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"
24 #include "check-integrity.h"
26 #include "rcu-string.h"
31 #include "block-group.h"
32 #include "compression.h"
34 #include "accessors.h"
35 #include "file-item.h"
37 #include "dev-replace.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 {
101 enum btrfs_compression_type compress_type;
102 u32 len_to_stripe_boundary;
103 u32 len_to_oe_boundary;
104 btrfs_bio_end_io_t end_io_func;
107 * This is for metadata read, to provide the extra needed verification
108 * info. This has to be provided for submit_one_bio(), as
109 * submit_one_bio() can submit a bio if it ends at stripe boundary. If
110 * no such parent_check is provided, the metadata can hit false alert at
113 struct btrfs_tree_parent_check *parent_check;
116 * Tell writepage not to lock the state bits for this range, it still
117 * does the unlocking.
121 /* Tell the submit_bio code to use REQ_SYNC */
125 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
129 struct btrfs_inode *inode;
136 bv = bio_first_bvec_all(bio);
137 inode = BTRFS_I(bv->bv_page->mapping->host);
138 mirror_num = bio_ctrl->mirror_num;
140 /* Caller should ensure the bio has at least some range added */
141 ASSERT(bio->bi_iter.bi_size);
143 btrfs_bio(bio)->file_offset = page_offset(bv->bv_page) + bv->bv_offset;
145 if (!is_data_inode(&inode->vfs_inode)) {
146 if (btrfs_op(bio) != BTRFS_MAP_WRITE) {
148 * For metadata read, we should have the parent_check,
149 * and copy it to bbio for metadata verification.
151 ASSERT(bio_ctrl->parent_check);
152 memcpy(&btrfs_bio(bio)->parent_check,
153 bio_ctrl->parent_check,
154 sizeof(struct btrfs_tree_parent_check));
156 btrfs_submit_metadata_bio(inode, bio, mirror_num);
157 } else if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
158 btrfs_submit_data_write_bio(inode, bio, mirror_num);
160 btrfs_submit_data_read_bio(inode, bio, mirror_num,
161 bio_ctrl->compress_type);
164 /* The bio is owned by the end_io handler now */
165 bio_ctrl->bio = NULL;
169 * Submit or fail the current bio in the bio_ctrl structure.
171 static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret)
173 struct bio *bio = bio_ctrl->bio;
180 btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret));
181 /* The bio is owned by the end_io handler now */
182 bio_ctrl->bio = NULL;
184 submit_one_bio(bio_ctrl);
188 int __init extent_buffer_init_cachep(void)
190 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
191 sizeof(struct extent_buffer), 0,
192 SLAB_MEM_SPREAD, NULL);
193 if (!extent_buffer_cache)
199 void __cold extent_buffer_free_cachep(void)
202 * Make sure all delayed rcu free are flushed before we
206 kmem_cache_destroy(extent_buffer_cache);
209 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
211 unsigned long index = start >> PAGE_SHIFT;
212 unsigned long end_index = end >> PAGE_SHIFT;
215 while (index <= end_index) {
216 page = find_get_page(inode->i_mapping, index);
217 BUG_ON(!page); /* Pages should be in the extent_io_tree */
218 clear_page_dirty_for_io(page);
224 void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end)
226 struct address_space *mapping = inode->i_mapping;
227 unsigned long index = start >> PAGE_SHIFT;
228 unsigned long end_index = end >> PAGE_SHIFT;
231 while (index <= end_index) {
232 folio = filemap_get_folio(mapping, index);
233 filemap_dirty_folio(mapping, folio);
234 folio_account_redirty(folio);
235 index += folio_nr_pages(folio);
241 * Process one page for __process_pages_contig().
243 * Return >0 if we hit @page == @locked_page.
244 * Return 0 if we updated the page status.
245 * Return -EGAIN if the we need to try again.
