1 /* FS-Cache worker operation management routines
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * See Documentation/filesystems/caching/operations.txt
12 */
13
14 #define FSCACHE_DEBUG_LEVEL OPERATION
15 #include <linux/module.h>
16 #include <linux/seq_file.h>
17 #include <linux/slab.h>
18 #include "internal.h"
19
20 atomic_t fscache_op_debug_id;
21 EXPORT_SYMBOL(fscache_op_debug_id);
22
23 /**
24 * fscache_enqueue_operation - Enqueue an operation for processing
25 * @op: The operation to enqueue
26 *
27 * Enqueue an operation for processing by the FS-Cache thread pool.
28 *
29 * This will get its own ref on the object.
30 */
31 void fscache_enqueue_operation(struct fscache_operation *op)
32 {
33 _enter("{OBJ%x OP%x,%u}",
34 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
35
36 ASSERT(list_empty(&op->pend_link));
37 ASSERT(op->processor != NULL);
38 ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
39 ASSERTCMP(atomic_read(&op->usage), >, 0);
40 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
41
42 fscache_stat(&fscache_n_op_enqueue);
43 switch (op->flags & FSCACHE_OP_TYPE) {
44 case FSCACHE_OP_ASYNC:
45 _debug("queue async");
46 atomic_inc(&op->usage);
47 if (!queue_work(fscache_op_wq, &op->work))
48 fscache_put_operation(op);
49 break;
50 case FSCACHE_OP_MYTHREAD:
51 _debug("queue for caller's attention");
52 break;
53 default:
54 printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
55 op->flags);
56 BUG();
57 break;
58 }
59 }
60 EXPORT_SYMBOL(fscache_enqueue_operation);
61
62 /*
63 * start an op running
64 */
65 static void fscache_run_op(struct fscache_object *object,
66 struct fscache_operation *op)
67 {
68 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
69
70 op->state = FSCACHE_OP_ST_IN_PROGRESS;
71 object->n_in_progress++;
72 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
73 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
74 if (op->processor)
75 fscache_enqueue_operation(op);
76 fscache_stat(&fscache_n_op_run);
77 }
78
79 /*
80 * submit an exclusive operation for an object
81 * - other ops are excluded from running simultaneously with this one
82 * - this gets any extra refs it needs on an op
83 */
84 int fscache_submit_exclusive_op(struct fscache_object *object,
85 struct fscache_operation *op)
86 {
87 int ret;
88
89 _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
90
91 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
92 ASSERTCMP(atomic_read(&op->usage), >, 0);
93
94 spin_lock(&object->lock);
95 ASSERTCMP(object->n_ops, >=, object->n_in_progress);
96 ASSERTCMP(object->n_ops, >=, object->n_exclusive);
97 ASSERT(list_empty(&op->pend_link));
98
99 op->state = FSCACHE_OP_ST_PENDING;
100 if (fscache_object_is_active(object)) {
101 op->object = object;
102 object->n_ops++;
103 object->n_exclusive++; /* reads and writes must wait */
104
105 if (object->n_in_progress > 0) {
106 atomic_inc(&op->usage);
107 list_add_tail(&op->pend_link, &object->pending_ops);
108 fscache_stat(&fscache_n_op_pend);
109 } else if (!list_empty(&object->pending_ops)) {
110 atomic_inc(&op->usage);
111 list_add_tail(&op->pend_link, &object->pending_ops);
112 fscache_stat(&fscache_n_op_pend);
113 fscache_start_operations(object);
114 } else {
115 ASSERTCMP(object->n_in_progress, ==, 0);
116 fscache_run_op(object, op);
117 }
118
119 /* need to issue a new write op after this */
120 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
121 ret = 0;
122 } else if (object->state == FSCACHE_OBJECT_CREATING) {
123 op->object = object;
124 object->n_ops++;
125 object->n_exclusive++; /* reads and writes must wait */
126 atomic_inc(&op->usage);
127 list_add_tail(&op->pend_link, &object->pending_ops);
128 fscache_stat(&fscache_n_op_pend);
129 ret = 0;
130 } else {
131 /* If we're in any other state, there must have been an I/O
132 * error of some nature.
