]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/net/sunrpc/sched.c | |
4 | * | |
5 | * Scheduling for synchronous and asynchronous RPC requests. | |
6 | * | |
7 | * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de> | |
cca5172a | 8 | * |
1da177e4 LT |
9 | * TCP NFS related read + write fixes |
10 | * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> | |
11 | */ | |
12 | ||
13 | #include <linux/module.h> | |
14 | ||
15 | #include <linux/sched.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/smp.h> | |
1da177e4 | 20 | #include <linux/spinlock.h> |
4a3e2f71 | 21 | #include <linux/mutex.h> |
d310310c | 22 | #include <linux/freezer.h> |
a1231fda | 23 | #include <linux/sched/mm.h> |
1da177e4 LT |
24 | |
25 | #include <linux/sunrpc/clnt.h> | |
1da177e4 | 26 | |
6951867b BH |
27 | #include "sunrpc.h" |
28 | ||
f895b252 | 29 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
1da177e4 | 30 | #define RPCDBG_FACILITY RPCDBG_SCHED |
1da177e4 LT |
31 | #endif |
32 | ||
82b0a4c3 TM |
33 | #define CREATE_TRACE_POINTS |
34 | #include <trace/events/sunrpc.h> | |
35 | ||
1da177e4 LT |
36 | /* |
37 | * RPC slabs and memory pools | |
38 | */ | |
39 | #define RPC_BUFFER_MAXSIZE (2048) | |
40 | #define RPC_BUFFER_POOLSIZE (8) | |
41 | #define RPC_TASK_POOLSIZE (8) | |
e18b890b CL |
42 | static struct kmem_cache *rpc_task_slabp __read_mostly; |
43 | static struct kmem_cache *rpc_buffer_slabp __read_mostly; | |
ba89966c ED |
44 | static mempool_t *rpc_task_mempool __read_mostly; |
45 | static mempool_t *rpc_buffer_mempool __read_mostly; | |
1da177e4 | 46 | |
65f27f38 | 47 | static void rpc_async_schedule(struct work_struct *); |
bde8f00c | 48 | static void rpc_release_task(struct rpc_task *task); |
ff861c4d | 49 | static void __rpc_queue_timer_fn(struct timer_list *t); |
1da177e4 | 50 | |
1da177e4 LT |
51 | /* |
52 | * RPC tasks sit here while waiting for conditions to improve. | |
53 | */ | |
a4a87499 | 54 | static struct rpc_wait_queue delay_queue; |
1da177e4 | 55 | |
1da177e4 LT |
56 | /* |
57 | * rpciod-related stuff | |
58 | */ | |
40a5f1b1 TM |
59 | struct workqueue_struct *rpciod_workqueue __read_mostly; |
60 | struct workqueue_struct *xprtiod_workqueue __read_mostly; | |
1da177e4 | 61 | |
5efd1876 TM |
62 | unsigned long |
63 | rpc_task_timeout(const struct rpc_task *task) | |
64 | { | |
65 | unsigned long timeout = READ_ONCE(task->tk_timeout); | |
66 | ||
67 | if (timeout != 0) { | |
68 | unsigned long now = jiffies; | |
69 | if (time_before(now, timeout)) | |
70 | return timeout - now; | |
71 | } | |
72 | return 0; | |
73 | } | |
74 | EXPORT_SYMBOL_GPL(rpc_task_timeout); | |
75 | ||
1da177e4 LT |
76 | /* |
77 | * Disable the timer for a given RPC task. Should be called with | |
78 | * queue->lock and bh_disabled in order to avoid races within | |
79 | * rpc_run_timer(). | |
80 | */ | |
5d00837b | 81 | static void |
eb276c0e | 82 | __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 83 | { |
6b2e6856 | 84 | if (list_empty(&task->u.tk_wait.timer_list)) |
36df9aae | 85 | return; |
46121cf7 | 86 | dprintk("RPC: %5u disabling timer\n", task->tk_pid); |
1da177e4 | 87 | task->tk_timeout = 0; |
36df9aae | 88 | list_del(&task->u.tk_wait.timer_list); |
eb276c0e TM |
89 | if (list_empty(&queue->timer_list.list)) |
90 | del_timer(&queue->timer_list.timer); | |
36df9aae TM |
91 | } |
92 | ||
93 | static void | |
94 | rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires) | |
95 | { | |
24a9d9a2 | 96 | timer_reduce(&queue->timer_list.timer, expires); |
1da177e4 LT |
97 | } |
98 | ||
1da177e4 LT |
99 | /* |
100 | * Set up a timer for the current task. | |
101 | */ | |
5d00837b | 102 | static void |
6b2e6856 TM |
103 | __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task, |
104 | unsigned long timeout) | |
1da177e4 | 105 | { |
55cc1d78 | 106 | dprintk("RPC: %5u setting alarm for %u ms\n", |
6b2e6856 | 107 | task->tk_pid, jiffies_to_msecs(timeout - jiffies)); |
1da177e4 | 108 | |
6b2e6856 | 109 | task->tk_timeout = timeout; |
24a9d9a2 | 110 | rpc_set_queue_timer(queue, timeout); |
eb276c0e | 111 | list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); |
1da177e4 LT |
112 | } |
113 | ||
c05eecf6 TM |
114 | static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) |
115 | { | |
edd2e36f | 116 | if (queue->priority != priority) { |
edd2e36f | 117 | queue->priority = priority; |
f42f7c28 | 118 | queue->nr = 1U << priority; |
edd2e36f | 119 | } |
c05eecf6 TM |
120 | } |
121 | ||
c05eecf6 TM |
122 | static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) |
123 | { | |
124 | rpc_set_waitqueue_priority(queue, queue->maxpriority); | |
c05eecf6 TM |
125 | } |
126 | ||
1da177e4 | 127 | /* |
f42f7c28 | 128 | * Add a request to a queue list |
1da177e4 | 129 | */ |
f42f7c28 TM |
130 | static void |
131 | __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task) | |
1da177e4 | 132 | { |
1da177e4 LT |
133 | struct rpc_task *t; |
134 | ||
1da177e4 | 135 | list_for_each_entry(t, q, u.tk_wait.list) { |
3ff7576d | 136 | if (t->tk_owner == task->tk_owner) { |
f42f7c28 TM |
137 | list_add_tail(&task->u.tk_wait.links, |
138 | &t->u.tk_wait.links); | |
139 | /* Cache the queue head in task->u.tk_wait.list */ | |
140 | task->u.tk_wait.list.next = q; | |
141 | task->u.tk_wait.list.prev = NULL; | |
1da177e4 LT |
142 | return; |
143 | } | |
144 | } | |
f42f7c28 | 145 | INIT_LIST_HEAD(&task->u.tk_wait.links); |
1da177e4 LT |
146 | list_add_tail(&task->u.tk_wait.list, q); |
147 | } | |
148 | ||
f42f7c28 TM |
149 | /* |
150 | * Remove request from a queue list | |
151 | */ | |
152 | static void | |
153 | __rpc_list_dequeue_task(struct rpc_task *task) | |
154 | { | |
155 | struct list_head *q; | |
156 | struct rpc_task *t; | |
157 | ||
158 | if (task->u.tk_wait.list.prev == NULL) { | |
159 | list_del(&task->u.tk_wait.links); | |
160 | return; | |
161 | } | |
162 | if (!list_empty(&task->u.tk_wait.links)) { | |
163 | t = list_first_entry(&task->u.tk_wait.links, | |
164 | struct rpc_task, | |
165 | u.tk_wait.links); | |
166 | /* Assume __rpc_list_enqueue_task() cached the queue head */ | |
167 | q = t->u.tk_wait.list.next; | |
168 | list_add_tail(&t->u.tk_wait.list, q); | |
169 | list_del(&task->u.tk_wait.links); | |
170 | } | |
171 | list_del(&task->u.tk_wait.list); | |
