Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Storage object read/write
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/mount.h>
9#include <linux/slab.h>
10#include <linux/file.h>
11#include <linux/swap.h>
12#include "internal.h"
13
14/*
15 * detect wake up events generated by the unlocking of pages in which we're
16 * interested
17 * - we use this to detect read completion of backing pages
18 * - the caller holds the waitqueue lock
19 */
20static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
21 int sync, void *_key)
22{
23 struct cachefiles_one_read *monitor =
24 container_of(wait, struct cachefiles_one_read, monitor);
25 struct cachefiles_object *object;
26 struct fscache_retrieval *op = monitor->op;
27 struct wait_bit_key *key = _key;
28 struct page *page = wait->private;
29
30 ASSERT(key);
31
32 _enter("{%lu},%u,%d,{%p,%u}",
33 monitor->netfs_page->index, mode, sync,
34 key->flags, key->bit_nr);
35
36 if (key->flags != &page->flags ||
37 key->bit_nr != PG_locked)
38 return 0;
39
40 _debug("--- monitor %p %lx ---", page, page->flags);
41
42 if (!PageUptodate(page) && !PageError(page)) {
43 /* unlocked, not uptodate and not erronous? */
44 _debug("page probably truncated");
45 }
46
47 /* remove from the waitqueue */
48 list_del(&wait->entry);
49
50 /* move onto the action list and queue for FS-Cache thread pool */
51 ASSERT(op);
52
53 /* We need to temporarily bump the usage count as we don't own a ref
54 * here otherwise cachefiles_read_copier() may free the op between the
55 * monitor being enqueued on the op->to_do list and the op getting
56 * enqueued on the work queue.
57 */
58 fscache_get_retrieval(op);
59
60 object = container_of(op->op.object, struct cachefiles_object, fscache);
61 spin_lock(&object->work_lock);
62 list_add_tail(&monitor->op_link, &op->to_do);
63 fscache_enqueue_retrieval(op);
64 spin_unlock(&object->work_lock);
65
66 fscache_put_retrieval(op);
67 return 0;
68}
69
70/*
71 * handle a probably truncated page
72 * - check to see if the page is still relevant and reissue the read if
73 * possible
74 * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
75 * must wait again and 0 if successful
76 */
77static int cachefiles_read_reissue(struct cachefiles_object *object,
78 struct cachefiles_one_read *monitor)
79{
80 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
81 struct page *backpage = monitor->back_page, *backpage2;
82 int ret;
83
84 _enter("{ino=%lx},{%lx,%lx}",
85 d_backing_inode(object->backer)->i_ino,
86 backpage->index, backpage->flags);
87
88 /* skip if the page was truncated away completely */
89 if (backpage->mapping != bmapping) {
90 _leave(" = -ENODATA [mapping]");
91 return -ENODATA;
92 }
93
94 backpage2 = find_get_page(bmapping, backpage->index);
95 if (!backpage2) {
96 _leave(" = -ENODATA [gone]");
97 return -ENODATA;
98 }
99
100 if (backpage != backpage2) {
101 put_page(backpage2);
102 _leave(" = -ENODATA [different]");
103 return -ENODATA;
104 }
105
106 /* the page is still there and we already have a ref on it, so we don't
107 * need a second */
108 put_page(backpage2);
109
110 INIT_LIST_HEAD(&monitor->op_link);
111 add_page_wait_queue(backpage, &monitor->monitor);
112
113 if (trylock_page(backpage)) {
114 ret = -EIO;
115 if (PageError(backpage))
116 goto unlock_discard;
117 ret = 0;
118 if (PageUptodate(backpage))
119 goto unlock_discard;
120
121 _debug("reissue read");
122 ret = bmapping->a_ops->readpage(NULL, backpage);
123 if (ret < 0)
124 goto unlock_discard;
125 }
126
127 /* but the page may have been read before the monitor was installed, so
