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>
22 #include <linux/sunrpc/clnt.h>
27 #define RPCDBG_FACILITY RPCDBG_SCHED
31 * RPC slabs and memory pools
33 #define RPC_BUFFER_MAXSIZE (2048)
34 #define RPC_BUFFER_POOLSIZE (8)
35 #define RPC_TASK_POOLSIZE (8)
36 static struct kmem_cache *rpc_task_slabp __read_mostly;
37 static struct kmem_cache *rpc_buffer_slabp __read_mostly;
38 static mempool_t *rpc_task_mempool __read_mostly;
39 static mempool_t *rpc_buffer_mempool __read_mostly;
41 static void rpc_async_schedule(struct work_struct *);
42 static void rpc_release_task(struct rpc_task *task);
43 static void __rpc_queue_timer_fn(unsigned long ptr);
46 * RPC tasks sit here while waiting for conditions to improve.
48 static struct rpc_wait_queue delay_queue;
51 * rpciod-related stuff
53 struct workqueue_struct *rpciod_workqueue;
56 * Disable the timer for a given RPC task. Should be called with
57 * queue->lock and bh_disabled in order to avoid races within
61 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
63 if (task->tk_timeout == 0)
65 dprintk("RPC: %5u disabling timer\n", task->tk_pid);
67 list_del(&task->u.tk_wait.timer_list);
68 if (list_empty(&queue->timer_list.list))
69 del_timer(&queue->timer_list.timer);
73 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
75 queue->timer_list.expires = expires;
76 mod_timer(&queue->timer_list.timer, expires);
80 * Set up a timer for the current task.
83 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
85 if (!task->tk_timeout)
88 dprintk("RPC: %5u setting alarm for %lu ms\n",
89 task->tk_pid, task->tk_timeout * 1000 / HZ);
91 task->u.tk_wait.expires = jiffies + task->tk_timeout;
92 if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
93 rpc_set_queue_timer(queue, task->u.tk_wait.expires);
94 list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
98 * Add new request to a priority queue.
100 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
105 INIT_LIST_HEAD(&task->u.tk_wait.links);
106 q = &queue->tasks[task->tk_priority];
107 if (unlikely(task->tk_priority > queue->maxpriority))
108 q = &queue->tasks[queue->maxpriority];
109 list_for_each_entry(t, q, u.tk_wait.list) {
110 if (t->tk_owner == task->tk_owner) {
111 list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
115 list_add_tail(&task->u.tk_wait.list, q);
119 * Add new request to wait queue.
121 * Swapper tasks always get inserted at the head of the queue.
122 * This should avoid many nasty memory deadlocks and hopefully
123 * improve overall performance.
124 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
126 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
128 BUG_ON (RPC_IS_QUEUED(task));
130 if (RPC_IS_PRIORITY(queue))
131 __rpc_add_wait_queue_priority(queue, task);
132 else if (RPC_IS_SWAPPER(task))
133 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
135 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
136 task->tk_waitqueue = queue;
138 rpc_set_queued(task);
140 dprintk("RPC: %5u added to queue %p \"%s\"\n",
141 task->tk_pid, queue, rpc_qname(queue));
145 * Remove request from a priority queue.
147 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
151 if (!list_empty(&task->u.tk_wait.links)) {
152 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
153 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
154 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
159 * Remove request from queue.
160 * Note: must be called with spin lock held.