133 */
134 ASSERT(test_bit(FSCACHE_IOERROR, &object->cache->flags));
135 ret = -EIO;
136 }
137
138 spin_unlock(&object->lock);
139 return ret;
140 }
141
142 /*
143 * report an unexpected submission
144 */
145 static void fscache_report_unexpected_submission(struct fscache_object *object,
146 struct fscache_operation *op,
147 unsigned long ostate)
148 {
149 static bool once_only;
150 struct fscache_operation *p;
151 unsigned n;
152
153 if (once_only)
154 return;
155 once_only = true;
156
157 kdebug("unexpected submission OP%x [OBJ%x %s]",
158 op->debug_id, object->debug_id,
159 fscache_object_states[object->state]);
160 kdebug("objstate=%s [%s]",
161 fscache_object_states[object->state],
162 fscache_object_states[ostate]);
163 kdebug("objflags=%lx", object->flags);
164 kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
165 kdebug("ops=%u inp=%u exc=%u",
166 object->n_ops, object->n_in_progress, object->n_exclusive);
167
168 if (!list_empty(&object->pending_ops)) {
169 n = 0;
170 list_for_each_entry(p, &object->pending_ops, pend_link) {
171 ASSERTCMP(p->object, ==, object);
172 kdebug("%p %p", op->processor, op->release);
173 n++;
174 }
175
176 kdebug("n=%u", n);
177 }
178
179 dump_stack();
180 }
181
182 /*
183 * submit an operation for an object
184 * - objects may be submitted only in the following states:
185 * - during object creation (write ops may be submitted)
186 * - whilst the object is active
187 * - after an I/O error incurred in one of the two above states (op rejected)
188 * - this gets any extra refs it needs on an op
189 */
190 int fscache_submit_op(struct fscache_object *object,
191 struct fscache_operation *op)
192 {
193 unsigned long ostate;
194 int ret;
195
196 _enter("{OBJ%x OP%x},{%u}",
197 object->debug_id, op->debug_id, atomic_read(&op->usage));
198
199 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_INITIALISED);
200 ASSERTCMP(atomic_read(&op->usage), >, 0);
201
202 spin_lock(&object->lock);
203 ASSERTCMP(object->n_ops, >=, object->n_in_progress);
204 ASSERTCMP(object->n_ops, >=, object->n_exclusive);
205 ASSERT(list_empty(&op->pend_link));
206
207 ostate = object->state;
208 smp_rmb();
209
210 op->state = FSCACHE_OP_ST_PENDING;
211 if (fscache_object_is_active(object)) {
212 op->object = object;
213 object->n_ops++;
214
215 if (object->n_exclusive > 0) {
216 atomic_inc(&op->usage);
217 list_add_tail(&op->pend_link, &object->pending_ops);
218 fscache_stat(&fscache_n_op_pend);
219 } else if (!list_empty(&object->pending_ops)) {
220 atomic_inc(&op->usage);
221 list_add_tail(&op->pend_link, &object->pending_ops);
222 fscache_stat(&fscache_n_op_pend);
223 fscache_start_operations(object);
224 } else {
225 ASSERTCMP(object->n_exclusive, ==, 0);
226 fscache_run_op(object, op);
227 }
228 ret = 0;
229 } else if (object->state == FSCACHE_OBJECT_CREATING) {
230 op->object = object;
231 object->n_ops++;
232 atomic_inc(&op->usage);
233 list_add_tail(&op->pend_link, &object->pending_ops);
234 fscache_stat(&fscache_n_op_pend);
235 ret = 0;
236 } else if (object->state == FSCACHE_OBJECT_DYING ||
237 object->state == FSCACHE_OBJECT_LC_DYING ||
238 object->state == FSCACHE_OBJECT_WITHDRAWING) {
239 fscache_stat(&fscache_n_op_rejected);
240 op->state = FSCACHE_OP_ST_CANCELLED;
241 ret = -ENOBUFS;
242 } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
243 fscache_report_unexpected_submission(object, op, ostate);
244 ASSERT(!fscache_object_is_active(object));
245 op->state = FSCACHE_OP_ST_CANCELLED;
246 ret = -ENOBUFS;
247 } else {
248 op->state = FSCACHE_OP_ST_CANCELLED;
249 ret = -ENOBUFS;
250 }
251
252 spin_unlock(&object->lock);
253 return ret;
254 }
255
256 /*
257 * queue an object for withdrawal on error, aborting all following asynchronous
