1 /* Storage object read/write
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #include <linux/mount.h>
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/swap.h>
19 * detect wake up events generated by the unlocking of pages in which we're
21 * - we use this to detect read completion of backing pages
22 * - the caller holds the waitqueue lock
24 static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
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;
35 _enter("{%lu},%u,%d,{%p,%u}",
36 monitor->netfs_page->index, mode, sync,
37 key->flags, key->bit_nr);
39 if (key->flags != &page->flags ||
40 key->bit_nr != PG_locked)
43 _debug("--- monitor %p %lx ---", page, page->flags);
45 if (!PageUptodate(page) && !PageError(page)) {
46 /* unlocked, not uptodate and not erronous? */
47 _debug("page probably truncated");
50 /* remove from the waitqueue */
51 list_del(&wait->entry);
53 /* move onto the action list and queue for FS-Cache thread pool */
56 object = container_of(monitor->op->op.object,
57 struct cachefiles_object, fscache);
59 spin_lock(&object->work_lock);
60 list_add_tail(&monitor->op_link, &monitor->op->to_do);
61 spin_unlock(&object->work_lock);
63 fscache_enqueue_retrieval(monitor->op);
68 * handle a probably truncated page
69 * - check to see if the page is still relevant and reissue the read if
71 * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
72 * must wait again and 0 if successful
74 static int cachefiles_read_reissue(struct cachefiles_object *object,
75 struct cachefiles_one_read *monitor)
77 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
78 struct page *backpage = monitor->back_page, *backpage2;
81 _enter("{ino=%lx},{%lx,%lx}",
82 d_backing_inode(object->backer)->i_ino,
83 backpage->index, backpage->flags);
85 /* skip if the page was truncated away completely */
86 if (backpage->mapping != bmapping) {
87 _leave(" = -ENODATA [mapping]");
91 backpage2 = find_get_page(bmapping, backpage->index);
93 _leave(" = -ENODATA [gone]");
97 if (backpage != backpage2) {
99 _leave(" = -ENODATA [different]");
103 /* the page is still there and we already have a ref on it, so we don't
107 INIT_LIST_HEAD(&monitor->op_link);
108 add_page_wait_queue(backpage, &monitor->monitor);
110 if (trylock_page(backpage)) {
112 if (PageError(backpage))
115 if (PageUptodate(backpage))
118 _debug("reissue read");
119 ret = bmapping->a_ops->readpage(NULL, backpage);
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);
132 /* it'll reappear on the todo list */
133 _leave(" = -EINPROGRESS");
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);
146 * copy data from backing pages to netfs pages to complete a read operation
147 * - driven by FS-Cache's thread pool
149 static void cachefiles_read_copier(struct fscache_operation *_op)
151 struct cachefiles_one_read *monitor;
152 struct cachefiles_object *object;
153 struct fscache_retrieval *op;
156 op = container_of(_op, struct fscache_retrieval, op);
157 object = container_of(op->op.object,
158 struct cachefiles_object, fscache);
160 _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
163 spin_lock_irq(&object->work_lock);
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);
170 spin_unlock_irq(&object->work_lock);
172 _debug("- copy {%lu}", monitor->back_page->index);
175 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
176 &object->fscache.cookie->flags)) {
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);
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)
190 cachefiles_io_error_obj(
192 "Readpage failed on backing file %lx",
193 (unsigned long) monitor->back_page->flags);
197 put_page(monitor->back_page);
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);
206 /* let the thread pool have some air occasionally */
208 if (max < 0 || need_resched()) {
209 if (!list_empty(&op->to_do))
210 fscache_enqueue_retrieval(op);
211 _leave(" [maxed out]");
