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}
v3.1
   1/*
   2 * linux/net/sunrpc/sched.c
   3 *
   4 * Scheduling for synchronous and asynchronous RPC requests.
   5 *
   6 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
   7 *
   8 * TCP NFS related read + write fixes
   9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10 */
  11
  12#include <linux/module.h>
  13
  14#include <linux/sched.h>
  15#include <linux/interrupt.h>
  16#include <linux/slab.h>
  17#include <linux/mempool.h>
  18#include <linux/smp.h>
  19#include <linux/spinlock.h>
  20#include <linux/mutex.h>
 
  21
  22#include <linux/sunrpc/clnt.h>
  23
  24#include "sunrpc.h"
  25
  26#ifdef RPC_DEBUG
  27#define RPCDBG_FACILITY		RPCDBG_SCHED
  28#endif
  29
 
 
 
  30/*
  31 * RPC slabs and memory pools
  32 */
  33#define RPC_BUFFER_MAXSIZE	(2048)
  34#define RPC_BUFFER_POOLSIZE	(8)
  35#define RPC_TASK_POOLSIZE	(8)
  36static struct kmem_cache	*rpc_task_slabp __read_mostly;
  37static struct kmem_cache	*rpc_buffer_slabp __read_mostly;
  38static mempool_t	*rpc_task_mempool __read_mostly;
  39static mempool_t	*rpc_buffer_mempool __read_mostly;
  40
  41static void			rpc_async_schedule(struct work_struct *);
  42static void			 rpc_release_task(struct rpc_task *task);
  43static void __rpc_queue_timer_fn(unsigned long ptr);
  44
  45/*
  46 * RPC tasks sit here while waiting for conditions to improve.
  47 */
  48static struct rpc_wait_queue delay_queue;
  49
  50/*
  51 * rpciod-related stuff
  52 */
  53struct workqueue_struct *rpciod_workqueue;
  54
  55/*
  56 * Disable the timer for a given RPC task. Should be called with
  57 * queue->lock and bh_disabled in order to avoid races within
  58 * rpc_run_timer().
  59 */
  60static void
  61__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  62{
  63	if (task->tk_timeout == 0)
  64		return;
  65	dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  66	task->tk_timeout = 0;
  67	list_del(&task->u.tk_wait.timer_list);
  68	if (list_empty(&queue->timer_list.list))
  69		del_timer(&queue->timer_list.timer);
  70}
  71
  72static void
  73rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  74{
  75	queue->timer_list.expires = expires;
  76	mod_timer(&queue->timer_list.timer, expires);
  77}
  78
  79/*
  80 * Set up a timer for the current task.
  81 */
  82static void
  83__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  84{
  85	if (!task->tk_timeout)
  86		return;
  87
  88	dprintk("RPC: %5u setting alarm for %lu ms\n",
  89			task->tk_pid, task->tk_timeout * 1000 / HZ);
  90
  91	task->u.tk_wait.expires = jiffies + task->tk_timeout;
  92	if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
  93		rpc_set_queue_timer(queue, task->u.tk_wait.expires);
  94	list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
  95}
  96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  97/*
  98 * Add new request to a priority queue.
  99 */
 100static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
 101		struct rpc_task *task,
 102		unsigned char queue_priority)
 103{
 104	struct list_head *q;
 105	struct rpc_task *t;
 106
 107	INIT_LIST_HEAD(&task->u.tk_wait.links);
 
 
 
 
 108	q = &queue->tasks[queue_priority];
 109	if (unlikely(queue_priority > queue->maxpriority))
 110		q = &queue->tasks[queue->maxpriority];
 111	list_for_each_entry(t, q, u.tk_wait.list) {
 112		if (t->tk_owner == task->tk_owner) {
 113			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 114			return;
 115		}
 116	}
 117	list_add_tail(&task->u.tk_wait.list, q);
 118}
 119
 120/*
 121 * Add new request to wait queue.
 122 *
 123 * Swapper tasks always get inserted at the head of the queue.
 124 * This should avoid many nasty memory deadlocks and hopefully
 125 * improve overall performance.
 126 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 127 */
 128static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
 129		struct rpc_task *task,
 130		unsigned char queue_priority)
 131{
 132	BUG_ON (RPC_IS_QUEUED(task));
 
