1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/net/sunrpc/sched.c
5 * Scheduling for synchronous and asynchronous RPC requests.
7 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
9 * TCP NFS related read + write fixes
10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
13 #include <linux/module.h>
15 #include <linux/sched.h>
16 #include <linux/interrupt.h>
17 #include <linux/slab.h>
18 #include <linux/mempool.h>
19 #include <linux/smp.h>
20 #include <linux/spinlock.h>
21 #include <linux/mutex.h>
22 #include <linux/freezer.h>
23 #include <linux/sched/mm.h>
25 #include <linux/sunrpc/clnt.h>
26 #include <linux/sunrpc/metrics.h>
30 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
31 #define RPCDBG_FACILITY RPCDBG_SCHED
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/sunrpc.h>
38 * RPC slabs and memory pools
40 #define RPC_BUFFER_MAXSIZE (2048)
41 #define RPC_BUFFER_POOLSIZE (8)
42 #define RPC_TASK_POOLSIZE (8)
43 static struct kmem_cache
*rpc_task_slabp __read_mostly
;
44 static struct kmem_cache
*rpc_buffer_slabp __read_mostly
;
45 static mempool_t
*rpc_task_mempool __read_mostly
;
46 static mempool_t
*rpc_buffer_mempool __read_mostly
;
48 static void rpc_async_schedule(struct work_struct
*);
49 static void rpc_release_task(struct rpc_task
*task
);
50 static void __rpc_queue_timer_fn(struct work_struct
*);
53 * RPC tasks sit here while waiting for conditions to improve.
55 static struct rpc_wait_queue delay_queue
;
58 * rpciod-related stuff
60 struct workqueue_struct
*rpciod_workqueue __read_mostly
;
61 struct workqueue_struct
*xprtiod_workqueue __read_mostly
;
62 EXPORT_SYMBOL_GPL(xprtiod_workqueue
);
65 rpc_task_timeout(const struct rpc_task
*task
)
67 unsigned long timeout
= READ_ONCE(task
->tk_timeout
);
70 unsigned long now
= jiffies
;
71 if (time_before(now
, timeout
))
76 EXPORT_SYMBOL_GPL(rpc_task_timeout
);
79 * Disable the timer for a given RPC task. Should be called with
80 * queue->lock and bh_disabled in order to avoid races within
84 __rpc_disable_timer(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
86 if (list_empty(&task
->u
.tk_wait
.timer_list
))
88 dprintk("RPC: %5u disabling timer\n", task
->tk_pid
);
90 list_del(&task
->u
.tk_wait
.timer_list
);
91 if (list_empty(&queue
->timer_list
.list
))
92 cancel_delayed_work(&queue
->timer_list
.dwork
);
96 rpc_set_queue_timer(struct rpc_wait_queue
*queue
, unsigned long expires
)
98 unsigned long now
= jiffies
;
99 queue
->timer_list
.expires
= expires
;
100 if (time_before_eq(expires
, now
))
104 mod_delayed_work(rpciod_workqueue
, &queue
->timer_list
.dwork
, expires
);
108 * Set up a timer for the current task.
111 __rpc_add_timer(struct rpc_wait_queue
*queue
, struct rpc_task
*task
,
112 unsigned long timeout
)
114 dprintk("RPC: %5u setting alarm for %u ms\n",
115 task
->tk_pid
, jiffies_to_msecs(timeout
- jiffies
));
117 task
->tk_timeout
= timeout
;
118 if (list_empty(&queue
->timer_list
.list
) || time_before(timeout
, queue
->timer_list
.expires
))
119 rpc_set_queue_timer(queue
, timeout
);
120 list_add(&task
->u
.tk_wait
.timer_list
, &queue
->timer_list
.list
);
123 static void rpc_set_waitqueue_priority(struct rpc_wait_queue
*queue
, int priority
)
125 if (queue
->priority
!= priority
) {
126 queue
->priority
= priority
;
127 queue
->nr
= 1U << priority
;
131 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue
*queue
)
133 rpc_set_waitqueue_priority(queue
, queue
->maxpriority
);
137 * Add a request to a queue list
140 __rpc_list_enqueue_task(struct list_head
*q
, struct rpc_task
*task
)
144 list_for_each_entry(t
, q
, u
.tk_wait
.list
) {
145 if (t
->tk_owner
== task
->tk_owner
) {
146 list_add_tail(&task
->u
.tk_wait
.links
,
147 &t
->u
.tk_wait
.links
);
148 /* Cache the queue head in task->u.tk_wait.list */
149 task
->u
.tk_wait
.list
.next
= q
;
150 task
->u
.tk_wait
.list
.prev
= NULL
;
154 INIT_LIST_HEAD(&task
->u
.tk_wait
.links
);
155 list_add_tail(&task
->u
.tk_wait
.list
, q
);
159 * Remove request from a queue list
162 __rpc_list_dequeue_task(struct rpc_task
*task
)
167 if (task
->u
.tk_wait
.list
.prev
== NULL
) {
168 list_del(&task
->u
.tk_wait
.links
);
171 if (!list_empty(&task
->u
.tk_wait
.links
)) {
172 t
= list_first_entry(&task
->u
.tk_wait
.links
,
175 /* Assume __rpc_list_enqueue_task() cached the queue head */
176 q
= t
->u
.tk_wait
.list
.next
;
177 list_add_tail(&t
->u
.tk_wait
.list
, q
);
178 list_del(&task
->u
.tk_wait
.links
);
180 list_del(&task
->u
.tk_wait
.list
);
184 * Add new request to a priority queue.
