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);
102 * Add new request to a priority queue.
104 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
105 struct rpc_task *task,
106 unsigned char queue_priority)
111 INIT_LIST_HEAD(&task->u.tk_wait.links);
112 q = &queue->tasks[queue_priority];
113 if (unlikely(queue_priority > queue->maxpriority))
114 q = &queue->tasks[queue->maxpriority];
115 list_for_each_entry(t, q, u.tk_wait.list) {
116 if (t->tk_owner == task->tk_owner) {
117 list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
121 list_add_tail(&task->u.tk_wait.list, q);
125 * Add new request to wait queue.
127 * Swapper tasks always get inserted at the head of the queue.
128 * This should avoid many nasty memory deadlocks and hopefully
129 * improve overall performance.
130 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
132 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
133 struct rpc_task *task,
134 unsigned char queue_priority)
136 WARN_ON_ONCE(RPC_IS_QUEUED(task));
137 if (RPC_IS_QUEUED(task))
140 if (RPC_IS_PRIORITY(queue))
141 __rpc_add_wait_queue_priority(queue, task, queue_priority);
142 else if (RPC_IS_SWAPPER(task))
143 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
145 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
146 task->tk_waitqueue = queue;
148 rpc_set_queued(task);
150 dprintk("RPC: %5u added to queue %p \"%s\"\n",
151 task->tk_pid, queue, rpc_qname(queue));
155 * Remove request from a priority queue.
157 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
161 if (!list_empty(&task->u.tk_wait.links)) {
162 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
163 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
164 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
169 * Remove request from queue.
170 * Note: must be called with spin lock held.
172 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
174 __rpc_disable_timer(queue, task);
175 if (RPC_IS_PRIORITY(queue))
176 __rpc_remove_wait_queue_priority(task);
177 list_del(&task->u.tk_wait.list);
179 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
180 task->tk_pid, queue, rpc_qname(queue));
183 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
185 queue->priority = priority;
186 queue->count = 1 << (priority * 2);
189 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
192 queue->nr = RPC_BATCH_COUNT;
195 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
197 rpc_set_waitqueue_priority(queue, queue->maxpriority);
198 rpc_set_waitqueue_owner(queue, 0);
201 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
205 spin_lock_init(&queue->lock);
206 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
207 INIT_LIST_HEAD(&queue->tasks[i]);
208 queue->maxpriority = nr_queues - 1;
209 rpc_reset_waitqueue_priority(queue);
211 setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
212 INIT_LIST_HEAD(&queue->timer_list.list);
213 rpc_assign_waitqueue_name(queue, qname);
216 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
218 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
220 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
222 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
224 __rpc_init_priority_wait_queue(queue, qname, 1);
226 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
228 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
230 del_timer_sync(&queue->timer_list.timer);
232 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
234 static int rpc_wait_bit_killable(void *word)
236 if (fatal_signal_pending(current))
238 freezable_schedule();
243 static void rpc_task_set_debuginfo(struct rpc_task *task)
245 static atomic_t rpc_pid;
247 task->tk_pid = atomic_inc_return(&rpc_pid);
250 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
255 static void rpc_set_active(struct rpc_task *task)
257 trace_rpc_task_begin(task->tk_client, task, NULL);
259 rpc_task_set_debuginfo(task);
260 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
264 * Mark an RPC call as having completed by clearing the 'active' bit
265 * and then waking up all tasks that were sleeping.
267 static int rpc_complete_task(struct rpc_task *task)
269 void *m = &task->tk_runstate;
270 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
271 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
275 trace_rpc_task_complete(task->tk_client, task, NULL);
277 spin_lock_irqsave(&wq->lock, flags);
278 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
279 ret = atomic_dec_and_test(&task->tk_count);
280 if (waitqueue_active(wq))
281 __wake_up_locked_key(wq, TASK_NORMAL, &k);
282 spin_unlock_irqrestore(&wq->lock, flags);
287 * Allow callers to wait for completion of an RPC call
289 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
290 * to enforce taking of the wq->lock and hence avoid races with
291 * rpc_complete_task().
293 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
296 action = rpc_wait_bit_killable;
297 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
298 action, TASK_KILLABLE);
300 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
303 * Make an RPC task runnable.
305 * Note: If the task is ASYNC, and is being made runnable after sitting on an
306 * rpc_wait_queue, this must be called with the queue spinlock held to protect
307 * the wait queue operation.
