]>
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 | { |
6b2e6856 | 207 | INIT_LIST_HEAD(&task->u.tk_wait.timer_list); |
1da177e4 | 208 | if (RPC_IS_PRIORITY(queue)) |
3b27bad7 | 209 | __rpc_add_wait_queue_priority(queue, task, queue_priority); |
1da177e4 LT |
210 | else if (RPC_IS_SWAPPER(task)) |
211 | list_add(&task->u.tk_wait.list, &queue->tasks[0]); | |
212 | else | |
213 | list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]); | |
96ef13b2 | 214 | task->tk_waitqueue = queue; |
e19b63da | 215 | queue->qlen++; |
1166fde6 TM |
216 | /* barrier matches the read in rpc_wake_up_task_queue_locked() */ |
217 | smp_wmb(); | |
1da177e4 LT |
218 | rpc_set_queued(task); |
219 | ||
46121cf7 CL |
220 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
221 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
222 | } |
223 | ||
224 | /* | |
225 | * Remove request from a priority queue. | |
226 | */ | |
227 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
228 | { | |
f42f7c28 | 229 | __rpc_list_dequeue_task(task); |
1da177e4 LT |
230 | } |
231 | ||
232 | /* | |
233 | * Remove request from queue. | |
234 | * Note: must be called with spin lock held. | |
235 | */ | |
96ef13b2 | 236 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 237 | { |
eb276c0e | 238 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
239 | if (RPC_IS_PRIORITY(queue)) |
240 | __rpc_remove_wait_queue_priority(task); | |
f42f7c28 TM |
241 | else |
242 | list_del(&task->u.tk_wait.list); | |
e19b63da | 243 | queue->qlen--; |
46121cf7 CL |
244 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
245 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
246 | } |
247 | ||
3ff7576d | 248 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
249 | { |
250 | int i; | |
251 | ||
252 | spin_lock_init(&queue->lock); | |
253 | for (i = 0; i < ARRAY_SIZE(queue->tasks); i++) | |
254 | INIT_LIST_HEAD(&queue->tasks[i]); | |
3ff7576d | 255 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 256 | rpc_reset_waitqueue_priority(queue); |
36df9aae | 257 | queue->qlen = 0; |
7e0a0e38 | 258 | queue->timer_list.expires = 0; |
f063ed21 | 259 | INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn); |
36df9aae | 260 | INIT_LIST_HEAD(&queue->timer_list.list); |
2f09c242 | 261 | rpc_assign_waitqueue_name(queue, qname); |
1da177e4 LT |
262 | } |
263 | ||
264 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
265 | { | |
3ff7576d | 266 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 267 | } |
689cf5c1 | 268 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
269 | |
270 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
271 | { | |
3ff7576d | 272 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 273 | } |
e8914c65 | 274 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 275 | |
f6a1cc89 TM |
276 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
277 | { | |
7e0a0e38 | 278 | cancel_delayed_work_sync(&queue->timer_list.dwork); |
f6a1cc89 TM |
279 | } |
280 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
281 | ||
dfd01f02 | 282 | static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode) |
44c28873 | 283 | { |
416ad3c9 | 284 | freezable_schedule_unsafe(); |
dfd01f02 PZ |
285 | if (signal_pending_state(mode, current)) |
286 | return -ERESTARTSYS; | |
44c28873 TM |
287 | return 0; |
288 | } | |
289 | ||
1306729b | 290 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS) |
c44fe705 TM |
291 | static void rpc_task_set_debuginfo(struct rpc_task *task) |
292 | { | |
293 | static atomic_t rpc_pid; | |
294 | ||
c44fe705 TM |
295 | task->tk_pid = atomic_inc_return(&rpc_pid); |
296 | } | |
297 | #else | |
298 | static inline void rpc_task_set_debuginfo(struct rpc_task *task) | |
299 | { | |
300 | } | |
301 | #endif | |
302 | ||
e6b3c4db TM |
303 | static void rpc_set_active(struct rpc_task *task) |
304 | { | |
c44fe705 | 305 | rpc_task_set_debuginfo(task); |
58f9612c | 306 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e671edb9 | 307 | trace_rpc_task_begin(task, NULL); |
e6b3c4db TM |
308 | } |
309 | ||
44c28873 TM |
310 | /* |
311 | * Mark an RPC call as having completed by clearing the 'active' bit | |
bf294b41 | 312 | * and then waking up all tasks that were sleeping. |
44c28873 | 313 | */ |
bf294b41 | 314 | static int rpc_complete_task(struct rpc_task *task) |
44c28873 | 315 | { |
bf294b41 TM |
316 | void *m = &task->tk_runstate; |
317 | wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE); | |
318 | struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE); | |
319 | unsigned long flags; | |
320 | int ret; | |
321 | ||
e671edb9 | 322 | trace_rpc_task_complete(task, NULL); |
82b0a4c3 | 323 | |
bf294b41 | 324 | spin_lock_irqsave(&wq->lock, flags); |
e6b3c4db | 325 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
bf294b41 TM |
326 | ret = atomic_dec_and_test(&task->tk_count); |
327 | if (waitqueue_active(wq)) | |
ac5be6b4 | 328 | __wake_up_locked_key(wq, TASK_NORMAL, &k); |
bf294b41 TM |
329 | spin_unlock_irqrestore(&wq->lock, flags); |
330 | return ret; | |
44c28873 TM |
331 | } |
332 | ||
333 | /* | |
334 | * Allow callers to wait for completion of an RPC call | |
bf294b41 TM |
335 | * |
336 | * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit() | |
337 | * to enforce taking of the wq->lock and hence avoid races with | |
338 | * rpc_complete_task(). | |
44c28873 | 339 | */ |
c1221321 | 340 | int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action) |