162 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
164 __rpc_disable_timer(queue, task);
165 if (RPC_IS_PRIORITY(queue))
166 __rpc_remove_wait_queue_priority(task);
167 list_del(&task->u.tk_wait.list);
169 dprintk("RPC: %5u removed from queue %p \"%s\"\n",
170 task->tk_pid, queue, rpc_qname(queue));
173 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
175 queue->priority = priority;
176 queue->count = 1 << (priority * 2);
179 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
182 queue->nr = RPC_BATCH_COUNT;
185 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
187 rpc_set_waitqueue_priority(queue, queue->maxpriority);
188 rpc_set_waitqueue_owner(queue, 0);
191 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
195 spin_lock_init(&queue->lock);
196 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
197 INIT_LIST_HEAD(&queue->tasks[i]);
198 queue->maxpriority = nr_queues - 1;
199 rpc_reset_waitqueue_priority(queue);
201 setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
202 INIT_LIST_HEAD(&queue->timer_list.list);
208 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
210 __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
212 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
214 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
216 __rpc_init_priority_wait_queue(queue, qname, 1);
218 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
220 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
222 del_timer_sync(&queue->timer_list.timer);
224 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
226 static int rpc_wait_bit_killable(void *word)
228 if (fatal_signal_pending(current))
235 static void rpc_task_set_debuginfo(struct rpc_task *task)
237 static atomic_t rpc_pid;
239 task->tk_pid = atomic_inc_return(&rpc_pid);
242 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
247 static void rpc_set_active(struct rpc_task *task)
249 rpc_task_set_debuginfo(task);
250 set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
254 * Mark an RPC call as having completed by clearing the 'active' bit
255 * and then waking up all tasks that were sleeping.
257 static int rpc_complete_task(struct rpc_task *task)
259 void *m = &task->tk_runstate;
260 wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
261 struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
265 spin_lock_irqsave(&wq->lock, flags);
266 clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
267 ret = atomic_dec_and_test(&task->tk_count);
268 if (waitqueue_active(wq))
269 __wake_up_locked_key(wq, TASK_NORMAL, &k);
270 spin_unlock_irqrestore(&wq->lock, flags);
275 * Allow callers to wait for completion of an RPC call
277 * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
278 * to enforce taking of the wq->lock and hence avoid races with
279 * rpc_complete_task().
281 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
284 action = rpc_wait_bit_killable;
285 return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
286 action, TASK_KILLABLE);
288 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
291 * Make an RPC task runnable.
293 * Note: If the task is ASYNC, this must be called with
294 * the spinlock held to protect the wait queue operation.
296 static void rpc_make_runnable(struct rpc_task *task)
298 rpc_clear_queued(task);
299 if (rpc_test_and_set_running(task))
301 if (RPC_IS_ASYNC(task)) {
302 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
303 queue_work(rpciod_workqueue, &task->u.tk_work);
305 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
309 * Prepare for sleeping on a wait queue.
310 * By always appending tasks to the list we ensure FIFO behavior.
311 * NB: An RPC task will only receive interrupt-driven events as long
312 * as it's on a wait queue.
314 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
317 dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
318 task->tk_pid, rpc_qname(q), jiffies);
320 __rpc_add_wait_queue(q, task);
322 BUG_ON(task->tk_callback != NULL);
323 task->tk_callback = action;
324 __rpc_add_timer(q, task);
327 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
330 /* We shouldn't ever put an inactive task to sleep */
331 BUG_ON(!RPC_IS_ACTIVATED(task));
334 * Protect the queue operations.
336 spin_lock_bh(&q->lock);
337 __rpc_sleep_on(q, task, action);
338 spin_unlock_bh(&q->lock);
340 EXPORT_SYMBOL_GPL(rpc_sleep_on);
343 * __rpc_do_wake_up_task - wake up a single rpc_task
345 * @task: task to be woken up
347 * Caller must hold queue->lock, and have cleared the task queued flag.
349 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
351 dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
352 task->tk_pid, jiffies);
354 /* Has the task been executed yet? If not, we cannot wake it up! */
355 if (!RPC_IS_ACTIVATED(task)) {
356 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
360 __rpc_remove_wait_queue(queue, task);
362 rpc_make_runnable(task);
364 dprintk("RPC: __rpc_wake_up_task done\n");
368 * Wake up a queued task while the queue lock is being held
370 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
372 if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
373 __rpc_do_wake_up_task(queue, task);
377 * Tests whether rpc queue is empty
379 int rpc_queue_empty(struct rpc_wait_queue *queue)
383 spin_lock_bh(&queue->lock);
385 spin_unlock_bh(&queue->lock);
388 EXPORT_SYMBOL_GPL(rpc_queue_empty);
391 * Wake up a task on a specific queue
393 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
395 spin_lock_bh(&queue->lock);
396 rpc_wake_up_task_queue_locked(queue, task);
397 spin_unlock_bh(&queue->lock);
399 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
402 * Wake up the next task on a priority queue.