172 | } | |
173 | ||
174 | /* | |
175 | * Add new request to a priority queue. | |
176 | */ | |
177 | static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, | |
178 | struct rpc_task *task, | |
179 | unsigned char queue_priority) | |
180 | { | |
181 | if (unlikely(queue_priority > queue->maxpriority)) | |
182 | queue_priority = queue->maxpriority; | |
183 | __rpc_list_enqueue_task(&queue->tasks[queue_priority], task); | |
184 | } | |
185 | ||
1da177e4 LT |
186 | /* |
187 | * Add new request to wait queue. | |
188 | * | |
189 | * Swapper tasks always get inserted at the head of the queue. | |
190 | * This should avoid many nasty memory deadlocks and hopefully | |
191 | * improve overall performance. | |
192 | * Everyone else gets appended to the queue to ensure proper FIFO behavior. | |
193 | */ | |
3b27bad7 TM |
194 | static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, |
195 | struct rpc_task *task, | |
196 | unsigned char queue_priority) | |
1da177e4 | 197 | { |
2bd4eef8 WAA |
198 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
199 | if (RPC_IS_QUEUED(task)) | |
200 | return; | |
1da177e4 | 201 | |
6b2e6856 | 202 | INIT_LIST_HEAD(&task->u.tk_wait.timer_list); |
1da177e4 | 203 | if (RPC_IS_PRIORITY(queue)) |
3b27bad7 | 204 | __rpc_add_wait_queue_priority(queue, task, queue_priority); |
1da177e4 LT |
205 | else if (RPC_IS_SWAPPER(task)) |
206 | list_add(&task->u.tk_wait.list, &queue->tasks[0]); | |
207 | else | |
208 | list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]); | |
96ef13b2 | 209 | task->tk_waitqueue = queue; |
e19b63da | 210 | queue->qlen++; |
1166fde6 TM |
211 | /* barrier matches the read in rpc_wake_up_task_queue_locked() */ |
212 | smp_wmb(); | |
1da177e4 LT |
213 | rpc_set_queued(task); |
214 | ||
46121cf7 CL |
215 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
216 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
217 | } |
218 | ||
219 | /* | |
220 | * Remove request from a priority queue. | |
221 | */ | |
222 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
223 | { | |
f42f7c28 | 224 | __rpc_list_dequeue_task(task); |
1da177e4 LT |
225 | } |
226 | ||
227 | /* | |
228 | * Remove request from queue. | |
229 | * Note: must be called with spin lock held. | |
230 | */ | |
96ef13b2 | 231 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 232 | { |
eb276c0e | 233 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
234 | if (RPC_IS_PRIORITY(queue)) |
235 | __rpc_remove_wait_queue_priority(task); | |
f42f7c28 TM |
236 | else |
237 | list_del(&task->u.tk_wait.list); | |
e19b63da | 238 | queue->qlen--; |
46121cf7 CL |
239 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
240 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
241 | } |
242 | ||
3ff7576d | 243 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
244 | { |
245 | int i; | |
246 | ||
247 | spin_lock_init(&queue->lock); | |
248 | for (i = 0; i < ARRAY_SIZE(queue->tasks); i++) | |
249 | INIT_LIST_HEAD(&queue->tasks[i]); | |
3ff7576d | 250 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 251 | rpc_reset_waitqueue_priority(queue); |
36df9aae | 252 | queue->qlen = 0; |
43123581 TM |
253 | timer_setup(&queue->timer_list.timer, |
254 | __rpc_queue_timer_fn, | |
255 | TIMER_DEFERRABLE); | |
36df9aae | 256 | INIT_LIST_HEAD(&queue->timer_list.list); |
2f09c242 | 257 | rpc_assign_waitqueue_name(queue, qname); |
1da177e4 LT |
258 | } |
259 | ||
260 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
261 | { | |
3ff7576d | 262 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 263 | } |
689cf5c1 | 264 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
265 | |
266 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
267 | { | |
3ff7576d | 268 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 269 | } |
e8914c65 | 270 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 271 | |
f6a1cc89 TM |
272 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
273 | { | |
36df9aae | 274 | del_timer_sync(&queue->timer_list.timer); |
f6a1cc89 TM |
275 | } |
276 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
277 | ||
dfd01f02 | 278 | static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode) |
44c28873 | 279 | { |
416ad3c9 | 280 | freezable_schedule_unsafe(); |
dfd01f02 PZ |
281 | if (signal_pending_state(mode, current)) |
282 | return -ERESTARTSYS; | |
44c28873 TM |
283 | return 0; |
284 | } | |
285 | ||
1306729b | 286 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS) |
c44fe705 TM |
287 | static void rpc_task_set_debuginfo(struct rpc_task *task) |
288 | { | |
289 | static atomic_t rpc_pid; | |
290 | ||
c44fe705 TM |
291 | task->tk_pid = atomic_inc_return(&rpc_pid); |
292 | } | |
293 | #else | |
294 | static inline void rpc_task_set_debuginfo(struct rpc_task *task) | |
295 | { | |
296 | } | |
297 | #endif | |
298 | ||
e6b3c4db TM |
299 | static void rpc_set_active(struct rpc_task *task) |
300 | { | |
c44fe705 | 301 | rpc_task_set_debuginfo(task); |
58f9612c | 302 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e671edb9 | 303 | trace_rpc_task_begin(task, NULL); |
e6b3c4db TM |
304 | } |
305 | ||
44c28873 TM |
306 | /* |
307 | * Mark an RPC call as having completed by clearing the 'active' bit | |
bf294b41 | 308 | * and then waking up all tasks that were sleeping. |
44c28873 | 309 | */ |
bf294b41 | 310 | static int rpc_complete_task(struct rpc_task *task) |
44c28873 | 311 | { |
bf294b41 TM |
312 | void *m = &task->tk_runstate; |
313 | wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE); | |
314 | struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE); | |
315 | unsigned long flags; | |
316 | int ret; | |
317 | ||
e671edb9 | 318 | trace_rpc_task_complete(task, NULL); |
82b0a4c3 | 319 | |
bf294b41 | 320 | spin_lock_irqsave(&wq->lock, flags); |
e6b3c4db | 321 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
bf294b41 TM |
322 | ret = atomic_dec_and_test(&task->tk_count); |
323 | if (waitqueue_active(wq)) | |
ac5be6b4 | 324 | __wake_up_locked_key(wq, TASK_NORMAL, &k); |
bf294b41 TM |
325 | spin_unlock_irqrestore(&wq->lock, flags); |
326 | return ret; | |
44c28873 TM |
327 | } |
328 | ||
329 | /* | |
330 | * Allow callers to wait for completion of an RPC call | |
bf294b41 TM |
331 | * |
