These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / fs / cachefiles / rdwr.c
1 /* Storage object read/write
2  *
3  * Copyright (C) 2007 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 Licence
8  * as published by the Free Software Foundation; either version
9  * 2 of the Licence, or (at your option) any later version.
10  */
11
12 #include <linux/mount.h>
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/swap.h>
16 #include "internal.h"
17
18 /*
19  * detect wake up events generated by the unlocking of pages in which we're
20  * interested
21  * - we use this to detect read completion of backing pages
22  * - the caller holds the waitqueue lock
23  */
24 static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode,
25                                   int sync, void *_key)
26 {
27         struct cachefiles_one_read *monitor =
28                 container_of(wait, struct cachefiles_one_read, monitor);
29         struct cachefiles_object *object;
30         struct wait_bit_key *key = _key;
31         struct page *page = wait->private;
32
33         ASSERT(key);
34
35         _enter("{%lu},%u,%d,{%p,%u}",
36                monitor->netfs_page->index, mode, sync,
37                key->flags, key->bit_nr);
38
39         if (key->flags != &page->flags ||
40             key->bit_nr != PG_locked)
41                 return 0;
42
43         _debug("--- monitor %p %lx ---", page, page->flags);
44
45         if (!PageUptodate(page) && !PageError(page)) {
46                 /* unlocked, not uptodate and not erronous? */
47                 _debug("page probably truncated");
48         }
49
50         /* remove from the waitqueue */
51         list_del(&wait->task_list);
52
53         /* move onto the action list and queue for FS-Cache thread pool */
54         ASSERT(monitor->op);
55
56         object = container_of(monitor->op->op.object,
57                               struct cachefiles_object, fscache);
58
59         spin_lock(&object->work_lock);
60         list_add_tail(&monitor->op_link, &monitor->op->to_do);
61         spin_unlock(&object->work_lock);
62
63         fscache_enqueue_retrieval(monitor->op);
64         return 0;
65 }
66
67 /*
68  * handle a probably truncated page
69  * - check to see if the page is still relevant and reissue the read if
70  *   possible
71  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
72  *   must wait again and 0 if successful
73  */
74 static int cachefiles_read_reissue(struct cachefiles_object *object,
75                                    struct cachefiles_one_read *monitor)
76 {
77         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
78         struct page *backpage = monitor->back_page, *backpage2;
79         int ret;
80
81         _enter("{ino=%lx},{%lx,%lx}",
82                d_backing_inode(object->backer)->i_ino,
83                backpage->index, backpage->flags);
84
85         /* skip if the page was truncated away completely */
86         if (backpage->mapping != bmapping) {
87                 _leave(" = -ENODATA [mapping]");
88                 return -ENODATA;
89         }
90
91         backpage2 = find_get_page(bmapping, backpage->index);
92         if (!backpage2) {
93                 _leave(" = -ENODATA [gone]");
94                 return -ENODATA;
95         }
96
97         if (backpage != backpage2) {
98                 put_page(backpage2);
99                 _leave(" = -ENODATA [different]");
100                 return -ENODATA;
101         }
102
103         /* the page is still there and we already have a ref on it, so we don't
104          * need a second */
105         put_page(backpage2);
106
107         INIT_LIST_HEAD(&monitor->op_link);
108         add_page_wait_queue(backpage, &monitor->monitor);
109
110         if (trylock_page(backpage)) {
111                 ret = -EIO;
112                 if (PageError(backpage))
113                         goto unlock_discard;
114                 ret = 0;
115                 if (PageUptodate(backpage))
116                         goto unlock_discard;
117
118                 _debug("reissue read");
119                 ret = bmapping->a_ops->readpage(NULL, backpage);
120                 if (ret < 0)
121                         goto unlock_discard;
122         }
123
124         /* but the page may have been read before the monitor was installed, so
125          * the monitor may miss the event - so we have to ensure that we do get
126          * one in such a case */
127         if (trylock_page(backpage)) {
128                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
129                 unlock_page(backpage);
130         }
131
132         /* it'll reappear on the todo list */
133         _leave(" = -EINPROGRESS");
134         return -EINPROGRESS;
135
136 unlock_discard:
137         unlock_page(backpage);
138         spin_lock_irq(&object->work_lock);
139         list_del(&monitor->op_link);
140         spin_unlock_irq(&object->work_lock);
141         _leave(" = %d", ret);
142         return ret;
143 }
144
145 /*
146  * copy data from backing pages to netfs pages to complete a read operation
147  * - driven by FS-Cache's thread pool
148  */
