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>
18 #include "extent_io.h"
19 #include "extent-io-tree.h"
20 #include "extent_map.h"
22 #include "btrfs_inode.h"
25 #include "rcu-string.h"
30 #include "block-group.h"
31 #include "compression.h"
33 #include "accessors.h"
34 #include "file-item.h"
36 #include "dev-replace.h"
38 #include "transaction.h"
40 static struct kmem_cache *extent_buffer_cache;
42 #ifdef CONFIG_BTRFS_DEBUG
43 static inline void btrfs_leak_debug_add_eb(struct extent_buffer *eb)
45 struct btrfs_fs_info *fs_info = eb->fs_info;
48 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
49 list_add(&eb->leak_list, &fs_info->allocated_ebs);
50 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
53 static inline void btrfs_leak_debug_del_eb(struct extent_buffer *eb)
55 struct btrfs_fs_info *fs_info = eb->fs_info;
58 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
59 list_del(&eb->leak_list);
60 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
63 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
65 struct extent_buffer *eb;
69 * If we didn't get into open_ctree our allocated_ebs will not be
70 * initialized, so just skip this.
72 if (!fs_info->allocated_ebs.next)
75 WARN_ON(!list_empty(&fs_info->allocated_ebs));
76 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
77 while (!list_empty(&fs_info->allocated_ebs)) {
78 eb = list_first_entry(&fs_info->allocated_ebs,
79 struct extent_buffer, leak_list);
81 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
82 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
83 btrfs_header_owner(eb));
84 list_del(&eb->leak_list);
85 kmem_cache_free(extent_buffer_cache, eb);
87 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
90 #define btrfs_leak_debug_add_eb(eb) do {} while (0)
91 #define btrfs_leak_debug_del_eb(eb) do {} while (0)
95 * Structure to record info about the bio being assembled, and other info like
96 * how many bytes are there before stripe/ordered extent boundary.
98 struct btrfs_bio_ctrl {
99 struct btrfs_bio *bbio;
100 enum btrfs_compression_type compress_type;
101 u32 len_to_oe_boundary;
103 btrfs_bio_end_io_t end_io_func;
104 struct writeback_control *wbc;
107 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
109 struct btrfs_bio *bbio = bio_ctrl->bbio;
114 /* Caller should ensure the bio has at least some range added */
115 ASSERT(bbio->bio.bi_iter.bi_size);
117 if (btrfs_op(&bbio->bio) == BTRFS_MAP_READ &&
118 bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
119 btrfs_submit_compressed_read(bbio);
121 btrfs_submit_bio(bbio, 0);
123 /* The bbio is owned by the end_io handler now */
124 bio_ctrl->bbio = NULL;
128 * Submit or fail the current bio in the bio_ctrl structure.
130 static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret)
132 struct btrfs_bio *bbio = bio_ctrl->bbio;
139 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
140 /* The bio is owned by the end_io handler now */
141 bio_ctrl->bbio = NULL;
143 submit_one_bio(bio_ctrl);
147 int __init extent_buffer_init_cachep(void)
149 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
150 sizeof(struct extent_buffer), 0,
151 SLAB_MEM_SPREAD, NULL);
152 if (!extent_buffer_cache)
158 void __cold extent_buffer_free_cachep(void)
161 * Make sure all delayed rcu free are flushed before we
165 kmem_cache_destroy(extent_buffer_cache);
168 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
170 unsigned long index = start >> PAGE_SHIFT;
171 unsigned long end_index = end >> PAGE_SHIFT;
174 while (index <= end_index) {
175 page = find_get_page(inode->i_mapping, index);
176 BUG_ON(!page); /* Pages should be in the extent_io_tree */
177 clear_page_dirty_for_io(page);
183 static void process_one_page(struct btrfs_fs_info *fs_info,
184 struct page *page, struct page *locked_page,
185 unsigned long page_ops, u64 start, u64 end)
187 struct folio *folio = page_folio(page);
190 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
191 len = end + 1 - start;
193 if (page_ops & PAGE_SET_ORDERED)
194 btrfs_folio_clamp_set_ordered(fs_info, folio, start, len);
195 if (page_ops & PAGE_START_WRITEBACK) {
196 btrfs_folio_clamp_clear_dirty(fs_info, folio, start, len);
197 btrfs_folio_clamp_set_writeback(fs_info, folio, start, len);
199 if (page_ops & PAGE_END_WRITEBACK)
200 btrfs_folio_clamp_clear_writeback(fs_info, folio, start, len);
202 if (page != locked_page && (page_ops & PAGE_UNLOCK))
203 btrfs_folio_end_writer_lock(fs_info, folio, start, len);
206 static void __process_pages_contig(struct address_space *mapping,
207 struct page *locked_page, u64 start, u64 end,
208 unsigned long page_ops)
210 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
211 pgoff_t start_index = start >> PAGE_SHIFT;
212 pgoff_t end_index = end >> PAGE_SHIFT;
213 pgoff_t index = start_index;
214 struct folio_batch fbatch;
217 folio_batch_init(&fbatch);
218 while (index <= end_index) {
221 found_folios = filemap_get_folios_contig(mapping, &index,
223 for (i = 0; i < found_folios; i++) {
224 struct folio *folio = fbatch.folios[i];
226 process_one_page(fs_info, &folio->page, locked_page,
227 page_ops, start, end);
229 folio_batch_release(&fbatch);
234 static noinline void __unlock_for_delalloc(struct inode *inode,
235 struct page *locked_page,
238 unsigned long index = start >> PAGE_SHIFT;
239 unsigned long end_index = end >> PAGE_SHIFT;
242 if (index == locked_page->index && end_index == index)
245 __process_pages_contig(inode->i_mapping, locked_page, start, end,
249 static noinline int lock_delalloc_pages(struct inode *inode,
250 struct page *locked_page,
254 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
255 struct address_space *mapping = inode->i_mapping;
256 pgoff_t start_index = start >> PAGE_SHIFT;
257 pgoff_t end_index = end >> PAGE_SHIFT;
258 pgoff_t index = start_index;
259 u64 processed_end = start;
260 struct folio_batch fbatch;
262 if (index == locked_page->index && index == end_index)
265 folio_batch_init(&fbatch);
266 while (index <= end_index) {
267 unsigned int found_folios, i;
269 found_folios = filemap_get_folios_contig(mapping, &index,
271 if (found_folios == 0)
274 for (i = 0; i < found_folios; i++) {
275 struct folio *folio = fbatch.folios[i];
276 struct page *page = folio_page(folio, 0);
277 u32 len = end + 1 - start;
279 if (page == locked_page)
282 if (btrfs_folio_start_writer_lock(fs_info, folio, start,
286 if (!PageDirty(page) || page->mapping != mapping) {
287 btrfs_folio_end_writer_lock(fs_info, folio, start,
292 processed_end = page_offset(page) + PAGE_SIZE - 1;
294 folio_batch_release(&fbatch);
300 folio_batch_release(&fbatch);
301 if (processed_end > start)
302 __unlock_for_delalloc(inode, locked_page, start, processed_end);
307 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
308 * more than @max_bytes.
310 * @start: The original start bytenr to search.
311 * Will store the extent range start bytenr.
312 * @end: The original end bytenr of the search range
313 * Will store the extent range end bytenr.
315 * Return true if we find a delalloc range which starts inside the original
316 * range, and @start/@end will store the delalloc range start/end.
318 * Return false if we can't find any delalloc range which starts inside the
319 * original range, and @start/@end will be the non-delalloc range start/end.
322 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
323 struct page *locked_page, u64 *start,
326 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
327 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
328 const u64 orig_start = *start;
329 const u64 orig_end = *end;
330 /* The sanity tests may not set a valid fs_info. */
331 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
335 struct extent_state *cached_state = NULL;
339 /* Caller should pass a valid @end to indicate the search range end */
340 ASSERT(orig_end > orig_start);
342 /* The range should at least cover part of the page */
343 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
344 orig_end <= page_offset(locked_page)));
346 /* step one, find a bunch of delalloc bytes starting at start */
347 delalloc_start = *start;
349 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
350 max_bytes, &cached_state);
351 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
352 *start = delalloc_start;
354 /* @delalloc_end can be -1, never go beyond @orig_end */
355 *end = min(delalloc_end, orig_end);
356 free_extent_state(cached_state);
361 * start comes from the offset of locked_page. We have to lock
362 * pages in order, so we can't process delalloc bytes before
365 if (delalloc_start < *start)
366 delalloc_start = *start;
369 * make sure to limit the number of pages we try to lock down
371 if (delalloc_end + 1 - delalloc_start > max_bytes)
372 delalloc_end = delalloc_start + max_bytes - 1;
374 /* step two, lock all the pages after the page that has start */
375 ret = lock_delalloc_pages(inode, locked_page,
376 delalloc_start, delalloc_end);
377 ASSERT(!ret || ret == -EAGAIN);
378 if (ret == -EAGAIN) {
379 /* some of the pages are gone, lets avoid looping by
380 * shortening the size of the delalloc range we're searching
382 free_extent_state(cached_state);
385 max_bytes = PAGE_SIZE;
394 /* step three, lock the state bits for the whole range */
395 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
397 /* then test to make sure it is all still delalloc */
398 ret = test_range_bit(tree, delalloc_start, delalloc_end,
399 EXTENT_DELALLOC, cached_state);
401 unlock_extent(tree, delalloc_start, delalloc_end,
403 __unlock_for_delalloc(inode, locked_page,
404 delalloc_start, delalloc_end);
408 free_extent_state(cached_state);
409 *start = delalloc_start;
415 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
416 struct page *locked_page,
417 u32 clear_bits, unsigned long page_ops)
419 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
421 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
422 start, end, page_ops);
425 static bool btrfs_verify_page(struct page *page, u64 start)
427 if (!fsverity_active(page->mapping->host) ||
428 PageUptodate(page) ||
429 start >= i_size_read(page->mapping->host))
431 return fsverity_verify_page(page);
434 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
436 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
437 struct folio *folio = page_folio(page);
439 ASSERT(page_offset(page) <= start &&
440 start + len <= page_offset(page) + PAGE_SIZE);
442 if (uptodate && btrfs_verify_page(page, start))
443 btrfs_folio_set_uptodate(fs_info, folio, start, len);
445 btrfs_folio_clear_uptodate(fs_info, folio, start, len);
447 if (!btrfs_is_subpage(fs_info, page->mapping))
450 btrfs_subpage_end_reader(fs_info, folio, start, len);
454 * After a write IO is done, we need to:
456 * - clear the uptodate bits on error
457 * - clear the writeback bits in the extent tree for the range
458 * - filio_end_writeback() if there is no more pending io for the folio
460 * Scheduling is not allowed, so the extent state tree is expected
461 * to have one and only one object corresponding to this IO.
463 static void end_bbio_data_write(struct btrfs_bio *bbio)
465 struct bio *bio = &bbio->bio;
466 int error = blk_status_to_errno(bio->bi_status);
467 struct folio_iter fi;
469 ASSERT(!bio_flagged(bio, BIO_CLONED));
470 bio_for_each_folio_all(fi, bio) {
471 struct folio *folio = fi.folio;
472 struct inode *inode = folio->mapping->host;
473 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
474 const u32 sectorsize = fs_info->sectorsize;
475 u64 start = folio_pos(folio) + fi.offset;
478 /* Only order 0 (single page) folios are allowed for data. */
479 ASSERT(folio_order(folio) == 0);
481 /* Our read/write should always be sector aligned. */
482 if (!IS_ALIGNED(fi.offset, sectorsize))
484 "partial page write in btrfs with offset %zu and length %zu",
485 fi.offset, fi.length);
486 else if (!IS_ALIGNED(fi.length, sectorsize))
488 "incomplete page write with offset %zu and length %zu",
489 fi.offset, fi.length);
491 btrfs_finish_ordered_extent(bbio->ordered,
492 folio_page(folio, 0), start, len, !error);
494 mapping_set_error(folio->mapping, error);
495 btrfs_folio_clear_writeback(fs_info, folio, start, len);
502 * Record previously processed extent range
504 * For endio_readpage_release_extent() to handle a full extent range, reducing
505 * the extent io operations.
507 struct processed_extent {
508 struct btrfs_inode *inode;
509 /* Start of the range in @inode */
511 /* End of the range in @inode */
517 * Try to release processed extent range
519 * May not release the extent range right now if the current range is
520 * contiguous to processed extent.
522 * Will release processed extent when any of @inode, @uptodate, the range is
523 * no longer contiguous to the processed range.
525 * Passing @inode == NULL will force processed extent to be released.
527 static void endio_readpage_release_extent(struct processed_extent *processed,
528 struct btrfs_inode *inode, u64 start, u64 end,
531 struct extent_state *cached = NULL;
532 struct extent_io_tree *tree;
534 /* The first extent, initialize @processed */
535 if (!processed->inode)
539 * Contiguous to processed extent, just uptodate the end.
541 * Several things to notice:
543 * - bio can be merged as long as on-disk bytenr is contiguous
544 * This means we can have page belonging to other inodes, thus need to
545 * check if the inode still matches.
546 * - bvec can contain range beyond current page for multi-page bvec
547 * Thus we need to do processed->end + 1 >= start check
549 if (processed->inode == inode && processed->uptodate == uptodate &&
550 processed->end + 1 >= start && end >= processed->end) {
551 processed->end = end;
555 tree = &processed->inode->io_tree;
557 * Now we don't have range contiguous to the processed range, release
558 * the processed range now.
560 unlock_extent(tree, processed->start, processed->end, &cached);
563 /* Update processed to current range */
564 processed->inode = inode;
565 processed->start = start;
566 processed->end = end;
567 processed->uptodate = uptodate;
570 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
572 struct folio *folio = page_folio(page);
574 ASSERT(folio_test_locked(folio));
575 if (!btrfs_is_subpage(fs_info, folio->mapping))
578 ASSERT(folio_test_private(folio));
579 btrfs_subpage_start_reader(fs_info, folio, page_offset(page), PAGE_SIZE);
583 * After a data read IO is done, we need to:
585 * - clear the uptodate bits on error
586 * - set the uptodate bits if things worked
587 * - set the folio up to date if all extents in the tree are uptodate
588 * - clear the lock bit in the extent tree
589 * - unlock the folio if there are no other extents locked for it
591 * Scheduling is not allowed, so the extent state tree is expected
592 * to have one and only one object corresponding to this IO.
594 static void end_bbio_data_read(struct btrfs_bio *bbio)
596 struct bio *bio = &bbio->bio;
597 struct processed_extent processed = { 0 };
598 struct folio_iter fi;
600 * The offset to the beginning of a bio, since one bio can never be
601 * larger than UINT_MAX, u32 here is enough.