246 * (For PAGE_LOCK case but got dirty page or page not belong to mapping)
248 static int process_one_page(struct btrfs_fs_info *fs_info,
249 struct address_space *mapping,
250 struct page *page, struct page *locked_page,
251 unsigned long page_ops, u64 start, u64 end)
255 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
256 len = end + 1 - start;
258 if (page_ops & PAGE_SET_ORDERED)
259 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
260 if (page_ops & PAGE_SET_ERROR)
261 btrfs_page_clamp_set_error(fs_info, page, start, len);
262 if (page_ops & PAGE_START_WRITEBACK) {
263 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
264 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
266 if (page_ops & PAGE_END_WRITEBACK)
267 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
269 if (page == locked_page)
272 if (page_ops & PAGE_LOCK) {
275 ret = btrfs_page_start_writer_lock(fs_info, page, start, len);
278 if (!PageDirty(page) || page->mapping != mapping) {
279 btrfs_page_end_writer_lock(fs_info, page, start, len);
283 if (page_ops & PAGE_UNLOCK)
284 btrfs_page_end_writer_lock(fs_info, page, start, len);
288 static int __process_pages_contig(struct address_space *mapping,
289 struct page *locked_page,
290 u64 start, u64 end, unsigned long page_ops,
293 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
294 pgoff_t start_index = start >> PAGE_SHIFT;
295 pgoff_t end_index = end >> PAGE_SHIFT;
296 pgoff_t index = start_index;
297 unsigned long pages_processed = 0;
298 struct folio_batch fbatch;
302 if (page_ops & PAGE_LOCK) {
303 ASSERT(page_ops == PAGE_LOCK);
304 ASSERT(processed_end && *processed_end == start);
307 if ((page_ops & PAGE_SET_ERROR) && start_index <= end_index)
308 mapping_set_error(mapping, -EIO);
310 folio_batch_init(&fbatch);
311 while (index <= end_index) {
314 found_folios = filemap_get_folios_contig(mapping, &index,
317 if (found_folios == 0) {
319 * Only if we're going to lock these pages, we can find
322 ASSERT(page_ops & PAGE_LOCK);
327 for (i = 0; i < found_folios; i++) {
329 struct folio *folio = fbatch.folios[i];
330 process_ret = process_one_page(fs_info, mapping,
331 &folio->page, locked_page, page_ops,
333 if (process_ret < 0) {
335 folio_batch_release(&fbatch);
338 pages_processed += folio_nr_pages(folio);
340 folio_batch_release(&fbatch);
344 if (err && processed_end) {
346 * Update @processed_end. I know this is awful since it has
347 * two different return value patterns (inclusive vs exclusive).
349 * But the exclusive pattern is necessary if @start is 0, or we
350 * underflow and check against processed_end won't work as
354 *processed_end = min(end,
355 ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1);
357 *processed_end = start;
362 static noinline void __unlock_for_delalloc(struct inode *inode,
363 struct page *locked_page,
366 unsigned long index = start >> PAGE_SHIFT;
367 unsigned long end_index = end >> PAGE_SHIFT;
370 if (index == locked_page->index && end_index == index)
373 __process_pages_contig(inode->i_mapping, locked_page, start, end,
377 static noinline int lock_delalloc_pages(struct inode *inode,
378 struct page *locked_page,
382 unsigned long index = delalloc_start >> PAGE_SHIFT;
383 unsigned long end_index = delalloc_end >> PAGE_SHIFT;
384 u64 processed_end = delalloc_start;
388 if (index == locked_page->index && index == end_index)
391 ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start,
392 delalloc_end, PAGE_LOCK, &processed_end);
393 if (ret == -EAGAIN && processed_end > delalloc_start)
394 __unlock_for_delalloc(inode, locked_page, delalloc_start,
400 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
401 * more than @max_bytes.
403 * @start: The original start bytenr to search.
404 * Will store the extent range start bytenr.
405 * @end: The original end bytenr of the search range
406 * Will store the extent range end bytenr.
408 * Return true if we find a delalloc range which starts inside the original
409 * range, and @start/@end will store the delalloc range start/end.
411 * Return false if we can't find any delalloc range which starts inside the
412 * original range, and @start/@end will be the non-delalloc range start/end.
415 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
416 struct page *locked_page, u64 *start,
419 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
420 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
421 const u64 orig_start = *start;
422 const u64 orig_end = *end;
423 /* The sanity tests may not set a valid fs_info. */
424 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
428 struct extent_state *cached_state = NULL;
432 /* Caller should pass a valid @end to indicate the search range end */
433 ASSERT(orig_end > orig_start);
435 /* The range should at least cover part of the page */
436 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
437 orig_end <= page_offset(locked_page)));
439 /* step one, find a bunch of delalloc bytes starting at start */
440 delalloc_start = *start;
442 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
443 max_bytes, &cached_state);
444 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
445 *start = delalloc_start;
447 /* @delalloc_end can be -1, never go beyond @orig_end */
448 *end = min(delalloc_end, orig_end);
449 free_extent_state(cached_state);
