Merge branch 'master' into for-next
[pandora-kernel.git] / fs / fscache / page.c
1 /* Cache page management and data I/O routines
2  *
3  * Copyright (C) 2004-2008 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
12 #define FSCACHE_DEBUG_LEVEL PAGE
13 #include <linux/module.h>
14 #include <linux/fscache-cache.h>
15 #include <linux/buffer_head.h>
16 #include <linux/pagevec.h>
17 #include <linux/slab.h>
18 #include "internal.h"
19
20 /*
21  * check to see if a page is being written to the cache
22  */
23 bool __fscache_check_page_write(struct fscache_cookie *cookie, struct page *page)
24 {
25         void *val;
26
27         rcu_read_lock();
28         val = radix_tree_lookup(&cookie->stores, page->index);
29         rcu_read_unlock();
30
31         return val != NULL;
32 }
33 EXPORT_SYMBOL(__fscache_check_page_write);
34
35 /*
36  * wait for a page to finish being written to the cache
37  */
38 void __fscache_wait_on_page_write(struct fscache_cookie *cookie, struct page *page)
39 {
40         wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
41
42         wait_event(*wq, !__fscache_check_page_write(cookie, page));
43 }
44 EXPORT_SYMBOL(__fscache_wait_on_page_write);
45
46 /*
47  * decide whether a page can be released, possibly by cancelling a store to it
48  * - we're allowed to sleep if __GFP_WAIT is flagged
49  */
50 bool __fscache_maybe_release_page(struct fscache_cookie *cookie,
51                                   struct page *page,
52                                   gfp_t gfp)
53 {
54         struct page *xpage;
55         void *val;
56
57         _enter("%p,%p,%x", cookie, page, gfp);
58
59 try_again:
60         rcu_read_lock();
61         val = radix_tree_lookup(&cookie->stores, page->index);
62         if (!val) {
63                 rcu_read_unlock();
64                 fscache_stat(&fscache_n_store_vmscan_not_storing);
65                 __fscache_uncache_page(cookie, page);
66                 return true;
67         }
68
69         /* see if the page is actually undergoing storage - if so we can't get
70          * rid of it till the cache has finished with it */
71         if (radix_tree_tag_get(&cookie->stores, page->index,
72                                FSCACHE_COOKIE_STORING_TAG)) {
73                 rcu_read_unlock();
74                 goto page_busy;
75         }
76
77         /* the page is pending storage, so we attempt to cancel the store and
78          * discard the store request so that the page can be reclaimed */
79         spin_lock(&cookie->stores_lock);
80         rcu_read_unlock();
81
82         if (radix_tree_tag_get(&cookie->stores, page->index,
83                                FSCACHE_COOKIE_STORING_TAG)) {
84                 /* the page started to undergo storage whilst we were looking,
85                  * so now we can only wait or return */
86                 spin_unlock(&cookie->stores_lock);
87                 goto page_busy;
88         }
89
90         xpage = radix_tree_delete(&cookie->stores, page->index);
91         spin_unlock(&cookie->stores_lock);
92
93         if (xpage) {
94                 fscache_stat(&fscache_n_store_vmscan_cancelled);
95                 fscache_stat(&fscache_n_store_radix_deletes);
96                 ASSERTCMP(xpage, ==, page);
97         } else {
98                 fscache_stat(&fscache_n_store_vmscan_gone);
99         }
100
101         wake_up_bit(&cookie->flags, 0);
102         if (xpage)
103                 page_cache_release(xpage);
104         __fscache_uncache_page(cookie, page);
105         return true;
106
107 page_busy:
108         /* We will wait here if we're allowed to, but that could deadlock the
109          * allocator as the work threads writing to the cache may all end up
110          * sleeping on memory allocation, so we may need to impose a timeout
111          * too. */
112         if (!(gfp & __GFP_WAIT) || !(gfp & __GFP_FS)) {
113                 fscache_stat(&fscache_n_store_vmscan_busy);
114                 return false;
115         }
116
117         fscache_stat(&fscache_n_store_vmscan_wait);
118         __fscache_wait_on_page_write(cookie, page);
119         gfp &= ~__GFP_WAIT;
120         goto try_again;
121 }
122 EXPORT_SYMBOL(__fscache_maybe_release_page);
123
124 /*
125  * note that a page has finished being written to the cache
126  */
127 static void fscache_end_page_write(struct fscache_object *object,
128                                    struct page *page)
129 {
130         struct fscache_cookie *cookie;
131         struct page *xpage = NULL;
132
133         spin_lock(&object->lock);
134         cookie = object->cookie;
135         if (cookie) {
136                 /* delete the page from the tree if it is now no longer
137                  * pending */
138                 spin_lock(&cookie->stores_lock);
139                 radix_tree_tag_clear(&cookie->stores, page->index,
140                                      FSCACHE_COOKIE_STORING_TAG);
141                 if (!radix_tree_tag_get(&cookie->stores, page->index,
142                                         FSCACHE_COOKIE_PENDING_TAG)) {
143                         fscache_stat(&fscache_n_store_radix_deletes);
144                         xpage = radix_tree_delete(&cookie->stores, page->index);
145                 }
146                 spin_unlock(&cookie->stores_lock);
