Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/net/sunrpc/sched.c | |
3 | * | |
4 | * Scheduling for synchronous and asynchronous RPC requests. | |
5 | * | |
6 | * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de> | |
cca5172a | 7 | * |
1da177e4 LT |
8 | * TCP NFS related read + write fixes |
9 | * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> | |
10 | */ | |
11 | ||
12 | #include <linux/module.h> | |
13 | ||
14 | #include <linux/sched.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/mempool.h> | |
18 | #include <linux/smp.h> | |
1da177e4 | 19 | #include <linux/spinlock.h> |
4a3e2f71 | 20 | #include <linux/mutex.h> |
d310310c | 21 | #include <linux/freezer.h> |
1da177e4 LT |
22 | |
23 | #include <linux/sunrpc/clnt.h> | |
1da177e4 | 24 | |
6951867b BH |
25 | #include "sunrpc.h" |
26 | ||
f895b252 | 27 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) |
1da177e4 | 28 | #define RPCDBG_FACILITY RPCDBG_SCHED |
1da177e4 LT |
29 | #endif |
30 | ||
82b0a4c3 TM |
31 | #define CREATE_TRACE_POINTS |
32 | #include <trace/events/sunrpc.h> | |
33 | ||
1da177e4 LT |
34 | /* |
35 | * RPC slabs and memory pools | |
36 | */ | |
37 | #define RPC_BUFFER_MAXSIZE (2048) | |
38 | #define RPC_BUFFER_POOLSIZE (8) | |
39 | #define RPC_TASK_POOLSIZE (8) | |
e18b890b CL |
40 | static struct kmem_cache *rpc_task_slabp __read_mostly; |
41 | static struct kmem_cache *rpc_buffer_slabp __read_mostly; | |
ba89966c ED |
42 | static mempool_t *rpc_task_mempool __read_mostly; |
43 | static mempool_t *rpc_buffer_mempool __read_mostly; | |
1da177e4 | 44 | |
65f27f38 | 45 | static void rpc_async_schedule(struct work_struct *); |
bde8f00c | 46 | static void rpc_release_task(struct rpc_task *task); |
36df9aae | 47 | static void __rpc_queue_timer_fn(unsigned long ptr); |
1da177e4 | 48 | |
1da177e4 LT |
49 | /* |
50 | * RPC tasks sit here while waiting for conditions to improve. | |
51 | */ | |
a4a87499 | 52 | static struct rpc_wait_queue delay_queue; |
1da177e4 | 53 | |
1da177e4 LT |
54 | /* |
55 | * rpciod-related stuff | |
56 | */ | |
40a5f1b1 TM |
57 | struct workqueue_struct *rpciod_workqueue __read_mostly; |
58 | struct workqueue_struct *xprtiod_workqueue __read_mostly; | |
1da177e4 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Disable the timer for a given RPC task. Should be called with | |
62 | * queue->lock and bh_disabled in order to avoid races within | |
63 | * rpc_run_timer(). | |
64 | */ | |
5d00837b | 65 | static void |
eb276c0e | 66 | __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 67 | { |
36df9aae TM |
68 | if (task->tk_timeout == 0) |
69 | return; | |
46121cf7 | 70 | dprintk("RPC: %5u disabling timer\n", task->tk_pid); |
1da177e4 | 71 | task->tk_timeout = 0; |
36df9aae | 72 | list_del(&task->u.tk_wait.timer_list); |
eb276c0e TM |
73 | if (list_empty(&queue->timer_list.list)) |
74 | del_timer(&queue->timer_list.timer); | |
36df9aae TM |
75 | } |
76 | ||
77 | static void | |
78 | rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires) | |
79 | { | |
80 | queue->timer_list.expires = expires; | |
81 | mod_timer(&queue->timer_list.timer, expires); | |
1da177e4 LT |
82 | } |
83 | ||
1da177e4 LT |
84 | /* |
85 | * Set up a timer for the current task. | |
86 | */ | |
5d00837b | 87 | static void |
eb276c0e | 88 | __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 LT |
89 | { |
90 | if (!task->tk_timeout) | |
91 | return; | |
92 | ||
55cc1d78 NMG |
93 | dprintk("RPC: %5u setting alarm for %u ms\n", |
94 | task->tk_pid, jiffies_to_msecs(task->tk_timeout)); | |
1da177e4 | 95 | |
eb276c0e TM |
96 | task->u.tk_wait.expires = jiffies + task->tk_timeout; |
97 | if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires)) | |
98 | rpc_set_queue_timer(queue, task->u.tk_wait.expires); | |
99 | list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list); | |
1da177e4 LT |
100 | } |
101 | ||
edd2e36f TM |
102 | static void rpc_rotate_queue_owner(struct rpc_wait_queue *queue) |
103 | { | |
104 | struct list_head *q = &queue->tasks[queue->priority]; | |
105 | struct rpc_task *task; | |
106 | ||
107 | if (!list_empty(q)) { | |
108 | task = list_first_entry(q, struct rpc_task, u.tk_wait.list); | |
109 | if (task->tk_owner == queue->owner) | |
110 | list_move_tail(&task->u.tk_wait.list, q); | |
111 | } | |
112 | } | |
113 | ||
c05eecf6 TM |
114 | static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority) |
115 | { | |
edd2e36f TM |
116 | if (queue->priority != priority) { |
117 | /* Fairness: rotate the list when changing priority */ | |
118 | rpc_rotate_queue_owner(queue); | |
119 | queue->priority = priority; | |
120 | } | |
c05eecf6 TM |
121 | } |
122 | ||
123 | static void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid) | |
124 | { | |
125 | queue->owner = pid; | |
126 | queue->nr = RPC_BATCH_COUNT; | |
127 | } | |
128 | ||
129 | static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue) | |
130 | { | |
131 | rpc_set_waitqueue_priority(queue, queue->maxpriority); | |
132 | rpc_set_waitqueue_owner(queue, 0); | |
133 | } | |
134 | ||
1da177e4 LT |
135 | /* |
136 | * Add new request to a priority queue. | |
137 | */ | |
3b27bad7 TM |
138 | static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, |
139 | struct rpc_task *task, | |
140 | unsigned char queue_priority) | |
1da177e4 LT |
141 | { |
142 | struct list_head *q; | |
143 | struct rpc_task *t; | |
144 | ||
145 | INIT_LIST_HEAD(&task->u.tk_wait.links); | |
3b27bad7 | 146 | if (unlikely(queue_priority > queue->maxpriority)) |
c05eecf6 TM |
147 | queue_priority = queue->maxpriority; |