44c28873 TM |
341 | { |
342 | if (action == NULL) | |
150030b7 | 343 | action = rpc_wait_bit_killable; |
bf294b41 | 344 | return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 345 | action, TASK_KILLABLE); |
44c28873 | 346 | } |
e8914c65 | 347 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 348 | |
1da177e4 LT |
349 | /* |
350 | * Make an RPC task runnable. | |
351 | * | |
506026c3 JL |
352 | * Note: If the task is ASYNC, and is being made runnable after sitting on an |
353 | * rpc_wait_queue, this must be called with the queue spinlock held to protect | |
354 | * the wait queue operation. | |
a3c3cac5 TM |
355 | * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(), |
356 | * which is needed to ensure that __rpc_execute() doesn't loop (due to the | |
357 | * lockless RPC_IS_QUEUED() test) before we've had a chance to test | |
358 | * the RPC_TASK_RUNNING flag. | |
1da177e4 | 359 | */ |
f1dc237c TM |
360 | static void rpc_make_runnable(struct workqueue_struct *wq, |
361 | struct rpc_task *task) | |
1da177e4 | 362 | { |
a3c3cac5 TM |
363 | bool need_wakeup = !rpc_test_and_set_running(task); |
364 | ||
1da177e4 | 365 | rpc_clear_queued(task); |
a3c3cac5 | 366 | if (!need_wakeup) |
cc4dc59e | 367 | return; |
1da177e4 | 368 | if (RPC_IS_ASYNC(task)) { |
65f27f38 | 369 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
f1dc237c | 370 | queue_work(wq, &task->u.tk_work); |
1da177e4 | 371 | } else |
96651ab3 | 372 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
373 | } |
374 | ||
1da177e4 LT |
375 | /* |
376 | * Prepare for sleeping on a wait queue. | |
377 | * By always appending tasks to the list we ensure FIFO behavior. | |
378 | * NB: An RPC task will only receive interrupt-driven events as long | |
379 | * as it's on a wait queue. | |
380 | */ | |
d40171df | 381 | static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q, |
3b27bad7 | 382 | struct rpc_task *task, |
3b27bad7 | 383 | unsigned char queue_priority) |
1da177e4 | 384 | { |
46121cf7 CL |
385 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
386 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 387 | |
e671edb9 | 388 | trace_rpc_task_sleep(task, q); |
82b0a4c3 | 389 | |
3b27bad7 | 390 | __rpc_add_wait_queue(q, task, queue_priority); |
1da177e4 | 391 | |
6b2e6856 TM |
392 | } |
393 | ||
d40171df TM |
394 | static void __rpc_sleep_on_priority(struct rpc_wait_queue *q, |
395 | struct rpc_task *task, | |
396 | unsigned char queue_priority) | |
397 | { | |
398 | if (WARN_ON_ONCE(RPC_IS_QUEUED(task))) | |
399 | return; | |
400 | __rpc_do_sleep_on_priority(q, task, queue_priority); | |
401 | } | |
402 | ||
6b2e6856 TM |
403 | static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, |
404 | struct rpc_task *task, unsigned long timeout, | |
405 | unsigned char queue_priority) | |
406 | { | |
d40171df TM |
407 | if (WARN_ON_ONCE(RPC_IS_QUEUED(task))) |
408 | return; | |
6b2e6856 | 409 | if (time_is_after_jiffies(timeout)) { |
d40171df | 410 | __rpc_do_sleep_on_priority(q, task, queue_priority); |
6b2e6856 TM |
411 | __rpc_add_timer(q, task, timeout); |
412 | } else | |
413 | task->tk_status = -ETIMEDOUT; | |
1da177e4 LT |
414 | } |
415 | ||
87150aae TM |
416 | static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action) |
417 | { | |
418 | if (action && !WARN_ON_ONCE(task->tk_callback != NULL)) | |
419 | task->tk_callback = action; | |
420 | } | |
421 | ||
422 | static bool rpc_sleep_check_activated(struct rpc_task *task) | |
1da177e4 | 423 | { |
58f9612c | 424 | /* We shouldn't ever put an inactive task to sleep */ |
87150aae | 425 | if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) { |
e454a7a8 WAA |
426 | task->tk_status = -EIO; |
427 | rpc_put_task_async(task); | |
87150aae | 428 | return false; |
e454a7a8 | 429 | } |
87150aae TM |
430 | return true; |
431 | } | |
432 | ||
6b2e6856 TM |
433 | void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task, |
434 | rpc_action action, unsigned long timeout) | |
435 | { | |
436 | if (!rpc_sleep_check_activated(task)) | |
437 | return; | |
438 | ||
439 | rpc_set_tk_callback(task, action); | |
440 | ||
441 | /* | |
442 | * Protect the queue operations. | |
443 | */ | |
c049f8ea | 444 | spin_lock(&q->lock); |
6b2e6856 | 445 | __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority); |
c049f8ea | 446 | spin_unlock(&q->lock); |
6b2e6856 TM |
447 | } |
448 | EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout); | |
449 | ||
87150aae TM |
450 | void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, |
451 | rpc_action action) | |
452 | { | |
453 | if (!rpc_sleep_check_activated(task)) | |
454 | return; | |
455 | ||
456 | rpc_set_tk_callback(task, action); | |
e6b3c4db | 457 | |
6b2e6856 | 458 | WARN_ON_ONCE(task->tk_timeout != 0); |
1da177e4 LT |
459 | /* |
460 | * Protect the queue operations. | |
461 | */ | |
c049f8ea | 462 | spin_lock(&q->lock); |
87150aae | 463 | __rpc_sleep_on_priority(q, task, task->tk_priority); |
c049f8ea | 464 | spin_unlock(&q->lock); |
1da177e4 | 465 | } |
e8914c65 | 466 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 | 467 | |
6b2e6856 TM |
468 | void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q, |
469 | struct rpc_task *task, unsigned long timeout, int priority) | |
470 | { | |
471 | if (!rpc_sleep_check_activated(task)) | |
472 | return; | |
473 | ||
474 | priority -= RPC_PRIORITY_LOW; | |
475 | /* | |
476 | * Protect the queue operations. | |
