1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/sched/mm.h>
10 #include <linux/spinlock.h>
11 #include <linux/blkdev.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/prefetch.h>
16 #include <linux/fsverity.h>
17 #include "extent_io.h"
18 #include "extent-io-tree.h"
19 #include "extent_map.h"
21 #include "btrfs_inode.h"
28 #include "block-group.h"
29 #include "compression.h"
31 #include "accessors.h"
32 #include "file-item.h"
34 #include "dev-replace.h"
36 #include "transaction.h"
38 static struct kmem_cache *extent_buffer_cache;
40 #ifdef CONFIG_BTRFS_DEBUG
41 static inline void btrfs_leak_debug_add_eb(struct extent_buffer *eb)
43 struct btrfs_fs_info *fs_info = eb->fs_info;
46 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
47 list_add(&eb->leak_list, &fs_info->allocated_ebs);
48 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
51 static inline void btrfs_leak_debug_del_eb(struct extent_buffer *eb)
53 struct btrfs_fs_info *fs_info = eb->fs_info;
56 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
57 list_del(&eb->leak_list);
58 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
61 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
63 struct extent_buffer *eb;
67 * If we didn't get into open_ctree our allocated_ebs will not be
68 * initialized, so just skip this.
70 if (!fs_info->allocated_ebs.next)
73 WARN_ON(!list_empty(&fs_info->allocated_ebs));
74 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
75 while (!list_empty(&fs_info->allocated_ebs)) {
76 eb = list_first_entry(&fs_info->allocated_ebs,
77 struct extent_buffer, leak_list);
79 "BTRFS: buffer leak start %llu len %u refs %d bflags %lu owner %llu\n",
80 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
81 btrfs_header_owner(eb));
82 list_del(&eb->leak_list);
84 kmem_cache_free(extent_buffer_cache, eb);
86 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
89 #define btrfs_leak_debug_add_eb(eb) do {} while (0)
90 #define btrfs_leak_debug_del_eb(eb) do {} while (0)
94 * Structure to record info about the bio being assembled, and other info like
95 * how many bytes are there before stripe/ordered extent boundary.
97 struct btrfs_bio_ctrl {
98 struct btrfs_bio *bbio;
99 enum btrfs_compression_type compress_type;
100 u32 len_to_oe_boundary;
102 btrfs_bio_end_io_t end_io_func;
103 struct writeback_control *wbc;
106 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
108 struct btrfs_bio *bbio = bio_ctrl->bbio;
113 /* Caller should ensure the bio has at least some range added */
114 ASSERT(bbio->bio.bi_iter.bi_size);
116 if (btrfs_op(&bbio->bio) == BTRFS_MAP_READ &&
117 bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
118 btrfs_submit_compressed_read(bbio);
120 btrfs_submit_bio(bbio, 0);
122 /* The bbio is owned by the end_io handler now */
123 bio_ctrl->bbio = NULL;
127 * Submit or fail the current bio in the bio_ctrl structure.
129 static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret)
131 struct btrfs_bio *bbio = bio_ctrl->bbio;
138 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
139 /* The bio is owned by the end_io handler now */
140 bio_ctrl->bbio = NULL;
142 submit_one_bio(bio_ctrl);
146 int __init extent_buffer_init_cachep(void)
148 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
149 sizeof(struct extent_buffer), 0, 0,
151 if (!extent_buffer_cache)
157 void __cold extent_buffer_free_cachep(void)
160 * Make sure all delayed rcu free are flushed before we
164 kmem_cache_destroy(extent_buffer_cache);
167 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
169 unsigned long index = start >> PAGE_SHIFT;
170 unsigned long end_index = end >> PAGE_SHIFT;
173 while (index <= end_index) {
174 page = find_get_page(inode->i_mapping, index);
175 BUG_ON(!page); /* Pages should be in the extent_io_tree */
176 clear_page_dirty_for_io(page);
182 static void process_one_page(struct btrfs_fs_info *fs_info,
183 struct page *page, struct page *locked_page,
184 unsigned long page_ops, u64 start, u64 end)
186 struct folio *folio = page_folio(page);
189 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
190 len = end + 1 - start;
192 if (page_ops & PAGE_SET_ORDERED)
193 btrfs_folio_clamp_set_ordered(fs_info, folio, start, len);
194 if (page_ops & PAGE_START_WRITEBACK) {
195 btrfs_folio_clamp_clear_dirty(fs_info, folio, start, len);
196 btrfs_folio_clamp_set_writeback(fs_info, folio, start, len);
198 if (page_ops & PAGE_END_WRITEBACK)
199 btrfs_folio_clamp_clear_writeback(fs_info, folio, start, len);
201 if (page != locked_page && (page_ops & PAGE_UNLOCK))
202 btrfs_folio_end_writer_lock(fs_info, folio, start, len);
205 static void __process_pages_contig(struct address_space *mapping,
206 struct page *locked_page, u64 start, u64 end,
207 unsigned long page_ops)
209 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
210 pgoff_t start_index = start >> PAGE_SHIFT;
211 pgoff_t end_index = end >> PAGE_SHIFT;
212 pgoff_t index = start_index;
213 struct folio_batch fbatch;
216 folio_batch_init(&fbatch);
217 while (index <= end_index) {
220 found_folios = filemap_get_folios_contig(mapping, &index,
222 for (i = 0; i < found_folios; i++) {
223 struct folio *folio = fbatch.folios[i];
225 process_one_page(fs_info, &folio->page, locked_page,
226 page_ops, start, end);
228 folio_batch_release(&fbatch);
233 static noinline void __unlock_for_delalloc(struct inode *inode,
234 struct page *locked_page,
237 unsigned long index = start >> PAGE_SHIFT;
238 unsigned long end_index = end >> PAGE_SHIFT;
241 if (index == locked_page->index && end_index == index)
244 __process_pages_contig(inode->i_mapping, locked_page, start, end,
248 static noinline int lock_delalloc_pages(struct inode *inode,
249 struct page *locked_page,
253 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
254 struct address_space *mapping = inode->i_mapping;
255 pgoff_t start_index = start >> PAGE_SHIFT;
256 pgoff_t end_index = end >> PAGE_SHIFT;
257 pgoff_t index = start_index;
258 u64 processed_end = start;
259 struct folio_batch fbatch;
261 if (index == locked_page->index && index == end_index)
264 folio_batch_init(&fbatch);
265 while (index <= end_index) {
266 unsigned int found_folios, i;
268 found_folios = filemap_get_folios_contig(mapping, &index,
270 if (found_folios == 0)
273 for (i = 0; i < found_folios; i++) {
274 struct folio *folio = fbatch.folios[i];
275 struct page *page = folio_page(folio, 0);
276 u32 len = end + 1 - start;
278 if (page == locked_page)
281 if (btrfs_folio_start_writer_lock(fs_info, folio, start,
285 if (!PageDirty(page) || page->mapping != mapping) {
286 btrfs_folio_end_writer_lock(fs_info, folio, start,
291 processed_end = page_offset(page) + PAGE_SIZE - 1;
293 folio_batch_release(&fbatch);
299 folio_batch_release(&fbatch);
300 if (processed_end > start)
301 __unlock_for_delalloc(inode, locked_page, start, processed_end);
306 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
307 * more than @max_bytes.
309 * @start: The original start bytenr to search.
310 * Will store the extent range start bytenr.
311 * @end: The original end bytenr of the search range
312 * Will store the extent range end bytenr.
314 * Return true if we find a delalloc range which starts inside the original
315 * range, and @start/@end will store the delalloc range start/end.
317 * Return false if we can't find any delalloc range which starts inside the
318 * original range, and @start/@end will be the non-delalloc range start/end.
321 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
322 struct page *locked_page, u64 *start,
325 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
326 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
327 const u64 orig_start = *start;
328 const u64 orig_end = *end;
329 /* The sanity tests may not set a valid fs_info. */
330 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
334 struct extent_state *cached_state = NULL;
338 /* Caller should pass a valid @end to indicate the search range end */
339 ASSERT(orig_end > orig_start);
341 /* The range should at least cover part of the page */
342 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
343 orig_end <= page_offset(locked_page)));
345 /* step one, find a bunch of delalloc bytes starting at start */
346 delalloc_start = *start;
348 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
349 max_bytes, &cached_state);
350 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
351 *start = delalloc_start;
353 /* @delalloc_end can be -1, never go beyond @orig_end */
354 *end = min(delalloc_end, orig_end);
355 free_extent_state(cached_state);
360 * start comes from the offset of locked_page. We have to lock
361 * pages in order, so we can't process delalloc bytes before
364 if (delalloc_start < *start)
365 delalloc_start = *start;
368 * make sure to limit the number of pages we try to lock down
370 if (delalloc_end + 1 - delalloc_start > max_bytes)
371 delalloc_end = delalloc_start + max_bytes - 1;
373 /* step two, lock all the pages after the page that has start */
374 ret = lock_delalloc_pages(inode, locked_page,
375 delalloc_start, delalloc_end);
376 ASSERT(!ret || ret == -EAGAIN);
377 if (ret == -EAGAIN) {
378 /* some of the pages are gone, lets avoid looping by
379 * shortening the size of the delalloc range we're searching
381 free_extent_state(cached_state);
384 max_bytes = PAGE_SIZE;
393 /* step three, lock the state bits for the whole range */
394 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
396 /* then test to make sure it is all still delalloc */
397 ret = test_range_bit(tree, delalloc_start, delalloc_end,
398 EXTENT_DELALLOC, cached_state);
400 unlock_extent(tree, delalloc_start, delalloc_end,
402 __unlock_for_delalloc(inode, locked_page,
403 delalloc_start, delalloc_end);
407 free_extent_state(cached_state);
408 *start = delalloc_start;
414 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
415 struct page *locked_page,
416 u32 clear_bits, unsigned long page_ops)
418 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
420 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
421 start, end, page_ops);
424 static bool btrfs_verify_page(struct page *page, u64 start)
426 if (!fsverity_active(page->mapping->host) ||
427 PageUptodate(page) ||
428 start >= i_size_read(page->mapping->host))
430 return fsverity_verify_page(page);
433 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
435 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
436 struct folio *folio = page_folio(page);
438 ASSERT(page_offset(page) <= start &&
439 start + len <= page_offset(page) + PAGE_SIZE);
441 if (uptodate && btrfs_verify_page(page, start))
442 btrfs_folio_set_uptodate(fs_info, folio, start, len);
444 btrfs_folio_clear_uptodate(fs_info, folio, start, len);
446 if (!btrfs_is_subpage(fs_info, page->mapping))
449 btrfs_subpage_end_reader(fs_info, folio, start, len);
453 * After a write IO is done, we need to:
455 * - clear the uptodate bits on error
456 * - clear the writeback bits in the extent tree for the range
457 * - filio_end_writeback() if there is no more pending io for the folio
459 * Scheduling is not allowed, so the extent state tree is expected
460 * to have one and only one object corresponding to this IO.
462 static void end_bbio_data_write(struct btrfs_bio *bbio)
464 struct btrfs_fs_info *fs_info = bbio->fs_info;
465 struct bio *bio = &bbio->bio;
466 int error = blk_status_to_errno(bio->bi_status);
467 struct folio_iter fi;
468 const u32 sectorsize = fs_info->sectorsize;
470 ASSERT(!bio_flagged(bio, BIO_CLONED));
471 bio_for_each_folio_all(fi, bio) {
472 struct folio *folio = fi.folio;
473 u64 start = folio_pos(folio) + fi.offset;
476 /* Only order 0 (single page) folios are allowed for data. */
477 ASSERT(folio_order(folio) == 0);
479 /* Our read/write should always be sector aligned. */
480 if (!IS_ALIGNED(fi.offset, sectorsize))
482 "partial page write in btrfs with offset %zu and length %zu",
483 fi.offset, fi.length);
484 else if (!IS_ALIGNED(fi.length, sectorsize))
486 "incomplete page write with offset %zu and length %zu",
487 fi.offset, fi.length);
489 btrfs_finish_ordered_extent(bbio->ordered,
490 folio_page(folio, 0), start, len, !error);
492 mapping_set_error(folio->mapping, error);
493 btrfs_folio_clear_writeback(fs_info, folio, start, len);
500 * Record previously processed extent range
502 * For endio_readpage_release_extent() to handle a full extent range, reducing
503 * the extent io operations.
505 struct processed_extent {
506 struct btrfs_inode *inode;
507 /* Start of the range in @inode */
509 /* End of the range in @inode */
515 * Try to release processed extent range
517 * May not release the extent range right now if the current range is
518 * contiguous to processed extent.
520 * Will release processed extent when any of @inode, @uptodate, the range is
521 * no longer contiguous to the processed range.
523 * Passing @inode == NULL will force processed extent to be released.
525 static void endio_readpage_release_extent(struct processed_extent *processed,
526 struct btrfs_inode *inode, u64 start, u64 end,
529 struct extent_state *cached = NULL;
530 struct extent_io_tree *tree;
532 /* The first extent, initialize @processed */
533 if (!processed->inode)
537 * Contiguous to processed extent, just uptodate the end.
539 * Several things to notice:
541 * - bio can be merged as long as on-disk bytenr is contiguous
542 * This means we can have page belonging to other inodes, thus need to
543 * check if the inode still matches.
544 * - bvec can contain range beyond current page for multi-page bvec
545 * Thus we need to do processed->end + 1 >= start check
547 if (processed->inode == inode && processed->uptodate == uptodate &&
548 processed->end + 1 >= start && end >= processed->end) {
549 processed->end = end;
553 tree = &processed->inode->io_tree;
555 * Now we don't have range contiguous to the processed range, release
556 * the processed range now.
558 unlock_extent(tree, processed->start, processed->end, &cached);
561 /* Update processed to current range */
562 processed->inode = inode;
563 processed->start = start;
564 processed->end = end;
565 processed->uptodate = uptodate;
568 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
570 struct folio *folio = page_folio(page);
572 ASSERT(folio_test_locked(folio));
573 if (!btrfs_is_subpage(fs_info, folio->mapping))
576 ASSERT(folio_test_private(folio));
577 btrfs_subpage_start_reader(fs_info, folio, page_offset(page), PAGE_SIZE);
581 * After a data read IO is done, we need to:
583 * - clear the uptodate bits on error
584 * - set the uptodate bits if things worked
585 * - set the folio up to date if all extents in the tree are uptodate
586 * - clear the lock bit in the extent tree
587 * - unlock the folio if there are no other extents locked for it
589 * Scheduling is not allowed, so the extent state tree is expected
590 * to have one and only one object corresponding to this IO.
592 static void end_bbio_data_read(struct btrfs_bio *bbio)
594 struct btrfs_fs_info *fs_info = bbio->fs_info;
595 struct bio *bio = &bbio->bio;
596 struct processed_extent processed = { 0 };
597 struct folio_iter fi;
598 const u32 sectorsize = fs_info->sectorsize;
600 ASSERT(!bio_flagged(bio, BIO_CLONED));
601 bio_for_each_folio_all(fi, &bbio->bio) {
602 bool uptodate = !bio->bi_status;
603 struct folio *folio = fi.folio;
604 struct inode *inode = folio->mapping->host;
609 /* For now only order 0 folios are supported for data. */
610 ASSERT(folio_order(folio) == 0);
612 "%s: bi_sector=%llu, err=%d, mirror=%u",
613 __func__, bio->bi_iter.bi_sector, bio->bi_status,
617 * We always issue full-sector reads, but if some block in a
618 * folio fails to read, blk_update_request() will advance
619 * bv_offset and adjust bv_len to compensate. Print a warning
620 * for unaligned offsets, and an error if they don't add up to
623 if (!IS_ALIGNED(fi.offset, sectorsize))
625 "partial page read in btrfs with offset %zu and length %zu",
626 fi.offset, fi.length);
627 else if (!IS_ALIGNED(fi.offset + fi.length, sectorsize))
629 "incomplete page read with offset %zu and length %zu",
630 fi.offset, fi.length);
632 start = folio_pos(folio) + fi.offset;
633 end = start + fi.length - 1;
636 if (likely(uptodate)) {
637 loff_t i_size = i_size_read(inode);
638 pgoff_t end_index = i_size >> folio_shift(folio);
641 * Zero out the remaining part if this range straddles
644 * Here we should only zero the range inside the folio,
645 * not touch anything else.
