]> asedeno.scripts.mit.edu Git - linux.git/blob - net/sunrpc/sched.c
xfrm: ressurrect "Fix uninitialized memory read in _decode_session4"
[linux.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 #include <linux/sched/mm.h>
23
24 #include <linux/sunrpc/clnt.h>
25
26 #include "sunrpc.h"
27
28 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
29 #define RPCDBG_FACILITY         RPCDBG_SCHED
30 #endif
31
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/sunrpc.h>
34
35 /*
36  * RPC slabs and memory pools
37  */
38 #define RPC_BUFFER_MAXSIZE      (2048)
39 #define RPC_BUFFER_POOLSIZE     (8)
40 #define RPC_TASK_POOLSIZE       (8)
41 static struct kmem_cache        *rpc_task_slabp __read_mostly;
42 static struct kmem_cache        *rpc_buffer_slabp __read_mostly;
43 static mempool_t        *rpc_task_mempool __read_mostly;
44 static mempool_t        *rpc_buffer_mempool __read_mostly;
45
46 static void                     rpc_async_schedule(struct work_struct *);
47 static void                      rpc_release_task(struct rpc_task *task);
48 static void __rpc_queue_timer_fn(struct timer_list *t);
49
50 /*
51  * RPC tasks sit here while waiting for conditions to improve.
52  */
53 static struct rpc_wait_queue delay_queue;
54
55 /*
56  * rpciod-related stuff
57  */
58 struct workqueue_struct *rpciod_workqueue __read_mostly;
59 struct workqueue_struct *xprtiod_workqueue __read_mostly;
60
61 unsigned long
62 rpc_task_timeout(const struct rpc_task *task)
63 {
64         unsigned long timeout = READ_ONCE(task->tk_timeout);
65
66         if (timeout != 0) {
67                 unsigned long now = jiffies;
68                 if (time_before(now, timeout))
69                         return timeout - now;
70         }
71         return 0;
72 }
73 EXPORT_SYMBOL_GPL(rpc_task_timeout);
74
75 /*
76  * Disable the timer for a given RPC task. Should be called with
77  * queue->lock and bh_disabled in order to avoid races within
78  * rpc_run_timer().
79  */
80 static void
81 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
82 {
83         if (list_empty(&task->u.tk_wait.timer_list))
84                 return;
85         dprintk("RPC: %5u disabling timer\n", task->tk_pid);
86         task->tk_timeout = 0;
87         list_del(&task->u.tk_wait.timer_list);
88         if (list_empty(&queue->timer_list.list))
89                 del_timer(&queue->timer_list.timer);
90 }
91
92 static void
93 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
94 {
95         timer_reduce(&queue->timer_list.timer, expires);
96 }
97
98 /*
99  * Set up a timer for the current task.
100  */
101 static void
102 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
103                 unsigned long timeout)
104 {
105         dprintk("RPC: %5u setting alarm for %u ms\n",
106                 task->tk_pid, jiffies_to_msecs(timeout - jiffies));
107
108         task->tk_timeout = timeout;
109         rpc_set_queue_timer(queue, timeout);
110         list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
111 }
112
113 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
114 {
115         if (queue->priority != priority) {
116                 queue->priority = priority;
117                 queue->nr = 1U << priority;
118         }
119 }
120
121 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
122 {
123         rpc_set_waitqueue_priority(queue, queue->maxpriority);
124 }
125
126 /*
127  * Add a request to a queue list
128  */
129 static void
130 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
131 {
132         struct rpc_task *t;
133
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.links,
137                                         &t->u.tk_wait.links);
138                         /* Cache the queue head in task->u.tk_wait.list */
139                         task->u.tk_wait.list.next = q;
140                         task->u.tk_wait.list.prev = NULL;
141                         return;
142                 }
143         }
144         INIT_LIST_HEAD(&task->u.tk_wait.links);
145         list_add_tail(&task->u.tk_wait.list, q);
146 }
147
148 /*
149  * Remove request from a queue list
150  */
151 static void
152 __rpc_list_dequeue_task(struct rpc_task *task)
153 {
154         struct list_head *q;
155         struct rpc_task *t;
156
157         if (task->u.tk_wait.list.prev == NULL) {
158                 list_del(&task->u.tk_wait.links);
159                 return;
160         }
161         if (!list_empty(&task->u.tk_wait.links)) {
162                 t = list_first_entry(&task->u.tk_wait.links,
163                                 struct rpc_task,
164                                 u.tk_wait.links);
165                 /* Assume __rpc_list_enqueue_task() cached the queue head */
166                 q = t->u.tk_wait.list.next;
167                 list_add_tail(&t->u.tk_wait.list, q);
168                 list_del(&task->u.tk_wait.links);
169         }
170         list_del(&task->u.tk_wait.list);
171 }
172
173 /*
174  * Add new request to a priority queue.