128 * the monitor may miss the event - so we have to ensure that we do get
129 * one in such a case */
130 if (trylock_page(backpage)) {
131 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
132 unlock_page(backpage);
133 }
134
135 /* it'll reappear on the todo list */
136 _leave(" = -EINPROGRESS");
137 return -EINPROGRESS;
138
139unlock_discard:
140 unlock_page(backpage);
141 spin_lock_irq(&object->work_lock);
142 list_del(&monitor->op_link);
143 spin_unlock_irq(&object->work_lock);
144 _leave(" = %d", ret);
145 return ret;
146}
147
148/*
149 * copy data from backing pages to netfs pages to complete a read operation
150 * - driven by FS-Cache's thread pool
151 */
152static void cachefiles_read_copier(struct fscache_operation *_op)
153{
154 struct cachefiles_one_read *monitor;
155 struct cachefiles_object *object;
156 struct fscache_retrieval *op;
157 int error, max;
158
159 op = container_of(_op, struct fscache_retrieval, op);
160 object = container_of(op->op.object,
161 struct cachefiles_object, fscache);
162
163 _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
164
165 max = 8;
166 spin_lock_irq(&object->work_lock);
167
168 while (!list_empty(&op->to_do)) {
169 monitor = list_entry(op->to_do.next,
170 struct cachefiles_one_read, op_link);
171 list_del(&monitor->op_link);
172
173 spin_unlock_irq(&object->work_lock);
174
175 _debug("- copy {%lu}", monitor->back_page->index);
176
177 recheck:
178 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
179 &object->fscache.cookie->flags)) {
180 error = -ESTALE;
181 } else if (PageUptodate(monitor->back_page)) {
182 copy_highpage(monitor->netfs_page, monitor->back_page);
183 fscache_mark_page_cached(monitor->op,
184 monitor->netfs_page);
185 error = 0;
186 } else if (!PageError(monitor->back_page)) {
187 /* the page has probably been truncated */
188 error = cachefiles_read_reissue(object, monitor);
189 if (error == -EINPROGRESS)
190 goto next;
191 goto recheck;
192 } else {
193 cachefiles_io_error_obj(
194 object,
195 "Readpage failed on backing file %lx",
196 (unsigned long) monitor->back_page->flags);
197 error = -EIO;
198 }
199
200 put_page(monitor->back_page);
201
202 fscache_end_io(op, monitor->netfs_page, error);
203 put_page(monitor->netfs_page);
204 fscache_retrieval_complete(op, 1);
205 fscache_put_retrieval(op);
206 kfree(monitor);
207
208 next:
209 /* let the thread pool have some air occasionally */
210 max--;
211 if (max < 0 || need_resched()) {
212 if (!list_empty(&op->to_do))
213 fscache_enqueue_retrieval(op);
214 _leave(" [maxed out]");
215 return;
216 }
217
218 spin_lock_irq(&object->work_lock);
219 }
220
221 spin_unlock_irq(&object->work_lock);
222 _leave("");
223}
224
225/*
226 * read the corresponding page to the given set from the backing file
227 * - an uncertain page is simply discarded, to be tried again another time
228 */
229static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
230 struct fscache_retrieval *op,
231 struct page *netpage)
232{
233 struct cachefiles_one_read *monitor;
234 struct address_space *bmapping;
235 struct page *newpage, *backpage;
236 int ret;
237
238 _enter("");
239
240 _debug("read back %p{%lu,%d}",
241 netpage, netpage->index, page_count(netpage));
242
243 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
244 if (!monitor)
245 goto nomem;
246
247 monitor->netfs_page = netpage;
248 monitor->op = fscache_get_retrieval(op);
249
250 init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
251
252 /* attempt to get hold of the backing page */
253 bmapping = d_backing_inode(object->backer)->i_mapping;