647 * NOTE: i_size is exclusive while end is inclusive.
649 if (folio_index(folio) == end_index && i_size <= end) {
650 u32 zero_start = max(offset_in_folio(folio, i_size),
651 offset_in_folio(folio, start));
652 u32 zero_len = offset_in_folio(folio, end) + 1 -
655 folio_zero_range(folio, zero_start, zero_len);
659 /* Update page status and unlock. */
660 end_page_read(folio_page(folio, 0), uptodate, start, len);
661 endio_readpage_release_extent(&processed, BTRFS_I(inode),
662 start, end, uptodate);
664 /* Release the last extent */
665 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
670 * Populate every free slot in a provided array with pages.
672 * @nr_pages: number of pages to allocate
673 * @page_array: the array to fill with pages; any existing non-null entries in
674 * the array will be skipped
675 * @extra_gfp: the extra GFP flags for the allocation.
677 * Return: 0 if all pages were able to be allocated;
678 * -ENOMEM otherwise, the partially allocated pages would be freed and
679 * the array slots zeroed
681 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
684 unsigned int allocated;
686 for (allocated = 0; allocated < nr_pages;) {
687 unsigned int last = allocated;
689 allocated = alloc_pages_bulk_array(GFP_NOFS | extra_gfp,
690 nr_pages, page_array);
692 if (allocated == nr_pages)
696 * During this iteration, no page could be allocated, even
697 * though alloc_pages_bulk_array() falls back to alloc_page()
698 * if it could not bulk-allocate. So we must be out of memory.
700 if (allocated == last) {
701 for (int i = 0; i < allocated; i++) {
702 __free_page(page_array[i]);
703 page_array[i] = NULL;
708 memalloc_retry_wait(GFP_NOFS);
714 * Populate needed folios for the extent buffer.
716 * For now, the folios populated are always in order 0 (aka, single page).
718 static int alloc_eb_folio_array(struct extent_buffer *eb, gfp_t extra_gfp)
720 struct page *page_array[INLINE_EXTENT_BUFFER_PAGES] = { 0 };
721 int num_pages = num_extent_pages(eb);
724 ret = btrfs_alloc_page_array(num_pages, page_array, extra_gfp);
728 for (int i = 0; i < num_pages; i++)
729 eb->folios[i] = page_folio(page_array[i]);
730 eb->folio_size = PAGE_SIZE;
731 eb->folio_shift = PAGE_SHIFT;
735 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
736 struct page *page, u64 disk_bytenr,
737 unsigned int pg_offset)
739 struct bio *bio = &bio_ctrl->bbio->bio;
740 struct bio_vec *bvec = bio_last_bvec_all(bio);
741 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
743 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
745 * For compression, all IO should have its logical bytenr set
746 * to the starting bytenr of the compressed extent.
748 return bio->bi_iter.bi_sector == sector;
752 * The contig check requires the following conditions to be met:
754 * 1) The pages are belonging to the same inode
755 * This is implied by the call chain.
757 * 2) The range has adjacent logical bytenr
759 * 3) The range has adjacent file offset
760 * This is required for the usage of btrfs_bio->file_offset.
762 return bio_end_sector(bio) == sector &&
763 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
764 page_offset(page) + pg_offset;
767 static void alloc_new_bio(struct btrfs_inode *inode,
768 struct btrfs_bio_ctrl *bio_ctrl,
769 u64 disk_bytenr, u64 file_offset)
771 struct btrfs_fs_info *fs_info = inode->root->fs_info;
772 struct btrfs_bio *bbio;
774 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
775 bio_ctrl->end_io_func, NULL);
776 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
778 bbio->file_offset = file_offset;
779 bio_ctrl->bbio = bbio;
780 bio_ctrl->len_to_oe_boundary = U32_MAX;
782 /* Limit data write bios to the ordered boundary. */
784 struct btrfs_ordered_extent *ordered;
786 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
788 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
789 ordered->file_offset +
790 ordered->disk_num_bytes - file_offset);
791 bbio->ordered = ordered;
795 * Pick the last added device to support cgroup writeback. For
796 * multi-device file systems this means blk-cgroup policies have
797 * to always be set on the last added/replaced device.
798 * This is a bit odd but has been like that for a long time.
800 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
801 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
806 * @disk_bytenr: logical bytenr where the write will be
807 * @page: page to add to the bio
808 * @size: portion of page that we want to write to
809 * @pg_offset: offset of the new bio or to check whether we are adding
810 * a contiguous page to the previous one
812 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
813 * new one in @bio_ctrl->bbio.
814 * The mirror number for this IO should already be initizlied in
815 * @bio_ctrl->mirror_num.
817 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
818 u64 disk_bytenr, struct page *page,
819 size_t size, unsigned long pg_offset)
821 struct btrfs_inode *inode = page_to_inode(page);
823 ASSERT(pg_offset + size <= PAGE_SIZE);
824 ASSERT(bio_ctrl->end_io_func);
826 if (bio_ctrl->bbio &&
827 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
828 submit_one_bio(bio_ctrl);
833 /* Allocate new bio if needed */
834 if (!bio_ctrl->bbio) {
835 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
836 page_offset(page) + pg_offset);
839 /* Cap to the current ordered extent boundary if there is one. */
840 if (len > bio_ctrl->len_to_oe_boundary) {
841 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
842 ASSERT(is_data_inode(&inode->vfs_inode));
843 len = bio_ctrl->len_to_oe_boundary;
846 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
847 /* bio full: move on to a new one */
848 submit_one_bio(bio_ctrl);
853 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
860 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
861 * sector aligned. alloc_new_bio() then sets it to the end of
862 * our ordered extent for writes into zoned devices.
864 * When len_to_oe_boundary is tracking an ordered extent, we
865 * trust the ordered extent code to align things properly, and
866 * the check above to cap our write to the ordered extent
867 * boundary is correct.
869 * When len_to_oe_boundary is U32_MAX, the cap above would
870 * result in a 4095 byte IO for the last page right before
871 * we hit the bio limit of UINT_MAX. bio_add_page() has all
872 * the checks required to make sure we don't overflow the bio,
873 * and we should just ignore len_to_oe_boundary completely
874 * unless we're using it to track an ordered extent.
876 * It's pretty hard to make a bio sized U32_MAX, but it can
877 * happen when the page cache is able to feed us contiguous
878 * pages for large extents.
880 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
881 bio_ctrl->len_to_oe_boundary -= len;
883 /* Ordered extent boundary: move on to a new bio. */
884 if (bio_ctrl->len_to_oe_boundary == 0)
885 submit_one_bio(bio_ctrl);
889 static int attach_extent_buffer_folio(struct extent_buffer *eb,
891 struct btrfs_subpage *prealloc)
893 struct btrfs_fs_info *fs_info = eb->fs_info;
897 * If the page is mapped to btree inode, we should hold the private
898 * lock to prevent race.
899 * For cloned or dummy extent buffers, their pages are not mapped and
900 * will not race with any other ebs.
903 lockdep_assert_held(&folio->mapping->i_private_lock);
905 if (fs_info->nodesize >= PAGE_SIZE) {
906 if (!folio_test_private(folio))
907 folio_attach_private(folio, eb);
909 WARN_ON(folio_get_private(folio) != eb);
913 /* Already mapped, just free prealloc */
914 if (folio_test_private(folio)) {
915 btrfs_free_subpage(prealloc);
920 /* Has preallocated memory for subpage */
921 folio_attach_private(folio, prealloc);
923 /* Do new allocation to attach subpage */
924 ret = btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_METADATA);
928 int set_page_extent_mapped(struct page *page)
930 return set_folio_extent_mapped(page_folio(page));
933 int set_folio_extent_mapped(struct folio *folio)
935 struct btrfs_fs_info *fs_info;
937 ASSERT(folio->mapping);
939 if (folio_test_private(folio))
942 fs_info = folio_to_fs_info(folio);
944 if (btrfs_is_subpage(fs_info, folio->mapping))
945 return btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_DATA);
947 folio_attach_private(folio, (void *)EXTENT_FOLIO_PRIVATE);
951 void clear_page_extent_mapped(struct page *page)
953 struct folio *folio = page_folio(page);
954 struct btrfs_fs_info *fs_info;
956 ASSERT(page->mapping);
958 if (!folio_test_private(folio))
961 fs_info = page_to_fs_info(page);
962 if (btrfs_is_subpage(fs_info, page->mapping))
963 return btrfs_detach_subpage(fs_info, folio);
965 folio_detach_private(folio);
968 static struct extent_map *__get_extent_map(struct inode *inode, struct page *page,
969 u64 start, u64 len, struct extent_map **em_cached)
971 struct extent_map *em;
977 if (extent_map_in_tree(em) && start >= em->start &&
978 start < extent_map_end(em)) {
979 refcount_inc(&em->refs);
987 em = btrfs_get_extent(BTRFS_I(inode), page, start, len);
990 refcount_inc(&em->refs);
996 * basic readpage implementation. Locked extent state structs are inserted
997 * into the tree that are removed when the IO is done (by the end_io
999 * XXX JDM: This needs looking at to ensure proper page locking
1000 * return 0 on success, otherwise return error
1002 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
1003 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
1005 struct inode *inode = page->mapping->host;
1006 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1007 u64 start = page_offset(page);
1008 const u64 end = start + PAGE_SIZE - 1;
1011 u64 last_byte = i_size_read(inode);
1013 struct extent_map *em;
1015 size_t pg_offset = 0;
1017 size_t blocksize = fs_info->sectorsize;
1018 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1020 ret = set_page_extent_mapped(page);
1022 unlock_extent(tree, start, end, NULL);
1027 if (page->index == last_byte >> PAGE_SHIFT) {
1028 size_t zero_offset = offset_in_page(last_byte);
1031 iosize = PAGE_SIZE - zero_offset;
1032 memzero_page(page, zero_offset, iosize);
1035 bio_ctrl->end_io_func = end_bbio_data_read;
1036 begin_page_read(fs_info, page);
1037 while (cur <= end) {
1038 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1039 bool force_bio_submit = false;
1042 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1043 if (cur >= last_byte) {
1044 iosize = PAGE_SIZE - pg_offset;
1045 memzero_page(page, pg_offset, iosize);
1046 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1047 end_page_read(page, true, cur, iosize);
1050 em = __get_extent_map(inode, page, cur, end - cur + 1, em_cached);
1052 unlock_extent(tree, cur, end, NULL);
1053 end_page_read(page, false, cur, end + 1 - cur);
1056 extent_offset = cur - em->start;
1057 BUG_ON(extent_map_end(em) <= cur);
1060 compress_type = extent_map_compression(em);
1062 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1063 iosize = ALIGN(iosize, blocksize);
1064 if (compress_type != BTRFS_COMPRESS_NONE)
1065 disk_bytenr = em->block_start;
1067 disk_bytenr = em->block_start + extent_offset;
1068 block_start = em->block_start;
1069 if (em->flags & EXTENT_FLAG_PREALLOC)
1070 block_start = EXTENT_MAP_HOLE;
1073 * If we have a file range that points to a compressed extent
1074 * and it's followed by a consecutive file range that points
1075 * to the same compressed extent (possibly with a different
1076 * offset and/or length, so it either points to the whole extent
1077 * or only part of it), we must make sure we do not submit a
1078 * single bio to populate the pages for the 2 ranges because
1079 * this makes the compressed extent read zero out the pages
1080 * belonging to the 2nd range. Imagine the following scenario:
1083 * [0 - 8K] [8K - 24K]
1086 * points to extent X, points to extent X,
1087 * offset 4K, length of 8K offset 0, length 16K
1089 * [extent X, compressed length = 4K uncompressed length = 16K]
1091 * If the bio to read the compressed extent covers both ranges,
1092 * it will decompress extent X into the pages belonging to the
1093 * first range and then it will stop, zeroing out the remaining
1094 * pages that belong to the other range that points to extent X.
1095 * So here we make sure we submit 2 bios, one for the first
1096 * range and another one for the third range. Both will target
1097 * the same physical extent from disk, but we can't currently
1098 * make the compressed bio endio callback populate the pages
1099 * for both ranges because each compressed bio is tightly
1100 * coupled with a single extent map, and each range can have
1101 * an extent map with a different offset value relative to the
1102 * uncompressed data of our extent and different lengths. This
1103 * is a corner case so we prioritize correctness over
1104 * non-optimal behavior (submitting 2 bios for the same extent).
1106 if (compress_type != BTRFS_COMPRESS_NONE &&
1107 prev_em_start && *prev_em_start != (u64)-1 &&
1108 *prev_em_start != em->start)
1109 force_bio_submit = true;
1112 *prev_em_start = em->start;
1114 free_extent_map(em);
1117 /* we've found a hole, just zero and go on */
1118 if (block_start == EXTENT_MAP_HOLE) {
1119 memzero_page(page, pg_offset, iosize);
1121 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1122 end_page_read(page, true, cur, iosize);
1124 pg_offset += iosize;
1127 /* the get_extent function already copied into the page */
1128 if (block_start == EXTENT_MAP_INLINE) {
1129 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1130 end_page_read(page, true, cur, iosize);
1132 pg_offset += iosize;
1136 if (bio_ctrl->compress_type != compress_type) {
1137 submit_one_bio(bio_ctrl);
1138 bio_ctrl->compress_type = compress_type;
1141 if (force_bio_submit)
1142 submit_one_bio(bio_ctrl);
1143 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1146 pg_offset += iosize;
1152 int btrfs_read_folio(struct file *file, struct folio *folio)
1154 struct page *page = &folio->page;
1155 struct btrfs_inode *inode = page_to_inode(page);
1156 u64 start = page_offset(page);
1157 u64 end = start + PAGE_SIZE - 1;
1158 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1159 struct extent_map *em_cached = NULL;
1162 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1164 ret = btrfs_do_readpage(page, &em_cached, &bio_ctrl, NULL);
1165 free_extent_map(em_cached);
1168 * If btrfs_do_readpage() failed we will want to submit the assembled
1169 * bio to do the cleanup.
1171 submit_one_bio(&bio_ctrl);
1175 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1177 struct extent_map **em_cached,
1178 struct btrfs_bio_ctrl *bio_ctrl,
1181 struct btrfs_inode *inode = page_to_inode(pages[0]);
1186 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1188 for (index = 0; index < nr_pages; index++) {
1189 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1191 put_page(pages[index]);
1196 * helper for __extent_writepage, doing all of the delayed allocation setup.
1198 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1199 * to write the page (copy into inline extent). In this case the IO has
1200 * been started and the page is already unlocked.
1202 * This returns 0 if all went well (page still locked)
1203 * This returns < 0 if there were errors (page still locked)
1205 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1206 struct page *page, struct writeback_control *wbc)
1208 const u64 page_start = page_offset(page);
1209 const u64 page_end = page_start + PAGE_SIZE - 1;
1210 u64 delalloc_start = page_start;
1211 u64 delalloc_end = page_end;
1212 u64 delalloc_to_write = 0;
1215 while (delalloc_start < page_end) {
1216 delalloc_end = page_end;
1217 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1218 &delalloc_start, &delalloc_end)) {
1219 delalloc_start = delalloc_end + 1;
1223 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1228 delalloc_start = delalloc_end + 1;
1232 * delalloc_end is already one less than the total length, so
1233 * we don't subtract one from PAGE_SIZE
1235 delalloc_to_write +=
1236 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1239 * If btrfs_run_dealloc_range() already started I/O and unlocked
1240 * the pages, we just need to account for them here.
