FS-Cache: Allow the current state of all objects to be dumped
[safe/jmp/linux-2.6] / 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 "internal.h"
18
19 atomic_t fscache_op_debug_id;
20 EXPORT_SYMBOL(fscache_op_debug_id);
21
22 /**
23  * fscache_enqueue_operation - Enqueue an operation for processing
24  * @op: The operation to enqueue
25  *
26  * Enqueue an operation for processing by the FS-Cache thread pool.
27  *
28  * This will get its own ref on the object.
29  */
30 void fscache_enqueue_operation(struct fscache_operation *op)
31 {
32         _enter("{OBJ%x OP%x,%u}",
33                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
34
35         fscache_set_op_state(op, "EnQ");
36
37         ASSERT(op->processor != NULL);
38         ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
39         ASSERTCMP(atomic_read(&op->usage), >, 0);
40
41         if (list_empty(&op->pend_link)) {
42                 switch (op->flags & FSCACHE_OP_TYPE) {
43                 case FSCACHE_OP_FAST:
44                         _debug("queue fast");
45                         atomic_inc(&op->usage);
46                         if (!schedule_work(&op->fast_work))
47                                 fscache_put_operation(op);
48                         break;
49                 case FSCACHE_OP_SLOW:
50                         _debug("queue slow");
51                         slow_work_enqueue(&op->slow_work);
52                         break;
53                 case FSCACHE_OP_MYTHREAD:
54                         _debug("queue for caller's attention");
55                         break;
56                 default:
57                         printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
58                                op->flags);
59                         BUG();
60                         break;
61                 }
62                 fscache_stat(&fscache_n_op_enqueue);
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
101         ret = -ENOBUFS;
102         if (fscache_object_is_active(object)) {
103                 op->object = object;
104                 object->n_ops++;
105                 object->n_exclusive++;  /* reads and writes must wait */
106
107                 if (object->n_ops > 0) {
108                         atomic_inc(&op->usage);
109                         list_add_tail(&op->pend_link, &object->pending_ops);
110                         fscache_stat(&fscache_n_op_pend);
111                 } else if (!list_empty(&object->pending_ops)) {
112                         atomic_inc(&op->usage);
113                         list_add_tail(&op->pend_link, &object->pending_ops);
114                         fscache_stat(&fscache_n_op_pend);
115                         fscache_start_operations(object);
116                 } else {
117                         ASSERTCMP(object->n_in_progress, ==, 0);
118                         fscache_run_op(object, op);
119                 }
120
121                 /* need to issue a new write op after this */
122                 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
123                 ret = 0;
124         } else if (object->state == FSCACHE_OBJECT_CREATING) {
125                 op->object = object;
126                 object->n_ops++;
127                 object->n_exclusive++;  /* reads and writes must wait */
128                 atomic_inc(&op->usage);
129                 list_add_tail(&op->pend_link, &object->pending_ops);
130                 fscache_stat(&fscache_n_op_pend);
131                 ret = 0;
132         } else {
133                 /* not allowed to submit ops in any other state */
134                 BUG();
135         }
136
137         spin_unlock(&object->lock);
138         return ret;
139 }
140
141 /*
142  * report an unexpected submission
143  */
144 static void fscache_report_unexpected_submission(struct fscache_object *object,
145                                                  struct fscache_operation *op,
146                                                  unsigned long ostate)
147 {
148         static bool once_only;
149         struct fscache_operation *p;
150         unsigned n;
151
152         if (once_only)
153                 return;
154         once_only = true;
155
156         kdebug("unexpected submission OP%x [OBJ%x %s]",
157                op->debug_id, object->debug_id,
158                fscache_object_states[object->state]);
159         kdebug("objstate=%s [%s]",
160                fscache_object_states[object->state],
161                fscache_object_states[ostate]);
162         kdebug("objflags=%lx", object->flags);
163         kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
164         kdebug("ops=%u inp=%u exc=%u",
165                object->n_ops, object->n_in_progress, object->n_exclusive);
166
167         if (!list_empty(&object->pending_ops)) {
168                 n = 0;
169                 list_for_each_entry(p, &object->pending_ops, pend_link) {
170                         ASSERTCMP(p->object, ==, object);
171                         kdebug("%p %p", op->processor, op->release);
172                         n++;
173                 }
174
175                 kdebug("n=%u", n);
176         }
177
178         dump_stack();
179 }
180
181 /*
182  * submit an operation for an object
183  * - objects may be submitted only in the following states:
184  *   - during object creation (write ops may be submitted)
185  *   - whilst the object is active
