xref: /openbmc/linux/net/sunrpc/sched.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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/smp_lock.h>
20 #include <linux/spinlock.h>
21 
22 #include <linux/sunrpc/clnt.h>
23 #include <linux/sunrpc/xprt.h>
24 
25 #ifdef RPC_DEBUG
26 #define RPCDBG_FACILITY		RPCDBG_SCHED
27 #define RPC_TASK_MAGIC_ID	0xf00baa
28 static int			rpc_task_id;
29 #endif
30 
31 /*
32  * RPC slabs and memory pools
33  */
34 #define RPC_BUFFER_MAXSIZE	(2048)
35 #define RPC_BUFFER_POOLSIZE	(8)
36 #define RPC_TASK_POOLSIZE	(8)
37 static kmem_cache_t	*rpc_task_slabp __read_mostly;
38 static kmem_cache_t	*rpc_buffer_slabp __read_mostly;
39 static mempool_t	*rpc_task_mempool __read_mostly;
40 static mempool_t	*rpc_buffer_mempool __read_mostly;
41 
42 static void			__rpc_default_timer(struct rpc_task *task);
43 static void			rpciod_killall(void);
44 static void			rpc_free(struct rpc_task *task);
45 
46 static void			rpc_async_schedule(void *);
47 
48 /*
49  * RPC tasks that create another task (e.g. for contacting the portmapper)
50  * will wait on this queue for their child's completion
51  */
52 static RPC_WAITQ(childq, "childq");
53 
54 /*
55  * RPC tasks sit here while waiting for conditions to improve.
56  */
57 static RPC_WAITQ(delay_queue, "delayq");
58 
59 /*
60  * All RPC tasks are linked into this list
61  */
62 static LIST_HEAD(all_tasks);
63 
64 /*
65  * rpciod-related stuff
66  */
67 static DECLARE_MUTEX(rpciod_sema);
68 static unsigned int		rpciod_users;
69 static struct workqueue_struct *rpciod_workqueue;
70 
71 /*
72  * Spinlock for other critical sections of code.
73  */
74 static DEFINE_SPINLOCK(rpc_sched_lock);
75 
76 /*
77  * Disable the timer for a given RPC task. Should be called with
78  * queue->lock and bh_disabled in order to avoid races within
79  * rpc_run_timer().
80  */
81 static inline void
82 __rpc_disable_timer(struct rpc_task *task)
83 {
84 	dprintk("RPC: %4d disabling timer\n", task->tk_pid);
85 	task->tk_timeout_fn = NULL;
86 	task->tk_timeout = 0;
87 }
88 
89 /*
90  * Run a timeout function.
91  * We use the callback in order to allow __rpc_wake_up_task()
92  * and friends to disable the timer synchronously on SMP systems
93  * without calling del_timer_sync(). The latter could cause a
94  * deadlock if called while we're holding spinlocks...
95  */
96 static void rpc_run_timer(struct rpc_task *task)
97 {
98 	void (*callback)(struct rpc_task *);
99 
100 	callback = task->tk_timeout_fn;
101 	task->tk_timeout_fn = NULL;
102 	if (callback && RPC_IS_QUEUED(task)) {
103 		dprintk("RPC: %4d running timer\n", task->tk_pid);
104 		callback(task);
105 	}
106 	smp_mb__before_clear_bit();
107 	clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
108 	smp_mb__after_clear_bit();
109 }
110 
111 /*
112  * Set up a timer for the current task.
113  */
114 static inline void
115 __rpc_add_timer(struct rpc_task *task, rpc_action timer)
116 {
117 	if (!task->tk_timeout)
118 		return;
119 
120 	dprintk("RPC: %4d setting alarm for %lu ms\n",
121 			task->tk_pid, task->tk_timeout * 1000 / HZ);
122 
123 	if (timer)
124 		task->tk_timeout_fn = timer;
125 	else
126 		task->tk_timeout_fn = __rpc_default_timer;
127 	set_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
128 	mod_timer(&task->tk_timer, jiffies + task->tk_timeout);
129 }
130 
131 /*
132  * Delete any timer for the current task. Because we use del_timer_sync(),
133  * this function should never be called while holding queue->lock.
134  */
135 static void
136 rpc_delete_timer(struct rpc_task *task)
137 {
138 	if (RPC_IS_QUEUED(task))
139 		return;
140 	if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
141 		del_singleshot_timer_sync(&task->tk_timer);
142 		dprintk("RPC: %4d deleting timer\n", task->tk_pid);
143 	}
144 }
145 
146 /*
147  * Add new request to a priority queue.