605 ASSERT(!bio_flagged(bio, BIO_CLONED));
606 bio_for_each_folio_all(fi, &bbio->bio) {
607 bool uptodate = !bio->bi_status;
608 struct folio *folio = fi.folio;
609 struct inode *inode = folio->mapping->host;
610 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
611 const u32 sectorsize = fs_info->sectorsize;
616 /* For now only order 0 folios are supported for data. */
617 ASSERT(folio_order(folio) == 0);
619 "%s: bi_sector=%llu, err=%d, mirror=%u",
620 __func__, bio->bi_iter.bi_sector, bio->bi_status,
624 * We always issue full-sector reads, but if some block in a
625 * folio fails to read, blk_update_request() will advance
626 * bv_offset and adjust bv_len to compensate. Print a warning
627 * for unaligned offsets, and an error if they don't add up to
630 if (!IS_ALIGNED(fi.offset, sectorsize))
632 "partial page read in btrfs with offset %zu and length %zu",
633 fi.offset, fi.length);
634 else if (!IS_ALIGNED(fi.offset + fi.length, sectorsize))
636 "incomplete page read with offset %zu and length %zu",
637 fi.offset, fi.length);
639 start = folio_pos(folio) + fi.offset;
640 end = start + fi.length - 1;
643 if (likely(uptodate)) {
644 loff_t i_size = i_size_read(inode);
645 pgoff_t end_index = i_size >> folio_shift(folio);
648 * Zero out the remaining part if this range straddles
651 * Here we should only zero the range inside the folio,
652 * not touch anything else.
654 * NOTE: i_size is exclusive while end is inclusive.
656 if (folio_index(folio) == end_index && i_size <= end) {
657 u32 zero_start = max(offset_in_folio(folio, i_size),
658 offset_in_folio(folio, start));
659 u32 zero_len = offset_in_folio(folio, end) + 1 -
662 folio_zero_range(folio, zero_start, zero_len);
666 /* Update page status and unlock. */
667 end_page_read(folio_page(folio, 0), uptodate, start, len);
668 endio_readpage_release_extent(&processed, BTRFS_I(inode),
669 start, end, uptodate);
671 ASSERT(bio_offset + len > bio_offset);
675 /* Release the last extent */
676 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
681 * Populate every free slot in a provided array with pages.
683 * @nr_pages: number of pages to allocate
684 * @page_array: the array to fill with pages; any existing non-null entries in
685 * the array will be skipped
686 * @extra_gfp: the extra GFP flags for the allocation.
688 * Return: 0 if all pages were able to be allocated;
689 * -ENOMEM otherwise, the partially allocated pages would be freed and
690 * the array slots zeroed
692 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
695 unsigned int allocated;
697 for (allocated = 0; allocated < nr_pages;) {
698 unsigned int last = allocated;
700 allocated = alloc_pages_bulk_array(GFP_NOFS | extra_gfp,
701 nr_pages, page_array);
703 if (allocated == nr_pages)
707 * During this iteration, no page could be allocated, even
708 * though alloc_pages_bulk_array() falls back to alloc_page()
709 * if it could not bulk-allocate. So we must be out of memory.
711 if (allocated == last) {
712 for (int i = 0; i < allocated; i++) {
713 __free_page(page_array[i]);
714 page_array[i] = NULL;
719 memalloc_retry_wait(GFP_NOFS);
725 * Populate needed folios for the extent buffer.
727 * For now, the folios populated are always in order 0 (aka, single page).
729 static int alloc_eb_folio_array(struct extent_buffer *eb, gfp_t extra_gfp)
731 struct page *page_array[INLINE_EXTENT_BUFFER_PAGES] = { 0 };
732 int num_pages = num_extent_pages(eb);
735 ret = btrfs_alloc_page_array(num_pages, page_array, extra_gfp);
739 for (int i = 0; i < num_pages; i++)
740 eb->folios[i] = page_folio(page_array[i]);
744 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
745 struct page *page, u64 disk_bytenr,
746 unsigned int pg_offset)
748 struct bio *bio = &bio_ctrl->bbio->bio;
749 struct bio_vec *bvec = bio_last_bvec_all(bio);
750 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
752 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
754 * For compression, all IO should have its logical bytenr set
755 * to the starting bytenr of the compressed extent.
757 return bio->bi_iter.bi_sector == sector;
761 * The contig check requires the following conditions to be met:
763 * 1) The pages are belonging to the same inode
764 * This is implied by the call chain.
766 * 2) The range has adjacent logical bytenr
768 * 3) The range has adjacent file offset
769 * This is required for the usage of btrfs_bio->file_offset.
771 return bio_end_sector(bio) == sector &&
772 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
773 page_offset(page) + pg_offset;
776 static void alloc_new_bio(struct btrfs_inode *inode,
777 struct btrfs_bio_ctrl *bio_ctrl,
778 u64 disk_bytenr, u64 file_offset)
780 struct btrfs_fs_info *fs_info = inode->root->fs_info;
781 struct btrfs_bio *bbio;
783 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
784 bio_ctrl->end_io_func, NULL);
785 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
787 bbio->file_offset = file_offset;
788 bio_ctrl->bbio = bbio;
789 bio_ctrl->len_to_oe_boundary = U32_MAX;
791 /* Limit data write bios to the ordered boundary. */
793 struct btrfs_ordered_extent *ordered;
795 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
797 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
798 ordered->file_offset +
799 ordered->disk_num_bytes - file_offset);
800 bbio->ordered = ordered;
804 * Pick the last added device to support cgroup writeback. For
805 * multi-device file systems this means blk-cgroup policies have
806 * to always be set on the last added/replaced device.
807 * This is a bit odd but has been like that for a long time.
809 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
810 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
815 * @disk_bytenr: logical bytenr where the write will be
816 * @page: page to add to the bio
817 * @size: portion of page that we want to write to
818 * @pg_offset: offset of the new bio or to check whether we are adding
819 * a contiguous page to the previous one
821 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
822 * new one in @bio_ctrl->bbio.
823 * The mirror number for this IO should already be initizlied in
824 * @bio_ctrl->mirror_num.
826 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
827 u64 disk_bytenr, struct page *page,
828 size_t size, unsigned long pg_offset)
830 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
832 ASSERT(pg_offset + size <= PAGE_SIZE);
833 ASSERT(bio_ctrl->end_io_func);
835 if (bio_ctrl->bbio &&
836 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
837 submit_one_bio(bio_ctrl);
842 /* Allocate new bio if needed */
843 if (!bio_ctrl->bbio) {
844 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
845 page_offset(page) + pg_offset);
848 /* Cap to the current ordered extent boundary if there is one. */
849 if (len > bio_ctrl->len_to_oe_boundary) {
850 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
851 ASSERT(is_data_inode(&inode->vfs_inode));
852 len = bio_ctrl->len_to_oe_boundary;
855 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
856 /* bio full: move on to a new one */
857 submit_one_bio(bio_ctrl);
862 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
869 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
870 * sector aligned. alloc_new_bio() then sets it to the end of
871 * our ordered extent for writes into zoned devices.
873 * When len_to_oe_boundary is tracking an ordered extent, we
874 * trust the ordered extent code to align things properly, and
875 * the check above to cap our write to the ordered extent
876 * boundary is correct.
878 * When len_to_oe_boundary is U32_MAX, the cap above would
879 * result in a 4095 byte IO for the last page right before
880 * we hit the bio limit of UINT_MAX. bio_add_page() has all
881 * the checks required to make sure we don't overflow the bio,
882 * and we should just ignore len_to_oe_boundary completely
883 * unless we're using it to track an ordered extent.
885 * It's pretty hard to make a bio sized U32_MAX, but it can
886 * happen when the page cache is able to feed us contiguous
887 * pages for large extents.
889 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
890 bio_ctrl->len_to_oe_boundary -= len;
892 /* Ordered extent boundary: move on to a new bio. */
893 if (bio_ctrl->len_to_oe_boundary == 0)
894 submit_one_bio(bio_ctrl);
898 static int attach_extent_buffer_folio(struct extent_buffer *eb,
900 struct btrfs_subpage *prealloc)
902 struct btrfs_fs_info *fs_info = eb->fs_info;
906 * If the page is mapped to btree inode, we should hold the private
907 * lock to prevent race.
908 * For cloned or dummy extent buffers, their pages are not mapped and
909 * will not race with any other ebs.
912 lockdep_assert_held(&folio->mapping->private_lock);
914 if (fs_info->nodesize >= PAGE_SIZE) {
915 if (!folio_test_private(folio))
916 folio_attach_private(folio, eb);
918 WARN_ON(folio_get_private(folio) != eb);
922 /* Already mapped, just free prealloc */
923 if (folio_test_private(folio)) {
924 btrfs_free_subpage(prealloc);
929 /* Has preallocated memory for subpage */
930 folio_attach_private(folio, prealloc);
932 /* Do new allocation to attach subpage */
933 ret = btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_METADATA);
937 int set_page_extent_mapped(struct page *page)
939 struct folio *folio = page_folio(page);
940 struct btrfs_fs_info *fs_info;
942 ASSERT(page->mapping);
944 if (folio_test_private(folio))
947 fs_info = btrfs_sb(page->mapping->host->i_sb);
949 if (btrfs_is_subpage(fs_info, page->mapping))
950 return btrfs_attach_subpage(fs_info, folio, BTRFS_SUBPAGE_DATA);
952 folio_attach_private(folio, (void *)EXTENT_FOLIO_PRIVATE);
956 void clear_page_extent_mapped(struct page *page)
958 struct folio *folio = page_folio(page);
959 struct btrfs_fs_info *fs_info;
961 ASSERT(page->mapping);
963 if (!folio_test_private(folio))
966 fs_info = btrfs_sb(page->mapping->host->i_sb);
967 if (btrfs_is_subpage(fs_info, page->mapping))
968 return btrfs_detach_subpage(fs_info, folio);
970 folio_detach_private(folio);
973 static struct extent_map *
974 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
975 u64 start, u64 len, struct extent_map **em_cached)
977 struct extent_map *em;
979 if (em_cached && *em_cached) {
981 if (extent_map_in_tree(em) && start >= em->start &&
982 start < extent_map_end(em)) {
983 refcount_inc(&em->refs);
991 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
992 if (em_cached && !IS_ERR(em)) {
994 refcount_inc(&em->refs);
1000 * basic readpage implementation. Locked extent state structs are inserted
1001 * into the tree that are removed when the IO is done (by the end_io
1003 * XXX JDM: This needs looking at to ensure proper page locking
1004 * return 0 on success, otherwise return error
1006 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
1007 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
1009 struct inode *inode = page->mapping->host;
1010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1011 u64 start = page_offset(page);
1012 const u64 end = start + PAGE_SIZE - 1;
1015 u64 last_byte = i_size_read(inode);
1017 struct extent_map *em;
1019 size_t pg_offset = 0;
1021 size_t blocksize = inode->i_sb->s_blocksize;
1022 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1024 ret = set_page_extent_mapped(page);
1026 unlock_extent(tree, start, end, NULL);
1031 if (page->index == last_byte >> PAGE_SHIFT) {
1032 size_t zero_offset = offset_in_page(last_byte);
1035 iosize = PAGE_SIZE - zero_offset;
1036 memzero_page(page, zero_offset, iosize);
1039 bio_ctrl->end_io_func = end_bbio_data_read;
1040 begin_page_read(fs_info, page);
1041 while (cur <= end) {
1042 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1043 bool force_bio_submit = false;
1046 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1047 if (cur >= last_byte) {
1048 iosize = PAGE_SIZE - pg_offset;
1049 memzero_page(page, pg_offset, iosize);
1050 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1051 end_page_read(page, true, cur, iosize);
1054 em = __get_extent_map(inode, page, pg_offset, cur,
1055 end - cur + 1, em_cached);
1057 unlock_extent(tree, cur, end, NULL);
1058 end_page_read(page, false, cur, end + 1 - cur);
1061 extent_offset = cur - em->start;
1062 BUG_ON(extent_map_end(em) <= cur);
1065 compress_type = extent_map_compression(em);
1067 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1068 iosize = ALIGN(iosize, blocksize);
1069 if (compress_type != BTRFS_COMPRESS_NONE)
1070 disk_bytenr = em->block_start;
1072 disk_bytenr = em->block_start + extent_offset;
1073 block_start = em->block_start;
1074 if (em->flags & EXTENT_FLAG_PREALLOC)
1075 block_start = EXTENT_MAP_HOLE;
1078 * If we have a file range that points to a compressed extent
1079 * and it's followed by a consecutive file range that points
1080 * to the same compressed extent (possibly with a different
1081 * offset and/or length, so it either points to the whole extent
1082 * or only part of it), we must make sure we do not submit a
1083 * single bio to populate the pages for the 2 ranges because
1084 * this makes the compressed extent read zero out the pages
1085 * belonging to the 2nd range. Imagine the following scenario:
1088 * [0 - 8K] [8K - 24K]
1091 * points to extent X, points to extent X,
1092 * offset 4K, length of 8K offset 0, length 16K
1094 * [extent X, compressed length = 4K uncompressed length = 16K]
1096 * If the bio to read the compressed extent covers both ranges,
1097 * it will decompress extent X into the pages belonging to the
1098 * first range and then it will stop, zeroing out the remaining
1099 * pages that belong to the other range that points to extent X.
1100 * So here we make sure we submit 2 bios, one for the first
1101 * range and another one for the third range. Both will target
1102 * the same physical extent from disk, but we can't currently
1103 * make the compressed bio endio callback populate the pages
1104 * for both ranges because each compressed bio is tightly
1105 * coupled with a single extent map, and each range can have
1106 * an extent map with a different offset value relative to the
1107 * uncompressed data of our extent and different lengths. This
1108 * is a corner case so we prioritize correctness over
1109 * non-optimal behavior (submitting 2 bios for the same extent).
1111 if (compress_type != BTRFS_COMPRESS_NONE &&
1112 prev_em_start && *prev_em_start != (u64)-1 &&
1113 *prev_em_start != em->start)
1114 force_bio_submit = true;
1117 *prev_em_start = em->start;
1119 free_extent_map(em);
1122 /* we've found a hole, just zero and go on */
1123 if (block_start == EXTENT_MAP_HOLE) {
1124 memzero_page(page, pg_offset, iosize);
1126 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1127 end_page_read(page, true, cur, iosize);
1129 pg_offset += iosize;
1132 /* the get_extent function already copied into the page */
1133 if (block_start == EXTENT_MAP_INLINE) {
1134 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1135 end_page_read(page, true, cur, iosize);
1137 pg_offset += iosize;
1141 if (bio_ctrl->compress_type != compress_type) {
1142 submit_one_bio(bio_ctrl);
1143 bio_ctrl->compress_type = compress_type;
1146 if (force_bio_submit)
1147 submit_one_bio(bio_ctrl);
1148 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1151 pg_offset += iosize;
1157 int btrfs_read_folio(struct file *file, struct folio *folio)
1159 struct page *page = &folio->page;
1160 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1161 u64 start = page_offset(page);
1162 u64 end = start + PAGE_SIZE - 1;
1163 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1166 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1168 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL);
1170 * If btrfs_do_readpage() failed we will want to submit the assembled
1171 * bio to do the cleanup.
