Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/sched.c | |
3 | * | |
4 | * Scheduling for synchronous and asynchronous RPC requests. | |
5 | * | |
6 | * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de> | |
cca5172a | 7 | * |
1da177e4 LT |
8 | * TCP NFS related read + write fixes |
9 | * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> | |
10 | */ | |
11 | ||
12 | #include <linux/module.h> | |
13 | ||
14 | #include <linux/sched.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/smp.h> | |
1da177e4 | 19 | #include <linux/spinlock.h> |
4a3e2f71 | 20 | #include <linux/mutex.h> |
1da177e4 LT |
21 | |
22 | #include <linux/sunrpc/clnt.h> | |
1da177e4 | 23 | |
6951867b BH |
24 | #include "sunrpc.h" |
25 | ||
1da177e4 LT |
26 | #ifdef RPC_DEBUG |
27 | #define RPCDBG_FACILITY RPCDBG_SCHED | |
1da177e4 LT |
28 | #endif |
29 | ||
30 | /* | |
31 | * RPC slabs and memory pools | |
32 | */ | |
33 | #define RPC_BUFFER_MAXSIZE (2048) | |
34 | #define RPC_BUFFER_POOLSIZE (8) | |
35 | #define RPC_TASK_POOLSIZE (8) | |
e18b890b CL |
36 | static struct kmem_cache *rpc_task_slabp __read_mostly; |
37 | static struct kmem_cache *rpc_buffer_slabp __read_mostly; | |
ba89966c ED |
38 | static mempool_t *rpc_task_mempool __read_mostly; |
39 | static mempool_t *rpc_buffer_mempool __read_mostly; | |
1da177e4 | 40 | |
65f27f38 | 41 | static void rpc_async_schedule(struct work_struct *); |
bde8f00c | 42 | static void rpc_release_task(struct rpc_task *task); |
36df9aae | 43 | static void __rpc_queue_timer_fn(unsigned long ptr); |
1da177e4 | 44 | |
1da177e4 LT |
45 | /* |
46 | * RPC tasks sit here while waiting for conditions to improve. | |
47 | */ | |
a4a87499 | 48 | static struct rpc_wait_queue delay_queue; |
1da177e4 | 49 | |
1da177e4 LT |
50 | /* |
51 | * rpciod-related stuff | |
52 | */ | |
24c5d9d7 | 53 | struct workqueue_struct *rpciod_workqueue; |
1da177e4 | 54 | |
1da177e4 LT |
55 | /* |
56 | * Disable the timer for a given RPC task. Should be called with | |
57 | * queue->lock and bh_disabled in order to avoid races within | |
58 | * rpc_run_timer(). | |
59 | */ | |
5d00837b | 60 | static void |
eb276c0e | 61 | __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 62 | { |
36df9aae TM |
63 | if (task->tk_timeout == 0) |
64 | return; | |
46121cf7 | 65 | dprintk("RPC: %5u disabling timer\n", task->tk_pid); |
1da177e4 | 66 | task->tk_timeout = 0; |
36df9aae | 67 | list_del(&task->u.tk_wait.timer_list); |
eb276c0e TM |
68 | if (list_empty(&queue->timer_list.list)) |
69 | del_timer(&queue->timer_list.timer); | |
36df9aae TM |
70 | } |
71 | ||
72 | static void | |
73 | rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires) | |
74 | { | |
75 | queue->timer_list.expires = expires; | |
76 | mod_timer(&queue->timer_list.timer, expires); | |
1da177e4 LT |
77 | } |
78 | ||
1da177e4 LT |
79 | /* |
80 | * Set up a timer for the current task. | |
81 | */ | |
5d00837b | 82 | static void |
eb276c0e | 83 | __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 LT |
84 | { |
85 | if (!task->tk_timeout) | |
86 | return; | |
87 | ||
46121cf7 | 88 | dprintk("RPC: %5u setting alarm for %lu ms\n", |
1da177e4 LT |
89 | task->tk_pid, task->tk_timeout * 1000 / HZ); |
90 | ||
eb276c0e TM |
91 | task->u.tk_wait.expires = jiffies + task->tk_timeout; |
92 | if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires)) | |
93 | rpc_set_queue_timer(queue, task->u.tk_wait.expires); | |
94 | list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); | |
1da177e4 LT |
95 | } |
96 | ||
97 | /* | |
98 | * Add new request to a priority queue. | |
99 | */ | |
100 | static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task) | |
101 | { | |
102 | struct list_head *q; | |
103 | struct rpc_task *t; | |
104 | ||
105 | INIT_LIST_HEAD(&task->u.tk_wait.links); | |
106 | q = &queue->tasks[task->tk_priority]; | |
107 | if (unlikely(task->tk_priority > queue->maxpriority)) | |
108 | q = &queue->tasks[queue->maxpriority]; | |
109 | list_for_each_entry(t, q, u.tk_wait.list) { | |
3ff7576d | 110 | if (t->tk_owner == task->tk_owner) { |
1da177e4 LT |
111 | list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links); |
112 | return; | |
113 | } | |
114 | } | |
115 | list_add_tail(&task->u.tk_wait.list, q); | |
116 | } | |
117 | ||
118 | /* | |
119 | * Add new request to wait queue. | |
120 | * | |
121 | * Swapper tasks always get inserted at the head of the queue. | |
122 | * This should avoid many nasty memory deadlocks and hopefully | |
