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 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
31 #define RPCDBG_FACILITY RPCDBG_SCHED
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/sunrpc.h>
38 * RPC slabs and memory pools
40 #define RPC_BUFFER_MAXSIZE (2048)
41 #define RPC_BUFFER_POOLSIZE (8)
42 #define RPC_TASK_POOLSIZE (8)
43 static struct kmem_cache *rpc_task_slabp __read_mostly;
44 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
45 static mempool_t *rpc_task_mempool __read_mostly;
46 static mempool_t *rpc_buffer_mempool __read_mostly;
48 static void rpc_async_schedule(struct work_struct *);
49 static void rpc_release_task(struct rpc_task *task);
50 static void __rpc_queue_timer_fn(struct work_struct *);
53 * RPC tasks sit here while waiting for conditions to improve.
55 static struct rpc_wait_queue delay_queue;
58 * rpciod-related stuff
60 struct workqueue_struct *rpciod_workqueue __read_mostly;
61 struct workqueue_struct *xprtiod_workqueue __read_mostly;
62 EXPORT_SYMBOL_GPL(xprtiod_workqueue);
65 rpc_task_timeout(const struct rpc_task *task)
67 unsigned long timeout = READ_ONCE(task->tk_timeout);
70 unsigned long now = jiffies;
71 if (time_before(now, timeout))
76 EXPORT_SYMBOL_GPL(rpc_task_timeout);
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
84 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
86 if (list_empty(&task->u.tk_wait.timer_list))
88 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
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);
96 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
98 unsigned long now = jiffies;
99 queue->timer_list.expires = expires;
100 if (time_before_eq(expires, now))
104 mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
108 * Set up a timer for the current task.
111 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
112 unsigned long timeout)
114 dprintk("RPC: %5u setting alarm for %u ms\n",
115 task->tk_pid, jiffies_to_msecs(timeout - jiffies));
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);
123 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
125 if (queue->priority != priority) {
126 queue->priority = priority;
127 queue->nr = 1U << priority;
131 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
133 rpc_set_waitqueue_priority(queue, queue->maxpriority);
137 * Add a request to a queue list
140 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
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;
154 INIT_LIST_HEAD(&task->u.tk_wait.links);
155 list_add_tail(&task->u.tk_wait.list, q);
159 * Remove request from a queue list
162 __rpc_list_dequeue_task(struct rpc_task *task)
167 if (task->u.tk_wait.list.prev == NULL) {
168 list_del(&task->u.tk_wait.links);
171 if (!list_empty(&task->u.tk_wait.links)) {
172 t = list_first_entry(&task->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);
180 list_del(&task->u.tk_wait.list);
184 * Add new request to a priority queue.
186 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
187 struct rpc_task *task,
188 unsigned char queue_priority)
190 if (unlikely(queue_priority > queue->maxpriority))
191 queue_priority = queue->maxpriority;
192 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
196 * Add new request to wait queue.
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.
203 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
204 struct rpc_task *task,
205 unsigned char queue_priority)
207 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
208 if (RPC_IS_PRIORITY(queue))
209 __rpc_add_wait_queue_priority(queue, task, queue_priority);
210 else if (RPC_IS_SWAPPER(task))
211 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
213 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
214 task->tk_waitqueue = queue;
216 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
218 rpc_set_queued(task);
220 dprintk("RPC: %5u added to queue %p \"%s\"\n",
221 task->tk_pid, queue, rpc_qname(queue));
225 * Remove request from a priority queue.
227 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
229 __rpc_list_dequeue_task(task);
233 * Remove request from queue.
234 * Note: must be called with spin lock held.
