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 BUG_ON (RPC_IS_QUEUED(task));
138 if (RPC_IS_PRIORITY(queue))
139 __rpc_add_wait_queue_priority(queue, task, queue_priority);
140 else if (RPC_IS_SWAPPER(task))
141 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
143 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
144 task->tk_waitqueue = queue;
146 rpc_set_queued(task);
148 dprintk("RPC: %5u added to queue %p \"%s\"\n",
149 task->tk_pid, queue, rpc_qname(queue));
153 * Remove request from a priority queue.
155 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
159 if (!list_empty(&task->u.tk_wait.links)) {
160 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
161 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
162 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
167 * Remove request from queue.
168 * Note: must be called with spin lock held.
170 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
172 __rpc_disable_timer(queue, task);
173 if (RPC_IS_PRIORITY(queue))
174 __rpc_remove_wait_queue_priority(task);
175 list_del(&task->u.tk_wait.list);
177 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
178 task->tk_pid, queue, rpc_qname(queue));
181 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
183 queue->priority = priority;
184 queue->count = 1 << (priority * 2);
187 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
190 queue->nr = RPC_BATCH_COUNT;
193 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
195 rpc_set_waitqueue_priority(queue, queue->maxpriority);
196 rpc_set_waitqueue_owner(queue, 0);
199 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
203 spin_lock_init(&queue->lock);
204 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
205 INIT_LIST_HEAD(&queue->tasks[i]);
206 queue->maxpriority = nr_queues - 1;
207 rpc_reset_waitqueue_priority(queue);
209 setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
210 INIT_LIST_HEAD(&queue->timer_list.list);
211 rpc_assign_waitqueue_name(queue, qname);
214 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
216 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
218 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
220 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
222 __rpc_init_priority_wait_queue(queue, qname, 1);
224 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
226 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
228 del_timer_sync(&queue->timer_list.timer);
230 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
232 static int rpc_wait_bit_killable(void *word)
234 if (fatal_signal_pending(current))
236 freezable_schedule();
241 static void rpc_task_set_debuginfo(struct rpc_task *task)
243 static atomic_t rpc_pid;
245 task->tk_pid = atomic_inc_return(&rpc_pid);
248 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
253 static void rpc_set_active(struct rpc_task *task)
255 trace_rpc_task_begin(task->tk_client, task, NULL);
257 rpc_task_set_debuginfo(task);
258 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
262 * Mark an RPC call as having completed by clearing the 'active' bit
263 * and then waking up all tasks that were sleeping.
265 static int rpc_complete_task(struct rpc_task *task)
267 void *m = &task->tk_runstate;
268 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
269 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
273 trace_rpc_task_complete(task->tk_client, task, NULL);
275 spin_lock_irqsave(&wq->lock, flags);
276 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
277 ret = atomic_dec_and_test(&task->tk_count);
278 if (waitqueue_active(wq))
279 __wake_up_locked_key(wq, TASK_NORMAL, &k);
280 spin_unlock_irqrestore(&wq->lock, flags);
285 * Allow callers to wait for completion of an RPC call
287 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
288 * to enforce taking of the wq->lock and hence avoid races with
289 * rpc_complete_task().
291 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
294 action = rpc_wait_bit_killable;
295 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
296 action, TASK_KILLABLE);
298 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
301 * Make an RPC task runnable.
303 * Note: If the task is ASYNC, this must be called with
304 * the spinlock held to protect the wait queue operation.
306 static void rpc_make_runnable(struct rpc_task *task)
308 rpc_clear_queued(task);
309 if (rpc_test_and_set_running(task))
311 if (RPC_IS_ASYNC(task)) {
312 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
313 queue_work(rpciod_workqueue, &task->u.tk_work);
315 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
319 * Prepare for sleeping on a wait queue.
320 * By always appending tasks to the list we ensure FIFO behavior.
321 * NB: An RPC task will only receive interrupt-driven events as long
322 * as it's on a wait queue.
324 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
325 struct rpc_task *task,
327 unsigned char queue_priority)
329 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
330 task->tk_pid, rpc_qname(q), jiffies);
332 trace_rpc_task_sleep(task->tk_client, task, q);
334 __rpc_add_wait_queue(q, task, queue_priority);
336 BUG_ON(task->tk_callback != NULL);
337 task->tk_callback = action;
338 __rpc_add_timer(q, task);
341 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
344 /* We shouldn't ever put an inactive task to sleep */
345 BUG_ON(!RPC_IS_ACTIVATED(task));
348 * Protect the queue operations.
