Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
cca5172a | 7 | * |
1da177e4 LT |
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> | |
1da177e4 | 19 | #include <linux/spinlock.h> |
4a3e2f71 | 20 | #include <linux/mutex.h> |
d310310c | 21 | #include <linux/freezer.h> |
a1231fda | 22 | #include <linux/sched/mm.h> |
1da177e4 LT |
23 | |
24 | #include <linux/sunrpc/clnt.h> | |
1da177e4 | 25 | |
6951867b BH |
26 | #include "sunrpc.h" |
27 | ||
f895b252 | 28 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
1da177e4 | 29 | #define RPCDBG_FACILITY RPCDBG_SCHED |
1da177e4 LT |
30 | #endif |
31 | ||
82b0a4c3 TM |
32 | #define CREATE_TRACE_POINTS |
33 | #include <trace/events/sunrpc.h> | |
34 | ||
1da177e4 LT |
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) | |
e18b890b CL |
41 | static struct kmem_cache *rpc_task_slabp __read_mostly; |
42 | static struct kmem_cache *rpc_buffer_slabp __read_mostly; | |
ba89966c ED |
43 | static mempool_t *rpc_task_mempool __read_mostly; |
44 | static mempool_t *rpc_buffer_mempool __read_mostly; | |
1da177e4 | 45 | |
65f27f38 | 46 | static void rpc_async_schedule(struct work_struct *); |
bde8f00c | 47 | static void rpc_release_task(struct rpc_task *task); |
ff861c4d | 48 | static void __rpc_queue_timer_fn(struct timer_list *t); |
1da177e4 | 49 | |
1da177e4 LT |
50 | /* |
51 | * RPC tasks sit here while waiting for conditions to improve. | |
52 | */ | |
a4a87499 | 53 | static struct rpc_wait_queue delay_queue; |
1da177e4 | 54 | |
1da177e4 LT |
55 | /* |
56 | * rpciod-related stuff | |
57 | */ | |
40a5f1b1 TM |
58 | struct workqueue_struct *rpciod_workqueue __read_mostly; |
59 | struct workqueue_struct *xprtiod_workqueue __read_mostly; | |
1da177e4 | 60 | |
5efd1876 TM |
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 | ||
1da177e4 LT |
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 | */ | |
5d00837b | 80 | static void |
eb276c0e | 81 | __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 82 | { |
6b2e6856 | 83 | if (list_empty(&task->u.tk_wait.timer_list)) |
36df9aae | 84 | return; |
46121cf7 | 85 | dprintk("RPC: %5u disabling timer\n", task->tk_pid); |
1da177e4 | 86 | task->tk_timeout = 0; |
36df9aae | 87 | list_del(&task->u.tk_wait.timer_list); |
eb276c0e TM |
88 | if (list_empty(&queue->timer_list.list)) |
89 | del_timer(&queue->timer_list.timer); | |
36df9aae TM |
90 | } |
91 | ||
92 | static void | |
93 | rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires) | |
94 | { | |
24a9d9a2 | 95 | timer_reduce(&queue->timer_list.timer, expires); |
1da177e4 LT |
96 | } |
97 | ||
1da177e4 LT |
98 | /* |
99 | * Set up a timer for the current task. | |
100 | */ | |
5d00837b | 101 | static void |
6b2e6856 TM |
102 | __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task, |
103 | unsigned long timeout) | |
1da177e4 | 104 | { |
55cc1d78 | 105 | dprintk("RPC: %5u setting alarm for %u ms\n", |
6b2e6856 | 106 | task->tk_pid, jiffies_to_msecs(timeout - jiffies)); |
1da177e4 | 107 | |
6b2e6856 | 108 | task->tk_timeout = timeout; |
24a9d9a2 | 109 | rpc_set_queue_timer(queue, timeout); |
eb276c0e | 110 | list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); |
1da177e4 LT |
111 | } |
112 | ||
c05eecf6 TM |
113 | static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) |
114 | { | |
edd2e36f | 115 | if (queue->priority != priority) { |
edd2e36f | 116 | queue->priority = priority; |
f42f7c28 | 117 | queue->nr = 1U << priority; |
edd2e36f | 118 | } |
c05eecf6 TM |
119 | } |
120 | ||
c05eecf6 TM |
121 | static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) |
122 | { | |
123 | rpc_set_waitqueue_priority(queue, queue->maxpriority); | |
c05eecf6 TM |
124 | } |
125 | ||
1da177e4 | 126 | /* |
f42f7c28 | 127 | * Add a request to a queue list |
1da177e4 | 128 | */ |
f42f7c28 TM |
129 | static void |
130 | __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task) | |
1da177e4 | 131 | { |
1da177e4 LT |
132 | struct rpc_task *t; |
133 | ||
1da177e4 | 134 | list_for_each_entry(t, q, u.tk_wait.list) { |
3ff7576d | 135 | if (t->tk_owner == task->tk_owner) { |
f42f7c28 TM |
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; | |
1da177e4 LT |
141 | return; |
142 | } | |
143 | } | |
f42f7c28 | 144 | INIT_LIST_HEAD(&task->u.tk_wait.links); |