477 | */ | |
c049f8ea | 478 | spin_lock(&q->lock); |
6b2e6856 | 479 | __rpc_sleep_on_priority_timeout(q, task, timeout, priority); |
c049f8ea | 480 | spin_unlock(&q->lock); |
6b2e6856 TM |
481 | } |
482 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout); | |
483 | ||
3b27bad7 | 484 | void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task, |
8357a9b6 | 485 | int priority) |
3b27bad7 | 486 | { |
87150aae | 487 | if (!rpc_sleep_check_activated(task)) |
e454a7a8 | 488 | return; |
87150aae | 489 | |
6b2e6856 | 490 | WARN_ON_ONCE(task->tk_timeout != 0); |
8357a9b6 | 491 | priority -= RPC_PRIORITY_LOW; |
3b27bad7 TM |
492 | /* |
493 | * Protect the queue operations. | |
494 | */ | |
c049f8ea | 495 | spin_lock(&q->lock); |
8357a9b6 | 496 | __rpc_sleep_on_priority(q, task, priority); |
c049f8ea | 497 | spin_unlock(&q->lock); |
3b27bad7 | 498 | } |
1e1093c7 | 499 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority); |
3b27bad7 | 500 | |
1da177e4 | 501 | /** |
f1dc237c TM |
502 | * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task |
503 | * @wq: workqueue on which to run task | |
96ef13b2 | 504 | * @queue: wait queue |
1da177e4 LT |
505 | * @task: task to be woken up |
506 | * | |
507 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
508 | */ | |
f1dc237c TM |
509 | static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq, |
510 | struct rpc_wait_queue *queue, | |
511 | struct rpc_task *task) | |
1da177e4 | 512 | { |
46121cf7 CL |
513 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
514 | task->tk_pid, jiffies); | |
1da177e4 | 515 | |
1da177e4 LT |
516 | /* Has the task been executed yet? If not, we cannot wake it up! */ |
517 | if (!RPC_IS_ACTIVATED(task)) { | |
518 | printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task); | |
519 | return; | |
520 | } | |
521 | ||
e671edb9 | 522 | trace_rpc_task_wakeup(task, queue); |
82b0a4c3 | 523 | |
96ef13b2 | 524 | __rpc_remove_wait_queue(queue, task); |
1da177e4 | 525 | |
f1dc237c | 526 | rpc_make_runnable(wq, task); |
1da177e4 | 527 | |
46121cf7 | 528 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
529 | } |
530 | ||
531 | /* | |
96ef13b2 | 532 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 533 | */ |
359c48c0 TM |
534 | static struct rpc_task * |
535 | rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq, | |
536 | struct rpc_wait_queue *queue, struct rpc_task *task, | |
537 | bool (*action)(struct rpc_task *, void *), void *data) | |
1da177e4 | 538 | { |
1166fde6 TM |
539 | if (RPC_IS_QUEUED(task)) { |
540 | smp_rmb(); | |
359c48c0 TM |
541 | if (task->tk_waitqueue == queue) { |
542 | if (action == NULL || action(task, data)) { | |
543 | __rpc_do_wake_up_task_on_wq(wq, queue, task); | |
544 | return task; | |
545 | } | |
546 | } | |
1166fde6 | 547 | } |
359c48c0 TM |
548 | return NULL; |
549 | } | |
550 | ||
f1dc237c TM |
551 | /* |
552 | * Wake up a queued task while the queue lock is being held | |
553 | */ | |
691b45dd CL |
554 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, |
555 | struct rpc_task *task) | |
f1dc237c | 556 | { |
691b45dd CL |
557 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, |
558 | task, NULL, NULL); | |
2275cde4 TM |
559 | } |
560 | ||
1da177e4 | 561 | /* |
96ef13b2 | 562 | * Wake up a task on a specific queue |
1da177e4 | 563 | */ |
96ef13b2 | 564 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 565 | { |
5ce97039 TM |
566 | if (!RPC_IS_QUEUED(task)) |
567 | return; | |
c049f8ea | 568 | spin_lock(&queue->lock); |
96ef13b2 | 569 | rpc_wake_up_task_queue_locked(queue, task); |
c049f8ea | 570 | spin_unlock(&queue->lock); |
1da177e4 | 571 | } |
96ef13b2 TM |
572 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
573 | ||
359c48c0 TM |
574 | static bool rpc_task_action_set_status(struct rpc_task *task, void *status) |
575 | { | |
576 | task->tk_status = *(int *)status; | |
577 | return true; | |
578 | } | |
579 | ||
580 | static void | |
581 | rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue, | |
582 | struct rpc_task *task, int status) | |
583 | { | |
584 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, | |
585 | task, rpc_task_action_set_status, &status); | |
586 | } | |
587 | ||
588 | /** | |
589 | * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status | |
590 | * @queue: pointer to rpc_wait_queue | |
591 | * @task: pointer to rpc_task | |
592 | * @status: integer error value | |
593 | * | |
594 | * If @task is queued on @queue, then it is woken up, and @task->tk_status is | |
595 | * set to the value of @status. | |
596 | */ | |
597 | void | |
598 | rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue, | |
599 | struct rpc_task *task, int status) | |
600 | { | |
601 | if (!RPC_IS_QUEUED(task)) | |
602 | return; | |
c049f8ea | 603 | spin_lock(&queue->lock); |
359c48c0 | 604 | rpc_wake_up_task_queue_set_status_locked(queue, task, status); |
c049f8ea | 605 | spin_unlock(&queue->lock); |
359c48c0 TM |
606 | } |
607 | ||
1da177e4 LT |
608 | /* |
609 | * Wake up the next task on a priority queue. | |
610 | */ | |
961a828d | 611 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
612 | { |
613 | struct list_head *q; | |
614 | struct rpc_task *task; | |
615 | ||
616 | /* | |
3ff7576d | 617 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
618 | */ |
619 | q = &queue->tasks[queue->priority]; | |
49ac5f2a ZX |
620 | if (!list_empty(q) && queue->nr) { |
621 | queue->nr--; | |
f42f7c28 TM |
