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
32 * RPC slabs and memory pools
34 #define RPC_BUFFER_MAXSIZE (2048)
35 #define RPC_BUFFER_POOLSIZE (8)
36 #define RPC_TASK_POOLSIZE (8)
37 static struct kmem_cache *rpc_task_slabp __read_mostly;
38 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
39 static mempool_t *rpc_task_mempool __read_mostly;
40 static mempool_t *rpc_buffer_mempool __read_mostly;
42 static void rpc_async_schedule(struct work_struct *);
43 static void rpc_release_task(struct rpc_task *task);
44 static void __rpc_queue_timer_fn(unsigned long ptr);
47 * RPC tasks sit here while waiting for conditions to improve.
49 static struct rpc_wait_queue delay_queue;
52 * rpciod-related stuff
54 struct workqueue_struct *rpciod_workqueue;
57 * Disable the timer for a given RPC task. Should be called with
58 * queue->lock and bh_disabled in order to avoid races within
62 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
64 if (task->tk_timeout == 0)
66 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
68 list_del(&task->u.tk_wait.timer_list);
69 if (list_empty(&queue->timer_list.list))
70 del_timer(&queue->timer_list.timer);
74 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
76 queue->timer_list.expires = expires;
77 mod_timer(&queue->timer_list.timer, expires);
81 * Set up a timer for the current task.
84 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
86 if (!task->tk_timeout)
89 dprintk("RPC: %5u setting alarm for %lu ms\n",
90 task->tk_pid, task->tk_timeout * 1000 / HZ);
92 task->u.tk_wait.expires = jiffies + task->tk_timeout;
93 if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
94 rpc_set_queue_timer(queue, task->u.tk_wait.expires);
95 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
99 * Add new request to a priority queue.
101 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
102 struct rpc_task *task,
103 unsigned char queue_priority)
108 INIT_LIST_HEAD(&task->u.tk_wait.links);
109 q = &queue->tasks[queue_priority];
110 if (unlikely(queue_priority > queue->maxpriority))
111 q = &queue->tasks[queue->maxpriority];
112 list_for_each_entry(t, q, u.tk_wait.list) {
113 if (t->tk_owner == task->tk_owner) {
114 list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
118 list_add_tail(&task->u.tk_wait.list, q);
122 * Add new request to wait queue.
124 * Swapper tasks always get inserted at the head of the queue.
125 * This should avoid many nasty memory deadlocks and hopefully
126 * improve overall performance.
127 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
129 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
130 struct rpc_task *task,
131 unsigned char queue_priority)
133 BUG_ON (RPC_IS_QUEUED(task));
135 if (RPC_IS_PRIORITY(queue))
136 __rpc_add_wait_queue_priority(queue, task, queue_priority);
137 else if (RPC_IS_SWAPPER(task))
138 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
140 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
141 task->tk_waitqueue = queue;
143 rpc_set_queued(task);
145 dprintk("RPC: %5u added to queue %p \"%s\"\n",
146 task->tk_pid, queue, rpc_qname(queue));
150 * Remove request from a priority queue.
152 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
156 if (!list_empty(&task->u.tk_wait.links)) {
157 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
158 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
159 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
164 * Remove request from queue.
165 * Note: must be called with spin lock held.
