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>
29 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
30 #define RPCDBG_FACILITY RPCDBG_SCHED
33 #define CREATE_TRACE_POINTS
34 #include <trace/events/sunrpc.h>
37 * RPC slabs and memory pools
39 #define RPC_BUFFER_MAXSIZE (2048)
40 #define RPC_BUFFER_POOLSIZE (8)
41 #define RPC_TASK_POOLSIZE (8)
42 static struct kmem_cache *rpc_task_slabp __read_mostly;
43 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
44 static mempool_t *rpc_task_mempool __read_mostly;
45 static mempool_t *rpc_buffer_mempool __read_mostly;
47 static void rpc_async_schedule(struct work_struct *);
48 static void rpc_release_task(struct rpc_task *task);
49 static void __rpc_queue_timer_fn(struct timer_list *t);
52 * RPC tasks sit here while waiting for conditions to improve.
54 static struct rpc_wait_queue delay_queue;
57 * rpciod-related stuff
59 struct workqueue_struct *rpciod_workqueue __read_mostly;
60 struct workqueue_struct *xprtiod_workqueue __read_mostly;
63 rpc_task_timeout(const struct rpc_task *task)
65 unsigned long timeout = READ_ONCE(task->tk_timeout);
68 unsigned long now = jiffies;
69 if (time_before(now, timeout))
74 EXPORT_SYMBOL_GPL(rpc_task_timeout);
77 * Disable the timer for a given RPC task. Should be called with
78 * queue->lock and bh_disabled in order to avoid races within
82 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
84 if (list_empty(&task->u.tk_wait.timer_list))
86 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
88 list_del(&task->u.tk_wait.timer_list);
89 if (list_empty(&queue->timer_list.list))
90 del_timer(&queue->timer_list.timer);
94 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
96 timer_reduce(&queue->timer_list.timer, expires);
100 * Set up a timer for the current task.
103 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
104 unsigned long timeout)
106 dprintk("RPC: %5u setting alarm for %u ms\n",
107 task->tk_pid, jiffies_to_msecs(timeout - jiffies));
109 task->tk_timeout = timeout;
110 rpc_set_queue_timer(queue, timeout);
111 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
114 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
116 if (queue->priority != priority) {
117 queue->priority = priority;
118 queue->nr = 1U << priority;
122 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
124 rpc_set_waitqueue_priority(queue, queue->maxpriority);
128 * Add a request to a queue list
131 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
135 list_for_each_entry(t, q, u.tk_wait.list) {
136 if (t->tk_owner == task->tk_owner) {
137 list_add_tail(&task->u.tk_wait.links,
138 &t->u.tk_wait.links);
139 /* Cache the queue head in task->u.tk_wait.list */
140 task->u.tk_wait.list.next = q;
141 task->u.tk_wait.list.prev = NULL;
145 INIT_LIST_HEAD(&task->u.tk_wait.links);
146 list_add_tail(&task->u.tk_wait.list, q);
150 * Remove request from a queue list
153 __rpc_list_dequeue_task(struct rpc_task *task)
158 if (task->u.tk_wait.list.prev == NULL) {
159 list_del(&task->u.tk_wait.links);
162 if (!list_empty(&task->u.tk_wait.links)) {
163 t = list_first_entry(&task->u.tk_wait.links,
166 /* Assume __rpc_list_enqueue_task() cached the queue head */
167 q = t->u.tk_wait.list.next;
168 list_add_tail(&t->u.tk_wait.list, q);
169 list_del(&task->u.tk_wait.links);
171 list_del(&task->u.tk_wait.list);
175 * Add new request to a priority queue.
177 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
178 struct rpc_task *task,
179 unsigned char queue_priority)
181 if (unlikely(queue_priority > queue->maxpriority))
182 queue_priority = queue->maxpriority;
183 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
187 * Add new request to wait queue.
189 * Swapper tasks always get inserted at the head of the queue.
190 * This should avoid many nasty memory deadlocks and hopefully
191 * improve overall performance.
