1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/net/sunrpc/sched.c
5 * Scheduling for synchronous and asynchronous RPC requests.
7 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
13 #include <linux/module.h>
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>
25 #include <linux/sunrpc/clnt.h>
26 #include <linux/sunrpc/metrics.h>
30 #define CREATE_TRACE_POINTS
31 #include <trace/events/sunrpc.h>
34 * RPC slabs and memory pools
36 #define RPC_BUFFER_MAXSIZE (2048)
37 #define RPC_BUFFER_POOLSIZE (8)
38 #define RPC_TASK_POOLSIZE (8)
39 static struct kmem_cache *rpc_task_slabp __read_mostly;
40 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
41 static mempool_t *rpc_task_mempool __read_mostly;
42 static mempool_t *rpc_buffer_mempool __read_mostly;
44 static void rpc_async_schedule(struct work_struct *);
45 static void rpc_release_task(struct rpc_task *task);
46 static void __rpc_queue_timer_fn(struct work_struct *);
49 * RPC tasks sit here while waiting for conditions to improve.
51 static struct rpc_wait_queue delay_queue;
54 * rpciod-related stuff
56 struct workqueue_struct *rpciod_workqueue __read_mostly;
57 struct workqueue_struct *xprtiod_workqueue __read_mostly;
58 EXPORT_SYMBOL_GPL(xprtiod_workqueue);
60 gfp_t rpc_task_gfp_mask(void)
62 if (current->flags & PF_WQ_WORKER)
63 return GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
68 rpc_task_timeout(const struct rpc_task *task)
70 unsigned long timeout = READ_ONCE(task->tk_timeout);
73 unsigned long now = jiffies;
74 if (time_before(now, timeout))
79 EXPORT_SYMBOL_GPL(rpc_task_timeout);
82 * Disable the timer for a given RPC task. Should be called with
83 * queue->lock and bh_disabled in order to avoid races within
87 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
89 if (list_empty(&task->u.tk_wait.timer_list))
92 list_del(&task->u.tk_wait.timer_list);
93 if (list_empty(&queue->timer_list.list))
94 cancel_delayed_work(&queue->timer_list.dwork);
98 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
100 unsigned long now = jiffies;
101 queue->timer_list.expires = expires;
102 if (time_before_eq(expires, now))
106 mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
110 * Set up a timer for the current task.
113 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
114 unsigned long timeout)
116 task->tk_timeout = timeout;
117 if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
118 rpc_set_queue_timer(queue, timeout);
119 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
122 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
124 if (queue->priority != priority) {
125 queue->priority = priority;
126 queue->nr = 1U << priority;
130 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
132 rpc_set_waitqueue_priority(queue, queue->maxpriority);
136 * Add a request to a queue list
139 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
143 list_for_each_entry(t, q, u.tk_wait.list) {
144 if (t->tk_owner == task->tk_owner) {
145 list_add_tail(&task->u.tk_wait.links,
146 &t->u.tk_wait.links);
147 /* Cache the queue head in task->u.tk_wait.list */
148 task->u.tk_wait.list.next = q;
149 task->u.tk_wait.list.prev = NULL;
153 INIT_LIST_HEAD(&task->u.tk_wait.links);
154 list_add_tail(&task->u.tk_wait.list, q);
158 * Remove request from a queue list
161 __rpc_list_dequeue_task(struct rpc_task *task)
166 if (task->u.tk_wait.list.prev == NULL) {
167 list_del(&task->u.tk_wait.links);
170 if (!list_empty(&task->u.tk_wait.links)) {
171 t = list_first_entry(&task->u.tk_wait.links,
174 /* Assume __rpc_list_enqueue_task() cached the queue head */
175 q = t->u.tk_wait.list.next;
176 list_add_tail(&t->u.tk_wait.list, q);
177 list_del(&task->u.tk_wait.links);
179 list_del(&task->u.tk_wait.list);
183 * Add new request to a priority queue.
185 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
186 struct rpc_task *task,
187 unsigned char queue_priority)
189 if (unlikely(queue_priority > queue->maxpriority))
190 queue_priority = queue->maxpriority;
191 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
195 * Add new request to wait queue.
