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)
189 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
190 len = end + 1 - start;
192 if (page_ops & PAGE_SET_ORDERED)
193 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
194 if (page_ops & PAGE_START_WRITEBACK) {
195 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
196 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
198 if (page_ops & PAGE_END_WRITEBACK)
199 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
201 if (page != locked_page && (page_ops & PAGE_UNLOCK))
202 btrfs_page_end_writer_lock(fs_info, page, start, len);
205 static void __process_pages_contig(struct address_space *mapping,
206 struct page *locked_page, u64 start, u64 end,
207 unsigned long page_ops)
209 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
210 pgoff_t start_index = start >> PAGE_SHIFT;
211 pgoff_t end_index = end >> PAGE_SHIFT;
212 pgoff_t index = start_index;
213 struct folio_batch fbatch;
216 folio_batch_init(&fbatch);
217 while (index <= end_index) {
220 found_folios = filemap_get_folios_contig(mapping, &index,
222 for (i = 0; i < found_folios; i++) {
223 struct folio *folio = fbatch.folios[i];
225 process_one_page(fs_info, &folio->page, locked_page,
226 page_ops, start, end);
228 folio_batch_release(&fbatch);
233 static noinline void __unlock_for_delalloc(struct inode *inode,
234 struct page *locked_page,
237 unsigned long index = start >> PAGE_SHIFT;
238 unsigned long end_index = end >> PAGE_SHIFT;
241 if (index == locked_page->index && end_index == index)
244 __process_pages_contig(inode->i_mapping, locked_page, start, end,
248 static noinline int lock_delalloc_pages(struct inode *inode,
249 struct page *locked_page,
253 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
254 struct address_space *mapping = inode->i_mapping;
255 pgoff_t start_index = start >> PAGE_SHIFT;
256 pgoff_t end_index = end >> PAGE_SHIFT;
257 pgoff_t index = start_index;
258 u64 processed_end = start;
259 struct folio_batch fbatch;
261 if (index == locked_page->index && index == end_index)
264 folio_batch_init(&fbatch);
265 while (index <= end_index) {
266 unsigned int found_folios, i;
268 found_folios = filemap_get_folios_contig(mapping, &index,
270 if (found_folios == 0)
273 for (i = 0; i < found_folios; i++) {
274 struct page *page = &fbatch.folios[i]->page;
275 u32 len = end + 1 - start;
277 if (page == locked_page)
280 if (btrfs_page_start_writer_lock(fs_info, page, start,
284 if (!PageDirty(page) || page->mapping != mapping) {
285 btrfs_page_end_writer_lock(fs_info, page, start,
290 processed_end = page_offset(page) + PAGE_SIZE - 1;
292 folio_batch_release(&fbatch);
298 folio_batch_release(&fbatch);
299 if (processed_end > start)
300 __unlock_for_delalloc(inode, locked_page, start, processed_end);
305 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
306 * more than @max_bytes.
308 * @start: The original start bytenr to search.
309 * Will store the extent range start bytenr.
310 * @end: The original end bytenr of the search range
311 * Will store the extent range end bytenr.
313 * Return true if we find a delalloc range which starts inside the original
314 * range, and @start/@end will store the delalloc range start/end.
316 * Return false if we can't find any delalloc range which starts inside the
317 * original range, and @start/@end will be the non-delalloc range start/end.
320 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
321 struct page *locked_page, u64 *start,
324 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
325 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
326 const u64 orig_start = *start;
327 const u64 orig_end = *end;
328 /* The sanity tests may not set a valid fs_info. */
329 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
333 struct extent_state *cached_state = NULL;
337 /* Caller should pass a valid @end to indicate the search range end */
338 ASSERT(orig_end > orig_start);
340 /* The range should at least cover part of the page */
341 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
342 orig_end <= page_offset(locked_page)));
344 /* step one, find a bunch of delalloc bytes starting at start */
345 delalloc_start = *start;
347 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
348 max_bytes, &cached_state);
349 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
350 *start = delalloc_start;
352 /* @delalloc_end can be -1, never go beyond @orig_end */
353 *end = min(delalloc_end, orig_end);
354 free_extent_state(cached_state);
359 * start comes from the offset of locked_page. We have to lock
360 * pages in order, so we can't process delalloc bytes before
363 if (delalloc_start < *start)
364 delalloc_start = *start;
367 * make sure to limit the number of pages we try to lock down
369 if (delalloc_end + 1 - delalloc_start > max_bytes)
370 delalloc_end = delalloc_start + max_bytes - 1;
372 /* step two, lock all the pages after the page that has start */
373 ret = lock_delalloc_pages(inode, locked_page,
374 delalloc_start, delalloc_end);
375 ASSERT(!ret || ret == -EAGAIN);
376 if (ret == -EAGAIN) {
377 /* some of the pages are gone, lets avoid looping by
378 * shortening the size of the delalloc range we're searching
380 free_extent_state(cached_state);
383 max_bytes = PAGE_SIZE;
392 /* step three, lock the state bits for the whole range */
393 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
395 /* then test to make sure it is all still delalloc */
396 ret = test_range_bit(tree, delalloc_start, delalloc_end,
397 EXTENT_DELALLOC, cached_state);
399 unlock_extent(tree, delalloc_start, delalloc_end,
401 __unlock_for_delalloc(inode, locked_page,
402 delalloc_start, delalloc_end);
406 free_extent_state(cached_state);
407 *start = delalloc_start;
413 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
414 struct page *locked_page,
415 u32 clear_bits, unsigned long page_ops)
417 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
419 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
420 start, end, page_ops);
423 static bool btrfs_verify_page(struct page *page, u64 start)
425 if (!fsverity_active(page->mapping->host) ||
426 PageUptodate(page) ||
427 start >= i_size_read(page->mapping->host))
429 return fsverity_verify_page(page);
432 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
434 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
436 ASSERT(page_offset(page) <= start &&
437 start + len <= page_offset(page) + PAGE_SIZE);
439 if (uptodate && btrfs_verify_page(page, start))
440 btrfs_page_set_uptodate(fs_info, page, start, len);
442 btrfs_page_clear_uptodate(fs_info, page, start, len);
444 if (!btrfs_is_subpage(fs_info, page->mapping))
447 btrfs_subpage_end_reader(fs_info, page, start, len);
451 * after a writepage IO is done, we need to:
452 * clear the uptodate bits on error
453 * clear the writeback bits in the extent tree for this IO
454 * end_page_writeback if the page has no more pending IO
456 * Scheduling is not allowed, so the extent state tree is expected
457 * to have one and only one object corresponding to this IO.
459 static void end_bio_extent_writepage(struct btrfs_bio *bbio)
461 struct bio *bio = &bbio->bio;
462 int error = blk_status_to_errno(bio->bi_status);
463 struct bio_vec *bvec;
464 struct bvec_iter_all iter_all;
466 ASSERT(!bio_flagged(bio, BIO_CLONED));
467 bio_for_each_segment_all(bvec, bio, iter_all) {
468 struct page *page = bvec->bv_page;
469 struct inode *inode = page->mapping->host;
470 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
471 const u32 sectorsize = fs_info->sectorsize;
472 u64 start = page_offset(page) + bvec->bv_offset;
473 u32 len = bvec->bv_len;
475 /* Our read/write should always be sector aligned. */
476 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
478 "partial page write in btrfs with offset %u and length %u",
479 bvec->bv_offset, bvec->bv_len);
480 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
482 "incomplete page write with offset %u and length %u",
483 bvec->bv_offset, bvec->bv_len);
485 btrfs_finish_ordered_extent(bbio->ordered, page, start, len, !error);
487 mapping_set_error(page->mapping, error);
488 btrfs_page_clear_writeback(fs_info, page, start, len);
495 * Record previously processed extent range
497 * For endio_readpage_release_extent() to handle a full extent range, reducing
498 * the extent io operations.
500 struct processed_extent {
501 struct btrfs_inode *inode;
502 /* Start of the range in @inode */
504 /* End of the range in @inode */
510 * Try to release processed extent range
512 * May not release the extent range right now if the current range is
513 * contiguous to processed extent.
515 * Will release processed extent when any of @inode, @uptodate, the range is
516 * no longer contiguous to the processed range.
518 * Passing @inode == NULL will force processed extent to be released.
520 static void endio_readpage_release_extent(struct processed_extent *processed,
521 struct btrfs_inode *inode, u64 start, u64 end,
524 struct extent_state *cached = NULL;
525 struct extent_io_tree *tree;
527 /* The first extent, initialize @processed */
528 if (!processed->inode)
532 * Contiguous to processed extent, just uptodate the end.
534 * Several things to notice:
536 * - bio can be merged as long as on-disk bytenr is contiguous
537 * This means we can have page belonging to other inodes, thus need to
538 * check if the inode still matches.
539 * - bvec can contain range beyond current page for multi-page bvec
540 * Thus we need to do processed->end + 1 >= start check
542 if (processed->inode == inode && processed->uptodate == uptodate &&
543 processed->end + 1 >= start && end >= processed->end) {
544 processed->end = end;
548 tree = &processed->inode->io_tree;
550 * Now we don't have range contiguous to the processed range, release
551 * the processed range now.
553 unlock_extent(tree, processed->start, processed->end, &cached);
556 /* Update processed to current range */
557 processed->inode = inode;
558 processed->start = start;
559 processed->end = end;
560 processed->uptodate = uptodate;
563 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
565 struct folio *folio = page_folio(page);
567 ASSERT(PageLocked(page));
568 if (!btrfs_is_subpage(fs_info, page->mapping))
571 ASSERT(folio_test_private(folio));
572 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
576 * after a readpage IO is done, we need to:
577 * clear the uptodate bits on error
578 * set the uptodate bits if things worked
579 * set the page up to date if all extents in the tree are uptodate
580 * clear the lock bit in the extent tree
581 * unlock the page if there are no other extents locked for it
583 * Scheduling is not allowed, so the extent state tree is expected
584 * to have one and only one object corresponding to this IO.
586 static void end_bio_extent_readpage(struct btrfs_bio *bbio)
588 struct bio *bio = &bbio->bio;
589 struct bio_vec *bvec;
590 struct processed_extent processed = { 0 };
592 * The offset to the beginning of a bio, since one bio can never be
593 * larger than UINT_MAX, u32 here is enough.
596 struct bvec_iter_all iter_all;
598 ASSERT(!bio_flagged(bio, BIO_CLONED));
599 bio_for_each_segment_all(bvec, bio, iter_all) {
600 bool uptodate = !bio->bi_status;
601 struct page *page = bvec->bv_page;
602 struct inode *inode = page->mapping->host;
603 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
604 const u32 sectorsize = fs_info->sectorsize;
610 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
611 bio->bi_iter.bi_sector, bio->bi_status,
615 * We always issue full-sector reads, but if some block in a
616 * page fails to read, blk_update_request() will advance
617 * bv_offset and adjust bv_len to compensate. Print a warning
618 * for unaligned offsets, and an error if they don't add up to
621 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
623 "partial page read in btrfs with offset %u and length %u",
624 bvec->bv_offset, bvec->bv_len);
625 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
628 "incomplete page read with offset %u and length %u",
629 bvec->bv_offset, bvec->bv_len);
631 start = page_offset(page) + bvec->bv_offset;
632 end = start + bvec->bv_len - 1;
635 if (likely(uptodate)) {
636 loff_t i_size = i_size_read(inode);
637 pgoff_t end_index = i_size >> PAGE_SHIFT;
640 * Zero out the remaining part if this range straddles
643 * Here we should only zero the range inside the bvec,
644 * not touch anything else.
646 * NOTE: i_size is exclusive while end is inclusive.
648 if (page->index == end_index && i_size <= end) {
649 u32 zero_start = max(offset_in_page(i_size),
650 offset_in_page(start));
652 zero_user_segment(page, zero_start,
653 offset_in_page(end) + 1);
657 /* Update page status and unlock. */
658 end_page_read(page, uptodate, start, len);
659 endio_readpage_release_extent(&processed, BTRFS_I(inode),
660 start, end, uptodate);
662 ASSERT(bio_offset + len > bio_offset);
666 /* Release the last extent */
667 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
672 * Populate every free slot in a provided array with pages.
674 * @nr_pages: number of pages to allocate
675 * @page_array: the array to fill with pages; any existing non-null entries in
676 * the array will be skipped
677 * @extra_gfp: the extra GFP flags for the allocation.
679 * Return: 0 if all pages were able to be allocated;
680 * -ENOMEM otherwise, the partially allocated pages would be freed and
681 * the array slots zeroed
683 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array,
686 unsigned int allocated;
688 for (allocated = 0; allocated < nr_pages;) {
689 unsigned int last = allocated;
691 allocated = alloc_pages_bulk_array(GFP_NOFS | extra_gfp,
692 nr_pages, page_array);
694 if (allocated == nr_pages)
698 * During this iteration, no page could be allocated, even
699 * though alloc_pages_bulk_array() falls back to alloc_page()
700 * if it could not bulk-allocate. So we must be out of memory.
702 if (allocated == last) {
703 for (int i = 0; i < allocated; i++) {
704 __free_page(page_array[i]);
705 page_array[i] = NULL;
710 memalloc_retry_wait(GFP_NOFS);
716 * Populate needed folios for the extent buffer.
718 * For now, the folios populated are always in order 0 (aka, single page).
720 static int alloc_eb_folio_array(struct extent_buffer *eb, gfp_t extra_gfp)
722 struct page *page_array[INLINE_EXTENT_BUFFER_PAGES] = { 0 };
723 int num_pages = num_extent_pages(eb);
726 ret = btrfs_alloc_page_array(num_pages, page_array, extra_gfp);
730 for (int i = 0; i < num_pages; i++)
731 eb->folios[i] = page_folio(page_array[i]);
735 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
736 struct page *page, u64 disk_bytenr,
737 unsigned int pg_offset)
739 struct bio *bio = &bio_ctrl->bbio->bio;
740 struct bio_vec *bvec = bio_last_bvec_all(bio);
741 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
743 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
745 * For compression, all IO should have its logical bytenr set
746 * to the starting bytenr of the compressed extent.
