Linux Audio

Check our new training course

Loading...
v4.6
   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#include <linux/freezer.h>
  22
  23#include <linux/sunrpc/clnt.h>
  24
  25#include "sunrpc.h"
  26
  27#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  28#define RPCDBG_FACILITY		RPCDBG_SCHED
  29#endif
  30
  31#define CREATE_TRACE_POINTS
  32#include <trace/events/sunrpc.h>
  33
  34/*
  35 * RPC slabs and memory pools
  36 */
  37#define RPC_BUFFER_MAXSIZE	(2048)
  38#define RPC_BUFFER_POOLSIZE	(8)
  39#define RPC_TASK_POOLSIZE	(8)
  40static struct kmem_cache	*rpc_task_slabp __read_mostly;
  41static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
  42static mempool_t	*rpc_task_mempool __read_mostly;
  43static mempool_t	*rpc_buffer_mempool __read_mostly;
  44
  45static void			rpc_async_schedule(struct work_struct *);
  46static void			 rpc_release_task(struct rpc_task *task);
  47static void __rpc_queue_timer_fn(unsigned long ptr);
  48
  49/*
  50 * RPC tasks sit here while waiting for conditions to improve.
  51 */
  52static struct rpc_wait_queue delay_queue;
  53
  54/*
  55 * rpciod-related stuff
  56 */
  57struct workqueue_struct *rpciod_workqueue;
  58
  59/*
  60 * Disable the timer for a given RPC task. Should be called with
  61 * queue->lock and bh_disabled in order to avoid races within
  62 * rpc_run_timer().
  63 */
  64static void
  65__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  66{
  67	if (task->tk_timeout == 0)
  68		return;
  69	dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  70	task->tk_timeout = 0;
  71	list_del(&task->u.tk_wait.timer_list);
  72	if (list_empty(&queue->timer_list.list))
  73		del_timer(&queue->timer_list.timer);
  74}
  75
  76static void
  77rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  78{
  79	queue->timer_list.expires = expires;
  80	mod_timer(&queue->timer_list.timer, expires);
  81}
  82
  83/*
  84 * Set up a timer for the current task.
  85 */
  86static void
  87__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  88{
  89	if (!task->tk_timeout)
  90		return;
  91
  92	dprintk("RPC: %5u setting alarm for %u ms\n",
  93		task->tk_pid, jiffies_to_msecs(task->tk_timeout));
  94
  95	task->u.tk_wait.expires = jiffies + task->tk_timeout;
  96	if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
  97		rpc_set_queue_timer(queue, task->u.tk_wait.expires);
  98	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
  99}
 100
 101static void rpc_rotate_queue_owner(struct rpc_wait_queue *queue)
 102{
 103	struct list_head *q = &queue->tasks[queue->priority];
 104	struct rpc_task *task;
 105
 106	if (!list_empty(q)) {
 107		task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 108		if (task->tk_owner == queue->owner)
 109			list_move_tail(&task->u.tk_wait.list, q);
 110	}
 111}
 112
 113static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 114{
 115	if (queue->priority != priority) {
 116		/* Fairness: rotate the list when changing priority */
 117		rpc_rotate_queue_owner(queue);
 118		queue->priority = priority;
 119	}
 120}
 121
 122static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 123{
 124	queue->owner = pid;
 125	queue->nr = RPC_BATCH_COUNT;
 126}
 127
 128static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 129{
 130	rpc_set_waitqueue_priority(queue, queue->maxpriority);
 131	rpc_set_waitqueue_owner(queue, 0);
 132}
 133
 134/*
 135 * Add new request to a priority queue.
 136 */
 137static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
 138		struct rpc_task *task,
 139		unsigned char queue_priority)
 140{
 141	struct list_head *q;
 142	struct rpc_task *t;
 143
 144	INIT_LIST_HEAD(&task->u.tk_wait.links);
 145	if (unlikely(queue_priority > queue->maxpriority))
 146		queue_priority = queue->maxpriority;
 147	if (queue_priority > queue->priority)
 148		rpc_set_waitqueue_priority(queue, queue_priority);
 149	q = &queue->tasks[queue_priority];
 
 
 150	list_for_each_entry(t, q, u.tk_wait.list) {
 151		if (t->tk_owner == task->tk_owner) {
 152			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 153			return;
 154		}
 155	}
 156	list_add_tail(&task->u.tk_wait.list, q);
 157}
 158
 159/*
 160 * Add new request to wait queue.
 161 *
 162 * Swapper tasks always get inserted at the head of the queue.
 163 * This should avoid many nasty memory deadlocks and hopefully
 164 * improve overall performance.
 165 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 166 */
 167static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
 168		struct rpc_task *task,
 169		unsigned char queue_priority)
 170{
 171	WARN_ON_ONCE(RPC_IS_QUEUED(task));
 172	if (RPC_IS_QUEUED(task))
 173		return;
 174
 175	if (RPC_IS_PRIORITY(queue))
 176		__rpc_add_wait_queue_priority(queue, task, queue_priority);
 177	else if (RPC_IS_SWAPPER(task))
 178		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 179	else
 180		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 181	task->tk_waitqueue = queue;
 182	queue->qlen++;
 183	/* barrier matches the read in rpc_wake_up_task_queue_locked() */
 184	smp_wmb();
 185	rpc_set_queued(task);
 186
 187	dprintk("RPC: %5u added to queue %p \"%s\"\n",
 188			task->tk_pid, queue, rpc_qname(queue));
 189}
 190
 191/*
 192 * Remove request from a priority queue.
 193 */
 194static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 195{
 196	struct rpc_task *t;
 197
 198	if (!list_empty(&task->u.tk_wait.links)) {
 199		t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 200		list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 201		list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 202	}
 203}
 204
 205/*
 206 * Remove request from queue.
 207 * Note: must be called with spin lock held.
 208 */
 209static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 210{
 211	__rpc_disable_timer(queue, task);
 212	if (RPC_IS_PRIORITY(queue))
 213		__rpc_remove_wait_queue_priority(task);
 214	list_del(&task->u.tk_wait.list);
 215	queue->qlen--;
 216	dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 217			task->tk_pid, queue, rpc_qname(queue));
 218}
 219
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 220static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 221{
 222	int i;
 223
 224	spin_lock_init(&queue->lock);
 225	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 226		INIT_LIST_HEAD(&queue->tasks[i]);
 227	queue->maxpriority = nr_queues - 1;
 228	rpc_reset_waitqueue_priority(queue);
 229	queue->qlen = 0;
 230	setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 231	INIT_LIST_HEAD(&queue->timer_list.list);
 232	rpc_assign_waitqueue_name(queue, qname);
 233}
 234
 235void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 236{
 237	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 238}
 239EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 240
 241void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 242{
 243	__rpc_init_priority_wait_queue(queue, qname, 1);
 244}
 245EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 246
 247void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 248{
 249	del_timer_sync(&queue->timer_list.timer);
 250}
 251EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 252
 253static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
 254{
 255	freezable_schedule_unsafe();
 256	if (signal_pending_state(mode, current))
 257		return -ERESTARTSYS;
 
