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 "internal.h"
16
17/*
18 * detect wake up events generated by the unlocking of pages in which we're
19 * interested
20 * - we use this to detect read completion of backing pages
21 * - the caller holds the waitqueue lock
22 */
23static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode,
24 int sync, void *_key)
25{
26 struct cachefiles_one_read *monitor =
27 container_of(wait, struct cachefiles_one_read, monitor);
28 struct cachefiles_object *object;
29 struct wait_bit_key *key = _key;
30 struct page *page = wait->private;
31
32 ASSERT(key);
33
34 _enter("{%lu},%u,%d,{%p,%u}",
35 monitor->netfs_page->index, mode, sync,
36 key->flags, key->bit_nr);
37
38 if (key->flags != &page->flags ||
39 key->bit_nr != PG_locked)
40 return 0;
41
42 _debug("--- monitor %p %lx ---", page, page->flags);
43
44 if (!PageUptodate(page) && !PageError(page)) {
45 /* unlocked, not uptodate and not erronous? */
46 _debug("page probably truncated");
47 }
48
49 /* remove from the waitqueue */
50 list_del(&wait->task_list);
51
52 /* move onto the action list and queue for FS-Cache thread pool */
53 ASSERT(monitor->op);
54
55 object = container_of(monitor->op->op.object,
56 struct cachefiles_object, fscache);
57
58 spin_lock(&object->work_lock);
59 list_add_tail(&monitor->op_link, &monitor->op->to_do);
60 spin_unlock(&object->work_lock);
61
62 fscache_enqueue_retrieval(monitor->op);
63 return 0;
64}
65
66/*
67 * handle a probably truncated page
68 * - check to see if the page is still relevant and reissue the read if
69 * possible
70 * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
71 * must wait again and 0 if successful
72 */
73static int cachefiles_read_reissue(struct cachefiles_object *object,
74 struct cachefiles_one_read *monitor)
75{
76 struct address_space *bmapping = object->backer->d_inode->i_mapping;
77 struct page *backpage = monitor->back_page, *backpage2;
78 int ret;
79
80 kenter("{ino=%lx},{%lx,%lx}",
81 object->backer->d_inode->i_ino,
82 backpage->index, backpage->flags);
83
84 /* skip if the page was truncated away completely */
85 if (backpage->mapping != bmapping) {
86 kleave(" = -ENODATA [mapping]");
87 return -ENODATA;
88 }
89
90 backpage2 = find_get_page(bmapping, backpage->index);
91 if (!backpage2) {
92 kleave(" = -ENODATA [gone]");
93 return -ENODATA;
94 }
95
96 if (backpage != backpage2) {
97 put_page(backpage2);
98 kleave(" = -ENODATA [different]");
99 return -ENODATA;
100 }
101
102 /* the page is still there and we already have a ref on it, so we don't
103 * need a second */
104 put_page(backpage2);
105
106 INIT_LIST_HEAD(&monitor->op_link);
107 add_page_wait_queue(backpage, &monitor->monitor);
108
109 if (trylock_page(backpage)) {
110 ret = -EIO;
111 if (PageError(backpage))
112 goto unlock_discard;
113 ret = 0;
114 if (PageUptodate(backpage))
115 goto unlock_discard;
116
117 kdebug("reissue read");
118 ret = bmapping->a_ops->readpage(NULL, backpage);
119 if (ret < 0)
120 goto unlock_discard;
121 }
122
123 /* but the page may have been read before the monitor was installed, so
124 * the monitor may miss the event - so we have to ensure that we do get
125 * one in such a case */
126 if (trylock_page(backpage)) {
127 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
128 unlock_page(backpage);
129 }
130
131 /* it'll reappear on the todo list */
132 kleave(" = -EINPROGRESS");
133 return -EINPROGRESS;
134
135unlock_discard:
136 unlock_page(backpage);
137 spin_lock_irq(&object->work_lock);
138 list_del(&monitor->op_link);
139 spin_unlock_irq(&object->work_lock);
140 kleave(" = %d", ret);
