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 const gfp_t gfp = GFP_NOFS | extra_gfp;
685 unsigned int allocated;
687 for (allocated = 0; allocated < nr_pages;) {
688 unsigned int last = allocated;
690 allocated = alloc_pages_bulk_array(gfp, nr_pages, page_array);
691 if (unlikely(allocated == last)) {
692 /* No progress, fail and do cleanup. */
693 for (int i = 0; i < allocated; i++) {
694 __free_page(page_array[i]);
695 page_array[i] = NULL;
704 * Populate needed folios for the extent buffer.
706 * For now, the folios populated are always in order 0 (aka, single page).
708 static int alloc_eb_folio_array(struct extent_buffer *eb, gfp_t extra_gfp)
710 struct page *page_array[INLINE_EXTENT_BUFFER_PAGES] = { 0 };
711 int num_pages = num_extent_pages(eb);
714 ret = btrfs_alloc_page_array(num_pages, page_array, extra_gfp);
718 for (int i = 0; i < num_pages; i++)
719 eb->folios[i] = page_folio(page_array[i]);
720 eb->folio_size = PAGE_SIZE;
721 eb->folio_shift = PAGE_SHIFT;
725 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
726 struct page *page, u64 disk_bytenr,
727 unsigned int pg_offset)
729 struct bio *bio = &bio_ctrl->bbio->bio;
730 struct bio_vec *bvec = bio_last_bvec_all(bio);
731 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
733 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
735 * For compression, all IO should have its logical bytenr set
736 * to the starting bytenr of the compressed extent.
738 return bio->bi_iter.bi_sector == sector;
742 * The contig check requires the following conditions to be met:
744 * 1) The pages are belonging to the same inode
745 * This is implied by the call chain.
747 * 2) The range has adjacent logical bytenr
749 * 3) The range has adjacent file offset
750 * This is required for the usage of btrfs_bio->file_offset.
752 return bio_end_sector(bio) == sector &&
753 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
754 page_offset(page) + pg_offset;
757 static void alloc_new_bio(struct btrfs_inode *inode,
758 struct btrfs_bio_ctrl *bio_ctrl,
759 u64 disk_bytenr, u64 file_offset)
761 struct btrfs_fs_info *fs_info = inode->root->fs_info;
762 struct btrfs_bio *bbio;
764 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
765 bio_ctrl->end_io_func, NULL);
766 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
768 bbio->file_offset = file_offset;
769 bio_ctrl->bbio = bbio;
770 bio_ctrl->len_to_oe_boundary = U32_MAX;
772 /* Limit data write bios to the ordered boundary. */
774 struct btrfs_ordered_extent *ordered;
776 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
778 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
779 ordered->file_offset +
780 ordered->disk_num_bytes - file_offset);
781 bbio->ordered = ordered;
785 * Pick the last added device to support cgroup writeback. For
786 * multi-device file systems this means blk-cgroup policies have
787 * to always be set on the last added/replaced device.
788 * This is a bit odd but has been like that for a long time.
790 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
791 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
796 * @disk_bytenr: logical bytenr where the write will be
797 * @page: page to add to the bio
798 * @size: portion of page that we want to write to
799 * @pg_offset: offset of the new bio or to check whether we are adding
800 * a contiguous page to the previous one
802 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
803 * new one in @bio_ctrl->bbio.
804 * The mirror number for this IO should already be initizlied in
805 * @bio_ctrl->mirror_num.
807 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
808 u64 disk_bytenr, struct page *page,
809 size_t size, unsigned long pg_offset)
811 struct btrfs_inode *inode = page_to_inode(page);
813 ASSERT(pg_offset + size <= PAGE_SIZE);
814 ASSERT(bio_ctrl->end_io_func);
816 if (bio_ctrl->bbio &&
817 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
818 submit_one_bio(bio_ctrl);
823 /* Allocate new bio if needed */
824 if (!bio_ctrl->bbio) {
825 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
826 page_offset(page) + pg_offset);
829 /* Cap to the current ordered extent boundary if there is one. */
830 if (len > bio_ctrl->len_to_oe_boundary) {
831 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
832 ASSERT(is_data_inode(&inode->vfs_inode));
833 len = bio_ctrl->len_to_oe_boundary;
836 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
837 /* bio full: move on to a new one */
838 submit_one_bio(bio_ctrl);
843 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
850 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
851 * sector aligned. alloc_new_bio() then sets it to the end of
852 * our ordered extent for writes into zoned devices.
854 * When len_to_oe_boundary is tracking an ordered extent, we
855 * trust the ordered extent code to align things properly, and
856 * the check above to cap our write to the ordered extent
857 * boundary is correct.
859 * When len_to_oe_boundary is U32_MAX, the cap above would
860 * result in a 4095 byte IO for the last page right before
861 * we hit the bio limit of UINT_MAX. bio_add_page() has all
862 * the checks required to make sure we don't overflow the bio,
863 * and we should just ignore len_to_oe_boundary completely
864 * unless we're using it to track an ordered extent.
866 * It's pretty hard to make a bio sized U32_MAX, but it can
867 * happen when the page cache is able to feed us contiguous
868 * pages for large extents.
870 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
871 bio_ctrl->len_to_oe_boundary -= len;
873 /* Ordered extent boundary: move on to a new bio. */
874 if (bio_ctrl->len_to_oe_boundary == 0)
875 submit_one_bio(bio_ctrl);
879 static int attach_extent_buffer_folio(struct extent_buffer *eb,
881 struct btrfs_subpage *prealloc)
883 struct btrfs_fs_info *fs_info = eb->fs_info;
887 * If the page is mapped to btree inode, we should hold the private
888 * lock to prevent race.
889 * For cloned or dummy extent buffers, their pages are not mapped and
890 * will not race with any other ebs.
893 lockdep_assert_held(&folio->mapping->i_private_lock);
895 if (fs_info->nodesize >= PAGE_SIZE) {
896 if (!folio_test_private(folio))
897 folio_attach_private(folio, eb);
899 WARN_ON(folio_get_private(folio) != eb);
903 /* Already mapped, just free prealloc */
904 if (folio_test_private(folio)) {
905 btrfs_free_subpage(prealloc);
910 /* Has preallocated memory for subpage */
911 folio_attach_private(folio, prealloc);
913 /* Do new allocation to attach subpage */
914 ret = btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_METADATA);
918 int set_page_extent_mapped(struct page *page)
920 return set_folio_extent_mapped(page_folio(page));
923 int set_folio_extent_mapped(struct folio *folio)
925 struct btrfs_fs_info *fs_info;
927 ASSERT(folio->mapping);
929 if (folio_test_private(folio))
932 fs_info = folio_to_fs_info(folio);
934 if (btrfs_is_subpage(fs_info, folio->mapping))
935 return btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_DATA);
937 folio_attach_private(folio, (void *)EXTENT_FOLIO_PRIVATE);
941 void clear_page_extent_mapped(struct page *page)
943 struct folio *folio = page_folio(page);
944 struct btrfs_fs_info *fs_info;
946 ASSERT(page->mapping);
948 if (!folio_test_private(folio))
951 fs_info = page_to_fs_info(page);
952 if (btrfs_is_subpage(fs_info, page->mapping))
953 return btrfs_detach_subpage(fs_info, folio);
955 folio_detach_private(folio);
958 static struct extent_map *__get_extent_map(struct inode *inode, struct page *page,
959 u64 start, u64 len, struct extent_map **em_cached)
961 struct extent_map *em;
967 if (extent_map_in_tree(em) && start >= em->start &&
968 start < extent_map_end(em)) {
969 refcount_inc(&em->refs);
977 em = btrfs_get_extent(BTRFS_I(inode), page, start, len);
980 refcount_inc(&em->refs);
986 * basic readpage implementation. Locked extent state structs are inserted
987 * into the tree that are removed when the IO is done (by the end_io
989 * XXX JDM: This needs looking at to ensure proper page locking
990 * return 0 on success, otherwise return error
992 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
993 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
995 struct inode *inode = page->mapping->host;
996 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
997 u64 start = page_offset(page);
998 const u64 end = start + PAGE_SIZE - 1;
1001 u64 last_byte = i_size_read(inode);
1003 struct extent_map *em;
1005 size_t pg_offset = 0;
1007 size_t blocksize = fs_info->sectorsize;
1008 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1010 ret = set_page_extent_mapped(page);
1012 unlock_extent(tree, start, end, NULL);
1017 if (page->index == last_byte >> PAGE_SHIFT) {
1018 size_t zero_offset = offset_in_page(last_byte);
1021 iosize = PAGE_SIZE - zero_offset;
1022 memzero_page(page, zero_offset, iosize);
1025 bio_ctrl->end_io_func = end_bbio_data_read;
1026 begin_page_read(fs_info, page);
1027 while (cur <= end) {
1028 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1029 bool force_bio_submit = false;
1032 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1033 if (cur >= last_byte) {
1034 iosize = PAGE_SIZE - pg_offset;
1035 memzero_page(page, pg_offset, iosize);
1036 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1037 end_page_read(page, true, cur, iosize);
1040 em = __get_extent_map(inode, page, cur, end - cur + 1, em_cached);
1042 unlock_extent(tree, cur, end, NULL);
1043 end_page_read(page, false, cur, end + 1 - cur);
1046 extent_offset = cur - em->start;
1047 BUG_ON(extent_map_end(em) <= cur);
1050 compress_type = extent_map_compression(em);
1052 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1053 iosize = ALIGN(iosize, blocksize);
1054 if (compress_type != BTRFS_COMPRESS_NONE)
1055 disk_bytenr = em->block_start;
1057 disk_bytenr = em->block_start + extent_offset;
1058 block_start = em->block_start;
1059 if (em->flags & EXTENT_FLAG_PREALLOC)
1060 block_start = EXTENT_MAP_HOLE;
1063 * If we have a file range that points to a compressed extent
1064 * and it's followed by a consecutive file range that points
1065 * to the same compressed extent (possibly with a different
1066 * offset and/or length, so it either points to the whole extent
1067 * or only part of it), we must make sure we do not submit a
1068 * single bio to populate the pages for the 2 ranges because
1069 * this makes the compressed extent read zero out the pages
1070 * belonging to the 2nd range. Imagine the following scenario:
1073 * [0 - 8K] [8K - 24K]
1076 * points to extent X, points to extent X,
1077 * offset 4K, length of 8K offset 0, length 16K
1079 * [extent X, compressed length = 4K uncompressed length = 16K]
1081 * If the bio to read the compressed extent covers both ranges,
1082 * it will decompress extent X into the pages belonging to the
1083 * first range and then it will stop, zeroing out the remaining
1084 * pages that belong to the other range that points to extent X.
1085 * So here we make sure we submit 2 bios, one for the first
1086 * range and another one for the third range. Both will target
1087 * the same physical extent from disk, but we can't currently
1088 * make the compressed bio endio callback populate the pages
1089 * for both ranges because each compressed bio is tightly
1090 * coupled with a single extent map, and each range can have
1091 * an extent map with a different offset value relative to the
1092 * uncompressed data of our extent and different lengths. This
1093 * is a corner case so we prioritize correctness over
1094 * non-optimal behavior (submitting 2 bios for the same extent).
1096 if (compress_type != BTRFS_COMPRESS_NONE &&
1097 prev_em_start && *prev_em_start != (u64)-1 &&
1098 *prev_em_start != em->start)
1099 force_bio_submit = true;
1102 *prev_em_start = em->start;
1104 free_extent_map(em);
1107 /* we've found a hole, just zero and go on */
1108 if (block_start == EXTENT_MAP_HOLE) {
1109 memzero_page(page, pg_offset, iosize);
1111 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1112 end_page_read(page, true, cur, iosize);
1114 pg_offset += iosize;
1117 /* the get_extent function already copied into the page */
1118 if (block_start == EXTENT_MAP_INLINE) {
1119 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1120 end_page_read(page, true, cur, iosize);
1122 pg_offset += iosize;
1126 if (bio_ctrl->compress_type != compress_type) {
1127 submit_one_bio(bio_ctrl);
1128 bio_ctrl->compress_type = compress_type;
1131 if (force_bio_submit)
1132 submit_one_bio(bio_ctrl);
1133 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1136 pg_offset += iosize;
1142 int btrfs_read_folio(struct file *file, struct folio *folio)
1144 struct page *page = &folio->page;
1145 struct btrfs_inode *inode = page_to_inode(page);
1146 u64 start = page_offset(page);
1147 u64 end = start + PAGE_SIZE - 1;
1148 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1149 struct extent_map *em_cached = NULL;
1152 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1154 ret = btrfs_do_readpage(page, &em_cached, &bio_ctrl, NULL);
1155 free_extent_map(em_cached);
1158 * If btrfs_do_readpage() failed we will want to submit the assembled
1159 * bio to do the cleanup.
1161 submit_one_bio(&bio_ctrl);
1165 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1167 struct extent_map **em_cached,
1168 struct btrfs_bio_ctrl *bio_ctrl,
1171 struct btrfs_inode *inode = page_to_inode(pages[0]);
1176 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1178 for (index = 0; index < nr_pages; index++) {
1179 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1181 put_page(pages[index]);
1186 * helper for __extent_writepage, doing all of the delayed allocation setup.
1188 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1189 * to write the page (copy into inline extent). In this case the IO has
1190 * been started and the page is already unlocked.
1192 * This returns 0 if all went well (page still locked)
1193 * This returns < 0 if there were errors (page still locked)
1195 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1196 struct page *page, struct writeback_control *wbc)
1198 const u64 page_start = page_offset(page);
1199 const u64 page_end = page_start + PAGE_SIZE - 1;
1200 u64 delalloc_start = page_start;
1201 u64 delalloc_end = page_end;
1202 u64 delalloc_to_write = 0;
1205 while (delalloc_start < page_end) {
1206 delalloc_end = page_end;
1207 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1208 &delalloc_start, &delalloc_end)) {
1209 delalloc_start = delalloc_end + 1;
1213 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1218 delalloc_start = delalloc_end + 1;
1222 * delalloc_end is already one less than the total length, so
1223 * we don't subtract one from PAGE_SIZE
1225 delalloc_to_write +=
1226 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1229 * If btrfs_run_dealloc_range() already started I/O and unlocked
1230 * the pages, we just need to account for them here.
