk_pid, rpc_qname(q), jiffies);
 387
 388	trace_rpc_task_sleep(task, q);
 389
 390	__rpc_add_wait_queue(q, task, queue_priority);
 391
 392}
 393
 394static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 395		struct rpc_task *task,
 396		unsigned char queue_priority)
 397{
 398	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 399		return;
 400	__rpc_do_sleep_on_priority(q, task, queue_priority);
 401}
 402
 403static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 404		struct rpc_task *task, unsigned long timeout,
 405		unsigned char queue_priority)
 406{
 407	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 408		return;
 409	if (time_is_after_jiffies(timeout)) {
 410		__rpc_do_sleep_on_priority(q, task, queue_priority);
 411		__rpc_add_timer(q, task, timeout);
 412	} else
 413		task->tk_status = -ETIMEDOUT;
 414}
 415
 416static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
 417{
 418	if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
 419		task->tk_callback = action;
 420}
 421
 422static bool rpc_sleep_check_activated(struct rpc_task *task)
 423{
 424	/* We shouldn't ever put an inactive task to sleep */
 425	if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
 426		task->tk_status = -EIO;
 427		rpc_put_task_async(task);
 428		return false;
 429	}
 430	return true;
 431}
 432
 433void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
 434				rpc_action action, unsigned long timeout)
 435{
 436	if (!rpc_sleep_check_activated(task))
 437		return;
 438
 439	rpc_set_tk_callback(task, action);
 440
 441	/*
 442	 * Protect the queue operations.
 443	 */
 444	spin_lock(&q->lock);
 445	__rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
 446	spin_unlock(&q->lock);
 447}
 448EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
 449
 450void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 451				rpc_action action)
 452{
 453	if (!rpc_sleep_check_activated(task))
 454		return;
 455
 456	rpc_set_tk_callback(task, action);
 457
 458	WARN_ON_ONCE(task->tk_timeout != 0);
 459	/*
 460	 * Protect the queue operations.
 461	 */
 462	spin_lock(&q->lock);
 463	__rpc_sleep_on_priority(q, task, task->tk_priority);
 464	spin_unlock(&q->lock);
 465}
 466EXPORT_SYMBOL_GPL(rpc_sleep_on);
 467
 468void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 469		struct rpc_task *task, unsigned long timeout, int priority)
 470{
 471	if (!rpc_sleep_check_activated(task))
 472		return;
 473
 474	priority -= RPC_PRIORITY_LOW;
 475	/*
 476	 * Protect the queue operations.
 477	 */
 478	spin_lock(&q->lock);
 479	__rpc_sleep_on_priority_timeout(q, task, timeout, priority);
 480	spin_unlock(&q->lock);
 481}
 482EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
 483
 484void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 485		int priority)
 486{
 487	if (!rpc_sleep_check_activated(task))
 488		return;
 489
 490	WARN_ON_ONCE(task->tk_timeout != 0);
 491	priority -= RPC_PRIORITY_LOW;
 492	/*
 493	 * Protect the queue operations.
 494	 */
 495	spin_lock(&q->lock);
 496	__rpc_sleep_on_priority(q, task, priority);
 497	spin_unlock(&q->lock);
 498}
 499EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
 500
 501/**
 502 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
 503 * @wq: workqueue on which to run task
 504 * @queue: wait queue
 505 * @task: task to be woken up
 506 *
 507 * Caller must hold queue->lock, and have cleared the task queued flag.
 508 */
 509static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
 510		struct rpc_wait_queue *queue,
 511		struct rpc_task *task)
 512{
 513	dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 514			task->tk_pid, jiffies);
 515
 516	/* Has the task been executed yet? If not, we cannot wake it up! */
 517	if (!RPC_IS_ACTIVATED(task)) {
 518		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 519		return;
 520	}
 521
 522	trace_rpc_task_wakeup(task, queue);
 523
 524	__rpc_remove_wait_queue(queue, task);
 525
 526	rpc_make_runnable(wq, task);
 527
 528	dprintk("RPC:       __rpc_wake_up_task done\n");
 529}
 530
 531/*
 532 * Wake up a queued task while the queue lock is being held
 533 */
 534static struct rpc_task *
 535rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
 536		struct rpc_wait_queue *queue, struct rpc_task *task,
 537		bool (*action)(struct rpc_task *, void *), void *data)
 538{
 539	if (RPC_IS_QUEUED(task)) {
 540		smp_rmb();
 541		if (task->tk_waitqueue == queue) {
 542			if (action == NULL || action(task, data)) {
 543				__rpc_do_wake_up_task_on_wq(wq, queue, task);
 544				return task;
 545			}
 546		}
 547	}
 548	return NULL;
 549}
 550
 551/*
 552 * Wake up a queued task while the queue lock is being held
 553 */
 554static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
 555					  struct rpc_task *task)
 556{
 557	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 558						   task, NULL, NULL);
 559}
 560
 561/*
 562 * Wake up a task on a specific queue
 563 */
 564void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 565{
 566	if (!RPC_IS_QUEUED(task))
 567		return;
 568	spin_lock(&queue->lock);
 569	rpc_wake_up_task_queue_locked(queue, task);
 570	spin_unlock(&queue->lock);
 571}
 572EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 573
 574static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
 575{
 576	task->tk_status = *(int *)status;
 577	return true;
 578}
 579
 580static void
 581rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
 582		struct rpc_task *task, int status)
 583{
 584	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 585			task, rpc_task_action_set_status, &status);
 586}
 587
 588/**
 589 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
 590 * @queue: pointer to rpc_wait_queue
 591 * @task: pointer to rpc_task
 592 * @status: integer error value
 593 *
 594 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
 595 * set to the value of @status.
 596 */
 597void
 598rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
 599		struct rpc_task *task, int status)
 600{
 601	if (!RPC_IS_QUEUED(task))
 602		return;
 603	spin_lock(&queue->lock);
 604	rpc_wake_up_task_queue_set_status_locked(queue, task, status);
 605	spin_unlock(&queue->lock);
 606}
 607
 608/*
 609 * Wake up the next task on a priority queue.
 610 */
 611static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 612{
 613	struct list_head *q;
 614	struct rpc_task *task;
 615
 616	/*
 