167 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
169 __rpc_disable_timer(queue, task);
170 if (RPC_IS_PRIORITY(queue))
171 __rpc_remove_wait_queue_priority(task);
172 list_del(&task->u.tk_wait.list);
174 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
175 task->tk_pid, queue, rpc_qname(queue));
178 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
180 queue->priority = priority;
181 queue->count = 1 << (priority * 2);
184 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
187 queue->nr = RPC_BATCH_COUNT;
190 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
192 rpc_set_waitqueue_priority(queue, queue->maxpriority);
193 rpc_set_waitqueue_owner(queue, 0);
196 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
200 spin_lock_init(&queue->lock);
201 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
202 INIT_LIST_HEAD(&queue->tasks[i]);
203 queue->maxpriority = nr_queues - 1;
204 rpc_reset_waitqueue_priority(queue);
206 setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
207 INIT_LIST_HEAD(&queue->timer_list.list);
213 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
215 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
217 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
219 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
221 __rpc_init_priority_wait_queue(queue, qname, 1);
223 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
225 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
227 del_timer_sync(&queue->timer_list.timer);
229 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
231 static int rpc_wait_bit_killable(void *word)
233 if (fatal_signal_pending(current))
235 freezable_schedule();
240 static void rpc_task_set_debuginfo(struct rpc_task *task)
242 static atomic_t rpc_pid;
244 task->tk_pid = atomic_inc_return(&rpc_pid);
247 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
252 static void rpc_set_active(struct rpc_task *task)
254 rpc_task_set_debuginfo(task);
255 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
259 * Mark an RPC call as having completed by clearing the 'active' bit
260 * and then waking up all tasks that were sleeping.
262 static int rpc_complete_task(struct rpc_task *task)
264 void *m = &task->tk_runstate;
265 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
266 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
270 spin_lock_irqsave(&wq->lock, flags);
271 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
272 ret = atomic_dec_and_test(&task->tk_count);
273 if (waitqueue_active(wq))
274 __wake_up_locked_key(wq, TASK_NORMAL, &k);
275 spin_unlock_irqrestore(&wq->lock, flags);
280 * Allow callers to wait for completion of an RPC call
282 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
283 * to enforce taking of the wq->lock and hence avoid races with
284 * rpc_complete_task().
286 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
289 action = rpc_wait_bit_killable;
290 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
291 action, TASK_KILLABLE);
293 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
296 * Make an RPC task runnable.
298 * Note: If the task is ASYNC, this must be called with
299 * the spinlock held to protect the wait queue operation.
301 static void rpc_make_runnable(struct rpc_task *task)
303 rpc_clear_queued(task);
304 if (rpc_test_and_set_running(task))
306 if (RPC_IS_ASYNC(task)) {
307 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
308 queue_work(rpciod_workqueue, &task->u.tk_work);
310 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
314 * Prepare for sleeping on a wait queue.
315 * By always appending tasks to the list we ensure FIFO behavior.
316 * NB: An RPC task will only receive interrupt-driven events as long
317 * as it's on a wait queue.
319 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
320 struct rpc_task *task,
322 unsigned char queue_priority)
324 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
325 task->tk_pid, rpc_qname(q), jiffies);
327 __rpc_add_wait_queue(q, task, queue_priority);
329 BUG_ON(task->tk_callback != NULL);
330 task->tk_callback = action;
331 __rpc_add_timer(q, task);
334 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
337 /* We shouldn't ever put an inactive task to sleep */
338 BUG_ON(!RPC_IS_ACTIVATED(task));
341 * Protect the queue operations.
343 spin_lock_bh(&q->lock);
344 __rpc_sleep_on_priority(q, task, action, task->tk_priority);
345 spin_unlock_bh(&q->lock);
347 EXPORT_SYMBOL_GPL(rpc_sleep_on);
349 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
350 rpc_action action, int priority)
352 /* We shouldn't ever put an inactive task to sleep */
353 BUG_ON(!RPC_IS_ACTIVATED(task));
356 * Protect the queue operations.
358 spin_lock_bh(&q->lock);
359 __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
360 spin_unlock_bh(&q->lock);
364 * __rpc_do_wake_up_task - wake up a single rpc_task
366 * @task: task to be woken up
368 * Caller must hold queue->lock, and have cleared the task queued flag.