 258	return 0;
 259}
 260
 261#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
 262static void rpc_task_set_debuginfo(struct rpc_task *task)
 263{
 264	static atomic_t rpc_pid;
 265
 266	task->tk_pid = atomic_inc_return(&rpc_pid);
 267}
 268#else
 269static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 270{
 271}
 272#endif
 273
 274static void rpc_set_active(struct rpc_task *task)
 275{
 276	trace_rpc_task_begin(task->tk_client, task, NULL);
 277
 278	rpc_task_set_debuginfo(task);
 279	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 280}
 281
 282/*
 283 * Mark an RPC call as having completed by clearing the 'active' bit
 284 * and then waking up all tasks that were sleeping.
 285 */
 286static int rpc_complete_task(struct rpc_task *task)
 287{
 288	void *m = &task->tk_runstate;
 289	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
 290	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
 291	unsigned long flags;
 292	int ret;
 293
 294	trace_rpc_task_complete(task->tk_client, task, NULL);
 295
 296	spin_lock_irqsave(&wq->lock, flags);
 297	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 298	ret = atomic_dec_and_test(&task->tk_count);
 299	if (waitqueue_active(wq))
 300		__wake_up_locked_key(wq, TASK_NORMAL, &k);
 301	spin_unlock_irqrestore(&wq->lock, flags);
 302	return ret;
 303}
 304
 305/*
 306 * Allow callers to wait for completion of an RPC call
 307 *
 308 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
 309 * to enforce taking of the wq->lock and hence avoid races with
 310 * rpc_complete_task().
 311 */
 312int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
 313{
 314	if (action == NULL)
 315		action = rpc_wait_bit_killable;
 316	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 317			action, TASK_KILLABLE);
 318}
 319EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 320
 321/*
 322 * Make an RPC task runnable.
 323 *
 324 * Note: If the task is ASYNC, and is being made runnable after sitting on an
 325 * rpc_wait_queue, this must be called with the queue spinlock held to protect
 326 * the wait queue operation.
 327 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
 328 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
 329 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
 330 * the RPC_TASK_RUNNING flag.
 331 */
 332static void rpc_make_runnable(struct rpc_task *task)
 333{
 334	bool need_wakeup = !rpc_test_and_set_running(task);
 335
 336	rpc_clear_queued(task);
 337	if (!need_wakeup)
 338		return;
 339	if (RPC_IS_ASYNC(task)) {
 340		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 341		queue_work(rpciod_workqueue, &task->u.tk_work);
 342	} else
 343		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 344}
 345
 346/*
 347 * Prepare for sleeping on a wait queue.
 348 * By always appending tasks to the list we ensure FIFO behavior.
 349 * NB: An RPC task will only receive interrupt-driven events as long
 350 * as it's on a wait queue.
 351 */
 352static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 353		struct rpc_task *task,
 354		rpc_action action,
 355		unsigned char queue_priority)
 356{
 357	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 358			task->tk_pid, rpc_qname(q), jiffies);
 359
 360	trace_rpc_task_sleep(task->tk_client, task, q);
 361
 362	__rpc_add_wait_queue(q, task, queue_priority);
 363
 364	WARN_ON_ONCE(task->tk_callback != NULL);
 365	task->tk_callback = action;
 366	__rpc_add_timer(q, task);
 367}
 368
 369void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 370				rpc_action action)
 371{
 372	/* We shouldn't ever put an inactive task to sleep */
 373	WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
 374	if (!RPC_IS_ACTIVATED(task)) {
 375		task->tk_status = -EIO;
 376		rpc_put_task_async(task);
 377		return;
 378	}
 379
 380	/*
 381	 * Protect the queue operations.
 382	 */
 383	spin_lock_bh(&q->lock);
 384	__rpc_sleep_on_priority(q, task, action, task->tk_priority);
 385	spin_unlock_bh(&q->lock);
 386}
 387EXPORT_SYMBOL_GPL(rpc_sleep_on);
 388
 389void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 390		rpc_action action, int priority)
 391{
 392	/* We shouldn't ever put an inactive task to sleep */
 393	WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
 394	if (!RPC_IS_ACTIVATED(task)) {
 395		task->tk_status = -EIO;
 396		rpc_put_task_async(task);
 397		return;
 398	}
 399
 400	/*
 401	 * Protect the queue operations.
 402	 */
 403	spin_lock_bh(&q->lock);
 404	__rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
 405	spin_unlock_bh(&q->lock);
 406}
 407EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
 408
 409/**
 410 * __rpc_do_wake_up_task - wake up a single rpc_task
 411 * @queue: wait queue
 412 * @task: task to be woken up
 413 *
 414 * Caller must hold queue->lock, and have cleared the task queued flag.
 415 */
 416static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 417{
 418	dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 419			task->tk_pid, jiffies);
 420
 421	/* Has the task been executed yet? If not, we cannot wake it up! */
 422	if (!RPC_IS_ACTIVATED(task)) {
 423		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 424		return;
 425	}
 426
 427	trace_rpc_task_wakeup(task->tk_client, task, queue);
 428
 429	__rpc_remove_wait_queue(queue, task);
 430
 431	rpc_make_runnable(task);
 432
 433	dprintk("RPC:       __rpc_wake_up_task done\n");
 434}
 435
 436/*
 437 * Wake up a queued task while the queue lock is being held
 438 */
 439static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 440{
 441	if (RPC_IS_QUEUED(task)) {
 442		smp_rmb();
 443		if (task->tk_waitqueue == queue)
 444			__rpc_do_wake_up_task(queue, task);
 445	}
 446}
 447
 448/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 449 * Wake up a task on a specific queue
 450 */
 451void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 452{
 453	spin_lock_bh(&queue->lock);
 454	rpc_wake_up_task_queue_locked(queue, task);
 455	spin_unlock_bh(&queue->lock);
 456}
 457EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 458
 459/*
 460 * Wake up the next task on a priority queue.
 461 */
 462static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 463{
 464	struct list_head *q;
 465	struct rpc_task *task;
 466
 467	/*
 468	 * Service a batch of tasks from a single owner.
 469	 */
 470	q = &queue->tasks[queue->priority];
 471	if (!list_empty(q)) {
 472		task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 473		if (queue->owner == task->tk_owner) {
 474			if (--queue->nr)
 475				goto out;
 476			list_move_tail(&task->u.tk_wait.list, q);
 477		}
 478		/*
 479		 * Check if we need to switch queues.
 480		 */
 481		goto new_owner;
 