236 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
238 __rpc_disable_timer(queue, task);
239 if (RPC_IS_PRIORITY(queue))
240 __rpc_remove_wait_queue_priority(task);
242 list_del(&task->u.tk_wait.list);
244 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
245 task->tk_pid, queue, rpc_qname(queue));
248 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
252 spin_lock_init(&queue->lock);
253 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
254 INIT_LIST_HEAD(&queue->tasks[i]);
255 queue->maxpriority = nr_queues - 1;
256 rpc_reset_waitqueue_priority(queue);
258 queue->timer_list.expires = 0;
259 INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
260 INIT_LIST_HEAD(&queue->timer_list.list);
261 rpc_assign_waitqueue_name(queue, qname);
264 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
266 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
268 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
270 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
272 __rpc_init_priority_wait_queue(queue, qname, 1);
274 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
276 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
278 cancel_delayed_work_sync(&queue->timer_list.dwork);
280 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
282 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
284 freezable_schedule_unsafe();
285 if (signal_pending_state(mode, current))
290 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
291 static void rpc_task_set_debuginfo(struct rpc_task *task)
293 static atomic_t rpc_pid;
295 task->tk_pid = atomic_inc_return(&rpc_pid);
298 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
303 static void rpc_set_active(struct rpc_task *task)
305 rpc_task_set_debuginfo(task);
306 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
307 trace_rpc_task_begin(task, NULL);
311 * Mark an RPC call as having completed by clearing the 'active' bit
312 * and then waking up all tasks that were sleeping.
314 static int rpc_complete_task(struct rpc_task *task)
316 void *m = &task->tk_runstate;
317 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
318 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
322 trace_rpc_task_complete(task, NULL);
324 spin_lock_irqsave(&wq->lock, flags);
325 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
326 ret = atomic_dec_and_test(&task->tk_count);
327 if (waitqueue_active(wq))
328 __wake_up_locked_key(wq, TASK_NORMAL, &k);
329 spin_unlock_irqrestore(&wq->lock, flags);
334 * Allow callers to wait for completion of an RPC call
336 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
337 * to enforce taking of the wq->lock and hence avoid races with
338 * rpc_complete_task().
340 int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
343 action = rpc_wait_bit_killable;
344 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
345 action, TASK_KILLABLE);
347 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
350 * Make an RPC task runnable.
352 * Note: If the task is ASYNC, and is being made runnable after sitting on an
353 * rpc_wait_queue, this must be called with the queue spinlock held to protect
354 * the wait queue operation.
355 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
356 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
357 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
358 * the RPC_TASK_RUNNING flag.
360 static void rpc_make_runnable(struct workqueue_struct *wq,
361 struct rpc_task *task)
363 bool need_wakeup = !rpc_test_and_set_running(task);
365 rpc_clear_queued(task);
368 if (RPC_IS_ASYNC(task)) {
369 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
370 queue_work(wq, &task->u.tk_work);
372 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
376 * Prepare for sleeping on a wait queue.
377 * By always appending tasks to the list we ensure FIFO behavior.
378 * NB: An RPC task will only receive interrupt-driven events as long
379 * as it's on a wait queue.
381 static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
382 struct rpc_task *task,
383 unsigned char queue_priority)
385 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
386 task->tk_pid, rpc_qname(q), jiffies);
388 trace_rpc_task_sleep(task, q);
390 __rpc_add_wait_queue(q, task, queue_priority);
394 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
395 struct rpc_task *task,
396 unsigned char queue_priority)
398 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
400 __rpc_do_sleep_on_priority(q, task, queue_priority);
403 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
404 struct rpc_task *task, unsigned long timeout,
405 unsigned char queue_priority)
407 if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
409 if (time_is_after_jiffies(timeout)) {
410 __rpc_do_sleep_on_priority(q, task, queue_priority);
411 __rpc_add_timer(q, task, timeout);
413 task->tk_status = -ETIMEDOUT;
416 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
418 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
419 task->tk_callback = action;
422 static bool rpc_sleep_check_activated(struct rpc_task *task)
424 /* We shouldn't ever put an inactive task to sleep */
425 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
426 task->tk_status = -EIO;
427 rpc_put_task_async(task);
433 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
434 rpc_action action, unsigned long timeout)
436 if (!rpc_sleep_check_activated(task))
439 rpc_set_tk_callback(task, action);
442 * Protect the queue operations.
445 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
446 spin_unlock(&q->lock);
448 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
450 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
453 if (!rpc_sleep_check_activated(task))
456 rpc_set_tk_callback(task, action);
458 WARN_ON_ONCE(task->tk_timeout != 0);
460 * Protect the queue operations.
