1 /* Storage object read/write
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
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 fscache_retrieval *op = monitor->op;
31 struct wait_bit_key *key = _key;
32 struct page *page = wait->private;
36 _enter("{%lu},%u,%d,{%p,%u}",
37 monitor->netfs_page->index, mode, sync,
38 key->flags, key->bit_nr);
40 if (key->flags != &page->flags ||
41 key->bit_nr != PG_locked)
44 _debug("--- monitor %p %lx ---", page, page->flags);
46 if (!PageUptodate(page) && !PageError(page)) {
47 /* unlocked, not uptodate and not erronous? */
48 _debug("page probably truncated");
51 /* remove from the waitqueue */
52 list_del(&wait->entry);
54 /* move onto the action list and queue for FS-Cache thread pool */
57 /* We need to temporarily bump the usage count as we don't own a ref
58 * here otherwise cachefiles_read_copier() may free the op between the
59 * monitor being enqueued on the op->to_do list and the op getting
60 * enqueued on the work queue.
62 fscache_get_retrieval(op);
64 object = container_of(op->op.object, struct cachefiles_object, fscache);
65 spin_lock(&object->work_lock);
66 list_add_tail(&monitor->op_link, &op->to_do);
67 spin_unlock(&object->work_lock);
69 fscache_enqueue_retrieval(op);
70 fscache_put_retrieval(op);
75 * handle a probably truncated page
76 * - check to see if the page is still relevant and reissue the read if
78 * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
79 * must wait again and 0 if successful
81 static int cachefiles_read_reissue(struct cachefiles_object *object,
82 struct cachefiles_one_read *monitor)
84 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
85 struct page *backpage = monitor->back_page, *backpage2;
88 _enter("{ino=%lx},{%lx,%lx}",
89 d_backing_inode(object->backer)->i_ino,
90 backpage->index, backpage->flags);
92 /* skip if the page was truncated away completely */
93 if (backpage->mapping != bmapping) {
94 _leave(" = -ENODATA [mapping]");
98 backpage2 = find_get_page(bmapping, backpage->index);
100 _leave(" = -ENODATA [gone]");
104 if (backpage != backpage2) {
106 _leave(" = -ENODATA [different]");
110 /* the page is still there and we already have a ref on it, so we don't
114 INIT_LIST_HEAD(&monitor->op_link);
115 add_page_wait_queue(backpage, &monitor->monitor);
117 if (trylock_page(backpage)) {
119 if (PageError(backpage))
122 if (PageUptodate(backpage))
125 _debug("reissue read");
126 ret = bmapping->a_ops->readpage(NULL, backpage);
131 /* but the page may have been read before the monitor was installed, so
132 * the monitor may miss the event - so we have to ensure that we do get
133 * one in such a case */
134 if (trylock_page(backpage)) {
135 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
136 unlock_page(backpage);
139 /* it'll reappear on the todo list */
140 _leave(" = -EINPROGRESS");
144 unlock_page(backpage);
145 spin_lock_irq(&object->work_lock);
146 list_del(&monitor->op_link);
147 spin_unlock_irq(&object->work_lock);
148 _leave(" = %d", ret);
153 * copy data from backing pages to netfs pages to complete a read operation
154 * - driven by FS-Cache's thread pool
156 static void cachefiles_read_copier(struct fscache_operation *_op)
158 struct cachefiles_one_read *monitor;
159 struct cachefiles_object *object;
160 struct fscache_retrieval *op;
163 op = container_of(_op, struct fscache_retrieval, op);
164 object = container_of(op->op.object,
165 struct cachefiles_object, fscache);
167 _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
170 spin_lock_irq(&object->work_lock);
172 while (!list_empty(&op->to_do)) {
173 monitor = list_entry(op->to_do.next,
174 struct cachefiles_one_read, op_link);
175 list_del(&monitor->op_link);
177 spin_unlock_irq(&object->work_lock);
179 _debug("- copy {%lu}", monitor->back_page->index);
182 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
183 &object->fscache.cookie->flags)) {
185 } else if (PageUptodate(monitor->back_page)) {