149 static void cachefiles_read_copier(struct fscache_operation *_op)
150 {
151         struct cachefiles_one_read *monitor;
152         struct cachefiles_object *object;
153         struct fscache_retrieval *op;
154         int error, max;
155
156         op = container_of(_op, struct fscache_retrieval, op);
157         object = container_of(op->op.object,
158                               struct cachefiles_object, fscache);
159
160         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
161
162         max = 8;
163         spin_lock_irq(&object->work_lock);
164
165         while (!list_empty(&op->to_do)) {
166                 monitor = list_entry(op->to_do.next,
167                                      struct cachefiles_one_read, op_link);
168                 list_del(&monitor->op_link);
169
170                 spin_unlock_irq(&object->work_lock);
171
172                 _debug("- copy {%lu}", monitor->back_page->index);
173
174         recheck:
175                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
176                              &object->fscache.cookie->flags)) {
177                         error = -ESTALE;
178                 } else if (PageUptodate(monitor->back_page)) {
179                         copy_highpage(monitor->netfs_page, monitor->back_page);
180                         fscache_mark_page_cached(monitor->op,
181                                                  monitor->netfs_page);
182                         error = 0;
183                 } else if (!PageError(monitor->back_page)) {
184                         /* the page has probably been truncated */
185                         error = cachefiles_read_reissue(object, monitor);
186                         if (error == -EINPROGRESS)
187                                 goto next;
188                         goto recheck;
189                 } else {
190                         cachefiles_io_error_obj(
191                                 object,
192                                 "Readpage failed on backing file %lx",
193                                 (unsigned long) monitor->back_page->flags);
194                         error = -EIO;
195                 }
196
197                 page_cache_release(monitor->back_page);
198
199                 fscache_end_io(op, monitor->netfs_page, error);
200                 page_cache_release(monitor->netfs_page);
201                 fscache_retrieval_complete(op, 1);
202                 fscache_put_retrieval(op);
203                 kfree(monitor);
204
205         next:
206                 /* let the thread pool have some air occasionally */
207                 max--;
208                 if (max < 0 || need_resched()) {
209                         if (!list_empty(&op->to_do))
210                                 fscache_enqueue_retrieval(op);
211                         _leave(" [maxed out]");
212                         return;
213                 }
214
215                 spin_lock_irq(&object->work_lock);
216         }
217
218         spin_unlock_irq(&object->work_lock);
219         _leave("");
220 }
221
222 /*
223  * read the corresponding page to the given set from the backing file
224  * - an uncertain page is simply discarded, to be tried again another time
225  */
226 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
227                                             struct fscache_retrieval *op,
228                                             struct page *netpage)
229 {
230         struct cachefiles_one_read *monitor;
231         struct address_space *bmapping;
232         struct page *newpage, *backpage;
233         int ret;
234
235         _enter("");
236
237         _debug("read back %p{%lu,%d}",
238                netpage, netpage->index, page_count(netpage));
239
240         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
241         if (!monitor)
242                 goto nomem;
243
244         monitor->netfs_page = netpage;
245         monitor->op = fscache_get_retrieval(op);
246
247         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
248
249         /* attempt to get hold of the backing page */
250         bmapping = d_backing_inode(object->backer)->i_mapping;
251         newpage = NULL;
252
253         for (;;) {
254                 backpage = find_get_page(bmapping, netpage->index);
255                 if (backpage)
256                         goto backing_page_already_present;
257
258                 if (!newpage) {
259                         newpage = __page_cache_alloc(cachefiles_gfp |
260                                                      __GFP_COLD);
261                         if (!newpage)
262                                 goto nomem_monitor;
263                 }
264
265                 ret = add_to_page_cache_lru(newpage, bmapping,
266                                             netpage->index, cachefiles_gfp);
267                 if (ret == 0)
268                         goto installed_new_backing_page;
269                 if (ret != -EEXIST)
270                         goto nomem_page;
271         }
272
273         /* we've installed a new backing page, so now we need to start
274          * it reading */