1173 submit_one_bio(&bio_ctrl);
1177 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1179 struct extent_map **em_cached,
1180 struct btrfs_bio_ctrl *bio_ctrl,
1183 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1186 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1188 for (index = 0; index < nr_pages; index++) {
1189 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1191 put_page(pages[index]);
1196 * helper for __extent_writepage, doing all of the delayed allocation setup.
1198 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1199 * to write the page (copy into inline extent). In this case the IO has
1200 * been started and the page is already unlocked.
1202 * This returns 0 if all went well (page still locked)
1203 * This returns < 0 if there were errors (page still locked)
1205 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1206 struct page *page, struct writeback_control *wbc)
1208 const u64 page_start = page_offset(page);
1209 const u64 page_end = page_start + PAGE_SIZE - 1;
1210 u64 delalloc_start = page_start;
1211 u64 delalloc_end = page_end;
1212 u64 delalloc_to_write = 0;
1215 while (delalloc_start < page_end) {
1216 delalloc_end = page_end;
1217 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1218 &delalloc_start, &delalloc_end)) {
1219 delalloc_start = delalloc_end + 1;
1223 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1228 delalloc_start = delalloc_end + 1;
1232 * delalloc_end is already one less than the total length, so
1233 * we don't subtract one from PAGE_SIZE
1235 delalloc_to_write +=
1236 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1239 * If btrfs_run_dealloc_range() already started I/O and unlocked
1240 * the pages, we just need to account for them here.
1243 wbc->nr_to_write -= delalloc_to_write;
1247 if (wbc->nr_to_write < delalloc_to_write) {
1250 if (delalloc_to_write < thresh * 2)
1251 thresh = delalloc_to_write;
1252 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1260 * Find the first byte we need to write.
1262 * For subpage, one page can contain several sectors, and
1263 * __extent_writepage_io() will just grab all extent maps in the page
1264 * range and try to submit all non-inline/non-compressed extents.
1266 * This is a big problem for subpage, we shouldn't re-submit already written
1268 * This function will lookup subpage dirty bit to find which range we really
1271 * Return the next dirty range in [@start, @end).
1272 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1274 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1275 struct page *page, u64 *start, u64 *end)
1277 struct folio *folio = page_folio(page);
1278 struct btrfs_subpage *subpage = folio_get_private(folio);
1279 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1280 u64 orig_start = *start;
1281 /* Declare as unsigned long so we can use bitmap ops */
1282 unsigned long flags;
1283 int range_start_bit;
1287 * For regular sector size == page size case, since one page only
1288 * contains one sector, we return the page offset directly.
1290 if (!btrfs_is_subpage(fs_info, page->mapping)) {
1291 *start = page_offset(page);
1292 *end = page_offset(page) + PAGE_SIZE;
1296 range_start_bit = spi->dirty_offset +
1297 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1299 /* We should have the page locked, but just in case */
1300 spin_lock_irqsave(&subpage->lock, flags);
1301 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1302 spi->dirty_offset + spi->bitmap_nr_bits);
1303 spin_unlock_irqrestore(&subpage->lock, flags);
1305 range_start_bit -= spi->dirty_offset;
1306 range_end_bit -= spi->dirty_offset;
1308 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1309 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1313 * helper for __extent_writepage. This calls the writepage start hooks,
1314 * and does the loop to map the page into extents and bios.
1316 * We return 1 if the IO is started and the page is unlocked,
1317 * 0 if all went well (page still locked)
1318 * < 0 if there were errors (page still locked)
1320 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1322 struct btrfs_bio_ctrl *bio_ctrl,
1326 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1327 u64 cur = page_offset(page);
1328 u64 end = cur + PAGE_SIZE - 1;
1331 struct extent_map *em;
1335 ret = btrfs_writepage_cow_fixup(page);
1337 /* Fixup worker will requeue */
1338 redirty_page_for_writepage(bio_ctrl->wbc, page);
1343 bio_ctrl->end_io_func = end_bbio_data_write;
1344 while (cur <= end) {
1345 u32 len = end - cur + 1;
1348 u64 dirty_range_start = cur;
1349 u64 dirty_range_end;
1352 if (cur >= i_size) {
1353 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1356 * This range is beyond i_size, thus we don't need to
1357 * bother writing back.
1358 * But we still need to clear the dirty subpage bit, or
1359 * the next time the page gets dirtied, we will try to
1360 * writeback the sectors with subpage dirty bits,
1361 * causing writeback without ordered extent.
1363 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, len);
1367 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1369 if (cur < dirty_range_start) {
1370 cur = dirty_range_start;
1374 em = btrfs_get_extent(inode, NULL, 0, cur, len);
1376 ret = PTR_ERR_OR_ZERO(em);
1380 extent_offset = cur - em->start;
1381 em_end = extent_map_end(em);
1382 ASSERT(cur <= em_end);
1384 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1385 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1387 block_start = em->block_start;
1388 disk_bytenr = em->block_start + extent_offset;
1390 ASSERT(!extent_map_is_compressed(em));
1391 ASSERT(block_start != EXTENT_MAP_HOLE);
1392 ASSERT(block_start != EXTENT_MAP_INLINE);
1395 * Note that em_end from extent_map_end() and dirty_range_end from
1396 * find_next_dirty_byte() are all exclusive
1398 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1399 free_extent_map(em);
1402 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1403 if (!PageWriteback(page)) {
1404 btrfs_err(inode->root->fs_info,
1405 "page %lu not writeback, cur %llu end %llu",
1406 page->index, cur, end);
1410 * Although the PageDirty bit is cleared before entering this
1411 * function, subpage dirty bit is not cleared.
1412 * So clear subpage dirty bit here so next time we won't submit
1413 * page for range already written to disk.
1415 btrfs_folio_clear_dirty(fs_info, page_folio(page), cur, iosize);
1417 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1418 cur - page_offset(page));
1423 btrfs_folio_assert_not_dirty(fs_info, page_folio(page));
1429 * If we finish without problem, we should not only clear page dirty,
1430 * but also empty subpage dirty bits
1437 * the writepage semantics are similar to regular writepage. extent
1438 * records are inserted to lock ranges in the tree, and as dirty areas
1439 * are found, they are marked writeback. Then the lock bits are removed
1440 * and the end_io handler clears the writeback ranges
1442 * Return 0 if everything goes well.
1443 * Return <0 for error.
1445 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1447 struct folio *folio = page_folio(page);
1448 struct inode *inode = page->mapping->host;
1449 const u64 page_start = page_offset(page);
1453 loff_t i_size = i_size_read(inode);
1454 unsigned long end_index = i_size >> PAGE_SHIFT;
1456 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1458 WARN_ON(!PageLocked(page));
1460 pg_offset = offset_in_page(i_size);
1461 if (page->index > end_index ||
1462 (page->index == end_index && !pg_offset)) {
1463 folio_invalidate(folio, 0, folio_size(folio));
1464 folio_unlock(folio);
1468 if (page->index == end_index)
1469 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1471 ret = set_page_extent_mapped(page);
1475 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1481 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1485 bio_ctrl->wbc->nr_to_write--;
1489 /* make sure the mapping tag for page dirty gets cleared */
1490 set_page_writeback(page);
1491 end_page_writeback(page);
1494 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1496 mapping_set_error(page->mapping, ret);
1503 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1505 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1506 TASK_UNINTERRUPTIBLE);
1510 * Lock extent buffer status and pages for writeback.
1512 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1513 * extent buffer is not dirty)
1514 * Return %true is the extent buffer is submitted to bio.
1516 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1517 struct writeback_control *wbc)
1519 struct btrfs_fs_info *fs_info = eb->fs_info;
1522 btrfs_tree_lock(eb);
1523 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1524 btrfs_tree_unlock(eb);
1525 if (wbc->sync_mode != WB_SYNC_ALL)
1527 wait_on_extent_buffer_writeback(eb);
1528 btrfs_tree_lock(eb);
1532 * We need to do this to prevent races in people who check if the eb is
1533 * under IO since we can end up having no IO bits set for a short period
1536 spin_lock(&eb->refs_lock);
1537 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1538 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1539 spin_unlock(&eb->refs_lock);
1540 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1541 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1543 fs_info->dirty_metadata_batch);
1546 spin_unlock(&eb->refs_lock);
1548 btrfs_tree_unlock(eb);
1552 static void set_btree_ioerr(struct extent_buffer *eb)
1554 struct btrfs_fs_info *fs_info = eb->fs_info;
1556 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1559 * A read may stumble upon this buffer later, make sure that it gets an
1560 * error and knows there was an error.
1562 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1565 * We need to set the mapping with the io error as well because a write
1566 * error will flip the file system readonly, and then syncfs() will
1567 * return a 0 because we are readonly if we don't modify the err seq for
1570 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1573 * If writeback for a btree extent that doesn't belong to a log tree
1574 * failed, increment the counter transaction->eb_write_errors.
1575 * We do this because while the transaction is running and before it's
1576 * committing (when we call filemap_fdata[write|wait]_range against
1577 * the btree inode), we might have
1578 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1579 * returns an error or an error happens during writeback, when we're
1580 * committing the transaction we wouldn't know about it, since the pages
1581 * can be no longer dirty nor marked anymore for writeback (if a
1582 * subsequent modification to the extent buffer didn't happen before the
1583 * transaction commit), which makes filemap_fdata[write|wait]_range not
1584 * able to find the pages tagged with SetPageError at transaction
1585 * commit time. So if this happens we must abort the transaction,
1586 * otherwise we commit a super block with btree roots that point to
1587 * btree nodes/leafs whose content on disk is invalid - either garbage
1588 * or the content of some node/leaf from a past generation that got
1589 * cowed or deleted and is no longer valid.
1591 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1592 * not be enough - we need to distinguish between log tree extents vs
1593 * non-log tree extents, and the next filemap_fdatawait_range() call
1594 * will catch and clear such errors in the mapping - and that call might
1595 * be from a log sync and not from a transaction commit. Also, checking
1596 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1597 * not done and would not be reliable - the eb might have been released
1598 * from memory and reading it back again means that flag would not be
1599 * set (since it's a runtime flag, not persisted on disk).
1601 * Using the flags below in the btree inode also makes us achieve the
1602 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1603 * writeback for all dirty pages and before filemap_fdatawait_range()
1604 * is called, the writeback for all dirty pages had already finished
1605 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1606 * filemap_fdatawait_range() would return success, as it could not know
1607 * that writeback errors happened (the pages were no longer tagged for
1610 switch (eb->log_index) {
1612 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1615 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1618 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1621 BUG(); /* unexpected, logic error */
1626 * The endio specific version which won't touch any unsafe spinlock in endio
1629 static struct extent_buffer *find_extent_buffer_nolock(
1630 struct btrfs_fs_info *fs_info, u64 start)
1632 struct extent_buffer *eb;
1635 eb = radix_tree_lookup(&fs_info->buffer_radix,
1636 start >> fs_info->sectorsize_bits);
1637 if (eb && atomic_inc_not_zero(&eb->refs)) {
1645 static void end_bbio_meta_write(struct btrfs_bio *bbio)
1647 struct extent_buffer *eb = bbio->private;
1648 struct btrfs_fs_info *fs_info = eb->fs_info;
1649 bool uptodate = !bbio->bio.bi_status;
1650 struct folio_iter fi;
1654 set_btree_ioerr(eb);
1656 bio_for_each_folio_all(fi, &bbio->bio) {
1657 u64 start = eb->start + bio_offset;
1658 struct folio *folio = fi.folio;
1659 u32 len = fi.length;
1661 btrfs_folio_clear_writeback(fs_info, folio, start, len);
1665 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1666 smp_mb__after_atomic();
1667 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1669 bio_put(&bbio->bio);
1672 static void prepare_eb_write(struct extent_buffer *eb)
1675 unsigned long start;
1678 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1680 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1681 nritems = btrfs_header_nritems(eb);
1682 if (btrfs_header_level(eb) > 0) {
1683 end = btrfs_node_key_ptr_offset(eb, nritems);
1684 memzero_extent_buffer(eb, end, eb->len - end);
1688 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1690 start = btrfs_item_nr_offset(eb, nritems);
1691 end = btrfs_item_nr_offset(eb, 0);
1693 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1695 end += btrfs_item_offset(eb, nritems - 1);
1696 memzero_extent_buffer(eb, start, end - start);
1700 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1701 struct writeback_control *wbc)
1703 struct btrfs_fs_info *fs_info = eb->fs_info;
1704 struct btrfs_bio *bbio;
1706 prepare_eb_write(eb);
1708 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1709 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1710 eb->fs_info, end_bbio_meta_write, eb);
1711 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1712 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1713 wbc_init_bio(wbc, &bbio->bio);
1714 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1715 bbio->file_offset = eb->start;
1716 if (fs_info->nodesize < PAGE_SIZE) {
1717 struct folio *folio = eb->folios[0];
1721 btrfs_subpage_set_writeback(fs_info, folio, eb->start, eb->len);
1722 if (btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start,
1724 folio_clear_dirty_for_io(folio);
1727 ret = bio_add_folio(&bbio->bio, folio, eb->len,
1728 eb->start - folio_pos(folio));
1730 wbc_account_cgroup_owner(wbc, folio_page(folio, 0), eb->len);
1731 folio_unlock(folio);
1733 int num_folios = num_extent_folios(eb);
1735 for (int i = 0; i < num_folios; i++) {
1736 struct folio *folio = eb->folios[i];
1740 folio_clear_dirty_for_io(folio);
1741 folio_start_writeback(folio);
1742 ret = bio_add_folio(&bbio->bio, folio, folio_size(folio), 0);
1744 wbc_account_cgroup_owner(wbc, folio_page(folio, 0),
1746 wbc->nr_to_write -= folio_nr_pages(folio);
1747 folio_unlock(folio);
1750 btrfs_submit_bio(bbio, 0);
1754 * Submit one subpage btree page.
1756 * The main difference to submit_eb_page() is:
1758 * For subpage, we don't rely on page locking at all.
1761 * We only flush bio if we may be unable to fit current extent buffers into
1764 * Return >=0 for the number of submitted extent buffers.
1765 * Return <0 for fatal error.