215 spin_lock_irq(&object->work_lock);
218 spin_unlock_irq(&object->work_lock);
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
226 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
227 struct fscache_retrieval *op,
228 struct page *netpage)
230 struct cachefiles_one_read *monitor;
231 struct address_space *bmapping;
232 struct page *newpage, *backpage;
237 _debug("read back %p{%lu,%d}",
238 netpage, netpage->index, page_count(netpage));
240 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
244 monitor->netfs_page = netpage;
245 monitor->op = fscache_get_retrieval(op);
247 init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
249 /* attempt to get hold of the backing page */
250 bmapping = d_backing_inode(object->backer)->i_mapping;
254 backpage = find_get_page(bmapping, netpage->index);
256 goto backing_page_already_present;
259 newpage = __page_cache_alloc(cachefiles_gfp);
264 ret = add_to_page_cache_lru(newpage, bmapping,
265 netpage->index, cachefiles_gfp);
267 goto installed_new_backing_page;
272 /* we've installed a new backing page, so now we need to start
274 installed_new_backing_page:
275 _debug("- new %p", newpage);
281 ret = bmapping->a_ops->readpage(NULL, backpage);
285 /* set the monitor to transfer the data across */
286 monitor_backing_page:
287 _debug("- monitor add");
289 /* install the monitor */
290 get_page(monitor->netfs_page);
292 monitor->back_page = backpage;
293 monitor->monitor.private = backpage;
294 add_page_wait_queue(backpage, &monitor->monitor);
297 /* but the page may have been read before the monitor was installed, so
298 * the monitor may miss the event - so we have to ensure that we do get
299 * one in such a case */
300 if (trylock_page(backpage)) {
301 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
302 unlock_page(backpage);
306 /* if the backing page is already present, it can be in one of
307 * three states: read in progress, read failed or read okay */
308 backing_page_already_present:
316 if (PageError(backpage))
319 if (PageUptodate(backpage))
320 goto backing_page_already_uptodate;
322 if (!trylock_page(backpage))
323 goto monitor_backing_page;
324 _debug("read %p {%lx}", backpage, backpage->flags);
325 goto read_backing_page;
327 /* the backing page is already up to date, attach the netfs
328 * page to the pagecache and LRU and copy the data across */
329 backing_page_already_uptodate:
330 _debug("- uptodate");
332 fscache_mark_page_cached(op, netpage);
334 copy_highpage(netpage, backpage);
335 fscache_end_io(op, netpage, 0);
336 fscache_retrieval_complete(op, 1);
346 fscache_put_retrieval(monitor->op);
349 _leave(" = %d", ret);
353 _debug("read error %d", ret);
354 if (ret == -ENOMEM) {
355 fscache_retrieval_complete(op, 1);
359 cachefiles_io_error_obj(object, "Page read error on backing file");
360 fscache_retrieval_complete(op, 1);
367 fscache_put_retrieval(monitor->op);
370 fscache_retrieval_complete(op, 1);
371 _leave(" = -ENOMEM");
376 * read a page from the cache or allocate a block in which to store it
377 * - cache withdrawal is prevented by the caller
378 * - returns -EINTR if interrupted
379 * - returns -ENOMEM if ran out of memory
380 * - returns -ENOBUFS if no buffers can be made available
381 * - returns -ENOBUFS if page is beyond EOF
382 * - if the page is backed by a block in the cache:
383 * - a read will be started which will call the callback on completion
384 * - 0 will be returned
385 * - else if the page is unbacked:
386 * - the metadata will be retained
387 * - -ENODATA will be returned
389 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
393 struct cachefiles_object *object;
394 struct cachefiles_cache *cache;
396 sector_t block0, block;
400 object = container_of(op->op.object,
401 struct cachefiles_object, fscache);
402 cache = container_of(object->fscache.cache,
403 struct cachefiles_cache, cache);