622 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
623 | goto out; | |
1da177e4 LT |
624 | } |
625 | ||
626 | /* | |
627 | * Service the next queue. | |
628 | */ | |
629 | do { | |
630 | if (q == &queue->tasks[0]) | |
631 | q = &queue->tasks[queue->maxpriority]; | |
632 | else | |
633 | q = q - 1; | |
634 | if (!list_empty(q)) { | |
f42f7c28 | 635 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
1da177e4 LT |
636 | goto new_queue; |
637 | } | |
638 | } while (q != &queue->tasks[queue->priority]); | |
639 | ||
640 | rpc_reset_waitqueue_priority(queue); | |
641 | return NULL; | |
642 | ||
643 | new_queue: | |
644 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
1da177e4 | 645 | out: |
1da177e4 LT |
646 | return task; |
647 | } | |
648 | ||
961a828d TM |
649 | static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue) |
650 | { | |
651 | if (RPC_IS_PRIORITY(queue)) | |
652 | return __rpc_find_next_queued_priority(queue); | |
653 | if (!list_empty(&queue->tasks[0])) | |
654 | return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list); | |
655 | return NULL; | |
656 | } | |
657 | ||
1da177e4 | 658 | /* |
961a828d | 659 | * Wake up the first task on the wait queue. |
1da177e4 | 660 | */ |
f1dc237c TM |
661 | struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq, |
662 | struct rpc_wait_queue *queue, | |
961a828d | 663 | bool (*func)(struct rpc_task *, void *), void *data) |
1da177e4 LT |
664 | { |
665 | struct rpc_task *task = NULL; | |
666 | ||
961a828d | 667 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 668 | queue, rpc_qname(queue)); |
c049f8ea | 669 | spin_lock(&queue->lock); |
961a828d | 670 | task = __rpc_find_next_queued(queue); |
359c48c0 TM |
671 | if (task != NULL) |
672 | task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, | |
673 | task, func, data); | |
c049f8ea | 674 | spin_unlock(&queue->lock); |
1da177e4 LT |
675 | |
676 | return task; | |
677 | } | |
f1dc237c TM |
678 | |
679 | /* | |
680 | * Wake up the first task on the wait queue. | |
681 | */ | |
682 | struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue, | |
683 | bool (*func)(struct rpc_task *, void *), void *data) | |
684 | { | |
685 | return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data); | |
686 | } | |
961a828d TM |
687 | EXPORT_SYMBOL_GPL(rpc_wake_up_first); |
688 | ||
689 | static bool rpc_wake_up_next_func(struct rpc_task *task, void *data) | |
690 | { | |
691 | return true; | |
692 | } | |
693 | ||
694 | /* | |
695 | * Wake up the next task on the wait queue. | |
696 | */ | |
697 | struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue) | |
698 | { | |
699 | return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL); | |
700 | } | |
e8914c65 | 701 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 | 702 | |
e106db03 TM |
703 | /** |
704 | * rpc_wake_up_locked - wake up all rpc_tasks | |
705 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
706 | * | |
707 | */ | |
708 | static void rpc_wake_up_locked(struct rpc_wait_queue *queue) | |
709 | { | |
710 | struct rpc_task *task; | |
711 | ||
712 | for (;;) { | |
713 | task = __rpc_find_next_queued(queue); | |
714 | if (task == NULL) | |
715 | break; | |
716 | rpc_wake_up_task_queue_locked(queue, task); | |
717 | } | |
718 | } | |
719 | ||
1da177e4 LT |
720 | /** |
721 | * rpc_wake_up - wake up all rpc_tasks | |
722 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
723 | * | |
724 | * Grabs queue->lock | |
725 | */ | |
726 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
727 | { | |
c049f8ea | 728 | spin_lock(&queue->lock); |
e106db03 TM |
729 | rpc_wake_up_locked(queue); |
730 | spin_unlock(&queue->lock); | |
731 | } | |
732 | EXPORT_SYMBOL_GPL(rpc_wake_up); | |
733 | ||
734 | /** | |
735 | * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value. | |
736 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
737 | * @status: status value to set | |
738 | */ | |
739 | static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status) | |
740 | { | |
741 | struct rpc_task *task; | |
742 | ||
1da177e4 | 743 | for (;;) { |
e106db03 TM |
744 | task = __rpc_find_next_queued(queue); |
745 | if (task == NULL) | |
1da177e4 | 746 | break; |
e106db03 | 747 | rpc_wake_up_task_queue_set_status_locked(queue, task, status); |
1da177e4 | 748 | } |
1da177e4 LT |
749 | } |
750 | ||
751 | /** | |
752 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
753 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
754 | * @status: status value to set | |
755 | * | |
756 | * Grabs queue->lock | |
757 | */ | |
758 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
759 | { | |
c049f8ea | 760 | spin_lock(&queue->lock); |
e106db03 | 761 | rpc_wake_up_status_locked(queue, status); |
c049f8ea | 762 | spin_unlock(&queue->lock); |
1da177e4 | 763 | } |
e8914c65 | 764 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 765 | |
7e0a0e38 | 766 | static void __rpc_queue_timer_fn(struct work_struct *work) |
36df9aae | 767 | { |
7e0a0e38 TM |
768 | struct rpc_wait_queue *queue = container_of(work, |
769 | struct rpc_wait_queue, | |
770 | timer_list.dwork.work); | |
36df9aae TM |
771 | struct rpc_task *task, *n; |
772 | unsigned long expires, now, timeo; | |
773 | ||
774 | spin_lock(&queue->lock); | |
775 | expires = now = jiffies; | |
776 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
6b2e6856 | 777 | timeo = task->tk_timeout; |
36df9aae | 778 | if (time_after_eq(now, timeo)) { |