258 * operations
259 */
260 void fscache_abort_object(struct fscache_object *object)
261 {
262 _enter("{OBJ%x}", object->debug_id);
263
264 fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
265 }
266
267 /*
268 * jump start the operation processing on an object
269 * - caller must hold object->lock
270 */
271 void fscache_start_operations(struct fscache_object *object)
272 {
273 struct fscache_operation *op;
274 bool stop = false;
275
276 while (!list_empty(&object->pending_ops) && !stop) {
277 op = list_entry(object->pending_ops.next,
278 struct fscache_operation, pend_link);
279
280 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
281 if (object->n_in_progress > 0)
282 break;
283 stop = true;
284 }
285 list_del_init(&op->pend_link);
286 fscache_run_op(object, op);
287
288 /* the pending queue was holding a ref on the object */
289 fscache_put_operation(op);
290 }
291
292 ASSERTCMP(object->n_in_progress, <=, object->n_ops);
293
294 _debug("woke %d ops on OBJ%x",
295 object->n_in_progress, object->debug_id);
296 }
297
298 /*
299 * cancel an operation that's pending on an object
300 */
301 int fscache_cancel_op(struct fscache_operation *op,
302 void (*do_cancel)(struct fscache_operation *))
303 {
304 struct fscache_object *object = op->object;
305 int ret;
306
307 _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
308
309 ASSERTCMP(op->state, >=, FSCACHE_OP_ST_PENDING);
310 ASSERTCMP(op->state, !=, FSCACHE_OP_ST_CANCELLED);
311 ASSERTCMP(atomic_read(&op->usage), >, 0);
312
313 spin_lock(&object->lock);
314
315 ret = -EBUSY;
316 if (op->state == FSCACHE_OP_ST_PENDING) {
317 ASSERT(!list_empty(&op->pend_link));
318 fscache_stat(&fscache_n_op_cancelled);
319 list_del_init(&op->pend_link);
320 if (do_cancel)
321 do_cancel(op);
322 op->state = FSCACHE_OP_ST_CANCELLED;
323 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
324 object->n_exclusive--;
325 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
326 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
327 fscache_put_operation(op);
328 ret = 0;
329 }
330
331 spin_unlock(&object->lock);
332 _leave(" = %d", ret);
333 return ret;
334 }
335
336 /*
337 * Cancel all pending operations on an object
338 */
339 void fscache_cancel_all_ops(struct fscache_object *object)
340 {
341 struct fscache_operation *op;
342
343 _enter("OBJ%x", object->debug_id);
344
345 spin_lock(&object->lock);
346
347 while (!list_empty(&object->pending_ops)) {
348 op = list_entry(object->pending_ops.next,
349 struct fscache_operation, pend_link);
350 fscache_stat(&fscache_n_op_cancelled);
351 list_del_init(&op->pend_link);
352
353 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
354 op->state = FSCACHE_OP_ST_CANCELLED;
355
356 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
357 object->n_exclusive--;
358 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
359 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
360 fscache_put_operation(op);
361 cond_resched_lock(&object->lock);
362 }
363
364 spin_unlock(&object->lock);
365 _leave("");
366 }
367
368 /*
369 * Record the completion or cancellation of an in-progress operation.
370 */
371 void fscache_op_complete(struct fscache_operation *op, bool cancelled)
372 {
373 struct fscache_object *object = op->object;
374
375 _enter("OBJ%x", object->debug_id);
376
377 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_IN_PROGRESS);
378 ASSERTCMP(object->n_in_progress, >, 0);
379 ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
380 object->n_exclusive, >, 0);
381 ASSERTIFCMP(test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags),
382 object->n_in_progress, ==, 1);
383
384 spin_lock(&object->lock);
385
386 op->state = cancelled ?