192 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
194 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
195 struct rpc_task *task,
196 unsigned char queue_priority)
198 WARN_ON_ONCE(RPC_IS_QUEUED(task));
199 if (RPC_IS_QUEUED(task))
202 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
203 if (RPC_IS_PRIORITY(queue))
204 __rpc_add_wait_queue_priority(queue, task, queue_priority);
205 else if (RPC_IS_SWAPPER(task))
206 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
208 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
209 task->tk_waitqueue = queue;
211 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
213 rpc_set_queued(task);
215 dprintk("RPC: %5u added to queue %p \"%s\"\n",
216 task->tk_pid, queue, rpc_qname(queue));
220 * Remove request from a priority queue.
222 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
224 __rpc_list_dequeue_task(task);
228 * Remove request from queue.
229 * Note: must be called with spin lock held.
231 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
233 __rpc_disable_timer(queue, task);
234 if (RPC_IS_PRIORITY(queue))
235 __rpc_remove_wait_queue_priority(task);
237 list_del(&task->u.tk_wait.list);
239 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
240 task->tk_pid, queue, rpc_qname(queue));
243 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
247 spin_lock_init(&queue->lock);
248 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
249 INIT_LIST_HEAD(&queue->tasks[i]);
250 queue->maxpriority = nr_queues - 1;
251 rpc_reset_waitqueue_priority(queue);
253 timer_setup(&queue->timer_list.timer,
254 __rpc_queue_timer_fn,
256 INIT_LIST_HEAD(&queue->timer_list.list);
257 rpc_assign_waitqueue_name(queue, qname);
260 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
262 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
264 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
266 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
268 __rpc_init_priority_wait_queue(queue, qname, 1);
270 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
272 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
274 del_timer_sync(&queue->timer_list.timer);
276 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
278 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
280 freezable_schedule_unsafe();
281 if (signal_pending_state(mode, current))
286 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
287 static void rpc_task_set_debuginfo(struct rpc_task *task)
289 static atomic_t rpc_pid;
291 task->tk_pid = atomic_inc_return(&rpc_pid);
294 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
299 static void rpc_set_active(struct rpc_task *task)
301 rpc_task_set_debuginfo(task);
302 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
303 trace_rpc_task_begin(task, NULL);
307 * Mark an RPC call as having completed by clearing the 'active' bit
308 * and then waking up all tasks that were sleeping.
310 static int rpc_complete_task(struct rpc_task *task)
312 void *m = &task->tk_runstate;
313 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
314 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
318 trace_rpc_task_complete(task, NULL);
320 spin_lock_irqsave(&wq->lock, flags);
321 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
322 ret = atomic_dec_and_test(&task->tk_count);
323 if (waitqueue_active(wq))
324 __wake_up_locked_key(wq, TASK_NORMAL, &k);
325 spin_unlock_irqrestore(&wq->lock, flags);
330 * Allow callers to wait for completion of an RPC call
332 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
333 * to enforce taking of the wq->lock and hence avoid races with
334 * rpc_complete_task().
336 int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
339 action = rpc_wait_bit_killable;
340 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
341 action, TASK_KILLABLE);
343 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
346 * Make an RPC task runnable.
348 * Note: If the task is ASYNC, and is being made runnable after sitting on an
349 * rpc_wait_queue, this must be called with the queue spinlock held to protect
350 * the wait queue operation.
351 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
352 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
353 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
354 * the RPC_TASK_RUNNING flag.
356 static void rpc_make_runnable(struct workqueue_struct *wq,
357 struct rpc_task *task)
359 bool need_wakeup = !rpc_test_and_set_running(task);
361 rpc_clear_queued(task);
364 if (RPC_IS_ASYNC(task)) {
365 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
366 queue_work(wq, &task->u.tk_work);
368 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
372 * Prepare for sleeping on a wait queue.
373 * By always appending tasks to the list we ensure FIFO behavior.
374 * NB: An RPC task will only receive interrupt-driven events as long
375 * as it's on a wait queue.