1da177e4 LT |
145 | list_add_tail(&task->u.tk_wait.list, q); |
146 | } | |
147 | ||
f42f7c28 TM |
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 | ||
1da177e4 LT |
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 | */ | |
3b27bad7 TM |
193 | static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, |
194 | struct rpc_task *task, | |
195 | unsigned char queue_priority) | |
1da177e4 | 196 | { |
2bd4eef8 WAA |
197 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
198 | if (RPC_IS_QUEUED(task)) | |
199 | return; | |
1da177e4 | 200 | |
6b2e6856 | 201 | INIT_LIST_HEAD(&task->u.tk_wait.timer_list); |
1da177e4 | 202 | if (RPC_IS_PRIORITY(queue)) |
3b27bad7 | 203 | __rpc_add_wait_queue_priority(queue, task, queue_priority); |
1da177e4 LT |
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]); | |
96ef13b2 | 208 | task->tk_waitqueue = queue; |
e19b63da | 209 | queue->qlen++; |
1166fde6 TM |
210 | /* barrier matches the read in rpc_wake_up_task_queue_locked() */ |
211 | smp_wmb(); | |
1da177e4 LT |
212 | rpc_set_queued(task); |
213 | ||
46121cf7 CL |
214 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
215 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
216 | } |
217 | ||
218 | /* | |
219 | * Remove request from a priority queue. | |
220 | */ | |
221 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
222 | { | |
f42f7c28 | 223 | __rpc_list_dequeue_task(task); |
1da177e4 LT |
224 | } |
225 | ||
226 | /* | |
227 | * Remove request from queue. | |
228 | * Note: must be called with spin lock held. | |
229 | */ | |
96ef13b2 | 230 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 231 | { |
eb276c0e | 232 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
233 | if (RPC_IS_PRIORITY(queue)) |
234 | __rpc_remove_wait_queue_priority(task); | |
f42f7c28 TM |
235 | else |
236 | list_del(&task->u.tk_wait.list); | |
e19b63da | 237 | queue->qlen--; |
46121cf7 CL |
238 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
239 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
240 | } |
241 | ||
3ff7576d | 242 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
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]); | |
3ff7576d | 249 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 250 | rpc_reset_waitqueue_priority(queue); |
36df9aae | 251 | queue->qlen = 0; |
43123581 TM |
252 | timer_setup(&queue->timer_list.timer, |
253 | __rpc_queue_timer_fn, | |
254 | TIMER_DEFERRABLE); | |
36df9aae | 255 | INIT_LIST_HEAD(&queue->timer_list.list); |
2f09c242 | 256 | rpc_assign_waitqueue_name(queue, qname); |
1da177e4 LT |
257 | } |
258 | ||
259 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
260 | { | |
3ff7576d | 261 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 262 | } |
689cf5c1 | 263 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
264 | |
265 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
266 | { | |
3ff7576d | 267 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 268 | } |
e8914c65 | 269 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 270 | |
f6a1cc89 TM |
271 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
272 | { | |
36df9aae | 273 | del_timer_sync(&queue->timer_list.timer); |
f6a1cc89 TM |
274 | } |
275 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
276 | ||
dfd01f02 | 277 | static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode) |
44c28873 | 278 | { |
416ad3c9 | 279 | freezable_schedule_unsafe(); |
dfd01f02 PZ |
280 | if (signal_pending_state(mode, current)) |
281 | return -ERESTARTSYS; | |
44c28873 TM |
282 | return 0; |
283 | } | |
284 | ||
1306729b | 285 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS) |
c44fe705 TM |
286 | static void rpc_task_set_debuginfo(struct rpc_task *task) |
287 | { | |
288 | static atomic_t rpc_pid; | |
289 | ||
c44fe705 TM |
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 | ||
e6b3c4db TM |
298 | static void rpc_set_active(struct rpc_task *task) |
299 | { | |
c44fe705 | 300 | rpc_task_set_debuginfo(task); |
58f9612c | 301 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e671edb9 | 302 | trace_rpc_task_begin(task, NULL); |
e6b3c4db TM |
303 | } |
304 | ||
44c28873 TM |
305 | /* |
306 | * Mark an RPC call as having completed by clearing the 'active' bit | |