1243 wbc->nr_to_write -= delalloc_to_write;
1247 if (wbc->nr_to_write < delalloc_to_write) {
1250 if (delalloc_to_write < thresh * 2)
1251 thresh = delalloc_to_write;
1252 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1260 * Find the first byte we need to write.
1262 * For subpage, one page can contain several sectors, and
1263 * __extent_writepage_io() will just grab all extent maps in the page
1264 * range and try to submit all non-inline/non-compressed extents.
1266 * This is a big problem for subpage, we shouldn't re-submit already written
1268 * This function will lookup subpage dirty bit to find which range we really
1271 * Return the next dirty range in [@start, @end).
1272 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1274 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1275 struct page *page, u64 *start, u64 *end)
1277 struct folio *folio = page_folio(page);
1278 struct btrfs_subpage *subpage = folio_get_private(folio);
1279 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1280 u64 orig_start = *start;
1281 /* Declare as unsigned long so we can use bitmap ops */
1282 unsigned long flags;
1283 int range_start_bit;
1287 * For regular sector size == page size case, since one page only
1288 * contains one sector, we return the page offset directly.
1290 if (!btrfs_is_subpage(fs_info, page->mapping)) {
1291 *start = page_offset(page);
1292 *end = page_offset(page) + PAGE_SIZE;
1296 range_start_bit = spi->dirty_offset +
1297 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1299 /* We should have the page locked, but just in case */
1300 spin_lock_irqsave(&subpage->lock, flags);
1301 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1302 spi->dirty_offset + spi->bitmap_nr_bits);
1303 spin_unlock_irqrestore(&subpage->lock, flags);
1305 range_start_bit -= spi->dirty_offset;
1306 range_end_bit -= spi->dirty_offset;
1308 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1309 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1313 * helper for __extent_writepage. This calls the writepage start hooks,
1314 * and does the loop to map the page into extents and bios.
1316 * We return 1 if the IO is started and the page is unlocked,
1317 * 0 if all went well (page still locked)
1318 * < 0 if there were errors (page still locked)
1320 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1322 struct btrfs_bio_ctrl *bio_ctrl,
1326 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1327 u64 cur = page_offset(page);
1328 u64 end = cur + PAGE_SIZE - 1;
1331 struct extent_map *em;
1335 ret = btrfs_writepage_cow_fixup(page);
1337 /* Fixup worker will requeue */
1338 redirty_page_for_writepage(bio_ctrl->wbc, page);
1343 bio_ctrl->end_io_func = end_bbio_data_write;
1344 while (cur <= end) {
1345 u32 len = end - cur + 1;
1348 u64 dirty_range_start = cur;
1349 u64 dirty_range_end;
1352 if (cur >= i_size) {
1353 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1356 * This range is beyond i_size, thus we don't need to
1357 * bother writing back.
1358 * But we still need to clear the dirty subpage bit, or
1359 * the next time the page gets dirtied, we will try to
1360 * writeback the sectors with subpage dirty bits,
1361 * causing writeback without ordered extent.
1363 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, len);
1367 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1369 if (cur < dirty_range_start) {
1370 cur = dirty_range_start;
1374 em = btrfs_get_extent(inode, NULL, cur, len);
1376 ret = PTR_ERR_OR_ZERO(em);
1380 extent_offset = cur - em->start;
1381 em_end = extent_map_end(em);
1382 ASSERT(cur <= em_end);
1384 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1385 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1387 block_start = em->block_start;
1388 disk_bytenr = em->block_start + extent_offset;
1390 ASSERT(!extent_map_is_compressed(em));
1391 ASSERT(block_start != EXTENT_MAP_HOLE);
1392 ASSERT(block_start != EXTENT_MAP_INLINE);
1395 * Note that em_end from extent_map_end() and dirty_range_end from
1396 * find_next_dirty_byte() are all exclusive
1398 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1399 free_extent_map(em);
1402 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1403 if (!PageWriteback(page)) {
1404 btrfs_err(inode->root->fs_info,
1405 "page %lu not writeback, cur %llu end %llu",
1406 page->index, cur, end);
1410 * Although the PageDirty bit is cleared before entering this
1411 * function, subpage dirty bit is not cleared.
1412 * So clear subpage dirty bit here so next time we won't submit
1413 * page for range already written to disk.
1415 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, iosize);
1417 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1418 cur - page_offset(page));
1423 btrfs_folio_assert_not_dirty(fs_info, page_folio(page));
1429 * If we finish without problem, we should not only clear page dirty,
1430 * but also empty subpage dirty bits
1437 * the writepage semantics are similar to regular writepage. extent
1438 * records are inserted to lock ranges in the tree, and as dirty areas
1439 * are found, they are marked writeback. Then the lock bits are removed
1440 * and the end_io handler clears the writeback ranges
1442 * Return 0 if everything goes well.
1443 * Return <0 for error.
1445 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1447 struct folio *folio = page_folio(page);
1448 struct inode *inode = page->mapping->host;
1449 const u64 page_start = page_offset(page);
1453 loff_t i_size = i_size_read(inode);
1454 unsigned long end_index = i_size >> PAGE_SHIFT;
1456 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1458 WARN_ON(!PageLocked(page));
1460 pg_offset = offset_in_page(i_size);
1461 if (page->index > end_index ||
1462 (page->index == end_index && !pg_offset)) {
1463 folio_invalidate(folio, 0, folio_size(folio));
1464 folio_unlock(folio);
1468 if (page->index == end_index)
1469 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1471 ret = set_page_extent_mapped(page);
1475 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1481 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1485 bio_ctrl->wbc->nr_to_write--;
1489 /* make sure the mapping tag for page dirty gets cleared */
1490 set_page_writeback(page);
1491 end_page_writeback(page);
1494 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1496 mapping_set_error(page->mapping, ret);
1503 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1505 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1506 TASK_UNINTERRUPTIBLE);
1510 * Lock extent buffer status and pages for writeback.
1512 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1513 * extent buffer is not dirty)
1514 * Return %true is the extent buffer is submitted to bio.
1516 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1517 struct writeback_control *wbc)
1519 struct btrfs_fs_info *fs_info = eb->fs_info;
1522 btrfs_tree_lock(eb);
1523 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1524 btrfs_tree_unlock(eb);
1525 if (wbc->sync_mode != WB_SYNC_ALL)
1527 wait_on_extent_buffer_writeback(eb);
1528 btrfs_tree_lock(eb);
1532 * We need to do this to prevent races in people who check if the eb is
1533 * under IO since we can end up having no IO bits set for a short period
1536 spin_lock(&eb->refs_lock);
1537 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1538 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1539 spin_unlock(&eb->refs_lock);
1540 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1541 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1543 fs_info->dirty_metadata_batch);
1546 spin_unlock(&eb->refs_lock);
1548 btrfs_tree_unlock(eb);
1552 static void set_btree_ioerr(struct extent_buffer *eb)
1554 struct btrfs_fs_info *fs_info = eb->fs_info;
1556 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1559 * A read may stumble upon this buffer later, make sure that it gets an
1560 * error and knows there was an error.
1562 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1565 * We need to set the mapping with the io error as well because a write
1566 * error will flip the file system readonly, and then syncfs() will
1567 * return a 0 because we are readonly if we don't modify the err seq for
1570 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1573 * If writeback for a btree extent that doesn't belong to a log tree
1574 * failed, increment the counter transaction->eb_write_errors.
1575 * We do this because while the transaction is running and before it's
1576 * committing (when we call filemap_fdata[write|wait]_range against
1577 * the btree inode), we might have
1578 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1579 * returns an error or an error happens during writeback, when we're
1580 * committing the transaction we wouldn't know about it, since the pages
1581 * can be no longer dirty nor marked anymore for writeback (if a
1582 * subsequent modification to the extent buffer didn't happen before the
1583 * transaction commit), which makes filemap_fdata[write|wait]_range not
1584 * able to find the pages tagged with SetPageError at transaction
1585 * commit time. So if this happens we must abort the transaction,
1586 * otherwise we commit a super block with btree roots that point to
1587 * btree nodes/leafs whose content on disk is invalid - either garbage
1588 * or the content of some node/leaf from a past generation that got
1589 * cowed or deleted and is no longer valid.
1591 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1592 * not be enough - we need to distinguish between log tree extents vs
1593 * non-log tree extents, and the next filemap_fdatawait_range() call
1594 * will catch and clear such errors in the mapping - and that call might
1595 * be from a log sync and not from a transaction commit. Also, checking
1596 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1597 * not done and would not be reliable - the eb might have been released
1598 * from memory and reading it back again means that flag would not be
1599 * set (since it's a runtime flag, not persisted on disk).
1601 * Using the flags below in the btree inode also makes us achieve the
1602 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1603 * writeback for all dirty pages and before filemap_fdatawait_range()
1604 * is called, the writeback for all dirty pages had already finished
1605 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1606 * filemap_fdatawait_range() would return success, as it could not know
1607 * that writeback errors happened (the pages were no longer tagged for
1610 switch (eb->log_index) {
1612 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1615 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1618 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1621 BUG(); /* unexpected, logic error */
1626 * The endio specific version which won't touch any unsafe spinlock in endio
1629 static struct extent_buffer *find_extent_buffer_nolock(
1630 struct btrfs_fs_info *fs_info, u64 start)
1632 struct extent_buffer *eb;
1635 eb = radix_tree_lookup(&fs_info->buffer_radix,
1636 start >> fs_info->sectorsize_bits);
1637 if (eb && atomic_inc_not_zero(&eb->refs)) {
1645 static void end_bbio_meta_write(struct btrfs_bio *bbio)
1647 struct extent_buffer *eb = bbio->private;
1648 struct btrfs_fs_info *fs_info = eb->fs_info;
1649 bool uptodate = !bbio->bio.bi_status;
1650 struct folio_iter fi;
1654 set_btree_ioerr(eb);
1656 bio_for_each_folio_all(fi, &bbio->bio) {
1657 u64 start = eb->start + bio_offset;
1658 struct folio *folio = fi.folio;
1659 u32 len = fi.length;
1661 btrfs_folio_clear_writeback(fs_info, folio, start, len);
1665 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1666 smp_mb__after_atomic();
1667 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1669 bio_put(&bbio->bio);
1672 static void prepare_eb_write(struct extent_buffer *eb)
1675 unsigned long start;
1678 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1680 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1681 nritems = btrfs_header_nritems(eb);
1682 if (btrfs_header_level(eb) > 0) {
1683 end = btrfs_node_key_ptr_offset(eb, nritems);
1684 memzero_extent_buffer(eb, end, eb->len - end);
1688 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1690 start = btrfs_item_nr_offset(eb, nritems);
1691 end = btrfs_item_nr_offset(eb, 0);
1693 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1695 end += btrfs_item_offset(eb, nritems - 1);
1696 memzero_extent_buffer(eb, start, end - start);
1700 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1701 struct writeback_control *wbc)
1703 struct btrfs_fs_info *fs_info = eb->fs_info;
1704 struct btrfs_bio *bbio;
1706 prepare_eb_write(eb);
1708 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1709 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1710 eb->fs_info, end_bbio_meta_write, eb);
1711 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1712 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1713 wbc_init_bio(wbc, &bbio->bio);
1714 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1715 bbio->file_offset = eb->start;
1716 if (fs_info->nodesize < PAGE_SIZE) {
1717 struct folio *folio = eb->folios[0];
1721 btrfs_subpage_set_writeback(fs_info, folio, eb->start, eb->len);
1722 if (btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start,
1724 folio_clear_dirty_for_io(folio);
1727 ret = bio_add_folio(&bbio->bio, folio, eb->len,
1728 eb->start - folio_pos(folio));
1730 wbc_account_cgroup_owner(wbc, folio_page(folio, 0), eb->len);
1731 folio_unlock(folio);
1733 int num_folios = num_extent_folios(eb);
1735 for (int i = 0; i < num_folios; i++) {
1736 struct folio *folio = eb->folios[i];
1740 folio_clear_dirty_for_io(folio);
1741 folio_start_writeback(folio);
1742 ret = bio_add_folio(&bbio->bio, folio, eb->folio_size, 0);
1744 wbc_account_cgroup_owner(wbc, folio_page(folio, 0),
1746 wbc->nr_to_write -= folio_nr_pages(folio);
1747 folio_unlock(folio);
1750 btrfs_submit_bio(bbio, 0);
1754 * Submit one subpage btree page.
1756 * The main difference to submit_eb_page() is:
1758 * For subpage, we don't rely on page locking at all.
1761 * We only flush bio if we may be unable to fit current extent buffers into
1764 * Return >=0 for the number of submitted extent buffers.
1765 * Return <0 for fatal error.
1767 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1769 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
1770 struct folio *folio = page_folio(page);
1772 u64 page_start = page_offset(page);
1774 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1776 /* Lock and write each dirty extent buffers in the range */
1777 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1778 struct btrfs_subpage *subpage = folio_get_private(folio);
1779 struct extent_buffer *eb;
1780 unsigned long flags;
1784 * Take private lock to ensure the subpage won't be detached
1787 spin_lock(&page->mapping->i_private_lock);
1788 if (!folio_test_private(folio)) {
1789 spin_unlock(&page->mapping->i_private_lock);
1792 spin_lock_irqsave(&subpage->lock, flags);
1793 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1794 subpage->bitmaps)) {
1795 spin_unlock_irqrestore(&subpage->lock, flags);
1796 spin_unlock(&page->mapping->i_private_lock);
1801 start = page_start + bit_start * fs_info->sectorsize;
1802 bit_start += sectors_per_node;
1805 * Here we just want to grab the eb without touching extra
1806 * spin locks, so call find_extent_buffer_nolock().
1808 eb = find_extent_buffer_nolock(fs_info, start);
1809 spin_unlock_irqrestore(&subpage->lock, flags);
1810 spin_unlock(&page->mapping->i_private_lock);
1813 * The eb has already reached 0 refs thus find_extent_buffer()
1814 * doesn't return it. We don't need to write back such eb
1820 if (lock_extent_buffer_for_io(eb, wbc)) {
1821 write_one_eb(eb, wbc);
1824 free_extent_buffer(eb);
1830 * Submit all page(s) of one extent buffer.
1832 * @page: the page of one extent buffer
1833 * @eb_context: to determine if we need to submit this page, if current page
1834 * belongs to this eb, we don't need to submit
1836 * The caller should pass each page in their bytenr order, and here we use
1837 * @eb_context to determine if we have submitted pages of one extent buffer.
1839 * If we have, we just skip until we hit a new page that doesn't belong to
1840 * current @eb_context.
1842 * If not, we submit all the page(s) of the extent buffer.
1844 * Return >0 if we have submitted the extent buffer successfully.
1845 * Return 0 if we don't need to submit the page, as it's already submitted by
1847 * Return <0 for fatal error.
1849 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1851 struct writeback_control *wbc = ctx->wbc;
1852 struct address_space *mapping = page->mapping;
1853 struct folio *folio = page_folio(page);
1854 struct extent_buffer *eb;
1857 if (!folio_test_private(folio))
1860 if (page_to_fs_info(page)->nodesize < PAGE_SIZE)
1861 return submit_eb_subpage(page, wbc);
1863 spin_lock(&mapping->i_private_lock);
1864 if (!folio_test_private(folio)) {
1865 spin_unlock(&mapping->i_private_lock);
1869 eb = folio_get_private(folio);
1872 * Shouldn't happen and normally this would be a BUG_ON but no point
1873 * crashing the machine for something we can survive anyway.