350 spin_lock_bh(&q->lock);
351 __rpc_sleep_on_priority(q, task, action, task->tk_priority);
352 spin_unlock_bh(&q->lock);
354 EXPORT_SYMBOL_GPL(rpc_sleep_on);
356 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
357 rpc_action action, int priority)
359 /* We shouldn't ever put an inactive task to sleep */
360 BUG_ON(!RPC_IS_ACTIVATED(task));
363 * Protect the queue operations.
365 spin_lock_bh(&q->lock);
366 __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
367 spin_unlock_bh(&q->lock);
371 * __rpc_do_wake_up_task - wake up a single rpc_task
373 * @task: task to be woken up
375 * Caller must hold queue->lock, and have cleared the task queued flag.
377 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
379 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
380 task->tk_pid, jiffies);
382 /* Has the task been executed yet? If not, we cannot wake it up! */
383 if (!RPC_IS_ACTIVATED(task)) {
384 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
388 trace_rpc_task_wakeup(task->tk_client, task, queue);
390 __rpc_remove_wait_queue(queue, task);
392 rpc_make_runnable(task);
394 dprintk("RPC: __rpc_wake_up_task done\n");
398 * Wake up a queued task while the queue lock is being held
400 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
402 if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
403 __rpc_do_wake_up_task(queue, task);
407 * Tests whether rpc queue is empty
409 int rpc_queue_empty(struct rpc_wait_queue *queue)
413 spin_lock_bh(&queue->lock);
415 spin_unlock_bh(&queue->lock);
418 EXPORT_SYMBOL_GPL(rpc_queue_empty);
421 * Wake up a task on a specific queue
423 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
425 spin_lock_bh(&queue->lock);
426 rpc_wake_up_task_queue_locked(queue, task);
427 spin_unlock_bh(&queue->lock);
429 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
432 * Wake up the next task on a priority queue.
434 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
437 struct rpc_task *task;
440 * Service a batch of tasks from a single owner.
442 q = &queue->tasks[queue->priority];
443 if (!list_empty(q)) {
444 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
445 if (queue->owner == task->tk_owner) {
448 list_move_tail(&task->u.tk_wait.list, q);
451 * Check if we need to switch queues.
458 * Service the next queue.
461 if (q == &queue->tasks[0])
462 q = &queue->tasks[queue->maxpriority];
465 if (!list_empty(q)) {
466 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
469 } while (q != &queue->tasks[queue->priority]);
471 rpc_reset_waitqueue_priority(queue);
475 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
477 rpc_set_waitqueue_owner(queue, task->tk_owner);
482 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
484 if (RPC_IS_PRIORITY(queue))
485 return __rpc_find_next_queued_priority(queue);
486 if (!list_empty(&queue->tasks[0]))
487 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
492 * Wake up the first task on the wait queue.
494 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
495 bool (*func)(struct rpc_task *, void *), void *data)
497 struct rpc_task *task = NULL;
499 dprintk("RPC: wake_up_first(%p \"%s\")\n",
500 queue, rpc_qname(queue));
501 spin_lock_bh(&queue->lock);
502 task = __rpc_find_next_queued(queue);
504 if (func(task, data))
505 rpc_wake_up_task_queue_locked(queue, task);
509 spin_unlock_bh(&queue->lock);
513 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
515 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
521 * Wake up the next task on the wait queue.