1767 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1769 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1770 struct folio *folio = page_folio(page);
1772 u64 page_start = page_offset(page);
1774 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1776 /* Lock and write each dirty extent buffers in the range */
1777 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1778 struct btrfs_subpage *subpage = folio_get_private(folio);
1779 struct extent_buffer *eb;
1780 unsigned long flags;
1784 * Take private lock to ensure the subpage won't be detached
1787 spin_lock(&page->mapping->private_lock);
1788 if (!folio_test_private(folio)) {
1789 spin_unlock(&page->mapping->private_lock);
1792 spin_lock_irqsave(&subpage->lock, flags);
1793 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1794 subpage->bitmaps)) {
1795 spin_unlock_irqrestore(&subpage->lock, flags);
1796 spin_unlock(&page->mapping->private_lock);
1801 start = page_start + bit_start * fs_info->sectorsize;
1802 bit_start += sectors_per_node;
1805 * Here we just want to grab the eb without touching extra
1806 * spin locks, so call find_extent_buffer_nolock().
1808 eb = find_extent_buffer_nolock(fs_info, start);
1809 spin_unlock_irqrestore(&subpage->lock, flags);
1810 spin_unlock(&page->mapping->private_lock);
1813 * The eb has already reached 0 refs thus find_extent_buffer()
1814 * doesn't return it. We don't need to write back such eb
1820 if (lock_extent_buffer_for_io(eb, wbc)) {
1821 write_one_eb(eb, wbc);
1824 free_extent_buffer(eb);
1830 * Submit all page(s) of one extent buffer.
1832 * @page: the page of one extent buffer
1833 * @eb_context: to determine if we need to submit this page, if current page
1834 * belongs to this eb, we don't need to submit
1836 * The caller should pass each page in their bytenr order, and here we use
1837 * @eb_context to determine if we have submitted pages of one extent buffer.
1839 * If we have, we just skip until we hit a new page that doesn't belong to
1840 * current @eb_context.
1842 * If not, we submit all the page(s) of the extent buffer.
1844 * Return >0 if we have submitted the extent buffer successfully.
1845 * Return 0 if we don't need to submit the page, as it's already submitted by
1847 * Return <0 for fatal error.
1849 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1851 struct writeback_control *wbc = ctx->wbc;
1852 struct address_space *mapping = page->mapping;
1853 struct folio *folio = page_folio(page);
1854 struct extent_buffer *eb;
1857 if (!folio_test_private(folio))
1860 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
1861 return submit_eb_subpage(page, wbc);
1863 spin_lock(&mapping->private_lock);
1864 if (!folio_test_private(folio)) {
1865 spin_unlock(&mapping->private_lock);
1869 eb = folio_get_private(folio);
1872 * Shouldn't happen and normally this would be a BUG_ON but no point
1873 * crashing the machine for something we can survive anyway.
1876 spin_unlock(&mapping->private_lock);
1880 if (eb == ctx->eb) {
1881 spin_unlock(&mapping->private_lock);
1884 ret = atomic_inc_not_zero(&eb->refs);
1885 spin_unlock(&mapping->private_lock);
1891 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1895 free_extent_buffer(eb);
1899 if (!lock_extent_buffer_for_io(eb, wbc)) {
1900 free_extent_buffer(eb);
1903 /* Implies write in zoned mode. */
1904 if (ctx->zoned_bg) {
1905 /* Mark the last eb in the block group. */
1906 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1907 ctx->zoned_bg->meta_write_pointer += eb->len;
1909 write_one_eb(eb, wbc);
1910 free_extent_buffer(eb);
1914 int btree_write_cache_pages(struct address_space *mapping,
1915 struct writeback_control *wbc)
1917 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1918 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
1921 int nr_to_write_done = 0;
1922 struct folio_batch fbatch;
1923 unsigned int nr_folios;
1925 pgoff_t end; /* Inclusive */
1929 folio_batch_init(&fbatch);
1930 if (wbc->range_cyclic) {
1931 index = mapping->writeback_index; /* Start from prev offset */
1934 * Start from the beginning does not need to cycle over the
1935 * range, mark it as scanned.
1937 scanned = (index == 0);
1939 index = wbc->range_start >> PAGE_SHIFT;
1940 end = wbc->range_end >> PAGE_SHIFT;
1943 if (wbc->sync_mode == WB_SYNC_ALL)
1944 tag = PAGECACHE_TAG_TOWRITE;
1946 tag = PAGECACHE_TAG_DIRTY;
1947 btrfs_zoned_meta_io_lock(fs_info);
1949 if (wbc->sync_mode == WB_SYNC_ALL)
1950 tag_pages_for_writeback(mapping, index, end);
1951 while (!done && !nr_to_write_done && (index <= end) &&
1952 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1956 for (i = 0; i < nr_folios; i++) {
1957 struct folio *folio = fbatch.folios[i];
1959 ret = submit_eb_page(&folio->page, &ctx);
1968 * the filesystem may choose to bump up nr_to_write.
1969 * We have to make sure to honor the new nr_to_write
1972 nr_to_write_done = wbc->nr_to_write <= 0;
1974 folio_batch_release(&fbatch);
1977 if (!scanned && !done) {
1979 * We hit the last page and there is more work to be done: wrap
1980 * back to the start of the file
1987 * If something went wrong, don't allow any metadata write bio to be
1990 * This would prevent use-after-free if we had dirty pages not
1991 * cleaned up, which can still happen by fuzzed images.
1994 * Allowing existing tree block to be allocated for other trees.
1996 * - Log tree operations
1997 * Exiting tree blocks get allocated to log tree, bumps its
1998 * generation, then get cleaned in tree re-balance.
1999 * Such tree block will not be written back, since it's clean,
2000 * thus no WRITTEN flag set.
2001 * And after log writes back, this tree block is not traced by
2002 * any dirty extent_io_tree.
2004 * - Offending tree block gets re-dirtied from its original owner
2005 * Since it has bumped generation, no WRITTEN flag, it can be
2006 * reused without COWing. This tree block will not be traced
2007 * by btrfs_transaction::dirty_pages.
2009 * Now such dirty tree block will not be cleaned by any dirty
2010 * extent io tree. Thus we don't want to submit such wild eb
2011 * if the fs already has error.
2013 * We can get ret > 0 from submit_extent_page() indicating how many ebs
2014 * were submitted. Reset it to 0 to avoid false alerts for the caller.
2018 if (!ret && BTRFS_FS_ERROR(fs_info))
2022 btrfs_put_block_group(ctx.zoned_bg);
2023 btrfs_zoned_meta_io_unlock(fs_info);
2028 * Walk the list of dirty pages of the given address space and write all of them.
2030 * @mapping: address space structure to write
2031 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2032 * @bio_ctrl: holds context for the write, namely the bio
2034 * If a page is already under I/O, write_cache_pages() skips it, even
2035 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2036 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2037 * and msync() need to guarantee that all the data which was dirty at the time
2038 * the call was made get new I/O started against them. If wbc->sync_mode is
2039 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2040 * existing IO to complete.
2042 static int extent_write_cache_pages(struct address_space *mapping,
2043 struct btrfs_bio_ctrl *bio_ctrl)
2045 struct writeback_control *wbc = bio_ctrl->wbc;
2046 struct inode *inode = mapping->host;
2049 int nr_to_write_done = 0;
2050 struct folio_batch fbatch;
2051 unsigned int nr_folios;
2053 pgoff_t end; /* Inclusive */
2055 int range_whole = 0;
2060 * We have to hold onto the inode so that ordered extents can do their
2061 * work when the IO finishes. The alternative to this is failing to add
2062 * an ordered extent if the igrab() fails there and that is a huge pain
2063 * to deal with, so instead just hold onto the inode throughout the
2064 * writepages operation. If it fails here we are freeing up the inode
2065 * anyway and we'd rather not waste our time writing out stuff that is
2066 * going to be truncated anyway.
2071 folio_batch_init(&fbatch);
2072 if (wbc->range_cyclic) {
2073 index = mapping->writeback_index; /* Start from prev offset */
2076 * Start from the beginning does not need to cycle over the
2077 * range, mark it as scanned.
2079 scanned = (index == 0);
2081 index = wbc->range_start >> PAGE_SHIFT;
2082 end = wbc->range_end >> PAGE_SHIFT;
2083 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2089 * We do the tagged writepage as long as the snapshot flush bit is set
2090 * and we are the first one who do the filemap_flush() on this inode.
2092 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2093 * not race in and drop the bit.
2095 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2096 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2097 &BTRFS_I(inode)->runtime_flags))
2098 wbc->tagged_writepages = 1;
2100 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2101 tag = PAGECACHE_TAG_TOWRITE;
2103 tag = PAGECACHE_TAG_DIRTY;
2105 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2106 tag_pages_for_writeback(mapping, index, end);
2108 while (!done && !nr_to_write_done && (index <= end) &&
2109 (nr_folios = filemap_get_folios_tag(mapping, &index,
2110 end, tag, &fbatch))) {
2113 for (i = 0; i < nr_folios; i++) {
2114 struct folio *folio = fbatch.folios[i];
2116 done_index = folio_next_index(folio);
2118 * At this point we hold neither the i_pages lock nor
2119 * the page lock: the page may be truncated or
2120 * invalidated (changing page->mapping to NULL),
2121 * or even swizzled back from swapper_space to
2122 * tmpfs file mapping
2124 if (!folio_trylock(folio)) {
2125 submit_write_bio(bio_ctrl, 0);
2129 if (unlikely(folio->mapping != mapping)) {
2130 folio_unlock(folio);
2134 if (!folio_test_dirty(folio)) {
2135 /* Someone wrote it for us. */
2136 folio_unlock(folio);
2140 if (wbc->sync_mode != WB_SYNC_NONE) {
2141 if (folio_test_writeback(folio))
2142 submit_write_bio(bio_ctrl, 0);
2143 folio_wait_writeback(folio);
2146 if (folio_test_writeback(folio) ||
2147 !folio_clear_dirty_for_io(folio)) {
2148 folio_unlock(folio);
2152 ret = __extent_writepage(&folio->page, bio_ctrl);
2159 * The filesystem may choose to bump up nr_to_write.
2160 * We have to make sure to honor the new nr_to_write
2163 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2164 wbc->nr_to_write <= 0);
2166 folio_batch_release(&fbatch);
2169 if (!scanned && !done) {
2171 * We hit the last page and there is more work to be done: wrap
2172 * back to the start of the file
2178 * If we're looping we could run into a page that is locked by a
2179 * writer and that writer could be waiting on writeback for a
2180 * page in our current bio, and thus deadlock, so flush the
2183 submit_write_bio(bio_ctrl, 0);
2187 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2188 mapping->writeback_index = done_index;
2190 btrfs_add_delayed_iput(BTRFS_I(inode));
2195 * Submit the pages in the range to bio for call sites which delalloc range has
2196 * already been ran (aka, ordered extent inserted) and all pages are still
2199 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2200 u64 start, u64 end, struct writeback_control *wbc,
2203 bool found_error = false;
2205 struct address_space *mapping = inode->i_mapping;
2206 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2207 const u32 sectorsize = fs_info->sectorsize;
2208 loff_t i_size = i_size_read(inode);
2210 struct btrfs_bio_ctrl bio_ctrl = {
2212 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2215 if (wbc->no_cgroup_owner)
2216 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2218 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2220 while (cur <= end) {
2221 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2222 u32 cur_len = cur_end + 1 - cur;
2226 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2227 ASSERT(PageLocked(page));
2228 if (pages_dirty && page != locked_page) {
2229 ASSERT(PageDirty(page));
2230 clear_page_dirty_for_io(page);
2233 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2238 /* Make sure the mapping tag for page dirty gets cleared. */
2240 set_page_writeback(page);
2241 end_page_writeback(page);
2244 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2245 cur, cur_len, !ret);
2246 mapping_set_error(page->mapping, ret);
2248 btrfs_folio_unlock_writer(fs_info, page_folio(page), cur, cur_len);
2256 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2259 int extent_writepages(struct address_space *mapping,
2260 struct writeback_control *wbc)
2262 struct inode *inode = mapping->host;
2264 struct btrfs_bio_ctrl bio_ctrl = {
2266 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2270 * Allow only a single thread to do the reloc work in zoned mode to
2271 * protect the write pointer updates.
2273 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2274 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2275 submit_write_bio(&bio_ctrl, ret);
2276 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2280 void extent_readahead(struct readahead_control *rac)
2282 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2283 struct page *pagepool[16];
2284 struct extent_map *em_cached = NULL;
2285 u64 prev_em_start = (u64)-1;
2288 while ((nr = readahead_page_batch(rac, pagepool))) {
2289 u64 contig_start = readahead_pos(rac);
2290 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2292 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2293 &em_cached, &bio_ctrl, &prev_em_start);
2297 free_extent_map(em_cached);
2298 submit_one_bio(&bio_ctrl);
2302 * basic invalidate_folio code, this waits on any locked or writeback
2303 * ranges corresponding to the folio, and then deletes any extent state
2304 * records from the tree
2306 int extent_invalidate_folio(struct extent_io_tree *tree,
2307 struct folio *folio, size_t offset)
2309 struct extent_state *cached_state = NULL;
2310 u64 start = folio_pos(folio);
2311 u64 end = start + folio_size(folio) - 1;
2312 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
2314 /* This function is only called for the btree inode */
2315 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2317 start += ALIGN(offset, blocksize);
2321 lock_extent(tree, start, end, &cached_state);
2322 folio_wait_writeback(folio);
2325 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2326 * so here we only need to unlock the extent range to free any
2327 * existing extent state.
2329 unlock_extent(tree, start, end, &cached_state);
2334 * a helper for release_folio, this tests for areas of the page that
2335 * are locked or under IO and drops the related state bits if it is safe
2338 static int try_release_extent_state(struct extent_io_tree *tree,
2339 struct page *page, gfp_t mask)
2341 u64 start = page_offset(page);
2342 u64 end = start + PAGE_SIZE - 1;
2345 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2348 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2349 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS |
2350 EXTENT_QGROUP_RESERVED);
2353 * At this point we can safely clear everything except the
2354 * locked bit, the nodatasum bit and the delalloc new bit.
2355 * The delalloc new bit will be cleared by ordered extent
2358 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2360 /* if clear_extent_bit failed for enomem reasons,
2361 * we can't allow the release to continue.
2372 * a helper for release_folio. As long as there are no locked extents
2373 * in the range corresponding to the page, both state records and extent
2374 * map records are removed
2376 int try_release_extent_mapping(struct page *page, gfp_t mask)
2378 struct extent_map *em;
2379 u64 start = page_offset(page);
2380 u64 end = start + PAGE_SIZE - 1;
2381 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
2382 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2383 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2385 if (gfpflags_allow_blocking(mask) &&
2386 page->mapping->host->i_size > SZ_16M) {
2388 while (start <= end) {
2389 struct btrfs_fs_info *fs_info;
2392 len = end - start + 1;
2393 write_lock(&map->lock);
2394 em = lookup_extent_mapping(map, start, len);
2396 write_unlock(&map->lock);
2399 if ((em->flags & EXTENT_FLAG_PINNED) ||
2400 em->start != start) {
2401 write_unlock(&map->lock);
2402 free_extent_map(em);
2405 if (test_range_bit_exists(tree, em->start,
2406 extent_map_end(em) - 1,
2410 * If it's not in the list of modified extents, used
2411 * by a fast fsync, we can remove it. If it's being
2412 * logged we can safely remove it since fsync took an
2413 * extra reference on the em.
