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>
23 #include <linux/sunrpc/clnt.h>
28 #define RPCDBG_FACILITY RPCDBG_SCHED
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/sunrpc.h>
35 * RPC slabs and memory pools
37 #define RPC_BUFFER_MAXSIZE (2048)
38 #define RPC_BUFFER_POOLSIZE (8)
39 #define RPC_TASK_POOLSIZE (8)
40 static struct kmem_cache
*rpc_task_slabp __read_mostly
;
41 static struct kmem_cache
*rpc_buffer_slabp __read_mostly
;
42 static mempool_t
*rpc_task_mempool __read_mostly
;
43 static mempool_t
*rpc_buffer_mempool __read_mostly
;
45 static void rpc_async_schedule(struct work_struct
*);
46 static void rpc_release_task(struct rpc_task
*task
);
47 static void __rpc_queue_timer_fn(unsigned long ptr
);
50 * RPC tasks sit here while waiting for conditions to improve.
52 static struct rpc_wait_queue delay_queue
;
55 * rpciod-related stuff
57 struct workqueue_struct
*rpciod_workqueue
;
60 * Disable the timer for a given RPC task. Should be called with
61 * queue->lock and bh_disabled in order to avoid races within
65 __rpc_disable_timer(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
67 if (task
->tk_timeout
== 0)
69 dprintk("RPC: %5u disabling timer\n", task
->tk_pid
);
71 list_del(&task
->u
.tk_wait
.timer_list
);
72 if (list_empty(&queue
->timer_list
.list
))
73 del_timer(&queue
->timer_list
.timer
);
77 rpc_set_queue_timer(struct rpc_wait_queue
*queue
, unsigned long expires
)
79 queue
->timer_list
.expires
= expires
;
80 mod_timer(&queue
->timer_list
.timer
, expires
);
84 * Set up a timer for the current task.
87 __rpc_add_timer(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
89 if (!task
->tk_timeout
)
92 dprintk("RPC: %5u setting alarm for %lu ms\n",
93 task
->tk_pid
, task
->tk_timeout
* 1000 / HZ
);
95 task
->u
.tk_wait
.expires
= jiffies
+ task
->tk_timeout
;
96 if (list_empty(&queue
->timer_list
.list
) || time_before(task
->u
.tk_wait
.expires
, queue
->timer_list
.expires
))
97 rpc_set_queue_timer(queue
, task
->u
.tk_wait
.expires
);
98 list_add(&task
->u
.tk_wait
.timer_list
, &queue
->timer_list
.list
);
101 static void rpc_rotate_queue_owner(struct rpc_wait_queue
*queue
)
103 struct list_head
*q
= &queue
->tasks
[queue
->priority
];
104 struct rpc_task
*task
;
106 if (!list_empty(q
)) {
107 task
= list_first_entry(q
, struct rpc_task
, u
.tk_wait
.list
);
108 if (task
->tk_owner
== queue
->owner
)
109 list_move_tail(&task
->u
.tk_wait
.list
, q
);
113 static void rpc_set_waitqueue_priority(struct rpc_wait_queue
*queue
, int priority
)
115 if (queue
->priority
!= priority
) {
116 /* Fairness: rotate the list when changing priority */
117 rpc_rotate_queue_owner(queue
);
118 queue
->priority
= priority
;
122 static void rpc_set_waitqueue_owner(struct rpc_wait_queue
*queue
, pid_t pid
)
125 queue
->nr
= RPC_BATCH_COUNT
;
128 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue
*queue
)
130 rpc_set_waitqueue_priority(queue
, queue
->maxpriority
);
131 rpc_set_waitqueue_owner(queue
, 0);
135 * Add new request to a priority queue.
137 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue
*queue
,
138 struct rpc_task
*task
,
139 unsigned char queue_priority
)
144 INIT_LIST_HEAD(&task
->u
.tk_wait
.links
);
145 if (unlikely(queue_priority
> queue
->maxpriority
))
146 queue_priority
= queue
->maxpriority
;
147 if (queue_priority
> queue
->priority
)
148 rpc_set_waitqueue_priority(queue
, queue_priority
);
149 q
= &queue
->tasks
[queue_priority
];
150 list_for_each_entry(t
, q
, u
.tk_wait
.list
) {
151 if (t
->tk_owner
== task
->tk_owner
) {
152 list_add_tail(&task
->u
.tk_wait
.list
, &t
->u
.tk_wait
.links
);
156 list_add_tail(&task
->u
.tk_wait
.list
, q
);
160 * Add new request to wait queue.
