Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/sched.c
4 *
5 * Scheduling for synchronous and asynchronous RPC requests.
6 *
7 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
8 *
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 */
12
13#include <linux/module.h>
14
15#include <linux/sched.h>
16#include <linux/interrupt.h>
17#include <linux/slab.h>
18#include <linux/mempool.h>
19#include <linux/smp.h>
20#include <linux/spinlock.h>
21#include <linux/mutex.h>
22#include <linux/freezer.h>
23#include <linux/sched/mm.h>
24
25#include <linux/sunrpc/clnt.h>
26#include <linux/sunrpc/metrics.h>
27
28#include "sunrpc.h"
29
30#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
31#define RPCDBG_FACILITY RPCDBG_SCHED
32#endif
33
34#define CREATE_TRACE_POINTS
35#include <trace/events/sunrpc.h>
36
37/*
38 * RPC slabs and memory pools
39 */
40#define RPC_BUFFER_MAXSIZE (2048)
41#define RPC_BUFFER_POOLSIZE (8)
42#define RPC_TASK_POOLSIZE (8)
43static struct kmem_cache *rpc_task_slabp __read_mostly;
44static struct kmem_cache *rpc_buffer_slabp __read_mostly;
45static mempool_t *rpc_task_mempool __read_mostly;
46static mempool_t *rpc_buffer_mempool __read_mostly;
47
48static void rpc_async_schedule(struct work_struct *);
49static void rpc_release_task(struct rpc_task *task);
50static void __rpc_queue_timer_fn(struct work_struct *);
51
52/*
53 * RPC tasks sit here while waiting for conditions to improve.
54 */
55static struct rpc_wait_queue delay_queue;
56
57/*
58 * rpciod-related stuff
59 */
60struct workqueue_struct *rpciod_workqueue __read_mostly;
61struct workqueue_struct *xprtiod_workqueue __read_mostly;
62EXPORT_SYMBOL_GPL(xprtiod_workqueue);
63
64unsigned long
65rpc_task_timeout(const struct rpc_task *task)
66{
67 unsigned long timeout = READ_ONCE(task->tk_timeout);
68
69 if (timeout != 0) {
70 unsigned long now = jiffies;
71 if (time_before(now, timeout))
72 return timeout - now;
73 }
74 return 0;
75}
76EXPORT_SYMBOL_GPL(rpc_task_timeout);
77
78/*
79 * Disable the timer for a given RPC task. Should be called with
80 * queue->lock and bh_disabled in order to avoid races within
81 * rpc_run_timer().
82 */
83static void
84__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
85{
86 if (list_empty(&task->u.tk_wait.timer_list))
87 return;
88 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
89 task->tk_timeout = 0;
90 list_del(&task->u.tk_wait.timer_list);
91 if (list_empty(&queue->timer_list.list))
92 cancel_delayed_work(&queue->timer_list.dwork);
93}
94
95static void
96rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
97{
98 unsigned long now = jiffies;
99 queue->timer_list.expires = expires;
100 if (time_before_eq(expires, now))
101 expires = 0;
102 else
103 expires -= now;
104 mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
105}
106
107/*
108 * Set up a timer for the current task.
109 */
110static void
111__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
112 unsigned long timeout)
113{
114 dprintk("RPC: %5u setting alarm for %u ms\n",
115 task->tk_pid, jiffies_to_msecs(timeout - jiffies));
116
117 task->tk_timeout = timeout;
118 if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
119 rpc_set_queue_timer(queue, timeout);
120 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
121}
122
123static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
124{
125 if (queue->priority != priority) {
126 queue->priority = priority;
127 queue->nr = 1U << priority;
128 }
129}
130
131static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
132{
133 rpc_set_waitqueue_priority(queue, queue->maxpriority);
134}
135
136/*
137 * Add a request to a queue list
138 */
139static void
140__rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
141{
142 struct rpc_task *t;
143
144 list_for_each_entry(t, q, u.tk_wait.list) {
145 if (t->tk_owner == task->tk_owner) {
146 list_add_tail(&task->u.tk_wait.links,
147 &t->u.tk_wait.links);
148 /* Cache the queue head in task->u.tk_wait.list */
149 task->u.tk_wait.list.next = q;
150 task->u.tk_wait.list.prev = NULL;
151 return;
152 }
153 }
154 INIT_LIST_HEAD(&task->u.tk_wait.links);
155 list_add_tail(&task->u.tk_wait.list, q);
156}
157
158/*
159 * Remove request from a queue list
160 */
161static void
162__rpc_list_dequeue_task(struct rpc_task *task)
163{
164 struct list_head *q;
165 struct rpc_task *t;
166
167 if (task->u.tk_wait.list.prev == NULL) {
168 list_del(&task->u.tk_wait.links);
169 return;
170 }
171 if (!list_empty(&task->u.tk_wait.links)) {
172 t = list_first_entry(&task->u.tk_wait.links,
173 struct rpc_task,
174 u.tk_wait.links);
175 /* Assume __rpc_list_enqueue_task() cached the queue head */
176 q = t->u.tk_wait.list.next;
177 list_add_tail(&t->u.tk_wait.list, q);
178 list_del(&task->u.tk_wait.links);
179 }
180 list_del(&task->u.tk_wait.list);
181}
182
183/*
184 * Add new request to a priority queue.
185 */
186static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
187 struct rpc_task *task,
188 unsigned char queue_priority)
189{
190 if (unlikely(queue_priority > queue->maxpriority))
191 queue_priority = queue->maxpriority;
192 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
193}
194
195/*
196 * Add new request to wait queue.
197 *
198 * Swapper tasks always get inserted at the head of the queue.
199 * This should avoid many nasty memory deadlocks and hopefully
200 * improve overall performance.
201 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
202 */
203static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
204 struct rpc_task *task,
205 unsigned char queue_priority)
206{
207 WARN_ON_ONCE(RPC_IS_QUEUED(task));
208 if (RPC_IS_QUEUED(task))
209 return;
210
211 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
212 if (RPC_IS_PRIORITY(queue))
213 __rpc_add_wait_queue_priority(queue, task, queue_priority);
214 else if (RPC_IS_SWAPPER(task))
215 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
216 else
217 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
218 task->tk_waitqueue = queue;
219 queue->qlen++;
220 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
221 smp_wmb();
222 rpc_set_queued(task);
223
224 dprintk("RPC: %5u added to queue %p \"%s\"\n",
225 task->tk_pid, queue, rpc_qname(queue));
226}
227
228/*
229 * Remove request from a priority queue.
230 */
231static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
232{
233 __rpc_list_dequeue_task(task);
234}
235
236/*
237 * Remove request from queue.
238 * Note: must be called with spin lock held.
239 */
240static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
241{
242 __rpc_disable_timer(queue, task);
243 if (RPC_IS_PRIORITY(queue))
244 __rpc_remove_wait_queue_priority(task);
245 else
246 list_del(&task->u.tk_wait.list);
247 queue->qlen--;
248 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
249 task->tk_pid, queue, rpc_qname(queue));
250}
251
252static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
253{
254 int i;
255
256 spin_lock_init(&queue->lock);
257 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
258 INIT_LIST_HEAD(&queue->tasks[i]);
259 queue->maxpriority = nr_queues - 1;
260 rpc_reset_waitqueue_priority(queue);
261 queue->qlen = 0;
262 queue->timer_list.expires = 0;
263 INIT_DEFERRABLE_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
264 INIT_LIST_HEAD(&queue->timer_list.list);
265 rpc_assign_waitqueue_name(queue, qname);
266}
267
268void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
269{
270 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
271}
272EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
273
274void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
275{
276 __rpc_init_priority_wait_queue(queue, qname, 1);
277}
278EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
279
280void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
281{
282 cancel_delayed_work_sync(&queue->timer_list.dwork);
283}
284EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
285
286static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
287{
288 freezable_schedule_unsafe();
289 if (signal_pending_state(mode, current))
290 return -ERESTARTSYS;
291 return 0;
292}
293
294#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
295static void rpc_task_set_debuginfo(struct rpc_task *task)
296{
297 static atomic_t rpc_pid;
298
299 task->tk_pid = atomic_inc_return(&rpc_pid);
300}
301#else
302static inline void rpc_task_set_debuginfo(struct rpc_task *task)
303{
304}
305#endif
306
307static void rpc_set_active(struct rpc_task *task)
308{
309 rpc_task_set_debuginfo(task);
310 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
311 trace_rpc_task_begin(task, NULL);
312}
313
314/*
315 * Mark an RPC call as having completed by clearing the 'active' bit
316 * and then waking up all tasks that were sleeping.
317 */
318static int rpc_complete_task(struct rpc_task *task)
319{
320 void *m = &task->tk_runstate;
321 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
322 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
323 unsigned long flags;
324 int ret;
325
326 trace_rpc_task_complete(task, NULL);
327
328 spin_lock_irqsave(&wq->lock, flags);
329 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
330 ret = atomic_dec_and_test(&task->tk_count);
331 if (waitqueue_active(wq))
332 __wake_up_locked_key(wq, TASK_NORMAL, &k);
333 spin_unlock_irqrestore(&wq->lock, flags);
334 return ret;
335}
336
337/*
338 * Allow callers to wait for completion of an RPC call
339 *
340 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
341 * to enforce taking of the wq->lock and hence avoid races with
342 * rpc_complete_task().