 
 
 
 
 
 
 
 
 617	 * Service a batch of tasks from a single owner.
 618	 */
 619	q = &queue->tasks[queue->priority];
 620	if (!list_empty(q) && --queue->nr) {
 
 621		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 622		goto out;
 623	}
 624
 625	/*
 626	 * Service the next queue.
 627	 */
 628	do {
 629		if (q == &queue->tasks[0])
 630			q = &queue->tasks[queue->maxpriority];
 631		else
 632			q = q - 1;
 633		if (!list_empty(q)) {
 634			task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 635			goto new_queue;
 636		}
 637	} while (q != &queue->tasks[queue->priority]);
 638
 639	rpc_reset_waitqueue_priority(queue);
 640	return NULL;
 641
 642new_queue:
 643	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 644out:
 645	return task;
 646}
 647
 648static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
 649{
 650	if (RPC_IS_PRIORITY(queue))
 651		return __rpc_find_next_queued_priority(queue);
 652	if (!list_empty(&queue->tasks[0]))
 653		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
 654	return NULL;
 655}
 656
 657/*
 658 * Wake up the first task on the wait queue.
 659 */
 660struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
 661		struct rpc_wait_queue *queue,
 662		bool (*func)(struct rpc_task *, void *), void *data)
 663{
 664	struct rpc_task	*task = NULL;
 665
 666	dprintk("RPC:       wake_up_first(%p \"%s\")\n",
 667			queue, rpc_qname(queue));
 668	spin_lock(&queue->lock);
 669	task = __rpc_find_next_queued(queue);
 670	if (task != NULL)
 671		task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
 672				task, func, data);
 673	spin_unlock(&queue->lock);
 674
 675	return task;
 676}
 677
 678/*
 679 * Wake up the first task on the wait queue.
 680 */
 681struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
 682		bool (*func)(struct rpc_task *, void *), void *data)
 683{
 684	return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
 685}
 686EXPORT_SYMBOL_GPL(rpc_wake_up_first);
 687
 688static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
 689{
 690	return true;
 691}
 692
 693/*
 694 * Wake up the next task on the wait queue.
 695*/
 696struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
 697{
 698	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
 699}
 700EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 701
 702/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 703 * rpc_wake_up - wake up all rpc_tasks
 704 * @queue: rpc_wait_queue on which the tasks are sleeping
 705 *
 706 * Grabs queue->lock
 707 */
 708void rpc_wake_up(struct rpc_wait_queue *queue)
 709{
 710	struct list_head *head;
 711
 712	spin_lock(&queue->lock);
 713	head = &queue->tasks[queue->maxpriority];
 
 
 
 
 
 
 
 
 
 
 
 
 
 714	for (;;) {
 715		while (!list_empty(head)) {
 716			struct rpc_task *task;
 717			task = list_first_entry(head,
 718					struct rpc_task,
 719					u.tk_wait.list);
 720			rpc_wake_up_task_queue_locked(queue, task);
 721		}
 722		if (head == &queue->tasks[0])
 723			break;
 724		head--;
 725	}
 726	spin_unlock(&queue->lock);
 727}
 728EXPORT_SYMBOL_GPL(rpc_wake_up);
 729
 730/**
 731 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 732 * @queue: rpc_wait_queue on which the tasks are sleeping
 733 * @status: status value to set
 734 *
 735 * Grabs queue->lock
 736 */
 737void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 738{
 739	struct list_head *head;
 740
 741	spin_lock(&queue->lock);
 742	head = &queue->tasks[queue->maxpriority];
 743	for (;;) {
 744		while (!list_empty(head)) {
 745			struct rpc_task *task;
 746			task = list_first_entry(head,
 747					struct rpc_task,
 748					u.tk_wait.list);
 749			task->tk_status = status;
 750			rpc_wake_up_task_queue_locked(queue, task);
 751		}
 752		if (head == &queue->tasks[0])
 753			break;
 754		head--;
 755	}
 756	spin_unlock(&queue->lock);
 757}
 758EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 759
 760static void __rpc_queue_timer_fn(struct work_struct *work)
 761{
 762	struct rpc_wait_queue *queue = container_of(work,
 763			struct rpc_wait_queue,
 764			timer_list.dwork.work);
 765	struct rpc_task *task, *n;
 766	unsigned long expires, now, timeo;
 767
 768	spin_lock(&queue->lock);
 769	expires = now = jiffies;
 770	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 771		timeo = task->tk_timeout;
 772		if (time_after_eq(now, timeo)) {
 773			dprintk("RPC: %5u timeout\n", task->tk_pid);
 774			task->tk_status = -ETIMEDOUT;
 775			rpc_wake_up_task_queue_locked(queue, task);
 776			continue;
 777		}
 778		if (expires == now || time_after(expires, timeo))
 779			expires = timeo;
 780	}
 781	if (!list_empty(&queue->timer_list.list))
 782		rpc_set_queue_timer(queue, expires);
 783	spin_unlock(&queue->lock);
 784}
 785
 786static void __rpc_atrun(struct rpc_task *task)
 787{
 788	if (task->tk_status == -ETIMEDOUT)
 789		task->tk_status = 0;
 790}
 791
 792/*
 793 * Run a task at a later time
 794 */
 795void rpc_delay(struct rpc_task *task, unsigned long delay)
 796{
 797	rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
 798}
 799EXPORT_SYMBOL_GPL(rpc_delay);
 800
 801/*
 802 * Helper to call task->tk_ops->rpc_call_prepare
 803 */
 804void rpc_prepare_task(struct rpc_task *task)
 805{
 806	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 807}
 808
 809static void
 810rpc_init_task_statistics(struct rpc_task *task)
 811{
 812	/* Initialize retry counters */
 813	task->tk_garb_retry = 2;
 814	task->tk_cred_retry = 2;
 815	task->tk_rebind_retry = 2;
 816
 817	/* starting timestamp */
 818	task->tk_start = ktime_get();
 819}
 820
 821static void
 822rpc_reset_task_statistics(struct rpc_task *task)
 823{
 824	task->tk_timeouts = 0;
 825	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
 826	rpc_init_task_statistics(task);
 827}
 828
 829/*
 830 * Helper that calls task->tk_ops->rpc_call_done if it exists
 831 */
 832void rpc_exit_task(struct rpc_task *task)
 833{
 834	trace_rpc_task_end(task, task->tk_action);
 835	task->tk_action = NULL;
 836	if (task->tk_ops->rpc_count_stats)
 837		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
 838	else if (task->tk_client)
 839		rpc_count_iostats(task, task->tk_client->cl_metrics);
 840	if (task->tk_ops->rpc_call_done != NULL) {
 