186 copy_highpage(monitor->netfs_page, monitor->back_page);
187 fscache_mark_page_cached(monitor->op,
188 monitor->netfs_page);
190 } else if (!PageError(monitor->back_page)) {
191 /* the page has probably been truncated */
192 error = cachefiles_read_reissue(object, monitor);
193 if (error == -EINPROGRESS)
197 cachefiles_io_error_obj(
199 "Readpage failed on backing file %lx",
200 (unsigned long) monitor->back_page->flags);
204 put_page(monitor->back_page);
206 fscache_end_io(op, monitor->netfs_page, error);
207 put_page(monitor->netfs_page);
208 fscache_retrieval_complete(op, 1);
209 fscache_put_retrieval(op);
213 /* let the thread pool have some air occasionally */
215 if (max < 0 || need_resched()) {
216 if (!list_empty(&op->to_do))
217 fscache_enqueue_retrieval(op);
218 _leave(" [maxed out]");
222 spin_lock_irq(&object->work_lock);
225 spin_unlock_irq(&object->work_lock);
230 * read the corresponding page to the given set from the backing file
231 * - an uncertain page is simply discarded, to be tried again another time
233 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
234 struct fscache_retrieval *op,
235 struct page *netpage)
237 struct cachefiles_one_read *monitor;
238 struct address_space *bmapping;
239 struct page *newpage, *backpage;
244 _debug("read back %p{%lu,%d}",
245 netpage, netpage->index, page_count(netpage));
247 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
251 monitor->netfs_page = netpage;
252 monitor->op = fscache_get_retrieval(op);
254 init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
256 /* attempt to get hold of the backing page */
257 bmapping = d_backing_inode(object->backer)->i_mapping;
261 backpage = find_get_page(bmapping, netpage->index);
263 goto backing_page_already_present;
266 newpage = __page_cache_alloc(cachefiles_gfp);
271 ret = add_to_page_cache_lru(newpage, bmapping,
272 netpage->index, cachefiles_gfp);
274 goto installed_new_backing_page;
279 /* we've installed a new backing page, so now we need to start
281 installed_new_backing_page:
282 _debug("- new %p", newpage);
288 ret = bmapping->a_ops->readpage(NULL, backpage);
292 /* set the monitor to transfer the data across */
293 monitor_backing_page:
294 _debug("- monitor add");
296 /* install the monitor */
297 get_page(monitor->netfs_page);
299 monitor->back_page = backpage;
300 monitor->monitor.private = backpage;
301 add_page_wait_queue(backpage, &monitor->monitor);
304 /* but the page may have been read before the monitor was installed, so
305 * the monitor may miss the event - so we have to ensure that we do get
306 * one in such a case */
307 if (trylock_page(backpage)) {
308 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
309 unlock_page(backpage);
313 /* if the backing page is already present, it can be in one of
314 * three states: read in progress, read failed or read okay */
315 backing_page_already_present:
323 if (PageError(backpage))
326 if (PageUptodate(backpage))
327 goto backing_page_already_uptodate;
329 if (!trylock_page(backpage))
330 goto monitor_backing_page;
331 _debug("read %p {%lx}", backpage, backpage->flags);
332 goto read_backing_page;
334 /* the backing page is already up to date, attach the netfs
335 * page to the pagecache and LRU and copy the data across */
336 backing_page_already_uptodate:
337 _debug("- uptodate");
339 fscache_mark_page_cached(op, netpage);
341 copy_highpage(netpage, backpage);
342 fscache_end_io(op, netpage, 0);
343 fscache_retrieval_complete(op, 1);
353 fscache_put_retrieval(monitor->op);
356 _leave(" = %d", ret);
360 _debug("read error %d", ret);
361 if (ret == -ENOMEM) {
362 fscache_retrieval_complete(op, 1);
366 cachefiles_io_error_obj(object, "Page read error on backing file");
367 fscache_retrieval_complete(op, 1);
374 fscache_put_retrieval(monitor->op);
377 fscache_retrieval_complete(op, 1);
378 _leave(" = -ENOMEM");
383 * read a page from the cache or allocate a block in which to store it