332 | * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit() | |
333 | * to enforce taking of the wq->lock and hence avoid races with | |
334 | * rpc_complete_task(). | |
44c28873 | 335 | */ |
c1221321 | 336 | int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action) |
44c28873 TM |
337 | { |
338 | if (action == NULL) | |
150030b7 | 339 | action = rpc_wait_bit_killable; |
bf294b41 | 340 | return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 341 | action, TASK_KILLABLE); |
44c28873 | 342 | } |
e8914c65 | 343 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 344 | |
1da177e4 LT |
345 | /* |
346 | * Make an RPC task runnable. | |
347 | * | |
506026c3 JL |
348 | * Note: If the task is ASYNC, and is being made runnable after sitting on an |
349 | * rpc_wait_queue, this must be called with the queue spinlock held to protect | |
350 | * the wait queue operation. | |
a3c3cac5 TM |
351 | * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(), |
352 | * which is needed to ensure that __rpc_execute() doesn't loop (due to the | |
353 | * lockless RPC_IS_QUEUED() test) before we've had a chance to test | |
354 | * the RPC_TASK_RUNNING flag. | |
1da177e4 | 355 | */ |
f1dc237c TM |
356 | static void rpc_make_runnable(struct workqueue_struct *wq, |
357 | struct rpc_task *task) | |
1da177e4 | 358 | { |
a3c3cac5 TM |
359 | bool need_wakeup = !rpc_test_and_set_running(task); |
360 | ||
1da177e4 | 361 | rpc_clear_queued(task); |
a3c3cac5 | 362 | if (!need_wakeup) |
cc4dc59e | 363 | return; |
1da177e4 | 364 | if (RPC_IS_ASYNC(task)) { |
65f27f38 | 365 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
f1dc237c | 366 | queue_work(wq, &task->u.tk_work); |
1da177e4 | 367 | } else |
96651ab3 | 368 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
369 | } |
370 | ||
1da177e4 LT |
371 | /* |
372 | * Prepare for sleeping on a wait queue. | |
373 | * By always appending tasks to the list we ensure FIFO behavior. | |
374 | * NB: An RPC task will only receive interrupt-driven events as long | |
375 | * as it's on a wait queue. | |
376 | */ | |
3b27bad7 TM |
377 | static void __rpc_sleep_on_priority(struct rpc_wait_queue *q, |
378 | struct rpc_task *task, | |
3b27bad7 | 379 | unsigned char queue_priority) |
1da177e4 | 380 | { |
46121cf7 CL |
381 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
382 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 383 | |
e671edb9 | 384 | trace_rpc_task_sleep(task, q); |
82b0a4c3 | 385 | |
3b27bad7 | 386 | __rpc_add_wait_queue(q, task, queue_priority); |
1da177e4 | 387 | |
6b2e6856 TM |
388 | } |
389 | ||
390 | static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, | |
391 | struct rpc_task *task, unsigned long timeout, | |
392 | unsigned char queue_priority) | |
393 | { | |
394 | if (time_is_after_jiffies(timeout)) { | |
395 | __rpc_sleep_on_priority(q, task, queue_priority); | |
396 | __rpc_add_timer(q, task, timeout); | |
397 | } else | |
398 | task->tk_status = -ETIMEDOUT; | |
1da177e4 LT |
399 | } |
400 | ||
87150aae TM |
401 | static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action) |
402 | { | |
403 | if (action && !WARN_ON_ONCE(task->tk_callback != NULL)) | |
404 | task->tk_callback = action; | |
405 | } | |
406 | ||
407 | static bool rpc_sleep_check_activated(struct rpc_task *task) | |
1da177e4 | 408 | { |
58f9612c | 409 | /* We shouldn't ever put an inactive task to sleep */ |
87150aae | 410 | if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) { |
e454a7a8 WAA |
411 | task->tk_status = -EIO; |
412 | rpc_put_task_async(task); | |
87150aae | 413 | return false; |
e454a7a8 | 414 | } |
87150aae TM |
415 | return true; |
416 | } | |
417 | ||
6b2e6856 TM |
418 | void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task, |
419 | rpc_action action, unsigned long timeout) | |
420 | { | |
421 | if (!rpc_sleep_check_activated(task)) | |
422 | return; | |
423 | ||
424 | rpc_set_tk_callback(task, action); | |
425 | ||
426 | /* | |
427 | * Protect the queue operations. | |
428 | */ | |
429 | spin_lock_bh(&q->lock); | |
430 | __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority); | |
431 | spin_unlock_bh(&q->lock); | |
432 | } | |
433 | EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout); | |
434 | ||
87150aae TM |
435 | void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, |
436 | rpc_action action) | |
437 | { | |
438 | if (!rpc_sleep_check_activated(task)) | |
439 | return; | |
440 | ||
441 | rpc_set_tk_callback(task, action); | |
e6b3c4db | 442 | |
6b2e6856 | 443 | WARN_ON_ONCE(task->tk_timeout != 0); |
1da177e4 LT |
444 | /* |
445 | * Protect the queue operations. | |
446 | */ | |
447 | spin_lock_bh(&q->lock); | |
87150aae | 448 | __rpc_sleep_on_priority(q, task, task->tk_priority); |
1da177e4 LT |
449 | spin_unlock_bh(&q->lock); |
450 | } | |
e8914c65 | 451 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 | 452 | |
6b2e6856 TM |
453 | void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, |
454 | struct rpc_task *task, unsigned long timeout, int priority) | |
455 | { | |
456 | if (!rpc_sleep_check_activated(task)) | |
457 | return; | |
458 | ||
459 | priority -= RPC_PRIORITY_LOW; | |
460 | /* | |
461 | * Protect the queue operations. | |
462 | */ | |
463 | spin_lock_bh(&q->lock); | |
464 | __rpc_sleep_on_priority_timeout(q, task, timeout, priority); | |
465 | spin_unlock_bh(&q->lock); | |
466 | } | |
467 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout); | |
468 | ||
3b27bad7 | 469 | void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task, |
8357a9b6 | 470 | int priority) |
3b27bad7 | 471 | { |
87150aae | 472 | if (!rpc_sleep_check_activated(task)) |
e454a7a8 | 473 | return; |
87150aae | 474 | |
6b2e6856 | 475 | WARN_ON_ONCE(task->tk_timeout != 0); |
8357a9b6 | 476 | priority -= RPC_PRIORITY_LOW; |
3b27bad7 TM |
477 | /* |
478 | * Protect the queue operations. | |
479 | */ | |
480 | spin_lock_bh(&q->lock); | |
8357a9b6 | 481 | __rpc_sleep_on_priority(q, task, priority); |
3b27bad7 TM |
482 | spin_unlock_bh(&q->lock); |
483 | } | |
1e1093c7 | 484 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority); |
3b27bad7 | 485 | |
1da177e4 | 486 | /** |
f1dc237c TM |
487 | * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task |
488 | * @wq: workqueue on which to run task | |