148  */
149 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
150 {
151 	struct list_head *q;
152 	struct rpc_task *t;
153 
154 	INIT_LIST_HEAD(&task->u.tk_wait.links);
155 	q = &queue->tasks[task->tk_priority];
156 	if (unlikely(task->tk_priority > queue->maxpriority))
157 		q = &queue->tasks[queue->maxpriority];
158 	list_for_each_entry(t, q, u.tk_wait.list) {
159 		if (t->tk_cookie == task->tk_cookie) {
160 			list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
161 			return;
162 		}
163 	}
164 	list_add_tail(&task->u.tk_wait.list, q);
165 }
166 
167 /*
168  * Add new request to wait queue.
169  *
170  * Swapper tasks always get inserted at the head of the queue.
171  * This should avoid many nasty memory deadlocks and hopefully
172  * improve overall performance.
173  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
174  */
175 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
176 {
177 	BUG_ON (RPC_IS_QUEUED(task));
178 
179 	if (RPC_IS_PRIORITY(queue))
180 		__rpc_add_wait_queue_priority(queue, task);
181 	else if (RPC_IS_SWAPPER(task))
182 		list_add(&task->u.tk_wait.list, &queue->tasks[0]);
183 	else
184 		list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
185 	task->u.tk_wait.rpc_waitq = queue;
186 	rpc_set_queued(task);
187 
188 	dprintk("RPC: %4d added to queue %p \"%s\"\n",
189 				task->tk_pid, queue, rpc_qname(queue));
190 }
191 
192 /*
193  * Remove request from a priority queue.
194  */
195 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
196 {
197 	struct rpc_task *t;
198 
199 	if (!list_empty(&task->u.tk_wait.links)) {
200 		t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
201 		list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
202 		list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
203 	}
204 	list_del(&task->u.tk_wait.list);
205 }
206 
207 /*
208  * Remove request from queue.
209  * Note: must be called with spin lock held.
210  */
211 static void __rpc_remove_wait_queue(struct rpc_task *task)
212 {
213 	struct rpc_wait_queue *queue;
214 	queue = task->u.tk_wait.rpc_waitq;
215 
216 	if (RPC_IS_PRIORITY(queue))
217 		__rpc_remove_wait_queue_priority(task);
218 	else
219 		list_del(&task->u.tk_wait.list);
220 	dprintk("RPC: %4d removed from queue %p \"%s\"\n",
221 				task->tk_pid, queue, rpc_qname(queue));
222 }
223 
224 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
225 {
226 	queue->priority = priority;
227 	queue->count = 1 << (priority * 2);
228 }
229 
230 static inline void rpc_set_waitqueue_cookie(struct rpc_wait_queue *queue, unsigned long cookie)
231 {
232 	queue->cookie = cookie;
233 	queue->nr = RPC_BATCH_COUNT;
234 }
235 
236 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
237 {
238 	rpc_set_waitqueue_priority(queue, queue->maxpriority);
239 	rpc_set_waitqueue_cookie(queue, 0);
240 }
241 
242 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, int maxprio)
243 {
244 	int i;
245 
246 	spin_lock_init(&queue->lock);
247 	for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
248 		INIT_LIST_HEAD(&queue->tasks[i]);
249 	queue->maxpriority = maxprio;
250 	rpc_reset_waitqueue_priority(queue);
251 #ifdef RPC_DEBUG
252 	queue->name = qname;
253 #endif
254 }
255 
256 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
257 {
258 	__rpc_init_priority_wait_queue(queue, qname, RPC_PRIORITY_HIGH);
259 }
260 
261 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
262 {
263 	__rpc_init_priority_wait_queue(queue, qname, 0);
264 }
265 EXPORT_SYMBOL(rpc_init_wait_queue);
266 
267 /*
268  * Make an RPC task runnable.
269  *
270  * Note: If the task is ASYNC, this must be called with
271  * the spinlock held to protect the wait queue operation.
272  */
273 static void rpc_make_runnable(struct rpc_task *task)
274 {
275 	int do_ret;
276 
277 	BUG_ON(task->tk_timeout_fn);
278 	do_ret = rpc_test_and_set_running(task);
279 	rpc_clear_queued(task);
280 	if (do_ret)
281 		return;
282 	if (RPC_IS_ASYNC(task)) {
283 		int status;
284 
285 		INIT_WORK(&task->u.tk_work, rpc_async_schedule, (void *)task);
286 		status = queue_work(task->tk_workqueue, &task->u.tk_work);
287 		if (status < 0) {
288 			printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
289 			task->tk_status = status;
290 			return;
291 		}
292 	} else
293 		wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
294 }
295 
296 /*
297  * Place a newly initialized task on the workqueue.
298  */
299 static inline void
300 rpc_schedule_run(struct rpc_task *task)
301 {
302 	/* Don't run a child twice! */
303 	if (RPC_IS_ACTIVATED(task))
304 		return;
305 	task->tk_active = 1;
306 	rpc_make_runnable(task);
307 }
308 
309 /*
310  * Prepare for sleeping on a wait queue.