384 * - cache withdrawal is prevented by the caller
385 * - returns -EINTR if interrupted
386 * - returns -ENOMEM if ran out of memory
387 * - returns -ENOBUFS if no buffers can be made available
388 * - returns -ENOBUFS if page is beyond EOF
389 * - if the page is backed by a block in the cache:
390 * - a read will be started which will call the callback on completion
391 * - 0 will be returned
392 * - else if the page is unbacked:
393 * - the metadata will be retained
394 * - -ENODATA will be returned
396 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
400 struct cachefiles_object *object;
401 struct cachefiles_cache *cache;
403 sector_t block0, block;
407 object = container_of(op->op.object,
408 struct cachefiles_object, fscache);
409 cache = container_of(object->fscache.cache,
410 struct cachefiles_cache, cache);
412 _enter("{%p},{%lx},,,", object, page->index);
417 inode = d_backing_inode(object->backer);
418 ASSERT(S_ISREG(inode->i_mode));
419 ASSERT(inode->i_mapping->a_ops->bmap);
420 ASSERT(inode->i_mapping->a_ops->readpages);
422 /* calculate the shift required to use bmap */
423 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
425 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
426 op->op.flags |= FSCACHE_OP_ASYNC;
427 op->op.processor = cachefiles_read_copier;
429 /* we assume the absence or presence of the first block is a good
430 * enough indication for the page as a whole
431 * - TODO: don't use bmap() for this as it is _not_ actually good
432 * enough for this as it doesn't indicate errors, but it's all we've
435 block0 = page->index;
438 block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
439 _debug("%llx -> %llx",
440 (unsigned long long) block0,
441 (unsigned long long) block);
444 /* submit the apparently valid page to the backing fs to be
446 ret = cachefiles_read_backing_file_one(object, op, page);
447 } else if (cachefiles_has_space(cache, 0, 1) == 0) {
448 /* there's space in the cache we can use */
449 fscache_mark_page_cached(op, page);
450 fscache_retrieval_complete(op, 1);
456 _leave(" = %d", ret);
460 fscache_retrieval_complete(op, 1);
461 _leave(" = -ENOBUFS");
466 * read the corresponding pages to the given set from the backing file
467 * - any uncertain pages are simply discarded, to be tried again another time
469 static int cachefiles_read_backing_file(struct cachefiles_object *object,
470 struct fscache_retrieval *op,
471 struct list_head *list)
473 struct cachefiles_one_read *monitor = NULL;
474 struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
475 struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
480 list_for_each_entry_safe(netpage, _n, list, lru) {
481 list_del(&netpage->lru);
483 _debug("read back %p{%lu,%d}",
484 netpage, netpage->index, page_count(netpage));
487 monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
491 monitor->op = fscache_get_retrieval(op);
492 init_waitqueue_func_entry(&monitor->monitor,
493 cachefiles_read_waiter);
497 backpage = find_get_page(bmapping, netpage->index);
499 goto backing_page_already_present;
502 newpage = __page_cache_alloc(cachefiles_gfp);
507 ret = add_to_page_cache_lru(newpage, bmapping,
511 goto installed_new_backing_page;
516 /* we've installed a new backing page, so now we need
517 * to start it reading */
518 installed_new_backing_page:
519 _debug("- new %p", newpage);
525 ret = bmapping->a_ops->readpage(NULL, backpage);
529 /* add the netfs page to the pagecache and LRU, and set the
530 * monitor to transfer the data across */
531 monitor_backing_page:
532 _debug("- monitor add");
534 ret = add_to_page_cache_lru(netpage, op->mapping,
535 netpage->index, cachefiles_gfp);
537 if (ret == -EEXIST) {
539 fscache_retrieval_complete(op, 1);
545 /* install a monitor */
547 monitor->netfs_page = netpage;
550 monitor->back_page = backpage;
551 monitor->monitor.private = backpage;
552 add_page_wait_queue(backpage, &monitor->monitor);
