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 | 262 | queue->timer_list.expires = 0; |
66eb3add | 263 | INIT_DELAYED_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 | ||
f1dc237c TM |
544 | /* |
545 | * Wake up a queued task while the queue lock is being held | |
546 | */ | |
691b45dd CL |
547 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, |
548 | struct rpc_task *task) | |
f1dc237c | 549 | { |
691b45dd CL |
550 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, |
551 | task, NULL, NULL); | |
2275cde4 TM |
552 | } |
553 | ||
1da177e4 | 554 | /* |
96ef13b2 | 555 | * Wake up a task on a specific queue |
1da177e4 | 556 | */ |
96ef13b2 | 557 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 558 | { |
5ce97039 TM |
559 | if (!RPC_IS_QUEUED(task)) |
560 | return; | |
c049f8ea | 561 | spin_lock(&queue->lock); |
96ef13b2 | 562 | rpc_wake_up_task_queue_locked(queue, task); |
c049f8ea | 563 | spin_unlock(&queue->lock); |
1da177e4 | 564 | } |
96ef13b2 TM |
565 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
566 | ||
359c48c0 TM |
567 | static bool rpc_task_action_set_status(struct rpc_task *task, void *status) |
568 | { | |
569 | task->tk_status = *(int *)status; | |
570 | return true; | |
571 | } | |
572 | ||
573 | static void | |
574 | rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue, | |
575 | struct rpc_task *task, int status) | |
576 | { | |
577 | rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue, | |
578 | task, rpc_task_action_set_status, &status); | |
579 | } | |
580 | ||
581 | /** | |
582 | * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status | |
583 | * @queue: pointer to rpc_wait_queue | |
584 | * @task: pointer to rpc_task | |
585 | * @status: integer error value | |
586 | * | |
587 | * If @task is queued on @queue, then it is woken up, and @task->tk_status is | |
588 | * set to the value of @status. | |
589 | */ | |
590 | void | |
591 | rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue, | |
592 | struct rpc_task *task, int status) | |
593 | { | |
594 | if (!RPC_IS_QUEUED(task)) | |
595 | return; | |
c049f8ea | 596 | spin_lock(&queue->lock); |
359c48c0 | 597 | rpc_wake_up_task_queue_set_status_locked(queue, task, status); |
c049f8ea | 598 | spin_unlock(&queue->lock); |
359c48c0 TM |
599 | } |
600 | ||
1da177e4 LT |
601 | /* |
602 | * Wake up the next task on a priority queue. | |
603 | */ | |
961a828d | 604 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
605 | { |
606 | struct list_head *q; | |
607 | struct rpc_task *task; | |
608 | ||
609 | /* | |
3ff7576d | 610 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
611 | */ |
612 | q = &queue->tasks[queue->priority]; | |
f42f7c28 TM |
613 | if (!list_empty(q) && --queue->nr) { |
614 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); | |
615 | goto out; | |
1da177e4 LT |
616 | } |
617 | ||
618 | /* | |
619 | * Service the next queue. | |
620 | */ | |
621 | do { | |
622 | if (q == &queue->tasks[0]) | |
623 | q = &queue->tasks[queue->maxpriority]; | |
624 | else | |
625 | q = q - 1; | |
626 | if (!list_empty(q)) { | |
f42f7c28 | 627 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); |
1da177e4 LT |
628 | goto new_queue; |
629 | } | |
630 | } while (q != &queue->tasks[queue->priority]); | |
631 | ||
632 | rpc_reset_waitqueue_priority(queue); | |
633 | return NULL; | |
634 | ||
635 | new_queue: | |
636 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
1da177e4 | 637 | out: |
1da177e4 LT |
638 | return task; |
639 | } | |
640 | ||
961a828d TM |
641 | static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue) |
642 | { | |
643 | if (RPC_IS_PRIORITY(queue)) | |
644 | return __rpc_find_next_queued_priority(queue); | |