275 installed_new_backing_page:
276         _debug("- new %p", newpage);
277
278         backpage = newpage;
279         newpage = NULL;
280
281 read_backing_page:
282         ret = bmapping->a_ops->readpage(NULL, backpage);
283         if (ret < 0)
284                 goto read_error;
285
286         /* set the monitor to transfer the data across */
287 monitor_backing_page:
288         _debug("- monitor add");
289
290         /* install the monitor */
291         page_cache_get(monitor->netfs_page);
292         page_cache_get(backpage);
293         monitor->back_page = backpage;
294         monitor->monitor.private = backpage;
295         add_page_wait_queue(backpage, &monitor->monitor);
296         monitor = NULL;
297
298         /* but the page may have been read before the monitor was installed, so
299          * the monitor may miss the event - so we have to ensure that we do get
300          * one in such a case */
301         if (trylock_page(backpage)) {
302                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
303                 unlock_page(backpage);
304         }
305         goto success;
306
307         /* if the backing page is already present, it can be in one of
308          * three states: read in progress, read failed or read okay */
309 backing_page_already_present:
310         _debug("- present");
311
312         if (newpage) {
313                 page_cache_release(newpage);
314                 newpage = NULL;
315         }
316
317         if (PageError(backpage))
318                 goto io_error;
319
320         if (PageUptodate(backpage))
321                 goto backing_page_already_uptodate;
322
323         if (!trylock_page(backpage))
324                 goto monitor_backing_page;
325         _debug("read %p {%lx}", backpage, backpage->flags);
326         goto read_backing_page;
327
328         /* the backing page is already up to date, attach the netfs
329          * page to the pagecache and LRU and copy the data across */
330 backing_page_already_uptodate:
331         _debug("- uptodate");
332
333         fscache_mark_page_cached(op, netpage);
334
335         copy_highpage(netpage, backpage);
336         fscache_end_io(op, netpage, 0);
337         fscache_retrieval_complete(op, 1);
338
339 success:
340         _debug("success");
341         ret = 0;
342
343 out:
344         if (backpage)
345                 page_cache_release(backpage);
346         if (monitor) {
347                 fscache_put_retrieval(monitor->op);
348                 kfree(monitor);
349         }
350         _leave(" = %d", ret);
351         return ret;
352
353 read_error:
354         _debug("read error %d", ret);
355         if (ret == -ENOMEM) {
356                 fscache_retrieval_complete(op, 1);
357                 goto out;
358         }
359 io_error:
360         cachefiles_io_error_obj(object, "Page read error on backing file");
361         fscache_retrieval_complete(op, 1);
362         ret = -ENOBUFS;
363         goto out;
364
365 nomem_page:
366         page_cache_release(newpage);
367 nomem_monitor:
368         fscache_put_retrieval(monitor->op);
369         kfree(monitor);
370 nomem:
371         fscache_retrieval_complete(op, 1);
372         _leave(" = -ENOMEM");
373         return -ENOMEM;
374 }
375
376 /*
377  * read a page from the cache or allocate a block in which to store it
378  * - cache withdrawal is prevented by the caller
379  * - returns -EINTR if interrupted
380  * - returns -ENOMEM if ran out of memory
381  * - returns -ENOBUFS if no buffers can be made available
382  * - returns -ENOBUFS if page is beyond EOF
383  * - if the page is backed by a block in the cache:
384  *   - a read will be started which will call the callback on completion
385  *   - 0 will be returned
386  * - else if the page is unbacked:
387  *   - the metadata will be retained
388  *   - -ENODATA will be returned
389  */
390 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
391                                   struct page *page,
392                                   gfp_t gfp)
393 {
394         struct cachefiles_object *object;
395         struct cachefiles_cache *cache;
396         struct inode *inode;
397         sector_t block0, block;
398         unsigned shift;
399         int ret;
400
401         object = container_of(op->op.object,
402                               struct cachefiles_object, fscache);
403         cache = container_of(object->fscache.cache,
404                              struct cachefiles_cache, cache);
405
406         _enter("{%p},{%lx},,,", object, page->index);
407
408         if (!object->backer)
409                 goto enobufs;
410
411         inode = d_backing_inode(object->backer);
412         ASSERT(S_ISREG(inode->i_mode));
413         ASSERT(inode->i_mapping->a_ops->bmap);
414         ASSERT(inode->i_mapping->a_ops->readpages);
415
416         /* calculate the shift required to use bmap */
417         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
418
419         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
420         op->op.flags |= FSCACHE_OP_ASYNC;
421         op->op.processor = cachefiles_read_copier;
422