1876 spin_unlock(&mapping->i_private_lock);
1880 if (eb == ctx->eb) {
1881 spin_unlock(&mapping->i_private_lock);
1884 ret = atomic_inc_not_zero(&eb->refs);
1885 spin_unlock(&mapping->i_private_lock);
1891 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1895 free_extent_buffer(eb);
1899 if (!lock_extent_buffer_for_io(eb, wbc)) {
1900 free_extent_buffer(eb);
1903 /* Implies write in zoned mode. */
1904 if (ctx->zoned_bg) {
1905 /* Mark the last eb in the block group. */
1906 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1907 ctx->zoned_bg->meta_write_pointer += eb->len;
1909 write_one_eb(eb, wbc);
1910 free_extent_buffer(eb);
1914 int btree_write_cache_pages(struct address_space *mapping,
1915 struct writeback_control *wbc)
1917 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1918 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
1921 int nr_to_write_done = 0;
1922 struct folio_batch fbatch;
1923 unsigned int nr_folios;
1925 pgoff_t end; /* Inclusive */
1929 folio_batch_init(&fbatch);
1930 if (wbc->range_cyclic) {
1931 index = mapping->writeback_index; /* Start from prev offset */
1934 * Start from the beginning does not need to cycle over the
1935 * range, mark it as scanned.
1937 scanned = (index == 0);
1939 index = wbc->range_start >> PAGE_SHIFT;
1940 end = wbc->range_end >> PAGE_SHIFT;
1943 if (wbc->sync_mode == WB_SYNC_ALL)
1944 tag = PAGECACHE_TAG_TOWRITE;
1946 tag = PAGECACHE_TAG_DIRTY;
1947 btrfs_zoned_meta_io_lock(fs_info);
1949 if (wbc->sync_mode == WB_SYNC_ALL)
1950 tag_pages_for_writeback(mapping, index, end);
1951 while (!done && !nr_to_write_done && (index <= end) &&
1952 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1956 for (i = 0; i < nr_folios; i++) {
1957 struct folio *folio = fbatch.folios[i];
1959 ret = submit_eb_page(&folio->page, &ctx);
1968 * the filesystem may choose to bump up nr_to_write.
1969 * We have to make sure to honor the new nr_to_write
1972 nr_to_write_done = wbc->nr_to_write <= 0;
1974 folio_batch_release(&fbatch);
1977 if (!scanned && !done) {
1979 * We hit the last page and there is more work to be done: wrap
1980 * back to the start of the file
1987 * If something went wrong, don't allow any metadata write bio to be
1990 * This would prevent use-after-free if we had dirty pages not
1991 * cleaned up, which can still happen by fuzzed images.
1994 * Allowing existing tree block to be allocated for other trees.
1996 * - Log tree operations
1997 * Exiting tree blocks get allocated to log tree, bumps its
1998 * generation, then get cleaned in tree re-balance.
1999 * Such tree block will not be written back, since it's clean,
2000 * thus no WRITTEN flag set.
2001 * And after log writes back, this tree block is not traced by
2002 * any dirty extent_io_tree.
2004 * - Offending tree block gets re-dirtied from its original owner
2005 * Since it has bumped generation, no WRITTEN flag, it can be
2006 * reused without COWing. This tree block will not be traced
2007 * by btrfs_transaction::dirty_pages.
2009 * Now such dirty tree block will not be cleaned by any dirty
2010 * extent io tree. Thus we don't want to submit such wild eb
2011 * if the fs already has error.
2013 * We can get ret > 0 from submit_extent_page() indicating how many ebs
2014 * were submitted. Reset it to 0 to avoid false alerts for the caller.
2018 if (!ret && BTRFS_FS_ERROR(fs_info))
2022 btrfs_put_block_group(ctx.zoned_bg);
2023 btrfs_zoned_meta_io_unlock(fs_info);
2028 * Walk the list of dirty pages of the given address space and write all of them.
2030 * @mapping: address space structure to write
2031 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2032 * @bio_ctrl: holds context for the write, namely the bio
2034 * If a page is already under I/O, write_cache_pages() skips it, even
2035 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2036 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2037 * and msync() need to guarantee that all the data which was dirty at the time
2038 * the call was made get new I/O started against them. If wbc->sync_mode is
2039 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2040 * existing IO to complete.
2042 static int extent_write_cache_pages(struct address_space *mapping,
2043 struct btrfs_bio_ctrl *bio_ctrl)
2045 struct writeback_control *wbc = bio_ctrl->wbc;
2046 struct inode *inode = mapping->host;
2049 int nr_to_write_done = 0;
2050 struct folio_batch fbatch;
2051 unsigned int nr_folios;
2053 pgoff_t end; /* Inclusive */
2055 int range_whole = 0;
2060 * We have to hold onto the inode so that ordered extents can do their
2061 * work when the IO finishes. The alternative to this is failing to add
2062 * an ordered extent if the igrab() fails there and that is a huge pain
2063 * to deal with, so instead just hold onto the inode throughout the
2064 * writepages operation. If it fails here we are freeing up the inode
2065 * anyway and we'd rather not waste our time writing out stuff that is
2066 * going to be truncated anyway.
2071 folio_batch_init(&fbatch);
2072 if (wbc->range_cyclic) {
2073 index = mapping->writeback_index; /* Start from prev offset */
2076 * Start from the beginning does not need to cycle over the
2077 * range, mark it as scanned.
2079 scanned = (index == 0);
2081 index = wbc->range_start >> PAGE_SHIFT;
2082 end = wbc->range_end >> PAGE_SHIFT;
2083 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2089 * We do the tagged writepage as long as the snapshot flush bit is set
2090 * and we are the first one who do the filemap_flush() on this inode.
2092 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2093 * not race in and drop the bit.
2095 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2096 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2097 &BTRFS_I(inode)->runtime_flags))
2098 wbc->tagged_writepages = 1;
2100 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2101 tag = PAGECACHE_TAG_TOWRITE;
2103 tag = PAGECACHE_TAG_DIRTY;
2105 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2106 tag_pages_for_writeback(mapping, index, end);
2108 while (!done && !nr_to_write_done && (index <= end) &&
2109 (nr_folios = filemap_get_folios_tag(mapping, &index,
2110 end, tag, &fbatch))) {
2113 for (i = 0; i < nr_folios; i++) {
2114 struct folio *folio = fbatch.folios[i];
2116 done_index = folio_next_index(folio);
2118 * At this point we hold neither the i_pages lock nor
2119 * the page lock: the page may be truncated or
2120 * invalidated (changing page->mapping to NULL),
2121 * or even swizzled back from swapper_space to
2122 * tmpfs file mapping
2124 if (!folio_trylock(folio)) {
2125 submit_write_bio(bio_ctrl, 0);
2129 if (unlikely(folio->mapping != mapping)) {
2130 folio_unlock(folio);
2134 if (!folio_test_dirty(folio)) {
2135 /* Someone wrote it for us. */
2136 folio_unlock(folio);
2140 if (wbc->sync_mode != WB_SYNC_NONE) {
2141 if (folio_test_writeback(folio))
2142 submit_write_bio(bio_ctrl, 0);
2143 folio_wait_writeback(folio);
2146 if (folio_test_writeback(folio) ||
2147 !folio_clear_dirty_for_io(folio)) {
2148 folio_unlock(folio);
2152 ret = __extent_writepage(&folio->page, bio_ctrl);
2159 * The filesystem may choose to bump up nr_to_write.
2160 * We have to make sure to honor the new nr_to_write
2163 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2164 wbc->nr_to_write <= 0);
2166 folio_batch_release(&fbatch);
2169 if (!scanned && !done) {
2171 * We hit the last page and there is more work to be done: wrap
2172 * back to the start of the file
2178 * If we're looping we could run into a page that is locked by a
2179 * writer and that writer could be waiting on writeback for a
2180 * page in our current bio, and thus deadlock, so flush the
2183 submit_write_bio(bio_ctrl, 0);
2187 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2188 mapping->writeback_index = done_index;
2190 btrfs_add_delayed_iput(BTRFS_I(inode));
2195 * Submit the pages in the range to bio for call sites which delalloc range has
2196 * already been ran (aka, ordered extent inserted) and all pages are still
2199 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2200 u64 start, u64 end, struct writeback_control *wbc,
2203 bool found_error = false;
2205 struct address_space *mapping = inode->i_mapping;
2206 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2207 const u32 sectorsize = fs_info->sectorsize;
2208 loff_t i_size = i_size_read(inode);
2210 struct btrfs_bio_ctrl bio_ctrl = {
2212 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2215 if (wbc->no_cgroup_owner)
2216 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2218 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2220 while (cur <= end) {
2221 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2222 u32 cur_len = cur_end + 1 - cur;
2226 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2227 ASSERT(PageLocked(page));
2228 if (pages_dirty && page != locked_page) {
2229 ASSERT(PageDirty(page));
2230 clear_page_dirty_for_io(page);
2233 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2238 /* Make sure the mapping tag for page dirty gets cleared. */
2240 set_page_writeback(page);
2241 end_page_writeback(page);
2244 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2245 cur, cur_len, !ret);
2246 mapping_set_error(page->mapping, ret);
2248 btrfs_folio_unlock_writer(fs_info, page_folio(page), cur, cur_len);
2256 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2259 int extent_writepages(struct address_space *mapping,
2260 struct writeback_control *wbc)
2262 struct inode *inode = mapping->host;
2264 struct btrfs_bio_ctrl bio_ctrl = {
2266 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2270 * Allow only a single thread to do the reloc work in zoned mode to
2271 * protect the write pointer updates.
2273 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2274 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2275 submit_write_bio(&bio_ctrl, ret);
2276 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2280 void extent_readahead(struct readahead_control *rac)
2282 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2283 struct page *pagepool[16];
2284 struct extent_map *em_cached = NULL;
2285 u64 prev_em_start = (u64)-1;
2288 while ((nr = readahead_page_batch(rac, pagepool))) {
2289 u64 contig_start = readahead_pos(rac);
2290 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2292 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2293 &em_cached, &bio_ctrl, &prev_em_start);
2297 free_extent_map(em_cached);
2298 submit_one_bio(&bio_ctrl);
2302 * basic invalidate_folio code, this waits on any locked or writeback
2303 * ranges corresponding to the folio, and then deletes any extent state
2304 * records from the tree
2306 int extent_invalidate_folio(struct extent_io_tree *tree,
2307 struct folio *folio, size_t offset)
2309 struct extent_state *cached_state = NULL;
2310 u64 start = folio_pos(folio);
2311 u64 end = start + folio_size(folio) - 1;
2312 size_t blocksize = folio_to_fs_info(folio)->sectorsize;
2314 /* This function is only called for the btree inode */
2315 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2317 start += ALIGN(offset, blocksize);
2321 lock_extent(tree, start, end, &cached_state);
2322 folio_wait_writeback(folio);
2325 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2326 * so here we only need to unlock the extent range to free any
2327 * existing extent state.
2329 unlock_extent(tree, start, end, &cached_state);
2334 * a helper for release_folio, this tests for areas of the page that
2335 * are locked or under IO and drops the related state bits if it is safe
2338 static int try_release_extent_state(struct extent_io_tree *tree,
2339 struct page *page, gfp_t mask)
2341 u64 start = page_offset(page);
2342 u64 end = start + PAGE_SIZE - 1;
2345 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2348 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2349 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS |
2350 EXTENT_QGROUP_RESERVED);
2353 * At this point we can safely clear everything except the
2354 * locked bit, the nodatasum bit and the delalloc new bit.
2355 * The delalloc new bit will be cleared by ordered extent
2358 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2360 /* if clear_extent_bit failed for enomem reasons,
2361 * we can't allow the release to continue.
2372 * a helper for release_folio. As long as there are no locked extents
2373 * in the range corresponding to the page, both state records and extent
2374 * map records are removed
2376 int try_release_extent_mapping(struct page *page, gfp_t mask)
2378 struct extent_map *em;
2379 u64 start = page_offset(page);
2380 u64 end = start + PAGE_SIZE - 1;
2381 struct btrfs_inode *btrfs_inode = page_to_inode(page);
2382 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2383 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2385 if (gfpflags_allow_blocking(mask) &&
2386 page->mapping->host->i_size > SZ_16M) {
2388 while (start <= end) {
2389 struct btrfs_fs_info *fs_info;
2392 len = end - start + 1;
2393 write_lock(&map->lock);
2394 em = lookup_extent_mapping(map, start, len);
2396 write_unlock(&map->lock);
2399 if ((em->flags & EXTENT_FLAG_PINNED) ||
2400 em->start != start) {
2401 write_unlock(&map->lock);
2402 free_extent_map(em);
2405 if (test_range_bit_exists(tree, em->start,
2406 extent_map_end(em) - 1,
2410 * If it's not in the list of modified extents, used
2411 * by a fast fsync, we can remove it. If it's being
2412 * logged we can safely remove it since fsync took an
2413 * extra reference on the em.
2415 if (list_empty(&em->list) ||
2416 (em->flags & EXTENT_FLAG_LOGGING))
2419 * If it's in the list of modified extents, remove it
2420 * only if its generation is older then the current one,
2421 * in which case we don't need it for a fast fsync.
2422 * Otherwise don't remove it, we could be racing with an
2423 * ongoing fast fsync that could miss the new extent.
2425 fs_info = btrfs_inode->root->fs_info;
2426 spin_lock(&fs_info->trans_lock);
2427 cur_gen = fs_info->generation;
2428 spin_unlock(&fs_info->trans_lock);
2429 if (em->generation >= cur_gen)
2433 * We only remove extent maps that are not in the list of
2434 * modified extents or that are in the list but with a
2435 * generation lower then the current generation, so there
2436 * is no need to set the full fsync flag on the inode (it
2437 * hurts the fsync performance for workloads with a data
2438 * size that exceeds or is close to the system's memory).
2440 remove_extent_mapping(map, em);
2441 /* once for the rb tree */
2442 free_extent_map(em);
2444 start = extent_map_end(em);
2445 write_unlock(&map->lock);
2448 free_extent_map(em);
2450 cond_resched(); /* Allow large-extent preemption. */
2453 return try_release_extent_state(tree, page, mask);
2456 struct btrfs_fiemap_entry {
2464 * Indicate the caller of emit_fiemap_extent() that it needs to unlock the file
2465 * range from the inode's io tree, unlock the subvolume tree search path, flush
2466 * the fiemap cache and relock the file range and research the subvolume tree.
2467 * The value here is something negative that can't be confused with a valid
2468 * errno value and different from 1 because that's also a return value from
2469 * fiemap_fill_next_extent() and also it's often used to mean some btree search
2470 * did not find a key, so make it some distinct negative value.
2472 #define BTRFS_FIEMAP_FLUSH_CACHE (-(MAX_ERRNO + 1))
2477 * - Cache the next entry to be emitted to the fiemap buffer, so that we can
2478 * merge extents that are contiguous and can be grouped as a single one;
2480 * - Store extents ready to be written to the fiemap buffer in an intermediary
2481 * buffer. This intermediary buffer is to ensure that in case the fiemap
2482 * buffer is memory mapped to the fiemap target file, we don't deadlock
2483 * during btrfs_page_mkwrite(). This is because during fiemap we are locking
2484 * an extent range in order to prevent races with delalloc flushing and
2485 * ordered extent completion, which is needed in order to reliably detect
2486 * delalloc in holes and prealloc extents. And this can lead to a deadlock
2487 * if the fiemap buffer is memory mapped to the file we are running fiemap
2488 * against (a silly, useless in practice scenario, but possible) because
2489 * btrfs_page_mkwrite() will try to lock the same extent range.
2491 struct fiemap_cache {
2492 /* An array of ready fiemap entries. */
2493 struct btrfs_fiemap_entry *entries;
2494 /* Number of entries in the entries array. */
2496 /* Index of the next entry in the entries array to write to. */
2499 * Once the entries array is full, this indicates what's the offset for
2500 * the next file extent item we must search for in the inode's subvolume
2501 * tree after unlocking the extent range in the inode's io tree and
2502 * releasing the search path.
2504 u64 next_search_offset;
2506 * This matches struct fiemap_extent_info::fi_mapped_extents, we use it
2507 * to count ourselves emitted extents and stop instead of relying on
2508 * fiemap_fill_next_extent() because we buffer ready fiemap entries at
2509 * the @entries array, and we want to stop as soon as we hit the max
2510 * amount of extents to map, not just to save time but also to make the
2511 * logic at extent_fiemap() simpler.