645 | if (!list_empty(&queue->tasks[0])) | |
646 | return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list); | |
647 | return NULL; | |
648 | } | |
649 | ||
1da177e4 | 650 | /* |
961a828d | 651 | * Wake up the first task on the wait queue. |
1da177e4 | 652 | */ |
f1dc237c TM |
653 | struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq, |
654 | struct rpc_wait_queue *queue, | |
961a828d | 655 | bool (*func)(struct rpc_task *, void *), void *data) |
1da177e4 LT |
656 | { |
657 | struct rpc_task *task = NULL; | |
658 | ||
961a828d | 659 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 660 | queue, rpc_qname(queue)); |
c049f8ea | 661 | spin_lock(&queue->lock); |
961a828d | 662 | task = __rpc_find_next_queued(queue); |
359c48c0 TM |
663 | if (task != NULL) |
664 | task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue, | |
665 | task, func, data); | |
c049f8ea | 666 | spin_unlock(&queue->lock); |
1da177e4 LT |
667 | |
668 | return task; | |
669 | } | |
f1dc237c TM |
670 | |
671 | /* | |
672 | * Wake up the first task on the wait queue. | |
673 | */ | |
674 | struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue, | |
675 | bool (*func)(struct rpc_task *, void *), void *data) | |
676 | { | |
677 | return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data); | |
678 | } | |
961a828d TM |
679 | EXPORT_SYMBOL_GPL(rpc_wake_up_first); |
680 | ||
681 | static bool rpc_wake_up_next_func(struct rpc_task *task, void *data) | |
682 | { | |
683 | return true; | |
684 | } | |
685 | ||
686 | /* | |
687 | * Wake up the next task on the wait queue. | |
688 | */ | |
689 | struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue) | |
690 | { | |
691 | return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL); | |
692 | } | |
e8914c65 | 693 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
694 | |
695 | /** | |
696 | * rpc_wake_up - wake up all rpc_tasks | |
697 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
698 | * | |
699 | * Grabs queue->lock | |
700 | */ | |
701 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
702 | { | |
1da177e4 | 703 | struct list_head *head; |
e6d83d55 | 704 | |
c049f8ea | 705 | spin_lock(&queue->lock); |
1da177e4 LT |
706 | head = &queue->tasks[queue->maxpriority]; |
707 | for (;;) { | |
540a0f75 TM |
708 | while (!list_empty(head)) { |
709 | struct rpc_task *task; | |
710 | task = list_first_entry(head, | |
711 | struct rpc_task, | |
712 | u.tk_wait.list); | |
96ef13b2 | 713 | rpc_wake_up_task_queue_locked(queue, task); |
540a0f75 | 714 | } |
1da177e4 LT |
715 | if (head == &queue->tasks[0]) |
716 | break; | |
717 | head--; | |
718 | } | |
c049f8ea | 719 | spin_unlock(&queue->lock); |
1da177e4 | 720 | } |
e8914c65 | 721 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
722 | |
723 | /** | |
724 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
725 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
726 | * @status: status value to set | |
727 | * | |
728 | * Grabs queue->lock | |
729 | */ | |
730 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
731 | { | |
732 | struct list_head *head; | |
1da177e4 | 733 | |
c049f8ea | 734 | spin_lock(&queue->lock); |
1da177e4 LT |
735 | head = &queue->tasks[queue->maxpriority]; |
736 | for (;;) { | |
540a0f75 TM |
737 | while (!list_empty(head)) { |
738 | struct rpc_task *task; | |
739 | task = list_first_entry(head, | |
740 | struct rpc_task, | |
741 | u.tk_wait.list); | |
1da177e4 | 742 | task->tk_status = status; |
96ef13b2 | 743 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
744 | } |
745 | if (head == &queue->tasks[0]) | |
746 | break; | |
747 | head--; | |
748 | } | |
c049f8ea | 749 | spin_unlock(&queue->lock); |
1da177e4 | 750 | } |