423         /* we assume the absence or presence of the first block is a good
424          * enough indication for the page as a whole
425          * - TODO: don't use bmap() for this as it is _not_ actually good
426          *   enough for this as it doesn't indicate errors, but it's all we've
427          *   got for the moment
428          */
429         block0 = page->index;
430         block0 <<= shift;
431
432         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
433         _debug("%llx -> %llx",
434                (unsigned long long) block0,
435                (unsigned long long) block);
436
437         if (block) {
438                 /* submit the apparently valid page to the backing fs to be
439                  * read from disk */
440                 ret = cachefiles_read_backing_file_one(object, op, page);
441         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
442                 /* there's space in the cache we can use */
443                 fscache_mark_page_cached(op, page);
444                 fscache_retrieval_complete(op, 1);
445                 ret = -ENODATA;
446         } else {
447                 goto enobufs;
448         }
449
450         _leave(" = %d", ret);
451         return ret;
452
453 enobufs:
454         fscache_retrieval_complete(op, 1);
455         _leave(" = -ENOBUFS");
456         return -ENOBUFS;
457 }
458
459 /*
460  * read the corresponding pages to the given set from the backing file
461  * - any uncertain pages are simply discarded, to be tried again another time
462  */
463 static int cachefiles_read_backing_file(struct cachefiles_object *object,
464                                         struct fscache_retrieval *op,
465                                         struct list_head *list)
466 {
467         struct cachefiles_one_read *monitor = NULL;
468         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
469         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
470         int ret = 0;
471
472         _enter("");
473
474         list_for_each_entry_safe(netpage, _n, list, lru) {
475                 list_del(&netpage->lru);
476
477                 _debug("read back %p{%lu,%d}",
478                        netpage, netpage->index, page_count(netpage));
479
480                 if (!monitor) {
481                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
482                         if (!monitor)
483                                 goto nomem;
484
485                         monitor->op = fscache_get_retrieval(op);
486                         init_waitqueue_func_entry(&monitor->monitor,
487                                                   cachefiles_read_waiter);
488                 }
489
490                 for (;;) {
491                         backpage = find_get_page(bmapping, netpage->index);
492                         if (backpage)
493                                 goto backing_page_already_present;
494
495                         if (!newpage) {
496                                 newpage = __page_cache_alloc(cachefiles_gfp |
497                                                              __GFP_COLD);
498                                 if (!newpage)
499                                         goto nomem;
500                         }
501
502                         ret = add_to_page_cache_lru(newpage, bmapping,
503                                                     netpage->index,
504                                                     cachefiles_gfp);
505                         if (ret == 0)
506                                 goto installed_new_backing_page;
507                         if (ret != -EEXIST)
508                                 goto nomem;
509                 }
510
511                 /* we've installed a new backing page, so now we need
512                  * to start it reading */
513         installed_new_backing_page:
514                 _debug("- new %p", newpage);
515
516                 backpage = newpage;
517                 newpage = NULL;
518
519         reread_backing_page:
520                 ret = bmapping->a_ops->readpage(NULL, backpage);
521                 if (ret < 0)
522                         goto read_error;
523
524                 /* add the netfs page to the pagecache and LRU, and set the
525                  * monitor to transfer the data across */
526         monitor_backing_page:
527                 _debug("- monitor add");
528
529                 ret = add_to_page_cache_lru(netpage, op->mapping,
530                                             netpage->index, cachefiles_gfp);
531                 if (ret < 0) {
532                         if (ret == -EEXIST) {
533                                 page_cache_release(netpage);
534                                 fscache_retrieval_complete(op, 1);
535                                 continue;
536                         }
537                         goto nomem;
538                 }
539
540                 /* install a monitor */
541                 page_cache_get(netpage);
542                 monitor->netfs_page = netpage;
543
544                 page_cache_get(backpage);
545                 monitor->back_page = backpage;
