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> | |
9dfe52a9 | 26 | #include <linux/sunrpc/metrics.h> |
1da177e4 | 27 | |
6951867b BH |
28 | #include "sunrpc.h" |
29 | ||
f895b252 | 30 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
1da177e4 | 31 | #define RPCDBG_FACILITY RPCDBG_SCHED |
1da177e4 LT |
32 | #endif |
33 | ||
82b0a4c3 TM |
34 | #define CREATE_TRACE_POINTS |
35 | #include <trace/events/sunrpc.h> | |
36 | ||
1da177e4 LT |
37 | /* |
38 | * RPC slabs and memory pools | |
39 | */ | |
40 | #define RPC_BUFFER_MAXSIZE (2048) | |
41 | #define RPC_BUFFER_POOLSIZE (8) | |
42 | #define RPC_TASK_POOLSIZE (8) | |
e18b890b CL |
43 | static struct kmem_cache *rpc_task_slabp __read_mostly; |
44 | static struct kmem_cache *rpc_buffer_slabp __read_mostly; | |
ba89966c ED |
45 | static mempool_t *rpc_task_mempool __read_mostly; |
46 | static mempool_t *rpc_buffer_mempool __read_mostly; | |
1da177e4 | 47 | |
65f27f38 | 48 | static void rpc_async_schedule(struct work_struct *); |
bde8f00c | 49 | static void rpc_release_task(struct rpc_task *task); |
7e0a0e38 | 50 | static void __rpc_queue_timer_fn(struct work_struct *); |
1da177e4 | 51 | |
1da177e4 LT |
52 | /* |
53 | * RPC tasks sit here while waiting for conditions to improve. | |
54 | */ | |
a4a87499 | 55 | static struct rpc_wait_queue delay_queue; |
1da177e4 | 56 | |
1da177e4 LT |
57 | /* |
58 | * rpciod-related stuff | |
59 | */ | |
40a5f1b1 TM |
60 | struct workqueue_struct *rpciod_workqueue __read_mostly; |
61 | struct workqueue_struct *xprtiod_workqueue __read_mostly; | |
675dd90a | 62 | EXPORT_SYMBOL_GPL(xprtiod_workqueue); |
1da177e4 | 63 | |
5efd1876 TM |
64 | unsigned long |
65 | rpc_task_timeout(const struct rpc_task *task) | |
66 | { | |
67 | unsigned long timeout = READ_ONCE(task->tk_timeout); | |
68 | ||
69 | if (timeout != 0) { | |
70 | unsigned long now = jiffies; | |
71 | if (time_before(now, timeout)) | |
72 | return timeout - now; | |
73 | } | |
74 | return 0; | |
75 | } | |
76 | EXPORT_SYMBOL_GPL(rpc_task_timeout); | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * Disable the timer for a given RPC task. Should be called with | |
80 | * queue->lock and bh_disabled in order to avoid races within | |
81 | * rpc_run_timer(). | |
82 | */ | |
5d00837b | 83 | static void |
eb276c0e | 84 | __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 85 | { |
6b2e6856 | 86 | if (list_empty(&task->u.tk_wait.timer_list)) |
36df9aae | 87 | return; |
46121cf7 | 88 | dprintk("RPC: %5u disabling timer\n", task->tk_pid); |
1da177e4 | 89 | task->tk_timeout = 0; |
36df9aae | 90 | list_del(&task->u.tk_wait.timer_list); |
eb276c0e | 91 | if (list_empty(&queue->timer_list.list)) |
7e0a0e38 | 92 | cancel_delayed_work(&queue->timer_list.dwork); |
36df9aae TM |
93 | } |
94 | ||
95 | static void | |
96 | rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires) | |
97 | { | |
7e0a0e38 TM |
98 | unsigned long now = jiffies; |
99 | queue->timer_list.expires = expires; | |
100 | if (time_before_eq(expires, now)) | |
101 | expires = 0; | |
102 | else | |
103 | expires -= now; | |
104 | mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires); | |
1da177e4 LT |
105 | } |
106 | ||
1da177e4 LT |
107 | /* |
108 | * Set up a timer for the current task. | |
109 | */ | |
5d00837b | 110 | static void |
6b2e6856 TM |
111 | __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task, |
112 | unsigned long timeout) | |
1da177e4 | 113 | { |
55cc1d78 | 114 | dprintk("RPC: %5u setting alarm for %u ms\n", |
6b2e6856 | 115 | task->tk_pid, jiffies_to_msecs(timeout - jiffies)); |
1da177e4 | 116 | |
6b2e6856 | 117 | task->tk_timeout = timeout; |
7e0a0e38 TM |
118 | if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires)) |
119 | rpc_set_queue_timer(queue, timeout); | |
eb276c0e | 120 | list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); |
1da177e4 LT |
121 | } |
122 | ||
c05eecf6 TM |
123 | static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) |
124 | { | |
edd2e36f | 125 | if (queue->priority != priority) { |
edd2e36f | 126 | queue->priority = priority; |
f42f7c28 | 127 | queue->nr = 1U << priority; |
edd2e36f | 128 | } |
c05eecf6 TM |
129 | } |
130 | ||
c05eecf6 TM |
131 | static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) |
132 | { | |
133 | rpc_set_waitqueue_priority(queue, queue->maxpriority); | |
c05eecf6 TM |
134 | } |
135 | ||
1da177e4 | 136 | /* |
f42f7c28 | 137 | * Add a request to a queue list |
1da177e4 | 138 | */ |
f42f7c28 TM |
139 | static void |
140 | __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task) | |
1da177e4 | 141 | { |
1da177e4 LT |
142 | struct rpc_task *t; |
143 | ||
1da177e4 | 144 | list_for_each_entry(t, q, u.tk_wait.list) { |
3ff7576d | 145 | if (t->tk_owner == task->tk_owner) { |
f42f7c28 TM |
146 | list_add_tail(&task->u.tk_wait.links, |
147 | &t->u.tk_wait.links); | |
148 | /* Cache the queue head in task->u.tk_wait.list */ | |
149 | task->u.tk_wait.list.next = q; | |
150 | task->u.tk_wait.list.prev = NULL; | |
1da177e4 LT |
151 | return; |
152 | } | |
153 | } | |
f42f7c28 | 154 | INIT_LIST_HEAD(&task->u.tk_wait.links); |
1da177e4 LT |
155 | list_add_tail(&task->u.tk_wait.list, q); |
156 | } | |
157 | ||
f42f7c28 TM |
158 | /* |
159 | * Remove request from a queue list | |
160 | */ | |
161 | static void | |
162 | __rpc_list_dequeue_task(struct rpc_task *task) | |
163 | { | |
164 | struct list_head *q; | |
165 | struct rpc_task *t; | |
166 | ||
167 | if (task->u.tk_wait.list.prev == NULL) { | |
168 | list_del(&task->u.tk_wait.links); | |
169 | return; | |
170 | } | |
171 | if (!list_empty(&task->u.tk_wait.links)) { | |
172 | t = list_first_entry(&task->u.tk_wait.links, | |
173 | struct rpc_task, | |
174 | u.tk_wait.links); | |
175 | /* Assume __rpc_list_enqueue_task() cached the queue head */ | |
176 | q = t->u.tk_wait.list.next; | |
177 | list_add_tail(&t->u.tk_wait.list, q); | |
178 | list_del(&task->u.tk_wait.links); | |
179 | } | |
180 | list_del(&task->u.tk_wait.list); | |
181 | } | |
182 | ||
183 | /* | |
184 | * Add new request to a priority queue. | |
185 | */ | |
186 | static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, | |
187 | struct rpc_task *task, | |
188 | unsigned char queue_priority) | |
189 | { | |
190 | if (unlikely(queue_priority > queue->maxpriority)) | |