162 * Swapper tasks always get inserted at the head of the queue.
163 * This should avoid many nasty memory deadlocks and hopefully
164 * improve overall performance.
165 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
167 static void __rpc_add_wait_queue(struct rpc_wait_queue
*queue
,
168 struct rpc_task
*task
,
169 unsigned char queue_priority
)
171 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
172 if (RPC_IS_QUEUED(task
))
175 if (RPC_IS_PRIORITY(queue
))
176 __rpc_add_wait_queue_priority(queue
, task
, queue_priority
);
177 else if (RPC_IS_SWAPPER(task
))
178 list_add(&task
->u
.tk_wait
.list
, &queue
->tasks
[0]);
180 list_add_tail(&task
->u
.tk_wait
.list
, &queue
->tasks
[0]);
181 task
->tk_waitqueue
= queue
;
183 /* barrier matches the read in rpc_wake_up_task_queue_locked() */
185 rpc_set_queued(task
);
187 dprintk("RPC: %5u added to queue %p \"%s\"\n",
188 task
->tk_pid
, queue
, rpc_qname(queue
));
192 * Remove request from a priority queue.
194 static void __rpc_remove_wait_queue_priority(struct rpc_task
*task
)
198 if (!list_empty(&task
->u
.tk_wait
.links
)) {
199 t
= list_entry(task
->u
.tk_wait
.links
.next
, struct rpc_task
, u
.tk_wait
.list
);
200 list_move(&t
->u
.tk_wait
.list
, &task
->u
.tk_wait
.list
);
201 list_splice_init(&task
->u
.tk_wait
.links
, &t
->u
.tk_wait
.links
);
206 * Remove request from queue.
207 * Note: must be called with spin lock held.
209 static void __rpc_remove_wait_queue(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
211 __rpc_disable_timer(queue
, task
);
212 if (RPC_IS_PRIORITY(queue
))
213 __rpc_remove_wait_queue_priority(task
);
214 list_del(&task
->u
.tk_wait
.list
);
216 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
217 task
->tk_pid
, queue
, rpc_qname(queue
));
220 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
, unsigned char nr_queues
)
224 spin_lock_init(&queue
->lock
);
225 for (i
= 0; i
< ARRAY_SIZE(queue
->tasks
); i
++)
226 INIT_LIST_HEAD(&queue
->tasks
[i
]);
227 queue
->maxpriority
= nr_queues
- 1;
228 rpc_reset_waitqueue_priority(queue
);
230 setup_timer(&queue
->timer_list
.timer
, __rpc_queue_timer_fn
, (unsigned long)queue
);
231 INIT_LIST_HEAD(&queue
->timer_list
.list
);
232 rpc_assign_waitqueue_name(queue
, qname
);
235 void rpc_init_priority_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
)
237 __rpc_init_priority_wait_queue(queue
, qname
, RPC_NR_PRIORITY
);
239 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue
);
241 void rpc_init_wait_queue(struct rpc_wait_queue
*queue
, const char *qname
)
243 __rpc_init_priority_wait_queue(queue
, qname
, 1);
245 EXPORT_SYMBOL_GPL(rpc_init_wait_queue
);
247 void rpc_destroy_wait_queue(struct rpc_wait_queue
*queue
)
249 del_timer_sync(&queue
->timer_list
.timer
);
251 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue
);
253 static int rpc_wait_bit_killable(void *word
)
255 if (fatal_signal_pending(current
))
257 freezable_schedule();
262 static void rpc_task_set_debuginfo(struct rpc_task
*task
)
264 static atomic_t rpc_pid
;
266 task
->tk_pid
= atomic_inc_return(&rpc_pid
);
269 static inline void rpc_task_set_debuginfo(struct rpc_task
*task
)
274 static void rpc_set_active(struct rpc_task
*task
)
276 trace_rpc_task_begin(task
->tk_client
, task
, NULL
);
278 rpc_task_set_debuginfo(task
);
279 set_bit(RPC_TASK_ACTIVE
, &task
->tk_runstate
);
283 * Mark an RPC call as having completed by clearing the 'active' bit
284 * and then waking up all tasks that were sleeping.