96ef13b2 | 489 | * @queue: wait queue |
1da177e4 LT |
490 | * @task: task to be woken up |
491 | * | |
492 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
493 | */ | |
f1dc237c TM |
494 | static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq, |
495 | struct rpc_wait_queue *queue, | |
496 | struct rpc_task *task) | |
1da177e4 | 497 | { |
46121cf7 CL |
498 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
499 | task->tk_pid, jiffies); | |
1da177e4 | 500 | |
1da177e4 LT |
501 | /* Has the task been executed yet? If not, we cannot wake it up! */ |
502 | if (!RPC_IS_ACTIVATED(task)) { | |
503 | printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task); | |
504 | return; | |
505 | } | |
506 | ||
e671edb9 | 507 | trace_rpc_task_wakeup(task, queue); |
82b0a4c3 | 508 | |
96ef13b2 | 509 | __rpc_remove_wait_queue(queue, task); |
1da177e4 | 510 | |
f1dc237c | 511 | rpc_make_runnable(wq, task); |
1da177e4 | 512 | |
46121cf7 | 513 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
514 | } |
515 | ||
516 | /* | |
96ef13b2 | 517 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 518 | */ |
359c48c0 TM |
519 | static struct rpc_task * |
520 | rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq, | |
521 | struct rpc_wait_queue *queue, struct rpc_task *task, | |
522 | bool (*action)(struct rpc_task *, void *), void *data) | |
1da177e4 | 523 | { |
1166fde6 TM |
524 | if (RPC_IS_QUEUED(task)) { |
525 | smp_rmb(); | |
359c48c0 TM |
526 | if (task->tk_waitqueue == queue) { |
527 | if (action == NULL || action(task, data)) { | |
528 | __rpc_do_wake_up_task_on_wq(wq, queue, task); | |
529 | return task; | |
530 | } | |
531 | } | |
1166fde6 | 532 | } |
359c48c0 TM |
533 | return NULL; |
534 | } | |
535 | ||
536 | static void | |
537 | rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq, | |
538 | struct rpc_wait_queue *queue, struct rpc_task *task) | |
539 | { | |
540 | rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL); | |
1da177e4 LT |
541 | } |
542 | ||
f1dc237c TM |
543 | /* |
544 | * Wake up a queued task while the queue lock is being held | |
545 | */ | |
546 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task) | |
547 | { | |
548 | rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task); | |
549 | } | |
550 | ||
2275cde4 TM |
551 | /* |
552 | * Wake up a task on a specific queue | |
553 | */ | |
554 | void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq, | |
555 | struct rpc_wait_queue *queue, | |
556 | struct rpc_task *task) | |
557 | { | |
5ce97039 TM |
558 | if (!RPC_IS_QUEUED(task)) |
559 | return; | |
2275cde4 TM |
560 | spin_lock_bh(&queue->lock); |
561 | rpc_wake_up_task_on_wq_queue_locked(wq, queue, task); | |
562 | spin_unlock_bh(&queue->lock); | |
563 | } | |
564 | ||
1da177e4 | 565 | /* |
96ef13b2 | 566 | * Wake up a task on a specific queue |
1da177e4 | 567 | */ |
96ef13b2 | 568 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 569 | { |
5ce97039 TM |
570 | if (!RPC_IS_QUEUED(task)) |
571 | return; | |
5e4424af | 572 | spin_lock_bh(&queue->lock); |
96ef13b2 | 573 | rpc_wake_up_task_queue_locked(queue, task); |
5e4424af | 574 | spin_unlock_bh(&queue->lock); |
1da177e4 | 575 | } |
96ef13b2 TM |
576 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
577 | ||
359c48c0 TM |
578 | static bool rpc_task_action_set_status(struct rpc_task *task, void *status) |
579 | { | |
580 | task->tk_status = *(int *)status; | |
581 | return true; | |
582 | } | |
583 | ||
584 | static void | |
585 | rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue, | |
586 | struct rpc_task *task, int status) | |
587 | { | |
588 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, | |
589 | task, rpc_task_action_set_status, &status); | |
590 | } | |
591 | ||
592 | /** | |
593 | * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status | |
594 | * @queue: pointer to rpc_wait_queue | |
595 | * @task: pointer to rpc_task | |
596 | * @status: integer error value | |
597 | * | |
598 | * If @task is queued on @queue, then it is woken up, and @task->tk_status is | |
599 | * set to the value of @status. | |
600 | */ | |
601 | void | |
602 | rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue, | |
603 | struct rpc_task *task, int status) | |
604 | { | |
605 | if (!RPC_IS_QUEUED(task)) | |
606 | return; | |
607 | spin_lock_bh(&queue->lock); | |
608 | rpc_wake_up_task_queue_set_status_locked(queue, task, status); | |
609 | spin_unlock_bh(&queue->lock); | |
610 | } | |
611 | ||
1da177e4 LT |
612 | /* |
613 | * Wake up the next task on a priority queue. | |
614 | */ | |
961a828d | 615 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
616 | { |
617 | struct list_head *q; | |
618 | struct rpc_task *task; | |
619 | ||
620 | /* | |
3ff7576d | 621 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
622 | */ |
623 | q = &queue->tasks[queue->priority]; | |
f42f7c28 TM |
624 | if (!list_empty(q) && --queue->nr) { |
625 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); | |
626 | goto out; | |
1da177e4 LT |
627 | } |
628 | ||
629 | /* | |
630 | * Service the next queue. | |
631 | */ | |
632 | do { | |
633 | if (q == &queue->tasks[0]) | |
634 | q = &queue->tasks[queue->maxpriority]; | |
635 | else | |
636 | q = q - 1; | |
637 | if (!list_empty(q)) { | |
f42f7c28 | 638 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
1da177e4 LT |
639 | goto new_queue; |
640 | } | |
641 | } while (q != &queue->tasks[queue->priority]); | |
642 | ||
643 | rpc_reset_waitqueue_priority(queue); | |
644 | return NULL; | |
645 | ||
646 | new_queue: | |
647 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
1da177e4 | 648 | out: |
1da177e4 LT |
649 | return task; |
650 | } | |
651 | ||
961a828d TM |
652 | static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue) |
653 | { | |
654 | if (RPC_IS_PRIORITY(queue)) | |
655 | return __rpc_find_next_queued_priority(queue); | |
656 | if (!list_empty(&queue->tasks[0])) | |
657 | return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list); | |
658 | return NULL; | |
659 | } | |
660 | ||
1da177e4 | 661 | /* |
961a828d | 662 | * Wake up the first task on the wait queue. |
1da177e4 | 663 | */ |
f1dc237c TM |
664 | struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq, |