405 _enter("{%p},{%lx},,,", object, page->index);
410 inode = d_backing_inode(object->backer);
411 ASSERT(S_ISREG(inode->i_mode));
412 ASSERT(inode->i_mapping->a_ops->bmap);
413 ASSERT(inode->i_mapping->a_ops->readpages);
415 /* calculate the shift required to use bmap */
416 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
418 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
419 op->op.flags |= FSCACHE_OP_ASYNC;
420 op->op.processor = cachefiles_read_copier;
422 /* we assume the absence or presence of the first block is a good
423 * enough indication for the page as a whole
424 * - TODO: don't use bmap() for this as it is _not_ actually good
425 * enough for this as it doesn't indicate errors, but it's all we've
428 block0 = page->index;
431 block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
432 _debug("%llx -> %llx",
433 (unsigned long long) block0,
434 (unsigned long long) block);
437 /* submit the apparently valid page to the backing fs to be
439 ret = cachefiles_read_backing_file_one(object, op, page);
440 } else if (cachefiles_has_space(cache, 0, 1) == 0) {
441 /* there's space in the cache we can use */
442 fscache_mark_page_cached(op, page);
443 fscache_retrieval_complete(op, 1);
449 _leave(" = %d", ret);
453 fscache_retrieval_complete(op, 1);
454 _leave(" = -ENOBUFS");
459 * read the corresponding pages to the given set from the backing file
460 * - any uncertain pages are simply discarded, to be tried again another time
462 static int cachefiles_read_backing_file(struct cachefiles_object *object,
463 struct fscache_retrieval *op,
464 struct list_head *list)
466 struct cachefiles_one_read *monitor = NULL;
467 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
468 struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
473 list_for_each_entry_safe(netpage, _n, list, lru) {
474 list_del(&netpage->lru);
476 _debug("read back %p{%lu,%d}",
477 netpage, netpage->index, page_count(netpage));
480 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
484 monitor->op = fscache_get_retrieval(op);
485 init_waitqueue_func_entry(&monitor->monitor,
486 cachefiles_read_waiter);
490 backpage = find_get_page(bmapping, netpage->index);
492 goto backing_page_already_present;
495 newpage = __page_cache_alloc(cachefiles_gfp);
500 ret = add_to_page_cache_lru(newpage, bmapping,
504 goto installed_new_backing_page;
509 /* we've installed a new backing page, so now we need
510 * to start it reading */
511 installed_new_backing_page:
512 _debug("- new %p", newpage);
518 ret = bmapping->a_ops->readpage(NULL, backpage);
522 /* add the netfs page to the pagecache and LRU, and set the
523 * monitor to transfer the data across */
524 monitor_backing_page:
525 _debug("- monitor add");
527 ret = add_to_page_cache_lru(netpage, op->mapping,
528 netpage->index, cachefiles_gfp);
530 if (ret == -EEXIST) {
532 fscache_retrieval_complete(op, 1);
538 /* install a monitor */
540 monitor->netfs_page = netpage;
543 monitor->back_page = backpage;
544 monitor->monitor.private = backpage;
545 add_page_wait_queue(backpage, &monitor->monitor);
548 /* but the page may have been read before the monitor was
549 * installed, so the monitor may miss the event - so we have to
550 * ensure that we do get one in such a case */
551 if (trylock_page(backpage)) {
552 _debug("2unlock %p {%lx}", backpage, backpage->flags);
553 unlock_page(backpage);
563 /* if the backing page is already present, it can be in one of
564 * three states: read in progress, read failed or read okay */
565 backing_page_already_present:
566 _debug("- present %p", backpage);
568 if (PageError(backpage))
571 if (PageUptodate(backpage))
572 goto backing_page_already_uptodate;
574 _debug("- not ready %p{%lx}", backpage, backpage->flags);
576 if (!trylock_page(backpage))
577 goto monitor_backing_page;
579 if (PageError(backpage)) {
580 _debug("error %lx", backpage->flags);
581 unlock_page(backpage);