147                 wake_up_bit(&cookie->flags, 0);
148         }
149         spin_unlock(&object->lock);
150         if (xpage)
151                 page_cache_release(xpage);
152 }
153
154 /*
155  * actually apply the changed attributes to a cache object
156  */
157 static void fscache_attr_changed_op(struct fscache_operation *op)
158 {
159         struct fscache_object *object = op->object;
160         int ret;
161
162         _enter("{OBJ%x OP%x}", object->debug_id, op->debug_id);
163
164         fscache_stat(&fscache_n_attr_changed_calls);
165
166         if (fscache_object_is_active(object) &&
167             fscache_use_cookie(object)) {
168                 fscache_stat(&fscache_n_cop_attr_changed);
169                 ret = object->cache->ops->attr_changed(object);
170                 fscache_stat_d(&fscache_n_cop_attr_changed);
171                 fscache_unuse_cookie(object);
172                 if (ret < 0)
173                         fscache_abort_object(object);
174         }
175
176         fscache_op_complete(op, true);
177         _leave("");
178 }
179
180 /*
181  * notification that the attributes on an object have changed
182  */
183 int __fscache_attr_changed(struct fscache_cookie *cookie)
184 {
185         struct fscache_operation *op;
186         struct fscache_object *object;
187
188         _enter("%p", cookie);
189
190         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
191
192         fscache_stat(&fscache_n_attr_changed);
193
194         op = kzalloc(sizeof(*op), GFP_KERNEL);
195         if (!op) {
196                 fscache_stat(&fscache_n_attr_changed_nomem);
197                 _leave(" = -ENOMEM");
198                 return -ENOMEM;
199         }
200
201         fscache_operation_init(op, fscache_attr_changed_op, NULL);
202         op->flags = FSCACHE_OP_ASYNC | (1 << FSCACHE_OP_EXCLUSIVE);
203
204         spin_lock(&cookie->lock);
205
206         if (hlist_empty(&cookie->backing_objects))
207                 goto nobufs;
208         object = hlist_entry(cookie->backing_objects.first,
209                              struct fscache_object, cookie_link);
210
211         if (fscache_submit_exclusive_op(object, op) < 0)
212                 goto nobufs;
213         spin_unlock(&cookie->lock);
214         fscache_stat(&fscache_n_attr_changed_ok);
215         fscache_put_operation(op);
216         _leave(" = 0");
217         return 0;
218
219 nobufs:
220         spin_unlock(&cookie->lock);
221         kfree(op);
222         fscache_stat(&fscache_n_attr_changed_nobufs);
223         _leave(" = %d", -ENOBUFS);
224         return -ENOBUFS;
225 }
226 EXPORT_SYMBOL(__fscache_attr_changed);
227
228 /*
229  * release a retrieval op reference
230  */
231 static void fscache_release_retrieval_op(struct fscache_operation *_op)
232 {
233         struct fscache_retrieval *op =
234                 container_of(_op, struct fscache_retrieval, op);
235
236         _enter("{OP%x}", op->op.debug_id);
237
238         ASSERTCMP(atomic_read(&op->n_pages), ==, 0);
239
240         fscache_hist(fscache_retrieval_histogram, op->start_time);
241         if (op->context)
242                 fscache_put_context(op->op.object->cookie, op->context);
243
244         _leave("");
245 }
246
247 /*
248  * allocate a retrieval op
249  */
250 static struct fscache_retrieval *fscache_alloc_retrieval(
251         struct fscache_cookie *cookie,
252         struct address_space *mapping,
253         fscache_rw_complete_t end_io_func,
254         void *context)
255 {
256         struct fscache_retrieval *op;
257
258         /* allocate a retrieval operation and attempt to submit it */
259         op = kzalloc(sizeof(*op), GFP_NOIO);
260         if (!op) {
261                 fscache_stat(&fscache_n_retrievals_nomem);
262                 return NULL;
263         }
264
265         fscache_operation_init(&op->op, NULL, fscache_release_retrieval_op);
266         atomic_inc(&cookie->n_active);
267         op->op.flags    = FSCACHE_OP_MYTHREAD |
268                 (1UL << FSCACHE_OP_WAITING) |
269                 (1UL << FSCACHE_OP_UNUSE_COOKIE);
270         op->mapping     = mapping;
271         op->end_io_func = end_io_func;
272         op->context     = context;
273         op->start_time  = jiffies;
274         INIT_LIST_HEAD(&op->to_do);
275         return op;
276 }
277
278 /*
279  * wait for a deferred lookup to complete
280  */
281 static int fscache_wait_for_deferred_lookup(struct fscache_cookie *cookie)
282 {
283         unsigned long jif;
284
285         _enter("");
286
287         if (!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags)) {
288                 _leave(" = 0 [imm]");
289                 return 0;
290         }
291
292         fscache_stat(&fscache_n_retrievals_wait);
293
294         jif = jiffies;
295         if (wait_on_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP,
296                         fscache_wait_bit_interruptible,
297                         TASK_INTERRUPTIBLE) != 0) {
298                 fscache_stat(&fscache_n_retrievals_intr);
299                 _leave(" = -ERESTARTSYS");
300                 return -ERESTARTSYS;
301         }
302
303         ASSERT(!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags));
304
305         smp_rmb();
306         fscache_hist(fscache_retrieval_delay_histogram, jif);