175  */
176 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
177                 struct rpc_task *task,
178                 unsigned char queue_priority)
179 {
180         if (unlikely(queue_priority > queue->maxpriority))
181                 queue_priority = queue->maxpriority;
182         __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
183 }
184
185 /*
186  * Add new request to wait queue.
187  *
188  * Swapper tasks always get inserted at the head of the queue.
189  * This should avoid many nasty memory deadlocks and hopefully
190  * improve overall performance.
191  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
192  */
193 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
194                 struct rpc_task *task,
195                 unsigned char queue_priority)
196 {
197         WARN_ON_ONCE(RPC_IS_QUEUED(task));
198         if (RPC_IS_QUEUED(task))
199                 return;
200
201         INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
202         if (RPC_IS_PRIORITY(queue))
203                 __rpc_add_wait_queue_priority(queue, task, queue_priority);
204         else if (RPC_IS_SWAPPER(task))
205                 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
206         else
207                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
208         task->tk_waitqueue = queue;
209         queue->qlen++;
210         /* barrier matches the read in rpc_wake_up_task_queue_locked() */
211         smp_wmb();
212         rpc_set_queued(task);
213
214         dprintk("RPC: %5u added to queue %p \"%s\"\n",
215                         task->tk_pid, queue, rpc_qname(queue));
216 }
217
218 /*
219  * Remove request from a priority queue.
220  */
221 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
222 {
223         __rpc_list_dequeue_task(task);
224 }
225
226 /*
227  * Remove request from queue.
228  * Note: must be called with spin lock held.
229  */
230 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
231 {
232         __rpc_disable_timer(queue, task);
233         if (RPC_IS_PRIORITY(queue))
234                 __rpc_remove_wait_queue_priority(task);
235         else
236                 list_del(&task->u.tk_wait.list);
237         queue->qlen--;
238         dprintk("RPC: %5u removed from queue %p \"%s\"\n",
239                         task->tk_pid, queue, rpc_qname(queue));
240 }
241
242 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
243 {
244         int i;
245
246         spin_lock_init(&queue->lock);
247         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
248                 INIT_LIST_HEAD(&queue->tasks[i]);
249         queue->maxpriority = nr_queues - 1;
250         rpc_reset_waitqueue_priority(queue);
251         queue->qlen = 0;
252         timer_setup(&queue->timer_list.timer,
253                         __rpc_queue_timer_fn,
254                         TIMER_DEFERRABLE);
255         INIT_LIST_HEAD(&queue->timer_list.list);
256         rpc_assign_waitqueue_name(queue, qname);
257 }
258
259 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
260 {
261         __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
262 }
263 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
264
265 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
266 {
267         __rpc_init_priority_wait_queue(queue, qname, 1);
268 }
269 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
270
271 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
272 {
273         del_timer_sync(&queue->timer_list.timer);
274 }
275 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
276
277 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
278 {
279         freezable_schedule_unsafe();
280         if (signal_pending_state(mode, current))
281                 return -ERESTARTSYS;
282         return 0;
283 }
284
285 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
286 static void rpc_task_set_debuginfo(struct rpc_task *task)
287 {
288         static atomic_t rpc_pid;
289
290         task->tk_pid = atomic_inc_return(&rpc_pid);
291 }
292 #else
293 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
294 {
295 }
296 #endif
297
298 static void rpc_set_active(struct rpc_task *task)
299 {
300         rpc_task_set_debuginfo(task);
301         set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
302         trace_rpc_task_begin(task, NULL);
303 }
304
305 /*
306  * Mark an RPC call as having completed by clearing the 'active' bit
307  * and then waking up all tasks that were sleeping.
308  */
309 static int rpc_complete_task(struct rpc_task *task)
310 {
311         void *m = &task->tk_runstate;
312         wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
313         struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
314         unsigned long flags;
315         int ret;
316
317         trace_rpc_task_complete(task, NULL);
318
319         spin_lock_irqsave(&wq->lock, flags);
320         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
321         ret = atomic_dec_and_test(&task->tk_count);
322         if (waitqueue_active(wq))
323                 __wake_up_locked_key(wq, TASK_NORMAL, &k);
324         spin_unlock_irqrestore(&wq->lock, flags);
325         return ret;
326 }
327
328 /*
329  * Allow callers to wait for completion of an RPC call
330  *
331  * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
332  * to enforce taking of the wq->lock and hence avoid races with
333  * rpc_complete_task().
334  */
335 int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action)
336 {
337         if (action == NULL)
338                 action = rpc_wait_bit_killable;
339         return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
340                         action, TASK_KILLABLE);
341 }
342 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
343
344 /*
345  * Make an RPC task runnable.
346  *
347  * Note: If the task is ASYNC, and is being made runnable after sitting on an
348  * rpc_wait_queue, this must be called with the queue spinlock held to protect
349  * the wait queue operation.