bf294b41 | 307 | * and then waking up all tasks that were sleeping. |
44c28873 | 308 | */ |
bf294b41 | 309 | static int rpc_complete_task(struct rpc_task *task) |
44c28873 | 310 | { |
bf294b41 TM |
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 | ||
e671edb9 | 317 | trace_rpc_task_complete(task, NULL); |
82b0a4c3 | 318 | |
bf294b41 | 319 | spin_lock_irqsave(&wq->lock, flags); |
e6b3c4db | 320 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
bf294b41 TM |
321 | ret = atomic_dec_and_test(&task->tk_count); |
322 | if (waitqueue_active(wq)) | |
ac5be6b4 | 323 | __wake_up_locked_key(wq, TASK_NORMAL, &k); |
bf294b41 TM |
324 | spin_unlock_irqrestore(&wq->lock, flags); |
325 | return ret; | |
44c28873 TM |
326 | } |
327 | ||
328 | /* | |
329 | * Allow callers to wait for completion of an RPC call | |
bf294b41 TM |
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(). | |
44c28873 | 334 | */ |
c1221321 | 335 | int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action) |
44c28873 TM |
336 | { |
337 | if (action == NULL) | |
150030b7 | 338 | action = rpc_wait_bit_killable; |
bf294b41 | 339 | return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 340 | action, TASK_KILLABLE); |
44c28873 | 341 | } |
e8914c65 | 342 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 343 | |
1da177e4 LT |
344 | /* |
345 | * Make an RPC task runnable. | |
346 | * | |
506026c3 JL |
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. | |
a3c3cac5 TM |
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. | |
1da177e4 | 354 | */ |
f1dc237c TM |
355 | static void rpc_make_runnable(struct workqueue_struct *wq, |
356 | struct rpc_task *task) | |
1da177e4 | 357 | { |
a3c3cac5 TM |
358 | bool need_wakeup = !rpc_test_and_set_running(task); |
359 | ||
1da177e4 | 360 | rpc_clear_queued(task); |
a3c3cac5 | 361 | if (!need_wakeup) |
cc4dc59e | 362 | return; |
1da177e4 | 363 | if (RPC_IS_ASYNC(task)) { |
65f27f38 | 364 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
f1dc237c | 365 | queue_work(wq, &task->u.tk_work); |
1da177e4 | 366 | } else |
96651ab3 | 367 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
368 | } |
369 | ||
1da177e4 LT |
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 | */ | |
3b27bad7 TM |
376 | static void __rpc_sleep_on_priority(struct rpc_wait_queue *q, |
377 | struct rpc_task *task, | |
3b27bad7 | 378 | unsigned char queue_priority) |
1da177e4 | 379 | { |
46121cf7 CL |
380 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
381 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 382 | |
e671edb9 | 383 | trace_rpc_task_sleep(task, q); |
82b0a4c3 | 384 | |
3b27bad7 | 385 | __rpc_add_wait_queue(q, task, queue_priority); |
1da177e4 | 386 | |
6b2e6856 TM |
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; | |
1da177e4 LT |
398 | } |
399 | ||
87150aae TM |
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) | |
1da177e4 | 407 | { |
58f9612c | 408 | /* We shouldn't ever put an inactive task to sleep */ |
87150aae | 409 | if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) { |
e454a7a8 WAA |
410 | task->tk_status = -EIO; |
411 | rpc_put_task_async(task); | |
87150aae | 412 | return false; |
e454a7a8 | 413 | } |
87150aae TM |
414 | return true; |
415 | } | |
416 | ||
6b2e6856 TM |
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 | ||
87150aae TM |
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); | |
e6b3c4db | 441 | |
6b2e6856 | 442 | WARN_ON_ONCE(task->tk_timeout != 0); |
1da177e4 LT |
443 | /* |
444 | * Protect the queue operations. | |
445 | */ | |
446 | spin_lock_bh(&q->lock); | |
87150aae | 447 | __rpc_sleep_on_priority(q, task, task->tk_priority); |
1da177e4 LT |
448 | spin_unlock_bh(&q->lock); |
449 | } | |
e8914c65 | 450 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 | 451 | |
6b2e6856 TM |
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 | ||
3b27bad7 | 468 | void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task, |
8357a9b6 | 469 | int priority) |
3b27bad7 | 470 | { |
87150aae | 471 | if (!rpc_sleep_check_activated(task)) |
e454a7a8 | 472 | return; |
87150aae | 473 | |
6b2e6856 | 474 | WARN_ON_ONCE(task->tk_timeout != 0); |
8357a9b6 | 475 | priority -= RPC_PRIORITY_LOW; |
3b27bad7 TM |
476 | /* |
477 | * Protect the queue operations. | |