191 | queue_priority = queue->maxpriority; | |
192 | __rpc_list_enqueue_task(&queue->tasks[queue_priority], task); | |
193 | } | |
194 | ||
1da177e4 LT |
195 | /* |
196 | * Add new request to wait queue. | |
197 | * | |
198 | * Swapper tasks always get inserted at the head of the queue. | |
199 | * This should avoid many nasty memory deadlocks and hopefully | |
200 | * improve overall performance. | |
201 | * Everyone else gets appended to the queue to ensure proper FIFO behavior. | |
202 | */ | |
3b27bad7 TM |
203 | static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, |
204 | struct rpc_task *task, | |
205 | unsigned char queue_priority) | |
1da177e4 | 206 | { |
2bd4eef8 WAA |
207 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
208 | if (RPC_IS_QUEUED(task)) | |
209 | return; | |
1da177e4 | 210 | |
6b2e6856 | 211 | INIT_LIST_HEAD(&task->u.tk_wait.timer_list); |
1da177e4 | 212 | if (RPC_IS_PRIORITY(queue)) |
3b27bad7 | 213 | __rpc_add_wait_queue_priority(queue, task, queue_priority); |
1da177e4 LT |
214 | else if (RPC_IS_SWAPPER(task)) |
215 | list_add(&task->u.tk_wait.list, &queue->tasks[0]); | |
216 | else | |
217 | list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]); | |
96ef13b2 | 218 | task->tk_waitqueue = queue; |
e19b63da | 219 | queue->qlen++; |
1166fde6 TM |
220 | /* barrier matches the read in rpc_wake_up_task_queue_locked() */ |
221 | smp_wmb(); | |
1da177e4 LT |
222 | rpc_set_queued(task); |
223 | ||
46121cf7 CL |
224 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
225 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
226 | } |
227 | ||
228 | /* | |
229 | * Remove request from a priority queue. | |
230 | */ | |
231 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
232 | { | |
f42f7c28 | 233 | __rpc_list_dequeue_task(task); |
1da177e4 LT |
234 | } |
235 | ||
236 | /* | |
237 | * Remove request from queue. | |
238 | * Note: must be called with spin lock held. | |
239 | */ | |
96ef13b2 | 240 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 241 | { |
eb276c0e | 242 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
243 | if (RPC_IS_PRIORITY(queue)) |
244 | __rpc_remove_wait_queue_priority(task); | |
f42f7c28 TM |
245 | else |
246 | list_del(&task->u.tk_wait.list); | |
e19b63da | 247 | queue->qlen--; |
46121cf7 CL |
248 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
249 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
250 | } |
251 | ||
3ff7576d | 252 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
253 | { |
254 | int i; | |
255 | ||
256 | spin_lock_init(&queue->lock); | |
257 | for (i = 0; i < ARRAY_SIZE(queue->tasks); i++) | |
258 | INIT_LIST_HEAD(&queue->tasks[i]); | |
3ff7576d | 259 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 260 | rpc_reset_waitqueue_priority(queue); |
36df9aae | 261 | queue->qlen = 0; |
7e0a0e38 TM |
262 | queue->timer_list.expires = 0; |
263 | INIT_DEFERRABLE_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn); | |
36df9aae | 264 | INIT_LIST_HEAD(&queue->timer_list.list); |
2f09c242 | 265 | rpc_assign_waitqueue_name(queue, qname); |
1da177e4 LT |
266 | } |
267 | ||
268 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
269 | { | |
3ff7576d | 270 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 271 | } |
689cf5c1 | 272 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
273 | |
274 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
275 | { | |
3ff7576d | 276 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 277 | } |
e8914c65 | 278 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 279 | |
f6a1cc89 TM |
280 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
281 | { | |
7e0a0e38 | 282 | cancel_delayed_work_sync(&queue->timer_list.dwork); |
f6a1cc89 TM |
283 | } |
284 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
285 | ||
dfd01f02 | 286 | static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode) |
44c28873 | 287 | { |
416ad3c9 | 288 | freezable_schedule_unsafe(); |
dfd01f02 PZ |
289 | if (signal_pending_state(mode, current)) |
290 | return -ERESTARTSYS; | |
44c28873 TM |
291 | return 0; |
292 | } | |
293 | ||
1306729b | 294 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS) |
c44fe705 TM |
295 | static void rpc_task_set_debuginfo(struct rpc_task *task) |
296 | { | |
297 | static atomic_t rpc_pid; | |
298 | ||
c44fe705 TM |
299 | task->tk_pid = atomic_inc_return(&rpc_pid); |
300 | } | |
301 | #else | |
302 | static inline void rpc_task_set_debuginfo(struct rpc_task *task) | |
303 | { | |
304 | } | |
305 | #endif | |
306 | ||
e6b3c4db TM |
307 | static void rpc_set_active(struct rpc_task *task) |
308 | { | |
c44fe705 | 309 | rpc_task_set_debuginfo(task); |
58f9612c | 310 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e671edb9 | 311 | trace_rpc_task_begin(task, NULL); |
e6b3c4db TM |
312 | } |
313 | ||
44c28873 TM |
314 | /* |
315 | * Mark an RPC call as having completed by clearing the 'active' bit | |
bf294b41 | 316 | * and then waking up all tasks that were sleeping. |
44c28873 | 317 | */ |
bf294b41 | 318 | static int rpc_complete_task(struct rpc_task *task) |
44c28873 | 319 | { |
bf294b41 TM |
320 | void *m = &task->tk_runstate; |
321 | wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE); | |
322 | struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE); | |
323 | unsigned long flags; | |
324 | int ret; | |
325 | ||
e671edb9 | 326 | trace_rpc_task_complete(task, NULL); |
82b0a4c3 | 327 | |
bf294b41 | 328 | spin_lock_irqsave(&wq->lock, flags); |
e6b3c4db | 329 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
bf294b41 TM |
330 | ret = atomic_dec_and_test(&task->tk_count); |
331 | if (waitqueue_active(wq)) | |
ac5be6b4 | 332 | __wake_up_locked_key(wq, TASK_NORMAL, &k); |
bf294b41 TM |
333 | spin_unlock_irqrestore(&wq->lock, flags); |
334 | return ret; | |
44c28873 TM |
335 | } |
336 | ||
337 | /* | |
338 | * Allow callers to wait for completion of an RPC call | |
bf294b41 TM |
339 | * |
340 | * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit() | |
341 | * to enforce taking of the wq->lock and hence avoid races with | |
342 | * rpc_complete_task(). | |
44c28873 | 343 | */ |
c1221321 | 344 | int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action) |
44c28873 TM |
345 | { |
346 | if (action == NULL) | |