148 | if (queue_priority > queue->priority) | |
149 | rpc_set_waitqueue_priority(queue, queue_priority); | |
150 | q = &queue->tasks[queue_priority]; | |
1da177e4 | 151 | list_for_each_entry(t, q, u.tk_wait.list) { |
3ff7576d | 152 | if (t->tk_owner == task->tk_owner) { |
1da177e4 LT |
153 | list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links); |
154 | return; | |
155 | } | |
156 | } | |
157 | list_add_tail(&task->u.tk_wait.list, q); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Add new request to wait queue. | |
162 | * | |
163 | * Swapper tasks always get inserted at the head of the queue. | |
164 | * This should avoid many nasty memory deadlocks and hopefully | |
165 | * improve overall performance. | |
166 | * Everyone else gets appended to the queue to ensure proper FIFO behavior. | |
167 | */ | |
3b27bad7 TM |
168 | static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, |
169 | struct rpc_task *task, | |
170 | unsigned char queue_priority) | |
1da177e4 | 171 | { |
2bd4eef8 WAA |
172 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
173 | if (RPC_IS_QUEUED(task)) | |
174 | return; | |
1da177e4 LT |
175 | |
176 | if (RPC_IS_PRIORITY(queue)) | |
3b27bad7 | 177 | __rpc_add_wait_queue_priority(queue, task, queue_priority); |
1da177e4 LT |
178 | else if (RPC_IS_SWAPPER(task)) |
179 | list_add(&task->u.tk_wait.list, &queue->tasks[0]); | |
180 | else | |
181 | list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]); | |
96ef13b2 | 182 | task->tk_waitqueue = queue; |
e19b63da | 183 | queue->qlen++; |
1166fde6 TM |
184 | /* barrier matches the read in rpc_wake_up_task_queue_locked() */ |
185 | smp_wmb(); | |
1da177e4 LT |
186 | rpc_set_queued(task); |
187 | ||
46121cf7 CL |
188 | dprintk("RPC: %5u added to queue %p \"%s\"\n", |
189 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
190 | } |
191 | ||
192 | /* | |
193 | * Remove request from a priority queue. | |
194 | */ | |
195 | static void __rpc_remove_wait_queue_priority(struct rpc_task *task) | |
196 | { | |
197 | struct rpc_task *t; | |
198 | ||
199 | if (!list_empty(&task->u.tk_wait.links)) { | |
200 | t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list); | |
201 | list_move(&t->u.tk_wait.list, &task->u.tk_wait.list); | |
202 | list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links); | |
203 | } | |
1da177e4 LT |
204 | } |
205 | ||
206 | /* | |
207 | * Remove request from queue. | |
208 | * Note: must be called with spin lock held. | |
209 | */ | |
96ef13b2 | 210 | static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 211 | { |
eb276c0e | 212 | __rpc_disable_timer(queue, task); |
1da177e4 LT |
213 | if (RPC_IS_PRIORITY(queue)) |
214 | __rpc_remove_wait_queue_priority(task); | |
36df9aae | 215 | list_del(&task->u.tk_wait.list); |
e19b63da | 216 | queue->qlen--; |
46121cf7 CL |
217 | dprintk("RPC: %5u removed from queue %p \"%s\"\n", |
218 | task->tk_pid, queue, rpc_qname(queue)); | |
1da177e4 LT |
219 | } |
220 | ||
3ff7576d | 221 | static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues) |
1da177e4 LT |
222 | { |
223 | int i; | |
224 | ||
225 | spin_lock_init(&queue->lock); | |
226 | for (i = 0; i < ARRAY_SIZE(queue->tasks); i++) | |
227 | INIT_LIST_HEAD(&queue->tasks[i]); | |
3ff7576d | 228 | queue->maxpriority = nr_queues - 1; |
1da177e4 | 229 | rpc_reset_waitqueue_priority(queue); |
36df9aae TM |
230 | queue->qlen = 0; |
231 | setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue); | |
232 | INIT_LIST_HEAD(&queue->timer_list.list); | |
2f09c242 | 233 | rpc_assign_waitqueue_name(queue, qname); |
1da177e4 LT |
234 | } |
235 | ||
236 | void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
237 | { | |
3ff7576d | 238 | __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY); |
1da177e4 | 239 | } |
689cf5c1 | 240 | EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue); |
1da177e4 LT |
241 | |
242 | void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname) | |
243 | { | |
3ff7576d | 244 | __rpc_init_priority_wait_queue(queue, qname, 1); |
1da177e4 | 245 | } |
e8914c65 | 246 | EXPORT_SYMBOL_GPL(rpc_init_wait_queue); |
1da177e4 | 247 | |
f6a1cc89 TM |
248 | void rpc_destroy_wait_queue(struct rpc_wait_queue *queue) |
249 | { | |
36df9aae | 250 | del_timer_sync(&queue->timer_list.timer); |
f6a1cc89 TM |
251 | } |
252 | EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue); | |
253 | ||
dfd01f02 | 254 | static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode) |
44c28873 | 255 | { |
416ad3c9 | 256 | freezable_schedule_unsafe(); |
dfd01f02 PZ |
257 | if (signal_pending_state(mode, current)) |
258 | return -ERESTARTSYS; | |
44c28873 TM |
259 | return 0; |
260 | } | |
261 | ||
1306729b | 262 | #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS) |
c44fe705 TM |
263 | static void rpc_task_set_debuginfo(struct rpc_task *task) |
264 | { | |
265 | static atomic_t rpc_pid; | |
266 | ||
c44fe705 TM |
267 | task->tk_pid = atomic_inc_return(&rpc_pid); |
268 | } | |
269 | #else | |
270 | static inline void rpc_task_set_debuginfo(struct rpc_task *task) | |
271 | { | |
272 | } | |
273 | #endif | |
274 | ||
e6b3c4db TM |
275 | static void rpc_set_active(struct rpc_task *task) |
276 | { | |
82b0a4c3 TM |
277 | trace_rpc_task_begin(task->tk_client, task, NULL); |
278 | ||
c44fe705 | 279 | rpc_task_set_debuginfo(task); |
58f9612c | 280 | set_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
e6b3c4db TM |
281 | } |
282 | ||
44c28873 TM |
283 | /* |
284 | * Mark an RPC call as having completed by clearing the 'active' bit | |
bf294b41 | 285 | * and then waking up all tasks that were sleeping. |