 
 133
 134	if (RPC_IS_PRIORITY(queue))
 135		__rpc_add_wait_queue_priority(queue, task, queue_priority);
 136	else if (RPC_IS_SWAPPER(task))
 137		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 138	else
 139		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 140	task->tk_waitqueue = queue;
 141	queue->qlen++;
 
 
 142	rpc_set_queued(task);
 143
 144	dprintk("RPC: %5u added to queue %p \"%s\"\n",
 145			task->tk_pid, queue, rpc_qname(queue));
 146}
 147
 148/*
 149 * Remove request from a priority queue.
 150 */
 151static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 152{
 153	struct rpc_task *t;
 154
 155	if (!list_empty(&task->u.tk_wait.links)) {
 156		t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 157		list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 158		list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 159	}
 160}
 161
 162/*
 163 * Remove request from queue.
 164 * Note: must be called with spin lock held.
 165 */
 166static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 167{
 168	__rpc_disable_timer(queue, task);
 169	if (RPC_IS_PRIORITY(queue))
 170		__rpc_remove_wait_queue_priority(task);
 171	list_del(&task->u.tk_wait.list);
 172	queue->qlen--;
 173	dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 174			task->tk_pid, queue, rpc_qname(queue));
 175}
 176
 177static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 178{
 179	queue->priority = priority;
 180	queue->count = 1 << (priority * 2);
 181}
 182
 183static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 184{
 185	queue->owner = pid;
 186	queue->nr = RPC_BATCH_COUNT;
 187}
 188
 189static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 190{
 191	rpc_set_waitqueue_priority(queue, queue->maxpriority);
 192	rpc_set_waitqueue_owner(queue, 0);
 193}
 194
 195static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 196{
 197	int i;
 198
 199	spin_lock_init(&queue->lock);
 200	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 201		INIT_LIST_HEAD(&queue->tasks[i]);
 202	queue->maxpriority = nr_queues - 1;
 203	rpc_reset_waitqueue_priority(queue);
 204	queue->qlen = 0;
 205	setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 206	INIT_LIST_HEAD(&queue->timer_list.list);
 207#ifdef RPC_DEBUG
 208	queue->name = qname;
 209#endif
 210}
 211
 212void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 213{
 214	__rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 215}
 216EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 217
 218void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 219{
 220	__rpc_init_priority_wait_queue(queue, qname, 1);
 221}
 222EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 223
 224void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 225{
 226	del_timer_sync(&queue->timer_list.timer);
 227}
 228EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 229
 230static int rpc_wait_bit_killable(void *word)
 231{
 232	if (fatal_signal_pending(current))
 
 233		return -ERESTARTSYS;
 234	schedule();
 235	return 0;
 236}
 237
 238#ifdef RPC_DEBUG
 239static void rpc_task_set_debuginfo(struct rpc_task *task)
 240{
 241	static atomic_t rpc_pid;
 242
 243	task->tk_pid = atomic_inc_return(&rpc_pid);
 244}
 245#else
 246static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 247{
 248}
 249#endif
 250
 251static void rpc_set_active(struct rpc_task *task)
 252{
 
 
 253	rpc_task_set_debuginfo(task);
 254	set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 255}
 256
 257/*
 258 * Mark an RPC call as having completed by clearing the 'active' bit
 259 * and then waking up all tasks that were sleeping.
 260 */
 261static int rpc_complete_task(struct rpc_task *task)
 262{
 263	void *m = &task->tk_runstate;
 264	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
 265	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
 266	unsigned long flags;
 267	int ret;
 268
 