286 static int rpc_complete_task(struct rpc_task
*task
)
288 void *m
= &task
->tk_runstate
;
289 wait_queue_head_t
*wq
= bit_waitqueue(m
, RPC_TASK_ACTIVE
);
290 struct wait_bit_key k
= __WAIT_BIT_KEY_INITIALIZER(m
, RPC_TASK_ACTIVE
);
294 trace_rpc_task_complete(task
->tk_client
, task
, NULL
);
296 spin_lock_irqsave(&wq
->lock
, flags
);
297 clear_bit(RPC_TASK_ACTIVE
, &task
->tk_runstate
);
298 ret
= atomic_dec_and_test(&task
->tk_count
);
299 if (waitqueue_active(wq
))
300 __wake_up_locked_key(wq
, TASK_NORMAL
, &k
);
301 spin_unlock_irqrestore(&wq
->lock
, flags
);
306 * Allow callers to wait for completion of an RPC call
308 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
309 * to enforce taking of the wq->lock and hence avoid races with
310 * rpc_complete_task().
312 int __rpc_wait_for_completion_task(struct rpc_task
*task
, int (*action
)(void *))
315 action
= rpc_wait_bit_killable
;
316 return out_of_line_wait_on_bit(&task
->tk_runstate
, RPC_TASK_ACTIVE
,
317 action
, TASK_KILLABLE
);
319 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task
);
322 * Make an RPC task runnable.
324 * Note: If the task is ASYNC, and is being made runnable after sitting on an
325 * rpc_wait_queue, this must be called with the queue spinlock held to protect
326 * the wait queue operation.
328 static void rpc_make_runnable(struct rpc_task
*task
)
330 rpc_clear_queued(task
);
331 if (rpc_test_and_set_running(task
))
333 if (RPC_IS_ASYNC(task
)) {
334 INIT_WORK(&task
->u
.tk_work
, rpc_async_schedule
);
335 queue_work(rpciod_workqueue
, &task
->u
.tk_work
);
337 wake_up_bit(&task
->tk_runstate
, RPC_TASK_QUEUED
);
341 * Prepare for sleeping on a wait queue.
342 * By always appending tasks to the list we ensure FIFO behavior.
343 * NB: An RPC task will only receive interrupt-driven events as long
344 * as it's on a wait queue.
346 static void __rpc_sleep_on_priority(struct rpc_wait_queue
*q
,
347 struct rpc_task
*task
,
349 unsigned char queue_priority
)
351 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
352 task
->tk_pid
, rpc_qname(q
), jiffies
);
354 trace_rpc_task_sleep(task
->tk_client
, task
, q
);
356 __rpc_add_wait_queue(q
, task
, queue_priority
);
358 WARN_ON_ONCE(task
->tk_callback
!= NULL
);
359 task
->tk_callback
= action
;
360 __rpc_add_timer(q
, task
);
363 void rpc_sleep_on(struct rpc_wait_queue
*q
, struct rpc_task
*task
,
366 /* We shouldn't ever put an inactive task to sleep */
367 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task
));
368 if (!RPC_IS_ACTIVATED(task
)) {
369 task
->tk_status
= -EIO
;
370 rpc_put_task_async(task
);
375 * Protect the queue operations.
377 spin_lock_bh(&q
->lock
);
378 __rpc_sleep_on_priority(q
, task
, action
, task
->tk_priority
);
379 spin_unlock_bh(&q
->lock
);
381 EXPORT_SYMBOL_GPL(rpc_sleep_on
);
383 void rpc_sleep_on_priority(struct rpc_wait_queue
*q
, struct rpc_task
*task
,
384 rpc_action action
, int priority
)
386 /* We shouldn't ever put an inactive task to sleep */
387 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task
));
388 if (!RPC_IS_ACTIVATED(task
)) {
389 task
->tk_status
= -EIO
;
390 rpc_put_task_async(task
);
395 * Protect the queue operations.
397 spin_lock_bh(&q
->lock
);
398 __rpc_sleep_on_priority(q
, task
, action
, priority
- RPC_PRIORITY_LOW
);
399 spin_unlock_bh(&q
->lock
);
401 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority
);
404 * __rpc_do_wake_up_task - wake up a single rpc_task
406 * @task: task to be woken up
408 * Caller must hold queue->lock, and have cleared the task queued flag.