44c28873 | 286 | */ |
bf294b41 | 287 | static int rpc_complete_task(struct rpc_task *task) |
44c28873 | 288 | { |
bf294b41 TM |
289 | void *m = &task->tk_runstate; |
290 | wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE); | |
291 | struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE); | |
292 | unsigned long flags; | |
293 | int ret; | |
294 | ||
82b0a4c3 TM |
295 | trace_rpc_task_complete(task->tk_client, task, NULL); |
296 | ||
bf294b41 | 297 | spin_lock_irqsave(&wq->lock, flags); |
e6b3c4db | 298 | clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate); |
bf294b41 TM |
299 | ret = atomic_dec_and_test(&task->tk_count); |
300 | if (waitqueue_active(wq)) | |
ac5be6b4 | 301 | __wake_up_locked_key(wq, TASK_NORMAL, &k); |
bf294b41 TM |
302 | spin_unlock_irqrestore(&wq->lock, flags); |
303 | return ret; | |
44c28873 TM |
304 | } |
305 | ||
306 | /* | |
307 | * Allow callers to wait for completion of an RPC call | |
bf294b41 TM |
308 | * |
309 | * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit() | |
310 | * to enforce taking of the wq->lock and hence avoid races with | |
311 | * rpc_complete_task(). | |
44c28873 | 312 | */ |
c1221321 | 313 | int __rpc_wait_for_completion_task(struct rpc_task *task, wait_bit_action_f *action) |
44c28873 TM |
314 | { |
315 | if (action == NULL) | |
150030b7 | 316 | action = rpc_wait_bit_killable; |
bf294b41 | 317 | return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE, |
150030b7 | 318 | action, TASK_KILLABLE); |
44c28873 | 319 | } |
e8914c65 | 320 | EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task); |
44c28873 | 321 | |
1da177e4 LT |
322 | /* |
323 | * Make an RPC task runnable. | |
324 | * | |
506026c3 JL |
325 | * Note: If the task is ASYNC, and is being made runnable after sitting on an |
326 | * rpc_wait_queue, this must be called with the queue spinlock held to protect | |
327 | * the wait queue operation. | |
a3c3cac5 TM |
328 | * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(), |
329 | * which is needed to ensure that __rpc_execute() doesn't loop (due to the | |
330 | * lockless RPC_IS_QUEUED() test) before we've had a chance to test | |
331 | * the RPC_TASK_RUNNING flag. | |
1da177e4 LT |
332 | */ |
333 | static void rpc_make_runnable(struct rpc_task *task) | |
334 | { | |
a3c3cac5 TM |
335 | bool need_wakeup = !rpc_test_and_set_running(task); |
336 | ||
1da177e4 | 337 | rpc_clear_queued(task); |
a3c3cac5 | 338 | if (!need_wakeup) |
cc4dc59e | 339 | return; |
1da177e4 | 340 | if (RPC_IS_ASYNC(task)) { |
65f27f38 | 341 | INIT_WORK(&task->u.tk_work, rpc_async_schedule); |
cee6a537 | 342 | queue_work(rpciod_workqueue, &task->u.tk_work); |
1da177e4 | 343 | } else |
96651ab3 | 344 | wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED); |
1da177e4 LT |
345 | } |
346 | ||
1da177e4 LT |
347 | /* |
348 | * Prepare for sleeping on a wait queue. | |
349 | * By always appending tasks to the list we ensure FIFO behavior. | |
350 | * NB: An RPC task will only receive interrupt-driven events as long | |
351 | * as it's on a wait queue. | |
352 | */ | |
3b27bad7 TM |
353 | static void __rpc_sleep_on_priority(struct rpc_wait_queue *q, |
354 | struct rpc_task *task, | |
355 | rpc_action action, | |
356 | unsigned char queue_priority) | |
1da177e4 | 357 | { |
46121cf7 CL |
358 | dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n", |
359 | task->tk_pid, rpc_qname(q), jiffies); | |
1da177e4 | 360 | |
82b0a4c3 TM |
361 | trace_rpc_task_sleep(task->tk_client, task, q); |
362 | ||
3b27bad7 | 363 | __rpc_add_wait_queue(q, task, queue_priority); |
1da177e4 | 364 | |
f50ad428 | 365 | WARN_ON_ONCE(task->tk_callback != NULL); |
1da177e4 | 366 | task->tk_callback = action; |
eb276c0e | 367 | __rpc_add_timer(q, task); |
1da177e4 LT |
368 | } |
369 | ||
370 | void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task, | |
5d00837b | 371 | rpc_action action) |
1da177e4 | 372 | { |
58f9612c | 373 | /* We shouldn't ever put an inactive task to sleep */ |
e454a7a8 WAA |
374 | WARN_ON_ONCE(!RPC_IS_ACTIVATED(task)); |
375 | if (!RPC_IS_ACTIVATED(task)) { | |
376 | task->tk_status = -EIO; | |
377 | rpc_put_task_async(task); | |
378 | return; | |
379 | } | |
e6b3c4db | 380 | |
1da177e4 LT |
381 | /* |
382 | * Protect the queue operations. | |
383 | */ | |
384 | spin_lock_bh(&q->lock); | |
3b27bad7 | 385 | __rpc_sleep_on_priority(q, task, action, task->tk_priority); |
1da177e4 LT |
386 | spin_unlock_bh(&q->lock); |
387 | } | |
e8914c65 | 388 | EXPORT_SYMBOL_GPL(rpc_sleep_on); |
1da177e4 | 389 | |
3b27bad7 TM |
390 | void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task, |
391 | rpc_action action, int priority) | |
392 | { | |
393 | /* We shouldn't ever put an inactive task to sleep */ | |
e454a7a8 WAA |
394 | WARN_ON_ONCE(!RPC_IS_ACTIVATED(task)); |
395 | if (!RPC_IS_ACTIVATED(task)) { | |
396 | task->tk_status = -EIO; | |
397 | rpc_put_task_async(task); | |
398 | return; | |
399 | } | |
3b27bad7 TM |
400 | |
401 | /* | |
402 | * Protect the queue operations. | |
403 | */ | |
404 | spin_lock_bh(&q->lock); | |
405 | __rpc_sleep_on_priority(q, task, action, priority - RPC_PRIORITY_LOW); | |
406 | spin_unlock_bh(&q->lock); | |
407 | } | |
1e1093c7 | 408 | EXPORT_SYMBOL_GPL(rpc_sleep_on_priority); |
3b27bad7 | 409 | |
1da177e4 LT |
410 | /** |
411 | * __rpc_do_wake_up_task - wake up a single rpc_task | |
96ef13b2 | 412 | * @queue: wait queue |
1da177e4 LT |
413 | * @task: task to be woken up |
414 | * | |
415 | * Caller must hold queue->lock, and have cleared the task queued flag. | |
