2 * linux/net/sunrpc/sched.c
4 * Scheduling for synchronous and asynchronous RPC requests.
6 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
12 #include <linux/module.h>
14 #include <linux/sched.h>
15 #include <linux/interrupt.h>
16 #include <linux/slab.h>
17 #include <linux/mempool.h>
18 #include <linux/smp.h>
19 #include <linux/spinlock.h>
20 #include <linux/mutex.h>
21 #include <linux/freezer.h>
22 #include <linux/sched/mm.h>
24 #include <linux/sunrpc/clnt.h>
28 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
29 #define RPCDBG_FACILITY RPCDBG_SCHED
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/sunrpc.h>
36 * RPC slabs and memory pools
38 #define RPC_BUFFER_MAXSIZE (2048)
39 #define RPC_BUFFER_POOLSIZE (8)
40 #define RPC_TASK_POOLSIZE (8)
41 static struct kmem_cache *rpc_task_slabp __read_mostly;
42 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
43 static mempool_t *rpc_task_mempool __read_mostly;
44 static mempool_t *rpc_buffer_mempool __read_mostly;
46 static void rpc_async_schedule(struct work_struct *);
47 static void rpc_release_task(struct rpc_task *task);
48 static void __rpc_queue_timer_fn(struct timer_list *t);
51 * RPC tasks sit here while waiting for conditions to improve.
53 static struct rpc_wait_queue delay_queue;
56 * rpciod-related stuff
58 struct workqueue_struct *rpciod_workqueue __read_mostly;
59 struct workqueue_struct *xprtiod_workqueue __read_mostly;
62 rpc_task_timeout(const struct rpc_task *task)
64 unsigned long timeout = READ_ONCE(task->tk_timeout);
67 unsigned long now = jiffies;
68 if (time_before(now, timeout))
73 EXPORT_SYMBOL_GPL(rpc_task_timeout);
76 * Disable the timer for a given RPC task. Should be called with
77 * queue->lock and bh_disabled in order to avoid races within
81 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
83 if (list_empty(&task->u.tk_wait.timer_list))
85 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
87 list_del(&task->u.tk_wait.timer_list);
88 if (list_empty(&queue->timer_list.list))
89 del_timer(&queue->timer_list.timer);
93 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
95 queue->timer_list.expires = expires;
96 mod_timer(&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 if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
111 rpc_set_queue_timer(queue, timeout);
112 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
115 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
117 if (queue->priority != priority) {
118 queue->priority = priority;
119 queue->nr = 1U << priority;
123 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
125 rpc_set_waitqueue_priority(queue, queue->maxpriority);
129 * Add a request to a queue list
132 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
136 list_for_each_entry(t, q, u.tk_wait.list) {
137 if (t->tk_owner == task->tk_owner) {
138 list_add_tail(&task->u.tk_wait.links,
139 &t->u.tk_wait.links);
140 /* Cache the queue head in task->u.tk_wait.list */
141 task->u.tk_wait.list.next = q;
142 task->u.tk_wait.list.prev = NULL;
146 INIT_LIST_HEAD(&task->u.tk_wait.links);
147 list_add_tail(&task->u.tk_wait.list, q);
151 * Remove request from a queue list
154 __rpc_list_dequeue_task(struct rpc_task *task)
159 if (task->u.tk_wait.list.prev == NULL) {
160 list_del(&task->u.tk_wait.links);
163 if (!list_empty(&task->u.tk_wait.links)) {
164 t = list_first_entry(&task->u.tk_wait.links,
167 /* Assume __rpc_list_enqueue_task() cached the queue head */
168 q = t->u.tk_wait.list.next;
169 list_add_tail(&t->u.tk_wait.list, q);
170 list_del(&task->u.tk_wait.links);
172 list_del(&task->u.tk_wait.list);
176 * Add new request to a priority queue.
178 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
179 struct rpc_task *task,
180 unsigned char queue_priority)
182 if (unlikely(queue_priority > queue->maxpriority))
183 queue_priority = queue->maxpriority;
184 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
188 * Add new request to wait queue.