311  * By always appending tasks to the list we ensure FIFO behavior.
312  * NB: An RPC task will only receive interrupt-driven events as long
313  * as it's on a wait queue.
314  */
315 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
316 			rpc_action action, rpc_action timer)
317 {
318 	dprintk("RPC: %4d sleep_on(queue \"%s\" time %ld)\n", task->tk_pid,
319 				rpc_qname(q), jiffies);
320 
321 	if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
322 		printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
323 		return;
324 	}
325 
326 	/* Mark the task as being activated if so needed */
327 	if (!RPC_IS_ACTIVATED(task))
328 		task->tk_active = 1;
329 
330 	__rpc_add_wait_queue(q, task);
331 
332 	BUG_ON(task->tk_callback != NULL);
333 	task->tk_callback = action;
334 	__rpc_add_timer(task, timer);
335 }
336 
337 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
338 				rpc_action action, rpc_action timer)
339 {
340 	/*
341 	 * Protect the queue operations.
342 	 */
343 	spin_lock_bh(&q->lock);
344 	__rpc_sleep_on(q, task, action, timer);
345 	spin_unlock_bh(&q->lock);
346 }
347 
348 /**
349  * __rpc_do_wake_up_task - wake up a single rpc_task
350  * @task: task to be woken up
351  *
352  * Caller must hold queue->lock, and have cleared the task queued flag.
353  */
354 static void __rpc_do_wake_up_task(struct rpc_task *task)
355 {
356 	dprintk("RPC: %4d __rpc_wake_up_task (now %ld)\n", task->tk_pid, jiffies);
357 
358 #ifdef RPC_DEBUG
359 	BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
360 #endif
361 	/* Has the task been executed yet? If not, we cannot wake it up! */
362 	if (!RPC_IS_ACTIVATED(task)) {
363 		printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
364 		return;
365 	}
366 
367 	__rpc_disable_timer(task);
368 	__rpc_remove_wait_queue(task);
369 
370 	rpc_make_runnable(task);
371 
372 	dprintk("RPC:      __rpc_wake_up_task done\n");
373 }
374 
375 /*
376  * Wake up the specified task
377  */
378 static void __rpc_wake_up_task(struct rpc_task *task)
379 {
380 	if (rpc_start_wakeup(task)) {
381 		if (RPC_IS_QUEUED(task))
382 			__rpc_do_wake_up_task(task);
383 		rpc_finish_wakeup(task);
384 	}
385 }
386 
387 /*
388  * Default timeout handler if none specified by user
389  */
390 static void
391 __rpc_default_timer(struct rpc_task *task)
392 {
393 	dprintk("RPC: %d timeout (default timer)\n", task->tk_pid);
394 	task->tk_status = -ETIMEDOUT;
395 	rpc_wake_up_task(task);
396 }
397 
398 /*
399  * Wake up the specified task
400  */
401 void rpc_wake_up_task(struct rpc_task *task)
402 {
403 	if (rpc_start_wakeup(task)) {
404 		if (RPC_IS_QUEUED(task)) {
405 			struct rpc_wait_queue *queue = task->u.tk_wait.rpc_waitq;
406 
407 			spin_lock_bh(&queue->lock);
408 			__rpc_do_wake_up_task(task);
409 			spin_unlock_bh(&queue->lock);
410 		}
411 		rpc_finish_wakeup(task);
412 	}
413 }
414 
415 /*
416  * Wake up the next task on a priority queue.
417  */
418 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
419 {
420 	struct list_head *q;
421 	struct rpc_task *task;
422 
423 	/*
424 	 * Service a batch of tasks from a single cookie.
425 	 */
426 	q = &queue->tasks[queue->priority];
427 	if (!list_empty(q)) {
428 		task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
429 		if (queue->cookie == task->tk_cookie) {
430 			if (--queue->nr)
431 				goto out;
432 			list_move_tail(&task->u.tk_wait.list, q);
433 		}
434 		/*
435 		 * Check if we need to switch queues.
436 		 */
437 		if (--queue->count)
438 			goto new_cookie;
439 	}
440 
441 	/*
442 	 * Service the next queue.
443 	 */
444 	do {
445 		if (q == &queue->tasks[0])
446 			q = &queue->tasks[queue->maxpriority];
447 		else
448 			q = q - 1;
449 		if (!list_empty(q)) {
450 			task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
451 			goto new_queue;
452 		}
453 	} while (q != &queue->tasks[queue->priority]);
454 
455 	rpc_reset_waitqueue_priority(queue);
456 	return NULL;
457 
458 new_queue:
459 	rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
460 new_cookie:
461 	rpc_set_waitqueue_cookie(queue, task->tk_cookie);
462 out:
463 	__rpc_wake_up_task(task);
464 	return task;
465 }
466 
467 /*
468  * Wake up the next task on the wait queue.