197 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
198 struct rpc_task *task,
199 unsigned char queue_priority)
201 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
202 if (RPC_IS_PRIORITY(queue))
203 __rpc_add_wait_queue_priority(queue, task, queue_priority);
205 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
206 task->tk_waitqueue = queue;
208 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
210 rpc_set_queued(task);
214 * Remove request from a priority queue.
216 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
218 __rpc_list_dequeue_task(task);
222 * Remove request from queue.
223 * Note: must be called with spin lock held.
225 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
227 __rpc_disable_timer(queue, task);
228 if (RPC_IS_PRIORITY(queue))
229 __rpc_remove_wait_queue_priority(task);
231 list_del(&task->u.tk_wait.list);
235 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
239 spin_lock_init(&queue->lock);
240 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
241 INIT_LIST_HEAD(&queue->tasks[i]);
242 queue->maxpriority = nr_queues - 1;
243 rpc_reset_waitqueue_priority(queue);
245 queue->timer_list.expires = 0;
246 INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
247 INIT_LIST_HEAD(&queue->timer_list.list);
248 rpc_assign_waitqueue_name(queue, qname);
251 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
253 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
255 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
257 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
259 __rpc_init_priority_wait_queue(queue, qname, 1);
261 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
263 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
265 cancel_delayed_work_sync(&queue->timer_list.dwork);
267 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
269 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
271 freezable_schedule_unsafe();
272 if (signal_pending_state(mode, current))
277 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
278 static void rpc_task_set_debuginfo(struct rpc_task *task)
280 struct rpc_clnt *clnt = task->tk_client;
282 /* Might be a task carrying a reverse-direction operation */
284 static atomic_t rpc_pid;
286 task->tk_pid = atomic_inc_return(&rpc_pid);
290 task->tk_pid = atomic_inc_return(&clnt->cl_pid);
293 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
298 static void rpc_set_active(struct rpc_task *task)
300 rpc_task_set_debuginfo(task);
301 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
302 trace_rpc_task_begin(task, NULL);
306 * Mark an RPC call as having completed by clearing the 'active' bit
307 * and then waking up all tasks that were sleeping.
309 static int rpc_complete_task(struct rpc_task *task)
311 void *m = &task->tk_runstate;
312 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
313 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
317 trace_rpc_task_complete(task, NULL);
319 spin_lock_irqsave(&wq->lock, flags);
320 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
321 ret = atomic_dec_and_test(&task->tk_count);
322 if (waitqueue_active(wq))
323 __wake_up_locked_key(wq, TASK_NORMAL, &k);
324 spin_unlock_irqrestore(&wq->lock, flags);
329 * Allow callers to wait for completion of an RPC call
331 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
332 * to enforce taking of the wq->lock and hence avoid races with
333 * rpc_complete_task().
335 int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
338 action = rpc_wait_bit_killable;
339 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
340 action, TASK_KILLABLE);
342 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
345 * Make an RPC task runnable.
347 * Note: If the task is ASYNC, and is being made runnable after sitting on an
348 * rpc_wait_queue, this must be called with the queue spinlock held to protect
349 * the wait queue operation.
350 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
351 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
352 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
353 * the RPC_TASK_RUNNING flag.
355 static void rpc_make_runnable(struct workqueue_struct *wq,
356 struct rpc_task *task)
358 bool need_wakeup = !rpc_test_and_set_running(task);
360 rpc_clear_queued(task);
363 if (RPC_IS_ASYNC(task)) {
364 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
365 queue_work(wq, &task->u.tk_work);
367 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
371 * Prepare for sleeping on a wait queue.
372 * By always appending tasks to the list we ensure FIFO behavior.
373 * NB: An RPC task will only receive interrupt-driven events as long
374 * as it's on a wait queue.
376 static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
377 struct rpc_task *task,
378 unsigned char queue_priority)
380 trace_rpc_task_sleep(task, q);
382 __rpc_add_wait_queue(q, task, queue_priority);
385 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
386 struct rpc_task *task,
387 unsigned char queue_priority)
389 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
391 __rpc_do_sleep_on_priority(q, task, queue_priority);
394 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
395 struct rpc_task *task, unsigned long timeout,
396 unsigned char queue_priority)
398 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
400 if (time_is_after_jiffies(timeout)) {
401 __rpc_do_sleep_on_priority(q, task, queue_priority);
402 __rpc_add_timer(q, task, timeout);
404 task->tk_status = -ETIMEDOUT;
407 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
409 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
410 task->tk_callback = action;
413 static bool rpc_sleep_check_activated(struct rpc_task *task)
415 /* We shouldn't ever put an inactive task to sleep */
416 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
417 task->tk_status = -EIO;
418 rpc_put_task_async(task);
424 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
425 rpc_action action, unsigned long timeout)
427 if (!rpc_sleep_check_activated(task))
430 rpc_set_tk_callback(task, action);
433 * Protect the queue operations.