123 | * improve overall performance. | |
124 | * Everyone else gets appended to the queue to ensure proper FIFO behavior. | |
125 | */ | |
126 | static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) | |
127 | { | |
128 | BUG_ON (RPC_IS_QUEUED(task)); | |
129 | ||
130 | if (RPC_IS_PRIORITY(queue)) | |
131 | __rpc_add_wait_queue_priority(queue, task); | |
132 | else if (RPC_IS_SWAPPER(task)) | |
133 | list_add(&task->u.tk_wait.list, &queue->tasks[0]); | |
134 | else | |
135 | list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]); | |
96ef13b2 | 136 | task->tk_waitqueue = queue; |
e19b63da | 137 | queue->qlen++; |
1da177e4 LT |
138 | rpc_set_queued(task); |
139 | ||
46121cf7 CL |
140 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
141 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
142 | } |
143 | ||
144 | /* | |
145 | * Remove request from a priority queue. | |
146 | */ | |
147 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
148 | { | |
149 | struct rpc_task *t; | |
150 | ||
151 | if (!list_empty(&task->u.tk_wait.links)) { | |
152 | t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list); | |
153 | list_move(&t->u.tk_wait.list, &task->u.tk_wait.list); | |
154 | list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links); | |
155 | } | |
1da177e4 LT |
156 | } |
157 | ||
158 | /* | |
159 | * Remove request from queue. | |
160 | * Note: must be called with spin lock held. | |
161 | */ | |
96ef13b2 | 162 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 163 | { |
eb276c0e | 164 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
165 | if (RPC_IS_PRIORITY(queue)) |
166 | __rpc_remove_wait_queue_priority(task); | |
36df9aae | 167 | list_del(&task->u.tk_wait.list); |
e19b63da | 168 | queue->qlen--; |
46121cf7 CL |
169 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
170 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
171 | } |
172 | ||
173 | static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) | |
174 | { | |
175 | queue->priority = priority; | |
176 | queue->count = 1 << (priority * 2); | |
177 | } | |
178 | ||
3ff7576d | 179 | static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid) |
1da177e4 | 180 | { |
3ff7576d | 181 | queue->owner = pid; |
1da177e4 LT |
182 | queue->nr = RPC_BATCH_COUNT; |
183 | } | |
184 | ||
185 | static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) | |
186 | { | |
187 | rpc_set_waitqueue_priority(queue, queue->maxpriority); | |
3ff7576d | 188 | rpc_set_waitqueue_owner(queue, 0); |
1da177e4 LT |
189 | } |
190 | ||
3ff7576d | 191 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
192 | { |
193 | int i; | |
194 | ||
195 | spin_lock_init(&queue->lock); | |
196 | for (i = 0; i < ARRAY_SIZE(queue->tasks); i++) | |
197 | INIT_LIST_HEAD(&queue->tasks[i]); | |
3ff7576d | 198 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 199 | rpc_reset_waitqueue_priority(queue); |
36df9aae TM |
200 | queue->qlen = 0; |
201 | setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue); | |
202 | INIT_LIST_HEAD(&queue->timer_list.list); | |
1da177e4 LT |
203 | #ifdef RPC_DEBUG |
204 | queue->name = qname; | |
205 | #endif | |
206 | } | |
207 | ||
208 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
209 | { | |
3ff7576d | 210 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 211 | } |
689cf5c1 | 212 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
213 | |
214 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
215 | { | |
3ff7576d | 216 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 217 | } |
e8914c65 | 218 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 219 | |
f6a1cc89 TM |
220 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
221 | { | |
36df9aae | 222 | del_timer_sync(&queue->timer_list.timer); |
f6a1cc89 TM |
223 | } |
224 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
225 | ||
150030b7 | 226 | static int rpc_wait_bit_killable(void *word) |
44c28873 | 227 | { |
150030b7 | 228 | if (fatal_signal_pending(current)) |
44c28873 TM |
229 | return -ERESTARTSYS; |
230 | schedule(); | |
231 | return 0; | |
232 | } | |
233 | ||
c44fe705 TM |
234 | #ifdef RPC_DEBUG |
235 | static void rpc_task_set_debuginfo(struct rpc_task *task) | |
236 | { | |
237 | static atomic_t rpc_pid; | |
238 | ||
c44fe705 TM |
239 | task->tk_pid = atomic_inc_return(&rpc_pid); |
240 | } | |
241 | #else | |
242 | static inline void rpc_task_set_debuginfo(struct rpc_task *task) | |