377 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
378 struct rpc_task *task,
379 unsigned char queue_priority)
381 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
382 task->tk_pid, rpc_qname(q), jiffies);
384 trace_rpc_task_sleep(task, q);
386 __rpc_add_wait_queue(q, task, queue_priority);
390 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
391 struct rpc_task *task, unsigned long timeout,
392 unsigned char queue_priority)
394 if (time_is_after_jiffies(timeout)) {
395 __rpc_sleep_on_priority(q, task, queue_priority);
396 __rpc_add_timer(q, task, timeout);
398 task->tk_status = -ETIMEDOUT;
401 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
403 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
404 task->tk_callback = action;
407 static bool rpc_sleep_check_activated(struct rpc_task *task)
409 /* We shouldn't ever put an inactive task to sleep */
410 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
411 task->tk_status = -EIO;
412 rpc_put_task_async(task);
418 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
419 rpc_action action, unsigned long timeout)
421 if (!rpc_sleep_check_activated(task))
424 rpc_set_tk_callback(task, action);
427 * Protect the queue operations.
429 spin_lock_bh(&q->lock);
430 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
431 spin_unlock_bh(&q->lock);
433 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
435 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
438 if (!rpc_sleep_check_activated(task))
441 rpc_set_tk_callback(task, action);
443 WARN_ON_ONCE(task->tk_timeout != 0);
445 * Protect the queue operations.
447 spin_lock_bh(&q->lock);
448 __rpc_sleep_on_priority(q, task, task->tk_priority);
449 spin_unlock_bh(&q->lock);
451 EXPORT_SYMBOL_GPL(rpc_sleep_on);
453 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
454 struct rpc_task *task, unsigned long timeout, int priority)
456 if (!rpc_sleep_check_activated(task))
459 priority -= RPC_PRIORITY_LOW;
461 * Protect the queue operations.
463 spin_lock_bh(&q->lock);
464 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
465 spin_unlock_bh(&q->lock);
467 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
469 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
472 if (!rpc_sleep_check_activated(task))
475 WARN_ON_ONCE(task->tk_timeout != 0);
476 priority -= RPC_PRIORITY_LOW;
478 * Protect the queue operations.
480 spin_lock_bh(&q->lock);
481 __rpc_sleep_on_priority(q, task, priority);
482 spin_unlock_bh(&q->lock);
484 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
487 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
488 * @wq: workqueue on which to run task
490 * @task: task to be woken up
492 * Caller must hold queue->lock, and have cleared the task queued flag.
494 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
495 struct rpc_wait_queue *queue,
496 struct rpc_task *task)
498 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
499 task->tk_pid, jiffies);
501 /* Has the task been executed yet? If not, we cannot wake it up! */
502 if (!RPC_IS_ACTIVATED(task)) {
503 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
507 trace_rpc_task_wakeup(task, queue);
509 __rpc_remove_wait_queue(queue, task);
511 rpc_make_runnable(wq, task);
513 dprintk("RPC: __rpc_wake_up_task done\n");
517 * Wake up a queued task while the queue lock is being held
519 static struct rpc_task *
520 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
521 struct rpc_wait_queue *queue, struct rpc_task *task,
522 bool (*action)(struct rpc_task *, void *), void *data)
524 if (RPC_IS_QUEUED(task)) {
526 if (task->tk_waitqueue == queue) {
527 if (action == NULL || action(task, data)) {
528 __rpc_do_wake_up_task_on_wq(wq, queue, task);
537 rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq,
538 struct rpc_wait_queue *queue, struct rpc_task *task)
540 rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL);
544 * Wake up a queued task while the queue lock is being held
546 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
548 rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task);
552 * Wake up a task on a specific queue
554 void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq,
555 struct rpc_wait_queue *queue,
556 struct rpc_task *task)
558 if (!RPC_IS_QUEUED(task))
560 spin_lock_bh(&queue->lock);
561 rpc_wake_up_task_on_wq_queue_locked(wq, queue, task);
562 spin_unlock_bh(&queue->lock);
566 * Wake up a task on a specific queue
568 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
570 if (!RPC_IS_QUEUED(task))
572 spin_lock_bh(&queue->lock);
573 rpc_wake_up_task_queue_locked(queue, task);
574 spin_unlock_bh(&queue->lock);
576 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
578 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
580 task->tk_status = *(int *)status;
585 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
586 struct rpc_task *task, int status)
588 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
589 task, rpc_task_action_set_status, &status);
593 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
594 * @queue: pointer to rpc_wait_queue
595 * @task: pointer to rpc_task
596 * @status: integer error value
598 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
599 * set to the value of @status.