469  */
470 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
471 {
472 	struct rpc_task	*task = NULL;
473 
474 	dprintk("RPC:      wake_up_next(%p \"%s\")\n", queue, rpc_qname(queue));
475 	spin_lock_bh(&queue->lock);
476 	if (RPC_IS_PRIORITY(queue))
477 		task = __rpc_wake_up_next_priority(queue);
478 	else {
479 		task_for_first(task, &queue->tasks[0])
480 			__rpc_wake_up_task(task);
481 	}
482 	spin_unlock_bh(&queue->lock);
483 
484 	return task;
485 }
486 
487 /**
488  * rpc_wake_up - wake up all rpc_tasks
489  * @queue: rpc_wait_queue on which the tasks are sleeping
490  *
491  * Grabs queue->lock
492  */
493 void rpc_wake_up(struct rpc_wait_queue *queue)
494 {
495 	struct rpc_task *task;
496 
497 	struct list_head *head;
498 	spin_lock_bh(&queue->lock);
499 	head = &queue->tasks[queue->maxpriority];
500 	for (;;) {
501 		while (!list_empty(head)) {
502 			task = list_entry(head->next, struct rpc_task, u.tk_wait.list);
503 			__rpc_wake_up_task(task);
504 		}
505 		if (head == &queue->tasks[0])
506 			break;
507 		head--;
508 	}
509 	spin_unlock_bh(&queue->lock);
510 }
511 
512 /**
513  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
514  * @queue: rpc_wait_queue on which the tasks are sleeping
515  * @status: status value to set
516  *
517  * Grabs queue->lock
518  */
519 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
520 {
521 	struct list_head *head;
522 	struct rpc_task *task;
523 
524 	spin_lock_bh(&queue->lock);
525 	head = &queue->tasks[queue->maxpriority];
526 	for (;;) {
527 		while (!list_empty(head)) {
528 			task = list_entry(head->next, struct rpc_task, u.tk_wait.list);
529 			task->tk_status = status;
530 			__rpc_wake_up_task(task);
531 		}
532 		if (head == &queue->tasks[0])
533 			break;
534 		head--;
535 	}
536 	spin_unlock_bh(&queue->lock);
537 }
538 
539 /*
540  * Run a task at a later time
541  */
542 static void	__rpc_atrun(struct rpc_task *);
543 void
544 rpc_delay(struct rpc_task *task, unsigned long delay)
545 {
546 	task->tk_timeout = delay;
547 	rpc_sleep_on(&delay_queue, task, NULL, __rpc_atrun);
548 }
549 
550 static void
551 __rpc_atrun(struct rpc_task *task)
552 {
553 	task->tk_status = 0;
554 	rpc_wake_up_task(task);
555 }
556 
557 /*
558  * Helper that calls task->tk_exit if it exists and then returns
559  * true if we should exit __rpc_execute.
560  */
561 static inline int __rpc_do_exit(struct rpc_task *task)
562 {
563 	if (task->tk_exit != NULL) {
564 		lock_kernel();
565 		task->tk_exit(task);
566 		unlock_kernel();
567 		/* If tk_action is non-null, we should restart the call */
568 		if (task->tk_action != NULL) {
569 			if (!RPC_ASSASSINATED(task)) {
570 				/* Release RPC slot and buffer memory */
571 				xprt_release(task);
572 				rpc_free(task);
573 				return 0;
574 			}
575 			printk(KERN_ERR "RPC: dead task tried to walk away.\n");
576 		}
577 	}
578 	return 1;
579 }
580 
581 static int rpc_wait_bit_interruptible(void *word)
582 {
583 	if (signal_pending(current))
584 		return -ERESTARTSYS;
585 	schedule();
586 	return 0;
587 }
588 
589 /*
590  * This is the RPC `scheduler' (or rather, the finite state machine).
591  */
592 static int __rpc_execute(struct rpc_task *task)
593 {
594 	int		status = 0;
595 
596 	dprintk("RPC: %4d rpc_execute flgs %x\n",
597 				task->tk_pid, task->tk_flags);
598 
599 	BUG_ON(RPC_IS_QUEUED(task));
600 
601 	for (;;) {
602 		/*
603 		 * Garbage collection of pending timers...
604 		 */
605 		rpc_delete_timer(task);
606 
607 		/*
608 		 * Execute any pending callback.
609 		 */
610 		if (RPC_DO_CALLBACK(task)) {
611 			/* Define a callback save pointer */
612 			void (*save_callback)(struct rpc_task *);
613 
614 			/*
615 			 * If a callback exists, save it, reset it,
616 			 * call it.
