blob: 802d4fe0f55cb5fc37b358b360a7bffb5ee3bbb5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/net/sunrpc/sched.c
3 *
4 * Scheduling for synchronous and asynchronous RPC requests.
5 *
6 * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
7 *
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 */
11
12#include <linux/module.h>
13
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/smp_lock.h>
20#include <linux/spinlock.h>
21
22#include <linux/sunrpc/clnt.h>
23#include <linux/sunrpc/xprt.h>
24
25#ifdef RPC_DEBUG
26#define RPCDBG_FACILITY RPCDBG_SCHED
27#define RPC_TASK_MAGIC_ID 0xf00baa
28static int rpc_task_id;
29#endif
30
31/*
32 * RPC slabs and memory pools
33 */
34#define RPC_BUFFER_MAXSIZE (2048)
35#define RPC_BUFFER_POOLSIZE (8)
36#define RPC_TASK_POOLSIZE (8)
Eric Dumazetba899662005-08-26 12:05:31 -070037static kmem_cache_t *rpc_task_slabp __read_mostly;
38static kmem_cache_t *rpc_buffer_slabp __read_mostly;
39static mempool_t *rpc_task_mempool __read_mostly;
40static mempool_t *rpc_buffer_mempool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42static void __rpc_default_timer(struct rpc_task *task);
43static void rpciod_killall(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static void rpc_async_schedule(void *);
45
46/*
47 * RPC tasks that create another task (e.g. for contacting the portmapper)
48 * will wait on this queue for their child's completion
49 */
50static RPC_WAITQ(childq, "childq");
51
52/*
53 * RPC tasks sit here while waiting for conditions to improve.
54 */
55static RPC_WAITQ(delay_queue, "delayq");
56
57/*
58 * All RPC tasks are linked into this list
59 */
60static LIST_HEAD(all_tasks);
61
62/*
63 * rpciod-related stuff
64 */
65static DECLARE_MUTEX(rpciod_sema);
66static unsigned int rpciod_users;
67static struct workqueue_struct *rpciod_workqueue;
68
69/*
70 * Spinlock for other critical sections of code.
71 */
72static DEFINE_SPINLOCK(rpc_sched_lock);
73
74/*
75 * Disable the timer for a given RPC task. Should be called with
76 * queue->lock and bh_disabled in order to avoid races within
77 * rpc_run_timer().
78 */
79static inline void
80__rpc_disable_timer(struct rpc_task *task)
81{
82 dprintk("RPC: %4d disabling timer\n", task->tk_pid);
83 task->tk_timeout_fn = NULL;
84 task->tk_timeout = 0;
85}
86
87/*
88 * Run a timeout function.
89 * We use the callback in order to allow __rpc_wake_up_task()
90 * and friends to disable the timer synchronously on SMP systems
91 * without calling del_timer_sync(). The latter could cause a
92 * deadlock if called while we're holding spinlocks...
93 */
94static void rpc_run_timer(struct rpc_task *task)
95{
96 void (*callback)(struct rpc_task *);
97
98 callback = task->tk_timeout_fn;
99 task->tk_timeout_fn = NULL;
100 if (callback && RPC_IS_QUEUED(task)) {
101 dprintk("RPC: %4d running timer\n", task->tk_pid);
102 callback(task);
103 }
104 smp_mb__before_clear_bit();
105 clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
106 smp_mb__after_clear_bit();
107}
108
109/*
110 * Set up a timer for the current task.
111 */
112static inline void
113__rpc_add_timer(struct rpc_task *task, rpc_action timer)
114{
115 if (!task->tk_timeout)
116 return;
117
118 dprintk("RPC: %4d setting alarm for %lu ms\n",
119 task->tk_pid, task->tk_timeout * 1000 / HZ);
120
121 if (timer)
122 task->tk_timeout_fn = timer;
123 else
124 task->tk_timeout_fn = __rpc_default_timer;
125 set_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
126 mod_timer(&task->tk_timer, jiffies + task->tk_timeout);
127}
128
129/*
130 * Delete any timer for the current task. Because we use del_timer_sync(),
131 * this function should never be called while holding queue->lock.
132 */
133static void
134rpc_delete_timer(struct rpc_task *task)
135{
136 if (RPC_IS_QUEUED(task))
137 return;
138 if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
139 del_singleshot_timer_sync(&task->tk_timer);
140 dprintk("RPC: %4d deleting timer\n", task->tk_pid);
141 }
142}
143
144/*
145 * Add new request to a priority queue.
146 */
147static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
148{
149 struct list_head *q;
150 struct rpc_task *t;
151
152 INIT_LIST_HEAD(&task->u.tk_wait.links);
153 q = &queue->tasks[task->tk_priority];
154 if (unlikely(task->tk_priority > queue->maxpriority))
155 q = &queue->tasks[queue->maxpriority];
156 list_for_each_entry(t, q, u.tk_wait.list) {
157 if (t->tk_cookie == task->tk_cookie) {
158 list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
159 return;
160 }
161 }
162 list_add_tail(&task->u.tk_wait.list, q);
163}
164
165/*
166 * Add new request to wait queue.
