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(head: &task->u.tk_wait.timer_list))
98 return;
99 task->tk_timeout = 0;
100 list_del(entry: &task->u.tk_wait.timer_list);
101 if (list_empty(head: &queue->timer_list.list))
102 cancel_delayed_work(dwork: &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(wq: rpciod_workqueue, dwork: &queue->timer_list.dwork, delay: 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(head: &queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
126 rpc_set_queue_timer(queue, expires: timeout);
127 list_add(new: &task->u.tk_wait.timer_list, head: &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, priority: 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(new: &task->u.tk_wait.links,
154 head: &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(list: &task->u.tk_wait.links);
162 list_add_tail(new: &task->u.tk_wait.list, head: 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(entry: &task->u.tk_wait.links);
176 return;
177 }
178 if (!list_empty(head: &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(new: &t->u.tk_wait.list, head: q);
185 list_del(entry: &task->u.tk_wait.links);
186 }
187 list_del(entry: &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(q: &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(list: &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(new: &task->u.tk_wait.list, head: &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(entry: &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(list: &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(list: &queue->timer_list.list);
256 rpc_assign_waitqueue_name(q: queue, name: 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, nr_queues: 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(dwork: &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(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(v: &rpc_pid);
295 return;
296 }
297
298 task->tk_pid = atomic_inc_return(v: &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(nr: RPC_TASK_ACTIVE, addr: &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(word: m, bit: 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(nr: RPC_TASK_ACTIVE, addr: &task->tk_runstate);
329 ret = atomic_dec_and_test(v: &task->tk_count);
330 if (waitqueue_active(wq_head: wq))
331 __wake_up_locked_key(wq_head: wq, TASK_NORMAL, key: &k);
332 spin_unlock_irqrestore(lock: &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(word: &task->tk_runstate, RPC_TASK_ACTIVE,
346 action: 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, work: &task->u.tk_work);
372 } else {
373 smp_mb__after_atomic();
374 wake_up_bit(word: &task->tk_runstate, bit: 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(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(queue: 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(lock: &q->lock);
444 __rpc_sleep_on_priority_timeout(q, task, timeout, queue_priority: task->tk_priority);
445 spin_unlock(lock: &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(lock: &q->lock);
462 __rpc_sleep_on_priority(q, task, queue_priority: task->tk_priority);
463 spin_unlock(lock: &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(lock: &q->lock);
478 __rpc_sleep_on_priority_timeout(q, task, timeout, queue_priority: priority);
479 spin_unlock(lock: &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(lock: &q->lock);
495 __rpc_sleep_on_priority(q, task, queue_priority: priority);
496 spin_unlock(lock: &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, q: 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(wq: 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(lock: &queue->lock);
563 rpc_wake_up_task_queue_locked(queue, task);
564 spin_unlock(lock: &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(wq: rpciod_workqueue, queue,
579 task, action: rpc_task_action_set_status, data: &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(lock: &queue->lock);
598 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
599 spin_unlock(lock: &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(head: 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(head: 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(head: 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, priority: (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(head: &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(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, action: func, data);
675 spin_unlock(lock: &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(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(lock: &queue->lock);
730 rpc_wake_up_locked(queue);
731 spin_unlock(lock: &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(lock: &queue->lock);
762 rpc_wake_up_status_locked(queue, status);
763 spin_unlock(lock: &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(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, action: 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(head: &queue->timer_list.list))