463 __rpc_sleep_on_priority(q, task, task->tk_priority);
464 spin_unlock(&q->lock);
466 EXPORT_SYMBOL_GPL(rpc_sleep_on);
468 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
469 struct rpc_task *task, unsigned long timeout, int priority)
471 if (!rpc_sleep_check_activated(task))
474 priority -= RPC_PRIORITY_LOW;
476 * Protect the queue operations.
479 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
480 spin_unlock(&q->lock);
482 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
484 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
487 if (!rpc_sleep_check_activated(task))
490 WARN_ON_ONCE(task->tk_timeout != 0);
491 priority -= RPC_PRIORITY_LOW;
493 * Protect the queue operations.
496 __rpc_sleep_on_priority(q, task, priority);
497 spin_unlock(&q->lock);
499 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
502 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
503 * @wq: workqueue on which to run task
505 * @task: task to be woken up
507 * Caller must hold queue->lock, and have cleared the task queued flag.
509 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
510 struct rpc_wait_queue *queue,
511 struct rpc_task *task)
513 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
514 task->tk_pid, jiffies);
516 /* Has the task been executed yet? If not, we cannot wake it up! */
517 if (!RPC_IS_ACTIVATED(task)) {
518 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
522 trace_rpc_task_wakeup(task, queue);
524 __rpc_remove_wait_queue(queue, task);
526 rpc_make_runnable(wq, task);
528 dprintk("RPC: __rpc_wake_up_task done\n");
532 * Wake up a queued task while the queue lock is being held
534 static struct rpc_task *
535 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
536 struct rpc_wait_queue *queue, struct rpc_task *task,
537 bool (*action)(struct rpc_task *, void *), void *data)
539 if (RPC_IS_QUEUED(task)) {
541 if (task->tk_waitqueue == queue) {
542 if (action == NULL || action(task, data)) {
543 __rpc_do_wake_up_task_on_wq(wq, queue, task);
552 * Wake up a queued task while the queue lock is being held
554 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
555 struct rpc_task *task)
557 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
562 * Wake up a task on a specific queue
564 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
566 if (!RPC_IS_QUEUED(task))
568 spin_lock(&queue->lock);
569 rpc_wake_up_task_queue_locked(queue, task);
570 spin_unlock(&queue->lock);
572 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
574 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
576 task->tk_status = *(int *)status;
581 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
582 struct rpc_task *task, int status)
584 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
585 task, rpc_task_action_set_status, &status);
589 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
590 * @queue: pointer to rpc_wait_queue
591 * @task: pointer to rpc_task
592 * @status: integer error value
594 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
595 * set to the value of @status.
598 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
599 struct rpc_task *task, int status)
601 if (!RPC_IS_QUEUED(task))
603 spin_lock(&queue->lock);
604 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
605 spin_unlock(&queue->lock);
609 * Wake up the next task on a priority queue.
611 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
614 struct rpc_task *task;
617 * Service the privileged queue.
619 q = &queue->tasks[RPC_NR_PRIORITY - 1];
620 if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
621 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
626 * Service a batch of tasks from a single owner.
628 q = &queue->tasks[queue->priority];
629 if (!list_empty(q) && queue->nr) {
631 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
636 * Service the next queue.
639 if (q == &queue->tasks[0])
640 q = &queue->tasks[queue->maxpriority];
643 if (!list_empty(q)) {
644 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
647 } while (q != &queue->tasks[queue->priority]);
649 rpc_reset_waitqueue_priority(queue);
653 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
658 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
660 if (RPC_IS_PRIORITY(queue))
661 return __rpc_find_next_queued_priority(queue);
662 if (!list_empty(&queue->tasks[0]))
663 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
668 * Wake up the first task on the wait queue.