309 static void rpc_make_runnable(struct rpc_task *task)
311 rpc_clear_queued(task);
312 if (rpc_test_and_set_running(task))
314 if (RPC_IS_ASYNC(task)) {
315 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
316 queue_work(rpciod_workqueue, &task->u.tk_work);
318 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
322 * Prepare for sleeping on a wait queue.
323 * By always appending tasks to the list we ensure FIFO behavior.
324 * NB: An RPC task will only receive interrupt-driven events as long
325 * as it's on a wait queue.
327 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
328 struct rpc_task *task,
330 unsigned char queue_priority)
332 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
333 task->tk_pid, rpc_qname(q), jiffies);
335 trace_rpc_task_sleep(task->tk_client, task, q);
337 __rpc_add_wait_queue(q, task, queue_priority);
339 WARN_ON_ONCE(task->tk_callback != NULL);
340 task->tk_callback = action;
341 __rpc_add_timer(q, task);
344 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
347 /* We shouldn't ever put an inactive task to sleep */
348 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
349 if (!RPC_IS_ACTIVATED(task)) {
350 task->tk_status = -EIO;
351 rpc_put_task_async(task);
356 * Protect the queue operations.
358 spin_lock_bh(&q->lock);
359 __rpc_sleep_on_priority(q, task, action, task->tk_priority);
360 spin_unlock_bh(&q->lock);
362 EXPORT_SYMBOL_GPL(rpc_sleep_on);
364 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
365 rpc_action action, int priority)
367 /* We shouldn't ever put an inactive task to sleep */
368 WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
369 if (!RPC_IS_ACTIVATED(task)) {
370 task->tk_status = -EIO;
371 rpc_put_task_async(task);
376 * Protect the queue operations.
378 spin_lock_bh(&q->lock);
379 __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
380 spin_unlock_bh(&q->lock);
382 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
385 * __rpc_do_wake_up_task - wake up a single rpc_task
387 * @task: task to be woken up
389 * Caller must hold queue->lock, and have cleared the task queued flag.
391 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
393 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
394 task->tk_pid, jiffies);
396 /* Has the task been executed yet? If not, we cannot wake it up! */
397 if (!RPC_IS_ACTIVATED(task)) {
398 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
402 trace_rpc_task_wakeup(task->tk_client, task, queue);
404 __rpc_remove_wait_queue(queue, task);
406 rpc_make_runnable(task);
408 dprintk("RPC: __rpc_wake_up_task done\n");
412 * Wake up a queued task while the queue lock is being held
414 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
416 if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
417 __rpc_do_wake_up_task(queue, task);
421 * Tests whether rpc queue is empty
423 int rpc_queue_empty(struct rpc_wait_queue *queue)
427 spin_lock_bh(&queue->lock);
429 spin_unlock_bh(&queue->lock);
432 EXPORT_SYMBOL_GPL(rpc_queue_empty);
435 * Wake up a task on a specific queue
437 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
439 spin_lock_bh(&queue->lock);
440 rpc_wake_up_task_queue_locked(queue, task);
441 spin_unlock_bh(&queue->lock);
443 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
446 * Wake up the next task on a priority queue.
448 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
451 struct rpc_task *task;
454 * Service a batch of tasks from a single owner.
456 q = &queue->tasks[queue->priority];
457 if (!list_empty(q)) {
458 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
459 if (queue->owner == task->tk_owner) {
462 list_move_tail(&task->u.tk_wait.list, q);
465 * Check if we need to switch queues.
472 * Service the next queue.
475 if (q == &queue->tasks[0])
476 q = &queue->tasks[queue->maxpriority];
479 if (!list_empty(q)) {
480 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
483 } while (q != &queue->tasks[queue->priority]);
485 rpc_reset_waitqueue_priority(queue);
489 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
491 rpc_set_waitqueue_owner(queue, task->tk_owner);
496 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
498 if (RPC_IS_PRIORITY(queue))
499 return __rpc_find_next_queued_priority(queue);
500 if (!list_empty(&queue->tasks[0]))
501 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
506 * Wake up the first task on the wait queue.
508 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
509 bool (*func)(struct rpc_task *, void *), void *data)
511 struct rpc_task *task = NULL;
513 dprintk("RPC: wake_up_first(%p \"%s\")\n",
514 queue, rpc_qname(queue));
515 spin_lock_bh(&queue->lock);
516 task = __rpc_find_next_queued(queue);
518 if (func(task, data))
519 rpc_wake_up_task_queue_locked(queue, task);
523 spin_unlock_bh(&queue->lock);
527 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
529 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
535 * Wake up the next task on the wait queue.