617 			 * The save is needed to stop from resetting
618 			 * another callback set within the callback handler
619 			 * - Dave
620 			 */
621 			save_callback=task->tk_callback;
622 			task->tk_callback=NULL;
623 			lock_kernel();
624 			save_callback(task);
625 			unlock_kernel();
626 		}
627 
628 		/*
629 		 * Perform the next FSM step.
630 		 * tk_action may be NULL when the task has been killed
631 		 * by someone else.
632 		 */
633 		if (!RPC_IS_QUEUED(task)) {
634 			if (task->tk_action != NULL) {
635 				lock_kernel();
636 				task->tk_action(task);
637 				unlock_kernel();
638 			} else if (__rpc_do_exit(task))
639 				break;
640 		}
641 
642 		/*
643 		 * Lockless check for whether task is sleeping or not.
644 		 */
645 		if (!RPC_IS_QUEUED(task))
646 			continue;
647 		rpc_clear_running(task);
648 		if (RPC_IS_ASYNC(task)) {
649 			/* Careful! we may have raced... */
650 			if (RPC_IS_QUEUED(task))
651 				return 0;
652 			if (rpc_test_and_set_running(task))
653 				return 0;
654 			continue;
655 		}
656 
657 		/* sync task: sleep here */
658 		dprintk("RPC: %4d sync task going to sleep\n", task->tk_pid);
659 		/* Note: Caller should be using rpc_clnt_sigmask() */
660 		status = out_of_line_wait_on_bit(&task->tk_runstate,
661 				RPC_TASK_QUEUED, rpc_wait_bit_interruptible,
662 				TASK_INTERRUPTIBLE);
663 		if (status == -ERESTARTSYS) {
664 			/*
665 			 * When a sync task receives a signal, it exits with
666 			 * -ERESTARTSYS. In order to catch any callbacks that
667 			 * clean up after sleeping on some queue, we don't
668 			 * break the loop here, but go around once more.
669 			 */
670 			dprintk("RPC: %4d got signal\n", task->tk_pid);
671 			task->tk_flags |= RPC_TASK_KILLED;
672 			rpc_exit(task, -ERESTARTSYS);
673 			rpc_wake_up_task(task);
674 		}
675 		rpc_set_running(task);
676 		dprintk("RPC: %4d sync task resuming\n", task->tk_pid);
677 	}
678 
679 	dprintk("RPC: %4d exit() = %d\n", task->tk_pid, task->tk_status);
680 	status = task->tk_status;
681 
682 	/* Release all resources associated with the task */
683 	rpc_release_task(task);
684 	return status;
685 }
686 
687 /*
688  * User-visible entry point to the scheduler.
689  *
690  * This may be called recursively if e.g. an async NFS task updates
691  * the attributes and finds that dirty pages must be flushed.
692  * NOTE: Upon exit of this function the task is guaranteed to be
693  *	 released. In particular note that tk_release() will have
694  *	 been called, so your task memory may have been freed.
695  */
696 int
697 rpc_execute(struct rpc_task *task)
698 {
699 	BUG_ON(task->tk_active);
700 
701 	task->tk_active = 1;
702 	rpc_set_running(task);
703 	return __rpc_execute(task);
704 }
705 
706 static void rpc_async_schedule(void *arg)
707 {
708 	__rpc_execute((struct rpc_task *)arg);
709 }
710 
711 /*
712  * Allocate memory for RPC purposes.
713  *
714  * We try to ensure that some NFS reads and writes can always proceed
715  * by using a mempool when allocating 'small' buffers.
716  * In order to avoid memory starvation triggering more writebacks of
717  * NFS requests, we use GFP_NOFS rather than GFP_KERNEL.