416 | */ | |
96ef13b2 | 417 | static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 418 | { |
46121cf7 CL |
419 | dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n", |
420 | task->tk_pid, jiffies); | |
1da177e4 | 421 | |
1da177e4 LT |
422 | /* Has the task been executed yet? If not, we cannot wake it up! */ |
423 | if (!RPC_IS_ACTIVATED(task)) { | |
424 | printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task); | |
425 | return; | |
426 | } | |
427 | ||
82b0a4c3 TM |
428 | trace_rpc_task_wakeup(task->tk_client, task, queue); |
429 | ||
96ef13b2 | 430 | __rpc_remove_wait_queue(queue, task); |
1da177e4 LT |
431 | |
432 | rpc_make_runnable(task); | |
433 | ||
46121cf7 | 434 | dprintk("RPC: __rpc_wake_up_task done\n"); |
1da177e4 LT |
435 | } |
436 | ||
437 | /* | |
96ef13b2 | 438 | * Wake up a queued task while the queue lock is being held |
1da177e4 | 439 | */ |
96ef13b2 | 440 | static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 441 | { |
1166fde6 TM |
442 | if (RPC_IS_QUEUED(task)) { |
443 | smp_rmb(); | |
444 | if (task->tk_waitqueue == queue) | |
445 | __rpc_do_wake_up_task(queue, task); | |
446 | } | |
1da177e4 LT |
447 | } |
448 | ||
1da177e4 | 449 | /* |
96ef13b2 | 450 | * Wake up a task on a specific queue |
1da177e4 | 451 | */ |
96ef13b2 | 452 | void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task) |
1da177e4 | 453 | { |
5e4424af | 454 | spin_lock_bh(&queue->lock); |
96ef13b2 | 455 | rpc_wake_up_task_queue_locked(queue, task); |
5e4424af | 456 | spin_unlock_bh(&queue->lock); |
1da177e4 | 457 | } |
96ef13b2 TM |
458 | EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task); |
459 | ||
1da177e4 LT |
460 | /* |
461 | * Wake up the next task on a priority queue. | |
462 | */ | |
961a828d | 463 | static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue) |
1da177e4 LT |
464 | { |
465 | struct list_head *q; | |
466 | struct rpc_task *task; | |
467 | ||
468 | /* | |
3ff7576d | 469 | * Service a batch of tasks from a single owner. |
1da177e4 LT |
470 | */ |
471 | q = &queue->tasks[queue->priority]; | |
472 | if (!list_empty(q)) { | |
473 | task = list_entry(q->next, struct rpc_task, u.tk_wait.list); | |
3ff7576d | 474 | if (queue->owner == task->tk_owner) { |
1da177e4 LT |
475 | if (--queue->nr) |
476 | goto out; | |
477 | list_move_tail(&task->u.tk_wait.list, q); | |
478 | } | |
479 | /* | |
480 | * Check if we need to switch queues. | |
481 | */ | |
c05eecf6 | 482 | goto new_owner; |
1da177e4 LT |
483 | } |
484 | ||
485 | /* | |
486 | * Service the next queue. | |
487 | */ | |
488 | do { | |
489 | if (q == &queue->tasks[0]) | |
490 | q = &queue->tasks[queue->maxpriority]; | |
491 | else | |
492 | q = q - 1; | |
493 | if (!list_empty(q)) { | |
494 | task = list_entry(q->next, struct rpc_task, u.tk_wait.list); | |
495 | goto new_queue; | |
496 | } | |
497 | } while (q != &queue->tasks[queue->priority]); | |
498 | ||
499 | rpc_reset_waitqueue_priority(queue); | |
500 | return NULL; | |
501 | ||
502 | new_queue: | |
503 | rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0])); | |
3ff7576d TM |
504 | new_owner: |
505 | rpc_set_waitqueue_owner(queue, task->tk_owner); | |
1da177e4 | 506 | out: |
1da177e4 LT |
507 | return task; |
508 | } | |
509 | ||
961a828d TM |
510 | static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue) |
511 | { | |
512 | if (RPC_IS_PRIORITY(queue)) | |
513 | return __rpc_find_next_queued_priority(queue); | |
514 | if (!list_empty(&queue->tasks[0])) | |
515 | return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list); | |
516 | return NULL; | |
517 | } | |
518 | ||
1da177e4 | 519 | /* |
961a828d | 520 | * Wake up the first task on the wait queue. |
1da177e4 | 521 | */ |
961a828d TM |
522 | struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue, |
523 | bool (*func)(struct rpc_task *, void *), void *data) | |
1da177e4 LT |
524 | { |
525 | struct rpc_task *task = NULL; | |
526 | ||
961a828d | 527 | dprintk("RPC: wake_up_first(%p \"%s\")\n", |
46121cf7 | 528 | queue, rpc_qname(queue)); |
5e4424af | 529 | spin_lock_bh(&queue->lock); |
961a828d TM |
530 | task = __rpc_find_next_queued(queue); |
531 | if (task != NULL) { | |
532 | if (func(task, data)) | |
96ef13b2 | 533 | rpc_wake_up_task_queue_locked(queue, task); |
961a828d TM |
534 | else |
535 | task = NULL; | |
1da177e4 | 536 | } |
5e4424af | 537 | spin_unlock_bh(&queue->lock); |
1da177e4 LT |
538 | |
539 | return task; | |
540 | } | |
961a828d TM |
541 | EXPORT_SYMBOL_GPL(rpc_wake_up_first); |
542 | ||
543 | static bool rpc_wake_up_next_func(struct rpc_task *task, void *data) | |
544 | { | |
545 | return true; | |
546 | } | |
547 | ||
548 | /* | |
549 | * Wake up the next task on the wait queue. | |
550 | */ | |
551 | struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue) | |
552 | { | |
553 | return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL); | |
554 | } | |
e8914c65 | 555 | EXPORT_SYMBOL_GPL(rpc_wake_up_next); |
1da177e4 LT |
556 | |
557 | /** | |
558 | * rpc_wake_up - wake up all rpc_tasks | |
559 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
560 | * | |
561 | * Grabs queue->lock | |
562 | */ | |
563 | void rpc_wake_up(struct rpc_wait_queue *queue) | |
564 | { | |
1da177e4 | 565 | struct list_head *head; |
e6d83d55 | 566 | |
5e4424af | 567 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
568 | head = &queue->tasks[queue->maxpriority]; |
569 | for (;;) { | |
540a0f75 TM |
570 | while (!list_empty(head)) { |
571 | struct rpc_task *task; | |
572 | task = list_first_entry(head, | |
573 | struct rpc_task, | |