537 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
539 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
541 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
544 * rpc_wake_up - wake up all rpc_tasks
545 * @queue: rpc_wait_queue on which the tasks are sleeping
549 void rpc_wake_up(struct rpc_wait_queue *queue)
551 struct list_head *head;
553 spin_lock_bh(&queue->lock);
554 head = &queue->tasks[queue->maxpriority];
556 while (!list_empty(head)) {
557 struct rpc_task *task;
558 task = list_first_entry(head,
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);
572 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
573 * @queue: rpc_wait_queue on which the tasks are sleeping
574 * @status: status value to set
578 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
580 struct list_head *head;
582 spin_lock_bh(&queue->lock);
583 head = &queue->tasks[queue->maxpriority];
585 while (!list_empty(head)) {
586 struct rpc_task *task;
587 task = list_first_entry(head,
590 task->tk_status = status;
591 rpc_wake_up_task_queue_locked(queue, task);
593 if (head == &queue->tasks[0])
597 spin_unlock_bh(&queue->lock);
599 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
601 static void __rpc_queue_timer_fn(unsigned long ptr)
603 struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
604 struct rpc_task *task, *n;
605 unsigned long expires, now, timeo;
607 spin_lock(&queue->lock);
608 expires = now = jiffies;
609 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
610 timeo = task->u.tk_wait.expires;
611 if (time_after_eq(now, timeo)) {
612 dprintk("RPC: %5u timeout\n", task->tk_pid);
613 task->tk_status = -ETIMEDOUT;
614 rpc_wake_up_task_queue_locked(queue, task);
617 if (expires == now || time_after(expires, timeo))
620 if (!list_empty(&queue->timer_list.list))
621 rpc_set_queue_timer(queue, expires);
622 spin_unlock(&queue->lock);
625 static void __rpc_atrun(struct rpc_task *task)
631 * Run a task at a later time
633 void rpc_delay(struct rpc_task *task, unsigned long delay)
635 task->tk_timeout = delay;
636 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
638 EXPORT_SYMBOL_GPL(rpc_delay);
641 * Helper to call task->tk_ops->rpc_call_prepare
643 void rpc_prepare_task(struct rpc_task *task)
645 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
649 rpc_init_task_statistics(struct rpc_task *task)
651 /* Initialize retry counters */
652 task->tk_garb_retry = 2;
653 task->tk_cred_retry = 2;
654 task->tk_rebind_retry = 2;
656 /* starting timestamp */
657 task->tk_start = ktime_get();
661 rpc_reset_task_statistics(struct rpc_task *task)
663 task->tk_timeouts = 0;
664 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
666 rpc_init_task_statistics(task);
670 * Helper that calls task->tk_ops->rpc_call_done if it exists
672 void rpc_exit_task(struct rpc_task *task)
674 task->tk_action = NULL;
675 if (task->tk_ops->rpc_call_done != NULL) {
676 task->tk_ops->rpc_call_done(task, task->tk_calldata);
677 if (task->tk_action != NULL) {
678 WARN_ON(RPC_ASSASSINATED(task));
679 /* Always release the RPC slot and buffer memory */
681 rpc_reset_task_statistics(task);
686 void rpc_exit(struct rpc_task *task, int status)
688 task->tk_status = status;
689 task->tk_action = rpc_exit_task;
690 if (RPC_IS_QUEUED(task))
691 rpc_wake_up_queued_task(task->tk_waitqueue, task);
693 EXPORT_SYMBOL_GPL(rpc_exit);
695 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
697 if (ops->rpc_release != NULL)
698 ops->rpc_release(calldata);
702 * This is the RPC `scheduler' (or rather, the finite state machine).
704 static void __rpc_execute(struct rpc_task *task)
706 struct rpc_wait_queue *queue;
707 int task_is_async = RPC_IS_ASYNC(task);
710 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
711 task->tk_pid, task->tk_flags);
713 WARN_ON_ONCE(RPC_IS_QUEUED(task));
714 if (RPC_IS_QUEUED(task))
718 void (*do_action)(struct rpc_task *);
721 * Execute any pending callback first.
723 do_action = task->tk_callback;
724 task->tk_callback = NULL;
725 if (do_action == NULL) {
727 * Perform the next FSM step.
728 * tk_action may be NULL if the task has been killed.
729 * In particular, note that rpc_killall_tasks may
730 * do this at any time, so beware when dereferencing.
732 do_action = task->tk_action;
733 if (do_action == NULL)
736 trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
740 * Lockless check for whether task is sleeping or not.
742 if (!RPC_IS_QUEUED(task))
745 * The queue->lock protects against races with
746 * rpc_make_runnable().