370 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
372 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
373 task->tk_pid, jiffies);
375 /* Has the task been executed yet? If not, we cannot wake it up! */
376 if (!RPC_IS_ACTIVATED(task)) {
377 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
381 __rpc_remove_wait_queue(queue, task);
383 rpc_make_runnable(task);
385 dprintk("RPC: __rpc_wake_up_task done\n");
389 * Wake up a queued task while the queue lock is being held
391 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
393 if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
394 __rpc_do_wake_up_task(queue, task);
398 * Tests whether rpc queue is empty
400 int rpc_queue_empty(struct rpc_wait_queue *queue)
404 spin_lock_bh(&queue->lock);
406 spin_unlock_bh(&queue->lock);
409 EXPORT_SYMBOL_GPL(rpc_queue_empty);
412 * Wake up a task on a specific queue
414 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
416 spin_lock_bh(&queue->lock);
417 rpc_wake_up_task_queue_locked(queue, task);
418 spin_unlock_bh(&queue->lock);
420 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
423 * Wake up the next task on a priority queue.
425 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
428 struct rpc_task *task;
431 * Service a batch of tasks from a single owner.
433 q = &queue->tasks[queue->priority];
434 if (!list_empty(q)) {
435 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
436 if (queue->owner == task->tk_owner) {
439 list_move_tail(&task->u.tk_wait.list, q);
442 * Check if we need to switch queues.
449 * Service the next queue.
452 if (q == &queue->tasks[0])
453 q = &queue->tasks[queue->maxpriority];
456 if (!list_empty(q)) {
457 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
460 } while (q != &queue->tasks[queue->priority]);
462 rpc_reset_waitqueue_priority(queue);
466 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
468 rpc_set_waitqueue_owner(queue, task->tk_owner);
473 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
475 if (RPC_IS_PRIORITY(queue))
476 return __rpc_find_next_queued_priority(queue);
477 if (!list_empty(&queue->tasks[0]))
478 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
483 * Wake up the first task on the wait queue.
485 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
486 bool (*func)(struct rpc_task *, void *), void *data)
488 struct rpc_task *task = NULL;
490 dprintk("RPC: wake_up_first(%p \"%s\")\n",
491 queue, rpc_qname(queue));
492 spin_lock_bh(&queue->lock);
493 task = __rpc_find_next_queued(queue);
495 if (func(task, data))
496 rpc_wake_up_task_queue_locked(queue, task);
500 spin_unlock_bh(&queue->lock);
504 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
506 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
512 * Wake up the next task on the wait queue.
514 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
516 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
518 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
521 * rpc_wake_up - wake up all rpc_tasks
522 * @queue: rpc_wait_queue on which the tasks are sleeping
526 void rpc_wake_up(struct rpc_wait_queue *queue)
528 struct rpc_task *task, *next;
529 struct list_head *head;
531 spin_lock_bh(&queue->lock);
532 head = &queue->tasks[queue->maxpriority];
534 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
535 rpc_wake_up_task_queue_locked(queue, task);
536 if (head == &queue->tasks[0])
540 spin_unlock_bh(&queue->lock);
542 EXPORT_SYMBOL_GPL(rpc_wake_up);
545 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
546 * @queue: rpc_wait_queue on which the tasks are sleeping
547 * @status: status value to set
551 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
553 struct rpc_task *task, *next;
554 struct list_head *head;
556 spin_lock_bh(&queue->lock);
557 head = &queue->tasks[queue->maxpriority];
559 list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
560 task->tk_status = status;
561 rpc_wake_up_task_queue_locked(queue, task);
563 if (head == &queue->tasks[0])
567 spin_unlock_bh(&queue->lock);
569 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
