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 * Disable the timer for a given RPC task. Should be called with
63 * queue->lock and bh_disabled in order to avoid races within
67 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
69 if (task->tk_timeout == 0)
71 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
73 list_del(&task->u.tk_wait.timer_list);
74 if (list_empty(&queue->timer_list.list))
75 del_timer(&queue->timer_list.timer);
79 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
81 queue->timer_list.expires = expires;
82 mod_timer(&queue->timer_list.timer, expires);
86 * Set up a timer for the current task.
89 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
91 if (!task->tk_timeout)
94 dprintk("RPC: %5u setting alarm for %u ms\n",
95 task->tk_pid, jiffies_to_msecs(task->tk_timeout));
97 task->u.tk_wait.expires = jiffies + task->tk_timeout;
98 if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
99 rpc_set_queue_timer(queue, task->u.tk_wait.expires);
100 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
103 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
105 if (queue->priority != priority) {
106 queue->priority = priority;
107 queue->nr = 1U << priority;
111 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
113 rpc_set_waitqueue_priority(queue, queue->maxpriority);
117 * Add a request to a queue list
120 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
124 list_for_each_entry(t, q, u.tk_wait.list) {
125 if (t->tk_owner == task->tk_owner) {
126 list_add_tail(&task->u.tk_wait.links,
127 &t->u.tk_wait.links);
128 /* Cache the queue head in task->u.tk_wait.list */
129 task->u.tk_wait.list.next = q;
130 task->u.tk_wait.list.prev = NULL;
134 INIT_LIST_HEAD(&task->u.tk_wait.links);
135 list_add_tail(&task->u.tk_wait.list, q);
139 * Remove request from a queue list
142 __rpc_list_dequeue_task(struct rpc_task *task)
147 if (task->u.tk_wait.list.prev == NULL) {
148 list_del(&task->u.tk_wait.links);
151 if (!list_empty(&task->u.tk_wait.links)) {
152 t = list_first_entry(&task->u.tk_wait.links,
155 /* Assume __rpc_list_enqueue_task() cached the queue head */
156 q = t->u.tk_wait.list.next;
157 list_add_tail(&t->u.tk_wait.list, q);
158 list_del(&task->u.tk_wait.links);
160 list_del(&task->u.tk_wait.list);
164 * Add new request to a priority queue.
166 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
167 struct rpc_task *task,
168 unsigned char queue_priority)
170 if (unlikely(queue_priority > queue->maxpriority))
171 queue_priority = queue->maxpriority;
172 __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
176 * Add new request to wait queue.
178 * Swapper tasks always get inserted at the head of the queue.
179 * This should avoid many nasty memory deadlocks and hopefully
180 * improve overall performance.
181 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
183 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
184 struct rpc_task *task,
185 unsigned char queue_priority)
187 WARN_ON_ONCE(RPC_IS_QUEUED(task));
188 if (RPC_IS_QUEUED(task))
191 if (RPC_IS_PRIORITY(queue))
192 __rpc_add_wait_queue_priority(queue, task, queue_priority);
193 else if (RPC_IS_SWAPPER(task))
194 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
196 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
197 task->tk_waitqueue = queue;
199 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
201 rpc_set_queued(task);
203 dprintk("RPC: %5u added to queue %p \"%s\"\n",
204 task->tk_pid, queue, rpc_qname(queue));
208 * Remove request from a priority queue.
210 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
212 __rpc_list_dequeue_task(task);
216 * Remove request from queue.
217 * Note: must be called with spin lock held.