665 | struct rpc_wait_queue *queue, | |
961a828d | 666 | bool (*func)(struct rpc_task *, void *), void *data) |
1da177e4 LT |
667 | { |
668 | struct rpc_task *task = NULL; | |
669 | ||
961a828d | 670 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 671 | queue, rpc_qname(queue)); |
5e4424af | 672 | spin_lock_bh(&queue->lock); |
961a828d | 673 | task = __rpc_find_next_queued(queue); |
359c48c0 TM |
674 | if (task != NULL) |
675 | task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, | |
676 | task, func, data); | |
5e4424af | 677 | spin_unlock_bh(&queue->lock); |
1da177e4 LT |
678 | |
679 | return task; | |
680 | } | |
f1dc237c TM |
681 | |
682 | /* | |
683 | * Wake up the first task on the wait queue. | |
684 | */ | |
685 | struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue, | |
686 | bool (*func)(struct rpc_task *, void *), void *data) | |
687 | { | |
688 | return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data); | |
689 | } | |
961a828d TM |
690 | EXPORT_SYMBOL_GPL(rpc_wake_up_first); |
691 | ||
692 | static bool rpc_wake_up_next_func(struct rpc_task *task, void *data) | |
693 | { | |
694 | return true; | |
695 | } | |
696 | ||
697 | /* | |
698 | * Wake up the next task on the wait queue. | |
699 | */ | |
700 | struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue) | |
701 | { | |
702 | return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL); | |
703 | } | |
e8914c65 | 704 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
705 | |
706 | /** | |
707 | * rpc_wake_up - wake up all rpc_tasks | |
708 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
709 | * | |
710 | * Grabs queue->lock | |
711 | */ | |
712 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
713 | { | |
1da177e4 | 714 | struct list_head *head; |
e6d83d55 | 715 | |
5e4424af | 716 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
717 | head = &queue->tasks[queue->maxpriority]; |
718 | for (;;) { | |
540a0f75 TM |
719 | while (!list_empty(head)) { |
720 | struct rpc_task *task; | |
721 | task = list_first_entry(head, | |
722 | struct rpc_task, | |
723 | u.tk_wait.list); | |
96ef13b2 | 724 | rpc_wake_up_task_queue_locked(queue, task); |
540a0f75 | 725 | } |
1da177e4 LT |
726 | if (head == &queue->tasks[0]) |
727 | break; | |
728 | head--; | |
729 | } | |
5e4424af | 730 | spin_unlock_bh(&queue->lock); |
1da177e4 | 731 | } |
e8914c65 | 732 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
733 | |
734 | /** | |
735 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
736 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
737 | * @status: status value to set | |
738 | * | |
739 | * Grabs queue->lock | |
740 | */ | |
741 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
742 | { | |
743 | struct list_head *head; | |
1da177e4 | 744 | |
5e4424af | 745 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
746 | head = &queue->tasks[queue->maxpriority]; |
747 | for (;;) { | |
540a0f75 TM |
748 | while (!list_empty(head)) { |
749 | struct rpc_task *task; | |
750 | task = list_first_entry(head, | |
751 | struct rpc_task, | |
752 | u.tk_wait.list); | |
1da177e4 | 753 | task->tk_status = status; |
96ef13b2 | 754 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
755 | } |
756 | if (head == &queue->tasks[0]) | |
757 | break; | |
758 | head--; | |
759 | } | |
5e4424af | 760 | spin_unlock_bh(&queue->lock); |
1da177e4 | 761 | } |
e8914c65 | 762 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 763 | |
ff861c4d | 764 | static void __rpc_queue_timer_fn(struct timer_list *t) |
36df9aae | 765 | { |
ff861c4d | 766 | struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer); |
36df9aae TM |
767 | struct rpc_task *task, *n; |
768 | unsigned long expires, now, timeo; | |
769 | ||
770 | spin_lock(&queue->lock); | |
771 | expires = now = jiffies; | |
772 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
6b2e6856 | 773 | timeo = task->tk_timeout; |
36df9aae | 774 | if (time_after_eq(now, timeo)) { |
36df9aae TM |
775 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
776 | task->tk_status = -ETIMEDOUT; | |
777 | rpc_wake_up_task_queue_locked(queue, task); | |
778 | continue; | |
779 | } | |
780 | if (expires == now || time_after(expires, timeo)) | |
781 | expires = timeo; | |
782 | } | |
783 | if (!list_empty(&queue->timer_list.list)) | |
784 | rpc_set_queue_timer(queue, expires); | |
785 | spin_unlock(&queue->lock); | |
786 | } | |
787 | ||
8014793b TM |
788 | static void __rpc_atrun(struct rpc_task *task) |
789 | { | |
6bd14416 TM |
790 | if (task->tk_status == -ETIMEDOUT) |
791 | task->tk_status = 0; | |
8014793b TM |
792 | } |
793 | ||
1da177e4 LT |
794 | /* |
795 | * Run a task at a later time | |
796 | */ | |
8014793b | 797 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 | 798 | { |
6b2e6856 | 799 | rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay); |
1da177e4 | 800 | } |
e8914c65 | 801 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 802 | |
4ce70ada TM |
803 | /* |
804 | * Helper to call task->tk_ops->rpc_call_prepare | |
805 | */ | |
aae2006e | 806 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
807 | { |
808 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
809 | } | |
810 | ||
7fdcf13b TM |
811 | static void |
812 | rpc_init_task_statistics(struct rpc_task *task) | |
813 | { | |
814 | /* Initialize retry counters */ | |
815 | task->tk_garb_retry = 2; | |
816 | task->tk_cred_retry = 2; | |
817 | task->tk_rebind_retry = 2; | |
818 | ||
819 | /* starting timestamp */ | |
820 | task->tk_start = ktime_get(); | |
821 | } | |
822 | ||
823 | static void | |
824 | rpc_reset_task_statistics(struct rpc_task *task) | |
825 | { | |
826 | task->tk_timeouts = 0; | |
ae67bd38 | 827 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT); |
7fdcf13b TM |
828 | rpc_init_task_statistics(task); |
829 | } | |
830 | ||
d05fdb0c | 831 | /* |
963d8fe5 | 832 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 833 | */ |
abbcf28f | 834 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 835 | { |
abbcf28f | 836 | task->tk_action = NULL; |
963d8fe5 TM |
837 | if (task->tk_ops->rpc_call_done != NULL) { |
838 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 839 | if (task->tk_action != NULL) { |
abbcf28f TM |