350  * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
351  * which is needed to ensure that __rpc_execute() doesn't loop (due to the
352  * lockless RPC_IS_QUEUED() test) before we've had a chance to test
353  * the RPC_TASK_RUNNING flag.
354  */
355 static void rpc_make_runnable(struct workqueue_struct *wq,
356                 struct rpc_task *task)
357 {
358         bool need_wakeup = !rpc_test_and_set_running(task);
359
360         rpc_clear_queued(task);
361         if (!need_wakeup)
362                 return;
363         if (RPC_IS_ASYNC(task)) {
364                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
365                 queue_work(wq, &task->u.tk_work);
366         } else
367                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
368 }
369
370 /*
371  * Prepare for sleeping on a wait queue.
372  * By always appending tasks to the list we ensure FIFO behavior.
373  * NB: An RPC task will only receive interrupt-driven events as long
374  * as it's on a wait queue.
375  */
376 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
377                 struct rpc_task *task,
378                 unsigned char queue_priority)
379 {
380         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
381                         task->tk_pid, rpc_qname(q), jiffies);
382
383         trace_rpc_task_sleep(task, q);
384
385         __rpc_add_wait_queue(q, task, queue_priority);
386
387 }
388
389 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
390                 struct rpc_task *task, unsigned long timeout,
391                 unsigned char queue_priority)
392 {
393         if (time_is_after_jiffies(timeout)) {
394                 __rpc_sleep_on_priority(q, task, queue_priority);
395                 __rpc_add_timer(q, task, timeout);
396         } else
397                 task->tk_status = -ETIMEDOUT;
398 }
399
400 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
401 {
402         if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
403                 task->tk_callback = action;
404 }
405
406 static bool rpc_sleep_check_activated(struct rpc_task *task)
407 {
408         /* We shouldn't ever put an inactive task to sleep */
409         if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
410                 task->tk_status = -EIO;
411                 rpc_put_task_async(task);
412                 return false;
413         }
414         return true;
415 }
416
417 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
418                                 rpc_action action, unsigned long timeout)
419 {
420         if (!rpc_sleep_check_activated(task))
421                 return;
422
423         rpc_set_tk_callback(task, action);
424
425         /*
426          * Protect the queue operations.
427          */
428         spin_lock_bh(&q->lock);
429         __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
430         spin_unlock_bh(&q->lock);
431 }
432 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
433
434 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
435                                 rpc_action action)
436 {
437         if (!rpc_sleep_check_activated(task))
438                 return;
439
440         rpc_set_tk_callback(task, action);
441
442         WARN_ON_ONCE(task->tk_timeout != 0);
443         /*
444          * Protect the queue operations.
445          */
446         spin_lock_bh(&q->lock);
447         __rpc_sleep_on_priority(q, task, task->tk_priority);
448         spin_unlock_bh(&q->lock);
449 }
450 EXPORT_SYMBOL_GPL(rpc_sleep_on);
451
452 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
453                 struct rpc_task *task, unsigned long timeout, int priority)
454 {
455         if (!rpc_sleep_check_activated(task))
456                 return;
457
458         priority -= RPC_PRIORITY_LOW;
459         /*
460          * Protect the queue operations.
461          */
462         spin_lock_bh(&q->lock);
463         __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
464         spin_unlock_bh(&q->lock);
465 }
466 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
467
468 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
469                 int priority)
470 {
471         if (!rpc_sleep_check_activated(task))
472                 return;
473
474         WARN_ON_ONCE(task->tk_timeout != 0);
475         priority -= RPC_PRIORITY_LOW;
476         /*
477          * Protect the queue operations.
478          */
479         spin_lock_bh(&q->lock);
480         __rpc_sleep_on_priority(q, task, priority);
481         spin_unlock_bh(&q->lock);
482 }
483 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
484
485 /**
486  * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
487  * @wq: workqueue on which to run task
488  * @queue: wait queue
489  * @task: task to be woken up
490  *
491  * Caller must hold queue->lock, and have cleared the task queued flag.