219 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
221 __rpc_disable_timer(queue, task);
222 if (RPC_IS_PRIORITY(queue))
223 __rpc_remove_wait_queue_priority(task);
225 list_del(&task->u.tk_wait.list);
227 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
228 task->tk_pid, queue, rpc_qname(queue));
231 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
235 spin_lock_init(&queue->lock);
236 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
237 INIT_LIST_HEAD(&queue->tasks[i]);
238 queue->maxpriority = nr_queues - 1;
239 rpc_reset_waitqueue_priority(queue);
241 timer_setup(&queue->timer_list.timer, __rpc_queue_timer_fn, 0);
242 INIT_LIST_HEAD(&queue->timer_list.list);
243 rpc_assign_waitqueue_name(queue, qname);
246 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
248 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
250 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
252 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
254 __rpc_init_priority_wait_queue(queue, qname, 1);
256 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
258 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
260 del_timer_sync(&queue->timer_list.timer);
262 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
264 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
266 freezable_schedule_unsafe();
267 if (signal_pending_state(mode, current))
272 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
273 static void rpc_task_set_debuginfo(struct rpc_task *task)
275 static atomic_t rpc_pid;
277 task->tk_pid = atomic_inc_return(&rpc_pid);
280 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
285 static void rpc_set_active(struct rpc_task *task)
287 rpc_task_set_debuginfo(task);
288 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
289 trace_rpc_task_begin(task, NULL);
293 * Mark an RPC call as having completed by clearing the 'active' bit
294 * and then waking up all tasks that were sleeping.
296 static int rpc_complete_task(struct rpc_task *task)
298 void *m = &task->tk_runstate;
299 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
300 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
304 trace_rpc_task_complete(task, NULL);
306 spin_lock_irqsave(&wq->lock, flags);
307 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
308 ret = atomic_dec_and_test(&task->tk_count);
309 if (waitqueue_active(wq))
310 __wake_up_locked_key(wq, TASK_NORMAL, &k);
311 spin_unlock_irqrestore(&wq->lock, flags);
316 * Allow callers to wait for completion of an RPC call
318 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
319 * to enforce taking of the wq->lock and hence avoid races with
320 * rpc_complete_task().
322 int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
325 action = rpc_wait_bit_killable;
326 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
327 action, TASK_KILLABLE);
329 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
332 * Make an RPC task runnable.
334 * Note: If the task is ASYNC, and is being made runnable after sitting on an
335 * rpc_wait_queue, this must be called with the queue spinlock held to protect
336 * the wait queue operation.
337 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
338 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
339 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
340 * the RPC_TASK_RUNNING flag.
342 static void rpc_make_runnable(struct workqueue_struct *wq,
343 struct rpc_task *task)
345 bool need_wakeup = !rpc_test_and_set_running(task);
347 rpc_clear_queued(task);
350 if (RPC_IS_ASYNC(task)) {
351 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
352 queue_work(wq, &task->u.tk_work);
354 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
358 * Prepare for sleeping on a wait queue.
359 * By always appending tasks to the list we ensure FIFO behavior.
360 * NB: An RPC task will only receive interrupt-driven events as long
361 * as it's on a wait queue.
363 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
364 struct rpc_task *task,
366 unsigned char queue_priority)
368 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
369 task->tk_pid, rpc_qname(q), jiffies);
371 trace_rpc_task_sleep(task, q);
373 __rpc_add_wait_queue(q, task, queue_priority);
375 WARN_ON_ONCE(task->tk_callback != NULL);
376 task->tk_callback = action;
377 __rpc_add_timer(q, task);
380 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
383 /* We shouldn't ever put an inactive task to sleep */
384 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
385 if (!RPC_IS_ACTIVATED(task)) {
386 task->tk_status = -EIO;
387 rpc_put_task_async(task);
392 * Protect the queue operations.
394 spin_lock_bh(&q->lock);
395 __rpc_sleep_on_priority(q, task, action, task->tk_priority);
396 spin_unlock_bh(&q->lock);
398 EXPORT_SYMBOL_GPL(rpc_sleep_on);
400 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
401 rpc_action action, int priority)
403 /* We shouldn't ever put an inactive task to sleep */
404 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
405 if (!RPC_IS_ACTIVATED(task)) {
406 task->tk_status = -EIO;
407 rpc_put_task_async(task);
412 * Protect the queue operations.