1233 wbc->nr_to_write -= delalloc_to_write;
1237 if (wbc->nr_to_write < delalloc_to_write) {
1240 if (delalloc_to_write < thresh * 2)
1241 thresh = delalloc_to_write;
1242 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1250 * Find the first byte we need to write.
1252 * For subpage, one page can contain several sectors, and
1253 * __extent_writepage_io() will just grab all extent maps in the page
1254 * range and try to submit all non-inline/non-compressed extents.
1256 * This is a big problem for subpage, we shouldn't re-submit already written
1258 * This function will lookup subpage dirty bit to find which range we really
1261 * Return the next dirty range in [@start, @end).
1262 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1264 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1265 struct page *page, u64 *start, u64 *end)
1267 struct folio *folio = page_folio(page);
1268 struct btrfs_subpage *subpage = folio_get_private(folio);
1269 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1270 u64 orig_start = *start;
1271 /* Declare as unsigned long so we can use bitmap ops */
1272 unsigned long flags;
1273 int range_start_bit;
1277 * For regular sector size == page size case, since one page only
1278 * contains one sector, we return the page offset directly.
1280 if (!btrfs_is_subpage(fs_info, page->mapping)) {
1281 *start = page_offset(page);
1282 *end = page_offset(page) + PAGE_SIZE;
1286 range_start_bit = spi->dirty_offset +
1287 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1289 /* We should have the page locked, but just in case */
1290 spin_lock_irqsave(&subpage->lock, flags);
1291 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1292 spi->dirty_offset + spi->bitmap_nr_bits);
1293 spin_unlock_irqrestore(&subpage->lock, flags);
1295 range_start_bit -= spi->dirty_offset;
1296 range_end_bit -= spi->dirty_offset;
1298 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1299 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1303 * helper for __extent_writepage. This calls the writepage start hooks,
1304 * and does the loop to map the page into extents and bios.
1306 * We return 1 if the IO is started and the page is unlocked,
1307 * 0 if all went well (page still locked)
1308 * < 0 if there were errors (page still locked)
1310 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1312 struct btrfs_bio_ctrl *bio_ctrl,
1316 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1317 u64 cur = page_offset(page);
1318 u64 end = cur + PAGE_SIZE - 1;
1321 struct extent_map *em;
1325 ret = btrfs_writepage_cow_fixup(page);
1327 /* Fixup worker will requeue */
1328 redirty_page_for_writepage(bio_ctrl->wbc, page);
1333 bio_ctrl->end_io_func = end_bbio_data_write;
1334 while (cur <= end) {
1335 u32 len = end - cur + 1;
1338 u64 dirty_range_start = cur;
1339 u64 dirty_range_end;
1342 if (cur >= i_size) {
1343 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1346 * This range is beyond i_size, thus we don't need to
1347 * bother writing back.
1348 * But we still need to clear the dirty subpage bit, or
1349 * the next time the page gets dirtied, we will try to
1350 * writeback the sectors with subpage dirty bits,
1351 * causing writeback without ordered extent.
1353 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, len);
1357 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1359 if (cur < dirty_range_start) {
1360 cur = dirty_range_start;
1364 em = btrfs_get_extent(inode, NULL, cur, len);
1366 ret = PTR_ERR_OR_ZERO(em);
1370 extent_offset = cur - em->start;
1371 em_end = extent_map_end(em);
1372 ASSERT(cur <= em_end);
1374 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1375 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1377 block_start = em->block_start;
1378 disk_bytenr = em->block_start + extent_offset;
1380 ASSERT(!extent_map_is_compressed(em));
1381 ASSERT(block_start != EXTENT_MAP_HOLE);
1382 ASSERT(block_start != EXTENT_MAP_INLINE);
1385 * Note that em_end from extent_map_end() and dirty_range_end from
1386 * find_next_dirty_byte() are all exclusive
1388 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1389 free_extent_map(em);
1392 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1393 if (!PageWriteback(page)) {
1394 btrfs_err(inode->root->fs_info,
1395 "page %lu not writeback, cur %llu end %llu",
1396 page->index, cur, end);
1400 * Although the PageDirty bit is cleared before entering this
1401 * function, subpage dirty bit is not cleared.
1402 * So clear subpage dirty bit here so next time we won't submit
1403 * page for range already written to disk.
1405 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, iosize);
1407 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1408 cur - page_offset(page));
1413 btrfs_folio_assert_not_dirty(fs_info, page_folio(page));
1419 * If we finish without problem, we should not only clear page dirty,
1420 * but also empty subpage dirty bits
1427 * the writepage semantics are similar to regular writepage. extent
1428 * records are inserted to lock ranges in the tree, and as dirty areas
1429 * are found, they are marked writeback. Then the lock bits are removed
1430 * and the end_io handler clears the writeback ranges
1432 * Return 0 if everything goes well.
1433 * Return <0 for error.
1435 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1437 struct folio *folio = page_folio(page);
1438 struct inode *inode = page->mapping->host;
1439 const u64 page_start = page_offset(page);
1443 loff_t i_size = i_size_read(inode);
1444 unsigned long end_index = i_size >> PAGE_SHIFT;
1446 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1448 WARN_ON(!PageLocked(page));
1450 pg_offset = offset_in_page(i_size);
1451 if (page->index > end_index ||
1452 (page->index == end_index && !pg_offset)) {
1453 folio_invalidate(folio, 0, folio_size(folio));
1454 folio_unlock(folio);
1458 if (page->index == end_index)
1459 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1461 ret = set_page_extent_mapped(page);
1465 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1471 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1475 bio_ctrl->wbc->nr_to_write--;
1479 /* make sure the mapping tag for page dirty gets cleared */
1480 set_page_writeback(page);
1481 end_page_writeback(page);
1484 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1486 mapping_set_error(page->mapping, ret);
1493 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1495 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1496 TASK_UNINTERRUPTIBLE);
1500 * Lock extent buffer status and pages for writeback.
1502 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1503 * extent buffer is not dirty)
1504 * Return %true is the extent buffer is submitted to bio.
1506 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1507 struct writeback_control *wbc)
1509 struct btrfs_fs_info *fs_info = eb->fs_info;
1512 btrfs_tree_lock(eb);
1513 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1514 btrfs_tree_unlock(eb);
1515 if (wbc->sync_mode != WB_SYNC_ALL)
1517 wait_on_extent_buffer_writeback(eb);
1518 btrfs_tree_lock(eb);
1522 * We need to do this to prevent races in people who check if the eb is
1523 * under IO since we can end up having no IO bits set for a short period
1526 spin_lock(&eb->refs_lock);
1527 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1528 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1529 spin_unlock(&eb->refs_lock);
1530 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1531 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1533 fs_info->dirty_metadata_batch);
1536 spin_unlock(&eb->refs_lock);
1538 btrfs_tree_unlock(eb);
1542 static void set_btree_ioerr(struct extent_buffer *eb)
1544 struct btrfs_fs_info *fs_info = eb->fs_info;
1546 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1549 * A read may stumble upon this buffer later, make sure that it gets an
1550 * error and knows there was an error.
1552 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1555 * We need to set the mapping with the io error as well because a write
1556 * error will flip the file system readonly, and then syncfs() will
1557 * return a 0 because we are readonly if we don't modify the err seq for
1560 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1563 * If writeback for a btree extent that doesn't belong to a log tree
1564 * failed, increment the counter transaction->eb_write_errors.
1565 * We do this because while the transaction is running and before it's
1566 * committing (when we call filemap_fdata[write|wait]_range against
1567 * the btree inode), we might have
1568 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1569 * returns an error or an error happens during writeback, when we're
1570 * committing the transaction we wouldn't know about it, since the pages
1571 * can be no longer dirty nor marked anymore for writeback (if a
1572 * subsequent modification to the extent buffer didn't happen before the
1573 * transaction commit), which makes filemap_fdata[write|wait]_range not
1574 * able to find the pages tagged with SetPageError at transaction
1575 * commit time. So if this happens we must abort the transaction,
1576 * otherwise we commit a super block with btree roots that point to
1577 * btree nodes/leafs whose content on disk is invalid - either garbage
1578 * or the content of some node/leaf from a past generation that got
1579 * cowed or deleted and is no longer valid.
1581 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1582 * not be enough - we need to distinguish between log tree extents vs
1583 * non-log tree extents, and the next filemap_fdatawait_range() call
1584 * will catch and clear such errors in the mapping - and that call might
1585 * be from a log sync and not from a transaction commit. Also, checking
1586 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1587 * not done and would not be reliable - the eb might have been released
1588 * from memory and reading it back again means that flag would not be
1589 * set (since it's a runtime flag, not persisted on disk).
1591 * Using the flags below in the btree inode also makes us achieve the
1592 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1593 * writeback for all dirty pages and before filemap_fdatawait_range()
1594 * is called, the writeback for all dirty pages had already finished
1595 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1596 * filemap_fdatawait_range() would return success, as it could not know
1597 * that writeback errors happened (the pages were no longer tagged for
1600 switch (eb->log_index) {
1602 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1605 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1608 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1611 BUG(); /* unexpected, logic error */
1616 * The endio specific version which won't touch any unsafe spinlock in endio
1619 static struct extent_buffer *find_extent_buffer_nolock(
1620 struct btrfs_fs_info *fs_info, u64 start)
1622 struct extent_buffer *eb;
1625 eb = radix_tree_lookup(&fs_info->buffer_radix,
1626 start >> fs_info->sectorsize_bits);
1627 if (eb && atomic_inc_not_zero(&eb->refs)) {
1635 static void end_bbio_meta_write(struct btrfs_bio *bbio)
1637 struct extent_buffer *eb = bbio->private;
1638 struct btrfs_fs_info *fs_info = eb->fs_info;
1639 bool uptodate = !bbio->bio.bi_status;
1640 struct folio_iter fi;
1644 set_btree_ioerr(eb);
1646 bio_for_each_folio_all(fi, &bbio->bio) {
1647 u64 start = eb->start + bio_offset;
1648 struct folio *folio = fi.folio;
1649 u32 len = fi.length;
1651 btrfs_folio_clear_writeback(fs_info, folio, start, len);
1655 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1656 smp_mb__after_atomic();
1657 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1659 bio_put(&bbio->bio);
1662 static void prepare_eb_write(struct extent_buffer *eb)
1665 unsigned long start;
1668 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1670 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1671 nritems = btrfs_header_nritems(eb);
1672 if (btrfs_header_level(eb) > 0) {
1673 end = btrfs_node_key_ptr_offset(eb, nritems);
1674 memzero_extent_buffer(eb, end, eb->len - end);
1678 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1680 start = btrfs_item_nr_offset(eb, nritems);
1681 end = btrfs_item_nr_offset(eb, 0);
1683 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1685 end += btrfs_item_offset(eb, nritems - 1);
1686 memzero_extent_buffer(eb, start, end - start);
1690 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1691 struct writeback_control *wbc)
1693 struct btrfs_fs_info *fs_info = eb->fs_info;
1694 struct btrfs_bio *bbio;
1696 prepare_eb_write(eb);
1698 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1699 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1700 eb->fs_info, end_bbio_meta_write, eb);
1701 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1702 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1703 wbc_init_bio(wbc, &bbio->bio);
1704 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1705 bbio->file_offset = eb->start;
1706 if (fs_info->nodesize < PAGE_SIZE) {
1707 struct folio *folio = eb->folios[0];
1711 btrfs_subpage_set_writeback(fs_info, folio, eb->start, eb->len);
1712 if (btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start,
1714 folio_clear_dirty_for_io(folio);
1717 ret = bio_add_folio(&bbio->bio, folio, eb->len,
1718 eb->start - folio_pos(folio));
1720 wbc_account_cgroup_owner(wbc, folio_page(folio, 0), eb->len);
1721 folio_unlock(folio);
1723 int num_folios = num_extent_folios(eb);
1725 for (int i = 0; i < num_folios; i++) {
1726 struct folio *folio = eb->folios[i];
1730 folio_clear_dirty_for_io(folio);
1731 folio_start_writeback(folio);
1732 ret = bio_add_folio(&bbio->bio, folio, eb->folio_size, 0);
1734 wbc_account_cgroup_owner(wbc, folio_page(folio, 0),
1736 wbc->nr_to_write -= folio_nr_pages(folio);
1737 folio_unlock(folio);
1740 btrfs_submit_bio(bbio, 0);
1744 * Submit one subpage btree page.
1746 * The main difference to submit_eb_page() is:
1748 * For subpage, we don't rely on page locking at all.
1751 * We only flush bio if we may be unable to fit current extent buffers into
1754 * Return >=0 for the number of submitted extent buffers.
1755 * Return <0 for fatal error.
1757 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1759 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
1760 struct folio *folio = page_folio(page);
1762 u64 page_start = page_offset(page);
1764 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1766 /* Lock and write each dirty extent buffers in the range */
1767 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1768 struct btrfs_subpage *subpage = folio_get_private(folio);
1769 struct extent_buffer *eb;
1770 unsigned long flags;
1774 * Take private lock to ensure the subpage won't be detached
1777 spin_lock(&page->mapping->i_private_lock);
1778 if (!folio_test_private(folio)) {
1779 spin_unlock(&page->mapping->i_private_lock);
1782 spin_lock_irqsave(&subpage->lock, flags);
1783 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1784 subpage->bitmaps)) {
1785 spin_unlock_irqrestore(&subpage->lock, flags);
1786 spin_unlock(&page->mapping->i_private_lock);
1791 start = page_start + bit_start * fs_info->sectorsize;
1792 bit_start += sectors_per_node;
1795 * Here we just want to grab the eb without touching extra
1796 * spin locks, so call find_extent_buffer_nolock().