840 | /* Always release the RPC slot and buffer memory */ |
841 | xprt_release(task); | |
7fdcf13b | 842 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
843 | } |
844 | } | |
d05fdb0c | 845 | } |
d9b6cd94 | 846 | |
ae67bd38 TM |
847 | void rpc_signal_task(struct rpc_task *task) |
848 | { | |
849 | struct rpc_wait_queue *queue; | |
850 | ||
851 | if (!RPC_IS_ACTIVATED(task)) | |
852 | return; | |
853 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); | |
854 | smp_mb__after_atomic(); | |
855 | queue = READ_ONCE(task->tk_waitqueue); | |
856 | if (queue) | |
857 | rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS); | |
858 | } | |
859 | ||
d9b6cd94 TM |
860 | void rpc_exit(struct rpc_task *task, int status) |
861 | { | |
862 | task->tk_status = status; | |
863 | task->tk_action = rpc_exit_task; | |
6b5f5900 | 864 | rpc_wake_up_queued_task(task->tk_waitqueue, task); |
d9b6cd94 TM |
865 | } |
866 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 867 | |
bbd5a1f9 TM |
868 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
869 | { | |
a86dc496 | 870 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 871 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
872 | } |
873 | ||
1da177e4 LT |
874 | /* |
875 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
876 | */ | |
2efef837 | 877 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 878 | { |
eb9b55ab TM |
879 | struct rpc_wait_queue *queue; |
880 | int task_is_async = RPC_IS_ASYNC(task); | |
881 | int status = 0; | |
1da177e4 | 882 | |
46121cf7 CL |
883 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
884 | task->tk_pid, task->tk_flags); | |
1da177e4 | 885 | |
2bd4eef8 WAA |
886 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
887 | if (RPC_IS_QUEUED(task)) | |
888 | return; | |
1da177e4 | 889 | |
d05fdb0c | 890 | for (;;) { |
b55c5989 | 891 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
892 | |
893 | /* | |
21ead9ff CL |
894 | * Perform the next FSM step or a pending callback. |
895 | * | |
896 | * tk_action may be NULL if the task has been killed. | |
897 | * In particular, note that rpc_killall_tasks may | |
898 | * do this at any time, so beware when dereferencing. | |
1da177e4 | 899 | */ |
21ead9ff CL |
900 | do_action = task->tk_action; |
901 | if (task->tk_callback) { | |
902 | do_action = task->tk_callback; | |
903 | task->tk_callback = NULL; | |
1da177e4 | 904 | } |
21ead9ff CL |
905 | if (!do_action) |
906 | break; | |
e671edb9 | 907 | trace_rpc_task_run_action(task, do_action); |
b55c5989 | 908 | do_action(task); |
1da177e4 LT |
909 | |
910 | /* | |
911 | * Lockless check for whether task is sleeping or not. | |
912 | */ | |
913 | if (!RPC_IS_QUEUED(task)) | |
914 | continue; | |
ae67bd38 TM |
915 | |
916 | /* | |
917 | * Signalled tasks should exit rather than sleep. | |
918 | */ | |
919 | if (RPC_SIGNALLED(task)) | |
920 | rpc_exit(task, -ERESTARTSYS); | |
921 | ||
eb9b55ab TM |
922 | /* |
923 | * The queue->lock protects against races with | |
924 | * rpc_make_runnable(). | |
925 | * | |
926 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
927 | * rpc_task, rpc_make_runnable() can assign it to a | |
928 | * different workqueue. We therefore cannot assume that the | |
929 | * rpc_task pointer may still be dereferenced. | |
930 | */ | |
931 | queue = task->tk_waitqueue; | |
932 | spin_lock_bh(&queue->lock); | |
933 | if (!RPC_IS_QUEUED(task)) { | |
934 | spin_unlock_bh(&queue->lock); | |
1da177e4 LT |
935 | continue; |
936 | } | |
eb9b55ab TM |
937 | rpc_clear_running(task); |
938 | spin_unlock_bh(&queue->lock); | |
939 | if (task_is_async) | |
940 | return; | |
1da177e4 LT |
941 | |
942 | /* sync task: sleep here */ | |
46121cf7 | 943 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 944 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
945 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
946 | TASK_KILLABLE); | |
ae67bd38 | 947 | if (status < 0) { |
1da177e4 LT |
948 | /* |
949 | * When a sync task receives a signal, it exits with | |
950 | * -ERESTARTSYS. In order to catch any callbacks that | |
951 | * clean up after sleeping on some queue, we don't | |
952 | * break the loop here, but go around once more. | |
953 | */ | |
46121cf7 | 954 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
ae67bd38 | 955 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); |
96651ab3 | 956 | rpc_exit(task, -ERESTARTSYS); |
1da177e4 | 957 | } |
46121cf7 | 958 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
959 | } |
960 | ||
46121cf7 CL |
961 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
962 | task->tk_status); | |
1da177e4 LT |
963 | /* Release all resources associated with the task */ |
964 | rpc_release_task(task); | |
1da177e4 LT |
965 | } |
966 | ||
967 | /* | |
968 | * User-visible entry point to the scheduler. | |
969 | * | |
970 | * This may be called recursively if e.g. an async NFS task updates | |
971 | * the attributes and finds that dirty pages must be flushed. | |
972 | * NOTE: Upon exit of this function the task is guaranteed to be | |
973 | * released. In particular note that tk_release() will have | |
974 | * been called, so your task memory may have been freed. | |
975 | */ | |
2efef837 | 976 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 977 | { |
a76580fb TM |
978 | bool is_async = RPC_IS_ASYNC(task); |
979 | ||
44c28873 | 980 | rpc_set_active(task); |
f1dc237c | 981 | rpc_make_runnable(rpciod_workqueue, task); |
a76580fb | 982 | if (!is_async) |
d6a1ed08 | 983 | __rpc_execute(task); |
1da177e4 LT |
984 | } |
985 | ||
65f27f38 | 986 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 987 | { |
a1231fda TM |
988 | unsigned int pflags = memalloc_nofs_save(); |
989 | ||
65f27f38 | 990 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 991 | memalloc_nofs_restore(pflags); |
1da177e4 LT |
992 | } |
993 | ||
02107148 | 994 | /** |
5fe6eaa1 CL |
995 | * rpc_malloc - allocate RPC buffer resources |
996 | * @task: RPC task | |
997 | * | |
998 | * A single memory region is allocated, which is split between the | |
999 | * RPC call and RPC reply that this task is being used for. When | |
1000 | * this RPC is retired, the memory is released by calling rpc_free. | |
1da177e4 | 1001 | * |
c5a4dd8b | 1002 | * To prevent rpciod from hanging, this allocator never sleeps, |