670 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
671 struct rpc_wait_queue *queue,
672 bool (*func)(struct rpc_task *, void *), void *data)
674 struct rpc_task *task = NULL;
676 dprintk("RPC: wake_up_first(%p \"%s\")\n",
677 queue, rpc_qname(queue));
678 spin_lock(&queue->lock);
679 task = __rpc_find_next_queued(queue);
681 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
683 spin_unlock(&queue->lock);
689 * Wake up the first task on the wait queue.
691 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
692 bool (*func)(struct rpc_task *, void *), void *data)
694 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
696 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
698 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
704 * Wake up the next task on the wait queue.
706 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
708 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
710 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
713 * rpc_wake_up_locked - wake up all rpc_tasks
714 * @queue: rpc_wait_queue on which the tasks are sleeping
717 static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
719 struct rpc_task *task;
722 task = __rpc_find_next_queued(queue);
725 rpc_wake_up_task_queue_locked(queue, task);
730 * rpc_wake_up - wake up all rpc_tasks
731 * @queue: rpc_wait_queue on which the tasks are sleeping
735 void rpc_wake_up(struct rpc_wait_queue *queue)
737 spin_lock(&queue->lock);
738 rpc_wake_up_locked(queue);
739 spin_unlock(&queue->lock);
741 EXPORT_SYMBOL_GPL(rpc_wake_up);
744 * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
745 * @queue: rpc_wait_queue on which the tasks are sleeping
746 * @status: status value to set
748 static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
750 struct rpc_task *task;
753 task = __rpc_find_next_queued(queue);
756 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
761 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
762 * @queue: rpc_wait_queue on which the tasks are sleeping
763 * @status: status value to set
767 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
769 spin_lock(&queue->lock);
770 rpc_wake_up_status_locked(queue, status);
771 spin_unlock(&queue->lock);
773 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
775 static void __rpc_queue_timer_fn(struct work_struct *work)
777 struct rpc_wait_queue *queue = container_of(work,
778 struct rpc_wait_queue,
779 timer_list.dwork.work);
780 struct rpc_task *task, *n;
781 unsigned long expires, now, timeo;
783 spin_lock(&queue->lock);
784 expires = now = jiffies;
785 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
786 timeo = task->tk_timeout;
787 if (time_after_eq(now, timeo)) {
788 dprintk("RPC: %5u timeout\n", task->tk_pid);
789 task->tk_status = -ETIMEDOUT;
790 rpc_wake_up_task_queue_locked(queue, task);
793 if (expires == now || time_after(expires, timeo))
796 if (!list_empty(&queue->timer_list.list))
797 rpc_set_queue_timer(queue, expires);
798 spin_unlock(&queue->lock);
801 static void __rpc_atrun(struct rpc_task *task)
803 if (task->tk_status == -ETIMEDOUT)
808 * Run a task at a later time
810 void rpc_delay(struct rpc_task *task, unsigned long delay)
812 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
814 EXPORT_SYMBOL_GPL(rpc_delay);
817 * Helper to call task->tk_ops->rpc_call_prepare
819 void rpc_prepare_task(struct rpc_task *task)
821 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
825 rpc_init_task_statistics(struct rpc_task *task)
827 /* Initialize retry counters */
828 task->tk_garb_retry = 2;
829 task->tk_cred_retry = 2;
831 /* starting timestamp */
832 task->tk_start = ktime_get();
836 rpc_reset_task_statistics(struct rpc_task *task)
838 task->tk_timeouts = 0;
839 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
840 rpc_init_task_statistics(task);
844 * Helper that calls task->tk_ops->rpc_call_done if it exists
846 void rpc_exit_task(struct rpc_task *task)
848 trace_rpc_task_end(task, task->tk_action);
849 task->tk_action = NULL;
850 if (task->tk_ops->rpc_count_stats)
851 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
852 else if (task->tk_client)
853 rpc_count_iostats(task, task->tk_client->cl_metrics);
854 if (task->tk_ops->rpc_call_done != NULL) {
855 task->tk_ops->rpc_call_done(task, task->tk_calldata);
856 if (task->tk_action != NULL) {
857 /* Always release the RPC slot and buffer memory */
859 rpc_reset_task_statistics(task);
864 void rpc_signal_task(struct rpc_task *task)
866 struct rpc_wait_queue *queue;
868 if (!RPC_IS_ACTIVATED(task))
870 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
871 smp_mb__after_atomic();
872 queue = READ_ONCE(task->tk_waitqueue);
874 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
877 void rpc_exit(struct rpc_task *task, int status)
879 task->tk_status = status;
880 task->tk_action = rpc_exit_task;
881 rpc_wake_up_queued_task(task->tk_waitqueue, task);
883 EXPORT_SYMBOL_GPL(rpc_exit);
885 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
887 if (ops->rpc_release != NULL)
888 ops->rpc_release(calldata);
892 * This is the RPC `scheduler' (or rather, the finite state machine).