190 * Swapper tasks always get inserted at the head of the queue.
191 * This should avoid many nasty memory deadlocks and hopefully
192 * improve overall performance.
193 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
195 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
196 struct rpc_task *task,
197 unsigned char queue_priority)
199 WARN_ON_ONCE(RPC_IS_QUEUED(task));
200 if (RPC_IS_QUEUED(task))
203 INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
204 if (RPC_IS_PRIORITY(queue))
205 __rpc_add_wait_queue_priority(queue, task, queue_priority);
206 else if (RPC_IS_SWAPPER(task))
207 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
209 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
210 task->tk_waitqueue = queue;
212 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
214 rpc_set_queued(task);
216 dprintk("RPC: %5u added to queue %p \"%s\"\n",
217 task->tk_pid, queue, rpc_qname(queue));
221 * Remove request from a priority queue.
223 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
225 __rpc_list_dequeue_task(task);
229 * Remove request from queue.
230 * Note: must be called with spin lock held.
232 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
234 __rpc_disable_timer(queue, task);
235 if (RPC_IS_PRIORITY(queue))
236 __rpc_remove_wait_queue_priority(task);
238 list_del(&task->u.tk_wait.list);
240 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
241 task->tk_pid, queue, rpc_qname(queue));
244 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
248 spin_lock_init(&queue->lock);
249 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
250 INIT_LIST_HEAD(&queue->tasks[i]);
251 queue->maxpriority = nr_queues - 1;
252 rpc_reset_waitqueue_priority(queue);
254 timer_setup(&queue->timer_list.timer, __rpc_queue_timer_fn, 0);
255 INIT_LIST_HEAD(&queue->timer_list.list);
256 rpc_assign_waitqueue_name(queue, qname);
259 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
261 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
263 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
265 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
267 __rpc_init_priority_wait_queue(queue, qname, 1);
269 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
271 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
273 del_timer_sync(&queue->timer_list.timer);
275 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
277 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
279 freezable_schedule_unsafe();
280 if (signal_pending_state(mode, current))
285 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
286 static void rpc_task_set_debuginfo(struct rpc_task *task)
288 static atomic_t rpc_pid;
290 task->tk_pid = atomic_inc_return(&rpc_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_sleep_on_priority(struct rpc_wait_queue *q,
377 struct rpc_task *task,
378 unsigned char queue_priority)
380 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
381 task->tk_pid, rpc_qname(q), jiffies);
383 trace_rpc_task_sleep(task, q);
385 __rpc_add_wait_queue(q, task, queue_priority);
389 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
390 struct rpc_task *task, unsigned long timeout,
391 unsigned char queue_priority)
393 if (time_is_after_jiffies(timeout)) {
394 __rpc_sleep_on_priority(q, task, queue_priority);
395 __rpc_add_timer(q, task, timeout);
397 task->tk_status = -ETIMEDOUT;
400 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
402 if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
403 task->tk_callback = action;
406 static bool rpc_sleep_check_activated(struct rpc_task *task)
408 /* We shouldn't ever put an inactive task to sleep */
409 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
410 task->tk_status = -EIO;
411 rpc_put_task_async(task);
417 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
418 rpc_action action, unsigned long timeout)
420 if (!rpc_sleep_check_activated(task))
423 rpc_set_tk_callback(task, action);
426 * Protect the queue operations.
428 spin_lock_bh(&q->lock);
429 __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
430 spin_unlock_bh(&q->lock);
432 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
434 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
437 if (!rpc_sleep_check_activated(task))
440 rpc_set_tk_callback(task, action);
442 WARN_ON_ONCE(task->tk_timeout != 0);
444 * Protect the queue operations.