410 static void __rpc_do_wake_up_task(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
412 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
413 task
->tk_pid
, jiffies
);
415 /* Has the task been executed yet? If not, we cannot wake it up! */
416 if (!RPC_IS_ACTIVATED(task
)) {
417 printk(KERN_ERR
"RPC: Inactive task (%p) being woken up!\n", task
);
421 trace_rpc_task_wakeup(task
->tk_client
, task
, queue
);
423 __rpc_remove_wait_queue(queue
, task
);
425 rpc_make_runnable(task
);
427 dprintk("RPC: __rpc_wake_up_task done\n");
431 * Wake up a queued task while the queue lock is being held
433 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
435 if (RPC_IS_QUEUED(task
)) {
437 if (task
->tk_waitqueue
== queue
)
438 __rpc_do_wake_up_task(queue
, task
);
443 * Tests whether rpc queue is empty
445 int rpc_queue_empty(struct rpc_wait_queue
*queue
)
449 spin_lock_bh(&queue
->lock
);
451 spin_unlock_bh(&queue
->lock
);
454 EXPORT_SYMBOL_GPL(rpc_queue_empty
);
457 * Wake up a task on a specific queue
459 void rpc_wake_up_queued_task(struct rpc_wait_queue
*queue
, struct rpc_task
*task
)
461 spin_lock_bh(&queue
->lock
);
462 rpc_wake_up_task_queue_locked(queue
, task
);
463 spin_unlock_bh(&queue
->lock
);
465 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task
);
468 * Wake up the next task on a priority queue.
470 static struct rpc_task
*__rpc_find_next_queued_priority(struct rpc_wait_queue
*queue
)
473 struct rpc_task
*task
;
476 * Service a batch of tasks from a single owner.
478 q
= &queue
->tasks
[queue
->priority
];
479 if (!list_empty(q
)) {
480 task
= list_entry(q
->next
, struct rpc_task
, u
.tk_wait
.list
);
481 if (queue
->owner
== task
->tk_owner
) {
484 list_move_tail(&task
->u
.tk_wait
.list
, q
);
487 * Check if we need to switch queues.
493 * Service the next queue.
496 if (q
== &queue
->tasks
[0])
497 q
= &queue
->tasks
[queue
->maxpriority
];
500 if (!list_empty(q
)) {
501 task
= list_entry(q
->next
, struct rpc_task
, u
.tk_wait
.list
);
504 } while (q
!= &queue
->tasks
[queue
->priority
]);
506 rpc_reset_waitqueue_priority(queue
);
510 rpc_set_waitqueue_priority(queue
, (unsigned int)(q
- &queue
->tasks
[0]));
512 rpc_set_waitqueue_owner(queue
, task
->tk_owner
);
517 static struct rpc_task
*__rpc_find_next_queued(struct rpc_wait_queue
*queue
)
519 if (RPC_IS_PRIORITY(queue
))
520 return __rpc_find_next_queued_priority(queue
);
521 if (!list_empty(&queue
->tasks
[0]))
522 return list_first_entry(&queue
->tasks
[0], struct rpc_task
, u
.tk_wait
.list
);
527 * Wake up the first task on the wait queue.
529 struct rpc_task
*rpc_wake_up_first(struct rpc_wait_queue
*queue
,
530 bool (*func
)(struct rpc_task
*, void *), void *data
)
532 struct rpc_task
*task
= NULL
;
534 dprintk("RPC: wake_up_first(%p \"%s\")\n",
535 queue
, rpc_qname(queue
));
536 spin_lock_bh(&queue
->lock
);
537 task
= __rpc_find_next_queued(queue
);
539 if (func(task
, data
))
540 rpc_wake_up_task_queue_locked(queue
, task
);
544 spin_unlock_bh(&queue
->lock
);
548 EXPORT_SYMBOL_GPL(rpc_wake_up_first
);
550 static bool rpc_wake_up_next_func(struct rpc_task
*task
, void *data
)
556 * Wake up the next task on the wait queue.