 841		task->tk_ops->rpc_call_done(task, task->tk_calldata);
 842		if (task->tk_action != NULL) {
 843			/* Always release the RPC slot and buffer memory */
 844			xprt_release(task);
 845			rpc_reset_task_statistics(task);
 846		}
 847	}
 848}
 849
 850void rpc_signal_task(struct rpc_task *task)
 851{
 852	struct rpc_wait_queue *queue;
 853
 854	if (!RPC_IS_ACTIVATED(task))
 855		return;
 856
 
 
 857	trace_rpc_task_signalled(task, task->tk_action);
 858	set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
 859	smp_mb__after_atomic();
 860	queue = READ_ONCE(task->tk_waitqueue);
 861	if (queue)
 862		rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
 
 
 
 
 
 
 
 
 
 
 
 863}
 864
 865void rpc_exit(struct rpc_task *task, int status)
 866{
 867	task->tk_status = status;
 868	task->tk_action = rpc_exit_task;
 869	rpc_wake_up_queued_task(task->tk_waitqueue, task);
 870}
 871EXPORT_SYMBOL_GPL(rpc_exit);
 872
 873void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 874{
 875	if (ops->rpc_release != NULL)
 876		ops->rpc_release(calldata);
 877}
 878
 
 
 
 
 
 
 
 
 
 879/*
 880 * This is the RPC `scheduler' (or rather, the finite state machine).
 881 */
 882static void __rpc_execute(struct rpc_task *task)
 883{
 884	struct rpc_wait_queue *queue;
 885	int task_is_async = RPC_IS_ASYNC(task);
 886	int status = 0;
 887
 888	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 889			task->tk_pid, task->tk_flags);
 890
 891	WARN_ON_ONCE(RPC_IS_QUEUED(task));
 892	if (RPC_IS_QUEUED(task))
 893		return;
 894
 895	for (;;) {
 896		void (*do_action)(struct rpc_task *);
 897
 898		/*
 899		 * Perform the next FSM step or a pending callback.
 900		 *
 901		 * tk_action may be NULL if the task has been killed.
 902		 * In particular, note that rpc_killall_tasks may
 903		 * do this at any time, so beware when dereferencing.
 904		 */
 905		do_action = task->tk_action;
 
 
 
 
 
 
 
 
 906		if (task->tk_callback) {
 907			do_action = task->tk_callback;
 908			task->tk_callback = NULL;
 909		}
 910		if (!do_action)
 911			break;
 
 
 
 
 912		trace_rpc_task_run_action(task, do_action);
 913		do_action(task);
 914
 915		/*
 916		 * Lockless check for whether task is sleeping or not.
 917		 */
 918		if (!RPC_IS_QUEUED(task))
 
 919			continue;
 920
 921		/*
 922		 * Signalled tasks should exit rather than sleep.
 923		 */
 924		if (RPC_SIGNALLED(task)) {
 925			task->tk_rpc_status = -ERESTARTSYS;
 926			rpc_exit(task, -ERESTARTSYS);
 927		}
 928
 929		/*
 930		 * The queue->lock protects against races with
 931		 * rpc_make_runnable().
 932		 *
 933		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 934		 * rpc_task, rpc_make_runnable() can assign it to a
 935		 * different workqueue. We therefore cannot assume that the
 936		 * rpc_task pointer may still be dereferenced.
 937		 */
 938		queue = task->tk_waitqueue;
 939		spin_lock(&queue->lock);
 940		if (!RPC_IS_QUEUED(task)) {
 941			spin_unlock(&queue->lock);
 942			continue;
 943		}
 
 
 
 
 
 
 944		rpc_clear_running(task);
 945		spin_unlock(&queue->lock);
 946		if (task_is_async)
 947			return;
 948
 949		/* sync task: sleep here */
 950		dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 951		status = out_of_line_wait_on_bit(&task->tk_runstate,
 952				RPC_TASK_QUEUED, rpc_wait_bit_killable,
 953				TASK_KILLABLE);
 954		if (status < 0) {
 955			/*
 956			 * When a sync task receives a signal, it exits with
 957			 * -ERESTARTSYS. In order to catch any callbacks that
 958			 * clean up after sleeping on some queue, we don't
 959			 * break the loop here, but go around once more.
 960			 */
 961			trace_rpc_task_signalled(task, task->tk_action);
 962			set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
 963			task->tk_rpc_status = -ERESTARTSYS;
 964			rpc_exit(task, -ERESTARTSYS);
 965		}
 966		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 967	}
 968
 969	dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 970			task->tk_status);
 971	/* Release all resources associated with the task */
 972	rpc_release_task(task);
 
 
 973}
 974
 975/*
 976 * User-visible entry point to the scheduler.
 977 *
 978 * This may be called recursively if e.g. an async NFS task updates
 979 * the attributes and finds that dirty pages must be flushed.
 980 * NOTE: Upon exit of this function the task is guaranteed to be
 981 *	 released. In particular note that tk_release() will have
 982 *	 been called, so your task memory may have been freed.
 983 */
 984void rpc_execute(struct rpc_task *task)
 985{
 986	bool is_async = RPC_IS_ASYNC(task);
 987
 988	rpc_set_active(task);
 989	rpc_make_runnable(rpciod_workqueue, task);
 990	if (!is_async)
 
 991		__rpc_execute(task);
 