523 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
525 return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
527 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
530 * rpc_wake_up - wake up all rpc_tasks
531 * @queue: rpc_wait_queue on which the tasks are sleeping
535 void rpc_wake_up(struct rpc_wait_queue *queue)
537 struct list_head *head;
539 spin_lock_bh(&queue->lock);
540 head = &queue->tasks[queue->maxpriority];
542 while (!list_empty(head)) {
543 struct rpc_task *task;
544 task = list_first_entry(head,
547 rpc_wake_up_task_queue_locked(queue, task);
549 if (head == &queue->tasks[0])
553 spin_unlock_bh(&queue->lock);
555 EXPORT_SYMBOL_GPL(rpc_wake_up);
558 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
559 * @queue: rpc_wait_queue on which the tasks are sleeping
560 * @status: status value to set
564 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
566 struct list_head *head;
568 spin_lock_bh(&queue->lock);
569 head = &queue->tasks[queue->maxpriority];
571 while (!list_empty(head)) {
572 struct rpc_task *task;
573 task = list_first_entry(head,
576 task->tk_status = status;
577 rpc_wake_up_task_queue_locked(queue, task);
579 if (head == &queue->tasks[0])
583 spin_unlock_bh(&queue->lock);
585 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
588 * rpc_wake_up_softconn_status - wake up all SOFTCONN rpc_tasks and set their
590 * @queue: rpc_wait_queue on which the tasks are sleeping
591 * @status: status value to set
595 void rpc_wake_up_softconn_status(struct rpc_wait_queue *queue, int status)
597 struct rpc_task *task, *next;
598 struct list_head *head;
600 spin_lock_bh(&queue->lock);
601 head = &queue->tasks[queue->maxpriority];
603 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
604 if (RPC_IS_SOFTCONN(task)) {
605 task->tk_status = status;
606 rpc_wake_up_task_queue_locked(queue, task);
608 if (head == &queue->tasks[0])
612 spin_unlock_bh(&queue->lock);
614 EXPORT_SYMBOL_GPL(rpc_wake_up_softconn_status);
616 static void __rpc_queue_timer_fn(unsigned long ptr)
618 struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
619 struct rpc_task *task, *n;
620 unsigned long expires, now, timeo;
622 spin_lock(&queue->lock);
623 expires = now = jiffies;
624 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
625 timeo = task->u.tk_wait.expires;
626 if (time_after_eq(now, timeo)) {
627 dprintk("RPC: %5u timeout\n", task->tk_pid);
628 task->tk_status = -ETIMEDOUT;
629 rpc_wake_up_task_queue_locked(queue, task);
632 if (expires == now || time_after(expires, timeo))
635 if (!list_empty(&queue->timer_list.list))
636 rpc_set_queue_timer(queue, expires);
637 spin_unlock(&queue->lock);
640 static void __rpc_atrun(struct rpc_task *task)
646 * Run a task at a later time
648 void rpc_delay(struct rpc_task *task, unsigned long delay)
650 task->tk_timeout = delay;
651 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
653 EXPORT_SYMBOL_GPL(rpc_delay);
656 * Helper to call task->tk_ops->rpc_call_prepare
658 void rpc_prepare_task(struct rpc_task *task)
660 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
664 rpc_init_task_statistics(struct rpc_task *task)
666 /* Initialize retry counters */
667 task->tk_garb_retry = 2;
668 task->tk_cred_retry = 2;
669 task->tk_rebind_retry = 2;
671 /* starting timestamp */
672 task->tk_start = ktime_get();
676 rpc_reset_task_statistics(struct rpc_task *task)
678 task->tk_timeouts = 0;
679 task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
681 rpc_init_task_statistics(task);
685 * Helper that calls task->tk_ops->rpc_call_done if it exists
687 void rpc_exit_task(struct rpc_task *task)
689 task->tk_action = NULL;
690 if (task->tk_ops->rpc_call_done != NULL) {
691 task->tk_ops->rpc_call_done(task, task->tk_calldata);
692 if (task->tk_action != NULL) {
693 WARN_ON(RPC_ASSASSINATED(task));
694 /* Always release the RPC slot and buffer memory */
696 rpc_reset_task_statistics(task);
701 void rpc_exit(struct rpc_task *task, int status)
703 task->tk_status = status;
704 task->tk_action = rpc_exit_task;
705 if (RPC_IS_QUEUED(task))
706 rpc_wake_up_queued_task(task->tk_waitqueue, task);
708 EXPORT_SYMBOL_GPL(rpc_exit);
710 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
712 if (ops->rpc_release != NULL)
713 ops->rpc_release(calldata);
717 * This is the RPC `scheduler' (or rather, the finite state machine).
719 static void __rpc_execute(struct rpc_task *task)
721 struct rpc_wait_queue *queue;
722 int task_is_async = RPC_IS_ASYNC(task);
725 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
726 task->tk_pid, task->tk_flags);
728 BUG_ON(RPC_IS_QUEUED(task));
731 void (*do_action)(struct rpc_task *);
734 * Execute any pending callback first.
736 do_action = task->tk_callback;
737 task->tk_callback = NULL;
738 if (do_action == NULL) {
740 * Perform the next FSM step.
741 * tk_action may be NULL if the task has been killed.
742 * In particular, note that rpc_killall_tasks may
743 * do this at any time, so beware when dereferencing.
745 do_action = task->tk_action;
746 if (do_action == NULL)
749 trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
753 * Lockless check for whether task is sleeping or not.