186 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue
*queue
,
187 struct rpc_task
*task
,
188 unsigned char queue_priority
)
190 if (unlikely(queue_priority
> queue
->maxpriority
))
191 queue_priority
= queue
->maxpriority
;
192 __rpc_list_enqueue_task(&queue
->tasks
[queue_priority
], task
);
196 * Add new request to wait queue.
198 * Swapper tasks always get inserted at the head of the queue.
199 * This should avoid many nasty memory deadlocks and hopefully
200 * improve overall performance.
201 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
203 static void __rpc_add_wait_queue(struct rpc_wait_queue
*queue
,
204 struct rpc_task
*task
,
205 unsigned char queue_priority
)
207 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
208 if (RPC_IS_QUEUED(task
))
211 INIT_LIST_HEAD(&task
->u
.tk_wait
.timer_list
);
212 if (RPC_IS_PRIORITY(queue
))
213 __rpc_add_wait_queue_priority(queue
, task
, queue_priority
);
214 else if (RPC_IS_SWAPPER(task
))
215 list_add(&task
->u
.tk_wait
.list
, &queue
->tasks
[0]);
217 list_add_tail(&task
->u
.tk_wait
.list
, &queue
->tasks
[0]);
218 task
->tk_waitqueue
= queue
;
220 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
222 rpc_set_queued(task
);
224 dprintk("RPC: %5u added to queue %p \"%s\"\n",
225 task
->tk_pid
, queue
, rpc_qname(queue
));
229 * Remove request from a priority queue.
231 static void __rpc_remove_wait_queue_priority(struct rpc_task
*task
)
233 __rpc_list_dequeue_task(task
);
237 * Remove request from queue.
238 * Note: must be called with spin lock held.
240 static void __rpc_remove_wait_queue(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
242 __rpc_disable_timer(queue
, task
);
243 if (RPC_IS_PRIORITY(queue
))
244 __rpc_remove_wait_queue_priority(task
);
246 list_del(&task
->u
.tk_wait
.list
);
248 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
249 task
->tk_pid
, queue
, rpc_qname(queue
));
252 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
, unsigned char nr_queues
)
256 spin_lock_init(&queue
->lock
);
257 for (i
= 0; i
< ARRAY_SIZE(queue
->tasks
); i
++)
258 INIT_LIST_HEAD(&queue
->tasks
[i
]);
259 queue
->maxpriority
= nr_queues
- 1;
260 rpc_reset_waitqueue_priority(queue
);
262 queue
->timer_list
.expires
= 0;
263 INIT_DEFERRABLE_WORK(&queue
->timer_list
.dwork
, __rpc_queue_timer_fn
);
264 INIT_LIST_HEAD(&queue
->timer_list
.list
);
265 rpc_assign_waitqueue_name(queue
, qname
);
268 void rpc_init_priority_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
)
270 __rpc_init_priority_wait_queue(queue
, qname
, RPC_NR_PRIORITY
);
272 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue
);
274 void rpc_init_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
)
276 __rpc_init_priority_wait_queue(queue
, qname
, 1);
278 EXPORT_SYMBOL_GPL(rpc_init_wait_queue
);
280 void rpc_destroy_wait_queue(struct rpc_wait_queue
*queue
)
282 cancel_delayed_work_sync(&queue
->timer_list
.dwork
);
284 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue
);
286 static int rpc_wait_bit_killable(struct wait_bit_key
*key
, int mode
)
288 freezable_schedule_unsafe();
289 if (signal_pending_state(mode
, current
))
294 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
295 static void rpc_task_set_debuginfo(struct rpc_task
*task
)
297 static atomic_t rpc_pid
;
299 task
->tk_pid
= atomic_inc_return(&rpc_pid
);
302 static inline void rpc_task_set_debuginfo(struct rpc_task
*task
)
307 static void rpc_set_active(struct rpc_task
*task
)
309 rpc_task_set_debuginfo(task
);
310 set_bit(RPC_TASK_ACTIVE
, &task
->tk_runstate
);
311 trace_rpc_task_begin(task
, NULL
);
315 * Mark an RPC call as having completed by clearing the 'active' bit
316 * and then waking up all tasks that were sleeping.
318 static int rpc_complete_task(struct rpc_task
*task
)
320 void *m
= &task
->tk_runstate
;
321 wait_queue_head_t
*wq
= bit_waitqueue(m
, RPC_TASK_ACTIVE
);
322 struct wait_bit_key k
= __WAIT_BIT_KEY_INITIALIZER(m
, RPC_TASK_ACTIVE
);
326 trace_rpc_task_complete(task
, NULL
);
328 spin_lock_irqsave(&wq
->lock
, flags
);
329 clear_bit(RPC_TASK_ACTIVE
, &task
->tk_runstate
);
330 ret
= atomic_dec_and_test(&task
->tk_count
);
331 if (waitqueue_active(wq
))
332 __wake_up_locked_key(wq
, TASK_NORMAL
, &k
);
333 spin_unlock_irqrestore(&wq
->lock
, flags
);
338 * Allow callers to wait for completion of an RPC call
340 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
341 * to enforce taking of the wq->lock and hence avoid races with
342 * rpc_complete_task().