558 struct rpc_task
*rpc_wake_up_next(struct rpc_wait_queue
*queue
)
560 return rpc_wake_up_first(queue
, rpc_wake_up_next_func
, NULL
);
562 EXPORT_SYMBOL_GPL(rpc_wake_up_next
);
565 * rpc_wake_up - wake up all rpc_tasks
566 * @queue: rpc_wait_queue on which the tasks are sleeping
570 void rpc_wake_up(struct rpc_wait_queue
*queue
)
572 struct list_head
*head
;
574 spin_lock_bh(&queue
->lock
);
575 head
= &queue
->tasks
[queue
->maxpriority
];
577 while (!list_empty(head
)) {
578 struct rpc_task
*task
;
579 task
= list_first_entry(head
,
582 rpc_wake_up_task_queue_locked(queue
, task
);
584 if (head
== &queue
->tasks
[0])
588 spin_unlock_bh(&queue
->lock
);
590 EXPORT_SYMBOL_GPL(rpc_wake_up
);
593 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
594 * @queue: rpc_wait_queue on which the tasks are sleeping
595 * @status: status value to set
599 void rpc_wake_up_status(struct rpc_wait_queue
*queue
, int status
)
601 struct list_head
*head
;
603 spin_lock_bh(&queue
->lock
);
604 head
= &queue
->tasks
[queue
->maxpriority
];
606 while (!list_empty(head
)) {
607 struct rpc_task
*task
;
608 task
= list_first_entry(head
,
611 task
->tk_status
= status
;
612 rpc_wake_up_task_queue_locked(queue
, task
);
614 if (head
== &queue
->tasks
[0])
618 spin_unlock_bh(&queue
->lock
);
620 EXPORT_SYMBOL_GPL(rpc_wake_up_status
);
622 static void __rpc_queue_timer_fn(unsigned long ptr
)
624 struct rpc_wait_queue
*queue
= (struct rpc_wait_queue
*)ptr
;
625 struct rpc_task
*task
, *n
;
626 unsigned long expires
, now
, timeo
;
628 spin_lock(&queue
->lock
);
629 expires
= now
= jiffies
;
630 list_for_each_entry_safe(task
, n
, &queue
->timer_list
.list
, u
.tk_wait
.timer_list
) {
631 timeo
= task
->u
.tk_wait
.expires
;
632 if (time_after_eq(now
, timeo
)) {
633 dprintk("RPC: %5u timeout\n", task
->tk_pid
);
634 task
->tk_status
= -ETIMEDOUT
;
635 rpc_wake_up_task_queue_locked(queue
, task
);
638 if (expires
== now
|| time_after(expires
, timeo
))
641 if (!list_empty(&queue
->timer_list
.list
))
642 rpc_set_queue_timer(queue
, expires
);
643 spin_unlock(&queue
->lock
);
646 static void __rpc_atrun(struct rpc_task
*task
)
652 * Run a task at a later time
654 void rpc_delay(struct rpc_task
*task
, unsigned long delay
)
656 task
->tk_timeout
= delay
;
657 rpc_sleep_on(&delay_queue
, task
, __rpc_atrun
);
659 EXPORT_SYMBOL_GPL(rpc_delay
);
662 * Helper to call task->tk_ops->rpc_call_prepare
664 void rpc_prepare_task(struct rpc_task
*task
)
666 task
->tk_ops
->rpc_call_prepare(task
, task
->tk_calldata
);
670 rpc_init_task_statistics(struct rpc_task
*task
)
672 /* Initialize retry counters */
673 task
->tk_garb_retry
= 2;
674 task
->tk_cred_retry
= 2;
675 task
->tk_rebind_retry
= 2;
677 /* starting timestamp */
678 task
->tk_start
= ktime_get();
682 rpc_reset_task_statistics(struct rpc_task
*task
)
684 task
->tk_timeouts
= 0;
685 task
->tk_flags
&= ~(RPC_CALL_MAJORSEEN
|RPC_TASK_KILLED
|RPC_TASK_SENT
);
687 rpc_init_task_statistics(task
);
691 * Helper that calls task->tk_ops->rpc_call_done if it exists
693 void rpc_exit_task(struct rpc_task
*task
)
695 task
->tk_action
= NULL
;
696 if (task
->tk_ops
->rpc_call_done
!= NULL
) {
697 task
->tk_ops
->rpc_call_done(task
, task
->tk_calldata
);
698 if (task
->tk_action
!= NULL
) {
699 WARN_ON(RPC_ASSASSINATED(task
));
700 /* Always release the RPC slot and buffer memory */
702 rpc_reset_task_statistics(task
);
707 void rpc_exit(struct rpc_task
*task
, int status
)
709 task
->tk_status
= status
;
710 task
->tk_action
= rpc_exit_task
;
711 if (RPC_IS_QUEUED(task
))
712 rpc_wake_up_queued_task(task
->tk_waitqueue
, task
);
714 EXPORT_SYMBOL_GPL(rpc_exit
);
716 void rpc_release_calldata(const struct rpc_call_ops
*ops
, void *calldata
)
718 if (ops
->rpc_release
!= NULL
)
719 ops
->rpc_release(calldata
);
723 * This is the RPC `scheduler' (or rather, the finite state machine).