789 rpc_set_queue_timer(queue, expires);
790 spin_unlock(lock: &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, action: 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, action: 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, rpc_status: -ERESTARTSYS))
865 return;
866 trace_rpc_task_signalled(task, action: 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, rpc_status: 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(v: &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(xprt: task->tk_xprt, tk: task))
944 current->flags |= PF_MEMALLOC;
945
946 trace_rpc_task_run_action(task, action: 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(lock: &queue->lock);
968 if (!RPC_IS_QUEUED(task)) {
969 spin_unlock(lock: &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(lock: &queue->lock);
976 continue;
977 }
978 rpc_clear_running(task);
979 spin_unlock(lock: &queue->lock);
980 if (task_is_async)
981 goto out;
982
983 /* sync task: sleep here */
984 trace_rpc_task_sync_sleep(task, action: task->tk_action);
985 status = out_of_line_wait_on_bit(word: &task->tk_runstate,
986 RPC_TASK_QUEUED, action: 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, action: task->tk_action);
998 }
999
1000 /* Release all resources associated with the task */
1001 rpc_release_task(task);
1002out:
1003 current_restore_flags(orig_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(wq: rpciod_workqueue, task);
1021 if (!is_async) {
1022 unsigned int pflags = memalloc_nofs_save();
1023 __rpc_execute(task);
1024 memalloc_nofs_restore(flags: 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(flags: 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}
1077
1078/**
1079 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1080 * @task: RPC task
1081 *
1082 */
1083void rpc_free(struct rpc_task *task)
1084{
1085 void *buffer = task->tk_rqstp->rq_buffer;
1086 size_t size;
1087 struct rpc_buffer *buf;
1088
1089 buf = container_of(buffer, struct rpc_buffer, data);
1090 size = buf->len;
1091
1092 if (size <= RPC_BUFFER_MAXSIZE)
1093 mempool_free(element: buf, pool: rpc_buffer_mempool);
1094 else
1095 kfree(objp: buf);
1096}
1097
1098/*
1099 * Creation and deletion of RPC task structures
1100 */
1101static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1102{
1103 memset(s: task, c: 0, n: sizeof(*task));
1104 atomic_set(v: &task->tk_count, i: 1);
1105 task->tk_flags = task_setup_data->flags;
1106 task->tk_ops = task_setup_data->callback_ops;
1107 task->tk_calldata = task_setup_data->callback_data;
1108 INIT_LIST_HEAD(list: &task->tk_task);
1109
1110 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1111 task->tk_owner = current->tgid;
1112
1113 /* Initialize workqueue for async tasks */
1114 task->tk_workqueue = task_setup_data->workqueue;
1115
1116 task->tk_xprt = rpc_task_get_xprt(clnt: task_setup_data->rpc_client,
1117 xprt: xprt_get(xprt: task_setup_data->rpc_xprt));
1118
1119 task->tk_op_cred = get_rpccred(cred: task_setup_data->rpc_op_cred);
1120
1121 if (task->tk_ops->rpc_call_prepare != NULL)
1122 task->tk_action = rpc_prepare_task;
1123
1124 rpc_init_task_statistics(task);
1125}
1126
1127static struct rpc_task *rpc_alloc_task(void)
1128{
1129 struct rpc_task *task;
1130
1131 task = kmem_cache_alloc(rpc_task_slabp, rpc_task_gfp_mask());
1132 if (task)
1133 return task;
1134 return mempool_alloc(rpc_task_mempool, GFP_NOWAIT);
1135}
1136
1137/*
1138 * Create a new task for the specified client.
1139 */
1140struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1141{
1142 struct rpc_task *task = setup_data->task;
1143 unsigned short flags = 0;
1144
1145 if (task == NULL) {
1146 task = rpc_alloc_task();
1147 if (task == NULL) {
1148 rpc_release_calldata(ops: setup_data->callback_ops,
1149 calldata: setup_data->callback_data);
1150 return ERR_PTR(error: -ENOMEM);
1151 }
1152 flags = RPC_TASK_DYNAMIC;
1153 }
1154
1155 rpc_init_task(task, task_setup_data: setup_data);
1156 task->tk_flags |= flags;
1157 return task;
1158}
1159
1160/*
1161 * rpc_free_task - release rpc task and perform cleanups
1162 *
1163 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1164 * in order to work around a workqueue dependency issue.
1165 *
1166 * Tejun Heo states:
1167 * "Workqueue currently considers two work items to be the same if they're
1168 * on the same address and won't execute them concurrently - ie. it
1169 * makes a work item which is queued again while being executed wait
1170 * for the previous execution to complete.
1171 *
1172 * If a work function frees the work item, and then waits for an event
1173 * which should be performed by another work item and *that* work item
1174 * recycles the freed work item, it can create a false dependency loop.
1175 * There really is no reliable way to detect this short of verifying
1176 * every memory free."
