b4133bd13915317a96d94de9dcf2e5d48551870e
[~shefty/rdma-dev.git] / net / sunrpc / sched.c
1 /*
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/spinlock.h>
20 #include <linux/mutex.h>
21 #include <linux/freezer.h>
22
23 #include <linux/sunrpc/clnt.h>
24
25 #include "sunrpc.h"
26
27 #ifdef RPC_DEBUG
28 #define RPCDBG_FACILITY         RPCDBG_SCHED
29 #endif
30
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/sunrpc.h>
33
34 /*
35  * RPC slabs and memory pools
36  */
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;
44
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);
48
49 /*
50  * RPC tasks sit here while waiting for conditions to improve.
51  */
52 static struct rpc_wait_queue delay_queue;
53
54 /*
55  * rpciod-related stuff
56  */
57 struct workqueue_struct *rpciod_workqueue;
58
59 /*
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
62  * rpc_run_timer().
63  */
64 static void
65 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
66 {
67         if (task->tk_timeout == 0)
68                 return;
69         dprintk("RPC: %5u disabling timer\n", task->tk_pid);
70         task->tk_timeout = 0;
71         list_del(&task->u.tk_wait.timer_list);
72         if (list_empty(&queue->timer_list.list))
73                 del_timer(&queue->timer_list.timer);
74 }
75
76 static void
77 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
78 {
79         queue->timer_list.expires = expires;
80         mod_timer(&queue->timer_list.timer, expires);
81 }
82
83 /*
84  * Set up a timer for the current task.
85  */
86 static void
87 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
88 {
89         if (!task->tk_timeout)
90                 return;
91
92         dprintk("RPC: %5u setting alarm for %lu ms\n",
93                         task->tk_pid, task->tk_timeout * 1000 / HZ);
94
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);
99 }
100
101 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
102 {
103         queue->priority = priority;
104 }
105
106 static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
107 {
108         queue->owner = pid;
109         queue->nr = RPC_BATCH_COUNT;
110 }
111
112 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
113 {
114         rpc_set_waitqueue_priority(queue, queue->maxpriority);
115         rpc_set_waitqueue_owner(queue, 0);
116 }
117
118 /*
119  * Add new request to a priority queue.
120  */
121 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
122                 struct rpc_task *task,
123                 unsigned char queue_priority)
124 {
125         struct list_head *q;
126         struct rpc_task *t;
127
128         INIT_LIST_HEAD(&task->u.tk_wait.links);
129         if (unlikely(queue_priority > queue->maxpriority))
130                 queue_priority = queue->maxpriority;
131         if (queue_priority > queue->priority)
132                 rpc_set_waitqueue_priority(queue, queue_priority);
133         q = &queue->tasks[queue_priority];
134         list_for_each_entry(t, q, u.tk_wait.list) {
135                 if (t->tk_owner == task->tk_owner) {
136                         list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
137                         return;
138                 }
139         }
140         list_add_tail(&task->u.tk_wait.list, q);
141 }
142
143 /*
144  * Add new request to wait queue.
145  *
146  * Swapper tasks always get inserted at the head of the queue.
147  * This should avoid many nasty memory deadlocks and hopefully
148  * improve overall performance.
149  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
150  */
151 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
152                 struct rpc_task *task,
153                 unsigned char queue_priority)
154 {
155         WARN_ON_ONCE(RPC_IS_QUEUED(task));
156         if (RPC_IS_QUEUED(task))
157                 return;
158
159         if (RPC_IS_PRIORITY(queue))
160                 __rpc_add_wait_queue_priority(queue, task, queue_priority);
161         else if (RPC_IS_SWAPPER(task))
162                 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
163         else
164                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
165         task->tk_waitqueue = queue;
166         queue->qlen++;
167         rpc_set_queued(task);
168
169         dprintk("RPC: %5u added to queue %p \"%s\"\n",
170                         task->tk_pid, queue, rpc_qname(queue));
171 }
172
173 /*
174  * Remove request from a priority queue.
175  */
176 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
177 {
178         struct rpc_task *t;
179
180         if (!list_empty(&task->u.tk_wait.links)) {
181                 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
182                 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
183                 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
184         }
185 }
186
187 /*
188  * Remove request from queue.