478 | */ | |
479 | spin_lock_bh(&q->lock); | |
8357a9b6 | 480 | __rpc_sleep_on_priority(q, task, priority); |
3b27bad7 TM |
481 | spin_unlock_bh(&q->lock); |
482 | } | |
1e1093c7 | 483 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority); |
3b27bad7 | 484 | |
1da177e4 | 485 | /** |
f1dc237c TM |
486 | * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task |
487 | * @wq: workqueue on which to run task | |
96ef13b2 | 488 | * @queue: wait queue |
1da177e4 LT |
489 | * @task: task to be woken up |
490 | * | |
491 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
492 | */ | |
f1dc237c TM |
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) | |
1da177e4 | 496 | { |
46121cf7 CL |
497 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
498 | task->tk_pid, jiffies); | |
1da177e4 | 499 | |
1da177e4 LT |
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 | ||
e671edb9 | 506 | trace_rpc_task_wakeup(task, queue); |
82b0a4c3 | 507 | |
96ef13b2 | 508 | __rpc_remove_wait_queue(queue, task); |
1da177e4 | 509 | |
f1dc237c | 510 | rpc_make_runnable(wq, task); |
1da177e4 | 511 | |
46121cf7 | 512 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
513 | } |
514 | ||
515 | /* | |
96ef13b2 | 516 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 517 | */ |
359c48c0 TM |
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) | |
1da177e4 | 522 | { |
1166fde6 TM |
523 | if (RPC_IS_QUEUED(task)) { |
524 | smp_rmb(); | |
359c48c0 TM |
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 | } | |
1166fde6 | 531 | } |
359c48c0 TM |
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); | |
1da177e4 LT |
540 | } |
541 | ||
f1dc237c TM |
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 | ||
2275cde4 TM |
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 | { | |
5ce97039 TM |
557 | if (!RPC_IS_QUEUED(task)) |
558 | return; | |
2275cde4 TM |
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 | ||
1da177e4 | 564 | /* |
96ef13b2 | 565 | * Wake up a task on a specific queue |
1da177e4 | 566 | */ |
96ef13b2 | 567 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 568 | { |
5ce97039 TM |
569 | if (!RPC_IS_QUEUED(task)) |
570 | return; | |
5e4424af | 571 | spin_lock_bh(&queue->lock); |
96ef13b2 | 572 | rpc_wake_up_task_queue_locked(queue, task); |
5e4424af | 573 | spin_unlock_bh(&queue->lock); |
1da177e4 | 574 | } |
96ef13b2 TM |
575 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
576 | ||
359c48c0 TM |
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 | ||
1da177e4 LT |
611 | /* |
612 | * Wake up the next task on a priority queue. | |
613 | */ | |
961a828d | 614 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
615 | { |
616 | struct list_head *q; | |
617 | struct rpc_task *task; | |
618 | ||
619 | /* | |
3ff7576d | 620 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
621 | */ |
622 | q = &queue->tasks[queue->priority]; | |
f42f7c28 TM |
623 | if (!list_empty(q) && --queue->nr) { |
624 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); | |
625 | goto out; | |
1da177e4 LT |
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)) { | |
f42f7c28 | 637 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
1da177e4 LT |
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])); | |
1da177e4 | 647 | out: |
1da177e4 LT |
648 | return task; |
649 | } | |
650 | ||
961a828d TM |
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 | ||
1da177e4 | 660 | /* |
961a828d | 661 | * Wake up the first task on the wait queue. |
1da177e4 | 662 | */ |
f1dc237c TM |
663 | struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq, |
664 | struct rpc_wait_queue *queue, | |
961a828d | 665 | bool (*func)(struct rpc_task *, void *), void *data) |
1da177e4 LT |
666 | { |
667 | struct rpc_task *task = NULL; | |
668 | ||
961a828d | 669 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 670 | queue, rpc_qname(queue)); |
5e4424af | 671 | spin_lock_bh(&queue->lock); |
961a828d | 672 | task = __rpc_find_next_queued(queue); |
359c48c0 TM |
673 | if (task != NULL) |
674 | task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, | |
675 | task, func, data); | |
5e4424af | 676 | spin_unlock_bh(&queue->lock); |
1da177e4 LT |
677 | |
678 | return task; | |
679 | } | |
f1dc237c TM |