748 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
749 * rpc_task, rpc_make_runnable() can assign it to a
750 * different workqueue. We therefore cannot assume that the
751 * rpc_task pointer may still be dereferenced.
753 queue = task->tk_waitqueue;
754 spin_lock_bh(&queue->lock);
755 if (!RPC_IS_QUEUED(task)) {
756 spin_unlock_bh(&queue->lock);
759 rpc_clear_running(task);
760 spin_unlock_bh(&queue->lock);
764 /* sync task: sleep here */
765 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
766 status = out_of_line_wait_on_bit(&task->tk_runstate,
767 RPC_TASK_QUEUED, rpc_wait_bit_killable,
769 if (status == -ERESTARTSYS) {
771 * When a sync task receives a signal, it exits with
772 * -ERESTARTSYS. In order to catch any callbacks that
773 * clean up after sleeping on some queue, we don't
774 * break the loop here, but go around once more.
776 dprintk("RPC: %5u got signal\n", task->tk_pid);
777 task->tk_flags |= RPC_TASK_KILLED;
778 rpc_exit(task, -ERESTARTSYS);
780 rpc_set_running(task);
781 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
784 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
786 /* Release all resources associated with the task */
787 rpc_release_task(task);
791 * User-visible entry point to the scheduler.
793 * This may be called recursively if e.g. an async NFS task updates
794 * the attributes and finds that dirty pages must be flushed.
795 * NOTE: Upon exit of this function the task is guaranteed to be
796 * released. In particular note that tk_release() will have
797 * been called, so your task memory may have been freed.
799 void rpc_execute(struct rpc_task *task)
801 rpc_set_active(task);
802 rpc_make_runnable(task);
803 if (!RPC_IS_ASYNC(task))
807 static void rpc_async_schedule(struct work_struct *work)
809 current->flags |= PF_FSTRANS;
810 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
811 current->flags &= ~PF_FSTRANS;
815 * rpc_malloc - allocate an RPC buffer
816 * @task: RPC task that will use this buffer
817 * @size: requested byte size
819 * To prevent rpciod from hanging, this allocator never sleeps,
820 * returning NULL if the request cannot be serviced immediately.
821 * The caller can arrange to sleep in a way that is safe for rpciod.
823 * Most requests are 'small' (under 2KiB) and can be serviced from a
824 * mempool, ensuring that NFS reads and writes can always proceed,
825 * and that there is good locality of reference for these buffers.
827 * In order to avoid memory starvation triggering more writebacks of
828 * NFS requests, we avoid using GFP_KERNEL.
830 void *rpc_malloc(struct rpc_task *task, size_t size)
832 struct rpc_buffer *buf;
833 gfp_t gfp = GFP_NOWAIT;
835 if (RPC_IS_SWAPPER(task))
836 gfp |= __GFP_MEMALLOC;
838 size += sizeof(struct rpc_buffer);
839 if (size <= RPC_BUFFER_MAXSIZE)
840 buf = mempool_alloc(rpc_buffer_mempool, gfp);
842 buf = kmalloc(size, gfp);
848 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
849 task->tk_pid, size, buf);
852 EXPORT_SYMBOL_GPL(rpc_malloc);
855 * rpc_free - free buffer allocated via rpc_malloc
856 * @buffer: buffer to free
859 void rpc_free(void *buffer)
862 struct rpc_buffer *buf;
867 buf = container_of(buffer, struct rpc_buffer, data);
870 dprintk("RPC: freeing buffer of size %zu at %p\n",
873 if (size <= RPC_BUFFER_MAXSIZE)
874 mempool_free(buf, rpc_buffer_mempool);
878 EXPORT_SYMBOL_GPL(rpc_free);
881 * Creation and deletion of RPC task structures
883 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
885 memset(task, 0, sizeof(*task));
886 atomic_set(&task->tk_count, 1);
887 task->tk_flags = task_setup_data->flags;
888 task->tk_ops = task_setup_data->callback_ops;
889 task->tk_calldata = task_setup_data->callback_data;
890 INIT_LIST_HEAD(&task->tk_task);
892 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
893 task->tk_owner = current->tgid;
895 /* Initialize workqueue for async tasks */
896 task->tk_workqueue = task_setup_data->workqueue;
898 if (task->tk_ops->rpc_call_prepare != NULL)
899 task->tk_action = rpc_prepare_task;
901 rpc_init_task_statistics(task);
903 dprintk("RPC: new task initialized, procpid %u\n",
904 task_pid_nr(current));
907 static struct rpc_task *
910 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO);
914 * Create a new task for the specified client.