 482	}
 483
 484	/*
 485	 * Service the next queue.
 486	 */
 487	do {
 488		if (q == &queue->tasks[0])
 489			q = &queue->tasks[queue->maxpriority];
 490		else
 491			q = q - 1;
 492		if (!list_empty(q)) {
 493			task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 494			goto new_queue;
 495		}
 496	} while (q != &queue->tasks[queue->priority]);
 497
 498	rpc_reset_waitqueue_priority(queue);
 499	return NULL;
 500
 501new_queue:
 502	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 503new_owner:
 504	rpc_set_waitqueue_owner(queue, task->tk_owner);
 505out:
 506	return task;
 507}
 508
 509static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
 510{
 511	if (RPC_IS_PRIORITY(queue))
 512		return __rpc_find_next_queued_priority(queue);
 513	if (!list_empty(&queue->tasks[0]))
 514		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
 515	return NULL;
 516}
 517
 518/*
 519 * Wake up the first task on the wait queue.
 520 */
 521struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
 522		bool (*func)(struct rpc_task *, void *), void *data)
 523{
 524	struct rpc_task	*task = NULL;
 525
 526	dprintk("RPC:       wake_up_first(%p \"%s\")\n",
 527			queue, rpc_qname(queue));
 528	spin_lock_bh(&queue->lock);
 529	task = __rpc_find_next_queued(queue);
 530	if (task != NULL) {
 531		if (func(task, data))
 532			rpc_wake_up_task_queue_locked(queue, task);
 533		else
 534			task = NULL;
 535	}
 536	spin_unlock_bh(&queue->lock);
 537
 538	return task;
 539}
 540EXPORT_SYMBOL_GPL(rpc_wake_up_first);
 541
 542static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
 543{
 544	return true;
 545}
 546
 547/*
 548 * Wake up the next task on the wait queue.
 549*/
 550struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
 551{
 552	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
 553}
 554EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 555
 556/**
 557 * rpc_wake_up - wake up all rpc_tasks
 558 * @queue: rpc_wait_queue on which the tasks are sleeping
 559 *
 560 * Grabs queue->lock
 561 */
 562void rpc_wake_up(struct rpc_wait_queue *queue)
 563{
 564	struct list_head *head;
 565
 566	spin_lock_bh(&queue->lock);
 567	head = &queue->tasks[queue->maxpriority];
 568	for (;;) {
 569		while (!list_empty(head)) {
 570			struct rpc_task *task;
 571			task = list_first_entry(head,
 572					struct rpc_task,
 573					u.tk_wait.list);
 574			rpc_wake_up_task_queue_locked(queue, task);
 575		}
 576		if (head == &queue->tasks[0])
 577			break;
 578		head--;
 579	}
 580	spin_unlock_bh(&queue->lock);
 581}
 582EXPORT_SYMBOL_GPL(rpc_wake_up);
 583
 584/**
 585 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 586 * @queue: rpc_wait_queue on which the tasks are sleeping
 587 * @status: status value to set
 588 *
 589 * Grabs queue->lock
 590 */
 591void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 592{
 593	struct list_head *head;
 594
 595	spin_lock_bh(&queue->lock);
 596	head = &queue->tasks[queue->maxpriority];
 597	for (;;) {
 598		while (!list_empty(head)) {
 599			struct rpc_task *task;
 600			task = list_first_entry(head,
 601					struct rpc_task,
 602					u.tk_wait.list);
 603			task->tk_status = status;
 604			rpc_wake_up_task_queue_locked(queue, task);
 605		}
 606		if (head == &queue->tasks[0])
 607			break;
 608		head--;
 609	}
 610	spin_unlock_bh(&queue->lock);
 611}
 612EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 613
 614static void __rpc_queue_timer_fn(unsigned long ptr)
 615{
 616	struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 617	struct rpc_task *task, *n;
 618	unsigned long expires, now, timeo;
 619
 620	spin_lock(&queue->lock);
 621	expires = now = jiffies;
 622	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 623		timeo = task->u.tk_wait.expires;
 624		if (time_after_eq(now, timeo)) {
 625			dprintk("RPC: %5u timeout\n", task->tk_pid);
 626			task->tk_status = -ETIMEDOUT;
 627			rpc_wake_up_task_queue_locked(queue, task);
 628			continue;
 629		}
 630		if (expires == now || time_after(expires, timeo))
 631			expires = timeo;
 632	}
 633	if (!list_empty(&queue->timer_list.list))
 634		rpc_set_queue_timer(queue, expires);
 635	spin_unlock(&queue->lock);
 636}
 637
 638static void __rpc_atrun(struct rpc_task *task)
 639{
 640	if (task->tk_status == -ETIMEDOUT)
 641		task->tk_status = 0;
 642}
 643
 644/*
 645 * Run a task at a later time
 646 */
 647void rpc_delay(struct rpc_task *task, unsigned long delay)
 648{
 649	task->tk_timeout = delay;
 650	rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 651}
 652EXPORT_SYMBOL_GPL(rpc_delay);
 653
 654/*
 655 * Helper to call task->tk_ops->rpc_call_prepare
 656 */
 657void rpc_prepare_task(struct rpc_task *task)
 658{
 659	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 660}
 661
 662static void
 663rpc_init_task_statistics(struct rpc_task *task)
 664{
 665	/* Initialize retry counters */
 666	task->tk_garb_retry = 2;
 667	task->tk_cred_retry = 2;
 668	task->tk_rebind_retry = 2;
 669
 670	/* starting timestamp */
 671	task->tk_start = ktime_get();
 672}
 673
 674static void
 675rpc_reset_task_statistics(struct rpc_task *task)
 676{
 677	task->tk_timeouts = 0;
 678	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
 679
 680	rpc_init_task_statistics(task);
 681}
 682
 683/*
 684 * Helper that calls task->tk_ops->rpc_call_done if it exists
 685 */
 686void rpc_exit_task(struct rpc_task *task)
 687{
 688	task->tk_action = NULL;
 689	if (task->tk_ops->rpc_call_done != NULL) {
 690		task->tk_ops->rpc_call_done(task, task->tk_calldata);
 691		if (task->tk_action != NULL) {
 692			WARN_ON(RPC_ASSASSINATED(task));
 693			/* Always release the RPC slot and buffer memory */
 694			xprt_release(task);
 695			rpc_reset_task_statistics(task);
 696		}
 697	}
 698}
 699
 700void rpc_exit(struct rpc_task *task, int status)
 701{
 702	task->tk_status = status;
 703	task->tk_action = rpc_exit_task;
 704	if (RPC_IS_QUEUED(task))
 705		rpc_wake_up_queued_task(task->tk_waitqueue, task);
 706}
 707EXPORT_SYMBOL_GPL(rpc_exit);
 708
 709void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 710{
 711	if (ops->rpc_release != NULL)
 712		ops->rpc_release(calldata);
 713}
 714
 715/*
 716 * This is the RPC `scheduler' (or rather, the finite state machine).
 717 */
 718static void __rpc_execute(struct rpc_task *task)
 719{
 720	struct rpc_wait_queue *queue;
 721	int task_is_async = RPC_IS_ASYNC(task);
 722	int status = 0;
 723
 724	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 725			task->tk_pid, task->tk_flags);
 726
 727	WARN_ON_ONCE(RPC_IS_QUEUED(task));
 728	if (RPC_IS_QUEUED(task))
 729		return;
 730
 731	for (;;) {
 732		void (*do_action)(struct rpc_task *);
 733
 734		/*
 735		 * Execute any pending callback first.
 736		 */
 737		do_action = task->tk_callback;
 738		task->tk_callback = NULL;
 739		if (do_action == NULL) {
 740			/*
 741			 * Perform the next FSM step.
 742			 * tk_action may be NULL if the task has been killed.
 743			 * In particular, note that rpc_killall_tasks may
 744			 * do this at any time, so beware when dereferencing.
 745			 */
 746			do_action = task->tk_action;
 747			if (do_action == NULL)
 748				break;
 749		}
 750		trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
 751		do_action(task);
 752
 753		/*
 754		 * Lockless check for whether task is sleeping or not.
 755		 */
 756		if (!RPC_IS_QUEUED(task))
 757			continue;
 758		/*
 759		 * The queue->lock protects against races with
 760		 * rpc_make_runnable().
 761		 *
 762		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 763		 * rpc_task, rpc_make_runnable() can assign it to a
 764		 * different workqueue. We therefore cannot assume that the
 765		 * rpc_task pointer may still be dereferenced.
 766		 */
 767		queue = task->tk_waitqueue;
 768		spin_lock_bh(&queue->lock);
 769		if (!RPC_IS_QUEUED(task)) {
 770			spin_unlock_bh(&queue->lock);
 771			continue;
 772		}
 773		rpc_clear_running(task);
 774		spin_unlock_bh(&queue->lock);
 775		if (task_is_async)
 776			return;
 777
 778		/* sync task: sleep here */
 779		dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 780		status = out_of_line_wait_on_bit(&task->tk_runstate,
 781				RPC_TASK_QUEUED, rpc_wait_bit_killable,
 782				TASK_KILLABLE);
 783		if (status == -ERESTARTSYS) {
 784			/*
 785			 * When a sync task receives a signal, it exits with
 786			 * -ERESTARTSYS. In order to catch any callbacks that
 787			 * clean up after sleeping on some queue, we don't
 788			 * break the loop here, but go around once more.
 789			 */
 790			dprintk("RPC: %5u got signal\n", task->tk_pid);
 791			task->tk_flags |= RPC_TASK_KILLED;
 792			rpc_exit(task, -ERESTARTSYS);
 793		}
 
 794		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 795	}
 796
 797	dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 798			task->tk_status);
 799	/* Release all resources associated with the task */
 800	rpc_release_task(task);
 801}
 802
 803/*
 804 * User-visible entry point to the scheduler.
 805 *
 806 * This may be called recursively if e.g. an async NFS task updates
 807 * the attributes and finds that dirty pages must be flushed.
 808 * NOTE: Upon exit of this function the task is guaranteed to be
 809 *	 released. In particular note that tk_release() will have
 810 *	 been called, so your task memory may have been freed.
 811 */
 812void rpc_execute(struct rpc_task *task)
 813{
 814	bool is_async = RPC_IS_ASYNC(task);
 815
 816	rpc_set_active(task);
 817	rpc_make_runnable(task);
 818	if (!is_async)
 819		__rpc_execute(task);
 820}
 821
 822static void rpc_async_schedule(struct work_struct *work)
 823{
 