492  */
493 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
494                 struct rpc_wait_queue *queue,
495                 struct rpc_task *task)
496 {
497         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
498                         task->tk_pid, jiffies);
499
500         /* Has the task been executed yet? If not, we cannot wake it up! */
501         if (!RPC_IS_ACTIVATED(task)) {
502                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
503                 return;
504         }
505
506         trace_rpc_task_wakeup(task, queue);
507
508         __rpc_remove_wait_queue(queue, task);
509
510         rpc_make_runnable(wq, task);
511
512         dprintk("RPC:       __rpc_wake_up_task done\n");
513 }
514
515 /*
516  * Wake up a queued task while the queue lock is being held
517  */
518 static struct rpc_task *
519 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
520                 struct rpc_wait_queue *queue, struct rpc_task *task,
521                 bool (*action)(struct rpc_task *, void *), void *data)
522 {
523         if (RPC_IS_QUEUED(task)) {
524                 smp_rmb();
525                 if (task->tk_waitqueue == queue) {
526                         if (action == NULL || action(task, data)) {
527                                 __rpc_do_wake_up_task_on_wq(wq, queue, task);
528                                 return task;
529                         }
530                 }
531         }
532         return NULL;
533 }
534
535 static void
536 rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq,
537                 struct rpc_wait_queue *queue, struct rpc_task *task)
538 {
539         rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL);
540 }
541
542 /*
543  * Wake up a queued task while the queue lock is being held
544  */
545 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
546 {
547         rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task);
548 }
549
550 /*
551  * Wake up a task on a specific queue
552  */
553 void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq,
554                 struct rpc_wait_queue *queue,
555                 struct rpc_task *task)
556 {
557         if (!RPC_IS_QUEUED(task))
558                 return;
559         spin_lock_bh(&queue->lock);
560         rpc_wake_up_task_on_wq_queue_locked(wq, queue, task);
561         spin_unlock_bh(&queue->lock);
562 }
563
564 /*
565  * Wake up a task on a specific queue
566  */
567 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
568 {
569         if (!RPC_IS_QUEUED(task))
570                 return;
571         spin_lock_bh(&queue->lock);
572         rpc_wake_up_task_queue_locked(queue, task);
573         spin_unlock_bh(&queue->lock);
574 }
575 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
576
577 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
578 {
579         task->tk_status = *(int *)status;
580         return true;
581 }
582
583 static void
584 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
585                 struct rpc_task *task, int status)
586 {
587         rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
588                         task, rpc_task_action_set_status, &status);
589 }
590
591 /**
592  * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
593  * @queue: pointer to rpc_wait_queue
594  * @task: pointer to rpc_task
595  * @status: integer error value
596  *
597  * If @task is queued on @queue, then it is woken up, and @task->tk_status is
598  * set to the value of @status.
599  */
600 void
601 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
602                 struct rpc_task *task, int status)
603 {
604         if (!RPC_IS_QUEUED(task))
605                 return;
606         spin_lock_bh(&queue->lock);
607         rpc_wake_up_task_queue_set_status_locked(queue, task, status);
608         spin_unlock_bh(&queue->lock);
609 }
610
611 /*
612  * Wake up the next task on a priority queue.
613  */
614 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
615 {
616         struct list_head *q;
617         struct rpc_task *task;
618
619         /*
620          * Service a batch of tasks from a single owner.
621          */
622         q = &queue->tasks[queue->priority];
623         if (!list_empty(q) && --queue->nr) {
624                 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
625                 goto out;
626         }
627
628         /*
629          * Service the next queue.
630          */
631         do {
632                 if (q == &queue->tasks[0])
633                         q = &queue->tasks[queue->maxpriority];
634                 else
635                         q = q - 1;
636                 if (!list_empty(q)) {
637                         task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
638                         goto new_queue;
639                 }
640         } while (q != &queue->tasks[queue->priority]);
641
642         rpc_reset_waitqueue_priority(queue);
643         return NULL;
644
645 new_queue:
646         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
647 out:
648         return task;
649 }
650
651 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
652 {
653         if (RPC_IS_PRIORITY(queue))
654                 return __rpc_find_next_queued_priority(queue);
655         if (!list_empty(&queue->tasks[0]))
656                 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
657         return NULL;
658 }
659
660 /*
661  * Wake up the first task on the wait queue.
662  */
663 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
664                 struct rpc_wait_queue *queue,
665                 bool (*func)(struct rpc_task *, void *), void *data)
666 {
667         struct rpc_task *task = NULL;
668
669         dprintk("RPC:       wake_up_first(%p \"%s\")\n",
670                         queue, rpc_qname(queue));
671         spin_lock_bh(&queue->lock);
672         task = __rpc_find_next_queued(queue);
673         if (task != NULL)
674                 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
675                                 task, func, data);
676         spin_unlock_bh(&queue->lock);
677
678         return task;
679 }
680
681 /*
682  * Wake up the first task on the wait queue.
683  */
684 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
685                 bool (*func)(struct rpc_task *, void *), void *data)
686 {
687         return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
688 }
689 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
690
691 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
692 {
693         return true;
694 }
695
696 /*
697  * Wake up the next task on the wait queue.