2513 unsigned int extents_mapped;
2514 /* Fields for the cached extent (unsubmitted, not ready, extent). */
2522 static int flush_fiemap_cache(struct fiemap_extent_info *fieinfo,
2523 struct fiemap_cache *cache)
2525 for (int i = 0; i < cache->entries_pos; i++) {
2526 struct btrfs_fiemap_entry *entry = &cache->entries[i];
2529 ret = fiemap_fill_next_extent(fieinfo, entry->offset,
2530 entry->phys, entry->len,
2533 * Ignore 1 (reached max entries) because we keep track of that
2534 * ourselves in emit_fiemap_extent().
2539 cache->entries_pos = 0;
2545 * Helper to submit fiemap extent.
2547 * Will try to merge current fiemap extent specified by @offset, @phys,
2548 * @len and @flags with cached one.
2549 * And only when we fails to merge, cached one will be submitted as
2552 * Return value is the same as fiemap_fill_next_extent().
2554 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2555 struct fiemap_cache *cache,
2556 u64 offset, u64 phys, u64 len, u32 flags)
2558 struct btrfs_fiemap_entry *entry;
2561 /* Set at the end of extent_fiemap(). */
2562 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2568 * When iterating the extents of the inode, at extent_fiemap(), we may
2569 * find an extent that starts at an offset behind the end offset of the
2570 * previous extent we processed. This happens if fiemap is called
2571 * without FIEMAP_FLAG_SYNC and there are ordered extents completing
2572 * after we had to unlock the file range, release the search path, emit
2573 * the fiemap extents stored in the buffer (cache->entries array) and
2574 * the lock the remainder of the range and re-search the btree.
2576 * For example we are in leaf X processing its last item, which is the
2577 * file extent item for file range [512K, 1M[, and after
2578 * btrfs_next_leaf() releases the path, there's an ordered extent that
2579 * completes for the file range [768K, 2M[, and that results in trimming
2580 * the file extent item so that it now corresponds to the file range
2581 * [512K, 768K[ and a new file extent item is inserted for the file
2582 * range [768K, 2M[, which may end up as the last item of leaf X or as
2583 * the first item of the next leaf - in either case btrfs_next_leaf()
2584 * will leave us with a path pointing to the new extent item, for the
2585 * file range [768K, 2M[, since that's the first key that follows the
2586 * last one we processed. So in order not to report overlapping extents
2587 * to user space, we trim the length of the previously cached extent and
2590 * Upon calling btrfs_next_leaf() we may also find an extent with an
2591 * offset smaller than or equals to cache->offset, and this happens
2592 * when we had a hole or prealloc extent with several delalloc ranges in
2593 * it, but after btrfs_next_leaf() released the path, delalloc was
2594 * flushed and the resulting ordered extents were completed, so we can
2595 * now have found a file extent item for an offset that is smaller than
2596 * or equals to what we have in cache->offset. We deal with this as
2599 cache_end = cache->offset + cache->len;
2600 if (cache_end > offset) {
2601 if (offset == cache->offset) {
2603 * We cached a dealloc range (found in the io tree) for
2604 * a hole or prealloc extent and we have now found a
2605 * file extent item for the same offset. What we have
2606 * now is more recent and up to date, so discard what
2607 * we had in the cache and use what we have just found.
2610 } else if (offset > cache->offset) {
2612 * The extent range we previously found ends after the
2613 * offset of the file extent item we found and that
2614 * offset falls somewhere in the middle of that previous
2615 * extent range. So adjust the range we previously found
2616 * to end at the offset of the file extent item we have
2617 * just found, since this extent is more up to date.
2618 * Emit that adjusted range and cache the file extent
2619 * item we have just found. This corresponds to the case
2620 * where a previously found file extent item was split
2621 * due to an ordered extent completing.
2623 cache->len = offset - cache->offset;
2626 const u64 range_end = offset + len;
2629 * The offset of the file extent item we have just found
2630 * is behind the cached offset. This means we were
2631 * processing a hole or prealloc extent for which we
2632 * have found delalloc ranges (in the io tree), so what
2633 * we have in the cache is the last delalloc range we
2634 * found while the file extent item we found can be
2635 * either for a whole delalloc range we previously
2636 * emmitted or only a part of that range.
2638 * We have two cases here:
2640 * 1) The file extent item's range ends at or behind the
2641 * cached extent's end. In this case just ignore the
2642 * current file extent item because we don't want to
2643 * overlap with previous ranges that may have been
2646 * 2) The file extent item starts behind the currently
2647 * cached extent but its end offset goes beyond the
2648 * end offset of the cached extent. We don't want to
2649 * overlap with a previous range that may have been
2650 * emmitted already, so we emit the currently cached
2651 * extent and then partially store the current file
2652 * extent item's range in the cache, for the subrange
2653 * going the cached extent's end to the end of the
2656 if (range_end <= cache_end)
2659 if (!(flags & (FIEMAP_EXTENT_ENCODED | FIEMAP_EXTENT_DELALLOC)))
2660 phys += cache_end - offset;
2663 len = range_end - cache_end;
2669 * Only merges fiemap extents if
2670 * 1) Their logical addresses are continuous
2672 * 2) Their physical addresses are continuous
2673 * So truly compressed (physical size smaller than logical size)
2674 * extents won't get merged with each other
2676 * 3) Share same flags
2678 if (cache->offset + cache->len == offset &&
2679 cache->phys + cache->len == phys &&
2680 cache->flags == flags) {
2686 /* Not mergeable, need to submit cached one */
2688 if (cache->entries_pos == cache->entries_size) {
2690 * We will need to research for the end offset of the last
2691 * stored extent and not from the current offset, because after
2692 * unlocking the range and releasing the path, if there's a hole
2693 * between that end offset and this current offset, a new extent
2694 * may have been inserted due to a new write, so we don't want
2697 entry = &cache->entries[cache->entries_size - 1];
2698 cache->next_search_offset = entry->offset + entry->len;
2699 cache->cached = false;
2701 return BTRFS_FIEMAP_FLUSH_CACHE;
2704 entry = &cache->entries[cache->entries_pos];
2705 entry->offset = cache->offset;
2706 entry->phys = cache->phys;
2707 entry->len = cache->len;
2708 entry->flags = cache->flags;
2709 cache->entries_pos++;
2710 cache->extents_mapped++;
2712 if (cache->extents_mapped == fieinfo->fi_extents_max) {
2713 cache->cached = false;
2717 cache->cached = true;
2718 cache->offset = offset;
2721 cache->flags = flags;
2727 * Emit last fiemap cache
2729 * The last fiemap cache may still be cached in the following case:
2731 * |<- Fiemap range ->|
2732 * |<------------ First extent ----------->|
2734 * In this case, the first extent range will be cached but not emitted.
2735 * So we must emit it before ending extent_fiemap().
2737 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2738 struct fiemap_cache *cache)
2745 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2746 cache->len, cache->flags);
2747 cache->cached = false;
2753 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2755 struct extent_buffer *clone = path->nodes[0];
2756 struct btrfs_key key;
2761 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2765 * Add a temporary extra ref to an already cloned extent buffer to
2766 * prevent btrfs_next_leaf() freeing it, we want to reuse it to avoid
2767 * the cost of allocating a new one.
2769 ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, &clone->bflags));
2770 atomic_inc(&clone->refs);
2772 ret = btrfs_next_leaf(inode->root, path);
2777 * Don't bother with cloning if there are no more file extent items for
2780 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2781 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY) {
2786 /* See the comment at fiemap_search_slot() about why we clone. */
2787 copy_extent_buffer_full(clone, path->nodes[0]);
2789 * Important to preserve the start field, for the optimizations when
2790 * checking if extents are shared (see extent_fiemap()).
2792 clone->start = path->nodes[0]->start;
2794 slot = path->slots[0];
2795 btrfs_release_path(path);
2796 path->nodes[0] = clone;
2797 path->slots[0] = slot;
2800 free_extent_buffer(clone);
2806 * Search for the first file extent item that starts at a given file offset or
2807 * the one that starts immediately before that offset.
2808 * Returns: 0 on success, < 0 on error, 1 if not found.
2810 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2813 const u64 ino = btrfs_ino(inode);
2814 struct btrfs_root *root = inode->root;
2815 struct extent_buffer *clone;
2816 struct btrfs_key key;
2821 key.type = BTRFS_EXTENT_DATA_KEY;
2822 key.offset = file_offset;
2824 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2828 if (ret > 0 && path->slots[0] > 0) {
2829 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2830 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2834 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2835 ret = btrfs_next_leaf(root, path);
2839 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2840 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2845 * We clone the leaf and use it during fiemap. This is because while
2846 * using the leaf we do expensive things like checking if an extent is
2847 * shared, which can take a long time. In order to prevent blocking
2848 * other tasks for too long, we use a clone of the leaf. We have locked
2849 * the file range in the inode's io tree, so we know none of our file
2850 * extent items can change. This way we avoid blocking other tasks that
2851 * want to insert items for other inodes in the same leaf or b+tree
2852 * rebalance operations (triggered for example when someone is trying
2853 * to push items into this leaf when trying to insert an item in a
2855 * We also need the private clone because holding a read lock on an
2856 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2857 * when we check if extents are shared, as backref walking may need to
2858 * lock the same leaf we are processing.
2860 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2864 slot = path->slots[0];
2865 btrfs_release_path(path);
2866 path->nodes[0] = clone;
2867 path->slots[0] = slot;
2873 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2874 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2875 * extent. The end offset (@end) is inclusive.
2877 static int fiemap_process_hole(struct btrfs_inode *inode,
2878 struct fiemap_extent_info *fieinfo,
2879 struct fiemap_cache *cache,
2880 struct extent_state **delalloc_cached_state,
2881 struct btrfs_backref_share_check_ctx *backref_ctx,
2882 u64 disk_bytenr, u64 extent_offset,
2886 const u64 i_size = i_size_read(&inode->vfs_inode);
2887 u64 cur_offset = start;
2888 u64 last_delalloc_end = 0;
2889 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2890 bool checked_extent_shared = false;
2894 * There can be no delalloc past i_size, so don't waste time looking for
2897 while (cur_offset < end && cur_offset < i_size) {
2901 u64 prealloc_len = 0;
2904 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2905 delalloc_cached_state,
2912 * If this is a prealloc extent we have to report every section
2913 * of it that has no delalloc.
2915 if (disk_bytenr != 0) {
2916 if (last_delalloc_end == 0) {
2917 prealloc_start = start;
2918 prealloc_len = delalloc_start - start;
2920 prealloc_start = last_delalloc_end + 1;
2921 prealloc_len = delalloc_start - prealloc_start;
2925 if (prealloc_len > 0) {
2926 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2927 ret = btrfs_is_data_extent_shared(inode,
2934 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2936 checked_extent_shared = true;
2938 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2939 disk_bytenr + extent_offset,
2940 prealloc_len, prealloc_flags);
2943 extent_offset += prealloc_len;
2946 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2947 delalloc_end + 1 - delalloc_start,
2948 FIEMAP_EXTENT_DELALLOC |
2949 FIEMAP_EXTENT_UNKNOWN);
2953 last_delalloc_end = delalloc_end;
2954 cur_offset = delalloc_end + 1;
2955 extent_offset += cur_offset - delalloc_start;
2960 * Either we found no delalloc for the whole prealloc extent or we have
2961 * a prealloc extent that spans i_size or starts at or after i_size.
2963 if (disk_bytenr != 0 && last_delalloc_end < end) {
2967 if (last_delalloc_end == 0) {
2968 prealloc_start = start;
2969 prealloc_len = end + 1 - start;
2971 prealloc_start = last_delalloc_end + 1;
2972 prealloc_len = end + 1 - prealloc_start;
2975 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2976 ret = btrfs_is_data_extent_shared(inode,
2983 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2985 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2986 disk_bytenr + extent_offset,
2987 prealloc_len, prealloc_flags);
2995 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2996 struct btrfs_path *path,
2997 u64 *last_extent_end_ret)
2999 const u64 ino = btrfs_ino(inode);
3000 struct btrfs_root *root = inode->root;
3001 struct extent_buffer *leaf;
3002 struct btrfs_file_extent_item *ei;
3003 struct btrfs_key key;
3008 * Lookup the last file extent. We're not using i_size here because
3009 * there might be preallocation past i_size.
3011 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
3012 /* There can't be a file extent item at offset (u64)-1 */
3018 * For a non-existing key, btrfs_search_slot() always leaves us at a
3019 * slot > 0, except if the btree is empty, which is impossible because
3020 * at least it has the inode item for this inode and all the items for
3021 * the root inode 256.
3023 ASSERT(path->slots[0] > 0);
3025 leaf = path->nodes[0];
3026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3027 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
3028 /* No file extent items in the subvolume tree. */
3029 *last_extent_end_ret = 0;
3034 * For an inline extent, the disk_bytenr is where inline data starts at,
3035 * so first check if we have an inline extent item before checking if we
3036 * have an implicit hole (disk_bytenr == 0).
3038 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
3039 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
3040 *last_extent_end_ret = btrfs_file_extent_end(path);
3045 * Find the last file extent item that is not a hole (when NO_HOLES is
3046 * not enabled). This should take at most 2 iterations in the worst
3047 * case: we have one hole file extent item at slot 0 of a leaf and
3048 * another hole file extent item as the last item in the previous leaf.
3049 * This is because we merge file extent items that represent holes.
3051 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3052 while (disk_bytenr == 0) {
3053 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
3056 } else if (ret > 0) {
3057 /* No file extent items that are not holes. */
3058 *last_extent_end_ret = 0;
3061 leaf = path->nodes[0];
3062 ei = btrfs_item_ptr(leaf, path->slots[0],
3063 struct btrfs_file_extent_item);
3064 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3067 *last_extent_end_ret = btrfs_file_extent_end(path);
3071 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
3074 const u64 ino = btrfs_ino(inode);
3075 struct extent_state *cached_state = NULL;
3076 struct extent_state *delalloc_cached_state = NULL;
3077 struct btrfs_path *path;
3078 struct fiemap_cache cache = { 0 };
3079 struct btrfs_backref_share_check_ctx *backref_ctx;
3080 u64 last_extent_end;
3081 u64 prev_extent_end;
3084 const u64 sectorsize = inode->root->fs_info->sectorsize;
3085 bool stopped = false;
3088 cache.entries_size = PAGE_SIZE / sizeof(struct btrfs_fiemap_entry);
3089 cache.entries = kmalloc_array(cache.entries_size,
3090 sizeof(struct btrfs_fiemap_entry),
3092 backref_ctx = btrfs_alloc_backref_share_check_ctx();
3093 path = btrfs_alloc_path();
3094 if (!cache.entries || !backref_ctx || !path) {
3100 range_start = round_down(start, sectorsize);
3101 range_end = round_up(start + len, sectorsize);
3102 prev_extent_end = range_start;
3104 lock_extent(&inode->io_tree, range_start, range_end, &cached_state);
3106 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
3109 btrfs_release_path(path);
3111 path->reada = READA_FORWARD;
3112 ret = fiemap_search_slot(inode, path, range_start);
3115 } else if (ret > 0) {
3117 * No file extent item found, but we may have delalloc between
3118 * the current offset and i_size. So check for that.
3121 goto check_eof_delalloc;
3124 while (prev_extent_end < range_end) {
3125 struct extent_buffer *leaf = path->nodes[0];
3126 struct btrfs_file_extent_item *ei;
3127 struct btrfs_key key;
3130 u64 extent_offset = 0;
3132 u64 disk_bytenr = 0;
3137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3138 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3141 extent_end = btrfs_file_extent_end(path);
3144 * The first iteration can leave us at an extent item that ends
3145 * before our range's start. Move to the next item.