150030b7 | 347 | action = rpc_wait_bit_killable; |
bf294b41 | 348 | return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 349 | action, TASK_KILLABLE); |
44c28873 | 350 | } |
e8914c65 | 351 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 352 | |
1da177e4 LT |
353 | /* |
354 | * Make an RPC task runnable. | |
355 | * | |
506026c3 JL |
356 | * Note: If the task is ASYNC, and is being made runnable after sitting on an |
357 | * rpc_wait_queue, this must be called with the queue spinlock held to protect | |
358 | * the wait queue operation. | |
a3c3cac5 TM |
359 | * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(), |
360 | * which is needed to ensure that __rpc_execute() doesn't loop (due to the | |
361 | * lockless RPC_IS_QUEUED() test) before we've had a chance to test | |
362 | * the RPC_TASK_RUNNING flag. | |
1da177e4 | 363 | */ |
f1dc237c TM |
364 | static void rpc_make_runnable(struct workqueue_struct *wq, |
365 | struct rpc_task *task) | |
1da177e4 | 366 | { |
a3c3cac5 TM |
367 | bool need_wakeup = !rpc_test_and_set_running(task); |
368 | ||
1da177e4 | 369 | rpc_clear_queued(task); |
a3c3cac5 | 370 | if (!need_wakeup) |
cc4dc59e | 371 | return; |
1da177e4 | 372 | if (RPC_IS_ASYNC(task)) { |
65f27f38 | 373 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
f1dc237c | 374 | queue_work(wq, &task->u.tk_work); |
1da177e4 | 375 | } else |
96651ab3 | 376 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
377 | } |
378 | ||
1da177e4 LT |
379 | /* |
380 | * Prepare for sleeping on a wait queue. | |
381 | * By always appending tasks to the list we ensure FIFO behavior. | |
382 | * NB: An RPC task will only receive interrupt-driven events as long | |
383 | * as it's on a wait queue. | |
384 | */ | |
3b27bad7 TM |
385 | static void __rpc_sleep_on_priority(struct rpc_wait_queue *q, |
386 | struct rpc_task *task, | |
3b27bad7 | 387 | unsigned char queue_priority) |
1da177e4 | 388 | { |
46121cf7 CL |
389 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
390 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 391 | |
e671edb9 | 392 | trace_rpc_task_sleep(task, q); |
82b0a4c3 | 393 | |
3b27bad7 | 394 | __rpc_add_wait_queue(q, task, queue_priority); |
1da177e4 | 395 | |
6b2e6856 TM |
396 | } |
397 | ||
398 | static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, | |
399 | struct rpc_task *task, unsigned long timeout, | |
400 | unsigned char queue_priority) | |
401 | { | |
402 | if (time_is_after_jiffies(timeout)) { | |
403 | __rpc_sleep_on_priority(q, task, queue_priority); | |
404 | __rpc_add_timer(q, task, timeout); | |
405 | } else | |
406 | task->tk_status = -ETIMEDOUT; | |
1da177e4 LT |
407 | } |
408 | ||
87150aae TM |
409 | static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action) |
410 | { | |
411 | if (action && !WARN_ON_ONCE(task->tk_callback != NULL)) | |
412 | task->tk_callback = action; | |
413 | } | |
414 | ||
415 | static bool rpc_sleep_check_activated(struct rpc_task *task) | |
1da177e4 | 416 | { |
58f9612c | 417 | /* We shouldn't ever put an inactive task to sleep */ |
87150aae | 418 | if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) { |
e454a7a8 WAA |
419 | task->tk_status = -EIO; |
420 | rpc_put_task_async(task); | |
87150aae | 421 | return false; |
e454a7a8 | 422 | } |
87150aae TM |
423 | return true; |
424 | } | |
425 | ||
6b2e6856 TM |
426 | void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task, |
427 | rpc_action action, unsigned long timeout) | |
428 | { | |
429 | if (!rpc_sleep_check_activated(task)) | |
430 | return; | |
431 | ||
432 | rpc_set_tk_callback(task, action); | |
433 | ||
434 | /* | |
435 | * Protect the queue operations. | |
436 | */ | |
c049f8ea | 437 | spin_lock(&q->lock); |
6b2e6856 | 438 | __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority); |
c049f8ea | 439 | spin_unlock(&q->lock); |
6b2e6856 TM |
440 | } |
441 | EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout); | |
442 | ||
87150aae TM |
443 | void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, |
444 | rpc_action action) | |
445 | { | |
446 | if (!rpc_sleep_check_activated(task)) | |
447 | return; | |
448 | ||
449 | rpc_set_tk_callback(task, action); | |
e6b3c4db | 450 | |
6b2e6856 | 451 | WARN_ON_ONCE(task->tk_timeout != 0); |
1da177e4 LT |
452 | /* |
453 | * Protect the queue operations. | |
454 | */ | |
c049f8ea | 455 | spin_lock(&q->lock); |
87150aae | 456 | __rpc_sleep_on_priority(q, task, task->tk_priority); |
c049f8ea | 457 | spin_unlock(&q->lock); |
1da177e4 | 458 | } |
e8914c65 | 459 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 | 460 | |
6b2e6856 TM |
461 | void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, |
462 | struct rpc_task *task, unsigned long timeout, int priority) | |
463 | { | |
464 | if (!rpc_sleep_check_activated(task)) | |
465 | return; | |
466 | ||
467 | priority -= RPC_PRIORITY_LOW; | |
468 | /* | |
469 | * Protect the queue operations. | |
470 | */ | |
c049f8ea | 471 | spin_lock(&q->lock); |
6b2e6856 | 472 | __rpc_sleep_on_priority_timeout(q, task, timeout, priority); |
c049f8ea | 473 | spin_unlock(&q->lock); |
6b2e6856 TM |
474 | } |
475 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout); | |
476 | ||
3b27bad7 | 477 | void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task, |
8357a9b6 | 478 | int priority) |
3b27bad7 | 479 | { |
87150aae | 480 | if (!rpc_sleep_check_activated(task)) |
e454a7a8 | 481 | return; |
87150aae | 482 | |
6b2e6856 | 483 | WARN_ON_ONCE(task->tk_timeout != 0); |
8357a9b6 | 484 | priority -= RPC_PRIORITY_LOW; |
3b27bad7 TM |
485 | /* |
486 | * Protect the queue operations. | |
487 | */ | |
c049f8ea | 488 | spin_lock(&q->lock); |
8357a9b6 | 489 | __rpc_sleep_on_priority(q, task, priority); |
c049f8ea | 490 | spin_unlock(&q->lock); |
3b27bad7 | 491 | } |
1e1093c7 | 492 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority); |
3b27bad7 | 493 | |
1da177e4 | 494 | /** |
f1dc237c TM |
495 | * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task |
496 | * @wq: workqueue on which to run task | |
96ef13b2 | 497 | * @queue: wait queue |
1da177e4 LT |
498 | * @task: task to be woken up |
499 | * | |
500 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
501 | */ | |
f1dc237c TM |
502 | static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq, |