189  * Note: must be called with spin lock held.
190  */
191 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
192 {
193         __rpc_disable_timer(queue, task);
194         if (RPC_IS_PRIORITY(queue))
195                 __rpc_remove_wait_queue_priority(task);
196         list_del(&task->u.tk_wait.list);
197         queue->qlen--;
198         dprintk("RPC: %5u removed from queue %p \"%s\"\n",
199                         task->tk_pid, queue, rpc_qname(queue));
200 }
201
202 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
203 {
204         int i;
205
206         spin_lock_init(&queue->lock);
207         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
208                 INIT_LIST_HEAD(&queue->tasks[i]);
209         queue->maxpriority = nr_queues - 1;
210         rpc_reset_waitqueue_priority(queue);
211         queue->qlen = 0;
212         setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
213         INIT_LIST_HEAD(&queue->timer_list.list);
214         rpc_assign_waitqueue_name(queue, qname);
215 }
216
217 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
218 {
219         __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
220 }
221 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
222
223 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
224 {
225         __rpc_init_priority_wait_queue(queue, qname, 1);
226 }
227 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
228
229 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
230 {
231         del_timer_sync(&queue->timer_list.timer);
232 }
233 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
234
235 static int rpc_wait_bit_killable(void *word)
236 {
237         if (fatal_signal_pending(current))
238                 return -ERESTARTSYS;
239         freezable_schedule();
240         return 0;
241 }
242
243 #ifdef RPC_DEBUG
244 static void rpc_task_set_debuginfo(struct rpc_task *task)
245 {
246         static atomic_t rpc_pid;
247
248         task->tk_pid = atomic_inc_return(&rpc_pid);
249 }
250 #else
251 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
252 {
253 }
254 #endif
255
256 static void rpc_set_active(struct rpc_task *task)
257 {
258         trace_rpc_task_begin(task->tk_client, task, NULL);
259
260         rpc_task_set_debuginfo(task);
261         set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
262 }
263
264 /*
265  * Mark an RPC call as having completed by clearing the 'active' bit
266  * and then waking up all tasks that were sleeping.
267  */
268 static int rpc_complete_task(struct rpc_task *task)
269 {
270         void *m = &task->tk_runstate;
271         wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
272         struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
273         unsigned long flags;
274         int ret;
275
276         trace_rpc_task_complete(task->tk_client, task, NULL);
277
278         spin_lock_irqsave(&wq->lock, flags);
279         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
280         ret = atomic_dec_and_test(&task->tk_count);
281         if (waitqueue_active(wq))
282                 __wake_up_locked_key(wq, TASK_NORMAL, &k);
283         spin_unlock_irqrestore(&wq->lock, flags);
284         return ret;
285 }
286
287 /*
288  * Allow callers to wait for completion of an RPC call
289  *
290  * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
291  * to enforce taking of the wq->lock and hence avoid races with
292  * rpc_complete_task().
293  */
294 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
295 {
296         if (action == NULL)
297                 action = rpc_wait_bit_killable;
298         return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
299                         action, TASK_KILLABLE);
300 }
301 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
302
303 /*
304  * Make an RPC task runnable.
305  *
306  * Note: If the task is ASYNC, and is being made runnable after sitting on an
307  * rpc_wait_queue, this must be called with the queue spinlock held to protect
308  * the wait queue operation.
309  */
310 static void rpc_make_runnable(struct rpc_task *task)
311 {
312         rpc_clear_queued(task);
313         if (rpc_test_and_set_running(task))
314                 return;
315         if (RPC_IS_ASYNC(task)) {
316                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
317                 queue_work(rpciod_workqueue, &task->u.tk_work);
318         } else
319                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
320 }
321
322 /*
323  * Prepare for sleeping on a wait queue.
324  * By always appending tasks to the list we ensure FIFO behavior.
325  * NB: An RPC task will only receive interrupt-driven events as long
326  * as it's on a wait queue.