254 newpage = NULL;
255
256 for (;;) {
257 backpage = find_get_page(bmapping, netpage->index);
258 if (backpage)
259 goto backing_page_already_present;
260
261 if (!newpage) {
262 newpage = __page_cache_alloc(cachefiles_gfp);
263 if (!newpage)
264 goto nomem_monitor;
265 }
266
267 ret = add_to_page_cache_lru(newpage, bmapping,
268 netpage->index, cachefiles_gfp);
269 if (ret == 0)
270 goto installed_new_backing_page;
271 if (ret != -EEXIST)
272 goto nomem_page;
273 }
274
275 /* we've installed a new backing page, so now we need to start
276 * it reading */
277installed_new_backing_page:
278 _debug("- new %p", newpage);
279
280 backpage = newpage;
281 newpage = NULL;
282
283read_backing_page:
284 ret = bmapping->a_ops->readpage(NULL, backpage);
285 if (ret < 0)
286 goto read_error;
287
288 /* set the monitor to transfer the data across */
289monitor_backing_page:
290 _debug("- monitor add");
291
292 /* install the monitor */
293 get_page(monitor->netfs_page);
294 get_page(backpage);
295 monitor->back_page = backpage;
296 monitor->monitor.private = backpage;
297 add_page_wait_queue(backpage, &monitor->monitor);
298 monitor = NULL;
299
300 /* but the page may have been read before the monitor was installed, so
301 * the monitor may miss the event - so we have to ensure that we do get
302 * one in such a case */
303 if (trylock_page(backpage)) {
304 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
305 unlock_page(backpage);
306 }
307 goto success;
308
309 /* if the backing page is already present, it can be in one of
310 * three states: read in progress, read failed or read okay */
311backing_page_already_present:
312 _debug("- present");
313
314 if (newpage) {
315 put_page(newpage);
316 newpage = NULL;
317 }
318
319 if (PageError(backpage))
320 goto io_error;
321
322 if (PageUptodate(backpage))
323 goto backing_page_already_uptodate;
324
325 if (!trylock_page(backpage))
326 goto monitor_backing_page;
327 _debug("read %p {%lx}", backpage, backpage->flags);
328 goto read_backing_page;
329
330 /* the backing page is already up to date, attach the netfs
331 * page to the pagecache and LRU and copy the data across */
332backing_page_already_uptodate:
333 _debug("- uptodate");
334
335 fscache_mark_page_cached(op, netpage);
336
337 copy_highpage(netpage, backpage);
338 fscache_end_io(op, netpage, 0);
339 fscache_retrieval_complete(op, 1);
340
341success:
342 _debug("success");
343 ret = 0;
344
345out:
346 if (backpage)
347 put_page(backpage);
348 if (monitor) {
349 fscache_put_retrieval(monitor->op);
350 kfree(monitor);
351 }
352 _leave(" = %d", ret);
353 return ret;
354
355read_error:
356 _debug("read error %d", ret);
357 if (ret == -ENOMEM) {
358 fscache_retrieval_complete(op, 1);
359 goto out;
360 }
361io_error:
362 cachefiles_io_error_obj(object, "Page read error on backing file");
363 fscache_retrieval_complete(op, 1);
364 ret = -ENOBUFS;
365 goto out;
366
367nomem_page:
368 put_page(newpage);
369nomem_monitor:
370 fscache_put_retrieval(monitor->op);
371 kfree(monitor);
372nomem:
373 fscache_retrieval_complete(op, 1);
374 _leave(" = -ENOMEM");
375 return -ENOMEM;
376}
377
378/*
379 * read a page from the cache or allocate a block in which to store it
380 * - cache withdrawal is prevented by the caller
381 * - returns -EINTR if interrupted
382 * - returns -ENOMEM if ran out of memory
383 * - returns -ENOBUFS if no buffers can be made available
384 * - returns -ENOBUFS if page is beyond EOF
385 * - if the page is backed by a block in the cache:
386 * - a read will be started which will call the callback on completion