 824	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 
 825}
 826
 827/**
 828 * rpc_malloc - allocate an RPC buffer
 829 * @task: RPC task that will use this buffer
 830 * @size: requested byte size
 831 *
 832 * To prevent rpciod from hanging, this allocator never sleeps,
 833 * returning NULL and suppressing warning if the request cannot be serviced
 834 * immediately.
 835 * The caller can arrange to sleep in a way that is safe for rpciod.
 836 *
 837 * Most requests are 'small' (under 2KiB) and can be serviced from a
 838 * mempool, ensuring that NFS reads and writes can always proceed,
 839 * and that there is good locality of reference for these buffers.
 840 *
 841 * In order to avoid memory starvation triggering more writebacks of
 842 * NFS requests, we avoid using GFP_KERNEL.
 843 */
 844void *rpc_malloc(struct rpc_task *task, size_t size)
 845{
 846	struct rpc_buffer *buf;
 847	gfp_t gfp = GFP_NOIO | __GFP_NOWARN;
 848
 849	if (RPC_IS_SWAPPER(task))
 850		gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
 851
 852	size += sizeof(struct rpc_buffer);
 853	if (size <= RPC_BUFFER_MAXSIZE)
 854		buf = mempool_alloc(rpc_buffer_mempool, gfp);
 855	else
 856		buf = kmalloc(size, gfp);
 857
 858	if (!buf)
 859		return NULL;
 860
 861	buf->len = size;
 862	dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 863			task->tk_pid, size, buf);
 864	return &buf->data;
 865}
 866EXPORT_SYMBOL_GPL(rpc_malloc);
 867
 868/**
 869 * rpc_free - free buffer allocated via rpc_malloc
 870 * @buffer: buffer to free
 871 *
 872 */
 873void rpc_free(void *buffer)
 874{
 875	size_t size;
 876	struct rpc_buffer *buf;
 877
 878	if (!buffer)
 879		return;
 880
 881	buf = container_of(buffer, struct rpc_buffer, data);
 882	size = buf->len;
 883
 884	dprintk("RPC:       freeing buffer of size %zu at %p\n",
 885			size, buf);
 886
 887	if (size <= RPC_BUFFER_MAXSIZE)
 888		mempool_free(buf, rpc_buffer_mempool);
 889	else
 890		kfree(buf);
 891}
 892EXPORT_SYMBOL_GPL(rpc_free);
 893
 894/*
 895 * Creation and deletion of RPC task structures
 896 */
 897static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 898{
 899	memset(task, 0, sizeof(*task));
 900	atomic_set(&task->tk_count, 1);
 901	task->tk_flags  = task_setup_data->flags;
 902	task->tk_ops = task_setup_data->callback_ops;
 903	task->tk_calldata = task_setup_data->callback_data;
 904	INIT_LIST_HEAD(&task->tk_task);
 905
 906	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 907	task->tk_owner = current->tgid;
 908
 909	/* Initialize workqueue for async tasks */
 910	task->tk_workqueue = task_setup_data->workqueue;
 911
 912	task->tk_xprt = xprt_get(task_setup_data->rpc_xprt);
 913
 914	if (task->tk_ops->rpc_call_prepare != NULL)
 915		task->tk_action = rpc_prepare_task;
 916
 917	rpc_init_task_statistics(task);
 918
 919	dprintk("RPC:       new task initialized, procpid %u\n",
 920				task_pid_nr(current));
 921}
 922
 923static struct rpc_task *
 924rpc_alloc_task(void)
 925{
 926	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO);
 927}
 928
 929/*
 930 * Create a new task for the specified client.
 931 */
 932struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 933{
 934	struct rpc_task	*task = setup_data->task;
 935	unsigned short flags = 0;
 936
 937	if (task == NULL) {
 938		task = rpc_alloc_task();
 939		if (task == NULL) {
 940			rpc_release_calldata(setup_data->callback_ops,
 941					setup_data->callback_data);
 942			return ERR_PTR(-ENOMEM);
 943		}
 944		flags = RPC_TASK_DYNAMIC;
 945	}
 946
 947	rpc_init_task(task, setup_data);
 948	task->tk_flags |= flags;
 949	dprintk("RPC:       allocated task %p\n", task);
 950	return task;
 951}
 952
 953/*
 954 * rpc_free_task - release rpc task and perform cleanups
 955 *
 956 * Note that we free up the rpc_task _after_ rpc_release_calldata()
 957 * in order to work around a workqueue dependency issue.
 958 *
 959 * Tejun Heo states:
 960 * "Workqueue currently considers two work items to be the same if they're
 961 * on the same address and won't execute them concurrently - ie. it
 962 * makes a work item which is queued again while being executed wait
 963 * for the previous execution to complete.
 964 *
 965 * If a work function frees the work item, and then waits for an event
 966 * which should be performed by another work item and *that* work item
 967 * recycles the freed work item, it can create a false dependency loop.
 968 * There really is no reliable way to detect this short of verifying
 969 * every memory free."
 970 *
 971 */
 972static void rpc_free_task(struct rpc_task *task)
 973{
 974	unsigned short tk_flags = task->tk_flags;
 975
 976	rpc_release_calldata(task->tk_ops, task->tk_calldata);
 977
 978	if (tk_flags & RPC_TASK_DYNAMIC) {
 979		dprintk("RPC: %5u freeing task\n", task->tk_pid);
 980		mempool_free(task, rpc_task_mempool);
 981	}
 
 982}
 983
 984static void rpc_async_release(struct work_struct *work)
 985{
 986	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 987}
 988
 989static void rpc_release_resources_task(struct rpc_task *task)
 990{
 991	xprt_release(task);
 
 992	if (task->tk_msg.rpc_cred) {
 993		put_rpccred(task->tk_msg.rpc_cred);
 994		task->tk_msg.rpc_cred = NULL;
 995	}
 996	rpc_task_release_client(task);
 997}
 998
 999static void rpc_final_put_task(struct rpc_task *task,
1000		struct workqueue_struct *q)
1001{
1002	if (q != NULL) {
1003		INIT_WORK(&task->u.tk_work, rpc_async_release);
1004		queue_work(q, &task->u.tk_work);
1005	} else
1006		rpc_free_task(task);
1007}
1008
1009static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1010{
1011	if (atomic_dec_and_test(&task->tk_count)) {
1012		rpc_release_resources_task(task);
1013		rpc_final_put_task(task, q);
1014	}
1015}
1016
1017void rpc_put_task(struct rpc_task *task)
1018{
1019	rpc_do_put_task(task, NULL);
1020}
1021EXPORT_SYMBOL_GPL(rpc_put_task);
1022
1023void rpc_put_task_async(struct rpc_task *task)
1024{
1025	rpc_do_put_task(task, task->tk_workqueue);
1026}
1027EXPORT_SYMBOL_GPL(rpc_put_task_async);
1028
1029static void rpc_release_task(struct rpc_task *task)
1030{
1031	dprintk("RPC: %5u release task\n", task->tk_pid);
1032
1033	WARN_ON_ONCE(RPC_IS_QUEUED(task));
1034
1035	rpc_release_resources_task(task);
1036
1037	/*
1038	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1039	 * so it should be safe to use task->tk_count as a test for whether
1040	 * or not any other processes still hold references to our rpc_task.
1041	 */
1042	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1043		/* Wake up anyone who may be waiting for task completion */
1044		if (!rpc_complete_task(task))
1045			return;
1046	} else {
1047		if (!atomic_dec_and_test(&task->tk_count))
1048			return;
1049	}
1050	rpc_final_put_task(task, task->tk_workqueue);
1051}
1052
1053int rpciod_up(void)
1054{
1055	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1056}
1057
1058void rpciod_down(void)
1059{
1060	module_put(THIS_MODULE);
1061}
1062
1063/*
1064 * Start up the rpciod workqueue.
1065 */
1066static int rpciod_start(void)
1067{
1068	struct workqueue_struct *wq;
1069
1070	/*
1071	 * Create the rpciod thread and wait for it to start.
1072	 */
1073	dprintk("RPC:       creating workqueue rpciod\n");
1074	/* Note: highpri because network receive is latency sensitive */
1075	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1076	rpciod_workqueue = wq;
1077	return rpciod_workqueue != NULL;
1078}
1079
1080static void rpciod_stop(void)
1081{
1082	struct workqueue_struct *wq = NULL;
1083
1084	if (rpciod_workqueue == NULL)
1085		return;
1086	dprintk("RPC:       destroying workqueue rpciod\n");
1087
1088	wq = rpciod_workqueue;
1089	rpciod_workqueue = NULL;
1090	destroy_workqueue(wq);
1091}
1092
1093void
1094rpc_destroy_mempool(void)
1095{
1096	rpciod_stop();
1097	mempool_destroy(rpc_buffer_mempool);
1098	mempool_destroy(rpc_task_mempool);
1099	kmem_cache_destroy(rpc_task_slabp);
1100	kmem_cache_destroy(rpc_buffer_slabp);
 
 
 