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);
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);
575 if (PageError(backpage))
578 if (PageUptodate(backpage))
579 goto backing_page_already_uptodate;
581 _debug("- not ready %p{%lx}", backpage, backpage->flags);
583 if (!trylock_page(backpage))
584 goto monitor_backing_page;
586 if (PageError(backpage)) {
587 _debug("error %lx", backpage->flags);
588 unlock_page(backpage);
592 if (PageUptodate(backpage))
593 goto backing_page_already_uptodate_unlock;
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;
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");
607 ret = add_to_page_cache_lru(netpage, op->mapping,
608 netpage->index, cachefiles_gfp);
610 if (ret == -EEXIST) {
612 fscache_retrieval_complete(op, 1);
618 copy_highpage(netpage, backpage);
623 fscache_mark_page_cached(op, netpage);
625 /* the netpage is unlocked and marked up to date here */
626 fscache_end_io(op, netpage, 0);
629 fscache_retrieval_complete(op, 1);
646 fscache_put_retrieval(op);
650 list_for_each_entry_safe(netpage, _n, list, lru) {
651 list_del(&netpage->lru);
653 fscache_retrieval_complete(op, 1);
656 _leave(" = %d", ret);
662 goto record_page_complete;
665 _debug("read error %d", ret);
667 goto record_page_complete;
669 cachefiles_io_error_obj(object, "Page read error on backing file");
671 record_page_complete:
672 fscache_retrieval_complete(op, 1);
677 * read a list of pages from the cache or allocate blocks in which to store
680 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
681 struct list_head *pages,
685 struct cachefiles_object *object;
686 struct cachefiles_cache *cache;
687 struct list_head backpages;
688 struct pagevec pagevec;
690 struct page *page, *_n;
691 unsigned shift, nrbackpages;
692 int ret, ret2, space;
694 object = container_of(op->op.object,
695 struct cachefiles_object, fscache);
696 cache = container_of(object->fscache.cache,
697 struct cachefiles_cache, cache);
699 _enter("{OBJ%x,%d},,%d,,",
700 object->fscache.debug_id, atomic_read(&op->op.usage),
707 if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
710 inode = d_backing_inode(object->backer);
711 ASSERT(S_ISREG(inode->i_mode));
712 ASSERT(inode->i_mapping->a_ops->bmap);
713 ASSERT(inode->i_mapping->a_ops->readpages);
715 /* calculate the shift required to use bmap */
716 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
718 pagevec_init(&pagevec);
720 op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
721 op->op.flags |= FSCACHE_OP_ASYNC;
722 op->op.processor = cachefiles_read_copier;
724 INIT_LIST_HEAD(&backpages);
727 ret = space ? -ENODATA : -ENOBUFS;
728 list_for_each_entry_safe(page, _n, pages, lru) {
729 sector_t block0, block;
731 /* we assume the absence or presence of the first block is a
732 * good enough indication for the page as a whole
733 * - TODO: don't use bmap() for this as it is _not_ actually
734 * good enough for this as it doesn't indicate errors, but
735 * it's all we've got for the moment
737 block0 = page->index;
740 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
742 _debug("%llx -> %llx",
743 (unsigned long long) block0,
744 (unsigned long long) block);
747 /* we have data - add it to the list to give to the
749 list_move(&page->lru, &backpages);
752 } else if (space && pagevec_add(&pagevec, page) == 0) {
753 fscache_mark_pages_cached(op, &pagevec);
754 fscache_retrieval_complete(op, 1);
757 fscache_retrieval_complete(op, 1);
761 if (pagevec_count(&pagevec) > 0)
762 fscache_mark_pages_cached(op, &pagevec);
764 if (list_empty(pages))
767 /* submit the apparently valid pages to the backing fs to be read from
769 if (nrbackpages > 0) {
770 ret2 = cachefiles_read_backing_file(object, op, &backpages);
771 if (ret2 == -ENOMEM || ret2 == -EINTR)
775 _leave(" = %d [nr=%u%s]",
776 ret, *nr_pages, list_empty(pages) ? " empty" : "");
780 fscache_retrieval_complete(op, *nr_pages);
785 * allocate a block in the cache in which to store a page