446 spin_lock_bh(&q->lock);
447 __rpc_sleep_on_priority(q, task, task->tk_priority);
448 spin_unlock_bh(&q->lock);
450 EXPORT_SYMBOL_GPL(rpc_sleep_on);
452 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
453 struct rpc_task *task, unsigned long timeout, int priority)
455 if (!rpc_sleep_check_activated(task))
458 priority -= RPC_PRIORITY_LOW;
460 * Protect the queue operations.
462 spin_lock_bh(&q->lock);
463 __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
464 spin_unlock_bh(&q->lock);
466 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
468 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
471 if (!rpc_sleep_check_activated(task))
474 WARN_ON_ONCE(task->tk_timeout != 0);
475 priority -= RPC_PRIORITY_LOW;
477 * Protect the queue operations.
479 spin_lock_bh(&q->lock);
480 __rpc_sleep_on_priority(q, task, priority);
481 spin_unlock_bh(&q->lock);
483 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
486 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
487 * @wq: workqueue on which to run task
489 * @task: task to be woken up
491 * Caller must hold queue->lock, and have cleared the task queued flag.
493 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
494 struct rpc_wait_queue *queue,
495 struct rpc_task *task)
497 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
498 task->tk_pid, jiffies);
500 /* Has the task been executed yet? If not, we cannot wake it up! */
501 if (!RPC_IS_ACTIVATED(task)) {
502 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
506 trace_rpc_task_wakeup(task, queue);
508 __rpc_remove_wait_queue(queue, task);
510 rpc_make_runnable(wq, task);
512 dprintk("RPC: __rpc_wake_up_task done\n");
516 * Wake up a queued task while the queue lock is being held
518 static struct rpc_task *
519 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
520 struct rpc_wait_queue *queue, struct rpc_task *task,
521 bool (*action)(struct rpc_task *, void *), void *data)
523 if (RPC_IS_QUEUED(task)) {
525 if (task->tk_waitqueue == queue) {
526 if (action == NULL || action(task, data)) {
527 __rpc_do_wake_up_task_on_wq(wq, queue, task);
536 rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq,
537 struct rpc_wait_queue *queue, struct rpc_task *task)
539 rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL);
543 * Wake up a queued task while the queue lock is being held
545 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
547 rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task);
551 * Wake up a task on a specific queue
553 void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq,
554 struct rpc_wait_queue *queue,
555 struct rpc_task *task)
557 if (!RPC_IS_QUEUED(task))
559 spin_lock_bh(&queue->lock);
560 rpc_wake_up_task_on_wq_queue_locked(wq, queue, task);
561 spin_unlock_bh(&queue->lock);
565 * Wake up a task on a specific queue
567 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
569 if (!RPC_IS_QUEUED(task))
571 spin_lock_bh(&queue->lock);
572 rpc_wake_up_task_queue_locked(queue, task);
573 spin_unlock_bh(&queue->lock);
575 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
577 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
579 task->tk_status = *(int *)status;
584 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
585 struct rpc_task *task, int status)
587 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
588 task, rpc_task_action_set_status, &status);
592 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
593 * @queue: pointer to rpc_wait_queue
594 * @task: pointer to rpc_task
595 * @status: integer error value
597 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
598 * set to the value of @status.
601 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
602 struct rpc_task *task, int status)
604 if (!RPC_IS_QUEUED(task))
606 spin_lock_bh(&queue->lock);
607 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
608 spin_unlock_bh(&queue->lock);
612 * Wake up the next task on a priority queue.
614 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
617 struct rpc_task *task;
620 * Service a batch of tasks from a single owner.
622 q = &queue->tasks[queue->priority];
623 if (!list_empty(q) && --queue->nr) {
624 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
629 * Service the next queue.
632 if (q == &queue->tasks[0])
633 q = &queue->tasks[queue->maxpriority];
636 if (!list_empty(q)) {
637 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
640 } while (q != &queue->tasks[queue->priority]);
642 rpc_reset_waitqueue_priority(queue);
646 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
651 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
653 if (RPC_IS_PRIORITY(queue))
654 return __rpc_find_next_queued_priority(queue);
655 if (!list_empty(&queue->tasks[0]))
656 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
661 * Wake up the first task on the wait queue.