e8914c65 | 751 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 752 | |
7e0a0e38 | 753 | static void __rpc_queue_timer_fn(struct work_struct *work) |
36df9aae | 754 | { |
7e0a0e38 TM |
755 | struct rpc_wait_queue *queue = container_of(work, |
756 | struct rpc_wait_queue, | |
757 | timer_list.dwork.work); | |
36df9aae TM |
758 | struct rpc_task *task, *n; |
759 | unsigned long expires, now, timeo; | |
760 | ||
761 | spin_lock(&queue->lock); | |
762 | expires = now = jiffies; | |
763 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
6b2e6856 | 764 | timeo = task->tk_timeout; |
36df9aae | 765 | if (time_after_eq(now, timeo)) { |
36df9aae TM |
766 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
767 | task->tk_status = -ETIMEDOUT; | |
768 | rpc_wake_up_task_queue_locked(queue, task); | |
769 | continue; | |
770 | } | |
771 | if (expires == now || time_after(expires, timeo)) | |
772 | expires = timeo; | |
773 | } | |
774 | if (!list_empty(&queue->timer_list.list)) | |
775 | rpc_set_queue_timer(queue, expires); | |
776 | spin_unlock(&queue->lock); | |
777 | } | |
778 | ||
8014793b TM |
779 | static void __rpc_atrun(struct rpc_task *task) |
780 | { | |
6bd14416 TM |
781 | if (task->tk_status == -ETIMEDOUT) |
782 | task->tk_status = 0; | |
8014793b TM |
783 | } |
784 | ||
1da177e4 LT |
785 | /* |
786 | * Run a task at a later time | |
787 | */ | |
8014793b | 788 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 | 789 | { |
6b2e6856 | 790 | rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay); |
1da177e4 | 791 | } |
e8914c65 | 792 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 793 | |
4ce70ada TM |
794 | /* |
795 | * Helper to call task->tk_ops->rpc_call_prepare | |
796 | */ | |
aae2006e | 797 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
798 | { |
799 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
800 | } | |
801 | ||
7fdcf13b TM |
802 | static void |
803 | rpc_init_task_statistics(struct rpc_task *task) | |
804 | { | |
805 | /* Initialize retry counters */ | |
806 | task->tk_garb_retry = 2; | |
807 | task->tk_cred_retry = 2; | |
808 | task->tk_rebind_retry = 2; | |
809 | ||
810 | /* starting timestamp */ | |
811 | task->tk_start = ktime_get(); | |
812 | } | |
813 | ||
814 | static void | |
815 | rpc_reset_task_statistics(struct rpc_task *task) | |
816 | { | |
817 | task->tk_timeouts = 0; | |
ae67bd38 | 818 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT); |
7fdcf13b TM |
819 | rpc_init_task_statistics(task); |
820 | } | |
821 | ||
d05fdb0c | 822 | /* |
963d8fe5 | 823 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 824 | */ |
abbcf28f | 825 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 826 | { |
a264abad | 827 | trace_rpc_task_end(task, task->tk_action); |
abbcf28f | 828 | task->tk_action = NULL; |
9dfe52a9 DW |
829 | if (task->tk_ops->rpc_count_stats) |
830 | task->tk_ops->rpc_count_stats(task, task->tk_calldata); | |
831 | else if (task->tk_client) | |
832 | rpc_count_iostats(task, task->tk_client->cl_metrics); | |
963d8fe5 TM |
833 | if (task->tk_ops->rpc_call_done != NULL) { |
834 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 835 | if (task->tk_action != NULL) { |
abbcf28f TM |
836 | /* Always release the RPC slot and buffer memory */ |
837 | xprt_release(task); | |
7fdcf13b | 838 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
839 | } |
840 | } | |
d05fdb0c | 841 | } |
d9b6cd94 | 842 | |
ae67bd38 TM |
843 | void rpc_signal_task(struct rpc_task *task) |
844 | { | |
845 | struct rpc_wait_queue *queue; | |
846 | ||
847 | if (!RPC_IS_ACTIVATED(task)) | |
848 | return; | |
849 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); | |
850 | smp_mb__after_atomic(); | |