343 */
344int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
345{
346 if (action == NULL)
347 action = rpc_wait_bit_killable;
348 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
349 action, TASK_KILLABLE);
350}
351EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
352
353/*
354 * Make an RPC task runnable.
355 *
356 * Note: If the task is ASYNC, and is being made runnable after sitting on an
357 * rpc_wait_queue, this must be called with the queue spinlock held to protect
358 * the wait queue operation.
359 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
360 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
361 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
362 * the RPC_TASK_RUNNING flag.
363 */
364static void rpc_make_runnable(struct workqueue_struct *wq,
365 struct rpc_task *task)
366{
367 bool need_wakeup = !rpc_test_and_set_running(task);
368
369 rpc_clear_queued(task);
370 if (!need_wakeup)
371 return;
372 if (RPC_IS_ASYNC(task)) {
373 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
374 queue_work(wq, &task->u.tk_work);
375 } else
376 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
377}
378
379/*
380 * Prepare for sleeping on a wait queue.
381 * By always appending tasks to the list we ensure FIFO behavior.
382 * NB: An RPC task will only receive interrupt-driven events as long
383 * as it's on a wait queue.
384 */
385static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
386 struct rpc_task *task,
387 unsigned char queue_priority)
388{
389 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
390 task->tk_pid, rpc_qname(q), jiffies);
391
392 trace_rpc_task_sleep(task, q);
393
394 __rpc_add_wait_queue(q, task, queue_priority);
395
396}
397
398static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
399 struct rpc_task *task, unsigned long timeout,
400 unsigned char queue_priority)
401{
402 if (time_is_after_jiffies(timeout)) {
403 __rpc_sleep_on_priority(q, task, queue_priority);
404 __rpc_add_timer(q, task, timeout);
405 } else
406 task->tk_status = -ETIMEDOUT;
407}
408
409static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
410{
411 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
412 task->tk_callback = action;
413}
414
415static bool rpc_sleep_check_activated(struct rpc_task *task)
416{
417 /* We shouldn't ever put an inactive task to sleep */
418 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
419 task->tk_status = -EIO;
420 rpc_put_task_async(task);
421 return false;
422 }
423 return true;
424}
425
426void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
427 rpc_action action, unsigned long timeout)
428{
429 if (!rpc_sleep_check_activated(task))
430 return;
431
432 rpc_set_tk_callback(task, action);
433
434 /*
435 * Protect the queue operations.
436 */
437 spin_lock(&q->lock);
438 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
439 spin_unlock(&q->lock);
440}
441EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
442
443void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
444 rpc_action action)
445{
446 if (!rpc_sleep_check_activated(task))
447 return;
448
449 rpc_set_tk_callback(task, action);
450
451 WARN_ON_ONCE(task->tk_timeout != 0);
452 /*
453 * Protect the queue operations.
454 */
455 spin_lock(&q->lock);
456 __rpc_sleep_on_priority(q, task, task->tk_priority);
457 spin_unlock(&q->lock);
458}
459EXPORT_SYMBOL_GPL(rpc_sleep_on);
460
461void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
462 struct rpc_task *task, unsigned long timeout, int priority)
463{
464 if (!rpc_sleep_check_activated(task))
465 return;
466
467 priority -= RPC_PRIORITY_LOW;
468 /*
469 * Protect the queue operations.
470 */
471 spin_lock(&q->lock);
472 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
473 spin_unlock(&q->lock);
474}
475EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
476
477void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
478 int priority)
479{
480 if (!rpc_sleep_check_activated(task))
481 return;
482
483 WARN_ON_ONCE(task->tk_timeout != 0);
484 priority -= RPC_PRIORITY_LOW;
485 /*
486 * Protect the queue operations.
487 */
488 spin_lock(&q->lock);
489 __rpc_sleep_on_priority(q, task, priority);
490 spin_unlock(&q->lock);
491}
492EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
493
494/**
495 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
496 * @wq: workqueue on which to run task
497 * @queue: wait queue
498 * @task: task to be woken up
499 *
500 * Caller must hold queue->lock, and have cleared the task queued flag.
501 */
502static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
503 struct rpc_wait_queue *queue,
504 struct rpc_task *task)
505{
506 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
507 task->tk_pid, jiffies);
508
509 /* Has the task been executed yet? If not, we cannot wake it up! */
510 if (!RPC_IS_ACTIVATED(task)) {
511 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
512 return;
513 }
514
515 trace_rpc_task_wakeup(task, queue);
516
517 __rpc_remove_wait_queue(queue, task);
518
519 rpc_make_runnable(wq, task);
520
521 dprintk("RPC: __rpc_wake_up_task done\n");
522}
523
524/*
525 * Wake up a queued task while the queue lock is being held
526 */
527static struct rpc_task *
528rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
529 struct rpc_wait_queue *queue, struct rpc_task *task,
530 bool (*action)(struct rpc_task *, void *), void *data)
531{
532 if (RPC_IS_QUEUED(task)) {
533 smp_rmb();
534 if (task->tk_waitqueue == queue) {
535 if (action == NULL || action(task, data)) {
536 __rpc_do_wake_up_task_on_wq(wq, queue, task);
537 return task;
538 }
539 }
540 }
541 return NULL;
542}
543
544/*
545 * Wake up a queued task while the queue lock is being held
546 */
547static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
548 struct rpc_task *task)
549{
550 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
551 task, NULL, NULL);
552}
553
554/*
555 * Wake up a task on a specific queue
556 */
557void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
558{
559 if (!RPC_IS_QUEUED(task))
560 return;
561 spin_lock(&queue->lock);
562 rpc_wake_up_task_queue_locked(queue, task);
563 spin_unlock(&queue->lock);
564}
565EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
566
567static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
568{
569 task->tk_status = *(int *)status;
570 return true;
571}
572
573static void
574rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
575 struct rpc_task *task, int status)
576{
577 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
578 task, rpc_task_action_set_status, &status);
579}
580
581/**
582 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
583 * @queue: pointer to rpc_wait_queue
584 * @task: pointer to rpc_task
585 * @status: integer error value
586 *
587 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
588 * set to the value of @status.
589 */
590void
591rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
592 struct rpc_task *task, int status)
593{
594 if (!RPC_IS_QUEUED(task))
595 return;
596 spin_lock(&queue->lock);
597 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
598 spin_unlock(&queue->lock);
599}
600
601/*
602 * Wake up the next task on a priority queue.
603 */
604static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
605{
606 struct list_head *q;
607 struct rpc_task *task;
608
609 /*
610 * Service a batch of tasks from a single owner.
611 */
612 q = &queue->tasks[queue->priority];
613 if (!list_empty(q) && --queue->nr) {
614 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
615 goto out;
616 }
617
618 /*
619 * Service the next queue.
620 */
621 do {
622 if (q == &queue->tasks[0])
623 q = &queue->tasks[queue->maxpriority];
624 else
625 q = q - 1;
626 if (!list_empty(q)) {
627 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
628 goto new_queue;
629 }
630 } while (q != &queue->tasks[queue->priority]);
631
632 rpc_reset_waitqueue_priority(queue);
633 return NULL;
634
635new_queue:
636 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
637out:
638 return task;
639}
640
641static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
642{
643 if (RPC_IS_PRIORITY(queue))
644 return __rpc_find_next_queued_priority(queue);
645 if (!list_empty(&queue->tasks[0]))
646 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
647 return NULL;
648}
649
650/*
651 * Wake up the first task on the wait queue.
652 */
653struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
654 struct rpc_wait_queue *queue,
655 bool (*func)(struct rpc_task *, void *), void *data)
656{
657 struct rpc_task *task = NULL;
658
659 dprintk("RPC: wake_up_first(%p \"%s\")\n",
660 queue, rpc_qname(queue));
661 spin_lock(&queue->lock);
662 task = __rpc_find_next_queued(queue);
663 if (task != NULL)
664 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
665 task, func, data);
666 spin_unlock(&queue->lock);
667
668 return task;
669}
670
671/*
672 * Wake up the first task on the wait queue.
673 */
674struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
675 bool (*func)(struct rpc_task *, void *), void *data)
676{
677 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
678}
679EXPORT_SYMBOL_GPL(rpc_wake_up_first);
680
681static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
682{
683 return true;
684}
685
686/*
687 * Wake up the next task on the wait queue.