663 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
664 struct rpc_wait_queue *queue,
665 bool (*func)(struct rpc_task *, void *), void *data)
667 struct rpc_task *task = NULL;
669 dprintk("RPC: wake_up_first(%p \"%s\")\n",
670 queue, rpc_qname(queue));
671 spin_lock_bh(&queue->lock);
672 task = __rpc_find_next_queued(queue);
674 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
676 spin_unlock_bh(&queue->lock);
682 * Wake up the first task on the wait queue.
684 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
685 bool (*func)(struct rpc_task *, void *), void *data)
687 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
689 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
691 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
697 * Wake up the next task on the wait queue.
699 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
701 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
703 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
706 * rpc_wake_up - wake up all rpc_tasks
707 * @queue: rpc_wait_queue on which the tasks are sleeping
711 void rpc_wake_up(struct rpc_wait_queue *queue)
713 struct list_head *head;
715 spin_lock_bh(&queue->lock);
716 head = &queue->tasks[queue->maxpriority];
718 while (!list_empty(head)) {
719 struct rpc_task *task;
720 task = list_first_entry(head,
723 rpc_wake_up_task_queue_locked(queue, task);
725 if (head == &queue->tasks[0])
729 spin_unlock_bh(&queue->lock);
731 EXPORT_SYMBOL_GPL(rpc_wake_up);
734 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
735 * @queue: rpc_wait_queue on which the tasks are sleeping
736 * @status: status value to set
740 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
742 struct list_head *head;
744 spin_lock_bh(&queue->lock);
745 head = &queue->tasks[queue->maxpriority];
747 while (!list_empty(head)) {
748 struct rpc_task *task;
749 task = list_first_entry(head,
752 task->tk_status = status;
753 rpc_wake_up_task_queue_locked(queue, task);
755 if (head == &queue->tasks[0])
759 spin_unlock_bh(&queue->lock);
761 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
763 static void __rpc_queue_timer_fn(struct timer_list *t)
765 struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer);
766 struct rpc_task *task, *n;
767 unsigned long expires, now, timeo;
769 spin_lock(&queue->lock);
770 expires = now = jiffies;
771 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
772 timeo = task->tk_timeout;
773 if (time_after_eq(now, timeo)) {
774 dprintk("RPC: %5u timeout\n", task->tk_pid);
775 task->tk_status = -ETIMEDOUT;
776 rpc_wake_up_task_queue_locked(queue, task);
779 if (expires == now || time_after(expires, timeo))
782 if (!list_empty(&queue->timer_list.list))
783 rpc_set_queue_timer(queue, expires);
784 spin_unlock(&queue->lock);
787 static void __rpc_atrun(struct rpc_task *task)
789 if (task->tk_status == -ETIMEDOUT)
794 * Run a task at a later time
796 void rpc_delay(struct rpc_task *task, unsigned long delay)
798 rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
800 EXPORT_SYMBOL_GPL(rpc_delay);
803 * Helper to call task->tk_ops->rpc_call_prepare
805 void rpc_prepare_task(struct rpc_task *task)
807 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
811 rpc_init_task_statistics(struct rpc_task *task)
813 /* Initialize retry counters */
814 task->tk_garb_retry = 2;
815 task->tk_cred_retry = 2;
816 task->tk_rebind_retry = 2;
818 /* starting timestamp */
819 task->tk_start = ktime_get();
823 rpc_reset_task_statistics(struct rpc_task *task)
825 task->tk_timeouts = 0;
826 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
827 rpc_init_task_statistics(task);
831 * Helper that calls task->tk_ops->rpc_call_done if it exists
833 void rpc_exit_task(struct rpc_task *task)
835 task->tk_action = NULL;
836 if (task->tk_ops->rpc_call_done != NULL) {
837 task->tk_ops->rpc_call_done(task, task->tk_calldata);
838 if (task->tk_action != NULL) {
839 /* Always release the RPC slot and buffer memory */
841 rpc_reset_task_statistics(task);
846 void rpc_signal_task(struct rpc_task *task)
848 struct rpc_wait_queue *queue;
850 if (!RPC_IS_ACTIVATED(task))
852 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
853 smp_mb__after_atomic();
854 queue = READ_ONCE(task->tk_waitqueue);
856 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
859 void rpc_exit(struct rpc_task *task, int status)
861 task->tk_status = status;
862 task->tk_action = rpc_exit_task;
863 rpc_wake_up_queued_task(task->tk_waitqueue, task);
865 EXPORT_SYMBOL_GPL(rpc_exit);
867 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
869 if (ops->rpc_release != NULL)
870 ops->rpc_release(calldata);
874 * This is the RPC `scheduler' (or rather, the finite state machine).