167 *
168 * Swapper tasks always get inserted at the head of the queue.
169 * This should avoid many nasty memory deadlocks and hopefully
170 * improve overall performance.
171 * Everyone else gets appended to the queue to ensure proper FIFO behavior.
172 */
173static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
174{
175 BUG_ON (RPC_IS_QUEUED(task));
176
177 if (RPC_IS_PRIORITY(queue))
178 __rpc_add_wait_queue_priority(queue, task);
179 else if (RPC_IS_SWAPPER(task))
180 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
181 else
182 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
183 task->u.tk_wait.rpc_waitq = queue;
184 rpc_set_queued(task);
185
186 dprintk("RPC: %4d added to queue %p \"%s\"\n",
187 task->tk_pid, queue, rpc_qname(queue));
188}
189
190/*
191 * Remove request from a priority queue.
192 */
193static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
194{
195 struct rpc_task *t;
196
197 if (!list_empty(&task->u.tk_wait.links)) {
198 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
199 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
200 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
201 }
202 list_del(&task->u.tk_wait.list);
203}
204
205/*
206 * Remove request from queue.
207 * Note: must be called with spin lock held.
208 */
209static void __rpc_remove_wait_queue(struct rpc_task *task)
210{
211 struct rpc_wait_queue *queue;
212 queue = task->u.tk_wait.rpc_waitq;
213
214 if (RPC_IS_PRIORITY(queue))
215 __rpc_remove_wait_queue_priority(task);
216 else
217 list_del(&task->u.tk_wait.list);
218 dprintk("RPC: %4d removed from queue %p \"%s\"\n",
219 task->tk_pid, queue, rpc_qname(queue));
220}
221
222static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
223{
224 queue->priority = priority;
225 queue->count = 1 << (priority * 2);
226}
227
228static inline void rpc_set_waitqueue_cookie(struct rpc_wait_queue *queue, unsigned long cookie)
229{
230 queue->cookie = cookie;
231 queue->nr = RPC_BATCH_COUNT;
232}
233
234static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
235{
236 rpc_set_waitqueue_priority(queue, queue->maxpriority);
237 rpc_set_waitqueue_cookie(queue, 0);
238}
239
240static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, int maxprio)
241{
242 int i;
243
244 spin_lock_init(&queue->lock);
245 for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
246 INIT_LIST_HEAD(&queue->tasks[i]);
247 queue->maxpriority = maxprio;
248 rpc_reset_waitqueue_priority(queue);
249#ifdef RPC_DEBUG
250 queue->name = qname;
251#endif
252}
253
254void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
255{
256 __rpc_init_priority_wait_queue(queue, qname, RPC_PRIORITY_HIGH);
257}
258
259void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
260{
261 __rpc_init_priority_wait_queue(queue, qname, 0);
262}
263EXPORT_SYMBOL(rpc_init_wait_queue);
264
Trond Myklebust44c28872006-01-03 09:55:06 +0100265static int rpc_wait_bit_interruptible(void *word)
266{
267 if (signal_pending(current))
268 return -ERESTARTSYS;
269 schedule();
270 return 0;
271}
272
273/*
274 * Mark an RPC call as having completed by clearing the 'active' bit
275 */
276static inline void rpc_mark_complete_task(struct rpc_task *task)
277{
278 rpc_clear_active(task);
279 wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
280}
281
282/*
283 * Allow callers to wait for completion of an RPC call
284 */
285int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
286{
287 if (action == NULL)
288 action = rpc_wait_bit_interruptible;
289 return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
290 action, TASK_INTERRUPTIBLE);
291}
292EXPORT_SYMBOL(__rpc_wait_for_completion_task);
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294/*
295 * Make an RPC task runnable.
296 *
297 * Note: If the task is ASYNC, this must be called with
298 * the spinlock held to protect the wait queue operation.
299 */
300static void rpc_make_runnable(struct rpc_task *task)
301{
302 int do_ret;
303
304 BUG_ON(task->tk_timeout_fn);
305 do_ret = rpc_test_and_set_running(task);
306 rpc_clear_queued(task);
307 if (do_ret)
308 return;
309 if (RPC_IS_ASYNC(task)) {
310 int status;
311
312 INIT_WORK(&task->u.tk_work, rpc_async_schedule, (void *)task);
313 status = queue_work(task->tk_workqueue, &task->u.tk_work);
314 if (status < 0) {
315 printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
316 task->tk_status = status;
317 return;
318 }
319 } else
Trond Myklebust96651ab2005-06-22 17:16:21 +0000320 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323/*
324 * Place a newly initialized task on the workqueue.