688*/
689struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
690{
691 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
692}
693EXPORT_SYMBOL_GPL(rpc_wake_up_next);
694
695/**
696 * rpc_wake_up - wake up all rpc_tasks
697 * @queue: rpc_wait_queue on which the tasks are sleeping
698 *
699 * Grabs queue->lock
700 */
701void rpc_wake_up(struct rpc_wait_queue *queue)
702{
703 struct list_head *head;
704
705 spin_lock(&queue->lock);
706 head = &queue->tasks[queue->maxpriority];
707 for (;;) {
708 while (!list_empty(head)) {
709 struct rpc_task *task;
710 task = list_first_entry(head,
711 struct rpc_task,
712 u.tk_wait.list);
713 rpc_wake_up_task_queue_locked(queue, task);
714 }
715 if (head == &queue->tasks[0])
716 break;
717 head--;
718 }
719 spin_unlock(&queue->lock);
720}
721EXPORT_SYMBOL_GPL(rpc_wake_up);
722
723/**
724 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
725 * @queue: rpc_wait_queue on which the tasks are sleeping
726 * @status: status value to set
727 *
728 * Grabs queue->lock
729 */
730void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
731{
732 struct list_head *head;
733
734 spin_lock(&queue->lock);
735 head = &queue->tasks[queue->maxpriority];
736 for (;;) {
737 while (!list_empty(head)) {
738 struct rpc_task *task;
739 task = list_first_entry(head,
740 struct rpc_task,
741 u.tk_wait.list);
742 task->tk_status = status;
743 rpc_wake_up_task_queue_locked(queue, task);
744 }
745 if (head == &queue->tasks[0])
746 break;
747 head--;
748 }
749 spin_unlock(&queue->lock);
750}
751EXPORT_SYMBOL_GPL(rpc_wake_up_status);
752
753static void __rpc_queue_timer_fn(struct work_struct *work)
754{
755 struct rpc_wait_queue *queue = container_of(work,
756 struct rpc_wait_queue,
757 timer_list.dwork.work);
758 struct rpc_task *task, *n;
759 unsigned long expires, now, timeo;
760
761 spin_lock(&queue->lock);
762 expires = now = jiffies;
763 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
764 timeo = task->tk_timeout;
765 if (time_after_eq(now, timeo)) {
766 dprintk("RPC: %5u timeout\n", task->tk_pid);
767 task->tk_status = -ETIMEDOUT;
768 rpc_wake_up_task_queue_locked(queue, task);
769 continue;
770 }
771 if (expires == now || time_after(expires, timeo))
772 expires = timeo;
773 }
774 if (!list_empty(&queue->timer_list.list))
775 rpc_set_queue_timer(queue, expires);
776 spin_unlock(&queue->lock);
777}
778
779static void __rpc_atrun(struct rpc_task *task)
780{
781 if (task->tk_status == -ETIMEDOUT)
782 task->tk_status = 0;
783}
784
785/*
786 * Run a task at a later time
787 */
788void rpc_delay(struct rpc_task *task, unsigned long delay)
789{
790 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
791}
792EXPORT_SYMBOL_GPL(rpc_delay);
793
794/*
795 * Helper to call task->tk_ops->rpc_call_prepare
796 */
797void rpc_prepare_task(struct rpc_task *task)
798{
799 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
800}
801
802static void
803rpc_init_task_statistics(struct rpc_task *task)
804{
805 /* Initialize retry counters */
806 task->tk_garb_retry = 2;
807 task->tk_cred_retry = 2;
808 task->tk_rebind_retry = 2;
809
810 /* starting timestamp */
811 task->tk_start = ktime_get();
812}
813
814static void
815rpc_reset_task_statistics(struct rpc_task *task)
816{
817 task->tk_timeouts = 0;
818 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
819 rpc_init_task_statistics(task);
820}
821
822/*
823 * Helper that calls task->tk_ops->rpc_call_done if it exists
824 */
825void rpc_exit_task(struct rpc_task *task)
826{
827 task->tk_action = NULL;
828 if (task->tk_ops->rpc_count_stats)
829 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
830 else if (task->tk_client)
831 rpc_count_iostats(task, task->tk_client->cl_metrics);
832 if (task->tk_ops->rpc_call_done != NULL) {
833 task->tk_ops->rpc_call_done(task, task->tk_calldata);
834 if (task->tk_action != NULL) {
835 /* Always release the RPC slot and buffer memory */
836 xprt_release(task);
837 rpc_reset_task_statistics(task);
838 }
839 }
840}
841
842void rpc_signal_task(struct rpc_task *task)
843{
844 struct rpc_wait_queue *queue;
845
846 if (!RPC_IS_ACTIVATED(task))
847 return;
848 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
849 smp_mb__after_atomic();
850 queue = READ_ONCE(task->tk_waitqueue);
851 if (queue)
852 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
853}
854
855void rpc_exit(struct rpc_task *task, int status)
856{
857 task->tk_status = status;
858 task->tk_action = rpc_exit_task;
859 rpc_wake_up_queued_task(task->tk_waitqueue, task);
860}
861EXPORT_SYMBOL_GPL(rpc_exit);
862
863void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
864{
865 if (ops->rpc_release != NULL)
866 ops->rpc_release(calldata);
867}
868
869/*
870 * This is the RPC `scheduler' (or rather, the finite state machine).
871 */
872static void __rpc_execute(struct rpc_task *task)
873{
874 struct rpc_wait_queue *queue;
875 int task_is_async = RPC_IS_ASYNC(task);
876 int status = 0;
877
878 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
879 task->tk_pid, task->tk_flags);
880
881 WARN_ON_ONCE(RPC_IS_QUEUED(task));
882 if (RPC_IS_QUEUED(task))
883 return;
884
885 for (;;) {
886 void (*do_action)(struct rpc_task *);
887
888 /*
889 * Perform the next FSM step or a pending callback.
890 *
891 * tk_action may be NULL if the task has been killed.
892 * In particular, note that rpc_killall_tasks may
893 * do this at any time, so beware when dereferencing.
894 */
895 do_action = task->tk_action;
896 if (task->tk_callback) {
897 do_action = task->tk_callback;
898 task->tk_callback = NULL;
899 }
900 if (!do_action)
901 break;
902 trace_rpc_task_run_action(task, do_action);
903 do_action(task);
904
905 /*
906 * Lockless check for whether task is sleeping or not.
907 */
908 if (!RPC_IS_QUEUED(task))
909 continue;
910
911 /*
912 * Signalled tasks should exit rather than sleep.
913 */
914 if (RPC_SIGNALLED(task)) {
915 task->tk_rpc_status = -ERESTARTSYS;
916 rpc_exit(task, -ERESTARTSYS);
917 }
918
919 /*
920 * The queue->lock protects against races with
921 * rpc_make_runnable().
922 *
923 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
924 * rpc_task, rpc_make_runnable() can assign it to a
925 * different workqueue. We therefore cannot assume that the
926 * rpc_task pointer may still be dereferenced.
927 */
928 queue = task->tk_waitqueue;
929 spin_lock(&queue->lock);
930 if (!RPC_IS_QUEUED(task)) {
931 spin_unlock(&queue->lock);
932 continue;
933 }
934 rpc_clear_running(task);
935 spin_unlock(&queue->lock);
936 if (task_is_async)
937 return;
938
939 /* sync task: sleep here */
940 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
941 status = out_of_line_wait_on_bit(&task->tk_runstate,
942 RPC_TASK_QUEUED, rpc_wait_bit_killable,
943 TASK_KILLABLE);
944 if (status < 0) {
945 /*
946 * When a sync task receives a signal, it exits with
947 * -ERESTARTSYS. In order to catch any callbacks that
948 * clean up after sleeping on some queue, we don't
949 * break the loop here, but go around once more.
950 */
951 dprintk("RPC: %5u got signal\n", task->tk_pid);
952 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
953 task->tk_rpc_status = -ERESTARTSYS;
954 rpc_exit(task, -ERESTARTSYS);
955 }
956 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
957 }
958
959 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
960 task->tk_status);
961 /* Release all resources associated with the task */
962 rpc_release_task(task);
963}
964
965/*
966 * User-visible entry point to the scheduler.
967 *
968 * This may be called recursively if e.g. an async NFS task updates
969 * the attributes and finds that dirty pages must be flushed.
970 * NOTE: Upon exit of this function the task is guaranteed to be
971 * released. In particular note that tk_release() will have
972 * been called, so your task memory may have been freed.
973 */
974void rpc_execute(struct rpc_task *task)
975{
976 bool is_async = RPC_IS_ASYNC(task);
977
978 rpc_set_active(task);
979 rpc_make_runnable(rpciod_workqueue, task);
980 if (!is_async)
981 __rpc_execute(task);
982}
983
984static void rpc_async_schedule(struct work_struct *work)
985{
986 unsigned int pflags = memalloc_nofs_save();
987
988 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
989 memalloc_nofs_restore(pflags);
990}
991
992/**
993 * rpc_malloc - allocate RPC buffer resources
994 * @task: RPC task
995 *
996 * A single memory region is allocated, which is split between the
997 * RPC call and RPC reply that this task is being used for. When
998 * this RPC is retired, the memory is released by calling rpc_free.
999 *
1000 * To prevent rpciod from hanging, this allocator never sleeps,
1001 * returning -ENOMEM and suppressing warning if the request cannot
1002 * be serviced immediately. The caller can arrange to sleep in a
1003 * way that is safe for rpciod.
1004 *
1005 * Most requests are 'small' (under 2KiB) and can be serviced from a
1006 * mempool, ensuring that NFS reads and writes can always proceed,
1007 * and that there is good locality of reference for these buffers.