585 if (PageUptodate(backpage))
586 goto backing_page_already_uptodate_unlock;
588 /* we've locked a page that's neither up to date nor erroneous,
589 * so we need to attempt to read it again */
590 goto reread_backing_page;
592 /* the backing page is already up to date, attach the netfs
593 * page to the pagecache and LRU and copy the data across */
594 backing_page_already_uptodate_unlock:
595 _debug("uptodate %lx", backpage->flags);
596 unlock_page(backpage);
597 backing_page_already_uptodate:
598 _debug("- uptodate");
600 ret = add_to_page_cache_lru(netpage, op->mapping,
601 netpage->index, cachefiles_gfp);
603 if (ret == -EEXIST) {
605 fscache_retrieval_complete(op, 1);
611 copy_highpage(netpage, backpage);
616 fscache_mark_page_cached(op, netpage);
618 /* the netpage is unlocked and marked up to date here */
619 fscache_end_io(op, netpage, 0);
622 fscache_retrieval_complete(op, 1);
639 fscache_put_retrieval(op);
643 list_for_each_entry_safe(netpage, _n, list, lru) {
644 list_del(&netpage->lru);
646 fscache_retrieval_complete(op, 1);
649 _leave(" = %d", ret);
655 goto record_page_complete;
658 _debug("read error %d", ret);
660 goto record_page_complete;
662 cachefiles_io_error_obj(object, "Page read error on backing file");
664 record_page_complete:
665 fscache_retrieval_complete(op, 1);
670 * read a list of pages from the cache or allocate blocks in which to store
673 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
674 struct list_head *pages,
678 struct cachefiles_object *object;
679 struct cachefiles_cache *cache;
680 struct list_head backpages;
681 struct pagevec pagevec;
683 struct page *page, *_n;
684 unsigned shift, nrbackpages;
685 int ret, ret2, space;
687 object = container_of(op->op.object,
688 struct cachefiles_object, fscache);
689 cache = container_of(object->fscache.cache,
690 struct cachefiles_cache, cache);
692 _enter("{OBJ%x,%d},,%d,,",
693 object->fscache.debug_id, atomic_read(&op->op.usage),
700 if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
703 inode = d_backing_inode(object->backer);
704 ASSERT(S_ISREG(inode->i_mode));
705 ASSERT(inode->i_mapping->a_ops->bmap);
706 ASSERT(inode->i_mapping->a_ops->readpages);
708 /* calculate the shift required to use bmap */
709 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
711 pagevec_init(&pagevec);
713 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
714 op->op.flags |= FSCACHE_OP_ASYNC;
715 op->op.processor = cachefiles_read_copier;
717 INIT_LIST_HEAD(&backpages);
720 ret = space ? -ENODATA : -ENOBUFS;
721 list_for_each_entry_safe(page, _n, pages, lru) {
722 sector_t block0, block;
724 /* we assume the absence or presence of the first block is a
725 * good enough indication for the page as a whole
726 * - TODO: don't use bmap() for this as it is _not_ actually
727 * good enough for this as it doesn't indicate errors, but
728 * it's all we've got for the moment
730 block0 = page->index;
733 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
735 _debug("%llx -> %llx",
736 (unsigned long long) block0,
737 (unsigned long long) block);
740 /* we have data - add it to the list to give to the
742 list_move(&page->lru, &backpages);
745 } else if (space && pagevec_add(&pagevec, page) == 0) {
746 fscache_mark_pages_cached(op, &pagevec);
747 fscache_retrieval_complete(op, 1);
750 fscache_retrieval_complete(op, 1);
754 if (pagevec_count(&pagevec) > 0)
755 fscache_mark_pages_cached(op, &pagevec);
757 if (list_empty(pages))
760 /* submit the apparently valid pages to the backing fs to be read from
762 if (nrbackpages > 0) {
763 ret2 = cachefiles_read_backing_file(object, op, &backpages);
764 if (ret2 == -ENOMEM || ret2 == -EINTR)
768 _leave(" = %d [nr=%u%s]",
769 ret, *nr_pages, list_empty(pages) ? " empty" : "");
773 fscache_retrieval_complete(op, *nr_pages);
778 * allocate a block in the cache in which to store a page
779 * - cache withdrawal is prevented by the caller