2415 if (list_empty(&em->list) ||
2416 (em->flags & EXTENT_FLAG_LOGGING))
2419 * If it's in the list of modified extents, remove it
2420 * only if its generation is older then the current one,
2421 * in which case we don't need it for a fast fsync.
2422 * Otherwise don't remove it, we could be racing with an
2423 * ongoing fast fsync that could miss the new extent.
2425 fs_info = btrfs_inode->root->fs_info;
2426 spin_lock(&fs_info->trans_lock);
2427 cur_gen = fs_info->generation;
2428 spin_unlock(&fs_info->trans_lock);
2429 if (em->generation >= cur_gen)
2433 * We only remove extent maps that are not in the list of
2434 * modified extents or that are in the list but with a
2435 * generation lower then the current generation, so there
2436 * is no need to set the full fsync flag on the inode (it
2437 * hurts the fsync performance for workloads with a data
2438 * size that exceeds or is close to the system's memory).
2440 remove_extent_mapping(map, em);
2441 /* once for the rb tree */
2442 free_extent_map(em);
2444 start = extent_map_end(em);
2445 write_unlock(&map->lock);
2448 free_extent_map(em);
2450 cond_resched(); /* Allow large-extent preemption. */
2453 return try_release_extent_state(tree, page, mask);
2457 * To cache previous fiemap extent
2459 * Will be used for merging fiemap extent
2461 struct fiemap_cache {
2470 * Helper to submit fiemap extent.
2472 * Will try to merge current fiemap extent specified by @offset, @phys,
2473 * @len and @flags with cached one.
2474 * And only when we fails to merge, cached one will be submitted as
2477 * Return value is the same as fiemap_fill_next_extent().
2479 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2480 struct fiemap_cache *cache,
2481 u64 offset, u64 phys, u64 len, u32 flags)
2485 /* Set at the end of extent_fiemap(). */
2486 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2492 * Sanity check, extent_fiemap() should have ensured that new
2493 * fiemap extent won't overlap with cached one.
2496 * NOTE: Physical address can overlap, due to compression
2498 if (cache->offset + cache->len > offset) {
2504 * Only merges fiemap extents if
2505 * 1) Their logical addresses are continuous
2507 * 2) Their physical addresses are continuous
2508 * So truly compressed (physical size smaller than logical size)
2509 * extents won't get merged with each other
2511 * 3) Share same flags
2513 if (cache->offset + cache->len == offset &&
2514 cache->phys + cache->len == phys &&
2515 cache->flags == flags) {
2520 /* Not mergeable, need to submit cached one */
2521 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2522 cache->len, cache->flags);
2523 cache->cached = false;
2527 cache->cached = true;
2528 cache->offset = offset;
2531 cache->flags = flags;
2537 * Emit last fiemap cache
2539 * The last fiemap cache may still be cached in the following case:
2541 * |<- Fiemap range ->|
2542 * |<------------ First extent ----------->|
2544 * In this case, the first extent range will be cached but not emitted.
2545 * So we must emit it before ending extent_fiemap().
2547 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2548 struct fiemap_cache *cache)
2555 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2556 cache->len, cache->flags);
2557 cache->cached = false;
2563 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2565 struct extent_buffer *clone;
2566 struct btrfs_key key;
2571 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2574 ret = btrfs_next_leaf(inode->root, path);
2579 * Don't bother with cloning if there are no more file extent items for
2582 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2583 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
2586 /* See the comment at fiemap_search_slot() about why we clone. */
2587 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2591 slot = path->slots[0];
2592 btrfs_release_path(path);
2593 path->nodes[0] = clone;
2594 path->slots[0] = slot;
2600 * Search for the first file extent item that starts at a given file offset or
2601 * the one that starts immediately before that offset.
2602 * Returns: 0 on success, < 0 on error, 1 if not found.
2604 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2607 const u64 ino = btrfs_ino(inode);
2608 struct btrfs_root *root = inode->root;
2609 struct extent_buffer *clone;
2610 struct btrfs_key key;
2615 key.type = BTRFS_EXTENT_DATA_KEY;
2616 key.offset = file_offset;
2618 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2622 if (ret > 0 && path->slots[0] > 0) {
2623 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2624 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2628 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2629 ret = btrfs_next_leaf(root, path);
2633 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2634 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2639 * We clone the leaf and use it during fiemap. This is because while
2640 * using the leaf we do expensive things like checking if an extent is
2641 * shared, which can take a long time. In order to prevent blocking
2642 * other tasks for too long, we use a clone of the leaf. We have locked
2643 * the file range in the inode's io tree, so we know none of our file
2644 * extent items can change. This way we avoid blocking other tasks that
2645 * want to insert items for other inodes in the same leaf or b+tree
2646 * rebalance operations (triggered for example when someone is trying
2647 * to push items into this leaf when trying to insert an item in a
2649 * We also need the private clone because holding a read lock on an
2650 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2651 * when we call fiemap_fill_next_extent(), because that may cause a page
2652 * fault when filling the user space buffer with fiemap data.
2654 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2658 slot = path->slots[0];
2659 btrfs_release_path(path);
2660 path->nodes[0] = clone;
2661 path->slots[0] = slot;
2667 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2668 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2669 * extent. The end offset (@end) is inclusive.
2671 static int fiemap_process_hole(struct btrfs_inode *inode,
2672 struct fiemap_extent_info *fieinfo,
2673 struct fiemap_cache *cache,
2674 struct extent_state **delalloc_cached_state,
2675 struct btrfs_backref_share_check_ctx *backref_ctx,
2676 u64 disk_bytenr, u64 extent_offset,
2680 const u64 i_size = i_size_read(&inode->vfs_inode);
2681 u64 cur_offset = start;
2682 u64 last_delalloc_end = 0;
2683 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2684 bool checked_extent_shared = false;
2688 * There can be no delalloc past i_size, so don't waste time looking for
2691 while (cur_offset < end && cur_offset < i_size) {
2695 u64 prealloc_len = 0;
2698 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2699 delalloc_cached_state,
2706 * If this is a prealloc extent we have to report every section
2707 * of it that has no delalloc.
2709 if (disk_bytenr != 0) {
2710 if (last_delalloc_end == 0) {
2711 prealloc_start = start;
2712 prealloc_len = delalloc_start - start;
2714 prealloc_start = last_delalloc_end + 1;
2715 prealloc_len = delalloc_start - prealloc_start;
2719 if (prealloc_len > 0) {
2720 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2721 ret = btrfs_is_data_extent_shared(inode,
2728 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2730 checked_extent_shared = true;
2732 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2733 disk_bytenr + extent_offset,
2734 prealloc_len, prealloc_flags);
2737 extent_offset += prealloc_len;
2740 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2741 delalloc_end + 1 - delalloc_start,
2742 FIEMAP_EXTENT_DELALLOC |
2743 FIEMAP_EXTENT_UNKNOWN);
2747 last_delalloc_end = delalloc_end;
2748 cur_offset = delalloc_end + 1;
2749 extent_offset += cur_offset - delalloc_start;
2754 * Either we found no delalloc for the whole prealloc extent or we have
2755 * a prealloc extent that spans i_size or starts at or after i_size.
2757 if (disk_bytenr != 0 && last_delalloc_end < end) {
2761 if (last_delalloc_end == 0) {
2762 prealloc_start = start;
2763 prealloc_len = end + 1 - start;
2765 prealloc_start = last_delalloc_end + 1;
2766 prealloc_len = end + 1 - prealloc_start;
2769 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2770 ret = btrfs_is_data_extent_shared(inode,
2777 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2779 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2780 disk_bytenr + extent_offset,
2781 prealloc_len, prealloc_flags);
2789 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2790 struct btrfs_path *path,
2791 u64 *last_extent_end_ret)
2793 const u64 ino = btrfs_ino(inode);
2794 struct btrfs_root *root = inode->root;
2795 struct extent_buffer *leaf;
2796 struct btrfs_file_extent_item *ei;
2797 struct btrfs_key key;
2802 * Lookup the last file extent. We're not using i_size here because
2803 * there might be preallocation past i_size.
2805 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
2806 /* There can't be a file extent item at offset (u64)-1 */
2812 * For a non-existing key, btrfs_search_slot() always leaves us at a
2813 * slot > 0, except if the btree is empty, which is impossible because
2814 * at least it has the inode item for this inode and all the items for
2815 * the root inode 256.
2817 ASSERT(path->slots[0] > 0);
2819 leaf = path->nodes[0];
2820 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2821 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
2822 /* No file extent items in the subvolume tree. */
2823 *last_extent_end_ret = 0;
2828 * For an inline extent, the disk_bytenr is where inline data starts at,
2829 * so first check if we have an inline extent item before checking if we
2830 * have an implicit hole (disk_bytenr == 0).
2832 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2833 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
2834 *last_extent_end_ret = btrfs_file_extent_end(path);
2839 * Find the last file extent item that is not a hole (when NO_HOLES is
2840 * not enabled). This should take at most 2 iterations in the worst
2841 * case: we have one hole file extent item at slot 0 of a leaf and
2842 * another hole file extent item as the last item in the previous leaf.
2843 * This is because we merge file extent items that represent holes.
2845 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2846 while (disk_bytenr == 0) {
2847 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
2850 } else if (ret > 0) {
2851 /* No file extent items that are not holes. */
2852 *last_extent_end_ret = 0;
2855 leaf = path->nodes[0];
2856 ei = btrfs_item_ptr(leaf, path->slots[0],
2857 struct btrfs_file_extent_item);
2858 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2861 *last_extent_end_ret = btrfs_file_extent_end(path);
2865 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
2868 const u64 ino = btrfs_ino(inode);
2869 struct extent_state *cached_state = NULL;
2870 struct extent_state *delalloc_cached_state = NULL;
2871 struct btrfs_path *path;
2872 struct fiemap_cache cache = { 0 };
2873 struct btrfs_backref_share_check_ctx *backref_ctx;
2874 u64 last_extent_end;
2875 u64 prev_extent_end;
2878 bool stopped = false;
2881 backref_ctx = btrfs_alloc_backref_share_check_ctx();
2882 path = btrfs_alloc_path();
2883 if (!backref_ctx || !path) {
2888 lockstart = round_down(start, inode->root->fs_info->sectorsize);
2889 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
2890 prev_extent_end = lockstart;
2892 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
2893 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
2895 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
2898 btrfs_release_path(path);
2900 path->reada = READA_FORWARD;
2901 ret = fiemap_search_slot(inode, path, lockstart);
2904 } else if (ret > 0) {
2906 * No file extent item found, but we may have delalloc between
2907 * the current offset and i_size. So check for that.
2910 goto check_eof_delalloc;
2913 while (prev_extent_end < lockend) {
2914 struct extent_buffer *leaf = path->nodes[0];
2915 struct btrfs_file_extent_item *ei;
2916 struct btrfs_key key;
2919 u64 extent_offset = 0;
2921 u64 disk_bytenr = 0;
2926 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2927 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2930 extent_end = btrfs_file_extent_end(path);
2933 * The first iteration can leave us at an extent item that ends
2934 * before our range's start. Move to the next item.
2936 if (extent_end <= lockstart)
2939 backref_ctx->curr_leaf_bytenr = leaf->start;
2941 /* We have in implicit hole (NO_HOLES feature enabled). */
2942 if (prev_extent_end < key.offset) {
2943 const u64 range_end = min(key.offset, lockend) - 1;
2945 ret = fiemap_process_hole(inode, fieinfo, &cache,
2946 &delalloc_cached_state,
2947 backref_ctx, 0, 0, 0,
2948 prev_extent_end, range_end);
2951 } else if (ret > 0) {
2952 /* fiemap_fill_next_extent() told us to stop. */
2957 /* We've reached the end of the fiemap range, stop. */
2958 if (key.offset >= lockend) {
2964 extent_len = extent_end - key.offset;
2965 ei = btrfs_item_ptr(leaf, path->slots[0],
2966 struct btrfs_file_extent_item);
2967 compression = btrfs_file_extent_compression(leaf, ei);
2968 extent_type = btrfs_file_extent_type(leaf, ei);
2969 extent_gen = btrfs_file_extent_generation(leaf, ei);
2971 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2972 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2973 if (compression == BTRFS_COMPRESS_NONE)
2974 extent_offset = btrfs_file_extent_offset(leaf, ei);
2977 if (compression != BTRFS_COMPRESS_NONE)
2978 flags |= FIEMAP_EXTENT_ENCODED;
2980 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2981 flags |= FIEMAP_EXTENT_DATA_INLINE;
2982 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
2983 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
2985 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2986 ret = fiemap_process_hole(inode, fieinfo, &cache,
2987 &delalloc_cached_state,
2989 disk_bytenr, extent_offset,
2990 extent_gen, key.offset,
2992 } else if (disk_bytenr == 0) {
2993 /* We have an explicit hole. */
2994 ret = fiemap_process_hole(inode, fieinfo, &cache,
2995 &delalloc_cached_state,
2996 backref_ctx, 0, 0, 0,
2997 key.offset, extent_end - 1);
2999 /* We have a regular extent. */
3000 if (fieinfo->fi_extents_max) {
3001 ret = btrfs_is_data_extent_shared(inode,
3008 flags |= FIEMAP_EXTENT_SHARED;
3011 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
3012 disk_bytenr + extent_offset,
3018 } else if (ret > 0) {
3019 /* fiemap_fill_next_extent() told us to stop. */
3024 prev_extent_end = extent_end;
3026 if (fatal_signal_pending(current)) {
3031 ret = fiemap_next_leaf_item(inode, path);
3034 } else if (ret > 0) {
3035 /* No more file extent items for this inode. */
3043 * Release (and free) the path before emitting any final entries to
3044 * fiemap_fill_next_extent() to keep lockdep happy. This is because
3045 * once we find no more file extent items exist, we may have a
3046 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
3047 * faults when copying data to the user space buffer.