404 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
407 struct rpc_task *task;
410 * Service a batch of tasks from a single owner.
412 q = &queue->tasks[queue->priority];
413 if (!list_empty(q)) {
414 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
415 if (queue->owner == task->tk_owner) {
418 list_move_tail(&task->u.tk_wait.list, q);
421 * Check if we need to switch queues.
428 * Service the next queue.
431 if (q == &queue->tasks[0])
432 q = &queue->tasks[queue->maxpriority];
435 if (!list_empty(q)) {
436 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
439 } while (q != &queue->tasks[queue->priority]);
441 rpc_reset_waitqueue_priority(queue);
445 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
447 rpc_set_waitqueue_owner(queue, task->tk_owner);
449 rpc_wake_up_task_queue_locked(queue, task);
454 * Wake up the next task on the wait queue.
456 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
458 struct rpc_task *task = NULL;
460 dprintk("RPC: wake_up_next(%p \"%s\")\n",
461 queue, rpc_qname(queue));
462 spin_lock_bh(&queue->lock);
463 if (RPC_IS_PRIORITY(queue))
464 task = __rpc_wake_up_next_priority(queue);
466 task_for_first(task, &queue->tasks[0])
467 rpc_wake_up_task_queue_locked(queue, task);
469 spin_unlock_bh(&queue->lock);
473 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
476 * rpc_wake_up - wake up all rpc_tasks
477 * @queue: rpc_wait_queue on which the tasks are sleeping
481 void rpc_wake_up(struct rpc_wait_queue *queue)
483 struct list_head *head;
485 spin_lock_bh(&queue->lock);
486 head = &queue->tasks[queue->maxpriority];
488 while (!list_empty(head)) {
489 struct rpc_task *task;
490 task = list_first_entry(head,
493 rpc_wake_up_task_queue_locked(queue, task);
495 if (head == &queue->tasks[0])
499 spin_unlock_bh(&queue->lock);
501 EXPORT_SYMBOL_GPL(rpc_wake_up);
504 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
505 * @queue: rpc_wait_queue on which the tasks are sleeping
506 * @status: status value to set
510 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
512 struct list_head *head;
514 spin_lock_bh(&queue->lock);
515 head = &queue->tasks[queue->maxpriority];
517 while (!list_empty(head)) {
518 struct rpc_task *task;
519 task = list_first_entry(head,
522 task->tk_status = status;
523 rpc_wake_up_task_queue_locked(queue, task);
525 if (head == &queue->tasks[0])
529 spin_unlock_bh(&queue->lock);
531 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
533 static void __rpc_queue_timer_fn(unsigned long ptr)
535 struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
536 struct rpc_task *task, *n;
537 unsigned long expires, now, timeo;
539 spin_lock(&queue->lock);
540 expires = now = jiffies;
541 list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
542 timeo = task->u.tk_wait.expires;
543 if (time_after_eq(now, timeo)) {
544 dprintk("RPC: %5u timeout\n", task->tk_pid);
545 task->tk_status = -ETIMEDOUT;
546 rpc_wake_up_task_queue_locked(queue, task);
549 if (expires == now || time_after(expires, timeo))
552 if (!list_empty(&queue->timer_list.list))
553 rpc_set_queue_timer(queue, expires);
554 spin_unlock(&queue->lock);
557 static void __rpc_atrun(struct rpc_task *task)
563 * Run a task at a later time
565 void rpc_delay(struct rpc_task *task, unsigned long delay)
567 task->tk_timeout = delay;
568 rpc_sleep_on(&delay_queue, task, __rpc_atrun);
570 EXPORT_SYMBOL_GPL(rpc_delay);