503 | struct rpc_wait_queue *queue, | |
504 | struct rpc_task *task) | |
1da177e4 | 505 | { |
46121cf7 CL |
506 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
507 | task->tk_pid, jiffies); | |
1da177e4 | 508 | |
1da177e4 LT |
509 | /* Has the task been executed yet? If not, we cannot wake it up! */ |
510 | if (!RPC_IS_ACTIVATED(task)) { | |
511 | printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task); | |
512 | return; | |
513 | } | |
514 | ||
e671edb9 | 515 | trace_rpc_task_wakeup(task, queue); |
82b0a4c3 | 516 | |
96ef13b2 | 517 | __rpc_remove_wait_queue(queue, task); |
1da177e4 | 518 | |
f1dc237c | 519 | rpc_make_runnable(wq, task); |
1da177e4 | 520 | |
46121cf7 | 521 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
522 | } |
523 | ||
524 | /* | |
96ef13b2 | 525 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 526 | */ |
359c48c0 TM |
527 | static struct rpc_task * |
528 | rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq, | |
529 | struct rpc_wait_queue *queue, struct rpc_task *task, | |
530 | bool (*action)(struct rpc_task *, void *), void *data) | |
1da177e4 | 531 | { |
1166fde6 TM |
532 | if (RPC_IS_QUEUED(task)) { |
533 | smp_rmb(); | |
359c48c0 TM |
534 | if (task->tk_waitqueue == queue) { |
535 | if (action == NULL || action(task, data)) { | |
536 | __rpc_do_wake_up_task_on_wq(wq, queue, task); | |
537 | return task; | |
538 | } | |
539 | } | |
1166fde6 | 540 | } |
359c48c0 TM |
541 | return NULL; |
542 | } | |
543 | ||
544 | static void | |
545 | rpc_wake_up_task_on_wq_queue_locked(struct workqueue_struct *wq, | |
546 | struct rpc_wait_queue *queue, struct rpc_task *task) | |
547 | { | |
548 | rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, task, NULL, NULL); | |
1da177e4 LT |
549 | } |
550 | ||
f1dc237c TM |
551 | /* |
552 | * Wake up a queued task while the queue lock is being held | |
553 | */ | |
554 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task) | |
555 | { | |
556 | rpc_wake_up_task_on_wq_queue_locked(rpciod_workqueue, queue, task); | |
557 | } | |
558 | ||
2275cde4 TM |
559 | /* |
560 | * Wake up a task on a specific queue | |
561 | */ | |
562 | void rpc_wake_up_queued_task_on_wq(struct workqueue_struct *wq, | |
563 | struct rpc_wait_queue *queue, | |
564 | struct rpc_task *task) | |
565 | { | |
5ce97039 TM |
566 | if (!RPC_IS_QUEUED(task)) |
567 | return; | |
c049f8ea | 568 | spin_lock(&queue->lock); |
2275cde4 | 569 | rpc_wake_up_task_on_wq_queue_locked(wq, queue, task); |
c049f8ea | 570 | spin_unlock(&queue->lock); |
2275cde4 TM |
571 | } |
572 | ||
1da177e4 | 573 | /* |
96ef13b2 | 574 | * Wake up a task on a specific queue |
1da177e4 | 575 | */ |
96ef13b2 | 576 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 577 | { |
5ce97039 TM |
578 | if (!RPC_IS_QUEUED(task)) |
579 | return; | |
c049f8ea | 580 | spin_lock(&queue->lock); |
96ef13b2 | 581 | rpc_wake_up_task_queue_locked(queue, task); |
c049f8ea | 582 | spin_unlock(&queue->lock); |
1da177e4 | 583 | } |
96ef13b2 TM |
584 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
585 | ||
359c48c0 TM |
586 | static bool rpc_task_action_set_status(struct rpc_task *task, void *status) |
587 | { | |
588 | task->tk_status = *(int *)status; | |
589 | return true; | |
590 | } | |
591 | ||
592 | static void | |
593 | rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue, | |
594 | struct rpc_task *task, int status) | |
595 | { | |
596 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, | |
597 | task, rpc_task_action_set_status, &status); | |
598 | } | |
599 | ||
600 | /** | |
601 | * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status | |
602 | * @queue: pointer to rpc_wait_queue | |
603 | * @task: pointer to rpc_task | |
604 | * @status: integer error value | |
605 | * | |
606 | * If @task is queued on @queue, then it is woken up, and @task->tk_status is | |
607 | * set to the value of @status. | |
608 | */ | |
609 | void | |
610 | rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue, | |
611 | struct rpc_task *task, int status) | |
612 | { | |
613 | if (!RPC_IS_QUEUED(task)) | |
614 | return; | |
c049f8ea | 615 | spin_lock(&queue->lock); |
359c48c0 | 616 | rpc_wake_up_task_queue_set_status_locked(queue, task, status); |
c049f8ea | 617 | spin_unlock(&queue->lock); |
359c48c0 TM |
618 | } |
619 | ||
1da177e4 LT |
620 | /* |
621 | * Wake up the next task on a priority queue. | |
622 | */ | |
961a828d | 623 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
624 | { |
625 | struct list_head *q; | |
626 | struct rpc_task *task; | |
627 | ||
628 | /* | |
3ff7576d | 629 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
630 | */ |
631 | q = &queue->tasks[queue->priority]; | |
f42f7c28 TM |
632 | if (!list_empty(q) && --queue->nr) { |
633 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); | |
634 | goto out; | |
1da177e4 LT |
635 | } |
636 | ||
637 | /* | |
638 | * Service the next queue. | |
639 | */ | |
640 | do { | |
641 | if (q == &queue->tasks[0]) | |
642 | q = &queue->tasks[queue->maxpriority]; | |
643 | else | |
644 | q = q - 1; | |
645 | if (!list_empty(q)) { | |
f42f7c28 | 646 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
1da177e4 LT |
647 | goto new_queue; |
648 | } | |
649 | } while (q != &queue->tasks[queue->priority]); | |
650 | ||
651 | rpc_reset_waitqueue_priority(queue); | |
652 | return NULL; | |
653 | ||
654 | new_queue: | |
655 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
1da177e4 | 656 | out: |
1da177e4 LT |
657 | return task; |
658 | } | |
659 | ||
961a828d TM |
660 | static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue) |
661 | { | |
662 | if (RPC_IS_PRIORITY(queue)) | |
663 | return __rpc_find_next_queued_priority(queue); | |
664 | if (!list_empty(&queue->tasks[0])) | |
665 | return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list); | |
666 | return NULL; | |
667 | } | |
668 | ||
1da177e4 | 669 | /* |
961a828d | 670 | * Wake up the first task on the wait queue. |
1da177e4 | 671 | */ |
f1dc237c TM |
672 | struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq, |
673 | struct rpc_wait_queue *queue, | |
961a828d | 674 | bool (*func)(struct rpc_task *, void *), void *data) |
1da177e4 LT |
675 | { |
676 | struct rpc_task *task = NULL; | |
677 | ||