141 return ret;
142}
143
144/*
145 * copy data from backing pages to netfs pages to complete a read operation
146 * - driven by FS-Cache's thread pool
147 */
148static void cachefiles_read_copier(struct fscache_operation *_op)
149{
150 struct cachefiles_one_read *monitor;
151 struct cachefiles_object *object;
152 struct fscache_retrieval *op;
153 struct pagevec pagevec;
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}", object->backer->d_inode->i_ino);
161
162 pagevec_init(&pagevec, 0);
163
164 max = 8;
165 spin_lock_irq(&object->work_lock);
166
167 while (!list_empty(&op->to_do)) {
168 monitor = list_entry(op->to_do.next,
169 struct cachefiles_one_read, op_link);
170 list_del(&monitor->op_link);
171
172 spin_unlock_irq(&object->work_lock);
173
174 _debug("- copy {%lu}", monitor->back_page->index);
175
176 recheck:
177 if (PageUptodate(monitor->back_page)) {
178 copy_highpage(monitor->netfs_page, monitor->back_page);
179
180 pagevec_add(&pagevec, monitor->netfs_page);
181 fscache_mark_pages_cached(monitor->op, &pagevec);
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_put_retrieval(op);
202 kfree(monitor);
203
204 next:
205 /* let the thread pool have some air occasionally */
206 max--;
207 if (max < 0 || need_resched()) {
208 if (!list_empty(&op->to_do))
209 fscache_enqueue_retrieval(op);
210 _leave(" [maxed out]");
211 return;
212 }
213
214 spin_lock_irq(&object->work_lock);
215 }
216
217 spin_unlock_irq(&object->work_lock);
218 _leave("");
219}
220
221/*
222 * read the corresponding page to the given set from the backing file
223 * - an uncertain page is simply discarded, to be tried again another time
224 */
225static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
226 struct fscache_retrieval *op,
227 struct page *netpage,
228 struct pagevec *pagevec)
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 pagevec_reinit(pagevec);
238
239 _debug("read back %p{%lu,%d}",
240 netpage, netpage->index, page_count(netpage));
241
242 monitor = kzalloc(sizeof(*monitor), GFP_KERNEL);
243 if (!monitor)
244 goto nomem;
245
246 monitor->netfs_page = netpage;
247 monitor->op = fscache_get_retrieval(op);
248
249 init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
250
251 /* attempt to get hold of the backing page */
252 bmapping = object->backer->d_inode->i_mapping;
253 newpage = NULL;
254
255 for (;;) {
256 backpage = find_get_page(bmapping, netpage->index);
257 if (backpage)
258 goto backing_page_already_present;
259
260 if (!newpage) {
261 newpage = page_cache_alloc_cold(bmapping);
262 if (!newpage)
263 goto nomem_monitor;
264 }
265
266 ret = add_to_page_cache(newpage, bmapping,
267 netpage->index, GFP_KERNEL);
268 if (ret == 0)
269 goto installed_new_backing_page;
270 if (ret != -EEXIST)
271 goto nomem_page;
272 }
273
274 /* we've installed a new backing page, so now we need to add it
275 * to the LRU list and start it reading */
276installed_new_backing_page:
277 _debug("- new %p", newpage);
278
279 backpage = newpage;
280 newpage = NULL;
281
282 page_cache_get(backpage);
283 pagevec_add(pagevec, backpage);
284 __pagevec_lru_add_file(pagevec);
285
286read_backing_page:
287 ret = bmapping->a_ops->readpage(NULL, backpage);
288 if (ret < 0)
289 goto read_error;
290
291 /* set the monitor to transfer the data across */
292monitor_backing_page:
293 _debug("- monitor add");
294
295 /* install the monitor */
296 page_cache_get(monitor->netfs_page);
297 page_cache_get(backpage);
298 monitor->back_page = backpage;
299 monitor->monitor.private = backpage;
300 add_page_wait_queue(backpage, &monitor->monitor);
301 monitor = NULL;
302