571 static void __rpc_queue_timer_fn(unsigned long ptr)
573 struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
574 struct rpc_task *task, *n;
575 unsigned long expires, now, timeo;
577 spin_lock(&queue->lock);
578 expires = now = jiffies;
579 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
580 timeo = task->u.tk_wait.expires;
581 if (time_after_eq(now, timeo)) {
582 dprintk("RPC: %5u timeout\n", task->tk_pid);
583 task->tk_status = -ETIMEDOUT;
584 rpc_wake_up_task_queue_locked(queue, task);
587 if (expires == now || time_after(expires, timeo))
590 if (!list_empty(&queue->timer_list.list))
591 rpc_set_queue_timer(queue, expires);
592 spin_unlock(&queue->lock);
595 static void __rpc_atrun(struct rpc_task *task)
601 * Run a task at a later time
603 void rpc_delay(struct rpc_task *task, unsigned long delay)
605 task->tk_timeout = delay;
606 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
608 EXPORT_SYMBOL_GPL(rpc_delay);
611 * Helper to call task->tk_ops->rpc_call_prepare
613 void rpc_prepare_task(struct rpc_task *task)
615 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
619 rpc_init_task_statistics(struct rpc_task *task)
621 /* Initialize retry counters */
622 task->tk_garb_retry = 2;
623 task->tk_cred_retry = 2;
624 task->tk_rebind_retry = 2;
626 /* starting timestamp */
627 task->tk_start = ktime_get();
631 rpc_reset_task_statistics(struct rpc_task *task)
633 task->tk_timeouts = 0;
634 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
636 rpc_init_task_statistics(task);
640 * Helper that calls task->tk_ops->rpc_call_done if it exists
642 void rpc_exit_task(struct rpc_task *task)
644 task->tk_action = NULL;
645 if (task->tk_ops->rpc_call_done != NULL) {
646 task->tk_ops->rpc_call_done(task, task->tk_calldata);
647 if (task->tk_action != NULL) {
648 WARN_ON(RPC_ASSASSINATED(task));
649 /* Always release the RPC slot and buffer memory */
651 rpc_reset_task_statistics(task);
656 void rpc_exit(struct rpc_task *task, int status)
658 task->tk_status = status;
659 task->tk_action = rpc_exit_task;
660 if (RPC_IS_QUEUED(task))
661 rpc_wake_up_queued_task(task->tk_waitqueue, task);
663 EXPORT_SYMBOL_GPL(rpc_exit);
665 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
667 if (ops->rpc_release != NULL)
668 ops->rpc_release(calldata);
672 * This is the RPC `scheduler' (or rather, the finite state machine).
674 static void __rpc_execute(struct rpc_task *task)
676 struct rpc_wait_queue *queue;
677 int task_is_async = RPC_IS_ASYNC(task);
680 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
681 task->tk_pid, task->tk_flags);
683 BUG_ON(RPC_IS_QUEUED(task));
686 void (*do_action)(struct rpc_task *);
689 * Execute any pending callback first.
691 do_action = task->tk_callback;
692 task->tk_callback = NULL;
693 if (do_action == NULL) {
695 * Perform the next FSM step.
696 * tk_action may be NULL if the task has been killed.
697 * In particular, note that rpc_killall_tasks may
698 * do this at any time, so beware when dereferencing.
700 do_action = task->tk_action;
701 if (do_action == NULL)
707 * Lockless check for whether task is sleeping or not.
709 if (!RPC_IS_QUEUED(task))
712 * The queue->lock protects against races with
713 * rpc_make_runnable().
715 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
716 * rpc_task, rpc_make_runnable() can assign it to a
717 * different workqueue. We therefore cannot assume that the
718 * rpc_task pointer may still be dereferenced.
720 queue = task->tk_waitqueue;
721 spin_lock_bh(&queue->lock);
722 if (!RPC_IS_QUEUED(task)) {
723 spin_unlock_bh(&queue->lock);
726 rpc_clear_running(task);
727 spin_unlock_bh(&queue->lock);
731 /* sync task: sleep here */
732 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
733 status = out_of_line_wait_on_bit(&task->tk_runstate,
734 RPC_TASK_QUEUED, rpc_wait_bit_killable,
736 if (status == -ERESTARTSYS) {
738 * When a sync task receives a signal, it exits with
739 * -ERESTARTSYS. In order to catch any callbacks that
740 * clean up after sleeping on some queue, we don't
741 * break the loop here, but go around once more.