3049 btrfs_free_path(path);
3052 if (!stopped && prev_extent_end < lockend) {
3053 ret = fiemap_process_hole(inode, fieinfo, &cache,
3054 &delalloc_cached_state, backref_ctx,
3055 0, 0, 0, prev_extent_end, lockend - 1);
3058 prev_extent_end = lockend;
3061 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3062 const u64 i_size = i_size_read(&inode->vfs_inode);
3064 if (prev_extent_end < i_size) {
3069 delalloc = btrfs_find_delalloc_in_range(inode,
3072 &delalloc_cached_state,
3076 cache.flags |= FIEMAP_EXTENT_LAST;
3078 cache.flags |= FIEMAP_EXTENT_LAST;
3082 ret = emit_last_fiemap_cache(fieinfo, &cache);
3085 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3086 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3088 free_extent_state(delalloc_cached_state);
3089 btrfs_free_backref_share_ctx(backref_ctx);
3090 btrfs_free_path(path);
3094 static void __free_extent_buffer(struct extent_buffer *eb)
3096 kmem_cache_free(extent_buffer_cache, eb);
3099 static int extent_buffer_under_io(const struct extent_buffer *eb)
3101 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3102 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3105 static bool folio_range_has_eb(struct btrfs_fs_info *fs_info, struct folio *folio)
3107 struct btrfs_subpage *subpage;
3109 lockdep_assert_held(&folio->mapping->private_lock);
3111 if (folio_test_private(folio)) {
3112 subpage = folio_get_private(folio);
3113 if (atomic_read(&subpage->eb_refs))
3116 * Even there is no eb refs here, we may still have
3117 * end_page_read() call relying on page::private.
3119 if (atomic_read(&subpage->readers))
3125 static void detach_extent_buffer_folio(struct extent_buffer *eb, struct folio *folio)
3127 struct btrfs_fs_info *fs_info = eb->fs_info;
3128 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3131 * For mapped eb, we're going to change the folio private, which should
3132 * be done under the private_lock.
3135 spin_lock(&folio->mapping->private_lock);
3137 if (!folio_test_private(folio)) {
3139 spin_unlock(&folio->mapping->private_lock);
3143 if (fs_info->nodesize >= PAGE_SIZE) {
3145 * We do this since we'll remove the pages after we've
3146 * removed the eb from the radix tree, so we could race
3147 * and have this page now attached to the new eb. So
3148 * only clear folio if it's still connected to
3151 if (folio_test_private(folio) && folio_get_private(folio) == eb) {
3152 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3153 BUG_ON(folio_test_dirty(folio));
3154 BUG_ON(folio_test_writeback(folio));
3155 /* We need to make sure we haven't be attached to a new eb. */
3156 folio_detach_private(folio);
3159 spin_unlock(&folio->mapping->private_lock);
3164 * For subpage, we can have dummy eb with folio private attached. In
3165 * this case, we can directly detach the private as such folio is only
3166 * attached to one dummy eb, no sharing.
3169 btrfs_detach_subpage(fs_info, folio);
3173 btrfs_folio_dec_eb_refs(fs_info, folio);
3176 * We can only detach the folio private if there are no other ebs in the
3177 * page range and no unfinished IO.
3179 if (!folio_range_has_eb(fs_info, folio))
3180 btrfs_detach_subpage(fs_info, folio);
3182 spin_unlock(&folio->mapping->private_lock);
3185 /* Release all pages attached to the extent buffer */
3186 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3188 ASSERT(!extent_buffer_under_io(eb));
3190 for (int i = 0; i < INLINE_EXTENT_BUFFER_PAGES; i++) {
3191 struct folio *folio = eb->folios[i];
3196 detach_extent_buffer_folio(eb, folio);
3198 /* One for when we allocated the folio. */
3204 * Helper for releasing the extent buffer.
3206 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3208 btrfs_release_extent_buffer_pages(eb);
3209 btrfs_leak_debug_del_eb(eb);
3210 __free_extent_buffer(eb);
3213 static struct extent_buffer *
3214 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3217 struct extent_buffer *eb = NULL;
3219 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3222 eb->fs_info = fs_info;
3223 init_rwsem(&eb->lock);
3225 btrfs_leak_debug_add_eb(eb);
3227 spin_lock_init(&eb->refs_lock);
3228 atomic_set(&eb->refs, 1);
3230 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3235 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3237 struct extent_buffer *new;
3238 int num_folios = num_extent_folios(src);
3241 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3246 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3247 * btrfs_release_extent_buffer() have different behavior for
3248 * UNMAPPED subpage extent buffer.
3250 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3252 ret = alloc_eb_folio_array(new, 0);
3254 btrfs_release_extent_buffer(new);
3258 for (int i = 0; i < num_folios; i++) {
3259 struct folio *folio = new->folios[i];
3262 ret = attach_extent_buffer_folio(new, folio, NULL);
3264 btrfs_release_extent_buffer(new);
3267 WARN_ON(folio_test_dirty(folio));
3269 copy_extent_buffer_full(new, src);
3270 set_extent_buffer_uptodate(new);
3275 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3276 u64 start, unsigned long len)
3278 struct extent_buffer *eb;
3282 eb = __alloc_extent_buffer(fs_info, start, len);
3286 ret = alloc_eb_folio_array(eb, 0);
3290 num_folios = num_extent_folios(eb);
3291 for (int i = 0; i < num_folios; i++) {
3292 ret = attach_extent_buffer_folio(eb, eb->folios[i], NULL);
3297 set_extent_buffer_uptodate(eb);
3298 btrfs_set_header_nritems(eb, 0);
3299 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3303 for (int i = 0; i < num_folios; i++) {
3304 if (eb->folios[i]) {
3305 detach_extent_buffer_folio(eb, eb->folios[i]);
3306 __folio_put(eb->folios[i]);
3309 __free_extent_buffer(eb);
3313 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3316 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3319 static void check_buffer_tree_ref(struct extent_buffer *eb)
3323 * The TREE_REF bit is first set when the extent_buffer is added
3324 * to the radix tree. It is also reset, if unset, when a new reference
3325 * is created by find_extent_buffer.
3327 * It is only cleared in two cases: freeing the last non-tree
3328 * reference to the extent_buffer when its STALE bit is set or
3329 * calling release_folio when the tree reference is the only reference.
3331 * In both cases, care is taken to ensure that the extent_buffer's
3332 * pages are not under io. However, release_folio can be concurrently
3333 * called with creating new references, which is prone to race
3334 * conditions between the calls to check_buffer_tree_ref in those
3335 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3337 * The actual lifetime of the extent_buffer in the radix tree is
3338 * adequately protected by the refcount, but the TREE_REF bit and
3339 * its corresponding reference are not. To protect against this
3340 * class of races, we call check_buffer_tree_ref from the codepaths
3341 * which trigger io. Note that once io is initiated, TREE_REF can no
3342 * longer be cleared, so that is the moment at which any such race is
3345 refs = atomic_read(&eb->refs);
3346 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3349 spin_lock(&eb->refs_lock);
3350 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3351 atomic_inc(&eb->refs);
3352 spin_unlock(&eb->refs_lock);
3355 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
3357 int num_folios= num_extent_folios(eb);
3359 check_buffer_tree_ref(eb);
3361 for (int i = 0; i < num_folios; i++)
3362 folio_mark_accessed(eb->folios[i]);
3365 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3368 struct extent_buffer *eb;
3370 eb = find_extent_buffer_nolock(fs_info, start);
3374 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3375 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3376 * another task running free_extent_buffer() might have seen that flag
3377 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3378 * writeback flags not set) and it's still in the tree (flag
3379 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3380 * decrementing the extent buffer's reference count twice. So here we
3381 * could race and increment the eb's reference count, clear its stale
3382 * flag, mark it as dirty and drop our reference before the other task
3383 * finishes executing free_extent_buffer, which would later result in
3384 * an attempt to free an extent buffer that is dirty.
3386 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3387 spin_lock(&eb->refs_lock);
3388 spin_unlock(&eb->refs_lock);
3390 mark_extent_buffer_accessed(eb);
3394 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3395 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3398 struct extent_buffer *eb, *exists = NULL;
3401 eb = find_extent_buffer(fs_info, start);
3404 eb = alloc_dummy_extent_buffer(fs_info, start);
3406 return ERR_PTR(-ENOMEM);
3407 eb->fs_info = fs_info;
3409 ret = radix_tree_preload(GFP_NOFS);
3411 exists = ERR_PTR(ret);
3414 spin_lock(&fs_info->buffer_lock);
3415 ret = radix_tree_insert(&fs_info->buffer_radix,
3416 start >> fs_info->sectorsize_bits, eb);
3417 spin_unlock(&fs_info->buffer_lock);
3418 radix_tree_preload_end();
3419 if (ret == -EEXIST) {
3420 exists = find_extent_buffer(fs_info, start);
3426 check_buffer_tree_ref(eb);
3427 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3431 btrfs_release_extent_buffer(eb);
3436 static struct extent_buffer *grab_extent_buffer(
3437 struct btrfs_fs_info *fs_info, struct page *page)
3439 struct folio *folio = page_folio(page);
3440 struct extent_buffer *exists;
3443 * For subpage case, we completely rely on radix tree to ensure we
3444 * don't try to insert two ebs for the same bytenr. So here we always
3445 * return NULL and just continue.
3447 if (fs_info->nodesize < PAGE_SIZE)
3450 /* Page not yet attached to an extent buffer */
3451 if (!folio_test_private(folio))
3455 * We could have already allocated an eb for this page and attached one
3456 * so lets see if we can get a ref on the existing eb, and if we can we
3457 * know it's good and we can just return that one, else we know we can
3458 * just overwrite folio private.
3460 exists = folio_get_private(folio);
3461 if (atomic_inc_not_zero(&exists->refs))
3464 WARN_ON(PageDirty(page));
3465 folio_detach_private(folio);
3469 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3471 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3472 btrfs_err(fs_info, "bad tree block start %llu", start);
3476 if (fs_info->nodesize < PAGE_SIZE &&
3477 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3479 "tree block crosses page boundary, start %llu nodesize %u",
3480 start, fs_info->nodesize);
3483 if (fs_info->nodesize >= PAGE_SIZE &&
3484 !PAGE_ALIGNED(start)) {
3486 "tree block is not page aligned, start %llu nodesize %u",
3487 start, fs_info->nodesize);
3490 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3491 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3493 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3494 start, fs_info->nodesize);
3501 * Return 0 if eb->folios[i] is attached to btree inode successfully.
3502 * Return >0 if there is already another extent buffer for the range,
3503 * and @found_eb_ret would be updated.
3504 * Return -EAGAIN if the filemap has an existing folio but with different size
3506 * The caller needs to free the existing folios and retry using the same order.
3508 static int attach_eb_folio_to_filemap(struct extent_buffer *eb, int i,
3509 struct extent_buffer **found_eb_ret)
3512 struct btrfs_fs_info *fs_info = eb->fs_info;
3513 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3514 const unsigned long index = eb->start >> PAGE_SHIFT;
3515 struct folio *existing_folio;
3518 ASSERT(found_eb_ret);
3520 /* Caller should ensure the folio exists. */
3521 ASSERT(eb->folios[i]);
3524 ret = filemap_add_folio(mapping, eb->folios[i], index + i,
3525 GFP_NOFS | __GFP_NOFAIL);
3529 existing_folio = filemap_lock_folio(mapping, index + i);
3530 /* The page cache only exists for a very short time, just retry. */
3531 if (IS_ERR(existing_folio))
3534 /* For now, we should only have single-page folios for btree inode. */
3535 ASSERT(folio_nr_pages(existing_folio) == 1);
3537 if (folio_size(existing_folio) != folio_size(eb->folios[0])) {
3538 folio_unlock(existing_folio);
3539 folio_put(existing_folio);
3543 if (fs_info->nodesize < PAGE_SIZE) {
3545 * We're going to reuse the existing page, can drop our page
3546 * and subpage structure now.
3548 __free_page(folio_page(eb->folios[i], 0));
3549 eb->folios[i] = existing_folio;
3551 struct extent_buffer *existing_eb;
3553 existing_eb = grab_extent_buffer(fs_info,
3554 folio_page(existing_folio, 0));
3556 /* The extent buffer still exists, we can use it directly. */
3557 *found_eb_ret = existing_eb;
3558 folio_unlock(existing_folio);
3559 folio_put(existing_folio);
3562 /* The extent buffer no longer exists, we can reuse the folio. */
3563 __free_page(folio_page(eb->folios[i], 0));
3564 eb->folios[i] = existing_folio;
3569 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3570 u64 start, u64 owner_root, int level)
3572 unsigned long len = fs_info->nodesize;
3575 struct extent_buffer *eb;
3576 struct extent_buffer *existing_eb = NULL;
3577 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3578 struct btrfs_subpage *prealloc = NULL;
3579 u64 lockdep_owner = owner_root;
3580 bool page_contig = true;
3584 if (check_eb_alignment(fs_info, start))
3585 return ERR_PTR(-EINVAL);
3587 #if BITS_PER_LONG == 32
3588 if (start >= MAX_LFS_FILESIZE) {
3589 btrfs_err_rl(fs_info,
3590 "extent buffer %llu is beyond 32bit page cache limit", start);
3591 btrfs_err_32bit_limit(fs_info);
3592 return ERR_PTR(-EOVERFLOW);
3594 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3595 btrfs_warn_32bit_limit(fs_info);
3598 eb = find_extent_buffer(fs_info, start);
3602 eb = __alloc_extent_buffer(fs_info, start, len);
3604 return ERR_PTR(-ENOMEM);
3607 * The reloc trees are just snapshots, so we need them to appear to be
3608 * just like any other fs tree WRT lockdep.
3610 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3611 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3613 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3616 * Preallocate folio private for subpage case, so that we won't
3617 * allocate memory with private_lock nor page lock hold.
3619 * The memory will be freed by attach_extent_buffer_page() or freed
3620 * manually if we exit earlier.
3622 if (fs_info->nodesize < PAGE_SIZE) {
3623 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3624 if (IS_ERR(prealloc)) {
3625 ret = PTR_ERR(prealloc);
3631 /* Allocate all pages first. */
3632 ret = alloc_eb_folio_array(eb, __GFP_NOFAIL);
3634 btrfs_free_subpage(prealloc);
3638 num_folios = num_extent_folios(eb);
3639 /* Attach all pages to the filemap. */
3640 for (int i = 0; i < num_folios; i++) {
3641 struct folio *folio;
3643 ret = attach_eb_folio_to_filemap(eb, i, &existing_eb);
3645 ASSERT(existing_eb);
3650 * TODO: Special handling for a corner case where the order of
3651 * folios mismatch between the new eb and filemap.
3653 * This happens when:
3655 * - the new eb is using higher order folio
3657 * - the filemap is still using 0-order folios for the range
3658 * This can happen at the previous eb allocation, and we don't
3659 * have higher order folio for the call.
3661 * - the existing eb has already been freed
3663 * In this case, we have to free the existing folios first, and
3664 * re-allocate using the same order.