5fe6eaa1 CL |
1003 | * returning -ENOMEM and suppressing warning if the request cannot |
1004 | * be serviced immediately. The caller can arrange to sleep in a | |
1005 | * way that is safe for rpciod. | |
c5a4dd8b CL |
1006 | * |
1007 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
1008 | * mempool, ensuring that NFS reads and writes can always proceed, | |
1009 | * and that there is good locality of reference for these buffers. | |
1da177e4 | 1010 | */ |
5fe6eaa1 | 1011 | int rpc_malloc(struct rpc_task *task) |
1da177e4 | 1012 | { |
5fe6eaa1 CL |
1013 | struct rpc_rqst *rqst = task->tk_rqstp; |
1014 | size_t size = rqst->rq_callsize + rqst->rq_rcvsize; | |
aa3d1fae | 1015 | struct rpc_buffer *buf; |
12a3ad61 | 1016 | gfp_t gfp = GFP_NOFS; |
a564b8f0 MG |
1017 | |
1018 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 1019 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 1020 | |
aa3d1fae | 1021 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
1022 | if (size <= RPC_BUFFER_MAXSIZE) |
1023 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 1024 | else |
c5a4dd8b | 1025 | buf = kmalloc(size, gfp); |
ddce40df PZ |
1026 | |
1027 | if (!buf) | |
5fe6eaa1 | 1028 | return -ENOMEM; |
ddce40df | 1029 | |
aa3d1fae | 1030 | buf->len = size; |
215d0678 | 1031 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 1032 | task->tk_pid, size, buf); |
5fe6eaa1 | 1033 | rqst->rq_buffer = buf->data; |
68778945 | 1034 | rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; |
5fe6eaa1 | 1035 | return 0; |
1da177e4 | 1036 | } |
12444809 | 1037 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 1038 | |
02107148 | 1039 | /** |
3435c74a CL |
1040 | * rpc_free - free RPC buffer resources allocated via rpc_malloc |
1041 | * @task: RPC task | |
02107148 CL |
1042 | * |
1043 | */ | |
3435c74a | 1044 | void rpc_free(struct rpc_task *task) |
1da177e4 | 1045 | { |
3435c74a | 1046 | void *buffer = task->tk_rqstp->rq_buffer; |
aa3d1fae CL |
1047 | size_t size; |
1048 | struct rpc_buffer *buf; | |
02107148 | 1049 | |
aa3d1fae CL |
1050 | buf = container_of(buffer, struct rpc_buffer, data); |
1051 | size = buf->len; | |
c5a4dd8b | 1052 | |
215d0678 | 1053 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 1054 | size, buf); |
aa3d1fae | 1055 | |
c5a4dd8b CL |
1056 | if (size <= RPC_BUFFER_MAXSIZE) |
1057 | mempool_free(buf, rpc_buffer_mempool); | |
1058 | else | |
1059 | kfree(buf); | |
1da177e4 | 1060 | } |
12444809 | 1061 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
1062 | |
1063 | /* | |
1064 | * Creation and deletion of RPC task structures | |
1065 | */ | |
47fe0648 | 1066 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
1067 | { |
1068 | memset(task, 0, sizeof(*task)); | |
44c28873 | 1069 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
1070 | task->tk_flags = task_setup_data->flags; |
1071 | task->tk_ops = task_setup_data->callback_ops; | |
1072 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 1073 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 1074 | |
3ff7576d TM |
1075 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
1076 | task->tk_owner = current->tgid; | |
1da177e4 LT |
1077 | |
1078 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 1079 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 1080 | |
9d61498d TM |
1081 | task->tk_xprt = xprt_get(task_setup_data->rpc_xprt); |
1082 | ||
1de7eea9 N |
1083 | task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred); |
1084 | ||
84115e1c TM |
1085 | if (task->tk_ops->rpc_call_prepare != NULL) |
1086 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 1087 | |
7fdcf13b | 1088 | rpc_init_task_statistics(task); |
ef759a2e | 1089 | |
46121cf7 | 1090 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 1091 | task_pid_nr(current)); |
1da177e4 LT |
1092 | } |
1093 | ||
1094 | static struct rpc_task * | |
1095 | rpc_alloc_task(void) | |
1096 | { | |
12a3ad61 | 1097 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); |
1da177e4 LT |
1098 | } |
1099 | ||
1da177e4 | 1100 | /* |
90c5755f | 1101 | * Create a new task for the specified client. |
1da177e4 | 1102 | */ |
84115e1c | 1103 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 1104 | { |
e8f5d77c TM |
1105 | struct rpc_task *task = setup_data->task; |
1106 | unsigned short flags = 0; | |
1107 | ||
1108 | if (task == NULL) { | |
1109 | task = rpc_alloc_task(); | |
e8f5d77c TM |
1110 | flags = RPC_TASK_DYNAMIC; |
1111 | } | |
1da177e4 | 1112 | |
84115e1c | 1113 | rpc_init_task(task, setup_data); |
e8f5d77c | 1114 | task->tk_flags |= flags; |
46121cf7 | 1115 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 1116 | return task; |
1da177e4 LT |
1117 | } |
1118 | ||
c6567ed1 TM |
1119 | /* |
1120 | * rpc_free_task - release rpc task and perform cleanups | |
1121 | * | |
1122 | * Note that we free up the rpc_task _after_ rpc_release_calldata() | |
1123 | * in order to work around a workqueue dependency issue. | |
1124 | * | |
1125 | * Tejun Heo states: | |
1126 | * "Workqueue currently considers two work items to be the same if they're | |
1127 | * on the same address and won't execute them concurrently - ie. it | |
1128 | * makes a work item which is queued again while being executed wait | |
1129 | * for the previous execution to complete. | |
1130 | * | |
1131 | * If a work function frees the work item, and then waits for an event | |
1132 | * which should be performed by another work item and *that* work item | |
1133 | * recycles the freed work item, it can create a false dependency loop. | |
1134 | * There really is no reliable way to detect this short of verifying | |
1135 | * every memory free." | |
1136 | * | |
1137 | */ | |
32bfb5c0 | 1138 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 1139 | { |
c6567ed1 TM |
1140 | unsigned short tk_flags = task->tk_flags; |
1141 | ||
1de7eea9 | 1142 | put_rpccred(task->tk_op_cred); |
c6567ed1 | 1143 | rpc_release_calldata(task->tk_ops, task->tk_calldata); |
1da177e4 | 1144 | |
c6567ed1 | 1145 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
1146 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
1147 | mempool_free(task, rpc_task_mempool); | |
1148 | } | |
32bfb5c0 TM |
1149 | } |
1150 | ||