243 | { | |
244 | } | |
245 | #endif | |
246 | ||
e6b3c4db TM |
247 | static void rpc_set_active(struct rpc_task *task) |
248 | { | |
c44fe705 | 249 | rpc_task_set_debuginfo(task); |
58f9612c | 250 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e6b3c4db TM |
251 | } |
252 | ||
44c28873 TM |
253 | /* |
254 | * Mark an RPC call as having completed by clearing the 'active' bit | |
255 | */ | |
e6b3c4db | 256 | static void rpc_mark_complete_task(struct rpc_task *task) |
44c28873 | 257 | { |
e6b3c4db TM |
258 | smp_mb__before_clear_bit(); |
259 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); | |
260 | smp_mb__after_clear_bit(); | |
44c28873 TM |
261 | wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE); |
262 | } | |
263 | ||
264 | /* | |
265 | * Allow callers to wait for completion of an RPC call | |
266 | */ | |
267 | int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *)) | |
268 | { | |
269 | if (action == NULL) | |
150030b7 | 270 | action = rpc_wait_bit_killable; |
44c28873 | 271 | return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 272 | action, TASK_KILLABLE); |
44c28873 | 273 | } |
e8914c65 | 274 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 275 | |
1da177e4 LT |
276 | /* |
277 | * Make an RPC task runnable. | |
278 | * | |
cca5172a | 279 | * Note: If the task is ASYNC, this must be called with |
1da177e4 LT |
280 | * the spinlock held to protect the wait queue operation. |
281 | */ | |
282 | static void rpc_make_runnable(struct rpc_task *task) | |
283 | { | |
1da177e4 | 284 | rpc_clear_queued(task); |
cc4dc59e CS |
285 | if (rpc_test_and_set_running(task)) |
286 | return; | |
1da177e4 LT |
287 | if (RPC_IS_ASYNC(task)) { |
288 | int status; | |
289 | ||
65f27f38 | 290 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
32bfb5c0 | 291 | status = queue_work(rpciod_workqueue, &task->u.tk_work); |
1da177e4 LT |
292 | if (status < 0) { |
293 | printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status); | |
294 | task->tk_status = status; | |
295 | return; | |
296 | } | |
297 | } else | |
96651ab3 | 298 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
299 | } |
300 | ||
1da177e4 LT |
301 | /* |
302 | * Prepare for sleeping on a wait queue. | |
303 | * By always appending tasks to the list we ensure FIFO behavior. | |
304 | * NB: An RPC task will only receive interrupt-driven events as long | |
305 | * as it's on a wait queue. | |
306 | */ | |
307 | static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, | |
5d00837b | 308 | rpc_action action) |
1da177e4 | 309 | { |
46121cf7 CL |
310 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
311 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 312 | |
1da177e4 LT |
313 | __rpc_add_wait_queue(q, task); |
314 | ||
315 | BUG_ON(task->tk_callback != NULL); | |
316 | task->tk_callback = action; | |
eb276c0e | 317 | __rpc_add_timer(q, task); |
1da177e4 LT |
318 | } |
319 | ||
320 | void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, | |
5d00837b | 321 | rpc_action action) |
1da177e4 | 322 | { |
58f9612c TM |
323 | /* We shouldn't ever put an inactive task to sleep */ |
324 | BUG_ON(!RPC_IS_ACTIVATED(task)); | |
e6b3c4db | 325 | |
1da177e4 LT |
326 | /* |
327 | * Protect the queue operations. | |
328 | */ | |
329 | spin_lock_bh(&q->lock); | |
5d00837b | 330 | __rpc_sleep_on(q, task, action); |
1da177e4 LT |
331 | spin_unlock_bh(&q->lock); |
332 | } | |
e8914c65 | 333 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 LT |
334 | |
335 | /** | |
336 | * __rpc_do_wake_up_task - wake up a single rpc_task | |
96ef13b2 | 337 | * @queue: wait queue |
1da177e4 LT |
338 | * @task: task to be woken up |
339 | * | |
340 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
341 | */ | |
96ef13b2 | 342 | static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 343 | { |
46121cf7 CL |
344 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
345 | task->tk_pid, jiffies); | |
1da177e4 | 346 | |
1da177e4 LT |
347 | /* Has the task been executed yet? If not, we cannot wake it up! */ |
348 | if (!RPC_IS_ACTIVATED(task)) { | |
349 | printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task); | |
350 | return; | |
351 | } | |
352 | ||
96ef13b2 | 353 | __rpc_remove_wait_queue(queue, task); |
1da177e4 LT |
354 | |
355 | rpc_make_runnable(task); | |
356 | ||
46121cf7 | 357 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
358 | } |
359 | ||
360 | /* | |