436 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
437 spin_unlock(&q->lock);
439 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
441 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
444 if (!rpc_sleep_check_activated(task))
447 rpc_set_tk_callback(task, action);
449 WARN_ON_ONCE(task->tk_timeout != 0);
451 * Protect the queue operations.
454 __rpc_sleep_on_priority(q, task, task->tk_priority);
455 spin_unlock(&q->lock);
457 EXPORT_SYMBOL_GPL(rpc_sleep_on);
459 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
460 struct rpc_task *task, unsigned long timeout, int priority)
462 if (!rpc_sleep_check_activated(task))
465 priority -= RPC_PRIORITY_LOW;
467 * Protect the queue operations.
470 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
471 spin_unlock(&q->lock);
473 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
475 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
478 if (!rpc_sleep_check_activated(task))
481 WARN_ON_ONCE(task->tk_timeout != 0);
482 priority -= RPC_PRIORITY_LOW;
484 * Protect the queue operations.
487 __rpc_sleep_on_priority(q, task, priority);
488 spin_unlock(&q->lock);
490 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
493 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
494 * @wq: workqueue on which to run task
496 * @task: task to be woken up
498 * Caller must hold queue->lock, and have cleared the task queued flag.
500 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
501 struct rpc_wait_queue *queue,
502 struct rpc_task *task)
504 /* Has the task been executed yet? If not, we cannot wake it up! */
505 if (!RPC_IS_ACTIVATED(task)) {
506 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
510 trace_rpc_task_wakeup(task, queue);
512 __rpc_remove_wait_queue(queue, task);
514 rpc_make_runnable(wq, task);
518 * Wake up a queued task while the queue lock is being held
520 static struct rpc_task *
521 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
522 struct rpc_wait_queue *queue, struct rpc_task *task,
523 bool (*action)(struct rpc_task *, void *), void *data)
525 if (RPC_IS_QUEUED(task)) {
527 if (task->tk_waitqueue == queue) {
528 if (action == NULL || action(task, data)) {
529 __rpc_do_wake_up_task_on_wq(wq, queue, task);
538 * Wake up a queued task while the queue lock is being held
540 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
541 struct rpc_task *task)
543 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
548 * Wake up a task on a specific queue
550 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
552 if (!RPC_IS_QUEUED(task))
554 spin_lock(&queue->lock);
555 rpc_wake_up_task_queue_locked(queue, task);
556 spin_unlock(&queue->lock);
558 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
560 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
562 task->tk_status = *(int *)status;
567 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
568 struct rpc_task *task, int status)
570 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
571 task, rpc_task_action_set_status, &status);
575 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
576 * @queue: pointer to rpc_wait_queue
577 * @task: pointer to rpc_task
578 * @status: integer error value
580 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
581 * set to the value of @status.
584 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
585 struct rpc_task *task, int status)
587 if (!RPC_IS_QUEUED(task))
589 spin_lock(&queue->lock);
590 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
591 spin_unlock(&queue->lock);
595 * Wake up the next task on a priority queue.
597 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
600 struct rpc_task *task;
603 * Service the privileged queue.
605 q = &queue->tasks[RPC_NR_PRIORITY - 1];
606 if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
607 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
612 * Service a batch of tasks from a single owner.
614 q = &queue->tasks[queue->priority];
615 if (!list_empty(q) && queue->nr) {
617 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
622 * Service the next queue.
625 if (q == &queue->tasks[0])
626 q = &queue->tasks[queue->maxpriority];
629 if (!list_empty(q)) {
630 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
633 } while (q != &queue->tasks[queue->priority]);
635 rpc_reset_waitqueue_priority(queue);
639 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
644 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
646 if (RPC_IS_PRIORITY(queue))
647 return __rpc_find_next_queued_priority(queue);
648 if (!list_empty(&queue->tasks[0]))
649 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
654 * Wake up the first task on the wait queue.