1798 eb = find_extent_buffer_nolock(fs_info, start);
1799 spin_unlock_irqrestore(&subpage->lock, flags);
1800 spin_unlock(&page->mapping->i_private_lock);
1803 * The eb has already reached 0 refs thus find_extent_buffer()
1804 * doesn't return it. We don't need to write back such eb
1810 if (lock_extent_buffer_for_io(eb, wbc)) {
1811 write_one_eb(eb, wbc);
1814 free_extent_buffer(eb);
1820 * Submit all page(s) of one extent buffer.
1822 * @page: the page of one extent buffer
1823 * @eb_context: to determine if we need to submit this page, if current page
1824 * belongs to this eb, we don't need to submit
1826 * The caller should pass each page in their bytenr order, and here we use
1827 * @eb_context to determine if we have submitted pages of one extent buffer.
1829 * If we have, we just skip until we hit a new page that doesn't belong to
1830 * current @eb_context.
1832 * If not, we submit all the page(s) of the extent buffer.
1834 * Return >0 if we have submitted the extent buffer successfully.
1835 * Return 0 if we don't need to submit the page, as it's already submitted by
1837 * Return <0 for fatal error.
1839 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1841 struct writeback_control *wbc = ctx->wbc;
1842 struct address_space *mapping = page->mapping;
1843 struct folio *folio = page_folio(page);
1844 struct extent_buffer *eb;
1847 if (!folio_test_private(folio))
1850 if (page_to_fs_info(page)->nodesize < PAGE_SIZE)
1851 return submit_eb_subpage(page, wbc);
1853 spin_lock(&mapping->i_private_lock);
1854 if (!folio_test_private(folio)) {
1855 spin_unlock(&mapping->i_private_lock);
1859 eb = folio_get_private(folio);
1862 * Shouldn't happen and normally this would be a BUG_ON but no point
1863 * crashing the machine for something we can survive anyway.
1866 spin_unlock(&mapping->i_private_lock);
1870 if (eb == ctx->eb) {
1871 spin_unlock(&mapping->i_private_lock);
1874 ret = atomic_inc_not_zero(&eb->refs);
1875 spin_unlock(&mapping->i_private_lock);
1881 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1885 free_extent_buffer(eb);
1889 if (!lock_extent_buffer_for_io(eb, wbc)) {
1890 free_extent_buffer(eb);
1893 /* Implies write in zoned mode. */
1894 if (ctx->zoned_bg) {
1895 /* Mark the last eb in the block group. */
1896 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1897 ctx->zoned_bg->meta_write_pointer += eb->len;
1899 write_one_eb(eb, wbc);
1900 free_extent_buffer(eb);
1904 int btree_write_cache_pages(struct address_space *mapping,
1905 struct writeback_control *wbc)
1907 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1908 struct btrfs_fs_info *fs_info = inode_to_fs_info(mapping->host);
1911 int nr_to_write_done = 0;
1912 struct folio_batch fbatch;
1913 unsigned int nr_folios;
1915 pgoff_t end; /* Inclusive */
1919 folio_batch_init(&fbatch);
1920 if (wbc->range_cyclic) {
1921 index = mapping->writeback_index; /* Start from prev offset */
1924 * Start from the beginning does not need to cycle over the
1925 * range, mark it as scanned.
1927 scanned = (index == 0);
1929 index = wbc->range_start >> PAGE_SHIFT;
1930 end = wbc->range_end >> PAGE_SHIFT;
1933 if (wbc->sync_mode == WB_SYNC_ALL)
1934 tag = PAGECACHE_TAG_TOWRITE;
1936 tag = PAGECACHE_TAG_DIRTY;
1937 btrfs_zoned_meta_io_lock(fs_info);
1939 if (wbc->sync_mode == WB_SYNC_ALL)
1940 tag_pages_for_writeback(mapping, index, end);
1941 while (!done && !nr_to_write_done && (index <= end) &&
1942 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1946 for (i = 0; i < nr_folios; i++) {
1947 struct folio *folio = fbatch.folios[i];
1949 ret = submit_eb_page(&folio->page, &ctx);
1958 * the filesystem may choose to bump up nr_to_write.
1959 * We have to make sure to honor the new nr_to_write
1962 nr_to_write_done = wbc->nr_to_write <= 0;
1964 folio_batch_release(&fbatch);
1967 if (!scanned && !done) {
1969 * We hit the last page and there is more work to be done: wrap
1970 * back to the start of the file
1977 * If something went wrong, don't allow any metadata write bio to be
1980 * This would prevent use-after-free if we had dirty pages not
1981 * cleaned up, which can still happen by fuzzed images.
1984 * Allowing existing tree block to be allocated for other trees.
1986 * - Log tree operations
1987 * Exiting tree blocks get allocated to log tree, bumps its
1988 * generation, then get cleaned in tree re-balance.
1989 * Such tree block will not be written back, since it's clean,
1990 * thus no WRITTEN flag set.
1991 * And after log writes back, this tree block is not traced by
1992 * any dirty extent_io_tree.
1994 * - Offending tree block gets re-dirtied from its original owner
1995 * Since it has bumped generation, no WRITTEN flag, it can be
1996 * reused without COWing. This tree block will not be traced
1997 * by btrfs_transaction::dirty_pages.
1999 * Now such dirty tree block will not be cleaned by any dirty
2000 * extent io tree. Thus we don't want to submit such wild eb
2001 * if the fs already has error.
2003 * We can get ret > 0 from submit_extent_page() indicating how many ebs
2004 * were submitted. Reset it to 0 to avoid false alerts for the caller.
2008 if (!ret && BTRFS_FS_ERROR(fs_info))
2012 btrfs_put_block_group(ctx.zoned_bg);
2013 btrfs_zoned_meta_io_unlock(fs_info);
2018 * Walk the list of dirty pages of the given address space and write all of them.
2020 * @mapping: address space structure to write
2021 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2022 * @bio_ctrl: holds context for the write, namely the bio
2024 * If a page is already under I/O, write_cache_pages() skips it, even
2025 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2026 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2027 * and msync() need to guarantee that all the data which was dirty at the time
2028 * the call was made get new I/O started against them. If wbc->sync_mode is
2029 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2030 * existing IO to complete.
2032 static int extent_write_cache_pages(struct address_space *mapping,
2033 struct btrfs_bio_ctrl *bio_ctrl)
2035 struct writeback_control *wbc = bio_ctrl->wbc;
2036 struct inode *inode = mapping->host;
2039 int nr_to_write_done = 0;
2040 struct folio_batch fbatch;
2041 unsigned int nr_folios;
2043 pgoff_t end; /* Inclusive */
2045 int range_whole = 0;
2050 * We have to hold onto the inode so that ordered extents can do their
2051 * work when the IO finishes. The alternative to this is failing to add
2052 * an ordered extent if the igrab() fails there and that is a huge pain
2053 * to deal with, so instead just hold onto the inode throughout the
2054 * writepages operation. If it fails here we are freeing up the inode
2055 * anyway and we'd rather not waste our time writing out stuff that is
2056 * going to be truncated anyway.
2061 folio_batch_init(&fbatch);
2062 if (wbc->range_cyclic) {
2063 index = mapping->writeback_index; /* Start from prev offset */
2066 * Start from the beginning does not need to cycle over the
2067 * range, mark it as scanned.
2069 scanned = (index == 0);
2071 index = wbc->range_start >> PAGE_SHIFT;
2072 end = wbc->range_end >> PAGE_SHIFT;
2073 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2079 * We do the tagged writepage as long as the snapshot flush bit is set
2080 * and we are the first one who do the filemap_flush() on this inode.
2082 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2083 * not race in and drop the bit.
2085 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2086 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2087 &BTRFS_I(inode)->runtime_flags))
2088 wbc->tagged_writepages = 1;
2090 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2091 tag = PAGECACHE_TAG_TOWRITE;
2093 tag = PAGECACHE_TAG_DIRTY;
2095 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2096 tag_pages_for_writeback(mapping, index, end);
2098 while (!done && !nr_to_write_done && (index <= end) &&
2099 (nr_folios = filemap_get_folios_tag(mapping, &index,
2100 end, tag, &fbatch))) {
2103 for (i = 0; i < nr_folios; i++) {
2104 struct folio *folio = fbatch.folios[i];
2106 done_index = folio_next_index(folio);
2108 * At this point we hold neither the i_pages lock nor
2109 * the page lock: the page may be truncated or
2110 * invalidated (changing page->mapping to NULL),
2111 * or even swizzled back from swapper_space to
2112 * tmpfs file mapping
2114 if (!folio_trylock(folio)) {
2115 submit_write_bio(bio_ctrl, 0);
2119 if (unlikely(folio->mapping != mapping)) {
2120 folio_unlock(folio);
2124 if (!folio_test_dirty(folio)) {
2125 /* Someone wrote it for us. */
2126 folio_unlock(folio);
2130 if (wbc->sync_mode != WB_SYNC_NONE) {
2131 if (folio_test_writeback(folio))
2132 submit_write_bio(bio_ctrl, 0);
2133 folio_wait_writeback(folio);
2136 if (folio_test_writeback(folio) ||
2137 !folio_clear_dirty_for_io(folio)) {
2138 folio_unlock(folio);
2142 ret = __extent_writepage(&folio->page, bio_ctrl);
2149 * The filesystem may choose to bump up nr_to_write.
2150 * We have to make sure to honor the new nr_to_write
2153 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2154 wbc->nr_to_write <= 0);
2156 folio_batch_release(&fbatch);
2159 if (!scanned && !done) {
2161 * We hit the last page and there is more work to be done: wrap
2162 * back to the start of the file
2168 * If we're looping we could run into a page that is locked by a
2169 * writer and that writer could be waiting on writeback for a
2170 * page in our current bio, and thus deadlock, so flush the
2173 submit_write_bio(bio_ctrl, 0);
2177 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2178 mapping->writeback_index = done_index;
2180 btrfs_add_delayed_iput(BTRFS_I(inode));
2185 * Submit the pages in the range to bio for call sites which delalloc range has
2186 * already been ran (aka, ordered extent inserted) and all pages are still
2189 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2190 u64 start, u64 end, struct writeback_control *wbc,
2193 bool found_error = false;
2195 struct address_space *mapping = inode->i_mapping;
2196 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2197 const u32 sectorsize = fs_info->sectorsize;
2198 loff_t i_size = i_size_read(inode);
2200 struct btrfs_bio_ctrl bio_ctrl = {
2202 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2205 if (wbc->no_cgroup_owner)
2206 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2208 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2210 while (cur <= end) {
2211 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2212 u32 cur_len = cur_end + 1 - cur;
2216 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2217 ASSERT(PageLocked(page));
2218 if (pages_dirty && page != locked_page) {
2219 ASSERT(PageDirty(page));
2220 clear_page_dirty_for_io(page);
2223 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2228 /* Make sure the mapping tag for page dirty gets cleared. */
2230 set_page_writeback(page);
2231 end_page_writeback(page);
2234 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2235 cur, cur_len, !ret);
2236 mapping_set_error(page->mapping, ret);
2238 btrfs_folio_unlock_writer(fs_info, page_folio(page), cur, cur_len);
2246 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2249 int extent_writepages(struct address_space *mapping,
2250 struct writeback_control *wbc)
2252 struct inode *inode = mapping->host;
2254 struct btrfs_bio_ctrl bio_ctrl = {
2256 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2260 * Allow only a single thread to do the reloc work in zoned mode to
2261 * protect the write pointer updates.
2263 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2264 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2265 submit_write_bio(&bio_ctrl, ret);
2266 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2270 void extent_readahead(struct readahead_control *rac)
2272 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2273 struct page *pagepool[16];
2274 struct extent_map *em_cached = NULL;
2275 u64 prev_em_start = (u64)-1;
2278 while ((nr = readahead_page_batch(rac, pagepool))) {
2279 u64 contig_start = readahead_pos(rac);
2280 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2282 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2283 &em_cached, &bio_ctrl, &prev_em_start);
2287 free_extent_map(em_cached);
2288 submit_one_bio(&bio_ctrl);
2292 * basic invalidate_folio code, this waits on any locked or writeback
2293 * ranges corresponding to the folio, and then deletes any extent state
2294 * records from the tree
2296 int extent_invalidate_folio(struct extent_io_tree *tree,
2297 struct folio *folio, size_t offset)
2299 struct extent_state *cached_state = NULL;
2300 u64 start = folio_pos(folio);
2301 u64 end = start + folio_size(folio) - 1;
2302 size_t blocksize = folio_to_fs_info(folio)->sectorsize;
2304 /* This function is only called for the btree inode */
2305 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2307 start += ALIGN(offset, blocksize);
2311 lock_extent(tree, start, end, &cached_state);
2312 folio_wait_writeback(folio);
2315 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2316 * so here we only need to unlock the extent range to free any
2317 * existing extent state.
2319 unlock_extent(tree, start, end, &cached_state);
2324 * a helper for release_folio, this tests for areas of the page that
2325 * are locked or under IO and drops the related state bits if it is safe
2328 static int try_release_extent_state(struct extent_io_tree *tree,
2329 struct page *page, gfp_t mask)
2331 u64 start = page_offset(page);
2332 u64 end = start + PAGE_SIZE - 1;
2335 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2338 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2339 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS |
2340 EXTENT_QGROUP_RESERVED);
2343 * At this point we can safely clear everything except the
2344 * locked bit, the nodatasum bit and the delalloc new bit.
2345 * The delalloc new bit will be cleared by ordered extent
2348 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2350 /* if clear_extent_bit failed for enomem reasons,
2351 * we can't allow the release to continue.