303 /* but the page may have been read before the monitor was installed, so
304 * the monitor may miss the event - so we have to ensure that we do get
305 * one in such a case */
306 if (trylock_page(backpage)) {
307 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
308 unlock_page(backpage);
309 }
310 goto success;
311
312 /* if the backing page is already present, it can be in one of
313 * three states: read in progress, read failed or read okay */
314backing_page_already_present:
315 _debug("- present");
316
317 if (newpage) {
318 page_cache_release(newpage);
319 newpage = NULL;
320 }
321
322 if (PageError(backpage))
323 goto io_error;
324
325 if (PageUptodate(backpage))
326 goto backing_page_already_uptodate;
327
328 if (!trylock_page(backpage))
329 goto monitor_backing_page;
330 _debug("read %p {%lx}", backpage, backpage->flags);
331 goto read_backing_page;
332
333 /* the backing page is already up to date, attach the netfs
334 * page to the pagecache and LRU and copy the data across */
335backing_page_already_uptodate:
336 _debug("- uptodate");
337
338 pagevec_add(pagevec, netpage);
339 fscache_mark_pages_cached(op, pagevec);
340
341 copy_highpage(netpage, backpage);
342 fscache_end_io(op, netpage, 0);
343
344success:
345 _debug("success");
346 ret = 0;
347
348out:
349 if (backpage)
350 page_cache_release(backpage);
351 if (monitor) {
352 fscache_put_retrieval(monitor->op);
353 kfree(monitor);
354 }
355 _leave(" = %d", ret);
356 return ret;
357
358read_error:
359 _debug("read error %d", ret);
360 if (ret == -ENOMEM)
361 goto out;
362io_error:
363 cachefiles_io_error_obj(object, "Page read error on backing file");
364 ret = -ENOBUFS;
365 goto out;
366
367nomem_page:
368 page_cache_release(newpage);
369nomem_monitor:
370 fscache_put_retrieval(monitor->op);
371 kfree(monitor);
372nomem:
373 _leave(" = -ENOMEM");
374 return -ENOMEM;
375}
376
377/*
378 * read a page from the cache or allocate a block in which to store it
379 * - cache withdrawal is prevented by the caller
380 * - returns -EINTR if interrupted
381 * - returns -ENOMEM if ran out of memory
382 * - returns -ENOBUFS if no buffers can be made available
383 * - returns -ENOBUFS if page is beyond EOF
384 * - if the page is backed by a block in the cache:
385 * - a read will be started which will call the callback on completion
386 * - 0 will be returned
387 * - else if the page is unbacked:
388 * - the metadata will be retained
389 * - -ENODATA will be returned
390 */
391int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
392 struct page *page,
393 gfp_t gfp)
394{
395 struct cachefiles_object *object;
396 struct cachefiles_cache *cache;
397 struct pagevec pagevec;
398 struct inode *inode;
399 sector_t block0, block;
400 unsigned shift;
401 int ret;
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 return -ENOBUFS;
412
413 inode = object->backer->d_inode;
414 ASSERT(S_ISREG(inode->i_mode));
415 ASSERT(inode->i_mapping->a_ops->bmap);
416 ASSERT(inode->i_mapping->a_ops->readpages);
417
418 /* calculate the shift required to use bmap */
419 if (inode->i_sb->s_blocksize > PAGE_SIZE)
420 return -ENOBUFS;
421
422 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
423
424 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
425 op->op.flags |= FSCACHE_OP_ASYNC;
426 op->op.processor = cachefiles_read_copier;
427
428 pagevec_init(&pagevec, 0);
429
430 /* we assume the absence or presence of the first block is a good
431 * enough indication for the page as a whole
432 * - TODO: don't use bmap() for this as it is _not_ actually good
433 * enough for this as it doesn't indicate errors, but it's all we've