725 static void __rpc_execute(struct rpc_task
*task
)
727 struct rpc_wait_queue
*queue
;
728 int task_is_async
= RPC_IS_ASYNC(task
);
731 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
732 task
->tk_pid
, task
->tk_flags
);
734 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
735 if (RPC_IS_QUEUED(task
))
739 void (*do_action
)(struct rpc_task
*);
742 * Execute any pending callback first.
744 do_action
= task
->tk_callback
;
745 task
->tk_callback
= NULL
;
746 if (do_action
== NULL
) {
748 * Perform the next FSM step.
749 * tk_action may be NULL if the task has been killed.
750 * In particular, note that rpc_killall_tasks may
751 * do this at any time, so beware when dereferencing.
753 do_action
= task
->tk_action
;
754 if (do_action
== NULL
)
757 trace_rpc_task_run_action(task
->tk_client
, task
, task
->tk_action
);
761 * Lockless check for whether task is sleeping or not.
763 if (!RPC_IS_QUEUED(task
))
766 * The queue->lock protects against races with
767 * rpc_make_runnable().
769 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
770 * rpc_task, rpc_make_runnable() can assign it to a
771 * different workqueue. We therefore cannot assume that the
772 * rpc_task pointer may still be dereferenced.
774 queue
= task
->tk_waitqueue
;
775 spin_lock_bh(&queue
->lock
);
776 if (!RPC_IS_QUEUED(task
)) {
777 spin_unlock_bh(&queue
->lock
);
780 rpc_clear_running(task
);
781 spin_unlock_bh(&queue
->lock
);
785 /* sync task: sleep here */
786 dprintk("RPC: %5u sync task going to sleep\n", task
->tk_pid
);
787 status
= out_of_line_wait_on_bit(&task
->tk_runstate
,
788 RPC_TASK_QUEUED
, rpc_wait_bit_killable
,
790 if (status
== -ERESTARTSYS
) {
792 * When a sync task receives a signal, it exits with
793 * -ERESTARTSYS. In order to catch any callbacks that
794 * clean up after sleeping on some queue, we don't
795 * break the loop here, but go around once more.
797 dprintk("RPC: %5u got signal\n", task
->tk_pid
);
798 task
->tk_flags
|= RPC_TASK_KILLED
;
799 rpc_exit(task
, -ERESTARTSYS
);
801 rpc_set_running(task
);
802 dprintk("RPC: %5u sync task resuming\n", task
->tk_pid
);
805 dprintk("RPC: %5u return %d, status %d\n", task
->tk_pid
, status
,
807 /* Release all resources associated with the task */
808 rpc_release_task(task
);
812 * User-visible entry point to the scheduler.
814 * This may be called recursively if e.g. an async NFS task updates
815 * the attributes and finds that dirty pages must be flushed.
816 * NOTE: Upon exit of this function the task is guaranteed to be
817 * released. In particular note that tk_release() will have
818 * been called, so your task memory may have been freed.
820 void rpc_execute(struct rpc_task
*task
)
822 rpc_set_active(task
);
823 rpc_make_runnable(task
);
824 if (!RPC_IS_ASYNC(task
))
828 static void rpc_async_schedule(struct work_struct
*work
)
830 current
->flags
|= PF_FSTRANS
;
831 __rpc_execute(container_of(work
, struct rpc_task
, u
.tk_work
));
832 current
->flags
&= ~PF_FSTRANS
;
836 * rpc_malloc - allocate an RPC buffer
837 * @task: RPC task that will use this buffer
838 * @size: requested byte size
840 * To prevent rpciod from hanging, this allocator never sleeps,
841 * returning NULL if the request cannot be serviced immediately.
842 * The caller can arrange to sleep in a way that is safe for rpciod.
844 * Most requests are 'small' (under 2KiB) and can be serviced from a
845 * mempool, ensuring that NFS reads and writes can always proceed,
846 * and that there is good locality of reference for these buffers.
848 * In order to avoid memory starvation triggering more writebacks of
849 * NFS requests, we avoid using GFP_KERNEL.