851 | queue = READ_ONCE(task->tk_waitqueue); | |
852 | if (queue) | |
853 | rpc_wake_up_queued_task_set_status(queue, task, -ERESTARTSYS); | |
854 | } | |
855 | ||
d9b6cd94 TM |
856 | void rpc_exit(struct rpc_task *task, int status) |
857 | { | |
858 | task->tk_status = status; | |
859 | task->tk_action = rpc_exit_task; | |
6b5f5900 | 860 | rpc_wake_up_queued_task(task->tk_waitqueue, task); |
d9b6cd94 TM |
861 | } |
862 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 863 | |
bbd5a1f9 TM |
864 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
865 | { | |
a86dc496 | 866 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 867 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
868 | } |
869 | ||
1da177e4 LT |
870 | /* |
871 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
872 | */ | |
2efef837 | 873 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 874 | { |
eb9b55ab TM |
875 | struct rpc_wait_queue *queue; |
876 | int task_is_async = RPC_IS_ASYNC(task); | |
877 | int status = 0; | |
1da177e4 | 878 | |
46121cf7 CL |
879 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
880 | task->tk_pid, task->tk_flags); | |
1da177e4 | 881 | |
2bd4eef8 WAA |
882 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
883 | if (RPC_IS_QUEUED(task)) | |
884 | return; | |
1da177e4 | 885 | |
d05fdb0c | 886 | for (;;) { |
b55c5989 | 887 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
888 | |
889 | /* | |
21ead9ff CL |
890 | * Perform the next FSM step or a pending callback. |
891 | * | |
892 | * tk_action may be NULL if the task has been killed. | |
893 | * In particular, note that rpc_killall_tasks may | |
894 | * do this at any time, so beware when dereferencing. | |
1da177e4 | 895 | */ |
21ead9ff CL |
896 | do_action = task->tk_action; |
897 | if (task->tk_callback) { | |
898 | do_action = task->tk_callback; | |
899 | task->tk_callback = NULL; | |
1da177e4 | 900 | } |
21ead9ff CL |
901 | if (!do_action) |
902 | break; | |
e671edb9 | 903 | trace_rpc_task_run_action(task, do_action); |
b55c5989 | 904 | do_action(task); |
1da177e4 LT |
905 | |
906 | /* | |
907 | * Lockless check for whether task is sleeping or not. | |
908 | */ | |
909 | if (!RPC_IS_QUEUED(task)) | |
910 | continue; | |
ae67bd38 TM |
911 | |
912 | /* | |
913 | * Signalled tasks should exit rather than sleep. | |
914 | */ | |
714fbc73 TM |
915 | if (RPC_SIGNALLED(task)) { |
916 | task->tk_rpc_status = -ERESTARTSYS; | |
ae67bd38 | 917 | rpc_exit(task, -ERESTARTSYS); |
714fbc73 | 918 | } |
ae67bd38 | 919 | |
eb9b55ab TM |
920 | /* |
921 | * The queue->lock protects against races with | |
922 | * rpc_make_runnable(). | |
923 | * | |
924 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
925 | * rpc_task, rpc_make_runnable() can assign it to a | |
926 | * different workqueue. We therefore cannot assume that the | |
927 | * rpc_task pointer may still be dereferenced. | |
928 | */ | |
929 | queue = task->tk_waitqueue; | |
c049f8ea | 930 | spin_lock(&queue->lock); |
eb9b55ab | 931 | if (!RPC_IS_QUEUED(task)) { |
c049f8ea | 932 | spin_unlock(&queue->lock); |
1da177e4 LT |
933 | continue; |
934 | } | |
eb9b55ab | 935 | rpc_clear_running(task); |
c049f8ea | 936 | spin_unlock(&queue->lock); |
eb9b55ab TM |
937 | if (task_is_async) |
938 | return; | |
1da177e4 LT |
939 | |
940 | /* sync task: sleep here */ | |
46121cf7 | 941 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 942 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
943 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
944 | TASK_KILLABLE); | |
ae67bd38 | 945 | if (status < 0) { |
1da177e4 LT |
946 | /* |
947 | * When a sync task receives a signal, it exits with | |
948 | * -ERESTARTSYS. In order to catch any callbacks that | |