434 * got for the moment
435 */
436 block0 = page->index;
437 block0 <<= shift;
438
439 block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
440 _debug("%llx -> %llx",
441 (unsigned long long) block0,
442 (unsigned long long) block);
443
444 if (block) {
445 /* submit the apparently valid page to the backing fs to be
446 * read from disk */
447 ret = cachefiles_read_backing_file_one(object, op, page,
448 &pagevec);
449 } else if (cachefiles_has_space(cache, 0, 1) == 0) {
450 /* there's space in the cache we can use */
451 pagevec_add(&pagevec, page);
452 fscache_mark_pages_cached(op, &pagevec);
453 ret = -ENODATA;
454 } else {
455 ret = -ENOBUFS;
456 }
457
458 _leave(" = %d", ret);
459 return ret;
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 struct pagevec *mark_pvec)
470{
471 struct cachefiles_one_read *monitor = NULL;
472 struct address_space *bmapping = object->backer->d_inode->i_mapping;
473 struct pagevec lru_pvec;
474 struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
475 int ret = 0;
476
477 _enter("");
478
479 pagevec_init(&lru_pvec, 0);
480
481 list_for_each_entry_safe(netpage, _n, list, lru) {
482 list_del(&netpage->lru);
483
484 _debug("read back %p{%lu,%d}",
485 netpage, netpage->index, page_count(netpage));
486
487 if (!monitor) {
488 monitor = kzalloc(sizeof(*monitor), GFP_KERNEL);
489 if (!monitor)
490 goto nomem;
491
492 monitor->op = fscache_get_retrieval(op);
493 init_waitqueue_func_entry(&monitor->monitor,
494 cachefiles_read_waiter);
495 }
496
497 for (;;) {
498 backpage = find_get_page(bmapping, netpage->index);
499 if (backpage)
500 goto backing_page_already_present;
501
502 if (!newpage) {
503 newpage = page_cache_alloc_cold(bmapping);
504 if (!newpage)
505 goto nomem;
506 }
507
508 ret = add_to_page_cache(newpage, bmapping,
509 netpage->index, GFP_KERNEL);
510 if (ret == 0)
511 goto installed_new_backing_page;
512 if (ret != -EEXIST)
513 goto nomem;
514 }
515
516 /* we've installed a new backing page, so now we need to add it
517 * to the LRU list and start it reading */
518 installed_new_backing_page:
519 _debug("- new %p", newpage);
520
521 backpage = newpage;
522 newpage = NULL;
523
524 page_cache_get(backpage);
525 if (!pagevec_add(&lru_pvec, backpage))
526 __pagevec_lru_add_file(&lru_pvec);
527
528 reread_backing_page:
529 ret = bmapping->a_ops->readpage(NULL, backpage);
530 if (ret < 0)
531 goto read_error;
532
533 /* add the netfs page to the pagecache and LRU, and set the
534 * monitor to transfer the data across */
535 monitor_backing_page:
536 _debug("- monitor add");
537
538 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
539 GFP_KERNEL);
540 if (ret < 0) {
541 if (ret == -EEXIST) {
542 page_cache_release(netpage);
543 continue;
544 }
545 goto nomem;
546 }
547
548 page_cache_get(netpage);
549 if (!pagevec_add(&lru_pvec, netpage))
550 __pagevec_lru_add_file(&lru_pvec);
551
552 /* install a monitor */
553 page_cache_get(netpage);
554 monitor->netfs_page = netpage;
555
556 page_cache_get(backpage);
557 monitor->back_page = backpage;
558 monitor->monitor.private = backpage;
559 add_page_wait_queue(backpage, &monitor->monitor);
560 monitor = NULL;
561
562 /* but the page may have been read before the monitor was
563 * installed, so the monitor may miss the event - so we have to
564 * ensure that we do get one in such a case */
565 if (trylock_page(backpage)) {
566 _debug("2unlock %p {%lx}", backpage, backpage->flags);
567 unlock_page(backpage);
568 }
569
570 page_cache_release(backpage);
571 backpage = NULL;
572
573 page_cache_release(netpage);
574 netpage = NULL;