1177 *
1178 */
1179static void rpc_free_task(struct rpc_task *task)
1180{
1181 unsigned short tk_flags = task->tk_flags;
1182
1183 put_rpccred(task->tk_op_cred);
1184 rpc_release_calldata(ops: task->tk_ops, calldata: task->tk_calldata);
1185
1186 if (tk_flags & RPC_TASK_DYNAMIC)
1187 mempool_free(element: task, pool: rpc_task_mempool);
1188}
1189
1190static void rpc_async_release(struct work_struct *work)
1191{
1192 unsigned int pflags = memalloc_nofs_save();
1193
1194 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1195 memalloc_nofs_restore(flags: pflags);
1196}
1197
1198static void rpc_release_resources_task(struct rpc_task *task)
1199{
1200 xprt_release(task);
1201 if (task->tk_msg.rpc_cred) {
1202 if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1203 put_cred(cred: task->tk_msg.rpc_cred);
1204 task->tk_msg.rpc_cred = NULL;
1205 }
1206 rpc_task_release_client(task);
1207}
1208
1209static void rpc_final_put_task(struct rpc_task *task,
1210 struct workqueue_struct *q)
1211{
1212 if (q != NULL) {
1213 INIT_WORK(&task->u.tk_work, rpc_async_release);
1214 queue_work(wq: q, work: &task->u.tk_work);
1215 } else
1216 rpc_free_task(task);
1217}
1218
1219static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1220{
1221 if (atomic_dec_and_test(v: &task->tk_count)) {
1222 rpc_release_resources_task(task);
1223 rpc_final_put_task(task, q);
1224 }
1225}
1226
1227void rpc_put_task(struct rpc_task *task)
1228{
1229 rpc_do_put_task(task, NULL);
1230}
1231EXPORT_SYMBOL_GPL(rpc_put_task);
1232
1233void rpc_put_task_async(struct rpc_task *task)
1234{
1235 rpc_do_put_task(task, q: task->tk_workqueue);
1236}
1237EXPORT_SYMBOL_GPL(rpc_put_task_async);
1238
1239static void rpc_release_task(struct rpc_task *task)
1240{
1241 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1242
1243 rpc_release_resources_task(task);
1244
1245 /*
1246 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1247 * so it should be safe to use task->tk_count as a test for whether
1248 * or not any other processes still hold references to our rpc_task.
1249 */
1250 if (atomic_read(v: &task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1251 /* Wake up anyone who may be waiting for task completion */
1252 if (!rpc_complete_task(task))
1253 return;
1254 } else {
1255 if (!atomic_dec_and_test(v: &task->tk_count))
1256 return;
1257 }
1258 rpc_final_put_task(task, q: task->tk_workqueue);
1259}
1260
1261int rpciod_up(void)
1262{
1263 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1264}
1265
1266void rpciod_down(void)
1267{
1268 module_put(THIS_MODULE);
1269}
1270
1271/*
1272 * Start up the rpciod workqueue.
1273 */
1274static int rpciod_start(void)
1275{
1276 struct workqueue_struct *wq;
1277
1278 /*
1279 * Create the rpciod thread and wait for it to start.
1280 */
1281 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1282 if (!wq)
1283 goto out_failed;
1284 rpciod_workqueue = wq;
1285 wq = alloc_workqueue("xprtiod", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
1286 if (!wq)
1287 goto free_rpciod;
1288 xprtiod_workqueue = wq;
1289 return 1;
1290free_rpciod:
1291 wq = rpciod_workqueue;
1292 rpciod_workqueue = NULL;
1293 destroy_workqueue(wq);
1294out_failed:
1295 return 0;
1296}
1297
1298static void rpciod_stop(void)
1299{
1300 struct workqueue_struct *wq = NULL;
1301
1302 if (rpciod_workqueue == NULL)
1303 return;
1304
1305 wq = rpciod_workqueue;
1306 rpciod_workqueue = NULL;
1307 destroy_workqueue(wq);
1308 wq = xprtiod_workqueue;
1309 xprtiod_workqueue = NULL;
1310 destroy_workqueue(wq);
1311}
1312
1313void
1314rpc_destroy_mempool(void)
1315{
1316 rpciod_stop();
1317 mempool_destroy(pool: rpc_buffer_mempool);
1318 mempool_destroy(pool: rpc_task_mempool);
1319 kmem_cache_destroy(s: rpc_task_slabp);
1320 kmem_cache_destroy(s: rpc_buffer_slabp);
1321 rpc_destroy_wait_queue(&delay_queue);
1322}
1323
1324int
1325rpc_init_mempool(void)
1326{
1327 /*
1328 * The following is not strictly a mempool initialisation,
1329 * but there is no harm in doing it here
1330 */
1331 rpc_init_wait_queue(&delay_queue, "delayq");
1332 if (!rpciod_start())
1333 goto err_nomem;
1334
1335 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1336 sizeof(struct rpc_task),
1337 0, SLAB_HWCACHE_ALIGN,
1338 NULL);
1339 if (!rpc_task_slabp)
1340 goto err_nomem;
1341 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1342 RPC_BUFFER_MAXSIZE,
1343 0, SLAB_HWCACHE_ALIGN,
1344 NULL);
1345 if (!rpc_buffer_slabp)
1346 goto err_nomem;
1347 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1348 rpc_task_slabp);
1349 if (!rpc_task_mempool)
1350 goto err_nomem;
1351 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1352 rpc_buffer_slabp);
1353 if (!rpc_buffer_mempool)
1354 goto err_nomem;
1355 return 0;
1356err_nomem:
1357 rpc_destroy_mempool();
1358 return -ENOMEM;
1359}
1360