851 void *rpc_malloc(struct rpc_task
*task
, size_t size
)
853 struct rpc_buffer
*buf
;
854 gfp_t gfp
= GFP_NOWAIT
;
856 if (RPC_IS_SWAPPER(task
))
857 gfp
|= __GFP_MEMALLOC
;
859 size
+= sizeof(struct rpc_buffer
);
860 if (size
<= RPC_BUFFER_MAXSIZE
)
861 buf
= mempool_alloc(rpc_buffer_mempool
, gfp
);
863 buf
= kmalloc(size
, gfp
);
869 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
870 task
->tk_pid
, size
, buf
);
873 EXPORT_SYMBOL_GPL(rpc_malloc
);
876 * rpc_free - free buffer allocated via rpc_malloc
877 * @buffer: buffer to free
880 void rpc_free(void *buffer
)
883 struct rpc_buffer
*buf
;
888 buf
= container_of(buffer
, struct rpc_buffer
, data
);
891 dprintk("RPC: freeing buffer of size %zu at %p\n",
894 if (size
<= RPC_BUFFER_MAXSIZE
)
895 mempool_free(buf
, rpc_buffer_mempool
);
899 EXPORT_SYMBOL_GPL(rpc_free
);
902 * Creation and deletion of RPC task structures
904 static void rpc_init_task(struct rpc_task
*task
, const struct rpc_task_setup
*task_setup_data
)
906 memset(task
, 0, sizeof(*task
));
907 atomic_set(&task
->tk_count
, 1);
908 task
->tk_flags
= task_setup_data
->flags
;
909 task
->tk_ops
= task_setup_data
->callback_ops
;
910 task
->tk_calldata
= task_setup_data
->callback_data
;
911 INIT_LIST_HEAD(&task
->tk_task
);
913 task
->tk_priority
= task_setup_data
->priority
- RPC_PRIORITY_LOW
;
914 task
->tk_owner
= current
->tgid
;
916 /* Initialize workqueue for async tasks */
917 task
->tk_workqueue
= task_setup_data
->workqueue
;
919 if (task
->tk_ops
->rpc_call_prepare
!= NULL
)
920 task
->tk_action
= rpc_prepare_task
;
922 rpc_init_task_statistics(task
);
924 dprintk("RPC: new task initialized, procpid %u\n",
925 task_pid_nr(current
));
928 static struct rpc_task
*
931 return (struct rpc_task
*)mempool_alloc(rpc_task_mempool
, GFP_NOIO
);
935 * Create a new task for the specified client.
937 struct rpc_task
*rpc_new_task(const struct rpc_task_setup
*setup_data
)
939 struct rpc_task
*task
= setup_data
->task
;
940 unsigned short flags
= 0;
943 task
= rpc_alloc_task();
945 rpc_release_calldata(setup_data
->callback_ops
,
946 setup_data
->callback_data
);
947 return ERR_PTR(-ENOMEM
);
949 flags
= RPC_TASK_DYNAMIC
;
952 rpc_init_task(task
, setup_data
);
953 task
->tk_flags
|= flags
;
954 dprintk("RPC: allocated task %p\n", task
);
959 * rpc_free_task - release rpc task and perform cleanups
961 * Note that we free up the rpc_task _after_ rpc_release_calldata()
962 * in order to work around a workqueue dependency issue.
965 * "Workqueue currently considers two work items to be the same if they're
966 * on the same address and won't execute them concurrently - ie. it
967 * makes a work item which is queued again while being executed wait
968 * for the previous execution to complete.
970 * If a work function frees the work item, and then waits for an event
971 * which should be performed by another work item and *that* work item
972 * recycles the freed work item, it can create a false dependency loop.
973 * There really is no reliable way to detect this short of verifying
974 * every memory free."