575 continue;
576
577 /* if the backing page is already present, it can be in one of
578 * three states: read in progress, read failed or read okay */
579 backing_page_already_present:
580 _debug("- present %p", backpage);
581
582 if (PageError(backpage))
583 goto io_error;
584
585 if (PageUptodate(backpage))
586 goto backing_page_already_uptodate;
587
588 _debug("- not ready %p{%lx}", backpage, backpage->flags);
589
590 if (!trylock_page(backpage))
591 goto monitor_backing_page;
592
593 if (PageError(backpage)) {
594 _debug("error %lx", backpage->flags);
595 unlock_page(backpage);
596 goto io_error;
597 }
598
599 if (PageUptodate(backpage))
600 goto backing_page_already_uptodate_unlock;
601
602 /* we've locked a page that's neither up to date nor erroneous,
603 * so we need to attempt to read it again */
604 goto reread_backing_page;
605
606 /* the backing page is already up to date, attach the netfs
607 * page to the pagecache and LRU and copy the data across */
608 backing_page_already_uptodate_unlock:
609 _debug("uptodate %lx", backpage->flags);
610 unlock_page(backpage);
611 backing_page_already_uptodate:
612 _debug("- uptodate");
613
614 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
615 GFP_KERNEL);
616 if (ret < 0) {
617 if (ret == -EEXIST) {
618 page_cache_release(netpage);
619 continue;
620 }
621 goto nomem;
622 }
623
624 copy_highpage(netpage, backpage);
625
626 page_cache_release(backpage);
627 backpage = NULL;
628
629 if (!pagevec_add(mark_pvec, netpage))
630 fscache_mark_pages_cached(op, mark_pvec);
631
632 page_cache_get(netpage);
633 if (!pagevec_add(&lru_pvec, netpage))
634 __pagevec_lru_add_file(&lru_pvec);
635
636 fscache_end_io(op, netpage, 0);
637 page_cache_release(netpage);
638 netpage = NULL;
639 continue;
640 }
641
642 netpage = NULL;
643
644 _debug("out");
645
646out:
647 /* tidy up */
648 pagevec_lru_add_file(&lru_pvec);
649
650 if (newpage)
651 page_cache_release(newpage);
652 if (netpage)
653 page_cache_release(netpage);
654 if (backpage)
655 page_cache_release(backpage);
656 if (monitor) {
657 fscache_put_retrieval(op);
658 kfree(monitor);
659 }
660
661 list_for_each_entry_safe(netpage, _n, list, lru) {
662 list_del(&netpage->lru);
663 page_cache_release(netpage);
664 }
665
666 _leave(" = %d", ret);
667 return ret;
668
669nomem:
670 _debug("nomem");
671 ret = -ENOMEM;
672 goto out;
673
674read_error:
675 _debug("read error %d", ret);
676 if (ret == -ENOMEM)
677 goto out;
678io_error:
679 cachefiles_io_error_obj(object, "Page read error on backing file");
680 ret = -ENOBUFS;
681 goto out;
682}
683
684/*
685 * read a list of pages from the cache or allocate blocks in which to store
686 * them
687 */
688int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
689 struct list_head *pages,
690 unsigned *nr_pages,
691 gfp_t gfp)
692{
693 struct cachefiles_object *object;
694 struct cachefiles_cache *cache;
695 struct list_head backpages;
696 struct pagevec pagevec;
697 struct inode *inode;
698 struct page *page, *_n;
699 unsigned shift, nrbackpages;
700 int ret, ret2, space;
701
702 object = container_of(op->op.object,
703 struct cachefiles_object, fscache);
704 cache = container_of(object->fscache.cache,
705 struct cachefiles_cache, cache);
706
707 _enter("{OBJ%x,%d},,%d,,",
708 object->fscache.debug_id, atomic_read(&op->op.usage),
709 *nr_pages);
710
711 if (!object->backer)
712 return -ENOBUFS;
713
714 space = 1;
715 if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
716 space = 0;
717
718 inode = object->backer->d_inode;
719 ASSERT(S_ISREG(inode->i_mode));
720 ASSERT(inode->i_mapping->a_ops->bmap);