656 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
657 struct rpc_wait_queue *queue,
658 bool (*func)(struct rpc_task *, void *), void *data)
660 struct rpc_task *task = NULL;
662 spin_lock(&queue->lock);
663 task = __rpc_find_next_queued(queue);
665 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
667 spin_unlock(&queue->lock);
673 * Wake up the first task on the wait queue.
675 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
676 bool (*func)(struct rpc_task *, void *), void *data)
678 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
680 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
682 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
688 * Wake up the next task on the wait queue.
690 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
692 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
694 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
697 * rpc_wake_up_locked - wake up all rpc_tasks
698 * @queue: rpc_wait_queue on which the tasks are sleeping
701 static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
703 struct rpc_task *task;
706 task = __rpc_find_next_queued(queue);
709 rpc_wake_up_task_queue_locked(queue, task);
714 * rpc_wake_up - wake up all rpc_tasks
715 * @queue: rpc_wait_queue on which the tasks are sleeping
719 void rpc_wake_up(struct rpc_wait_queue *queue)
721 spin_lock(&queue->lock);
722 rpc_wake_up_locked(queue);
723 spin_unlock(&queue->lock);
725 EXPORT_SYMBOL_GPL(rpc_wake_up);
728 * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
729 * @queue: rpc_wait_queue on which the tasks are sleeping
730 * @status: status value to set
732 static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
734 struct rpc_task *task;
737 task = __rpc_find_next_queued(queue);
740 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
745 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
746 * @queue: rpc_wait_queue on which the tasks are sleeping
747 * @status: status value to set
751 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
753 spin_lock(&queue->lock);
754 rpc_wake_up_status_locked(queue, status);
755 spin_unlock(&queue->lock);
757 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
759 static void __rpc_queue_timer_fn(struct work_struct *work)
761 struct rpc_wait_queue *queue = container_of(work,
762 struct rpc_wait_queue,
763 timer_list.dwork.work);
764 struct rpc_task *task, *n;
765 unsigned long expires, now, timeo;
767 spin_lock(&queue->lock);
768 expires = now = jiffies;
769 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
770 timeo = task->tk_timeout;
771 if (time_after_eq(now, timeo)) {
772 trace_rpc_task_timeout(task, task->tk_action);
773 task->tk_status = -ETIMEDOUT;
774 rpc_wake_up_task_queue_locked(queue, task);
777 if (expires == now || time_after(expires, timeo))
780 if (!list_empty(&queue->timer_list.list))
781 rpc_set_queue_timer(queue, expires);
782 spin_unlock(&queue->lock);
785 static void __rpc_atrun(struct rpc_task *task)
787 if (task->tk_status == -ETIMEDOUT)
792 * Run a task at a later time
794 void rpc_delay(struct rpc_task *task, unsigned long delay)
796 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
798 EXPORT_SYMBOL_GPL(rpc_delay);
801 * Helper to call task->tk_ops->rpc_call_prepare
803 void rpc_prepare_task(struct rpc_task *task)
805 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
809 rpc_init_task_statistics(struct rpc_task *task)
811 /* Initialize retry counters */
812 task->tk_garb_retry = 2;
813 task->tk_cred_retry = 2;
814 task->tk_rebind_retry = 2;
816 /* starting timestamp */
817 task->tk_start = ktime_get();
821 rpc_reset_task_statistics(struct rpc_task *task)
823 task->tk_timeouts = 0;
824 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
825 rpc_init_task_statistics(task);
829 * Helper that calls task->tk_ops->rpc_call_done if it exists
831 void rpc_exit_task(struct rpc_task *task)
833 trace_rpc_task_end(task, task->tk_action);
834 task->tk_action = NULL;
835 if (task->tk_ops->rpc_count_stats)
836 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
837 else if (task->tk_client)
838 rpc_count_iostats(task, task->tk_client->cl_metrics);
839 if (task->tk_ops->rpc_call_done != NULL) {
840 trace_rpc_task_call_done(task, task->tk_ops->rpc_call_done);
841 task->tk_ops->rpc_call_done(task, task->tk_calldata);
842 if (task->tk_action != NULL) {
843 /* Always release the RPC slot and buffer memory */
845 rpc_reset_task_statistics(task);
850 void rpc_signal_task(struct rpc_task *task)
852 struct rpc_wait_queue *queue;
854 if (!RPC_IS_ACTIVATED(task))
857 trace_rpc_task_signalled(task, task->tk_action);
858 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
859 smp_mb__after_atomic();
860 queue = READ_ONCE(task->tk_waitqueue);
862 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
865 void rpc_exit(struct rpc_task *task, int status)
867 task->tk_status = status;
868 task->tk_action = rpc_exit_task;
869 rpc_wake_up_queued_task(task->tk_waitqueue, task);
871 EXPORT_SYMBOL_GPL(rpc_exit);
873 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
875 if (ops->rpc_release != NULL)
876 ops->rpc_release(calldata);
879 static bool xprt_needs_memalloc(struct rpc_xprt *xprt, struct rpc_task *tk)
883 if (!atomic_read(&xprt->swapper))
885 return test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == tk;
889 * This is the RPC `scheduler' (or rather, the finite state machine).