 
1101	rpc_destroy_wait_queue(&delay_queue);
1102}
1103
1104int
1105rpc_init_mempool(void)
1106{
1107	/*
1108	 * The following is not strictly a mempool initialisation,
1109	 * but there is no harm in doing it here
1110	 */
1111	rpc_init_wait_queue(&delay_queue, "delayq");
1112	if (!rpciod_start())
1113		goto err_nomem;
1114
1115	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1116					     sizeof(struct rpc_task),
1117					     0, SLAB_HWCACHE_ALIGN,
1118					     NULL);
1119	if (!rpc_task_slabp)
1120		goto err_nomem;
1121	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1122					     RPC_BUFFER_MAXSIZE,
1123					     0, SLAB_HWCACHE_ALIGN,
1124					     NULL);
1125	if (!rpc_buffer_slabp)
1126		goto err_nomem;
1127	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1128						    rpc_task_slabp);
1129	if (!rpc_task_mempool)
1130		goto err_nomem;
1131	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1132						      rpc_buffer_slabp);
1133	if (!rpc_buffer_mempool)
1134		goto err_nomem;
1135	return 0;
1136err_nomem:
1137	rpc_destroy_mempool();
1138	return -ENOMEM;
1139}
v3.5.6
   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#include <linux/freezer.h>
  22
  23#include <linux/sunrpc/clnt.h>
  24
  25#include "sunrpc.h"
  26
  27#ifdef RPC_DEBUG
  28#define RPCDBG_FACILITY		RPCDBG_SCHED
  29#endif
  30
  31#define CREATE_TRACE_POINTS
  32#include <trace/events/sunrpc.h>
  33
  34/*
  35 * RPC slabs and memory pools
  36 */
  37#define RPC_BUFFER_MAXSIZE	(2048)
  38#define RPC_BUFFER_POOLSIZE	(8)
  39#define RPC_TASK_POOLSIZE	(8)
  40static struct kmem_cache	*rpc_task_slabp __read_mostly;
  41static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
  42static mempool_t	*rpc_task_mempool __read_mostly;
  43static mempool_t	*rpc_buffer_mempool __read_mostly;
  44
  45static void			rpc_async_schedule(struct work_struct *);
  46static void			 rpc_release_task(struct rpc_task *task);
  47static void __rpc_queue_timer_fn(unsigned long ptr);
  48
  49/*
  50 * RPC tasks sit here while waiting for conditions to improve.
  51 */
  52static struct rpc_wait_queue delay_queue;
  53
  54/*
  55 * rpciod-related stuff
  56 */
  57struct workqueue_struct *rpciod_workqueue;
  58
  59/*
  60 * Disable the timer for a given RPC task. Should be called with
  61 * queue->lock and bh_disabled in order to avoid races within
  62 * rpc_run_timer().
  63 */
  64static void
  65__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  66{
  67	if (task->tk_timeout == 0)
  68		return;
  69	dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  70	task->tk_timeout = 0;
  71	list_del(&task->u.tk_wait.timer_list);
  72	if (list_empty(&queue->timer_list.list))
  73		del_timer(&queue->timer_list.timer);
  74}
  75
  76static void
  77rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  78{
  79	queue->timer_list.expires = expires;
  80	mod_timer(&queue->timer_list.timer, expires);
  81}
  82
  83/*
  84 * Set up a timer for the current task.
  85 */
  86static void
  87__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  88{
  89	if (!task->tk_timeout)
  90		return;
  91
  92	dprintk("RPC: %5u setting alarm for %lu ms\n",
  93			task->tk_pid, task->tk_timeout * 1000 / HZ);
  94
  95	task->u.tk_wait.expires = jiffies + task->tk_timeout;
  96	if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
  97		rpc_set_queue_timer(queue, task->u.tk_wait.expires);
  98	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
  99}
 100
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 101/*
 102 * Add new request to a priority queue.
 103 */
 104static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
 105		struct rpc_task *task,
 106		unsigned char queue_priority)
 107{
 108	struct list_head *q;
 109	struct rpc_task *t;
 110
 111	INIT_LIST_HEAD(&task->u.tk_wait.links);
 
 
 
 
 112	q = &queue->tasks[queue_priority];
 113	if (unlikely(queue_priority > queue->maxpriority))
 114		q = &queue->tasks[queue->maxpriority];
 115	list_for_each_entry(t, q, u.tk_wait.list) {
 116		if (t->tk_owner == task->tk_owner) {
 117			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 118			return;
 119		}
 120	}
 121	list_add_tail(&task->u.tk_wait.list, q);
 122}
 123
 124/*
 125 * Add new request to wait queue.
 126 *
 127 * Swapper tasks always get inserted at the head of the queue.
 128 * This should avoid many nasty memory deadlocks and hopefully
 129 * improve overall performance.
 130 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 131 */
 132static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
 133		struct rpc_task *task,
 134		unsigned char queue_priority)
 135{
 136	BUG_ON (RPC_IS_QUEUED(task));
 
 
 137
 138	if (RPC_IS_PRIORITY(queue))
 139		__rpc_add_wait_queue_priority(queue, task, queue_priority);
 140	else if (RPC_IS_SWAPPER(task))
 141		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 142	else
 143		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 144	task->tk_waitqueue = queue;
 145	queue->qlen++;
 
 
 146	rpc_set_queued(task);
 147
 148	dprintk("RPC: %5u added to queue %p \"%s\"\n",
 149			task->tk_pid, queue, rpc_qname(queue));
 150}
 151
 152/*
 153 * Remove request from a priority queue.
 154 */
 155static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 156{
 157	struct rpc_task *t;
 158
 159	if (!list_empty(&task->u.tk_wait.links)) {
 160		t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 161		list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 162		list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 163	}
 164}
 165
 166/*
 167 * Remove request from queue.
 168 * Note: must be called with spin lock held.
 169 */
 170static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 171{
 172	__rpc_disable_timer(queue, task);
 173	if (RPC_IS_PRIORITY(queue))
 174		__rpc_remove_wait_queue_priority(task);
 175	list_del(&task->u.tk_wait.list);
 176	queue->qlen--;
 177	dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 178			task->tk_pid, queue, rpc_qname(queue));
 179}
 180
 181static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 182{
 183	queue->priority = priority;
 184	queue->count = 1 << (priority * 2);
 185}
 186
 187static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 188{
 189	queue->owner = pid;
 190	queue->nr = RPC_BATCH_COUNT;
 191}
 192
 193static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 194{
 195	rpc_set_waitqueue_priority(queue, queue->maxpriority);
 196	rpc_set_waitqueue_owner(queue, 0);
 197}
 198
 199static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 200{
 201	int i;
 202
 203	spin_lock_init(&queue->lock);
 204	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 205		INIT_LIST_HEAD(&queue->tasks[i]);
 206	queue->maxpriority = nr_queues - 1;
 207	rpc_reset_waitqueue_priority(queue);
 208	queue->qlen = 0;
 209	setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 210	INIT_LIST_HEAD(&queue->timer_list.list);
 211	rpc_assign_waitqueue_name(queue, qname);
 212}
 213
 214void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 215{
 216	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 217}
 218EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 219
 220void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 221{
 222	__rpc_init_priority_wait_queue(queue, qname, 1);
 223}
 224EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 225
 226void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 227{
 228	del_timer_sync(&queue->timer_list.timer);
 229}
 230EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 231
 232static int rpc_wait_bit_killable(void *word)
 233{
 234	if (fatal_signal_pending(current))
 
 235		return -ERESTARTSYS;
 236	freezable_schedule();
 237	return 0;
 238}
 239
 240#ifdef RPC_DEBUG
 241static void rpc_task_set_debuginfo(struct rpc_task *task)
 242{
 243	static atomic_t rpc_pid;
 244
 245	task->tk_pid = atomic_inc_return(&rpc_pid);
 246}
 247#else
 248static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 249{
 250}
 251#endif
 252
 253static void rpc_set_active(struct rpc_task *task)
 254{
 255	trace_rpc_task_begin(task->tk_client, task, NULL);
 256
 257	rpc_task_set_debuginfo(task);
 258	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 259}
 260
 261/*
 262 * Mark an RPC call as having completed by clearing the 'active' bit
 263 * and then waking up all tasks that were sleeping.
 264 */
 265static int rpc_complete_task(struct rpc_task *task)
 266{
 267	void *m = &task->tk_runstate;
 268	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
 269	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
 270	unsigned long flags;
 271	int ret;
 272
 273	trace_rpc_task_complete(task->tk_client, task, NULL);
 274
 275	spin_lock_irqsave(&wq->lock, flags);
 276	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 277	ret = atomic_dec_and_test(&task->tk_count);
 278	if (waitqueue_active(wq))
 279		__wake_up_locked_key(wq, TASK_NORMAL, &k);
 280	spin_unlock_irqrestore(&wq->lock, flags);
 281	return ret;
 282}
 283
 284/*
 285 * Allow callers to wait for completion of an RPC call
 286 *
 287 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
 288 * to enforce taking of the wq->lock and hence avoid races with
 289 * rpc_complete_task().
 290 */
 291int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 292{
 293	if (action == NULL)
 294		action = rpc_wait_bit_killable;
 295	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 296			action, TASK_KILLABLE);
 297}
 298EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 299
 300/*
 301 * Make an RPC task runnable.
 302 *
 303 * Note: If the task is ASYNC, this must be called with
 304 * the spinlock held to protect the wait queue operation.
 