96ef13b2 | 361 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 362 | */ |
96ef13b2 | 363 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 364 | { |
f5fb7b06 TM |
365 | if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue) |
366 | __rpc_do_wake_up_task(queue, task); | |
1da177e4 LT |
367 | } |
368 | ||
48f18612 AB |
369 | /* |
370 | * Tests whether rpc queue is empty | |
371 | */ | |
372 | int rpc_queue_empty(struct rpc_wait_queue *queue) | |
373 | { | |
374 | int res; | |
375 | ||
376 | spin_lock_bh(&queue->lock); | |
377 | res = queue->qlen; | |
378 | spin_unlock_bh(&queue->lock); | |
379 | return (res == 0); | |
380 | } | |
381 | EXPORT_SYMBOL_GPL(rpc_queue_empty); | |
382 | ||
1da177e4 | 383 | /* |
96ef13b2 | 384 | * Wake up a task on a specific queue |
1da177e4 | 385 | */ |
96ef13b2 | 386 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 387 | { |
5e4424af | 388 | spin_lock_bh(&queue->lock); |
96ef13b2 | 389 | rpc_wake_up_task_queue_locked(queue, task); |
5e4424af | 390 | spin_unlock_bh(&queue->lock); |
1da177e4 | 391 | } |
96ef13b2 TM |
392 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
393 | ||
1da177e4 LT |
394 | /* |
395 | * Wake up the next task on a priority queue. | |
396 | */ | |
397 | static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue) | |
398 | { | |
399 | struct list_head *q; | |
400 | struct rpc_task *task; | |
401 | ||
402 | /* | |
3ff7576d | 403 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
404 | */ |
405 | q = &queue->tasks[queue->priority]; | |
406 | if (!list_empty(q)) { | |
407 | task = list_entry(q->next, struct rpc_task, u.tk_wait.list); | |
3ff7576d | 408 | if (queue->owner == task->tk_owner) { |
1da177e4 LT |
409 | if (--queue->nr) |
410 | goto out; | |
411 | list_move_tail(&task->u.tk_wait.list, q); | |
412 | } | |
413 | /* | |
414 | * Check if we need to switch queues. | |
415 | */ | |
416 | if (--queue->count) | |
3ff7576d | 417 | goto new_owner; |
1da177e4 LT |
418 | } |
419 | ||
420 | /* | |
421 | * Service the next queue. | |
422 | */ | |
423 | do { | |
424 | if (q == &queue->tasks[0]) | |
425 | q = &queue->tasks[queue->maxpriority]; | |
426 | else | |
427 | q = q - 1; | |
428 | if (!list_empty(q)) { | |
429 | task = list_entry(q->next, struct rpc_task, u.tk_wait.list); | |
430 | goto new_queue; | |
431 | } | |
432 | } while (q != &queue->tasks[queue->priority]); | |
433 | ||
434 | rpc_reset_waitqueue_priority(queue); | |
435 | return NULL; | |
436 | ||
437 | new_queue: | |
438 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
3ff7576d TM |
439 | new_owner: |
440 | rpc_set_waitqueue_owner(queue, task->tk_owner); | |
1da177e4 | 441 | out: |
96ef13b2 | 442 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
443 | return task; |
444 | } | |
445 | ||
446 | /* | |
447 | * Wake up the next task on the wait queue. | |
448 | */ | |
449 | struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue) | |
450 | { | |
451 | struct rpc_task *task = NULL; | |
452 | ||
46121cf7 CL |
453 | dprintk("RPC: wake_up_next(%p \"%s\")\n", |
454 | queue, rpc_qname(queue)); | |
5e4424af | 455 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
456 | if (RPC_IS_PRIORITY(queue)) |
457 | task = __rpc_wake_up_next_priority(queue); | |
458 | else { | |
459 | task_for_first(task, &queue->tasks[0]) | |
96ef13b2 | 460 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 | 461 | } |
5e4424af | 462 | spin_unlock_bh(&queue->lock); |
1da177e4 LT |
463 | |
464 | return task; | |
465 | } | |
e8914c65 | 466 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
467 | |
468 | /** | |
469 | * rpc_wake_up - wake up all rpc_tasks | |
470 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
471 | * | |
472 | * Grabs queue->lock | |
473 | */ | |
474 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
475 | { | |
e6d83d55 | 476 | struct rpc_task *task, *next; |
1da177e4 | 477 | struct list_head *head; |
e6d83d55 | 478 | |
5e4424af | 479 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
480 | head = &queue->tasks[queue->maxpriority]; |
481 | for (;;) { | |
e6d83d55 | 482 | list_for_each_entry_safe(task, next, head, u.tk_wait.list) |
96ef13b2 | 483 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
484 | if (head == &queue->tasks[0]) |
485 | break; | |
486 | head--; | |
487 | } | |
5e4424af | 488 | spin_unlock_bh(&queue->lock); |
1da177e4 | 489 | } |
e8914c65 | 490 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
491 | |
492 | /** | |
493 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