307         _leave(" = 0 [dly]");
308         return 0;
309 }
310
311 /*
312  * Handle cancellation of a pending retrieval op
313  */
314 static void fscache_do_cancel_retrieval(struct fscache_operation *_op)
315 {
316         struct fscache_retrieval *op =
317                 container_of(_op, struct fscache_retrieval, op);
318
319         atomic_set(&op->n_pages, 0);
320 }
321
322 /*
323  * wait for an object to become active (or dead)
324  */
325 static int fscache_wait_for_retrieval_activation(struct fscache_object *object,
326                                                  struct fscache_retrieval *op,
327                                                  atomic_t *stat_op_waits,
328                                                  atomic_t *stat_object_dead)
329 {
330         int ret;
331
332         if (!test_bit(FSCACHE_OP_WAITING, &op->op.flags))
333                 goto check_if_dead;
334
335         _debug(">>> WT");
336         fscache_stat(stat_op_waits);
337         if (wait_on_bit(&op->op.flags, FSCACHE_OP_WAITING,
338                         fscache_wait_bit_interruptible,
339                         TASK_INTERRUPTIBLE) != 0) {
340                 ret = fscache_cancel_op(&op->op, fscache_do_cancel_retrieval);
341                 if (ret == 0)
342                         return -ERESTARTSYS;
343
344                 /* it's been removed from the pending queue by another party,
345                  * so we should get to run shortly */
346                 wait_on_bit(&op->op.flags, FSCACHE_OP_WAITING,
347                             fscache_wait_bit, TASK_UNINTERRUPTIBLE);
348         }
349         _debug("<<< GO");
350
351 check_if_dead:
352         if (op->op.state == FSCACHE_OP_ST_CANCELLED) {
353                 fscache_stat(stat_object_dead);
354                 _leave(" = -ENOBUFS [cancelled]");
355                 return -ENOBUFS;
356         }
357         if (unlikely(fscache_object_is_dead(object))) {
358                 pr_err("%s() = -ENOBUFS [obj dead %d]\n", __func__, op->op.state);
359                 fscache_cancel_op(&op->op, fscache_do_cancel_retrieval);
360                 fscache_stat(stat_object_dead);
361                 return -ENOBUFS;
362         }
363         return 0;
364 }
365
366 /*
367  * read a page from the cache or allocate a block in which to store it
368  * - we return:
369  *   -ENOMEM    - out of memory, nothing done
370  *   -ERESTARTSYS - interrupted
371  *   -ENOBUFS   - no backing object available in which to cache the block
372  *   -ENODATA   - no data available in the backing object for this block
373  *   0          - dispatched a read - it'll call end_io_func() when finished
374  */
375 int __fscache_read_or_alloc_page(struct fscache_cookie *cookie,
376                                  struct page *page,
377                                  fscache_rw_complete_t end_io_func,
378                                  void *context,
379                                  gfp_t gfp)
380 {
381         struct fscache_retrieval *op;
382         struct fscache_object *object;
383         int ret;
384
385         _enter("%p,%p,,,", cookie, page);
386
387         fscache_stat(&fscache_n_retrievals);
388
389         if (hlist_empty(&cookie->backing_objects))
390                 goto nobufs;
391
392         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
393                 _leave(" = -ENOBUFS [invalidating]");
394                 return -ENOBUFS;
395         }
396
397         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
398         ASSERTCMP(page, !=, NULL);
399
400         if (fscache_wait_for_deferred_lookup(cookie) < 0)
401                 return -ERESTARTSYS;
402
403         op = fscache_alloc_retrieval(cookie, page->mapping,
404                                      end_io_func,context);
405         if (!op) {
406                 _leave(" = -ENOMEM");
407                 return -ENOMEM;
408         }
409         atomic_set(&op->n_pages, 1);
410
411         spin_lock(&cookie->lock);
412
413         if (hlist_empty(&cookie->backing_objects))
414                 goto nobufs_unlock;
415         object = hlist_entry(cookie->backing_objects.first,
416                              struct fscache_object, cookie_link);
417
418         ASSERT(test_bit(FSCACHE_OBJECT_IS_LOOKED_UP, &object->flags));
419
420         atomic_inc(&object->n_reads);
421         __set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
422
423         if (fscache_submit_op(object, &op->op) < 0)
424                 goto nobufs_unlock_dec;
425         spin_unlock(&cookie->lock);
426
427         fscache_stat(&fscache_n_retrieval_ops);
428
429         /* pin the netfs read context in case we need to do the actual netfs
430          * read because we've encountered a cache read failure */
431         fscache_get_context(object->cookie, op->context);
432
433         /* we wait for the operation to become active, and then process it
434          * *here*, in this thread, and not in the thread pool */
435         ret = fscache_wait_for_retrieval_activation(
436                 object, op,
437                 __fscache_stat(&fscache_n_retrieval_op_waits),
438                 __fscache_stat(&fscache_n_retrievals_object_dead));
439         if (ret < 0)
440                 goto error;
441