573 * Helper to call task->tk_ops->rpc_call_prepare
575 void rpc_prepare_task(struct rpc_task *task)
577 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
581 * Helper that calls task->tk_ops->rpc_call_done if it exists
583 void rpc_exit_task(struct rpc_task *task)
585 task->tk_action = NULL;
586 if (task->tk_ops->rpc_call_done != NULL) {
587 task->tk_ops->rpc_call_done(task, task->tk_calldata);
588 if (task->tk_action != NULL) {
589 WARN_ON(RPC_ASSASSINATED(task));
590 /* Always release the RPC slot and buffer memory */
596 void rpc_exit(struct rpc_task *task, int status)
598 task->tk_status = status;
599 task->tk_action = rpc_exit_task;
600 if (RPC_IS_QUEUED(task))
601 rpc_wake_up_queued_task(task->tk_waitqueue, task);
603 EXPORT_SYMBOL_GPL(rpc_exit);
605 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
607 if (ops->rpc_release != NULL)
608 ops->rpc_release(calldata);
612 * This is the RPC `scheduler' (or rather, the finite state machine).
614 static void __rpc_execute(struct rpc_task *task)
616 struct rpc_wait_queue *queue;
617 int task_is_async = RPC_IS_ASYNC(task);
620 dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
621 task->tk_pid, task->tk_flags);
623 BUG_ON(RPC_IS_QUEUED(task));
626 void (*do_action)(struct rpc_task *);
629 * Execute any pending callback first.
631 do_action = task->tk_callback;
632 task->tk_callback = NULL;
633 if (do_action == NULL) {
635 * Perform the next FSM step.
636 * tk_action may be NULL if the task has been killed.
637 * In particular, note that rpc_killall_tasks may
638 * do this at any time, so beware when dereferencing.
640 do_action = task->tk_action;
641 if (do_action == NULL)
647 * Lockless check for whether task is sleeping or not.
649 if (!RPC_IS_QUEUED(task))
652 * The queue->lock protects against races with
653 * rpc_make_runnable().
655 * Note that once we clear RPC_TASK_RUNNING on an asynchronous
656 * rpc_task, rpc_make_runnable() can assign it to a
657 * different workqueue. We therefore cannot assume that the
658 * rpc_task pointer may still be dereferenced.
660 queue = task->tk_waitqueue;
661 spin_lock_bh(&queue->lock);
662 if (!RPC_IS_QUEUED(task)) {
663 spin_unlock_bh(&queue->lock);
666 rpc_clear_running(task);
667 spin_unlock_bh(&queue->lock);
671 /* sync task: sleep here */
672 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
673 status = out_of_line_wait_on_bit(&task->tk_runstate,
674 RPC_TASK_QUEUED, rpc_wait_bit_killable,
676 if (status == -ERESTARTSYS) {
678 * When a sync task receives a signal, it exits with
679 * -ERESTARTSYS. In order to catch any callbacks that
680 * clean up after sleeping on some queue, we don't
681 * break the loop here, but go around once more.
683 dprintk("RPC: %5u got signal\n", task->tk_pid);
684 task->tk_flags |= RPC_TASK_KILLED;
685 rpc_exit(task, -ERESTARTSYS);
687 rpc_set_running(task);
688 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
691 dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
693 /* Release all resources associated with the task */
694 rpc_release_task(task);
698 * User-visible entry point to the scheduler.
700 * This may be called recursively if e.g. an async NFS task updates
701 * the attributes and finds that dirty pages must be flushed.
702 * NOTE: Upon exit of this function the task is guaranteed to be
703 * released. In particular note that tk_release() will have
704 * been called, so your task memory may have been freed.