494 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
495 | * @status: status value to set | |
496 | * | |
497 | * Grabs queue->lock | |
498 | */ | |
499 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
500 | { | |
e6d83d55 | 501 | struct rpc_task *task, *next; |
1da177e4 | 502 | struct list_head *head; |
1da177e4 | 503 | |
5e4424af | 504 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
505 | head = &queue->tasks[queue->maxpriority]; |
506 | for (;;) { | |
e6d83d55 | 507 | list_for_each_entry_safe(task, next, head, u.tk_wait.list) { |
1da177e4 | 508 | task->tk_status = status; |
96ef13b2 | 509 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
510 | } |
511 | if (head == &queue->tasks[0]) | |
512 | break; | |
513 | head--; | |
514 | } | |
5e4424af | 515 | spin_unlock_bh(&queue->lock); |
1da177e4 | 516 | } |
e8914c65 | 517 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 518 | |
36df9aae TM |
519 | static void __rpc_queue_timer_fn(unsigned long ptr) |
520 | { | |
521 | struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr; | |
522 | struct rpc_task *task, *n; | |
523 | unsigned long expires, now, timeo; | |
524 | ||
525 | spin_lock(&queue->lock); | |
526 | expires = now = jiffies; | |
527 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
528 | timeo = task->u.tk_wait.expires; | |
529 | if (time_after_eq(now, timeo)) { | |
36df9aae TM |
530 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
531 | task->tk_status = -ETIMEDOUT; | |
532 | rpc_wake_up_task_queue_locked(queue, task); | |
533 | continue; | |
534 | } | |
535 | if (expires == now || time_after(expires, timeo)) | |
536 | expires = timeo; | |
537 | } | |
538 | if (!list_empty(&queue->timer_list.list)) | |
539 | rpc_set_queue_timer(queue, expires); | |
540 | spin_unlock(&queue->lock); | |
541 | } | |
542 | ||
8014793b TM |
543 | static void __rpc_atrun(struct rpc_task *task) |
544 | { | |
5d00837b | 545 | task->tk_status = 0; |
8014793b TM |
546 | } |
547 | ||
1da177e4 LT |
548 | /* |
549 | * Run a task at a later time | |
550 | */ | |
8014793b | 551 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 LT |
552 | { |
553 | task->tk_timeout = delay; | |
5d00837b | 554 | rpc_sleep_on(&delay_queue, task, __rpc_atrun); |
1da177e4 | 555 | } |
e8914c65 | 556 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 557 | |
4ce70ada TM |
558 | /* |
559 | * Helper to call task->tk_ops->rpc_call_prepare | |
560 | */ | |
aae2006e | 561 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
562 | { |
563 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
564 | } | |
565 | ||
d05fdb0c | 566 | /* |
963d8fe5 | 567 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 568 | */ |
abbcf28f | 569 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 570 | { |
abbcf28f | 571 | task->tk_action = NULL; |
963d8fe5 TM |
572 | if (task->tk_ops->rpc_call_done != NULL) { |
573 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 574 | if (task->tk_action != NULL) { |
abbcf28f TM |
575 | WARN_ON(RPC_ASSASSINATED(task)); |
576 | /* Always release the RPC slot and buffer memory */ | |
577 | xprt_release(task); | |
d05fdb0c TM |
578 | } |
579 | } | |
d05fdb0c | 580 | } |
d9b6cd94 TM |
581 | |
582 | void rpc_exit(struct rpc_task *task, int status) | |
583 | { | |
584 | task->tk_status = status; | |
585 | task->tk_action = rpc_exit_task; | |
586 | if (RPC_IS_QUEUED(task)) | |
587 | rpc_wake_up_queued_task(task->tk_waitqueue, task); | |
588 | } | |
589 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 590 | |
bbd5a1f9 TM |
591 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
592 | { | |
a86dc496 | 593 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 594 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
595 | } |
596 | ||
1da177e4 LT |
597 | /* |
598 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
599 | */ | |
2efef837 | 600 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 601 | { |
eb9b55ab TM |
602 | struct rpc_wait_queue *queue; |
603 | int task_is_async = RPC_IS_ASYNC(task); | |
604 | int status = 0; | |
1da177e4 | 605 | |
46121cf7 CL |
606 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
607 | task->tk_pid, task->tk_flags); | |
1da177e4 LT |
608 | |
609 | BUG_ON(RPC_IS_QUEUED(task)); | |
610 | ||
d05fdb0c | 611 | for (;;) { |
1da177e4 LT |
612 | |
613 | /* | |
614 | * Execute any pending callback. | |
615 | */ | |
a486aeda | 616 | if (task->tk_callback) { |