961a828d | 678 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 679 | queue, rpc_qname(queue)); |
c049f8ea | 680 | spin_lock(&queue->lock); |
961a828d | 681 | task = __rpc_find_next_queued(queue); |
359c48c0 TM |
682 | if (task != NULL) |
683 | task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, | |
684 | task, func, data); | |
c049f8ea | 685 | spin_unlock(&queue->lock); |
1da177e4 LT |
686 | |
687 | return task; | |
688 | } | |
f1dc237c TM |
689 | |
690 | /* | |
691 | * Wake up the first task on the wait queue. | |
692 | */ | |
693 | struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue, | |
694 | bool (*func)(struct rpc_task *, void *), void *data) | |
695 | { | |
696 | return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data); | |
697 | } | |
961a828d TM |
698 | EXPORT_SYMBOL_GPL(rpc_wake_up_first); |
699 | ||
700 | static bool rpc_wake_up_next_func(struct rpc_task *task, void *data) | |
701 | { | |
702 | return true; | |
703 | } | |
704 | ||
705 | /* | |
706 | * Wake up the next task on the wait queue. | |
707 | */ | |
708 | struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue) | |
709 | { | |
710 | return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL); | |
711 | } | |
e8914c65 | 712 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
713 | |
714 | /** | |
715 | * rpc_wake_up - wake up all rpc_tasks | |
716 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
717 | * | |
718 | * Grabs queue->lock | |
719 | */ | |
720 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
721 | { | |
1da177e4 | 722 | struct list_head *head; |
e6d83d55 | 723 | |
c049f8ea | 724 | spin_lock(&queue->lock); |
1da177e4 LT |
725 | head = &queue->tasks[queue->maxpriority]; |
726 | for (;;) { | |
540a0f75 TM |
727 | while (!list_empty(head)) { |
728 | struct rpc_task *task; | |
729 | task = list_first_entry(head, | |
730 | struct rpc_task, | |
731 | u.tk_wait.list); | |
96ef13b2 | 732 | rpc_wake_up_task_queue_locked(queue, task); |
540a0f75 | 733 | } |
1da177e4 LT |
734 | if (head == &queue->tasks[0]) |
735 | break; | |
736 | head--; | |
737 | } | |
c049f8ea | 738 | spin_unlock(&queue->lock); |
1da177e4 | 739 | } |
e8914c65 | 740 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
741 | |
742 | /** | |
743 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
744 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
745 | * @status: status value to set | |
746 | * | |
747 | * Grabs queue->lock | |
748 | */ | |
749 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
750 | { | |
751 | struct list_head *head; | |
1da177e4 | 752 | |
c049f8ea | 753 | spin_lock(&queue->lock); |
1da177e4 LT |
754 | head = &queue->tasks[queue->maxpriority]; |
755 | for (;;) { | |
540a0f75 TM |
756 | while (!list_empty(head)) { |
757 | struct rpc_task *task; | |
758 | task = list_first_entry(head, | |
759 | struct rpc_task, | |
760 | u.tk_wait.list); | |
1da177e4 | 761 | task->tk_status = status; |
96ef13b2 | 762 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
763 | } |
764 | if (head == &queue->tasks[0]) | |
765 | break; | |
766 | head--; | |
767 | } | |
c049f8ea | 768 | spin_unlock(&queue->lock); |
1da177e4 | 769 | } |
e8914c65 | 770 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 771 | |
7e0a0e38 | 772 | static void __rpc_queue_timer_fn(struct work_struct *work) |
36df9aae | 773 | { |
7e0a0e38 TM |
774 | struct rpc_wait_queue *queue = container_of(work, |
775 | struct rpc_wait_queue, | |
776 | timer_list.dwork.work); | |
36df9aae TM |
777 | struct rpc_task *task, *n; |
778 | unsigned long expires, now, timeo; | |
779 | ||
780 | spin_lock(&queue->lock); | |
781 | expires = now = jiffies; | |
782 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
6b2e6856 | 783 | timeo = task->tk_timeout; |
36df9aae | 784 | if (time_after_eq(now, timeo)) { |
36df9aae TM |
785 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
786 | task->tk_status = -ETIMEDOUT; | |
787 | rpc_wake_up_task_queue_locked(queue, task); | |
788 | continue; | |
789 | } | |
790 | if (expires == now || time_after(expires, timeo)) | |
791 | expires = timeo; | |
792 | } | |
793 | if (!list_empty(&queue->timer_list.list)) | |
794 | rpc_set_queue_timer(queue, expires); | |
795 | spin_unlock(&queue->lock); | |
796 | } | |
797 | ||
8014793b TM |
798 | static void __rpc_atrun(struct rpc_task *task) |
799 | { | |
6bd14416 TM |
800 | if (task->tk_status == -ETIMEDOUT) |
801 | task->tk_status = 0; | |
8014793b TM |
802 | } |
803 | ||
1da177e4 LT |
804 | /* |
805 | * Run a task at a later time | |
806 | */ | |
8014793b | 807 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 | 808 | { |
6b2e6856 | 809 | rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay); |
1da177e4 | 810 | } |
e8914c65 | 811 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 812 | |
4ce70ada TM |
813 | /* |
814 | * Helper to call task->tk_ops->rpc_call_prepare | |
815 | */ | |
aae2006e | 816 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
817 | { |
818 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
819 | } | |
820 | ||
7fdcf13b TM |
821 | static void |
822 | rpc_init_task_statistics(struct rpc_task *task) | |
823 | { | |
824 | /* Initialize retry counters */ | |
825 | task->tk_garb_retry = 2; | |
826 | task->tk_cred_retry = 2; | |
827 | task->tk_rebind_retry = 2; | |
828 | ||
829 | /* starting timestamp */ | |
830 | task->tk_start = ktime_get(); | |
831 | } | |
832 | ||
833 | static void | |
834 | rpc_reset_task_statistics(struct rpc_task *task) | |
835 | { | |
836 | task->tk_timeouts = 0; | |
ae67bd38 | 837 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT); |
7fdcf13b TM |
838 | rpc_init_task_statistics(task); |
839 | } | |
840 | ||
d05fdb0c | 841 | /* |
963d8fe5 | 842 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 843 | */ |
abbcf28f | 844 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 845 | { |
abbcf28f | 846 | task->tk_action = NULL; |
9dfe52a9 DW |
847 | if (task->tk_ops->rpc_count_stats) |
848 | task->tk_ops->rpc_count_stats(task, task->tk_calldata); | |
849 | else if (task->tk_client) | |
850 | rpc_count_iostats(task, task->tk_client->cl_metrics); | |
963d8fe5 TM |
851 | if (task->tk_ops->rpc_call_done != NULL) { |
852 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 853 | if (task->tk_action != NULL) { |
abbcf28f TM |
854 | /* Always release the RPC slot and buffer memory */ |