 
 992}
 993
 994static void rpc_async_schedule(struct work_struct *work)
 995{
 996	unsigned int pflags = memalloc_nofs_save();
 997
 998	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 999	memalloc_nofs_restore(pflags);
1000}
1001
1002/**
1003 * rpc_malloc - allocate RPC buffer resources
1004 * @task: RPC task
1005 *
1006 * A single memory region is allocated, which is split between the
1007 * RPC call and RPC reply that this task is being used for. When
1008 * this RPC is retired, the memory is released by calling rpc_free.
1009 *
1010 * To prevent rpciod from hanging, this allocator never sleeps,
1011 * returning -ENOMEM and suppressing warning if the request cannot
1012 * be serviced immediately. The caller can arrange to sleep in a
1013 * way that is safe for rpciod.
1014 *
1015 * Most requests are 'small' (under 2KiB) and can be serviced from a
1016 * mempool, ensuring that NFS reads and writes can always proceed,
1017 * and that there is good locality of reference for these buffers.
1018 */
1019int rpc_malloc(struct rpc_task *task)
1020{
1021	struct rpc_rqst *rqst = task->tk_rqstp;
1022	size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1023	struct rpc_buffer *buf;
1024	gfp_t gfp = GFP_NOFS;
1025
1026	if (RPC_IS_SWAPPER(task))
1027		gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
1028
1029	size += sizeof(struct rpc_buffer);
1030	if (size <= RPC_BUFFER_MAXSIZE)
1031		buf = mempool_alloc(rpc_buffer_mempool, gfp);
1032	else
 
 
 
1033		buf = kmalloc(size, gfp);
1034
1035	if (!buf)
1036		return -ENOMEM;
1037
1038	buf->len = size;
1039	dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1040			task->tk_pid, size, buf);
1041	rqst->rq_buffer = buf->data;
1042	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1043	return 0;
1044}
1045EXPORT_SYMBOL_GPL(rpc_malloc);
1046
1047/**
1048 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1049 * @task: RPC task
1050 *
1051 */
1052void rpc_free(struct rpc_task *task)
1053{
1054	void *buffer = task->tk_rqstp->rq_buffer;
1055	size_t size;
1056	struct rpc_buffer *buf;
1057
1058	buf = container_of(buffer, struct rpc_buffer, data);
1059	size = buf->len;
1060
1061	dprintk("RPC:       freeing buffer of size %zu at %p\n",
1062			size, buf);
1063
1064	if (size <= RPC_BUFFER_MAXSIZE)
1065		mempool_free(buf, rpc_buffer_mempool);
1066	else
1067		kfree(buf);
1068}
1069EXPORT_SYMBOL_GPL(rpc_free);
1070
1071/*
1072 * Creation and deletion of RPC task structures
1073 */
1074static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1075{
1076	memset(task, 0, sizeof(*task));
1077	atomic_set(&task->tk_count, 1);
1078	task->tk_flags  = task_setup_data->flags;
1079	task->tk_ops = task_setup_data->callback_ops;
1080	task->tk_calldata = task_setup_data->callback_data;
1081	INIT_LIST_HEAD(&task->tk_task);
1082
1083	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1084	task->tk_owner = current->tgid;
1085
1086	/* Initialize workqueue for async tasks */
1087	task->tk_workqueue = task_setup_data->workqueue;
1088
1089	task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1090			xprt_get(task_setup_data->rpc_xprt));
1091
1092	task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1093
1094	if (task->tk_ops->rpc_call_prepare != NULL)
1095		task->tk_action = rpc_prepare_task;
1096
1097	rpc_init_task_statistics(task);
1098
1099	dprintk("RPC:       new task initialized, procpid %u\n",
1100				task_pid_nr(current));
1101}
1102
1103static struct rpc_task *
1104rpc_alloc_task(void)
1105{
1106	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 
 
 
 
 
1107}
1108
1109/*
1110 * Create a new task for the specified client.
1111 */
1112struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1113{
1114	struct rpc_task	*task = setup_data->task;
1115	unsigned short flags = 0;
1116
1117	if (task == NULL) {
1118		task = rpc_alloc_task();
 
 
 
 
 