442         /* ask the cache to honour the operation */
443         if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
444                 fscache_stat(&fscache_n_cop_allocate_page);
445                 ret = object->cache->ops->allocate_page(op, page, gfp);
446                 fscache_stat_d(&fscache_n_cop_allocate_page);
447                 if (ret == 0)
448                         ret = -ENODATA;
449         } else {
450                 fscache_stat(&fscache_n_cop_read_or_alloc_page);
451                 ret = object->cache->ops->read_or_alloc_page(op, page, gfp);
452                 fscache_stat_d(&fscache_n_cop_read_or_alloc_page);
453         }
454
455 error:
456         if (ret == -ENOMEM)
457                 fscache_stat(&fscache_n_retrievals_nomem);
458         else if (ret == -ERESTARTSYS)
459                 fscache_stat(&fscache_n_retrievals_intr);
460         else if (ret == -ENODATA)
461                 fscache_stat(&fscache_n_retrievals_nodata);
462         else if (ret < 0)
463                 fscache_stat(&fscache_n_retrievals_nobufs);
464         else
465                 fscache_stat(&fscache_n_retrievals_ok);
466
467         fscache_put_retrieval(op);
468         _leave(" = %d", ret);
469         return ret;
470
471 nobufs_unlock_dec:
472         atomic_dec(&object->n_reads);
473 nobufs_unlock:
474         spin_unlock(&cookie->lock);
475         atomic_dec(&cookie->n_active);
476         kfree(op);
477 nobufs:
478         fscache_stat(&fscache_n_retrievals_nobufs);
479         _leave(" = -ENOBUFS");
480         return -ENOBUFS;
481 }
482 EXPORT_SYMBOL(__fscache_read_or_alloc_page);
483
484 /*
485  * read a list of page from the cache or allocate a block in which to store
486  * them
487  * - we return:
488  *   -ENOMEM    - out of memory, some pages may be being read
489  *   -ERESTARTSYS - interrupted, some pages may be being read
490  *   -ENOBUFS   - no backing object or space available in which to cache any
491  *                pages not being read
492  *   -ENODATA   - no data available in the backing object for some or all of
493  *                the pages
494  *   0          - dispatched a read on all pages
495  *
496  * end_io_func() will be called for each page read from the cache as it is
497  * finishes being read
498  *
499  * any pages for which a read is dispatched will be removed from pages and
500  * nr_pages
501  */
502 int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
503                                   struct address_space *mapping,
504                                   struct list_head *pages,
505                                   unsigned *nr_pages,
506                                   fscache_rw_complete_t end_io_func,
507                                   void *context,
508                                   gfp_t gfp)
509 {
510         struct fscache_retrieval *op;
511         struct fscache_object *object;
512         int ret;
513
514         _enter("%p,,%d,,,", cookie, *nr_pages);
515
516         fscache_stat(&fscache_n_retrievals);
517
518         if (hlist_empty(&cookie->backing_objects))
519                 goto nobufs;
520
521         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
522                 _leave(" = -ENOBUFS [invalidating]");
523                 return -ENOBUFS;
524         }
525
526         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
527         ASSERTCMP(*nr_pages, >, 0);
528         ASSERT(!list_empty(pages));
529
530         if (fscache_wait_for_deferred_lookup(cookie) < 0)
531                 return -ERESTARTSYS;
532
533         op = fscache_alloc_retrieval(cookie, mapping, end_io_func, context);
534         if (!op)
535                 return -ENOMEM;
536         atomic_set(&op->n_pages, *nr_pages);
537
538         spin_lock(&cookie->lock);
539
540         if (hlist_empty(&cookie->backing_objects))
541                 goto nobufs_unlock;
542         object = hlist_entry(cookie->backing_objects.first,
543                              struct fscache_object, cookie_link);
544
545         atomic_inc(&object->n_reads);
546         __set_bit(FSCACHE_OP_DEC_READ_CNT, &op->op.flags);
547
548         if (fscache_submit_op(object, &op->op) < 0)
549                 goto nobufs_unlock_dec;
550         spin_unlock(&cookie->lock);
551
552         fscache_stat(&fscache_n_retrieval_ops);
553
554         /* pin the netfs read context in case we need to do the actual netfs
555          * read because we've encountered a cache read failure */
556         fscache_get_context(object->cookie, op->context);
557
558         /* we wait for the operation to become active, and then process it
559          * *here*, in this thread, and not in the thread pool */
560         ret = fscache_wait_for_retrieval_activation(
561                 object, op,
562                 __fscache_stat(&fscache_n_retrieval_op_waits),
563                 __fscache_stat(&fscache_n_retrievals_object_dead));
564         if (ret < 0)
565                 goto error;
566
567         /* ask the cache to honour the operation */
568         if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
569                 fscache_stat(&fscache_n_cop_allocate_pages);
570                 ret = object->cache->ops->allocate_pages(
571                         op, pages, nr_pages, gfp);