1008 */
1009int rpc_malloc(struct rpc_task *task)
1010{
1011 struct rpc_rqst *rqst = task->tk_rqstp;
1012 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1013 struct rpc_buffer *buf;
1014 gfp_t gfp = GFP_NOFS;
1015
1016 if (RPC_IS_SWAPPER(task))
1017 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
1018
1019 size += sizeof(struct rpc_buffer);
1020 if (size <= RPC_BUFFER_MAXSIZE)
1021 buf = mempool_alloc(rpc_buffer_mempool, gfp);
1022 else
1023 buf = kmalloc(size, gfp);
1024
1025 if (!buf)
1026 return -ENOMEM;
1027
1028 buf->len = size;
1029 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1030 task->tk_pid, size, buf);
1031 rqst->rq_buffer = buf->data;
1032 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1033 return 0;
1034}
1035EXPORT_SYMBOL_GPL(rpc_malloc);
1036
1037/**
1038 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1039 * @task: RPC task
1040 *
1041 */
1042void rpc_free(struct rpc_task *task)
1043{
1044 void *buffer = task->tk_rqstp->rq_buffer;
1045 size_t size;
1046 struct rpc_buffer *buf;
1047
1048 buf = container_of(buffer, struct rpc_buffer, data);
1049 size = buf->len;
1050
1051 dprintk("RPC: freeing buffer of size %zu at %p\n",
1052 size, buf);
1053
1054 if (size <= RPC_BUFFER_MAXSIZE)
1055 mempool_free(buf, rpc_buffer_mempool);
1056 else
1057 kfree(buf);
1058}
1059EXPORT_SYMBOL_GPL(rpc_free);
1060
1061/*
1062 * Creation and deletion of RPC task structures
1063 */
1064static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1065{
1066 memset(task, 0, sizeof(*task));
1067 atomic_set(&task->tk_count, 1);
1068 task->tk_flags = task_setup_data->flags;
1069 task->tk_ops = task_setup_data->callback_ops;
1070 task->tk_calldata = task_setup_data->callback_data;
1071 INIT_LIST_HEAD(&task->tk_task);
1072
1073 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1074 task->tk_owner = current->tgid;
1075
1076 /* Initialize workqueue for async tasks */
1077 task->tk_workqueue = task_setup_data->workqueue;
1078
1079 task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1080 xprt_get(task_setup_data->rpc_xprt));
1081
1082 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1083
1084 if (task->tk_ops->rpc_call_prepare != NULL)
1085 task->tk_action = rpc_prepare_task;
1086
1087 rpc_init_task_statistics(task);
1088
1089 dprintk("RPC: new task initialized, procpid %u\n",
1090 task_pid_nr(current));
1091}
1092
1093static struct rpc_task *
1094rpc_alloc_task(void)
1095{
1096 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1097}
1098
1099/*
1100 * Create a new task for the specified client.
1101 */
1102struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1103{
1104 struct rpc_task *task = setup_data->task;
1105 unsigned short flags = 0;
1106
1107 if (task == NULL) {
1108 task = rpc_alloc_task();
1109 flags = RPC_TASK_DYNAMIC;
1110 }
1111
1112 rpc_init_task(task, setup_data);
1113 task->tk_flags |= flags;
1114 dprintk("RPC: allocated task %p\n", task);
1115 return task;
1116}
1117
1118/*
1119 * rpc_free_task - release rpc task and perform cleanups
1120 *
1121 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1122 * in order to work around a workqueue dependency issue.
1123 *
1124 * Tejun Heo states:
1125 * "Workqueue currently considers two work items to be the same if they're
1126 * on the same address and won't execute them concurrently - ie. it
1127 * makes a work item which is queued again while being executed wait
1128 * for the previous execution to complete.
1129 *
1130 * If a work function frees the work item, and then waits for an event
1131 * which should be performed by another work item and *that* work item
1132 * recycles the freed work item, it can create a false dependency loop.
1133 * There really is no reliable way to detect this short of verifying
1134 * every memory free."
1135 *
1136 */
1137static void rpc_free_task(struct rpc_task *task)
1138{
1139 unsigned short tk_flags = task->tk_flags;
1140
1141 put_rpccred(task->tk_op_cred);
1142 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1143
1144 if (tk_flags & RPC_TASK_DYNAMIC) {
1145 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1146 mempool_free(task, rpc_task_mempool);
1147 }
1148}
1149
1150static void rpc_async_release(struct work_struct *work)
1151{
1152 unsigned int pflags = memalloc_nofs_save();
1153
1154 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1155 memalloc_nofs_restore(pflags);
1156}
1157
1158static void rpc_release_resources_task(struct rpc_task *task)
1159{
1160 xprt_release(task);
1161 if (task->tk_msg.rpc_cred) {
1162 put_cred(task->tk_msg.rpc_cred);
1163 task->tk_msg.rpc_cred = NULL;
1164 }
1165 rpc_task_release_client(task);
1166}
1167
1168static void rpc_final_put_task(struct rpc_task *task,
1169 struct workqueue_struct *q)
1170{
1171 if (q != NULL) {
1172 INIT_WORK(&task->u.tk_work, rpc_async_release);
1173 queue_work(q, &task->u.tk_work);
1174 } else
1175 rpc_free_task(task);
1176}
1177
1178static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1179{
1180 if (atomic_dec_and_test(&task->tk_count)) {
1181 rpc_release_resources_task(task);
1182 rpc_final_put_task(task, q);
1183 }
1184}
1185
1186void rpc_put_task(struct rpc_task *task)
1187{
1188 rpc_do_put_task(task, NULL);
1189}
1190EXPORT_SYMBOL_GPL(rpc_put_task);
1191
1192void rpc_put_task_async(struct rpc_task *task)
1193{
1194 rpc_do_put_task(task, task->tk_workqueue);
1195}
1196EXPORT_SYMBOL_GPL(rpc_put_task_async);
1197
1198static void rpc_release_task(struct rpc_task *task)
1199{
1200 dprintk("RPC: %5u release task\n", task->tk_pid);
1201
1202 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1203
1204 rpc_release_resources_task(task);
1205
1206 /*
1207 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1208 * so it should be safe to use task->tk_count as a test for whether
1209 * or not any other processes still hold references to our rpc_task.
1210 */
1211 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1212 /* Wake up anyone who may be waiting for task completion */
1213 if (!rpc_complete_task(task))
1214 return;
1215 } else {
1216 if (!atomic_dec_and_test(&task->tk_count))
1217 return;
1218 }
1219 rpc_final_put_task(task, task->tk_workqueue);
1220}
1221
1222int rpciod_up(void)
1223{
1224 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1225}
1226
1227void rpciod_down(void)
1228{
1229 module_put(THIS_MODULE);
1230}
1231
1232/*
1233 * Start up the rpciod workqueue.
1234 */
1235static int rpciod_start(void)
1236{
1237 struct workqueue_struct *wq;
1238
1239 /*
1240 * Create the rpciod thread and wait for it to start.
1241 */
1242 dprintk("RPC: creating workqueue rpciod\n");
1243 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1244 if (!wq)
1245 goto out_failed;
1246 rpciod_workqueue = wq;
1247 /* Note: highpri because network receive is latency sensitive */
1248 wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1249 if (!wq)
1250 goto free_rpciod;
1251 xprtiod_workqueue = wq;
1252 return 1;
1253free_rpciod:
1254 wq = rpciod_workqueue;
1255 rpciod_workqueue = NULL;
1256 destroy_workqueue(wq);
1257out_failed:
1258 return 0;
1259}
1260
1261static void rpciod_stop(void)
1262{
1263 struct workqueue_struct *wq = NULL;
1264
1265 if (rpciod_workqueue == NULL)
1266 return;
1267 dprintk("RPC: destroying workqueue rpciod\n");
1268
1269 wq = rpciod_workqueue;
1270 rpciod_workqueue = NULL;
1271 destroy_workqueue(wq);
1272 wq = xprtiod_workqueue;
1273 xprtiod_workqueue = NULL;
1274 destroy_workqueue(wq);
1275}
1276
1277void
1278rpc_destroy_mempool(void)
1279{
1280 rpciod_stop();
1281 mempool_destroy(rpc_buffer_mempool);
1282 mempool_destroy(rpc_task_mempool);
1283 kmem_cache_destroy(rpc_task_slabp);
1284 kmem_cache_destroy(rpc_buffer_slabp);
1285 rpc_destroy_wait_queue(&delay_queue);
1286}
1287
1288int
1289rpc_init_mempool(void)
1290{
1291 /*
1292 * The following is not strictly a mempool initialisation,
1293 * but there is no harm in doing it here
1294 */
1295 rpc_init_wait_queue(&delay_queue, "delayq");
1296 if (!rpciod_start())
1297 goto err_nomem;
1298
1299 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1300 sizeof(struct rpc_task),
1301 0, SLAB_HWCACHE_ALIGN,
1302 NULL);
1303 if (!rpc_task_slabp)
1304 goto err_nomem;
1305 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1306 RPC_BUFFER_MAXSIZE,
1307 0, SLAB_HWCACHE_ALIGN,
1308 NULL);
1309 if (!rpc_buffer_slabp)
1310 goto err_nomem;
1311 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1312 rpc_task_slabp);
1313 if (!rpc_task_mempool)
1314 goto err_nomem;
1315 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1316 rpc_buffer_slabp);
1317 if (!rpc_buffer_mempool)
1318 goto err_nomem;
1319 return 0;
1320err_nomem:
1321 rpc_destroy_mempool();
1322 return -ENOMEM;
1323}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/sched.c
4 *
5 * Scheduling for synchronous and asynchronous RPC requests.