1119		flags = RPC_TASK_DYNAMIC;
1120	}
1121
1122	rpc_init_task(task, setup_data);
1123	task->tk_flags |= flags;
1124	dprintk("RPC:       allocated task %p\n", task);
1125	return task;
1126}
1127
1128/*
1129 * rpc_free_task - release rpc task and perform cleanups
1130 *
1131 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1132 * in order to work around a workqueue dependency issue.
1133 *
1134 * Tejun Heo states:
1135 * "Workqueue currently considers two work items to be the same if they're
1136 * on the same address and won't execute them concurrently - ie. it
1137 * makes a work item which is queued again while being executed wait
1138 * for the previous execution to complete.
1139 *
1140 * If a work function frees the work item, and then waits for an event
1141 * which should be performed by another work item and *that* work item
1142 * recycles the freed work item, it can create a false dependency loop.
1143 * There really is no reliable way to detect this short of verifying
1144 * every memory free."
1145 *
1146 */
1147static void rpc_free_task(struct rpc_task *task)
1148{
1149	unsigned short tk_flags = task->tk_flags;
1150
1151	put_rpccred(task->tk_op_cred);
1152	rpc_release_calldata(task->tk_ops, task->tk_calldata);
1153
1154	if (tk_flags & RPC_TASK_DYNAMIC) {
1155		dprintk("RPC: %5u freeing task\n", task->tk_pid);
1156		mempool_free(task, rpc_task_mempool);
1157	}
1158}
1159
1160static void rpc_async_release(struct work_struct *work)
1161{
1162	unsigned int pflags = memalloc_nofs_save();
1163
1164	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1165	memalloc_nofs_restore(pflags);
1166}
1167
1168static void rpc_release_resources_task(struct rpc_task *task)
1169{
1170	xprt_release(task);
1171	if (task->tk_msg.rpc_cred) {
1172		if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1173			put_cred(task->tk_msg.rpc_cred);
1174		task->tk_msg.rpc_cred = NULL;
1175	}
1176	rpc_task_release_client(task);
1177}
1178
1179static void rpc_final_put_task(struct rpc_task *task,
1180		struct workqueue_struct *q)
1181{
1182	if (q != NULL) {
1183		INIT_WORK(&task->u.tk_work, rpc_async_release);
1184		queue_work(q, &task->u.tk_work);
1185	} else
1186		rpc_free_task(task);
1187}
1188
1189static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1190{
1191	if (atomic_dec_and_test(&task->tk_count)) {
1192		rpc_release_resources_task(task);
1193		rpc_final_put_task(task, q);
1194	}
1195}
1196
1197void rpc_put_task(struct rpc_task *task)
1198{
1199	rpc_do_put_task(task, NULL);
1200}
1201EXPORT_SYMBOL_GPL(rpc_put_task);
1202
1203void rpc_put_task_async(struct rpc_task *task)
1204{
1205	rpc_do_put_task(task, task->tk_workqueue);
1206}
1207EXPORT_SYMBOL_GPL(rpc_put_task_async);
1208
1209static void rpc_release_task(struct rpc_task *task)
1210{
1211	dprintk("RPC: %5u release task\n", task->tk_pid);
1212
1213	WARN_ON_ONCE(RPC_IS_QUEUED(task));
1214
1215	rpc_release_resources_task(task);
1216
1217	/*
1218	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1219	 * so it should be safe to use task->tk_count as a test for whether
1220	 * or not any other processes still hold references to our rpc_task.
1221	 */
1222	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1223		/* Wake up anyone who may be waiting for task completion */
1224		if (!rpc_complete_task(task))
1225			return;
1226	} else {
1227		if (!atomic_dec_and_test(&task->tk_count))
1228			return;
1229	}
1230	rpc_final_put_task(task, task->tk_workqueue);
1231}
1232
1233int rpciod_up(void)
1234{
1235	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1236}
1237
1238void rpciod_down(void)
1239{
1240	module_put(THIS_MODULE);
1241}
1242
1243/*
1244 * Start up the rpciod workqueue.
1245 */
1246static int rpciod_start(void)
1247{
1248	struct workqueue_struct *wq;
1249
1250	/*
1251	 * Create the rpciod thread and wait for it to start.
1252	 */
1253	dprintk("RPC:       creating workqueue rpciod\n");
1254	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1255	if (!wq)
1256		goto out_failed;
1257	rpciod_workqueue = wq;
1258	/* Note: highpri because network receive is latency sensitive */
1259	wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1260	if (!wq)
1261		goto free_rpciod;
1262	xprtiod_workqueue = wq;
1263	return 1;
1264free_rpciod:
1265	wq = rpciod_workqueue;
1266	rpciod_workqueue = NULL;
1267	destroy_workqueue(wq);
1268out_failed:
1269	return 0;
1270}
1271
1272static void rpciod_stop(void)
1273{
1274	struct workqueue_struct *wq = NULL;
1275
1276	if (rpciod_workqueue == NULL)
1277		return;
1278	dprintk("RPC:       destroying workqueue rpciod\n");
1279
1280	wq = rpciod_workqueue;
1281	rpciod_workqueue = NULL;
1282	destroy_workqueue(wq);
1283	wq = xprtiod_workqueue;
1284	xprtiod_workqueue = NULL;
1285	destroy_workqueue(wq);
1286}
1287
1288void
1289rpc_destroy_mempool(void)
1290{
1291	rpciod_stop();
1292	mempool_destroy(rpc_buffer_mempool);
1293	mempool_destroy(rpc_task_mempool);
1294	kmem_cache_destroy(rpc_task_slabp);
1295	kmem_cache_destroy(rpc_buffer_slabp);
1296	rpc_destroy_wait_queue(&delay_queue);
1297}
1298
1299int
1300rpc_init_mempool(void)
1301{
1302	/*
1303	 * The following is not strictly a mempool initialisation,
1304	 * but there is no harm in doing it here
1305	 */
1306	rpc_init_wait_queue(&delay_queue, "delayq");
1307	if (!rpciod_start())
1308		goto err_nomem;
1309
1310	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1311					     sizeof(struct rpc_task),
1312					     0, SLAB_HWCACHE_ALIGN,
1313					     NULL);
1314	if (!rpc_task_slabp)
1315		goto err_nomem;
1316	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1317					     RPC_BUFFER_MAXSIZE,
1318					     0, SLAB_HWCACHE_ALIGN,
1319					     NULL);
1320	if (!rpc_buffer_slabp)
1321		goto err_nomem;
1322	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1323						    rpc_task_slabp);
1324	if (!rpc_task_mempool)
1325		goto err_nomem;
1326	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1327						      rpc_buffer_slabp);
1328	if (!rpc_buffer_mempool)
1329		goto err_nomem;
1330	return 0;
1331err_nomem:
1332	rpc_destroy_mempool();
1333	return -ENOMEM;
1334}
v6.2
   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)
  39static struct kmem_cache	*rpc_task_slabp __read_mostly;
  40static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
  41static mempool_t	*rpc_task_mempool __read_mostly;
  42static mempool_t	*rpc_buffer_mempool __read_mostly;
  43
  44static void			rpc_async_schedule(struct work_struct *);
  45static void			 rpc_release_task(struct rpc_task *task);
  46static void __rpc_queue_timer_fn(struct work_struct *);
  47
  48/*
  49 * RPC tasks sit here while waiting for conditions to improve.
  50 */
  51static struct rpc_wait_queue delay_queue;
  52
  53/*
  54 * rpciod-related stuff
  55 */
  56struct workqueue_struct *rpciod_workqueue __read_mostly;
  57struct workqueue_struct *xprtiod_workqueue __read_mostly;
  58EXPORT_SYMBOL_GPL(xprtiod_workqueue);
  59
  60gfp_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}
  66EXPORT_SYMBOL_GPL(rpc_task_gfp_mask);
  67
  68bool 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
  75unsigned long
  76rpc_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}
  87EXPORT_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 */
  94static 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
 105static void
 106rpc_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 */
 120static 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
 130static 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
 138static 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 */
 146static 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 */
 168static 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 */
 193static 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 */
 205static 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 */
 224static 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 */
 233static 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
 243static 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
 259void 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}
 263EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 264
 265void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 266{
 267	__rpc_init_priority_wait_queue(queue, qname, 1);
 268}
 269EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 270
 271void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 272{
 273	cancel_delayed_work_sync(&queue->timer_list.dwork);
 274}
 275EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 276
 277static 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)
 286static 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
 301static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 302{
 303}
 304#endif
 305
 306static 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 */
 317static 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 */
 343int 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}
 348EXPORT_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 */
 361static 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		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 374}
 375
 376/*
 377 * Prepare for sleeping on a wait queue.
 378 * By always appending tasks to the list we ensure FIFO behavior.
 379 * NB: An RPC task will only receive interrupt-driven events as long
 380 * as it's on a wait queue.
 381 */
 382static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
 383		struct rpc_task *task,
 384		unsigned char queue_priority)
 385{
 