 
 269	spin_lock_irqsave(&wq->lock, flags);
 270	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 271	ret = atomic_dec_and_test(&task->tk_count);
 272	if (waitqueue_active(wq))
 273		__wake_up_locked_key(wq, TASK_NORMAL, &k);
 274	spin_unlock_irqrestore(&wq->lock, flags);
 275	return ret;
 276}
 277
 278/*
 279 * Allow callers to wait for completion of an RPC call
 280 *
 281 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
 282 * to enforce taking of the wq->lock and hence avoid races with
 283 * rpc_complete_task().
 284 */
 285int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 286{
 287	if (action == NULL)
 288		action = rpc_wait_bit_killable;
 289	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 290			action, TASK_KILLABLE);
 291}
 292EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 293
 294/*
 295 * Make an RPC task runnable.
 296 *
 297 * Note: If the task is ASYNC, this must be called with
 298 * the spinlock held to protect the wait queue operation.
 
 
 
 
 
 299 */
 300static void rpc_make_runnable(struct rpc_task *task)
 301{
 
 
 302	rpc_clear_queued(task);
 303	if (rpc_test_and_set_running(task))
 304		return;
 305	if (RPC_IS_ASYNC(task)) {
 306		INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 307		queue_work(rpciod_workqueue, &task->u.tk_work);
 308	} else
 309		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 310}
 311
 312/*
 313 * Prepare for sleeping on a wait queue.
 314 * By always appending tasks to the list we ensure FIFO behavior.
 315 * NB: An RPC task will only receive interrupt-driven events as long
 316 * as it's on a wait queue.
 317 */
 318static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 319		struct rpc_task *task,
 320		rpc_action action,
 321		unsigned char queue_priority)
 322{
 323	dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 324			task->tk_pid, rpc_qname(q), jiffies);
 325
 
 
 326	__rpc_add_wait_queue(q, task, queue_priority);
 327
 328	BUG_ON(task->tk_callback != NULL);
 329	task->tk_callback = action;
 330	__rpc_add_timer(q, task);
 331}
 332
 333void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 334				rpc_action action)
 335{
 336	/* We shouldn't ever put an inactive task to sleep */
 337	BUG_ON(!RPC_IS_ACTIVATED(task));
 
 
 
 
 
 338
 339	/*
 340	 * Protect the queue operations.
 341	 */
 342	spin_lock_bh(&q->lock);
 343	__rpc_sleep_on_priority(q, task, action, task->tk_priority);
 344	spin_unlock_bh(&q->lock);
 345}
 346EXPORT_SYMBOL_GPL(rpc_sleep_on);
 347
 348void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 349		rpc_action action, int priority)
 350{
 351	/* We shouldn't ever put an inactive task to sleep */
 352	BUG_ON(!RPC_IS_ACTIVATED(task));
 
 
 
 
 
 353
 354	/*
 355	 * Protect the queue operations.
 356	 */
 357	spin_lock_bh(&q->lock);
 358	__rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
 359	spin_unlock_bh(&q->lock);
 360}
 
 361
 362/**
 363 * __rpc_do_wake_up_task - wake up a single rpc_task
 364 * @queue: wait queue
 365 * @task: task to be woken up
 366 *
 367 * Caller must hold queue->lock, and have cleared the task queued flag.
 368 */
 369static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 370{
 371	dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 372			task->tk_pid, jiffies);
 373
 374	/* Has the task been executed yet? If not, we cannot wake it up! */
 375	if (!RPC_IS_ACTIVATED(task)) {
 376		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 377		return;
 378	}
 379
 