6 *
7 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
8 *
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 */
12
13#include <linux/module.h>
14
15#include <linux/sched.h>
16#include <linux/interrupt.h>
17#include <linux/slab.h>
18#include <linux/mempool.h>
19#include <linux/smp.h>
20#include <linux/spinlock.h>
21#include <linux/mutex.h>
22#include <linux/freezer.h>
23#include <linux/sched/mm.h>
24
25#include <linux/sunrpc/clnt.h>
26#include <linux/sunrpc/metrics.h>
27
28#include "sunrpc.h"
29
30#define CREATE_TRACE_POINTS
31#include <trace/events/sunrpc.h>
32
33/*
34 * RPC slabs and memory pools
35 */
36#define RPC_BUFFER_MAXSIZE (2048)
37#define RPC_BUFFER_POOLSIZE (8)
38#define RPC_TASK_POOLSIZE (8)
39static struct kmem_cache *rpc_task_slabp __read_mostly;
40static struct kmem_cache *rpc_buffer_slabp __read_mostly;
41static mempool_t *rpc_task_mempool __read_mostly;
42static mempool_t *rpc_buffer_mempool __read_mostly;
43
44static void rpc_async_schedule(struct work_struct *);
45static void rpc_release_task(struct rpc_task *task);
46static void __rpc_queue_timer_fn(struct work_struct *);
47
48/*
49 * RPC tasks sit here while waiting for conditions to improve.
50 */
51static struct rpc_wait_queue delay_queue;
52
53/*
54 * rpciod-related stuff
55 */
56struct workqueue_struct *rpciod_workqueue __read_mostly;
57struct workqueue_struct *xprtiod_workqueue __read_mostly;
58EXPORT_SYMBOL_GPL(xprtiod_workqueue);
59
60gfp_t rpc_task_gfp_mask(void)
61{
62 if (current->flags & PF_WQ_WORKER)
63 return GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
64 return GFP_KERNEL;
65}
66EXPORT_SYMBOL_GPL(rpc_task_gfp_mask);
67
68bool rpc_task_set_rpc_status(struct rpc_task *task, int rpc_status)
69{
70 if (cmpxchg(&task->tk_rpc_status, 0, rpc_status) == 0)
71 return true;
72 return false;
73}
74
75unsigned long
76rpc_task_timeout(const struct rpc_task *task)
77{
78 unsigned long timeout = READ_ONCE(task->tk_timeout);
79
80 if (timeout != 0) {
81 unsigned long now = jiffies;
82 if (time_before(now, timeout))
83 return timeout - now;
84 }
85 return 0;
86}
87EXPORT_SYMBOL_GPL(rpc_task_timeout);
88
89/*
90 * Disable the timer for a given RPC task. Should be called with
91 * queue->lock and bh_disabled in order to avoid races within
92 * rpc_run_timer().
93 */
94static void
95__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
96{
97 if (list_empty(&task->u.tk_wait.timer_list))
98 return;
99 task->tk_timeout = 0;
100 list_del(&task->u.tk_wait.timer_list);
101 if (list_empty(&queue->timer_list.list))
102 cancel_delayed_work(&queue->timer_list.dwork);
103}
104
105static void
106rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
107{
108 unsigned long now = jiffies;
109 queue->timer_list.expires = expires;
110 if (time_before_eq(expires, now))
111 expires = 0;
112 else
113 expires -= now;
114 mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
115}
116
117/*
118 * Set up a timer for the current task.
119 */
120static void
121__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
122 unsigned long timeout)
123{
124 task->tk_timeout = timeout;
125 if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
126 rpc_set_queue_timer(queue, timeout);
127 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
128}
129
130static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
131{
132 if (queue->priority != priority) {
133 queue->priority = priority;
134 queue->nr = 1U << priority;
135 }
136}
137
138static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
139{
140 rpc_set_waitqueue_priority(queue, queue->maxpriority);
141}
142
143/*
144 * Add a request to a queue list
145 */
146static void
147__rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
148{
149 struct rpc_task *t;
150
151 list_for_each_entry(t, q, u.tk_wait.list) {
152 if (t->tk_owner == task->tk_owner) {
153 list_add_tail(&task->u.tk_wait.links,
154 &t->u.tk_wait.links);
155 /* Cache the queue head in task->u.tk_wait.list */
156 task->u.tk_wait.list.next = q;
157 task->u.tk_wait.list.prev = NULL;
158 return;
159 }
160 }
161 INIT_LIST_HEAD(&task->u.tk_wait.links);
162 list_add_tail(&task->u.tk_wait.list, q);
163}
164
165/*
166 * Remove request from a queue list
167 */
168static void
169__rpc_list_dequeue_task(struct rpc_task *task)
170{
171 struct list_head *q;
172 struct rpc_task *t;
173
174 if (task->u.tk_wait.list.prev == NULL) {
175 list_del(&task->u.tk_wait.links);
176 return;
177 }
178 if (!list_empty(&task->u.tk_wait.links)) {
179 t = list_first_entry(&task->u.tk_wait.links,
180 struct rpc_task,
181 u.tk_wait.links);
182 /* Assume __rpc_list_enqueue_task() cached the queue head */
183 q = t->u.tk_wait.list.next;
184 list_add_tail(&t->u.tk_wait.list, q);
185 list_del(&task->u.tk_wait.links);
186 }
187 list_del(&task->u.tk_wait.list);
188}
189
190/*
191 * Add new request to a priority queue.
192 */
193static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
194 struct rpc_task *task,
195 unsigned char queue_priority)
196{
197 if (unlikely(queue_priority > queue->maxpriority))
198 queue_priority = queue->maxpriority;
199 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
200}
201
202/*
203 * Add new request to wait queue.
204 */
205static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
206 struct rpc_task *task,
207 unsigned char queue_priority)
208{
209 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
210 if (RPC_IS_PRIORITY(queue))
211 __rpc_add_wait_queue_priority(queue, task, queue_priority);
212 else
213 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
214 task->tk_waitqueue = queue;
215 queue->qlen++;
216 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
217 smp_wmb();
218 rpc_set_queued(task);
219}
220
221/*
222 * Remove request from a priority queue.
223 */
224static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
225{
226 __rpc_list_dequeue_task(task);
227}
228
229/*
230 * Remove request from queue.
231 * Note: must be called with spin lock held.
232 */
233static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
234{
235 __rpc_disable_timer(queue, task);
236 if (RPC_IS_PRIORITY(queue))
237 __rpc_remove_wait_queue_priority(task);
238 else
239 list_del(&task->u.tk_wait.list);
240 queue->qlen--;
241}
242
243static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
244{
245 int i;
246
247 spin_lock_init(&queue->lock);
248 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
249 INIT_LIST_HEAD(&queue->tasks[i]);
250 queue->maxpriority = nr_queues - 1;
251 rpc_reset_waitqueue_priority(queue);
252 queue->qlen = 0;
253 queue->timer_list.expires = 0;
254 INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
255 INIT_LIST_HEAD(&queue->timer_list.list);
256 rpc_assign_waitqueue_name(queue, qname);
257}
258
259void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
260{
261 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
262}
263EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
264
265void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
266{
267 __rpc_init_priority_wait_queue(queue, qname, 1);
268}
269EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
270
271void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
272{
273 cancel_delayed_work_sync(&queue->timer_list.dwork);
274}
275EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
276
277static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
278{
279 schedule();
280 if (signal_pending_state(mode, current))
281 return -ERESTARTSYS;
282 return 0;
283}
284
285#if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
286static void rpc_task_set_debuginfo(struct rpc_task *task)
287{
288 struct rpc_clnt *clnt = task->tk_client;
289
290 /* Might be a task carrying a reverse-direction operation */
291 if (!clnt) {
292 static atomic_t rpc_pid;
293
294 task->tk_pid = atomic_inc_return(&rpc_pid);
295 return;
296 }
297
298 task->tk_pid = atomic_inc_return(&clnt->cl_pid);
299}
300#else
301static inline void rpc_task_set_debuginfo(struct rpc_task *task)
302{
303}
304#endif
305
306static void rpc_set_active(struct rpc_task *task)
307{
308 rpc_task_set_debuginfo(task);
309 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
310 trace_rpc_task_begin(task, NULL);
311}
312
313/*
314 * Mark an RPC call as having completed by clearing the 'active' bit
315 * and then waking up all tasks that were sleeping.
316 */
317static int rpc_complete_task(struct rpc_task *task)
318{
319 void *m = &task->tk_runstate;
320 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
321 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
322 unsigned long flags;
323 int ret;
324
325 trace_rpc_task_complete(task, NULL);
326
327 spin_lock_irqsave(&wq->lock, flags);
328 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
329 ret = atomic_dec_and_test(&task->tk_count);
330 if (waitqueue_active(wq))
331 __wake_up_locked_key(wq, TASK_NORMAL, &k);
332 spin_unlock_irqrestore(&wq->lock, flags);
333 return ret;
334}
335
336/*
337 * Allow callers to wait for completion of an RPC call
338 *
339 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
340 * to enforce taking of the wq->lock and hence avoid races with
341 * rpc_complete_task().
342 */
343int rpc_wait_for_completion_task(struct rpc_task *task)
344{
345 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
346 rpc_wait_bit_killable, TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
347}
348EXPORT_SYMBOL_GPL(rpc_wait_for_completion_task);
349
350/*
351 * Make an RPC task runnable.
352 *
353 * Note: If the task is ASYNC, and is being made runnable after sitting on an
354 * rpc_wait_queue, this must be called with the queue spinlock held to protect
355 * the wait queue operation.
356 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
357 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
358 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
359 * the RPC_TASK_RUNNING flag.
360 */
361static void rpc_make_runnable(struct workqueue_struct *wq,
362 struct rpc_task *task)
363{
364 bool need_wakeup = !rpc_test_and_set_running(task);
365
366 rpc_clear_queued(task);
367 if (!need_wakeup)
368 return;
369 if (RPC_IS_ASYNC(task)) {
370 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
371 queue_work(wq, &task->u.tk_work);
372 } else {
373 smp_mb__after_atomic();
374 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
375 }
376}
377
378/*
379 * Prepare for sleeping on a wait queue.