327  */
328 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
329                 struct rpc_task *task,
330                 rpc_action action,
331                 unsigned char queue_priority)
332 {
333         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
334                         task->tk_pid, rpc_qname(q), jiffies);
335
336         trace_rpc_task_sleep(task->tk_client, task, q);
337
338         __rpc_add_wait_queue(q, task, queue_priority);
339
340         WARN_ON_ONCE(task->tk_callback != NULL);
341         task->tk_callback = action;
342         __rpc_add_timer(q, task);
343 }
344
345 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
346                                 rpc_action action)
347 {
348         /* We shouldn't ever put an inactive task to sleep */
349         WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
350         if (!RPC_IS_ACTIVATED(task)) {
351                 task->tk_status = -EIO;
352                 rpc_put_task_async(task);
353                 return;
354         }
355
356         /*
357          * Protect the queue operations.
358          */
359         spin_lock_bh(&q->lock);
360         __rpc_sleep_on_priority(q, task, action, task->tk_priority);
361         spin_unlock_bh(&q->lock);
362 }
363 EXPORT_SYMBOL_GPL(rpc_sleep_on);
364
365 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
366                 rpc_action action, int priority)
367 {
368         /* We shouldn't ever put an inactive task to sleep */
369         WARN_ON_ONCE(!RPC_IS_ACTIVATED(task));
370         if (!RPC_IS_ACTIVATED(task)) {
371                 task->tk_status = -EIO;
372                 rpc_put_task_async(task);
373                 return;
374         }
375
376         /*
377          * Protect the queue operations.
378          */
379         spin_lock_bh(&q->lock);
380         __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW);
381         spin_unlock_bh(&q->lock);
382 }
383 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
384
385 /**
386  * __rpc_do_wake_up_task - wake up a single rpc_task
387  * @queue: wait queue
388  * @task: task to be woken up
389  *
390  * Caller must hold queue->lock, and have cleared the task queued flag.
391  */
392 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
393 {
394         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
395                         task->tk_pid, jiffies);
396
397         /* Has the task been executed yet? If not, we cannot wake it up! */
398         if (!RPC_IS_ACTIVATED(task)) {
399                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
400                 return;
401         }
402
403         trace_rpc_task_wakeup(task->tk_client, task, queue);
404
405         __rpc_remove_wait_queue(queue, task);
406
407         rpc_make_runnable(task);
408
409         dprintk("RPC:       __rpc_wake_up_task done\n");
410 }
411
412 /*
413  * Wake up a queued task while the queue lock is being held
414  */
415 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
416 {
417         if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
418                 __rpc_do_wake_up_task(queue, task);
419 }
420
421 /*
422  * Tests whether rpc queue is empty
423  */
424 int rpc_queue_empty(struct rpc_wait_queue *queue)
425 {
426         int res;
427
428         spin_lock_bh(&queue->lock);
429         res = queue->qlen;
430         spin_unlock_bh(&queue->lock);
431         return res == 0;
432 }
433 EXPORT_SYMBOL_GPL(rpc_queue_empty);
434
435 /*
436  * Wake up a task on a specific queue
437  */
438 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
439 {
440         spin_lock_bh(&queue->lock);
441         rpc_wake_up_task_queue_locked(queue, task);
442         spin_unlock_bh(&queue->lock);
443 }
444 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
445
446 /*
447  * Wake up the next task on a priority queue.
448  */
449 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
450 {
451         struct list_head *q;
452         struct rpc_task *task;
453
454         /*
455          * Service a batch of tasks from a single owner.
456          */
457         q = &queue->tasks[queue->priority];
458         if (!list_empty(q)) {
459                 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
460                 if (queue->owner == task->tk_owner) {
461                         if (--queue->nr)
462                                 goto out;
463                         list_move_tail(&task->u.tk_wait.list, q);
464                 }
465                 /*
466                  * Check if we need to switch queues.
467                  */
468                 goto new_owner;
469         }
470
471         /*
472          * Service the next queue.
473          */
474         do {
475                 if (q == &queue->tasks[0])
476                         q = &queue->tasks[queue->maxpriority];
477                 else
478                         q = q - 1;
479                 if (!list_empty(q)) {
480                         task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
481                         goto new_queue;
482                 }
483         } while (q != &queue->tasks[queue->priority]);
484
485         rpc_reset_waitqueue_priority(queue);
486         return NULL;
487
488 new_queue:
489         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
490 new_owner:
491         rpc_set_waitqueue_owner(queue, task->tk_owner);
492 out:
493         return task;
494 }
495
496 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
497 {
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);
502         return NULL;
503 }
504
505 /*
506  * Wake up the first task on the wait queue.