602 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
603 struct rpc_task *task, int status)
605 if (!RPC_IS_QUEUED(task))
607 spin_lock_bh(&queue->lock);
608 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
609 spin_unlock_bh(&queue->lock);
613 * Wake up the next task on a priority queue.
615 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
618 struct rpc_task *task;
621 * Service a batch of tasks from a single owner.
623 q = &queue->tasks[queue->priority];
624 if (!list_empty(q) && --queue->nr) {
625 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
630 * Service the next queue.
633 if (q == &queue->tasks[0])
634 q = &queue->tasks[queue->maxpriority];
637 if (!list_empty(q)) {
638 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
641 } while (q != &queue->tasks[queue->priority]);
643 rpc_reset_waitqueue_priority(queue);
647 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
652 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
654 if (RPC_IS_PRIORITY(queue))
655 return __rpc_find_next_queued_priority(queue);
656 if (!list_empty(&queue->tasks[0]))
657 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
662 * Wake up the first task on the wait queue.
664 struct 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)
668 struct rpc_task *task = NULL;
670 dprintk("RPC: wake_up_first(%p \"%s\")\n",
671 queue, rpc_qname(queue));
672 spin_lock_bh(&queue->lock);
673 task = __rpc_find_next_queued(queue);
675 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
677 spin_unlock_bh(&queue->lock);
683 * Wake up the first task on the wait queue.
685 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
686 bool (*func)(struct rpc_task *, void *), void *data)
688 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
690 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
692 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
698 * Wake up the next task on the wait queue.
700 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
702 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
704 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
707 * rpc_wake_up - wake up all rpc_tasks
708 * @queue: rpc_wait_queue on which the tasks are sleeping
712 void rpc_wake_up(struct rpc_wait_queue *queue)
714 struct list_head *head;
716 spin_lock_bh(&queue->lock);
717 head = &queue->tasks[queue->maxpriority];
719 while (!list_empty(head)) {
720 struct rpc_task *task;
721 task = list_first_entry(head,
724 rpc_wake_up_task_queue_locked(queue, task);
726 if (head == &queue->tasks[0])
730 spin_unlock_bh(&queue->lock);
732 EXPORT_SYMBOL_GPL(rpc_wake_up);
735 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
736 * @queue: rpc_wait_queue on which the tasks are sleeping
737 * @status: status value to set
741 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
743 struct list_head *head;
745 spin_lock_bh(&queue->lock);
746 head = &queue->tasks[queue->maxpriority];
748 while (!list_empty(head)) {
749 struct rpc_task *task;
750 task = list_first_entry(head,
753 task->tk_status = status;
754 rpc_wake_up_task_queue_locked(queue, task);
756 if (head == &queue->tasks[0])
760 spin_unlock_bh(&queue->lock);
762 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
764 static void __rpc_queue_timer_fn(struct timer_list *t)
766 struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer);
767 struct rpc_task *task, *n;
768 unsigned long expires, now, timeo;
770 spin_lock(&queue->lock);
771 expires = now = jiffies;
772 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
773 timeo = task->tk_timeout;
774 if (time_after_eq(now, timeo)) {
775 dprintk("RPC: %5u timeout\n", task->tk_pid);
776 task->tk_status = -ETIMEDOUT;
777 rpc_wake_up_task_queue_locked(queue, task);
780 if (expires == now || time_after(expires, timeo))
783 if (!list_empty(&queue->timer_list.list))
784 rpc_set_queue_timer(queue, expires);
785 spin_unlock(&queue->lock);
788 static void __rpc_atrun(struct rpc_task *task)
790 if (task->tk_status == -ETIMEDOUT)
795 * Run a task at a later time
797 void rpc_delay(struct rpc_task *task, unsigned long delay)
799 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
801 EXPORT_SYMBOL_GPL(rpc_delay);