546                 monitor->monitor.private = backpage;
547                 add_page_wait_queue(backpage, &monitor->monitor);
548                 monitor = NULL;
549
550                 /* but the page may have been read before the monitor was
551                  * installed, so the monitor may miss the event - so we have to
552                  * ensure that we do get one in such a case */
553                 if (trylock_page(backpage)) {
554                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
555                         unlock_page(backpage);
556                 }
557
558                 page_cache_release(backpage);
559                 backpage = NULL;
560
561                 page_cache_release(netpage);
562                 netpage = NULL;
563                 continue;
564
565                 /* if the backing page is already present, it can be in one of
566                  * three states: read in progress, read failed or read okay */
567         backing_page_already_present:
568                 _debug("- present %p", backpage);
569
570                 if (PageError(backpage))
571                         goto io_error;
572
573                 if (PageUptodate(backpage))
574                         goto backing_page_already_uptodate;
575
576                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
577
578                 if (!trylock_page(backpage))
579                         goto monitor_backing_page;
580
581                 if (PageError(backpage)) {
582                         _debug("error %lx", backpage->flags);
583                         unlock_page(backpage);
584                         goto io_error;
585                 }
586
587                 if (PageUptodate(backpage))
588                         goto backing_page_already_uptodate_unlock;
589
590                 /* we've locked a page that's neither up to date nor erroneous,
591                  * so we need to attempt to read it again */
592                 goto reread_backing_page;
593
594                 /* the backing page is already up to date, attach the netfs
595                  * page to the pagecache and LRU and copy the data across */
596         backing_page_already_uptodate_unlock:
597                 _debug("uptodate %lx", backpage->flags);
598                 unlock_page(backpage);
599         backing_page_already_uptodate:
600                 _debug("- uptodate");
601
602                 ret = add_to_page_cache_lru(netpage, op->mapping,
603                                             netpage->index, cachefiles_gfp);
604                 if (ret < 0) {
605                         if (ret == -EEXIST) {
606                                 page_cache_release(netpage);
607                                 fscache_retrieval_complete(op, 1);
608                                 continue;
609                         }
610                         goto nomem;
611                 }
612
613                 copy_highpage(netpage, backpage);
614
615                 page_cache_release(backpage);
616                 backpage = NULL;
617
618                 fscache_mark_page_cached(op, netpage);
619
620                 /* the netpage is unlocked and marked up to date here */
621                 fscache_end_io(op, netpage, 0);
622                 page_cache_release(netpage);
623                 netpage = NULL;
624                 fscache_retrieval_complete(op, 1);
625                 continue;
626         }
627
628         netpage = NULL;
629
630         _debug("out");
631
632 out:
633         /* tidy up */
634         if (newpage)
635                 page_cache_release(newpage);
636         if (netpage)
637                 page_cache_release(netpage);
638         if (backpage)
639                 page_cache_release(backpage);
640         if (monitor) {
641                 fscache_put_retrieval(op);
642                 kfree(monitor);
643         }
644
645         list_for_each_entry_safe(netpage, _n, list, lru) {
646                 list_del(&netpage->lru);
647                 page_cache_release(netpage);
648                 fscache_retrieval_complete(op, 1);
649         }
650
651         _leave(" = %d", ret);
652         return ret;
653
654 nomem:
655         _debug("nomem");
656         ret = -ENOMEM;
657         goto record_page_complete;
658
659 read_error:
660         _debug("read error %d", ret);
661         if (ret == -ENOMEM)
662                 goto record_page_complete;
663 io_error:
664         cachefiles_io_error_obj(object, "Page read error on backing file");
665         ret = -ENOBUFS;
666 record_page_complete:
667         fscache_retrieval_complete(op, 1);
668         goto out;
669 }
670
671 /*
672  * read a list of pages from the cache or allocate blocks in which to store
673  * them
674  */
675 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
676                                    struct list_head *pages,
677                                    unsigned *nr_pages,
678                                    gfp_t gfp)
679 {
680         struct cachefiles_object *object;
681         struct cachefiles_cache *cache;
682         struct list_head backpages;