748 return bio->bi_iter.bi_sector == sector;
752 * The contig check requires the following conditions to be met:
754 * 1) The pages are belonging to the same inode
755 * This is implied by the call chain.
757 * 2) The range has adjacent logical bytenr
759 * 3) The range has adjacent file offset
760 * This is required for the usage of btrfs_bio->file_offset.
762 return bio_end_sector(bio) == sector &&
763 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
764 page_offset(page) + pg_offset;
767 static void alloc_new_bio(struct btrfs_inode *inode,
768 struct btrfs_bio_ctrl *bio_ctrl,
769 u64 disk_bytenr, u64 file_offset)
771 struct btrfs_fs_info *fs_info = inode->root->fs_info;
772 struct btrfs_bio *bbio;
774 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
775 bio_ctrl->end_io_func, NULL);
776 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
778 bbio->file_offset = file_offset;
779 bio_ctrl->bbio = bbio;
780 bio_ctrl->len_to_oe_boundary = U32_MAX;
782 /* Limit data write bios to the ordered boundary. */
784 struct btrfs_ordered_extent *ordered;
786 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
788 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
789 ordered->file_offset +
790 ordered->disk_num_bytes - file_offset);
791 bbio->ordered = ordered;
795 * Pick the last added device to support cgroup writeback. For
796 * multi-device file systems this means blk-cgroup policies have
797 * to always be set on the last added/replaced device.
798 * This is a bit odd but has been like that for a long time.
800 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
801 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
806 * @disk_bytenr: logical bytenr where the write will be
807 * @page: page to add to the bio
808 * @size: portion of page that we want to write to
809 * @pg_offset: offset of the new bio or to check whether we are adding
810 * a contiguous page to the previous one
812 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
813 * new one in @bio_ctrl->bbio.
814 * The mirror number for this IO should already be initizlied in
815 * @bio_ctrl->mirror_num.
817 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
818 u64 disk_bytenr, struct page *page,
819 size_t size, unsigned long pg_offset)
821 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
823 ASSERT(pg_offset + size <= PAGE_SIZE);
824 ASSERT(bio_ctrl->end_io_func);
826 if (bio_ctrl->bbio &&
827 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
828 submit_one_bio(bio_ctrl);
833 /* Allocate new bio if needed */
834 if (!bio_ctrl->bbio) {
835 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
836 page_offset(page) + pg_offset);
839 /* Cap to the current ordered extent boundary if there is one. */
840 if (len > bio_ctrl->len_to_oe_boundary) {
841 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
842 ASSERT(is_data_inode(&inode->vfs_inode));
843 len = bio_ctrl->len_to_oe_boundary;
846 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
847 /* bio full: move on to a new one */
848 submit_one_bio(bio_ctrl);
853 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
860 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
861 * sector aligned. alloc_new_bio() then sets it to the end of
862 * our ordered extent for writes into zoned devices.
864 * When len_to_oe_boundary is tracking an ordered extent, we
865 * trust the ordered extent code to align things properly, and
866 * the check above to cap our write to the ordered extent
867 * boundary is correct.
869 * When len_to_oe_boundary is U32_MAX, the cap above would
870 * result in a 4095 byte IO for the last page right before
871 * we hit the bio limit of UINT_MAX. bio_add_page() has all
872 * the checks required to make sure we don't overflow the bio,
873 * and we should just ignore len_to_oe_boundary completely
874 * unless we're using it to track an ordered extent.
876 * It's pretty hard to make a bio sized U32_MAX, but it can
877 * happen when the page cache is able to feed us contiguous
878 * pages for large extents.
880 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
881 bio_ctrl->len_to_oe_boundary -= len;
883 /* Ordered extent boundary: move on to a new bio. */
884 if (bio_ctrl->len_to_oe_boundary == 0)
885 submit_one_bio(bio_ctrl);
889 static int attach_extent_buffer_folio(struct extent_buffer *eb,
891 struct btrfs_subpage *prealloc)
893 struct btrfs_fs_info *fs_info = eb->fs_info;
897 * If the page is mapped to btree inode, we should hold the private
898 * lock to prevent race.
899 * For cloned or dummy extent buffers, their pages are not mapped and
900 * will not race with any other ebs.
903 lockdep_assert_held(&folio->mapping->private_lock);
905 if (fs_info->nodesize >= PAGE_SIZE) {
906 if (!folio_test_private(folio))
907 folio_attach_private(folio, eb);
909 WARN_ON(folio_get_private(folio) != eb);
913 /* Already mapped, just free prealloc */
914 if (folio_test_private(folio)) {
915 btrfs_free_subpage(prealloc);
920 /* Has preallocated memory for subpage */
921 folio_attach_private(folio, prealloc);
923 /* Do new allocation to attach subpage */
924 ret = btrfs_attach_subpage(fs_info, folio_page(folio, 0),
925 BTRFS_SUBPAGE_METADATA);
929 int set_page_extent_mapped(struct page *page)
931 struct folio *folio = page_folio(page);
932 struct btrfs_fs_info *fs_info;
934 ASSERT(page->mapping);
936 if (folio_test_private(folio))
939 fs_info = btrfs_sb(page->mapping->host->i_sb);
941 if (btrfs_is_subpage(fs_info, page->mapping))
942 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
944 folio_attach_private(folio, (void *)EXTENT_FOLIO_PRIVATE);
948 void clear_page_extent_mapped(struct page *page)
950 struct folio *folio = page_folio(page);
951 struct btrfs_fs_info *fs_info;
953 ASSERT(page->mapping);
955 if (!folio_test_private(folio))
958 fs_info = btrfs_sb(page->mapping->host->i_sb);
959 if (btrfs_is_subpage(fs_info, page->mapping))
960 return btrfs_detach_subpage(fs_info, page);
962 folio_detach_private(folio);
965 static struct extent_map *
966 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
967 u64 start, u64 len, struct extent_map **em_cached)
969 struct extent_map *em;
971 if (em_cached && *em_cached) {
973 if (extent_map_in_tree(em) && start >= em->start &&
974 start < extent_map_end(em)) {
975 refcount_inc(&em->refs);
983 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
984 if (em_cached && !IS_ERR(em)) {
986 refcount_inc(&em->refs);
992 * basic readpage implementation. Locked extent state structs are inserted
993 * into the tree that are removed when the IO is done (by the end_io
995 * XXX JDM: This needs looking at to ensure proper page locking
996 * return 0 on success, otherwise return error
998 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
999 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
1001 struct inode *inode = page->mapping->host;
1002 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1003 u64 start = page_offset(page);
1004 const u64 end = start + PAGE_SIZE - 1;
1007 u64 last_byte = i_size_read(inode);
1009 struct extent_map *em;
1011 size_t pg_offset = 0;
1013 size_t blocksize = inode->i_sb->s_blocksize;
1014 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1016 ret = set_page_extent_mapped(page);
1018 unlock_extent(tree, start, end, NULL);
1023 if (page->index == last_byte >> PAGE_SHIFT) {
1024 size_t zero_offset = offset_in_page(last_byte);
1027 iosize = PAGE_SIZE - zero_offset;
1028 memzero_page(page, zero_offset, iosize);
1031 bio_ctrl->end_io_func = end_bio_extent_readpage;
1032 begin_page_read(fs_info, page);
1033 while (cur <= end) {
1034 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1035 bool force_bio_submit = false;
1038 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1039 if (cur >= last_byte) {
1040 iosize = PAGE_SIZE - pg_offset;
1041 memzero_page(page, pg_offset, iosize);
1042 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1043 end_page_read(page, true, cur, iosize);
1046 em = __get_extent_map(inode, page, pg_offset, cur,
1047 end - cur + 1, em_cached);
1049 unlock_extent(tree, cur, end, NULL);
1050 end_page_read(page, false, cur, end + 1 - cur);
1053 extent_offset = cur - em->start;
1054 BUG_ON(extent_map_end(em) <= cur);
1057 compress_type = extent_map_compression(em);
1059 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1060 iosize = ALIGN(iosize, blocksize);
1061 if (compress_type != BTRFS_COMPRESS_NONE)
1062 disk_bytenr = em->block_start;
1064 disk_bytenr = em->block_start + extent_offset;
1065 block_start = em->block_start;
1066 if (em->flags & EXTENT_FLAG_PREALLOC)
1067 block_start = EXTENT_MAP_HOLE;
1070 * If we have a file range that points to a compressed extent
1071 * and it's followed by a consecutive file range that points
1072 * to the same compressed extent (possibly with a different
1073 * offset and/or length, so it either points to the whole extent
1074 * or only part of it), we must make sure we do not submit a
1075 * single bio to populate the pages for the 2 ranges because
1076 * this makes the compressed extent read zero out the pages
1077 * belonging to the 2nd range. Imagine the following scenario:
1080 * [0 - 8K] [8K - 24K]
1083 * points to extent X, points to extent X,
1084 * offset 4K, length of 8K offset 0, length 16K
1086 * [extent X, compressed length = 4K uncompressed length = 16K]
1088 * If the bio to read the compressed extent covers both ranges,
1089 * it will decompress extent X into the pages belonging to the
1090 * first range and then it will stop, zeroing out the remaining
1091 * pages that belong to the other range that points to extent X.
1092 * So here we make sure we submit 2 bios, one for the first
1093 * range and another one for the third range. Both will target
1094 * the same physical extent from disk, but we can't currently
1095 * make the compressed bio endio callback populate the pages
1096 * for both ranges because each compressed bio is tightly
1097 * coupled with a single extent map, and each range can have
1098 * an extent map with a different offset value relative to the
1099 * uncompressed data of our extent and different lengths. This
1100 * is a corner case so we prioritize correctness over
1101 * non-optimal behavior (submitting 2 bios for the same extent).
1103 if (compress_type != BTRFS_COMPRESS_NONE &&
1104 prev_em_start && *prev_em_start != (u64)-1 &&
1105 *prev_em_start != em->start)
1106 force_bio_submit = true;
1109 *prev_em_start = em->start;
1111 free_extent_map(em);
1114 /* we've found a hole, just zero and go on */
1115 if (block_start == EXTENT_MAP_HOLE) {
1116 memzero_page(page, pg_offset, iosize);
1118 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1119 end_page_read(page, true, cur, iosize);
1121 pg_offset += iosize;
1124 /* the get_extent function already copied into the page */
1125 if (block_start == EXTENT_MAP_INLINE) {
1126 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1127 end_page_read(page, true, cur, iosize);
1129 pg_offset += iosize;
1133 if (bio_ctrl->compress_type != compress_type) {
1134 submit_one_bio(bio_ctrl);
1135 bio_ctrl->compress_type = compress_type;
1138 if (force_bio_submit)
1139 submit_one_bio(bio_ctrl);
1140 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1143 pg_offset += iosize;
1149 int btrfs_read_folio(struct file *file, struct folio *folio)
1151 struct page *page = &folio->page;
1152 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1153 u64 start = page_offset(page);
1154 u64 end = start + PAGE_SIZE - 1;
1155 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1158 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1160 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL);
1162 * If btrfs_do_readpage() failed we will want to submit the assembled
1163 * bio to do the cleanup.
1165 submit_one_bio(&bio_ctrl);
1169 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1171 struct extent_map **em_cached,
1172 struct btrfs_bio_ctrl *bio_ctrl,
1175 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1178 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1180 for (index = 0; index < nr_pages; index++) {
1181 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1183 put_page(pages[index]);
1188 * helper for __extent_writepage, doing all of the delayed allocation setup.
1190 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1191 * to write the page (copy into inline extent). In this case the IO has
1192 * been started and the page is already unlocked.
1194 * This returns 0 if all went well (page still locked)
1195 * This returns < 0 if there were errors (page still locked)
1197 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1198 struct page *page, struct writeback_control *wbc)
1200 const u64 page_start = page_offset(page);
1201 const u64 page_end = page_start + PAGE_SIZE - 1;
1202 u64 delalloc_start = page_start;
1203 u64 delalloc_end = page_end;
1204 u64 delalloc_to_write = 0;
1207 while (delalloc_start < page_end) {
1208 delalloc_end = page_end;
1209 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1210 &delalloc_start, &delalloc_end)) {
1211 delalloc_start = delalloc_end + 1;
1215 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1220 delalloc_start = delalloc_end + 1;
1224 * delalloc_end is already one less than the total length, so
1225 * we don't subtract one from PAGE_SIZE
1227 delalloc_to_write +=
1228 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1231 * If btrfs_run_dealloc_range() already started I/O and unlocked
1232 * the pages, we just need to account for them here.
1235 wbc->nr_to_write -= delalloc_to_write;
1239 if (wbc->nr_to_write < delalloc_to_write) {
1242 if (delalloc_to_write < thresh * 2)
1243 thresh = delalloc_to_write;
1244 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1252 * Find the first byte we need to write.
1254 * For subpage, one page can contain several sectors, and
1255 * __extent_writepage_io() will just grab all extent maps in the page
1256 * range and try to submit all non-inline/non-compressed extents.
1258 * This is a big problem for subpage, we shouldn't re-submit already written
1260 * This function will lookup subpage dirty bit to find which range we really
1263 * Return the next dirty range in [@start, @end).
1264 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1266 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1267 struct page *page, u64 *start, u64 *end)
1269 struct folio *folio = page_folio(page);
1270 struct btrfs_subpage *subpage = folio_get_private(folio);
1271 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1272 u64 orig_start = *start;
1273 /* Declare as unsigned long so we can use bitmap ops */
1274 unsigned long flags;
1275 int range_start_bit;
1279 * For regular sector size == page size case, since one page only
1280 * contains one sector, we return the page offset directly.
1282 if (!btrfs_is_subpage(fs_info, page->mapping)) {
1283 *start = page_offset(page);
1284 *end = page_offset(page) + PAGE_SIZE;
1288 range_start_bit = spi->dirty_offset +
1289 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1291 /* We should have the page locked, but just in case */
1292 spin_lock_irqsave(&subpage->lock, flags);
1293 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1294 spi->dirty_offset + spi->bitmap_nr_bits);
1295 spin_unlock_irqrestore(&subpage->lock, flags);
1297 range_start_bit -= spi->dirty_offset;
1298 range_end_bit -= spi->dirty_offset;
1300 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1301 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1305 * helper for __extent_writepage. This calls the writepage start hooks,
1306 * and does the loop to map the page into extents and bios.