743 dprintk("RPC: %5u got signal\n", task->tk_pid);
744 task->tk_flags |= RPC_TASK_KILLED;
745 rpc_exit(task, -ERESTARTSYS);
747 rpc_set_running(task);
748 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
751 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
753 /* Release all resources associated with the task */
754 rpc_release_task(task);
758 * User-visible entry point to the scheduler.
760 * This may be called recursively if e.g. an async NFS task updates
761 * the attributes and finds that dirty pages must be flushed.
762 * NOTE: Upon exit of this function the task is guaranteed to be
763 * released. In particular note that tk_release() will have
764 * been called, so your task memory may have been freed.
766 void rpc_execute(struct rpc_task *task)
768 rpc_set_active(task);
769 rpc_make_runnable(task);
770 if (!RPC_IS_ASYNC(task))
774 static void rpc_async_schedule(struct work_struct *work)
776 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
780 * rpc_malloc - allocate an RPC buffer
781 * @task: RPC task that will use this buffer
782 * @size: requested byte size
784 * To prevent rpciod from hanging, this allocator never sleeps,
785 * returning NULL if the request cannot be serviced immediately.
786 * The caller can arrange to sleep in a way that is safe for rpciod.
788 * Most requests are 'small' (under 2KiB) and can be serviced from a
789 * mempool, ensuring that NFS reads and writes can always proceed,
790 * and that there is good locality of reference for these buffers.
792 * In order to avoid memory starvation triggering more writebacks of
793 * NFS requests, we avoid using GFP_KERNEL.
795 void *rpc_malloc(struct rpc_task *task, size_t size)
797 struct rpc_buffer *buf;
798 gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
800 size += sizeof(struct rpc_buffer);
801 if (size <= RPC_BUFFER_MAXSIZE)
802 buf = mempool_alloc(rpc_buffer_mempool, gfp);
804 buf = kmalloc(size, gfp);
810 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
811 task->tk_pid, size, buf);
814 EXPORT_SYMBOL_GPL(rpc_malloc);
817 * rpc_free - free buffer allocated via rpc_malloc
818 * @buffer: buffer to free
821 void rpc_free(void *buffer)
824 struct rpc_buffer *buf;
829 buf = container_of(buffer, struct rpc_buffer, data);
832 dprintk("RPC: freeing buffer of size %zu at %p\n",
835 if (size <= RPC_BUFFER_MAXSIZE)
836 mempool_free(buf, rpc_buffer_mempool);
840 EXPORT_SYMBOL_GPL(rpc_free);
843 * Creation and deletion of RPC task structures
845 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
847 memset(task, 0, sizeof(*task));
848 atomic_set(&task->tk_count, 1);
849 task->tk_flags = task_setup_data->flags;
850 task->tk_ops = task_setup_data->callback_ops;
851 task->tk_calldata = task_setup_data->callback_data;
852 INIT_LIST_HEAD(&task->tk_task);
854 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
855 task->tk_owner = current->tgid;
857 /* Initialize workqueue for async tasks */
858 task->tk_workqueue = task_setup_data->workqueue;
860 if (task->tk_ops->rpc_call_prepare != NULL)
861 task->tk_action = rpc_prepare_task;
863 rpc_init_task_statistics(task);
865 dprintk("RPC: new task initialized, procpid %u\n",
866 task_pid_nr(current));
869 static struct rpc_task *
872 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
876 * Create a new task for the specified client.