344 int __rpc_wait_for_completion_task(struct rpc_task
*task
, wait_bit_action_f
*action
)
347 action
= rpc_wait_bit_killable
;
348 return out_of_line_wait_on_bit(&task
->tk_runstate
, RPC_TASK_ACTIVE
,
349 action
, TASK_KILLABLE
);
351 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task
);
354 * Make an RPC task runnable.
356 * Note: If the task is ASYNC, and is being made runnable after sitting on an
357 * rpc_wait_queue, this must be called with the queue spinlock held to protect
358 * the wait queue operation.
359 * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
360 * which is needed to ensure that __rpc_execute() doesn't loop (due to the
361 * lockless RPC_IS_QUEUED() test) before we've had a chance to test
362 * the RPC_TASK_RUNNING flag.
364 static void rpc_make_runnable(struct workqueue_struct
*wq
,
365 struct rpc_task
*task
)
367 bool need_wakeup
= !rpc_test_and_set_running(task
);
369 rpc_clear_queued(task
);
372 if (RPC_IS_ASYNC(task
)) {
373 INIT_WORK(&task
->u
.tk_work
, rpc_async_schedule
);
374 queue_work(wq
, &task
->u
.tk_work
);
376 wake_up_bit(&task
->tk_runstate
, RPC_TASK_QUEUED
);
380 * Prepare for sleeping on a wait queue.
381 * By always appending tasks to the list we ensure FIFO behavior.
382 * NB: An RPC task will only receive interrupt-driven events as long
383 * as it's on a wait queue.
385 static void __rpc_sleep_on_priority(struct rpc_wait_queue
*q
,
386 struct rpc_task
*task
,
387 unsigned char queue_priority
)
389 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
390 task
->tk_pid
, rpc_qname(q
), jiffies
);
392 trace_rpc_task_sleep(task
, q
);
394 __rpc_add_wait_queue(q
, task
, queue_priority
);
398 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue
*q
,
399 struct rpc_task
*task
, unsigned long timeout
,
400 unsigned char queue_priority
)
402 if (time_is_after_jiffies(timeout
)) {
403 __rpc_sleep_on_priority(q
, task
, queue_priority
);
404 __rpc_add_timer(q
, task
, timeout
);
406 task
->tk_status
= -ETIMEDOUT
;
409 static void rpc_set_tk_callback(struct rpc_task
*task
, rpc_action action
)
411 if (action
&& !WARN_ON_ONCE(task
->tk_callback
!= NULL
))
412 task
->tk_callback
= action
;
415 static bool rpc_sleep_check_activated(struct rpc_task
*task
)
417 /* We shouldn't ever put an inactive task to sleep */
418 if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task
))) {
419 task
->tk_status
= -EIO
;
420 rpc_put_task_async(task
);
426 void rpc_sleep_on_timeout(struct rpc_wait_queue
*q
, struct rpc_task
*task
,
427 rpc_action action
, unsigned long timeout
)
429 if (!rpc_sleep_check_activated(task
))
432 rpc_set_tk_callback(task
, action
);
435 * Protect the queue operations.
438 __rpc_sleep_on_priority_timeout(q
, task
, timeout
, task
->tk_priority
);
439 spin_unlock(&q
->lock
);
441 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout
);
443 void rpc_sleep_on(struct rpc_wait_queue
*q
, struct rpc_task
*task
,
446 if (!rpc_sleep_check_activated(task
))
449 rpc_set_tk_callback(task
, action
);
451 WARN_ON_ONCE(task
->tk_timeout
!= 0);
453 * Protect the queue operations.
456 __rpc_sleep_on_priority(q
, task
, task
->tk_priority
);
457 spin_unlock(&q
->lock
);
459 EXPORT_SYMBOL_GPL(rpc_sleep_on
);
461 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue
*q
,
462 struct rpc_task
*task
, unsigned long timeout
, int priority
)
464 if (!rpc_sleep_check_activated(task
))
467 priority
-= RPC_PRIORITY_LOW
;
469 * Protect the queue operations.
472 __rpc_sleep_on_priority_timeout(q
, task
, timeout
, priority
);
473 spin_unlock(&q
->lock
);
475 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout
);
477 void rpc_sleep_on_priority(struct rpc_wait_queue
*q
, struct rpc_task
*task
,
480 if (!rpc_sleep_check_activated(task
))
483 WARN_ON_ONCE(task
->tk_timeout
!= 0);
484 priority
-= RPC_PRIORITY_LOW
;
486 * Protect the queue operations.
489 __rpc_sleep_on_priority(q
, task
, priority
);
490 spin_unlock(&q
->lock
);
492 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority
);
495 * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
496 * @wq: workqueue on which to run task
498 * @task: task to be woken up
500 * Caller must hold queue->lock, and have cleared the task queued flag.