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 | } | |
961a828d TM |
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 | } | |
e8914c65 | 703 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
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 | { | |
1da177e4 | 713 | struct list_head *head; |
e6d83d55 | 714 | |
5e4424af | 715 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
716 | head = &queue->tasks[queue->maxpriority]; |
717 | for (;;) { | |
540a0f75 TM |
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); | |
96ef13b2 | 723 | rpc_wake_up_task_queue_locked(queue, task); |
540a0f75 | 724 | } |
1da177e4 LT |
725 | if (head == &queue->tasks[0]) |
726 | break; | |
727 | head--; | |
728 | } | |
5e4424af | 729 | spin_unlock_bh(&queue->lock); |
1da177e4 | 730 | } |
e8914c65 | 731 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
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; | |
1da177e4 | 743 | |
5e4424af | 744 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
745 | head = &queue->tasks[queue->maxpriority]; |
746 | for (;;) { | |
540a0f75 TM |
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); | |
1da177e4 | 752 | task->tk_status = status; |
96ef13b2 | 753 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
754 | } |
755 | if (head == &queue->tasks[0]) | |
756 | break; | |
757 | head--; | |
758 | } | |
5e4424af | 759 | spin_unlock_bh(&queue->lock); |
1da177e4 | 760 | } |
e8914c65 | 761 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 762 | |
ff861c4d | 763 | static void __rpc_queue_timer_fn(struct timer_list *t) |
36df9aae | 764 | { |
ff861c4d | 765 | struct rpc_wait_queue *queue = from_timer(queue, t, timer_list.timer); |
36df9aae TM |
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) { | |
6b2e6856 | 772 | timeo = task->tk_timeout; |
36df9aae | 773 | if (time_after_eq(now, timeo)) { |
36df9aae TM |
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 | ||
8014793b TM |
787 | static void __rpc_atrun(struct rpc_task *task) |
788 | { | |
6bd14416 TM |
789 | if (task->tk_status == -ETIMEDOUT) |
790 | task->tk_status = 0; | |
8014793b TM |
791 | } |
792 | ||
1da177e4 LT |
793 | /* |
794 | * Run a task at a later time | |
795 | */ | |
8014793b | 796 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 | 797 | { |
6b2e6856 | 798 | rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay); |
1da177e4 | 799 | } |
e8914c65 | 800 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 801 | |
4ce70ada TM |
802 | /* |
803 | * Helper to call task->tk_ops->rpc_call_prepare | |
804 | */ | |
aae2006e | 805 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
806 | { |
807 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
808 | } | |
809 | ||
7fdcf13b TM |
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; | |
ae67bd38 | 826 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT); |
7fdcf13b TM |
827 | rpc_init_task_statistics(task); |
828 | } | |
829 | ||
d05fdb0c | 830 | /* |
963d8fe5 | 831 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 832 | */ |
abbcf28f | 833 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 834 | { |
abbcf28f | 835 | task->tk_action = NULL; |
963d8fe5 TM |
836 | if (task->tk_ops->rpc_call_done != NULL) { |
837 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 838 | if (task->tk_action != NULL) { |
abbcf28f TM |
839 | /* Always release the RPC slot and buffer memory */ |
840 | xprt_release(task); | |
7fdcf13b | 841 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
842 | } |
843 | } | |
d05fdb0c | 844 | } |
d9b6cd94 | 845 | |
ae67bd38 TM |
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 | ||
d9b6cd94 TM |
859 | void rpc_exit(struct rpc_task *task, int status) |
860 | { | |
861 | task->tk_status = status; | |
862 | task->tk_action = rpc_exit_task; | |
6b5f5900 | 863 | rpc_wake_up_queued_task(task->tk_waitqueue, task); |
d9b6cd94 TM |
864 | } |
865 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 866 | |
bbd5a1f9 TM |
867 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
868 | { | |
a86dc496 | 869 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 870 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
871 | } |
872 | ||
1da177e4 LT |
873 | /* |
874 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