891 static void __rpc_execute(struct rpc_task *task)
893 struct rpc_wait_queue *queue;
894 int task_is_async = RPC_IS_ASYNC(task);
896 unsigned long pflags = current->flags;
898 WARN_ON_ONCE(RPC_IS_QUEUED(task));
899 if (RPC_IS_QUEUED(task))
903 void (*do_action)(struct rpc_task *);
906 * Perform the next FSM step or a pending callback.
908 * tk_action may be NULL if the task has been killed.
909 * In particular, note that rpc_killall_tasks may
910 * do this at any time, so beware when dereferencing.
912 do_action = task->tk_action;
913 if (task->tk_callback) {
914 do_action = task->tk_callback;
915 task->tk_callback = NULL;
919 if (RPC_IS_SWAPPER(task) ||
920 xprt_needs_memalloc(task->tk_xprt, task))
921 current->flags |= PF_MEMALLOC;
923 trace_rpc_task_run_action(task, do_action);
927 * Lockless check for whether task is sleeping or not.
929 if (!RPC_IS_QUEUED(task)) {
935 * Signalled tasks should exit rather than sleep.
937 if (RPC_SIGNALLED(task)) {
938 task->tk_rpc_status = -ERESTARTSYS;
939 rpc_exit(task, -ERESTARTSYS);
943 * The queue->lock protects against races with
944 * rpc_make_runnable().
946 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
947 * rpc_task, rpc_make_runnable() can assign it to a
948 * different workqueue. We therefore cannot assume that the
949 * rpc_task pointer may still be dereferenced.
951 queue = task->tk_waitqueue;
952 spin_lock(&queue->lock);
953 if (!RPC_IS_QUEUED(task)) {
954 spin_unlock(&queue->lock);
957 rpc_clear_running(task);
958 spin_unlock(&queue->lock);
962 /* sync task: sleep here */
963 trace_rpc_task_sync_sleep(task, task->tk_action);
964 status = out_of_line_wait_on_bit(&task->tk_runstate,
965 RPC_TASK_QUEUED, rpc_wait_bit_killable,
969 * When a sync task receives a signal, it exits with
970 * -ERESTARTSYS. In order to catch any callbacks that
971 * clean up after sleeping on some queue, we don't
972 * break the loop here, but go around once more.
974 trace_rpc_task_signalled(task, task->tk_action);
975 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
976 task->tk_rpc_status = -ERESTARTSYS;
977 rpc_exit(task, -ERESTARTSYS);
979 trace_rpc_task_sync_wake(task, task->tk_action);
982 /* Release all resources associated with the task */
983 rpc_release_task(task);
985 current_restore_flags(pflags, PF_MEMALLOC);
989 * User-visible entry point to the scheduler.
991 * This may be called recursively if e.g. an async NFS task updates
992 * the attributes and finds that dirty pages must be flushed.
993 * NOTE: Upon exit of this function the task is guaranteed to be
994 * released. In particular note that tk_release() will have
995 * been called, so your task memory may have been freed.