387 * - 0 will be returned
388 * - else if the page is unbacked:
389 * - the metadata will be retained
390 * - -ENODATA will be returned
391 */
392int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
393 struct page *page,
394 gfp_t gfp)
395{
396 struct cachefiles_object *object;
397 struct cachefiles_cache *cache;
398 struct inode *inode;
399 sector_t block;
400 unsigned shift;
401 int ret, ret2;
402
403 object = container_of(op->op.object,
404 struct cachefiles_object, fscache);
405 cache = container_of(object->fscache.cache,
406 struct cachefiles_cache, cache);
407
408 _enter("{%p},{%lx},,,", object, page->index);
409
410 if (!object->backer)
411 goto enobufs;
412
413 inode = d_backing_inode(object->backer);
414 ASSERT(S_ISREG(inode->i_mode));
415 ASSERT(inode->i_mapping->a_ops->readpages);
416
417 /* calculate the shift required to use bmap */
418 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
419
420 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
421 op->op.flags |= FSCACHE_OP_ASYNC;
422 op->op.processor = cachefiles_read_copier;
423
424 /* we assume the absence or presence of the first block is a good
425 * enough indication for the page as a whole
426 * - TODO: don't use bmap() for this as it is _not_ actually good
427 * enough for this as it doesn't indicate errors, but it's all we've
428 * got for the moment
429 */
430 block = page->index;
431 block <<= shift;
432
433 ret2 = bmap(inode, &block);
434 ASSERT(ret2 == 0);
435
436 _debug("%llx -> %llx",
437 (unsigned long long) (page->index << shift),
438 (unsigned long long) block);
439
440 if (block) {
441 /* submit the apparently valid page to the backing fs to be
442 * read from disk */
443 ret = cachefiles_read_backing_file_one(object, op, page);
444 } else if (cachefiles_has_space(cache, 0, 1) == 0) {
445 /* there's space in the cache we can use */
446 fscache_mark_page_cached(op, page);
447 fscache_retrieval_complete(op, 1);
448 ret = -ENODATA;
449 } else {
450 goto enobufs;
451 }
452
453 _leave(" = %d", ret);
454 return ret;
455
456enobufs:
457 fscache_retrieval_complete(op, 1);
458 _leave(" = -ENOBUFS");
459 return -ENOBUFS;
460}
461
462/*
463 * read the corresponding pages to the given set from the backing file
464 * - any uncertain pages are simply discarded, to be tried again another time
465 */
466static int cachefiles_read_backing_file(struct cachefiles_object *object,
467 struct fscache_retrieval *op,
468 struct list_head *list)
469{
470 struct cachefiles_one_read *monitor = NULL;
471 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
472 struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
473 int ret = 0;
474
475 _enter("");
476
477 list_for_each_entry_safe(netpage, _n, list, lru) {
478 list_del(&netpage->lru);
479
480 _debug("read back %p{%lu,%d}",
481 netpage, netpage->index, page_count(netpage));
482
483 if (!monitor) {
484 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
485 if (!monitor)
486 goto nomem;
487
488 monitor->op = fscache_get_retrieval(op);
489 init_waitqueue_func_entry(&monitor->monitor,
490 cachefiles_read_waiter);
491 }
492
493 for (;;) {
494 backpage = find_get_page(bmapping, netpage->index);
495 if (backpage)
496 goto backing_page_already_present;
497
498 if (!newpage) {
499 newpage = __page_cache_alloc(cachefiles_gfp);
500 if (!newpage)
501 goto nomem;
502 }
503
504 ret = add_to_page_cache_lru(newpage, bmapping,
505 netpage->index,
506 cachefiles_gfp);
507 if (ret == 0)
508 goto installed_new_backing_page;
509 if (ret != -EEXIST)
510 goto nomem;
511 }
512