325 */
326static inline void
327rpc_schedule_run(struct rpc_task *task)
328{
Trond Myklebust44c28872006-01-03 09:55:06 +0100329 rpc_set_active(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 rpc_make_runnable(task);
331}
332
333/*
334 * Prepare for sleeping on a wait queue.
335 * By always appending tasks to the list we ensure FIFO behavior.
336 * NB: An RPC task will only receive interrupt-driven events as long
337 * as it's on a wait queue.
338 */
339static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
340 rpc_action action, rpc_action timer)
341{
342 dprintk("RPC: %4d sleep_on(queue \"%s\" time %ld)\n", task->tk_pid,
343 rpc_qname(q), jiffies);
344
345 if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
346 printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
347 return;
348 }
349
350 /* Mark the task as being activated if so needed */
Trond Myklebust44c28872006-01-03 09:55:06 +0100351 rpc_set_active(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 __rpc_add_wait_queue(q, task);
354
355 BUG_ON(task->tk_callback != NULL);
356 task->tk_callback = action;
357 __rpc_add_timer(task, timer);
358}
359
360void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
361 rpc_action action, rpc_action timer)
362{
363 /*
364 * Protect the queue operations.
365 */
366 spin_lock_bh(&q->lock);
367 __rpc_sleep_on(q, task, action, timer);
368 spin_unlock_bh(&q->lock);
369}
370
371/**
372 * __rpc_do_wake_up_task - wake up a single rpc_task
373 * @task: task to be woken up
374 *
375 * Caller must hold queue->lock, and have cleared the task queued flag.
376 */
377static void __rpc_do_wake_up_task(struct rpc_task *task)
378{
379 dprintk("RPC: %4d __rpc_wake_up_task (now %ld)\n", task->tk_pid, jiffies);
380
381#ifdef RPC_DEBUG
382 BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
383#endif
384 /* Has the task been executed yet? If not, we cannot wake it up! */
385 if (!RPC_IS_ACTIVATED(task)) {
386 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
387 return;
388 }
389
390 __rpc_disable_timer(task);
391 __rpc_remove_wait_queue(task);
392
393 rpc_make_runnable(task);
394
395 dprintk("RPC: __rpc_wake_up_task done\n");
396}
397
398/*
399 * Wake up the specified task
400 */
401static void __rpc_wake_up_task(struct rpc_task *task)
402{
403 if (rpc_start_wakeup(task)) {
404 if (RPC_IS_QUEUED(task))
405 __rpc_do_wake_up_task(task);
406 rpc_finish_wakeup(task);
407 }
408}
409
410/*
411 * Default timeout handler if none specified by user
412 */
413static void
414__rpc_default_timer(struct rpc_task *task)
415{
416 dprintk("RPC: %d timeout (default timer)\n", task->tk_pid);
417 task->tk_status = -ETIMEDOUT;
418 rpc_wake_up_task(task);
419}
420
421/*
422 * Wake up the specified task
423 */
424void rpc_wake_up_task(struct rpc_task *task)
425{
426 if (rpc_start_wakeup(task)) {
427 if (RPC_IS_QUEUED(task)) {
428 struct rpc_wait_queue *queue = task->u.tk_wait.rpc_waitq;
429
430 spin_lock_bh(&queue->lock);
431 __rpc_do_wake_up_task(task);
432 spin_unlock_bh(&queue->lock);
433 }
434 rpc_finish_wakeup(task);
435 }
436}
437
438/*
439 * Wake up the next task on a priority queue.
440 */
441static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
442{
443 struct list_head *q;
444 struct rpc_task *task;
445
446 /*
447 * Service a batch of tasks from a single cookie.
448 */
449 q = &queue->tasks[queue->priority];
450 if (!list_empty(q)) {
451 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
452 if (queue->cookie == task->tk_cookie) {
453 if (--queue->nr)
454 goto out;
455 list_move_tail(&task->u.tk_wait.list, q);
456 }
457 /*
458 * Check if we need to switch queues.
459 */
460 if (--queue->count)
461 goto new_cookie;
462 }
463
464 /*
465 * Service the next queue.
466 */
467 do {
468 if (q == &queue->tasks[0])
469 q = &queue->tasks[queue->maxpriority];
470 else
471 q = q - 1;
472 if (!list_empty(q)) {
473 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
474 goto new_queue;
475 }
476 } while (q != &queue->tasks[queue->priority]);
477
478 rpc_reset_waitqueue_priority(queue);
479 return NULL;
480
481new_queue:
482 rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
483new_cookie:
484 rpc_set_waitqueue_cookie(queue, task->tk_cookie);
485out:
486 __rpc_wake_up_task(task);
487 return task;
488}
489
490/*
491 * Wake up the next task on the wait queue.