1308 * We return 1 if the IO is started and the page is unlocked,
1309 * 0 if all went well (page still locked)
1310 * < 0 if there were errors (page still locked)
1312 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1314 struct btrfs_bio_ctrl *bio_ctrl,
1318 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1319 u64 cur = page_offset(page);
1320 u64 end = cur + PAGE_SIZE - 1;
1323 struct extent_map *em;
1327 ret = btrfs_writepage_cow_fixup(page);
1329 /* Fixup worker will requeue */
1330 redirty_page_for_writepage(bio_ctrl->wbc, page);
1335 bio_ctrl->end_io_func = end_bio_extent_writepage;
1336 while (cur <= end) {
1337 u32 len = end - cur + 1;
1340 u64 dirty_range_start = cur;
1341 u64 dirty_range_end;
1344 if (cur >= i_size) {
1345 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1348 * This range is beyond i_size, thus we don't need to
1349 * bother writing back.
1350 * But we still need to clear the dirty subpage bit, or
1351 * the next time the page gets dirtied, we will try to
1352 * writeback the sectors with subpage dirty bits,
1353 * causing writeback without ordered extent.
1355 btrfs_page_clear_dirty(fs_info, page, cur, len);
1359 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1361 if (cur < dirty_range_start) {
1362 cur = dirty_range_start;
1366 em = btrfs_get_extent(inode, NULL, 0, cur, len);
1368 ret = PTR_ERR_OR_ZERO(em);
1372 extent_offset = cur - em->start;
1373 em_end = extent_map_end(em);
1374 ASSERT(cur <= em_end);
1376 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1377 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1379 block_start = em->block_start;
1380 disk_bytenr = em->block_start + extent_offset;
1382 ASSERT(!extent_map_is_compressed(em));
1383 ASSERT(block_start != EXTENT_MAP_HOLE);
1384 ASSERT(block_start != EXTENT_MAP_INLINE);
1387 * Note that em_end from extent_map_end() and dirty_range_end from
1388 * find_next_dirty_byte() are all exclusive
1390 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1391 free_extent_map(em);
1394 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1395 if (!PageWriteback(page)) {
1396 btrfs_err(inode->root->fs_info,
1397 "page %lu not writeback, cur %llu end %llu",
1398 page->index, cur, end);
1402 * Although the PageDirty bit is cleared before entering this
1403 * function, subpage dirty bit is not cleared.
1404 * So clear subpage dirty bit here so next time we won't submit
1405 * page for range already written to disk.
1407 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
1409 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1410 cur - page_offset(page));
1415 btrfs_page_assert_not_dirty(fs_info, page);
1421 * If we finish without problem, we should not only clear page dirty,
1422 * but also empty subpage dirty bits
1429 * the writepage semantics are similar to regular writepage. extent
1430 * records are inserted to lock ranges in the tree, and as dirty areas
1431 * are found, they are marked writeback. Then the lock bits are removed
1432 * and the end_io handler clears the writeback ranges
1434 * Return 0 if everything goes well.
1435 * Return <0 for error.
1437 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1439 struct folio *folio = page_folio(page);
1440 struct inode *inode = page->mapping->host;
1441 const u64 page_start = page_offset(page);
1445 loff_t i_size = i_size_read(inode);
1446 unsigned long end_index = i_size >> PAGE_SHIFT;
1448 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1450 WARN_ON(!PageLocked(page));
1452 pg_offset = offset_in_page(i_size);
1453 if (page->index > end_index ||
1454 (page->index == end_index && !pg_offset)) {
1455 folio_invalidate(folio, 0, folio_size(folio));
1456 folio_unlock(folio);
1460 if (page->index == end_index)
1461 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1463 ret = set_page_extent_mapped(page);
1467 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1473 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1477 bio_ctrl->wbc->nr_to_write--;
1481 /* make sure the mapping tag for page dirty gets cleared */
1482 set_page_writeback(page);
1483 end_page_writeback(page);
1486 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1488 mapping_set_error(page->mapping, ret);
1495 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1497 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1498 TASK_UNINTERRUPTIBLE);
1502 * Lock extent buffer status and pages for writeback.
1504 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1505 * extent buffer is not dirty)
1506 * Return %true is the extent buffer is submitted to bio.
1508 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1509 struct writeback_control *wbc)
1511 struct btrfs_fs_info *fs_info = eb->fs_info;
1514 btrfs_tree_lock(eb);
1515 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1516 btrfs_tree_unlock(eb);
1517 if (wbc->sync_mode != WB_SYNC_ALL)
1519 wait_on_extent_buffer_writeback(eb);
1520 btrfs_tree_lock(eb);
1524 * We need to do this to prevent races in people who check if the eb is
1525 * under IO since we can end up having no IO bits set for a short period
1528 spin_lock(&eb->refs_lock);
1529 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1530 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1531 spin_unlock(&eb->refs_lock);
1532 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1533 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1535 fs_info->dirty_metadata_batch);
1538 spin_unlock(&eb->refs_lock);
1540 btrfs_tree_unlock(eb);
1544 static void set_btree_ioerr(struct extent_buffer *eb)
1546 struct btrfs_fs_info *fs_info = eb->fs_info;
1548 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1551 * A read may stumble upon this buffer later, make sure that it gets an
1552 * error and knows there was an error.
1554 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1557 * We need to set the mapping with the io error as well because a write
1558 * error will flip the file system readonly, and then syncfs() will
1559 * return a 0 because we are readonly if we don't modify the err seq for
1562 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1565 * If writeback for a btree extent that doesn't belong to a log tree
1566 * failed, increment the counter transaction->eb_write_errors.
1567 * We do this because while the transaction is running and before it's
1568 * committing (when we call filemap_fdata[write|wait]_range against
1569 * the btree inode), we might have
1570 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1571 * returns an error or an error happens during writeback, when we're
1572 * committing the transaction we wouldn't know about it, since the pages
1573 * can be no longer dirty nor marked anymore for writeback (if a
1574 * subsequent modification to the extent buffer didn't happen before the
1575 * transaction commit), which makes filemap_fdata[write|wait]_range not
1576 * able to find the pages tagged with SetPageError at transaction
1577 * commit time. So if this happens we must abort the transaction,
1578 * otherwise we commit a super block with btree roots that point to
1579 * btree nodes/leafs whose content on disk is invalid - either garbage
1580 * or the content of some node/leaf from a past generation that got
1581 * cowed or deleted and is no longer valid.
1583 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1584 * not be enough - we need to distinguish between log tree extents vs
1585 * non-log tree extents, and the next filemap_fdatawait_range() call
1586 * will catch and clear such errors in the mapping - and that call might
1587 * be from a log sync and not from a transaction commit. Also, checking
1588 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1589 * not done and would not be reliable - the eb might have been released
1590 * from memory and reading it back again means that flag would not be
1591 * set (since it's a runtime flag, not persisted on disk).
1593 * Using the flags below in the btree inode also makes us achieve the
1594 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1595 * writeback for all dirty pages and before filemap_fdatawait_range()
1596 * is called, the writeback for all dirty pages had already finished
1597 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1598 * filemap_fdatawait_range() would return success, as it could not know
1599 * that writeback errors happened (the pages were no longer tagged for
1602 switch (eb->log_index) {
1604 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1607 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1610 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1613 BUG(); /* unexpected, logic error */
1618 * The endio specific version which won't touch any unsafe spinlock in endio
1621 static struct extent_buffer *find_extent_buffer_nolock(
1622 struct btrfs_fs_info *fs_info, u64 start)
1624 struct extent_buffer *eb;
1627 eb = radix_tree_lookup(&fs_info->buffer_radix,
1628 start >> fs_info->sectorsize_bits);
1629 if (eb && atomic_inc_not_zero(&eb->refs)) {
1637 static void extent_buffer_write_end_io(struct btrfs_bio *bbio)
1639 struct extent_buffer *eb = bbio->private;
1640 struct btrfs_fs_info *fs_info = eb->fs_info;
1641 bool uptodate = !bbio->bio.bi_status;
1642 struct bvec_iter_all iter_all;
1643 struct bio_vec *bvec;
1647 set_btree_ioerr(eb);
1649 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
1650 u64 start = eb->start + bio_offset;
1651 struct page *page = bvec->bv_page;
1652 u32 len = bvec->bv_len;
1654 btrfs_page_clear_writeback(fs_info, page, start, len);
1658 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1659 smp_mb__after_atomic();
1660 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1662 bio_put(&bbio->bio);
1665 static void prepare_eb_write(struct extent_buffer *eb)
1668 unsigned long start;
1671 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1673 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1674 nritems = btrfs_header_nritems(eb);
1675 if (btrfs_header_level(eb) > 0) {
1676 end = btrfs_node_key_ptr_offset(eb, nritems);
1677 memzero_extent_buffer(eb, end, eb->len - end);
1681 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1683 start = btrfs_item_nr_offset(eb, nritems);
1684 end = btrfs_item_nr_offset(eb, 0);
1686 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1688 end += btrfs_item_offset(eb, nritems - 1);
1689 memzero_extent_buffer(eb, start, end - start);
1693 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1694 struct writeback_control *wbc)
1696 struct btrfs_fs_info *fs_info = eb->fs_info;
1697 struct btrfs_bio *bbio;
1699 prepare_eb_write(eb);
1701 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1702 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1703 eb->fs_info, extent_buffer_write_end_io, eb);
1704 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1705 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1706 wbc_init_bio(wbc, &bbio->bio);
1707 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1708 bbio->file_offset = eb->start;
1709 if (fs_info->nodesize < PAGE_SIZE) {
1710 struct page *p = folio_page(eb->folios[0], 0);
1713 btrfs_subpage_set_writeback(fs_info, p, eb->start, eb->len);
1714 if (btrfs_subpage_clear_and_test_dirty(fs_info, p, eb->start,
1716 clear_page_dirty_for_io(p);
1719 __bio_add_page(&bbio->bio, p, eb->len, eb->start - page_offset(p));
1720 wbc_account_cgroup_owner(wbc, p, eb->len);
1723 int num_folios = num_extent_folios(eb);
1725 for (int i = 0; i < num_folios; i++) {
1726 struct folio *folio = eb->folios[i];
1730 folio_clear_dirty_for_io(folio);
1731 folio_start_writeback(folio);
1732 ret = bio_add_folio(&bbio->bio, folio, folio_size(folio), 0);
1734 wbc_account_cgroup_owner(wbc, folio_page(folio, 0),
1736 wbc->nr_to_write -= folio_nr_pages(folio);
1737 folio_unlock(folio);
1740 btrfs_submit_bio(bbio, 0);
1744 * Submit one subpage btree page.
1746 * The main difference to submit_eb_page() is:
1748 * For subpage, we don't rely on page locking at all.
1751 * We only flush bio if we may be unable to fit current extent buffers into
1754 * Return >=0 for the number of submitted extent buffers.
1755 * Return <0 for fatal error.
1757 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1759 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1760 struct folio *folio = page_folio(page);
1762 u64 page_start = page_offset(page);
1764 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1766 /* Lock and write each dirty extent buffers in the range */
1767 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1768 struct btrfs_subpage *subpage = folio_get_private(folio);
1769 struct extent_buffer *eb;
1770 unsigned long flags;
1774 * Take private lock to ensure the subpage won't be detached
1777 spin_lock(&page->mapping->private_lock);
1778 if (!folio_test_private(folio)) {
1779 spin_unlock(&page->mapping->private_lock);
1782 spin_lock_irqsave(&subpage->lock, flags);
1783 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1784 subpage->bitmaps)) {
1785 spin_unlock_irqrestore(&subpage->lock, flags);
1786 spin_unlock(&page->mapping->private_lock);
1791 start = page_start + bit_start * fs_info->sectorsize;
1792 bit_start += sectors_per_node;
1795 * Here we just want to grab the eb without touching extra
1796 * spin locks, so call find_extent_buffer_nolock().
1798 eb = find_extent_buffer_nolock(fs_info, start);
1799 spin_unlock_irqrestore(&subpage->lock, flags);
1800 spin_unlock(&page->mapping->private_lock);
1803 * The eb has already reached 0 refs thus find_extent_buffer()
1804 * doesn't return it. We don't need to write back such eb
1810 if (lock_extent_buffer_for_io(eb, wbc)) {
1811 write_one_eb(eb, wbc);
1814 free_extent_buffer(eb);
1820 * Submit all page(s) of one extent buffer.
1822 * @page: the page of one extent buffer
1823 * @eb_context: to determine if we need to submit this page, if current page
1824 * belongs to this eb, we don't need to submit
1826 * The caller should pass each page in their bytenr order, and here we use
1827 * @eb_context to determine if we have submitted pages of one extent buffer.
1829 * If we have, we just skip until we hit a new page that doesn't belong to
1830 * current @eb_context.
1832 * If not, we submit all the page(s) of the extent buffer.
1834 * Return >0 if we have submitted the extent buffer successfully.
1835 * Return 0 if we don't need to submit the page, as it's already submitted by
1837 * Return <0 for fatal error.
1839 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1841 struct writeback_control *wbc = ctx->wbc;
1842 struct address_space *mapping = page->mapping;
1843 struct folio *folio = page_folio(page);
1844 struct extent_buffer *eb;
1847 if (!folio_test_private(folio))
1850 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
1851 return submit_eb_subpage(page, wbc);
1853 spin_lock(&mapping->private_lock);
1854 if (!folio_test_private(folio)) {
1855 spin_unlock(&mapping->private_lock);
1859 eb = folio_get_private(folio);
1862 * Shouldn't happen and normally this would be a BUG_ON but no point
1863 * crashing the machine for something we can survive anyway.