780 * - returns -EINTR if interrupted
781 * - returns -ENOMEM if ran out of memory
782 * - returns -ENOBUFS if no buffers can be made available
783 * - returns -ENOBUFS if page is beyond EOF
785 * - the metadata will be retained
786 * - 0 will be returned
788 int cachefiles_allocate_page(struct fscache_retrieval *op,
792 struct cachefiles_object *object;
793 struct cachefiles_cache *cache;
796 object = container_of(op->op.object,
797 struct cachefiles_object, fscache);
798 cache = container_of(object->fscache.cache,
799 struct cachefiles_cache, cache);
801 _enter("%p,{%lx},", object, page->index);
803 ret = cachefiles_has_space(cache, 0, 1);
805 fscache_mark_page_cached(op, page);
809 fscache_retrieval_complete(op, 1);
810 _leave(" = %d", ret);
815 * allocate blocks in the cache in which to store a set of pages
816 * - cache withdrawal is prevented by the caller
817 * - returns -EINTR if interrupted
818 * - returns -ENOMEM if ran out of memory
819 * - returns -ENOBUFS if some buffers couldn't be made available
820 * - returns -ENOBUFS if some pages are beyond EOF
822 * - -ENODATA will be returned
823 * - metadata will be retained for any page marked
825 int cachefiles_allocate_pages(struct fscache_retrieval *op,
826 struct list_head *pages,
830 struct cachefiles_object *object;
831 struct cachefiles_cache *cache;
832 struct pagevec pagevec;
836 object = container_of(op->op.object,
837 struct cachefiles_object, fscache);
838 cache = container_of(object->fscache.cache,
839 struct cachefiles_cache, cache);
841 _enter("%p,,,%d,", object, *nr_pages);
843 ret = cachefiles_has_space(cache, 0, *nr_pages);
845 pagevec_init(&pagevec);
847 list_for_each_entry(page, pages, lru) {
848 if (pagevec_add(&pagevec, page) == 0)
849 fscache_mark_pages_cached(op, &pagevec);
852 if (pagevec_count(&pagevec) > 0)
853 fscache_mark_pages_cached(op, &pagevec);
859 fscache_retrieval_complete(op, *nr_pages);
860 _leave(" = %d", ret);
865 * request a page be stored in the cache
866 * - cache withdrawal is prevented by the caller
867 * - this request may be ignored if there's no cache block available, in which
868 * case -ENOBUFS will be returned
869 * - if the op is in progress, 0 will be returned
871 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
873 struct cachefiles_object *object;
874 struct cachefiles_cache *cache;
883 ASSERT(page != NULL);
885 object = container_of(op->op.object,
886 struct cachefiles_object, fscache);
888 _enter("%p,%p{%lx},,,", object, page, page->index);
890 if (!object->backer) {
891 _leave(" = -ENOBUFS");
895 ASSERT(d_is_reg(object->backer));
897 cache = container_of(object->fscache.cache,
898 struct cachefiles_cache, cache);
900 pos = (loff_t)page->index << PAGE_SHIFT;
902 /* We mustn't write more data than we have, so we have to beware of a
903 * partial page at EOF.
905 eof = object->fscache.store_limit_l;
909 /* write the page to the backing filesystem and let it store it in its
911 path.mnt = cache->mnt;
912 path.dentry = object->backer;
913 file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
920 if (eof & ~PAGE_MASK) {
921 if (eof - pos < PAGE_SIZE) {
922 _debug("cut short %llx to %llx",
925 ASSERTCMP(pos + len, ==, eof);
930 ret = __kernel_write(file, data, len, &pos);
943 cachefiles_io_error_obj(object,
944 "Write page to backing file failed");
946 _leave(" = -ENOBUFS [%d]", ret);
951 * detach a backing block from a page
952 * - cache withdrawal is prevented by the caller
954 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
955 __releases(&object->fscache.cookie->lock)
957 struct cachefiles_object *object;
958 struct cachefiles_cache *cache;
960 object = container_of(_object, struct cachefiles_object, fscache);
961 cache = container_of(object->fscache.cache,
962 struct cachefiles_cache, cache);
964 _enter("%p,{%lu}", object, page->index);
966 spin_unlock(&object->fscache.cookie->lock);