572                 fscache_stat_d(&fscache_n_cop_allocate_pages);
573         } else {
574                 fscache_stat(&fscache_n_cop_read_or_alloc_pages);
575                 ret = object->cache->ops->read_or_alloc_pages(
576                         op, pages, nr_pages, gfp);
577                 fscache_stat_d(&fscache_n_cop_read_or_alloc_pages);
578         }
579
580 error:
581         if (ret == -ENOMEM)
582                 fscache_stat(&fscache_n_retrievals_nomem);
583         else if (ret == -ERESTARTSYS)
584                 fscache_stat(&fscache_n_retrievals_intr);
585         else if (ret == -ENODATA)
586                 fscache_stat(&fscache_n_retrievals_nodata);
587         else if (ret < 0)
588                 fscache_stat(&fscache_n_retrievals_nobufs);
589         else
590                 fscache_stat(&fscache_n_retrievals_ok);
591
592         fscache_put_retrieval(op);
593         _leave(" = %d", ret);
594         return ret;
595
596 nobufs_unlock_dec:
597         atomic_dec(&object->n_reads);
598 nobufs_unlock:
599         spin_unlock(&cookie->lock);
600         atomic_dec(&cookie->n_active);
601         kfree(op);
602 nobufs:
603         fscache_stat(&fscache_n_retrievals_nobufs);
604         _leave(" = -ENOBUFS");
605         return -ENOBUFS;
606 }
607 EXPORT_SYMBOL(__fscache_read_or_alloc_pages);
608
609 /*
610  * allocate a block in the cache on which to store a page
611  * - we return:
612  *   -ENOMEM    - out of memory, nothing done
613  *   -ERESTARTSYS - interrupted
614  *   -ENOBUFS   - no backing object available in which to cache the block
615  *   0          - block allocated
616  */
617 int __fscache_alloc_page(struct fscache_cookie *cookie,
618                          struct page *page,
619                          gfp_t gfp)
620 {
621         struct fscache_retrieval *op;
622         struct fscache_object *object;
623         int ret;
624
625         _enter("%p,%p,,,", cookie, page);
626
627         fscache_stat(&fscache_n_allocs);
628
629         if (hlist_empty(&cookie->backing_objects))
630                 goto nobufs;
631
632         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
633         ASSERTCMP(page, !=, NULL);
634
635         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
636                 _leave(" = -ENOBUFS [invalidating]");
637                 return -ENOBUFS;
638         }
639
640         if (fscache_wait_for_deferred_lookup(cookie) < 0)
641                 return -ERESTARTSYS;
642
643         op = fscache_alloc_retrieval(cookie, page->mapping, NULL, NULL);
644         if (!op)
645                 return -ENOMEM;
646         atomic_set(&op->n_pages, 1);
647
648         spin_lock(&cookie->lock);
649
650         if (hlist_empty(&cookie->backing_objects))
651                 goto nobufs_unlock;
652         object = hlist_entry(cookie->backing_objects.first,
653                              struct fscache_object, cookie_link);
654
655         if (fscache_submit_op(object, &op->op) < 0)
656                 goto nobufs_unlock;
657         spin_unlock(&cookie->lock);
658
659         fscache_stat(&fscache_n_alloc_ops);
660
661         ret = fscache_wait_for_retrieval_activation(
662                 object, op,
663                 __fscache_stat(&fscache_n_alloc_op_waits),
664                 __fscache_stat(&fscache_n_allocs_object_dead));
665         if (ret < 0)
666                 goto error;
667
668         /* ask the cache to honour the operation */
669         fscache_stat(&fscache_n_cop_allocate_page);
670         ret = object->cache->ops->allocate_page(op, page, gfp);
671         fscache_stat_d(&fscache_n_cop_allocate_page);
672
673 error:
674         if (ret == -ERESTARTSYS)
675                 fscache_stat(&fscache_n_allocs_intr);
676         else if (ret < 0)
677                 fscache_stat(&fscache_n_allocs_nobufs);
678         else
679                 fscache_stat(&fscache_n_allocs_ok);
680
681         fscache_put_retrieval(op);
682         _leave(" = %d", ret);
683         return ret;
684
685 nobufs_unlock:
686         spin_unlock(&cookie->lock);
687         atomic_dec(&cookie->n_active);
688         kfree(op);
689 nobufs:
690         fscache_stat(&fscache_n_allocs_nobufs);
691         _leave(" = -ENOBUFS");
692         return -ENOBUFS;
693 }
694 EXPORT_SYMBOL(__fscache_alloc_page);
695
696 /*
697  * release a write op reference
698  */
699 static void fscache_release_write_op(struct fscache_operation *_op)
700 {
701         _enter("{OP%x}", _op->debug_id);
702 }
703
704 /*
705  * perform the background storage of a page into the cache
706  */
707 static void fscache_write_op(struct fscache_operation *_op)
708 {
709         struct fscache_storage *op =
710                 container_of(_op, struct fscache_storage, op);
711         struct fscache_object *object = op->op.object;
712         struct fscache_cookie *cookie;
713         struct page *page;
714         unsigned n;
715         void *results[1];
716         int ret;
717
718         _enter("{OP%x,%d}", op->op.debug_id, atomic_read(&op->op.usage));
719
720         spin_lock(&object->lock);
721         cookie = object->cookie;
722
723         if (!fscache_object_is_active(object)) {
724                 /* If we get here, then the on-disk cache object likely longer
725                  * exists, so we should just cancel this write operation.