949 | * clean up after sleeping on some queue, we don't | |
950 | * break the loop here, but go around once more. | |
951 | */ | |
46121cf7 | 952 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
ae67bd38 | 953 | set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate); |
714fbc73 | 954 | task->tk_rpc_status = -ERESTARTSYS; |
96651ab3 | 955 | rpc_exit(task, -ERESTARTSYS); |
1da177e4 | 956 | } |
46121cf7 | 957 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
958 | } |
959 | ||
46121cf7 CL |
960 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
961 | task->tk_status); | |
1da177e4 LT |
962 | /* Release all resources associated with the task */ |
963 | rpc_release_task(task); | |
1da177e4 LT |
964 | } |
965 | ||
966 | /* | |
967 | * User-visible entry point to the scheduler. | |
968 | * | |
969 | * This may be called recursively if e.g. an async NFS task updates | |
970 | * the attributes and finds that dirty pages must be flushed. | |
971 | * NOTE: Upon exit of this function the task is guaranteed to be | |
972 | * released. In particular note that tk_release() will have | |
973 | * been called, so your task memory may have been freed. | |
974 | */ | |
2efef837 | 975 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 976 | { |
a76580fb TM |
977 | bool is_async = RPC_IS_ASYNC(task); |
978 | ||
44c28873 | 979 | rpc_set_active(task); |
f1dc237c | 980 | rpc_make_runnable(rpciod_workqueue, task); |
a76580fb | 981 | if (!is_async) |
d6a1ed08 | 982 | __rpc_execute(task); |
1da177e4 LT |
983 | } |
984 | ||
65f27f38 | 985 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 986 | { |
a1231fda TM |
987 | unsigned int pflags = memalloc_nofs_save(); |
988 | ||
65f27f38 | 989 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 990 | memalloc_nofs_restore(pflags); |
1da177e4 LT |
991 | } |
992 | ||
02107148 | 993 | /** |
5fe6eaa1 CL |
994 | * rpc_malloc - allocate RPC buffer resources |
995 | * @task: RPC task | |
996 | * | |
997 | * A single memory region is allocated, which is split between the | |
998 | * RPC call and RPC reply that this task is being used for. When | |
999 | * this RPC is retired, the memory is released by calling rpc_free. | |
1da177e4 | 1000 | * |
c5a4dd8b | 1001 | * To prevent rpciod from hanging, this allocator never sleeps, |
5fe6eaa1 CL |
1002 | * returning -ENOMEM and suppressing warning if the request cannot |
1003 | * be serviced immediately. The caller can arrange to sleep in a | |
1004 | * way that is safe for rpciod. | |
c5a4dd8b CL |
1005 | * |
1006 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
1007 | * mempool, ensuring that NFS reads and writes can always proceed, | |
1008 | * and that there is good locality of reference for these buffers. | |
1da177e4 | 1009 | */ |
5fe6eaa1 | 1010 | int rpc_malloc(struct rpc_task *task) |
1da177e4 | 1011 | { |
5fe6eaa1 CL |
1012 | struct rpc_rqst *rqst = task->tk_rqstp; |
1013 | size_t size = rqst->rq_callsize + rqst->rq_rcvsize; | |
aa3d1fae | 1014 | struct rpc_buffer *buf; |
12a3ad61 | 1015 | gfp_t gfp = GFP_NOFS; |
a564b8f0 MG |
1016 | |
1017 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 1018 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 1019 | |
aa3d1fae | 1020 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
1021 | if (size <= RPC_BUFFER_MAXSIZE) |
1022 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 1023 | else |
c5a4dd8b | 1024 | buf = kmalloc(size, gfp); |
ddce40df PZ |
1025 | |
1026 | if (!buf) | |
5fe6eaa1 | 1027 | return -ENOMEM; |
ddce40df | 1028 | |
aa3d1fae | 1029 | buf->len = size; |
215d0678 | 1030 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 1031 | task->tk_pid, size, buf); |
5fe6eaa1 | 1032 | rqst->rq_buffer = buf->data; |