1151 | static void rpc_async_release(struct work_struct *work) | |
1152 | { | |
a1231fda TM |
1153 | unsigned int pflags = memalloc_nofs_save(); |
1154 | ||
32bfb5c0 | 1155 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1156 | memalloc_nofs_restore(pflags); |
32bfb5c0 TM |
1157 | } |
1158 | ||
bf294b41 | 1159 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 1160 | { |
87ed5003 | 1161 | xprt_release(task); |
a271c5a0 | 1162 | if (task->tk_msg.rpc_cred) { |
a52458b4 | 1163 | put_cred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
1164 | task->tk_msg.rpc_cred = NULL; |
1165 | } | |
58f9612c | 1166 | rpc_task_release_client(task); |
bf294b41 TM |
1167 | } |
1168 | ||
1169 | static void rpc_final_put_task(struct rpc_task *task, | |
1170 | struct workqueue_struct *q) | |
1171 | { | |
1172 | if (q != NULL) { | |
32bfb5c0 | 1173 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1174 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1175 | } else |
1176 | rpc_free_task(task); | |
e6b3c4db | 1177 | } |
bf294b41 TM |
1178 | |
1179 | static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q) | |
1180 | { | |
1181 | if (atomic_dec_and_test(&task->tk_count)) { | |
1182 | rpc_release_resources_task(task); | |
1183 | rpc_final_put_task(task, q); | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | void rpc_put_task(struct rpc_task *task) | |
1188 | { | |
1189 | rpc_do_put_task(task, NULL); | |
1190 | } | |
e8914c65 | 1191 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1192 | |
bf294b41 TM |
1193 | void rpc_put_task_async(struct rpc_task *task) |
1194 | { | |
1195 | rpc_do_put_task(task, task->tk_workqueue); | |
1196 | } | |
1197 | EXPORT_SYMBOL_GPL(rpc_put_task_async); | |
1198 | ||
bde8f00c | 1199 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1200 | { |
46121cf7 | 1201 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1202 | |
0a0c2a57 | 1203 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1204 | |
bf294b41 | 1205 | rpc_release_resources_task(task); |
e6b3c4db | 1206 | |
bf294b41 TM |
1207 | /* |
1208 | * Note: at this point we have been removed from rpc_clnt->cl_tasks, | |
1209 | * so it should be safe to use task->tk_count as a test for whether | |
1210 | * or not any other processes still hold references to our rpc_task. | |
1211 | */ | |
1212 | if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) { | |
1213 | /* Wake up anyone who may be waiting for task completion */ | |
1214 | if (!rpc_complete_task(task)) | |
1215 | return; | |
1216 | } else { | |
1217 | if (!atomic_dec_and_test(&task->tk_count)) | |
1218 | return; | |
1219 | } | |
1220 | rpc_final_put_task(task, task->tk_workqueue); | |
1da177e4 LT |
1221 | } |
1222 | ||
b247bbf1 TM |
1223 | int rpciod_up(void) |
1224 | { | |
1225 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
1226 | } | |
1227 | ||
1228 | void rpciod_down(void) | |
1229 | { | |
1230 | module_put(THIS_MODULE); | |
1231 | } | |
1232 | ||
1da177e4 | 1233 | /* |
b247bbf1 | 1234 | * Start up the rpciod workqueue. |
1da177e4 | 1235 | */ |
b247bbf1 | 1236 | static int rpciod_start(void) |
1da177e4 LT |
1237 | { |
1238 | struct workqueue_struct *wq; | |
ab418d70 | 1239 | |
1da177e4 LT |
1240 | /* |
1241 | * Create the rpciod thread and wait for it to start. | |
1242 | */ | |
ab418d70 | 1243 | dprintk("RPC: creating workqueue rpciod\n"); |
f515f86b | 1244 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); |
40a5f1b1 TM |
1245 | if (!wq) |
1246 | goto out_failed; | |
1da177e4 | 1247 | rpciod_workqueue = wq; |
40a5f1b1 | 1248 | /* Note: highpri because network receive is latency sensitive */ |
90ea9f1b | 1249 | wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0); |
40a5f1b1 TM |
1250 | if (!wq) |
1251 | goto free_rpciod; | |
1252 | xprtiod_workqueue = wq; | |
1253 | return 1; | |
1254 | free_rpciod: | |
1255 | wq = rpciod_workqueue; | |
1256 | rpciod_workqueue = NULL; | |
1257 | destroy_workqueue(wq); | |
1258 | out_failed: | |
1259 | return 0; | |
1da177e4 LT |
1260 | } |
1261 | ||
b247bbf1 | 1262 | static void rpciod_stop(void) |
1da177e4 | 1263 | { |
b247bbf1 | 1264 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1265 | |
b247bbf1 TM |
1266 | if (rpciod_workqueue == NULL) |
1267 | return; | |
ab418d70 | 1268 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1269 | |
b247bbf1 TM |
1270 | wq = rpciod_workqueue; |
1271 | rpciod_workqueue = NULL; | |
1272 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1273 | wq = xprtiod_workqueue; |
1274 | xprtiod_workqueue = NULL; | |
1275 | destroy_workqueue(wq); | |
1da177e4 LT |
1276 | } |
1277 | ||
1da177e4 LT |
1278 | void |
1279 | rpc_destroy_mempool(void) | |
1280 | { | |
b247bbf1 | 1281 | rpciod_stop(); |
17a9618e JL |
1282 | mempool_destroy(rpc_buffer_mempool); |
1283 | mempool_destroy(rpc_task_mempool); | |
1284 | kmem_cache_destroy(rpc_task_slabp); | |
1285 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 1286 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | int | |
1290 | rpc_init_mempool(void) | |
1291 | { | |
f6a1cc89 TM |
1292 | /* |
1293 | * The following is not strictly a mempool initialisation, | |
1294 | * but there is no harm in doing it here | |
1295 | */ | |
1296 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
1297 | if (!rpciod_start()) | |
1298 | goto err_nomem; | |
1299 | ||
1da177e4 LT |
1300 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1301 | sizeof(struct rpc_task), | |
1302 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1303 | NULL); |
1da177e4 LT |
1304 | if (!rpc_task_slabp) |
1305 | goto err_nomem; | |
1306 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
1307 | RPC_BUFFER_MAXSIZE, | |
1308 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1309 | NULL); |
1da177e4 LT |
1310 | if (!rpc_buffer_slabp) |
1311 | goto err_nomem; | |
93d2341c MD |
1312 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1313 | rpc_task_slabp); | |
1da177e4 LT |
1314 | if (!rpc_task_mempool) |
1315 | goto err_nomem; | |
93d2341c MD |
1316 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1317 | rpc_buffer_slabp); | |
1da177e4 LT |
1318 | if (!rpc_buffer_mempool) |
1319 | goto err_nomem; | |
1320 | return 0; | |
1321 | err_nomem: | |
1322 | rpc_destroy_mempool(); | |
1323 | return -ENOMEM; | |
1324 | } |