]> bbs.cooldavid.org Git - net-next-2.6.git/blob - fs/fscache/operation.c
Merge branch 'kbuild/clean' into kbuild/kbuild
[net-next-2.6.git] / fs / fscache / operation.c
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         fscache_set_op_state(op, "EnQ");
37
38         ASSERT(list_empty(&op->pend_link));
39         ASSERT(op->processor != NULL);
40         ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
41         ASSERTCMP(atomic_read(&op->usage), >, 0);
42
43         fscache_stat(&fscache_n_op_enqueue);
44         switch (op->flags & FSCACHE_OP_TYPE) {
45         case FSCACHE_OP_FAST:
46                 _debug("queue fast");
47                 atomic_inc(&op->usage);
48                 if (!schedule_work(&op->fast_work))
49                         fscache_put_operation(op);
50                 break;
51         case FSCACHE_OP_SLOW:
52                 _debug("queue slow");
53                 slow_work_enqueue(&op->slow_work);
54                 break;
55         case FSCACHE_OP_MYTHREAD:
56                 _debug("queue for caller's attention");
57                 break;
58         default:
59                 printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
60                        op->flags);
61                 BUG();
62                 break;
63         }
64 }
65 EXPORT_SYMBOL(fscache_enqueue_operation);
66
67 /*
68  * start an op running
69  */
70 static void fscache_run_op(struct fscache_object *object,
71                            struct fscache_operation *op)
72 {
73         fscache_set_op_state(op, "Run");
74
75         object->n_in_progress++;
76         if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
77                 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
78         if (op->processor)
79                 fscache_enqueue_operation(op);
80         fscache_stat(&fscache_n_op_run);
81 }
82
83 /*
84  * submit an exclusive operation for an object
85  * - other ops are excluded from running simultaneously with this one
86  * - this gets any extra refs it needs on an op
87  */
88 int fscache_submit_exclusive_op(struct fscache_object *object,
89                                 struct fscache_operation *op)
90 {
91         int ret;
92
93         _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
94
95         fscache_set_op_state(op, "SubmitX");
96
97         spin_lock(&object->lock);
98         ASSERTCMP(object->n_ops, >=, object->n_in_progress);
99         ASSERTCMP(object->n_ops, >=, object->n_exclusive);
100         ASSERT(list_empty(&op->pend_link));
101
102         ret = -ENOBUFS;
103         if (fscache_object_is_active(object)) {
104                 op->object = object;
105                 object->n_ops++;
106                 object->n_exclusive++;  /* reads and writes must wait */
107
108                 if (object->n_ops > 0) {
109                         atomic_inc(&op->usage);
110                         list_add_tail(&op->pend_link, &object->pending_ops);
111                         fscache_stat(&fscache_n_op_pend);
112                 } else if (!list_empty(&object->pending_ops)) {
113                         atomic_inc(&op->usage);
114                         list_add_tail(&op->pend_link, &object->pending_ops);
115                         fscache_stat(&fscache_n_op_pend);
116                         fscache_start_operations(object);
117                 } else {
118                         ASSERTCMP(object->n_in_progress, ==, 0);
119                         fscache_run_op(object, op);
120                 }
121
122                 /* need to issue a new write op after this */
123                 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
124                 ret = 0;
125         } else if (object->state == FSCACHE_OBJECT_CREATING) {
126                 op->object = object;
127                 object->n_ops++;
128                 object->n_exclusive++;  /* reads and writes must wait */
129                 atomic_inc(&op->usage);
130                 list_add_tail(&op->pend_link, &object->pending_ops);
131                 fscache_stat(&fscache_n_op_pend);
132                 ret = 0;
133         } else {
134                 /* not allowed to submit ops in any other state */
135                 BUG();
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(atomic_read(&op->usage), >, 0);
200
201         fscache_set_op_state(op, "Submit");
202
203         spin_lock(&object->lock);
204         ASSERTCMP(object->n_ops, >=, object->n_in_progress);
205         ASSERTCMP(object->n_ops, >=, object->n_exclusive);
206         ASSERT(list_empty(&op->pend_link));
207
208         ostate = object->state;
209         smp_rmb();
210
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                 ret = -ENOBUFS;
241         } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