1866 spin_unlock(&mapping->private_lock);
1870 if (eb == ctx->eb) {
1871 spin_unlock(&mapping->private_lock);
1874 ret = atomic_inc_not_zero(&eb->refs);
1875 spin_unlock(&mapping->private_lock);
1881 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1885 free_extent_buffer(eb);
1889 if (!lock_extent_buffer_for_io(eb, wbc)) {
1890 free_extent_buffer(eb);
1893 /* Implies write in zoned mode. */
1894 if (ctx->zoned_bg) {
1895 /* Mark the last eb in the block group. */
1896 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1897 ctx->zoned_bg->meta_write_pointer += eb->len;
1899 write_one_eb(eb, wbc);
1900 free_extent_buffer(eb);
1904 int btree_write_cache_pages(struct address_space *mapping,
1905 struct writeback_control *wbc)
1907 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1908 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
1911 int nr_to_write_done = 0;
1912 struct folio_batch fbatch;
1913 unsigned int nr_folios;
1915 pgoff_t end; /* Inclusive */
1919 folio_batch_init(&fbatch);
1920 if (wbc->range_cyclic) {
1921 index = mapping->writeback_index; /* Start from prev offset */
1924 * Start from the beginning does not need to cycle over the
1925 * range, mark it as scanned.
1927 scanned = (index == 0);
1929 index = wbc->range_start >> PAGE_SHIFT;
1930 end = wbc->range_end >> PAGE_SHIFT;
1933 if (wbc->sync_mode == WB_SYNC_ALL)
1934 tag = PAGECACHE_TAG_TOWRITE;
1936 tag = PAGECACHE_TAG_DIRTY;
1937 btrfs_zoned_meta_io_lock(fs_info);
1939 if (wbc->sync_mode == WB_SYNC_ALL)
1940 tag_pages_for_writeback(mapping, index, end);
1941 while (!done && !nr_to_write_done && (index <= end) &&
1942 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1946 for (i = 0; i < nr_folios; i++) {
1947 struct folio *folio = fbatch.folios[i];
1949 ret = submit_eb_page(&folio->page, &ctx);
1958 * the filesystem may choose to bump up nr_to_write.
1959 * We have to make sure to honor the new nr_to_write
1962 nr_to_write_done = wbc->nr_to_write <= 0;
1964 folio_batch_release(&fbatch);
1967 if (!scanned && !done) {
1969 * We hit the last page and there is more work to be done: wrap
1970 * back to the start of the file
1977 * If something went wrong, don't allow any metadata write bio to be
1980 * This would prevent use-after-free if we had dirty pages not
1981 * cleaned up, which can still happen by fuzzed images.
1984 * Allowing existing tree block to be allocated for other trees.
1986 * - Log tree operations
1987 * Exiting tree blocks get allocated to log tree, bumps its
1988 * generation, then get cleaned in tree re-balance.
1989 * Such tree block will not be written back, since it's clean,
1990 * thus no WRITTEN flag set.
1991 * And after log writes back, this tree block is not traced by
1992 * any dirty extent_io_tree.
1994 * - Offending tree block gets re-dirtied from its original owner
1995 * Since it has bumped generation, no WRITTEN flag, it can be
1996 * reused without COWing. This tree block will not be traced
1997 * by btrfs_transaction::dirty_pages.
1999 * Now such dirty tree block will not be cleaned by any dirty
2000 * extent io tree. Thus we don't want to submit such wild eb
2001 * if the fs already has error.
2003 * We can get ret > 0 from submit_extent_page() indicating how many ebs
2004 * were submitted. Reset it to 0 to avoid false alerts for the caller.
2008 if (!ret && BTRFS_FS_ERROR(fs_info))
2012 btrfs_put_block_group(ctx.zoned_bg);
2013 btrfs_zoned_meta_io_unlock(fs_info);
2018 * Walk the list of dirty pages of the given address space and write all of them.
2020 * @mapping: address space structure to write
2021 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2022 * @bio_ctrl: holds context for the write, namely the bio
2024 * If a page is already under I/O, write_cache_pages() skips it, even
2025 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2026 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2027 * and msync() need to guarantee that all the data which was dirty at the time
2028 * the call was made get new I/O started against them. If wbc->sync_mode is
2029 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2030 * existing IO to complete.
2032 static int extent_write_cache_pages(struct address_space *mapping,
2033 struct btrfs_bio_ctrl *bio_ctrl)
2035 struct writeback_control *wbc = bio_ctrl->wbc;
2036 struct inode *inode = mapping->host;
2039 int nr_to_write_done = 0;
2040 struct folio_batch fbatch;
2041 unsigned int nr_folios;
2043 pgoff_t end; /* Inclusive */
2045 int range_whole = 0;
2050 * We have to hold onto the inode so that ordered extents can do their
2051 * work when the IO finishes. The alternative to this is failing to add
2052 * an ordered extent if the igrab() fails there and that is a huge pain
2053 * to deal with, so instead just hold onto the inode throughout the
2054 * writepages operation. If it fails here we are freeing up the inode
2055 * anyway and we'd rather not waste our time writing out stuff that is
2056 * going to be truncated anyway.
2061 folio_batch_init(&fbatch);
2062 if (wbc->range_cyclic) {
2063 index = mapping->writeback_index; /* Start from prev offset */
2066 * Start from the beginning does not need to cycle over the
2067 * range, mark it as scanned.
2069 scanned = (index == 0);
2071 index = wbc->range_start >> PAGE_SHIFT;
2072 end = wbc->range_end >> PAGE_SHIFT;
2073 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2079 * We do the tagged writepage as long as the snapshot flush bit is set
2080 * and we are the first one who do the filemap_flush() on this inode.
2082 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2083 * not race in and drop the bit.
2085 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2086 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2087 &BTRFS_I(inode)->runtime_flags))
2088 wbc->tagged_writepages = 1;
2090 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2091 tag = PAGECACHE_TAG_TOWRITE;
2093 tag = PAGECACHE_TAG_DIRTY;
2095 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2096 tag_pages_for_writeback(mapping, index, end);
2098 while (!done && !nr_to_write_done && (index <= end) &&
2099 (nr_folios = filemap_get_folios_tag(mapping, &index,
2100 end, tag, &fbatch))) {
2103 for (i = 0; i < nr_folios; i++) {
2104 struct folio *folio = fbatch.folios[i];
2106 done_index = folio_next_index(folio);
2108 * At this point we hold neither the i_pages lock nor
2109 * the page lock: the page may be truncated or
2110 * invalidated (changing page->mapping to NULL),
2111 * or even swizzled back from swapper_space to
2112 * tmpfs file mapping
2114 if (!folio_trylock(folio)) {
2115 submit_write_bio(bio_ctrl, 0);
2119 if (unlikely(folio->mapping != mapping)) {
2120 folio_unlock(folio);
2124 if (!folio_test_dirty(folio)) {
2125 /* Someone wrote it for us. */
2126 folio_unlock(folio);
2130 if (wbc->sync_mode != WB_SYNC_NONE) {
2131 if (folio_test_writeback(folio))
2132 submit_write_bio(bio_ctrl, 0);
2133 folio_wait_writeback(folio);
2136 if (folio_test_writeback(folio) ||
2137 !folio_clear_dirty_for_io(folio)) {
2138 folio_unlock(folio);
2142 ret = __extent_writepage(&folio->page, bio_ctrl);
2149 * The filesystem may choose to bump up nr_to_write.
2150 * We have to make sure to honor the new nr_to_write
2153 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2154 wbc->nr_to_write <= 0);
2156 folio_batch_release(&fbatch);
2159 if (!scanned && !done) {
2161 * We hit the last page and there is more work to be done: wrap
2162 * back to the start of the file
2168 * If we're looping we could run into a page that is locked by a
2169 * writer and that writer could be waiting on writeback for a
2170 * page in our current bio, and thus deadlock, so flush the
2173 submit_write_bio(bio_ctrl, 0);
2177 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2178 mapping->writeback_index = done_index;
2180 btrfs_add_delayed_iput(BTRFS_I(inode));
2185 * Submit the pages in the range to bio for call sites which delalloc range has
2186 * already been ran (aka, ordered extent inserted) and all pages are still
2189 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2190 u64 start, u64 end, struct writeback_control *wbc,
2193 bool found_error = false;
2195 struct address_space *mapping = inode->i_mapping;
2196 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2197 const u32 sectorsize = fs_info->sectorsize;
2198 loff_t i_size = i_size_read(inode);
2200 struct btrfs_bio_ctrl bio_ctrl = {
2202 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2205 if (wbc->no_cgroup_owner)
2206 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2208 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2210 while (cur <= end) {
2211 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2212 u32 cur_len = cur_end + 1 - cur;
2216 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2217 ASSERT(PageLocked(page));
2218 if (pages_dirty && page != locked_page) {
2219 ASSERT(PageDirty(page));
2220 clear_page_dirty_for_io(page);
2223 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2228 /* Make sure the mapping tag for page dirty gets cleared. */
2230 set_page_writeback(page);
2231 end_page_writeback(page);
2234 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2235 cur, cur_len, !ret);
2236 mapping_set_error(page->mapping, ret);
2238 btrfs_page_unlock_writer(fs_info, page, cur, cur_len);
2246 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2249 int extent_writepages(struct address_space *mapping,
2250 struct writeback_control *wbc)
2252 struct inode *inode = mapping->host;
2254 struct btrfs_bio_ctrl bio_ctrl = {
2256 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2260 * Allow only a single thread to do the reloc work in zoned mode to
2261 * protect the write pointer updates.
2263 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2264 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2265 submit_write_bio(&bio_ctrl, ret);
2266 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2270 void extent_readahead(struct readahead_control *rac)
2272 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2273 struct page *pagepool[16];
2274 struct extent_map *em_cached = NULL;
2275 u64 prev_em_start = (u64)-1;
2278 while ((nr = readahead_page_batch(rac, pagepool))) {
2279 u64 contig_start = readahead_pos(rac);
2280 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2282 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2283 &em_cached, &bio_ctrl, &prev_em_start);
2287 free_extent_map(em_cached);
2288 submit_one_bio(&bio_ctrl);
2292 * basic invalidate_folio code, this waits on any locked or writeback
2293 * ranges corresponding to the folio, and then deletes any extent state
2294 * records from the tree
2296 int extent_invalidate_folio(struct extent_io_tree *tree,
2297 struct folio *folio, size_t offset)
2299 struct extent_state *cached_state = NULL;
2300 u64 start = folio_pos(folio);
2301 u64 end = start + folio_size(folio) - 1;
2302 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
2304 /* This function is only called for the btree inode */
2305 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2307 start += ALIGN(offset, blocksize);
2311 lock_extent(tree, start, end, &cached_state);
2312 folio_wait_writeback(folio);
2315 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2316 * so here we only need to unlock the extent range to free any
2317 * existing extent state.
2319 unlock_extent(tree, start, end, &cached_state);
2324 * a helper for release_folio, this tests for areas of the page that
2325 * are locked or under IO and drops the related state bits if it is safe
2328 static int try_release_extent_state(struct extent_io_tree *tree,
2329 struct page *page, gfp_t mask)
2331 u64 start = page_offset(page);
2332 u64 end = start + PAGE_SIZE - 1;
2335 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2338 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2339 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS |
2340 EXTENT_QGROUP_RESERVED);
2343 * At this point we can safely clear everything except the
2344 * locked bit, the nodatasum bit and the delalloc new bit.
2345 * The delalloc new bit will be cleared by ordered extent
2348 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2350 /* if clear_extent_bit failed for enomem reasons,
2351 * we can't allow the release to continue.
2362 * a helper for release_folio. As long as there are no locked extents
2363 * in the range corresponding to the page, both state records and extent
2364 * map records are removed
2366 int try_release_extent_mapping(struct page *page, gfp_t mask)
2368 struct extent_map *em;
2369 u64 start = page_offset(page);
2370 u64 end = start + PAGE_SIZE - 1;
2371 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
2372 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2373 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2375 if (gfpflags_allow_blocking(mask) &&
2376 page->mapping->host->i_size > SZ_16M) {
2378 while (start <= end) {
2379 struct btrfs_fs_info *fs_info;
2382 len = end - start + 1;
2383 write_lock(&map->lock);
2384 em = lookup_extent_mapping(map, start, len);
2386 write_unlock(&map->lock);
2389 if ((em->flags & EXTENT_FLAG_PINNED) ||
2390 em->start != start) {
2391 write_unlock(&map->lock);
2392 free_extent_map(em);
2395 if (test_range_bit_exists(tree, em->start,
2396 extent_map_end(em) - 1,
2400 * If it's not in the list of modified extents, used
2401 * by a fast fsync, we can remove it. If it's being
2402 * logged we can safely remove it since fsync took an
2403 * extra reference on the em.
2405 if (list_empty(&em->list) ||
2406 (em->flags & EXTENT_FLAG_LOGGING))
2409 * If it's in the list of modified extents, remove it
2410 * only if its generation is older then the current one,
2411 * in which case we don't need it for a fast fsync.
2412 * Otherwise don't remove it, we could be racing with an
2413 * ongoing fast fsync that could miss the new extent.
2415 fs_info = btrfs_inode->root->fs_info;
2416 spin_lock(&fs_info->trans_lock);
2417 cur_gen = fs_info->generation;
2418 spin_unlock(&fs_info->trans_lock);
2419 if (em->generation >= cur_gen)
2423 * We only remove extent maps that are not in the list of
2424 * modified extents or that are in the list but with a
2425 * generation lower then the current generation, so there
2426 * is no need to set the full fsync flag on the inode (it
2427 * hurts the fsync performance for workloads with a data
2428 * size that exceeds or is close to the system's memory).
2430 remove_extent_mapping(map, em);
2431 /* once for the rb tree */
2432 free_extent_map(em);
2434 start = extent_map_end(em);
2435 write_unlock(&map->lock);
2438 free_extent_map(em);
2440 cond_resched(); /* Allow large-extent preemption. */
2443 return try_release_extent_state(tree, page, mask);
2447 * To cache previous fiemap extent
2449 * Will be used for merging fiemap extent
2451 struct fiemap_cache {
2460 * Helper to submit fiemap extent.
2462 * Will try to merge current fiemap extent specified by @offset, @phys,
2463 * @len and @flags with cached one.
2464 * And only when we fails to merge, cached one will be submitted as
2467 * Return value is the same as fiemap_fill_next_extent().