 
 
 
 
 305 */
 306static void rpc_make_runnable(struct rpc_task *task)
 307{
 
 
 308	rpc_clear_queued(task);
 309	if (rpc_test_and_set_running(task))
 310		return;
 311	if (RPC_IS_ASYNC(task)) {
 312		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 313		queue_work(rpciod_workqueue, &task->u.tk_work);
 314	} else
 315		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 316}
 317
 318/*
 319 * Prepare for sleeping on a wait queue.
 320 * By always appending tasks to the list we ensure FIFO behavior.
 321 * NB: An RPC task will only receive interrupt-driven events as long
 322 * as it's on a wait queue.
 323 */
 324static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 325		struct rpc_task *task,
 326		rpc_action action,
 327		unsigned char queue_priority)
 328{
 329	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 330			task->tk_pid, rpc_qname(q), jiffies);
 331
 332	trace_rpc_task_sleep(task->tk_client, task, q);
 333
 334	__rpc_add_wait_queue(q, task, queue_priority);
 335
 336	BUG_ON(task->tk_callback != NULL);
 337	task->tk_callback = action;
 338	__rpc_add_timer(q, task);
 339}
 340
 341void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 342				rpc_action action)
 343{
 344	/* We shouldn't ever put an inactive task to sleep */
 345	BUG_ON(!RPC_IS_ACTIVATED(task));
 
 
 
 
 
 346
 347	/*
 348	 * Protect the queue operations.
 349	 */
 350	spin_lock_bh(&q->lock);
 351	__rpc_sleep_on_priority(q, task, action, task->tk_priority);
 352	spin_unlock_bh(&q->lock);
 353}
 354EXPORT_SYMBOL_GPL(rpc_sleep_on);
 355
 356void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 357		rpc_action action, int priority)
 358{
 359	/* We shouldn't ever put an inactive task to sleep */
 360	BUG_ON(!RPC_IS_ACTIVATED(task));
 
 
 
 
 
 361
 362	/*
 363	 * Protect the queue operations.
 364	 */
 365	spin_lock_bh(&q->lock);
 366	__rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
 367	spin_unlock_bh(&q->lock);
 368}
 
 369
 370/**
 371 * __rpc_do_wake_up_task - wake up a single rpc_task
 372 * @queue: wait queue
 373 * @task: task to be woken up
 374 *
 375 * Caller must hold queue->lock, and have cleared the task queued flag.
 376 */
 377static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 378{
 379	dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 380			task->tk_pid, jiffies);
 381
 382	/* Has the task been executed yet? If not, we cannot wake it up! */
 383	if (!RPC_IS_ACTIVATED(task)) {
 384		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 385		return;
 386	}
 387
 388	trace_rpc_task_wakeup(task->tk_client, task, queue);
 389
 390	__rpc_remove_wait_queue(queue, task);
 391
 392	rpc_make_runnable(task);
 393
 394	dprintk("RPC:       __rpc_wake_up_task done\n");
 395}
 396
 397/*
 398 * Wake up a queued task while the queue lock is being held
 399 */
 400static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 401{
 402	if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
 403		__rpc_do_wake_up_task(queue, task);
 
 
 
 404}
 405
 406/*
 407 * Tests whether rpc queue is empty
 408 */
 409int rpc_queue_empty(struct rpc_wait_queue *queue)
 410{
 411	int res;
 412
 413	spin_lock_bh(&queue->lock);
 414	res = queue->qlen;
 415	spin_unlock_bh(&queue->lock);
 416	return res == 0;
 417}
 418EXPORT_SYMBOL_GPL(rpc_queue_empty);
 419
 420/*
 421 * Wake up a task on a specific queue
 422 */
 423void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 424{
 425	spin_lock_bh(&queue->lock);
 426	rpc_wake_up_task_queue_locked(queue, task);
 427	spin_unlock_bh(&queue->lock);
 428}
 429EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 430
 431/*
 432 * Wake up the next task on a priority queue.
 433 */
 434static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 435{
 436	struct list_head *q;
 437	struct rpc_task *task;
 438
 439	/*
 440	 * Service a batch of tasks from a single owner.
 441	 */
 442	q = &queue->tasks[queue->priority];
 443	if (!list_empty(q)) {
 444		task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 445		if (queue->owner == task->tk_owner) {
 446			if (--queue->nr)
 447				goto out;
 448			list_move_tail(&task->u.tk_wait.list, q);
 449		}
 450		/*
 451		 * Check if we need to switch queues.
 452		 */
 453		if (--queue->count)
 454			goto new_owner;
 455	}
 456
 457	/*
 458	 * Service the next queue.
 459	 */
 460	do {
 461		if (q == &queue->tasks[0])
 462			q = &queue->tasks[queue->maxpriority];
 463		else
 464			q = q - 1;
 465		if (!list_empty(q)) {
 466			task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 467			goto new_queue;
 468		}
 469	} while (q != &queue->tasks[queue->priority]);
 470
 471	rpc_reset_waitqueue_priority(queue);
 472	return NULL;
 473
 474new_queue:
 475	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 476new_owner:
 477	rpc_set_waitqueue_owner(queue, task->tk_owner);
 478out:
 479	return task;
 480}
 481
 482static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
 483{
 484	if (RPC_IS_PRIORITY(queue))
 485		return __rpc_find_next_queued_priority(queue);
 486	if (!list_empty(&queue->tasks[0]))
 487		return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
 488	return NULL;
 489}
 490
 491/*
 492 * Wake up the first task on the wait queue.
 493 */
 494struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
 495		bool (*func)(struct rpc_task *, void *), void *data)
 496{
 497	struct rpc_task	*task = NULL;
 498
 499	dprintk("RPC:       wake_up_first(%p \"%s\")\n",
 500			queue, rpc_qname(queue));
 501	spin_lock_bh(&queue->lock);
 502	task = __rpc_find_next_queued(queue);
 503	if (task != NULL) {
 504		if (func(task, data))
 505			rpc_wake_up_task_queue_locked(queue, task);
 506		else
 507			task = NULL;
 508	}
 509	spin_unlock_bh(&queue->lock);
 510
 511	return task;
 512}
 513EXPORT_SYMBOL_GPL(rpc_wake_up_first);
 514
 515static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
 516{
 517	return true;
 518}
 519
 520/*
 521 * Wake up the next task on the wait queue.
 522*/
 523struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
 524{
 525	return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
 526}
 527EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 528
 529/**
 530 * rpc_wake_up - wake up all rpc_tasks
 531 * @queue: rpc_wait_queue on which the tasks are sleeping
 532 *
 533 * Grabs queue->lock
 534 */
 535void rpc_wake_up(struct rpc_wait_queue *queue)
 536{
 537	struct list_head *head;
 538
 539	spin_lock_bh(&queue->lock);
 540	head = &queue->tasks[queue->maxpriority];
 541	for (;;) {
 542		while (!list_empty(head)) {
 543			struct rpc_task *task;
 544			task = list_first_entry(head,
 545					struct rpc_task,
 546					u.tk_wait.list);
 547			rpc_wake_up_task_queue_locked(queue, task);
 548		}
 549		if (head == &queue->tasks[0])
 550			break;
 551		head--;
 552	}
 553	spin_unlock_bh(&queue->lock);
 554}
 555EXPORT_SYMBOL_GPL(rpc_wake_up);
 556
 557/**
 558 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 559 * @queue: rpc_wait_queue on which the tasks are sleeping
 560 * @status: status value to set
 561 *
 562 * Grabs queue->lock
 563 */
 564void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 565{
 566	struct list_head *head;
 567
 568	spin_lock_bh(&queue->lock);
 569	head = &queue->tasks[queue->maxpriority];
 570	for (;;) {
 571		while (!list_empty(head)) {
 572			struct rpc_task *task;
 573			task = list_first_entry(head,
 574					struct rpc_task,
 575					u.tk_wait.list);
 576			task->tk_status = status;
 577			rpc_wake_up_task_queue_locked(queue, task);
 578		}
 579		if (head == &queue->tasks[0])
 580			break;
 581		head--;
 582	}
 583	spin_unlock_bh(&queue->lock);
 584}
 585EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 586
 587static void __rpc_queue_timer_fn(unsigned long ptr)
 588{
 589	struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 590	struct rpc_task *task, *n;
 591	unsigned long expires, now, timeo;
 592
 593	spin_lock(&queue->lock);
 594	expires = now = jiffies;
 595	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 596		timeo = task->u.tk_wait.expires;
 597		if (time_after_eq(now, timeo)) {
 598			dprintk("RPC: %5u timeout\n", task->tk_pid);
 599			task->tk_status = -ETIMEDOUT;
 600			rpc_wake_up_task_queue_locked(queue, task);
 601			continue;
 602		}
 603		if (expires == now || time_after(expires, timeo))
 604			expires = timeo;
 605	}
 606	if (!list_empty(&queue->timer_list.list))
 607		rpc_set_queue_timer(queue, expires);
 608	spin_unlock(&queue->lock);
 609}
 610
 611static void __rpc_atrun(struct rpc_task *task)
 612{
 613	task->tk_status = 0;
 