414 spin_lock_bh(&q->lock);
415 __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
416 spin_unlock_bh(&q->lock);
418 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
421 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
422 * @wq: workqueue on which to run task
424 * @task: task to be woken up
426 * Caller must hold queue->lock, and have cleared the task queued flag.
428 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
429 struct rpc_wait_queue *queue,
430 struct rpc_task *task)
432 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
433 task->tk_pid, jiffies);
435 /* Has the task been executed yet? If not, we cannot wake it up! */
436 if (!RPC_IS_ACTIVATED(task)) {
437 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
441 trace_rpc_task_wakeup(task, queue);
443 __rpc_remove_wait_queue(queue, task);
445 rpc_make_runnable(wq, task);
447 dprintk("RPC: __rpc_wake_up_task done\n");
451 * Wake up a queued task while the queue lock is being held
453 static struct rpc_task *
454 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
455 struct rpc_wait_queue *queue, struct rpc_task *task,
456 bool (*action)(struct rpc_task *, void *), void *data)
458 if (RPC_IS_QUEUED(task)) {
460 if (task->tk_waitqueue == queue) {
461 if (action == NULL || action(task, data)) {
462 __rpc_do_wake_up_task_on_wq(wq, queue, task);
471 rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq,
472 struct rpc_wait_queue *queue, struct rpc_task *task)
474 rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL);
478 * Wake up a queued task while the queue lock is being held
480 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
482 rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task);
486 * Wake up a task on a specific queue
488 void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq,
489 struct rpc_wait_queue *queue,
490 struct rpc_task *task)
492 if (!RPC_IS_QUEUED(task))
494 spin_lock_bh(&queue->lock);
495 rpc_wake_up_task_on_wq_queue_locked(wq, queue, task);
496 spin_unlock_bh(&queue->lock);
500 * Wake up a task on a specific queue
502 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
504 if (!RPC_IS_QUEUED(task))
506 spin_lock_bh(&queue->lock);
507 rpc_wake_up_task_queue_locked(queue, task);
508 spin_unlock_bh(&queue->lock);
510 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
512 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
514 task->tk_status = *(int *)status;
519 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
520 struct rpc_task *task, int status)
522 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
523 task, rpc_task_action_set_status, &status);
527 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
528 * @queue: pointer to rpc_wait_queue
529 * @task: pointer to rpc_task
530 * @status: integer error value
532 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
533 * set to the value of @status.
536 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
537 struct rpc_task *task, int status)
539 if (!RPC_IS_QUEUED(task))
541 spin_lock_bh(&queue->lock);
542 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
543 spin_unlock_bh(&queue->lock);
547 * Wake up the next task on a priority queue.
549 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
552 struct rpc_task *task;
555 * Service a batch of tasks from a single owner.
557 q = &queue->tasks[queue->priority];
558 if (!list_empty(q) && --queue->nr) {
559 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
564 * Service the next queue.
567 if (q == &queue->tasks[0])
568 q = &queue->tasks[queue->maxpriority];
571 if (!list_empty(q)) {
572 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
575 } while (q != &queue->tasks[queue->priority]);
577 rpc_reset_waitqueue_priority(queue);
581 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
586 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
588 if (RPC_IS_PRIORITY(queue))
589 return __rpc_find_next_queued_priority(queue);
590 if (!list_empty(&queue->tasks[0]))
591 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
596 * Wake up the first task on the wait queue.
598 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
599 struct rpc_wait_queue *queue,
600 bool (*func)(struct rpc_task *, void *), void *data)
602 struct rpc_task *task = NULL;
604 dprintk("RPC: wake_up_first(%p \"%s\")\n",
605 queue, rpc_qname(queue));
606 spin_lock_bh(&queue->lock);
607 task = __rpc_find_next_queued(queue);
609 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
611 spin_unlock_bh(&queue->lock);
617 * Wake up the first task on the wait queue.