855 | xprt_release(task); | |
7fdcf13b | 856 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
857 | } |
858 | } | |
d05fdb0c | 859 | } |
d9b6cd94 | 860 | |
ae67bd38 TM |
861 | void rpc_signal_task(struct rpc_task *task) |
862 | { | |
863 | struct rpc_wait_queue *queue; | |
864 | ||
865 | if (!RPC_IS_ACTIVATED(task)) | |
866 | return; | |
867 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); | |
868 | smp_mb__after_atomic(); | |
869 | queue = READ_ONCE(task->tk_waitqueue); | |
870 | if (queue) | |
871 | rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS); | |
872 | } | |
873 | ||
d9b6cd94 TM |
874 | void rpc_exit(struct rpc_task *task, int status) |
875 | { | |
876 | task->tk_status = status; | |
877 | task->tk_action = rpc_exit_task; | |
6b5f5900 | 878 | rpc_wake_up_queued_task(task->tk_waitqueue, task); |
d9b6cd94 TM |
879 | } |
880 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 881 | |
bbd5a1f9 TM |
882 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
883 | { | |
a86dc496 | 884 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 885 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
886 | } |
887 | ||
1da177e4 LT |
888 | /* |
889 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
890 | */ | |
2efef837 | 891 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 892 | { |
eb9b55ab TM |
893 | struct rpc_wait_queue *queue; |
894 | int task_is_async = RPC_IS_ASYNC(task); | |
895 | int status = 0; | |
1da177e4 | 896 | |
46121cf7 CL |
897 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
898 | task->tk_pid, task->tk_flags); | |
1da177e4 | 899 | |
2bd4eef8 WAA |
900 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
901 | if (RPC_IS_QUEUED(task)) | |
902 | return; | |
1da177e4 | 903 | |
d05fdb0c | 904 | for (;;) { |
b55c5989 | 905 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
906 | |
907 | /* | |
21ead9ff CL |
908 | * Perform the next FSM step or a pending callback. |
909 | * | |
910 | * tk_action may be NULL if the task has been killed. | |
911 | * In particular, note that rpc_killall_tasks may | |
912 | * do this at any time, so beware when dereferencing. | |
1da177e4 | 913 | */ |
21ead9ff CL |
914 | do_action = task->tk_action; |
915 | if (task->tk_callback) { | |
916 | do_action = task->tk_callback; | |
917 | task->tk_callback = NULL; | |
1da177e4 | 918 | } |
21ead9ff CL |
919 | if (!do_action) |
920 | break; | |
e671edb9 | 921 | trace_rpc_task_run_action(task, do_action); |
b55c5989 | 922 | do_action(task); |
1da177e4 LT |
923 | |
924 | /* | |
925 | * Lockless check for whether task is sleeping or not. | |
926 | */ | |
927 | if (!RPC_IS_QUEUED(task)) | |
928 | continue; | |
ae67bd38 TM |
929 | |
930 | /* | |
931 | * Signalled tasks should exit rather than sleep. | |
932 | */ | |
933 | if (RPC_SIGNALLED(task)) | |
934 | rpc_exit(task, -ERESTARTSYS); | |
935 | ||
eb9b55ab TM |
936 | /* |
937 | * The queue->lock protects against races with | |
938 | * rpc_make_runnable(). | |
939 | * | |
940 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
941 | * rpc_task, rpc_make_runnable() can assign it to a | |
942 | * different workqueue. We therefore cannot assume that the | |
943 | * rpc_task pointer may still be dereferenced. | |
944 | */ | |
945 | queue = task->tk_waitqueue; | |
c049f8ea | 946 | spin_lock(&queue->lock); |
eb9b55ab | 947 | if (!RPC_IS_QUEUED(task)) { |
c049f8ea | 948 | spin_unlock(&queue->lock); |
1da177e4 LT |
949 | continue; |
950 | } | |
eb9b55ab | 951 | rpc_clear_running(task); |
c049f8ea | 952 | spin_unlock(&queue->lock); |
eb9b55ab TM |
953 | if (task_is_async) |
954 | return; | |
1da177e4 LT |
955 | |
956 | /* sync task: sleep here */ | |
46121cf7 | 957 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 958 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
959 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
960 | TASK_KILLABLE); | |
ae67bd38 | 961 | if (status < 0) { |
1da177e4 LT |
962 | /* |
963 | * When a sync task receives a signal, it exits with | |
964 | * -ERESTARTSYS. In order to catch any callbacks that | |
965 | * clean up after sleeping on some queue, we don't | |
966 | * break the loop here, but go around once more. | |
967 | */ | |
46121cf7 | 968 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
ae67bd38 | 969 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); |
96651ab3 | 970 | rpc_exit(task, -ERESTARTSYS); |
1da177e4 | 971 | } |
46121cf7 | 972 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
973 | } |
974 | ||
46121cf7 CL |
975 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
976 | task->tk_status); | |
1da177e4 LT |
977 | /* Release all resources associated with the task */ |
978 | rpc_release_task(task); | |
1da177e4 LT |
979 | } |
980 | ||
981 | /* | |
982 | * User-visible entry point to the scheduler. | |
983 | * | |
984 | * This may be called recursively if e.g. an async NFS task updates | |
985 | * the attributes and finds that dirty pages must be flushed. | |
986 | * NOTE: Upon exit of this function the task is guaranteed to be | |
987 | * released. In particular note that tk_release() will have | |
988 | * been called, so your task memory may have been freed. | |
989 | */ | |
2efef837 | 990 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 991 | { |
a76580fb TM |
992 | bool is_async = RPC_IS_ASYNC(task); |
993 | ||
44c28873 | 994 | rpc_set_active(task); |
f1dc237c | 995 | rpc_make_runnable(rpciod_workqueue, task); |
a76580fb | 996 | if (!is_async) |
d6a1ed08 | 997 | __rpc_execute(task); |
1da177e4 LT |
998 | } |
999 | ||
65f27f38 | 1000 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 1001 | { |
a1231fda TM |
1002 | unsigned int pflags = memalloc_nofs_save(); |
1003 | ||
65f27f38 | 1004 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1005 | memalloc_nofs_restore(pflags); |
1da177e4 LT |
1006 | } |
1007 | ||
02107148 | 1008 | /** |
5fe6eaa1 CL |
1009 | * rpc_malloc - allocate RPC buffer resources |
1010 | * @task: RPC task | |
1011 | * | |
1012 | * A single memory region is allocated, which is split between the | |
1013 | * RPC call and RPC reply that this task is being used for. When | |
1014 | * this RPC is retired, the memory is released by calling rpc_free. | |
1da177e4 | 1015 | * |