513 /* we've installed a new backing page, so now we need
514 * to start it reading */
515 installed_new_backing_page:
516 _debug("- new %p", newpage);
517
518 backpage = newpage;
519 newpage = NULL;
520
521 reread_backing_page:
522 ret = bmapping->a_ops->readpage(NULL, backpage);
523 if (ret < 0)
524 goto read_error;
525
526 /* add the netfs page to the pagecache and LRU, and set the
527 * monitor to transfer the data across */
528 monitor_backing_page:
529 _debug("- monitor add");
530
531 ret = add_to_page_cache_lru(netpage, op->mapping,
532 netpage->index, cachefiles_gfp);
533 if (ret < 0) {
534 if (ret == -EEXIST) {
535 put_page(backpage);
536 backpage = NULL;
537 put_page(netpage);
538 netpage = NULL;
539 fscache_retrieval_complete(op, 1);
540 continue;
541 }
542 goto nomem;
543 }
544
545 /* install a monitor */
546 get_page(netpage);
547 monitor->netfs_page = netpage;
548
549 get_page(backpage);
550 monitor->back_page = backpage;
551 monitor->monitor.private = backpage;
552 add_page_wait_queue(backpage, &monitor->monitor);
553 monitor = NULL;
554
555 /* but the page may have been read before the monitor was
556 * installed, so the monitor may miss the event - so we have to
557 * ensure that we do get one in such a case */
558 if (trylock_page(backpage)) {
559 _debug("2unlock %p {%lx}", backpage, backpage->flags);
560 unlock_page(backpage);
561 }
562
563 put_page(backpage);
564 backpage = NULL;
565
566 put_page(netpage);
567 netpage = NULL;
568 continue;
569
570 /* if the backing page is already present, it can be in one of
571 * three states: read in progress, read failed or read okay */
572 backing_page_already_present:
573 _debug("- present %p", backpage);
574
575 if (PageError(backpage))
576 goto io_error;
577
578 if (PageUptodate(backpage))
579 goto backing_page_already_uptodate;
580
581 _debug("- not ready %p{%lx}", backpage, backpage->flags);
582
583 if (!trylock_page(backpage))
584 goto monitor_backing_page;
585
586 if (PageError(backpage)) {
587 _debug("error %lx", backpage->flags);
588 unlock_page(backpage);
589 goto io_error;
590 }
591
592 if (PageUptodate(backpage))
593 goto backing_page_already_uptodate_unlock;
594
595 /* we've locked a page that's neither up to date nor erroneous,
596 * so we need to attempt to read it again */
597 goto reread_backing_page;
598
599 /* the backing page is already up to date, attach the netfs
600 * page to the pagecache and LRU and copy the data across */
601 backing_page_already_uptodate_unlock:
602 _debug("uptodate %lx", backpage->flags);
603 unlock_page(backpage);
604 backing_page_already_uptodate:
605 _debug("- uptodate");
606
607 ret = add_to_page_cache_lru(netpage, op->mapping,
608 netpage->index, cachefiles_gfp);
609 if (ret < 0) {
610 if (ret == -EEXIST) {
611 put_page(backpage);
612 backpage = NULL;
613 put_page(netpage);
614 netpage = NULL;
615 fscache_retrieval_complete(op, 1);
616 continue;
617 }
618 goto nomem;
619 }
620
621 copy_highpage(netpage, backpage);
622
623 put_page(backpage);
624 backpage = NULL;
625
626 fscache_mark_page_cached(op, netpage);
627
628 /* the netpage is unlocked and marked up to date here */
629 fscache_end_io(op, netpage, 0);
630 put_page(netpage);
631 netpage = NULL;
632 fscache_retrieval_complete(op, 1);
633 continue;
634 }
635
636 netpage = NULL;
637
638 _debug("out");
639
640out:
641 /* tidy up */
642 if (newpage)
643 put_page(newpage);
644 if (netpage)
645 put_page(netpage);
646 if (backpage)
647 put_page(backpage);
648 if (monitor) {
649 fscache_put_retrieval(op);
650 kfree(monitor);
651 }
652
653 list_for_each_entry_safe(netpage, _n, list, lru) {