492 */
493struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
494{
495 struct rpc_task *task = NULL;
496
497 dprintk("RPC: wake_up_next(%p \"%s\")\n", queue, rpc_qname(queue));
498 spin_lock_bh(&queue->lock);
499 if (RPC_IS_PRIORITY(queue))
500 task = __rpc_wake_up_next_priority(queue);
501 else {
502 task_for_first(task, &queue->tasks[0])
503 __rpc_wake_up_task(task);
504 }
505 spin_unlock_bh(&queue->lock);
506
507 return task;
508}
509
510/**
511 * rpc_wake_up - wake up all rpc_tasks
512 * @queue: rpc_wait_queue on which the tasks are sleeping
513 *
514 * Grabs queue->lock
515 */
516void rpc_wake_up(struct rpc_wait_queue *queue)
517{
518 struct rpc_task *task;
519
520 struct list_head *head;
521 spin_lock_bh(&queue->lock);
522 head = &queue->tasks[queue->maxpriority];
523 for (;;) {
524 while (!list_empty(head)) {
525 task = list_entry(head->next, struct rpc_task, u.tk_wait.list);
526 __rpc_wake_up_task(task);
527 }
528 if (head == &queue->tasks[0])
529 break;
530 head--;
531 }
532 spin_unlock_bh(&queue->lock);
533}
534
535/**
536 * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
537 * @queue: rpc_wait_queue on which the tasks are sleeping
538 * @status: status value to set
539 *
540 * Grabs queue->lock
541 */
542void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
543{
544 struct list_head *head;
545 struct rpc_task *task;
546
547 spin_lock_bh(&queue->lock);
548 head = &queue->tasks[queue->maxpriority];
549 for (;;) {
550 while (!list_empty(head)) {
551 task = list_entry(head->next, struct rpc_task, u.tk_wait.list);
552 task->tk_status = status;
553 __rpc_wake_up_task(task);
554 }
555 if (head == &queue->tasks[0])
556 break;
557 head--;
558 }
559 spin_unlock_bh(&queue->lock);
560}
561
562/*
563 * Run a task at a later time
564 */
565static void __rpc_atrun(struct rpc_task *);
566void
567rpc_delay(struct rpc_task *task, unsigned long delay)
568{
569 task->tk_timeout = delay;
570 rpc_sleep_on(&delay_queue, task, NULL, __rpc_atrun);
571}
572
573static void
574__rpc_atrun(struct rpc_task *task)
575{
576 task->tk_status = 0;
577 rpc_wake_up_task(task);
578}
579
580/*
Trond Myklebust4ce70ad2006-01-03 09:55:05 +0100581 * Helper to call task->tk_ops->rpc_call_prepare
582 */
583static void rpc_prepare_task(struct rpc_task *task)
584{
585 task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
586}
587
588/*
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100589 * Helper that calls task->tk_ops->rpc_call_done if it exists
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000590 */
Trond Myklebustabbcf282006-01-03 09:55:03 +0100591void rpc_exit_task(struct rpc_task *task)
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000592{
Trond Myklebustabbcf282006-01-03 09:55:03 +0100593 task->tk_action = NULL;
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100594 if (task->tk_ops->rpc_call_done != NULL) {
595 task->tk_ops->rpc_call_done(task, task->tk_calldata);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000596 if (task->tk_action != NULL) {
Trond Myklebustabbcf282006-01-03 09:55:03 +0100597 WARN_ON(RPC_ASSASSINATED(task));
598 /* Always release the RPC slot and buffer memory */
599 xprt_release(task);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000600 }
601 }
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000602}
Trond Myklebustabbcf282006-01-03 09:55:03 +0100603EXPORT_SYMBOL(rpc_exit_task);
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000604
605/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 * This is the RPC `scheduler' (or rather, the finite state machine).
607 */
608static int __rpc_execute(struct rpc_task *task)
609{
610 int status = 0;
611
612 dprintk("RPC: %4d rpc_execute flgs %x\n",
613 task->tk_pid, task->tk_flags);
614
615 BUG_ON(RPC_IS_QUEUED(task));
616
Trond Myklebustd05fdb02005-06-22 17:16:19 +0000617 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 /*
619 * Garbage collection of pending timers...
620 */
621 rpc_delete_timer(task);
622
623 /*
624 * Execute any pending callback.
625 */
626 if (RPC_DO_CALLBACK(task)) {
627 /* Define a callback save pointer */
628 void (*save_callback)(struct rpc_task *);
629
630 /*
631 * If a callback exists, save it, reset it,
632 * call it.
633 * The save is needed to stop from resetting
634 * another callback set within the callback handler
635 * - Dave
636 */
637 save_callback=task->tk_callback;
638 task->tk_callback=NULL;
639 lock_kernel();
640 save_callback(task);
641 unlock_kernel();
642 }
643
644 /*
645 * Perform the next FSM step.
646 * tk_action may be NULL when the task has been killed
647 * by someone else.
648 */
649 if (!RPC_IS_QUEUED(task)) {
Trond Myklebustabbcf282006-01-03 09:55:03 +0100650 if (task->tk_action == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 break;
Trond Myklebustabbcf282006-01-03 09:55:03 +0100652 lock_kernel();
653 task->tk_action(task);
654 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 }
656
657 /*
658 * Lockless check for whether task is sleeping or not.