242                 fscache_report_unexpected_submission(object, op, ostate);
243                 ASSERT(!fscache_object_is_active(object));
244                 ret = -ENOBUFS;
245         } else {
246                 ret = -ENOBUFS;
247         }
248
249         spin_unlock(&object->lock);
250         return ret;
251 }
252
253 /*
254  * queue an object for withdrawal on error, aborting all following asynchronous
255  * operations
256  */
257 void fscache_abort_object(struct fscache_object *object)
258 {
259         _enter("{OBJ%x}", object->debug_id);
260
261         fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
262 }
263
264 /*
265  * jump start the operation processing on an object
266  * - caller must hold object->lock
267  */
268 void fscache_start_operations(struct fscache_object *object)
269 {
270         struct fscache_operation *op;
271         bool stop = false;
272
273         while (!list_empty(&object->pending_ops) && !stop) {
274                 op = list_entry(object->pending_ops.next,
275                                 struct fscache_operation, pend_link);
276
277                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
278                         if (object->n_in_progress > 0)
279                                 break;
280                         stop = true;
281                 }
282                 list_del_init(&op->pend_link);
283                 fscache_run_op(object, op);
284
285                 /* the pending queue was holding a ref on the object */
286                 fscache_put_operation(op);
287         }
288
289         ASSERTCMP(object->n_in_progress, <=, object->n_ops);
290
291         _debug("woke %d ops on OBJ%x",
292                object->n_in_progress, object->debug_id);
293 }
294
295 /*
296  * cancel an operation that's pending on an object
297  */
298 int fscache_cancel_op(struct fscache_operation *op)
299 {
300         struct fscache_object *object = op->object;
301         int ret;
302
303         _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
304
305         spin_lock(&object->lock);
306
307         ret = -EBUSY;
308         if (!list_empty(&op->pend_link)) {
309                 fscache_stat(&fscache_n_op_cancelled);
310                 list_del_init(&op->pend_link);
311                 object->n_ops--;
312                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
313                         object->n_exclusive--;
314                 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
315                         wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
316                 fscache_put_operation(op);
317                 ret = 0;
318         }
319
320         spin_unlock(&object->lock);
321         _leave(" = %d", ret);
322         return ret;
323 }
324
325 /*
326  * release an operation
327  * - queues pending ops if this is the last in-progress op
328  */
329 void fscache_put_operation(struct fscache_operation *op)
330 {
331         struct fscache_object *object;
332         struct fscache_cache *cache;
333
334         _enter("{OBJ%x OP%x,%d}",
335                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
336
337         ASSERTCMP(atomic_read(&op->usage), >, 0);
338
339         if (!atomic_dec_and_test(&op->usage))
340                 return;
341
342         fscache_set_op_state(op, "Put");
343
344         _debug("PUT OP");
345         if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
346                 BUG();
347
348         fscache_stat(&fscache_n_op_release);
349
350         if (op->release) {
351                 op->release(op);
352                 op->release = NULL;
353         }
354
355         object = op->object;
356
357         if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
358                 atomic_dec(&object->n_reads);
359
360         /* now... we may get called with the object spinlock held, so we
361          * complete the cleanup here only if we can immediately acquire the
362          * lock, and defer it otherwise */
363         if (!spin_trylock(&object->lock)) {
364                 _debug("defer put");
365                 fscache_stat(&fscache_n_op_deferred_release);
366
367                 cache = object->cache;
368                 spin_lock(&cache->op_gc_list_lock);
369                 list_add_tail(&op->pend_link, &cache->op_gc_list);
370                 spin_unlock(&cache->op_gc_list_lock);
371                 schedule_work(&cache->op_gc);
372                 _leave(" [defer]");
373                 return;
374         }
375
376         if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
377                 ASSERTCMP(object->n_exclusive, >, 0);
378                 object->n_exclusive--;
379         }
380
381         ASSERTCMP(object->n_in_progress, >, 0);