507  */
508 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
509                 bool (*func)(struct rpc_task *, void *), void *data)
510 {
511         struct rpc_task *task = NULL;
512
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);
517         if (task != NULL) {
518                 if (func(task, data))
519                         rpc_wake_up_task_queue_locked(queue, task);
520                 else
521                         task = NULL;
522         }
523         spin_unlock_bh(&queue->lock);
524
525         return task;
526 }
527 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
528
529 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
530 {
531         return true;
532 }
533
534 /*
535  * Wake up the next task on the wait queue.
536 */
537 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
538 {
539         return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
540 }
541 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
542
543 /**
544  * rpc_wake_up - wake up all rpc_tasks
545  * @queue: rpc_wait_queue on which the tasks are sleeping
546  *
547  * Grabs queue->lock
548  */
549 void rpc_wake_up(struct rpc_wait_queue *queue)
550 {
551         struct list_head *head;
552
553         spin_lock_bh(&queue->lock);
554         head = &queue->tasks[queue->maxpriority];
555         for (;;) {
556                 while (!list_empty(head)) {
557                         struct rpc_task *task;
558                         task = list_first_entry(head,
559                                         struct rpc_task,
560                                         u.tk_wait.list);
561                         rpc_wake_up_task_queue_locked(queue, task);
562                 }
563                 if (head == &queue->tasks[0])
564                         break;
565                 head--;
566         }
567         spin_unlock_bh(&queue->lock);
568 }
569 EXPORT_SYMBOL_GPL(rpc_wake_up);
570
571 /**
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
575  *
576  * Grabs queue->lock
577  */
578 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
579 {
580         struct list_head *head;
581
582         spin_lock_bh(&queue->lock);
583         head = &queue->tasks[queue->maxpriority];
584         for (;;) {
585                 while (!list_empty(head)) {
586                         struct rpc_task *task;
587                         task = list_first_entry(head,
588                                         struct rpc_task,
589                                         u.tk_wait.list);
590                         task->tk_status = status;
591                         rpc_wake_up_task_queue_locked(queue, task);
592                 }
593                 if (head == &queue->tasks[0])
594                         break;
595                 head--;
596         }
597         spin_unlock_bh(&queue->lock);
598 }
599 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
600
601 static void __rpc_queue_timer_fn(unsigned long ptr)
602 {
603         struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
604         struct rpc_task *task, *n;
605         unsigned long expires, now, timeo;
606
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);
615                         continue;
616                 }
617                 if (expires == now || time_after(expires, timeo))
618                         expires = timeo;
619         }
620         if (!list_empty(&queue->timer_list.list))
621                 rpc_set_queue_timer(queue, expires);
622         spin_unlock(&queue->lock);
623 }
624
625 static void __rpc_atrun(struct rpc_task *task)
626 {
627         task->tk_status = 0;
628 }
629
630 /*
631  * Run a task at a later time
632  */
633 void rpc_delay(struct rpc_task *task, unsigned long delay)
634 {
635         task->tk_timeout = delay;
636         rpc_sleep_on(&delay_queue, task, __rpc_atrun);
637 }
638 EXPORT_SYMBOL_GPL(rpc_delay);
639
640 /*
641  * Helper to call task->tk_ops->rpc_call_prepare
642  */
643 void rpc_prepare_task(struct rpc_task *task)
644 {
645         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
646 }
647
648 static void
649 rpc_init_task_statistics(struct rpc_task *task)
650 {
651         /* Initialize retry counters */
652         task->tk_garb_retry = 2;
653         task->tk_cred_retry = 2;
654         task->tk_rebind_retry = 2;
655
656         /* starting timestamp */
657         task->tk_start = ktime_get();
658 }
659
660 static void
661 rpc_reset_task_statistics(struct rpc_task *task)
662 {
663         task->tk_timeouts = 0;
664         task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT);
665
666         rpc_init_task_statistics(task);
667 }
668
669 /*
670  * Helper that calls task->tk_ops->rpc_call_done if it exists
671  */
672 void rpc_exit_task(struct rpc_task *task)
673 {
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 */
680                         xprt_release(task);
681                         rpc_reset_task_statistics(task);
682                 }
683         }
684 }
685
686 void rpc_exit(struct rpc_task *task, int status)
687 {
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);
692 }
693 EXPORT_SYMBOL_GPL(rpc_exit);
694
695 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
696 {
697         if (ops->rpc_release != NULL)
698                 ops->rpc_release(calldata);
699 }
700
701 /*
702  * This is the RPC `scheduler' (or rather, the finite state machine).