997 void rpc_execute(struct rpc_task *task)
999 bool is_async = RPC_IS_ASYNC(task);
1001 rpc_set_active(task);
1002 rpc_make_runnable(rpciod_workqueue, task);
1004 unsigned int pflags = memalloc_nofs_save();
1005 __rpc_execute(task);
1006 memalloc_nofs_restore(pflags);
1010 static void rpc_async_schedule(struct work_struct *work)
1012 unsigned int pflags = memalloc_nofs_save();
1014 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
1015 memalloc_nofs_restore(pflags);
1019 * rpc_malloc - allocate RPC buffer resources
1022 * A single memory region is allocated, which is split between the
1023 * RPC call and RPC reply that this task is being used for. When
1024 * this RPC is retired, the memory is released by calling rpc_free.
1026 * To prevent rpciod from hanging, this allocator never sleeps,
1027 * returning -ENOMEM and suppressing warning if the request cannot
1028 * be serviced immediately. The caller can arrange to sleep in a
1029 * way that is safe for rpciod.
1031 * Most requests are 'small' (under 2KiB) and can be serviced from a
1032 * mempool, ensuring that NFS reads and writes can always proceed,
1033 * and that there is good locality of reference for these buffers.
1035 int rpc_malloc(struct rpc_task *task)
1037 struct rpc_rqst *rqst = task->tk_rqstp;
1038 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1039 struct rpc_buffer *buf;
1040 gfp_t gfp = rpc_task_gfp_mask();
1042 size += sizeof(struct rpc_buffer);
1043 if (size <= RPC_BUFFER_MAXSIZE) {
1044 buf = kmem_cache_alloc(rpc_buffer_slabp, gfp);
1045 /* Reach for the mempool if dynamic allocation fails */
1046 if (!buf && RPC_IS_ASYNC(task))
1047 buf = mempool_alloc(rpc_buffer_mempool, GFP_NOWAIT);
1049 buf = kmalloc(size, gfp);
1055 rqst->rq_buffer = buf->data;
1056 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1059 EXPORT_SYMBOL_GPL(rpc_malloc);
1062 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1066 void rpc_free(struct rpc_task *task)
1068 void *buffer = task->tk_rqstp->rq_buffer;
1070 struct rpc_buffer *buf;
1072 buf = container_of(buffer, struct rpc_buffer, data);
1075 if (size <= RPC_BUFFER_MAXSIZE)
1076 mempool_free(buf, rpc_buffer_mempool);
1080 EXPORT_SYMBOL_GPL(rpc_free);
1083 * Creation and deletion of RPC task structures
1085 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1087 memset(task, 0, sizeof(*task));
1088 atomic_set(&task->tk_count, 1);
1089 task->tk_flags = task_setup_data->flags;
1090 task->tk_ops = task_setup_data->callback_ops;
1091 task->tk_calldata = task_setup_data->callback_data;
1092 INIT_LIST_HEAD(&task->tk_task);
1094 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1095 task->tk_owner = current->tgid;
1097 /* Initialize workqueue for async tasks */
1098 task->tk_workqueue = task_setup_data->workqueue;
1100 task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1101 xprt_get(task_setup_data->rpc_xprt));
1103 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1105 if (task->tk_ops->rpc_call_prepare != NULL)
1106 task->tk_action = rpc_prepare_task;
1108 rpc_init_task_statistics(task);
1111 static struct rpc_task *rpc_alloc_task(void)
1113 struct rpc_task *task;
1115 task = kmem_cache_alloc(rpc_task_slabp, rpc_task_gfp_mask());
1118 return mempool_alloc(rpc_task_mempool, GFP_NOWAIT);
1122 * Create a new task for the specified client.
1124 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1126 struct rpc_task *task = setup_data->task;
1127 unsigned short flags = 0;
1130 task = rpc_alloc_task();
1132 rpc_release_calldata(setup_data->callback_ops,
1133 setup_data->callback_data);
1134 return ERR_PTR(-ENOMEM);
1136 flags = RPC_TASK_DYNAMIC;
1139 rpc_init_task(task, setup_data);
1140 task->tk_flags |= flags;
1145 * rpc_free_task - release rpc task and perform cleanups
1147 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1148 * in order to work around a workqueue dependency issue.
1151 * "Workqueue currently considers two work items to be the same if they're
1152 * on the same address and won't execute them concurrently - ie. it
1153 * makes a work item which is queued again while being executed wait
1154 * for the previous execution to complete.