755 if (!RPC_IS_QUEUED(task))
758 * The queue->lock protects against races with
759 * rpc_make_runnable().
761 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
762 * rpc_task, rpc_make_runnable() can assign it to a
763 * different workqueue. We therefore cannot assume that the
764 * rpc_task pointer may still be dereferenced.
766 queue = task->tk_waitqueue;
767 spin_lock_bh(&queue->lock);
768 if (!RPC_IS_QUEUED(task)) {
769 spin_unlock_bh(&queue->lock);
772 rpc_clear_running(task);
773 spin_unlock_bh(&queue->lock);
777 /* sync task: sleep here */
778 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
779 status = out_of_line_wait_on_bit(&task->tk_runstate,
780 RPC_TASK_QUEUED, rpc_wait_bit_killable,
782 if (status == -ERESTARTSYS) {
784 * When a sync task receives a signal, it exits with
785 * -ERESTARTSYS. In order to catch any callbacks that
786 * clean up after sleeping on some queue, we don't
787 * break the loop here, but go around once more.
789 dprintk("RPC: %5u got signal\n", task->tk_pid);
790 task->tk_flags |= RPC_TASK_KILLED;
791 rpc_exit(task, -ERESTARTSYS);
793 rpc_set_running(task);
794 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
797 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
799 /* Release all resources associated with the task */
800 rpc_release_task(task);
804 * User-visible entry point to the scheduler.
806 * This may be called recursively if e.g. an async NFS task updates
807 * the attributes and finds that dirty pages must be flushed.
808 * NOTE: Upon exit of this function the task is guaranteed to be
809 * released. In particular note that tk_release() will have
810 * been called, so your task memory may have been freed.
812 void rpc_execute(struct rpc_task *task)
814 rpc_set_active(task);
815 rpc_make_runnable(task);
816 if (!RPC_IS_ASYNC(task))
820 static void rpc_async_schedule(struct work_struct *work)
822 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
826 * rpc_malloc - allocate an RPC buffer
827 * @task: RPC task that will use this buffer
828 * @size: requested byte size
830 * To prevent rpciod from hanging, this allocator never sleeps,
831 * returning NULL if the request cannot be serviced immediately.
832 * The caller can arrange to sleep in a way that is safe for rpciod.
834 * Most requests are 'small' (under 2KiB) and can be serviced from a
835 * mempool, ensuring that NFS reads and writes can always proceed,
836 * and that there is good locality of reference for these buffers.
838 * In order to avoid memory starvation triggering more writebacks of
839 * NFS requests, we avoid using GFP_KERNEL.
841 void *rpc_malloc(struct rpc_task *task, size_t size)
843 struct rpc_buffer *buf;
844 gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
846 size += sizeof(struct rpc_buffer);
847 if (size <= RPC_BUFFER_MAXSIZE)
848 buf = mempool_alloc(rpc_buffer_mempool, gfp);
850 buf = kmalloc(size, gfp);
856 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
857 task->tk_pid, size, buf);
860 EXPORT_SYMBOL_GPL(rpc_malloc);
863 * rpc_free - free buffer allocated via rpc_malloc
864 * @buffer: buffer to free
867 void rpc_free(void *buffer)
870 struct rpc_buffer *buf;
875 buf = container_of(buffer, struct rpc_buffer, data);
878 dprintk("RPC: freeing buffer of size %zu at %p\n",
881 if (size <= RPC_BUFFER_MAXSIZE)
882 mempool_free(buf, rpc_buffer_mempool);
886 EXPORT_SYMBOL_GPL(rpc_free);
889 * Creation and deletion of RPC task structures
891 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
893 memset(task, 0, sizeof(*task));
894 atomic_set(&task->tk_count, 1);
895 task->tk_flags = task_setup_data->flags;
896 task->tk_ops = task_setup_data->callback_ops;
897 task->tk_calldata = task_setup_data->callback_data;
898 INIT_LIST_HEAD(&task->tk_task);
900 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
901 task->tk_owner = current->tgid;
903 /* Initialize workqueue for async tasks */
904 task->tk_workqueue = task_setup_data->workqueue;
906 if (task->tk_ops->rpc_call_prepare != NULL)
907 task->tk_action = rpc_prepare_task;
909 rpc_init_task_statistics(task);
911 dprintk("RPC: new task initialized, procpid %u\n",
912 task_pid_nr(current));
915 static struct rpc_task *
918 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
922 * Create a new task for the specified client.