387 FSCACHE_OP_ST_CANCELLED : FSCACHE_OP_ST_COMPLETE;
388
389 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
390 object->n_exclusive--;
391 object->n_in_progress--;
392 if (object->n_in_progress == 0)
393 fscache_start_operations(object);
394
395 spin_unlock(&object->lock);
396 _leave("");
397 }
398 EXPORT_SYMBOL(fscache_op_complete);
399
400 /*
401 * release an operation
402 * - queues pending ops if this is the last in-progress op
403 */
404 void fscache_put_operation(struct fscache_operation *op)
405 {
406 struct fscache_object *object;
407 struct fscache_cache *cache;
408
409 _enter("{OBJ%x OP%x,%d}",
410 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
411
412 ASSERTCMP(atomic_read(&op->usage), >, 0);
413
414 if (!atomic_dec_and_test(&op->usage))
415 return;
416
417 _debug("PUT OP");
418 ASSERTIFCMP(op->state != FSCACHE_OP_ST_COMPLETE,
419 op->state, ==, FSCACHE_OP_ST_CANCELLED);
420 op->state = FSCACHE_OP_ST_DEAD;
421
422 fscache_stat(&fscache_n_op_release);
423
424 if (op->release) {
425 op->release(op);
426 op->release = NULL;
427 }
428
429 object = op->object;
430
431 if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags)) {
432 if (atomic_dec_and_test(&object->n_reads)) {
433 clear_bit(FSCACHE_COOKIE_WAITING_ON_READS,
434 &object->cookie->flags);
435 wake_up_bit(&object->cookie->flags,
436 FSCACHE_COOKIE_WAITING_ON_READS);
437 }
438 }
439
440 /* now... we may get called with the object spinlock held, so we
441 * complete the cleanup here only if we can immediately acquire the
442 * lock, and defer it otherwise */
443 if (!spin_trylock(&object->lock)) {
444 _debug("defer put");
445 fscache_stat(&fscache_n_op_deferred_release);
446
447 cache = object->cache;
448 spin_lock(&cache->op_gc_list_lock);
449 list_add_tail(&op->pend_link, &cache->op_gc_list);
450 spin_unlock(&cache->op_gc_list_lock);
451 schedule_work(&cache->op_gc);
452 _leave(" [defer]");
453 return;
454 }
455
456 ASSERTCMP(object->n_ops, >, 0);
457 object->n_ops--;
458 if (object->n_ops == 0)
459 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
460
461 spin_unlock(&object->lock);
462
463 kfree(op);
464 _leave(" [done]");
465 }
466 EXPORT_SYMBOL(fscache_put_operation);
467
468 /*
469 * garbage collect operations that have had their release deferred
470 */
471 void fscache_operation_gc(struct work_struct *work)
472 {
473 struct fscache_operation *op;
474 struct fscache_object *object;
475 struct fscache_cache *cache =
476 container_of(work, struct fscache_cache, op_gc);
477 int count = 0;
478
479 _enter("");
480
481 do {
482 spin_lock(&cache->op_gc_list_lock);
483 if (list_empty(&cache->op_gc_list)) {
484 spin_unlock(&cache->op_gc_list_lock);
485 break;
486 }
487
488 op = list_entry(cache->op_gc_list.next,
489 struct fscache_operation, pend_link);
490 list_del(&op->pend_link);
491 spin_unlock(&cache->op_gc_list_lock);
492
493 object = op->object;
494 spin_lock(&object->lock);
495
496 _debug("GC DEFERRED REL OBJ%x OP%x",
497 object->debug_id, op->debug_id);
498 fscache_stat(&fscache_n_op_gc);
499
500 ASSERTCMP(atomic_read(&op->usage), ==, 0);
501 ASSERTCMP(op->state, ==, FSCACHE_OP_ST_DEAD);
502
503 ASSERTCMP(object->n_ops, >, 0);
504 object->n_ops--;
505 if (object->n_ops == 0)
506 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
507
508 spin_unlock(&object->lock);
509 kfree(op);
510
511 } while (count++ < 20);
512
513 if (!list_empty(&cache->op_gc_list))
514 schedule_work(&cache->op_gc);
515
516 _leave("");
517 }
518
519 /*
520 * execute an operation using fs_op_wq to provide processing context -
521 * the caller holds a ref to this object, so we don't need to hold one
522 */
523 void fscache_op_work_func(struct work_struct *work)
524 {
525 struct fscache_operation *op =
526 container_of(work, struct fscache_operation, work);
527 unsigned long start;
528
529 _enter("{OBJ%x OP%x,%d}",
530 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
531
532 ASSERT(op->processor != NULL);
533 start = jiffies;
534 op->processor(op);
535 fscache_hist(fscache_ops_histogram, start);
536 fscache_put_operation(op);
537
538 _leave("");
539 }
540
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.