380 * By always appending tasks to the list we ensure FIFO behavior.
381 * NB: An RPC task will only receive interrupt-driven events as long
382 * as it's on a wait queue.
383 */
384static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
385 struct rpc_task *task,
386 unsigned char queue_priority)
387{
388 trace_rpc_task_sleep(task, q);
389
390 __rpc_add_wait_queue(q, task, queue_priority);
391}
392
393static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
394 struct rpc_task *task,
395 unsigned char queue_priority)
396{
397 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
398 return;
399 __rpc_do_sleep_on_priority(q, task, queue_priority);
400}
401
402static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
403 struct rpc_task *task, unsigned long timeout,
404 unsigned char queue_priority)
405{
406 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
407 return;
408 if (time_is_after_jiffies(timeout)) {
409 __rpc_do_sleep_on_priority(q, task, queue_priority);
410 __rpc_add_timer(q, task, timeout);
411 } else
412 task->tk_status = -ETIMEDOUT;
413}
414
415static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
416{
417 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
418 task->tk_callback = action;
419}
420
421static bool rpc_sleep_check_activated(struct rpc_task *task)
422{
423 /* We shouldn't ever put an inactive task to sleep */
424 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
425 task->tk_status = -EIO;
426 rpc_put_task_async(task);
427 return false;
428 }
429 return true;
430}
431
432void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
433 rpc_action action, unsigned long timeout)
434{
435 if (!rpc_sleep_check_activated(task))
436 return;
437
438 rpc_set_tk_callback(task, action);
439
440 /*
441 * Protect the queue operations.
442 */
443 spin_lock(&q->lock);
444 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
445 spin_unlock(&q->lock);
446}
447EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
448
449void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
450 rpc_action action)
451{
452 if (!rpc_sleep_check_activated(task))
453 return;
454
455 rpc_set_tk_callback(task, action);
456
457 WARN_ON_ONCE(task->tk_timeout != 0);
458 /*
459 * Protect the queue operations.
460 */
461 spin_lock(&q->lock);
462 __rpc_sleep_on_priority(q, task, task->tk_priority);
463 spin_unlock(&q->lock);
464}
465EXPORT_SYMBOL_GPL(rpc_sleep_on);
466
467void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
468 struct rpc_task *task, unsigned long timeout, int priority)
469{
470 if (!rpc_sleep_check_activated(task))
471 return;
472
473 priority -= RPC_PRIORITY_LOW;
474 /*
475 * Protect the queue operations.
476 */
477 spin_lock(&q->lock);
478 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
479 spin_unlock(&q->lock);
480}
481EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
482
483void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
484 int priority)
485{
486 if (!rpc_sleep_check_activated(task))
487 return;
488
489 WARN_ON_ONCE(task->tk_timeout != 0);
490 priority -= RPC_PRIORITY_LOW;
491 /*
492 * Protect the queue operations.
493 */
494 spin_lock(&q->lock);
495 __rpc_sleep_on_priority(q, task, priority);
496 spin_unlock(&q->lock);
497}
498EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
499
500/**
501 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
502 * @wq: workqueue on which to run task
503 * @queue: wait queue
504 * @task: task to be woken up
505 *
506 * Caller must hold queue->lock, and have cleared the task queued flag.
507 */
508static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
509 struct rpc_wait_queue *queue,
510 struct rpc_task *task)
511{
512 /* Has the task been executed yet? If not, we cannot wake it up! */
513 if (!RPC_IS_ACTIVATED(task)) {
514 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
515 return;
516 }
517
518 trace_rpc_task_wakeup(task, queue);
519
520 __rpc_remove_wait_queue(queue, task);
521
522 rpc_make_runnable(wq, task);
523}
524
525/*
526 * Wake up a queued task while the queue lock is being held
527 */
528static struct rpc_task *
529rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
530 struct rpc_wait_queue *queue, struct rpc_task *task,
531 bool (*action)(struct rpc_task *, void *), void *data)
532{
533 if (RPC_IS_QUEUED(task)) {
534 smp_rmb();
535 if (task->tk_waitqueue == queue) {
536 if (action == NULL || action(task, data)) {
537 __rpc_do_wake_up_task_on_wq(wq, queue, task);
538 return task;
539 }
540 }
541 }
542 return NULL;
543}
544
545/*
546 * Wake up a queued task while the queue lock is being held
547 */
548static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
549 struct rpc_task *task)
550{
551 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
552 task, NULL, NULL);
553}
554
555/*
556 * Wake up a task on a specific queue
557 */
558void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
559{
560 if (!RPC_IS_QUEUED(task))
561 return;
562 spin_lock(&queue->lock);
563 rpc_wake_up_task_queue_locked(queue, task);
564 spin_unlock(&queue->lock);
565}
566EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
567
568static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
569{
570 task->tk_status = *(int *)status;
571 return true;
572}
573
574static void
575rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
576 struct rpc_task *task, int status)
577{
578 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
579 task, rpc_task_action_set_status, &status);
580}
581
582/**
583 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
584 * @queue: pointer to rpc_wait_queue
585 * @task: pointer to rpc_task
586 * @status: integer error value
587 *
588 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
589 * set to the value of @status.
590 */
591void
592rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
593 struct rpc_task *task, int status)
594{
595 if (!RPC_IS_QUEUED(task))
596 return;
597 spin_lock(&queue->lock);
598 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
599 spin_unlock(&queue->lock);
600}
601
602/*
603 * Wake up the next task on a priority queue.
604 */
605static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
606{
607 struct list_head *q;
608 struct rpc_task *task;
609
610 /*
611 * Service the privileged queue.
612 */
613 q = &queue->tasks[RPC_NR_PRIORITY - 1];
614 if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
615 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
616 goto out;
617 }
618
619 /*
620 * Service a batch of tasks from a single owner.
621 */
622 q = &queue->tasks[queue->priority];
623 if (!list_empty(q) && queue->nr) {
624 queue->nr--;
625 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
626 goto out;
627 }
628
629 /*
630 * Service the next queue.
631 */
632 do {
633 if (q == &queue->tasks[0])
634 q = &queue->tasks[queue->maxpriority];
635 else
636 q = q - 1;
637 if (!list_empty(q)) {
638 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
639 goto new_queue;
640 }
641 } while (q != &queue->tasks[queue->priority]);
642
643 rpc_reset_waitqueue_priority(queue);
644 return NULL;
645
646new_queue:
647 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
648out:
649 return task;
650}
651
652static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
653{
654 if (RPC_IS_PRIORITY(queue))
655 return __rpc_find_next_queued_priority(queue);
656 if (!list_empty(&queue->tasks[0]))
657 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
658 return NULL;
659}
660
661/*
662 * Wake up the first task on the wait queue.
663 */
664struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
665 struct rpc_wait_queue *queue,
666 bool (*func)(struct rpc_task *, void *), void *data)
667{
668 struct rpc_task *task = NULL;
669
670 spin_lock(&queue->lock);
671 task = __rpc_find_next_queued(queue);
672 if (task != NULL)
673 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
674 task, func, data);
675 spin_unlock(&queue->lock);
676
677 return task;
678}
679
680/*
681 * Wake up the first task on the wait queue.
682 */
683struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
684 bool (*func)(struct rpc_task *, void *), void *data)
685{
686 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
687}
688EXPORT_SYMBOL_GPL(rpc_wake_up_first);
689
690static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
691{
692 return true;
693}
694
695/*
696 * Wake up the next task on the wait queue.