502 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct
*wq
,
503 struct rpc_wait_queue
*queue
,
504 struct rpc_task
*task
)
506 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
507 task
->tk_pid
, jiffies
);
509 /* Has the task been executed yet? If not, we cannot wake it up! */
510 if (!RPC_IS_ACTIVATED(task
)) {
511 printk(KERN_ERR
"RPC: Inactive task (%p) being woken up!\n", task
);
515 trace_rpc_task_wakeup(task
, queue
);
517 __rpc_remove_wait_queue(queue
, task
);
519 rpc_make_runnable(wq
, task
);
521 dprintk("RPC: __rpc_wake_up_task done\n");
525 * Wake up a queued task while the queue lock is being held
527 static struct rpc_task
*
528 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct
*wq
,
529 struct rpc_wait_queue
*queue
, struct rpc_task
*task
,
530 bool (*action
)(struct rpc_task
*, void *), void *data
)
532 if (RPC_IS_QUEUED(task
)) {
534 if (task
->tk_waitqueue
== queue
) {
535 if (action
== NULL
|| action(task
, data
)) {
536 __rpc_do_wake_up_task_on_wq(wq
, queue
, task
);
545 rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct
*wq
,
546 struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
548 rpc_wake_up_task_on_wq_queue_action_locked(wq
, queue
, task
, NULL
, NULL
);
552 * Wake up a queued task while the queue lock is being held
554 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
556 rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue
, queue
, task
);
560 * Wake up a task on a specific queue
562 void rpc_wake_up_queued_task_on_wq(struct workqueue_struct
*wq
,
563 struct rpc_wait_queue
*queue
,
564 struct rpc_task
*task
)
566 if (!RPC_IS_QUEUED(task
))
568 spin_lock(&queue
->lock
);
569 rpc_wake_up_task_on_wq_queue_locked(wq
, queue
, task
);
570 spin_unlock(&queue
->lock
);
574 * Wake up a task on a specific queue
576 void rpc_wake_up_queued_task(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
578 if (!RPC_IS_QUEUED(task
))
580 spin_lock(&queue
->lock
);
581 rpc_wake_up_task_queue_locked(queue
, task
);
582 spin_unlock(&queue
->lock
);
584 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task
);
586 static bool rpc_task_action_set_status(struct rpc_task
*task
, void *status
)
588 task
->tk_status
= *(int *)status
;
593 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue
*queue
,
594 struct rpc_task
*task
, int status
)
596 rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue
, queue
,
597 task
, rpc_task_action_set_status
, &status
);
601 * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
602 * @queue: pointer to rpc_wait_queue
603 * @task: pointer to rpc_task
604 * @status: integer error value
606 * If @task is queued on @queue, then it is woken up, and @task->tk_status is
607 * set to the value of @status.
610 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue
*queue
,
611 struct rpc_task
*task
, int status
)
613 if (!RPC_IS_QUEUED(task
))
615 spin_lock(&queue
->lock
);
616 rpc_wake_up_task_queue_set_status_locked(queue
, task
, status
);
617 spin_unlock(&queue
->lock
);
621 * Wake up the next task on a priority queue.
623 static struct rpc_task
*__rpc_find_next_queued_priority(struct rpc_wait_queue
*queue
)
626 struct rpc_task
*task
;
629 * Service a batch of tasks from a single owner.
631 q
= &queue
->tasks
[queue
->priority
];
632 if (!list_empty(q
) && --queue
->nr
) {
633 task
= list_first_entry(q
, struct rpc_task
, u
.tk_wait
.list
);
638 * Service the next queue.
641 if (q
== &queue
->tasks
[0])
642 q
= &queue
->tasks
[queue
->maxpriority
];
645 if (!list_empty(q
)) {
646 task
= list_first_entry(q
, struct rpc_task
, u
.tk_wait
.list
);
649 } while (q
!= &queue
->tasks
[queue
->priority
]);
651 rpc_reset_waitqueue_priority(queue
);
655 rpc_set_waitqueue_priority(queue
, (unsigned int)(q
- &queue
->tasks
[0]));
660 static struct rpc_task
*__rpc_find_next_queued(struct rpc_wait_queue
*queue
)
662 if (RPC_IS_PRIORITY(queue
))
663 return __rpc_find_next_queued_priority(queue
);
664 if (!list_empty(&queue
->tasks
[0]))
665 return list_first_entry(&queue
->tasks
[0], struct rpc_task
, u
.tk_wait
.list
);
670 * Wake up the first task on the wait queue.
672 struct rpc_task
*rpc_wake_up_first_on_wq(struct workqueue_struct
*wq
,
673 struct rpc_wait_queue
*queue
,
674 bool (*func
)(struct rpc_task
*, void *), void *data
)
676 struct rpc_task
*task
= NULL
;
678 dprintk("RPC: wake_up_first(%p \"%s\")\n",
679 queue
, rpc_qname(queue
));
680 spin_lock(&queue
->lock
);
681 task
= __rpc_find_next_queued(queue
);
683 task
= rpc_wake_up_task_on_wq_queue_action_locked(wq
, queue
,
685 spin_unlock(&queue
->lock
);
691 * Wake up the first task on the wait queue.