 
 
 386	trace_rpc_task_sleep(task, q);
 387
 388	__rpc_add_wait_queue(q, task, queue_priority);
 
 389}
 390
 391static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 392		struct rpc_task *task,
 393		unsigned char queue_priority)
 394{
 395	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 396		return;
 397	__rpc_do_sleep_on_priority(q, task, queue_priority);
 398}
 399
 400static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 401		struct rpc_task *task, unsigned long timeout,
 402		unsigned char queue_priority)
 403{
 404	if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 405		return;
 406	if (time_is_after_jiffies(timeout)) {
 407		__rpc_do_sleep_on_priority(q, task, queue_priority);
 408		__rpc_add_timer(q, task, timeout);
 409	} else
 410		task->tk_status = -ETIMEDOUT;
 411}
 412
 413static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
 414{
 415	if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
 416		task->tk_callback = action;
 417}
 418
 419static bool rpc_sleep_check_activated(struct rpc_task *task)
 420{
 421	/* We shouldn't ever put an inactive task to sleep */
 422	if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
 423		task->tk_status = -EIO;
 424		rpc_put_task_async(task);
 425		return false;
 426	}
 427	return true;
 428}
 429
 430void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
 431				rpc_action action, unsigned long timeout)
 432{
 433	if (!rpc_sleep_check_activated(task))
 434		return;
 435
 436	rpc_set_tk_callback(task, action);
 437
 438	/*
 439	 * Protect the queue operations.
 440	 */
 441	spin_lock(&q->lock);
 442	__rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
 443	spin_unlock(&q->lock);
 444}
 445EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
 446
 447void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 448				rpc_action action)
 449{
 450	if (!rpc_sleep_check_activated(task))
 451		return;
 452
 453	rpc_set_tk_callback(task, action);
 454
 455	WARN_ON_ONCE(task->tk_timeout != 0);
 456	/*
 457	 * Protect the queue operations.
 458	 */
 459	spin_lock(&q->lock);
 460	__rpc_sleep_on_priority(q, task, task->tk_priority);
 461	spin_unlock(&q->lock);
 462}
 463EXPORT_SYMBOL_GPL(rpc_sleep_on);
 464
 465void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 466		struct rpc_task *task, unsigned long timeout, int priority)
 467{
 468	if (!rpc_sleep_check_activated(task))
 469		return;
 470
 471	priority -= RPC_PRIORITY_LOW;
 472	/*
 473	 * Protect the queue operations.
 474	 */
 475	spin_lock(&q->lock);
 476	__rpc_sleep_on_priority_timeout(q, task, timeout, priority);
 477	spin_unlock(&q->lock);
 478}
 479EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
 480
 481void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 482		int priority)
 483{
 484	if (!rpc_sleep_check_activated(task))
 485		return;
 486
 487	WARN_ON_ONCE(task->tk_timeout != 0);
 488	priority -= RPC_PRIORITY_LOW;
 489	/*
 490	 * Protect the queue operations.
 491	 */
 492	spin_lock(&q->lock);
 493	__rpc_sleep_on_priority(q, task, priority);
 494	spin_unlock(&q->lock);
 495}
 496EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
 497
 498/**
 499 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
 500 * @wq: workqueue on which to run task
 501 * @queue: wait queue
 502 * @task: task to be woken up
 503 *
 504 * Caller must hold queue->lock, and have cleared the task queued flag.
 505 */
 506static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
 507		struct rpc_wait_queue *queue,
 508		struct rpc_task *task)
 509{
 
 
 
 510	/* Has the task been executed yet? If not, we cannot wake it up! */
 511	if (!RPC_IS_ACTIVATED(task)) {
 512		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 513		return;
 514	}
 515
 516	trace_rpc_task_wakeup(task, queue);
 517
 518	__rpc_remove_wait_queue(queue, task);
 519
 520	rpc_make_runnable(wq, task);
 