697*/
698struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
699{
700 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
701}
702EXPORT_SYMBOL_GPL(rpc_wake_up_next);
703
704/**
705 * rpc_wake_up_locked - wake up all rpc_tasks
706 * @queue: rpc_wait_queue on which the tasks are sleeping
707 *
708 */
709static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
710{
711 struct rpc_task *task;
712
713 for (;;) {
714 task = __rpc_find_next_queued(queue);
715 if (task == NULL)
716 break;
717 rpc_wake_up_task_queue_locked(queue, task);
718 }
719}
720
721/**
722 * rpc_wake_up - wake up all rpc_tasks
723 * @queue: rpc_wait_queue on which the tasks are sleeping
724 *
725 * Grabs queue->lock
726 */
727void rpc_wake_up(struct rpc_wait_queue *queue)
728{
729 spin_lock(&queue->lock);
730 rpc_wake_up_locked(queue);
731 spin_unlock(&queue->lock);
732}
733EXPORT_SYMBOL_GPL(rpc_wake_up);
734
735/**
736 * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
737 * @queue: rpc_wait_queue on which the tasks are sleeping
738 * @status: status value to set
739 */
740static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
741{
742 struct rpc_task *task;
743
744 for (;;) {
745 task = __rpc_find_next_queued(queue);
746 if (task == NULL)
747 break;
748 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
749 }
750}
751
752/**
753 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
754 * @queue: rpc_wait_queue on which the tasks are sleeping
755 * @status: status value to set
756 *
757 * Grabs queue->lock
758 */
759void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
760{
761 spin_lock(&queue->lock);
762 rpc_wake_up_status_locked(queue, status);
763 spin_unlock(&queue->lock);
764}
765EXPORT_SYMBOL_GPL(rpc_wake_up_status);
766
767static void __rpc_queue_timer_fn(struct work_struct *work)
768{
769 struct rpc_wait_queue *queue = container_of(work,
770 struct rpc_wait_queue,
771 timer_list.dwork.work);
772 struct rpc_task *task, *n;
773 unsigned long expires, now, timeo;
774
775 spin_lock(&queue->lock);
776 expires = now = jiffies;
777 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
778 timeo = task->tk_timeout;
779 if (time_after_eq(now, timeo)) {
780 trace_rpc_task_timeout(task, task->tk_action);
781 task->tk_status = -ETIMEDOUT;
782 rpc_wake_up_task_queue_locked(queue, task);
783 continue;
784 }
785 if (expires == now || time_after(expires, timeo))
786 expires = timeo;
787 }
788 if (!list_empty(&queue->timer_list.list))
789 rpc_set_queue_timer(queue, expires);
790 spin_unlock(&queue->lock);
791}
792
793static void __rpc_atrun(struct rpc_task *task)
794{
795 if (task->tk_status == -ETIMEDOUT)
796 task->tk_status = 0;
797}
798
799/*
800 * Run a task at a later time
801 */
802void rpc_delay(struct rpc_task *task, unsigned long delay)
803{
804 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
805}
806EXPORT_SYMBOL_GPL(rpc_delay);
807
808/*
809 * Helper to call task->tk_ops->rpc_call_prepare
810 */
811void rpc_prepare_task(struct rpc_task *task)
812{
813 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
814}
815
816static void
817rpc_init_task_statistics(struct rpc_task *task)
818{
819 /* Initialize retry counters */
820 task->tk_garb_retry = 2;
821 task->tk_cred_retry = 2;
822
823 /* starting timestamp */
824 task->tk_start = ktime_get();
825}
826
827static void
828rpc_reset_task_statistics(struct rpc_task *task)
829{
830 task->tk_timeouts = 0;
831 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
832 rpc_init_task_statistics(task);
833}
834
835/*
836 * Helper that calls task->tk_ops->rpc_call_done if it exists
837 */
838void rpc_exit_task(struct rpc_task *task)
839{
840 trace_rpc_task_end(task, task->tk_action);
841 task->tk_action = NULL;
842 if (task->tk_ops->rpc_count_stats)
843 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
844 else if (task->tk_client)
845 rpc_count_iostats(task, task->tk_client->cl_metrics);
846 if (task->tk_ops->rpc_call_done != NULL) {
847 trace_rpc_task_call_done(task, task->tk_ops->rpc_call_done);
848 task->tk_ops->rpc_call_done(task, task->tk_calldata);
849 if (task->tk_action != NULL) {
850 /* Always release the RPC slot and buffer memory */
851 xprt_release(task);
852 rpc_reset_task_statistics(task);
853 }
854 }
855}
856
857void rpc_signal_task(struct rpc_task *task)
858{
859 struct rpc_wait_queue *queue;
860
861 if (!RPC_IS_ACTIVATED(task))
862 return;
863
864 if (!rpc_task_set_rpc_status(task, -ERESTARTSYS))
865 return;
866 trace_rpc_task_signalled(task, task->tk_action);
867 queue = READ_ONCE(task->tk_waitqueue);
868 if (queue)
869 rpc_wake_up_queued_task(queue, task);
870}
871
872void rpc_task_try_cancel(struct rpc_task *task, int error)
873{
874 struct rpc_wait_queue *queue;
875
876 if (!rpc_task_set_rpc_status(task, error))
877 return;
878 queue = READ_ONCE(task->tk_waitqueue);
879 if (queue)
880 rpc_wake_up_queued_task(queue, task);
881}
882
883void rpc_exit(struct rpc_task *task, int status)
884{
885 task->tk_status = status;
886 task->tk_action = rpc_exit_task;
887 rpc_wake_up_queued_task(task->tk_waitqueue, task);
888}
889EXPORT_SYMBOL_GPL(rpc_exit);
890
891void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
892{
893 if (ops->rpc_release != NULL)
894 ops->rpc_release(calldata);
895}
896
897static bool xprt_needs_memalloc(struct rpc_xprt *xprt, struct rpc_task *tk)
898{
899 if (!xprt)
900 return false;
901 if (!atomic_read(&xprt->swapper))
902 return false;
903 return test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == tk;
904}
905
906/*
907 * This is the RPC `scheduler' (or rather, the finite state machine).
908 */
909static void __rpc_execute(struct rpc_task *task)
910{
911 struct rpc_wait_queue *queue;
912 int task_is_async = RPC_IS_ASYNC(task);
913 int status = 0;
914 unsigned long pflags = current->flags;
915
916 WARN_ON_ONCE(RPC_IS_QUEUED(task));
917 if (RPC_IS_QUEUED(task))
918 return;
919
920 for (;;) {
921 void (*do_action)(struct rpc_task *);
922
923 /*
924 * Perform the next FSM step or a pending callback.
925 *
926 * tk_action may be NULL if the task has been killed.
927 */
928 do_action = task->tk_action;
929 /* Tasks with an RPC error status should exit */
930 if (do_action && do_action != rpc_exit_task &&
931 (status = READ_ONCE(task->tk_rpc_status)) != 0) {
932 task->tk_status = status;
933 do_action = rpc_exit_task;
934 }
935 /* Callbacks override all actions */
936 if (task->tk_callback) {
937 do_action = task->tk_callback;
938 task->tk_callback = NULL;
939 }
940 if (!do_action)
941 break;
942 if (RPC_IS_SWAPPER(task) ||
943 xprt_needs_memalloc(task->tk_xprt, task))
944 current->flags |= PF_MEMALLOC;
945
946 trace_rpc_task_run_action(task, do_action);
947 do_action(task);
948
949 /*
950 * Lockless check for whether task is sleeping or not.
951 */
952 if (!RPC_IS_QUEUED(task)) {
953 cond_resched();
954 continue;
955 }
956
957 /*
958 * The queue->lock protects against races with
959 * rpc_make_runnable().
960 *
961 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
962 * rpc_task, rpc_make_runnable() can assign it to a
963 * different workqueue. We therefore cannot assume that the
964 * rpc_task pointer may still be dereferenced.
965 */
966 queue = task->tk_waitqueue;
967 spin_lock(&queue->lock);
968 if (!RPC_IS_QUEUED(task)) {
969 spin_unlock(&queue->lock);
970 continue;
971 }
972 /* Wake up any task that has an exit status */
973 if (READ_ONCE(task->tk_rpc_status) != 0) {
974 rpc_wake_up_task_queue_locked(queue, task);
975 spin_unlock(&queue->lock);
976 continue;
977 }
978 rpc_clear_running(task);
979 spin_unlock(&queue->lock);
980 if (task_is_async)
981 goto out;
982
983 /* sync task: sleep here */
984 trace_rpc_task_sync_sleep(task, task->tk_action);
985 status = out_of_line_wait_on_bit(&task->tk_runstate,
986 RPC_TASK_QUEUED, rpc_wait_bit_killable,
987 TASK_KILLABLE|TASK_FREEZABLE);
988 if (status < 0) {
989 /*
990 * When a sync task receives a signal, it exits with
991 * -ERESTARTSYS. In order to catch any callbacks that
992 * clean up after sleeping on some queue, we don't
993 * break the loop here, but go around once more.
994 */
995 rpc_signal_task(task);
996 }
997 trace_rpc_task_sync_wake(task, task->tk_action);
998 }
999
1000 /* Release all resources associated with the task */
1001 rpc_release_task(task);
1002out:
1003 current_restore_flags(pflags, PF_MEMALLOC);
1004}
1005
1006/*
1007 * User-visible entry point to the scheduler.
1008 *
1009 * This may be called recursively if e.g. an async NFS task updates
1010 * the attributes and finds that dirty pages must be flushed.
1011 * NOTE: Upon exit of this function the task is guaranteed to be
1012 * released. In particular note that tk_release() will have
1013 * been called, so your task memory may have been freed.
1014 */
1015void rpc_execute(struct rpc_task *task)
1016{
1017 bool is_async = RPC_IS_ASYNC(task);
1018
1019 rpc_set_active(task);
1020 rpc_make_runnable(rpciod_workqueue, task);
1021 if (!is_async) {
1022 unsigned int pflags = memalloc_nofs_save();
1023 __rpc_execute(task);
1024 memalloc_nofs_restore(pflags);
1025 }
1026}
1027
1028static void rpc_async_schedule(struct work_struct *work)
1029{
1030 unsigned int pflags = memalloc_nofs_save();
1031
1032 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
1033 memalloc_nofs_restore(pflags);
1034}
1035
1036/**
1037 * rpc_malloc - allocate RPC buffer resources
1038 * @task: RPC task
1039 *
1040 * A single memory region is allocated, which is split between the
1041 * RPC call and RPC reply that this task is being used for. When
1042 * this RPC is retired, the memory is released by calling rpc_free.
1043 *
1044 * To prevent rpciod from hanging, this allocator never sleeps,
1045 * returning -ENOMEM and suppressing warning if the request cannot
1046 * be serviced immediately. The caller can arrange to sleep in a
1047 * way that is safe for rpciod.
1048 *
1049 * Most requests are 'small' (under 2KiB) and can be serviced from a
1050 * mempool, ensuring that NFS reads and writes can always proceed,
1051 * and that there is good locality of reference for these buffers.