718  */
719 void *
720 rpc_malloc(struct rpc_task *task, size_t size)
721 {
722 	gfp_t	gfp;
723 
724 	if (task->tk_flags & RPC_TASK_SWAPPER)
725 		gfp = GFP_ATOMIC;
726 	else
727 		gfp = GFP_NOFS;
728 
729 	if (size > RPC_BUFFER_MAXSIZE) {
730 		task->tk_buffer =  kmalloc(size, gfp);
731 		if (task->tk_buffer)
732 			task->tk_bufsize = size;
733 	} else {
734 		task->tk_buffer =  mempool_alloc(rpc_buffer_mempool, gfp);
735 		if (task->tk_buffer)
736 			task->tk_bufsize = RPC_BUFFER_MAXSIZE;
737 	}
738 	return task->tk_buffer;
739 }
740 
741 static void
742 rpc_free(struct rpc_task *task)
743 {
744 	if (task->tk_buffer) {
745 		if (task->tk_bufsize == RPC_BUFFER_MAXSIZE)
746 			mempool_free(task->tk_buffer, rpc_buffer_mempool);
747 		else
748 			kfree(task->tk_buffer);
749 		task->tk_buffer = NULL;
750 		task->tk_bufsize = 0;
751 	}
752 }
753 
754 /*
755  * Creation and deletion of RPC task structures
756  */
757 void rpc_init_task(struct rpc_task *task, struct rpc_clnt *clnt, rpc_action callback, int flags)
758 {
759 	memset(task, 0, sizeof(*task));
760 	init_timer(&task->tk_timer);
761 	task->tk_timer.data     = (unsigned long) task;
762 	task->tk_timer.function = (void (*)(unsigned long)) rpc_run_timer;
763 	task->tk_client = clnt;
764 	task->tk_flags  = flags;
765 	task->tk_exit   = callback;
766 
767 	/* Initialize retry counters */
768 	task->tk_garb_retry = 2;
769 	task->tk_cred_retry = 2;
770 
771 	task->tk_priority = RPC_PRIORITY_NORMAL;
772 	task->tk_cookie = (unsigned long)current;
773 
774 	/* Initialize workqueue for async tasks */
775 	task->tk_workqueue = rpciod_workqueue;
776 
777 	if (clnt) {
778 		atomic_inc(&clnt->cl_users);
779 		if (clnt->cl_softrtry)
780 			task->tk_flags |= RPC_TASK_SOFT;
781 		if (!clnt->cl_intr)
782 			task->tk_flags |= RPC_TASK_NOINTR;
783 	}
784 
785 #ifdef RPC_DEBUG
786 	task->tk_magic = RPC_TASK_MAGIC_ID;
787 	task->tk_pid = rpc_task_id++;
788 #endif
789 	/* Add to global list of all tasks */
790 	spin_lock(&rpc_sched_lock);
791 	list_add_tail(&task->tk_task, &all_tasks);
792 	spin_unlock(&rpc_sched_lock);
793 
794 	dprintk("RPC: %4d new task procpid %d\n", task->tk_pid,
795 				current->pid);
796 }
797 
798 static struct rpc_task *
799 rpc_alloc_task(void)
800 {
801 	return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
802 }
803 
804 static void
805 rpc_default_free_task(struct rpc_task *task)
806 {
807 	dprintk("RPC: %4d freeing task\n", task->tk_pid);
808 	mempool_free(task, rpc_task_mempool);
809 }
810 
811 /*
812  * Create a new task for the specified client.  We have to
813  * clean up after an allocation failure, as the client may
814  * have specified "oneshot".
815  */
816 struct rpc_task *
817 rpc_new_task(struct rpc_clnt *clnt, rpc_action callback, int flags)
818 {
819 	struct rpc_task	*task;
820 
821 	task = rpc_alloc_task();
822 	if (!task)
823 		goto cleanup;
824 
825 	rpc_init_task(task, clnt, callback, flags);
826 
827 	/* Replace tk_release */
828 	task->tk_release = rpc_default_free_task;
829 
830 	dprintk("RPC: %4d allocated task\n", task->tk_pid);
831 	task->tk_flags |= RPC_TASK_DYNAMIC;
832 out:
833 	return task;
834 
835 cleanup:
836 	/* Check whether to release the client */
837 	if (clnt) {
838 		printk("rpc_new_task: failed, users=%d, oneshot=%d\n",
839 			atomic_read(&clnt->cl_users), clnt->cl_oneshot);
840 		atomic_inc(&clnt->cl_users); /* pretend we were used ... */
841 		rpc_release_client(clnt);
842 	}
843 	goto out;
844 }
845 
846 void rpc_release_task(struct rpc_task *task)
847 {
848 	dprintk("RPC: %4d release task\n", task->tk_pid);
849 
850 #ifdef RPC_DEBUG
851 	BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
852 #endif
853 
854 	/* Remove from global task list */
855 	spin_lock(&rpc_sched_lock);
856 	list_del(&task->tk_task);
857 	spin_unlock(&rpc_sched_lock);
858 
859 	BUG_ON (RPC_IS_QUEUED(task));
860 	task->tk_active = 0;
861 
862 	/* Synchronously delete any running timer */
863 	rpc_delete_timer(task);
864 
865 	/* Release resources */
866 	if (task->tk_rqstp)
867 		xprt_release(task);
868 	if (task->tk_msg.rpc_cred)
869 		rpcauth_unbindcred(task);
870 	rpc_free(task);
871 	if (task->tk_client) {
872 		rpc_release_client(task->tk_client);
873 		task->tk_client = NULL;
874 	}
875 
876 #ifdef RPC_DEBUG
877 	task->tk_magic = 0;
878 #endif
879 	if (task->tk_release)
880 		task->tk_release(task);
881 }
882 
883 /**
884  * rpc_find_parent - find the parent of a child task.