36df9aae TM |
779 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
780 | task->tk_status = -ETIMEDOUT; | |
781 | rpc_wake_up_task_queue_locked(queue, task); | |
782 | continue; | |
783 | } | |
784 | if (expires == now || time_after(expires, timeo)) | |
785 | expires = timeo; | |
786 | } | |
787 | if (!list_empty(&queue->timer_list.list)) | |
788 | rpc_set_queue_timer(queue, expires); | |
789 | spin_unlock(&queue->lock); | |
790 | } | |
791 | ||
8014793b TM |
792 | static void __rpc_atrun(struct rpc_task *task) |
793 | { | |
6bd14416 TM |
794 | if (task->tk_status == -ETIMEDOUT) |
795 | task->tk_status = 0; | |
8014793b TM |
796 | } |
797 | ||
1da177e4 LT |
798 | /* |
799 | * Run a task at a later time | |
800 | */ | |
8014793b | 801 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 | 802 | { |
6b2e6856 | 803 | rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay); |
1da177e4 | 804 | } |
e8914c65 | 805 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 806 | |
4ce70ada TM |
807 | /* |
808 | * Helper to call task->tk_ops->rpc_call_prepare | |
809 | */ | |
aae2006e | 810 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
811 | { |
812 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
813 | } | |
814 | ||
7fdcf13b TM |
815 | static void |
816 | rpc_init_task_statistics(struct rpc_task *task) | |
817 | { | |
818 | /* Initialize retry counters */ | |
819 | task->tk_garb_retry = 2; | |
820 | task->tk_cred_retry = 2; | |
821 | task->tk_rebind_retry = 2; | |
822 | ||
823 | /* starting timestamp */ | |
824 | task->tk_start = ktime_get(); | |
825 | } | |
826 | ||
827 | static void | |
828 | rpc_reset_task_statistics(struct rpc_task *task) | |
829 | { | |
830 | task->tk_timeouts = 0; | |
ae67bd38 | 831 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT); |
7fdcf13b TM |
832 | rpc_init_task_statistics(task); |
833 | } | |
834 | ||
d05fdb0c | 835 | /* |
963d8fe5 | 836 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 837 | */ |
abbcf28f | 838 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 839 | { |
1d8a0e13 | 840 | trace_rpc_task_end(task, task->tk_action); |
abbcf28f | 841 | task->tk_action = NULL; |
9dfe52a9 DW |
842 | if (task->tk_ops->rpc_count_stats) |
843 | task->tk_ops->rpc_count_stats(task, task->tk_calldata); | |
844 | else if (task->tk_client) | |
845 | rpc_count_iostats(task, task->tk_client->cl_metrics); | |
963d8fe5 TM |
846 | if (task->tk_ops->rpc_call_done != NULL) { |
847 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 848 | if (task->tk_action != NULL) { |
abbcf28f TM |
849 | /* Always release the RPC slot and buffer memory */ |
850 | xprt_release(task); | |
7fdcf13b | 851 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
852 | } |
853 | } | |
d05fdb0c | 854 | } |
d9b6cd94 | 855 | |
ae67bd38 TM |
856 | void rpc_signal_task(struct rpc_task *task) |
857 | { | |
858 | struct rpc_wait_queue *queue; | |
859 | ||
860 | if (!RPC_IS_ACTIVATED(task)) | |
861 | return; | |
862 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); | |
863 | smp_mb__after_atomic(); | |
864 | queue = READ_ONCE(task->tk_waitqueue); | |
865 | if (queue) | |
866 | rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS); | |
867 | } | |
868 | ||
d9b6cd94 TM |
869 | void rpc_exit(struct rpc_task *task, int status) |
870 | { | |
871 | task->tk_status = status; | |
872 | task->tk_action = rpc_exit_task; | |
6b5f5900 | 873 | rpc_wake_up_queued_task(task->tk_waitqueue, task); |
d9b6cd94 TM |
874 | } |
875 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 876 | |
bbd5a1f9 TM |
877 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
878 | { | |
a86dc496 | 879 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 880 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
881 | } |
882 | ||
1da177e4 LT |
883 | /* |
884 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
885 | */ | |
2efef837 | 886 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 887 | { |
eb9b55ab TM |
888 | struct rpc_wait_queue *queue; |
889 | int task_is_async = RPC_IS_ASYNC(task); | |
890 | int status = 0; | |
1da177e4 | 891 | |
46121cf7 CL |
892 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
893 | task->tk_pid, task->tk_flags); | |
1da177e4 | 894 | |
2bd4eef8 WAA |
895 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
896 | if (RPC_IS_QUEUED(task)) | |
897 | return; | |
1da177e4 | 898 | |
d05fdb0c | 899 | for (;;) { |
b55c5989 | 900 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
901 | |
902 | /* | |
21ead9ff CL |
903 | * Perform the next FSM step or a pending callback. |
904 | * | |
905 | * tk_action may be NULL if the task has been killed. | |
906 | * In particular, note that rpc_killall_tasks may | |
907 | * do this at any time, so beware when dereferencing. | |
1da177e4 | 908 | */ |
21ead9ff CL |
909 | do_action = task->tk_action; |
910 | if (task->tk_callback) { | |
911 | do_action = task->tk_callback; | |
912 | task->tk_callback = NULL; | |
1da177e4 | 913 | } |
21ead9ff CL |
914 | if (!do_action) |
915 | break; | |
e671edb9 | 916 | trace_rpc_task_run_action(task, do_action); |
b55c5989 | 917 | do_action(task); |
1da177e4 LT |
918 | |
919 | /* | |
920 | * Lockless check for whether task is sleeping or not. | |
921 | */ | |
922 | if (!RPC_IS_QUEUED(task)) | |
923 | continue; | |
ae67bd38 TM |
924 | |
925 | /* | |
926 | * Signalled tasks should exit rather than sleep. | |
927 | */ | |
714fbc73 TM |
928 | if (RPC_SIGNALLED(task)) { |