1052 */
1053int rpc_malloc(struct rpc_task *task)
1054{
1055 struct rpc_rqst *rqst = task->tk_rqstp;
1056 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1057 struct rpc_buffer *buf;
1058 gfp_t gfp = rpc_task_gfp_mask();
1059
1060 size += sizeof(struct rpc_buffer);
1061 if (size <= RPC_BUFFER_MAXSIZE) {
1062 buf = kmem_cache_alloc(rpc_buffer_slabp, gfp);
1063 /* Reach for the mempool if dynamic allocation fails */
1064 if (!buf && RPC_IS_ASYNC(task))
1065 buf = mempool_alloc(rpc_buffer_mempool, GFP_NOWAIT);
1066 } else
1067 buf = kmalloc(size, gfp);
1068
1069 if (!buf)
1070 return -ENOMEM;
1071
1072 buf->len = size;
1073 rqst->rq_buffer = buf->data;
1074 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1075 return 0;
1076}
1077EXPORT_SYMBOL_GPL(rpc_malloc);
1078
1079/**
1080 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1081 * @task: RPC task
1082 *
1083 */
1084void rpc_free(struct rpc_task *task)
1085{
1086 void *buffer = task->tk_rqstp->rq_buffer;
1087 size_t size;
1088 struct rpc_buffer *buf;
1089
1090 buf = container_of(buffer, struct rpc_buffer, data);
1091 size = buf->len;
1092
1093 if (size <= RPC_BUFFER_MAXSIZE)
1094 mempool_free(buf, rpc_buffer_mempool);
1095 else
1096 kfree(buf);
1097}
1098EXPORT_SYMBOL_GPL(rpc_free);
1099
1100/*
1101 * Creation and deletion of RPC task structures
1102 */
1103static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1104{
1105 memset(task, 0, sizeof(*task));
1106 atomic_set(&task->tk_count, 1);
1107 task->tk_flags = task_setup_data->flags;
1108 task->tk_ops = task_setup_data->callback_ops;
1109 task->tk_calldata = task_setup_data->callback_data;
1110 INIT_LIST_HEAD(&task->tk_task);
1111
1112 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1113 task->tk_owner = current->tgid;
1114
1115 /* Initialize workqueue for async tasks */
1116 task->tk_workqueue = task_setup_data->workqueue;
1117
1118 task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1119 xprt_get(task_setup_data->rpc_xprt));
1120
1121 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1122
1123 if (task->tk_ops->rpc_call_prepare != NULL)
1124 task->tk_action = rpc_prepare_task;
1125
1126 rpc_init_task_statistics(task);
1127}
1128
1129static struct rpc_task *rpc_alloc_task(void)
1130{
1131 struct rpc_task *task;
1132
1133 task = kmem_cache_alloc(rpc_task_slabp, rpc_task_gfp_mask());
1134 if (task)
1135 return task;
1136 return mempool_alloc(rpc_task_mempool, GFP_NOWAIT);
1137}
1138
1139/*
1140 * Create a new task for the specified client.
1141 */
1142struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1143{
1144 struct rpc_task *task = setup_data->task;
1145 unsigned short flags = 0;
1146
1147 if (task == NULL) {
1148 task = rpc_alloc_task();
1149 if (task == NULL) {
1150 rpc_release_calldata(setup_data->callback_ops,
1151 setup_data->callback_data);
1152 return ERR_PTR(-ENOMEM);
1153 }
1154 flags = RPC_TASK_DYNAMIC;
1155 }
1156
1157 rpc_init_task(task, setup_data);
1158 task->tk_flags |= flags;
1159 return task;
1160}
1161
1162/*
1163 * rpc_free_task - release rpc task and perform cleanups
1164 *
1165 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1166 * in order to work around a workqueue dependency issue.
1167 *
1168 * Tejun Heo states:
1169 * "Workqueue currently considers two work items to be the same if they're
1170 * on the same address and won't execute them concurrently - ie. it
1171 * makes a work item which is queued again while being executed wait
1172 * for the previous execution to complete.
1173 *
1174 * If a work function frees the work item, and then waits for an event
1175 * which should be performed by another work item and *that* work item
1176 * recycles the freed work item, it can create a false dependency loop.
1177 * There really is no reliable way to detect this short of verifying
1178 * every memory free."
1179 *
1180 */
1181static void rpc_free_task(struct rpc_task *task)
1182{
1183 unsigned short tk_flags = task->tk_flags;
1184
1185 put_rpccred(task->tk_op_cred);
1186 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1187
1188 if (tk_flags & RPC_TASK_DYNAMIC)
1189 mempool_free(task, rpc_task_mempool);
1190}
1191
1192static void rpc_async_release(struct work_struct *work)
1193{
1194 unsigned int pflags = memalloc_nofs_save();
1195
1196 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1197 memalloc_nofs_restore(pflags);
1198}
1199
1200static void rpc_release_resources_task(struct rpc_task *task)
1201{
1202 xprt_release(task);
1203 if (task->tk_msg.rpc_cred) {
1204 if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1205 put_cred(task->tk_msg.rpc_cred);
1206 task->tk_msg.rpc_cred = NULL;
1207 }
1208 rpc_task_release_client(task);
1209}
1210
1211static void rpc_final_put_task(struct rpc_task *task,
1212 struct workqueue_struct *q)
1213{
1214 if (q != NULL) {
1215 INIT_WORK(&task->u.tk_work, rpc_async_release);
1216 queue_work(q, &task->u.tk_work);
1217 } else
1218 rpc_free_task(task);
1219}
1220
1221static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1222{
1223 if (atomic_dec_and_test(&task->tk_count)) {
1224 rpc_release_resources_task(task);
1225 rpc_final_put_task(task, q);
1226 }
1227}
1228
1229void rpc_put_task(struct rpc_task *task)
1230{
1231 rpc_do_put_task(task, NULL);
1232}
1233EXPORT_SYMBOL_GPL(rpc_put_task);
1234
1235void rpc_put_task_async(struct rpc_task *task)
1236{
1237 rpc_do_put_task(task, task->tk_workqueue);
1238}
1239EXPORT_SYMBOL_GPL(rpc_put_task_async);
1240
1241static void rpc_release_task(struct rpc_task *task)
1242{
1243 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1244
1245 rpc_release_resources_task(task);
1246
1247 /*
1248 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1249 * so it should be safe to use task->tk_count as a test for whether
1250 * or not any other processes still hold references to our rpc_task.
1251 */
1252 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1253 /* Wake up anyone who may be waiting for task completion */
1254 if (!rpc_complete_task(task))
1255 return;
1256 } else {
1257 if (!atomic_dec_and_test(&task->tk_count))
1258 return;
1259 }
1260 rpc_final_put_task(task, task->tk_workqueue);
1261}
1262
1263int rpciod_up(void)
1264{
1265 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1266}
1267
1268void rpciod_down(void)
1269{
1270 module_put(THIS_MODULE);
1271}
1272
1273/*
1274 * Start up the rpciod workqueue.
1275 */
1276static int rpciod_start(void)
1277{
1278 struct workqueue_struct *wq;
1279
1280 /*
1281 * Create the rpciod thread and wait for it to start.
1282 */
1283 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1284 if (!wq)
1285 goto out_failed;
1286 rpciod_workqueue = wq;
1287 wq = alloc_workqueue("xprtiod", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
1288 if (!wq)
1289 goto free_rpciod;
1290 xprtiod_workqueue = wq;
1291 return 1;
1292free_rpciod:
1293 wq = rpciod_workqueue;
1294 rpciod_workqueue = NULL;
1295 destroy_workqueue(wq);
1296out_failed:
1297 return 0;
1298}
1299
1300static void rpciod_stop(void)
1301{
1302 struct workqueue_struct *wq = NULL;
1303
1304 if (rpciod_workqueue == NULL)
1305 return;
1306
1307 wq = rpciod_workqueue;
1308 rpciod_workqueue = NULL;
1309 destroy_workqueue(wq);
1310 wq = xprtiod_workqueue;
1311 xprtiod_workqueue = NULL;
1312 destroy_workqueue(wq);
1313}
1314
1315void
1316rpc_destroy_mempool(void)
1317{
1318 rpciod_stop();
1319 mempool_destroy(rpc_buffer_mempool);
1320 mempool_destroy(rpc_task_mempool);
1321 kmem_cache_destroy(rpc_task_slabp);
1322 kmem_cache_destroy(rpc_buffer_slabp);
1323 rpc_destroy_wait_queue(&delay_queue);
1324}
1325
1326int
1327rpc_init_mempool(void)
1328{
1329 /*
1330 * The following is not strictly a mempool initialisation,
1331 * but there is no harm in doing it here
1332 */
1333 rpc_init_wait_queue(&delay_queue, "delayq");
1334 if (!rpciod_start())
1335 goto err_nomem;
1336
1337 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1338 sizeof(struct rpc_task),
1339 0, SLAB_HWCACHE_ALIGN,
1340 NULL);
1341 if (!rpc_task_slabp)
1342 goto err_nomem;
1343 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1344 RPC_BUFFER_MAXSIZE,
1345 0, SLAB_HWCACHE_ALIGN,
1346 NULL);
1347 if (!rpc_buffer_slabp)
1348 goto err_nomem;
1349 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1350 rpc_task_slabp);
1351 if (!rpc_task_mempool)
1352 goto err_nomem;
1353 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1354 rpc_buffer_slabp);
1355 if (!rpc_buffer_mempool)
1356 goto err_nomem;
1357 return 0;
1358err_nomem:
1359 rpc_destroy_mempool();
1360 return -ENOMEM;
1361}