693 struct rpc_task
*rpc_wake_up_first(struct rpc_wait_queue
*queue
,
694 bool (*func
)(struct rpc_task
*, void *), void *data
)
696 return rpc_wake_up_first_on_wq(rpciod_workqueue
, queue
, func
, data
);
698 EXPORT_SYMBOL_GPL(rpc_wake_up_first
);
700 static bool rpc_wake_up_next_func(struct rpc_task
*task
, void *data
)
706 * Wake up the next task on the wait queue.
708 struct rpc_task
*rpc_wake_up_next(struct rpc_wait_queue
*queue
)
710 return rpc_wake_up_first(queue
, rpc_wake_up_next_func
, NULL
);
712 EXPORT_SYMBOL_GPL(rpc_wake_up_next
);
715 * rpc_wake_up - wake up all rpc_tasks
716 * @queue: rpc_wait_queue on which the tasks are sleeping
720 void rpc_wake_up(struct rpc_wait_queue
*queue
)
722 struct list_head
*head
;
724 spin_lock(&queue
->lock
);
725 head
= &queue
->tasks
[queue
->maxpriority
];
727 while (!list_empty(head
)) {
728 struct rpc_task
*task
;
729 task
= list_first_entry(head
,
732 rpc_wake_up_task_queue_locked(queue
, task
);
734 if (head
== &queue
->tasks
[0])
738 spin_unlock(&queue
->lock
);
740 EXPORT_SYMBOL_GPL(rpc_wake_up
);
743 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
744 * @queue: rpc_wait_queue on which the tasks are sleeping
745 * @status: status value to set
749 void rpc_wake_up_status(struct rpc_wait_queue
*queue
, int status
)
751 struct list_head
*head
;
753 spin_lock(&queue
->lock
);
754 head
= &queue
->tasks
[queue
->maxpriority
];
756 while (!list_empty(head
)) {
757 struct rpc_task
*task
;
758 task
= list_first_entry(head
,
761 task
->tk_status
= status
;
762 rpc_wake_up_task_queue_locked(queue
, task
);
764 if (head
== &queue
->tasks
[0])
768 spin_unlock(&queue
->lock
);
770 EXPORT_SYMBOL_GPL(rpc_wake_up_status
);
772 static void __rpc_queue_timer_fn(struct work_struct
*work
)
774 struct rpc_wait_queue
*queue
= container_of(work
,
775 struct rpc_wait_queue
,
776 timer_list
.dwork
.work
);
777 struct rpc_task
*task
, *n
;
778 unsigned long expires
, now
, timeo
;
780 spin_lock(&queue
->lock
);
781 expires
= now
= jiffies
;
782 list_for_each_entry_safe(task
, n
, &queue
->timer_list
.list
, u
.tk_wait
.timer_list
) {
783 timeo
= task
->tk_timeout
;
784 if (time_after_eq(now
, timeo
)) {
785 dprintk("RPC: %5u timeout\n", task
->tk_pid
);
786 task
->tk_status
= -ETIMEDOUT
;
787 rpc_wake_up_task_queue_locked(queue
, task
);
790 if (expires
== now
|| time_after(expires
, timeo
))
793 if (!list_empty(&queue
->timer_list
.list
))
794 rpc_set_queue_timer(queue
, expires
);
795 spin_unlock(&queue
->lock
);
798 static void __rpc_atrun(struct rpc_task
*task
)
800 if (task
->tk_status
== -ETIMEDOUT
)
805 * Run a task at a later time
807 void rpc_delay(struct rpc_task
*task
, unsigned long delay
)
809 rpc_sleep_on_timeout(&delay_queue
, task
, __rpc_atrun
, jiffies
+ delay
);
811 EXPORT_SYMBOL_GPL(rpc_delay
);
814 * Helper to call task->tk_ops->rpc_call_prepare
816 void rpc_prepare_task(struct rpc_task
*task
)
818 task
->tk_ops
->rpc_call_prepare(task
, task
->tk_calldata
);
822 rpc_init_task_statistics(struct rpc_task
*task
)
824 /* Initialize retry counters */
825 task
->tk_garb_retry
= 2;
826 task
->tk_cred_retry
= 2;
827 task
->tk_rebind_retry
= 2;
829 /* starting timestamp */
830 task
->tk_start
= ktime_get();
834 rpc_reset_task_statistics(struct rpc_task
*task
)
836 task
->tk_timeouts
= 0;
837 task
->tk_flags
&= ~(RPC_CALL_MAJORSEEN
|RPC_TASK_SENT
);
838 rpc_init_task_statistics(task
);
842 * Helper that calls task->tk_ops->rpc_call_done if it exists
844 void rpc_exit_task(struct rpc_task
*task
)
846 task
->tk_action
= NULL
;
847 if (task
->tk_ops
->rpc_count_stats
)
848 task
->tk_ops
->rpc_count_stats(task
, task
->tk_calldata
);
849 else if (task
->tk_client
)
850 rpc_count_iostats(task
, task
->tk_client
->cl_metrics
);
851 if (task
->tk_ops
->rpc_call_done
!= NULL
) {
852 task
->tk_ops
->rpc_call_done(task
, task
->tk_calldata
);
853 if (task
->tk_action
!= NULL
) {
854 /* Always release the RPC slot and buffer memory */
856 rpc_reset_task_statistics(task
);
861 void rpc_signal_task(struct rpc_task
*task
)
863 struct rpc_wait_queue
*queue
;
865 if (!RPC_IS_ACTIVATED(task
))
867 set_bit(RPC_TASK_SIGNALLED
, &task
->tk_runstate
);
868 smp_mb__after_atomic();
869 queue
= READ_ONCE(task
->tk_waitqueue
);
871 rpc_wake_up_queued_task_set_status(queue
, task
, -ERESTARTSYS
);
874 void rpc_exit(struct rpc_task
*task
, int status
)
876 task
->tk_status
= status
;
877 task
->tk_action
= rpc_exit_task
;
878 rpc_wake_up_queued_task(task
->tk_waitqueue
, task
);
880 EXPORT_SYMBOL_GPL(rpc_exit
);
882 void rpc_release_calldata(const struct rpc_call_ops
*ops
, void *calldata
)
884 if (ops
->rpc_release
!= NULL
)
885 ops
->rpc_release(calldata
);
889 * This is the RPC `scheduler' (or rather, the finite state machine).
