]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/fscache/operation.c
Merge commit 'v3.1-rc4' into next
[karo-tx-linux.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         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
41         fscache_stat(&fscache_n_op_enqueue);
42         switch (op->flags & FSCACHE_OP_TYPE) {
43         case FSCACHE_OP_ASYNC:
44                 _debug("queue async");
45                 atomic_inc(&op->usage);
46                 if (!queue_work(fscache_op_wq, &op->work))
47                         fscache_put_operation(op);
48                 break;
49         case FSCACHE_OP_MYTHREAD:
50                 _debug("queue for caller's attention");
51                 break;
52         default:
53                 printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
54                        op->flags);
55                 BUG();
56                 break;
57         }
58 }
59 EXPORT_SYMBOL(fscache_enqueue_operation);
60
61 /*
62  * start an op running
63  */
64 static void fscache_run_op(struct fscache_object *object,
65                            struct fscache_operation *op)
66 {
67         object->n_in_progress++;
68         if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
69                 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
70         if (op->processor)
71                 fscache_enqueue_operation(op);
72         fscache_stat(&fscache_n_op_run);
73 }
74
75 /*
76  * submit an exclusive operation for an object
77  * - other ops are excluded from running simultaneously with this one
78  * - this gets any extra refs it needs on an op
79  */
80 int fscache_submit_exclusive_op(struct fscache_object *object,
81                                 struct fscache_operation *op)
82 {
83         int ret;
84
85         _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
86
87         spin_lock(&object->lock);
88         ASSERTCMP(object->n_ops, >=, object->n_in_progress);
89         ASSERTCMP(object->n_ops, >=, object->n_exclusive);
90         ASSERT(list_empty(&op->pend_link));
91
92         ret = -ENOBUFS;
93         if (fscache_object_is_active(object)) {
94                 op->object = object;
95                 object->n_ops++;
96                 object->n_exclusive++;  /* reads and writes must wait */
97
98                 if (object->n_ops > 1) {
99                         atomic_inc(&op->usage);
100                         list_add_tail(&op->pend_link, &object->pending_ops);
101                         fscache_stat(&fscache_n_op_pend);
102                 } else if (!list_empty(&object->pending_ops)) {
103                         atomic_inc(&op->usage);
104                         list_add_tail(&op->pend_link, &object->pending_ops);
105                         fscache_stat(&fscache_n_op_pend);
106                         fscache_start_operations(object);
107                 } else {
108                         ASSERTCMP(object->n_in_progress, ==, 0);
109                         fscache_run_op(object, op);
110                 }
111
112                 /* need to issue a new write op after this */
113                 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
114                 ret = 0;
115         } else if (object->state == FSCACHE_OBJECT_CREATING) {
116                 op->object = object;
117                 object->n_ops++;
118                 object->n_exclusive++;  /* reads and writes must wait */
119                 atomic_inc(&op->usage);
120                 list_add_tail(&op->pend_link, &object->pending_ops);
121                 fscache_stat(&fscache_n_op_pend);
122                 ret = 0;
123         } else {
124                 /* not allowed to submit ops in any other state */
125                 BUG();
126         }
127
128         spin_unlock(&object->lock);
129         return ret;
130 }
131
132 /*
133  * report an unexpected submission
134  */
135 static void fscache_report_unexpected_submission(struct fscache_object *object,
136                                                  struct fscache_operation *op,
137                                                  unsigned long ostate)
138 {
139         static bool once_only;
140         struct fscache_operation *p;
141         unsigned n;
142
143         if (once_only)
144                 return;
145         once_only = true;
146
147         kdebug("unexpected submission OP%x [OBJ%x %s]",
148                op->debug_id, object->debug_id,
149                fscache_object_states[object->state]);
150         kdebug("objstate=%s [%s]",
151                fscache_object_states[object->state],
152                fscache_object_states[ostate]);
153         kdebug("objflags=%lx", object->flags);
154         kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
155         kdebug("ops=%u inp=%u exc=%u",
156                object->n_ops, object->n_in_progress, object->n_exclusive);
157
158         if (!list_empty(&object->pending_ops)) {
159                 n = 0;
160                 list_for_each_entry(p, &object->pending_ops, pend_link) {
161                         ASSERTCMP(p->object, ==, object);
162                         kdebug("%p %p", op->processor, op->release);
163                         n++;
164                 }
165
166                 kdebug("n=%u", n);
167         }
168
169         dump_stack();
170 }