876 static void __rpc_execute(struct rpc_task *task)
878 struct rpc_wait_queue *queue;
879 int task_is_async = RPC_IS_ASYNC(task);
882 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
883 task->tk_pid, task->tk_flags);
885 WARN_ON_ONCE(RPC_IS_QUEUED(task));
886 if (RPC_IS_QUEUED(task))
890 void (*do_action)(struct rpc_task *);
893 * Perform the next FSM step or a pending callback.
895 * tk_action may be NULL if the task has been killed.
896 * In particular, note that rpc_killall_tasks may
897 * do this at any time, so beware when dereferencing.
899 do_action = task->tk_action;
900 if (task->tk_callback) {
901 do_action = task->tk_callback;
902 task->tk_callback = NULL;
906 trace_rpc_task_run_action(task, do_action);
910 * Lockless check for whether task is sleeping or not.
912 if (!RPC_IS_QUEUED(task))
916 * Signalled tasks should exit rather than sleep.
918 if (RPC_SIGNALLED(task))
919 rpc_exit(task, -ERESTARTSYS);
922 * The queue->lock protects against races with
923 * rpc_make_runnable().
925 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
926 * rpc_task, rpc_make_runnable() can assign it to a
927 * different workqueue. We therefore cannot assume that the
928 * rpc_task pointer may still be dereferenced.
930 queue = task->tk_waitqueue;
931 spin_lock_bh(&queue->lock);
932 if (!RPC_IS_QUEUED(task)) {
933 spin_unlock_bh(&queue->lock);
936 rpc_clear_running(task);
937 spin_unlock_bh(&queue->lock);
941 /* sync task: sleep here */
942 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
943 status = out_of_line_wait_on_bit(&task->tk_runstate,
944 RPC_TASK_QUEUED, rpc_wait_bit_killable,
948 * When a sync task receives a signal, it exits with
949 * -ERESTARTSYS. In order to catch any callbacks that
950 * clean up after sleeping on some queue, we don't
951 * break the loop here, but go around once more.
953 dprintk("RPC: %5u got signal\n", task->tk_pid);
954 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
955 rpc_exit(task, -ERESTARTSYS);
957 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
960 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
962 /* Release all resources associated with the task */
963 rpc_release_task(task);
967 * User-visible entry point to the scheduler.
969 * This may be called recursively if e.g. an async NFS task updates
970 * the attributes and finds that dirty pages must be flushed.
971 * NOTE: Upon exit of this function the task is guaranteed to be
972 * released. In particular note that tk_release() will have
973 * been called, so your task memory may have been freed.
975 void rpc_execute(struct rpc_task *task)
977 bool is_async = RPC_IS_ASYNC(task);
979 rpc_set_active(task);
980 rpc_make_runnable(rpciod_workqueue, task);
985 static void rpc_async_schedule(struct work_struct *work)
987 unsigned int pflags = memalloc_nofs_save();
989 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
990 memalloc_nofs_restore(pflags);
994 * rpc_malloc - allocate RPC buffer resources
997 * A single memory region is allocated, which is split between the
998 * RPC call and RPC reply that this task is being used for. When
999 * this RPC is retired, the memory is released by calling rpc_free.