885  * @child: child task
886  *
887  * Checks that the parent task is still sleeping on the
888  * queue 'childq'. If so returns a pointer to the parent.
889  * Upon failure returns NULL.
890  *
891  * Caller must hold childq.lock
892  */
893 static inline struct rpc_task *rpc_find_parent(struct rpc_task *child)
894 {
895 	struct rpc_task	*task, *parent;
896 	struct list_head *le;
897 
898 	parent = (struct rpc_task *) child->tk_calldata;
899 	task_for_each(task, le, &childq.tasks[0])
900 		if (task == parent)
901 			return parent;
902 
903 	return NULL;
904 }
905 
906 static void rpc_child_exit(struct rpc_task *child)
907 {
908 	struct rpc_task	*parent;
909 
910 	spin_lock_bh(&childq.lock);
911 	if ((parent = rpc_find_parent(child)) != NULL) {
912 		parent->tk_status = child->tk_status;
913 		__rpc_wake_up_task(parent);
914 	}
915 	spin_unlock_bh(&childq.lock);
916 }
917 
918 /*
919  * Note: rpc_new_task releases the client after a failure.
920  */
921 struct rpc_task *
922 rpc_new_child(struct rpc_clnt *clnt, struct rpc_task *parent)
923 {
924 	struct rpc_task	*task;
925 
926 	task = rpc_new_task(clnt, NULL, RPC_TASK_ASYNC | RPC_TASK_CHILD);
927 	if (!task)
928 		goto fail;
929 	task->tk_exit = rpc_child_exit;
930 	task->tk_calldata = parent;
931 	return task;
932 
933 fail:
934 	parent->tk_status = -ENOMEM;
935 	return NULL;
936 }
937 
938 void rpc_run_child(struct rpc_task *task, struct rpc_task *child, rpc_action func)
939 {
940 	spin_lock_bh(&childq.lock);
941 	/* N.B. Is it possible for the child to have already finished? */
942 	__rpc_sleep_on(&childq, task, func, NULL);
943 	rpc_schedule_run(child);
944 	spin_unlock_bh(&childq.lock);
945 }
946 
947 /*
948  * Kill all tasks for the given client.
949  * XXX: kill their descendants as well?
950  */
951 void rpc_killall_tasks(struct rpc_clnt *clnt)
952 {
953 	struct rpc_task	*rovr;
954 	struct list_head *le;
955 
956 	dprintk("RPC:      killing all tasks for client %p\n", clnt);
957 
958 	/*
959 	 * Spin lock all_tasks to prevent changes...
960 	 */
961 	spin_lock(&rpc_sched_lock);
962 	alltask_for_each(rovr, le, &all_tasks) {
963 		if (! RPC_IS_ACTIVATED(rovr))
964 			continue;
965 		if (!clnt || rovr->tk_client == clnt) {
966 			rovr->tk_flags |= RPC_TASK_KILLED;
967 			rpc_exit(rovr, -EIO);
968 			rpc_wake_up_task(rovr);
969 		}
970 	}
971 	spin_unlock(&rpc_sched_lock);
972 }
973 
974 static DECLARE_MUTEX_LOCKED(rpciod_running);
975 
976 static void rpciod_killall(void)
977 {
978 	unsigned long flags;
979 
980 	while (!list_empty(&all_tasks)) {
981 		clear_thread_flag(TIF_SIGPENDING);
982 		rpc_killall_tasks(NULL);
983 		flush_workqueue(rpciod_workqueue);
984 		if (!list_empty(&all_tasks)) {
985 			dprintk("rpciod_killall: waiting for tasks to exit\n");
986 			yield();
987 		}
988 	}
989 
990 	spin_lock_irqsave(&current->sighand->siglock, flags);
991 	recalc_sigpending();
992 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
993 }
994 
995 /*
996  * Start up the rpciod process if it's not already running.
997  */
998 int
999 rpciod_up(void)
1000 {
1001 	struct workqueue_struct *wq;
1002 	int error = 0;
1003 
1004 	down(&rpciod_sema);
1005 	dprintk("rpciod_up: users %d\n", rpciod_users);
1006 	rpciod_users++;
1007 	if (rpciod_workqueue)
1008 		goto out;
1009 	/*
1010 	 * If there's no pid, we should be the first user.
1011 	 */
1012 	if (rpciod_users > 1)
1013 		printk(KERN_WARNING "rpciod_up: no workqueue, %d users??\n", rpciod_users);
1014 	/*
1015 	 * Create the rpciod thread and wait for it to start.