875 | */ | |
2efef837 | 876 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 877 | { |
eb9b55ab TM |
878 | struct rpc_wait_queue *queue; |
879 | int task_is_async = RPC_IS_ASYNC(task); | |
880 | int status = 0; | |
1da177e4 | 881 | |
46121cf7 CL |
882 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
883 | task->tk_pid, task->tk_flags); | |
1da177e4 | 884 | |
2bd4eef8 WAA |
885 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
886 | if (RPC_IS_QUEUED(task)) | |
887 | return; | |
1da177e4 | 888 | |
d05fdb0c | 889 | for (;;) { |
b55c5989 | 890 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
891 | |
892 | /* | |
21ead9ff CL |
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. | |
1da177e4 | 898 | */ |
21ead9ff CL |
899 | do_action = task->tk_action; |
900 | if (task->tk_callback) { | |
901 | do_action = task->tk_callback; | |
902 | task->tk_callback = NULL; | |
1da177e4 | 903 | } |
21ead9ff CL |
904 | if (!do_action) |
905 | break; | |
e671edb9 | 906 | trace_rpc_task_run_action(task, do_action); |
b55c5989 | 907 | do_action(task); |
1da177e4 LT |
908 | |
909 | /* | |
910 | * Lockless check for whether task is sleeping or not. | |
911 | */ | |
912 | if (!RPC_IS_QUEUED(task)) | |
913 | continue; | |
ae67bd38 TM |
914 | |
915 | /* | |
916 | * Signalled tasks should exit rather than sleep. | |
917 | */ | |
918 | if (RPC_SIGNALLED(task)) | |
919 | rpc_exit(task, -ERESTARTSYS); | |
920 | ||
eb9b55ab TM |
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); | |
1da177e4 LT |
934 | continue; |
935 | } | |
eb9b55ab TM |
936 | rpc_clear_running(task); |
937 | spin_unlock_bh(&queue->lock); | |
938 | if (task_is_async) | |
939 | return; | |
1da177e4 LT |
940 | |
941 | /* sync task: sleep here */ | |
46121cf7 | 942 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 943 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
944 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
945 | TASK_KILLABLE); | |
ae67bd38 | 946 | if (status < 0) { |
1da177e4 LT |
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 | */ | |
46121cf7 | 953 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
ae67bd38 | 954 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); |
96651ab3 | 955 | rpc_exit(task, -ERESTARTSYS); |
1da177e4 | 956 | } |
46121cf7 | 957 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
958 | } |
959 | ||
46121cf7 CL |
960 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
961 | task->tk_status); | |
1da177e4 LT |
962 | /* Release all resources associated with the task */ |
963 | rpc_release_task(task); | |
1da177e4 LT |
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 | */ | |
2efef837 | 975 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 976 | { |
a76580fb TM |
977 | bool is_async = RPC_IS_ASYNC(task); |
978 | ||
44c28873 | 979 | rpc_set_active(task); |
f1dc237c | 980 | rpc_make_runnable(rpciod_workqueue, task); |
a76580fb | 981 | if (!is_async) |
d6a1ed08 | 982 | __rpc_execute(task); |
1da177e4 LT |
983 | } |
984 | ||
65f27f38 | 985 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 986 | { |
a1231fda TM |
987 | unsigned int pflags = memalloc_nofs_save(); |
988 | ||
65f27f38 | 989 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 990 | memalloc_nofs_restore(pflags); |
1da177e4 LT |
991 | } |
992 | ||
02107148 | 993 | /** |
5fe6eaa1 CL |
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. | |
1da177e4 | 1000 | * |
c5a4dd8b | 1001 | * To prevent rpciod from hanging, this allocator never sleeps, |
5fe6eaa1 CL |
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. | |
c5a4dd8b CL |
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. | |
1da177e4 | 1009 | */ |
5fe6eaa1 | 1010 | int rpc_malloc(struct rpc_task *task) |
1da177e4 | 1011 | { |
5fe6eaa1 CL |
1012 | struct rpc_rqst *rqst = task->tk_rqstp; |
1013 | size_t size = rqst->rq_callsize + rqst->rq_rcvsize; | |
aa3d1fae | 1014 | struct rpc_buffer *buf; |
12a3ad61 | 1015 | gfp_t gfp = GFP_NOFS; |
a564b8f0 MG |
1016 | |
1017 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 1018 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 1019 | |
aa3d1fae | 1020 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
1021 | if (size <= RPC_BUFFER_MAXSIZE) |
1022 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 1023 | else |