698 */
699 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
700 {
701         return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
702 }
703 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
704
705 /**
706  * rpc_wake_up - wake up all rpc_tasks
707  * @queue: rpc_wait_queue on which the tasks are sleeping
708  *
709  * Grabs queue->lock
710  */
711 void rpc_wake_up(struct rpc_wait_queue *queue)
712 {
713         struct list_head *head;
714
715         spin_lock_bh(&queue->lock);
716         head = &queue->tasks[queue->maxpriority];
717         for (;;) {
718                 while (!list_empty(head)) {
719                         struct rpc_task *task;
720                         task = list_first_entry(head,
721                                         struct rpc_task,
722                                         u.tk_wait.list);
723                         rpc_wake_up_task_queue_locked(queue, task);
724                 }
725                 if (head == &queue->tasks[0])
726                         break;
727                 head--;
728         }
729         spin_unlock_bh(&queue->lock);
730 }
731 EXPORT_SYMBOL_GPL(rpc_wake_up);
732
733 /**
734  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
735  * @queue: rpc_wait_queue on which the tasks are sleeping
736  * @status: status value to set
737  *
738  * Grabs queue->lock
739  */
740 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
741 {
742         struct list_head *head;
743
744         spin_lock_bh(&queue->lock);
745         head = &queue->tasks[queue->maxpriority];
746         for (;;) {
747                 while (!list_empty(head)) {
748                         struct rpc_task *task;
749                         task = list_first_entry(head,
750                                         struct rpc_task,
751                                         u.tk_wait.list);
752                         task->tk_status = status;
753                         rpc_wake_up_task_queue_locked(queue, task);
754                 }
755                 if (head == &queue->tasks[0])
756                         break;
757                 head--;
758         }
759         spin_unlock_bh(&queue->lock);
760 }
761 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
762
763 static void __rpc_queue_timer_fn(struct timer_list *t)
764 {
765         struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer);
766         struct rpc_task *task, *n;
767         unsigned long expires, now, timeo;
768
769         spin_lock(&queue->lock);
770         expires = now = jiffies;
771         list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
772                 timeo = task->tk_timeout;
773                 if (time_after_eq(now, timeo)) {
774                         dprintk("RPC: %5u timeout\n", task->tk_pid);
775                         task->tk_status = -ETIMEDOUT;
776                         rpc_wake_up_task_queue_locked(queue, task);
777                         continue;
778                 }
779                 if (expires == now || time_after(expires, timeo))
780                         expires = timeo;
781         }
782         if (!list_empty(&queue->timer_list.list))
783                 rpc_set_queue_timer(queue, expires);
784         spin_unlock(&queue->lock);
785 }
786
787 static void __rpc_atrun(struct rpc_task *task)
788 {
789         if (task->tk_status == -ETIMEDOUT)
790                 task->tk_status = 0;
791 }
792
793 /*
794  * Run a task at a later time
795  */
796 void rpc_delay(struct rpc_task *task, unsigned long delay)
797 {
798         rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
799 }
800 EXPORT_SYMBOL_GPL(rpc_delay);
801
802 /*
803  * Helper to call task->tk_ops->rpc_call_prepare
804  */
805 void rpc_prepare_task(struct rpc_task *task)
806 {
807         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
808 }
809
810 static void
811 rpc_init_task_statistics(struct rpc_task *task)
812 {
813         /* Initialize retry counters */
814         task->tk_garb_retry = 2;
815         task->tk_cred_retry = 2;
816         task->tk_rebind_retry = 2;
817
818         /* starting timestamp */
819         task->tk_start = ktime_get();
820 }
821
822 static void
823 rpc_reset_task_statistics(struct rpc_task *task)
824 {
825         task->tk_timeouts = 0;
826         task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
827         rpc_init_task_statistics(task);
828 }
829
830 /*
831  * Helper that calls task->tk_ops->rpc_call_done if it exists
832  */
833 void rpc_exit_task(struct rpc_task *task)
834 {
835         task->tk_action = NULL;
836         if (task->tk_ops->rpc_call_done != NULL) {
837                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
838                 if (task->tk_action != NULL) {
839                         /* Always release the RPC slot and buffer memory */
840                         xprt_release(task);
841                         rpc_reset_task_statistics(task);
842                 }
843         }
844 }
845
846 void rpc_signal_task(struct rpc_task *task)
847 {
848         struct rpc_wait_queue *queue;
849
850         if (!RPC_IS_ACTIVATED(task))
851                 return;
852         set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
853         smp_mb__after_atomic();
854         queue = READ_ONCE(task->tk_waitqueue);
855         if (queue)
856                 rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS);
857 }
858
859 void rpc_exit(struct rpc_task *task, int status)
860 {
861         task->tk_status = status;
862         task->tk_action = rpc_exit_task;
863         rpc_wake_up_queued_task(task->tk_waitqueue, task);
864 }
865 EXPORT_SYMBOL_GPL(rpc_exit);
866
867 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
868 {
869         if (ops->rpc_release != NULL)
870                 ops->rpc_release(calldata);
871 }
872
873 /*
874  * This is the RPC `scheduler' (or rather, the finite state machine).
875  */
876 static void __rpc_execute(struct rpc_task *task)
877 {
878         struct rpc_wait_queue *queue;
879         int task_is_async = RPC_IS_ASYNC(task);
880         int status = 0;
881
882         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
883                         task->tk_pid, task->tk_flags);
884
885         WARN_ON_ONCE(RPC_IS_QUEUED(task));
886         if (RPC_IS_QUEUED(task))
887                 return;
888
889         for (;;) {
890                 void (*do_action)(struct rpc_task *);
891
892                 /*
893                  * Perform the next FSM step or a pending callback.