 
 380	__rpc_remove_wait_queue(queue, task);
 381
 382	rpc_make_runnable(task);
 383
 384	dprintk("RPC:       __rpc_wake_up_task done\n");
 385}
 386
 387/*
 388 * Wake up a queued task while the queue lock is being held
 389 */
 390static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 391{
 392	if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
 393		__rpc_do_wake_up_task(queue, task);
 394}
 395
 396/*
 397 * Tests whether rpc queue is empty
 398 */
 399int rpc_queue_empty(struct rpc_wait_queue *queue)
 400{
 401	int res;
 402
 403	spin_lock_bh(&queue->lock);
 404	res = queue->qlen;
 405	spin_unlock_bh(&queue->lock);
 406	return res == 0;
 407}
 408EXPORT_SYMBOL_GPL(rpc_queue_empty);
 409
 410/*
 411 * Wake up a task on a specific queue
 412 */
 413void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 414{
 415	spin_lock_bh(&queue->lock);
 416	rpc_wake_up_task_queue_locked(queue, task);
 417	spin_unlock_bh(&queue->lock);
 418}
 419EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 420
 421/*
 422 * Wake up the next task on a priority queue.
 423 */
 424static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
 425{
 426	struct list_head *q;
 427	struct rpc_task *task;
 428
 429	/*
 430	 * Service a batch of tasks from a single owner.
 431	 */
 432	q = &queue->tasks[queue->priority];
 433	if (!list_empty(q)) {
 434		task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 435		if (queue->owner == task->tk_owner) {
 436			if (--queue->nr)
 437				goto out;
 438			list_move_tail(&task->u.tk_wait.list, q);
 439		}
 440		/*
 441		 * Check if we need to switch queues.
 442		 */
 443		if (--queue->count)
 444			goto new_owner;
 445	}
 446
 447	/*
 448	 * Service the next queue.
 449	 */
 450	do {
 451		if (q == &queue->tasks[0])
 452			q = &queue->tasks[queue->maxpriority];
 453		else
 454			q = q - 1;
 455		if (!list_empty(q)) {
 456			task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 457			goto new_queue;
 458		}
 459	} while (q != &queue->tasks[queue->priority]);
 460
 461	rpc_reset_waitqueue_priority(queue);
 462	return NULL;
 463
 464new_queue:
 465	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 466new_owner:
 467	rpc_set_waitqueue_owner(queue, task->tk_owner);
 468out:
 469	rpc_wake_up_task_queue_locked(queue, task);
 470	return task;
 471}
 472
 
 
 
 
 
 
 
 
 
 473/*
 474 * Wake up the next task on the wait queue.
 475 */
 476struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 
 477{
 478	struct rpc_task	*task = NULL;
 479
 480	dprintk("RPC:       wake_up_next(%p \"%s\")\n",
 481			queue, rpc_qname(queue));
 482	spin_lock_bh(&queue->lock);
 483	if (RPC_IS_PRIORITY(queue))
 484		task = __rpc_wake_up_next_priority(queue);
 485	else {
 486		task_for_first(task, &queue->tasks[0])
 487			rpc_wake_up_task_queue_locked(queue, task);
 
 
 488	}
 489	spin_unlock_bh(&queue->lock);
 490
 491	return task;
 492}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 493EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 494
 495/**
 496 * rpc_wake_up - wake up all rpc_tasks
 497 * @queue: rpc_wait_queue on which the tasks are sleeping
 498 *
 499 * Grabs queue->lock
 500 */
 501void rpc_wake_up(struct rpc_wait_queue *queue)
 502{
 503	struct rpc_task *task, *next;
 504	struct list_head *head;
 505
 506	spin_lock_bh(&queue->lock);
 507	head = &queue->tasks[queue->maxpriority];
 508	for (;;) {
 509		list_for_each_entry_safe(task, next, head, u.tk_wait.list)
 
 
 
 
 510			rpc_wake_up_task_queue_locked(queue, task);
 
 511		if (head == &queue->tasks[0])
 512			break;
 513		head--;
 514	}
 515	spin_unlock_bh(&queue->lock);
 516}
 517EXPORT_SYMBOL_GPL(rpc_wake_up);
 518
 519/**
 520 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 521 * @queue: rpc_wait_queue on which the tasks are sleeping
 522 * @status: status value to set
 523 *
 524 * Grabs queue->lock
 525 */
 526void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 527{
 528	struct rpc_task *task, *next;
 529	struct list_head *head;
 530
 531	spin_lock_bh(&queue->lock);
 532	head = &queue->tasks[queue->maxpriority];
 533	for (;;) {
 534		list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
 
 
 