 614}
 615
 616/*
 617 * Run a task at a later time
 618 */
 619void rpc_delay(struct rpc_task *task, unsigned long delay)
 620{
 621	task->tk_timeout = delay;
 622	rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 623}
 624EXPORT_SYMBOL_GPL(rpc_delay);
 625
 626/*
 627 * Helper to call task->tk_ops->rpc_call_prepare
 628 */
 629void rpc_prepare_task(struct rpc_task *task)
 630{
 631	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 632}
 633
 634static void
 635rpc_init_task_statistics(struct rpc_task *task)
 636{
 637	/* Initialize retry counters */
 638	task->tk_garb_retry = 2;
 639	task->tk_cred_retry = 2;
 640	task->tk_rebind_retry = 2;
 641
 642	/* starting timestamp */
 643	task->tk_start = ktime_get();
 644}
 645
 646static void
 647rpc_reset_task_statistics(struct rpc_task *task)
 648{
 649	task->tk_timeouts = 0;
 650	task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
 651
 652	rpc_init_task_statistics(task);
 653}
 654
 655/*
 656 * Helper that calls task->tk_ops->rpc_call_done if it exists
 657 */
 658void rpc_exit_task(struct rpc_task *task)
 659{
 660	task->tk_action = NULL;
 661	if (task->tk_ops->rpc_call_done != NULL) {
 662		task->tk_ops->rpc_call_done(task, task->tk_calldata);
 663		if (task->tk_action != NULL) {
 664			WARN_ON(RPC_ASSASSINATED(task));
 665			/* Always release the RPC slot and buffer memory */
 666			xprt_release(task);
 667			rpc_reset_task_statistics(task);
 668		}
 669	}
 670}
 671
 672void rpc_exit(struct rpc_task *task, int status)
 673{
 674	task->tk_status = status;
 675	task->tk_action = rpc_exit_task;
 676	if (RPC_IS_QUEUED(task))
 677		rpc_wake_up_queued_task(task->tk_waitqueue, task);
 678}
 679EXPORT_SYMBOL_GPL(rpc_exit);
 680
 681void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 682{
 683	if (ops->rpc_release != NULL)
 684		ops->rpc_release(calldata);
 685}
 686
 687/*
 688 * This is the RPC `scheduler' (or rather, the finite state machine).
 689 */
 690static void __rpc_execute(struct rpc_task *task)
 691{
 692	struct rpc_wait_queue *queue;
 693	int task_is_async = RPC_IS_ASYNC(task);
 694	int status = 0;
 695
 696	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 697			task->tk_pid, task->tk_flags);
 698
 699	BUG_ON(RPC_IS_QUEUED(task));
 
 
 700
 701	for (;;) {
 702		void (*do_action)(struct rpc_task *);
 703
 704		/*
 705		 * Execute any pending callback first.
 706		 */
 707		do_action = task->tk_callback;
 708		task->tk_callback = NULL;
 709		if (do_action == NULL) {
 710			/*
 711			 * Perform the next FSM step.
 712			 * tk_action may be NULL if the task has been killed.
 713			 * In particular, note that rpc_killall_tasks may
 714			 * do this at any time, so beware when dereferencing.
 715			 */
 716			do_action = task->tk_action;
 717			if (do_action == NULL)
 718				break;
 719		}
 720		trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
 721		do_action(task);
 722
 723		/*
 724		 * Lockless check for whether task is sleeping or not.
 725		 */
 726		if (!RPC_IS_QUEUED(task))
 727			continue;
 728		/*
 729		 * The queue->lock protects against races with
 730		 * rpc_make_runnable().
 731		 *
 732		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 733		 * rpc_task, rpc_make_runnable() can assign it to a
 734		 * different workqueue. We therefore cannot assume that the
 735		 * rpc_task pointer may still be dereferenced.
 736		 */
 737		queue = task->tk_waitqueue;
 738		spin_lock_bh(&queue->lock);
 739		if (!RPC_IS_QUEUED(task)) {
 740			spin_unlock_bh(&queue->lock);
 741			continue;
 742		}
 743		rpc_clear_running(task);
 744		spin_unlock_bh(&queue->lock);
 745		if (task_is_async)
 746			return;
 747
 748		/* sync task: sleep here */
 749		dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 750		status = out_of_line_wait_on_bit(&task->tk_runstate,
 751				RPC_TASK_QUEUED, rpc_wait_bit_killable,
 752				TASK_KILLABLE);
 753		if (status == -ERESTARTSYS) {
 754			/*
 755			 * When a sync task receives a signal, it exits with
 756			 * -ERESTARTSYS. In order to catch any callbacks that
 757			 * clean up after sleeping on some queue, we don't
 758			 * break the loop here, but go around once more.
 759			 */
 760			dprintk("RPC: %5u got signal\n", task->tk_pid);
 761			task->tk_flags |= RPC_TASK_KILLED;
 762			rpc_exit(task, -ERESTARTSYS);
 763		}
 764		rpc_set_running(task);
 765		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 766	}
 767
 768	dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 769			task->tk_status);
 770	/* Release all resources associated with the task */
 771	rpc_release_task(task);
 772}
 773
 774/*
 775 * User-visible entry point to the scheduler.
 776 *
 777 * This may be called recursively if e.g. an async NFS task updates
 778 * the attributes and finds that dirty pages must be flushed.
 779 * NOTE: Upon exit of this function the task is guaranteed to be
 780 *	 released. In particular note that tk_release() will have
 781 *	 been called, so your task memory may have been freed.
 782 */
 783void rpc_execute(struct rpc_task *task)
 784{
 
 
 785	rpc_set_active(task);
 786	rpc_make_runnable(task);
 787	if (!RPC_IS_ASYNC(task))
 788		__rpc_execute(task);
 789}
 790
 791static void rpc_async_schedule(struct work_struct *work)
 792{
 793	current->flags |= PF_FSTRANS;
 794	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 795	current->flags &= ~PF_FSTRANS;
 796}
 797
 798/**
 799 * rpc_malloc - allocate an RPC buffer
 800 * @task: RPC task that will use this buffer
 801 * @size: requested byte size
 802 *
 803 * To prevent rpciod from hanging, this allocator never sleeps,
 804 * returning NULL if the request cannot be serviced immediately.
 