654 list_del(&netpage->lru);
655 put_page(netpage);
656 fscache_retrieval_complete(op, 1);
657 }
658
659 _leave(" = %d", ret);
660 return ret;
661
662nomem:
663 _debug("nomem");
664 ret = -ENOMEM;
665 goto record_page_complete;
666
667read_error:
668 _debug("read error %d", ret);
669 if (ret == -ENOMEM)
670 goto record_page_complete;
671io_error:
672 cachefiles_io_error_obj(object, "Page read error on backing file");
673 ret = -ENOBUFS;
674record_page_complete:
675 fscache_retrieval_complete(op, 1);
676 goto out;
677}
678
679/*
680 * read a list of pages from the cache or allocate blocks in which to store
681 * them
682 */
683int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
684 struct list_head *pages,
685 unsigned *nr_pages,
686 gfp_t gfp)
687{
688 struct cachefiles_object *object;
689 struct cachefiles_cache *cache;
690 struct list_head backpages;
691 struct pagevec pagevec;
692 struct inode *inode;
693 struct page *page, *_n;
694 unsigned shift, nrbackpages;
695 int ret, ret2, space;
696
697 object = container_of(op->op.object,
698 struct cachefiles_object, fscache);
699 cache = container_of(object->fscache.cache,
700 struct cachefiles_cache, cache);
701
702 _enter("{OBJ%x,%d},,%d,,",
703 object->fscache.debug_id, atomic_read(&op->op.usage),
704 *nr_pages);
705
706 if (!object->backer)
707 goto all_enobufs;
708
709 space = 1;
710 if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
711 space = 0;
712
713 inode = d_backing_inode(object->backer);
714 ASSERT(S_ISREG(inode->i_mode));
715 ASSERT(inode->i_mapping->a_ops->readpages);
716
717 /* calculate the shift required to use bmap */
718 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
719
720 pagevec_init(&pagevec);
721
722 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
723 op->op.flags |= FSCACHE_OP_ASYNC;
724 op->op.processor = cachefiles_read_copier;
725
726 INIT_LIST_HEAD(&backpages);
727 nrbackpages = 0;
728
729 ret = space ? -ENODATA : -ENOBUFS;
730 list_for_each_entry_safe(page, _n, pages, lru) {
731 sector_t block;
732
733 /* we assume the absence or presence of the first block is a
734 * good enough indication for the page as a whole
735 * - TODO: don't use bmap() for this as it is _not_ actually
736 * good enough for this as it doesn't indicate errors, but
737 * it's all we've got for the moment
738 */
739 block = page->index;
740 block <<= shift;
741
742 ret2 = bmap(inode, &block);
743 ASSERT(ret2 == 0);
744
745 _debug("%llx -> %llx",
746 (unsigned long long) (page->index << shift),
747 (unsigned long long) block);
748
749 if (block) {
750 /* we have data - add it to the list to give to the
751 * backing fs */
752 list_move(&page->lru, &backpages);
753 (*nr_pages)--;
754 nrbackpages++;
755 } else if (space && pagevec_add(&pagevec, page) == 0) {
756 fscache_mark_pages_cached(op, &pagevec);
757 fscache_retrieval_complete(op, 1);
758 ret = -ENODATA;
759 } else {
760 fscache_retrieval_complete(op, 1);
761 }
762 }
763
764 if (pagevec_count(&pagevec) > 0)
765 fscache_mark_pages_cached(op, &pagevec);
766
767 if (list_empty(pages))
768 ret = 0;
769
770 /* submit the apparently valid pages to the backing fs to be read from
771 * disk */
772 if (nrbackpages > 0) {
773 ret2 = cachefiles_read_backing_file(object, op, &backpages);
774 if (ret2 == -ENOMEM || ret2 == -EINTR)
775 ret = ret2;
776 }
777
778 _leave(" = %d [nr=%u%s]",
779 ret, *nr_pages, list_empty(pages) ? " empty" : "");
780 return ret;
781
782all_enobufs:
783 fscache_retrieval_complete(op, *nr_pages);
784 return -ENOBUFS;
785}
786
787/*