894 static void __rpc_execute(struct rpc_task *task)
896 struct rpc_wait_queue *queue;
897 int task_is_async = RPC_IS_ASYNC(task);
900 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
901 task->tk_pid, task->tk_flags);
903 WARN_ON_ONCE(RPC_IS_QUEUED(task));
904 if (RPC_IS_QUEUED(task))
908 void (*do_action)(struct rpc_task *);
911 * Perform the next FSM step or a pending callback.
913 * tk_action may be NULL if the task has been killed.
914 * In particular, note that rpc_killall_tasks may
915 * do this at any time, so beware when dereferencing.
917 do_action = task->tk_action;
918 if (task->tk_callback) {
919 do_action = task->tk_callback;
920 task->tk_callback = NULL;
924 trace_rpc_task_run_action(task, do_action);
928 * Lockless check for whether task is sleeping or not.
930 if (!RPC_IS_QUEUED(task))
934 * Signalled tasks should exit rather than sleep.
936 if (RPC_SIGNALLED(task)) {
937 task->tk_rpc_status = -ERESTARTSYS;
938 rpc_exit(task, -ERESTARTSYS);
942 * The queue->lock protects against races with
943 * rpc_make_runnable().
945 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
946 * rpc_task, rpc_make_runnable() can assign it to a
947 * different workqueue. We therefore cannot assume that the
948 * rpc_task pointer may still be dereferenced.
950 queue = task->tk_waitqueue;
951 spin_lock(&queue->lock);
952 if (!RPC_IS_QUEUED(task)) {
953 spin_unlock(&queue->lock);
956 rpc_clear_running(task);
957 spin_unlock(&queue->lock);
961 /* sync task: sleep here */
962 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
963 status = out_of_line_wait_on_bit(&task->tk_runstate,
964 RPC_TASK_QUEUED, rpc_wait_bit_killable,
968 * When a sync task receives a signal, it exits with
969 * -ERESTARTSYS. In order to catch any callbacks that
970 * clean up after sleeping on some queue, we don't
971 * break the loop here, but go around once more.
973 dprintk("RPC: %5u got signal\n", task->tk_pid);
974 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
975 task->tk_rpc_status = -ERESTARTSYS;
976 rpc_exit(task, -ERESTARTSYS);
978 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
981 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
983 /* Release all resources associated with the task */
984 rpc_release_task(task);
988 * User-visible entry point to the scheduler.
990 * This may be called recursively if e.g. an async NFS task updates
991 * the attributes and finds that dirty pages must be flushed.
992 * NOTE: Upon exit of this function the task is guaranteed to be
993 * released. In particular note that tk_release() will have
994 * been called, so your task memory may have been freed.
996 void rpc_execute(struct rpc_task *task)
998 bool is_async = RPC_IS_ASYNC(task);
1000 rpc_set_active(task);
1001 rpc_make_runnable(rpciod_workqueue, task);
1003 unsigned int pflags = memalloc_nofs_save();
1004 __rpc_execute(task);
1005 memalloc_nofs_restore(pflags);
1009 static void rpc_async_schedule(struct work_struct *work)
1011 unsigned int pflags = memalloc_nofs_save();
1013 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
1014 memalloc_nofs_restore(pflags);
1018 * rpc_malloc - allocate RPC buffer resources
1021 * A single memory region is allocated, which is split between the
1022 * RPC call and RPC reply that this task is being used for. When
1023 * this RPC is retired, the memory is released by calling rpc_free.