3147 if (extent_end <= range_start)
3150 backref_ctx->curr_leaf_bytenr = leaf->start;
3152 /* We have in implicit hole (NO_HOLES feature enabled). */
3153 if (prev_extent_end < key.offset) {
3154 const u64 hole_end = min(key.offset, range_end) - 1;
3156 ret = fiemap_process_hole(inode, fieinfo, &cache,
3157 &delalloc_cached_state,
3158 backref_ctx, 0, 0, 0,
3159 prev_extent_end, hole_end);
3162 } else if (ret > 0) {
3163 /* fiemap_fill_next_extent() told us to stop. */
3168 /* We've reached the end of the fiemap range, stop. */
3169 if (key.offset >= range_end) {
3175 extent_len = extent_end - key.offset;
3176 ei = btrfs_item_ptr(leaf, path->slots[0],
3177 struct btrfs_file_extent_item);
3178 compression = btrfs_file_extent_compression(leaf, ei);
3179 extent_type = btrfs_file_extent_type(leaf, ei);
3180 extent_gen = btrfs_file_extent_generation(leaf, ei);
3182 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3183 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3184 if (compression == BTRFS_COMPRESS_NONE)
3185 extent_offset = btrfs_file_extent_offset(leaf, ei);
3188 if (compression != BTRFS_COMPRESS_NONE)
3189 flags |= FIEMAP_EXTENT_ENCODED;
3191 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3192 flags |= FIEMAP_EXTENT_DATA_INLINE;
3193 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
3194 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
3196 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
3197 ret = fiemap_process_hole(inode, fieinfo, &cache,
3198 &delalloc_cached_state,
3200 disk_bytenr, extent_offset,
3201 extent_gen, key.offset,
3203 } else if (disk_bytenr == 0) {
3204 /* We have an explicit hole. */
3205 ret = fiemap_process_hole(inode, fieinfo, &cache,
3206 &delalloc_cached_state,
3207 backref_ctx, 0, 0, 0,
3208 key.offset, extent_end - 1);
3210 /* We have a regular extent. */
3211 if (fieinfo->fi_extents_max) {
3212 ret = btrfs_is_data_extent_shared(inode,
3219 flags |= FIEMAP_EXTENT_SHARED;
3222 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
3223 disk_bytenr + extent_offset,
3229 } else if (ret > 0) {
3230 /* emit_fiemap_extent() told us to stop. */
3235 prev_extent_end = extent_end;
3237 if (fatal_signal_pending(current)) {
3242 ret = fiemap_next_leaf_item(inode, path);
3245 } else if (ret > 0) {
3246 /* No more file extent items for this inode. */
3253 if (!stopped && prev_extent_end < range_end) {
3254 ret = fiemap_process_hole(inode, fieinfo, &cache,
3255 &delalloc_cached_state, backref_ctx,
3256 0, 0, 0, prev_extent_end, range_end - 1);
3259 prev_extent_end = range_end;
3262 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3263 const u64 i_size = i_size_read(&inode->vfs_inode);
3265 if (prev_extent_end < i_size) {
3270 delalloc = btrfs_find_delalloc_in_range(inode,
3273 &delalloc_cached_state,
3277 cache.flags |= FIEMAP_EXTENT_LAST;
3279 cache.flags |= FIEMAP_EXTENT_LAST;
3284 unlock_extent(&inode->io_tree, range_start, range_end, &cached_state);
3286 if (ret == BTRFS_FIEMAP_FLUSH_CACHE) {
3287 btrfs_release_path(path);
3288 ret = flush_fiemap_cache(fieinfo, &cache);
3291 len -= cache.next_search_offset - start;
3292 start = cache.next_search_offset;
3294 } else if (ret < 0) {
3299 * Must free the path before emitting to the fiemap buffer because we
3300 * may have a non-cloned leaf and if the fiemap buffer is memory mapped
3301 * to a file, a write into it (through btrfs_page_mkwrite()) may trigger
3302 * waiting for an ordered extent that in order to complete needs to
3303 * modify that leaf, therefore leading to a deadlock.
3305 btrfs_free_path(path);
3308 ret = flush_fiemap_cache(fieinfo, &cache);
3312 ret = emit_last_fiemap_cache(fieinfo, &cache);
3314 free_extent_state(delalloc_cached_state);
3315 kfree(cache.entries);
3316 btrfs_free_backref_share_ctx(backref_ctx);
3317 btrfs_free_path(path);
3321 static void __free_extent_buffer(struct extent_buffer *eb)
3323 kmem_cache_free(extent_buffer_cache, eb);
3326 static int extent_buffer_under_io(const struct extent_buffer *eb)
3328 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3329 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3332 static bool folio_range_has_eb(struct btrfs_fs_info *fs_info, struct folio *folio)
3334 struct btrfs_subpage *subpage;
3336 lockdep_assert_held(&folio->mapping->i_private_lock);
3338 if (folio_test_private(folio)) {
3339 subpage = folio_get_private(folio);
3340 if (atomic_read(&subpage->eb_refs))
3343 * Even there is no eb refs here, we may still have
3344 * end_page_read() call relying on page::private.
3346 if (atomic_read(&subpage->readers))
3352 static void detach_extent_buffer_folio(struct extent_buffer *eb, struct folio *folio)
3354 struct btrfs_fs_info *fs_info = eb->fs_info;
3355 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3358 * For mapped eb, we're going to change the folio private, which should
3359 * be done under the i_private_lock.
3362 spin_lock(&folio->mapping->i_private_lock);
3364 if (!folio_test_private(folio)) {
3366 spin_unlock(&folio->mapping->i_private_lock);
3370 if (fs_info->nodesize >= PAGE_SIZE) {
3372 * We do this since we'll remove the pages after we've
3373 * removed the eb from the radix tree, so we could race
3374 * and have this page now attached to the new eb. So
3375 * only clear folio if it's still connected to
3378 if (folio_test_private(folio) && folio_get_private(folio) == eb) {
3379 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3380 BUG_ON(folio_test_dirty(folio));
3381 BUG_ON(folio_test_writeback(folio));
3382 /* We need to make sure we haven't be attached to a new eb. */
3383 folio_detach_private(folio);
3386 spin_unlock(&folio->mapping->i_private_lock);
3391 * For subpage, we can have dummy eb with folio private attached. In
3392 * this case, we can directly detach the private as such folio is only
3393 * attached to one dummy eb, no sharing.
3396 btrfs_detach_subpage(fs_info, folio);
3400 btrfs_folio_dec_eb_refs(fs_info, folio);
3403 * We can only detach the folio private if there are no other ebs in the
3404 * page range and no unfinished IO.
3406 if (!folio_range_has_eb(fs_info, folio))
3407 btrfs_detach_subpage(fs_info, folio);
3409 spin_unlock(&folio->mapping->i_private_lock);
3412 /* Release all pages attached to the extent buffer */
3413 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3415 ASSERT(!extent_buffer_under_io(eb));
3417 for (int i = 0; i < INLINE_EXTENT_BUFFER_PAGES; i++) {
3418 struct folio *folio = eb->folios[i];
3423 detach_extent_buffer_folio(eb, folio);
3425 /* One for when we allocated the folio. */
3431 * Helper for releasing the extent buffer.
3433 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3435 btrfs_release_extent_buffer_pages(eb);
3436 btrfs_leak_debug_del_eb(eb);
3437 __free_extent_buffer(eb);
3440 static struct extent_buffer *
3441 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3444 struct extent_buffer *eb = NULL;
3446 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3449 eb->fs_info = fs_info;
3450 init_rwsem(&eb->lock);
3452 btrfs_leak_debug_add_eb(eb);
3454 spin_lock_init(&eb->refs_lock);
3455 atomic_set(&eb->refs, 1);
3457 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3462 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3464 struct extent_buffer *new;
3465 int num_folios = num_extent_folios(src);
3468 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3473 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3474 * btrfs_release_extent_buffer() have different behavior for
3475 * UNMAPPED subpage extent buffer.
3477 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3479 ret = alloc_eb_folio_array(new, 0);
3481 btrfs_release_extent_buffer(new);
3485 for (int i = 0; i < num_folios; i++) {
3486 struct folio *folio = new->folios[i];
3489 ret = attach_extent_buffer_folio(new, folio, NULL);
3491 btrfs_release_extent_buffer(new);
3494 WARN_ON(folio_test_dirty(folio));
3496 copy_extent_buffer_full(new, src);
3497 set_extent_buffer_uptodate(new);
3502 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3503 u64 start, unsigned long len)
3505 struct extent_buffer *eb;
3509 eb = __alloc_extent_buffer(fs_info, start, len);
3513 ret = alloc_eb_folio_array(eb, 0);
3517 num_folios = num_extent_folios(eb);
3518 for (int i = 0; i < num_folios; i++) {
3519 ret = attach_extent_buffer_folio(eb, eb->folios[i], NULL);
3524 set_extent_buffer_uptodate(eb);
3525 btrfs_set_header_nritems(eb, 0);
3526 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3530 for (int i = 0; i < num_folios; i++) {
3531 if (eb->folios[i]) {
3532 detach_extent_buffer_folio(eb, eb->folios[i]);
3533 __folio_put(eb->folios[i]);
3536 __free_extent_buffer(eb);
3540 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3543 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3546 static void check_buffer_tree_ref(struct extent_buffer *eb)
3550 * The TREE_REF bit is first set when the extent_buffer is added
3551 * to the radix tree. It is also reset, if unset, when a new reference
3552 * is created by find_extent_buffer.
3554 * It is only cleared in two cases: freeing the last non-tree
3555 * reference to the extent_buffer when its STALE bit is set or
3556 * calling release_folio when the tree reference is the only reference.
3558 * In both cases, care is taken to ensure that the extent_buffer's
3559 * pages are not under io. However, release_folio can be concurrently
3560 * called with creating new references, which is prone to race
3561 * conditions between the calls to check_buffer_tree_ref in those
3562 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3564 * The actual lifetime of the extent_buffer in the radix tree is
3565 * adequately protected by the refcount, but the TREE_REF bit and
3566 * its corresponding reference are not. To protect against this
3567 * class of races, we call check_buffer_tree_ref from the codepaths
3568 * which trigger io. Note that once io is initiated, TREE_REF can no
3569 * longer be cleared, so that is the moment at which any such race is
3572 refs = atomic_read(&eb->refs);
3573 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3576 spin_lock(&eb->refs_lock);
3577 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3578 atomic_inc(&eb->refs);
3579 spin_unlock(&eb->refs_lock);
3582 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
3584 int num_folios= num_extent_folios(eb);
3586 check_buffer_tree_ref(eb);
3588 for (int i = 0; i < num_folios; i++)
3589 folio_mark_accessed(eb->folios[i]);
3592 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3595 struct extent_buffer *eb;
3597 eb = find_extent_buffer_nolock(fs_info, start);
3601 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3602 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3603 * another task running free_extent_buffer() might have seen that flag
3604 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3605 * writeback flags not set) and it's still in the tree (flag
3606 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3607 * decrementing the extent buffer's reference count twice. So here we
3608 * could race and increment the eb's reference count, clear its stale
3609 * flag, mark it as dirty and drop our reference before the other task
3610 * finishes executing free_extent_buffer, which would later result in
3611 * an attempt to free an extent buffer that is dirty.
3613 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3614 spin_lock(&eb->refs_lock);
3615 spin_unlock(&eb->refs_lock);
3617 mark_extent_buffer_accessed(eb);
3621 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3622 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3625 struct extent_buffer *eb, *exists = NULL;
3628 eb = find_extent_buffer(fs_info, start);
3631 eb = alloc_dummy_extent_buffer(fs_info, start);
3633 return ERR_PTR(-ENOMEM);
3634 eb->fs_info = fs_info;
3636 ret = radix_tree_preload(GFP_NOFS);
3638 exists = ERR_PTR(ret);
3641 spin_lock(&fs_info->buffer_lock);
3642 ret = radix_tree_insert(&fs_info->buffer_radix,
3643 start >> fs_info->sectorsize_bits, eb);
3644 spin_unlock(&fs_info->buffer_lock);
3645 radix_tree_preload_end();
3646 if (ret == -EEXIST) {
3647 exists = find_extent_buffer(fs_info, start);
3653 check_buffer_tree_ref(eb);
3654 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3658 btrfs_release_extent_buffer(eb);
3663 static struct extent_buffer *grab_extent_buffer(
3664 struct btrfs_fs_info *fs_info, struct page *page)
3666 struct folio *folio = page_folio(page);
3667 struct extent_buffer *exists;
3670 * For subpage case, we completely rely on radix tree to ensure we
3671 * don't try to insert two ebs for the same bytenr. So here we always
3672 * return NULL and just continue.
3674 if (fs_info->nodesize < PAGE_SIZE)
3677 /* Page not yet attached to an extent buffer */
3678 if (!folio_test_private(folio))
3682 * We could have already allocated an eb for this page and attached one
3683 * so lets see if we can get a ref on the existing eb, and if we can we
3684 * know it's good and we can just return that one, else we know we can
3685 * just overwrite folio private.
3687 exists = folio_get_private(folio);
3688 if (atomic_inc_not_zero(&exists->refs))
3691 WARN_ON(PageDirty(page));
3692 folio_detach_private(folio);
3696 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3698 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3699 btrfs_err(fs_info, "bad tree block start %llu", start);
3703 if (fs_info->nodesize < PAGE_SIZE &&
3704 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3706 "tree block crosses page boundary, start %llu nodesize %u",
3707 start, fs_info->nodesize);
3710 if (fs_info->nodesize >= PAGE_SIZE &&
3711 !PAGE_ALIGNED(start)) {
3713 "tree block is not page aligned, start %llu nodesize %u",
3714 start, fs_info->nodesize);
3717 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3718 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3720 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3721 start, fs_info->nodesize);
3728 * Return 0 if eb->folios[i] is attached to btree inode successfully.
3729 * Return >0 if there is already another extent buffer for the range,
3730 * and @found_eb_ret would be updated.
3731 * Return -EAGAIN if the filemap has an existing folio but with different size
3733 * The caller needs to free the existing folios and retry using the same order.
3735 static int attach_eb_folio_to_filemap(struct extent_buffer *eb, int i,
3736 struct extent_buffer **found_eb_ret)
3739 struct btrfs_fs_info *fs_info = eb->fs_info;
3740 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3741 const unsigned long index = eb->start >> PAGE_SHIFT;
3742 struct folio *existing_folio;
3745 ASSERT(found_eb_ret);
3747 /* Caller should ensure the folio exists. */
3748 ASSERT(eb->folios[i]);
3751 ret = filemap_add_folio(mapping, eb->folios[i], index + i,
3752 GFP_NOFS | __GFP_NOFAIL);
3756 existing_folio = filemap_lock_folio(mapping, index + i);
3757 /* The page cache only exists for a very short time, just retry. */
3758 if (IS_ERR(existing_folio))
3761 /* For now, we should only have single-page folios for btree inode. */
3762 ASSERT(folio_nr_pages(existing_folio) == 1);
3764 if (folio_size(existing_folio) != eb->folio_size) {
3765 folio_unlock(existing_folio);
3766 folio_put(existing_folio);
3770 if (fs_info->nodesize < PAGE_SIZE) {
3772 * We're going to reuse the existing page, can drop our page
3773 * and subpage structure now.