68778945 | 1033 | rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize; |
5fe6eaa1 | 1034 | return 0; |
1da177e4 | 1035 | } |
12444809 | 1036 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 1037 | |
02107148 | 1038 | /** |
3435c74a CL |
1039 | * rpc_free - free RPC buffer resources allocated via rpc_malloc |
1040 | * @task: RPC task | |
02107148 CL |
1041 | * |
1042 | */ | |
3435c74a | 1043 | void rpc_free(struct rpc_task *task) |
1da177e4 | 1044 | { |
3435c74a | 1045 | void *buffer = task->tk_rqstp->rq_buffer; |
aa3d1fae CL |
1046 | size_t size; |
1047 | struct rpc_buffer *buf; | |
02107148 | 1048 | |
aa3d1fae CL |
1049 | buf = container_of(buffer, struct rpc_buffer, data); |
1050 | size = buf->len; | |
c5a4dd8b | 1051 | |
215d0678 | 1052 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 1053 | size, buf); |
aa3d1fae | 1054 | |
c5a4dd8b CL |
1055 | if (size <= RPC_BUFFER_MAXSIZE) |
1056 | mempool_free(buf, rpc_buffer_mempool); | |
1057 | else | |
1058 | kfree(buf); | |
1da177e4 | 1059 | } |
12444809 | 1060 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
1061 | |
1062 | /* | |
1063 | * Creation and deletion of RPC task structures | |
1064 | */ | |
47fe0648 | 1065 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
1066 | { |
1067 | memset(task, 0, sizeof(*task)); | |
44c28873 | 1068 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
1069 | task->tk_flags = task_setup_data->flags; |
1070 | task->tk_ops = task_setup_data->callback_ops; | |
1071 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 1072 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 1073 | |
3ff7576d TM |
1074 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
1075 | task->tk_owner = current->tgid; | |
1da177e4 LT |
1076 | |
1077 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 1078 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 1079 | |
a101b043 TM |
1080 | task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client, |
1081 | xprt_get(task_setup_data->rpc_xprt)); | |
9d61498d | 1082 | |
1de7eea9 N |
1083 | task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred); |
1084 | ||
84115e1c TM |
1085 | if (task->tk_ops->rpc_call_prepare != NULL) |
1086 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 1087 | |
7fdcf13b | 1088 | rpc_init_task_statistics(task); |
ef759a2e | 1089 | |
46121cf7 | 1090 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 1091 | task_pid_nr(current)); |
1da177e4 LT |
1092 | } |
1093 | ||
1094 | static struct rpc_task * | |
1095 | rpc_alloc_task(void) | |
1096 | { | |
12a3ad61 | 1097 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); |
1da177e4 LT |
1098 | } |
1099 | ||
1da177e4 | 1100 | /* |
90c5755f | 1101 | * Create a new task for the specified client. |
1da177e4 | 1102 | */ |
84115e1c | 1103 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 1104 | { |
e8f5d77c TM |
1105 | struct rpc_task *task = setup_data->task; |
1106 | unsigned short flags = 0; | |
1107 | ||
1108 | if (task == NULL) { | |
1109 | task = rpc_alloc_task(); | |
e8f5d77c TM |
1110 | flags = RPC_TASK_DYNAMIC; |
1111 | } | |
1da177e4 | 1112 | |
84115e1c | 1113 | rpc_init_task(task, setup_data); |
e8f5d77c | 1114 | task->tk_flags |= flags; |
46121cf7 | 1115 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 1116 | return task; |
1da177e4 LT |
1117 | } |
1118 | ||
c6567ed1 TM |
1119 | /* |
1120 | * rpc_free_task - release rpc task and perform cleanups | |
1121 | * | |
1122 | * Note that we free up the rpc_task _after_ rpc_release_calldata() | |
1123 | * in order to work around a workqueue dependency issue. | |
1124 | * | |
1125 | * Tejun Heo states: | |
1126 | * "Workqueue currently considers two work items to be the same if they're | |