 
 521}
 522
 523/*
 524 * Wake up a queued task while the queue lock is being held
 525 */
 526static struct rpc_task *
 527rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
 528		struct rpc_wait_queue *queue, struct rpc_task *task,
 529		bool (*action)(struct rpc_task *, void *), void *data)
 530{
 531	if (RPC_IS_QUEUED(task)) {
 532		smp_rmb();
 533		if (task->tk_waitqueue == queue) {
 534			if (action == NULL || action(task, data)) {
 535				__rpc_do_wake_up_task_on_wq(wq, queue, task);
 536				return task;
 537			}
 538		}
 539	}
 540	return NULL;
 541}
 542
 543/*
 544 * Wake up a queued task while the queue lock is being held
 545 */
 546static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
 547					  struct rpc_task *task)
 548{
 549	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 550						   task, NULL, NULL);
 551}
 552
 553/*
 554 * Wake up a task on a specific queue
 555 */
 556void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 557{
 558	if (!RPC_IS_QUEUED(task))
 559		return;
 560	spin_lock(&queue->lock);
 561	rpc_wake_up_task_queue_locked(queue, task);
 562	spin_unlock(&queue->lock);
 563}
 564EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 565
 566static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
 567{
 568	task->tk_status = *(int *)status;
 569	return true;
 570}
 571
 572static void
 573rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
 574		struct rpc_task *task, int status)
 575{
 576	rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 577			task, rpc_task_action_set_status, &status);
 578}
 579
 580/**
 581 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
 582 * @queue: pointer to rpc_wait_queue
 583 * @task: pointer to rpc_task
 584 * @status: integer error value
 585 *
 586 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
 587 * set to the value of @status.
 588 */
 589void
 590rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
 591		struct rpc_task *task, int status)
 592{
 593	if (!RPC_IS_QUEUED(task))
 594		return;
 595	spin_lock(&queue->lock);
 596	rpc_wake_up_task_queue_set_status_locked(queue, task, status);
 597	spin_unlock(&queue->lock);
 598}
 599
 600/*
 601 * Wake up the next task on a priority queue.
 602 */
 603static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 604{
 605	struct list_head *q;
 606	struct rpc_task *task;
 607
 608	/*
 609	 * Service the privileged queue.
 610	 */
 611	q = &queue->tasks[RPC_NR_PRIORITY - 1];
 612	if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
 613		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 614		goto out;
 615	}
 616
 617	/*
 618	 * Service a batch of tasks from a single owner.
 619	 */
 620	q = &queue->tasks[queue->priority];
 621	if (!list_empty(q) && queue->nr) {
 622		queue->nr--;
 623		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 624		goto out;
 625	}
 626
 627	/*
 628	 * Service the next queue.
 629	 */
 630	do {
 631		if (q == &queue->tasks[0])
 632			q = &queue->tasks[queue->maxpriority];
 633		else
 634			q = q - 1;
 635		if (!list_empty(q)) {
 636			task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 637			goto new_queue;
 638		}
 639	} while (q != &queue->tasks[queue->priority]);
 640
 641	rpc_reset_waitqueue_priority(queue);
 642	return NULL;
 643
 644new_queue:
 645	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 646out:
 647	return task;
 648}
 649
 650static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
 651{
 652	if (RPC_IS_PRIORITY(queue))
 653		return __rpc_find_next_queued_priority(queue);
 654	if (!list_empty(&queue->tasks[0]))
 655		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
 656	return NULL;
 657}
 658
 659/*
 660 * Wake up the first task on the wait queue.
 661 */
 662struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
 663		struct rpc_wait_queue *queue,
 664		bool (*func)(struct rpc_task *, void *), void *data)
 665{
 666	struct rpc_task	*task = NULL;
 667
 
 
 668	spin_lock(&queue->lock);
 669	task = __rpc_find_next_queued(queue);
 670	if (task != NULL)
 671		task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
 672				task, func, data);
 673	spin_unlock(&queue->lock);
 674
 675	return task;
 676}
 677
 678/*
 679 * Wake up the first task on the wait queue.
 680 */
 681struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
 682		bool (*func)(struct rpc_task *, void *), void *data)
 683{
 684	return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
 685}
 686EXPORT_SYMBOL_GPL(rpc_wake_up_first);
 687
 688static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
 689{
 690	return true;
 691}
 692
 693/*
 694 * Wake up the next task on the wait queue.
 695*/
 696struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
 697{
 698	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
 699}
 700EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 701
 702/**
 703 * rpc_wake_up_locked - wake up all rpc_tasks
 704 * @queue: rpc_wait_queue on which the tasks are sleeping
 705 *
 706 */
 707static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
 708{
 709	struct rpc_task *task;
 710
 711	for (;;) {
 712		task = __rpc_find_next_queued(queue);
 713		if (task == NULL)
 714			break;
 715		rpc_wake_up_task_queue_locked(queue, task);
 716	}
 717}
 718
 719/**
 720 * rpc_wake_up - wake up all rpc_tasks
 721 * @queue: rpc_wait_queue on which the tasks are sleeping
 722 *
 723 * Grabs queue->lock
 724 */
 725void rpc_wake_up(struct rpc_wait_queue *queue)
 726{
 
 
 727	spin_lock(&queue->lock);
 728	rpc_wake_up_locked(queue);
 729	spin_unlock(&queue->lock);
 730}
 731EXPORT_SYMBOL_GPL(rpc_wake_up);
 732
 733/**
 734 * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
 735 * @queue: rpc_wait_queue on which the tasks are sleeping
 736 * @status: status value to set
 737 */
 738static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
 739{
 740	struct rpc_task *task;
 741
 742	for (;;) {
 743		task = __rpc_find_next_queued(queue);
 744		if (task == NULL)
 
 
 
 
 
 
 745			break;
 746		rpc_wake_up_task_queue_set_status_locked(queue, task, status);
 747	}
 
 748}
 
 749
 750/**
 751 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 752 * @queue: rpc_wait_queue on which the tasks are sleeping
 753 * @status: status value to set
 754 *
 755 * Grabs queue->lock
 756 */
 757void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 758{
 
 
 759	spin_lock(&queue->lock);
 760	rpc_wake_up_status_locked(queue, status);
 
 
 
 
 
 
 
 
 
 
 
 
 