894                  *
895                  * tk_action may be NULL if the task has been killed.
896                  * In particular, note that rpc_killall_tasks may
897                  * do this at any time, so beware when dereferencing.
898                  */
899                 do_action = task->tk_action;
900                 if (task->tk_callback) {
901                         do_action = task->tk_callback;
902                         task->tk_callback = NULL;
903                 }
904                 if (!do_action)
905                         break;
906                 trace_rpc_task_run_action(task, do_action);
907                 do_action(task);
908
909                 /*
910                  * Lockless check for whether task is sleeping or not.
911                  */
912                 if (!RPC_IS_QUEUED(task))
913                         continue;
914
915                 /*
916                  * Signalled tasks should exit rather than sleep.
917                  */
918                 if (RPC_SIGNALLED(task))
919                         rpc_exit(task, -ERESTARTSYS);
920
921                 /*
922                  * The queue->lock protects against races with
923                  * rpc_make_runnable().
924                  *
925                  * Note that once we clear RPC_TASK_RUNNING on an asynchronous
926                  * rpc_task, rpc_make_runnable() can assign it to a
927                  * different workqueue. We therefore cannot assume that the
928                  * rpc_task pointer may still be dereferenced.
929                  */
930                 queue = task->tk_waitqueue;
931                 spin_lock_bh(&queue->lock);
932                 if (!RPC_IS_QUEUED(task)) {
933                         spin_unlock_bh(&queue->lock);
934                         continue;
935                 }
936                 rpc_clear_running(task);
937                 spin_unlock_bh(&queue->lock);
938                 if (task_is_async)
939                         return;
940
941                 /* sync task: sleep here */
942                 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
943                 status = out_of_line_wait_on_bit(&task->tk_runstate,
944                                 RPC_TASK_QUEUED, rpc_wait_bit_killable,
945                                 TASK_KILLABLE);
946                 if (status < 0) {
947                         /*
948                          * When a sync task receives a signal, it exits with
949                          * -ERESTARTSYS. In order to catch any callbacks that
950                          * clean up after sleeping on some queue, we don't
951                          * break the loop here, but go around once more.
952                          */
953                         dprintk("RPC: %5u got signal\n", task->tk_pid);
954                         set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
955                         rpc_exit(task, -ERESTARTSYS);
956                 }
957                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
958         }
959
960         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
961                         task->tk_status);
962         /* Release all resources associated with the task */
963         rpc_release_task(task);
964 }
965
966 /*
967  * User-visible entry point to the scheduler.
968  *
969  * This may be called recursively if e.g. an async NFS task updates
970  * the attributes and finds that dirty pages must be flushed.
971  * NOTE: Upon exit of this function the task is guaranteed to be
972  *       released. In particular note that tk_release() will have
973  *       been called, so your task memory may have been freed.
974  */
975 void rpc_execute(struct rpc_task *task)
976 {
977         bool is_async = RPC_IS_ASYNC(task);
978
979         rpc_set_active(task);
980         rpc_make_runnable(rpciod_workqueue, task);
981         if (!is_async)
982                 __rpc_execute(task);
983 }
984
985 static void rpc_async_schedule(struct work_struct *work)
986 {
987         unsigned int pflags = memalloc_nofs_save();
988
989         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
990         memalloc_nofs_restore(pflags);
991 }
992
993 /**
994  * rpc_malloc - allocate RPC buffer resources
995  * @task: RPC task
996  *
997  * A single memory region is allocated, which is split between the
998  * RPC call and RPC reply that this task is being used for. When
999  * this RPC is retired, the memory is released by calling rpc_free.
1000  *
1001  * To prevent rpciod from hanging, this allocator never sleeps,
1002  * returning -ENOMEM and suppressing warning if the request cannot
1003  * be serviced immediately. The caller can arrange to sleep in a
1004  * way that is safe for rpciod.
1005  *
1006  * Most requests are 'small' (under 2KiB) and can be serviced from a
1007  * mempool, ensuring that NFS reads and writes can always proceed,
1008  * and that there is good locality of reference for these buffers.