3775 __free_page(folio_page(eb->folios[i], 0));
3776 eb->folios[i] = existing_folio;
3778 struct extent_buffer *existing_eb;
3780 existing_eb = grab_extent_buffer(fs_info,
3781 folio_page(existing_folio, 0));
3783 /* The extent buffer still exists, we can use it directly. */
3784 *found_eb_ret = existing_eb;
3785 folio_unlock(existing_folio);
3786 folio_put(existing_folio);
3789 /* The extent buffer no longer exists, we can reuse the folio. */
3790 __free_page(folio_page(eb->folios[i], 0));
3791 eb->folios[i] = existing_folio;
3796 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3797 u64 start, u64 owner_root, int level)
3799 unsigned long len = fs_info->nodesize;
3802 struct extent_buffer *eb;
3803 struct extent_buffer *existing_eb = NULL;
3804 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3805 struct btrfs_subpage *prealloc = NULL;
3806 u64 lockdep_owner = owner_root;
3807 bool page_contig = true;
3811 if (check_eb_alignment(fs_info, start))
3812 return ERR_PTR(-EINVAL);
3814 #if BITS_PER_LONG == 32
3815 if (start >= MAX_LFS_FILESIZE) {
3816 btrfs_err_rl(fs_info,
3817 "extent buffer %llu is beyond 32bit page cache limit", start);
3818 btrfs_err_32bit_limit(fs_info);
3819 return ERR_PTR(-EOVERFLOW);
3821 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3822 btrfs_warn_32bit_limit(fs_info);
3825 eb = find_extent_buffer(fs_info, start);
3829 eb = __alloc_extent_buffer(fs_info, start, len);
3831 return ERR_PTR(-ENOMEM);
3834 * The reloc trees are just snapshots, so we need them to appear to be
3835 * just like any other fs tree WRT lockdep.
3837 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3838 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3840 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3843 * Preallocate folio private for subpage case, so that we won't
3844 * allocate memory with i_private_lock nor page lock hold.
3846 * The memory will be freed by attach_extent_buffer_page() or freed
3847 * manually if we exit earlier.
3849 if (fs_info->nodesize < PAGE_SIZE) {
3850 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3851 if (IS_ERR(prealloc)) {
3852 ret = PTR_ERR(prealloc);
3858 /* Allocate all pages first. */
3859 ret = alloc_eb_folio_array(eb, __GFP_NOFAIL);
3861 btrfs_free_subpage(prealloc);
3865 num_folios = num_extent_folios(eb);
3866 /* Attach all pages to the filemap. */
3867 for (int i = 0; i < num_folios; i++) {
3868 struct folio *folio;
3870 ret = attach_eb_folio_to_filemap(eb, i, &existing_eb);
3872 ASSERT(existing_eb);
3877 * TODO: Special handling for a corner case where the order of
3878 * folios mismatch between the new eb and filemap.
3880 * This happens when:
3882 * - the new eb is using higher order folio
3884 * - the filemap is still using 0-order folios for the range
3885 * This can happen at the previous eb allocation, and we don't
3886 * have higher order folio for the call.
3888 * - the existing eb has already been freed
3890 * In this case, we have to free the existing folios first, and
3891 * re-allocate using the same order.
3892 * Thankfully this is not going to happen yet, as we're still
3893 * using 0-order folios.
3895 if (unlikely(ret == -EAGAIN)) {
3902 * Only after attach_eb_folio_to_filemap(), eb->folios[] is
3903 * reliable, as we may choose to reuse the existing page cache
3904 * and free the allocated page.
3906 folio = eb->folios[i];
3907 eb->folio_size = folio_size(folio);
3908 eb->folio_shift = folio_shift(folio);
3909 spin_lock(&mapping->i_private_lock);
3910 /* Should not fail, as we have preallocated the memory */
3911 ret = attach_extent_buffer_folio(eb, folio, prealloc);
3914 * To inform we have extra eb under allocation, so that
3915 * detach_extent_buffer_page() won't release the folio private
3916 * when the eb hasn't yet been inserted into radix tree.
3918 * The ref will be decreased when the eb released the page, in
3919 * detach_extent_buffer_page().
3920 * Thus needs no special handling in error path.
3922 btrfs_folio_inc_eb_refs(fs_info, folio);
3923 spin_unlock(&mapping->i_private_lock);
3925 WARN_ON(btrfs_folio_test_dirty(fs_info, folio, eb->start, eb->len));
3928 * Check if the current page is physically contiguous with previous eb
3930 * At this stage, either we allocated a large folio, thus @i
3931 * would only be 0, or we fall back to per-page allocation.
3933 if (i && folio_page(eb->folios[i - 1], 0) + 1 != folio_page(folio, 0))
3934 page_contig = false;
3936 if (!btrfs_folio_test_uptodate(fs_info, folio, eb->start, eb->len))
3940 * We can't unlock the pages just yet since the extent buffer
3941 * hasn't been properly inserted in the radix tree, this
3942 * opens a race with btree_release_folio which can free a page
3943 * while we are still filling in all pages for the buffer and
3948 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3949 /* All pages are physically contiguous, can skip cross page handling. */
3951 eb->addr = folio_address(eb->folios[0]) + offset_in_page(eb->start);
3953 ret = radix_tree_preload(GFP_NOFS);
3957 spin_lock(&fs_info->buffer_lock);
3958 ret = radix_tree_insert(&fs_info->buffer_radix,
3959 start >> fs_info->sectorsize_bits, eb);
3960 spin_unlock(&fs_info->buffer_lock);
3961 radix_tree_preload_end();
3962 if (ret == -EEXIST) {
3964 existing_eb = find_extent_buffer(fs_info, start);
3970 /* add one reference for the tree */
3971 check_buffer_tree_ref(eb);
3972 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3975 * Now it's safe to unlock the pages because any calls to
3976 * btree_release_folio will correctly detect that a page belongs to a
3977 * live buffer and won't free them prematurely.
3979 for (int i = 0; i < num_folios; i++)
3980 unlock_page(folio_page(eb->folios[i], 0));
3984 WARN_ON(!atomic_dec_and_test(&eb->refs));
3987 * Any attached folios need to be detached before we unlock them. This
3988 * is because when we're inserting our new folios into the mapping, and
3989 * then attaching our eb to that folio. If we fail to insert our folio
3990 * we'll lookup the folio for that index, and grab that EB. We do not
3991 * want that to grab this eb, as we're getting ready to free it. So we
3992 * have to detach it first and then unlock it.
3994 * We have to drop our reference and NULL it out here because in the
3995 * subpage case detaching does a btrfs_folio_dec_eb_refs() for our eb.
3996 * Below when we call btrfs_release_extent_buffer() we will call
3997 * detach_extent_buffer_folio() on our remaining pages in the !subpage
3998 * case. If we left eb->folios[i] populated in the subpage case we'd
3999 * double put our reference and be super sad.
4001 for (int i = 0; i < attached; i++) {
4002 ASSERT(eb->folios[i]);
4003 detach_extent_buffer_folio(eb, eb->folios[i]);
4004 unlock_page(folio_page(eb->folios[i], 0));
4005 folio_put(eb->folios[i]);
4006 eb->folios[i] = NULL;
4009 * Now all pages of that extent buffer is unmapped, set UNMAPPED flag,
4010 * so it can be cleaned up without utlizing page->mapping.
4012 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4014 btrfs_release_extent_buffer(eb);
4016 return ERR_PTR(ret);
4017 ASSERT(existing_eb);
4021 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4023 struct extent_buffer *eb =
4024 container_of(head, struct extent_buffer, rcu_head);
4026 __free_extent_buffer(eb);
4029 static int release_extent_buffer(struct extent_buffer *eb)
4030 __releases(&eb->refs_lock)
4032 lockdep_assert_held(&eb->refs_lock);
4034 WARN_ON(atomic_read(&eb->refs) == 0);
4035 if (atomic_dec_and_test(&eb->refs)) {
4036 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4037 struct btrfs_fs_info *fs_info = eb->fs_info;
4039 spin_unlock(&eb->refs_lock);
4041 spin_lock(&fs_info->buffer_lock);
4042 radix_tree_delete(&fs_info->buffer_radix,
4043 eb->start >> fs_info->sectorsize_bits);
4044 spin_unlock(&fs_info->buffer_lock);
4046 spin_unlock(&eb->refs_lock);
4049 btrfs_leak_debug_del_eb(eb);
4050 /* Should be safe to release our pages at this point */
4051 btrfs_release_extent_buffer_pages(eb);
4052 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4053 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
4054 __free_extent_buffer(eb);
4058 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4061 spin_unlock(&eb->refs_lock);
4066 void free_extent_buffer(struct extent_buffer *eb)
4072 refs = atomic_read(&eb->refs);
4074 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
4075 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
4078 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
4082 spin_lock(&eb->refs_lock);
4083 if (atomic_read(&eb->refs) == 2 &&
4084 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4085 !extent_buffer_under_io(eb) &&
4086 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4087 atomic_dec(&eb->refs);
4090 * I know this is terrible, but it's temporary until we stop tracking
4091 * the uptodate bits and such for the extent buffers.
4093 release_extent_buffer(eb);
4096 void free_extent_buffer_stale(struct extent_buffer *eb)
4101 spin_lock(&eb->refs_lock);
4102 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4104 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4105 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4106 atomic_dec(&eb->refs);
4107 release_extent_buffer(eb);
4110 static void btree_clear_folio_dirty(struct folio *folio)
4112 ASSERT(folio_test_dirty(folio));
4113 ASSERT(folio_test_locked(folio));
4114 folio_clear_dirty_for_io(folio);
4115 xa_lock_irq(&folio->mapping->i_pages);
4116 if (!folio_test_dirty(folio))
4117 __xa_clear_mark(&folio->mapping->i_pages,
4118 folio_index(folio), PAGECACHE_TAG_DIRTY);
4119 xa_unlock_irq(&folio->mapping->i_pages);
4122 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
4124 struct btrfs_fs_info *fs_info = eb->fs_info;
4125 struct folio *folio = eb->folios[0];
4128 /* btree_clear_folio_dirty() needs page locked. */
4130 last = btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start, eb->len);
4132 btree_clear_folio_dirty(folio);
4133 folio_unlock(folio);
4134 WARN_ON(atomic_read(&eb->refs) == 0);
4137 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
4138 struct extent_buffer *eb)
4140 struct btrfs_fs_info *fs_info = eb->fs_info;
4143 btrfs_assert_tree_write_locked(eb);
4145 if (trans && btrfs_header_generation(eb) != trans->transid)
4149 * Instead of clearing the dirty flag off of the buffer, mark it as
4150 * EXTENT_BUFFER_ZONED_ZEROOUT. This allows us to preserve
4151 * write-ordering in zoned mode, without the need to later re-dirty
4152 * the extent_buffer.
4154 * The actual zeroout of the buffer will happen later in
4155 * btree_csum_one_bio.
4157 if (btrfs_is_zoned(fs_info)) {
4158 set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
4162 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
4165 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
4166 fs_info->dirty_metadata_batch);
4168 if (eb->fs_info->nodesize < PAGE_SIZE)
4169 return clear_subpage_extent_buffer_dirty(eb);
4171 num_folios = num_extent_folios(eb);
4172 for (int i = 0; i < num_folios; i++) {
4173 struct folio *folio = eb->folios[i];
4175 if (!folio_test_dirty(folio))
4178 btree_clear_folio_dirty(folio);
4179 folio_unlock(folio);
4181 WARN_ON(atomic_read(&eb->refs) == 0);
4184 void set_extent_buffer_dirty(struct extent_buffer *eb)
4189 check_buffer_tree_ref(eb);
4191 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4193 num_folios = num_extent_folios(eb);
4194 WARN_ON(atomic_read(&eb->refs) == 0);
4195 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4198 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
4201 * For subpage case, we can have other extent buffers in the
4202 * same page, and in clear_subpage_extent_buffer_dirty() we
4203 * have to clear page dirty without subpage lock held.
4204 * This can cause race where our page gets dirty cleared after
4207 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
4208 * its page for other reasons, we can use page lock to prevent
4212 lock_page(folio_page(eb->folios[0], 0));
4213 for (int i = 0; i < num_folios; i++)
4214 btrfs_folio_set_dirty(eb->fs_info, eb->folios[i],
4215 eb->start, eb->len);
4217 unlock_page(folio_page(eb->folios[0], 0));
4218 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
4220 eb->fs_info->dirty_metadata_batch);
4222 #ifdef CONFIG_BTRFS_DEBUG
4223 for (int i = 0; i < num_folios; i++)
4224 ASSERT(folio_test_dirty(eb->folios[i]));
4228 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
4230 struct btrfs_fs_info *fs_info = eb->fs_info;
4231 int num_folios = num_extent_folios(eb);
4233 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4234 for (int i = 0; i < num_folios; i++) {
4235 struct folio *folio = eb->folios[i];
4241 * This is special handling for metadata subpage, as regular
4242 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4244 if (fs_info->nodesize >= PAGE_SIZE)
4245 folio_clear_uptodate(folio);
4247 btrfs_subpage_clear_uptodate(fs_info, folio,
4248 eb->start, eb->len);
4252 void set_extent_buffer_uptodate(struct extent_buffer *eb)
4254 struct btrfs_fs_info *fs_info = eb->fs_info;
4255 int num_folios = num_extent_folios(eb);
4257 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4258 for (int i = 0; i < num_folios; i++) {
4259 struct folio *folio = eb->folios[i];
4262 * This is special handling for metadata subpage, as regular
4263 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4265 if (fs_info->nodesize >= PAGE_SIZE)
4266 folio_mark_uptodate(folio);
4268 btrfs_subpage_set_uptodate(fs_info, folio,
4269 eb->start, eb->len);
4273 static void end_bbio_meta_read(struct btrfs_bio *bbio)
4275 struct extent_buffer *eb = bbio->private;
4276 struct btrfs_fs_info *fs_info = eb->fs_info;
4277 bool uptodate = !bbio->bio.bi_status;
4278 struct folio_iter fi;
4281 eb->read_mirror = bbio->mirror_num;
4284 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
4288 set_extent_buffer_uptodate(eb);
4290 clear_extent_buffer_uptodate(eb);
4291 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4294 bio_for_each_folio_all(fi, &bbio->bio) {
4295 struct folio *folio = fi.folio;
4296 u64 start = eb->start + bio_offset;
4297 u32 len = fi.length;
4300 btrfs_folio_set_uptodate(fs_info, folio, start, len);
4302 btrfs_folio_clear_uptodate(fs_info, folio, start, len);
4307 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
4308 smp_mb__after_atomic();
4309 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
4310 free_extent_buffer(eb);
4312 bio_put(&bbio->bio);
4315 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
4316 struct btrfs_tree_parent_check *check)
4318 struct btrfs_bio *bbio;
4321 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4325 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
4326 * operation, which could potentially still be in flight. In this case
4327 * we simply want to return an error.
4329 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
4332 /* Someone else is already reading the buffer, just wait for it. */
4333 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
4336 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4337 eb->read_mirror = 0;
4338 check_buffer_tree_ref(eb);
4339 atomic_inc(&eb->refs);
4341 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
4342 REQ_OP_READ | REQ_META, eb->fs_info,
4343 end_bbio_meta_read, eb);
4344 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
4345 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
4346 bbio->file_offset = eb->start;
4347 memcpy(&bbio->parent_check, check, sizeof(*check));
4348 if (eb->fs_info->nodesize < PAGE_SIZE) {
4349 ret = bio_add_folio(&bbio->bio, eb->folios[0], eb->len,
4350 eb->start - folio_pos(eb->folios[0]));
4353 int num_folios = num_extent_folios(eb);
4355 for (int i = 0; i < num_folios; i++) {
4356 struct folio *folio = eb->folios[i];
4358 ret = bio_add_folio(&bbio->bio, folio, eb->folio_size, 0);
4362 btrfs_submit_bio(bbio, mirror_num);
4365 if (wait == WAIT_COMPLETE) {
4366 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
4367 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4374 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
4377 btrfs_warn(eb->fs_info,
4378 "access to eb bytenr %llu len %u out of range start %lu len %lu",
4379 eb->start, eb->len, start, len);
4380 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4386 * Check if the [start, start + len) range is valid before reading/writing
4388 * NOTE: @start and @len are offset inside the eb, not logical address.
4390 * Caller should not touch the dst/src memory if this function returns error.
4392 static inline int check_eb_range(const struct extent_buffer *eb,
4393 unsigned long start, unsigned long len)
4395 unsigned long offset;
4397 /* start, start + len should not go beyond eb->len nor overflow */
4398 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
4399 return report_eb_range(eb, start, len);
4404 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4405 unsigned long start, unsigned long len)
4407 const int unit_size = eb->folio_size;
4410 char *dst = (char *)dstv;
4411 unsigned long i = get_eb_folio_index(eb, start);
4413 if (check_eb_range(eb, start, len)) {
4415 * Invalid range hit, reset the memory, so callers won't get
4416 * some random garbage for their uninitialized memory.