659 */
660 if (!RPC_IS_QUEUED(task))
661 continue;
662 rpc_clear_running(task);
663 if (RPC_IS_ASYNC(task)) {
664 /* Careful! we may have raced... */
665 if (RPC_IS_QUEUED(task))
666 return 0;
667 if (rpc_test_and_set_running(task))
668 return 0;
669 continue;
670 }
671
672 /* sync task: sleep here */
673 dprintk("RPC: %4d sync task going to sleep\n", task->tk_pid);
Trond Myklebust96651ab2005-06-22 17:16:21 +0000674 /* Note: Caller should be using rpc_clnt_sigmask() */
675 status = out_of_line_wait_on_bit(&task->tk_runstate,
676 RPC_TASK_QUEUED, rpc_wait_bit_interruptible,
677 TASK_INTERRUPTIBLE);
678 if (status == -ERESTARTSYS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 /*
680 * When a sync task receives a signal, it exits with
681 * -ERESTARTSYS. In order to catch any callbacks that
682 * clean up after sleeping on some queue, we don't
683 * break the loop here, but go around once more.
684 */
Trond Myklebust96651ab2005-06-22 17:16:21 +0000685 dprintk("RPC: %4d got signal\n", task->tk_pid);
686 task->tk_flags |= RPC_TASK_KILLED;
687 rpc_exit(task, -ERESTARTSYS);
688 rpc_wake_up_task(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
690 rpc_set_running(task);
691 dprintk("RPC: %4d sync task resuming\n", task->tk_pid);
692 }
693
Trond Myklebuste60859a2006-01-03 09:55:10 +0100694 dprintk("RPC: %4d, return %d, status %d\n", task->tk_pid, status, task->tk_status);
Trond Myklebust44c28872006-01-03 09:55:06 +0100695 /* Wake up anyone who is waiting for task completion */
696 rpc_mark_complete_task(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 /* Release all resources associated with the task */
698 rpc_release_task(task);
699 return status;
700}
701
702/*
703 * User-visible entry point to the scheduler.
704 *
705 * This may be called recursively if e.g. an async NFS task updates
706 * the attributes and finds that dirty pages must be flushed.
707 * NOTE: Upon exit of this function the task is guaranteed to be
708 * released. In particular note that tk_release() will have
709 * been called, so your task memory may have been freed.
710 */
711int
712rpc_execute(struct rpc_task *task)
713{
Trond Myklebust44c28872006-01-03 09:55:06 +0100714 rpc_set_active(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 rpc_set_running(task);
716 return __rpc_execute(task);
717}
718
719static void rpc_async_schedule(void *arg)
720{
721 __rpc_execute((struct rpc_task *)arg);
722}
723
Chuck Lever02107142006-01-03 09:55:49 +0100724/**
725 * rpc_malloc - allocate an RPC buffer
726 * @task: RPC task that will use this buffer
727 * @size: requested byte size
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 *
729 * We try to ensure that some NFS reads and writes can always proceed
730 * by using a mempool when allocating 'small' buffers.
731 * In order to avoid memory starvation triggering more writebacks of
732 * NFS requests, we use GFP_NOFS rather than GFP_KERNEL.
733 */
Chuck Lever02107142006-01-03 09:55:49 +0100734void * rpc_malloc(struct rpc_task *task, size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Chuck Lever02107142006-01-03 09:55:49 +0100736 struct rpc_rqst *req = task->tk_rqstp;
Al Virodd0fc662005-10-07 07:46:04 +0100737 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 if (task->tk_flags & RPC_TASK_SWAPPER)
740 gfp = GFP_ATOMIC;
741 else
742 gfp = GFP_NOFS;
743
744 if (size > RPC_BUFFER_MAXSIZE) {
Chuck Lever02107142006-01-03 09:55:49 +0100745 req->rq_buffer = kmalloc(size, gfp);
746 if (req->rq_buffer)
747 req->rq_bufsize = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 } else {
Chuck Lever02107142006-01-03 09:55:49 +0100749 req->rq_buffer = mempool_alloc(rpc_buffer_mempool, gfp);
750 if (req->rq_buffer)
751 req->rq_bufsize = RPC_BUFFER_MAXSIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 }
Chuck Lever02107142006-01-03 09:55:49 +0100753 return req->rq_buffer;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
Chuck Lever02107142006-01-03 09:55:49 +0100756/**
757 * rpc_free - free buffer allocated via rpc_malloc
758 * @task: RPC task with a buffer to be freed
759 *
760 */
761void rpc_free(struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
Chuck Lever02107142006-01-03 09:55:49 +0100763 struct rpc_rqst *req = task->tk_rqstp;
764
765 if (req->rq_buffer) {
766 if (req->rq_bufsize == RPC_BUFFER_MAXSIZE)
767 mempool_free(req->rq_buffer, rpc_buffer_mempool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 else
Chuck Lever02107142006-01-03 09:55:49 +0100769 kfree(req->rq_buffer);
770 req->rq_buffer = NULL;
771 req->rq_bufsize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
773}
774
775/*
776 * Creation and deletion of RPC task structures
777 */
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100778void rpc_init_task(struct rpc_task *task, struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
780 memset(task, 0, sizeof(*task));
781 init_timer(&task->tk_timer);
782 task->tk_timer.data = (unsigned long) task;
783 task->tk_timer.function = (void (*)(unsigned long)) rpc_run_timer;