2469 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2470 struct fiemap_cache *cache,
2471 u64 offset, u64 phys, u64 len, u32 flags)
2475 /* Set at the end of extent_fiemap(). */
2476 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2482 * Sanity check, extent_fiemap() should have ensured that new
2483 * fiemap extent won't overlap with cached one.
2486 * NOTE: Physical address can overlap, due to compression
2488 if (cache->offset + cache->len > offset) {
2494 * Only merges fiemap extents if
2495 * 1) Their logical addresses are continuous
2497 * 2) Their physical addresses are continuous
2498 * So truly compressed (physical size smaller than logical size)
2499 * extents won't get merged with each other
2501 * 3) Share same flags
2503 if (cache->offset + cache->len == offset &&
2504 cache->phys + cache->len == phys &&
2505 cache->flags == flags) {
2510 /* Not mergeable, need to submit cached one */
2511 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2512 cache->len, cache->flags);
2513 cache->cached = false;
2517 cache->cached = true;
2518 cache->offset = offset;
2521 cache->flags = flags;
2527 * Emit last fiemap cache
2529 * The last fiemap cache may still be cached in the following case:
2531 * |<- Fiemap range ->|
2532 * |<------------ First extent ----------->|
2534 * In this case, the first extent range will be cached but not emitted.
2535 * So we must emit it before ending extent_fiemap().
2537 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2538 struct fiemap_cache *cache)
2545 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2546 cache->len, cache->flags);
2547 cache->cached = false;
2553 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2555 struct extent_buffer *clone;
2556 struct btrfs_key key;
2561 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2564 ret = btrfs_next_leaf(inode->root, path);
2569 * Don't bother with cloning if there are no more file extent items for
2572 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2573 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
2576 /* See the comment at fiemap_search_slot() about why we clone. */
2577 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2581 slot = path->slots[0];
2582 btrfs_release_path(path);
2583 path->nodes[0] = clone;
2584 path->slots[0] = slot;
2590 * Search for the first file extent item that starts at a given file offset or
2591 * the one that starts immediately before that offset.
2592 * Returns: 0 on success, < 0 on error, 1 if not found.
2594 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2597 const u64 ino = btrfs_ino(inode);
2598 struct btrfs_root *root = inode->root;
2599 struct extent_buffer *clone;
2600 struct btrfs_key key;
2605 key.type = BTRFS_EXTENT_DATA_KEY;
2606 key.offset = file_offset;
2608 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2612 if (ret > 0 && path->slots[0] > 0) {
2613 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2614 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2618 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2619 ret = btrfs_next_leaf(root, path);
2623 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2624 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2629 * We clone the leaf and use it during fiemap. This is because while
2630 * using the leaf we do expensive things like checking if an extent is
2631 * shared, which can take a long time. In order to prevent blocking
2632 * other tasks for too long, we use a clone of the leaf. We have locked
2633 * the file range in the inode's io tree, so we know none of our file
2634 * extent items can change. This way we avoid blocking other tasks that
2635 * want to insert items for other inodes in the same leaf or b+tree
2636 * rebalance operations (triggered for example when someone is trying
2637 * to push items into this leaf when trying to insert an item in a
2639 * We also need the private clone because holding a read lock on an
2640 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2641 * when we call fiemap_fill_next_extent(), because that may cause a page
2642 * fault when filling the user space buffer with fiemap data.
2644 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2648 slot = path->slots[0];
2649 btrfs_release_path(path);
2650 path->nodes[0] = clone;
2651 path->slots[0] = slot;
2657 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2658 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2659 * extent. The end offset (@end) is inclusive.
2661 static int fiemap_process_hole(struct btrfs_inode *inode,
2662 struct fiemap_extent_info *fieinfo,
2663 struct fiemap_cache *cache,
2664 struct extent_state **delalloc_cached_state,
2665 struct btrfs_backref_share_check_ctx *backref_ctx,
2666 u64 disk_bytenr, u64 extent_offset,
2670 const u64 i_size = i_size_read(&inode->vfs_inode);
2671 u64 cur_offset = start;
2672 u64 last_delalloc_end = 0;
2673 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2674 bool checked_extent_shared = false;
2678 * There can be no delalloc past i_size, so don't waste time looking for
2681 while (cur_offset < end && cur_offset < i_size) {
2685 u64 prealloc_len = 0;
2688 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2689 delalloc_cached_state,
2696 * If this is a prealloc extent we have to report every section
2697 * of it that has no delalloc.
2699 if (disk_bytenr != 0) {
2700 if (last_delalloc_end == 0) {
2701 prealloc_start = start;
2702 prealloc_len = delalloc_start - start;
2704 prealloc_start = last_delalloc_end + 1;
2705 prealloc_len = delalloc_start - prealloc_start;
2709 if (prealloc_len > 0) {
2710 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2711 ret = btrfs_is_data_extent_shared(inode,
2718 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2720 checked_extent_shared = true;
2722 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2723 disk_bytenr + extent_offset,
2724 prealloc_len, prealloc_flags);
2727 extent_offset += prealloc_len;
2730 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2731 delalloc_end + 1 - delalloc_start,
2732 FIEMAP_EXTENT_DELALLOC |
2733 FIEMAP_EXTENT_UNKNOWN);
2737 last_delalloc_end = delalloc_end;
2738 cur_offset = delalloc_end + 1;
2739 extent_offset += cur_offset - delalloc_start;
2744 * Either we found no delalloc for the whole prealloc extent or we have
2745 * a prealloc extent that spans i_size or starts at or after i_size.
2747 if (disk_bytenr != 0 && last_delalloc_end < end) {
2751 if (last_delalloc_end == 0) {
2752 prealloc_start = start;
2753 prealloc_len = end + 1 - start;
2755 prealloc_start = last_delalloc_end + 1;
2756 prealloc_len = end + 1 - prealloc_start;
2759 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2760 ret = btrfs_is_data_extent_shared(inode,
2767 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2769 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2770 disk_bytenr + extent_offset,
2771 prealloc_len, prealloc_flags);
2779 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2780 struct btrfs_path *path,
2781 u64 *last_extent_end_ret)
2783 const u64 ino = btrfs_ino(inode);
2784 struct btrfs_root *root = inode->root;
2785 struct extent_buffer *leaf;
2786 struct btrfs_file_extent_item *ei;
2787 struct btrfs_key key;
2792 * Lookup the last file extent. We're not using i_size here because
2793 * there might be preallocation past i_size.
2795 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
2796 /* There can't be a file extent item at offset (u64)-1 */
2802 * For a non-existing key, btrfs_search_slot() always leaves us at a
2803 * slot > 0, except if the btree is empty, which is impossible because
2804 * at least it has the inode item for this inode and all the items for
2805 * the root inode 256.
2807 ASSERT(path->slots[0] > 0);
2809 leaf = path->nodes[0];
2810 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2811 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
2812 /* No file extent items in the subvolume tree. */
2813 *last_extent_end_ret = 0;
2818 * For an inline extent, the disk_bytenr is where inline data starts at,
2819 * so first check if we have an inline extent item before checking if we
2820 * have an implicit hole (disk_bytenr == 0).
2822 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2823 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
2824 *last_extent_end_ret = btrfs_file_extent_end(path);
2829 * Find the last file extent item that is not a hole (when NO_HOLES is
2830 * not enabled). This should take at most 2 iterations in the worst
2831 * case: we have one hole file extent item at slot 0 of a leaf and
2832 * another hole file extent item as the last item in the previous leaf.
2833 * This is because we merge file extent items that represent holes.
2835 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2836 while (disk_bytenr == 0) {
2837 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
2840 } else if (ret > 0) {
2841 /* No file extent items that are not holes. */
2842 *last_extent_end_ret = 0;
2845 leaf = path->nodes[0];
2846 ei = btrfs_item_ptr(leaf, path->slots[0],
2847 struct btrfs_file_extent_item);
2848 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2851 *last_extent_end_ret = btrfs_file_extent_end(path);
2855 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
2858 const u64 ino = btrfs_ino(inode);
2859 struct extent_state *cached_state = NULL;
2860 struct extent_state *delalloc_cached_state = NULL;
2861 struct btrfs_path *path;
2862 struct fiemap_cache cache = { 0 };
2863 struct btrfs_backref_share_check_ctx *backref_ctx;
2864 u64 last_extent_end;
2865 u64 prev_extent_end;
2868 bool stopped = false;
2871 backref_ctx = btrfs_alloc_backref_share_check_ctx();
2872 path = btrfs_alloc_path();
2873 if (!backref_ctx || !path) {
2878 lockstart = round_down(start, inode->root->fs_info->sectorsize);
2879 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
2880 prev_extent_end = lockstart;
2882 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
2883 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
2885 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
2888 btrfs_release_path(path);
2890 path->reada = READA_FORWARD;
2891 ret = fiemap_search_slot(inode, path, lockstart);
2894 } else if (ret > 0) {
2896 * No file extent item found, but we may have delalloc between
2897 * the current offset and i_size. So check for that.
2900 goto check_eof_delalloc;
2903 while (prev_extent_end < lockend) {
2904 struct extent_buffer *leaf = path->nodes[0];
2905 struct btrfs_file_extent_item *ei;
2906 struct btrfs_key key;
2909 u64 extent_offset = 0;
2911 u64 disk_bytenr = 0;
2916 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2917 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2920 extent_end = btrfs_file_extent_end(path);
2923 * The first iteration can leave us at an extent item that ends
2924 * before our range's start. Move to the next item.
2926 if (extent_end <= lockstart)
2929 backref_ctx->curr_leaf_bytenr = leaf->start;
2931 /* We have in implicit hole (NO_HOLES feature enabled). */
2932 if (prev_extent_end < key.offset) {
2933 const u64 range_end = min(key.offset, lockend) - 1;
2935 ret = fiemap_process_hole(inode, fieinfo, &cache,
2936 &delalloc_cached_state,
2937 backref_ctx, 0, 0, 0,
2938 prev_extent_end, range_end);
2941 } else if (ret > 0) {
2942 /* fiemap_fill_next_extent() told us to stop. */
2947 /* We've reached the end of the fiemap range, stop. */
2948 if (key.offset >= lockend) {
2954 extent_len = extent_end - key.offset;
2955 ei = btrfs_item_ptr(leaf, path->slots[0],
2956 struct btrfs_file_extent_item);
2957 compression = btrfs_file_extent_compression(leaf, ei);
2958 extent_type = btrfs_file_extent_type(leaf, ei);
2959 extent_gen = btrfs_file_extent_generation(leaf, ei);
2961 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2962 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2963 if (compression == BTRFS_COMPRESS_NONE)
2964 extent_offset = btrfs_file_extent_offset(leaf, ei);
2967 if (compression != BTRFS_COMPRESS_NONE)
2968 flags |= FIEMAP_EXTENT_ENCODED;
2970 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2971 flags |= FIEMAP_EXTENT_DATA_INLINE;
2972 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
2973 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
2975 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2976 ret = fiemap_process_hole(inode, fieinfo, &cache,
2977 &delalloc_cached_state,
2979 disk_bytenr, extent_offset,
2980 extent_gen, key.offset,
2982 } else if (disk_bytenr == 0) {
2983 /* We have an explicit hole. */
2984 ret = fiemap_process_hole(inode, fieinfo, &cache,
2985 &delalloc_cached_state,
2986 backref_ctx, 0, 0, 0,
2987 key.offset, extent_end - 1);
2989 /* We have a regular extent. */
2990 if (fieinfo->fi_extents_max) {
2991 ret = btrfs_is_data_extent_shared(inode,
2998 flags |= FIEMAP_EXTENT_SHARED;
3001 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
3002 disk_bytenr + extent_offset,
3008 } else if (ret > 0) {
3009 /* fiemap_fill_next_extent() told us to stop. */
3014 prev_extent_end = extent_end;
3016 if (fatal_signal_pending(current)) {
3021 ret = fiemap_next_leaf_item(inode, path);
3024 } else if (ret > 0) {
3025 /* No more file extent items for this inode. */
3033 * Release (and free) the path before emitting any final entries to
3034 * fiemap_fill_next_extent() to keep lockdep happy. This is because
3035 * once we find no more file extent items exist, we may have a
3036 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
3037 * faults when copying data to the user space buffer.
3039 btrfs_free_path(path);
3042 if (!stopped && prev_extent_end < lockend) {
3043 ret = fiemap_process_hole(inode, fieinfo, &cache,
3044 &delalloc_cached_state, backref_ctx,
3045 0, 0, 0, prev_extent_end, lockend - 1);
3048 prev_extent_end = lockend;
3051 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3052 const u64 i_size = i_size_read(&inode->vfs_inode);
3054 if (prev_extent_end < i_size) {
3059 delalloc = btrfs_find_delalloc_in_range(inode,
3062 &delalloc_cached_state,
3066 cache.flags |= FIEMAP_EXTENT_LAST;
3068 cache.flags |= FIEMAP_EXTENT_LAST;
3072 ret = emit_last_fiemap_cache(fieinfo, &cache);
3075 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3076 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3078 free_extent_state(delalloc_cached_state);
3079 btrfs_free_backref_share_ctx(backref_ctx);
3080 btrfs_free_path(path);
3084 static void __free_extent_buffer(struct extent_buffer *eb)
3086 kmem_cache_free(extent_buffer_cache, eb);
3089 static int extent_buffer_under_io(const struct extent_buffer *eb)
3091 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3092 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3095 static bool folio_range_has_eb(struct btrfs_fs_info *fs_info, struct folio *folio)
3097 struct btrfs_subpage *subpage;
3099 lockdep_assert_held(&folio->mapping->private_lock);
3101 if (folio_test_private(folio)) {
3102 subpage = folio_get_private(folio);
3103 if (atomic_read(&subpage->eb_refs))
3106 * Even there is no eb refs here, we may still have
3107 * end_page_read() call relying on page::private.
3109 if (atomic_read(&subpage->readers))
3115 static void detach_extent_buffer_folio(struct extent_buffer *eb, struct folio *folio)
3117 struct btrfs_fs_info *fs_info = eb->fs_info;
3118 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3121 * For mapped eb, we're going to change the folio private, which should
3122 * be done under the private_lock.