 761	spin_unlock(&queue->lock);
 762}
 763EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 764
 765static void __rpc_queue_timer_fn(struct work_struct *work)
 766{
 767	struct rpc_wait_queue *queue = container_of(work,
 768			struct rpc_wait_queue,
 769			timer_list.dwork.work);
 770	struct rpc_task *task, *n;
 771	unsigned long expires, now, timeo;
 772
 773	spin_lock(&queue->lock);
 774	expires = now = jiffies;
 775	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 776		timeo = task->tk_timeout;
 777		if (time_after_eq(now, timeo)) {
 778			trace_rpc_task_timeout(task, task->tk_action);
 779			task->tk_status = -ETIMEDOUT;
 780			rpc_wake_up_task_queue_locked(queue, task);
 781			continue;
 782		}
 783		if (expires == now || time_after(expires, timeo))
 784			expires = timeo;
 785	}
 786	if (!list_empty(&queue->timer_list.list))
 787		rpc_set_queue_timer(queue, expires);
 788	spin_unlock(&queue->lock);
 789}
 790
 791static void __rpc_atrun(struct rpc_task *task)
 792{
 793	if (task->tk_status == -ETIMEDOUT)
 794		task->tk_status = 0;
 795}
 796
 797/*
 798 * Run a task at a later time
 799 */
 800void rpc_delay(struct rpc_task *task, unsigned long delay)
 801{
 802	rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
 803}
 804EXPORT_SYMBOL_GPL(rpc_delay);
 805
 806/*
 807 * Helper to call task->tk_ops->rpc_call_prepare
 808 */
 809void rpc_prepare_task(struct rpc_task *task)
 810{
 811	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 812}
 813
 814static void
 815rpc_init_task_statistics(struct rpc_task *task)
 816{
 817	/* Initialize retry counters */
 818	task->tk_garb_retry = 2;
 819	task->tk_cred_retry = 2;
 820	task->tk_rebind_retry = 2;
 821
 822	/* starting timestamp */
 823	task->tk_start = ktime_get();
 824}
 825
 826static void
 827rpc_reset_task_statistics(struct rpc_task *task)
 828{
 829	task->tk_timeouts = 0;
 830	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
 831	rpc_init_task_statistics(task);
 832}
 833
 834/*
 835 * Helper that calls task->tk_ops->rpc_call_done if it exists
 836 */
 837void rpc_exit_task(struct rpc_task *task)
 838{
 839	trace_rpc_task_end(task, task->tk_action);
 840	task->tk_action = NULL;
 841	if (task->tk_ops->rpc_count_stats)
 842		task->tk_ops->rpc_count_stats(task, task->tk_calldata);
 843	else if (task->tk_client)
 844		rpc_count_iostats(task, task->tk_client->cl_metrics);
 845	if (task->tk_ops->rpc_call_done != NULL) {
 846		trace_rpc_task_call_done(task, task->tk_ops->rpc_call_done);
 847		task->tk_ops->rpc_call_done(task, task->tk_calldata);
 848		if (task->tk_action != NULL) {
 849			/* Always release the RPC slot and buffer memory */
 850			xprt_release(task);
 851			rpc_reset_task_statistics(task);
 852		}
 853	}
 854}
 855
 856void rpc_signal_task(struct rpc_task *task)
 857{
 858	struct rpc_wait_queue *queue;
 859
 860	if (!RPC_IS_ACTIVATED(task))
 861		return;
 862
 863	if (!rpc_task_set_rpc_status(task, -ERESTARTSYS))
 864		return;
 865	trace_rpc_task_signalled(task, task->tk_action);
 866	set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
 867	smp_mb__after_atomic();
 868	queue = READ_ONCE(task->tk_waitqueue);
 869	if (queue)
 870		rpc_wake_up_queued_task(queue, task);
 871}
 872
 873void rpc_task_try_cancel(struct rpc_task *task, int error)
 874{
 875	struct rpc_wait_queue *queue;
 876
 877	if (!rpc_task_set_rpc_status(task, error))
 878		return;
 879	queue = READ_ONCE(task->tk_waitqueue);
 880	if (queue)
 881		rpc_wake_up_queued_task(queue, task);
 882}
 883
 884void rpc_exit(struct rpc_task *task, int status)
 885{
 886	task->tk_status = status;
 887	task->tk_action = rpc_exit_task;
 888	rpc_wake_up_queued_task(task->tk_waitqueue, task);
 889}
 890EXPORT_SYMBOL_GPL(rpc_exit);
 891
 892void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 893{
 894	if (ops->rpc_release != NULL)
 895		ops->rpc_release(calldata);
 896}
 897
 898static bool xprt_needs_memalloc(struct rpc_xprt *xprt, struct rpc_task *tk)
 899{
 900	if (!xprt)
 901		return false;
 902	if (!atomic_read(&xprt->swapper))
 903		return false;
 904	return test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == tk;
 905}
 906
 907/*
 908 * This is the RPC `scheduler' (or rather, the finite state machine).
 909 */
 910static void __rpc_execute(struct rpc_task *task)
 911{
 912	struct rpc_wait_queue *queue;
 913	int task_is_async = RPC_IS_ASYNC(task);
 914	int status = 0;
 915	unsigned long pflags = current->flags;
 
 
 916
 917	WARN_ON_ONCE(RPC_IS_QUEUED(task));
 918	if (RPC_IS_QUEUED(task))
 919		return;
 920
 921	for (;;) {
 922		void (*do_action)(struct rpc_task *);
 923
 924		/*
 925		 * Perform the next FSM step or a pending callback.
 926		 *
 927		 * tk_action may be NULL if the task has been killed.
 
 
 928		 */
 929		do_action = task->tk_action;
 930		/* Tasks with an RPC error status should exit */
 931		if (do_action != rpc_exit_task &&
 932		    (status = READ_ONCE(task->tk_rpc_status)) != 0) {
 933			task->tk_status = status;
 934			if (do_action != NULL)
 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);
1004out:
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 */
1017void 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
1030static 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 */
1055int 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}
1079EXPORT_SYMBOL_GPL(rpc_malloc);
1080
1081/**
1082 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1083 * @task: RPC task
1084 *
1085 */
1086void 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}
1100EXPORT_SYMBOL_GPL(rpc_free);
1101
1102/*
1103 * Creation and deletion of RPC task structures
1104 */
1105static 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
1131static 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 */
1144struct 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 */
1183static 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
1194static 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
1202static 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
1213static 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
1223static 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
1231void rpc_put_task(struct rpc_task *task)
1232{
1233	rpc_do_put_task(task, NULL);
1234}
1235EXPORT_SYMBOL_GPL(rpc_put_task);
1236
1237void rpc_put_task_async(struct rpc_task *task)
1238{
1239	rpc_do_put_task(task, task->tk_workqueue);
1240}
1241EXPORT_SYMBOL_GPL(rpc_put_task_async);
1242
1243static 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
1265int rpciod_up(void)
1266{
1267	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1268}
1269
1270void rpciod_down(void)
1271{
1272	module_put(THIS_MODULE);
1273}
1274
1275/*
1276 * Start up the rpciod workqueue.
1277 */
1278static 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;
1294free_rpciod:
1295	wq = rpciod_workqueue;
1296	rpciod_workqueue = NULL;
1297	destroy_workqueue(wq);
1298out_failed:
1299	return 0;
1300}
1301
1302static 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
1317void
1318rpc_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
1328int
1329rpc_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;
1360err_nomem:
1361	rpc_destroy_mempool();
1362	return -ENOMEM;
1363}