c5a4dd8b | 1024 | buf = kmalloc(size, gfp); |
ddce40df PZ |
1025 | |
1026 | if (!buf) | |
5fe6eaa1 | 1027 | return -ENOMEM; |
ddce40df | 1028 | |
aa3d1fae | 1029 | buf->len = size; |
215d0678 | 1030 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 1031 | task->tk_pid, size, buf); |
5fe6eaa1 | 1032 | rqst->rq_buffer = buf->data; |
68778945 | 1033 | rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; |
5fe6eaa1 | 1034 | return 0; |
1da177e4 | 1035 | } |
12444809 | 1036 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 1037 | |
02107148 | 1038 | /** |
3435c74a CL |
1039 | * rpc_free - free RPC buffer resources allocated via rpc_malloc |
1040 | * @task: RPC task | |
02107148 CL |
1041 | * |
1042 | */ | |
3435c74a | 1043 | void rpc_free(struct rpc_task *task) |
1da177e4 | 1044 | { |
3435c74a | 1045 | void *buffer = task->tk_rqstp->rq_buffer; |
aa3d1fae CL |
1046 | size_t size; |
1047 | struct rpc_buffer *buf; | |
02107148 | 1048 | |
aa3d1fae CL |
1049 | buf = container_of(buffer, struct rpc_buffer, data); |
1050 | size = buf->len; | |
c5a4dd8b | 1051 | |
215d0678 | 1052 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 1053 | size, buf); |
aa3d1fae | 1054 | |
c5a4dd8b CL |
1055 | if (size <= RPC_BUFFER_MAXSIZE) |
1056 | mempool_free(buf, rpc_buffer_mempool); | |
1057 | else | |
1058 | kfree(buf); | |
1da177e4 | 1059 | } |
12444809 | 1060 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
1061 | |
1062 | /* | |
1063 | * Creation and deletion of RPC task structures | |
1064 | */ | |
47fe0648 | 1065 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
1066 | { |
1067 | memset(task, 0, sizeof(*task)); | |
44c28873 | 1068 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
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; | |
6529eba0 | 1072 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 1073 | |
3ff7576d TM |
1074 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
1075 | task->tk_owner = current->tgid; | |
1da177e4 LT |
1076 | |
1077 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 1078 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 1079 | |
9d61498d TM |
1080 | task->tk_xprt = xprt_get(task_setup_data->rpc_xprt); |
1081 | ||
1de7eea9 N |
1082 | task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred); |
1083 | ||
84115e1c TM |
1084 | if (task->tk_ops->rpc_call_prepare != NULL) |
1085 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 1086 | |
7fdcf13b | 1087 | rpc_init_task_statistics(task); |
ef759a2e | 1088 | |
46121cf7 | 1089 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 1090 | task_pid_nr(current)); |
1da177e4 LT |
1091 | } |
1092 | ||
1093 | static struct rpc_task * | |
1094 | rpc_alloc_task(void) | |
1095 | { | |
12a3ad61 | 1096 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); |
1da177e4 LT |
1097 | } |
1098 | ||
1da177e4 | 1099 | /* |
90c5755f | 1100 | * Create a new task for the specified client. |
1da177e4 | 1101 | */ |
84115e1c | 1102 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 1103 | { |
e8f5d77c TM |
1104 | struct rpc_task *task = setup_data->task; |
1105 | unsigned short flags = 0; | |
1106 | ||
1107 | if (task == NULL) { | |
1108 | task = rpc_alloc_task(); | |
e8f5d77c TM |
1109 | flags = RPC_TASK_DYNAMIC; |
1110 | } | |
1da177e4 | 1111 | |
84115e1c | 1112 | rpc_init_task(task, setup_data); |
e8f5d77c | 1113 | task->tk_flags |= flags; |
46121cf7 | 1114 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 1115 | return task; |
1da177e4 LT |
1116 | } |
1117 | ||
c6567ed1 TM |
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 | */ | |
32bfb5c0 | 1137 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 1138 | { |
c6567ed1 TM |
1139 | unsigned short tk_flags = task->tk_flags; |
1140 | ||
1de7eea9 | 1141 | put_rpccred(task->tk_op_cred); |
c6567ed1 | 1142 | rpc_release_calldata(task->tk_ops, task->tk_calldata); |
1da177e4 | 1143 | |
c6567ed1 | 1144 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
1145 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
1146 | mempool_free(task, rpc_task_mempool); | |
1147 | } | |
32bfb5c0 TM |
1148 | } |
1149 | ||
1150 | static void rpc_async_release(struct work_struct *work) | |
1151 | { | |
a1231fda TM |
1152 | unsigned int pflags = memalloc_nofs_save(); |