2362 * a helper for release_folio. As long as there are no locked extents
2363 * in the range corresponding to the page, both state records and extent
2364 * map records are removed
2366 int try_release_extent_mapping(struct page *page, gfp_t mask)
2368 struct extent_map *em;
2369 u64 start = page_offset(page);
2370 u64 end = start + PAGE_SIZE - 1;
2371 struct btrfs_inode *btrfs_inode = page_to_inode(page);
2372 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2373 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2375 if (gfpflags_allow_blocking(mask) &&
2376 page->mapping->host->i_size > SZ_16M) {
2378 while (start <= end) {
2379 struct btrfs_fs_info *fs_info;
2382 len = end - start + 1;
2383 write_lock(&map->lock);
2384 em = lookup_extent_mapping(map, start, len);
2386 write_unlock(&map->lock);
2389 if ((em->flags & EXTENT_FLAG_PINNED) ||
2390 em->start != start) {
2391 write_unlock(&map->lock);
2392 free_extent_map(em);
2395 if (test_range_bit_exists(tree, em->start,
2396 extent_map_end(em) - 1,
2400 * If it's not in the list of modified extents, used
2401 * by a fast fsync, we can remove it. If it's being
2402 * logged we can safely remove it since fsync took an
2403 * extra reference on the em.
2405 if (list_empty(&em->list) ||
2406 (em->flags & EXTENT_FLAG_LOGGING))
2409 * If it's in the list of modified extents, remove it
2410 * only if its generation is older then the current one,
2411 * in which case we don't need it for a fast fsync.
2412 * Otherwise don't remove it, we could be racing with an
2413 * ongoing fast fsync that could miss the new extent.
2415 fs_info = btrfs_inode->root->fs_info;
2416 spin_lock(&fs_info->trans_lock);
2417 cur_gen = fs_info->generation;
2418 spin_unlock(&fs_info->trans_lock);
2419 if (em->generation >= cur_gen)
2423 * We only remove extent maps that are not in the list of
2424 * modified extents or that are in the list but with a
2425 * generation lower then the current generation, so there
2426 * is no need to set the full fsync flag on the inode (it
2427 * hurts the fsync performance for workloads with a data
2428 * size that exceeds or is close to the system's memory).
2430 remove_extent_mapping(map, em);
2431 /* once for the rb tree */
2432 free_extent_map(em);
2434 start = extent_map_end(em);
2435 write_unlock(&map->lock);
2438 free_extent_map(em);
2440 cond_resched(); /* Allow large-extent preemption. */
2443 return try_release_extent_state(tree, page, mask);
2446 struct btrfs_fiemap_entry {
2454 * Indicate the caller of emit_fiemap_extent() that it needs to unlock the file
2455 * range from the inode's io tree, unlock the subvolume tree search path, flush
2456 * the fiemap cache and relock the file range and research the subvolume tree.
2457 * The value here is something negative that can't be confused with a valid
2458 * errno value and different from 1 because that's also a return value from
2459 * fiemap_fill_next_extent() and also it's often used to mean some btree search
2460 * did not find a key, so make it some distinct negative value.
2462 #define BTRFS_FIEMAP_FLUSH_CACHE (-(MAX_ERRNO + 1))
2467 * - Cache the next entry to be emitted to the fiemap buffer, so that we can
2468 * merge extents that are contiguous and can be grouped as a single one;
2470 * - Store extents ready to be written to the fiemap buffer in an intermediary
2471 * buffer. This intermediary buffer is to ensure that in case the fiemap
2472 * buffer is memory mapped to the fiemap target file, we don't deadlock
2473 * during btrfs_page_mkwrite(). This is because during fiemap we are locking
2474 * an extent range in order to prevent races with delalloc flushing and
2475 * ordered extent completion, which is needed in order to reliably detect
2476 * delalloc in holes and prealloc extents. And this can lead to a deadlock
2477 * if the fiemap buffer is memory mapped to the file we are running fiemap
2478 * against (a silly, useless in practice scenario, but possible) because
2479 * btrfs_page_mkwrite() will try to lock the same extent range.
2481 struct fiemap_cache {
2482 /* An array of ready fiemap entries. */
2483 struct btrfs_fiemap_entry *entries;
2484 /* Number of entries in the entries array. */
2486 /* Index of the next entry in the entries array to write to. */
2489 * Once the entries array is full, this indicates what's the offset for
2490 * the next file extent item we must search for in the inode's subvolume
2491 * tree after unlocking the extent range in the inode's io tree and
2492 * releasing the search path.
2494 u64 next_search_offset;
2496 * This matches struct fiemap_extent_info::fi_mapped_extents, we use it
2497 * to count ourselves emitted extents and stop instead of relying on
2498 * fiemap_fill_next_extent() because we buffer ready fiemap entries at
2499 * the @entries array, and we want to stop as soon as we hit the max
2500 * amount of extents to map, not just to save time but also to make the
2501 * logic at extent_fiemap() simpler.
2503 unsigned int extents_mapped;
2504 /* Fields for the cached extent (unsubmitted, not ready, extent). */
2512 static int flush_fiemap_cache(struct fiemap_extent_info *fieinfo,
2513 struct fiemap_cache *cache)
2515 for (int i = 0; i < cache->entries_pos; i++) {
2516 struct btrfs_fiemap_entry *entry = &cache->entries[i];
2519 ret = fiemap_fill_next_extent(fieinfo, entry->offset,
2520 entry->phys, entry->len,
2523 * Ignore 1 (reached max entries) because we keep track of that
2524 * ourselves in emit_fiemap_extent().
2529 cache->entries_pos = 0;
2535 * Helper to submit fiemap extent.
2537 * Will try to merge current fiemap extent specified by @offset, @phys,
2538 * @len and @flags with cached one.
2539 * And only when we fails to merge, cached one will be submitted as
2542 * Return value is the same as fiemap_fill_next_extent().
2544 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2545 struct fiemap_cache *cache,
2546 u64 offset, u64 phys, u64 len, u32 flags)
2548 struct btrfs_fiemap_entry *entry;
2551 /* Set at the end of extent_fiemap(). */
2552 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2558 * When iterating the extents of the inode, at extent_fiemap(), we may
2559 * find an extent that starts at an offset behind the end offset of the
2560 * previous extent we processed. This happens if fiemap is called
2561 * without FIEMAP_FLAG_SYNC and there are ordered extents completing
2562 * after we had to unlock the file range, release the search path, emit
2563 * the fiemap extents stored in the buffer (cache->entries array) and
2564 * the lock the remainder of the range and re-search the btree.
2566 * For example we are in leaf X processing its last item, which is the
2567 * file extent item for file range [512K, 1M[, and after
2568 * btrfs_next_leaf() releases the path, there's an ordered extent that
2569 * completes for the file range [768K, 2M[, and that results in trimming
2570 * the file extent item so that it now corresponds to the file range
2571 * [512K, 768K[ and a new file extent item is inserted for the file
2572 * range [768K, 2M[, which may end up as the last item of leaf X or as
2573 * the first item of the next leaf - in either case btrfs_next_leaf()
2574 * will leave us with a path pointing to the new extent item, for the
2575 * file range [768K, 2M[, since that's the first key that follows the
2576 * last one we processed. So in order not to report overlapping extents
2577 * to user space, we trim the length of the previously cached extent and
2580 * Upon calling btrfs_next_leaf() we may also find an extent with an
2581 * offset smaller than or equals to cache->offset, and this happens
2582 * when we had a hole or prealloc extent with several delalloc ranges in
2583 * it, but after btrfs_next_leaf() released the path, delalloc was
2584 * flushed and the resulting ordered extents were completed, so we can
2585 * now have found a file extent item for an offset that is smaller than
2586 * or equals to what we have in cache->offset. We deal with this as
2589 cache_end = cache->offset + cache->len;
2590 if (cache_end > offset) {
2591 if (offset == cache->offset) {
2593 * We cached a dealloc range (found in the io tree) for
2594 * a hole or prealloc extent and we have now found a
2595 * file extent item for the same offset. What we have
2596 * now is more recent and up to date, so discard what
2597 * we had in the cache and use what we have just found.
2600 } else if (offset > cache->offset) {
2602 * The extent range we previously found ends after the
2603 * offset of the file extent item we found and that
2604 * offset falls somewhere in the middle of that previous
2605 * extent range. So adjust the range we previously found
2606 * to end at the offset of the file extent item we have
2607 * just found, since this extent is more up to date.
2608 * Emit that adjusted range and cache the file extent
2609 * item we have just found. This corresponds to the case
2610 * where a previously found file extent item was split
2611 * due to an ordered extent completing.
2613 cache->len = offset - cache->offset;
2616 const u64 range_end = offset + len;
2619 * The offset of the file extent item we have just found
2620 * is behind the cached offset. This means we were
2621 * processing a hole or prealloc extent for which we
2622 * have found delalloc ranges (in the io tree), so what
2623 * we have in the cache is the last delalloc range we
2624 * found while the file extent item we found can be
2625 * either for a whole delalloc range we previously
2626 * emmitted or only a part of that range.
2628 * We have two cases here:
2630 * 1) The file extent item's range ends at or behind the
2631 * cached extent's end. In this case just ignore the
2632 * current file extent item because we don't want to
2633 * overlap with previous ranges that may have been
2636 * 2) The file extent item starts behind the currently
2637 * cached extent but its end offset goes beyond the
2638 * end offset of the cached extent. We don't want to
2639 * overlap with a previous range that may have been
2640 * emmitted already, so we emit the currently cached
2641 * extent and then partially store the current file
2642 * extent item's range in the cache, for the subrange
2643 * going the cached extent's end to the end of the
2646 if (range_end <= cache_end)
2649 if (!(flags & (FIEMAP_EXTENT_ENCODED | FIEMAP_EXTENT_DELALLOC)))
2650 phys += cache_end - offset;
2653 len = range_end - cache_end;
2659 * Only merges fiemap extents if
2660 * 1) Their logical addresses are continuous
2662 * 2) Their physical addresses are continuous
2663 * So truly compressed (physical size smaller than logical size)
2664 * extents won't get merged with each other
2666 * 3) Share same flags
2668 if (cache->offset + cache->len == offset &&
2669 cache->phys + cache->len == phys &&
2670 cache->flags == flags) {
2676 /* Not mergeable, need to submit cached one */
2678 if (cache->entries_pos == cache->entries_size) {
2680 * We will need to research for the end offset of the last
2681 * stored extent and not from the current offset, because after
2682 * unlocking the range and releasing the path, if there's a hole
2683 * between that end offset and this current offset, a new extent
2684 * may have been inserted due to a new write, so we don't want
2687 entry = &cache->entries[cache->entries_size - 1];
2688 cache->next_search_offset = entry->offset + entry->len;
2689 cache->cached = false;
2691 return BTRFS_FIEMAP_FLUSH_CACHE;
2694 entry = &cache->entries[cache->entries_pos];
2695 entry->offset = cache->offset;
2696 entry->phys = cache->phys;
2697 entry->len = cache->len;
2698 entry->flags = cache->flags;
2699 cache->entries_pos++;
2700 cache->extents_mapped++;
2702 if (cache->extents_mapped == fieinfo->fi_extents_max) {
2703 cache->cached = false;
2707 cache->cached = true;
2708 cache->offset = offset;
2711 cache->flags = flags;
2717 * Emit last fiemap cache
2719 * The last fiemap cache may still be cached in the following case:
2721 * |<- Fiemap range ->|
2722 * |<------------ First extent ----------->|
2724 * In this case, the first extent range will be cached but not emitted.
2725 * So we must emit it before ending extent_fiemap().
2727 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2728 struct fiemap_cache *cache)
2735 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2736 cache->len, cache->flags);
2737 cache->cached = false;
2743 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2745 struct extent_buffer *clone = path->nodes[0];
2746 struct btrfs_key key;
2751 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2755 * Add a temporary extra ref to an already cloned extent buffer to
2756 * prevent btrfs_next_leaf() freeing it, we want to reuse it to avoid
2757 * the cost of allocating a new one.
2759 ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, &clone->bflags));
2760 atomic_inc(&clone->refs);
2762 ret = btrfs_next_leaf(inode->root, path);
2767 * Don't bother with cloning if there are no more file extent items for
2770 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2771 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY) {
2776 /* See the comment at fiemap_search_slot() about why we clone. */
2777 copy_extent_buffer_full(clone, path->nodes[0]);
2779 * Important to preserve the start field, for the optimizations when
2780 * checking if extents are shared (see extent_fiemap()).
2782 clone->start = path->nodes[0]->start;
2784 slot = path->slots[0];
2785 btrfs_release_path(path);
2786 path->nodes[0] = clone;
2787 path->slots[0] = slot;
2790 free_extent_buffer(clone);
2796 * Search for the first file extent item that starts at a given file offset or
2797 * the one that starts immediately before that offset.
2798 * Returns: 0 on success, < 0 on error, 1 if not found.
2800 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2803 const u64 ino = btrfs_ino(inode);
2804 struct btrfs_root *root = inode->root;
2805 struct extent_buffer *clone;
2806 struct btrfs_key key;
2811 key.type = BTRFS_EXTENT_DATA_KEY;
2812 key.offset = file_offset;
2814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2818 if (ret > 0 && path->slots[0] > 0) {
2819 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2820 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2824 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2825 ret = btrfs_next_leaf(root, path);
2829 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2830 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2835 * We clone the leaf and use it during fiemap. This is because while
2836 * using the leaf we do expensive things like checking if an extent is
2837 * shared, which can take a long time. In order to prevent blocking
2838 * other tasks for too long, we use a clone of the leaf. We have locked
2839 * the file range in the inode's io tree, so we know none of our file
2840 * extent items can change. This way we avoid blocking other tasks that
2841 * want to insert items for other inodes in the same leaf or b+tree
2842 * rebalance operations (triggered for example when someone is trying
2843 * to push items into this leaf when trying to insert an item in a
2845 * We also need the private clone because holding a read lock on an
2846 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2847 * when we check if extents are shared, as backref walking may need to
2848 * lock the same leaf we are processing.