c5a4dd8b | 1016 | * To prevent rpciod from hanging, this allocator never sleeps, |
5fe6eaa1 CL |
1017 | * returning -ENOMEM and suppressing warning if the request cannot |
1018 | * be serviced immediately. The caller can arrange to sleep in a | |
1019 | * way that is safe for rpciod. | |
c5a4dd8b CL |
1020 | * |
1021 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
1022 | * mempool, ensuring that NFS reads and writes can always proceed, | |
1023 | * and that there is good locality of reference for these buffers. | |
1da177e4 | 1024 | */ |
5fe6eaa1 | 1025 | int rpc_malloc(struct rpc_task *task) |
1da177e4 | 1026 | { |
5fe6eaa1 CL |
1027 | struct rpc_rqst *rqst = task->tk_rqstp; |
1028 | size_t size = rqst->rq_callsize + rqst->rq_rcvsize; | |
aa3d1fae | 1029 | struct rpc_buffer *buf; |
12a3ad61 | 1030 | gfp_t gfp = GFP_NOFS; |
a564b8f0 MG |
1031 | |
1032 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 1033 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 1034 | |
aa3d1fae | 1035 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
1036 | if (size <= RPC_BUFFER_MAXSIZE) |
1037 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 1038 | else |
c5a4dd8b | 1039 | buf = kmalloc(size, gfp); |
ddce40df PZ |
1040 | |
1041 | if (!buf) | |
5fe6eaa1 | 1042 | return -ENOMEM; |
ddce40df | 1043 | |
aa3d1fae | 1044 | buf->len = size; |
215d0678 | 1045 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 1046 | task->tk_pid, size, buf); |
5fe6eaa1 | 1047 | rqst->rq_buffer = buf->data; |
68778945 | 1048 | rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; |
5fe6eaa1 | 1049 | return 0; |
1da177e4 | 1050 | } |
12444809 | 1051 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 1052 | |
02107148 | 1053 | /** |
3435c74a CL |
1054 | * rpc_free - free RPC buffer resources allocated via rpc_malloc |
1055 | * @task: RPC task | |
02107148 CL |
1056 | * |
1057 | */ | |
3435c74a | 1058 | void rpc_free(struct rpc_task *task) |
1da177e4 | 1059 | { |
3435c74a | 1060 | void *buffer = task->tk_rqstp->rq_buffer; |
aa3d1fae CL |
1061 | size_t size; |
1062 | struct rpc_buffer *buf; | |
02107148 | 1063 | |
aa3d1fae CL |
1064 | buf = container_of(buffer, struct rpc_buffer, data); |
1065 | size = buf->len; | |
c5a4dd8b | 1066 | |
215d0678 | 1067 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 1068 | size, buf); |
aa3d1fae | 1069 | |
c5a4dd8b CL |
1070 | if (size <= RPC_BUFFER_MAXSIZE) |
1071 | mempool_free(buf, rpc_buffer_mempool); | |
1072 | else | |
1073 | kfree(buf); | |
1da177e4 | 1074 | } |
12444809 | 1075 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
1076 | |
1077 | /* | |
1078 | * Creation and deletion of RPC task structures | |
1079 | */ | |
47fe0648 | 1080 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
1081 | { |
1082 | memset(task, 0, sizeof(*task)); | |
44c28873 | 1083 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
1084 | task->tk_flags = task_setup_data->flags; |
1085 | task->tk_ops = task_setup_data->callback_ops; | |
1086 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 1087 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 1088 | |
3ff7576d TM |
1089 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
1090 | task->tk_owner = current->tgid; | |
1da177e4 LT |
1091 | |
1092 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 1093 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 1094 | |
a101b043 TM |
1095 | task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client, |
1096 | xprt_get(task_setup_data->rpc_xprt)); | |
9d61498d | 1097 | |
1de7eea9 N |
1098 | task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred); |
1099 | ||
84115e1c TM |
1100 | if (task->tk_ops->rpc_call_prepare != NULL) |
1101 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 1102 | |
7fdcf13b | 1103 | rpc_init_task_statistics(task); |
ef759a2e | 1104 | |
46121cf7 | 1105 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 1106 | task_pid_nr(current)); |
1da177e4 LT |
1107 | } |
1108 | ||
1109 | static struct rpc_task * | |
1110 | rpc_alloc_task(void) | |
1111 | { | |
12a3ad61 | 1112 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); |
1da177e4 LT |
1113 | } |
1114 | ||
1da177e4 | 1115 | /* |
90c5755f | 1116 | * Create a new task for the specified client. |
1da177e4 | 1117 | */ |
84115e1c | 1118 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 1119 | { |
e8f5d77c TM |
1120 | struct rpc_task *task = setup_data->task; |
1121 | unsigned short flags = 0; | |
1122 | ||
1123 | if (task == NULL) { | |
1124 | task = rpc_alloc_task(); | |
e8f5d77c TM |
1125 | flags = RPC_TASK_DYNAMIC; |
1126 | } | |
1da177e4 | 1127 | |
84115e1c | 1128 | rpc_init_task(task, setup_data); |
e8f5d77c | 1129 | task->tk_flags |= flags; |
46121cf7 | 1130 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 1131 | return task; |
1da177e4 LT |
1132 | } |
1133 | ||
c6567ed1 TM |
1134 | /* |
1135 | * rpc_free_task - release rpc task and perform cleanups | |
1136 | * | |
1137 | * Note that we free up the rpc_task _after_ rpc_release_calldata() | |
1138 | * in order to work around a workqueue dependency issue. | |
1139 | * | |
1140 | * Tejun Heo states: | |
1141 | * "Workqueue currently considers two work items to be the same if they're | |
1142 | * on the same address and won't execute them concurrently - ie. it | |
1143 | * makes a work item which is queued again while being executed wait | |
1144 | * for the previous execution to complete. | |
1145 | * | |
1146 | * If a work function frees the work item, and then waits for an event | |
1147 | * which should be performed by another work item and *that* work item | |
1148 | * recycles the freed work item, it can create a false dependency loop. | |
1149 | * There really is no reliable way to detect this short of verifying | |
1150 | * every memory free." | |
1151 | * | |
1152 | */ | |
32bfb5c0 | 1153 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 1154 | { |
c6567ed1 TM |
1155 | unsigned short tk_flags = task->tk_flags; |
1156 | ||
1de7eea9 | 1157 | put_rpccred(task->tk_op_cred); |
c6567ed1 | 1158 | rpc_release_calldata(task->tk_ops, task->tk_calldata); |
1da177e4 | 1159 | |
c6567ed1 | 1160 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
1161 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
1162 | mempool_free(task, rpc_task_mempool); | |
1163 | } | |
32bfb5c0 TM |