1025 * To prevent rpciod from hanging, this allocator never sleeps,
1026 * returning -ENOMEM and suppressing warning if the request cannot
1027 * be serviced immediately. The caller can arrange to sleep in a
1028 * way that is safe for rpciod.
1030 * Most requests are 'small' (under 2KiB) and can be serviced from a
1031 * mempool, ensuring that NFS reads and writes can always proceed,
1032 * and that there is good locality of reference for these buffers.
1034 int rpc_malloc(struct rpc_task *task)
1036 struct rpc_rqst *rqst = task->tk_rqstp;
1037 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1038 struct rpc_buffer *buf;
1039 gfp_t gfp = GFP_NOFS;
1041 if (RPC_IS_ASYNC(task))
1042 gfp = GFP_NOWAIT | __GFP_NOWARN;
1043 if (RPC_IS_SWAPPER(task))
1044 gfp |= __GFP_MEMALLOC;
1046 size += sizeof(struct rpc_buffer);
1047 if (size <= RPC_BUFFER_MAXSIZE)
1048 buf = mempool_alloc(rpc_buffer_mempool, gfp);
1050 buf = kmalloc(size, gfp);
1056 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1057 task->tk_pid, size, buf);
1058 rqst->rq_buffer = buf->data;
1059 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1062 EXPORT_SYMBOL_GPL(rpc_malloc);
1065 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1069 void rpc_free(struct rpc_task *task)
1071 void *buffer = task->tk_rqstp->rq_buffer;
1073 struct rpc_buffer *buf;
1075 buf = container_of(buffer, struct rpc_buffer, data);
1078 dprintk("RPC: freeing buffer of size %zu at %p\n",
1081 if (size <= RPC_BUFFER_MAXSIZE)
1082 mempool_free(buf, rpc_buffer_mempool);
1086 EXPORT_SYMBOL_GPL(rpc_free);
1089 * Creation and deletion of RPC task structures
1091 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1093 memset(task, 0, sizeof(*task));
1094 atomic_set(&task->tk_count, 1);
1095 task->tk_flags = task_setup_data->flags;
1096 task->tk_ops = task_setup_data->callback_ops;
1097 task->tk_calldata = task_setup_data->callback_data;
1098 INIT_LIST_HEAD(&task->tk_task);
1100 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1101 task->tk_owner = current->tgid;
1103 /* Initialize workqueue for async tasks */
1104 task->tk_workqueue = task_setup_data->workqueue;
1106 task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1107 xprt_get(task_setup_data->rpc_xprt));
1109 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1111 if (task->tk_ops->rpc_call_prepare != NULL)
1112 task->tk_action = rpc_prepare_task;
1114 rpc_init_task_statistics(task);
1116 dprintk("RPC: new task initialized, procpid %u\n",
1117 task_pid_nr(current));
1120 static struct rpc_task *
1121 rpc_alloc_task(void)
1123 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1127 * Create a new task for the specified client.
1129 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1131 struct rpc_task *task = setup_data->task;
1132 unsigned short flags = 0;
1135 task = rpc_alloc_task();
1136 flags = RPC_TASK_DYNAMIC;
1139 rpc_init_task(task, setup_data);
1140 task->tk_flags |= flags;
1141 dprintk("RPC: allocated task %p\n", task);
1146 * rpc_free_task - release rpc task and perform cleanups
1148 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1149 * in order to work around a workqueue dependency issue.
1152 * "Workqueue currently considers two work items to be the same if they're
1153 * on the same address and won't execute them concurrently - ie. it
1154 * makes a work item which is queued again while being executed wait
1155 * for the previous execution to complete.
1157 * If a work function frees the work item, and then waits for an event
1158 * which should be performed by another work item and *that* work item
1159 * recycles the freed work item, it can create a false dependency loop.
1160 * There really is no reliable way to detect this short of verifying
1161 * every memory free."