2850 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2854 slot = path->slots[0];
2855 btrfs_release_path(path);
2856 path->nodes[0] = clone;
2857 path->slots[0] = slot;
2863 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2864 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2865 * extent. The end offset (@end) is inclusive.
2867 static int fiemap_process_hole(struct btrfs_inode *inode,
2868 struct fiemap_extent_info *fieinfo,
2869 struct fiemap_cache *cache,
2870 struct extent_state **delalloc_cached_state,
2871 struct btrfs_backref_share_check_ctx *backref_ctx,
2872 u64 disk_bytenr, u64 extent_offset,
2876 const u64 i_size = i_size_read(&inode->vfs_inode);
2877 u64 cur_offset = start;
2878 u64 last_delalloc_end = 0;
2879 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2880 bool checked_extent_shared = false;
2884 * There can be no delalloc past i_size, so don't waste time looking for
2887 while (cur_offset < end && cur_offset < i_size) {
2891 u64 prealloc_len = 0;
2894 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2895 delalloc_cached_state,
2902 * If this is a prealloc extent we have to report every section
2903 * of it that has no delalloc.
2905 if (disk_bytenr != 0) {
2906 if (last_delalloc_end == 0) {
2907 prealloc_start = start;
2908 prealloc_len = delalloc_start - start;
2910 prealloc_start = last_delalloc_end + 1;
2911 prealloc_len = delalloc_start - prealloc_start;
2915 if (prealloc_len > 0) {
2916 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2917 ret = btrfs_is_data_extent_shared(inode,
2924 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2926 checked_extent_shared = true;
2928 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2929 disk_bytenr + extent_offset,
2930 prealloc_len, prealloc_flags);
2933 extent_offset += prealloc_len;
2936 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2937 delalloc_end + 1 - delalloc_start,
2938 FIEMAP_EXTENT_DELALLOC |
2939 FIEMAP_EXTENT_UNKNOWN);
2943 last_delalloc_end = delalloc_end;
2944 cur_offset = delalloc_end + 1;
2945 extent_offset += cur_offset - delalloc_start;
2950 * Either we found no delalloc for the whole prealloc extent or we have
2951 * a prealloc extent that spans i_size or starts at or after i_size.
2953 if (disk_bytenr != 0 && last_delalloc_end < end) {
2957 if (last_delalloc_end == 0) {
2958 prealloc_start = start;
2959 prealloc_len = end + 1 - start;
2961 prealloc_start = last_delalloc_end + 1;
2962 prealloc_len = end + 1 - prealloc_start;
2965 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2966 ret = btrfs_is_data_extent_shared(inode,
2973 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2975 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2976 disk_bytenr + extent_offset,
2977 prealloc_len, prealloc_flags);
2985 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2986 struct btrfs_path *path,
2987 u64 *last_extent_end_ret)
2989 const u64 ino = btrfs_ino(inode);
2990 struct btrfs_root *root = inode->root;
2991 struct extent_buffer *leaf;
2992 struct btrfs_file_extent_item *ei;
2993 struct btrfs_key key;
2998 * Lookup the last file extent. We're not using i_size here because
2999 * there might be preallocation past i_size.
3001 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
3002 /* There can't be a file extent item at offset (u64)-1 */
3008 * For a non-existing key, btrfs_search_slot() always leaves us at a
3009 * slot > 0, except if the btree is empty, which is impossible because
3010 * at least it has the inode item for this inode and all the items for
3011 * the root inode 256.
3013 ASSERT(path->slots[0] > 0);
3015 leaf = path->nodes[0];
3016 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3017 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
3018 /* No file extent items in the subvolume tree. */
3019 *last_extent_end_ret = 0;
3024 * For an inline extent, the disk_bytenr is where inline data starts at,
3025 * so first check if we have an inline extent item before checking if we
3026 * have an implicit hole (disk_bytenr == 0).
3028 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
3029 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
3030 *last_extent_end_ret = btrfs_file_extent_end(path);
3035 * Find the last file extent item that is not a hole (when NO_HOLES is
3036 * not enabled). This should take at most 2 iterations in the worst
3037 * case: we have one hole file extent item at slot 0 of a leaf and
3038 * another hole file extent item as the last item in the previous leaf.
3039 * This is because we merge file extent items that represent holes.
3041 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3042 while (disk_bytenr == 0) {
3043 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
3046 } else if (ret > 0) {
3047 /* No file extent items that are not holes. */
3048 *last_extent_end_ret = 0;
3051 leaf = path->nodes[0];
3052 ei = btrfs_item_ptr(leaf, path->slots[0],
3053 struct btrfs_file_extent_item);
3054 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3057 *last_extent_end_ret = btrfs_file_extent_end(path);
3061 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
3064 const u64 ino = btrfs_ino(inode);
3065 struct extent_state *cached_state = NULL;
3066 struct extent_state *delalloc_cached_state = NULL;
3067 struct btrfs_path *path;
3068 struct fiemap_cache cache = { 0 };
3069 struct btrfs_backref_share_check_ctx *backref_ctx;
3070 u64 last_extent_end;
3071 u64 prev_extent_end;
3074 const u64 sectorsize = inode->root->fs_info->sectorsize;
3075 bool stopped = false;
3078 cache.entries_size = PAGE_SIZE / sizeof(struct btrfs_fiemap_entry);
3079 cache.entries = kmalloc_array(cache.entries_size,
3080 sizeof(struct btrfs_fiemap_entry),
3082 backref_ctx = btrfs_alloc_backref_share_check_ctx();
3083 path = btrfs_alloc_path();
3084 if (!cache.entries || !backref_ctx || !path) {
3090 range_start = round_down(start, sectorsize);
3091 range_end = round_up(start + len, sectorsize);
3092 prev_extent_end = range_start;
3094 lock_extent(&inode->io_tree, range_start, range_end, &cached_state);
3096 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
3099 btrfs_release_path(path);
3101 path->reada = READA_FORWARD;
3102 ret = fiemap_search_slot(inode, path, range_start);
3105 } else if (ret > 0) {
3107 * No file extent item found, but we may have delalloc between
3108 * the current offset and i_size. So check for that.
3111 goto check_eof_delalloc;
3114 while (prev_extent_end < range_end) {
3115 struct extent_buffer *leaf = path->nodes[0];
3116 struct btrfs_file_extent_item *ei;
3117 struct btrfs_key key;
3120 u64 extent_offset = 0;
3122 u64 disk_bytenr = 0;
3127 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3128 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3131 extent_end = btrfs_file_extent_end(path);
3134 * The first iteration can leave us at an extent item that ends
3135 * before our range's start. Move to the next item.
3137 if (extent_end <= range_start)
3140 backref_ctx->curr_leaf_bytenr = leaf->start;
3142 /* We have in implicit hole (NO_HOLES feature enabled). */
3143 if (prev_extent_end < key.offset) {
3144 const u64 hole_end = min(key.offset, range_end) - 1;
3146 ret = fiemap_process_hole(inode, fieinfo, &cache,
3147 &delalloc_cached_state,
3148 backref_ctx, 0, 0, 0,
3149 prev_extent_end, hole_end);
3152 } else if (ret > 0) {
3153 /* fiemap_fill_next_extent() told us to stop. */
3158 /* We've reached the end of the fiemap range, stop. */
3159 if (key.offset >= range_end) {
3165 extent_len = extent_end - key.offset;
3166 ei = btrfs_item_ptr(leaf, path->slots[0],
3167 struct btrfs_file_extent_item);
3168 compression = btrfs_file_extent_compression(leaf, ei);
3169 extent_type = btrfs_file_extent_type(leaf, ei);
3170 extent_gen = btrfs_file_extent_generation(leaf, ei);
3172 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3173 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3174 if (compression == BTRFS_COMPRESS_NONE)
3175 extent_offset = btrfs_file_extent_offset(leaf, ei);
3178 if (compression != BTRFS_COMPRESS_NONE)
3179 flags |= FIEMAP_EXTENT_ENCODED;
3181 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3182 flags |= FIEMAP_EXTENT_DATA_INLINE;
3183 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
3184 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
3186 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
3187 ret = fiemap_process_hole(inode, fieinfo, &cache,
3188 &delalloc_cached_state,
3190 disk_bytenr, extent_offset,
3191 extent_gen, key.offset,
3193 } else if (disk_bytenr == 0) {
3194 /* We have an explicit hole. */
3195 ret = fiemap_process_hole(inode, fieinfo, &cache,
3196 &delalloc_cached_state,
3197 backref_ctx, 0, 0, 0,
3198 key.offset, extent_end - 1);
3200 /* We have a regular extent. */
3201 if (fieinfo->fi_extents_max) {
3202 ret = btrfs_is_data_extent_shared(inode,
3209 flags |= FIEMAP_EXTENT_SHARED;
3212 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
3213 disk_bytenr + extent_offset,
3219 } else if (ret > 0) {
3220 /* emit_fiemap_extent() told us to stop. */
3225 prev_extent_end = extent_end;
3227 if (fatal_signal_pending(current)) {
3232 ret = fiemap_next_leaf_item(inode, path);
3235 } else if (ret > 0) {
3236 /* No more file extent items for this inode. */
3243 if (!stopped && prev_extent_end < range_end) {
3244 ret = fiemap_process_hole(inode, fieinfo, &cache,
3245 &delalloc_cached_state, backref_ctx,
3246 0, 0, 0, prev_extent_end, range_end - 1);
3249 prev_extent_end = range_end;
3252 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3253 const u64 i_size = i_size_read(&inode->vfs_inode);
3255 if (prev_extent_end < i_size) {
3260 delalloc = btrfs_find_delalloc_in_range(inode,
3263 &delalloc_cached_state,
3267 cache.flags |= FIEMAP_EXTENT_LAST;
3269 cache.flags |= FIEMAP_EXTENT_LAST;
3274 unlock_extent(&inode->io_tree, range_start, range_end, &cached_state);
3276 if (ret == BTRFS_FIEMAP_FLUSH_CACHE) {
3277 btrfs_release_path(path);
3278 ret = flush_fiemap_cache(fieinfo, &cache);
3281 len -= cache.next_search_offset - start;
3282 start = cache.next_search_offset;
3284 } else if (ret < 0) {
3289 * Must free the path before emitting to the fiemap buffer because we
3290 * may have a non-cloned leaf and if the fiemap buffer is memory mapped
3291 * to a file, a write into it (through btrfs_page_mkwrite()) may trigger
3292 * waiting for an ordered extent that in order to complete needs to
3293 * modify that leaf, therefore leading to a deadlock.
3295 btrfs_free_path(path);
3298 ret = flush_fiemap_cache(fieinfo, &cache);
3302 ret = emit_last_fiemap_cache(fieinfo, &cache);
3304 free_extent_state(delalloc_cached_state);
3305 kfree(cache.entries);
3306 btrfs_free_backref_share_ctx(backref_ctx);
3307 btrfs_free_path(path);
3311 static void __free_extent_buffer(struct extent_buffer *eb)
3313 kmem_cache_free(extent_buffer_cache, eb);
3316 static int extent_buffer_under_io(const struct extent_buffer *eb)
3318 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3319 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3322 static bool folio_range_has_eb(struct btrfs_fs_info *fs_info, struct folio *folio)
3324 struct btrfs_subpage *subpage;
3326 lockdep_assert_held(&folio->mapping->i_private_lock);
3328 if (folio_test_private(folio)) {
3329 subpage = folio_get_private(folio);
3330 if (atomic_read(&subpage->eb_refs))
3333 * Even there is no eb refs here, we may still have
3334 * end_page_read() call relying on page::private.
3336 if (atomic_read(&subpage->readers))
3342 static void detach_extent_buffer_folio(struct extent_buffer *eb, struct folio *folio)
3344 struct btrfs_fs_info *fs_info = eb->fs_info;
3345 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3348 * For mapped eb, we're going to change the folio private, which should
3349 * be done under the i_private_lock.
3352 spin_lock(&folio->mapping->i_private_lock);
3354 if (!folio_test_private(folio)) {
3356 spin_unlock(&folio->mapping->i_private_lock);
3360 if (fs_info->nodesize >= PAGE_SIZE) {
3362 * We do this since we'll remove the pages after we've
3363 * removed the eb from the radix tree, so we could race
3364 * and have this page now attached to the new eb. So
3365 * only clear folio if it's still connected to
3368 if (folio_test_private(folio) && folio_get_private(folio) == eb) {
3369 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3370 BUG_ON(folio_test_dirty(folio));
3371 BUG_ON(folio_test_writeback(folio));
3372 /* We need to make sure we haven't be attached to a new eb. */
3373 folio_detach_private(folio);
3376 spin_unlock(&folio->mapping->i_private_lock);
3381 * For subpage, we can have dummy eb with folio private attached. In
3382 * this case, we can directly detach the private as such folio is only
3383 * attached to one dummy eb, no sharing.
3386 btrfs_detach_subpage(fs_info, folio);
3390 btrfs_folio_dec_eb_refs(fs_info, folio);
3393 * We can only detach the folio private if there are no other ebs in the
3394 * page range and no unfinished IO.
3396 if (!folio_range_has_eb(fs_info, folio))
3397 btrfs_detach_subpage(fs_info, folio);
3399 spin_unlock(&folio->mapping->i_private_lock);
3402 /* Release all pages attached to the extent buffer */
3403 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3405 ASSERT(!extent_buffer_under_io(eb));
3407 for (int i = 0; i < INLINE_EXTENT_BUFFER_PAGES; i++) {
3408 struct folio *folio = eb->folios[i];
3413 detach_extent_buffer_folio(eb, folio);
3415 /* One for when we allocated the folio. */
3421 * Helper for releasing the extent buffer.