1156 * If a work function frees the work item, and then waits for an event
1157 * which should be performed by another work item and *that* work item
1158 * recycles the freed work item, it can create a false dependency loop.
1159 * There really is no reliable way to detect this short of verifying
1160 * every memory free."
1163 static void rpc_free_task(struct rpc_task *task)
1165 unsigned short tk_flags = task->tk_flags;
1167 put_rpccred(task->tk_op_cred);
1168 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1170 if (tk_flags & RPC_TASK_DYNAMIC)
1171 mempool_free(task, rpc_task_mempool);
1174 static void rpc_async_release(struct work_struct *work)
1176 unsigned int pflags = memalloc_nofs_save();
1178 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1179 memalloc_nofs_restore(pflags);
1182 static void rpc_release_resources_task(struct rpc_task *task)
1185 if (task->tk_msg.rpc_cred) {
1186 if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1187 put_cred(task->tk_msg.rpc_cred);
1188 task->tk_msg.rpc_cred = NULL;
1190 rpc_task_release_client(task);
1193 static void rpc_final_put_task(struct rpc_task *task,
1194 struct workqueue_struct *q)
1197 INIT_WORK(&task->u.tk_work, rpc_async_release);
1198 queue_work(q, &task->u.tk_work);
1200 rpc_free_task(task);
1203 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1205 if (atomic_dec_and_test(&task->tk_count)) {
1206 rpc_release_resources_task(task);
1207 rpc_final_put_task(task, q);
1211 void rpc_put_task(struct rpc_task *task)
1213 rpc_do_put_task(task, NULL);
1215 EXPORT_SYMBOL_GPL(rpc_put_task);
1217 void rpc_put_task_async(struct rpc_task *task)
1219 rpc_do_put_task(task, task->tk_workqueue);
1221 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1223 static void rpc_release_task(struct rpc_task *task)
1225 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1227 rpc_release_resources_task(task);
1230 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1231 * so it should be safe to use task->tk_count as a test for whether
1232 * or not any other processes still hold references to our rpc_task.
1234 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1235 /* Wake up anyone who may be waiting for task completion */
1236 if (!rpc_complete_task(task))
1239 if (!atomic_dec_and_test(&task->tk_count))
1242 rpc_final_put_task(task, task->tk_workqueue);
1247 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1250 void rpciod_down(void)
1252 module_put(THIS_MODULE);
1256 * Start up the rpciod workqueue.
1258 static int rpciod_start(void)
1260 struct workqueue_struct *wq;
1263 * Create the rpciod thread and wait for it to start.
1265 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1268 rpciod_workqueue = wq;
1269 wq = alloc_workqueue("xprtiod", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
1272 xprtiod_workqueue = wq;
1275 wq = rpciod_workqueue;
1276 rpciod_workqueue = NULL;
1277 destroy_workqueue(wq);
1282 static void rpciod_stop(void)
1284 struct workqueue_struct *wq = NULL;
1286 if (rpciod_workqueue == NULL)
1289 wq = rpciod_workqueue;
1290 rpciod_workqueue = NULL;
1291 destroy_workqueue(wq);
1292 wq = xprtiod_workqueue;
1293 xprtiod_workqueue = NULL;
1294 destroy_workqueue(wq);
1298 rpc_destroy_mempool(void)
1301 mempool_destroy(rpc_buffer_mempool);
1302 mempool_destroy(rpc_task_mempool);
1303 kmem_cache_destroy(rpc_task_slabp);
1304 kmem_cache_destroy(rpc_buffer_slabp);
1305 rpc_destroy_wait_queue(&delay_queue);
1309 rpc_init_mempool(void)
1312 * The following is not strictly a mempool initialisation,
1313 * but there is no harm in doing it here
1315 rpc_init_wait_queue(&delay_queue, "delayq");
1316 if (!rpciod_start())
1319 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1320 sizeof(struct rpc_task),
1321 0, SLAB_HWCACHE_ALIGN,
1323 if (!rpc_task_slabp)
1325 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1327 0, SLAB_HWCACHE_ALIGN,
1329 if (!rpc_buffer_slabp)
1331 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1333 if (!rpc_task_mempool)
1335 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1337 if (!rpc_buffer_mempool)
1341 rpc_destroy_mempool();