3125 spin_lock(&folio->mapping->private_lock);
3127 if (!folio_test_private(folio)) {
3129 spin_unlock(&folio->mapping->private_lock);
3133 if (fs_info->nodesize >= PAGE_SIZE) {
3135 * We do this since we'll remove the pages after we've
3136 * removed the eb from the radix tree, so we could race
3137 * and have this page now attached to the new eb. So
3138 * only clear folio if it's still connected to
3141 if (folio_test_private(folio) && folio_get_private(folio) == eb) {
3142 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3143 BUG_ON(folio_test_dirty(folio));
3144 BUG_ON(folio_test_writeback(folio));
3145 /* We need to make sure we haven't be attached to a new eb. */
3146 folio_detach_private(folio);
3149 spin_unlock(&folio->mapping->private_lock);
3154 * For subpage, we can have dummy eb with folio private attached. In
3155 * this case, we can directly detach the private as such folio is only
3156 * attached to one dummy eb, no sharing.
3159 btrfs_detach_subpage(fs_info, folio_page(folio, 0));
3163 btrfs_folio_dec_eb_refs(fs_info, folio);
3166 * We can only detach the folio private if there are no other ebs in the
3167 * page range and no unfinished IO.
3169 if (!folio_range_has_eb(fs_info, folio))
3170 btrfs_detach_subpage(fs_info, folio_page(folio, 0));
3172 spin_unlock(&folio->mapping->private_lock);
3175 /* Release all pages attached to the extent buffer */
3176 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3178 ASSERT(!extent_buffer_under_io(eb));
3180 for (int i = 0; i < INLINE_EXTENT_BUFFER_PAGES; i++) {
3181 struct folio *folio = eb->folios[i];
3186 detach_extent_buffer_folio(eb, folio);
3188 /* One for when we allocated the folio. */
3194 * Helper for releasing the extent buffer.
3196 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3198 btrfs_release_extent_buffer_pages(eb);
3199 btrfs_leak_debug_del_eb(eb);
3200 __free_extent_buffer(eb);
3203 static struct extent_buffer *
3204 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3207 struct extent_buffer *eb = NULL;
3209 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3212 eb->fs_info = fs_info;
3213 init_rwsem(&eb->lock);
3215 btrfs_leak_debug_add_eb(eb);
3217 spin_lock_init(&eb->refs_lock);
3218 atomic_set(&eb->refs, 1);
3220 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3225 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3227 struct extent_buffer *new;
3228 int num_folios = num_extent_folios(src);
3231 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3236 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3237 * btrfs_release_extent_buffer() have different behavior for
3238 * UNMAPPED subpage extent buffer.
3240 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3242 ret = alloc_eb_folio_array(new, 0);
3244 btrfs_release_extent_buffer(new);
3248 for (int i = 0; i < num_folios; i++) {
3249 struct folio *folio = new->folios[i];
3252 ret = attach_extent_buffer_folio(new, folio, NULL);
3254 btrfs_release_extent_buffer(new);
3257 WARN_ON(folio_test_dirty(folio));
3259 copy_extent_buffer_full(new, src);
3260 set_extent_buffer_uptodate(new);
3265 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3266 u64 start, unsigned long len)
3268 struct extent_buffer *eb;
3272 eb = __alloc_extent_buffer(fs_info, start, len);
3276 ret = alloc_eb_folio_array(eb, 0);
3280 num_folios = num_extent_folios(eb);
3281 for (int i = 0; i < num_folios; i++) {
3282 ret = attach_extent_buffer_folio(eb, eb->folios[i], NULL);
3287 set_extent_buffer_uptodate(eb);
3288 btrfs_set_header_nritems(eb, 0);
3289 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3293 for (int i = 0; i < num_folios; i++) {
3294 if (eb->folios[i]) {
3295 detach_extent_buffer_folio(eb, eb->folios[i]);
3296 __folio_put(eb->folios[i]);
3299 __free_extent_buffer(eb);
3303 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3306 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3309 static void check_buffer_tree_ref(struct extent_buffer *eb)
3313 * The TREE_REF bit is first set when the extent_buffer is added
3314 * to the radix tree. It is also reset, if unset, when a new reference
3315 * is created by find_extent_buffer.
3317 * It is only cleared in two cases: freeing the last non-tree
3318 * reference to the extent_buffer when its STALE bit is set or
3319 * calling release_folio when the tree reference is the only reference.
3321 * In both cases, care is taken to ensure that the extent_buffer's
3322 * pages are not under io. However, release_folio can be concurrently
3323 * called with creating new references, which is prone to race
3324 * conditions between the calls to check_buffer_tree_ref in those
3325 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3327 * The actual lifetime of the extent_buffer in the radix tree is
3328 * adequately protected by the refcount, but the TREE_REF bit and
3329 * its corresponding reference are not. To protect against this
3330 * class of races, we call check_buffer_tree_ref from the codepaths
3331 * which trigger io. Note that once io is initiated, TREE_REF can no
3332 * longer be cleared, so that is the moment at which any such race is
3335 refs = atomic_read(&eb->refs);
3336 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3339 spin_lock(&eb->refs_lock);
3340 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3341 atomic_inc(&eb->refs);
3342 spin_unlock(&eb->refs_lock);
3345 static void mark_extent_buffer_accessed(struct extent_buffer *eb)
3347 int num_folios= num_extent_folios(eb);
3349 check_buffer_tree_ref(eb);
3351 for (int i = 0; i < num_folios; i++)
3352 folio_mark_accessed(eb->folios[i]);
3355 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3358 struct extent_buffer *eb;
3360 eb = find_extent_buffer_nolock(fs_info, start);
3364 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3365 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3366 * another task running free_extent_buffer() might have seen that flag
3367 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3368 * writeback flags not set) and it's still in the tree (flag
3369 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3370 * decrementing the extent buffer's reference count twice. So here we
3371 * could race and increment the eb's reference count, clear its stale
3372 * flag, mark it as dirty and drop our reference before the other task
3373 * finishes executing free_extent_buffer, which would later result in
3374 * an attempt to free an extent buffer that is dirty.
3376 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3377 spin_lock(&eb->refs_lock);
3378 spin_unlock(&eb->refs_lock);
3380 mark_extent_buffer_accessed(eb);
3384 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3385 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3388 struct extent_buffer *eb, *exists = NULL;
3391 eb = find_extent_buffer(fs_info, start);
3394 eb = alloc_dummy_extent_buffer(fs_info, start);
3396 return ERR_PTR(-ENOMEM);
3397 eb->fs_info = fs_info;
3399 ret = radix_tree_preload(GFP_NOFS);
3401 exists = ERR_PTR(ret);
3404 spin_lock(&fs_info->buffer_lock);
3405 ret = radix_tree_insert(&fs_info->buffer_radix,
3406 start >> fs_info->sectorsize_bits, eb);
3407 spin_unlock(&fs_info->buffer_lock);
3408 radix_tree_preload_end();
3409 if (ret == -EEXIST) {
3410 exists = find_extent_buffer(fs_info, start);
3416 check_buffer_tree_ref(eb);
3417 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3421 btrfs_release_extent_buffer(eb);
3426 static struct extent_buffer *grab_extent_buffer(
3427 struct btrfs_fs_info *fs_info, struct page *page)
3429 struct folio *folio = page_folio(page);
3430 struct extent_buffer *exists;
3433 * For subpage case, we completely rely on radix tree to ensure we
3434 * don't try to insert two ebs for the same bytenr. So here we always
3435 * return NULL and just continue.
3437 if (fs_info->nodesize < PAGE_SIZE)
3440 /* Page not yet attached to an extent buffer */
3441 if (!folio_test_private(folio))
3445 * We could have already allocated an eb for this page and attached one
3446 * so lets see if we can get a ref on the existing eb, and if we can we
3447 * know it's good and we can just return that one, else we know we can
3448 * just overwrite folio private.
3450 exists = folio_get_private(folio);
3451 if (atomic_inc_not_zero(&exists->refs))
3454 WARN_ON(PageDirty(page));
3455 folio_detach_private(folio);
3459 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3461 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3462 btrfs_err(fs_info, "bad tree block start %llu", start);
3466 if (fs_info->nodesize < PAGE_SIZE &&
3467 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3469 "tree block crosses page boundary, start %llu nodesize %u",
3470 start, fs_info->nodesize);
3473 if (fs_info->nodesize >= PAGE_SIZE &&
3474 !PAGE_ALIGNED(start)) {
3476 "tree block is not page aligned, start %llu nodesize %u",
3477 start, fs_info->nodesize);
3480 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3481 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3483 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3484 start, fs_info->nodesize);
3491 * Return 0 if eb->folios[i] is attached to btree inode successfully.
3492 * Return >0 if there is already another extent buffer for the range,
3493 * and @found_eb_ret would be updated.
3494 * Return -EAGAIN if the filemap has an existing folio but with different size
3496 * The caller needs to free the existing folios and retry using the same order.
3498 static int attach_eb_folio_to_filemap(struct extent_buffer *eb, int i,
3499 struct extent_buffer **found_eb_ret)
3502 struct btrfs_fs_info *fs_info = eb->fs_info;
3503 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3504 const unsigned long index = eb->start >> PAGE_SHIFT;
3505 struct folio *existing_folio;
3508 ASSERT(found_eb_ret);
3510 /* Caller should ensure the folio exists. */
3511 ASSERT(eb->folios[i]);
3514 ret = filemap_add_folio(mapping, eb->folios[i], index + i,
3515 GFP_NOFS | __GFP_NOFAIL);
3519 existing_folio = filemap_lock_folio(mapping, index + i);
3520 /* The page cache only exists for a very short time, just retry. */
3521 if (IS_ERR(existing_folio))
3524 /* For now, we should only have single-page folios for btree inode. */
3525 ASSERT(folio_nr_pages(existing_folio) == 1);
3527 if (folio_size(existing_folio) != folio_size(eb->folios[0])) {
3528 folio_unlock(existing_folio);
3529 folio_put(existing_folio);
3533 if (fs_info->nodesize < PAGE_SIZE) {
3535 * We're going to reuse the existing page, can drop our page
3536 * and subpage structure now.
3538 __free_page(folio_page(eb->folios[i], 0));
3539 eb->folios[i] = existing_folio;
3541 struct extent_buffer *existing_eb;
3543 existing_eb = grab_extent_buffer(fs_info,
3544 folio_page(existing_folio, 0));
3546 /* The extent buffer still exists, we can use it directly. */
3547 *found_eb_ret = existing_eb;
3548 folio_unlock(existing_folio);
3549 folio_put(existing_folio);
3552 /* The extent buffer no longer exists, we can reuse the folio. */
3553 __free_page(folio_page(eb->folios[i], 0));
3554 eb->folios[i] = existing_folio;
3559 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3560 u64 start, u64 owner_root, int level)
3562 unsigned long len = fs_info->nodesize;
3565 struct extent_buffer *eb;
3566 struct extent_buffer *existing_eb = NULL;
3567 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3568 struct btrfs_subpage *prealloc = NULL;
3569 u64 lockdep_owner = owner_root;
3570 bool page_contig = true;
3574 if (check_eb_alignment(fs_info, start))
3575 return ERR_PTR(-EINVAL);
3577 #if BITS_PER_LONG == 32
3578 if (start >= MAX_LFS_FILESIZE) {
3579 btrfs_err_rl(fs_info,
3580 "extent buffer %llu is beyond 32bit page cache limit", start);
3581 btrfs_err_32bit_limit(fs_info);
3582 return ERR_PTR(-EOVERFLOW);
3584 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3585 btrfs_warn_32bit_limit(fs_info);
3588 eb = find_extent_buffer(fs_info, start);
3592 eb = __alloc_extent_buffer(fs_info, start, len);
3594 return ERR_PTR(-ENOMEM);
3597 * The reloc trees are just snapshots, so we need them to appear to be
3598 * just like any other fs tree WRT lockdep.
3600 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3601 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3603 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3606 * Preallocate folio private for subpage case, so that we won't
3607 * allocate memory with private_lock nor page lock hold.
3609 * The memory will be freed by attach_extent_buffer_page() or freed
3610 * manually if we exit earlier.
3612 if (fs_info->nodesize < PAGE_SIZE) {
3613 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3614 if (IS_ERR(prealloc)) {
3615 ret = PTR_ERR(prealloc);
3621 /* Allocate all pages first. */
3622 ret = alloc_eb_folio_array(eb, __GFP_NOFAIL);
3624 btrfs_free_subpage(prealloc);
3628 num_folios = num_extent_folios(eb);
3629 /* Attach all pages to the filemap. */
3630 for (int i = 0; i < num_folios; i++) {
3631 struct folio *folio;
3633 ret = attach_eb_folio_to_filemap(eb, i, &existing_eb);
3635 ASSERT(existing_eb);
3640 * TODO: Special handling for a corner case where the order of
3641 * folios mismatch between the new eb and filemap.
3643 * This happens when:
3645 * - the new eb is using higher order folio
3647 * - the filemap is still using 0-order folios for the range
3648 * This can happen at the previous eb allocation, and we don't
3649 * have higher order folio for the call.
3651 * - the existing eb has already been freed
3653 * In this case, we have to free the existing folios first, and
3654 * re-allocate using the same order.
3655 * Thankfully this is not going to happen yet, as we're still
3656 * using 0-order folios.
3658 if (unlikely(ret == -EAGAIN)) {
3665 * Only after attach_eb_folio_to_filemap(), eb->folios[] is
3666 * reliable, as we may choose to reuse the existing page cache
3667 * and free the allocated page.
3669 folio = eb->folios[i];
3670 spin_lock(&mapping->private_lock);
3671 /* Should not fail, as we have preallocated the memory */
3672 ret = attach_extent_buffer_folio(eb, folio, prealloc);
3675 * To inform we have extra eb under allocation, so that
3676 * detach_extent_buffer_page() won't release the folio private
3677 * when the eb hasn't yet been inserted into radix tree.
3679 * The ref will be decreased when the eb released the page, in
3680 * detach_extent_buffer_page().
3681 * Thus needs no special handling in error path.