574 | u.tk_wait.list); | |
96ef13b2 | 575 | rpc_wake_up_task_queue_locked(queue, task); |
540a0f75 | 576 | } |
1da177e4 LT |
577 | if (head == &queue->tasks[0]) |
578 | break; | |
579 | head--; | |
580 | } | |
5e4424af | 581 | spin_unlock_bh(&queue->lock); |
1da177e4 | 582 | } |
e8914c65 | 583 | EXPORT_SYMBOL_GPL(rpc_wake_up); |
1da177e4 LT |
584 | |
585 | /** | |
586 | * rpc_wake_up_status - wake up all rpc_tasks and set their status value. | |
587 | * @queue: rpc_wait_queue on which the tasks are sleeping | |
588 | * @status: status value to set | |
589 | * | |
590 | * Grabs queue->lock | |
591 | */ | |
592 | void rpc_wake_up_status(struct rpc_wait_queue *queue, int status) | |
593 | { | |
594 | struct list_head *head; | |
1da177e4 | 595 | |
5e4424af | 596 | spin_lock_bh(&queue->lock); |
1da177e4 LT |
597 | head = &queue->tasks[queue->maxpriority]; |
598 | for (;;) { | |
540a0f75 TM |
599 | while (!list_empty(head)) { |
600 | struct rpc_task *task; | |
601 | task = list_first_entry(head, | |
602 | struct rpc_task, | |
603 | u.tk_wait.list); | |
1da177e4 | 604 | task->tk_status = status; |
96ef13b2 | 605 | rpc_wake_up_task_queue_locked(queue, task); |
1da177e4 LT |
606 | } |
607 | if (head == &queue->tasks[0]) | |
608 | break; | |
609 | head--; | |
610 | } | |
5e4424af | 611 | spin_unlock_bh(&queue->lock); |
1da177e4 | 612 | } |
e8914c65 | 613 | EXPORT_SYMBOL_GPL(rpc_wake_up_status); |
1da177e4 | 614 | |
36df9aae TM |
615 | static void __rpc_queue_timer_fn(unsigned long ptr) |
616 | { | |
617 | struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr; | |
618 | struct rpc_task *task, *n; | |
619 | unsigned long expires, now, timeo; | |
620 | ||
621 | spin_lock(&queue->lock); | |
622 | expires = now = jiffies; | |
623 | list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) { | |
624 | timeo = task->u.tk_wait.expires; | |
625 | if (time_after_eq(now, timeo)) { | |
36df9aae TM |
626 | dprintk("RPC: %5u timeout\n", task->tk_pid); |
627 | task->tk_status = -ETIMEDOUT; | |
628 | rpc_wake_up_task_queue_locked(queue, task); | |
629 | continue; | |
630 | } | |
631 | if (expires == now || time_after(expires, timeo)) | |
632 | expires = timeo; | |
633 | } | |
634 | if (!list_empty(&queue->timer_list.list)) | |
635 | rpc_set_queue_timer(queue, expires); | |
636 | spin_unlock(&queue->lock); | |
637 | } | |
638 | ||
8014793b TM |
639 | static void __rpc_atrun(struct rpc_task *task) |
640 | { | |
6bd14416 TM |
641 | if (task->tk_status == -ETIMEDOUT) |
642 | task->tk_status = 0; | |
8014793b TM |
643 | } |
644 | ||
1da177e4 LT |
645 | /* |
646 | * Run a task at a later time | |
647 | */ | |
8014793b | 648 | void rpc_delay(struct rpc_task *task, unsigned long delay) |
1da177e4 LT |
649 | { |
650 | task->tk_timeout = delay; | |
5d00837b | 651 | rpc_sleep_on(&delay_queue, task, __rpc_atrun); |
1da177e4 | 652 | } |
e8914c65 | 653 | EXPORT_SYMBOL_GPL(rpc_delay); |
1da177e4 | 654 | |
4ce70ada TM |
655 | /* |
656 | * Helper to call task->tk_ops->rpc_call_prepare | |
657 | */ | |
aae2006e | 658 | void rpc_prepare_task(struct rpc_task *task) |
4ce70ada TM |
659 | { |
660 | task->tk_ops->rpc_call_prepare(task, task->tk_calldata); | |
661 | } | |
662 | ||
7fdcf13b TM |
663 | static void |
664 | rpc_init_task_statistics(struct rpc_task *task) | |
665 | { | |
666 | /* Initialize retry counters */ | |
667 | task->tk_garb_retry = 2; | |
668 | task->tk_cred_retry = 2; | |
669 | task->tk_rebind_retry = 2; | |
670 | ||
671 | /* starting timestamp */ | |
672 | task->tk_start = ktime_get(); | |
673 | } | |
674 | ||
675 | static void | |
676 | rpc_reset_task_statistics(struct rpc_task *task) | |
677 | { | |
678 | task->tk_timeouts = 0; | |
679 | task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_KILLED|RPC_TASK_SENT); | |
680 | ||
681 | rpc_init_task_statistics(task); | |
682 | } | |
683 | ||
d05fdb0c | 684 | /* |
963d8fe5 | 685 | * Helper that calls task->tk_ops->rpc_call_done if it exists |
d05fdb0c | 686 | */ |
abbcf28f | 687 | void rpc_exit_task(struct rpc_task *task) |
d05fdb0c | 688 | { |
abbcf28f | 689 | task->tk_action = NULL; |
963d8fe5 TM |
690 | if (task->tk_ops->rpc_call_done != NULL) { |
691 | task->tk_ops->rpc_call_done(task, task->tk_calldata); | |
d05fdb0c | 692 | if (task->tk_action != NULL) { |
abbcf28f TM |
693 | WARN_ON(RPC_ASSASSINATED(task)); |
694 | /* Always release the RPC slot and buffer memory */ | |
695 | xprt_release(task); | |
7fdcf13b | 696 | rpc_reset_task_statistics(task); |
d05fdb0c TM |
697 | } |
698 | } | |
d05fdb0c | 699 | } |
d9b6cd94 TM |
700 | |
701 | void rpc_exit(struct rpc_task *task, int status) | |
702 | { | |
703 | task->tk_status = status; | |
704 | task->tk_action = rpc_exit_task; | |
705 | if (RPC_IS_QUEUED(task)) | |
706 | rpc_wake_up_queued_task(task->tk_waitqueue, task); | |
707 | } | |
708 | EXPORT_SYMBOL_GPL(rpc_exit); | |
d05fdb0c | 709 | |
bbd5a1f9 TM |
710 | void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata) |
711 | { | |
a86dc496 | 712 | if (ops->rpc_release != NULL) |
bbd5a1f9 | 713 | ops->rpc_release(calldata); |
bbd5a1f9 TM |
714 | } |
715 | ||
1da177e4 LT |
716 | /* |
717 | * This is the RPC `scheduler' (or rather, the finite state machine). | |
718 | */ | |
2efef837 | 719 | static void __rpc_execute(struct rpc_task *task) |
1da177e4 | 720 | { |
eb9b55ab TM |
721 | struct rpc_wait_queue *queue; |
722 | int task_is_async = RPC_IS_ASYNC(task); | |
723 | int status = 0; | |
1da177e4 | 724 | |
46121cf7 CL |
725 | dprintk("RPC: %5u __rpc_execute flags=0x%x\n", |
726 | task->tk_pid, task->tk_flags); | |
1da177e4 | 727 | |
2bd4eef8 WAA |