891 static void __rpc_execute(struct rpc_task
*task
)
893 struct rpc_wait_queue
*queue
;
894 int task_is_async
= RPC_IS_ASYNC(task
);
897 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
898 task
->tk_pid
, task
->tk_flags
);
900 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
901 if (RPC_IS_QUEUED(task
))
905 void (*do_action
)(struct rpc_task
*);
908 * Perform the next FSM step or a pending callback.
910 * tk_action may be NULL if the task has been killed.
911 * In particular, note that rpc_killall_tasks may
912 * do this at any time, so beware when dereferencing.
914 do_action
= task
->tk_action
;
915 if (task
->tk_callback
) {
916 do_action
= task
->tk_callback
;
917 task
->tk_callback
= NULL
;
921 trace_rpc_task_run_action(task
, do_action
);
925 * Lockless check for whether task is sleeping or not.
927 if (!RPC_IS_QUEUED(task
))
931 * Signalled tasks should exit rather than sleep.
933 if (RPC_SIGNALLED(task
))
934 rpc_exit(task
, -ERESTARTSYS
);
937 * The queue->lock protects against races with
938 * rpc_make_runnable().
940 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
941 * rpc_task, rpc_make_runnable() can assign it to a
942 * different workqueue. We therefore cannot assume that the
943 * rpc_task pointer may still be dereferenced.
945 queue
= task
->tk_waitqueue
;
946 spin_lock(&queue
->lock
);
947 if (!RPC_IS_QUEUED(task
)) {
948 spin_unlock(&queue
->lock
);
951 rpc_clear_running(task
);
952 spin_unlock(&queue
->lock
);
956 /* sync task: sleep here */
957 dprintk("RPC: %5u sync task going to sleep\n", task
->tk_pid
);
958 status
= out_of_line_wait_on_bit(&task
->tk_runstate
,
959 RPC_TASK_QUEUED
, rpc_wait_bit_killable
,
963 * When a sync task receives a signal, it exits with
964 * -ERESTARTSYS. In order to catch any callbacks that
965 * clean up after sleeping on some queue, we don't
966 * break the loop here, but go around once more.
968 dprintk("RPC: %5u got signal\n", task
->tk_pid
);
969 set_bit(RPC_TASK_SIGNALLED
, &task
->tk_runstate
);
970 rpc_exit(task
, -ERESTARTSYS
);
972 dprintk("RPC: %5u sync task resuming\n", task
->tk_pid
);
975 dprintk("RPC: %5u return %d, status %d\n", task
->tk_pid
, status
,
977 /* Release all resources associated with the task */
978 rpc_release_task(task
);
982 * User-visible entry point to the scheduler.
984 * This may be called recursively if e.g. an async NFS task updates
985 * the attributes and finds that dirty pages must be flushed.
986 * NOTE: Upon exit of this function the task is guaranteed to be
987 * released. In particular note that tk_release() will have
988 * been called, so your task memory may have been freed.
990 void rpc_execute(struct rpc_task
*task
)
992 bool is_async
= RPC_IS_ASYNC(task
);
994 rpc_set_active(task
);
995 rpc_make_runnable(rpciod_workqueue
, task
);
1000 static void rpc_async_schedule(struct work_struct
*work
)
1002 unsigned int pflags
= memalloc_nofs_save();
1004 __rpc_execute(container_of(work
, struct rpc_task
, u
.tk_work
));
1005 memalloc_nofs_restore(pflags
);
1009 * rpc_malloc - allocate RPC buffer resources
1012 * A single memory region is allocated, which is split between the
1013 * RPC call and RPC reply that this task is being used for. When
1014 * this RPC is retired, the memory is released by calling rpc_free.
1016 * To prevent rpciod from hanging, this allocator never sleeps,
1017 * returning -ENOMEM and suppressing warning if the request cannot
1018 * be serviced immediately. The caller can arrange to sleep in a
1019 * way that is safe for rpciod.