924 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
926 struct rpc_task *task = setup_data->task;
927 unsigned short flags = 0;
930 task = rpc_alloc_task();
932 rpc_release_calldata(setup_data->callback_ops,
933 setup_data->callback_data);
934 return ERR_PTR(-ENOMEM);
936 flags = RPC_TASK_DYNAMIC;
939 rpc_init_task(task, setup_data);
940 task->tk_flags |= flags;
941 dprintk("RPC: allocated task %p\n", task);
945 static void rpc_free_task(struct rpc_task *task)
947 const struct rpc_call_ops *tk_ops = task->tk_ops;
948 void *calldata = task->tk_calldata;
950 if (task->tk_flags & RPC_TASK_DYNAMIC) {
951 dprintk("RPC: %5u freeing task\n", task->tk_pid);
952 mempool_free(task, rpc_task_mempool);
954 rpc_release_calldata(tk_ops, calldata);
957 static void rpc_async_release(struct work_struct *work)
959 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
962 static void rpc_release_resources_task(struct rpc_task *task)
966 if (task->tk_msg.rpc_cred) {
967 put_rpccred(task->tk_msg.rpc_cred);
968 task->tk_msg.rpc_cred = NULL;
970 rpc_task_release_client(task);
973 static void rpc_final_put_task(struct rpc_task *task,
974 struct workqueue_struct *q)
977 INIT_WORK(&task->u.tk_work, rpc_async_release);
978 queue_work(q, &task->u.tk_work);
983 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
985 if (atomic_dec_and_test(&task->tk_count)) {
986 rpc_release_resources_task(task);
987 rpc_final_put_task(task, q);
991 void rpc_put_task(struct rpc_task *task)
993 rpc_do_put_task(task, NULL);
995 EXPORT_SYMBOL_GPL(rpc_put_task);
997 void rpc_put_task_async(struct rpc_task *task)
999 rpc_do_put_task(task, task->tk_workqueue);
1001 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1003 static void rpc_release_task(struct rpc_task *task)
1005 dprintk("RPC: %5u release task\n", task->tk_pid);
1007 BUG_ON (RPC_IS_QUEUED(task));
1009 rpc_release_resources_task(task);
1012 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1013 * so it should be safe to use task->tk_count as a test for whether
1014 * or not any other processes still hold references to our rpc_task.
1016 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1017 /* Wake up anyone who may be waiting for task completion */
1018 if (!rpc_complete_task(task))
1021 if (!atomic_dec_and_test(&task->tk_count))
1024 rpc_final_put_task(task, task->tk_workqueue);
1029 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1032 void rpciod_down(void)
1034 module_put(THIS_MODULE);
1038 * Start up the rpciod workqueue.
1040 static int rpciod_start(void)
1042 struct workqueue_struct *wq;
1045 * Create the rpciod thread and wait for it to start.
1047 dprintk("RPC: creating workqueue rpciod\n");
1048 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
1049 rpciod_workqueue = wq;
1050 return rpciod_workqueue != NULL;
1053 static void rpciod_stop(void)
1055 struct workqueue_struct *wq = NULL;
1057 if (rpciod_workqueue == NULL)
1059 dprintk("RPC: destroying workqueue rpciod\n");
1061 wq = rpciod_workqueue;
1062 rpciod_workqueue = NULL;
1063 destroy_workqueue(wq);
1067 rpc_destroy_mempool(void)
1070 if (rpc_buffer_mempool)
1071 mempool_destroy(rpc_buffer_mempool);
1072 if (rpc_task_mempool)
1073 mempool_destroy(rpc_task_mempool);
1075 kmem_cache_destroy(rpc_task_slabp);
1076 if (rpc_buffer_slabp)
1077 kmem_cache_destroy(rpc_buffer_slabp);
1078 rpc_destroy_wait_queue(&delay_queue);
1082 rpc_init_mempool(void)
1085 * The following is not strictly a mempool initialisation,
1086 * but there is no harm in doing it here
1088 rpc_init_wait_queue(&delay_queue, "delayq");
1089 if (!rpciod_start())
1092 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1093 sizeof(struct rpc_task),
1094 0, SLAB_HWCACHE_ALIGN,
1096 if (!rpc_task_slabp)
1098 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1100 0, SLAB_HWCACHE_ALIGN,
1102 if (!rpc_buffer_slabp)
1104 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1106 if (!rpc_task_mempool)
1108 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1110 if (!rpc_buffer_mempool)
1114 rpc_destroy_mempool();