1da177e4 | 617 | void (*save_callback)(struct rpc_task *); |
cca5172a YH |
618 | |
619 | /* | |
a486aeda BF |
620 | * We set tk_callback to NULL before calling it, |
621 | * in case it sets the tk_callback field itself: | |
1da177e4 | 622 | */ |
a486aeda BF |
623 | save_callback = task->tk_callback; |
624 | task->tk_callback = NULL; | |
1da177e4 | 625 | save_callback(task); |
1da177e4 LT |
626 | } |
627 | ||
628 | /* | |
629 | * Perform the next FSM step. | |
630 | * tk_action may be NULL when the task has been killed | |
631 | * by someone else. | |
632 | */ | |
633 | if (!RPC_IS_QUEUED(task)) { | |
abbcf28f | 634 | if (task->tk_action == NULL) |
1da177e4 | 635 | break; |
abbcf28f | 636 | task->tk_action(task); |
1da177e4 LT |
637 | } |
638 | ||
639 | /* | |
640 | * Lockless check for whether task is sleeping or not. | |
641 | */ | |
642 | if (!RPC_IS_QUEUED(task)) | |
643 | continue; | |
eb9b55ab TM |
644 | /* |
645 | * The queue->lock protects against races with | |
646 | * rpc_make_runnable(). | |
647 | * | |
648 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
649 | * rpc_task, rpc_make_runnable() can assign it to a | |
650 | * different workqueue. We therefore cannot assume that the | |
651 | * rpc_task pointer may still be dereferenced. | |
652 | */ | |
653 | queue = task->tk_waitqueue; | |
654 | spin_lock_bh(&queue->lock); | |
655 | if (!RPC_IS_QUEUED(task)) { | |
656 | spin_unlock_bh(&queue->lock); | |
1da177e4 LT |
657 | continue; |
658 | } | |
eb9b55ab TM |
659 | rpc_clear_running(task); |
660 | spin_unlock_bh(&queue->lock); | |
661 | if (task_is_async) | |
662 | return; | |
1da177e4 LT |
663 | |
664 | /* sync task: sleep here */ | |
46121cf7 | 665 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 666 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
667 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
668 | TASK_KILLABLE); | |
96651ab3 | 669 | if (status == -ERESTARTSYS) { |
1da177e4 LT |
670 | /* |
671 | * When a sync task receives a signal, it exits with | |
672 | * -ERESTARTSYS. In order to catch any callbacks that | |
673 | * clean up after sleeping on some queue, we don't | |
674 | * break the loop here, but go around once more. | |
675 | */ | |
46121cf7 | 676 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
96651ab3 TM |
677 | task->tk_flags |= RPC_TASK_KILLED; |
678 | rpc_exit(task, -ERESTARTSYS); | |
1da177e4 LT |
679 | } |
680 | rpc_set_running(task); | |
46121cf7 | 681 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
682 | } |
683 | ||
46121cf7 CL |
684 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
685 | task->tk_status); | |
1da177e4 LT |
686 | /* Release all resources associated with the task */ |
687 | rpc_release_task(task); | |
1da177e4 LT |
688 | } |
689 | ||
690 | /* | |
691 | * User-visible entry point to the scheduler. | |
692 | * | |
693 | * This may be called recursively if e.g. an async NFS task updates | |
694 | * the attributes and finds that dirty pages must be flushed. | |
695 | * NOTE: Upon exit of this function the task is guaranteed to be | |
696 | * released. In particular note that tk_release() will have | |
697 | * been called, so your task memory may have been freed. | |
698 | */ | |
2efef837 | 699 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 700 | { |
44c28873 | 701 | rpc_set_active(task); |
d6a1ed08 TM |
702 | rpc_make_runnable(task); |
703 | if (!RPC_IS_ASYNC(task)) | |
704 | __rpc_execute(task); | |
1da177e4 LT |
705 | } |
706 | ||
65f27f38 | 707 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 708 | { |
65f27f38 | 709 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
1da177e4 LT |
710 | } |
711 | ||
02107148 CL |
712 | /** |
713 | * rpc_malloc - allocate an RPC buffer | |
714 | * @task: RPC task that will use this buffer | |
715 | * @size: requested byte size | |
1da177e4 | 716 | * |
c5a4dd8b CL |
717 | * To prevent rpciod from hanging, this allocator never sleeps, |
718 | * returning NULL if the request cannot be serviced immediately. | |
719 | * The caller can arrange to sleep in a way that is safe for rpciod. | |
720 | * | |
721 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
722 | * mempool, ensuring that NFS reads and writes can always proceed, | |
723 | * and that there is good locality of reference for these buffers. | |
724 | * | |
1da177e4 | 725 | * In order to avoid memory starvation triggering more writebacks of |
c5a4dd8b | 726 | * NFS requests, we avoid using GFP_KERNEL. |