1009  */
1010 int rpc_malloc(struct rpc_task *task)
1011 {
1012         struct rpc_rqst *rqst = task->tk_rqstp;
1013         size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1014         struct rpc_buffer *buf;
1015         gfp_t gfp = GFP_NOFS;
1016
1017         if (RPC_IS_SWAPPER(task))
1018                 gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
1019
1020         size += sizeof(struct rpc_buffer);
1021         if (size <= RPC_BUFFER_MAXSIZE)
1022                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
1023         else
1024                 buf = kmalloc(size, gfp);
1025
1026         if (!buf)
1027                 return -ENOMEM;
1028
1029         buf->len = size;
1030         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
1031                         task->tk_pid, size, buf);
1032         rqst->rq_buffer = buf->data;
1033         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1034         return 0;
1035 }
1036 EXPORT_SYMBOL_GPL(rpc_malloc);
1037
1038 /**
1039  * rpc_free - free RPC buffer resources allocated via rpc_malloc
1040  * @task: RPC task
1041  *
1042  */
1043 void rpc_free(struct rpc_task *task)
1044 {
1045         void *buffer = task->tk_rqstp->rq_buffer;
1046         size_t size;
1047         struct rpc_buffer *buf;
1048
1049         buf = container_of(buffer, struct rpc_buffer, data);
1050         size = buf->len;
1051
1052         dprintk("RPC:       freeing buffer of size %zu at %p\n",
1053                         size, buf);
1054
1055         if (size <= RPC_BUFFER_MAXSIZE)
1056                 mempool_free(buf, rpc_buffer_mempool);
1057         else
1058                 kfree(buf);
1059 }
1060 EXPORT_SYMBOL_GPL(rpc_free);
1061
1062 /*
1063  * Creation and deletion of RPC task structures
1064  */
1065 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1066 {
1067         memset(task, 0, sizeof(*task));
1068         atomic_set(&task->tk_count, 1);
1069         task->tk_flags  = task_setup_data->flags;
1070         task->tk_ops = task_setup_data->callback_ops;
1071         task->tk_calldata = task_setup_data->callback_data;
1072         INIT_LIST_HEAD(&task->tk_task);
1073
1074         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1075         task->tk_owner = current->tgid;
1076
1077         /* Initialize workqueue for async tasks */
1078         task->tk_workqueue = task_setup_data->workqueue;
1079
1080         task->tk_xprt = xprt_get(task_setup_data->rpc_xprt);
1081
1082         task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1083
1084         if (task->tk_ops->rpc_call_prepare != NULL)
1085                 task->tk_action = rpc_prepare_task;
1086
1087         rpc_init_task_statistics(task);
1088
1089         dprintk("RPC:       new task initialized, procpid %u\n",
1090                                 task_pid_nr(current));
1091 }
1092
1093 static struct rpc_task *
1094 rpc_alloc_task(void)
1095 {
1096         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
1097 }
1098
1099 /*
1100  * Create a new task for the specified client.
1101  */
1102 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1103 {
1104         struct rpc_task *task = setup_data->task;
1105         unsigned short flags = 0;
1106
1107         if (task == NULL) {
1108                 task = rpc_alloc_task();
1109                 flags = RPC_TASK_DYNAMIC;
1110         }
1111
1112         rpc_init_task(task, setup_data);
1113         task->tk_flags |= flags;
1114         dprintk("RPC:       allocated task %p\n", task);
1115         return task;
1116 }
1117
1118 /*
1119  * rpc_free_task - release rpc task and perform cleanups
1120  *
1121  * Note that we free up the rpc_task _after_ rpc_release_calldata()
1122  * in order to work around a workqueue dependency issue.
1123  *
1124  * Tejun Heo states:
1125  * "Workqueue currently considers two work items to be the same if they're
1126  * on the same address and won't execute them concurrently - ie. it
1127  * makes a work item which is queued again while being executed wait
1128  * for the previous execution to complete.
1129  *
1130  * If a work function frees the work item, and then waits for an event
1131  * which should be performed by another work item and *that* work item
1132  * recycles the freed work item, it can create a false dependency loop.
1133  * There really is no reliable way to detect this short of verifying
1134  * every memory free."