619 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
620 bool (*func)(struct rpc_task *, void *), void *data)
622 return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
624 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
626 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
632 * Wake up the next task on the wait queue.
634 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
636 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
638 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
641 * rpc_wake_up - wake up all rpc_tasks
642 * @queue: rpc_wait_queue on which the tasks are sleeping
646 void rpc_wake_up(struct rpc_wait_queue *queue)
648 struct list_head *head;
650 spin_lock_bh(&queue->lock);
651 head = &queue->tasks[queue->maxpriority];
653 while (!list_empty(head)) {
654 struct rpc_task *task;
655 task = list_first_entry(head,
658 rpc_wake_up_task_queue_locked(queue, task);
660 if (head == &queue->tasks[0])
664 spin_unlock_bh(&queue->lock);
666 EXPORT_SYMBOL_GPL(rpc_wake_up);
669 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
670 * @queue: rpc_wait_queue on which the tasks are sleeping
671 * @status: status value to set
675 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
677 struct list_head *head;
679 spin_lock_bh(&queue->lock);
680 head = &queue->tasks[queue->maxpriority];
682 while (!list_empty(head)) {
683 struct rpc_task *task;
684 task = list_first_entry(head,
687 task->tk_status = status;
688 rpc_wake_up_task_queue_locked(queue, task);
690 if (head == &queue->tasks[0])
694 spin_unlock_bh(&queue->lock);
696 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
698 static void __rpc_queue_timer_fn(struct timer_list *t)
700 struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer);
701 struct rpc_task *task, *n;
702 unsigned long expires, now, timeo;
704 spin_lock(&queue->lock);
705 expires = now = jiffies;
706 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
707 timeo = task->u.tk_wait.expires;
708 if (time_after_eq(now, timeo)) {
709 dprintk("RPC: %5u timeout\n", task->tk_pid);
710 task->tk_status = -ETIMEDOUT;
711 rpc_wake_up_task_queue_locked(queue, task);
714 if (expires == now || time_after(expires, timeo))
717 if (!list_empty(&queue->timer_list.list))
718 rpc_set_queue_timer(queue, expires);
719 spin_unlock(&queue->lock);
722 static void __rpc_atrun(struct rpc_task *task)
724 if (task->tk_status == -ETIMEDOUT)
729 * Run a task at a later time
731 void rpc_delay(struct rpc_task *task, unsigned long delay)
733 task->tk_timeout = delay;
734 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
736 EXPORT_SYMBOL_GPL(rpc_delay);
739 * Helper to call task->tk_ops->rpc_call_prepare
741 void rpc_prepare_task(struct rpc_task *task)
743 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
747 rpc_init_task_statistics(struct rpc_task *task)
749 /* Initialize retry counters */
750 task->tk_garb_retry = 2;
751 task->tk_cred_retry = 2;
752 task->tk_rebind_retry = 2;
754 /* starting timestamp */
755 task->tk_start = ktime_get();
759 rpc_reset_task_statistics(struct rpc_task *task)
761 task->tk_timeouts = 0;
762 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
763 rpc_init_task_statistics(task);
767 * Helper that calls task->tk_ops->rpc_call_done if it exists
769 void rpc_exit_task(struct rpc_task *task)
771 task->tk_action = NULL;
772 if (task->tk_ops->rpc_call_done != NULL) {
773 task->tk_ops->rpc_call_done(task, task->tk_calldata);
774 if (task->tk_action != NULL) {
775 /* Always release the RPC slot and buffer memory */
777 rpc_reset_task_statistics(task);
782 void rpc_signal_task(struct rpc_task *task)
784 struct rpc_wait_queue *queue;
786 if (!RPC_IS_ACTIVATED(task))
788 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
789 smp_mb__after_atomic();
790 queue = READ_ONCE(task->tk_waitqueue);
792 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
795 void rpc_exit(struct rpc_task *task, int status)
797 task->tk_status = status;
798 task->tk_action = rpc_exit_task;
799 rpc_wake_up_queued_task(task->tk_waitqueue, task);
801 EXPORT_SYMBOL_GPL(rpc_exit);
803 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
805 if (ops->rpc_release != NULL)
806 ops->rpc_release(calldata);
810 * This is the RPC `scheduler' (or rather, the finite state machine).