3423 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3425 btrfs_release_extent_buffer_pages(eb);
3426 btrfs_leak_debug_del_eb(eb);
3427 __free_extent_buffer(eb);
3430 static struct extent_buffer *
3431 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3434 struct extent_buffer *eb = NULL;
3436 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3439 eb->fs_info = fs_info;
3440 init_rwsem(&eb->lock);
3442 btrfs_leak_debug_add_eb(eb);
3444 spin_lock_init(&eb->refs_lock);
3445 atomic_set(&eb->refs, 1);
3447 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3452 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3454 struct extent_buffer *new;
3455 int num_folios = num_extent_folios(src);
3458 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3463 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3464 * btrfs_release_extent_buffer() have different behavior for
3465 * UNMAPPED subpage extent buffer.
3467 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3469 ret = alloc_eb_folio_array(new, 0);
3471 btrfs_release_extent_buffer(new);
3475 for (int i = 0; i < num_folios; i++) {
3476 struct folio *folio = new->folios[i];
3479 ret = attach_extent_buffer_folio(new, folio, NULL);
3481 btrfs_release_extent_buffer(new);
3484 WARN_ON(folio_test_dirty(folio));
3486 copy_extent_buffer_full(new, src);
3487 set_extent_buffer_uptodate(new);
3492 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3493 u64 start, unsigned long len)
3495 struct extent_buffer *eb;
3499 eb = __alloc_extent_buffer(fs_info, start, len);
3503 ret = alloc_eb_folio_array(eb, 0);
3507 num_folios = num_extent_folios(eb);
3508 for (int i = 0; i < num_folios; i++) {
3509 ret = attach_extent_buffer_folio(eb, eb->folios[i], NULL);
3514 set_extent_buffer_uptodate(eb);
3515 btrfs_set_header_nritems(eb, 0);
3516 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3520 for (int i = 0; i < num_folios; i++) {
3521 if (eb->folios[i]) {
3522 detach_extent_buffer_folio(eb, eb->folios[i]);
3523 __folio_put(eb->folios[i]);
3526 __free_extent_buffer(eb);
3530 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3533 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3536 static void check_buffer_tree_ref(struct extent_buffer *eb)
3540 * The TREE_REF bit is first set when the extent_buffer is added
3541 * to the radix tree. It is also reset, if unset, when a new reference
3542 * is created by find_extent_buffer.
3544 * It is only cleared in two cases: freeing the last non-tree
3545 * reference to the extent_buffer when its STALE bit is set or
3546 * calling release_folio when the tree reference is the only reference.
3548 * In both cases, care is taken to ensure that the extent_buffer's
3549 * pages are not under io. However, release_folio can be concurrently
3550 * called with creating new references, which is prone to race
3551 * conditions between the calls to check_buffer_tree_ref in those
3552 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3554 * The actual lifetime of the extent_buffer in the radix tree is
3555 * adequately protected by the refcount, but the TREE_REF bit and
3556 * its corresponding reference are not. To protect against this
3557 * class of races, we call check_buffer_tree_ref from the codepaths
3558 * which trigger io. Note that once io is initiated, TREE_REF can no
3559 * longer be cleared, so that is the moment at which any such race is
3562 refs = atomic_read(&eb->refs);
3563 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3566 spin_lock(&eb->refs_lock);
3567 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3568 atomic_inc(&eb->refs);
3569 spin_unlock(&eb->refs_lock);
3572 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
3574 int num_folios= num_extent_folios(eb);
3576 check_buffer_tree_ref(eb);
3578 for (int i = 0; i < num_folios; i++)
3579 folio_mark_accessed(eb->folios[i]);
3582 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3585 struct extent_buffer *eb;
3587 eb = find_extent_buffer_nolock(fs_info, start);
3591 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3592 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3593 * another task running free_extent_buffer() might have seen that flag
3594 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3595 * writeback flags not set) and it's still in the tree (flag
3596 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3597 * decrementing the extent buffer's reference count twice. So here we
3598 * could race and increment the eb's reference count, clear its stale
3599 * flag, mark it as dirty and drop our reference before the other task
3600 * finishes executing free_extent_buffer, which would later result in
3601 * an attempt to free an extent buffer that is dirty.
3603 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3604 spin_lock(&eb->refs_lock);
3605 spin_unlock(&eb->refs_lock);
3607 mark_extent_buffer_accessed(eb);
3611 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3612 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3615 struct extent_buffer *eb, *exists = NULL;
3618 eb = find_extent_buffer(fs_info, start);
3621 eb = alloc_dummy_extent_buffer(fs_info, start);
3623 return ERR_PTR(-ENOMEM);
3624 eb->fs_info = fs_info;
3626 ret = radix_tree_preload(GFP_NOFS);
3628 exists = ERR_PTR(ret);
3631 spin_lock(&fs_info->buffer_lock);
3632 ret = radix_tree_insert(&fs_info->buffer_radix,
3633 start >> fs_info->sectorsize_bits, eb);
3634 spin_unlock(&fs_info->buffer_lock);
3635 radix_tree_preload_end();
3636 if (ret == -EEXIST) {
3637 exists = find_extent_buffer(fs_info, start);
3643 check_buffer_tree_ref(eb);
3644 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3648 btrfs_release_extent_buffer(eb);
3653 static struct extent_buffer *grab_extent_buffer(
3654 struct btrfs_fs_info *fs_info, struct page *page)
3656 struct folio *folio = page_folio(page);
3657 struct extent_buffer *exists;
3660 * For subpage case, we completely rely on radix tree to ensure we
3661 * don't try to insert two ebs for the same bytenr. So here we always
3662 * return NULL and just continue.
3664 if (fs_info->nodesize < PAGE_SIZE)
3667 /* Page not yet attached to an extent buffer */
3668 if (!folio_test_private(folio))
3672 * We could have already allocated an eb for this page and attached one
3673 * so lets see if we can get a ref on the existing eb, and if we can we
3674 * know it's good and we can just return that one, else we know we can
3675 * just overwrite folio private.
3677 exists = folio_get_private(folio);
3678 if (atomic_inc_not_zero(&exists->refs))
3681 WARN_ON(PageDirty(page));
3682 folio_detach_private(folio);
3686 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3688 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3689 btrfs_err(fs_info, "bad tree block start %llu", start);
3693 if (fs_info->nodesize < PAGE_SIZE &&
3694 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3696 "tree block crosses page boundary, start %llu nodesize %u",
3697 start, fs_info->nodesize);
3700 if (fs_info->nodesize >= PAGE_SIZE &&
3701 !PAGE_ALIGNED(start)) {
3703 "tree block is not page aligned, start %llu nodesize %u",
3704 start, fs_info->nodesize);
3707 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3708 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3710 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3711 start, fs_info->nodesize);
3718 * Return 0 if eb->folios[i] is attached to btree inode successfully.
3719 * Return >0 if there is already another extent buffer for the range,
3720 * and @found_eb_ret would be updated.
3721 * Return -EAGAIN if the filemap has an existing folio but with different size
3723 * The caller needs to free the existing folios and retry using the same order.
3725 static int attach_eb_folio_to_filemap(struct extent_buffer *eb, int i,
3726 struct extent_buffer **found_eb_ret)
3729 struct btrfs_fs_info *fs_info = eb->fs_info;
3730 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3731 const unsigned long index = eb->start >> PAGE_SHIFT;
3732 struct folio *existing_folio;
3735 ASSERT(found_eb_ret);
3737 /* Caller should ensure the folio exists. */
3738 ASSERT(eb->folios[i]);
3741 ret = filemap_add_folio(mapping, eb->folios[i], index + i,
3742 GFP_NOFS | __GFP_NOFAIL);
3746 existing_folio = filemap_lock_folio(mapping, index + i);
3747 /* The page cache only exists for a very short time, just retry. */
3748 if (IS_ERR(existing_folio))
3751 /* For now, we should only have single-page folios for btree inode. */
3752 ASSERT(folio_nr_pages(existing_folio) == 1);
3754 if (folio_size(existing_folio) != eb->folio_size) {
3755 folio_unlock(existing_folio);
3756 folio_put(existing_folio);
3760 if (fs_info->nodesize < PAGE_SIZE) {
3762 * We're going to reuse the existing page, can drop our page
3763 * and subpage structure now.
3765 __free_page(folio_page(eb->folios[i], 0));
3766 eb->folios[i] = existing_folio;
3768 struct extent_buffer *existing_eb;
3770 existing_eb = grab_extent_buffer(fs_info,
3771 folio_page(existing_folio, 0));
3773 /* The extent buffer still exists, we can use it directly. */
3774 *found_eb_ret = existing_eb;
3775 folio_unlock(existing_folio);
3776 folio_put(existing_folio);
3779 /* The extent buffer no longer exists, we can reuse the folio. */
3780 __free_page(folio_page(eb->folios[i], 0));
3781 eb->folios[i] = existing_folio;
3786 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3787 u64 start, u64 owner_root, int level)
3789 unsigned long len = fs_info->nodesize;
3792 struct extent_buffer *eb;
3793 struct extent_buffer *existing_eb = NULL;
3794 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3795 struct btrfs_subpage *prealloc = NULL;
3796 u64 lockdep_owner = owner_root;
3797 bool page_contig = true;
3801 if (check_eb_alignment(fs_info, start))
3802 return ERR_PTR(-EINVAL);
3804 #if BITS_PER_LONG == 32
3805 if (start >= MAX_LFS_FILESIZE) {
3806 btrfs_err_rl(fs_info,
3807 "extent buffer %llu is beyond 32bit page cache limit", start);
3808 btrfs_err_32bit_limit(fs_info);
3809 return ERR_PTR(-EOVERFLOW);
3811 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3812 btrfs_warn_32bit_limit(fs_info);
3815 eb = find_extent_buffer(fs_info, start);
3819 eb = __alloc_extent_buffer(fs_info, start, len);
3821 return ERR_PTR(-ENOMEM);
3824 * The reloc trees are just snapshots, so we need them to appear to be
3825 * just like any other fs tree WRT lockdep.
3827 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3828 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3830 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3833 * Preallocate folio private for subpage case, so that we won't
3834 * allocate memory with i_private_lock nor page lock hold.
3836 * The memory will be freed by attach_extent_buffer_page() or freed
3837 * manually if we exit earlier.
3839 if (fs_info->nodesize < PAGE_SIZE) {
3840 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3841 if (IS_ERR(prealloc)) {
3842 ret = PTR_ERR(prealloc);
3848 /* Allocate all pages first. */
3849 ret = alloc_eb_folio_array(eb, __GFP_NOFAIL);
3851 btrfs_free_subpage(prealloc);
3855 num_folios = num_extent_folios(eb);
3856 /* Attach all pages to the filemap. */
3857 for (int i = 0; i < num_folios; i++) {
3858 struct folio *folio;
3860 ret = attach_eb_folio_to_filemap(eb, i, &existing_eb);
3862 ASSERT(existing_eb);
3867 * TODO: Special handling for a corner case where the order of
3868 * folios mismatch between the new eb and filemap.
3870 * This happens when:
3872 * - the new eb is using higher order folio
3874 * - the filemap is still using 0-order folios for the range
3875 * This can happen at the previous eb allocation, and we don't
3876 * have higher order folio for the call.
3878 * - the existing eb has already been freed
3880 * In this case, we have to free the existing folios first, and
3881 * re-allocate using the same order.
3882 * Thankfully this is not going to happen yet, as we're still
3883 * using 0-order folios.
3885 if (unlikely(ret == -EAGAIN)) {
3892 * Only after attach_eb_folio_to_filemap(), eb->folios[] is
3893 * reliable, as we may choose to reuse the existing page cache
3894 * and free the allocated page.
3896 folio = eb->folios[i];
3897 eb->folio_size = folio_size(folio);
3898 eb->folio_shift = folio_shift(folio);
3899 spin_lock(&mapping->i_private_lock);
3900 /* Should not fail, as we have preallocated the memory */
3901 ret = attach_extent_buffer_folio(eb, folio, prealloc);
3904 * To inform we have extra eb under allocation, so that
3905 * detach_extent_buffer_page() won't release the folio private
3906 * when the eb hasn't yet been inserted into radix tree.
3908 * The ref will be decreased when the eb released the page, in
3909 * detach_extent_buffer_page().
3910 * Thus needs no special handling in error path.
3912 btrfs_folio_inc_eb_refs(fs_info, folio);
3913 spin_unlock(&mapping->i_private_lock);
3915 WARN_ON(btrfs_folio_test_dirty(fs_info, folio, eb->start, eb->len));
3918 * Check if the current page is physically contiguous with previous eb
3920 * At this stage, either we allocated a large folio, thus @i
3921 * would only be 0, or we fall back to per-page allocation.
3923 if (i && folio_page(eb->folios[i - 1], 0) + 1 != folio_page(folio, 0))
3924 page_contig = false;
3926 if (!btrfs_folio_test_uptodate(fs_info, folio, eb->start, eb->len))
3930 * We can't unlock the pages just yet since the extent buffer
3931 * hasn't been properly inserted in the radix tree, this
3932 * opens a race with btree_release_folio which can free a page
3933 * while we are still filling in all pages for the buffer and
3938 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3939 /* All pages are physically contiguous, can skip cross page handling. */
3941 eb->addr = folio_address(eb->folios[0]) + offset_in_page(eb->start);
3943 ret = radix_tree_preload(GFP_NOFS);
3947 spin_lock(&fs_info->buffer_lock);
3948 ret = radix_tree_insert(&fs_info->buffer_radix,
3949 start >> fs_info->sectorsize_bits, eb);
3950 spin_unlock(&fs_info->buffer_lock);
3951 radix_tree_preload_end();
3952 if (ret == -EEXIST) {
3954 existing_eb = find_extent_buffer(fs_info, start);
3960 /* add one reference for the tree */
3961 check_buffer_tree_ref(eb);
3962 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3965 * Now it's safe to unlock the pages because any calls to
3966 * btree_release_folio will correctly detect that a page belongs to a
3967 * live buffer and won't free them prematurely.