 
 535			task->tk_status = status;
 536			rpc_wake_up_task_queue_locked(queue, task);
 537		}
 538		if (head == &queue->tasks[0])
 539			break;
 540		head--;
 541	}
 542	spin_unlock_bh(&queue->lock);
 543}
 544EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 545
 546static void __rpc_queue_timer_fn(unsigned long ptr)
 547{
 548	struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 549	struct rpc_task *task, *n;
 550	unsigned long expires, now, timeo;
 551
 552	spin_lock(&queue->lock);
 553	expires = now = jiffies;
 554	list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 555		timeo = task->u.tk_wait.expires;
 556		if (time_after_eq(now, timeo)) {
 557			dprintk("RPC: %5u timeout\n", task->tk_pid);
 558			task->tk_status = -ETIMEDOUT;
 559			rpc_wake_up_task_queue_locked(queue, task);
 560			continue;
 561		}
 562		if (expires == now || time_after(expires, timeo))
 563			expires = timeo;
 564	}
 565	if (!list_empty(&queue->timer_list.list))
 566		rpc_set_queue_timer(queue, expires);
 567	spin_unlock(&queue->lock);
 568}
 569
 570static void __rpc_atrun(struct rpc_task *task)
 571{
 572	task->tk_status = 0;
 
 573}
 574
 575/*
 576 * Run a task at a later time
 577 */
 578void rpc_delay(struct rpc_task *task, unsigned long delay)
 579{
 580	task->tk_timeout = delay;
 581	rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 582}
 583EXPORT_SYMBOL_GPL(rpc_delay);
 584
 585/*
 586 * Helper to call task->tk_ops->rpc_call_prepare
 587 */
 588void rpc_prepare_task(struct rpc_task *task)
 589{
 590	task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 591}
 592
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 593/*
 594 * Helper that calls task->tk_ops->rpc_call_done if it exists
 595 */
 596void rpc_exit_task(struct rpc_task *task)
 597{
 598	task->tk_action = NULL;
 599	if (task->tk_ops->rpc_call_done != NULL) {
 600		task->tk_ops->rpc_call_done(task, task->tk_calldata);
 601		if (task->tk_action != NULL) {
 602			WARN_ON(RPC_ASSASSINATED(task));
 603			/* Always release the RPC slot and buffer memory */
 604			xprt_release(task);
 
 605		}
 606	}
 607}
 608
 609void rpc_exit(struct rpc_task *task, int status)
 610{
 611	task->tk_status = status;
 612	task->tk_action = rpc_exit_task;
 613	if (RPC_IS_QUEUED(task))
 614		rpc_wake_up_queued_task(task->tk_waitqueue, task);
 615}
 616EXPORT_SYMBOL_GPL(rpc_exit);
 617
 618void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 619{
 620	if (ops->rpc_release != NULL)
 621		ops->rpc_release(calldata);
 622}
 623
 624/*
 625 * This is the RPC `scheduler' (or rather, the finite state machine).
 626 */
 627static void __rpc_execute(struct rpc_task *task)
 628{
 629	struct rpc_wait_queue *queue;
 630	int task_is_async = RPC_IS_ASYNC(task);
 631	int status = 0;
 632
 633	dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 634			task->tk_pid, task->tk_flags);
 635
 636	BUG_ON(RPC_IS_QUEUED(task));
 
 
 637
 638	for (;;) {
 639		void (*do_action)(struct rpc_task *);
 640
 641		/*
 642		 * Execute any pending callback first.
 643		 */
 644		do_action = task->tk_callback;
 645		task->tk_callback = NULL;
 646		if (do_action == NULL) {
 647			/*
 648			 * Perform the next FSM step.
 649			 * tk_action may be NULL if the task has been killed.
 650			 * In particular, note that rpc_killall_tasks may
 651			 * do this at any time, so beware when dereferencing.
 652			 */
 653			do_action = task->tk_action;
 654			if (do_action == NULL)
 655				break;
 656		}
 