703  */
704 static void __rpc_execute(struct rpc_task *task)
705 {
706         struct rpc_wait_queue *queue;
707         int task_is_async = RPC_IS_ASYNC(task);
708         int status = 0;
709
710         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
711                         task->tk_pid, task->tk_flags);
712
713         WARN_ON_ONCE(RPC_IS_QUEUED(task));
714         if (RPC_IS_QUEUED(task))
715                 return;
716
717         for (;;) {
718                 void (*do_action)(struct rpc_task *);
719
720                 /*
721                  * Execute any pending callback first.
722                  */
723                 do_action = task->tk_callback;
724                 task->tk_callback = NULL;
725                 if (do_action == NULL) {
726                         /*
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.
731                          */
732                         do_action = task->tk_action;
733                         if (do_action == NULL)
734                                 break;
735                 }
736                 trace_rpc_task_run_action(task->tk_client, task, task->tk_action);
737                 do_action(task);
738
739                 /*
740                  * Lockless check for whether task is sleeping or not.
741                  */
742                 if (!RPC_IS_QUEUED(task))
743                         continue;
744                 /*
745                  * The queue->lock protects against races with
746                  * rpc_make_runnable().
747                  *
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.
752                  */
753                 queue = task->tk_waitqueue;
754                 spin_lock_bh(&queue->lock);
755                 if (!RPC_IS_QUEUED(task)) {
756                         spin_unlock_bh(&queue->lock);
757                         continue;
758                 }
759                 rpc_clear_running(task);
760                 spin_unlock_bh(&queue->lock);
761                 if (task_is_async)
762                         return;
763
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,
768                                 TASK_KILLABLE);
769                 if (status == -ERESTARTSYS) {
770                         /*
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.
775                          */
776                         dprintk("RPC: %5u got signal\n", task->tk_pid);
777                         task->tk_flags |= RPC_TASK_KILLED;
778                         rpc_exit(task, -ERESTARTSYS);
779                 }
780                 rpc_set_running(task);
781                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
782         }
783
784         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
785                         task->tk_status);
786         /* Release all resources associated with the task */
787         rpc_release_task(task);
788 }
789
790 /*
791  * User-visible entry point to the scheduler.
792  *
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.
798  */
799 void rpc_execute(struct rpc_task *task)
800 {
801         rpc_set_active(task);
802         rpc_make_runnable(task);
803         if (!RPC_IS_ASYNC(task))
804                 __rpc_execute(task);
805 }
806
807 static void rpc_async_schedule(struct work_struct *work)
808 {
809         current->flags |= PF_FSTRANS;
810         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
811         current->flags &= ~PF_FSTRANS;
812 }
813
814 /**
815  * rpc_malloc - allocate an RPC buffer
816  * @task: RPC task that will use this buffer
817  * @size: requested byte size
818  *
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.
822  *
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.
826  *
827  * In order to avoid memory starvation triggering more writebacks of
828  * NFS requests, we avoid using GFP_KERNEL.