812 static void __rpc_execute(struct rpc_task *task)
814 struct rpc_wait_queue *queue;
815 int task_is_async = RPC_IS_ASYNC(task);
818 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
819 task->tk_pid, task->tk_flags);
821 WARN_ON_ONCE(RPC_IS_QUEUED(task));
822 if (RPC_IS_QUEUED(task))
826 void (*do_action)(struct rpc_task *);
829 * Perform the next FSM step or a pending callback.
831 * tk_action may be NULL if the task has been killed.
832 * In particular, note that rpc_killall_tasks may
833 * do this at any time, so beware when dereferencing.
835 do_action = task->tk_action;
836 if (task->tk_callback) {
837 do_action = task->tk_callback;
838 task->tk_callback = NULL;
842 trace_rpc_task_run_action(task, do_action);
846 * Lockless check for whether task is sleeping or not.
848 if (!RPC_IS_QUEUED(task))
852 * Signalled tasks should exit rather than sleep.
854 if (RPC_SIGNALLED(task))
855 rpc_exit(task, -ERESTARTSYS);
858 * The queue->lock protects against races with
859 * rpc_make_runnable().
861 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
862 * rpc_task, rpc_make_runnable() can assign it to a
863 * different workqueue. We therefore cannot assume that the
864 * rpc_task pointer may still be dereferenced.
866 queue = task->tk_waitqueue;
867 spin_lock_bh(&queue->lock);
868 if (!RPC_IS_QUEUED(task)) {
869 spin_unlock_bh(&queue->lock);
872 rpc_clear_running(task);
873 spin_unlock_bh(&queue->lock);
877 /* sync task: sleep here */
878 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
879 status = out_of_line_wait_on_bit(&task->tk_runstate,
880 RPC_TASK_QUEUED, rpc_wait_bit_killable,
884 * When a sync task receives a signal, it exits with
885 * -ERESTARTSYS. In order to catch any callbacks that
886 * clean up after sleeping on some queue, we don't
887 * break the loop here, but go around once more.
889 dprintk("RPC: %5u got signal\n", task->tk_pid);
890 set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
891 rpc_exit(task, -ERESTARTSYS);
893 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
896 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
898 /* Release all resources associated with the task */
899 rpc_release_task(task);
903 * User-visible entry point to the scheduler.
905 * This may be called recursively if e.g. an async NFS task updates
906 * the attributes and finds that dirty pages must be flushed.
907 * NOTE: Upon exit of this function the task is guaranteed to be
908 * released. In particular note that tk_release() will have
909 * been called, so your task memory may have been freed.
911 void rpc_execute(struct rpc_task *task)
913 bool is_async = RPC_IS_ASYNC(task);
915 rpc_set_active(task);
916 rpc_make_runnable(rpciod_workqueue, task);
921 static void rpc_async_schedule(struct work_struct *work)
923 unsigned int pflags = memalloc_nofs_save();
925 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
926 memalloc_nofs_restore(pflags);
930 * rpc_malloc - allocate RPC buffer resources
933 * A single memory region is allocated, which is split between the
934 * RPC call and RPC reply that this task is being used for. When
935 * this RPC is retired, the memory is released by calling rpc_free.
937 * To prevent rpciod from hanging, this allocator never sleeps,
938 * returning -ENOMEM and suppressing warning if the request cannot
939 * be serviced immediately. The caller can arrange to sleep in a
940 * way that is safe for rpciod.