726                  */
727                 spin_unlock(&object->lock);
728                 fscache_op_complete(&op->op, false);
729                 _leave(" [inactive]");
730                 return;
731         }
732
733         if (!cookie) {
734                 /* If we get here, then the cookie belonging to the object was
735                  * detached, probably by the cookie being withdrawn due to
736                  * memory pressure, which means that the pages we might write
737                  * to the cache from no longer exist - therefore, we can just
738                  * cancel this write operation.
739                  */
740                 spin_unlock(&object->lock);
741                 fscache_op_complete(&op->op, false);
742                 _leave(" [cancel] op{f=%lx s=%u} obj{s=%s f=%lx}",
743                        _op->flags, _op->state, object->state->short_name,
744                        object->flags);
745                 return;
746         }
747
748         spin_lock(&cookie->stores_lock);
749
750         fscache_stat(&fscache_n_store_calls);
751
752         /* find a page to store */
753         page = NULL;
754         n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0, 1,
755                                        FSCACHE_COOKIE_PENDING_TAG);
756         if (n != 1)
757                 goto superseded;
758         page = results[0];
759         _debug("gang %d [%lx]", n, page->index);
760         if (page->index > op->store_limit) {
761                 fscache_stat(&fscache_n_store_pages_over_limit);
762                 goto superseded;
763         }
764
765         radix_tree_tag_set(&cookie->stores, page->index,
766                            FSCACHE_COOKIE_STORING_TAG);
767         radix_tree_tag_clear(&cookie->stores, page->index,
768                              FSCACHE_COOKIE_PENDING_TAG);
769
770         spin_unlock(&cookie->stores_lock);
771         spin_unlock(&object->lock);
772
773         fscache_stat(&fscache_n_store_pages);
774         fscache_stat(&fscache_n_cop_write_page);
775         ret = object->cache->ops->write_page(op, page);
776         fscache_stat_d(&fscache_n_cop_write_page);
777         fscache_end_page_write(object, page);
778         if (ret < 0) {
779                 fscache_abort_object(object);
780                 fscache_op_complete(&op->op, true);
781         } else {
782                 fscache_enqueue_operation(&op->op);
783         }
784
785         _leave("");
786         return;
787
788 superseded:
789         /* this writer is going away and there aren't any more things to
790          * write */
791         _debug("cease");
792         spin_unlock(&cookie->stores_lock);
793         clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
794         spin_unlock(&object->lock);
795         fscache_op_complete(&op->op, true);
796         _leave("");
797 }
798
799 /*
800  * Clear the pages pending writing for invalidation
801  */
802 void fscache_invalidate_writes(struct fscache_cookie *cookie)
803 {
804         struct page *page;
805         void *results[16];
806         int n, i;
807
808         _enter("");
809
810         for (;;) {
811                 spin_lock(&cookie->stores_lock);
812                 n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0,
813                                                ARRAY_SIZE(results),
814                                                FSCACHE_COOKIE_PENDING_TAG);
815                 if (n == 0) {
816                         spin_unlock(&cookie->stores_lock);
817                         break;
818                 }
819
820                 for (i = n - 1; i >= 0; i--) {
821                         page = results[i];
822                         radix_tree_delete(&cookie->stores, page->index);
823                 }
824
825                 spin_unlock(&cookie->stores_lock);
826
827                 for (i = n - 1; i >= 0; i--)
828                         page_cache_release(results[i]);
829         }
830
831         _leave("");
832 }
833
834 /*
835  * request a page be stored in the cache
836  * - returns:
837  *   -ENOMEM    - out of memory, nothing done
838  *   -ENOBUFS   - no backing object available in which to cache the page
839  *   0          - dispatched a write - it'll call end_io_func() when finished
840  *
841  * if the cookie still has a backing object at this point, that object can be
842  * in one of a few states with respect to storage processing:
843  *
844  *  (1) negative lookup, object not yet created (FSCACHE_COOKIE_CREATING is
845  *      set)
846  *
847  *      (a) no writes yet
848  *
849  *      (b) writes deferred till post-creation (mark page for writing and
850  *          return immediately)
851  *
852  *  (2) negative lookup, object created, initial fill being made from netfs
853  *
854  *      (a) fill point not yet reached this page (mark page for writing and
855  *          return)
856  *
857  *      (b) fill point passed this page (queue op to store this page)
858  *