829  */
830 void *rpc_malloc(struct rpc_task *task, size_t size)
831 {
832         struct rpc_buffer *buf;
833         gfp_t gfp = GFP_NOWAIT;
834
835         if (RPC_IS_SWAPPER(task))
836                 gfp |= __GFP_MEMALLOC;
837
838         size += sizeof(struct rpc_buffer);
839         if (size <= RPC_BUFFER_MAXSIZE)
840                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
841         else
842                 buf = kmalloc(size, gfp);
843
844         if (!buf)
845                 return NULL;
846
847         buf->len = size;
848         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
849                         task->tk_pid, size, buf);
850         return &buf->data;
851 }
852 EXPORT_SYMBOL_GPL(rpc_malloc);
853
854 /**
855  * rpc_free - free buffer allocated via rpc_malloc
856  * @buffer: buffer to free
857  *
858  */
859 void rpc_free(void *buffer)
860 {
861         size_t size;
862         struct rpc_buffer *buf;
863
864         if (!buffer)
865                 return;
866
867         buf = container_of(buffer, struct rpc_buffer, data);
868         size = buf->len;
869
870         dprintk("RPC:       freeing buffer of size %zu at %p\n",
871                         size, buf);
872
873         if (size <= RPC_BUFFER_MAXSIZE)
874                 mempool_free(buf, rpc_buffer_mempool);
875         else
876                 kfree(buf);
877 }
878 EXPORT_SYMBOL_GPL(rpc_free);
879
880 /*
881  * Creation and deletion of RPC task structures
882  */
883 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
884 {
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);
891
892         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
893         task->tk_owner = current->tgid;
894
895         /* Initialize workqueue for async tasks */
896         task->tk_workqueue = task_setup_data->workqueue;
897
898         if (task->tk_ops->rpc_call_prepare != NULL)
899                 task->tk_action = rpc_prepare_task;
900
901         rpc_init_task_statistics(task);
902
903         dprintk("RPC:       new task initialized, procpid %u\n",
904                                 task_pid_nr(current));
905 }
906
907 static struct rpc_task *
908 rpc_alloc_task(void)
909 {
910         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO);
911 }
912
913 /*
914  * Create a new task for the specified client.
915  */
916 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
917 {
918         struct rpc_task *task = setup_data->task;
919         unsigned short flags = 0;
920
921         if (task == NULL) {
922                 task = rpc_alloc_task();
923                 if (task == NULL) {
924                         rpc_release_calldata(setup_data->callback_ops,
925                                         setup_data->callback_data);
926                         return ERR_PTR(-ENOMEM);
927                 }
928                 flags = RPC_TASK_DYNAMIC;
929         }
930
931         rpc_init_task(task, setup_data);
932         task->tk_flags |= flags;
933         dprintk("RPC:       allocated task %p\n", task);
934         return task;
935 }
936
937 /*
938  * rpc_free_task - release rpc task and perform cleanups
939  *
940  * Note that we free up the rpc_task _after_ rpc_release_calldata()
941  * in order to work around a workqueue dependency issue.
942  *
943  * Tejun Heo states:
944  * "Workqueue currently considers two work items to be the same if they're
945  * on the same address and won't execute them concurrently - ie. it
946  * makes a work item which is queued again while being executed wait
947  * for the previous execution to complete.
948  *
949  * If a work function frees the work item, and then waits for an event
950  * which should be performed by another work item and *that* work item
951  * recycles the freed work item, it can create a false dependency loop.
952  * There really is no reliable way to detect this short of verifying
953  * every memory free."
954  *
955  */
956 static void rpc_free_task(struct rpc_task *task)
957 {
958         unsigned short tk_flags = task->tk_flags;
959
960         rpc_release_calldata(task->tk_ops, task->tk_calldata);
961
962         if (tk_flags & RPC_TASK_DYNAMIC) {
963                 dprintk("RPC: %5u freeing task\n", task->tk_pid);
964                 mempool_free(task, rpc_task_mempool);
965         }
966 }
967
968 static void rpc_async_release(struct work_struct *work)
969 {
970         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
971 }
972
973 static void rpc_release_resources_task(struct rpc_task *task)
974 {
975         if (task->tk_rqstp)
976                 xprt_release(task);
977         if (task->tk_msg.rpc_cred) {
978                 put_rpccred(task->tk_msg.rpc_cred);
979                 task->tk_msg.rpc_cred = NULL;
980         }
981         rpc_task_release_client(task);
982 }
983
984 static void rpc_final_put_task(struct rpc_task *task,
985                 struct workqueue_struct *q)
986 {
987         if (q != NULL) {
988                 INIT_WORK(&task->u.tk_work, rpc_async_release);
989                 queue_work(q, &task->u.tk_work);
990         } else
991                 rpc_free_task(task);
992 }
993
994 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
995 {
996         if (atomic_dec_and_test(&task->tk_count)) {
997                 rpc_release_resources_task(task);
998                 rpc_final_put_task(task, q);
999         }
1000 }
1001
1002 void rpc_put_task(struct rpc_task *task)
1003 {
1004         rpc_do_put_task(task, NULL);
1005 }
1006 EXPORT_SYMBOL_GPL(rpc_put_task);
1007
1008 void rpc_put_task_async(struct rpc_task *task)
1009 {
1010         rpc_do_put_task(task, task->tk_workqueue);
1011 }
1012 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1013
1014 static void rpc_release_task(struct rpc_task *task)
1015 {
1016         dprintk("RPC: %5u release task\n", task->tk_pid);
1017
1018         WARN_ON_ONCE(RPC_IS_QUEUED(task));
1019
1020         rpc_release_resources_task(task);
1021
1022         /*
1023          * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1024          * so it should be safe to use task->tk_count as a test for whether
1025          * or not any other processes still hold references to our rpc_task.