706 void rpc_execute(struct rpc_task *task)
708 rpc_set_active(task);
709 rpc_make_runnable(task);
710 if (!RPC_IS_ASYNC(task))
714 static void rpc_async_schedule(struct work_struct *work)
716 current->flags |= PF_FSTRANS;
717 __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
718 current->flags &= ~PF_FSTRANS;
722 * rpc_malloc - allocate an RPC buffer
723 * @task: RPC task that will use this buffer
724 * @size: requested byte size
726 * To prevent rpciod from hanging, this allocator never sleeps,
727 * returning NULL if the request cannot be serviced immediately.
728 * The caller can arrange to sleep in a way that is safe for rpciod.
730 * Most requests are 'small' (under 2KiB) and can be serviced from a
731 * mempool, ensuring that NFS reads and writes can always proceed,
732 * and that there is good locality of reference for these buffers.
734 * In order to avoid memory starvation triggering more writebacks of
735 * NFS requests, we avoid using GFP_KERNEL.
737 void *rpc_malloc(struct rpc_task *task, size_t size)
739 struct rpc_buffer *buf;
740 gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
742 size += sizeof(struct rpc_buffer);
743 if (size <= RPC_BUFFER_MAXSIZE)
744 buf = mempool_alloc(rpc_buffer_mempool, gfp);
746 buf = kmalloc(size, gfp);
752 dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
753 task->tk_pid, size, buf);
756 EXPORT_SYMBOL_GPL(rpc_malloc);
759 * rpc_free - free buffer allocated via rpc_malloc
760 * @buffer: buffer to free
763 void rpc_free(void *buffer)
766 struct rpc_buffer *buf;
771 buf = container_of(buffer, struct rpc_buffer, data);
774 dprintk("RPC: freeing buffer of size %zu at %p\n",
777 if (size <= RPC_BUFFER_MAXSIZE)
778 mempool_free(buf, rpc_buffer_mempool);
782 EXPORT_SYMBOL_GPL(rpc_free);
785 * Creation and deletion of RPC task structures
787 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
789 memset(task, 0, sizeof(*task));
790 atomic_set(&task->tk_count, 1);
791 task->tk_flags = task_setup_data->flags;
792 task->tk_ops = task_setup_data->callback_ops;
793 task->tk_calldata = task_setup_data->callback_data;
794 INIT_LIST_HEAD(&task->tk_task);
796 /* Initialize retry counters */
797 task->tk_garb_retry = 2;
798 task->tk_cred_retry = 2;
799 task->tk_rebind_retry = 2;
801 task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
802 task->tk_owner = current->tgid;
804 /* Initialize workqueue for async tasks */
805 task->tk_workqueue = task_setup_data->workqueue;
807 if (task->tk_ops->rpc_call_prepare != NULL)
808 task->tk_action = rpc_prepare_task;
810 /* starting timestamp */
811 task->tk_start = ktime_get();
813 dprintk("RPC: new task initialized, procpid %u\n",
814 task_pid_nr(current));
817 static struct rpc_task *
820 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
824 * Create a new task for the specified client.
826 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
828 struct rpc_task *task = setup_data->task;
829 unsigned short flags = 0;
832 task = rpc_alloc_task();
834 rpc_release_calldata(setup_data->callback_ops,
835 setup_data->callback_data);
836 return ERR_PTR(-ENOMEM);
838 flags = RPC_TASK_DYNAMIC;
841 rpc_init_task(task, setup_data);
842 task->tk_flags |= flags;
843 dprintk("RPC: allocated task %p\n", task);
848 * rpc_free_task - release rpc task and perform cleanups
850 * Note that we free up the rpc_task _after_ rpc_release_calldata()
851 * in order to work around a workqueue dependency issue.
854 * "Workqueue currently considers two work items to be the same if they're
855 * on the same address and won't execute them concurrently - ie. it
856 * makes a work item which is queued again while being executed wait
857 * for the previous execution to complete.
859 * If a work function frees the work item, and then waits for an event
860 * which should be performed by another work item and *that* work item
861 * recycles the freed work item, it can create a false dependency loop.
862 * There really is no reliable way to detect this short of verifying
863 * every memory free."