804 * Helper to call task->tk_ops->rpc_call_prepare
806 void rpc_prepare_task(struct rpc_task *task)
808 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
812 rpc_init_task_statistics(struct rpc_task *task)
814 /* Initialize retry counters */
815 task->tk_garb_retry = 2;
816 task->tk_cred_retry = 2;
817 task->tk_rebind_retry = 2;
819 /* starting timestamp */
820 task->tk_start = ktime_get();
824 rpc_reset_task_statistics(struct rpc_task *task)
826 task->tk_timeouts = 0;
827 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
828 rpc_init_task_statistics(task);
832 * Helper that calls task->tk_ops->rpc_call_done if it exists
834 void rpc_exit_task(struct rpc_task *task)
836 task->tk_action = NULL;
837 if (task->tk_ops->rpc_call_done != NULL) {
838 task->tk_ops->rpc_call_done(task, task->tk_calldata);
839 if (task->tk_action != NULL) {
840 /* Always release the RPC slot and buffer memory */
842 rpc_reset_task_statistics(task);
847 void rpc_signal_task(struct rpc_task *task)
849 struct rpc_wait_queue *queue;
851 if (!RPC_IS_ACTIVATED(task))
853 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
854 smp_mb__after_atomic();
855 queue = READ_ONCE(task->tk_waitqueue);
857 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
860 void rpc_exit(struct rpc_task *task, int status)
862 task->tk_status = status;
863 task->tk_action = rpc_exit_task;
864 rpc_wake_up_queued_task(task->tk_waitqueue, task);
866 EXPORT_SYMBOL_GPL(rpc_exit);
868 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
870 if (ops->rpc_release != NULL)
871 ops->rpc_release(calldata);
875 * This is the RPC `scheduler' (or rather, the finite state machine).
877 static void __rpc_execute(struct rpc_task *task)
879 struct rpc_wait_queue *queue;
880 int task_is_async = RPC_IS_ASYNC(task);
883 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
884 task->tk_pid, task->tk_flags);
886 WARN_ON_ONCE(RPC_IS_QUEUED(task));
887 if (RPC_IS_QUEUED(task))
891 void (*do_action)(struct rpc_task *);
894 * Perform the next FSM step or a pending callback.
896 * tk_action may be NULL if the task has been killed.
897 * In particular, note that rpc_killall_tasks may
898 * do this at any time, so beware when dereferencing.
900 do_action = task->tk_action;
901 if (task->tk_callback) {
902 do_action = task->tk_callback;
903 task->tk_callback = NULL;
907 trace_rpc_task_run_action(task, do_action);
911 * Lockless check for whether task is sleeping or not.
913 if (!RPC_IS_QUEUED(task))
917 * Signalled tasks should exit rather than sleep.
919 if (RPC_SIGNALLED(task))
920 rpc_exit(task, -ERESTARTSYS);
923 * The queue->lock protects against races with
924 * rpc_make_runnable().
926 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
927 * rpc_task, rpc_make_runnable() can assign it to a
928 * different workqueue. We therefore cannot assume that the
929 * rpc_task pointer may still be dereferenced.
931 queue = task->tk_waitqueue;
932 spin_lock_bh(&queue->lock);
933 if (!RPC_IS_QUEUED(task)) {
934 spin_unlock_bh(&queue->lock);
937 rpc_clear_running(task);
938 spin_unlock_bh(&queue->lock);
942 /* sync task: sleep here */
943 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
944 status = out_of_line_wait_on_bit(&task->tk_runstate,
945 RPC_TASK_QUEUED, rpc_wait_bit_killable,
949 * When a sync task receives a signal, it exits with
950 * -ERESTARTSYS. In order to catch any callbacks that
951 * clean up after sleeping on some queue, we don't
952 * break the loop here, but go around once more.
954 dprintk("RPC: %5u got signal\n", task->tk_pid);
955 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
956 rpc_exit(task, -ERESTARTSYS);
958 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
961 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
963 /* Release all resources associated with the task */
964 rpc_release_task(task);
968 * User-visible entry point to the scheduler.
970 * This may be called recursively if e.g. an async NFS task updates
971 * the attributes and finds that dirty pages must be flushed.
972 * NOTE: Upon exit of this function the task is guaranteed to be
973 * released. In particular note that tk_release() will have
974 * been called, so your task memory may have been freed.