1164 | } |
1165 | ||
1166 | static void rpc_async_release(struct work_struct *work) | |
1167 | { | |
a1231fda TM |
1168 | unsigned int pflags = memalloc_nofs_save(); |
1169 | ||
32bfb5c0 | 1170 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1171 | memalloc_nofs_restore(pflags); |
32bfb5c0 TM |
1172 | } |
1173 | ||
bf294b41 | 1174 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 1175 | { |
87ed5003 | 1176 | xprt_release(task); |
a271c5a0 | 1177 | if (task->tk_msg.rpc_cred) { |
a52458b4 | 1178 | put_cred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
1179 | task->tk_msg.rpc_cred = NULL; |
1180 | } | |
58f9612c | 1181 | rpc_task_release_client(task); |
bf294b41 TM |
1182 | } |
1183 | ||
1184 | static void rpc_final_put_task(struct rpc_task *task, | |
1185 | struct workqueue_struct *q) | |
1186 | { | |
1187 | if (q != NULL) { | |
32bfb5c0 | 1188 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1189 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1190 | } else |
1191 | rpc_free_task(task); | |
e6b3c4db | 1192 | } |
bf294b41 TM |
1193 | |
1194 | static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q) | |
1195 | { | |
1196 | if (atomic_dec_and_test(&task->tk_count)) { | |
1197 | rpc_release_resources_task(task); | |
1198 | rpc_final_put_task(task, q); | |
1199 | } | |
1200 | } | |
1201 | ||
1202 | void rpc_put_task(struct rpc_task *task) | |
1203 | { | |
1204 | rpc_do_put_task(task, NULL); | |
1205 | } | |
e8914c65 | 1206 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1207 | |
bf294b41 TM |
1208 | void rpc_put_task_async(struct rpc_task *task) |
1209 | { | |
1210 | rpc_do_put_task(task, task->tk_workqueue); | |
1211 | } | |
1212 | EXPORT_SYMBOL_GPL(rpc_put_task_async); | |
1213 | ||
bde8f00c | 1214 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1215 | { |
46121cf7 | 1216 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1217 | |
0a0c2a57 | 1218 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1219 | |
bf294b41 | 1220 | rpc_release_resources_task(task); |
e6b3c4db | 1221 | |
bf294b41 TM |
1222 | /* |
1223 | * Note: at this point we have been removed from rpc_clnt->cl_tasks, | |
1224 | * so it should be safe to use task->tk_count as a test for whether | |
1225 | * or not any other processes still hold references to our rpc_task. | |
1226 | */ | |
1227 | if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) { | |
1228 | /* Wake up anyone who may be waiting for task completion */ | |
1229 | if (!rpc_complete_task(task)) | |
1230 | return; | |
1231 | } else { | |
1232 | if (!atomic_dec_and_test(&task->tk_count)) | |
1233 | return; | |
1234 | } | |
1235 | rpc_final_put_task(task, task->tk_workqueue); | |
1da177e4 LT |
1236 | } |
1237 | ||
b247bbf1 TM |
1238 | int rpciod_up(void) |
1239 | { | |
1240 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
1241 | } | |
1242 | ||
1243 | void rpciod_down(void) | |
1244 | { | |
1245 | module_put(THIS_MODULE); | |
1246 | } | |
1247 | ||
1da177e4 | 1248 | /* |
b247bbf1 | 1249 | * Start up the rpciod workqueue. |
1da177e4 | 1250 | */ |
b247bbf1 | 1251 | static int rpciod_start(void) |
1da177e4 LT |
1252 | { |
1253 | struct workqueue_struct *wq; | |
ab418d70 | 1254 | |
1da177e4 LT |
1255 | /* |
1256 | * Create the rpciod thread and wait for it to start. | |
1257 | */ | |
ab418d70 | 1258 | dprintk("RPC: creating workqueue rpciod\n"); |
f515f86b | 1259 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); |
40a5f1b1 TM |
1260 | if (!wq) |
1261 | goto out_failed; | |
1da177e4 | 1262 | rpciod_workqueue = wq; |
40a5f1b1 | 1263 | /* Note: highpri because network receive is latency sensitive */ |
90ea9f1b | 1264 | wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0); |
40a5f1b1 TM |
1265 | if (!wq) |
1266 | goto free_rpciod; | |
1267 | xprtiod_workqueue = wq; | |
1268 | return 1; | |
1269 | free_rpciod: | |
1270 | wq = rpciod_workqueue; | |
1271 | rpciod_workqueue = NULL; | |
1272 | destroy_workqueue(wq); | |
1273 | out_failed: | |
1274 | return 0; | |
1da177e4 LT |
1275 | } |
1276 | ||
b247bbf1 | 1277 | static void rpciod_stop(void) |
1da177e4 | 1278 | { |
b247bbf1 | 1279 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1280 | |
b247bbf1 TM |
1281 | if (rpciod_workqueue == NULL) |
1282 | return; | |
ab418d70 | 1283 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1284 | |
b247bbf1 TM |
1285 | wq = rpciod_workqueue; |
1286 | rpciod_workqueue = NULL; | |
1287 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1288 | wq = xprtiod_workqueue; |
1289 | xprtiod_workqueue = NULL; | |
1290 | destroy_workqueue(wq); | |
1da177e4 LT |
1291 | } |
1292 | ||
1da177e4 LT |
1293 | void |
1294 | rpc_destroy_mempool(void) | |
1295 | { | |
b247bbf1 | 1296 | rpciod_stop(); |
17a9618e JL |
1297 | mempool_destroy(rpc_buffer_mempool); |
1298 | mempool_destroy(rpc_task_mempool); | |
1299 | kmem_cache_destroy(rpc_task_slabp); | |
1300 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 1301 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | int | |
1305 | rpc_init_mempool(void) | |
1306 | { | |
f6a1cc89 TM |
1307 | /* |
1308 | * The following is not strictly a mempool initialisation, | |
1309 | * but there is no harm in doing it here | |
1310 | */ | |
1311 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
1312 | if (!rpciod_start()) | |
1313 | goto err_nomem; | |
1314 | ||
1da177e4 LT |
1315 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1316 | sizeof(struct rpc_task), | |
1317 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1318 | NULL); |
1da177e4 LT |
1319 | if (!rpc_task_slabp) |
1320 | goto err_nomem; | |
1321 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
1322 | RPC_BUFFER_MAXSIZE, | |
1323 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1324 | NULL); |
1da177e4 LT |
1325 | if (!rpc_buffer_slabp) |
1326 | goto err_nomem; | |
93d2341c MD |
1327 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1328 | rpc_task_slabp); | |
1da177e4 LT |
1329 | if (!rpc_task_mempool) |
1330 | goto err_nomem; | |
93d2341c MD |
1331 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1332 | rpc_buffer_slabp); | |
1da177e4 LT |
1333 | if (!rpc_buffer_mempool) |
1334 | goto err_nomem; | |
1335 | return 0; | |
1336 | err_nomem: | |
1337 | rpc_destroy_mempool(); | |
1338 | return -ENOMEM; | |
1339 | } |