 805 * The caller can arrange to sleep in a way that is safe for rpciod.
 806 *
 807 * Most requests are 'small' (under 2KiB) and can be serviced from a
 808 * mempool, ensuring that NFS reads and writes can always proceed,
 809 * and that there is good locality of reference for these buffers.
 810 *
 811 * In order to avoid memory starvation triggering more writebacks of
 812 * NFS requests, we avoid using GFP_KERNEL.
 813 */
 814void *rpc_malloc(struct rpc_task *task, size_t size)
 815{
 816	struct rpc_buffer *buf;
 817	gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 
 
 
 818
 819	size += sizeof(struct rpc_buffer);
 820	if (size <= RPC_BUFFER_MAXSIZE)
 821		buf = mempool_alloc(rpc_buffer_mempool, gfp);
 822	else
 823		buf = kmalloc(size, gfp);
 824
 825	if (!buf)
 826		return NULL;
 827
 828	buf->len = size;
 829	dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 830			task->tk_pid, size, buf);
 831	return &buf->data;
 832}
 833EXPORT_SYMBOL_GPL(rpc_malloc);
 834
 835/**
 836 * rpc_free - free buffer allocated via rpc_malloc
 837 * @buffer: buffer to free
 838 *
 839 */
 840void rpc_free(void *buffer)
 841{
 842	size_t size;
 843	struct rpc_buffer *buf;
 844
 845	if (!buffer)
 846		return;
 847
 848	buf = container_of(buffer, struct rpc_buffer, data);
 849	size = buf->len;
 850
 851	dprintk("RPC:       freeing buffer of size %zu at %p\n",
 852			size, buf);
 853
 854	if (size <= RPC_BUFFER_MAXSIZE)
 855		mempool_free(buf, rpc_buffer_mempool);
 856	else
 857		kfree(buf);
 858}
 859EXPORT_SYMBOL_GPL(rpc_free);
 860
 861/*
 862 * Creation and deletion of RPC task structures
 863 */
 864static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 865{
 866	memset(task, 0, sizeof(*task));
 867	atomic_set(&task->tk_count, 1);
 868	task->tk_flags  = task_setup_data->flags;
 869	task->tk_ops = task_setup_data->callback_ops;
 870	task->tk_calldata = task_setup_data->callback_data;
 871	INIT_LIST_HEAD(&task->tk_task);
 872
 873	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 874	task->tk_owner = current->tgid;
 875
 876	/* Initialize workqueue for async tasks */
 877	task->tk_workqueue = task_setup_data->workqueue;
 878
 
 
 879	if (task->tk_ops->rpc_call_prepare != NULL)
 880		task->tk_action = rpc_prepare_task;
 881
 882	rpc_init_task_statistics(task);
 883
 884	dprintk("RPC:       new task initialized, procpid %u\n",
 885				task_pid_nr(current));
 886}
 887
 888static struct rpc_task *
 889rpc_alloc_task(void)
 890{
 891	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 892}
 893
 894/*
 895 * Create a new task for the specified client.
 896 */
 897struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 898{
 899	struct rpc_task	*task = setup_data->task;
 900	unsigned short flags = 0;
 901
 902	if (task == NULL) {
 903		task = rpc_alloc_task();
 904		if (task == NULL) {
 905			rpc_release_calldata(setup_data->callback_ops,
 906					setup_data->callback_data);
 907			return ERR_PTR(-ENOMEM);
 908		}
 909		flags = RPC_TASK_DYNAMIC;
 910	}
 911
 912	rpc_init_task(task, setup_data);
 913	task->tk_flags |= flags;
 914	dprintk("RPC:       allocated task %p\n", task);
 915	return task;
 916}
 917
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 918static void rpc_free_task(struct rpc_task *task)
 919{
 920	const struct rpc_call_ops *tk_ops = task->tk_ops;
 921	void *calldata = task->tk_calldata;
 
 922
 923	if (task->tk_flags & RPC_TASK_DYNAMIC) {
 924		dprintk("RPC: %5u freeing task\n", task->tk_pid);
 925		mempool_free(task, rpc_task_mempool);
 926	}
 927	rpc_release_calldata(tk_ops, calldata);
 928}
 929
 930static void rpc_async_release(struct work_struct *work)
 931{
 932	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 933}
 934
 935static void rpc_release_resources_task(struct rpc_task *task)
 936{
 937	if (task->tk_rqstp)
 938		xprt_release(task);
 939	if (task->tk_msg.rpc_cred) {
 940		put_rpccred(task->tk_msg.rpc_cred);
 941		task->tk_msg.rpc_cred = NULL;
 942	}
 943	rpc_task_release_client(task);
 944}
 945
 946static void rpc_final_put_task(struct rpc_task *task,
 947		struct workqueue_struct *q)
 948{
 949	if (q != NULL) {
 950		INIT_WORK(&task->u.tk_work, rpc_async_release);
 951		queue_work(q, &task->u.tk_work);
 952	} else
 953		rpc_free_task(task);
 954}
 955
 956static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
 957{
 958	if (atomic_dec_and_test(&task->tk_count)) {
 959		rpc_release_resources_task(task);
 960		rpc_final_put_task(task, q);
 961	}
 962}
 963
 964void rpc_put_task(struct rpc_task *task)
 965{
 966	rpc_do_put_task(task, NULL);
 967}
 968EXPORT_SYMBOL_GPL(rpc_put_task);
 969
 970void rpc_put_task_async(struct rpc_task *task)
 971{
 972	rpc_do_put_task(task, task->tk_workqueue);
 973}
 974EXPORT_SYMBOL_GPL(rpc_put_task_async);
 975
 976static void rpc_release_task(struct rpc_task *task)
 977{
 978	dprintk("RPC: %5u release task\n", task->tk_pid);
 979
 980	BUG_ON (RPC_IS_QUEUED(task));
 981
 982	rpc_release_resources_task(task);
 983
 984	/*
 985	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
 986	 * so it should be safe to use task->tk_count as a test for whether
 987	 * or not any other processes still hold references to our rpc_task.
 988	 */
 989	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
 990		/* Wake up anyone who may be waiting for task completion */
 991		if (!rpc_complete_task(task))
 992			return;
 993	} else {
 994		if (!atomic_dec_and_test(&task->tk_count))
 995			return;
 996	}
 997	rpc_final_put_task(task, task->tk_workqueue);
 998}
 999
1000int rpciod_up(void)
1001{
1002	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1003}
1004
1005void rpciod_down(void)
1006{
1007	module_put(THIS_MODULE);
1008}
1009
1010/*
1011 * Start up the rpciod workqueue.
1012 */
1013static int rpciod_start(void)
1014{
1015	struct workqueue_struct *wq;
1016
1017	/*
1018	 * Create the rpciod thread and wait for it to start.
1019	 */
1020	dprintk("RPC:       creating workqueue rpciod\n");
1021	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
 
1022	rpciod_workqueue = wq;
1023	return rpciod_workqueue != NULL;
1024}
1025
1026static void rpciod_stop(void)
1027{
1028	struct workqueue_struct *wq = NULL;
1029
1030	if (rpciod_workqueue == NULL)
1031		return;
1032	dprintk("RPC:       destroying workqueue rpciod\n");
1033
1034	wq = rpciod_workqueue;
1035	rpciod_workqueue = NULL;
1036	destroy_workqueue(wq);
1037}
1038
1039void
1040rpc_destroy_mempool(void)
1041{
1042	rpciod_stop();
1043	if (rpc_buffer_mempool)
1044		mempool_destroy(rpc_buffer_mempool);
1045	if (rpc_task_mempool)
1046		mempool_destroy(rpc_task_mempool);
1047	if (rpc_task_slabp)
1048		kmem_cache_destroy(rpc_task_slabp);
1049	if (rpc_buffer_slabp)
1050		kmem_cache_destroy(rpc_buffer_slabp);
1051	rpc_destroy_wait_queue(&delay_queue);
1052}
1053
1054int
1055rpc_init_mempool(void)
1056{
1057	/*
1058	 * The following is not strictly a mempool initialisation,
1059	 * but there is no harm in doing it here
1060	 */
1061	rpc_init_wait_queue(&delay_queue, "delayq");
1062	if (!rpciod_start())
1063		goto err_nomem;
1064
1065	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1066					     sizeof(struct rpc_task),
1067					     0, SLAB_HWCACHE_ALIGN,
1068					     NULL);
1069	if (!rpc_task_slabp)
1070		goto err_nomem;
1071	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1072					     RPC_BUFFER_MAXSIZE,
1073					     0, SLAB_HWCACHE_ALIGN,
1074					     NULL);
1075	if (!rpc_buffer_slabp)
1076		goto err_nomem;
1077	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1078						    rpc_task_slabp);
1079	if (!rpc_task_mempool)
1080		goto err_nomem;
1081	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1082						      rpc_buffer_slabp);
1083	if (!rpc_buffer_mempool)
1084		goto err_nomem;
1085	return 0;
1086err_nomem:
1087	rpc_destroy_mempool();
1088	return -ENOMEM;
1089}