859  *  (3) object extant (queue op to store this page)
860  *
861  * any other state is invalid
862  */
863 int __fscache_write_page(struct fscache_cookie *cookie,
864                          struct page *page,
865                          gfp_t gfp)
866 {
867         struct fscache_storage *op;
868         struct fscache_object *object;
869         int ret;
870
871         _enter("%p,%x,", cookie, (u32) page->flags);
872
873         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
874         ASSERT(PageFsCache(page));
875
876         fscache_stat(&fscache_n_stores);
877
878         if (test_bit(FSCACHE_COOKIE_INVALIDATING, &cookie->flags)) {
879                 _leave(" = -ENOBUFS [invalidating]");
880                 return -ENOBUFS;
881         }
882
883         op = kzalloc(sizeof(*op), GFP_NOIO | __GFP_NOMEMALLOC | __GFP_NORETRY);
884         if (!op)
885                 goto nomem;
886
887         fscache_operation_init(&op->op, fscache_write_op,
888                                fscache_release_write_op);
889         op->op.flags = FSCACHE_OP_ASYNC |
890                 (1 << FSCACHE_OP_WAITING) |
891                 (1 << FSCACHE_OP_UNUSE_COOKIE);
892
893         ret = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
894         if (ret < 0)
895                 goto nomem_free;
896
897         ret = -ENOBUFS;
898         spin_lock(&cookie->lock);
899
900         if (hlist_empty(&cookie->backing_objects))
901                 goto nobufs;
902         object = hlist_entry(cookie->backing_objects.first,
903                              struct fscache_object, cookie_link);
904         if (test_bit(FSCACHE_IOERROR, &object->cache->flags))
905                 goto nobufs;
906
907         /* add the page to the pending-storage radix tree on the backing
908          * object */
909         spin_lock(&object->lock);
910         spin_lock(&cookie->stores_lock);
911
912         _debug("store limit %llx", (unsigned long long) object->store_limit);
913
914         ret = radix_tree_insert(&cookie->stores, page->index, page);
915         if (ret < 0) {
916                 if (ret == -EEXIST)
917                         goto already_queued;
918                 _debug("insert failed %d", ret);
919                 goto nobufs_unlock_obj;
920         }
921
922         radix_tree_tag_set(&cookie->stores, page->index,
923                            FSCACHE_COOKIE_PENDING_TAG);
924         page_cache_get(page);
925
926         /* we only want one writer at a time, but we do need to queue new
927          * writers after exclusive ops */
928         if (test_and_set_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags))
929                 goto already_pending;
930
931         spin_unlock(&cookie->stores_lock);
932         spin_unlock(&object->lock);
933
934         op->op.debug_id = atomic_inc_return(&fscache_op_debug_id);
935         op->store_limit = object->store_limit;
936
937         atomic_inc(&cookie->n_active);
938         if (fscache_submit_op(object, &op->op) < 0)
939                 goto submit_failed;
940
941         spin_unlock(&cookie->lock);
942         radix_tree_preload_end();
943         fscache_stat(&fscache_n_store_ops);
944         fscache_stat(&fscache_n_stores_ok);
945
946         /* the work queue now carries its own ref on the object */
947         fscache_put_operation(&op->op);
948         _leave(" = 0");
949         return 0;
950
951 already_queued:
952         fscache_stat(&fscache_n_stores_again);
953 already_pending:
954         spin_unlock(&cookie->stores_lock);
955         spin_unlock(&object->lock);
956         spin_unlock(&cookie->lock);
957         radix_tree_preload_end();
958         kfree(op);
959         fscache_stat(&fscache_n_stores_ok);
960         _leave(" = 0");
961         return 0;
962
963 submit_failed:
964         atomic_dec(&cookie->n_active);
965         spin_lock(&cookie->stores_lock);
966         radix_tree_delete(&cookie->stores, page->index);
967         spin_unlock(&cookie->stores_lock);
968         page_cache_release(page);
969         ret = -ENOBUFS;
970         goto nobufs;
971
972 nobufs_unlock_obj:
973         spin_unlock(&cookie->stores_lock);
974         spin_unlock(&object->lock);
975 nobufs:
976         spin_unlock(&cookie->lock);
977         radix_tree_preload_end();
978         kfree(op);
979         fscache_stat(&fscache_n_stores_nobufs);
980         _leave(" = -ENOBUFS");
981         return -ENOBUFS;
982
983 nomem_free:
984         kfree(op);
985 nomem:
986         fscache_stat(&fscache_n_stores_oom);
987         _leave(" = -ENOMEM");
988         return -ENOMEM;
989 }
990 EXPORT_SYMBOL(__fscache_write_page);
991
992 /*
993  * remove a page from the cache
994  */
995 void __fscache_uncache_page(struct fscache_cookie *cookie, struct page *page)
996 {
997         struct fscache_object *object;
998
999         _enter(",%p", page);
1000
1001         ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
1002         ASSERTCMP(page, !=, NULL);
1003