382         object->n_in_progress--;
383         if (object->n_in_progress == 0)
384                 fscache_start_operations(object);
385
386         ASSERTCMP(object->n_ops, >, 0);
387         object->n_ops--;
388         if (object->n_ops == 0)
389                 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
390
391         spin_unlock(&object->lock);
392
393         kfree(op);
394         _leave(" [done]");
395 }
396 EXPORT_SYMBOL(fscache_put_operation);
397
398 /*
399  * garbage collect operations that have had their release deferred
400  */
401 void fscache_operation_gc(struct work_struct *work)
402 {
403         struct fscache_operation *op;
404         struct fscache_object *object;
405         struct fscache_cache *cache =
406                 container_of(work, struct fscache_cache, op_gc);
407         int count = 0;
408
409         _enter("");
410
411         do {
412                 spin_lock(&cache->op_gc_list_lock);
413                 if (list_empty(&cache->op_gc_list)) {
414                         spin_unlock(&cache->op_gc_list_lock);
415                         break;
416                 }
417
418                 op = list_entry(cache->op_gc_list.next,
419                                 struct fscache_operation, pend_link);
420                 list_del(&op->pend_link);
421                 spin_unlock(&cache->op_gc_list_lock);
422
423                 object = op->object;
424
425                 _debug("GC DEFERRED REL OBJ%x OP%x",
426                        object->debug_id, op->debug_id);
427                 fscache_stat(&fscache_n_op_gc);
428
429                 ASSERTCMP(atomic_read(&op->usage), ==, 0);
430
431                 spin_lock(&object->lock);
432                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
433                         ASSERTCMP(object->n_exclusive, >, 0);
434                         object->n_exclusive--;
435                 }
436
437                 ASSERTCMP(object->n_in_progress, >, 0);
438                 object->n_in_progress--;
439                 if (object->n_in_progress == 0)
440                         fscache_start_operations(object);
441
442                 ASSERTCMP(object->n_ops, >, 0);
443                 object->n_ops--;
444                 if (object->n_ops == 0)
445                         fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
446
447                 spin_unlock(&object->lock);
448
449         } while (count++ < 20);
450
451         if (!list_empty(&cache->op_gc_list))
452                 schedule_work(&cache->op_gc);
453
454         _leave("");
455 }
456
457 /*
458  * allow the slow work item processor to get a ref on an operation
459  */
460 static int fscache_op_get_ref(struct slow_work *work)
461 {
462         struct fscache_operation *op =
463                 container_of(work, struct fscache_operation, slow_work);
464
465         atomic_inc(&op->usage);
466         return 0;
467 }
468
469 /*
470  * allow the slow work item processor to discard a ref on an operation
471  */
472 static void fscache_op_put_ref(struct slow_work *work)
473 {
474         struct fscache_operation *op =
475                 container_of(work, struct fscache_operation, slow_work);
476
477         fscache_put_operation(op);
478 }
479
480 /*
481  * execute an operation using the slow thread pool to provide processing context
482  * - the caller holds a ref to this object, so we don't need to hold one
483  */
484 static void fscache_op_execute(struct slow_work *work)
485 {
486         struct fscache_operation *op =
487                 container_of(work, struct fscache_operation, slow_work);
488         unsigned long start;
489
490         _enter("{OBJ%x OP%x,%d}",
491                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
492
493         ASSERT(op->processor != NULL);
494         start = jiffies;
495         op->processor(op);
496         fscache_hist(fscache_ops_histogram, start);
497
498         _leave("");
499 }
500
501 /*
502  * describe an operation for slow-work debugging
503  */
504 #ifdef CONFIG_SLOW_WORK_DEBUG
505 static void fscache_op_desc(struct slow_work *work, struct seq_file *m)
506 {
507         struct fscache_operation *op =
508                 container_of(work, struct fscache_operation, slow_work);
509
510         seq_printf(m, "FSC: OBJ%x OP%x: %s/%s fl=%lx",
511                    op->object->debug_id, op->debug_id,
512                    op->name, op->state, op->flags);
513 }
514 #endif
515
516 const struct slow_work_ops fscache_op_slow_work_ops = {
517         .owner          = THIS_MODULE,
518         .get_ref        = fscache_op_get_ref,
519         .put_ref        = fscache_op_put_ref,
520         .execute        = fscache_op_execute,
521 #ifdef CONFIG_SLOW_WORK_DEBUG
522         .desc           = fscache_op_desc,
523 #endif
524 };