683         struct pagevec pagevec;
684         struct inode *inode;
685         struct page *page, *_n;
686         unsigned shift, nrbackpages;
687         int ret, ret2, space;
688
689         object = container_of(op->op.object,
690                               struct cachefiles_object, fscache);
691         cache = container_of(object->fscache.cache,
692                              struct cachefiles_cache, cache);
693
694         _enter("{OBJ%x,%d},,%d,,",
695                object->fscache.debug_id, atomic_read(&op->op.usage),
696                *nr_pages);
697
698         if (!object->backer)
699                 goto all_enobufs;
700
701         space = 1;
702         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
703                 space = 0;
704
705         inode = d_backing_inode(object->backer);
706         ASSERT(S_ISREG(inode->i_mode));
707         ASSERT(inode->i_mapping->a_ops->bmap);
708         ASSERT(inode->i_mapping->a_ops->readpages);
709
710         /* calculate the shift required to use bmap */
711         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
712
713         pagevec_init(&pagevec, 0);
714
715         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
716         op->op.flags |= FSCACHE_OP_ASYNC;
717         op->op.processor = cachefiles_read_copier;
718
719         INIT_LIST_HEAD(&backpages);
720         nrbackpages = 0;
721
722         ret = space ? -ENODATA : -ENOBUFS;
723         list_for_each_entry_safe(page, _n, pages, lru) {
724                 sector_t block0, block;
725
726                 /* we assume the absence or presence of the first block is a
727                  * good enough indication for the page as a whole
728                  * - TODO: don't use bmap() for this as it is _not_ actually
729                  *   good enough for this as it doesn't indicate errors, but
730                  *   it's all we've got for the moment
731                  */
732                 block0 = page->index;
733                 block0 <<= shift;
734
735                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
736                                                       block0);
737                 _debug("%llx -> %llx",
738                        (unsigned long long) block0,
739                        (unsigned long long) block);
740
741                 if (block) {
742                         /* we have data - add it to the list to give to the
743                          * backing fs */
744                         list_move(&page->lru, &backpages);
745                         (*nr_pages)--;
746                         nrbackpages++;
747                 } else if (space && pagevec_add(&pagevec, page) == 0) {
748                         fscache_mark_pages_cached(op, &pagevec);
749                         fscache_retrieval_complete(op, 1);
750                         ret = -ENODATA;
751                 } else {
752                         fscache_retrieval_complete(op, 1);
753                 }
754         }
755
756         if (pagevec_count(&pagevec) > 0)
757                 fscache_mark_pages_cached(op, &pagevec);
758
759         if (list_empty(pages))
760                 ret = 0;
761
762         /* submit the apparently valid pages to the backing fs to be read from
763          * disk */
764         if (nrbackpages > 0) {
765                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
766                 if (ret2 == -ENOMEM || ret2 == -EINTR)
767                         ret = ret2;
768         }
769
770         _leave(" = %d [nr=%u%s]",
771                ret, *nr_pages, list_empty(pages) ? " empty" : "");
772         return ret;
773
774 all_enobufs:
775         fscache_retrieval_complete(op, *nr_pages);
776         return -ENOBUFS;
777 }
778
779 /*
780  * allocate a block in the cache in which to store a page
781  * - cache withdrawal is prevented by the caller
782  * - returns -EINTR if interrupted
783  * - returns -ENOMEM if ran out of memory
784  * - returns -ENOBUFS if no buffers can be made available
785  * - returns -ENOBUFS if page is beyond EOF
786  * - otherwise:
787  *   - the metadata will be retained
788  *   - 0 will be returned
789  */
790 int cachefiles_allocate_page(struct fscache_retrieval *op,
791                              struct page *page,
792                              gfp_t gfp)
793 {
794         struct cachefiles_object *object;
795         struct cachefiles_cache *cache;
796         int ret;
797
798         object = container_of(op->op.object,
799                               struct cachefiles_object, fscache);
800         cache = container_of(object->fscache.cache,
801                              struct cachefiles_cache, cache);
802
803         _enter("%p,{%lx},", object, page->index);
804
805         ret = cachefiles_has_space(cache, 0, 1);
806         if (ret == 0)
807                 fscache_mark_page_cached(op, page);
808         else
809                 ret = -ENOBUFS;
810
811         fscache_retrieval_complete(op, 1);
812         _leave(" = %d", ret);
813         return ret;
814 }
815
816 /*
817  * allocate blocks in the cache in which to store a set of pages
818  * - cache withdrawal is prevented by the caller