1004         fscache_stat(&fscache_n_uncaches);
1005
1006         /* cache withdrawal may beat us to it */
1007         if (!PageFsCache(page))
1008                 goto done;
1009
1010         /* get the object */
1011         spin_lock(&cookie->lock);
1012
1013         if (hlist_empty(&cookie->backing_objects)) {
1014                 ClearPageFsCache(page);
1015                 goto done_unlock;
1016         }
1017
1018         object = hlist_entry(cookie->backing_objects.first,
1019                              struct fscache_object, cookie_link);
1020
1021         /* there might now be stuff on disk we could read */
1022         clear_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
1023
1024         /* only invoke the cache backend if we managed to mark the page
1025          * uncached here; this deals with synchronisation vs withdrawal */
1026         if (TestClearPageFsCache(page) &&
1027             object->cache->ops->uncache_page) {
1028                 /* the cache backend releases the cookie lock */
1029                 fscache_stat(&fscache_n_cop_uncache_page);
1030                 object->cache->ops->uncache_page(object, page);
1031                 fscache_stat_d(&fscache_n_cop_uncache_page);
1032                 goto done;
1033         }
1034
1035 done_unlock:
1036         spin_unlock(&cookie->lock);
1037 done:
1038         _leave("");
1039 }
1040 EXPORT_SYMBOL(__fscache_uncache_page);
1041
1042 /**
1043  * fscache_mark_page_cached - Mark a page as being cached
1044  * @op: The retrieval op pages are being marked for
1045  * @page: The page to be marked
1046  *
1047  * Mark a netfs page as being cached.  After this is called, the netfs
1048  * must call fscache_uncache_page() to remove the mark.
1049  */
1050 void fscache_mark_page_cached(struct fscache_retrieval *op, struct page *page)
1051 {
1052         struct fscache_cookie *cookie = op->op.object->cookie;
1053
1054 #ifdef CONFIG_FSCACHE_STATS
1055         atomic_inc(&fscache_n_marks);
1056 #endif
1057
1058         _debug("- mark %p{%lx}", page, page->index);
1059         if (TestSetPageFsCache(page)) {
1060                 static bool once_only;
1061                 if (!once_only) {
1062                         once_only = true;
1063                         printk(KERN_WARNING "FS-Cache:"
1064                                " Cookie type %s marked page %lx"
1065                                " multiple times\n",
1066                                cookie->def->name, page->index);
1067                 }
1068         }
1069
1070         if (cookie->def->mark_page_cached)
1071                 cookie->def->mark_page_cached(cookie->netfs_data,
1072                                               op->mapping, page);
1073 }
1074 EXPORT_SYMBOL(fscache_mark_page_cached);
1075
1076 /**
1077  * fscache_mark_pages_cached - Mark pages as being cached
1078  * @op: The retrieval op pages are being marked for
1079  * @pagevec: The pages to be marked
1080  *
1081  * Mark a bunch of netfs pages as being cached.  After this is called,
1082  * the netfs must call fscache_uncache_page() to remove the mark.
1083  */
1084 void fscache_mark_pages_cached(struct fscache_retrieval *op,
1085                                struct pagevec *pagevec)
1086 {
1087         unsigned long loop;
1088
1089         for (loop = 0; loop < pagevec->nr; loop++)
1090                 fscache_mark_page_cached(op, pagevec->pages[loop]);
1091
1092         pagevec_reinit(pagevec);
1093 }
1094 EXPORT_SYMBOL(fscache_mark_pages_cached);
1095
1096 /*
1097  * Uncache all the pages in an inode that are marked PG_fscache, assuming them
1098  * to be associated with the given cookie.
1099  */
1100 void __fscache_uncache_all_inode_pages(struct fscache_cookie *cookie,
1101                                        struct inode *inode)
1102 {
1103         struct address_space *mapping = inode->i_mapping;
1104         struct pagevec pvec;
1105         pgoff_t next;
1106         int i;
1107
1108         _enter("%p,%p", cookie, inode);
1109
1110         if (!mapping || mapping->nrpages == 0) {
1111                 _leave(" [no pages]");
1112                 return;
1113         }
1114
1115         pagevec_init(&pvec, 0);
1116         next = 0;
1117         do {
1118                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE))
1119                         break;
1120                 for (i = 0; i < pagevec_count(&pvec); i++) {
1121                         struct page *page = pvec.pages[i];
1122                         next = page->index;
1123                         if (PageFsCache(page)) {
1124                                 __fscache_wait_on_page_write(cookie, page);
1125                                 __fscache_uncache_page(cookie, page);
1126                         }
1127                 }
1128                 pagevec_release(&pvec);
1129                 cond_resched();
1130         } while (++next);
1131
1132         _leave("");
1133 }
1134 EXPORT_SYMBOL(__fscache_uncache_all_inode_pages);