1016 	 */
1017 	error = -ENOMEM;
1018 	wq = create_workqueue("rpciod");
1019 	if (wq == NULL) {
1020 		printk(KERN_WARNING "rpciod_up: create workqueue failed, error=%d\n", error);
1021 		rpciod_users--;
1022 		goto out;
1023 	}
1024 	rpciod_workqueue = wq;
1025 	error = 0;
1026 out:
1027 	up(&rpciod_sema);
1028 	return error;
1029 }
1030 
1031 void
1032 rpciod_down(void)
1033 {
1034 	down(&rpciod_sema);
1035 	dprintk("rpciod_down sema %d\n", rpciod_users);
1036 	if (rpciod_users) {
1037 		if (--rpciod_users)
1038 			goto out;
1039 	} else
1040 		printk(KERN_WARNING "rpciod_down: no users??\n");
1041 
1042 	if (!rpciod_workqueue) {
1043 		dprintk("rpciod_down: Nothing to do!\n");
1044 		goto out;
1045 	}
1046 	rpciod_killall();
1047 
1048 	destroy_workqueue(rpciod_workqueue);
1049 	rpciod_workqueue = NULL;
1050  out:
1051 	up(&rpciod_sema);
1052 }
1053 
1054 #ifdef RPC_DEBUG
1055 void rpc_show_tasks(void)
1056 {
1057 	struct list_head *le;
1058 	struct rpc_task *t;
1059 
1060 	spin_lock(&rpc_sched_lock);
1061 	if (list_empty(&all_tasks)) {
1062 		spin_unlock(&rpc_sched_lock);
1063 		return;
1064 	}
1065 	printk("-pid- proc flgs status -client- -prog- --rqstp- -timeout "
1066 		"-rpcwait -action- --exit--\n");
1067 	alltask_for_each(t, le, &all_tasks) {
1068 		const char *rpc_waitq = "none";
1069 
1070 		if (RPC_IS_QUEUED(t))
1071 			rpc_waitq = rpc_qname(t->u.tk_wait.rpc_waitq);
1072 
1073 		printk("%05d %04d %04x %06d %8p %6d %8p %08ld %8s %8p %8p\n",
1074 			t->tk_pid,
1075 			(t->tk_msg.rpc_proc ? t->tk_msg.rpc_proc->p_proc : -1),
1076 			t->tk_flags, t->tk_status,
1077 			t->tk_client,
1078 			(t->tk_client ? t->tk_client->cl_prog : 0),
1079 			t->tk_rqstp, t->tk_timeout,
1080 			rpc_waitq,
1081 			t->tk_action, t->tk_exit);
1082 	}
1083 	spin_unlock(&rpc_sched_lock);
1084 }
1085 #endif
1086 
1087 void
1088 rpc_destroy_mempool(void)
1089 {
1090 	if (rpc_buffer_mempool)
1091 		mempool_destroy(rpc_buffer_mempool);
1092 	if (rpc_task_mempool)
1093 		mempool_destroy(rpc_task_mempool);
1094 	if (rpc_task_slabp && kmem_cache_destroy(rpc_task_slabp))
1095 		printk(KERN_INFO "rpc_task: not all structures were freed\n");
1096 	if (rpc_buffer_slabp && kmem_cache_destroy(rpc_buffer_slabp))
1097 		printk(KERN_INFO "rpc_buffers: not all structures were freed\n");
1098 }
1099 
1100 int
1101 rpc_init_mempool(void)
1102 {
1103 	rpc_task_slabp = kmem_cache_create("rpc_tasks",
1104 					     sizeof(struct rpc_task),
1105 					     0, SLAB_HWCACHE_ALIGN,
1106 					     NULL, NULL);
1107 	if (!rpc_task_slabp)
1108 		goto err_nomem;
1109 	rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1110 					     RPC_BUFFER_MAXSIZE,
1111 					     0, SLAB_HWCACHE_ALIGN,
1112 					     NULL, NULL);
1113 	if (!rpc_buffer_slabp)
1114 		goto err_nomem;
1115 	rpc_task_mempool = mempool_create(RPC_TASK_POOLSIZE,
1116 					    mempool_alloc_slab,
1117 					    mempool_free_slab,
1118 					    rpc_task_slabp);
1119 	if (!rpc_task_mempool)
1120 		goto err_nomem;
1121 	rpc_buffer_mempool = mempool_create(RPC_BUFFER_POOLSIZE,
1122 					    mempool_alloc_slab,
1123 					    mempool_free_slab,
1124 					    rpc_buffer_slabp);
1125 	if (!rpc_buffer_mempool)
1126 		goto err_nomem;
1127 	return 0;
1128 err_nomem:
1129 	rpc_destroy_mempool();
1130 	return -ENOMEM;
1131 }
1132