819  * - returns -EINTR if interrupted
820  * - returns -ENOMEM if ran out of memory
821  * - returns -ENOBUFS if some buffers couldn't be made available
822  * - returns -ENOBUFS if some pages are beyond EOF
823  * - otherwise:
824  *   - -ENODATA will be returned
825  * - metadata will be retained for any page marked
826  */
827 int cachefiles_allocate_pages(struct fscache_retrieval *op,
828                               struct list_head *pages,
829                               unsigned *nr_pages,
830                               gfp_t gfp)
831 {
832         struct cachefiles_object *object;
833         struct cachefiles_cache *cache;
834         struct pagevec pagevec;
835         struct page *page;
836         int ret;
837
838         object = container_of(op->op.object,
839                               struct cachefiles_object, fscache);
840         cache = container_of(object->fscache.cache,
841                              struct cachefiles_cache, cache);
842
843         _enter("%p,,,%d,", object, *nr_pages);
844
845         ret = cachefiles_has_space(cache, 0, *nr_pages);
846         if (ret == 0) {
847                 pagevec_init(&pagevec, 0);
848
849                 list_for_each_entry(page, pages, lru) {
850                         if (pagevec_add(&pagevec, page) == 0)
851                                 fscache_mark_pages_cached(op, &pagevec);
852                 }
853
854                 if (pagevec_count(&pagevec) > 0)
855                         fscache_mark_pages_cached(op, &pagevec);
856                 ret = -ENODATA;
857         } else {
858                 ret = -ENOBUFS;
859         }
860
861         fscache_retrieval_complete(op, *nr_pages);
862         _leave(" = %d", ret);
863         return ret;
864 }
865
866 /*
867  * request a page be stored in the cache
868  * - cache withdrawal is prevented by the caller
869  * - this request may be ignored if there's no cache block available, in which
870  *   case -ENOBUFS will be returned
871  * - if the op is in progress, 0 will be returned
872  */
873 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
874 {
875         struct cachefiles_object *object;
876         struct cachefiles_cache *cache;
877         struct file *file;
878         struct path path;
879         loff_t pos, eof;
880         size_t len;
881         void *data;
882         int ret = -ENOBUFS;
883
884         ASSERT(op != NULL);
885         ASSERT(page != NULL);
886
887         object = container_of(op->op.object,
888                               struct cachefiles_object, fscache);
889
890         _enter("%p,%p{%lx},,,", object, page, page->index);
891
892         if (!object->backer) {
893                 _leave(" = -ENOBUFS");
894                 return -ENOBUFS;
895         }
896
897         ASSERT(d_is_reg(object->backer));
898
899         cache = container_of(object->fscache.cache,
900                              struct cachefiles_cache, cache);
901
902         pos = (loff_t)page->index << PAGE_SHIFT;
903
904         /* We mustn't write more data than we have, so we have to beware of a
905          * partial page at EOF.
906          */
907         eof = object->fscache.store_limit_l;
908         if (pos >= eof)
909                 goto error;
910
911         /* write the page to the backing filesystem and let it store it in its
912          * own time */
913         path.mnt = cache->mnt;
914         path.dentry = object->backer;
915         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
916         if (IS_ERR(file)) {
917                 ret = PTR_ERR(file);
918                 goto error_2;
919         }
920
921         len = PAGE_SIZE;
922         if (eof & ~PAGE_MASK) {
923                 if (eof - pos < PAGE_SIZE) {
924                         _debug("cut short %llx to %llx",
925                                pos, eof);
926                         len = eof - pos;
927                         ASSERTCMP(pos + len, ==, eof);
928                 }
929         }
930
931         data = kmap(page);
932         ret = __kernel_write(file, data, len, &pos);
933         kunmap(page);
934         fput(file);
935         if (ret != len)
936                 goto error_eio;
937
938         _leave(" = 0");
939         return 0;
940
941 error_eio:
942         ret = -EIO;
943 error_2:
944         if (ret == -EIO)
945                 cachefiles_io_error_obj(object,
946                                         "Write page to backing file failed");
947 error:
948         _leave(" = -ENOBUFS [%d]", ret);
949         return -ENOBUFS;
950 }
951
952 /*
953  * detach a backing block from a page
954  * - cache withdrawal is prevented by the caller
955  */
956 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
957 {
958         struct cachefiles_object *object;
959         struct cachefiles_cache *cache;
960
961         object = container_of(_object, struct cachefiles_object, fscache);
962         cache = container_of(object->fscache.cache,
963                              struct cachefiles_cache, cache);
964
965         _enter("%p,{%lu}", object, page->index);
966
967         spin_unlock(&object->fscache.cookie->lock);
968 }