728 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
729 | if (RPC_IS_QUEUED(task)) | |
730 | return; | |
1da177e4 | 731 | |
d05fdb0c | 732 | for (;;) { |
b55c5989 | 733 | void (*do_action)(struct rpc_task *); |
1da177e4 LT |
734 | |
735 | /* | |
b55c5989 | 736 | * Execute any pending callback first. |
1da177e4 | 737 | */ |
b55c5989 TM |
738 | do_action = task->tk_callback; |
739 | task->tk_callback = NULL; | |
740 | if (do_action == NULL) { | |
e020c680 TM |
741 | /* |
742 | * Perform the next FSM step. | |
b55c5989 TM |
743 | * tk_action may be NULL if the task has been killed. |
744 | * In particular, note that rpc_killall_tasks may | |
745 | * do this at any time, so beware when dereferencing. | |
e020c680 | 746 | */ |
b55c5989 TM |
747 | do_action = task->tk_action; |
748 | if (do_action == NULL) | |
1da177e4 | 749 | break; |
1da177e4 | 750 | } |
82b0a4c3 | 751 | trace_rpc_task_run_action(task->tk_client, task, task->tk_action); |
b55c5989 | 752 | do_action(task); |
1da177e4 LT |
753 | |
754 | /* | |
755 | * Lockless check for whether task is sleeping or not. | |
756 | */ | |
757 | if (!RPC_IS_QUEUED(task)) | |
758 | continue; | |
eb9b55ab TM |
759 | /* |
760 | * The queue->lock protects against races with | |
761 | * rpc_make_runnable(). | |
762 | * | |
763 | * Note that once we clear RPC_TASK_RUNNING on an asynchronous | |
764 | * rpc_task, rpc_make_runnable() can assign it to a | |
765 | * different workqueue. We therefore cannot assume that the | |
766 | * rpc_task pointer may still be dereferenced. | |
767 | */ | |
768 | queue = task->tk_waitqueue; | |
769 | spin_lock_bh(&queue->lock); | |
770 | if (!RPC_IS_QUEUED(task)) { | |
771 | spin_unlock_bh(&queue->lock); | |
1da177e4 LT |
772 | continue; |
773 | } | |
eb9b55ab TM |
774 | rpc_clear_running(task); |
775 | spin_unlock_bh(&queue->lock); | |
776 | if (task_is_async) | |
777 | return; | |
1da177e4 LT |
778 | |
779 | /* sync task: sleep here */ | |
46121cf7 | 780 | dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid); |
96651ab3 | 781 | status = out_of_line_wait_on_bit(&task->tk_runstate, |
150030b7 MW |
782 | RPC_TASK_QUEUED, rpc_wait_bit_killable, |
783 | TASK_KILLABLE); | |
96651ab3 | 784 | if (status == -ERESTARTSYS) { |
1da177e4 LT |
785 | /* |
786 | * When a sync task receives a signal, it exits with | |
787 | * -ERESTARTSYS. In order to catch any callbacks that | |
788 | * clean up after sleeping on some queue, we don't | |
789 | * break the loop here, but go around once more. | |
790 | */ | |
46121cf7 | 791 | dprintk("RPC: %5u got signal\n", task->tk_pid); |
96651ab3 TM |
792 | task->tk_flags |= RPC_TASK_KILLED; |
793 | rpc_exit(task, -ERESTARTSYS); | |
1da177e4 | 794 | } |
46121cf7 | 795 | dprintk("RPC: %5u sync task resuming\n", task->tk_pid); |
1da177e4 LT |
796 | } |
797 | ||
46121cf7 CL |
798 | dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status, |
799 | task->tk_status); | |
1da177e4 LT |
800 | /* Release all resources associated with the task */ |
801 | rpc_release_task(task); | |
1da177e4 LT |
802 | } |
803 | ||
804 | /* | |
805 | * User-visible entry point to the scheduler. | |
806 | * | |
807 | * This may be called recursively if e.g. an async NFS task updates | |
808 | * the attributes and finds that dirty pages must be flushed. | |
809 | * NOTE: Upon exit of this function the task is guaranteed to be | |
810 | * released. In particular note that tk_release() will have | |
811 | * been called, so your task memory may have been freed. | |
812 | */ | |
2efef837 | 813 | void rpc_execute(struct rpc_task *task) |
1da177e4 | 814 | { |
a76580fb TM |
815 | bool is_async = RPC_IS_ASYNC(task); |
816 | ||
44c28873 | 817 | rpc_set_active(task); |
d6a1ed08 | 818 | rpc_make_runnable(task); |
a76580fb | 819 | if (!is_async) |
d6a1ed08 | 820 | __rpc_execute(task); |
1da177e4 LT |
821 | } |
822 | ||
65f27f38 | 823 | static void rpc_async_schedule(struct work_struct *work) |
1da177e4 | 824 | { |
65f27f38 | 825 | __rpc_execute(container_of(work, struct rpc_task, u.tk_work)); |
1da177e4 LT |
826 | } |
827 | ||
02107148 CL |
828 | /** |
829 | * rpc_malloc - allocate an RPC buffer | |
830 | * @task: RPC task that will use this buffer | |
831 | * @size: requested byte size | |
1da177e4 | 832 | * |
c5a4dd8b | 833 | * To prevent rpciod from hanging, this allocator never sleeps, |
c6c8fe79 DR |
834 | * returning NULL and suppressing warning if the request cannot be serviced |
835 | * immediately. | |
c5a4dd8b CL |
836 | * The caller can arrange to sleep in a way that is safe for rpciod. |
837 | * | |
838 | * Most requests are 'small' (under 2KiB) and can be serviced from a | |
839 | * mempool, ensuring that NFS reads and writes can always proceed, | |
840 | * and that there is good locality of reference for these buffers. | |
841 | * | |
1da177e4 | 842 | * In order to avoid memory starvation triggering more writebacks of |
c5a4dd8b | 843 | * NFS requests, we avoid using GFP_KERNEL. |
1da177e4 | 844 | */ |
c5a4dd8b | 845 | void *rpc_malloc(struct rpc_task *task, size_t size) |
1da177e4 | 846 | { |
aa3d1fae | 847 | struct rpc_buffer *buf; |
c4a7ca77 | 848 | gfp_t gfp = GFP_NOIO | __GFP_NOWARN; |
a564b8f0 MG |
849 | |
850 | if (RPC_IS_SWAPPER(task)) | |
c4a7ca77 | 851 | gfp = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN; |
1da177e4 | 852 | |
aa3d1fae | 853 | size += sizeof(struct rpc_buffer); |
c5a4dd8b CL |
854 | if (size <= RPC_BUFFER_MAXSIZE) |
855 | buf = mempool_alloc(rpc_buffer_mempool, gfp); | |
1da177e4 | 856 | else |
c5a4dd8b | 857 | buf = kmalloc(size, gfp); |
ddce40df PZ |
858 | |