1da177e4 | 727 | */ |
c5a4dd8b | 728 | void *rpc_malloc(struct rpc_task *task, size_t size) |
1da177e4 | 729 | { |
aa3d1fae | 730 | struct rpc_buffer *buf; |
c5a4dd8b | 731 | gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT; |
1da177e4 | 732 | |
aa3d1fae | 733 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
734 | if (size <= RPC_BUFFER_MAXSIZE) |
735 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 736 | else |
c5a4dd8b | 737 | buf = kmalloc(size, gfp); |
ddce40df PZ |
738 | |
739 | if (!buf) | |
740 | return NULL; | |
741 | ||
aa3d1fae | 742 | buf->len = size; |
215d0678 | 743 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 744 | task->tk_pid, size, buf); |
aa3d1fae | 745 | return &buf->data; |
1da177e4 | 746 | } |
12444809 | 747 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 748 | |
02107148 CL |
749 | /** |
750 | * rpc_free - free buffer allocated via rpc_malloc | |
c5a4dd8b | 751 | * @buffer: buffer to free |
02107148 CL |
752 | * |
753 | */ | |
c5a4dd8b | 754 | void rpc_free(void *buffer) |
1da177e4 | 755 | { |
aa3d1fae CL |
756 | size_t size; |
757 | struct rpc_buffer *buf; | |
02107148 | 758 | |
c5a4dd8b CL |
759 | if (!buffer) |
760 | return; | |
aa3d1fae CL |
761 | |
762 | buf = container_of(buffer, struct rpc_buffer, data); | |
763 | size = buf->len; | |
c5a4dd8b | 764 | |
215d0678 | 765 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 766 | size, buf); |
aa3d1fae | 767 | |
c5a4dd8b CL |
768 | if (size <= RPC_BUFFER_MAXSIZE) |
769 | mempool_free(buf, rpc_buffer_mempool); | |
770 | else | |
771 | kfree(buf); | |
1da177e4 | 772 | } |
12444809 | 773 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
774 | |
775 | /* | |
776 | * Creation and deletion of RPC task structures | |
777 | */ | |
47fe0648 | 778 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
779 | { |
780 | memset(task, 0, sizeof(*task)); | |
44c28873 | 781 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
782 | task->tk_flags = task_setup_data->flags; |
783 | task->tk_ops = task_setup_data->callback_ops; | |
784 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 785 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 LT |
786 | |
787 | /* Initialize retry counters */ | |
788 | task->tk_garb_retry = 2; | |
789 | task->tk_cred_retry = 2; | |
790 | ||
3ff7576d TM |
791 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
792 | task->tk_owner = current->tgid; | |
1da177e4 LT |
793 | |
794 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 795 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 796 | |
84115e1c TM |
797 | if (task->tk_ops->rpc_call_prepare != NULL) |
798 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 799 | |
ef759a2e | 800 | /* starting timestamp */ |
ff839970 | 801 | task->tk_start = ktime_get(); |
ef759a2e | 802 | |
46121cf7 | 803 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 804 | task_pid_nr(current)); |
1da177e4 LT |
805 | } |
806 | ||
807 | static struct rpc_task * | |
808 | rpc_alloc_task(void) | |
809 | { | |
810 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS); | |
811 | } | |
812 | ||
1da177e4 | 813 | /* |
90c5755f | 814 | * Create a new task for the specified client. |
1da177e4 | 815 | */ |
84115e1c | 816 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 817 | { |
e8f5d77c TM |
818 | struct rpc_task *task = setup_data->task; |
819 | unsigned short flags = 0; | |
820 | ||
821 | if (task == NULL) { | |
822 | task = rpc_alloc_task(); | |
19445b99 TM |
823 | if (task == NULL) { |
824 | rpc_release_calldata(setup_data->callback_ops, | |
825 | setup_data->callback_data); | |
826 | return ERR_PTR(-ENOMEM); | |
827 | } | |
e8f5d77c TM |
828 | flags = RPC_TASK_DYNAMIC; |
829 | } | |
1da177e4 | 830 | |
84115e1c | 831 | rpc_init_task(task, setup_data); |
19445b99 TM |
832 | if (task->tk_status < 0) { |
833 | int err = task->tk_status; | |
834 | rpc_put_task(task); | |
835 | return ERR_PTR(err); | |
836 | } | |
1da177e4 | 837 | |
e8f5d77c | 838 | task->tk_flags |= flags; |
46121cf7 | 839 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 840 | return task; |
1da177e4 LT |
841 | } |
842 | ||
32bfb5c0 | 843 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 844 | { |
963d8fe5 TM |
845 | const struct rpc_call_ops *tk_ops = task->tk_ops; |
846 | void *calldata = task->tk_calldata; | |
1da177e4 | 847 | |
5e4424af TM |
848 | if (task->tk_flags & RPC_TASK_DYNAMIC) { |
849 | dprintk("RPC: %5u freeing task\n", task->tk_pid); | |