 657		do_action(task);
 658
 659		/*
 660		 * Lockless check for whether task is sleeping or not.
 661		 */
 662		if (!RPC_IS_QUEUED(task))
 663			continue;
 664		/*
 665		 * The queue->lock protects against races with
 666		 * rpc_make_runnable().
 667		 *
 668		 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 669		 * rpc_task, rpc_make_runnable() can assign it to a
 670		 * different workqueue. We therefore cannot assume that the
 671		 * rpc_task pointer may still be dereferenced.
 672		 */
 673		queue = task->tk_waitqueue;
 674		spin_lock_bh(&queue->lock);
 675		if (!RPC_IS_QUEUED(task)) {
 676			spin_unlock_bh(&queue->lock);
 677			continue;
 678		}
 679		rpc_clear_running(task);
 680		spin_unlock_bh(&queue->lock);
 681		if (task_is_async)
 682			return;
 683
 684		/* sync task: sleep here */
 685		dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 686		status = out_of_line_wait_on_bit(&task->tk_runstate,
 687				RPC_TASK_QUEUED, rpc_wait_bit_killable,
 688				TASK_KILLABLE);
 689		if (status == -ERESTARTSYS) {
 690			/*
 691			 * When a sync task receives a signal, it exits with
 692			 * -ERESTARTSYS. In order to catch any callbacks that
 693			 * clean up after sleeping on some queue, we don't
 694			 * break the loop here, but go around once more.
 695			 */
 696			dprintk("RPC: %5u got signal\n", task->tk_pid);
 697			task->tk_flags |= RPC_TASK_KILLED;
 698			rpc_exit(task, -ERESTARTSYS);
 699		}
 700		rpc_set_running(task);
 701		dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 702	}
 703
 704	dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 705			task->tk_status);
 706	/* Release all resources associated with the task */
 707	rpc_release_task(task);
 708}
 709
 710/*
 711 * User-visible entry point to the scheduler.
 712 *
 713 * This may be called recursively if e.g. an async NFS task updates
 714 * the attributes and finds that dirty pages must be flushed.
 715 * NOTE: Upon exit of this function the task is guaranteed to be
 716 *	 released. In particular note that tk_release() will have
 717 *	 been called, so your task memory may have been freed.
 718 */
 719void rpc_execute(struct rpc_task *task)
 720{
 
 
 721	rpc_set_active(task);
 722	rpc_make_runnable(task);
 723	if (!RPC_IS_ASYNC(task))
 724		__rpc_execute(task);
 725}
 726
 727static void rpc_async_schedule(struct work_struct *work)
 728{
 729	__rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 730}
 731
 732/**
 733 * rpc_malloc - allocate an RPC buffer
 734 * @task: RPC task that will use this buffer
 735 * @size: requested byte size
 736 *
 737 * To prevent rpciod from hanging, this allocator never sleeps,
 738 * returning NULL if the request cannot be serviced immediately.
 
 739 * The caller can arrange to sleep in a way that is safe for rpciod.
 740 *
 741 * Most requests are 'small' (under 2KiB) and can be serviced from a
 742 * mempool, ensuring that NFS reads and writes can always proceed,
 743 * and that there is good locality of reference for these buffers.
 744 *
 745 * In order to avoid memory starvation triggering more writebacks of
 746 * NFS requests, we avoid using GFP_KERNEL.
 747 */
 748void *rpc_malloc(struct rpc_task *task, size_t size)
 749{
 750	struct rpc_buffer *buf;
 751	gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 
 