859 | if (!buf) | |
860 | return NULL; | |
861 | ||
aa3d1fae | 862 | buf->len = size; |
215d0678 | 863 | dprintk("RPC: %5u allocated buffer of size %zu at %p\n", |
c5a4dd8b | 864 | task->tk_pid, size, buf); |
aa3d1fae | 865 | return &buf->data; |
1da177e4 | 866 | } |
12444809 | 867 | EXPORT_SYMBOL_GPL(rpc_malloc); |
1da177e4 | 868 | |
02107148 CL |
869 | /** |
870 | * rpc_free - free buffer allocated via rpc_malloc | |
c5a4dd8b | 871 | * @buffer: buffer to free |
02107148 CL |
872 | * |
873 | */ | |
c5a4dd8b | 874 | void rpc_free(void *buffer) |
1da177e4 | 875 | { |
aa3d1fae CL |
876 | size_t size; |
877 | struct rpc_buffer *buf; | |
02107148 | 878 | |
c5a4dd8b CL |
879 | if (!buffer) |
880 | return; | |
aa3d1fae CL |
881 | |
882 | buf = container_of(buffer, struct rpc_buffer, data); | |
883 | size = buf->len; | |
c5a4dd8b | 884 | |
215d0678 | 885 | dprintk("RPC: freeing buffer of size %zu at %p\n", |
c5a4dd8b | 886 | size, buf); |
aa3d1fae | 887 | |
c5a4dd8b CL |
888 | if (size <= RPC_BUFFER_MAXSIZE) |
889 | mempool_free(buf, rpc_buffer_mempool); | |
890 | else | |
891 | kfree(buf); | |
1da177e4 | 892 | } |
12444809 | 893 | EXPORT_SYMBOL_GPL(rpc_free); |
1da177e4 LT |
894 | |
895 | /* | |
896 | * Creation and deletion of RPC task structures | |
897 | */ | |
47fe0648 | 898 | static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data) |
1da177e4 LT |
899 | { |
900 | memset(task, 0, sizeof(*task)); | |
44c28873 | 901 | atomic_set(&task->tk_count, 1); |
84115e1c TM |
902 | task->tk_flags = task_setup_data->flags; |
903 | task->tk_ops = task_setup_data->callback_ops; | |
904 | task->tk_calldata = task_setup_data->callback_data; | |
6529eba0 | 905 | INIT_LIST_HEAD(&task->tk_task); |
1da177e4 | 906 | |
3ff7576d TM |
907 | task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW; |
908 | task->tk_owner = current->tgid; | |
1da177e4 LT |
909 | |
910 | /* Initialize workqueue for async tasks */ | |
32bfb5c0 | 911 | task->tk_workqueue = task_setup_data->workqueue; |
1da177e4 | 912 | |
9d61498d TM |
913 | task->tk_xprt = xprt_get(task_setup_data->rpc_xprt); |
914 | ||
84115e1c TM |
915 | if (task->tk_ops->rpc_call_prepare != NULL) |
916 | task->tk_action = rpc_prepare_task; | |
963d8fe5 | 917 | |
7fdcf13b | 918 | rpc_init_task_statistics(task); |
ef759a2e | 919 | |
46121cf7 | 920 | dprintk("RPC: new task initialized, procpid %u\n", |
ba25f9dc | 921 | task_pid_nr(current)); |
1da177e4 LT |
922 | } |
923 | ||
924 | static struct rpc_task * | |
925 | rpc_alloc_task(void) | |
926 | { | |
a564b8f0 | 927 | return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOIO); |
1da177e4 LT |
928 | } |
929 | ||
1da177e4 | 930 | /* |
90c5755f | 931 | * Create a new task for the specified client. |
1da177e4 | 932 | */ |
84115e1c | 933 | struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data) |
1da177e4 | 934 | { |
e8f5d77c TM |
935 | struct rpc_task *task = setup_data->task; |
936 | unsigned short flags = 0; | |
937 | ||
938 | if (task == NULL) { | |
939 | task = rpc_alloc_task(); | |
19445b99 TM |
940 | if (task == NULL) { |
941 | rpc_release_calldata(setup_data->callback_ops, | |
942 | setup_data->callback_data); | |
943 | return ERR_PTR(-ENOMEM); | |
944 | } | |
e8f5d77c TM |
945 | flags = RPC_TASK_DYNAMIC; |
946 | } | |
1da177e4 | 947 | |
84115e1c | 948 | rpc_init_task(task, setup_data); |
e8f5d77c | 949 | task->tk_flags |= flags; |
46121cf7 | 950 | dprintk("RPC: allocated task %p\n", task); |
1da177e4 | 951 | return task; |
1da177e4 LT |
952 | } |
953 | ||
c6567ed1 TM |
954 | /* |
955 | * rpc_free_task - release rpc task and perform cleanups | |
956 | * | |
957 | * Note that we free up the rpc_task _after_ rpc_release_calldata() | |
958 | * in order to work around a workqueue dependency issue. | |
959 | * | |
960 | * Tejun Heo states: | |
961 | * "Workqueue currently considers two work items to be the same if they're | |
962 | * on the same address and won't execute them concurrently - ie. it | |
963 | * makes a work item which is queued again while being executed wait | |
964 | * for the previous execution to complete. | |
965 | * | |
966 | * If a work function frees the work item, and then waits for an event | |
967 | * which should be performed by another work item and *that* work item | |
968 | * recycles the freed work item, it can create a false dependency loop. | |
969 | * There really is no reliable way to detect this short of verifying | |
970 | * every memory free." | |
971 | * | |
972 | */ | |
32bfb5c0 | 973 | static void rpc_free_task(struct rpc_task *task) |
1da177e4 | 974 | { |
c6567ed1 TM |
975 | unsigned short tk_flags = task->tk_flags; |
976 | ||
977 | rpc_release_calldata(task->tk_ops, task->tk_calldata); | |
1da177e4 | 978 | |
c6567ed1 | 979 | if (tk_flags & RPC_TASK_DYNAMIC) { |
5e4424af TM |
980 | dprintk("RPC: %5u freeing task\n", task->tk_pid); |
981 | mempool_free(task, rpc_task_mempool); | |
982 | } | |
32bfb5c0 TM |
983 | } |
984 | ||
985 | static void rpc_async_release(struct work_struct *work) | |
986 | { | |
987 | rpc_free_task(container_of(work, struct rpc_task, u.tk_work)); | |
988 | } | |
989 | ||
bf294b41 | 990 | static void rpc_release_resources_task(struct rpc_task *task) |
32bfb5c0 | 991 | { |
87ed5003 | 992 | xprt_release(task); |
a271c5a0 | 993 | if (task->tk_msg.rpc_cred) { |
a17c2153 | 994 | put_rpccred(task->tk_msg.rpc_cred); |
a271c5a0 OH |
995 | task->tk_msg.rpc_cred = NULL; |
996 | } | |
58f9612c | 997 | rpc_task_release_client(task); |
bf294b41 TM |
998 | } |
999 | ||
1000 | static void rpc_final_put_task(struct rpc_task *task, | |