1021 * Most requests are 'small' (under 2KiB) and can be serviced from a
1022 * mempool, ensuring that NFS reads and writes can always proceed,
1023 * and that there is good locality of reference for these buffers.
1025 int rpc_malloc(struct rpc_task
*task
)
1027 struct rpc_rqst
*rqst
= task
->tk_rqstp
;
1028 size_t size
= rqst
->rq_callsize
+ rqst
->rq_rcvsize
;
1029 struct rpc_buffer
*buf
;
1030 gfp_t gfp
= GFP_NOFS
;
1032 if (RPC_IS_SWAPPER(task
))
1033 gfp
= __GFP_MEMALLOC
| GFP_NOWAIT
| __GFP_NOWARN
;
1035 size
+= sizeof(struct rpc_buffer
);
1036 if (size
<= RPC_BUFFER_MAXSIZE
)
1037 buf
= mempool_alloc(rpc_buffer_mempool
, gfp
);
1039 buf
= kmalloc(size
, gfp
);
1045 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1046 task
->tk_pid
, size
, buf
);
1047 rqst
->rq_buffer
= buf
->data
;
1048 rqst
->rq_rbuffer
= (char *)rqst
->rq_buffer
+ rqst
->rq_callsize
;
1051 EXPORT_SYMBOL_GPL(rpc_malloc
);
1054 * rpc_free - free RPC buffer resources allocated via rpc_malloc
1058 void rpc_free(struct rpc_task
*task
)
1060 void *buffer
= task
->tk_rqstp
->rq_buffer
;
1062 struct rpc_buffer
*buf
;
1064 buf
= container_of(buffer
, struct rpc_buffer
, data
);
1067 dprintk("RPC: freeing buffer of size %zu at %p\n",
1070 if (size
<= RPC_BUFFER_MAXSIZE
)
1071 mempool_free(buf
, rpc_buffer_mempool
);
1075 EXPORT_SYMBOL_GPL(rpc_free
);
1078 * Creation and deletion of RPC task structures
1080 static void rpc_init_task(struct rpc_task
*task
, const struct rpc_task_setup
*task_setup_data
)
1082 memset(task
, 0, sizeof(*task
));
1083 atomic_set(&task
->tk_count
, 1);
1084 task
->tk_flags
= task_setup_data
->flags
;
1085 task
->tk_ops
= task_setup_data
->callback_ops
;
1086 task
->tk_calldata
= task_setup_data
->callback_data
;
1087 INIT_LIST_HEAD(&task
->tk_task
);
1089 task
->tk_priority
= task_setup_data
->priority
- RPC_PRIORITY_LOW
;
1090 task
->tk_owner
= current
->tgid
;
1092 /* Initialize workqueue for async tasks */
1093 task
->tk_workqueue
= task_setup_data
->workqueue
;
1095 task
->tk_xprt
= xprt_get(task_setup_data
->rpc_xprt
);
1097 task
->tk_op_cred
= get_rpccred(task_setup_data
->rpc_op_cred
);
1099 if (task
->tk_ops
->rpc_call_prepare
!= NULL
)
1100 task
->tk_action
= rpc_prepare_task
;
1102 rpc_init_task_statistics(task
);
1104 dprintk("RPC: new task initialized, procpid %u\n",
1105 task_pid_nr(current
));
1108 static struct rpc_task
*
1109 rpc_alloc_task(void)
1111 return (struct rpc_task
*)mempool_alloc(rpc_task_mempool
, GFP_NOFS
);
1115 * Create a new task for the specified client.
1117 struct rpc_task
*rpc_new_task(const struct rpc_task_setup
*setup_data
)
1119 struct rpc_task
*task
= setup_data
->task
;
1120 unsigned short flags
= 0;
1123 task
= rpc_alloc_task();
1124 flags
= RPC_TASK_DYNAMIC
;
1127 rpc_init_task(task
, setup_data
);
1128 task
->tk_flags
|= flags
;
1129 dprintk("RPC: allocated task %p\n", task
);
1134 * rpc_free_task - release rpc task and perform cleanups
1136 * Note that we free up the rpc_task _after_ rpc_release_calldata()
1137 * in order to work around a workqueue dependency issue.
1140 * "Workqueue currently considers two work items to be the same if they're
1141 * on the same address and won't execute them concurrently - ie. it
1142 * makes a work item which is queued again while being executed wait
1143 * for the previous execution to complete.
1145 * If a work function frees the work item, and then waits for an event
1146 * which should be performed by another work item and *that* work item
1147 * recycles the freed work item, it can create a false dependency loop.
1148 * There really is no reliable way to detect this short of verifying
1149 * every memory free."