171
172 /*
173  * submit an operation for an object
174  * - objects may be submitted only in the following states:
175  *   - during object creation (write ops may be submitted)
176  *   - whilst the object is active
177  *   - after an I/O error incurred in one of the two above states (op rejected)
178  * - this gets any extra refs it needs on an op
179  */
180 int fscache_submit_op(struct fscache_object *object,
181                       struct fscache_operation *op)
182 {
183         unsigned long ostate;
184         int ret;
185
186         _enter("{OBJ%x OP%x},{%u}",
187                object->debug_id, op->debug_id, atomic_read(&op->usage));
188
189         ASSERTCMP(atomic_read(&op->usage), >, 0);
190
191         spin_lock(&object->lock);
192         ASSERTCMP(object->n_ops, >=, object->n_in_progress);
193         ASSERTCMP(object->n_ops, >=, object->n_exclusive);
194         ASSERT(list_empty(&op->pend_link));
195
196         ostate = object->state;
197         smp_rmb();
198
199         if (fscache_object_is_active(object)) {
200                 op->object = object;
201                 object->n_ops++;
202
203                 if (object->n_exclusive > 0) {
204                         atomic_inc(&op->usage);
205                         list_add_tail(&op->pend_link, &object->pending_ops);
206                         fscache_stat(&fscache_n_op_pend);
207                 } else if (!list_empty(&object->pending_ops)) {
208                         atomic_inc(&op->usage);
209                         list_add_tail(&op->pend_link, &object->pending_ops);
210                         fscache_stat(&fscache_n_op_pend);
211                         fscache_start_operations(object);
212                 } else {
213                         ASSERTCMP(object->n_exclusive, ==, 0);
214                         fscache_run_op(object, op);
215                 }
216                 ret = 0;
217         } else if (object->state == FSCACHE_OBJECT_CREATING) {
218                 op->object = object;
219                 object->n_ops++;
220                 atomic_inc(&op->usage);
221                 list_add_tail(&op->pend_link, &object->pending_ops);
222                 fscache_stat(&fscache_n_op_pend);
223                 ret = 0;
224         } else if (object->state == FSCACHE_OBJECT_DYING ||
225                    object->state == FSCACHE_OBJECT_LC_DYING ||
226                    object->state == FSCACHE_OBJECT_WITHDRAWING) {
227                 fscache_stat(&fscache_n_op_rejected);
228                 ret = -ENOBUFS;
229         } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
230                 fscache_report_unexpected_submission(object, op, ostate);
231                 ASSERT(!fscache_object_is_active(object));
232                 ret = -ENOBUFS;
233         } else {
234                 ret = -ENOBUFS;
235         }
236
237         spin_unlock(&object->lock);
238         return ret;
239 }
240
241 /*
242  * queue an object for withdrawal on error, aborting all following asynchronous
243  * operations
244  */
245 void fscache_abort_object(struct fscache_object *object)
246 {
247         _enter("{OBJ%x}", object->debug_id);
248
249         fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
250 }
251
252 /*
253  * jump start the operation processing on an object
254  * - caller must hold object->lock
255  */
256 void fscache_start_operations(struct fscache_object *object)
257 {
258         struct fscache_operation *op;
259         bool stop = false;
260
261         while (!list_empty(&object->pending_ops) && !stop) {
262                 op = list_entry(object->pending_ops.next,
263                                 struct fscache_operation, pend_link);
264
265                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
266                         if (object->n_in_progress > 0)
267                                 break;
268                         stop = true;
269                 }
270                 list_del_init(&op->pend_link);
271                 fscache_run_op(object, op);
272
273                 /* the pending queue was holding a ref on the object */
274                 fscache_put_operation(op);
275         }
276
277         ASSERTCMP(object->n_in_progress, <=, object->n_ops);
278
279         _debug("woke %d ops on OBJ%x",
280                object->n_in_progress, object->debug_id);
281 }
282
283 /*
284  * cancel an operation that's pending on an object
285  */
286 int fscache_cancel_op(struct fscache_operation *op)
287 {
288         struct fscache_object *object = op->object;
289         int ret;
290
291         _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
292
293         spin_lock(&object->lock);
294
295         ret = -EBUSY;
296         if (!list_empty(&op->pend_link)) {
297                 fscache_stat(&fscache_n_op_cancelled);
298                 list_del_init(&op->pend_link);
299                 object->n_ops--;
300                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
301                         object->n_exclusive--;
302                 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
303                         wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
304                 fscache_put_operation(op);
305                 ret = 0;
306         }
307
308         spin_unlock(&object->lock);