1001 | struct workqueue_struct *q) | |
1002 | { | |
1003 | if (q != NULL) { | |
32bfb5c0 | 1004 | INIT_WORK(&task->u.tk_work, rpc_async_release); |
bf294b41 | 1005 | queue_work(q, &task->u.tk_work); |
32bfb5c0 TM |
1006 | } else |
1007 | rpc_free_task(task); | |
e6b3c4db | 1008 | } |
bf294b41 TM |
1009 | |
1010 | static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q) | |
1011 | { | |
1012 | if (atomic_dec_and_test(&task->tk_count)) { | |
1013 | rpc_release_resources_task(task); | |
1014 | rpc_final_put_task(task, q); | |
1015 | } | |
1016 | } | |
1017 | ||
1018 | void rpc_put_task(struct rpc_task *task) | |
1019 | { | |
1020 | rpc_do_put_task(task, NULL); | |
1021 | } | |
e8914c65 | 1022 | EXPORT_SYMBOL_GPL(rpc_put_task); |
e6b3c4db | 1023 | |
bf294b41 TM |
1024 | void rpc_put_task_async(struct rpc_task *task) |
1025 | { | |
1026 | rpc_do_put_task(task, task->tk_workqueue); | |
1027 | } | |
1028 | EXPORT_SYMBOL_GPL(rpc_put_task_async); | |
1029 | ||
bde8f00c | 1030 | static void rpc_release_task(struct rpc_task *task) |
e6b3c4db | 1031 | { |
46121cf7 | 1032 | dprintk("RPC: %5u release task\n", task->tk_pid); |
1da177e4 | 1033 | |
0a0c2a57 | 1034 | WARN_ON_ONCE(RPC_IS_QUEUED(task)); |
1da177e4 | 1035 | |
bf294b41 | 1036 | rpc_release_resources_task(task); |
e6b3c4db | 1037 | |
bf294b41 TM |
1038 | /* |
1039 | * Note: at this point we have been removed from rpc_clnt->cl_tasks, | |
1040 | * so it should be safe to use task->tk_count as a test for whether | |
1041 | * or not any other processes still hold references to our rpc_task. | |
1042 | */ | |
1043 | if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) { | |
1044 | /* Wake up anyone who may be waiting for task completion */ | |
1045 | if (!rpc_complete_task(task)) | |
1046 | return; | |
1047 | } else { | |
1048 | if (!atomic_dec_and_test(&task->tk_count)) | |
1049 | return; | |
1050 | } | |
1051 | rpc_final_put_task(task, task->tk_workqueue); | |
1da177e4 LT |
1052 | } |
1053 | ||
b247bbf1 TM |
1054 | int rpciod_up(void) |
1055 | { | |
1056 | return try_module_get(THIS_MODULE) ? 0 : -EINVAL; | |
1057 | } | |
1058 | ||
1059 | void rpciod_down(void) | |
1060 | { | |
1061 | module_put(THIS_MODULE); | |
1062 | } | |
1063 | ||
1da177e4 | 1064 | /* |
b247bbf1 | 1065 | * Start up the rpciod workqueue. |
1da177e4 | 1066 | */ |
b247bbf1 | 1067 | static int rpciod_start(void) |
1da177e4 LT |
1068 | { |
1069 | struct workqueue_struct *wq; | |
ab418d70 | 1070 | |
1da177e4 LT |
1071 | /* |
1072 | * Create the rpciod thread and wait for it to start. | |
1073 | */ | |
ab418d70 | 1074 | dprintk("RPC: creating workqueue rpciod\n"); |
40a5f1b1 TM |
1075 | wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0); |
1076 | if (!wq) | |
1077 | goto out_failed; | |
1da177e4 | 1078 | rpciod_workqueue = wq; |
40a5f1b1 TM |
1079 | /* Note: highpri because network receive is latency sensitive */ |
1080 | wq = alloc_workqueue("xprtiod", WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1081 | if (!wq) | |
1082 | goto free_rpciod; | |
1083 | xprtiod_workqueue = wq; | |
1084 | return 1; | |
1085 | free_rpciod: | |
1086 | wq = rpciod_workqueue; | |
1087 | rpciod_workqueue = NULL; | |
1088 | destroy_workqueue(wq); | |
1089 | out_failed: | |
1090 | return 0; | |
1da177e4 LT |
1091 | } |
1092 | ||
b247bbf1 | 1093 | static void rpciod_stop(void) |
1da177e4 | 1094 | { |
b247bbf1 | 1095 | struct workqueue_struct *wq = NULL; |
ab418d70 | 1096 | |
b247bbf1 TM |
1097 | if (rpciod_workqueue == NULL) |
1098 | return; | |
ab418d70 | 1099 | dprintk("RPC: destroying workqueue rpciod\n"); |
1da177e4 | 1100 | |
b247bbf1 TM |
1101 | wq = rpciod_workqueue; |
1102 | rpciod_workqueue = NULL; | |
1103 | destroy_workqueue(wq); | |
40a5f1b1 TM |
1104 | wq = xprtiod_workqueue; |
1105 | xprtiod_workqueue = NULL; | |
1106 | destroy_workqueue(wq); | |
1da177e4 LT |
1107 | } |
1108 | ||
1da177e4 LT |
1109 | void |
1110 | rpc_destroy_mempool(void) | |
1111 | { | |
b247bbf1 | 1112 | rpciod_stop(); |
17a9618e JL |
1113 | mempool_destroy(rpc_buffer_mempool); |
1114 | mempool_destroy(rpc_task_mempool); | |
1115 | kmem_cache_destroy(rpc_task_slabp); | |
1116 | kmem_cache_destroy(rpc_buffer_slabp); | |
f6a1cc89 | 1117 | rpc_destroy_wait_queue(&delay_queue); |
1da177e4 LT |
1118 | } |
1119 | ||
1120 | int | |
1121 | rpc_init_mempool(void) | |
1122 | { | |
f6a1cc89 TM |
1123 | /* |
1124 | * The following is not strictly a mempool initialisation, | |
1125 | * but there is no harm in doing it here | |
1126 | */ | |
1127 | rpc_init_wait_queue(&delay_queue, "delayq"); | |
1128 | if (!rpciod_start()) | |
1129 | goto err_nomem; | |
1130 | ||
1da177e4 LT |
1131 | rpc_task_slabp = kmem_cache_create("rpc_tasks", |
1132 | sizeof(struct rpc_task), | |
1133 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1134 | NULL); |
1da177e4 LT |
1135 | if (!rpc_task_slabp) |
1136 | goto err_nomem; | |
1137 | rpc_buffer_slabp = kmem_cache_create("rpc_buffers", | |
1138 | RPC_BUFFER_MAXSIZE, | |
1139 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 1140 | NULL); |
1da177e4 LT |
1141 | if (!rpc_buffer_slabp) |
1142 | goto err_nomem; | |
93d2341c MD |
1143 | rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE, |
1144 | rpc_task_slabp); | |
1da177e4 LT |
1145 | if (!rpc_task_mempool) |
1146 | goto err_nomem; | |
93d2341c MD |
1147 | rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE, |
1148 | rpc_buffer_slabp); | |
1da177e4 LT |
1149 | if (!rpc_buffer_mempool) |
1150 | goto err_nomem; | |
1151 | return 0; | |
1152 | err_nomem: | |
1153 | rpc_destroy_mempool(); | |
1154 | return -ENOMEM; | |
1155 | } |