788 * allocate a block in the cache in which to store a page
789 * - cache withdrawal is prevented by the caller
790 * - returns -EINTR if interrupted
791 * - returns -ENOMEM if ran out of memory
792 * - returns -ENOBUFS if no buffers can be made available
793 * - returns -ENOBUFS if page is beyond EOF
794 * - otherwise:
795 * - the metadata will be retained
796 * - 0 will be returned
797 */
798int cachefiles_allocate_page(struct fscache_retrieval *op,
799 struct page *page,
800 gfp_t gfp)
801{
802 struct cachefiles_object *object;
803 struct cachefiles_cache *cache;
804 int ret;
805
806 object = container_of(op->op.object,
807 struct cachefiles_object, fscache);
808 cache = container_of(object->fscache.cache,
809 struct cachefiles_cache, cache);
810
811 _enter("%p,{%lx},", object, page->index);
812
813 ret = cachefiles_has_space(cache, 0, 1);
814 if (ret == 0)
815 fscache_mark_page_cached(op, page);
816 else
817 ret = -ENOBUFS;
818
819 fscache_retrieval_complete(op, 1);
820 _leave(" = %d", ret);
821 return ret;
822}
823
824/*
825 * allocate blocks in the cache in which to store a set of pages
826 * - cache withdrawal is prevented by the caller
827 * - returns -EINTR if interrupted
828 * - returns -ENOMEM if ran out of memory
829 * - returns -ENOBUFS if some buffers couldn't be made available
830 * - returns -ENOBUFS if some pages are beyond EOF
831 * - otherwise:
832 * - -ENODATA will be returned
833 * - metadata will be retained for any page marked
834 */
835int cachefiles_allocate_pages(struct fscache_retrieval *op,
836 struct list_head *pages,
837 unsigned *nr_pages,
838 gfp_t gfp)
839{
840 struct cachefiles_object *object;
841 struct cachefiles_cache *cache;
842 struct pagevec pagevec;
843 struct page *page;
844 int ret;
845
846 object = container_of(op->op.object,
847 struct cachefiles_object, fscache);
848 cache = container_of(object->fscache.cache,
849 struct cachefiles_cache, cache);
850
851 _enter("%p,,,%d,", object, *nr_pages);
852
853 ret = cachefiles_has_space(cache, 0, *nr_pages);
854 if (ret == 0) {
855 pagevec_init(&pagevec);
856
857 list_for_each_entry(page, pages, lru) {
858 if (pagevec_add(&pagevec, page) == 0)
859 fscache_mark_pages_cached(op, &pagevec);
860 }
861
862 if (pagevec_count(&pagevec) > 0)
863 fscache_mark_pages_cached(op, &pagevec);
864 ret = -ENODATA;
865 } else {
866 ret = -ENOBUFS;
867 }
868
869 fscache_retrieval_complete(op, *nr_pages);
870 _leave(" = %d", ret);
871 return ret;
872}
873
874/*
875 * request a page be stored in the cache
876 * - cache withdrawal is prevented by the caller
877 * - this request may be ignored if there's no cache block available, in which
878 * case -ENOBUFS will be returned
879 * - if the op is in progress, 0 will be returned
880 */
881int cachefiles_write_page(struct fscache_storage *op, struct page *page)
882{
883 struct cachefiles_object *object;
884 struct cachefiles_cache *cache;
885 struct file *file;
886 struct path path;
887 loff_t pos, eof;
888 size_t len;
889 void *data;
890 int ret = -ENOBUFS;
891
892 ASSERT(op != NULL);
893 ASSERT(page != NULL);
894
895 object = container_of(op->op.object,
896 struct cachefiles_object, fscache);
897
898 _enter("%p,%p{%lx},,,", object, page, page->index);
899
900 if (!object->backer) {
901 _leave(" = -ENOBUFS");
902 return -ENOBUFS;
903 }
904
905 ASSERT(d_is_reg(object->backer));
906
907 cache = container_of(object->fscache.cache,
908 struct cachefiles_cache, cache);
909
910 pos = (loff_t)page->index << PAGE_SHIFT;
911
912 /* We mustn't write more data than we have, so we have to beware of a
913 * partial page at EOF.