976 void rpc_execute(struct rpc_task *task)
978 bool is_async = RPC_IS_ASYNC(task);
980 rpc_set_active(task);
981 rpc_make_runnable(rpciod_workqueue, task);
986 static void rpc_async_schedule(struct work_struct *work)
988 unsigned int pflags = memalloc_nofs_save();
990 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
991 memalloc_nofs_restore(pflags);
995 * rpc_malloc - allocate RPC buffer resources
998 * A single memory region is allocated, which is split between the
999 * RPC call and RPC reply that this task is being used for. When
1000 * this RPC is retired, the memory is released by calling rpc_free.
1002 * To prevent rpciod from hanging, this allocator never sleeps,
1003 * returning -ENOMEM and suppressing warning if the request cannot
1004 * be serviced immediately. The caller can arrange to sleep in a
1005 * way that is safe for rpciod.
1007 * Most requests are 'small' (under 2KiB) and can be serviced from a
1008 * mempool, ensuring that NFS reads and writes can always proceed,
1009 * and that there is good locality of reference for these buffers.
1011 int rpc_malloc(struct rpc_task *task)
1013 struct rpc_rqst *rqst = task->tk_rqstp;
1014 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1015 struct rpc_buffer *buf;
1016 gfp_t gfp = GFP_NOFS;
1018 if (RPC_IS_SWAPPER(task))
1019 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
1021 size += sizeof(struct rpc_buffer);
1022 if (size <= RPC_BUFFER_MAXSIZE)
1023 buf = mempool_alloc(rpc_buffer_mempool, gfp);
1025 buf = kmalloc(size, gfp);
1031 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1032 task->tk_pid, size, buf);
1033 rqst->rq_buffer = buf->data;
1034 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1037 EXPORT_SYMBOL_GPL(rpc_malloc);
1040 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1044 void rpc_free(struct rpc_task *task)
1046 void *buffer = task->tk_rqstp->rq_buffer;
1048 struct rpc_buffer *buf;
1050 buf = container_of(buffer, struct rpc_buffer, data);
1053 dprintk("RPC: freeing buffer of size %zu at %p\n",
1056 if (size <= RPC_BUFFER_MAXSIZE)
1057 mempool_free(buf, rpc_buffer_mempool);
1061 EXPORT_SYMBOL_GPL(rpc_free);
1064 * Creation and deletion of RPC task structures
1066 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1068 memset(task, 0, sizeof(*task));
1069 atomic_set(&task->tk_count, 1);
1070 task->tk_flags = task_setup_data->flags;
1071 task->tk_ops = task_setup_data->callback_ops;
1072 task->tk_calldata = task_setup_data->callback_data;
1073 INIT_LIST_HEAD(&task->tk_task);
1075 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1076 task->tk_owner = current->tgid;
1078 /* Initialize workqueue for async tasks */
1079 task->tk_workqueue = task_setup_data->workqueue;
1081 task->tk_xprt = xprt_get(task_setup_data->rpc_xprt);
1083 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1085 if (task->tk_ops->rpc_call_prepare != NULL)
1086 task->tk_action = rpc_prepare_task;
1088 rpc_init_task_statistics(task);
1090 dprintk("RPC: new task initialized, procpid %u\n",
1091 task_pid_nr(current));
1094 static struct rpc_task *
1095 rpc_alloc_task(void)
1097 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1101 * Create a new task for the specified client.
1103 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1105 struct rpc_task *task = setup_data->task;
1106 unsigned short flags = 0;
1109 task = rpc_alloc_task();
1110 flags = RPC_TASK_DYNAMIC;
1113 rpc_init_task(task, setup_data);
1114 task->tk_flags |= flags;
1115 dprintk("RPC: allocated task %p\n", task);
1120 * rpc_free_task - release rpc task and perform cleanups
1122 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1123 * in order to work around a workqueue dependency issue.
1126 * "Workqueue currently considers two work items to be the same if they're
1127 * on the same address and won't execute them concurrently - ie. it
1128 * makes a work item which is queued again while being executed wait
1129 * for the previous execution to complete.
1131 * If a work function frees the work item, and then waits for an event
1132 * which should be performed by another work item and *that* work item
1133 * recycles the freed work item, it can create a false dependency loop.
1134 * There really is no reliable way to detect this short of verifying
1135 * every memory free."