1152 static void rpc_free_task(struct rpc_task
*task
)
1154 unsigned short tk_flags
= task
->tk_flags
;
1156 put_rpccred(task
->tk_op_cred
);
1157 rpc_release_calldata(task
->tk_ops
, task
->tk_calldata
);
1159 if (tk_flags
& RPC_TASK_DYNAMIC
) {
1160 dprintk("RPC: %5u freeing task\n", task
->tk_pid
);
1161 mempool_free(task
, rpc_task_mempool
);
1165 static void rpc_async_release(struct work_struct
*work
)
1167 unsigned int pflags
= memalloc_nofs_save();
1169 rpc_free_task(container_of(work
, struct rpc_task
, u
.tk_work
));
1170 memalloc_nofs_restore(pflags
);
1173 static void rpc_release_resources_task(struct rpc_task
*task
)
1176 if (task
->tk_msg
.rpc_cred
) {
1177 put_cred(task
->tk_msg
.rpc_cred
);
1178 task
->tk_msg
.rpc_cred
= NULL
;
1180 rpc_task_release_client(task
);
1183 static void rpc_final_put_task(struct rpc_task
*task
,
1184 struct workqueue_struct
*q
)
1187 INIT_WORK(&task
->u
.tk_work
, rpc_async_release
);
1188 queue_work(q
, &task
->u
.tk_work
);
1190 rpc_free_task(task
);
1193 static void rpc_do_put_task(struct rpc_task
*task
, struct workqueue_struct
*q
)
1195 if (atomic_dec_and_test(&task
->tk_count
)) {
1196 rpc_release_resources_task(task
);
1197 rpc_final_put_task(task
, q
);
1201 void rpc_put_task(struct rpc_task
*task
)
1203 rpc_do_put_task(task
, NULL
);
1205 EXPORT_SYMBOL_GPL(rpc_put_task
);
1207 void rpc_put_task_async(struct rpc_task
*task
)
1209 rpc_do_put_task(task
, task
->tk_workqueue
);
1211 EXPORT_SYMBOL_GPL(rpc_put_task_async
);
1213 static void rpc_release_task(struct rpc_task
*task
)
1215 dprintk("RPC: %5u release task\n", task
->tk_pid
);
1217 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
1219 rpc_release_resources_task(task
);
1222 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1223 * so it should be safe to use task->tk_count as a test for whether
1224 * or not any other processes still hold references to our rpc_task.
1226 if (atomic_read(&task
->tk_count
) != 1 + !RPC_IS_ASYNC(task
)) {
1227 /* Wake up anyone who may be waiting for task completion */
1228 if (!rpc_complete_task(task
))
1231 if (!atomic_dec_and_test(&task
->tk_count
))
1234 rpc_final_put_task(task
, task
->tk_workqueue
);
1239 return try_module_get(THIS_MODULE
) ? 0 : -EINVAL
;
1242 void rpciod_down(void)
1244 module_put(THIS_MODULE
);
1248 * Start up the rpciod workqueue.
1250 static int rpciod_start(void)
1252 struct workqueue_struct
*wq
;
1255 * Create the rpciod thread and wait for it to start.
1257 dprintk("RPC: creating workqueue rpciod\n");
1258 wq
= alloc_workqueue("rpciod", WQ_MEM_RECLAIM
| WQ_UNBOUND
, 0);
1261 rpciod_workqueue
= wq
;
1262 /* Note: highpri because network receive is latency sensitive */
1263 wq
= alloc_workqueue("xprtiod", WQ_UNBOUND
|WQ_MEM_RECLAIM
|WQ_HIGHPRI
, 0);
1266 xprtiod_workqueue
= wq
;
1269 wq
= rpciod_workqueue
;
1270 rpciod_workqueue
= NULL
;
1271 destroy_workqueue(wq
);
1276 static void rpciod_stop(void)
1278 struct workqueue_struct
*wq
= NULL
;
1280 if (rpciod_workqueue
== NULL
)
1282 dprintk("RPC: destroying workqueue rpciod\n");
1284 wq
= rpciod_workqueue
;
1285 rpciod_workqueue
= NULL
;
1286 destroy_workqueue(wq
);
1287 wq
= xprtiod_workqueue
;
1288 xprtiod_workqueue
= NULL
;
1289 destroy_workqueue(wq
);
1293 rpc_destroy_mempool(void)
1296 mempool_destroy(rpc_buffer_mempool
);
1297 mempool_destroy(rpc_task_mempool
);
1298 kmem_cache_destroy(rpc_task_slabp
);
1299 kmem_cache_destroy(rpc_buffer_slabp
);
1300 rpc_destroy_wait_queue(&delay_queue
);
1304 rpc_init_mempool(void)
1307 * The following is not strictly a mempool initialisation,
1308 * but there is no harm in doing it here
1310 rpc_init_wait_queue(&delay_queue
, "delayq");
1311 if (!rpciod_start())
1314 rpc_task_slabp
= kmem_cache_create("rpc_tasks",
1315 sizeof(struct rpc_task
),
1316 0, SLAB_HWCACHE_ALIGN
,
1318 if (!rpc_task_slabp
)
1320 rpc_buffer_slabp
= kmem_cache_create("rpc_buffers",
1322 0, SLAB_HWCACHE_ALIGN
,
1324 if (!rpc_buffer_slabp
)
1326 rpc_task_mempool
= mempool_create_slab_pool(RPC_TASK_POOLSIZE
,
1328 if (!rpc_task_mempool
)
1330 rpc_buffer_mempool
= mempool_create_slab_pool(RPC_BUFFER_POOLSIZE
,
1332 if (!rpc_buffer_mempool
)
1336 rpc_destroy_mempool();