3665 * Thankfully this is not going to happen yet, as we're still
3666 * using 0-order folios.
3668 if (unlikely(ret == -EAGAIN)) {
3675 * Only after attach_eb_folio_to_filemap(), eb->folios[] is
3676 * reliable, as we may choose to reuse the existing page cache
3677 * and free the allocated page.
3679 folio = eb->folios[i];
3680 spin_lock(&mapping->private_lock);
3681 /* Should not fail, as we have preallocated the memory */
3682 ret = attach_extent_buffer_folio(eb, folio, prealloc);
3685 * To inform we have extra eb under allocation, so that
3686 * detach_extent_buffer_page() won't release the folio private
3687 * when the eb hasn't yet been inserted into radix tree.
3689 * The ref will be decreased when the eb released the page, in
3690 * detach_extent_buffer_page().
3691 * Thus needs no special handling in error path.
3693 btrfs_folio_inc_eb_refs(fs_info, folio);
3694 spin_unlock(&mapping->private_lock);
3696 WARN_ON(btrfs_folio_test_dirty(fs_info, folio, eb->start, eb->len));
3699 * Check if the current page is physically contiguous with previous eb
3701 * At this stage, either we allocated a large folio, thus @i
3702 * would only be 0, or we fall back to per-page allocation.
3704 if (i && folio_page(eb->folios[i - 1], 0) + 1 != folio_page(folio, 0))
3705 page_contig = false;
3707 if (!btrfs_folio_test_uptodate(fs_info, folio, eb->start, eb->len))
3711 * We can't unlock the pages just yet since the extent buffer
3712 * hasn't been properly inserted in the radix tree, this
3713 * opens a race with btree_release_folio which can free a page
3714 * while we are still filling in all pages for the buffer and
3719 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3720 /* All pages are physically contiguous, can skip cross page handling. */
3722 eb->addr = folio_address(eb->folios[0]) + offset_in_page(eb->start);
3724 ret = radix_tree_preload(GFP_NOFS);
3728 spin_lock(&fs_info->buffer_lock);
3729 ret = radix_tree_insert(&fs_info->buffer_radix,
3730 start >> fs_info->sectorsize_bits, eb);
3731 spin_unlock(&fs_info->buffer_lock);
3732 radix_tree_preload_end();
3733 if (ret == -EEXIST) {
3735 existing_eb = find_extent_buffer(fs_info, start);
3741 /* add one reference for the tree */
3742 check_buffer_tree_ref(eb);
3743 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3746 * Now it's safe to unlock the pages because any calls to
3747 * btree_release_folio will correctly detect that a page belongs to a
3748 * live buffer and won't free them prematurely.
3750 for (int i = 0; i < num_folios; i++)
3751 unlock_page(folio_page(eb->folios[i], 0));
3755 WARN_ON(!atomic_dec_and_test(&eb->refs));
3758 * Any attached folios need to be detached before we unlock them. This
3759 * is because when we're inserting our new folios into the mapping, and
3760 * then attaching our eb to that folio. If we fail to insert our folio
3761 * we'll lookup the folio for that index, and grab that EB. We do not
3762 * want that to grab this eb, as we're getting ready to free it. So we
3763 * have to detach it first and then unlock it.
3765 * We have to drop our reference and NULL it out here because in the
3766 * subpage case detaching does a btrfs_folio_dec_eb_refs() for our eb.
3767 * Below when we call btrfs_release_extent_buffer() we will call
3768 * detach_extent_buffer_folio() on our remaining pages in the !subpage
3769 * case. If we left eb->folios[i] populated in the subpage case we'd
3770 * double put our reference and be super sad.
3772 for (int i = 0; i < attached; i++) {
3773 ASSERT(eb->folios[i]);
3774 detach_extent_buffer_folio(eb, eb->folios[i]);
3775 unlock_page(folio_page(eb->folios[i], 0));
3776 folio_put(eb->folios[i]);
3777 eb->folios[i] = NULL;
3780 * Now all pages of that extent buffer is unmapped, set UNMAPPED flag,
3781 * so it can be cleaned up without utlizing page->mapping.
3783 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3785 btrfs_release_extent_buffer(eb);
3787 return ERR_PTR(ret);
3788 ASSERT(existing_eb);
3792 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3794 struct extent_buffer *eb =
3795 container_of(head, struct extent_buffer, rcu_head);
3797 __free_extent_buffer(eb);
3800 static int release_extent_buffer(struct extent_buffer *eb)
3801 __releases(&eb->refs_lock)
3803 lockdep_assert_held(&eb->refs_lock);
3805 WARN_ON(atomic_read(&eb->refs) == 0);
3806 if (atomic_dec_and_test(&eb->refs)) {
3807 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
3808 struct btrfs_fs_info *fs_info = eb->fs_info;
3810 spin_unlock(&eb->refs_lock);
3812 spin_lock(&fs_info->buffer_lock);
3813 radix_tree_delete(&fs_info->buffer_radix,
3814 eb->start >> fs_info->sectorsize_bits);
3815 spin_unlock(&fs_info->buffer_lock);
3817 spin_unlock(&eb->refs_lock);
3820 btrfs_leak_debug_del_eb(eb);
3821 /* Should be safe to release our pages at this point */
3822 btrfs_release_extent_buffer_pages(eb);
3823 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3824 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
3825 __free_extent_buffer(eb);
3829 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
3832 spin_unlock(&eb->refs_lock);
3837 void free_extent_buffer(struct extent_buffer *eb)
3843 refs = atomic_read(&eb->refs);
3845 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
3846 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
3849 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
3853 spin_lock(&eb->refs_lock);
3854 if (atomic_read(&eb->refs) == 2 &&
3855 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
3856 !extent_buffer_under_io(eb) &&
3857 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3858 atomic_dec(&eb->refs);
3861 * I know this is terrible, but it's temporary until we stop tracking
3862 * the uptodate bits and such for the extent buffers.
3864 release_extent_buffer(eb);
3867 void free_extent_buffer_stale(struct extent_buffer *eb)
3872 spin_lock(&eb->refs_lock);
3873 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
3875 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3876 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3877 atomic_dec(&eb->refs);
3878 release_extent_buffer(eb);
3881 static void btree_clear_folio_dirty(struct folio *folio)
3883 ASSERT(folio_test_dirty(folio));
3884 ASSERT(folio_test_locked(folio));
3885 folio_clear_dirty_for_io(folio);
3886 xa_lock_irq(&folio->mapping->i_pages);
3887 if (!folio_test_dirty(folio))
3888 __xa_clear_mark(&folio->mapping->i_pages,
3889 folio_index(folio), PAGECACHE_TAG_DIRTY);
3890 xa_unlock_irq(&folio->mapping->i_pages);
3893 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
3895 struct btrfs_fs_info *fs_info = eb->fs_info;
3896 struct folio *folio = eb->folios[0];
3899 /* btree_clear_folio_dirty() needs page locked. */
3901 last = btrfs_subpage_clear_and_test_dirty(fs_info, folio, eb->start, eb->len);
3903 btree_clear_folio_dirty(folio);
3904 folio_unlock(folio);
3905 WARN_ON(atomic_read(&eb->refs) == 0);
3908 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
3909 struct extent_buffer *eb)
3911 struct btrfs_fs_info *fs_info = eb->fs_info;
3914 btrfs_assert_tree_write_locked(eb);
3916 if (trans && btrfs_header_generation(eb) != trans->transid)
3920 * Instead of clearing the dirty flag off of the buffer, mark it as
3921 * EXTENT_BUFFER_ZONED_ZEROOUT. This allows us to preserve
3922 * write-ordering in zoned mode, without the need to later re-dirty
3923 * the extent_buffer.
3925 * The actual zeroout of the buffer will happen later in
3926 * btree_csum_one_bio.
3928 if (btrfs_is_zoned(fs_info)) {
3929 set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
3933 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
3936 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
3937 fs_info->dirty_metadata_batch);
3939 if (eb->fs_info->nodesize < PAGE_SIZE)
3940 return clear_subpage_extent_buffer_dirty(eb);
3942 num_folios = num_extent_folios(eb);
3943 for (int i = 0; i < num_folios; i++) {
3944 struct folio *folio = eb->folios[i];
3946 if (!folio_test_dirty(folio))
3949 btree_clear_folio_dirty(folio);
3950 folio_unlock(folio);
3952 WARN_ON(atomic_read(&eb->refs) == 0);
3955 void set_extent_buffer_dirty(struct extent_buffer *eb)
3960 check_buffer_tree_ref(eb);
3962 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3964 num_folios = num_extent_folios(eb);
3965 WARN_ON(atomic_read(&eb->refs) == 0);
3966 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
3969 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
3972 * For subpage case, we can have other extent buffers in the
3973 * same page, and in clear_subpage_extent_buffer_dirty() we
3974 * have to clear page dirty without subpage lock held.
3975 * This can cause race where our page gets dirty cleared after
3978 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
3979 * its page for other reasons, we can use page lock to prevent
3983 lock_page(folio_page(eb->folios[0], 0));
3984 for (int i = 0; i < num_folios; i++)
3985 btrfs_folio_set_dirty(eb->fs_info, eb->folios[i],
3986 eb->start, eb->len);
3988 unlock_page(folio_page(eb->folios[0], 0));
3989 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
3991 eb->fs_info->dirty_metadata_batch);
3993 #ifdef CONFIG_BTRFS_DEBUG
3994 for (int i = 0; i < num_folios; i++)
3995 ASSERT(folio_test_dirty(eb->folios[i]));
3999 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
4001 struct btrfs_fs_info *fs_info = eb->fs_info;
4002 int num_folios = num_extent_folios(eb);
4004 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4005 for (int i = 0; i < num_folios; i++) {
4006 struct folio *folio = eb->folios[i];
4012 * This is special handling for metadata subpage, as regular
4013 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4015 if (fs_info->nodesize >= PAGE_SIZE)
4016 folio_clear_uptodate(folio);
4018 btrfs_subpage_clear_uptodate(fs_info, folio,
4019 eb->start, eb->len);
4023 void set_extent_buffer_uptodate(struct extent_buffer *eb)
4025 struct btrfs_fs_info *fs_info = eb->fs_info;
4026 int num_folios = num_extent_folios(eb);
4028 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4029 for (int i = 0; i < num_folios; i++) {
4030 struct folio *folio = eb->folios[i];
4033 * This is special handling for metadata subpage, as regular
4034 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4036 if (fs_info->nodesize >= PAGE_SIZE)
4037 folio_mark_uptodate(folio);
4039 btrfs_subpage_set_uptodate(fs_info, folio,
4040 eb->start, eb->len);
4044 static void end_bbio_meta_read(struct btrfs_bio *bbio)
4046 struct extent_buffer *eb = bbio->private;
4047 struct btrfs_fs_info *fs_info = eb->fs_info;
4048 bool uptodate = !bbio->bio.bi_status;
4049 struct folio_iter fi;
4052 eb->read_mirror = bbio->mirror_num;
4055 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
4059 set_extent_buffer_uptodate(eb);
4061 clear_extent_buffer_uptodate(eb);
4062 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4065 bio_for_each_folio_all(fi, &bbio->bio) {
4066 struct folio *folio = fi.folio;
4067 u64 start = eb->start + bio_offset;
4068 u32 len = fi.length;
4071 btrfs_folio_set_uptodate(fs_info, folio, start, len);
4073 btrfs_folio_clear_uptodate(fs_info, folio, start, len);
4078 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
4079 smp_mb__after_atomic();
4080 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
4081 free_extent_buffer(eb);
4083 bio_put(&bbio->bio);
4086 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
4087 struct btrfs_tree_parent_check *check)
4089 struct btrfs_bio *bbio;
4092 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4096 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
4097 * operation, which could potentially still be in flight. In this case
4098 * we simply want to return an error.
4100 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
4103 /* Someone else is already reading the buffer, just wait for it. */
4104 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
4107 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4108 eb->read_mirror = 0;
4109 check_buffer_tree_ref(eb);
4110 atomic_inc(&eb->refs);
4112 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
4113 REQ_OP_READ | REQ_META, eb->fs_info,
4114 end_bbio_meta_read, eb);
4115 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
4116 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
4117 bbio->file_offset = eb->start;
4118 memcpy(&bbio->parent_check, check, sizeof(*check));
4119 if (eb->fs_info->nodesize < PAGE_SIZE) {
4120 ret = bio_add_folio(&bbio->bio, eb->folios[0], eb->len,
4121 eb->start - folio_pos(eb->folios[0]));
4124 int num_folios = num_extent_folios(eb);
4126 for (int i = 0; i < num_folios; i++) {
4127 struct folio *folio = eb->folios[i];
4129 ret = bio_add_folio(&bbio->bio, folio, folio_size(folio), 0);
4133 btrfs_submit_bio(bbio, mirror_num);
4136 if (wait == WAIT_COMPLETE) {
4137 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
4138 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4145 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
4148 btrfs_warn(eb->fs_info,
4149 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
4150 eb->start, eb->len, start, len);
4151 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4157 * Check if the [start, start + len) range is valid before reading/writing
4159 * NOTE: @start and @len are offset inside the eb, not logical address.
4161 * Caller should not touch the dst/src memory if this function returns error.
4163 static inline int check_eb_range(const struct extent_buffer *eb,
4164 unsigned long start, unsigned long len)
4166 unsigned long offset;
4168 /* start, start + len should not go beyond eb->len nor overflow */
4169 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
4170 return report_eb_range(eb, start, len);
4175 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4176 unsigned long start, unsigned long len)
4178 const int unit_size = folio_size(eb->folios[0]);
4181 char *dst = (char *)dstv;
4182 unsigned long i = get_eb_folio_index(eb, start);
4184 if (check_eb_range(eb, start, len)) {
4186 * Invalid range hit, reset the memory, so callers won't get
4187 * some random garbage for their uninitialized memory.
4189 memset(dstv, 0, len);
4194 memcpy(dstv, eb->addr + start, len);
4198 offset = get_eb_offset_in_folio(eb, start);
4203 cur = min(len, unit_size - offset);
4204 kaddr = folio_address(eb->folios[i]);
4205 memcpy(dst, kaddr + offset, cur);
4214 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4216 unsigned long start, unsigned long len)
4218 const int unit_size = folio_size(eb->folios[0]);
4221 char __user *dst = (char __user *)dstv;
4222 unsigned long i = get_eb_folio_index(eb, start);
4225 WARN_ON(start > eb->len);
4226 WARN_ON(start + len > eb->start + eb->len);
4229 if (copy_to_user_nofault(dstv, eb->addr + start, len))
4234 offset = get_eb_offset_in_folio(eb, start);
4239 cur = min(len, unit_size - offset);
4240 kaddr = folio_address(eb->folios[i]);
4241 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4255 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4256 unsigned long start, unsigned long len)
4258 const int unit_size = folio_size(eb->folios[0]);
4262 char *ptr = (char *)ptrv;
4263 unsigned long i = get_eb_folio_index(eb, start);
4266 if (check_eb_range(eb, start, len))
4270 return memcmp(ptrv, eb->addr + start, len);
4272 offset = get_eb_offset_in_folio(eb, start);
4275 cur = min(len, unit_size - offset);
4276 kaddr = folio_address(eb->folios[i]);
4277 ret = memcmp(ptr, kaddr + offset, cur);
4290 * Check that the extent buffer is uptodate.