 
 752
 753	size += sizeof(struct rpc_buffer);
 754	if (size <= RPC_BUFFER_MAXSIZE)
 755		buf = mempool_alloc(rpc_buffer_mempool, gfp);
 756	else
 757		buf = kmalloc(size, gfp);
 758
 759	if (!buf)
 760		return NULL;
 761
 762	buf->len = size;
 763	dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 764			task->tk_pid, size, buf);
 765	return &buf->data;
 766}
 767EXPORT_SYMBOL_GPL(rpc_malloc);
 768
 769/**
 770 * rpc_free - free buffer allocated via rpc_malloc
 771 * @buffer: buffer to free
 772 *
 773 */
 774void rpc_free(void *buffer)
 775{
 776	size_t size;
 777	struct rpc_buffer *buf;
 778
 779	if (!buffer)
 780		return;
 781
 782	buf = container_of(buffer, struct rpc_buffer, data);
 783	size = buf->len;
 784
 785	dprintk("RPC:       freeing buffer of size %zu at %p\n",
 786			size, buf);
 787
 788	if (size <= RPC_BUFFER_MAXSIZE)
 789		mempool_free(buf, rpc_buffer_mempool);
 790	else
 791		kfree(buf);
 792}
 793EXPORT_SYMBOL_GPL(rpc_free);
 794
 795/*
 796 * Creation and deletion of RPC task structures
 797 */
 798static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 799{
 800	memset(task, 0, sizeof(*task));
 801	atomic_set(&task->tk_count, 1);
 802	task->tk_flags  = task_setup_data->flags;
 803	task->tk_ops = task_setup_data->callback_ops;
 804	task->tk_calldata = task_setup_data->callback_data;
 805	INIT_LIST_HEAD(&task->tk_task);
 806
 807	/* Initialize retry counters */
 808	task->tk_garb_retry = 2;
 809	task->tk_cred_retry = 2;
 810	task->tk_rebind_retry = 2;
 811
 812	task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 813	task->tk_owner = current->tgid;
 814
 815	/* Initialize workqueue for async tasks */
 816	task->tk_workqueue = task_setup_data->workqueue;
 817
 
 
 818	if (task->tk_ops->rpc_call_prepare != NULL)
 819		task->tk_action = rpc_prepare_task;
 820
 821	/* starting timestamp */
 822	task->tk_start = ktime_get();
 823
 824	dprintk("RPC:       new task initialized, procpid %u\n",
 825				task_pid_nr(current));
 826}
 827
 828static struct rpc_task *
 829rpc_alloc_task(void)
 830{
 831	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 832}
 833
 834/*
 835 * Create a new task for the specified client.
 836 */
 837struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 838{
 839	struct rpc_task	*task = setup_data->task;
 840	unsigned short flags = 0;
 841
 842	if (task == NULL) {
 843		task = rpc_alloc_task();
 844		if (task == NULL) {
 845			rpc_release_calldata(setup_data->callback_ops,
 846					setup_data->callback_data);
 847			return ERR_PTR(-ENOMEM);
 848		}
 849		flags = RPC_TASK_DYNAMIC;
 850	}
 851
 852	rpc_init_task(task, setup_data);
 853	task->tk_flags |= flags;
 854	dprintk("RPC:       allocated task %p\n", task);
 855	return task;
 856}
 857
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 858static void rpc_free_task(struct rpc_task *task)
 859{
 860	const struct rpc_call_ops *tk_ops = task->tk_ops;
 861	void *calldata = task->tk_calldata;
 