3969 for (int i = 0; i < num_folios; i++)
3970 unlock_page(folio_page(eb->folios[i], 0));
3974 WARN_ON(!atomic_dec_and_test(&eb->refs));
3977 * Any attached folios need to be detached before we unlock them. This
3978 * is because when we're inserting our new folios into the mapping, and
3979 * then attaching our eb to that folio. If we fail to insert our folio
3980 * we'll lookup the folio for that index, and grab that EB. We do not
3981 * want that to grab this eb, as we're getting ready to free it. So we
3982 * have to detach it first and then unlock it.
3984 * We have to drop our reference and NULL it out here because in the
3985 * subpage case detaching does a btrfs_folio_dec_eb_refs() for our eb.
3986 * Below when we call btrfs_release_extent_buffer() we will call
3987 * detach_extent_buffer_folio() on our remaining pages in the !subpage
3988 * case. If we left eb->folios[i] populated in the subpage case we'd
3989 * double put our reference and be super sad.
3991 for (int i = 0; i < attached; i++) {
3992 ASSERT(eb->folios[i]);
3993 detach_extent_buffer_folio(eb, eb->folios[i]);
3994 unlock_page(folio_page(eb->folios[i], 0));
3995 folio_put(eb->folios[i]);
3996 eb->folios[i] = NULL;
3999 * Now all pages of that extent buffer is unmapped, set UNMAPPED flag,
4000 * so it can be cleaned up without utlizing page->mapping.
4002 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4004 btrfs_release_extent_buffer(eb);
4006 return ERR_PTR(ret);
4007 ASSERT(existing_eb);
4011 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4013 struct extent_buffer *eb =
4014 container_of(head, struct extent_buffer, rcu_head);
4016 __free_extent_buffer(eb);
4019 static int release_extent_buffer(struct extent_buffer *eb)
4020 __releases(&eb->refs_lock)
4022 lockdep_assert_held(&eb->refs_lock);
4024 WARN_ON(atomic_read(&eb->refs) == 0);
4025 if (atomic_dec_and_test(&eb->refs)) {
4026 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4027 struct btrfs_fs_info *fs_info = eb->fs_info;
4029 spin_unlock(&eb->refs_lock);
4031 spin_lock(&fs_info->buffer_lock);
4032 radix_tree_delete(&fs_info->buffer_radix,
4033 eb->start >> fs_info->sectorsize_bits);
4034 spin_unlock(&fs_info->buffer_lock);
4036 spin_unlock(&eb->refs_lock);
4039 btrfs_leak_debug_del_eb(eb);
4040 /* Should be safe to release our pages at this point */
4041 btrfs_release_extent_buffer_pages(eb);
4042 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4043 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
4044 __free_extent_buffer(eb);
4048 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4051 spin_unlock(&eb->refs_lock);
4056 void free_extent_buffer(struct extent_buffer *eb)
4062 refs = atomic_read(&eb->refs);
4064 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
4065 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
4068 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
4072 spin_lock(&eb->refs_lock);
4073 if (atomic_read(&eb->refs) == 2 &&
4074 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4075 !extent_buffer_under_io(eb) &&
4076 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4077 atomic_dec(&eb->refs);
4080 * I know this is terrible, but it's temporary until we stop tracking
4081 * the uptodate bits and such for the extent buffers.
4083 release_extent_buffer(eb);
4086 void free_extent_buffer_stale(struct extent_buffer *eb)
4091 spin_lock(&eb->refs_lock);
4092 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4094 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4095 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4096 atomic_dec(&eb->refs);
4097 release_extent_buffer(eb);
4100 static void btree_clear_folio_dirty(struct folio *folio)
4102 ASSERT(folio_test_dirty(folio));
4103 ASSERT(folio_test_locked(folio));
4104 folio_clear_dirty_for_io(folio);
4105 xa_lock_irq(&folio->mapping->i_pages);
4106 if (!folio_test_dirty(folio))
4107 __xa_clear_mark(&folio->mapping->i_pages,
4108 folio_index(folio), PAGECACHE_TAG_DIRTY);
4109 xa_unlock_irq(&folio->mapping->i_pages);
4112 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
4114 struct btrfs_fs_info *fs_info = eb->fs_info;
4115 struct folio *folio = eb->folios[0];
4118 /* btree_clear_folio_dirty() needs page locked. */
4120 last = btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start, eb->len);
4122 btree_clear_folio_dirty(folio);
4123 folio_unlock(folio);
4124 WARN_ON(atomic_read(&eb->refs) == 0);
4127 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
4128 struct extent_buffer *eb)
4130 struct btrfs_fs_info *fs_info = eb->fs_info;
4133 btrfs_assert_tree_write_locked(eb);
4135 if (trans && btrfs_header_generation(eb) != trans->transid)
4139 * Instead of clearing the dirty flag off of the buffer, mark it as
4140 * EXTENT_BUFFER_ZONED_ZEROOUT. This allows us to preserve
4141 * write-ordering in zoned mode, without the need to later re-dirty
4142 * the extent_buffer.
4144 * The actual zeroout of the buffer will happen later in
4145 * btree_csum_one_bio.
4147 if (btrfs_is_zoned(fs_info) && test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
4148 set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
4152 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
4155 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
4156 fs_info->dirty_metadata_batch);
4158 if (eb->fs_info->nodesize < PAGE_SIZE)
4159 return clear_subpage_extent_buffer_dirty(eb);
4161 num_folios = num_extent_folios(eb);
4162 for (int i = 0; i < num_folios; i++) {
4163 struct folio *folio = eb->folios[i];
4165 if (!folio_test_dirty(folio))
4168 btree_clear_folio_dirty(folio);
4169 folio_unlock(folio);
4171 WARN_ON(atomic_read(&eb->refs) == 0);
4174 void set_extent_buffer_dirty(struct extent_buffer *eb)
4179 check_buffer_tree_ref(eb);
4181 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4183 num_folios = num_extent_folios(eb);
4184 WARN_ON(atomic_read(&eb->refs) == 0);
4185 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4186 WARN_ON(test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags));
4189 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
4192 * For subpage case, we can have other extent buffers in the
4193 * same page, and in clear_subpage_extent_buffer_dirty() we
4194 * have to clear page dirty without subpage lock held.
4195 * This can cause race where our page gets dirty cleared after
4198 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
4199 * its page for other reasons, we can use page lock to prevent
4203 lock_page(folio_page(eb->folios[0], 0));
4204 for (int i = 0; i < num_folios; i++)
4205 btrfs_folio_set_dirty(eb->fs_info, eb->folios[i],
4206 eb->start, eb->len);
4208 unlock_page(folio_page(eb->folios[0], 0));
4209 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
4211 eb->fs_info->dirty_metadata_batch);
4213 #ifdef CONFIG_BTRFS_DEBUG
4214 for (int i = 0; i < num_folios; i++)
4215 ASSERT(folio_test_dirty(eb->folios[i]));
4219 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
4221 struct btrfs_fs_info *fs_info = eb->fs_info;
4222 int num_folios = num_extent_folios(eb);
4224 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4225 for (int i = 0; i < num_folios; i++) {
4226 struct folio *folio = eb->folios[i];
4232 * This is special handling for metadata subpage, as regular
4233 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4235 if (fs_info->nodesize >= PAGE_SIZE)
4236 folio_clear_uptodate(folio);
4238 btrfs_subpage_clear_uptodate(fs_info, folio,
4239 eb->start, eb->len);
4243 void set_extent_buffer_uptodate(struct extent_buffer *eb)
4245 struct btrfs_fs_info *fs_info = eb->fs_info;
4246 int num_folios = num_extent_folios(eb);
4248 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4249 for (int i = 0; i < num_folios; i++) {
4250 struct folio *folio = eb->folios[i];
4253 * This is special handling for metadata subpage, as regular
4254 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4256 if (fs_info->nodesize >= PAGE_SIZE)
4257 folio_mark_uptodate(folio);
4259 btrfs_subpage_set_uptodate(fs_info, folio,
4260 eb->start, eb->len);
4264 static void end_bbio_meta_read(struct btrfs_bio *bbio)
4266 struct extent_buffer *eb = bbio->private;
4267 struct btrfs_fs_info *fs_info = eb->fs_info;
4268 bool uptodate = !bbio->bio.bi_status;
4269 struct folio_iter fi;
4272 eb->read_mirror = bbio->mirror_num;
4275 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
4279 set_extent_buffer_uptodate(eb);
4281 clear_extent_buffer_uptodate(eb);
4282 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4285 bio_for_each_folio_all(fi, &bbio->bio) {
4286 struct folio *folio = fi.folio;
4287 u64 start = eb->start + bio_offset;
4288 u32 len = fi.length;
4291 btrfs_folio_set_uptodate(fs_info, folio, start, len);
4293 btrfs_folio_clear_uptodate(fs_info, folio, start, len);
4298 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
4299 smp_mb__after_atomic();
4300 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
4301 free_extent_buffer(eb);
4303 bio_put(&bbio->bio);
4306 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
4307 struct btrfs_tree_parent_check *check)
4309 struct btrfs_bio *bbio;
4312 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4316 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
4317 * operation, which could potentially still be in flight. In this case
4318 * we simply want to return an error.
4320 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
4323 /* Someone else is already reading the buffer, just wait for it. */
4324 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
4328 * Between the initial test_bit(EXTENT_BUFFER_UPTODATE) and the above
4329 * test_and_set_bit(EXTENT_BUFFER_READING), someone else could have
4330 * started and finished reading the same eb. In this case, UPTODATE
4331 * will now be set, and we shouldn't read it in again.
4333 if (unlikely(test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))) {
4334 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
4335 smp_mb__after_atomic();
4336 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
4340 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4341 eb->read_mirror = 0;
4342 check_buffer_tree_ref(eb);
4343 atomic_inc(&eb->refs);
4345 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
4346 REQ_OP_READ | REQ_META, eb->fs_info,
4347 end_bbio_meta_read, eb);
4348 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
4349 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
4350 bbio->file_offset = eb->start;
4351 memcpy(&bbio->parent_check, check, sizeof(*check));
4352 if (eb->fs_info->nodesize < PAGE_SIZE) {
4353 ret = bio_add_folio(&bbio->bio, eb->folios[0], eb->len,
4354 eb->start - folio_pos(eb->folios[0]));
4357 int num_folios = num_extent_folios(eb);
4359 for (int i = 0; i < num_folios; i++) {
4360 struct folio *folio = eb->folios[i];
4362 ret = bio_add_folio(&bbio->bio, folio, eb->folio_size, 0);
4366 btrfs_submit_bio(bbio, mirror_num);
4369 if (wait == WAIT_COMPLETE) {
4370 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
4371 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4378 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
4381 btrfs_warn(eb->fs_info,
4382 "access to eb bytenr %llu len %u out of range start %lu len %lu",
4383 eb->start, eb->len, start, len);
4384 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4390 * Check if the [start, start + len) range is valid before reading/writing
4392 * NOTE: @start and @len are offset inside the eb, not logical address.
4394 * Caller should not touch the dst/src memory if this function returns error.
4396 static inline int check_eb_range(const struct extent_buffer *eb,
4397 unsigned long start, unsigned long len)
4399 unsigned long offset;
4401 /* start, start + len should not go beyond eb->len nor overflow */
4402 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
4403 return report_eb_range(eb, start, len);
4408 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4409 unsigned long start, unsigned long len)
4411 const int unit_size = eb->folio_size;
4414 char *dst = (char *)dstv;
4415 unsigned long i = get_eb_folio_index(eb, start);
4417 if (check_eb_range(eb, start, len)) {
4419 * Invalid range hit, reset the memory, so callers won't get
4420 * some random garbage for their uninitialized memory.
4422 memset(dstv, 0, len);
4427 memcpy(dstv, eb->addr + start, len);
4431 offset = get_eb_offset_in_folio(eb, start);
4436 cur = min(len, unit_size - offset);
4437 kaddr = folio_address(eb->folios[i]);
4438 memcpy(dst, kaddr + offset, cur);
4447 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4449 unsigned long start, unsigned long len)
4451 const int unit_size = eb->folio_size;
4454 char __user *dst = (char __user *)dstv;
4455 unsigned long i = get_eb_folio_index(eb, start);
4458 WARN_ON(start > eb->len);
4459 WARN_ON(start + len > eb->start + eb->len);
4462 if (copy_to_user_nofault(dstv, eb->addr + start, len))
4467 offset = get_eb_offset_in_folio(eb, start);
4472 cur = min(len, unit_size - offset);
4473 kaddr = folio_address(eb->folios[i]);
4474 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4488 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4489 unsigned long start, unsigned long len)
4491 const int unit_size = eb->folio_size;
4495 char *ptr = (char *)ptrv;
4496 unsigned long i = get_eb_folio_index(eb, start);
4499 if (check_eb_range(eb, start, len))
4503 return memcmp(ptrv, eb->addr + start, len);
4505 offset = get_eb_offset_in_folio(eb, start);
4508 cur = min(len, unit_size - offset);
4509 kaddr = folio_address(eb->folios[i]);
4510 ret = memcmp(ptr, kaddr + offset, cur);
4523 * Check that the extent buffer is uptodate.