1127 | * on the same address and won't execute them concurrently - ie. it | |
1128 | * makes a work item which is queued again while being executed wait | |
1129 | * for the previous execution to complete. | |
1130 | * | |
1131 | * If a work function frees the work item, and then waits for an event | |
1132 | * which should be performed by another work item and *that* work item | |
1133 | * recycles the freed work item, it can create a false dependency loop. | |
1134 | * There really is no reliable way to detect this short of verifying | |
1135 | * every memory free." | |
1136 | * | |
1137 | */ | |
32bfb5c0 | 1138 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 1139 | { |
c6567ed1 TM |
1140 | unsigned short tk_flags = task->tk_flags; |
1141 | ||
1de7eea9 | 1142 | put_rpccred(task->tk_op_cred); |
c6567ed1 | 1143 | rpc_release_calldata(task->tk_ops, task->tk_calldata); |
1da177e4 | 1144 | |
c6567ed1 | 1145 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
1146 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
1147 | mempool_free(task, rpc_task_mempool); | |
1148 | } | |
32bfb5c0 TM |
1149 | } |
1150 | ||
1151 | static void rpc_async_release(struct work_struct *work) | |
1152 | { | |
a1231fda TM |
1153 | unsigned int pflags = memalloc_nofs_save(); |
1154 | ||
32bfb5c0 | 1155 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); |
a1231fda | 1156 | memalloc_nofs_restore(pflags); |
32bfb5c0 TM |
1157 | } |
1158 | ||
bf294b41 | 1159 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 1160 | { |
87ed5003 | 1161 | xprt_release(task); |
a271c5a0 | 1162 | if (task->tk_msg.rpc_cred) { |
a52458b4 | 1163 | put_cred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
1164 | task->tk_msg.rpc_cred = NULL; |
1165 | } | |
58f9612c | 1166 | rpc_task_release_client(task); |
bf294b41 TM |
1167 | } |
1168 | ||
1169 | static void rpc_final_put_task(struct rpc_task *task, | |
1170 | struct workqueue_struct *q) | |
1171 | { | |
1172 | if (q != NULL) { | |
32bfb5c0 | 1173 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1174 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1175 | } else |
1176 | rpc_free_task(task); | |
e6b3c4db | 1177 | } |
bf294b41 TM |
1178 | |
1179 | static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q) | |
1180 | { | |
1181 | if (atomic_dec_and_test(&task->tk_count)) { | |
1182 | rpc_release_resources_task(task); | |
1183 | rpc_final_put_task(task, q); | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | void rpc_put_task(struct rpc_task *task) | |
1188 | { | |
1189 | rpc_do_put_task(task, NULL); | |
1190 | } | |
e8914c65 | 1191 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1192 | |
bf294b41 TM |
1193 | void rpc_put_task_async(struct rpc_task *task) |
1194 | { | |
1195 | rpc_do_put_task(task, task->tk_workqueue); | |
1196 | } | |
1197 | EXPORT_SYMBOL_GPL(rpc_put_task_async); | |
1198 | ||
bde8f00c | 1199 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1200 | { |
46121cf7 | 1201 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1202 | |
0a0c2a57 | 1203 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1204 | |
bf294b41 | 1205 | rpc_release_resources_task(task); |
e6b3c4db | 1206 | |
bf294b41 TM |
1207 | /* |
1208 | * Note: at this point we have been removed from rpc_clnt->cl_tasks, | |
1209 | * so it should be safe to use task->tk_count as a test for whether | |
1210 | * or not any other processes still hold references to our rpc_task. | |
1211 | */ | |
1212 | if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) { | |
1213 | /* Wake up anyone who may be waiting for task completion */ | |
1214 | if (!rpc_complete_task(task)) | |
1215 | return; | |
1216 | } else { | |
1217 | if (!atomic_dec_and_test(&task->tk_count)) | |