1153 | ||
32bfb5c0 | 1154 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1155 | memalloc_nofs_restore(pflags); |
32bfb5c0 TM |
1156 | } |
1157 | ||
bf294b41 | 1158 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 1159 | { |
87ed5003 | 1160 | xprt_release(task); |
a271c5a0 | 1161 | if (task->tk_msg.rpc_cred) { |
a52458b4 | 1162 | put_cred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
1163 | task->tk_msg.rpc_cred = NULL; |
1164 | } | |
58f9612c | 1165 | rpc_task_release_client(task); |
bf294b41 TM |
1166 | } |
1167 | ||
1168 | static void rpc_final_put_task(struct rpc_task *task, | |
1169 | struct workqueue_struct *q) | |
1170 | { | |
1171 | if (q != NULL) { | |
32bfb5c0 | 1172 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1173 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1174 | } else |
1175 | rpc_free_task(task); | |
e6b3c4db | 1176 | } |
bf294b41 TM |
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 | } | |
e8914c65 | 1190 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1191 | |
bf294b41 TM |
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 | ||
bde8f00c | 1198 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1199 | { |
46121cf7 | 1200 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1201 | |
0a0c2a57 | 1202 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1203 | |
bf294b41 | 1204 | rpc_release_resources_task(task); |
e6b3c4db | 1205 | |
bf294b41 TM |
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); | |
1da177e4 LT |
1220 | } |
1221 | ||
b247bbf1 TM |
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 | ||
1da177e4 | 1232 | /* |
b247bbf1 | 1233 | * Start up the rpciod workqueue. |
1da177e4 | 1234 | */ |
b247bbf1 | 1235 | static int rpciod_start(void) |
1da177e4 LT |
1236 | { |
1237 | struct workqueue_struct *wq; | |
ab418d70 | 1238 | |
1da177e4 LT |
1239 | /* |
1240 | * Create the rpciod thread and wait for it to start. | |
1241 | */ | |
ab418d70 | 1242 | dprintk("RPC: creating workqueue rpciod\n"); |
f515f86b | 1243 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); |
40a5f1b1 TM |
1244 | if (!wq) |
1245 | goto out_failed; | |
1da177e4 | 1246 | rpciod_workqueue = wq; |
40a5f1b1 | 1247 | /* Note: highpri because network receive is latency sensitive */ |
90ea9f1b | 1248 | wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0); |
40a5f1b1 TM |
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; | |
1da177e4 LT |
1259 | } |
1260 | ||
b247bbf1 | 1261 | static void rpciod_stop(void) |
1da177e4 | 1262 | { |
b247bbf1 | 1263 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1264 | |
b247bbf1 TM |
1265 | if (rpciod_workqueue == NULL) |
1266 | return; | |
ab418d70 | 1267 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1268 | |
b247bbf1 TM |
1269 | wq = rpciod_workqueue; |
1270 | rpciod_workqueue = NULL; | |
1271 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1272 | wq = xprtiod_workqueue; |
1273 | xprtiod_workqueue = NULL; | |
1274 | destroy_workqueue(wq); | |
1da177e4 LT |
1275 | } |
1276 | ||
1da177e4 LT |
1277 | void |
1278 | rpc_destroy_mempool(void) | |
1279 | { | |
b247bbf1 | 1280 | rpciod_stop(); |
17a9618e JL |
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); | |
f6a1cc89 | 1285 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1286 | } |
1287 | ||
1288 | int | |
1289 | rpc_init_mempool(void) | |
1290 | { | |
f6a1cc89 TM |
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 | ||
1da177e4 LT |
1299 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1300 | sizeof(struct rpc_task), | |
1301 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1302 | NULL); |
1da177e4 LT |
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, | |
20c2df83 | 1308 | NULL); |
1da177e4 LT |
1309 | if (!rpc_buffer_slabp) |
1310 | goto err_nomem; | |
93d2341c MD |
1311 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1312 | rpc_task_slabp); | |
1da177e4 LT |
1313 | if (!rpc_task_mempool) |
1314 | goto err_nomem; | |
93d2341c MD |
1315 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1316 | rpc_buffer_slabp); | |
1da177e4 LT |
1317 | if (!rpc_buffer_mempool) |
1318 | goto err_nomem; | |
1319 | return 0; | |
1320 | err_nomem: | |
1321 | rpc_destroy_mempool(); | |
1322 | return -ENOMEM; | |
1323 | } |