1138 static void rpc_free_task(struct rpc_task *task)
1140 unsigned short tk_flags = task->tk_flags;
1142 put_rpccred(task->tk_op_cred);
1143 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1145 if (tk_flags & RPC_TASK_DYNAMIC) {
1146 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1147 mempool_free(task, rpc_task_mempool);
1151 static void rpc_async_release(struct work_struct *work)
1153 unsigned int pflags = memalloc_nofs_save();
1155 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1156 memalloc_nofs_restore(pflags);
1159 static void rpc_release_resources_task(struct rpc_task *task)
1162 if (task->tk_msg.rpc_cred) {
1163 put_cred(task->tk_msg.rpc_cred);
1164 task->tk_msg.rpc_cred = NULL;
1166 rpc_task_release_client(task);
1169 static void rpc_final_put_task(struct rpc_task *task,
1170 struct workqueue_struct *q)
1173 INIT_WORK(&task->u.tk_work, rpc_async_release);
1174 queue_work(q, &task->u.tk_work);
1176 rpc_free_task(task);
1179 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1181 if (atomic_dec_and_test(&task->tk_count)) {
1182 rpc_release_resources_task(task);
1183 rpc_final_put_task(task, q);
1187 void rpc_put_task(struct rpc_task *task)
1189 rpc_do_put_task(task, NULL);
1191 EXPORT_SYMBOL_GPL(rpc_put_task);
1193 void rpc_put_task_async(struct rpc_task *task)
1195 rpc_do_put_task(task, task->tk_workqueue);
1197 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1199 static void rpc_release_task(struct rpc_task *task)
1201 dprintk("RPC: %5u release task\n", task->tk_pid);
1203 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1205 rpc_release_resources_task(task);
1208 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1209 * so it should be safe to use task->tk_count as a test for whether
1210 * or not any other processes still hold references to our rpc_task.
1212 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1213 /* Wake up anyone who may be waiting for task completion */
1214 if (!rpc_complete_task(task))
1217 if (!atomic_dec_and_test(&task->tk_count))
1220 rpc_final_put_task(task, task->tk_workqueue);
1225 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1228 void rpciod_down(void)
1230 module_put(THIS_MODULE);
1234 * Start up the rpciod workqueue.
1236 static int rpciod_start(void)
1238 struct workqueue_struct *wq;
1241 * Create the rpciod thread and wait for it to start.
1243 dprintk("RPC: creating workqueue rpciod\n");
1244 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1247 rpciod_workqueue = wq;
1248 /* Note: highpri because network receive is latency sensitive */
1249 wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1252 xprtiod_workqueue = wq;
1255 wq = rpciod_workqueue;
1256 rpciod_workqueue = NULL;
1257 destroy_workqueue(wq);
1262 static void rpciod_stop(void)
1264 struct workqueue_struct *wq = NULL;
1266 if (rpciod_workqueue == NULL)
1268 dprintk("RPC: destroying workqueue rpciod\n");
1270 wq = rpciod_workqueue;
1271 rpciod_workqueue = NULL;
1272 destroy_workqueue(wq);
1273 wq = xprtiod_workqueue;
1274 xprtiod_workqueue = NULL;
1275 destroy_workqueue(wq);
1279 rpc_destroy_mempool(void)
1282 mempool_destroy(rpc_buffer_mempool);
1283 mempool_destroy(rpc_task_mempool);
1284 kmem_cache_destroy(rpc_task_slabp);
1285 kmem_cache_destroy(rpc_buffer_slabp);
1286 rpc_destroy_wait_queue(&delay_queue);
1290 rpc_init_mempool(void)
1293 * The following is not strictly a mempool initialisation,
1294 * but there is no harm in doing it here
1296 rpc_init_wait_queue(&delay_queue, "delayq");
1297 if (!rpciod_start())
1300 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1301 sizeof(struct rpc_task),
1302 0, SLAB_HWCACHE_ALIGN,
1304 if (!rpc_task_slabp)
1306 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1308 0, SLAB_HWCACHE_ALIGN,
1310 if (!rpc_buffer_slabp)
1312 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1314 if (!rpc_task_mempool)
1316 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1318 if (!rpc_buffer_mempool)
1322 rpc_destroy_mempool();