929 | task->tk_rpc_status = -ERESTARTSYS; | |
ae67bd38 | 930 | rpc_exit(task, -ERESTARTSYS); |
714fbc73 | 931 | } |
ae67bd38 | 932 | |
eb9b55ab TM |
933 | /* |
934 | * The queue->lock protects against races with | |
935 | * rpc_make_runnable(). | |
936 | * | |
937 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
938 | * rpc_task, rpc_make_runnable() can assign it to a | |
939 | * different workqueue. We therefore cannot assume that the | |
940 | * rpc_task pointer may still be dereferenced. | |
941 | */ | |
942 | queue = task->tk_waitqueue; | |
c049f8ea | 943 | spin_lock(&queue->lock); |
eb9b55ab | 944 | if (!RPC_IS_QUEUED(task)) { |
c049f8ea | 945 | spin_unlock(&queue->lock); |
1da177e4 LT |
946 | continue; |
947 | } | |
eb9b55ab | 948 | rpc_clear_running(task); |
c049f8ea | 949 | spin_unlock(&queue->lock); |
eb9b55ab TM |
950 | if (task_is_async) |
951 | return; | |
1da177e4 LT |
952 | |
953 | /* sync task: sleep here */ | |
46121cf7 | 954 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 955 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
956 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
957 | TASK_KILLABLE); | |
ae67bd38 | 958 | if (status < 0) { |
1da177e4 LT |
959 | /* |
960 | * When a sync task receives a signal, it exits with | |
961 | * -ERESTARTSYS. In order to catch any callbacks that | |
962 | * clean up after sleeping on some queue, we don't | |
963 | * break the loop here, but go around once more. | |
964 | */ | |
46121cf7 | 965 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
ae67bd38 | 966 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); |
714fbc73 | 967 | task->tk_rpc_status = -ERESTARTSYS; |
96651ab3 | 968 | rpc_exit(task, -ERESTARTSYS); |
1da177e4 | 969 | } |
46121cf7 | 970 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
971 | } |
972 | ||
46121cf7 CL |
973 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
974 | task->tk_status); | |
1da177e4 LT |
975 | /* Release all resources associated with the task */ |
976 | rpc_release_task(task); | |
1da177e4 LT |
977 | } |
978 | ||
979 | /* | |
980 | * User-visible entry point to the scheduler. | |
981 | * | |
982 | * This may be called recursively if e.g. an async NFS task updates | |
983 | * the attributes and finds that dirty pages must be flushed. | |
984 | * NOTE: Upon exit of this function the task is guaranteed to be | |
985 | * released. In particular note that tk_release() will have | |
986 | * been called, so your task memory may have been freed. | |
987 | */ | |
2efef837 | 988 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 989 | { |
a76580fb TM |
990 | bool is_async = RPC_IS_ASYNC(task); |
991 | ||
44c28873 | 992 | rpc_set_active(task); |
f1dc237c | 993 | rpc_make_runnable(rpciod_workqueue, task); |
dca15c73 BC |
994 | if (!is_async) { |
995 | unsigned int pflags = memalloc_nofs_save(); | |
d6a1ed08 | 996 | __rpc_execute(task); |
dca15c73 BC |
997 | memalloc_nofs_restore(pflags); |
998 | } | |
1da177e4 LT |
999 | } |
1000 | ||
65f27f38 | 1001 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 1002 | { |
a1231fda TM |
1003 | unsigned int pflags = memalloc_nofs_save(); |
1004 | ||
65f27f38 | 1005 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1006 | memalloc_nofs_restore(pflags); |
1da177e4 LT |
1007 | } |
1008 | ||
02107148 | 1009 | /** |
5fe6eaa1 CL |
1010 | * rpc_malloc - allocate RPC buffer resources |
1011 | * @task: RPC task | |
1012 | * | |
1013 | * A single memory region is allocated, which is split between the | |
1014 | * RPC call and RPC reply that this task is being used for. When | |
1015 | * this RPC is retired, the memory is released by calling rpc_free. | |
1da177e4 | 1016 | * |
c5a4dd8b | 1017 | * To prevent rpciod from hanging, this allocator never sleeps, |
5fe6eaa1 CL |
1018 | * returning -ENOMEM and suppressing warning if the request cannot |
1019 | * be serviced immediately. The caller can arrange to sleep in a | |
1020 | * way that is safe for rpciod. | |
c5a4dd8b CL |
1021 | * |
1022 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
1023 | * mempool, ensuring that NFS reads and writes can always proceed, | |
1024 | * and that there is good locality of reference for these buffers. | |
1da177e4 | 1025 | */ |
5fe6eaa1 | 1026 | int rpc_malloc(struct rpc_task *task) |
1da177e4 | 1027 | { |
5fe6eaa1 CL |
1028 | struct rpc_rqst *rqst = task->tk_rqstp; |
1029 | size_t size = rqst->rq_callsize + rqst->rq_rcvsize; | |
aa3d1fae | 1030 | struct rpc_buffer *buf; |
12a3ad61 | 1031 | gfp_t gfp = GFP_NOFS; |
a564b8f0 MG |
1032 | |
1033 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 1034 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 1035 | |
aa3d1fae | 1036 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
1037 | if (size <= RPC_BUFFER_MAXSIZE) |
1038 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 1039 | else |
c5a4dd8b | 1040 | buf = kmalloc(size, gfp); |
ddce40df PZ |
1041 | |
1042 | if (!buf) | |
5fe6eaa1 | 1043 | return -ENOMEM; |
ddce40df | 1044 | |
aa3d1fae | 1045 | buf->len = size; |
215d0678 | 1046 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 1047 | task->tk_pid, size, buf); |
5fe6eaa1 | 1048 | rqst->rq_buffer = buf->data; |
68778945 | 1049 | rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; |
5fe6eaa1 | 1050 | return 0; |
1da177e4 | 1051 | } |
12444809 | 1052 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 1053 | |
02107148 | 1054 | /** |
3435c74a CL |