878 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
880 struct rpc_task *task = setup_data->task;
881 unsigned short flags = 0;
884 task = rpc_alloc_task();
886 rpc_release_calldata(setup_data->callback_ops,
887 setup_data->callback_data);
888 return ERR_PTR(-ENOMEM);
890 flags = RPC_TASK_DYNAMIC;
893 rpc_init_task(task, setup_data);
894 task->tk_flags |= flags;
895 dprintk("RPC: allocated task %p\n", task);
899 static void rpc_free_task(struct rpc_task *task)
901 const struct rpc_call_ops *tk_ops = task->tk_ops;
902 void *calldata = task->tk_calldata;
904 if (task->tk_flags & RPC_TASK_DYNAMIC) {
905 dprintk("RPC: %5u freeing task\n", task->tk_pid);
906 mempool_free(task, rpc_task_mempool);
908 rpc_release_calldata(tk_ops, calldata);
911 static void rpc_async_release(struct work_struct *work)
913 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
916 static void rpc_release_resources_task(struct rpc_task *task)
920 if (task->tk_msg.rpc_cred) {
921 put_rpccred(task->tk_msg.rpc_cred);
922 task->tk_msg.rpc_cred = NULL;
924 rpc_task_release_client(task);
927 static void rpc_final_put_task(struct rpc_task *task,
928 struct workqueue_struct *q)
931 INIT_WORK(&task->u.tk_work, rpc_async_release);
932 queue_work(q, &task->u.tk_work);
937 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
939 if (atomic_dec_and_test(&task->tk_count)) {
940 rpc_release_resources_task(task);
941 rpc_final_put_task(task, q);
945 void rpc_put_task(struct rpc_task *task)
947 rpc_do_put_task(task, NULL);
949 EXPORT_SYMBOL_GPL(rpc_put_task);
951 void rpc_put_task_async(struct rpc_task *task)
953 rpc_do_put_task(task, task->tk_workqueue);
955 EXPORT_SYMBOL_GPL(rpc_put_task_async);
957 static void rpc_release_task(struct rpc_task *task)
959 dprintk("RPC: %5u release task\n", task->tk_pid);
961 BUG_ON (RPC_IS_QUEUED(task));
963 rpc_release_resources_task(task);
966 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
967 * so it should be safe to use task->tk_count as a test for whether
968 * or not any other processes still hold references to our rpc_task.
970 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
971 /* Wake up anyone who may be waiting for task completion */
972 if (!rpc_complete_task(task))
975 if (!atomic_dec_and_test(&task->tk_count))
978 rpc_final_put_task(task, task->tk_workqueue);
983 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
986 void rpciod_down(void)
988 module_put(THIS_MODULE);
992 * Start up the rpciod workqueue.
994 static int rpciod_start(void)
996 struct workqueue_struct *wq;
999 * Create the rpciod thread and wait for it to start.
1001 dprintk("RPC: creating workqueue rpciod\n");
1002 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
1003 rpciod_workqueue = wq;
1004 return rpciod_workqueue != NULL;
1007 static void rpciod_stop(void)
1009 struct workqueue_struct *wq = NULL;
1011 if (rpciod_workqueue == NULL)
1013 dprintk("RPC: destroying workqueue rpciod\n");
1015 wq = rpciod_workqueue;
1016 rpciod_workqueue = NULL;
1017 destroy_workqueue(wq);
1021 rpc_destroy_mempool(void)
1024 if (rpc_buffer_mempool)
1025 mempool_destroy(rpc_buffer_mempool);
1026 if (rpc_task_mempool)
1027 mempool_destroy(rpc_task_mempool);
1029 kmem_cache_destroy(rpc_task_slabp);
1030 if (rpc_buffer_slabp)
1031 kmem_cache_destroy(rpc_buffer_slabp);
1032 rpc_destroy_wait_queue(&delay_queue);
1036 rpc_init_mempool(void)
1039 * The following is not strictly a mempool initialisation,
1040 * but there is no harm in doing it here
1042 rpc_init_wait_queue(&delay_queue, "delayq");
1043 if (!rpciod_start())
1046 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1047 sizeof(struct rpc_task),
1048 0, SLAB_HWCACHE_ALIGN,
1050 if (!rpc_task_slabp)
1052 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1054 0, SLAB_HWCACHE_ALIGN,
1056 if (!rpc_buffer_slabp)
1058 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1060 if (!rpc_task_mempool)
1062 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1064 if (!rpc_buffer_mempool)
1068 rpc_destroy_mempool();