309         _leave(" = %d", ret);
310         return ret;
311 }
312
313 /*
314  * release an operation
315  * - queues pending ops if this is the last in-progress op
316  */
317 void fscache_put_operation(struct fscache_operation *op)
318 {
319         struct fscache_object *object;
320         struct fscache_cache *cache;
321
322         _enter("{OBJ%x OP%x,%d}",
323                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
324
325         ASSERTCMP(atomic_read(&op->usage), >, 0);
326
327         if (!atomic_dec_and_test(&op->usage))
328                 return;
329
330         _debug("PUT OP");
331         if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
332                 BUG();
333
334         fscache_stat(&fscache_n_op_release);
335
336         if (op->release) {
337                 op->release(op);
338                 op->release = NULL;
339         }
340
341         object = op->object;
342
343         if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
344                 atomic_dec(&object->n_reads);
345
346         /* now... we may get called with the object spinlock held, so we
347          * complete the cleanup here only if we can immediately acquire the
348          * lock, and defer it otherwise */
349         if (!spin_trylock(&object->lock)) {
350                 _debug("defer put");
351                 fscache_stat(&fscache_n_op_deferred_release);
352
353                 cache = object->cache;
354                 spin_lock(&cache->op_gc_list_lock);
355                 list_add_tail(&op->pend_link, &cache->op_gc_list);
356                 spin_unlock(&cache->op_gc_list_lock);
357                 schedule_work(&cache->op_gc);
358                 _leave(" [defer]");
359                 return;
360         }
361
362         if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
363                 ASSERTCMP(object->n_exclusive, >, 0);
364                 object->n_exclusive--;
365         }
366
367         ASSERTCMP(object->n_in_progress, >, 0);
368         object->n_in_progress--;
369         if (object->n_in_progress == 0)
370                 fscache_start_operations(object);
371
372         ASSERTCMP(object->n_ops, >, 0);
373         object->n_ops--;
374         if (object->n_ops == 0)
375                 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
376
377         spin_unlock(&object->lock);
378
379         kfree(op);
380         _leave(" [done]");
381 }
382 EXPORT_SYMBOL(fscache_put_operation);
383
384 /*
385  * garbage collect operations that have had their release deferred
386  */
387 void fscache_operation_gc(struct work_struct *work)
388 {
389         struct fscache_operation *op;
390         struct fscache_object *object;
391         struct fscache_cache *cache =
392                 container_of(work, struct fscache_cache, op_gc);
393         int count = 0;
394
395         _enter("");
396
397         do {
398                 spin_lock(&cache->op_gc_list_lock);
399                 if (list_empty(&cache->op_gc_list)) {
400                         spin_unlock(&cache->op_gc_list_lock);
401                         break;
402                 }
403
404                 op = list_entry(cache->op_gc_list.next,
405                                 struct fscache_operation, pend_link);
406                 list_del(&op->pend_link);
407                 spin_unlock(&cache->op_gc_list_lock);
408
409                 object = op->object;
410
411                 _debug("GC DEFERRED REL OBJ%x OP%x",
412                        object->debug_id, op->debug_id);
413                 fscache_stat(&fscache_n_op_gc);
414
415                 ASSERTCMP(atomic_read(&op->usage), ==, 0);
416
417                 spin_lock(&object->lock);
418                 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
419                         ASSERTCMP(object->n_exclusive, >, 0);
420                         object->n_exclusive--;
421                 }
422
423                 ASSERTCMP(object->n_in_progress, >, 0);
424                 object->n_in_progress--;
425                 if (object->n_in_progress == 0)
426                         fscache_start_operations(object);
427
428                 ASSERTCMP(object->n_ops, >, 0);
429                 object->n_ops--;
430                 if (object->n_ops == 0)
431                         fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
432
433                 spin_unlock(&object->lock);
434
435         } while (count++ < 20);
436
437         if (!list_empty(&cache->op_gc_list))
438                 schedule_work(&cache->op_gc);
439
440         _leave("");
441 }
442
443 /*
444  * execute an operation using fs_op_wq to provide processing context -
445  * the caller holds a ref to this object, so we don't need to hold one
446  */
447 void fscache_op_work_func(struct work_struct *work)
448 {
449         struct fscache_operation *op =
450                 container_of(work, struct fscache_operation, work);
451         unsigned long start;
452
453         _enter("{OBJ%x OP%x,%d}",
454                op->object->debug_id, op->debug_id, atomic_read(&op->usage));
455
456         ASSERT(op->processor != NULL);
457         start = jiffies;
458         op->processor(op);
459         fscache_hist(fscache_ops_histogram, start);
460         fscache_put_operation(op);
461
462         _leave("");
463 }