1001 * To prevent rpciod from hanging, this allocator never sleeps,
1002 * returning -ENOMEM and suppressing warning if the request cannot
1003 * be serviced immediately. The caller can arrange to sleep in a
1004 * way that is safe for rpciod.
1006 * Most requests are 'small' (under 2KiB) and can be serviced from a
1007 * mempool, ensuring that NFS reads and writes can always proceed,
1008 * and that there is good locality of reference for these buffers.
1010 int rpc_malloc(struct rpc_task *task)
1012 struct rpc_rqst *rqst = task->tk_rqstp;
1013 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1014 struct rpc_buffer *buf;
1015 gfp_t gfp = GFP_NOFS;
1017 if (RPC_IS_SWAPPER(task))
1018 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
1020 size += sizeof(struct rpc_buffer);
1021 if (size <= RPC_BUFFER_MAXSIZE)
1022 buf = mempool_alloc(rpc_buffer_mempool, gfp);
1024 buf = kmalloc(size, gfp);
1030 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1031 task->tk_pid, size, buf);
1032 rqst->rq_buffer = buf->data;
1033 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1036 EXPORT_SYMBOL_GPL(rpc_malloc);
1039 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1043 void rpc_free(struct rpc_task *task)
1045 void *buffer = task->tk_rqstp->rq_buffer;
1047 struct rpc_buffer *buf;
1049 buf = container_of(buffer, struct rpc_buffer, data);
1052 dprintk("RPC: freeing buffer of size %zu at %p\n",
1055 if (size <= RPC_BUFFER_MAXSIZE)
1056 mempool_free(buf, rpc_buffer_mempool);
1060 EXPORT_SYMBOL_GPL(rpc_free);
1063 * Creation and deletion of RPC task structures
1065 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1067 memset(task, 0, sizeof(*task));
1068 atomic_set(&task->tk_count, 1);
1069 task->tk_flags = task_setup_data->flags;
1070 task->tk_ops = task_setup_data->callback_ops;
1071 task->tk_calldata = task_setup_data->callback_data;
1072 INIT_LIST_HEAD(&task->tk_task);
1074 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1075 task->tk_owner = current->tgid;
1077 /* Initialize workqueue for async tasks */
1078 task->tk_workqueue = task_setup_data->workqueue;
1080 task->tk_xprt = xprt_get(task_setup_data->rpc_xprt);
1082 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1084 if (task->tk_ops->rpc_call_prepare != NULL)
1085 task->tk_action = rpc_prepare_task;
1087 rpc_init_task_statistics(task);
1089 dprintk("RPC: new task initialized, procpid %u\n",
1090 task_pid_nr(current));
1093 static struct rpc_task *
1094 rpc_alloc_task(void)
1096 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1100 * Create a new task for the specified client.
1102 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1104 struct rpc_task *task = setup_data->task;
1105 unsigned short flags = 0;
1108 task = rpc_alloc_task();
1109 flags = RPC_TASK_DYNAMIC;
1112 rpc_init_task(task, setup_data);
1113 task->tk_flags |= flags;
1114 dprintk("RPC: allocated task %p\n", task);
1119 * rpc_free_task - release rpc task and perform cleanups
1121 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1122 * in order to work around a workqueue dependency issue.
1125 * "Workqueue currently considers two work items to be the same if they're
1126 * on the same address and won't execute them concurrently - ie. it
1127 * makes a work item which is queued again while being executed wait
1128 * for the previous execution to complete.
1130 * If a work function frees the work item, and then waits for an event
1131 * which should be performed by another work item and *that* work item
1132 * recycles the freed work item, it can create a false dependency loop.
1133 * There really is no reliable way to detect this short of verifying
1134 * every memory free."