850 | mempool_free(task, rpc_task_mempool); | |
851 | } | |
32bfb5c0 TM |
852 | rpc_release_calldata(tk_ops, calldata); |
853 | } | |
854 | ||
855 | static void rpc_async_release(struct work_struct *work) | |
856 | { | |
857 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); | |
858 | } | |
859 | ||
860 | void rpc_put_task(struct rpc_task *task) | |
861 | { | |
e6b3c4db TM |
862 | if (!atomic_dec_and_test(&task->tk_count)) |
863 | return; | |
864 | /* Release resources */ | |
865 | if (task->tk_rqstp) | |
866 | xprt_release(task); | |
867 | if (task->tk_msg.rpc_cred) | |
a17c2153 | 868 | put_rpccred(task->tk_msg.rpc_cred); |
58f9612c | 869 | rpc_task_release_client(task); |
32bfb5c0 TM |
870 | if (task->tk_workqueue != NULL) { |
871 | INIT_WORK(&task->u.tk_work, rpc_async_release); | |
872 | queue_work(task->tk_workqueue, &task->u.tk_work); | |
873 | } else | |
874 | rpc_free_task(task); | |
e6b3c4db | 875 | } |
e8914c65 | 876 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 877 | |
bde8f00c | 878 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 879 | { |
46121cf7 | 880 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 881 | |
1da177e4 | 882 | BUG_ON (RPC_IS_QUEUED(task)); |
1da177e4 | 883 | |
e6b3c4db TM |
884 | /* Wake up anyone who is waiting for task completion */ |
885 | rpc_mark_complete_task(task); | |
886 | ||
887 | rpc_put_task(task); | |
1da177e4 LT |
888 | } |
889 | ||
b247bbf1 TM |
890 | int rpciod_up(void) |
891 | { | |
892 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
893 | } | |
894 | ||
895 | void rpciod_down(void) | |
896 | { | |
897 | module_put(THIS_MODULE); | |
898 | } | |
899 | ||
1da177e4 | 900 | /* |
b247bbf1 | 901 | * Start up the rpciod workqueue. |
1da177e4 | 902 | */ |
b247bbf1 | 903 | static int rpciod_start(void) |
1da177e4 LT |
904 | { |
905 | struct workqueue_struct *wq; | |
ab418d70 | 906 | |
1da177e4 LT |
907 | /* |
908 | * Create the rpciod thread and wait for it to start. | |
909 | */ | |
ab418d70 | 910 | dprintk("RPC: creating workqueue rpciod\n"); |
1da177e4 | 911 | wq = create_workqueue("rpciod"); |
1da177e4 | 912 | rpciod_workqueue = wq; |
b247bbf1 | 913 | return rpciod_workqueue != NULL; |
1da177e4 LT |
914 | } |
915 | ||
b247bbf1 | 916 | static void rpciod_stop(void) |
1da177e4 | 917 | { |
b247bbf1 | 918 | struct workqueue_struct *wq = NULL; |
ab418d70 | 919 | |
b247bbf1 TM |
920 | if (rpciod_workqueue == NULL) |
921 | return; | |
ab418d70 | 922 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 923 | |
b247bbf1 TM |
924 | wq = rpciod_workqueue; |
925 | rpciod_workqueue = NULL; | |
926 | destroy_workqueue(wq); | |
1da177e4 LT |
927 | } |
928 | ||
1da177e4 LT |
929 | void |
930 | rpc_destroy_mempool(void) | |
931 | { | |
b247bbf1 | 932 | rpciod_stop(); |
1da177e4 LT |
933 | if (rpc_buffer_mempool) |
934 | mempool_destroy(rpc_buffer_mempool); | |
935 | if (rpc_task_mempool) | |
936 | mempool_destroy(rpc_task_mempool); | |
1a1d92c1 AD |
937 | if (rpc_task_slabp) |
938 | kmem_cache_destroy(rpc_task_slabp); | |
939 | if (rpc_buffer_slabp) | |
940 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 941 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
942 | } |
943 | ||
944 | int | |
945 | rpc_init_mempool(void) | |
946 | { | |
f6a1cc89 TM |
947 | /* |
948 | * The following is not strictly a mempool initialisation, | |
949 | * but there is no harm in doing it here | |
950 | */ | |
951 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
952 | if (!rpciod_start()) | |
953 | goto err_nomem; | |
954 | ||
1da177e4 LT |
955 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
956 | sizeof(struct rpc_task), | |
957 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 958 | NULL); |
1da177e4 LT |
959 | if (!rpc_task_slabp) |
960 | goto err_nomem; | |
961 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
962 | RPC_BUFFER_MAXSIZE, | |
963 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 964 | NULL); |
1da177e4 LT |
965 | if (!rpc_buffer_slabp) |
966 | goto err_nomem; | |
93d2341c MD |
967 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
968 | rpc_task_slabp); | |
1da177e4 LT |
969 | if (!rpc_task_mempool) |
970 | goto err_nomem; | |
93d2341c MD |
971 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
972 | rpc_buffer_slabp); | |
1da177e4 LT |
973 | if (!rpc_buffer_mempool) |
974 | goto err_nomem; | |
975 | return 0; | |
976 | err_nomem: | |
977 | rpc_destroy_mempool(); | |
978 | return -ENOMEM; | |
979 | } |