4525 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4526 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4528 static void assert_eb_folio_uptodate(const struct extent_buffer *eb, int i)
4530 struct btrfs_fs_info *fs_info = eb->fs_info;
4531 struct folio *folio = eb->folios[i];
4536 * If we are using the commit root we could potentially clear a page
4537 * Uptodate while we're using the extent buffer that we've previously
4538 * looked up. We don't want to complain in this case, as the page was
4539 * valid before, we just didn't write it out. Instead we want to catch
4540 * the case where we didn't actually read the block properly, which
4541 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4543 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4546 if (fs_info->nodesize < PAGE_SIZE) {
4547 struct folio *folio = eb->folios[0];
4550 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, folio,
4551 eb->start, eb->len)))
4552 btrfs_subpage_dump_bitmap(fs_info, folio, eb->start, eb->len);
4554 WARN_ON(!folio_test_uptodate(folio));
4558 static void __write_extent_buffer(const struct extent_buffer *eb,
4559 const void *srcv, unsigned long start,
4560 unsigned long len, bool use_memmove)
4562 const int unit_size = eb->folio_size;
4566 char *src = (char *)srcv;
4567 unsigned long i = get_eb_folio_index(eb, start);
4568 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4569 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4571 if (check_eb_range(eb, start, len))
4576 memmove(eb->addr + start, srcv, len);
4578 memcpy(eb->addr + start, srcv, len);
4582 offset = get_eb_offset_in_folio(eb, start);
4586 assert_eb_folio_uptodate(eb, i);
4588 cur = min(len, unit_size - offset);
4589 kaddr = folio_address(eb->folios[i]);
4591 memmove(kaddr + offset, src, cur);
4593 memcpy(kaddr + offset, src, cur);
4602 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4603 unsigned long start, unsigned long len)
4605 return __write_extent_buffer(eb, srcv, start, len, false);
4608 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4609 unsigned long start, unsigned long len)
4611 const int unit_size = eb->folio_size;
4612 unsigned long cur = start;
4615 memset(eb->addr + start, c, len);
4619 while (cur < start + len) {
4620 unsigned long index = get_eb_folio_index(eb, cur);
4621 unsigned int offset = get_eb_offset_in_folio(eb, cur);
4622 unsigned int cur_len = min(start + len - cur, unit_size - offset);
4624 assert_eb_folio_uptodate(eb, index);
4625 memset(folio_address(eb->folios[index]) + offset, c, cur_len);
4631 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4634 if (check_eb_range(eb, start, len))
4636 return memset_extent_buffer(eb, 0, start, len);
4639 void copy_extent_buffer_full(const struct extent_buffer *dst,
4640 const struct extent_buffer *src)
4642 const int unit_size = src->folio_size;
4643 unsigned long cur = 0;
4645 ASSERT(dst->len == src->len);
4647 while (cur < src->len) {
4648 unsigned long index = get_eb_folio_index(src, cur);
4649 unsigned long offset = get_eb_offset_in_folio(src, cur);
4650 unsigned long cur_len = min(src->len, unit_size - offset);
4651 void *addr = folio_address(src->folios[index]) + offset;
4653 write_extent_buffer(dst, addr, cur, cur_len);
4659 void copy_extent_buffer(const struct extent_buffer *dst,
4660 const struct extent_buffer *src,
4661 unsigned long dst_offset, unsigned long src_offset,
4664 const int unit_size = dst->folio_size;
4665 u64 dst_len = dst->len;
4669 unsigned long i = get_eb_folio_index(dst, dst_offset);
4671 if (check_eb_range(dst, dst_offset, len) ||
4672 check_eb_range(src, src_offset, len))
4675 WARN_ON(src->len != dst_len);
4677 offset = get_eb_offset_in_folio(dst, dst_offset);
4680 assert_eb_folio_uptodate(dst, i);
4682 cur = min(len, (unsigned long)(unit_size - offset));
4684 kaddr = folio_address(dst->folios[i]);
4685 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4695 * Calculate the folio and offset of the byte containing the given bit number.
4697 * @eb: the extent buffer
4698 * @start: offset of the bitmap item in the extent buffer
4700 * @folio_index: return index of the folio in the extent buffer that contains
4701 * the given bit number
4702 * @folio_offset: return offset into the folio given by folio_index
4704 * This helper hides the ugliness of finding the byte in an extent buffer which
4705 * contains a given bit.
4707 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4708 unsigned long start, unsigned long nr,
4709 unsigned long *folio_index,
4710 size_t *folio_offset)
4712 size_t byte_offset = BIT_BYTE(nr);
4716 * The byte we want is the offset of the extent buffer + the offset of
4717 * the bitmap item in the extent buffer + the offset of the byte in the
4720 offset = start + offset_in_eb_folio(eb, eb->start) + byte_offset;
4722 *folio_index = offset >> eb->folio_shift;
4723 *folio_offset = offset_in_eb_folio(eb, offset);
4727 * Determine whether a bit in a bitmap item is set.
4729 * @eb: the extent buffer
4730 * @start: offset of the bitmap item in the extent buffer
4731 * @nr: bit number to test
4733 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4740 eb_bitmap_offset(eb, start, nr, &i, &offset);
4741 assert_eb_folio_uptodate(eb, i);
4742 kaddr = folio_address(eb->folios[i]);
4743 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4746 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4748 unsigned long index = get_eb_folio_index(eb, bytenr);
4750 if (check_eb_range(eb, bytenr, 1))
4752 return folio_address(eb->folios[index]) + get_eb_offset_in_folio(eb, bytenr);
4756 * Set an area of a bitmap to 1.
4758 * @eb: the extent buffer
4759 * @start: offset of the bitmap item in the extent buffer
4760 * @pos: bit number of the first bit
4761 * @len: number of bits to set
4763 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4764 unsigned long pos, unsigned long len)
4766 unsigned int first_byte = start + BIT_BYTE(pos);
4767 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4768 const bool same_byte = (first_byte == last_byte);
4769 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4773 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4775 /* Handle the first byte. */
4776 kaddr = extent_buffer_get_byte(eb, first_byte);
4781 /* Handle the byte aligned part. */
4782 ASSERT(first_byte + 1 <= last_byte);
4783 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4785 /* Handle the last byte. */
4786 kaddr = extent_buffer_get_byte(eb, last_byte);
4787 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4792 * Clear an area of a bitmap.
4794 * @eb: the extent buffer
4795 * @start: offset of the bitmap item in the extent buffer
4796 * @pos: bit number of the first bit
4797 * @len: number of bits to clear
4799 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4800 unsigned long start, unsigned long pos,
4803 unsigned int first_byte = start + BIT_BYTE(pos);
4804 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4805 const bool same_byte = (first_byte == last_byte);
4806 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4810 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4812 /* Handle the first byte. */
4813 kaddr = extent_buffer_get_byte(eb, first_byte);
4818 /* Handle the byte aligned part. */
4819 ASSERT(first_byte + 1 <= last_byte);
4820 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4822 /* Handle the last byte. */
4823 kaddr = extent_buffer_get_byte(eb, last_byte);
4824 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4827 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4829 unsigned long distance = (src > dst) ? src - dst : dst - src;
4830 return distance < len;
4833 void memcpy_extent_buffer(const struct extent_buffer *dst,
4834 unsigned long dst_offset, unsigned long src_offset,
4837 const int unit_size = dst->folio_size;
4838 unsigned long cur_off = 0;
4840 if (check_eb_range(dst, dst_offset, len) ||
4841 check_eb_range(dst, src_offset, len))
4845 const bool use_memmove = areas_overlap(src_offset, dst_offset, len);
4848 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4850 memcpy(dst->addr + dst_offset, dst->addr + src_offset, len);
4854 while (cur_off < len) {
4855 unsigned long cur_src = cur_off + src_offset;
4856 unsigned long folio_index = get_eb_folio_index(dst, cur_src);
4857 unsigned long folio_off = get_eb_offset_in_folio(dst, cur_src);
4858 unsigned long cur_len = min(src_offset + len - cur_src,
4859 unit_size - folio_off);
4860 void *src_addr = folio_address(dst->folios[folio_index]) + folio_off;
4861 const bool use_memmove = areas_overlap(src_offset + cur_off,
4862 dst_offset + cur_off, cur_len);
4864 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4870 void memmove_extent_buffer(const struct extent_buffer *dst,
4871 unsigned long dst_offset, unsigned long src_offset,
4874 unsigned long dst_end = dst_offset + len - 1;
4875 unsigned long src_end = src_offset + len - 1;
4877 if (check_eb_range(dst, dst_offset, len) ||
4878 check_eb_range(dst, src_offset, len))
4881 if (dst_offset < src_offset) {
4882 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4887 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4892 unsigned long src_i;
4894 size_t dst_off_in_folio;
4895 size_t src_off_in_folio;
4899 src_i = get_eb_folio_index(dst, src_end);
4901 dst_off_in_folio = get_eb_offset_in_folio(dst, dst_end);
4902 src_off_in_folio = get_eb_offset_in_folio(dst, src_end);
4904 cur = min_t(unsigned long, len, src_off_in_folio + 1);
4905 cur = min(cur, dst_off_in_folio + 1);
4907 src_addr = folio_address(dst->folios[src_i]) + src_off_in_folio -
4909 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4912 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4921 #define GANG_LOOKUP_SIZE 16
4922 static struct extent_buffer *get_next_extent_buffer(
4923 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4925 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4926 struct extent_buffer *found = NULL;
4927 u64 page_start = page_offset(page);
4928 u64 cur = page_start;
4930 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4931 lockdep_assert_held(&fs_info->buffer_lock);
4933 while (cur < page_start + PAGE_SIZE) {
4937 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4938 (void **)gang, cur >> fs_info->sectorsize_bits,
4939 min_t(unsigned int, GANG_LOOKUP_SIZE,
4940 PAGE_SIZE / fs_info->nodesize));
4943 for (i = 0; i < ret; i++) {
4944 /* Already beyond page end */
4945 if (gang[i]->start >= page_start + PAGE_SIZE)
4948 if (gang[i]->start >= bytenr) {
4953 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4959 static int try_release_subpage_extent_buffer(struct page *page)
4961 struct btrfs_fs_info *fs_info = page_to_fs_info(page);
4962 u64 cur = page_offset(page);
4963 const u64 end = page_offset(page) + PAGE_SIZE;
4967 struct extent_buffer *eb = NULL;
4970 * Unlike try_release_extent_buffer() which uses folio private
4971 * to grab buffer, for subpage case we rely on radix tree, thus
4972 * we need to ensure radix tree consistency.
4974 * We also want an atomic snapshot of the radix tree, thus go
4975 * with spinlock rather than RCU.
4977 spin_lock(&fs_info->buffer_lock);
4978 eb = get_next_extent_buffer(fs_info, page, cur);
4980 /* No more eb in the page range after or at cur */
4981 spin_unlock(&fs_info->buffer_lock);
4984 cur = eb->start + eb->len;
4987 * The same as try_release_extent_buffer(), to ensure the eb
4988 * won't disappear out from under us.
4990 spin_lock(&eb->refs_lock);
4991 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4992 spin_unlock(&eb->refs_lock);
4993 spin_unlock(&fs_info->buffer_lock);
4996 spin_unlock(&fs_info->buffer_lock);
4999 * If tree ref isn't set then we know the ref on this eb is a
5000 * real ref, so just return, this eb will likely be freed soon
5003 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5004 spin_unlock(&eb->refs_lock);
5009 * Here we don't care about the return value, we will always
5010 * check the folio private at the end. And
5011 * release_extent_buffer() will release the refs_lock.
5013 release_extent_buffer(eb);
5016 * Finally to check if we have cleared folio private, as if we have
5017 * released all ebs in the page, the folio private should be cleared now.
5019 spin_lock(&page->mapping->i_private_lock);
5020 if (!folio_test_private(page_folio(page)))
5024 spin_unlock(&page->mapping->i_private_lock);
5029 int try_release_extent_buffer(struct page *page)
5031 struct folio *folio = page_folio(page);
5032 struct extent_buffer *eb;
5034 if (page_to_fs_info(page)->nodesize < PAGE_SIZE)
5035 return try_release_subpage_extent_buffer(page);
5038 * We need to make sure nobody is changing folio private, as we rely on
5039 * folio private as the pointer to extent buffer.
5041 spin_lock(&page->mapping->i_private_lock);
5042 if (!folio_test_private(folio)) {
5043 spin_unlock(&page->mapping->i_private_lock);
5047 eb = folio_get_private(folio);
5051 * This is a little awful but should be ok, we need to make sure that
5052 * the eb doesn't disappear out from under us while we're looking at
5055 spin_lock(&eb->refs_lock);
5056 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5057 spin_unlock(&eb->refs_lock);
5058 spin_unlock(&page->mapping->i_private_lock);
5061 spin_unlock(&page->mapping->i_private_lock);
5064 * If tree ref isn't set then we know the ref on this eb is a real ref,
5065 * so just return, this page will likely be freed soon anyway.
5067 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5068 spin_unlock(&eb->refs_lock);
5072 return release_extent_buffer(eb);
5076 * Attempt to readahead a child block.
5078 * @fs_info: the fs_info
5079 * @bytenr: bytenr to read
5080 * @owner_root: objectid of the root that owns this eb
5081 * @gen: generation for the uptodate check, can be 0
5082 * @level: level for the eb
5084 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
5085 * normal uptodate check of the eb, without checking the generation. If we have
5086 * to read the block we will not block on anything.
5088 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
5089 u64 bytenr, u64 owner_root, u64 gen, int level)
5091 struct btrfs_tree_parent_check check = {
5096 struct extent_buffer *eb;
5099 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
5103 if (btrfs_buffer_uptodate(eb, gen, 1)) {
5104 free_extent_buffer(eb);
5108 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
5110 free_extent_buffer_stale(eb);
5112 free_extent_buffer(eb);
5116 * Readahead a node's child block.
5118 * @node: parent node we're reading from
5119 * @slot: slot in the parent node for the child we want to read
5121 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
5122 * the slot in the node provided.
5124 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
5126 btrfs_readahead_tree_block(node->fs_info,
5127 btrfs_node_blockptr(node, slot),
5128 btrfs_header_owner(node),
5129 btrfs_node_ptr_generation(node, slot),
5130 btrfs_header_level(node) - 1);