942 * Most requests are 'small' (under 2KiB) and can be serviced from a
943 * mempool, ensuring that NFS reads and writes can always proceed,
944 * and that there is good locality of reference for these buffers.
946 int rpc_malloc(struct rpc_task *task)
948 struct rpc_rqst *rqst = task->tk_rqstp;
949 size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
950 struct rpc_buffer *buf;
951 gfp_t gfp = GFP_NOFS;
953 if (RPC_IS_SWAPPER(task))
954 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
956 size += sizeof(struct rpc_buffer);
957 if (size <= RPC_BUFFER_MAXSIZE)
958 buf = mempool_alloc(rpc_buffer_mempool, gfp);
960 buf = kmalloc(size, gfp);
966 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
967 task->tk_pid, size, buf);
968 rqst->rq_buffer = buf->data;
969 rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
972 EXPORT_SYMBOL_GPL(rpc_malloc);
975 * rpc_free - free RPC buffer resources allocated via rpc_malloc
979 void rpc_free(struct rpc_task *task)
981 void *buffer = task->tk_rqstp->rq_buffer;
983 struct rpc_buffer *buf;
985 buf = container_of(buffer, struct rpc_buffer, data);
988 dprintk("RPC: freeing buffer of size %zu at %p\n",
991 if (size <= RPC_BUFFER_MAXSIZE)
992 mempool_free(buf, rpc_buffer_mempool);
996 EXPORT_SYMBOL_GPL(rpc_free);
999 * Creation and deletion of RPC task structures
1001 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1003 memset(task, 0, sizeof(*task));
1004 atomic_set(&task->tk_count, 1);
1005 task->tk_flags = task_setup_data->flags;
1006 task->tk_ops = task_setup_data->callback_ops;
1007 task->tk_calldata = task_setup_data->callback_data;
1008 INIT_LIST_HEAD(&task->tk_task);
1010 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1011 task->tk_owner = current->tgid;
1013 /* Initialize workqueue for async tasks */
1014 task->tk_workqueue = task_setup_data->workqueue;
1016 task->tk_xprt = xprt_get(task_setup_data->rpc_xprt);
1018 task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1020 if (task->tk_ops->rpc_call_prepare != NULL)
1021 task->tk_action = rpc_prepare_task;
1023 rpc_init_task_statistics(task);
1025 dprintk("RPC: new task initialized, procpid %u\n",
1026 task_pid_nr(current));
1029 static struct rpc_task *
1030 rpc_alloc_task(void)
1032 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1036 * Create a new task for the specified client.
1038 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1040 struct rpc_task *task = setup_data->task;
1041 unsigned short flags = 0;
1044 task = rpc_alloc_task();
1045 flags = RPC_TASK_DYNAMIC;
1048 rpc_init_task(task, setup_data);
1049 task->tk_flags |= flags;
1050 dprintk("RPC: allocated task %p\n", task);
1055 * rpc_free_task - release rpc task and perform cleanups
1057 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1058 * in order to work around a workqueue dependency issue.
1061 * "Workqueue currently considers two work items to be the same if they're
1062 * on the same address and won't execute them concurrently - ie. it
1063 * makes a work item which is queued again while being executed wait
1064 * for the previous execution to complete.
1066 * If a work function frees the work item, and then waits for an event
1067 * which should be performed by another work item and *that* work item
1068 * recycles the freed work item, it can create a false dependency loop.
1069 * There really is no reliable way to detect this short of verifying
1070 * every memory free."