 862
 863	if (task->tk_flags & RPC_TASK_DYNAMIC) {
 864		dprintk("RPC: %5u freeing task\n", task->tk_pid);
 865		mempool_free(task, rpc_task_mempool);
 866	}
 867	rpc_release_calldata(tk_ops, calldata);
 868}
 869
 870static void rpc_async_release(struct work_struct *work)
 871{
 872	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 873}
 874
 875static void rpc_release_resources_task(struct rpc_task *task)
 876{
 877	if (task->tk_rqstp)
 878		xprt_release(task);
 879	if (task->tk_msg.rpc_cred) {
 880		put_rpccred(task->tk_msg.rpc_cred);
 881		task->tk_msg.rpc_cred = NULL;
 882	}
 883	rpc_task_release_client(task);
 884}
 885
 886static void rpc_final_put_task(struct rpc_task *task,
 887		struct workqueue_struct *q)
 888{
 889	if (q != NULL) {
 890		INIT_WORK(&task->u.tk_work, rpc_async_release);
 891		queue_work(q, &task->u.tk_work);
 892	} else
 893		rpc_free_task(task);
 894}
 895
 896static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
 897{
 898	if (atomic_dec_and_test(&task->tk_count)) {
 899		rpc_release_resources_task(task);
 900		rpc_final_put_task(task, q);
 901	}
 902}
 903
 904void rpc_put_task(struct rpc_task *task)
 905{
 906	rpc_do_put_task(task, NULL);
 907}
 908EXPORT_SYMBOL_GPL(rpc_put_task);
 909
 910void rpc_put_task_async(struct rpc_task *task)
 911{
 912	rpc_do_put_task(task, task->tk_workqueue);
 913}
 914EXPORT_SYMBOL_GPL(rpc_put_task_async);
 915
 916static void rpc_release_task(struct rpc_task *task)
 917{
 918	dprintk("RPC: %5u release task\n", task->tk_pid);
 919
 920	BUG_ON (RPC_IS_QUEUED(task));
 921
 922	rpc_release_resources_task(task);
 923
 924	/*
 925	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
 926	 * so it should be safe to use task->tk_count as a test for whether
 927	 * or not any other processes still hold references to our rpc_task.
 928	 */
 929	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
 930		/* Wake up anyone who may be waiting for task completion */
 931		if (!rpc_complete_task(task))
 932			return;
 933	} else {
 934		if (!atomic_dec_and_test(&task->tk_count))
 935			return;
 936	}
 937	rpc_final_put_task(task, task->tk_workqueue);
 938}
 939
 940int rpciod_up(void)
 941{
 942	return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
 943}
 944
 945void rpciod_down(void)
 946{
 947	module_put(THIS_MODULE);
 948}
 949
 950/*
 951 * Start up the rpciod workqueue.
 952 */
 953static int rpciod_start(void)
 954{
 955	struct workqueue_struct *wq;
 956
 957	/*
 958	 * Create the rpciod thread and wait for it to start.
 959	 */
 960	dprintk("RPC:       creating workqueue rpciod\n");
 961	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
 
 962	rpciod_workqueue = wq;
 963	return rpciod_workqueue != NULL;
 964}
 965
 966static void rpciod_stop(void)
 967{
 968	struct workqueue_struct *wq = NULL;
 969
 970	if (rpciod_workqueue == NULL)
 971		return;
 972	dprintk("RPC:       destroying workqueue rpciod\n");
 973
 974	wq = rpciod_workqueue;
 975	rpciod_workqueue = NULL;
 976	destroy_workqueue(wq);
 977}
 978
 979void
 980rpc_destroy_mempool(void)
 981{
 982	rpciod_stop();
 983	if (rpc_buffer_mempool)
 984		mempool_destroy(rpc_buffer_mempool);
 985	if (rpc_task_mempool)
 986		mempool_destroy(rpc_task_mempool);
 987	if (rpc_task_slabp)
 988		kmem_cache_destroy(rpc_task_slabp);
 989	if (rpc_buffer_slabp)
 990		kmem_cache_destroy(rpc_buffer_slabp);
 991	rpc_destroy_wait_queue(&delay_queue);
 992}
 993
 994int
 995rpc_init_mempool(void)
 996{
 997	/*
 998	 * The following is not strictly a mempool initialisation,
 999	 * but there is no harm in doing it here
1000	 */
1001	rpc_init_wait_queue(&delay_queue, "delayq");
1002	if (!rpciod_start())
1003		goto err_nomem;
1004
1005	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1006					     sizeof(struct rpc_task),
1007					     0, SLAB_HWCACHE_ALIGN,
1008					     NULL);
1009	if (!rpc_task_slabp)
1010		goto err_nomem;
1011	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1012					     RPC_BUFFER_MAXSIZE,
1013					     0, SLAB_HWCACHE_ALIGN,
1014					     NULL);
1015	if (!rpc_buffer_slabp)
1016		goto err_nomem;
1017	rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1018						    rpc_task_slabp);
1019	if (!rpc_task_mempool)
1020		goto err_nomem;
1021	rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1022						      rpc_buffer_slabp);
1023	if (!rpc_buffer_mempool)
1024		goto err_nomem;
1025	return 0;
1026err_nomem:
1027	rpc_destroy_mempool();
1028	return -ENOMEM;
1029}