916 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
918 struct rpc_task *task = setup_data->task;
919 unsigned short flags = 0;
922 task = rpc_alloc_task();
924 rpc_release_calldata(setup_data->callback_ops,
925 setup_data->callback_data);
926 return ERR_PTR(-ENOMEM);
928 flags = RPC_TASK_DYNAMIC;
931 rpc_init_task(task, setup_data);
932 task->tk_flags |= flags;
933 dprintk("RPC: allocated task %p\n", task);
937 static void rpc_free_task(struct rpc_task *task)
939 const struct rpc_call_ops *tk_ops = task->tk_ops;
940 void *calldata = task->tk_calldata;
942 if (task->tk_flags & RPC_TASK_DYNAMIC) {
943 dprintk("RPC: %5u freeing task\n", task->tk_pid);
944 mempool_free(task, rpc_task_mempool);
946 rpc_release_calldata(tk_ops, calldata);
949 static void rpc_async_release(struct work_struct *work)
951 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
954 static void rpc_release_resources_task(struct rpc_task *task)
958 if (task->tk_msg.rpc_cred) {
959 put_rpccred(task->tk_msg.rpc_cred);
960 task->tk_msg.rpc_cred = NULL;
962 rpc_task_release_client(task);
965 static void rpc_final_put_task(struct rpc_task *task,
966 struct workqueue_struct *q)
969 INIT_WORK(&task->u.tk_work, rpc_async_release);
970 queue_work(q, &task->u.tk_work);
975 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
977 if (atomic_dec_and_test(&task->tk_count)) {
978 rpc_release_resources_task(task);
979 rpc_final_put_task(task, q);
983 void rpc_put_task(struct rpc_task *task)
985 rpc_do_put_task(task, NULL);
987 EXPORT_SYMBOL_GPL(rpc_put_task);
989 void rpc_put_task_async(struct rpc_task *task)
991 rpc_do_put_task(task, task->tk_workqueue);
993 EXPORT_SYMBOL_GPL(rpc_put_task_async);
995 static void rpc_release_task(struct rpc_task *task)
997 dprintk("RPC: %5u release task\n", task->tk_pid);
999 WARN_ON_ONCE(RPC_IS_QUEUED(task));
1001 rpc_release_resources_task(task);
1004 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1005 * so it should be safe to use task->tk_count as a test for whether
1006 * or not any other processes still hold references to our rpc_task.
1008 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1009 /* Wake up anyone who may be waiting for task completion */
1010 if (!rpc_complete_task(task))
1013 if (!atomic_dec_and_test(&task->tk_count))
1016 rpc_final_put_task(task, task->tk_workqueue);
1021 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1024 void rpciod_down(void)
1026 module_put(THIS_MODULE);
1030 * Start up the rpciod workqueue.
1032 static int rpciod_start(void)
1034 struct workqueue_struct *wq;
1037 * Create the rpciod thread and wait for it to start.
1039 dprintk("RPC: creating workqueue rpciod\n");
1040 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 1);
1041 rpciod_workqueue = wq;
1042 return rpciod_workqueue != NULL;
1045 static void rpciod_stop(void)
1047 struct workqueue_struct *wq = NULL;
1049 if (rpciod_workqueue == NULL)
1051 dprintk("RPC: destroying workqueue rpciod\n");
1053 wq = rpciod_workqueue;
1054 rpciod_workqueue = NULL;
1055 destroy_workqueue(wq);
1059 rpc_destroy_mempool(void)
1062 if (rpc_buffer_mempool)
1063 mempool_destroy(rpc_buffer_mempool);
1064 if (rpc_task_mempool)
1065 mempool_destroy(rpc_task_mempool);
1067 kmem_cache_destroy(rpc_task_slabp);
1068 if (rpc_buffer_slabp)
1069 kmem_cache_destroy(rpc_buffer_slabp);
1070 rpc_destroy_wait_queue(&delay_queue);
1074 rpc_init_mempool(void)
1077 * The following is not strictly a mempool initialisation,
1078 * but there is no harm in doing it here
1080 rpc_init_wait_queue(&delay_queue, "delayq");
1081 if (!rpciod_start())
1084 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1085 sizeof(struct rpc_task),
1086 0, SLAB_HWCACHE_ALIGN,
1088 if (!rpc_task_slabp)
1090 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1092 0, SLAB_HWCACHE_ALIGN,
1094 if (!rpc_buffer_slabp)
1096 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1098 if (!rpc_task_mempool)
1100 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1102 if (!rpc_buffer_mempool)
1106 rpc_destroy_mempool();