1218 | return; | |
1219 | } | |
1220 | rpc_final_put_task(task, task->tk_workqueue); | |
1da177e4 LT |
1221 | } |
1222 | ||
b247bbf1 TM |
1223 | int rpciod_up(void) |
1224 | { | |
1225 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
1226 | } | |
1227 | ||
1228 | void rpciod_down(void) | |
1229 | { | |
1230 | module_put(THIS_MODULE); | |
1231 | } | |
1232 | ||
1da177e4 | 1233 | /* |
b247bbf1 | 1234 | * Start up the rpciod workqueue. |
1da177e4 | 1235 | */ |
b247bbf1 | 1236 | static int rpciod_start(void) |
1da177e4 LT |
1237 | { |
1238 | struct workqueue_struct *wq; | |
ab418d70 | 1239 | |
1da177e4 LT |
1240 | /* |
1241 | * Create the rpciod thread and wait for it to start. | |
1242 | */ | |
ab418d70 | 1243 | dprintk("RPC: creating workqueue rpciod\n"); |
f515f86b | 1244 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0); |
40a5f1b1 TM |
1245 | if (!wq) |
1246 | goto out_failed; | |
1da177e4 | 1247 | rpciod_workqueue = wq; |
40a5f1b1 | 1248 | /* Note: highpri because network receive is latency sensitive */ |
90ea9f1b | 1249 | wq = alloc_workqueue("xprtiod", WQ_UNBOUND|WQ_MEM_RECLAIM|WQ_HIGHPRI, 0); |
40a5f1b1 TM |
1250 | if (!wq) |
1251 | goto free_rpciod; | |
1252 | xprtiod_workqueue = wq; | |
1253 | return 1; | |
1254 | free_rpciod: | |
1255 | wq = rpciod_workqueue; | |
1256 | rpciod_workqueue = NULL; | |
1257 | destroy_workqueue(wq); | |
1258 | out_failed: | |
1259 | return 0; | |
1da177e4 LT |
1260 | } |
1261 | ||
b247bbf1 | 1262 | static void rpciod_stop(void) |
1da177e4 | 1263 | { |
b247bbf1 | 1264 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1265 | |
b247bbf1 TM |
1266 | if (rpciod_workqueue == NULL) |
1267 | return; | |
ab418d70 | 1268 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1269 | |
b247bbf1 TM |
1270 | wq = rpciod_workqueue; |
1271 | rpciod_workqueue = NULL; | |
1272 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1273 | wq = xprtiod_workqueue; |
1274 | xprtiod_workqueue = NULL; | |
1275 | destroy_workqueue(wq); | |
1da177e4 LT |
1276 | } |
1277 | ||
1da177e4 LT |
1278 | void |
1279 | rpc_destroy_mempool(void) | |
1280 | { | |
b247bbf1 | 1281 | rpciod_stop(); |
17a9618e JL |
1282 | mempool_destroy(rpc_buffer_mempool); |
1283 | mempool_destroy(rpc_task_mempool); | |
1284 | kmem_cache_destroy(rpc_task_slabp); | |
1285 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 1286 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | int | |
1290 | rpc_init_mempool(void) | |
1291 | { | |
f6a1cc89 TM |
1292 | /* |
1293 | * The following is not strictly a mempool initialisation, | |
1294 | * but there is no harm in doing it here | |
1295 | */ | |
1296 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
1297 | if (!rpciod_start()) | |
1298 | goto err_nomem; | |
1299 | ||
1da177e4 LT |
1300 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1301 | sizeof(struct rpc_task), | |
1302 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1303 | NULL); |
1da177e4 LT |
1304 | if (!rpc_task_slabp) |
1305 | goto err_nomem; | |
1306 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
1307 | RPC_BUFFER_MAXSIZE, | |
1308 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1309 | NULL); |
1da177e4 LT |
1310 | if (!rpc_buffer_slabp) |
1311 | goto err_nomem; | |
93d2341c MD |
1312 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1313 | rpc_task_slabp); | |
1da177e4 LT |
1314 | if (!rpc_task_mempool) |
1315 | goto err_nomem; | |
93d2341c MD |
1316 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1317 | rpc_buffer_slabp); | |
1da177e4 LT |
1318 | if (!rpc_buffer_mempool) |
1319 | goto err_nomem; | |
1320 | return 0; | |
1321 | err_nomem: | |
1322 | rpc_destroy_mempool(); | |
1323 | return -ENOMEM; | |
1324 | } |