Trond Myklebust44c28872006-01-03 09:55:06 +0100784 atomic_set(&task->tk_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 task->tk_client = clnt;
786 task->tk_flags = flags;
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100787 task->tk_ops = tk_ops;
Trond Myklebust4ce70ad2006-01-03 09:55:05 +0100788 if (tk_ops->rpc_call_prepare != NULL)
789 task->tk_action = rpc_prepare_task;
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100790 task->tk_calldata = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792 /* Initialize retry counters */
793 task->tk_garb_retry = 2;
794 task->tk_cred_retry = 2;
795
796 task->tk_priority = RPC_PRIORITY_NORMAL;
797 task->tk_cookie = (unsigned long)current;
798
799 /* Initialize workqueue for async tasks */
800 task->tk_workqueue = rpciod_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 if (clnt) {
803 atomic_inc(&clnt->cl_users);
804 if (clnt->cl_softrtry)
805 task->tk_flags |= RPC_TASK_SOFT;
806 if (!clnt->cl_intr)
807 task->tk_flags |= RPC_TASK_NOINTR;
808 }
809
810#ifdef RPC_DEBUG
811 task->tk_magic = RPC_TASK_MAGIC_ID;
812 task->tk_pid = rpc_task_id++;
813#endif
814 /* Add to global list of all tasks */
815 spin_lock(&rpc_sched_lock);
816 list_add_tail(&task->tk_task, &all_tasks);
817 spin_unlock(&rpc_sched_lock);
818
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100819 BUG_ON(task->tk_ops == NULL);
820
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 dprintk("RPC: %4d new task procpid %d\n", task->tk_pid,
822 current->pid);
823}
824
825static struct rpc_task *
826rpc_alloc_task(void)
827{
828 return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
829}
830
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100831static void rpc_free_task(struct rpc_task *task)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 dprintk("RPC: %4d freeing task\n", task->tk_pid);
834 mempool_free(task, rpc_task_mempool);
835}
836
837/*
838 * Create a new task for the specified client. We have to
839 * clean up after an allocation failure, as the client may
840 * have specified "oneshot".
841 */
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100842struct rpc_task *rpc_new_task(struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 struct rpc_task *task;
845
846 task = rpc_alloc_task();
847 if (!task)
848 goto cleanup;
849
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100850 rpc_init_task(task, clnt, flags, tk_ops, calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 dprintk("RPC: %4d allocated task\n", task->tk_pid);
853 task->tk_flags |= RPC_TASK_DYNAMIC;
854out:
855 return task;
856
857cleanup:
858 /* Check whether to release the client */
859 if (clnt) {
860 printk("rpc_new_task: failed, users=%d, oneshot=%d\n",
861 atomic_read(&clnt->cl_users), clnt->cl_oneshot);
862 atomic_inc(&clnt->cl_users); /* pretend we were used ... */
863 rpc_release_client(clnt);
864 }
865 goto out;
866}
867
868void rpc_release_task(struct rpc_task *task)
869{
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100870 const struct rpc_call_ops *tk_ops = task->tk_ops;
871 void *calldata = task->tk_calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873#ifdef RPC_DEBUG
874 BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
875#endif
Trond Myklebust44c28872006-01-03 09:55:06 +0100876 if (!atomic_dec_and_test(&task->tk_count))
877 return;
878 dprintk("RPC: %4d release task\n", task->tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 /* Remove from global task list */
881 spin_lock(&rpc_sched_lock);
882 list_del(&task->tk_task);
883 spin_unlock(&rpc_sched_lock);
884
885 BUG_ON (RPC_IS_QUEUED(task));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 /* Synchronously delete any running timer */
888 rpc_delete_timer(task);
889
890 /* Release resources */
891 if (task->tk_rqstp)
892 xprt_release(task);
893 if (task->tk_msg.rpc_cred)
894 rpcauth_unbindcred(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 if (task->tk_client) {
896 rpc_release_client(task->tk_client);
897 task->tk_client = NULL;
898 }
899
900#ifdef RPC_DEBUG
901 task->tk_magic = 0;
902#endif
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100903 if (task->tk_flags & RPC_TASK_DYNAMIC)
904 rpc_free_task(task);
905 if (tk_ops->rpc_release)
906 tk_ops->rpc_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
909/**
Trond Myklebust44c28872006-01-03 09:55:06 +0100910 * rpc_run_task - Allocate a new RPC task, then run rpc_execute against it
Martin Waitz99acf042006-02-01 03:06:56 -0800911 * @clnt: pointer to RPC client
912 * @flags: RPC flags
913 * @ops: RPC call ops
914 * @data: user call data
Trond Myklebust44c28872006-01-03 09:55:06 +0100915 */
916struct rpc_task *rpc_run_task(struct rpc_clnt *clnt, int flags,
917 const struct rpc_call_ops *ops,
918 void *data)
919{
920 struct rpc_task *task;
921 task = rpc_new_task(clnt, flags, ops, data);
922 if (task == NULL)
923 return ERR_PTR(-ENOMEM);
924 atomic_inc(&task->tk_count);
925 rpc_execute(task);
926 return task;
927}
928EXPORT_SYMBOL(rpc_run_task);
929
930/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 * rpc_find_parent - find the parent of a child task.