1135  *
1136  */
1137 static void rpc_free_task(struct rpc_task *task)
1138 {
1139         unsigned short tk_flags = task->tk_flags;
1140
1141         put_rpccred(task->tk_op_cred);
1142         rpc_release_calldata(task->tk_ops, task->tk_calldata);
1143
1144         if (tk_flags & RPC_TASK_DYNAMIC) {
1145                 dprintk("RPC: %5u freeing task\n", task->tk_pid);
1146                 mempool_free(task, rpc_task_mempool);
1147         }
1148 }
1149
1150 static void rpc_async_release(struct work_struct *work)
1151 {
1152         unsigned int pflags = memalloc_nofs_save();
1153
1154         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1155         memalloc_nofs_restore(pflags);
1156 }
1157
1158 static void rpc_release_resources_task(struct rpc_task *task)
1159 {
1160         xprt_release(task);
1161         if (task->tk_msg.rpc_cred) {
1162                 put_cred(task->tk_msg.rpc_cred);
1163                 task->tk_msg.rpc_cred = NULL;
1164         }
1165         rpc_task_release_client(task);
1166 }
1167
1168 static void rpc_final_put_task(struct rpc_task *task,
1169                 struct workqueue_struct *q)
1170 {
1171         if (q != NULL) {
1172                 INIT_WORK(&task->u.tk_work, rpc_async_release);
1173                 queue_work(q, &task->u.tk_work);
1174         } else
1175                 rpc_free_task(task);
1176 }
1177
1178 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1179 {
1180         if (atomic_dec_and_test(&task->tk_count)) {
1181                 rpc_release_resources_task(task);
1182                 rpc_final_put_task(task, q);
1183         }
1184 }
1185
1186 void rpc_put_task(struct rpc_task *task)
1187 {
1188         rpc_do_put_task(task, NULL);
1189 }
1190 EXPORT_SYMBOL_GPL(rpc_put_task);
1191
1192 void rpc_put_task_async(struct rpc_task *task)
1193 {
1194         rpc_do_put_task(task, task->tk_workqueue);
1195 }
1196 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1197
1198 static void rpc_release_task(struct rpc_task *task)
1199 {
1200         dprintk("RPC: %5u release task\n", task->tk_pid);
1201
1202         WARN_ON_ONCE(RPC_IS_QUEUED(task));
1203
1204         rpc_release_resources_task(task);
1205
1206         /*
1207          * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1208          * so it should be safe to use task->tk_count as a test for whether
1209          * or not any other processes still hold references to our rpc_task.
1210          */
1211         if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1212                 /* Wake up anyone who may be waiting for task completion */
1213                 if (!rpc_complete_task(task))
1214                         return;
1215         } else {
1216                 if (!atomic_dec_and_test(&task->tk_count))
1217                         return;
1218         }
1219         rpc_final_put_task(task, task->tk_workqueue);
1220 }
1221
1222 int rpciod_up(void)
1223 {
1224         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1225 }
1226
1227 void rpciod_down(void)
1228 {
1229         module_put(THIS_MODULE);
1230 }
1231
1232 /*
1233  * Start up the rpciod workqueue.
1234  */
1235 static int rpciod_start(void)
1236 {
1237         struct workqueue_struct *wq;
1238
1239         /*
1240          * Create the rpciod thread and wait for it to start.
1241          */
1242         dprintk("RPC:       creating workqueue rpciod\n");
1243         wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1244         if (!wq)
1245                 goto out_failed;
1246         rpciod_workqueue = wq;
1247         /* Note: highpri because network receive is latency sensitive */
1248         wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0);
1249         if (!wq)
1250                 goto free_rpciod;
1251         xprtiod_workqueue = wq;
1252         return 1;
1253 free_rpciod:
1254         wq = rpciod_workqueue;
1255         rpciod_workqueue = NULL;
1256         destroy_workqueue(wq);
1257 out_failed:
1258         return 0;
1259 }
1260
1261 static void rpciod_stop(void)
1262 {
1263         struct workqueue_struct *wq = NULL;
1264
1265         if (rpciod_workqueue == NULL)
1266                 return;
1267         dprintk("RPC:       destroying workqueue rpciod\n");
1268
1269         wq = rpciod_workqueue;
1270         rpciod_workqueue = NULL;
1271         destroy_workqueue(wq);
1272         wq = xprtiod_workqueue;
1273         xprtiod_workqueue = NULL;
1274         destroy_workqueue(wq);
1275 }
1276
1277 void
1278 rpc_destroy_mempool(void)
1279 {
1280         rpciod_stop();
1281         mempool_destroy(rpc_buffer_mempool);
1282         mempool_destroy(rpc_task_mempool);
1283         kmem_cache_destroy(rpc_task_slabp);
1284         kmem_cache_destroy(rpc_buffer_slabp);
1285         rpc_destroy_wait_queue(&delay_queue);
1286 }
1287
1288 int
1289 rpc_init_mempool(void)
1290 {
1291         /*
1292          * The following is not strictly a mempool initialisation,
1293          * but there is no harm in doing it here
1294          */
1295         rpc_init_wait_queue(&delay_queue, "delayq");
1296         if (!rpciod_start())
1297                 goto err_nomem;
1298
1299         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1300                                              sizeof(struct rpc_task),
1301                                              0, SLAB_HWCACHE_ALIGN,
1302                                              NULL);
1303         if (!rpc_task_slabp)
1304                 goto err_nomem;
1305         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1306                                              RPC_BUFFER_MAXSIZE,
1307                                              0, SLAB_HWCACHE_ALIGN,
1308                                              NULL);
1309         if (!rpc_buffer_slabp)
1310                 goto err_nomem;
1311         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1312                                                     rpc_task_slabp);
1313         if (!rpc_task_mempool)
1314                 goto err_nomem;
1315         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1316                                                       rpc_buffer_slabp);
1317         if (!rpc_buffer_mempool)
1318                 goto err_nomem;
1319         return 0;
1320 err_nomem:
1321         rpc_destroy_mempool();
1322         return -ENOMEM;
1323 }