186  *   - after an I/O error incurred in one of the two above states (op rejected)
187  * - this gets any extra refs it needs on an op
188  */
189 int fscache_submit_op(struct fscache_object *object,
190                       struct fscache_operation *op)
191 {
192         unsigned long ostate;
193         int ret;
194
195         _enter("{OBJ%x OP%x},{%u}",
196                object->debug_id, op->debug_id, atomic_read(&op->usage));
197
198         ASSERTCMP(atomic_read(&op->usage), >, 0);
199
200         fscache_set_op_state(op, "Submit");
201
202         spin_lock(&object->lock);
203         ASSERTCMP(object->n_ops, >=, object->n_in_progress);
204         ASSERTCMP(object->n_ops, >=, object->n_exclusive);
205
206         ostate = object->state;
207         smp_rmb();
208
209         if (fscache_object_is_active(object)) {
210                 op->object = object;
211                 object->n_ops++;
212
213                 if (object->n_exclusive > 0) {
214                         atomic_inc(&op->usage);
215                         list_add_tail(&op->pend_link, &object->pending_ops);
216                         fscache_stat(&fscache_n_op_pend);
217                 } else if (!list_empty(&object->pending_ops)) {
218                         atomic_inc(&op->usage);
219                         list_add_tail(&op->pend_link, &object->pending_ops);
220                         fscache_stat(&fscache_n_op_pend);
221                         fscache_start_operations(object);
222                 } else {
223                         ASSERTCMP(object->n_exclusive, ==, 0);
224                         fscache_run_op(object, op);
225                 }
226                 ret = 0;
227         } else if (object->state == FSCACHE_OBJECT_CREATING) {
228                 op->object = object;
229                 object->n_ops++;
230                 atomic_inc(&op->usage);
231                 list_add_tail(&op->pend_link, &object->pending_ops);
232                 fscache_stat(&fscache_n_op_pend);
233                 ret = 0;
234         } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
235                 fscache_report_unexpected_submission(object, op, ostate);
236                 ASSERT(!fscache_object_is_active(object));
237                 ret = -ENOBUFS;
238         } else {
239                 ret = -ENOBUFS;
240         }
241
242         spin_unlock(&object->lock);
243         return ret;
244 }
245
246 /*
247  * queue an object for withdrawal on error, aborting all following asynchronous
248  * operations
249  */
250 void fscache_abort_object(struct fscache_object *object)
251 {
252         _enter("{OBJ%x}", object->debug_id);
253
254         fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
255 }
256
257 /*
258  * jump start the operation processing on an object
259  * - caller must hold object->lock
260  */
261 void fscache_start_operations(struct fscache_object *object)
262 {
263         struct fscache_operation *op;
264         bool stop = false;
265
266         while (!list_empty(&object->pending_ops) && !stop) {
267                 op = list_entry(object->pending_ops.next,
268                                 struct fscache_operation, pend_link);
269
270                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
271                         if (object->n_in_progress > 0)
272                                 break;
273                         stop = true;
274                 }
275                 list_del_init(&op->pend_link);
276                 object->n_in_progress++;
277
278                 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
279                         wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
280                 if (op->processor)
281                         fscache_enqueue_operation(op);
282
283                 /* the pending queue was holding a ref on the object */
284                 fscache_put_operation(op);
285         }
286
287         ASSERTCMP(object->n_in_progress, <=, object->n_ops);
288
289         _debug("woke %d ops on OBJ%x",
290                object->n_in_progress, object->debug_id);
291 }
292
293 /*
294  * release an operation
295  * - queues pending ops if this is the last in-progress op
296  */
297 void fscache_put_operation(struct fscache_operation *op)
298 {
299         struct fscache_object *object;
300         struct fscache_cache *cache;
301
302         _enter("{OBJ%x OP%x,%d}",
303                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
304
305         ASSERTCMP(atomic_read(&op->usage), >, 0);
306
307         if (!atomic_dec_and_test(&op->usage))
308                 return;
309
310         fscache_set_op_state(op, "Put");
311
312         _debug("PUT OP");
313         if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
314                 BUG();
315
316         fscache_stat(&fscache_n_op_release);
317
318         if (op->release) {
319                 op->release(op);
320                 op->release = NULL;
321         }
322
323         object = op->object;
324
325         if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
326                 atomic_dec(&object->n_reads);
327
328         /* now... we may get called with the object spinlock held, so we
329          * complete the cleanup here only if we can immediately acquire the
330          * lock, and defer it otherwise */
331         if (!spin_trylock(&object->lock)) {
332                 _debug("defer put");