932 * @child: child task
Martin Waitz99acf042006-02-01 03:06:56 -0800933 * @parent: parent task
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 *
935 * Checks that the parent task is still sleeping on the
936 * queue 'childq'. If so returns a pointer to the parent.
937 * Upon failure returns NULL.
938 *
939 * Caller must hold childq.lock
940 */
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100941static inline struct rpc_task *rpc_find_parent(struct rpc_task *child, struct rpc_task *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100943 struct rpc_task *task;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 struct list_head *le;
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 task_for_each(task, le, &childq.tasks[0])
947 if (task == parent)
948 return parent;
949
950 return NULL;
951}
952
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100953static void rpc_child_exit(struct rpc_task *child, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
955 struct rpc_task *parent;
956
957 spin_lock_bh(&childq.lock);
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100958 if ((parent = rpc_find_parent(child, calldata)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 parent->tk_status = child->tk_status;
960 __rpc_wake_up_task(parent);
961 }
962 spin_unlock_bh(&childq.lock);
963}
964
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100965static const struct rpc_call_ops rpc_child_ops = {
966 .rpc_call_done = rpc_child_exit,
967};
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969/*
970 * Note: rpc_new_task releases the client after a failure.
971 */
972struct rpc_task *
973rpc_new_child(struct rpc_clnt *clnt, struct rpc_task *parent)
974{
975 struct rpc_task *task;
976
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100977 task = rpc_new_task(clnt, RPC_TASK_ASYNC | RPC_TASK_CHILD, &rpc_child_ops, parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (!task)
979 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 return task;
981
982fail:
983 parent->tk_status = -ENOMEM;
984 return NULL;
985}
986
987void rpc_run_child(struct rpc_task *task, struct rpc_task *child, rpc_action func)
988{
989 spin_lock_bh(&childq.lock);
990 /* N.B. Is it possible for the child to have already finished? */
991 __rpc_sleep_on(&childq, task, func, NULL);
992 rpc_schedule_run(child);
993 spin_unlock_bh(&childq.lock);
994}
995
996/*
997 * Kill all tasks for the given client.
998 * XXX: kill their descendants as well?
999 */
1000void rpc_killall_tasks(struct rpc_clnt *clnt)
1001{
1002 struct rpc_task *rovr;
1003 struct list_head *le;
1004
1005 dprintk("RPC: killing all tasks for client %p\n", clnt);
1006
1007 /*
1008 * Spin lock all_tasks to prevent changes...
1009 */
1010 spin_lock(&rpc_sched_lock);
1011 alltask_for_each(rovr, le, &all_tasks) {
1012 if (! RPC_IS_ACTIVATED(rovr))
1013 continue;
1014 if (!clnt || rovr->tk_client == clnt) {
1015 rovr->tk_flags |= RPC_TASK_KILLED;
1016 rpc_exit(rovr, -EIO);
1017 rpc_wake_up_task(rovr);
1018 }
1019 }
1020 spin_unlock(&rpc_sched_lock);
1021}
1022
1023static DECLARE_MUTEX_LOCKED(rpciod_running);
1024
1025static void rpciod_killall(void)
1026{
1027 unsigned long flags;
1028
1029 while (!list_empty(&all_tasks)) {
1030 clear_thread_flag(TIF_SIGPENDING);
1031 rpc_killall_tasks(NULL);
1032 flush_workqueue(rpciod_workqueue);
1033 if (!list_empty(&all_tasks)) {
1034 dprintk("rpciod_killall: waiting for tasks to exit\n");
1035 yield();
1036 }
1037 }
1038
1039 spin_lock_irqsave(&current->sighand->siglock, flags);
1040 recalc_sigpending();
1041 spin_unlock_irqrestore(&current->sighand->siglock, flags);
1042}
1043
1044/*
1045 * Start up the rpciod process if it's not already running.