721 ASSERT(inode->i_mapping->a_ops->readpages);
722
723 /* calculate the shift required to use bmap */
724 if (inode->i_sb->s_blocksize > PAGE_SIZE)
725 return -ENOBUFS;
726
727 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
728
729 pagevec_init(&pagevec, 0);
730
731 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
732 op->op.flags |= FSCACHE_OP_ASYNC;
733 op->op.processor = cachefiles_read_copier;
734
735 INIT_LIST_HEAD(&backpages);
736 nrbackpages = 0;
737
738 ret = space ? -ENODATA : -ENOBUFS;
739 list_for_each_entry_safe(page, _n, pages, lru) {
740 sector_t block0, block;
741
742 /* we assume the absence or presence of the first block is a
743 * good enough indication for the page as a whole
744 * - TODO: don't use bmap() for this as it is _not_ actually
745 * good enough for this as it doesn't indicate errors, but
746 * it's all we've got for the moment
747 */
748 block0 = page->index;
749 block0 <<= shift;
750
751 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
752 block0);
753 _debug("%llx -> %llx",
754 (unsigned long long) block0,
755 (unsigned long long) block);
756
757 if (block) {
758 /* we have data - add it to the list to give to the
759 * backing fs */
760 list_move(&page->lru, &backpages);
761 (*nr_pages)--;
762 nrbackpages++;
763 } else if (space && pagevec_add(&pagevec, page) == 0) {
764 fscache_mark_pages_cached(op, &pagevec);
765 ret = -ENODATA;
766 }
767 }
768
769 if (pagevec_count(&pagevec) > 0)
770 fscache_mark_pages_cached(op, &pagevec);
771
772 if (list_empty(pages))
773 ret = 0;
774
775 /* submit the apparently valid pages to the backing fs to be read from
776 * disk */
777 if (nrbackpages > 0) {
778 ret2 = cachefiles_read_backing_file(object, op, &backpages,
779 &pagevec);
780 if (ret2 == -ENOMEM || ret2 == -EINTR)
781 ret = ret2;
782 }
783
784 if (pagevec_count(&pagevec) > 0)
785 fscache_mark_pages_cached(op, &pagevec);
786
787 _leave(" = %d [nr=%u%s]",
788 ret, *nr_pages, list_empty(pages) ? " empty" : "");
789 return ret;
790}
791
792/*
793 * allocate a block in the cache in which to store a page
794 * - cache withdrawal is prevented by the caller
795 * - returns -EINTR if interrupted
796 * - returns -ENOMEM if ran out of memory
797 * - returns -ENOBUFS if no buffers can be made available
798 * - returns -ENOBUFS if page is beyond EOF
799 * - otherwise:
800 * - the metadata will be retained
801 * - 0 will be returned
802 */
803int cachefiles_allocate_page(struct fscache_retrieval *op,
804 struct page *page,
805 gfp_t gfp)
806{
807 struct cachefiles_object *object;
808 struct cachefiles_cache *cache;
809 struct pagevec pagevec;
810 int ret;
811
812 object = container_of(op->op.object,
813 struct cachefiles_object, fscache);
814 cache = container_of(object->fscache.cache,
815 struct cachefiles_cache, cache);
816
817 _enter("%p,{%lx},", object, page->index);
818
819 ret = cachefiles_has_space(cache, 0, 1);
820 if (ret == 0) {
821 pagevec_init(&pagevec, 0);
822 pagevec_add(&pagevec, page);
823 fscache_mark_pages_cached(op, &pagevec);
824 } else {
825 ret = -ENOBUFS;
826 }
827
828 _leave(" = %d", ret);
829 return ret;
830}
831
832/*
833 * allocate blocks in the cache in which to store a set of pages
834 * - cache withdrawal is prevented by the caller
835 * - returns -EINTR if interrupted
836 * - returns -ENOMEM if ran out of memory
837 * - returns -ENOBUFS if some buffers couldn't be made available
838 * - returns -ENOBUFS if some pages are beyond EOF
839 * - otherwise:
840 * - -ENODATA will be returned
841 * - metadata will be retained for any page marked
842 */
843int cachefiles_allocate_pages(struct fscache_retrieval *op,