1164 static void rpc_free_task(struct rpc_task *task)
1166 unsigned short tk_flags = task->tk_flags;
1168 put_rpccred(task->tk_op_cred);
1169 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1171 if (tk_flags & RPC_TASK_DYNAMIC) {
1172 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1173 mempool_free(task, rpc_task_mempool);
1177 static void rpc_async_release(struct work_struct *work)
1179 unsigned int pflags = memalloc_nofs_save();
1181 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1182 memalloc_nofs_restore(pflags);
1185 static void rpc_release_resources_task(struct rpc_task *task)
1188 if (task->tk_msg.rpc_cred) {
1189 put_cred(task->tk_msg.rpc_cred);
1190 task->tk_msg.rpc_cred = NULL;
1192 rpc_task_release_client(task);
1195 static void rpc_final_put_task(struct rpc_task *task,
1196 struct workqueue_struct *q)
1199 INIT_WORK(&task->u.tk_work, rpc_async_release);
1200 queue_work(q, &task->u.tk_work);
1202 rpc_free_task(task);
1205 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1207 if (atomic_dec_and_test(&task->tk_count)) {
1208 rpc_release_resources_task(task);
1209 rpc_final_put_task(task, q);
1213 void rpc_put_task(struct rpc_task *task)
1215 rpc_do_put_task(task, NULL);
1217 EXPORT_SYMBOL_GPL(rpc_put_task);
1219 void rpc_put_task_async(struct rpc_task *task)
1221 rpc_do_put_task(task, task->tk_workqueue);
1223 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1225 static void rpc_release_task(struct rpc_task *task)
1227 dprintk("RPC: %5u release task\n", task->tk_pid);
1229 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1231 rpc_release_resources_task(task);
1234 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1235 * so it should be safe to use task->tk_count as a test for whether
1236 * or not any other processes still hold references to our rpc_task.
1238 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1239 /* Wake up anyone who may be waiting for task completion */
1240 if (!rpc_complete_task(task))
1243 if (!atomic_dec_and_test(&task->tk_count))
1246 rpc_final_put_task(task, task->tk_workqueue);
1251 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1254 void rpciod_down(void)
1256 module_put(THIS_MODULE);
1260 * Start up the rpciod workqueue.
1262 static int rpciod_start(void)
1264 struct workqueue_struct *wq;
1267 * Create the rpciod thread and wait for it to start.
1269 dprintk("RPC: creating workqueue rpciod\n");
1270 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1273 rpciod_workqueue = wq;
1274 /* Note: highpri because network receive is latency sensitive */
1275 wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1278 xprtiod_workqueue = wq;
1281 wq = rpciod_workqueue;
1282 rpciod_workqueue = NULL;
1283 destroy_workqueue(wq);
1288 static void rpciod_stop(void)
1290 struct workqueue_struct *wq = NULL;
1292 if (rpciod_workqueue == NULL)
1294 dprintk("RPC: destroying workqueue rpciod\n");
1296 wq = rpciod_workqueue;
1297 rpciod_workqueue = NULL;
1298 destroy_workqueue(wq);
1299 wq = xprtiod_workqueue;
1300 xprtiod_workqueue = NULL;
1301 destroy_workqueue(wq);
1305 rpc_destroy_mempool(void)
1308 mempool_destroy(rpc_buffer_mempool);
1309 mempool_destroy(rpc_task_mempool);
1310 kmem_cache_destroy(rpc_task_slabp);
1311 kmem_cache_destroy(rpc_buffer_slabp);
1312 rpc_destroy_wait_queue(&delay_queue);
1316 rpc_init_mempool(void)
1319 * The following is not strictly a mempool initialisation,
1320 * but there is no harm in doing it here
1322 rpc_init_wait_queue(&delay_queue, "delayq");
1323 if (!rpciod_start())
1326 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1327 sizeof(struct rpc_task),
1328 0, SLAB_HWCACHE_ALIGN,
1330 if (!rpc_task_slabp)
1332 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1334 0, SLAB_HWCACHE_ALIGN,
1336 if (!rpc_buffer_slabp)
1338 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1340 if (!rpc_task_mempool)
1342 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1344 if (!rpc_buffer_mempool)
1348 rpc_destroy_mempool();