3683 btrfs_folio_inc_eb_refs(fs_info, folio);
3684 spin_unlock(&mapping->private_lock);
3686 WARN_ON(btrfs_page_test_dirty(fs_info, folio_page(folio, 0),
3687 eb->start, eb->len));
3690 * Check if the current page is physically contiguous with previous eb
3692 * At this stage, either we allocated a large folio, thus @i
3693 * would only be 0, or we fall back to per-page allocation.
3695 if (i && folio_page(eb->folios[i - 1], 0) + 1 != folio_page(folio, 0))
3696 page_contig = false;
3698 if (!btrfs_page_test_uptodate(fs_info, folio_page(folio, 0),
3699 eb->start, eb->len))
3703 * We can't unlock the pages just yet since the extent buffer
3704 * hasn't been properly inserted in the radix tree, this
3705 * opens a race with btree_release_folio which can free a page
3706 * while we are still filling in all pages for the buffer and
3711 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3712 /* All pages are physically contiguous, can skip cross page handling. */
3714 eb->addr = folio_address(eb->folios[0]) + offset_in_page(eb->start);
3716 ret = radix_tree_preload(GFP_NOFS);
3720 spin_lock(&fs_info->buffer_lock);
3721 ret = radix_tree_insert(&fs_info->buffer_radix,
3722 start >> fs_info->sectorsize_bits, eb);
3723 spin_unlock(&fs_info->buffer_lock);
3724 radix_tree_preload_end();
3725 if (ret == -EEXIST) {
3727 existing_eb = find_extent_buffer(fs_info, start);
3733 /* add one reference for the tree */
3734 check_buffer_tree_ref(eb);
3735 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3738 * Now it's safe to unlock the pages because any calls to
3739 * btree_release_folio will correctly detect that a page belongs to a
3740 * live buffer and won't free them prematurely.
3742 for (int i = 0; i < num_folios; i++)
3743 unlock_page(folio_page(eb->folios[i], 0));
3747 WARN_ON(!atomic_dec_and_test(&eb->refs));
3750 * Any attached folios need to be detached before we unlock them. This
3751 * is because when we're inserting our new folios into the mapping, and
3752 * then attaching our eb to that folio. If we fail to insert our folio
3753 * we'll lookup the folio for that index, and grab that EB. We do not
3754 * want that to grab this eb, as we're getting ready to free it. So we
3755 * have to detach it first and then unlock it.
3757 * We have to drop our reference and NULL it out here because in the
3758 * subpage case detaching does a btrfs_folio_dec_eb_refs() for our eb.
3759 * Below when we call btrfs_release_extent_buffer() we will call
3760 * detach_extent_buffer_folio() on our remaining pages in the !subpage
3761 * case. If we left eb->folios[i] populated in the subpage case we'd
3762 * double put our reference and be super sad.
3764 for (int i = 0; i < attached; i++) {
3765 ASSERT(eb->folios[i]);
3766 detach_extent_buffer_folio(eb, eb->folios[i]);
3767 unlock_page(folio_page(eb->folios[i], 0));
3768 folio_put(eb->folios[i]);
3769 eb->folios[i] = NULL;
3772 * Now all pages of that extent buffer is unmapped, set UNMAPPED flag,
3773 * so it can be cleaned up without utlizing page->mapping.
3775 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3777 btrfs_release_extent_buffer(eb);
3779 return ERR_PTR(ret);
3780 ASSERT(existing_eb);
3784 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3786 struct extent_buffer *eb =
3787 container_of(head, struct extent_buffer, rcu_head);
3789 __free_extent_buffer(eb);
3792 static int release_extent_buffer(struct extent_buffer *eb)
3793 __releases(&eb->refs_lock)
3795 lockdep_assert_held(&eb->refs_lock);
3797 WARN_ON(atomic_read(&eb->refs) == 0);
3798 if (atomic_dec_and_test(&eb->refs)) {
3799 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
3800 struct btrfs_fs_info *fs_info = eb->fs_info;
3802 spin_unlock(&eb->refs_lock);
3804 spin_lock(&fs_info->buffer_lock);
3805 radix_tree_delete(&fs_info->buffer_radix,
3806 eb->start >> fs_info->sectorsize_bits);
3807 spin_unlock(&fs_info->buffer_lock);
3809 spin_unlock(&eb->refs_lock);
3812 btrfs_leak_debug_del_eb(eb);
3813 /* Should be safe to release our pages at this point */
3814 btrfs_release_extent_buffer_pages(eb);
3815 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3816 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
3817 __free_extent_buffer(eb);
3821 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
3824 spin_unlock(&eb->refs_lock);
3829 void free_extent_buffer(struct extent_buffer *eb)
3835 refs = atomic_read(&eb->refs);
3837 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
3838 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
3841 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
3845 spin_lock(&eb->refs_lock);
3846 if (atomic_read(&eb->refs) == 2 &&
3847 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
3848 !extent_buffer_under_io(eb) &&
3849 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3850 atomic_dec(&eb->refs);
3853 * I know this is terrible, but it's temporary until we stop tracking
3854 * the uptodate bits and such for the extent buffers.
3856 release_extent_buffer(eb);
3859 void free_extent_buffer_stale(struct extent_buffer *eb)
3864 spin_lock(&eb->refs_lock);
3865 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
3867 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3868 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3869 atomic_dec(&eb->refs);
3870 release_extent_buffer(eb);
3873 static void btree_clear_folio_dirty(struct folio *folio)
3875 ASSERT(folio_test_dirty(folio));
3876 ASSERT(folio_test_locked(folio));
3877 folio_clear_dirty_for_io(folio);
3878 xa_lock_irq(&folio->mapping->i_pages);
3879 if (!folio_test_dirty(folio))
3880 __xa_clear_mark(&folio->mapping->i_pages,
3881 folio_index(folio), PAGECACHE_TAG_DIRTY);
3882 xa_unlock_irq(&folio->mapping->i_pages);
3885 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
3887 struct btrfs_fs_info *fs_info = eb->fs_info;
3888 struct folio *folio = eb->folios[0];
3891 /* btree_clear_folio_dirty() needs page locked. */
3893 last = btrfs_subpage_clear_and_test_dirty(fs_info, folio_page(folio, 0),
3894 eb->start, eb->len);
3896 btree_clear_folio_dirty(folio);
3897 folio_unlock(folio);
3898 WARN_ON(atomic_read(&eb->refs) == 0);
3901 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
3902 struct extent_buffer *eb)
3904 struct btrfs_fs_info *fs_info = eb->fs_info;
3907 btrfs_assert_tree_write_locked(eb);
3909 if (trans && btrfs_header_generation(eb) != trans->transid)
3913 * Instead of clearing the dirty flag off of the buffer, mark it as
3914 * EXTENT_BUFFER_ZONED_ZEROOUT. This allows us to preserve
3915 * write-ordering in zoned mode, without the need to later re-dirty
3916 * the extent_buffer.
3918 * The actual zeroout of the buffer will happen later in
3919 * btree_csum_one_bio.
3921 if (btrfs_is_zoned(fs_info)) {
3922 set_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags);
3926 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
3929 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
3930 fs_info->dirty_metadata_batch);
3932 if (eb->fs_info->nodesize < PAGE_SIZE)
3933 return clear_subpage_extent_buffer_dirty(eb);
3935 num_folios = num_extent_folios(eb);
3936 for (int i = 0; i < num_folios; i++) {
3937 struct folio *folio = eb->folios[i];
3939 if (!folio_test_dirty(folio))
3942 btree_clear_folio_dirty(folio);
3943 folio_unlock(folio);
3945 WARN_ON(atomic_read(&eb->refs) == 0);
3948 void set_extent_buffer_dirty(struct extent_buffer *eb)
3953 check_buffer_tree_ref(eb);
3955 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3957 num_folios = num_extent_folios(eb);
3958 WARN_ON(atomic_read(&eb->refs) == 0);
3959 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
3962 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
3965 * For subpage case, we can have other extent buffers in the
3966 * same page, and in clear_subpage_extent_buffer_dirty() we
3967 * have to clear page dirty without subpage lock held.
3968 * This can cause race where our page gets dirty cleared after
3971 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
3972 * its page for other reasons, we can use page lock to prevent
3976 lock_page(folio_page(eb->folios[0], 0));
3977 for (int i = 0; i < num_folios; i++)
3978 btrfs_page_set_dirty(eb->fs_info, folio_page(eb->folios[i], 0),
3979 eb->start, eb->len);
3981 unlock_page(folio_page(eb->folios[0], 0));
3982 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
3984 eb->fs_info->dirty_metadata_batch);
3986 #ifdef CONFIG_BTRFS_DEBUG
3987 for (int i = 0; i < num_folios; i++)
3988 ASSERT(folio_test_dirty(eb->folios[i]));
3992 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
3994 struct btrfs_fs_info *fs_info = eb->fs_info;
3995 int num_folios = num_extent_folios(eb);
3997 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3998 for (int i = 0; i < num_folios; i++) {
3999 struct folio *folio = eb->folios[i];
4005 * This is special handling for metadata subpage, as regular
4006 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4008 if (fs_info->nodesize >= PAGE_SIZE)
4009 folio_clear_uptodate(folio);
4011 btrfs_subpage_clear_uptodate(fs_info, folio_page(folio, 0),
4012 eb->start, eb->len);
4016 void set_extent_buffer_uptodate(struct extent_buffer *eb)
4018 struct btrfs_fs_info *fs_info = eb->fs_info;
4019 int num_folios = num_extent_folios(eb);
4021 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4022 for (int i = 0; i < num_folios; i++) {
4023 struct folio *folio = eb->folios[i];
4026 * This is special handling for metadata subpage, as regular
4027 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4029 if (fs_info->nodesize >= PAGE_SIZE)
4030 folio_mark_uptodate(folio);
4032 btrfs_subpage_set_uptodate(fs_info, folio_page(folio, 0),
4033 eb->start, eb->len);
4037 static void extent_buffer_read_end_io(struct btrfs_bio *bbio)
4039 struct extent_buffer *eb = bbio->private;
4040 struct btrfs_fs_info *fs_info = eb->fs_info;
4041 bool uptodate = !bbio->bio.bi_status;
4042 struct bvec_iter_all iter_all;
4043 struct bio_vec *bvec;
4046 eb->read_mirror = bbio->mirror_num;
4049 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
4053 set_extent_buffer_uptodate(eb);
4055 clear_extent_buffer_uptodate(eb);
4056 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4059 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
4060 u64 start = eb->start + bio_offset;
4061 struct page *page = bvec->bv_page;
4062 u32 len = bvec->bv_len;
4065 btrfs_page_set_uptodate(fs_info, page, start, len);
4067 btrfs_page_clear_uptodate(fs_info, page, start, len);
4072 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
4073 smp_mb__after_atomic();
4074 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
4075 free_extent_buffer(eb);
4077 bio_put(&bbio->bio);
4080 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
4081 struct btrfs_tree_parent_check *check)
4083 struct btrfs_bio *bbio;
4086 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4090 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
4091 * operation, which could potentially still be in flight. In this case
4092 * we simply want to return an error.
4094 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
4097 /* Someone else is already reading the buffer, just wait for it. */
4098 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
4101 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4102 eb->read_mirror = 0;
4103 check_buffer_tree_ref(eb);
4104 atomic_inc(&eb->refs);
4106 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
4107 REQ_OP_READ | REQ_META, eb->fs_info,
4108 extent_buffer_read_end_io, eb);
4109 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
4110 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
4111 bbio->file_offset = eb->start;
4112 memcpy(&bbio->parent_check, check, sizeof(*check));
4113 if (eb->fs_info->nodesize < PAGE_SIZE) {
4114 ret = bio_add_folio(&bbio->bio, eb->folios[0], eb->len,
4115 eb->start - folio_pos(eb->folios[0]));
4118 int num_folios = num_extent_folios(eb);
4120 for (int i = 0; i < num_folios; i++) {
4121 struct folio *folio = eb->folios[i];
4123 ret = bio_add_folio(&bbio->bio, folio, folio_size(folio), 0);
4127 btrfs_submit_bio(bbio, mirror_num);
4130 if (wait == WAIT_COMPLETE) {
4131 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
4132 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4139 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
4142 btrfs_warn(eb->fs_info,
4143 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
4144 eb->start, eb->len, start, len);
4145 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4151 * Check if the [start, start + len) range is valid before reading/writing
4153 * NOTE: @start and @len are offset inside the eb, not logical address.
4155 * Caller should not touch the dst/src memory if this function returns error.
4157 static inline int check_eb_range(const struct extent_buffer *eb,
4158 unsigned long start, unsigned long len)
4160 unsigned long offset;
4162 /* start, start + len should not go beyond eb->len nor overflow */
4163 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
4164 return report_eb_range(eb, start, len);
4169 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4170 unsigned long start, unsigned long len)
4176 char *dst = (char *)dstv;
4177 unsigned long i = get_eb_page_index(start);
4179 if (check_eb_range(eb, start, len)) {
4181 * Invalid range hit, reset the memory, so callers won't get
4182 * some random garbage for their uninitialized memory.
4184 memset(dstv, 0, len);
4189 memcpy(dstv, eb->addr + start, len);
4193 offset = get_eb_offset_in_page(eb, start);
4196 page = folio_page(eb->folios[i], 0);
4198 cur = min(len, (PAGE_SIZE - offset));
4199 kaddr = page_address(page);
4200 memcpy(dst, kaddr + offset, cur);
4209 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4211 unsigned long start, unsigned long len)
4217 char __user *dst = (char __user *)dstv;
4218 unsigned long i = get_eb_page_index(start);
4221 WARN_ON(start > eb->len);
4222 WARN_ON(start + len > eb->start + eb->len);
4225 if (copy_to_user_nofault(dstv, eb->addr + start, len))
4230 offset = get_eb_offset_in_page(eb, start);
4233 page = folio_page(eb->folios[i], 0);
4235 cur = min(len, (PAGE_SIZE - offset));
4236 kaddr = page_address(page);
4237 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4251 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4252 unsigned long start, unsigned long len)
4258 char *ptr = (char *)ptrv;
4259 unsigned long i = get_eb_page_index(start);
4262 if (check_eb_range(eb, start, len))
4266 return memcmp(ptrv, eb->addr + start, len);
4268 offset = get_eb_offset_in_page(eb, start);
4271 page = folio_page(eb->folios[i], 0);
4273 cur = min(len, (PAGE_SIZE - offset));
4275 kaddr = page_address(page);
4276 ret = memcmp(ptr, kaddr + offset, cur);
4289 * Check that the extent buffer is uptodate.