786 * - cache withdrawal is prevented by the caller
787 * - returns -EINTR if interrupted
788 * - returns -ENOMEM if ran out of memory
789 * - returns -ENOBUFS if no buffers can be made available
790 * - returns -ENOBUFS if page is beyond EOF
792 * - the metadata will be retained
793 * - 0 will be returned
795 int cachefiles_allocate_page(struct fscache_retrieval *op,
799 struct cachefiles_object *object;
800 struct cachefiles_cache *cache;
803 object = container_of(op->op.object,
804 struct cachefiles_object, fscache);
805 cache = container_of(object->fscache.cache,
806 struct cachefiles_cache, cache);
808 _enter("%p,{%lx},", object, page->index);
810 ret = cachefiles_has_space(cache, 0, 1);
812 fscache_mark_page_cached(op, page);
816 fscache_retrieval_complete(op, 1);
817 _leave(" = %d", ret);
822 * allocate blocks in the cache in which to store a set of pages
823 * - cache withdrawal is prevented by the caller
824 * - returns -EINTR if interrupted
825 * - returns -ENOMEM if ran out of memory
826 * - returns -ENOBUFS if some buffers couldn't be made available
827 * - returns -ENOBUFS if some pages are beyond EOF
829 * - -ENODATA will be returned
830 * - metadata will be retained for any page marked
832 int cachefiles_allocate_pages(struct fscache_retrieval *op,
833 struct list_head *pages,
837 struct cachefiles_object *object;
838 struct cachefiles_cache *cache;
839 struct pagevec pagevec;
843 object = container_of(op->op.object,
844 struct cachefiles_object, fscache);
845 cache = container_of(object->fscache.cache,
846 struct cachefiles_cache, cache);
848 _enter("%p,,,%d,", object, *nr_pages);
850 ret = cachefiles_has_space(cache, 0, *nr_pages);
852 pagevec_init(&pagevec);
854 list_for_each_entry(page, pages, lru) {
855 if (pagevec_add(&pagevec, page) == 0)
856 fscache_mark_pages_cached(op, &pagevec);
859 if (pagevec_count(&pagevec) > 0)
860 fscache_mark_pages_cached(op, &pagevec);
866 fscache_retrieval_complete(op, *nr_pages);
867 _leave(" = %d", ret);
872 * request a page be stored in the cache
873 * - cache withdrawal is prevented by the caller
874 * - this request may be ignored if there's no cache block available, in which
875 * case -ENOBUFS will be returned
876 * - if the op is in progress, 0 will be returned
878 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
880 struct cachefiles_object *object;
881 struct cachefiles_cache *cache;
890 ASSERT(page != NULL);
892 object = container_of(op->op.object,
893 struct cachefiles_object, fscache);
895 _enter("%p,%p{%lx},,,", object, page, page->index);
897 if (!object->backer) {
898 _leave(" = -ENOBUFS");
902 ASSERT(d_is_reg(object->backer));
904 cache = container_of(object->fscache.cache,
905 struct cachefiles_cache, cache);
907 pos = (loff_t)page->index << PAGE_SHIFT;
909 /* We mustn't write more data than we have, so we have to beware of a
910 * partial page at EOF.
912 eof = object->fscache.store_limit_l;
916 /* write the page to the backing filesystem and let it store it in its
918 path.mnt = cache->mnt;
919 path.dentry = object->backer;
920 file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
927 if (eof & ~PAGE_MASK) {
928 if (eof - pos < PAGE_SIZE) {
929 _debug("cut short %llx to %llx",
932 ASSERTCMP(pos + len, ==, eof);
937 ret = __kernel_write(file, data, len, &pos);
950 cachefiles_io_error_obj(object,
951 "Write page to backing file failed");
953 _leave(" = -ENOBUFS [%d]", ret);
958 * detach a backing block from a page
959 * - cache withdrawal is prevented by the caller
961 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
962 __releases(&object->fscache.cookie->lock)
964 struct cachefiles_object *object;
965 struct cachefiles_cache *cache;
967 object = container_of(_object, struct cachefiles_object, fscache);
968 cache = container_of(object->fscache.cache,
969 struct cachefiles_cache, cache);
971 _enter("%p,{%lu}", object, page->index);
973 spin_unlock(&object->fscache.cookie->lock);