977 static void rpc_free_task(struct rpc_task
*task
)
979 unsigned short tk_flags
= task
->tk_flags
;
981 rpc_release_calldata(task
->tk_ops
, task
->tk_calldata
);
983 if (tk_flags
& RPC_TASK_DYNAMIC
) {
984 dprintk("RPC: %5u freeing task\n", task
->tk_pid
);
985 mempool_free(task
, rpc_task_mempool
);
989 static void rpc_async_release(struct work_struct
*work
)
991 rpc_free_task(container_of(work
, struct rpc_task
, u
.tk_work
));
994 static void rpc_release_resources_task(struct rpc_task
*task
)
997 if (task
->tk_msg
.rpc_cred
) {
998 put_rpccred(task
->tk_msg
.rpc_cred
);
999 task
->tk_msg
.rpc_cred
= NULL
;
1001 rpc_task_release_client(task
);
1004 static void rpc_final_put_task(struct rpc_task
*task
,
1005 struct workqueue_struct
*q
)
1008 INIT_WORK(&task
->u
.tk_work
, rpc_async_release
);
1009 queue_work(q
, &task
->u
.tk_work
);
1011 rpc_free_task(task
);
1014 static void rpc_do_put_task(struct rpc_task
*task
, struct workqueue_struct
*q
)
1016 if (atomic_dec_and_test(&task
->tk_count
)) {
1017 rpc_release_resources_task(task
);
1018 rpc_final_put_task(task
, q
);
1022 void rpc_put_task(struct rpc_task
*task
)
1024 rpc_do_put_task(task
, NULL
);
1026 EXPORT_SYMBOL_GPL(rpc_put_task
);
1028 void rpc_put_task_async(struct rpc_task
*task
)
1030 rpc_do_put_task(task
, task
->tk_workqueue
);
1032 EXPORT_SYMBOL_GPL(rpc_put_task_async
);
1034 static void rpc_release_task(struct rpc_task
*task
)
1036 dprintk("RPC: %5u release task\n", task
->tk_pid
);
1038 WARN_ON_ONCE(RPC_IS_QUEUED(task
));
1040 rpc_release_resources_task(task
);
1043 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1044 * so it should be safe to use task->tk_count as a test for whether
1045 * or not any other processes still hold references to our rpc_task.
1047 if (atomic_read(&task
->tk_count
) != 1 + !RPC_IS_ASYNC(task
)) {
1048 /* Wake up anyone who may be waiting for task completion */
1049 if (!rpc_complete_task(task
))
1052 if (!atomic_dec_and_test(&task
->tk_count
))
1055 rpc_final_put_task(task
, task
->tk_workqueue
);
1060 return try_module_get(THIS_MODULE
) ? 0 : -EINVAL
;
1063 void rpciod_down(void)
1065 module_put(THIS_MODULE
);
1069 * Start up the rpciod workqueue.
1071 static int rpciod_start(void)
1073 struct workqueue_struct
*wq
;
1076 * Create the rpciod thread and wait for it to start.
1078 dprintk("RPC: creating workqueue rpciod\n");
1079 wq
= alloc_workqueue("rpciod", WQ_MEM_RECLAIM
, 1);
1080 rpciod_workqueue
= wq
;
1081 return rpciod_workqueue
!= NULL
;
1084 static void rpciod_stop(void)
1086 struct workqueue_struct
*wq
= NULL
;
1088 if (rpciod_workqueue
== NULL
)
1090 dprintk("RPC: destroying workqueue rpciod\n");
1092 wq
= rpciod_workqueue
;
1093 rpciod_workqueue
= NULL
;
1094 destroy_workqueue(wq
);
1098 rpc_destroy_mempool(void)
1101 if (rpc_buffer_mempool
)
1102 mempool_destroy(rpc_buffer_mempool
);
1103 if (rpc_task_mempool
)
1104 mempool_destroy(rpc_task_mempool
);
1106 kmem_cache_destroy(rpc_task_slabp
);
1107 if (rpc_buffer_slabp
)
1108 kmem_cache_destroy(rpc_buffer_slabp
);
1109 rpc_destroy_wait_queue(&delay_queue
);
1113 rpc_init_mempool(void)
1116 * The following is not strictly a mempool initialisation,
1117 * but there is no harm in doing it here
1119 rpc_init_wait_queue(&delay_queue
, "delayq");
1120 if (!rpciod_start())
1123 rpc_task_slabp
= kmem_cache_create("rpc_tasks",
1124 sizeof(struct rpc_task
),
1125 0, SLAB_HWCACHE_ALIGN
,
1127 if (!rpc_task_slabp
)
1129 rpc_buffer_slabp
= kmem_cache_create("rpc_buffers",
1131 0, SLAB_HWCACHE_ALIGN
,
1133 if (!rpc_buffer_slabp
)
1135 rpc_task_mempool
= mempool_create_slab_pool(RPC_TASK_POOLSIZE
,
1137 if (!rpc_task_mempool
)
1139 rpc_buffer_mempool
= mempool_create_slab_pool(RPC_BUFFER_POOLSIZE
,
1141 if (!rpc_buffer_mempool
)
1145 rpc_destroy_mempool();