914 */
915 eof = object->fscache.store_limit_l;
916 if (pos >= eof)
917 goto error;
918
919 /* write the page to the backing filesystem and let it store it in its
920 * own time */
921 path.mnt = cache->mnt;
922 path.dentry = object->backer;
923 file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
924 if (IS_ERR(file)) {
925 ret = PTR_ERR(file);
926 goto error_2;
927 }
928
929 len = PAGE_SIZE;
930 if (eof & ~PAGE_MASK) {
931 if (eof - pos < PAGE_SIZE) {
932 _debug("cut short %llx to %llx",
933 pos, eof);
934 len = eof - pos;
935 ASSERTCMP(pos + len, ==, eof);
936 }
937 }
938
939 data = kmap(page);
940 ret = kernel_write(file, data, len, &pos);
941 kunmap(page);
942 fput(file);
943 if (ret != len)
944 goto error_eio;
945
946 _leave(" = 0");
947 return 0;
948
949error_eio:
950 ret = -EIO;
951error_2:
952 if (ret == -EIO)
953 cachefiles_io_error_obj(object,
954 "Write page to backing file failed");
955error:
956 _leave(" = -ENOBUFS [%d]", ret);
957 return -ENOBUFS;
958}
959
960/*
961 * detach a backing block from a page
962 * - cache withdrawal is prevented by the caller
963 */
964void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
965 __releases(&object->fscache.cookie->lock)
966{
967 struct cachefiles_object *object;
968
969 object = container_of(_object, struct cachefiles_object, fscache);
970
971 _enter("%p,{%lu}", object, page->index);
972
973 spin_unlock(&object->fscache.cookie->lock);
974}
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 */
24static 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 */
74static 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
136unlock_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 */
149static 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 put_page(monitor->back_page);
198
199 fscache_end_io(op, monitor->netfs_page, error);
200 put_page(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 */
226static 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 */
275installed_new_backing_page:
276 _debug("- new %p", newpage);
277
278 backpage = newpage;
279 newpage = NULL;
280
281read_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 */
287monitor_backing_page:
288 _debug("- monitor add");
289
290 /* install the monitor */
291 get_page(monitor->netfs_page);
292 get_page(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 */
309backing_page_already_present:
310 _debug("- present");
311
312 if (newpage) {
313 put_page(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 */
330backing_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
339success:
340 _debug("success");
341 ret = 0;
342
343out:
344 if (backpage)
345 put_page(backpage);
346 if (monitor) {
347 fscache_put_retrieval(monitor->op);
348 kfree(monitor);
349 }
350 _leave(" = %d", ret);
351 return ret;
352
353read_error:
354 _debug("read error %d", ret);
355 if (ret == -ENOMEM) {
356 fscache_retrieval_complete(op, 1);
357 goto out;
358 }
359io_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
365nomem_page:
366 put_page(newpage);
367nomem_monitor:
368 fscache_put_retrieval(monitor->op);
369 kfree(monitor);
370nomem:
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 */
390int 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
453enobufs:
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 */
463static 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 put_page(netpage);
534 fscache_retrieval_complete(op, 1);
535 continue;
536 }
537 goto nomem;
538 }
539
540 /* install a monitor */
541 get_page(netpage);
542 monitor->netfs_page = netpage;
543
544 get_page(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 put_page(backpage);
559 backpage = NULL;
560
561 put_page(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 put_page(netpage);
607 fscache_retrieval_complete(op, 1);
608 continue;
609 }
610 goto nomem;
611 }
612
613 copy_highpage(netpage, backpage);
614
615 put_page(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 put_page(netpage);
623 netpage = NULL;
624 fscache_retrieval_complete(op, 1);
625 continue;
626 }
627
628 netpage = NULL;
629
630 _debug("out");
631
632out:
633 /* tidy up */
634 if (newpage)
635 put_page(newpage);
636 if (netpage)
637 put_page(netpage);
638 if (backpage)
639 put_page(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 put_page(netpage);
648 fscache_retrieval_complete(op, 1);
649 }
650
651 _leave(" = %d", ret);
652 return ret;
653
654nomem:
655 _debug("nomem");
656 ret = -ENOMEM;
657 goto record_page_complete;
658
659read_error:
660 _debug("read error %d", ret);
661 if (ret == -ENOMEM)
662 goto record_page_complete;
663io_error:
664 cachefiles_io_error_obj(object, "Page read error on backing file");
665 ret = -ENOBUFS;
666record_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 */
675int 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
774all_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 */
790int 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 */
827int 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 */
873int 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
941error_eio:
942 ret = -EIO;
943error_2:
944 if (ret == -EIO)
945 cachefiles_io_error_obj(object,
946 "Write page to backing file failed");
947error:
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 */
956void 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}