1055 | * rpc_free - free RPC buffer resources allocated via rpc_malloc |
1056 | * @task: RPC task | |
02107148 CL |
1057 | * |
1058 | */ | |
3435c74a | 1059 | void rpc_free(struct rpc_task *task) |
1da177e4 | 1060 | { |
3435c74a | 1061 | void *buffer = task->tk_rqstp->rq_buffer; |
aa3d1fae CL |
1062 | size_t size; |
1063 | struct rpc_buffer *buf; | |
02107148 | 1064 | |
aa3d1fae CL |
1065 | buf = container_of(buffer, struct rpc_buffer, data); |
1066 | size = buf->len; | |
c5a4dd8b | 1067 | |
215d0678 | 1068 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 1069 | size, buf); |
aa3d1fae | 1070 | |
c5a4dd8b CL |
1071 | if (size <= RPC_BUFFER_MAXSIZE) |
1072 | mempool_free(buf, rpc_buffer_mempool); | |
1073 | else | |
1074 | kfree(buf); | |
1da177e4 | 1075 | } |
12444809 | 1076 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
1077 | |
1078 | /* | |
1079 | * Creation and deletion of RPC task structures | |
1080 | */ | |
47fe0648 | 1081 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
1082 | { |
1083 | memset(task, 0, sizeof(*task)); | |
44c28873 | 1084 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
1085 | task->tk_flags = task_setup_data->flags; |
1086 | task->tk_ops = task_setup_data->callback_ops; | |
1087 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 1088 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 1089 | |
3ff7576d TM |
1090 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
1091 | task->tk_owner = current->tgid; | |
1da177e4 LT |
1092 | |
1093 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 1094 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 1095 | |
a101b043 TM |
1096 | task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client, |
1097 | xprt_get(task_setup_data->rpc_xprt)); | |
9d61498d | 1098 | |
1de7eea9 N |
1099 | task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred); |
1100 | ||
84115e1c TM |
1101 | if (task->tk_ops->rpc_call_prepare != NULL) |
1102 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 1103 | |
7fdcf13b | 1104 | rpc_init_task_statistics(task); |
ef759a2e | 1105 | |
46121cf7 | 1106 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 1107 | task_pid_nr(current)); |
1da177e4 LT |
1108 | } |
1109 | ||
1110 | static struct rpc_task * | |
1111 | rpc_alloc_task(void) | |
1112 | { | |
12a3ad61 | 1113 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); |
1da177e4 LT |
1114 | } |
1115 | ||
1da177e4 | 1116 | /* |
90c5755f | 1117 | * Create a new task for the specified client. |
1da177e4 | 1118 | */ |
84115e1c | 1119 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 1120 | { |
e8f5d77c TM |
1121 | struct rpc_task *task = setup_data->task; |
1122 | unsigned short flags = 0; | |
1123 | ||
1124 | if (task == NULL) { | |
1125 | task = rpc_alloc_task(); | |
e8f5d77c TM |
1126 | flags = RPC_TASK_DYNAMIC; |
1127 | } | |
1da177e4 | 1128 | |
84115e1c | 1129 | rpc_init_task(task, setup_data); |
e8f5d77c | 1130 | task->tk_flags |= flags; |
46121cf7 | 1131 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 1132 | return task; |
1da177e4 LT |
1133 | } |
1134 | ||
c6567ed1 TM |
1135 | /* |
1136 | * rpc_free_task - release rpc task and perform cleanups | |
1137 | * | |
1138 | * Note that we free up the rpc_task _after_ rpc_release_calldata() | |
1139 | * in order to work around a workqueue dependency issue. | |
1140 | * | |
1141 | * Tejun Heo states: | |
1142 | * "Workqueue currently considers two work items to be the same if they're | |
1143 | * on the same address and won't execute them concurrently - ie. it | |
1144 | * makes a work item which is queued again while being executed wait | |
1145 | * for the previous execution to complete. | |
1146 | * | |
1147 | * If a work function frees the work item, and then waits for an event | |
1148 | * which should be performed by another work item and *that* work item | |
1149 | * recycles the freed work item, it can create a false dependency loop. | |
1150 | * There really is no reliable way to detect this short of verifying | |
1151 | * every memory free." | |
1152 | * | |
1153 | */ | |
32bfb5c0 | 1154 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 1155 | { |
c6567ed1 TM |
1156 | unsigned short tk_flags = task->tk_flags; |
1157 | ||
1de7eea9 | 1158 | put_rpccred(task->tk_op_cred); |
c6567ed1 | 1159 | rpc_release_calldata(task->tk_ops, task->tk_calldata); |
1da177e4 | 1160 | |
c6567ed1 | 1161 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
1162 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
1163 | mempool_free(task, rpc_task_mempool); | |
1164 | } | |
32bfb5c0 TM |
1165 | } |
1166 | ||
1167 | static void rpc_async_release(struct work_struct *work) | |
1168 | { | |
a1231fda TM |
1169 | unsigned int pflags = memalloc_nofs_save(); |
1170 | ||
32bfb5c0 | 1171 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1172 | memalloc_nofs_restore(pflags); |
32bfb5c0 TM |
1173 | } |
1174 | ||
bf294b41 | 1175 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 1176 | { |
87ed5003 | 1177 | xprt_release(task); |
a271c5a0 | 1178 | if (task->tk_msg.rpc_cred) { |
a52458b4 | 1179 | put_cred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
1180 | task->tk_msg.rpc_cred = NULL; |
1181 | } | |
58f9612c | 1182 | rpc_task_release_client(task); |
bf294b41 TM |
1183 | } |
1184 | ||
1185 | static void rpc_final_put_task(struct rpc_task *task, | |
1186 | struct workqueue_struct *q) | |
1187 | { | |
1188 | if (q != NULL) { | |