4418 memset(dstv, 0, len);
4423 memcpy(dstv, eb->addr + start, len);
4427 offset = get_eb_offset_in_folio(eb, start);
4432 cur = min(len, unit_size - offset);
4433 kaddr = folio_address(eb->folios[i]);
4434 memcpy(dst, kaddr + offset, cur);
4443 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4445 unsigned long start, unsigned long len)
4447 const int unit_size = eb->folio_size;
4450 char __user *dst = (char __user *)dstv;
4451 unsigned long i = get_eb_folio_index(eb, start);
4454 WARN_ON(start > eb->len);
4455 WARN_ON(start + len > eb->start + eb->len);
4458 if (copy_to_user_nofault(dstv, eb->addr + start, len))
4463 offset = get_eb_offset_in_folio(eb, start);
4468 cur = min(len, unit_size - offset);
4469 kaddr = folio_address(eb->folios[i]);
4470 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4484 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4485 unsigned long start, unsigned long len)
4487 const int unit_size = eb->folio_size;
4491 char *ptr = (char *)ptrv;
4492 unsigned long i = get_eb_folio_index(eb, start);
4495 if (check_eb_range(eb, start, len))
4499 return memcmp(ptrv, eb->addr + start, len);
4501 offset = get_eb_offset_in_folio(eb, start);
4504 cur = min(len, unit_size - offset);
4505 kaddr = folio_address(eb->folios[i]);
4506 ret = memcmp(ptr, kaddr + offset, cur);
4519 * Check that the extent buffer is uptodate.
4521 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4522 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4524 static void assert_eb_folio_uptodate(const struct extent_buffer *eb, int i)
4526 struct btrfs_fs_info *fs_info = eb->fs_info;
4527 struct folio *folio = eb->folios[i];
4532 * If we are using the commit root we could potentially clear a page
4533 * Uptodate while we're using the extent buffer that we've previously
4534 * looked up. We don't want to complain in this case, as the page was
4535 * valid before, we just didn't write it out. Instead we want to catch
4536 * the case where we didn't actually read the block properly, which
4537 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4539 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4542 if (fs_info->nodesize < PAGE_SIZE) {
4543 struct folio *folio = eb->folios[0];
4546 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, folio,
4547 eb->start, eb->len)))
4548 btrfs_subpage_dump_bitmap(fs_info, folio, eb->start, eb->len);
4550 WARN_ON(!folio_test_uptodate(folio));
4554 static void __write_extent_buffer(const struct extent_buffer *eb,
4555 const void *srcv, unsigned long start,
4556 unsigned long len, bool use_memmove)
4558 const int unit_size = eb->folio_size;
4562 char *src = (char *)srcv;
4563 unsigned long i = get_eb_folio_index(eb, start);
4564 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4565 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4567 if (check_eb_range(eb, start, len))
4572 memmove(eb->addr + start, srcv, len);
4574 memcpy(eb->addr + start, srcv, len);
4578 offset = get_eb_offset_in_folio(eb, start);
4582 assert_eb_folio_uptodate(eb, i);
4584 cur = min(len, unit_size - offset);
4585 kaddr = folio_address(eb->folios[i]);
4587 memmove(kaddr + offset, src, cur);
4589 memcpy(kaddr + offset, src, cur);
4598 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4599 unsigned long start, unsigned long len)
4601 return __write_extent_buffer(eb, srcv, start, len, false);
4604 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4605 unsigned long start, unsigned long len)
4607 const int unit_size = eb->folio_size;
4608 unsigned long cur = start;
4611 memset(eb->addr + start, c, len);
4615 while (cur < start + len) {
4616 unsigned long index = get_eb_folio_index(eb, cur);
4617 unsigned int offset = get_eb_offset_in_folio(eb, cur);
4618 unsigned int cur_len = min(start + len - cur, unit_size - offset);
4620 assert_eb_folio_uptodate(eb, index);
4621 memset(folio_address(eb->folios[index]) + offset, c, cur_len);
4627 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4630 if (check_eb_range(eb, start, len))
4632 return memset_extent_buffer(eb, 0, start, len);
4635 void copy_extent_buffer_full(const struct extent_buffer *dst,
4636 const struct extent_buffer *src)
4638 const int unit_size = src->folio_size;
4639 unsigned long cur = 0;
4641 ASSERT(dst->len == src->len);
4643 while (cur < src->len) {
4644 unsigned long index = get_eb_folio_index(src, cur);
4645 unsigned long offset = get_eb_offset_in_folio(src, cur);
4646 unsigned long cur_len = min(src->len, unit_size - offset);
4647 void *addr = folio_address(src->folios[index]) + offset;
4649 write_extent_buffer(dst, addr, cur, cur_len);
4655 void copy_extent_buffer(const struct extent_buffer *dst,
4656 const struct extent_buffer *src,
4657 unsigned long dst_offset, unsigned long src_offset,
4660 const int unit_size = dst->folio_size;
4661 u64 dst_len = dst->len;
4665 unsigned long i = get_eb_folio_index(dst, dst_offset);
4667 if (check_eb_range(dst, dst_offset, len) ||
4668 check_eb_range(src, src_offset, len))
4671 WARN_ON(src->len != dst_len);
4673 offset = get_eb_offset_in_folio(dst, dst_offset);
4676 assert_eb_folio_uptodate(dst, i);
4678 cur = min(len, (unsigned long)(unit_size - offset));
4680 kaddr = folio_address(dst->folios[i]);
4681 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4691 * Calculate the folio and offset of the byte containing the given bit number.
4693 * @eb: the extent buffer
4694 * @start: offset of the bitmap item in the extent buffer
4696 * @folio_index: return index of the folio in the extent buffer that contains
4697 * the given bit number
4698 * @folio_offset: return offset into the folio given by folio_index
4700 * This helper hides the ugliness of finding the byte in an extent buffer which
4701 * contains a given bit.
4703 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4704 unsigned long start, unsigned long nr,
4705 unsigned long *folio_index,
4706 size_t *folio_offset)
4708 size_t byte_offset = BIT_BYTE(nr);
4712 * The byte we want is the offset of the extent buffer + the offset of
4713 * the bitmap item in the extent buffer + the offset of the byte in the
4716 offset = start + offset_in_eb_folio(eb, eb->start) + byte_offset;
4718 *folio_index = offset >> eb->folio_shift;
4719 *folio_offset = offset_in_eb_folio(eb, offset);
4723 * Determine whether a bit in a bitmap item is set.
4725 * @eb: the extent buffer
4726 * @start: offset of the bitmap item in the extent buffer
4727 * @nr: bit number to test
4729 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4736 eb_bitmap_offset(eb, start, nr, &i, &offset);
4737 assert_eb_folio_uptodate(eb, i);
4738 kaddr = folio_address(eb->folios[i]);
4739 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4742 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4744 unsigned long index = get_eb_folio_index(eb, bytenr);
4746 if (check_eb_range(eb, bytenr, 1))
4748 return folio_address(eb->folios[index]) + get_eb_offset_in_folio(eb, bytenr);
4752 * Set an area of a bitmap to 1.
4754 * @eb: the extent buffer
4755 * @start: offset of the bitmap item in the extent buffer
4756 * @pos: bit number of the first bit
4757 * @len: number of bits to set
4759 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4760 unsigned long pos, unsigned long len)
4762 unsigned int first_byte = start + BIT_BYTE(pos);
4763 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4764 const bool same_byte = (first_byte == last_byte);
4765 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4769 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4771 /* Handle the first byte. */
4772 kaddr = extent_buffer_get_byte(eb, first_byte);
4777 /* Handle the byte aligned part. */
4778 ASSERT(first_byte + 1 <= last_byte);
4779 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4781 /* Handle the last byte. */
4782 kaddr = extent_buffer_get_byte(eb, last_byte);
4783 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4788 * Clear an area of a bitmap.
4790 * @eb: the extent buffer
4791 * @start: offset of the bitmap item in the extent buffer
4792 * @pos: bit number of the first bit
4793 * @len: number of bits to clear
4795 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4796 unsigned long start, unsigned long pos,
4799 unsigned int first_byte = start + BIT_BYTE(pos);
4800 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4801 const bool same_byte = (first_byte == last_byte);
4802 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4806 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4808 /* Handle the first byte. */
4809 kaddr = extent_buffer_get_byte(eb, first_byte);
4814 /* Handle the byte aligned part. */
4815 ASSERT(first_byte + 1 <= last_byte);
4816 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4818 /* Handle the last byte. */
4819 kaddr = extent_buffer_get_byte(eb, last_byte);
4820 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4823 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4825 unsigned long distance = (src > dst) ? src - dst : dst - src;
4826 return distance < len;
4829 void memcpy_extent_buffer(const struct extent_buffer *dst,
4830 unsigned long dst_offset, unsigned long src_offset,
4833 const int unit_size = dst->folio_size;
4834 unsigned long cur_off = 0;
4836 if (check_eb_range(dst, dst_offset, len) ||
4837 check_eb_range(dst, src_offset, len))
4841 const bool use_memmove = areas_overlap(src_offset, dst_offset, len);
4844 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4846 memcpy(dst->addr + dst_offset, dst->addr + src_offset, len);
4850 while (cur_off < len) {
4851 unsigned long cur_src = cur_off + src_offset;
4852 unsigned long folio_index = get_eb_folio_index(dst, cur_src);
4853 unsigned long folio_off = get_eb_offset_in_folio(dst, cur_src);
4854 unsigned long cur_len = min(src_offset + len - cur_src,
4855 unit_size - folio_off);
4856 void *src_addr = folio_address(dst->folios[folio_index]) + folio_off;
4857 const bool use_memmove = areas_overlap(src_offset + cur_off,
4858 dst_offset + cur_off, cur_len);
4860 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4866 void memmove_extent_buffer(const struct extent_buffer *dst,
4867 unsigned long dst_offset, unsigned long src_offset,
4870 unsigned long dst_end = dst_offset + len - 1;
4871 unsigned long src_end = src_offset + len - 1;
4873 if (check_eb_range(dst, dst_offset, len) ||
4874 check_eb_range(dst, src_offset, len))
4877 if (dst_offset < src_offset) {
4878 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4883 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4888 unsigned long src_i;
4890 size_t dst_off_in_folio;
4891 size_t src_off_in_folio;
4895 src_i = get_eb_folio_index(dst, src_end);
4897 dst_off_in_folio = get_eb_offset_in_folio(dst, dst_end);
4898 src_off_in_folio = get_eb_offset_in_folio(dst, src_end);
4900 cur = min_t(unsigned long, len, src_off_in_folio + 1);
4901 cur = min(cur, dst_off_in_folio + 1);
4903 src_addr = folio_address(dst->folios[src_i]) + src_off_in_folio -
4905 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4908 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4917 #define GANG_LOOKUP_SIZE 16
4918 static struct extent_buffer *get_next_extent_buffer(
4919 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4921 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4922 struct extent_buffer *found = NULL;
4923 u64 page_start = page_offset(page);
4924 u64 cur = page_start;
4926 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4927 lockdep_assert_held(&fs_info->buffer_lock);
4929 while (cur < page_start + PAGE_SIZE) {
4933 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4934 (void **)gang, cur >> fs_info->sectorsize_bits,
4935 min_t(unsigned int, GANG_LOOKUP_SIZE,
4936 PAGE_SIZE / fs_info->nodesize));
4939 for (i = 0; i < ret; i++) {
4940 /* Already beyond page end */
4941 if (gang[i]->start >= page_start + PAGE_SIZE)
4944 if (gang[i]->start >= bytenr) {
4949 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4955 static int try_release_subpage_extent_buffer(struct page *page)
4957 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
4958 u64 cur = page_offset(page);
4959 const u64 end = page_offset(page) + PAGE_SIZE;
4963 struct extent_buffer *eb = NULL;
4966 * Unlike try_release_extent_buffer() which uses folio private
4967 * to grab buffer, for subpage case we rely on radix tree, thus
4968 * we need to ensure radix tree consistency.
4970 * We also want an atomic snapshot of the radix tree, thus go
4971 * with spinlock rather than RCU.
4973 spin_lock(&fs_info->buffer_lock);
4974 eb = get_next_extent_buffer(fs_info, page, cur);
4976 /* No more eb in the page range after or at cur */
4977 spin_unlock(&fs_info->buffer_lock);
4980 cur = eb->start + eb->len;
4983 * The same as try_release_extent_buffer(), to ensure the eb
4984 * won't disappear out from under us.
4986 spin_lock(&eb->refs_lock);
4987 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4988 spin_unlock(&eb->refs_lock);
4989 spin_unlock(&fs_info->buffer_lock);
4992 spin_unlock(&fs_info->buffer_lock);
4995 * If tree ref isn't set then we know the ref on this eb is a
4996 * real ref, so just return, this eb will likely be freed soon
4999 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5000 spin_unlock(&eb->refs_lock);
5005 * Here we don't care about the return value, we will always
5006 * check the folio private at the end. And
5007 * release_extent_buffer() will release the refs_lock.
5009 release_extent_buffer(eb);
5012 * Finally to check if we have cleared folio private, as if we have
5013 * released all ebs in the page, the folio private should be cleared now.
5015 spin_lock(&page->mapping->i_private_lock);
5016 if (!folio_test_private(page_folio(page)))
5020 spin_unlock(&page->mapping->i_private_lock);
5025 int try_release_extent_buffer(struct page *page)
5027 struct folio *folio = page_folio(page);
5028 struct extent_buffer *eb;
5030 if (page_to_fs_info(page)->nodesize < PAGE_SIZE)
5031 return try_release_subpage_extent_buffer(page);
5034 * We need to make sure nobody is changing folio private, as we rely on
5035 * folio private as the pointer to extent buffer.
5037 spin_lock(&page->mapping->i_private_lock);
5038 if (!folio_test_private(folio)) {
5039 spin_unlock(&page->mapping->i_private_lock);
5043 eb = folio_get_private(folio);
5047 * This is a little awful but should be ok, we need to make sure that
5048 * the eb doesn't disappear out from under us while we're looking at
5051 spin_lock(&eb->refs_lock);
5052 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5053 spin_unlock(&eb->refs_lock);
5054 spin_unlock(&page->mapping->i_private_lock);
5057 spin_unlock(&page->mapping->i_private_lock);
5060 * If tree ref isn't set then we know the ref on this eb is a real ref,
5061 * so just return, this page will likely be freed soon anyway.
5063 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5064 spin_unlock(&eb->refs_lock);
5068 return release_extent_buffer(eb);
5072 * Attempt to readahead a child block.
5074 * @fs_info: the fs_info
5075 * @bytenr: bytenr to read
5076 * @owner_root: objectid of the root that owns this eb
5077 * @gen: generation for the uptodate check, can be 0
5078 * @level: level for the eb
5080 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
5081 * normal uptodate check of the eb, without checking the generation. If we have
5082 * to read the block we will not block on anything.
5084 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
5085 u64 bytenr, u64 owner_root, u64 gen, int level)
5087 struct btrfs_tree_parent_check check = {
5092 struct extent_buffer *eb;
5095 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
5099 if (btrfs_buffer_uptodate(eb, gen, 1)) {
5100 free_extent_buffer(eb);
5104 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
5106 free_extent_buffer_stale(eb);
5108 free_extent_buffer(eb);
5112 * Readahead a node's child block.
5114 * @node: parent node we're reading from
5115 * @slot: slot in the parent node for the child we want to read
5117 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
5118 * the slot in the node provided.
5120 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
5122 btrfs_readahead_tree_block(node->fs_info,
5123 btrfs_node_blockptr(node, slot),
5124 btrfs_header_owner(node),
5125 btrfs_node_ptr_generation(node, slot),
5126 btrfs_header_level(node) - 1);