1137 static void rpc_free_task(struct rpc_task *task)
1139 unsigned short tk_flags = task->tk_flags;
1141 put_rpccred(task->tk_op_cred);
1142 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1144 if (tk_flags & RPC_TASK_DYNAMIC) {
1145 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1146 mempool_free(task, rpc_task_mempool);
1150 static void rpc_async_release(struct work_struct *work)
1152 unsigned int pflags = memalloc_nofs_save();
1154 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1155 memalloc_nofs_restore(pflags);
1158 static void rpc_release_resources_task(struct rpc_task *task)
1161 if (task->tk_msg.rpc_cred) {
1162 put_cred(task->tk_msg.rpc_cred);
1163 task->tk_msg.rpc_cred = NULL;
1165 rpc_task_release_client(task);
1168 static void rpc_final_put_task(struct rpc_task *task,
1169 struct workqueue_struct *q)
1172 INIT_WORK(&task->u.tk_work, rpc_async_release);
1173 queue_work(q, &task->u.tk_work);
1175 rpc_free_task(task);
1178 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1180 if (atomic_dec_and_test(&task->tk_count)) {
1181 rpc_release_resources_task(task);
1182 rpc_final_put_task(task, q);
1186 void rpc_put_task(struct rpc_task *task)
1188 rpc_do_put_task(task, NULL);
1190 EXPORT_SYMBOL_GPL(rpc_put_task);
1192 void rpc_put_task_async(struct rpc_task *task)
1194 rpc_do_put_task(task, task->tk_workqueue);
1196 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1198 static void rpc_release_task(struct rpc_task *task)
1200 dprintk("RPC: %5u release task\n", task->tk_pid);
1202 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1204 rpc_release_resources_task(task);
1207 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1208 * so it should be safe to use task->tk_count as a test for whether
1209 * or not any other processes still hold references to our rpc_task.
1211 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1212 /* Wake up anyone who may be waiting for task completion */
1213 if (!rpc_complete_task(task))
1216 if (!atomic_dec_and_test(&task->tk_count))
1219 rpc_final_put_task(task, task->tk_workqueue);
1224 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1227 void rpciod_down(void)
1229 module_put(THIS_MODULE);
1233 * Start up the rpciod workqueue.
1235 static int rpciod_start(void)
1237 struct workqueue_struct *wq;
1240 * Create the rpciod thread and wait for it to start.
1242 dprintk("RPC: creating workqueue rpciod\n");
1243 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1246 rpciod_workqueue = wq;
1247 /* Note: highpri because network receive is latency sensitive */
1248 wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1251 xprtiod_workqueue = wq;
1254 wq = rpciod_workqueue;
1255 rpciod_workqueue = NULL;
1256 destroy_workqueue(wq);
1261 static void rpciod_stop(void)
1263 struct workqueue_struct *wq = NULL;
1265 if (rpciod_workqueue == NULL)
1267 dprintk("RPC: destroying workqueue rpciod\n");
1269 wq = rpciod_workqueue;
1270 rpciod_workqueue = NULL;
1271 destroy_workqueue(wq);
1272 wq = xprtiod_workqueue;
1273 xprtiod_workqueue = NULL;
1274 destroy_workqueue(wq);
1278 rpc_destroy_mempool(void)
1281 mempool_destroy(rpc_buffer_mempool);
1282 mempool_destroy(rpc_task_mempool);
1283 kmem_cache_destroy(rpc_task_slabp);
1284 kmem_cache_destroy(rpc_buffer_slabp);
1285 rpc_destroy_wait_queue(&delay_queue);
1289 rpc_init_mempool(void)
1292 * The following is not strictly a mempool initialisation,
1293 * but there is no harm in doing it here
1295 rpc_init_wait_queue(&delay_queue, "delayq");
1296 if (!rpciod_start())
1299 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1300 sizeof(struct rpc_task),
1301 0, SLAB_HWCACHE_ALIGN,
1303 if (!rpc_task_slabp)
1305 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1307 0, SLAB_HWCACHE_ALIGN,
1309 if (!rpc_buffer_slabp)
1311 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1313 if (!rpc_task_mempool)
1315 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1317 if (!rpc_buffer_mempool)
1321 rpc_destroy_mempool();