333                 fscache_stat(&fscache_n_op_deferred_release);
334
335                 cache = object->cache;
336                 spin_lock(&cache->op_gc_list_lock);
337                 list_add_tail(&op->pend_link, &cache->op_gc_list);
338                 spin_unlock(&cache->op_gc_list_lock);
339                 schedule_work(&cache->op_gc);
340                 _leave(" [defer]");
341                 return;
342         }
343
344         if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
345                 ASSERTCMP(object->n_exclusive, >, 0);
346                 object->n_exclusive--;
347         }
348
349         ASSERTCMP(object->n_in_progress, >, 0);
350         object->n_in_progress--;
351         if (object->n_in_progress == 0)
352                 fscache_start_operations(object);
353
354         ASSERTCMP(object->n_ops, >, 0);
355         object->n_ops--;
356         if (object->n_ops == 0)
357                 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
358
359         spin_unlock(&object->lock);
360
361         kfree(op);
362         _leave(" [done]");
363 }
364 EXPORT_SYMBOL(fscache_put_operation);
365
366 /*
367  * garbage collect operations that have had their release deferred
368  */
369 void fscache_operation_gc(struct work_struct *work)
370 {
371         struct fscache_operation *op;
372         struct fscache_object *object;
373         struct fscache_cache *cache =
374                 container_of(work, struct fscache_cache, op_gc);
375         int count = 0;
376
377         _enter("");
378
379         do {
380                 spin_lock(&cache->op_gc_list_lock);
381                 if (list_empty(&cache->op_gc_list)) {
382                         spin_unlock(&cache->op_gc_list_lock);
383                         break;
384                 }
385
386                 op = list_entry(cache->op_gc_list.next,
387                                 struct fscache_operation, pend_link);
388                 list_del(&op->pend_link);
389                 spin_unlock(&cache->op_gc_list_lock);
390
391                 object = op->object;
392
393                 _debug("GC DEFERRED REL OBJ%x OP%x",
394                        object->debug_id, op->debug_id);
395                 fscache_stat(&fscache_n_op_gc);
396
397                 ASSERTCMP(atomic_read(&op->usage), ==, 0);
398
399                 spin_lock(&object->lock);
400                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
401                         ASSERTCMP(object->n_exclusive, >, 0);
402                         object->n_exclusive--;
403                 }
404
405                 ASSERTCMP(object->n_in_progress, >, 0);
406                 object->n_in_progress--;
407                 if (object->n_in_progress == 0)
408                         fscache_start_operations(object);
409
410                 ASSERTCMP(object->n_ops, >, 0);
411                 object->n_ops--;
412                 if (object->n_ops == 0)
413                         fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
414
415                 spin_unlock(&object->lock);
416
417         } while (count++ < 20);
418
419         if (!list_empty(&cache->op_gc_list))
420                 schedule_work(&cache->op_gc);
421
422         _leave("");
423 }
424
425 /*
426  * allow the slow work item processor to get a ref on an operation
427  */
428 static int fscache_op_get_ref(struct slow_work *work)
429 {
430         struct fscache_operation *op =
431                 container_of(work, struct fscache_operation, slow_work);
432
433         atomic_inc(&op->usage);
434         return 0;
435 }
436
437 /*
438  * allow the slow work item processor to discard a ref on an operation
439  */
440 static void fscache_op_put_ref(struct slow_work *work)
441 {
442         struct fscache_operation *op =
443                 container_of(work, struct fscache_operation, slow_work);
444
445         fscache_put_operation(op);
446 }
447
448 /*
449  * execute an operation using the slow thread pool to provide processing context
450  * - the caller holds a ref to this object, so we don't need to hold one
451  */
452 static void fscache_op_execute(struct slow_work *work)
453 {
454         struct fscache_operation *op =
455                 container_of(work, struct fscache_operation, slow_work);
456         unsigned long start;
457
458         _enter("{OBJ%x OP%x,%d}",
459                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
460
461         ASSERT(op->processor != NULL);
462         start = jiffies;
463         op->processor(op);
464         fscache_hist(fscache_ops_histogram, start);
465
466         _leave("");
467 }
468
469 /*
470  * describe an operation for slow-work debugging
471  */
472 #ifdef CONFIG_SLOW_WORK_PROC
473 static void fscache_op_desc(struct slow_work *work, struct seq_file *m)
474 {
475         struct fscache_operation *op =
476                 container_of(work, struct fscache_operation, slow_work);
477
478         seq_printf(m, "FSC: OBJ%x OP%x: %s/%s fl=%lx",
479                    op->object->debug_id, op->debug_id,
480                    op->name, op->state, op->flags);
481 }
482 #endif
483
484 const struct slow_work_ops fscache_op_slow_work_ops = {
485         .owner          = THIS_MODULE,
486         .get_ref        = fscache_op_get_ref,
487         .put_ref        = fscache_op_put_ref,
488         .execute        = fscache_op_execute,
489 #ifdef CONFIG_SLOW_WORK_PROC
490         .desc           = fscache_op_desc,
491 #endif
492 };