4291 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4292 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4294 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
4297 struct btrfs_fs_info *fs_info = eb->fs_info;
4300 * If we are using the commit root we could potentially clear a page
4301 * Uptodate while we're using the extent buffer that we've previously
4302 * looked up. We don't want to complain in this case, as the page was
4303 * valid before, we just didn't write it out. Instead we want to catch
4304 * the case where we didn't actually read the block properly, which
4305 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4307 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4310 if (fs_info->nodesize < PAGE_SIZE) {
4311 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, page,
4312 eb->start, eb->len)))
4313 btrfs_subpage_dump_bitmap(fs_info, page, eb->start, eb->len);
4315 WARN_ON(!PageUptodate(page));
4319 static void __write_extent_buffer(const struct extent_buffer *eb,
4320 const void *srcv, unsigned long start,
4321 unsigned long len, bool use_memmove)
4327 char *src = (char *)srcv;
4328 unsigned long i = get_eb_page_index(start);
4329 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4330 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4332 if (check_eb_range(eb, start, len))
4337 memmove(eb->addr + start, srcv, len);
4339 memcpy(eb->addr + start, srcv, len);
4343 offset = get_eb_offset_in_page(eb, start);
4346 page = folio_page(eb->folios[i], 0);
4348 assert_eb_page_uptodate(eb, page);
4350 cur = min(len, PAGE_SIZE - offset);
4351 kaddr = page_address(page);
4353 memmove(kaddr + offset, src, cur);
4355 memcpy(kaddr + offset, src, cur);
4364 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4365 unsigned long start, unsigned long len)
4367 return __write_extent_buffer(eb, srcv, start, len, false);
4370 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4371 unsigned long start, unsigned long len)
4373 unsigned long cur = start;
4376 memset(eb->addr + start, c, len);
4380 while (cur < start + len) {
4381 unsigned long index = get_eb_page_index(cur);
4382 unsigned int offset = get_eb_offset_in_page(eb, cur);
4383 unsigned int cur_len = min(start + len - cur, PAGE_SIZE - offset);
4384 struct page *page = folio_page(eb->folios[index], 0);
4386 assert_eb_page_uptodate(eb, page);
4387 memset_page(page, offset, c, cur_len);
4393 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4396 if (check_eb_range(eb, start, len))
4398 return memset_extent_buffer(eb, 0, start, len);
4401 void copy_extent_buffer_full(const struct extent_buffer *dst,
4402 const struct extent_buffer *src)
4404 unsigned long cur = 0;
4406 ASSERT(dst->len == src->len);
4408 while (cur < src->len) {
4409 unsigned long index = get_eb_page_index(cur);
4410 unsigned long offset = get_eb_offset_in_page(src, cur);
4411 unsigned long cur_len = min(src->len, PAGE_SIZE - offset);
4412 void *addr = folio_address(src->folios[index]) + offset;
4414 write_extent_buffer(dst, addr, cur, cur_len);
4420 void copy_extent_buffer(const struct extent_buffer *dst,
4421 const struct extent_buffer *src,
4422 unsigned long dst_offset, unsigned long src_offset,
4425 u64 dst_len = dst->len;
4430 unsigned long i = get_eb_page_index(dst_offset);
4432 if (check_eb_range(dst, dst_offset, len) ||
4433 check_eb_range(src, src_offset, len))
4436 WARN_ON(src->len != dst_len);
4438 offset = get_eb_offset_in_page(dst, dst_offset);
4441 page = folio_page(dst->folios[i], 0);
4442 assert_eb_page_uptodate(dst, page);
4444 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
4446 kaddr = page_address(page);
4447 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4457 * Calculate the page and offset of the byte containing the given bit number.
4459 * @eb: the extent buffer
4460 * @start: offset of the bitmap item in the extent buffer
4462 * @page_index: return index of the page in the extent buffer that contains
4463 * the given bit number
4464 * @page_offset: return offset into the page given by page_index
4466 * This helper hides the ugliness of finding the byte in an extent buffer which
4467 * contains a given bit.
4469 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4470 unsigned long start, unsigned long nr,
4471 unsigned long *page_index,
4472 size_t *page_offset)
4474 size_t byte_offset = BIT_BYTE(nr);
4478 * The byte we want is the offset of the extent buffer + the offset of
4479 * the bitmap item in the extent buffer + the offset of the byte in the
4482 offset = start + offset_in_page(eb->start) + byte_offset;
4484 *page_index = offset >> PAGE_SHIFT;
4485 *page_offset = offset_in_page(offset);
4489 * Determine whether a bit in a bitmap item is set.
4491 * @eb: the extent buffer
4492 * @start: offset of the bitmap item in the extent buffer
4493 * @nr: bit number to test
4495 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4503 eb_bitmap_offset(eb, start, nr, &i, &offset);
4504 page = folio_page(eb->folios[i], 0);
4505 assert_eb_page_uptodate(eb, page);
4506 kaddr = page_address(page);
4507 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4510 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4512 unsigned long index = get_eb_page_index(bytenr);
4514 if (check_eb_range(eb, bytenr, 1))
4516 return folio_address(eb->folios[index]) + get_eb_offset_in_page(eb, bytenr);
4520 * Set an area of a bitmap to 1.
4522 * @eb: the extent buffer
4523 * @start: offset of the bitmap item in the extent buffer
4524 * @pos: bit number of the first bit
4525 * @len: number of bits to set
4527 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4528 unsigned long pos, unsigned long len)
4530 unsigned int first_byte = start + BIT_BYTE(pos);
4531 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4532 const bool same_byte = (first_byte == last_byte);
4533 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4537 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4539 /* Handle the first byte. */
4540 kaddr = extent_buffer_get_byte(eb, first_byte);
4545 /* Handle the byte aligned part. */
4546 ASSERT(first_byte + 1 <= last_byte);
4547 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4549 /* Handle the last byte. */
4550 kaddr = extent_buffer_get_byte(eb, last_byte);
4551 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4556 * Clear an area of a bitmap.
4558 * @eb: the extent buffer
4559 * @start: offset of the bitmap item in the extent buffer
4560 * @pos: bit number of the first bit
4561 * @len: number of bits to clear
4563 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4564 unsigned long start, unsigned long pos,
4567 unsigned int first_byte = start + BIT_BYTE(pos);
4568 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4569 const bool same_byte = (first_byte == last_byte);
4570 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4574 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4576 /* Handle the first byte. */
4577 kaddr = extent_buffer_get_byte(eb, first_byte);
4582 /* Handle the byte aligned part. */
4583 ASSERT(first_byte + 1 <= last_byte);
4584 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4586 /* Handle the last byte. */
4587 kaddr = extent_buffer_get_byte(eb, last_byte);
4588 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4591 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4593 unsigned long distance = (src > dst) ? src - dst : dst - src;
4594 return distance < len;
4597 void memcpy_extent_buffer(const struct extent_buffer *dst,
4598 unsigned long dst_offset, unsigned long src_offset,
4601 unsigned long cur_off = 0;
4603 if (check_eb_range(dst, dst_offset, len) ||
4604 check_eb_range(dst, src_offset, len))
4608 const bool use_memmove = areas_overlap(src_offset, dst_offset, len);
4611 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4613 memcpy(dst->addr + dst_offset, dst->addr + src_offset, len);
4617 while (cur_off < len) {
4618 unsigned long cur_src = cur_off + src_offset;
4619 unsigned long pg_index = get_eb_page_index(cur_src);
4620 unsigned long pg_off = get_eb_offset_in_page(dst, cur_src);
4621 unsigned long cur_len = min(src_offset + len - cur_src,
4622 PAGE_SIZE - pg_off);
4623 void *src_addr = folio_address(dst->folios[pg_index]) + pg_off;
4624 const bool use_memmove = areas_overlap(src_offset + cur_off,
4625 dst_offset + cur_off, cur_len);
4627 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4633 void memmove_extent_buffer(const struct extent_buffer *dst,
4634 unsigned long dst_offset, unsigned long src_offset,
4637 unsigned long dst_end = dst_offset + len - 1;
4638 unsigned long src_end = src_offset + len - 1;
4640 if (check_eb_range(dst, dst_offset, len) ||
4641 check_eb_range(dst, src_offset, len))
4644 if (dst_offset < src_offset) {
4645 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4650 memmove(dst->addr + dst_offset, dst->addr + src_offset, len);
4655 unsigned long src_i;
4657 size_t dst_off_in_page;
4658 size_t src_off_in_page;
4662 src_i = get_eb_page_index(src_end);
4664 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
4665 src_off_in_page = get_eb_offset_in_page(dst, src_end);
4667 cur = min_t(unsigned long, len, src_off_in_page + 1);
4668 cur = min(cur, dst_off_in_page + 1);
4670 src_addr = folio_address(dst->folios[src_i]) + src_off_in_page -
4672 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4675 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4684 #define GANG_LOOKUP_SIZE 16
4685 static struct extent_buffer *get_next_extent_buffer(
4686 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4688 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4689 struct extent_buffer *found = NULL;
4690 u64 page_start = page_offset(page);
4691 u64 cur = page_start;
4693 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4694 lockdep_assert_held(&fs_info->buffer_lock);
4696 while (cur < page_start + PAGE_SIZE) {
4700 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4701 (void **)gang, cur >> fs_info->sectorsize_bits,
4702 min_t(unsigned int, GANG_LOOKUP_SIZE,
4703 PAGE_SIZE / fs_info->nodesize));
4706 for (i = 0; i < ret; i++) {
4707 /* Already beyond page end */
4708 if (gang[i]->start >= page_start + PAGE_SIZE)
4711 if (gang[i]->start >= bytenr) {
4716 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4722 static int try_release_subpage_extent_buffer(struct page *page)
4724 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4725 u64 cur = page_offset(page);
4726 const u64 end = page_offset(page) + PAGE_SIZE;
4730 struct extent_buffer *eb = NULL;
4733 * Unlike try_release_extent_buffer() which uses folio private
4734 * to grab buffer, for subpage case we rely on radix tree, thus
4735 * we need to ensure radix tree consistency.
4737 * We also want an atomic snapshot of the radix tree, thus go
4738 * with spinlock rather than RCU.
4740 spin_lock(&fs_info->buffer_lock);
4741 eb = get_next_extent_buffer(fs_info, page, cur);
4743 /* No more eb in the page range after or at cur */
4744 spin_unlock(&fs_info->buffer_lock);
4747 cur = eb->start + eb->len;
4750 * The same as try_release_extent_buffer(), to ensure the eb
4751 * won't disappear out from under us.
4753 spin_lock(&eb->refs_lock);
4754 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4755 spin_unlock(&eb->refs_lock);
4756 spin_unlock(&fs_info->buffer_lock);
4759 spin_unlock(&fs_info->buffer_lock);
4762 * If tree ref isn't set then we know the ref on this eb is a
4763 * real ref, so just return, this eb will likely be freed soon
4766 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4767 spin_unlock(&eb->refs_lock);
4772 * Here we don't care about the return value, we will always
4773 * check the folio private at the end. And
4774 * release_extent_buffer() will release the refs_lock.
4776 release_extent_buffer(eb);
4779 * Finally to check if we have cleared folio private, as if we have
4780 * released all ebs in the page, the folio private should be cleared now.
4782 spin_lock(&page->mapping->private_lock);
4783 if (!folio_test_private(page_folio(page)))
4787 spin_unlock(&page->mapping->private_lock);
4792 int try_release_extent_buffer(struct page *page)
4794 struct folio *folio = page_folio(page);
4795 struct extent_buffer *eb;
4797 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4798 return try_release_subpage_extent_buffer(page);
4801 * We need to make sure nobody is changing folio private, as we rely on
4802 * folio private as the pointer to extent buffer.
4804 spin_lock(&page->mapping->private_lock);
4805 if (!folio_test_private(folio)) {
4806 spin_unlock(&page->mapping->private_lock);
4810 eb = folio_get_private(folio);
4814 * This is a little awful but should be ok, we need to make sure that
4815 * the eb doesn't disappear out from under us while we're looking at
4818 spin_lock(&eb->refs_lock);
4819 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4820 spin_unlock(&eb->refs_lock);
4821 spin_unlock(&page->mapping->private_lock);
4824 spin_unlock(&page->mapping->private_lock);
4827 * If tree ref isn't set then we know the ref on this eb is a real ref,
4828 * so just return, this page will likely be freed soon anyway.
4830 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4831 spin_unlock(&eb->refs_lock);
4835 return release_extent_buffer(eb);
4839 * Attempt to readahead a child block.
4841 * @fs_info: the fs_info
4842 * @bytenr: bytenr to read
4843 * @owner_root: objectid of the root that owns this eb
4844 * @gen: generation for the uptodate check, can be 0
4845 * @level: level for the eb
4847 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
4848 * normal uptodate check of the eb, without checking the generation. If we have
4849 * to read the block we will not block on anything.
4851 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
4852 u64 bytenr, u64 owner_root, u64 gen, int level)
4854 struct btrfs_tree_parent_check check = {
4859 struct extent_buffer *eb;
4862 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
4866 if (btrfs_buffer_uptodate(eb, gen, 1)) {
4867 free_extent_buffer(eb);
4871 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
4873 free_extent_buffer_stale(eb);
4875 free_extent_buffer(eb);
4879 * Readahead a node's child block.
4881 * @node: parent node we're reading from
4882 * @slot: slot in the parent node for the child we want to read
4884 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
4885 * the slot in the node provided.
4887 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
4889 btrfs_readahead_tree_block(node->fs_info,
4890 btrfs_node_blockptr(node, slot),
4891 btrfs_header_owner(node),
4892 btrfs_node_ptr_generation(node, slot),
4893 btrfs_header_level(node) - 1);