866 static void rpc_free_task(struct rpc_task *task)
868 unsigned short tk_flags = task->tk_flags;
870 rpc_release_calldata(task->tk_ops, task->tk_calldata);
872 if (tk_flags & RPC_TASK_DYNAMIC) {
873 dprintk("RPC: %5u freeing task\n", task->tk_pid);
874 mempool_free(task, rpc_task_mempool);
878 static void rpc_async_release(struct work_struct *work)
880 rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
883 static void rpc_release_resources_task(struct rpc_task *task)
887 if (task->tk_msg.rpc_cred) {
888 put_rpccred(task->tk_msg.rpc_cred);
889 task->tk_msg.rpc_cred = NULL;
891 rpc_task_release_client(task);
894 static void rpc_final_put_task(struct rpc_task *task,
895 struct workqueue_struct *q)
898 INIT_WORK(&task->u.tk_work, rpc_async_release);
899 queue_work(q, &task->u.tk_work);
904 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
906 if (atomic_dec_and_test(&task->tk_count)) {
907 rpc_release_resources_task(task);
908 rpc_final_put_task(task, q);
912 void rpc_put_task(struct rpc_task *task)
914 rpc_do_put_task(task, NULL);
916 EXPORT_SYMBOL_GPL(rpc_put_task);
918 void rpc_put_task_async(struct rpc_task *task)
920 rpc_do_put_task(task, task->tk_workqueue);
922 EXPORT_SYMBOL_GPL(rpc_put_task_async);
924 static void rpc_release_task(struct rpc_task *task)
926 dprintk("RPC: %5u release task\n", task->tk_pid);
928 BUG_ON (RPC_IS_QUEUED(task));
930 rpc_release_resources_task(task);
933 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
934 * so it should be safe to use task->tk_count as a test for whether
935 * or not any other processes still hold references to our rpc_task.
937 if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
938 /* Wake up anyone who may be waiting for task completion */
939 if (!rpc_complete_task(task))
942 if (!atomic_dec_and_test(&task->tk_count))
945 rpc_final_put_task(task, task->tk_workqueue);
950 return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
953 void rpciod_down(void)
955 module_put(THIS_MODULE);
959 * Start up the rpciod workqueue.
961 static int rpciod_start(void)
963 struct workqueue_struct *wq;
966 * Create the rpciod thread and wait for it to start.
968 dprintk("RPC: creating workqueue rpciod\n");
969 wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
970 rpciod_workqueue = wq;
971 return rpciod_workqueue != NULL;
974 static void rpciod_stop(void)
976 struct workqueue_struct *wq = NULL;
978 if (rpciod_workqueue == NULL)
980 dprintk("RPC: destroying workqueue rpciod\n");
982 wq = rpciod_workqueue;
983 rpciod_workqueue = NULL;
984 destroy_workqueue(wq);
988 rpc_destroy_mempool(void)
991 if (rpc_buffer_mempool)
992 mempool_destroy(rpc_buffer_mempool);
993 if (rpc_task_mempool)
994 mempool_destroy(rpc_task_mempool);
996 kmem_cache_destroy(rpc_task_slabp);
997 if (rpc_buffer_slabp)
998 kmem_cache_destroy(rpc_buffer_slabp);
999 rpc_destroy_wait_queue(&delay_queue);
1003 rpc_init_mempool(void)
1006 * The following is not strictly a mempool initialisation,
1007 * but there is no harm in doing it here
1009 rpc_init_wait_queue(&delay_queue, "delayq");
1010 if (!rpciod_start())
1013 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1014 sizeof(struct rpc_task),
1015 0, SLAB_HWCACHE_ALIGN,
1017 if (!rpc_task_slabp)
1019 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1021 0, SLAB_HWCACHE_ALIGN,
1023 if (!rpc_buffer_slabp)
1025 rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1027 if (!rpc_task_mempool)
1029 rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1031 if (!rpc_buffer_mempool)
1035 rpc_destroy_mempool();