1026          */
1027         if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1028                 /* Wake up anyone who may be waiting for task completion */
1029                 if (!rpc_complete_task(task))
1030                         return;
1031         } else {
1032                 if (!atomic_dec_and_test(&task->tk_count))
1033                         return;
1034         }
1035         rpc_final_put_task(task, task->tk_workqueue);
1036 }
1037
1038 int rpciod_up(void)
1039 {
1040         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1041 }
1042
1043 void rpciod_down(void)
1044 {
1045         module_put(THIS_MODULE);
1046 }
1047
1048 /*
1049  * Start up the rpciod workqueue.
1050  */
1051 static int rpciod_start(void)
1052 {
1053         struct workqueue_struct *wq;
1054
1055         /*
1056          * Create the rpciod thread and wait for it to start.
1057          */
1058         dprintk("RPC:       creating workqueue rpciod\n");
1059         wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 1);
1060         rpciod_workqueue = wq;
1061         return rpciod_workqueue != NULL;
1062 }
1063
1064 static void rpciod_stop(void)
1065 {
1066         struct workqueue_struct *wq = NULL;
1067
1068         if (rpciod_workqueue == NULL)
1069                 return;
1070         dprintk("RPC:       destroying workqueue rpciod\n");
1071
1072         wq = rpciod_workqueue;
1073         rpciod_workqueue = NULL;
1074         destroy_workqueue(wq);
1075 }
1076
1077 void
1078 rpc_destroy_mempool(void)
1079 {
1080         rpciod_stop();
1081         if (rpc_buffer_mempool)
1082                 mempool_destroy(rpc_buffer_mempool);
1083         if (rpc_task_mempool)
1084                 mempool_destroy(rpc_task_mempool);
1085         if (rpc_task_slabp)
1086                 kmem_cache_destroy(rpc_task_slabp);
1087         if (rpc_buffer_slabp)
1088                 kmem_cache_destroy(rpc_buffer_slabp);
1089         rpc_destroy_wait_queue(&delay_queue);
1090 }
1091
1092 int
1093 rpc_init_mempool(void)
1094 {
1095         /*
1096          * The following is not strictly a mempool initialisation,
1097          * but there is no harm in doing it here
1098          */
1099         rpc_init_wait_queue(&delay_queue, "delayq");
1100         if (!rpciod_start())
1101                 goto err_nomem;
1102
1103         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1104                                              sizeof(struct rpc_task),
1105                                              0, SLAB_HWCACHE_ALIGN,
1106                                              NULL);
1107         if (!rpc_task_slabp)
1108                 goto err_nomem;
1109         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1110                                              RPC_BUFFER_MAXSIZE,
1111                                              0, SLAB_HWCACHE_ALIGN,
1112                                              NULL);
1113         if (!rpc_buffer_slabp)
1114                 goto err_nomem;
1115         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1116                                                     rpc_task_slabp);
1117         if (!rpc_task_mempool)
1118                 goto err_nomem;
1119         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1120                                                       rpc_buffer_slabp);
1121         if (!rpc_buffer_mempool)
1122                 goto err_nomem;
1123         return 0;
1124 err_nomem:
1125         rpc_destroy_mempool();
1126         return -ENOMEM;
1127 }