1046 */
1047int
1048rpciod_up(void)
1049{
1050 struct workqueue_struct *wq;
1051 int error = 0;
1052
1053 down(&rpciod_sema);
1054 dprintk("rpciod_up: users %d\n", rpciod_users);
1055 rpciod_users++;
1056 if (rpciod_workqueue)
1057 goto out;
1058 /*
1059 * If there's no pid, we should be the first user.
1060 */
1061 if (rpciod_users > 1)
1062 printk(KERN_WARNING "rpciod_up: no workqueue, %d users??\n", rpciod_users);
1063 /*
1064 * Create the rpciod thread and wait for it to start.
1065 */
1066 error = -ENOMEM;
1067 wq = create_workqueue("rpciod");
1068 if (wq == NULL) {
1069 printk(KERN_WARNING "rpciod_up: create workqueue failed, error=%d\n", error);
1070 rpciod_users--;
1071 goto out;
1072 }
1073 rpciod_workqueue = wq;
1074 error = 0;
1075out:
1076 up(&rpciod_sema);
1077 return error;
1078}
1079
1080void
1081rpciod_down(void)
1082{
1083 down(&rpciod_sema);
1084 dprintk("rpciod_down sema %d\n", rpciod_users);
1085 if (rpciod_users) {
1086 if (--rpciod_users)
1087 goto out;
1088 } else
1089 printk(KERN_WARNING "rpciod_down: no users??\n");
1090
1091 if (!rpciod_workqueue) {
1092 dprintk("rpciod_down: Nothing to do!\n");
1093 goto out;
1094 }
1095 rpciod_killall();
1096
1097 destroy_workqueue(rpciod_workqueue);
1098 rpciod_workqueue = NULL;
1099 out:
1100 up(&rpciod_sema);
1101}
1102
1103#ifdef RPC_DEBUG
1104void rpc_show_tasks(void)
1105{
1106 struct list_head *le;
1107 struct rpc_task *t;
1108
1109 spin_lock(&rpc_sched_lock);
1110 if (list_empty(&all_tasks)) {
1111 spin_unlock(&rpc_sched_lock);
1112 return;
1113 }
1114 printk("-pid- proc flgs status -client- -prog- --rqstp- -timeout "
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001115 "-rpcwait -action- ---ops--\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 alltask_for_each(t, le, &all_tasks) {
1117 const char *rpc_waitq = "none";
1118
1119 if (RPC_IS_QUEUED(t))
1120 rpc_waitq = rpc_qname(t->u.tk_wait.rpc_waitq);
1121
1122 printk("%05d %04d %04x %06d %8p %6d %8p %08ld %8s %8p %8p\n",
1123 t->tk_pid,
1124 (t->tk_msg.rpc_proc ? t->tk_msg.rpc_proc->p_proc : -1),
1125 t->tk_flags, t->tk_status,
1126 t->tk_client,
1127 (t->tk_client ? t->tk_client->cl_prog : 0),
1128 t->tk_rqstp, t->tk_timeout,
1129 rpc_waitq,
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001130 t->tk_action, t->tk_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
1132 spin_unlock(&rpc_sched_lock);
1133}
1134#endif
1135
1136void
1137rpc_destroy_mempool(void)
1138{
1139 if (rpc_buffer_mempool)
1140 mempool_destroy(rpc_buffer_mempool);
1141 if (rpc_task_mempool)
1142 mempool_destroy(rpc_task_mempool);
1143 if (rpc_task_slabp && kmem_cache_destroy(rpc_task_slabp))
1144 printk(KERN_INFO "rpc_task: not all structures were freed\n");
1145 if (rpc_buffer_slabp && kmem_cache_destroy(rpc_buffer_slabp))
1146 printk(KERN_INFO "rpc_buffers: not all structures were freed\n");
1147}
1148
1149int
1150rpc_init_mempool(void)
1151{
1152 rpc_task_slabp = kmem_cache_create("rpc_tasks",
1153 sizeof(struct rpc_task),
1154 0, SLAB_HWCACHE_ALIGN,
1155 NULL, NULL);
1156 if (!rpc_task_slabp)
1157 goto err_nomem;
1158 rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1159 RPC_BUFFER_MAXSIZE,
1160 0, SLAB_HWCACHE_ALIGN,
1161 NULL, NULL);
1162 if (!rpc_buffer_slabp)
1163 goto err_nomem;
1164 rpc_task_mempool = mempool_create(RPC_TASK_POOLSIZE,
1165 mempool_alloc_slab,
1166 mempool_free_slab,
1167 rpc_task_slabp);
1168 if (!rpc_task_mempool)
1169 goto err_nomem;
1170 rpc_buffer_mempool = mempool_create(RPC_BUFFER_POOLSIZE,
1171 mempool_alloc_slab,
1172 mempool_free_slab,
1173 rpc_buffer_slabp);
1174 if (!rpc_buffer_mempool)
1175 goto err_nomem;
1176 return 0;
1177err_nomem:
1178 rpc_destroy_mempool();
1179 return -ENOMEM;
1180}