32bfb5c0 | 1189 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1190 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1191 | } else |
1192 | rpc_free_task(task); | |
e6b3c4db | 1193 | } |
bf294b41 TM |
1194 | |
1195 | static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q) | |
1196 | { | |
1197 | if (atomic_dec_and_test(&task->tk_count)) { | |
1198 | rpc_release_resources_task(task); | |
1199 | rpc_final_put_task(task, q); | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | void rpc_put_task(struct rpc_task *task) | |
1204 | { | |
1205 | rpc_do_put_task(task, NULL); | |
1206 | } | |
e8914c65 | 1207 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1208 | |
bf294b41 TM |
1209 | void rpc_put_task_async(struct rpc_task *task) |
1210 | { | |
1211 | rpc_do_put_task(task, task->tk_workqueue); | |
1212 | } | |
1213 | EXPORT_SYMBOL_GPL(rpc_put_task_async); | |
1214 | ||
bde8f00c | 1215 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1216 | { |
46121cf7 | 1217 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1218 | |
0a0c2a57 | 1219 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1220 | |
bf294b41 | 1221 | rpc_release_resources_task(task); |
e6b3c4db | 1222 | |
bf294b41 TM |
1223 | /* |
1224 | * Note: at this point we have been removed from rpc_clnt->cl_tasks, | |
1225 | * so it should be safe to use task->tk_count as a test for whether | |
1226 | * or not any other processes still hold references to our rpc_task. | |
1227 | */ | |
1228 | if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) { | |
1229 | /* Wake up anyone who may be waiting for task completion */ | |
1230 | if (!rpc_complete_task(task)) | |
1231 | return; | |
1232 | } else { | |
1233 | if (!atomic_dec_and_test(&task->tk_count)) | |
1234 | return; | |
1235 | } | |
1236 | rpc_final_put_task(task, task->tk_workqueue); | |
1da177e4 LT |
1237 | } |
1238 | ||
b247bbf1 TM |
1239 | int rpciod_up(void) |
1240 | { | |
1241 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
1242 | } | |
1243 | ||
1244 | void rpciod_down(void) | |
1245 | { | |
1246 | module_put(THIS_MODULE); | |
1247 | } | |
1248 | ||
1da177e4 | 1249 | /* |
b247bbf1 | 1250 | * Start up the rpciod workqueue. |
1da177e4 | 1251 | */ |
b247bbf1 | 1252 | static int rpciod_start(void) |
1da177e4 LT |
1253 | { |
1254 | struct workqueue_struct *wq; | |
ab418d70 | 1255 | |
1da177e4 LT |
1256 | /* |
1257 | * Create the rpciod thread and wait for it to start. | |
1258 | */ | |
ab418d70 | 1259 | dprintk("RPC: creating workqueue rpciod\n"); |
f515f86b | 1260 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); |
40a5f1b1 TM |
1261 | if (!wq) |
1262 | goto out_failed; | |
1da177e4 | 1263 | rpciod_workqueue = wq; |
40a5f1b1 | 1264 | /* Note: highpri because network receive is latency sensitive */ |
90ea9f1b | 1265 | wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0); |
40a5f1b1 TM |
1266 | if (!wq) |
1267 | goto free_rpciod; | |
1268 | xprtiod_workqueue = wq; | |
1269 | return 1; | |
1270 | free_rpciod: | |
1271 | wq = rpciod_workqueue; | |
1272 | rpciod_workqueue = NULL; | |
1273 | destroy_workqueue(wq); | |
1274 | out_failed: | |
1275 | return 0; | |
1da177e4 LT |
1276 | } |
1277 | ||
b247bbf1 | 1278 | static void rpciod_stop(void) |
1da177e4 | 1279 | { |
b247bbf1 | 1280 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1281 | |
b247bbf1 TM |
1282 | if (rpciod_workqueue == NULL) |
1283 | return; | |
ab418d70 | 1284 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1285 | |
b247bbf1 TM |
1286 | wq = rpciod_workqueue; |
1287 | rpciod_workqueue = NULL; | |
1288 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1289 | wq = xprtiod_workqueue; |
1290 | xprtiod_workqueue = NULL; | |
1291 | destroy_workqueue(wq); | |
1da177e4 LT |
1292 | } |
1293 | ||
1da177e4 LT |
1294 | void |
1295 | rpc_destroy_mempool(void) | |
1296 | { | |
b247bbf1 | 1297 | rpciod_stop(); |
17a9618e JL |
1298 | mempool_destroy(rpc_buffer_mempool); |
1299 | mempool_destroy(rpc_task_mempool); | |
1300 | kmem_cache_destroy(rpc_task_slabp); | |
1301 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 1302 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1303 | } |
1304 | ||
1305 | int | |
1306 | rpc_init_mempool(void) | |
1307 | { | |
f6a1cc89 TM |
1308 | /* |
1309 | * The following is not strictly a mempool initialisation, | |
1310 | * but there is no harm in doing it here | |
1311 | */ | |
1312 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
1313 | if (!rpciod_start()) | |
1314 | goto err_nomem; | |
1315 | ||
1da177e4 LT |
1316 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1317 | sizeof(struct rpc_task), | |
1318 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1319 | NULL); |
1da177e4 LT |
1320 | if (!rpc_task_slabp) |
1321 | goto err_nomem; | |
1322 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
1323 | RPC_BUFFER_MAXSIZE, | |
1324 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1325 | NULL); |
1da177e4 LT |
1326 | if (!rpc_buffer_slabp) |
1327 | goto err_nomem; | |
93d2341c MD |
1328 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1329 | rpc_task_slabp); | |
1da177e4 LT |
1330 | if (!rpc_task_mempool) |
1331 | goto err_nomem; | |
93d2341c MD |
1332 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1333 | rpc_buffer_slabp); | |
1da177e4 LT |
1334 | if (!rpc_buffer_mempool) |
1335 | goto err_nomem; | |
1336 | return 0; | |
1337 | err_nomem: | |
1338 | rpc_destroy_mempool(); | |
1339 | return -ENOMEM; | |
1340 | } |