844 struct list_head *pages,
845 unsigned *nr_pages,
846 gfp_t gfp)
847{
848 struct cachefiles_object *object;
849 struct cachefiles_cache *cache;
850 struct pagevec pagevec;
851 struct page *page;
852 int ret;
853
854 object = container_of(op->op.object,
855 struct cachefiles_object, fscache);
856 cache = container_of(object->fscache.cache,
857 struct cachefiles_cache, cache);
858
859 _enter("%p,,,%d,", object, *nr_pages);
860
861 ret = cachefiles_has_space(cache, 0, *nr_pages);
862 if (ret == 0) {
863 pagevec_init(&pagevec, 0);
864
865 list_for_each_entry(page, pages, lru) {
866 if (pagevec_add(&pagevec, page) == 0)
867 fscache_mark_pages_cached(op, &pagevec);
868 }
869
870 if (pagevec_count(&pagevec) > 0)
871 fscache_mark_pages_cached(op, &pagevec);
872 ret = -ENODATA;
873 } else {
874 ret = -ENOBUFS;
875 }
876
877 _leave(" = %d", ret);
878 return ret;
879}
880
881/*
882 * request a page be stored in the cache
883 * - cache withdrawal is prevented by the caller
884 * - this request may be ignored if there's no cache block available, in which
885 * case -ENOBUFS will be returned
886 * - if the op is in progress, 0 will be returned
887 */
888int cachefiles_write_page(struct fscache_storage *op, struct page *page)
889{
890 struct cachefiles_object *object;
891 struct cachefiles_cache *cache;
892 mm_segment_t old_fs;
893 struct file *file;
894 loff_t pos, eof;
895 size_t len;
896 void *data;
897 int ret;
898
899 ASSERT(op != NULL);
900 ASSERT(page != NULL);
901
902 object = container_of(op->op.object,
903 struct cachefiles_object, fscache);
904
905 _enter("%p,%p{%lx},,,", object, page, page->index);
906
907 if (!object->backer) {
908 _leave(" = -ENOBUFS");
909 return -ENOBUFS;
910 }
911
912 ASSERT(S_ISREG(object->backer->d_inode->i_mode));
913
914 cache = container_of(object->fscache.cache,
915 struct cachefiles_cache, cache);
916
917 /* write the page to the backing filesystem and let it store it in its
918 * own time */
919 dget(object->backer);
920 mntget(cache->mnt);
921 file = dentry_open(object->backer, cache->mnt, O_RDWR,
922 cache->cache_cred);
923 if (IS_ERR(file)) {
924 ret = PTR_ERR(file);
925 } else {
926 ret = -EIO;
927 if (file->f_op->write) {
928 pos = (loff_t) page->index << PAGE_SHIFT;
929
930 /* we mustn't write more data than we have, so we have
931 * to beware of a partial page at EOF */
932 eof = object->fscache.store_limit_l;
933 len = PAGE_SIZE;
934 if (eof & ~PAGE_MASK) {
935 ASSERTCMP(pos, <, eof);
936 if (eof - pos < PAGE_SIZE) {
937 _debug("cut short %llx to %llx",
938 pos, eof);
939 len = eof - pos;
940 ASSERTCMP(pos + len, ==, eof);
941 }
942 }
943
944 data = kmap(page);
945 old_fs = get_fs();
946 set_fs(KERNEL_DS);
947 ret = file->f_op->write(
948 file, (const void __user *) data, len, &pos);
949 set_fs(old_fs);
950 kunmap(page);
951 if (ret != len)
952 ret = -EIO;
953 }
954 fput(file);
955 }
956
957 if (ret < 0) {
958 if (ret == -EIO)
959 cachefiles_io_error_obj(
960 object, "Write page to backing file failed");
961 ret = -ENOBUFS;
962 }
963
964 _leave(" = %d", ret);
965 return ret;
966}
967
968/*
969 * detach a backing block from a page
970 * - cache withdrawal is prevented by the caller
971 */
972void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
973{
974 struct cachefiles_object *object;
975 struct cachefiles_cache *cache;
976
977 object = container_of(_object, struct cachefiles_object, fscache);
978 cache = container_of(object->fscache.cache,
979 struct cachefiles_cache, cache);
980
981 _enter("%p,{%lu}", object, page->index);
982
983 spin_unlock(&object->fscache.cookie->lock);
984}