1073 static void rpc_free_task(struct rpc_task *task)
1075 unsigned short tk_flags = task->tk_flags;
1077 put_rpccred(task->tk_op_cred);
1078 rpc_release_calldata(task->tk_ops, task->tk_calldata);
1080 if (tk_flags & RPC_TASK_DYNAMIC) {
1081 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1082 mempool_free(task, rpc_task_mempool);
1086 static void rpc_async_release(struct work_struct *work)
1088 unsigned int pflags = memalloc_nofs_save();
1090 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1091 memalloc_nofs_restore(pflags);
1094 static void rpc_release_resources_task(struct rpc_task *task)
1097 if (task->tk_msg.rpc_cred) {
1098 put_cred(task->tk_msg.rpc_cred);
1099 task->tk_msg.rpc_cred = NULL;
1101 rpc_task_release_client(task);
1104 static void rpc_final_put_task(struct rpc_task *task,
1105 struct workqueue_struct *q)
1108 INIT_WORK(&task->u.tk_work, rpc_async_release);
1109 queue_work(q, &task->u.tk_work);
1111 rpc_free_task(task);
1114 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1116 if (atomic_dec_and_test(&task->tk_count)) {
1117 rpc_release_resources_task(task);
1118 rpc_final_put_task(task, q);
1122 void rpc_put_task(struct rpc_task *task)
1124 rpc_do_put_task(task, NULL);
1126 EXPORT_SYMBOL_GPL(rpc_put_task);
1128 void rpc_put_task_async(struct rpc_task *task)
1130 rpc_do_put_task(task, task->tk_workqueue);
1132 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1134 static void rpc_release_task(struct rpc_task *task)
1136 dprintk("RPC: %5u release task\n", task->tk_pid);
1138 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1140 rpc_release_resources_task(task);
1143 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1144 * so it should be safe to use task->tk_count as a test for whether
1145 * or not any other processes still hold references to our rpc_task.
1147 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1148 /* Wake up anyone who may be waiting for task completion */
1149 if (!rpc_complete_task(task))
1152 if (!atomic_dec_and_test(&task->tk_count))
1155 rpc_final_put_task(task, task->tk_workqueue);
1160 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1163 void rpciod_down(void)
1165 module_put(THIS_MODULE);
1169 * Start up the rpciod workqueue.
1171 static int rpciod_start(void)
1173 struct workqueue_struct *wq;
1176 * Create the rpciod thread and wait for it to start.
1178 dprintk("RPC: creating workqueue rpciod\n");
1179 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1182 rpciod_workqueue = wq;
1183 /* Note: highpri because network receive is latency sensitive */
1184 wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1187 xprtiod_workqueue = wq;
1190 wq = rpciod_workqueue;
1191 rpciod_workqueue = NULL;
1192 destroy_workqueue(wq);
1197 static void rpciod_stop(void)
1199 struct workqueue_struct *wq = NULL;
1201 if (rpciod_workqueue == NULL)
1203 dprintk("RPC: destroying workqueue rpciod\n");
1205 wq = rpciod_workqueue;
1206 rpciod_workqueue = NULL;
1207 destroy_workqueue(wq);
1208 wq = xprtiod_workqueue;
1209 xprtiod_workqueue = NULL;
1210 destroy_workqueue(wq);
1214 rpc_destroy_mempool(void)
1217 mempool_destroy(rpc_buffer_mempool);
1218 mempool_destroy(rpc_task_mempool);
1219 kmem_cache_destroy(rpc_task_slabp);
1220 kmem_cache_destroy(rpc_buffer_slabp);
1221 rpc_destroy_wait_queue(&delay_queue);
1225 rpc_init_mempool(void)
1228 * The following is not strictly a mempool initialisation,
1229 * but there is no harm in doing it here
1231 rpc_init_wait_queue(&delay_queue, "delayq");
1232 if (!rpciod_start())
1235 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1236 sizeof(struct rpc_task),
1237 0, SLAB_HWCACHE_ALIGN,
1239 if (!rpc_task_slabp)
1241 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1243 0, SLAB_HWCACHE_ALIGN,
1245 if (!rpc_buffer_slabp)
1247 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1249 if (!rpc_task_mempool)
1251 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1253 if (!rpc_buffer_mempool)
1257 rpc_destroy_mempool();