4292 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4293 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4295 static void assert_eb_folio_uptodate(const struct extent_buffer *eb, int i)
4297 struct btrfs_fs_info *fs_info = eb->fs_info;
4298 struct folio *folio = eb->folios[i];
4303 * If we are using the commit root we could potentially clear a page
4304 * Uptodate while we're using the extent buffer that we've previously
4305 * looked up. We don't want to complain in this case, as the page was
4306 * valid before, we just didn't write it out. Instead we want to catch
4307 * the case where we didn't actually read the block properly, which
4308 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4310 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4313 if (fs_info->nodesize < PAGE_SIZE) {
4314 struct folio *folio = eb->folios[0];
4317 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, folio,
4318 eb->start, eb->len)))
4319 btrfs_subpage_dump_bitmap(fs_info, folio, eb->start, eb->len);
4321 WARN_ON(!folio_test_uptodate(folio));
4325 static void __write_extent_buffer(const struct extent_buffer *eb,
4326 const void *srcv, unsigned long start,
4327 unsigned long len, bool use_memmove)
4329 const int unit_size = folio_size(eb->folios[0]);
4333 char *src = (char *)srcv;
4334 unsigned long i = get_eb_folio_index(eb, start);
4335 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4336 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4338 if (check_eb_range(eb, start, len))
4343 memmove(eb->addr + start, srcv, len);
4345 memcpy(eb->addr + start, srcv, len);
4349 offset = get_eb_offset_in_folio(eb, start);
4353 assert_eb_folio_uptodate(eb, i);
4355 cur = min(len, unit_size - offset);
4356 kaddr = folio_address(eb->folios[i]);
4358 memmove(kaddr + offset, src, cur);
4360 memcpy(kaddr + offset, src, cur);
4369 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4370 unsigned long start, unsigned long len)
4372 return __write_extent_buffer(eb, srcv, start, len, false);
4375 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4376 unsigned long start, unsigned long len)
4378 const int unit_size = folio_size(eb->folios[0]);
4379 unsigned long cur = start;
4382 memset(eb->addr + start, c, len);
4386 while (cur < start + len) {
4387 unsigned long index = get_eb_folio_index(eb, cur);
4388 unsigned int offset = get_eb_offset_in_folio(eb, cur);
4389 unsigned int cur_len = min(start + len - cur, unit_size - offset);
4391 assert_eb_folio_uptodate(eb, index);
4392 memset(folio_address(eb->folios[index]) + offset, c, cur_len);
4398 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4401 if (check_eb_range(eb, start, len))
4403 return memset_extent_buffer(eb, 0, start, len);
4406 void copy_extent_buffer_full(const struct extent_buffer *dst,
4407 const struct extent_buffer *src)
4409 const int unit_size = folio_size(src->folios[0]);
4410 unsigned long cur = 0;
4412 ASSERT(dst->len == src->len);
4414 while (cur < src->len) {
4415 unsigned long index = get_eb_folio_index(src, cur);
4416 unsigned long offset = get_eb_offset_in_folio(src, cur);
4417 unsigned long cur_len = min(src->len, unit_size - offset);
4418 void *addr = folio_address(src->folios[index]) + offset;
4420 write_extent_buffer(dst, addr, cur, cur_len);
4426 void copy_extent_buffer(const struct extent_buffer *dst,
4427 const struct extent_buffer *src,
4428 unsigned long dst_offset, unsigned long src_offset,
4431 const int unit_size = folio_size(dst->folios[0]);
4432 u64 dst_len = dst->len;
4436 unsigned long i = get_eb_folio_index(dst, dst_offset);
4438 if (check_eb_range(dst, dst_offset, len) ||
4439 check_eb_range(src, src_offset, len))
4442 WARN_ON(src->len != dst_len);
4444 offset = get_eb_offset_in_folio(dst, dst_offset);
4447 assert_eb_folio_uptodate(dst, i);
4449 cur = min(len, (unsigned long)(unit_size - offset));
4451 kaddr = folio_address(dst->folios[i]);
4452 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4462 * Calculate the page and offset of the byte containing the given bit number.
4464 * @eb: the extent buffer
4465 * @start: offset of the bitmap item in the extent buffer
4467 * @page_index: return index of the page in the extent buffer that contains
4468 * the given bit number
4469 * @page_offset: return offset into the page given by page_index
4471 * This helper hides the ugliness of finding the byte in an extent buffer which
4472 * contains a given bit.
4474 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4475 unsigned long start, unsigned long nr,
4476 unsigned long *page_index,
4477 size_t *page_offset)
4479 size_t byte_offset = BIT_BYTE(nr);
4483 * The byte we want is the offset of the extent buffer + the offset of
4484 * the bitmap item in the extent buffer + the offset of the byte in the
4487 offset = start + offset_in_page(eb->start) + byte_offset;
4489 *page_index = offset >> PAGE_SHIFT;
4490 *page_offset = offset_in_page(offset);
4494 * Determine whether a bit in a bitmap item is set.
4496 * @eb: the extent buffer
4497 * @start: offset of the bitmap item in the extent buffer
4498 * @nr: bit number to test
4500 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4508 eb_bitmap_offset(eb, start, nr, &i, &offset);
4509 page = folio_page(eb->folios[i], 0);
4510 assert_eb_folio_uptodate(eb, i);
4511 kaddr = page_address(page);
4512 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4515 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4517 unsigned long index = get_eb_folio_index(eb, bytenr);
4519 if (check_eb_range(eb, bytenr, 1))
4521 return folio_address(eb->folios[index]) + get_eb_offset_in_folio(eb, bytenr);
4525 * Set an area of a bitmap to 1.
4527 * @eb: the extent buffer
4528 * @start: offset of the bitmap item in the extent buffer
4529 * @pos: bit number of the first bit
4530 * @len: number of bits to set
4532 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4533 unsigned long pos, unsigned long len)
4535 unsigned int first_byte = start + BIT_BYTE(pos);
4536 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4537 const bool same_byte = (first_byte == last_byte);
4538 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4542 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4544 /* Handle the first byte. */
4545 kaddr = extent_buffer_get_byte(eb, first_byte);
4550 /* Handle the byte aligned part. */
4551 ASSERT(first_byte + 1 <= last_byte);
4552 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4554 /* Handle the last byte. */
4555 kaddr = extent_buffer_get_byte(eb, last_byte);
4556 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4561 * Clear an area of a bitmap.
4563 * @eb: the extent buffer
4564 * @start: offset of the bitmap item in the extent buffer
4565 * @pos: bit number of the first bit
4566 * @len: number of bits to clear
4568 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4569 unsigned long start, unsigned long pos,
4572 unsigned int first_byte = start + BIT_BYTE(pos);
4573 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4574 const bool same_byte = (first_byte == last_byte);
4575 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4579 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4581 /* Handle the first byte. */
4582 kaddr = extent_buffer_get_byte(eb, first_byte);
4587 /* Handle the byte aligned part. */
4588 ASSERT(first_byte + 1 <= last_byte);
4589 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4591 /* Handle the last byte. */
4592 kaddr = extent_buffer_get_byte(eb, last_byte);
4593 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4596 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4598 unsigned long distance = (src > dst) ? src - dst : dst - src;
4599 return distance < len;
4602 void memcpy_extent_buffer(const struct extent_buffer *dst,
4603 unsigned long dst_offset, unsigned long src_offset,
4606 const int unit_size = folio_size(dst->folios[0]);
4607 unsigned long cur_off = 0;
4609 if (check_eb_range(dst, dst_offset, len) ||
4610 check_eb_range(dst, src_offset, len))
4614 const bool use_memmove = areas_overlap(src_offset, dst_offset, len);
4617 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4619 memcpy(dst->addr + dst_offset, dst->addr + src_offset, len);
4623 while (cur_off < len) {
4624 unsigned long cur_src = cur_off + src_offset;
4625 unsigned long folio_index = get_eb_folio_index(dst, cur_src);
4626 unsigned long folio_off = get_eb_offset_in_folio(dst, cur_src);
4627 unsigned long cur_len = min(src_offset + len - cur_src,
4628 unit_size - folio_off);
4629 void *src_addr = folio_address(dst->folios[folio_index]) + folio_off;
4630 const bool use_memmove = areas_overlap(src_offset + cur_off,
4631 dst_offset + cur_off, cur_len);
4633 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4639 void memmove_extent_buffer(const struct extent_buffer *dst,
4640 unsigned long dst_offset, unsigned long src_offset,
4643 unsigned long dst_end = dst_offset + len - 1;
4644 unsigned long src_end = src_offset + len - 1;
4646 if (check_eb_range(dst, dst_offset, len) ||
4647 check_eb_range(dst, src_offset, len))
4650 if (dst_offset < src_offset) {
4651 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4656 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4661 unsigned long src_i;
4663 size_t dst_off_in_folio;
4664 size_t src_off_in_folio;
4668 src_i = get_eb_folio_index(dst, src_end);
4670 dst_off_in_folio = get_eb_offset_in_folio(dst, dst_end);
4671 src_off_in_folio = get_eb_offset_in_folio(dst, src_end);
4673 cur = min_t(unsigned long, len, src_off_in_folio + 1);
4674 cur = min(cur, dst_off_in_folio + 1);
4676 src_addr = folio_address(dst->folios[src_i]) + src_off_in_folio -
4678 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4681 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4690 #define GANG_LOOKUP_SIZE 16
4691 static struct extent_buffer *get_next_extent_buffer(
4692 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4694 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4695 struct extent_buffer *found = NULL;
4696 u64 page_start = page_offset(page);
4697 u64 cur = page_start;
4699 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4700 lockdep_assert_held(&fs_info->buffer_lock);
4702 while (cur < page_start + PAGE_SIZE) {
4706 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4707 (void **)gang, cur >> fs_info->sectorsize_bits,
4708 min_t(unsigned int, GANG_LOOKUP_SIZE,
4709 PAGE_SIZE / fs_info->nodesize));
4712 for (i = 0; i < ret; i++) {
4713 /* Already beyond page end */
4714 if (gang[i]->start >= page_start + PAGE_SIZE)
4717 if (gang[i]->start >= bytenr) {
4722 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4728 static int try_release_subpage_extent_buffer(struct page *page)
4730 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4731 u64 cur = page_offset(page);
4732 const u64 end = page_offset(page) + PAGE_SIZE;
4736 struct extent_buffer *eb = NULL;
4739 * Unlike try_release_extent_buffer() which uses folio private
4740 * to grab buffer, for subpage case we rely on radix tree, thus
4741 * we need to ensure radix tree consistency.
4743 * We also want an atomic snapshot of the radix tree, thus go
4744 * with spinlock rather than RCU.
4746 spin_lock(&fs_info->buffer_lock);
4747 eb = get_next_extent_buffer(fs_info, page, cur);
4749 /* No more eb in the page range after or at cur */
4750 spin_unlock(&fs_info->buffer_lock);
4753 cur = eb->start + eb->len;
4756 * The same as try_release_extent_buffer(), to ensure the eb
4757 * won't disappear out from under us.
4759 spin_lock(&eb->refs_lock);
4760 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4761 spin_unlock(&eb->refs_lock);
4762 spin_unlock(&fs_info->buffer_lock);
4765 spin_unlock(&fs_info->buffer_lock);
4768 * If tree ref isn't set then we know the ref on this eb is a
4769 * real ref, so just return, this eb will likely be freed soon
4772 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4773 spin_unlock(&eb->refs_lock);
4778 * Here we don't care about the return value, we will always
4779 * check the folio private at the end. And
4780 * release_extent_buffer() will release the refs_lock.
4782 release_extent_buffer(eb);
4785 * Finally to check if we have cleared folio private, as if we have
4786 * released all ebs in the page, the folio private should be cleared now.
4788 spin_lock(&page->mapping->private_lock);
4789 if (!folio_test_private(page_folio(page)))
4793 spin_unlock(&page->mapping->private_lock);
4798 int try_release_extent_buffer(struct page *page)
4800 struct folio *folio = page_folio(page);
4801 struct extent_buffer *eb;
4803 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4804 return try_release_subpage_extent_buffer(page);
4807 * We need to make sure nobody is changing folio private, as we rely on
4808 * folio private as the pointer to extent buffer.
4810 spin_lock(&page->mapping->private_lock);
4811 if (!folio_test_private(folio)) {
4812 spin_unlock(&page->mapping->private_lock);
4816 eb = folio_get_private(folio);
4820 * This is a little awful but should be ok, we need to make sure that
4821 * the eb doesn't disappear out from under us while we're looking at
4824 spin_lock(&eb->refs_lock);
4825 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4826 spin_unlock(&eb->refs_lock);
4827 spin_unlock(&page->mapping->private_lock);
4830 spin_unlock(&page->mapping->private_lock);
4833 * If tree ref isn't set then we know the ref on this eb is a real ref,
4834 * so just return, this page will likely be freed soon anyway.
4836 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4837 spin_unlock(&eb->refs_lock);
4841 return release_extent_buffer(eb);
4845 * Attempt to readahead a child block.
4847 * @fs_info: the fs_info
4848 * @bytenr: bytenr to read
4849 * @owner_root: objectid of the root that owns this eb
4850 * @gen: generation for the uptodate check, can be 0
4851 * @level: level for the eb
4853 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
4854 * normal uptodate check of the eb, without checking the generation. If we have
4855 * to read the block we will not block on anything.
4857 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
4858 u64 bytenr, u64 owner_root, u64 gen, int level)
4860 struct btrfs_tree_parent_check check = {
4865 struct extent_buffer *eb;
4868 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
4872 if (btrfs_buffer_uptodate(eb, gen, 1)) {
4873 free_extent_buffer(eb);
4877 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
4879 free_extent_buffer_stale(eb);
4881 free_extent_buffer(eb);
4885 * Readahead a node's child block.
4887 * @node: parent node we're reading from
4888 * @slot: slot in the parent node for the child we want to read
4890 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
4891 * the slot in the node provided.
4893 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
4895 btrfs_readahead_tree_block(node->fs_info,
4896 btrfs_node_blockptr(node, slot),
4897 btrfs_header_owner(node),
4898 btrfs_node_ptr_generation(node, slot),
4899 btrfs_header_level(node) - 1);