454 * start comes from the offset of locked_page. We have to lock
455 * pages in order, so we can't process delalloc bytes before
458 if (delalloc_start < *start)
459 delalloc_start = *start;
462 * make sure to limit the number of pages we try to lock down
464 if (delalloc_end + 1 - delalloc_start > max_bytes)
465 delalloc_end = delalloc_start + max_bytes - 1;
467 /* step two, lock all the pages after the page that has start */
468 ret = lock_delalloc_pages(inode, locked_page,
469 delalloc_start, delalloc_end);
470 ASSERT(!ret || ret == -EAGAIN);
471 if (ret == -EAGAIN) {
472 /* some of the pages are gone, lets avoid looping by
473 * shortening the size of the delalloc range we're searching
475 free_extent_state(cached_state);
478 max_bytes = PAGE_SIZE;
487 /* step three, lock the state bits for the whole range */
488 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
490 /* then test to make sure it is all still delalloc */
491 ret = test_range_bit(tree, delalloc_start, delalloc_end,
492 EXTENT_DELALLOC, 1, cached_state);
494 unlock_extent(tree, delalloc_start, delalloc_end,
496 __unlock_for_delalloc(inode, locked_page,
497 delalloc_start, delalloc_end);
501 free_extent_state(cached_state);
502 *start = delalloc_start;
508 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
509 struct page *locked_page,
510 u32 clear_bits, unsigned long page_ops)
512 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
514 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
515 start, end, page_ops, NULL);
518 static int insert_failrec(struct btrfs_inode *inode,
519 struct io_failure_record *failrec)
521 struct rb_node *exist;
523 spin_lock(&inode->io_failure_lock);
524 exist = rb_simple_insert(&inode->io_failure_tree, failrec->bytenr,
526 spin_unlock(&inode->io_failure_lock);
528 return (exist == NULL) ? 0 : -EEXIST;
531 static struct io_failure_record *get_failrec(struct btrfs_inode *inode, u64 start)
533 struct rb_node *node;
534 struct io_failure_record *failrec = ERR_PTR(-ENOENT);
536 spin_lock(&inode->io_failure_lock);
537 node = rb_simple_search(&inode->io_failure_tree, start);
539 failrec = rb_entry(node, struct io_failure_record, rb_node);
540 spin_unlock(&inode->io_failure_lock);
544 static void free_io_failure(struct btrfs_inode *inode,
545 struct io_failure_record *rec)
547 spin_lock(&inode->io_failure_lock);
548 rb_erase(&rec->rb_node, &inode->io_failure_tree);
549 spin_unlock(&inode->io_failure_lock);
554 static int next_mirror(const struct io_failure_record *failrec, int cur_mirror)
556 if (cur_mirror == failrec->num_copies)
557 return cur_mirror + 1 - failrec->num_copies;
558 return cur_mirror + 1;
561 static int prev_mirror(const struct io_failure_record *failrec, int cur_mirror)
564 return failrec->num_copies;
565 return cur_mirror - 1;
569 * each time an IO finishes, we do a fast check in the IO failure tree
570 * to see if we need to process or clean up an io_failure_record
572 int btrfs_clean_io_failure(struct btrfs_inode *inode, u64 start,
573 struct page *page, unsigned int pg_offset)
575 struct btrfs_fs_info *fs_info = inode->root->fs_info;
576 struct extent_io_tree *io_tree = &inode->io_tree;
577 u64 ino = btrfs_ino(inode);
578 u64 locked_start, locked_end;
579 struct io_failure_record *failrec;
583 failrec = get_failrec(inode, start);
587 BUG_ON(!failrec->this_mirror);
589 if (sb_rdonly(fs_info->sb))
592 ret = find_first_extent_bit(io_tree, failrec->bytenr, &locked_start,
593 &locked_end, EXTENT_LOCKED, NULL);
594 if (ret || locked_start > failrec->bytenr ||
595 locked_end < failrec->bytenr + failrec->len - 1)
598 mirror = failrec->this_mirror;
600 mirror = prev_mirror(failrec, mirror);
601 btrfs_repair_io_failure(fs_info, ino, start, failrec->len,
602 failrec->logical, page, pg_offset, mirror);
603 } while (mirror != failrec->failed_mirror);
606 free_io_failure(inode, failrec);
613 * - under ordered extent
614 * - the inode is freeing
616 void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end)
618 struct io_failure_record *failrec;
619 struct rb_node *node, *next;
621 if (RB_EMPTY_ROOT(&inode->io_failure_tree))
624 spin_lock(&inode->io_failure_lock);
625 node = rb_simple_search_first(&inode->io_failure_tree, start);
627 failrec = rb_entry(node, struct io_failure_record, rb_node);
628 if (failrec->bytenr > end)
631 next = rb_next(node);
632 rb_erase(&failrec->rb_node, &inode->io_failure_tree);
637 spin_unlock(&inode->io_failure_lock);
640 static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode,
641 struct btrfs_bio *bbio,
642 unsigned int bio_offset)
644 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
645 u64 start = bbio->file_offset + bio_offset;
646 struct io_failure_record *failrec;
647 const u32 sectorsize = fs_info->sectorsize;
650 failrec = get_failrec(BTRFS_I(inode), start);
651 if (!IS_ERR(failrec)) {
653 "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu",
654 failrec->logical, failrec->bytenr, failrec->len);
656 * when data can be on disk more than twice, add to failrec here
657 * (e.g. with a list for failed_mirror) to make
658 * clean_io_failure() clean all those errors at once.
660 ASSERT(failrec->this_mirror == bbio->mirror_num);
661 ASSERT(failrec->len == fs_info->sectorsize);
665 failrec = kzalloc(sizeof(*failrec), GFP_NOFS);
667 return ERR_PTR(-ENOMEM);
669 RB_CLEAR_NODE(&failrec->rb_node);
670 failrec->bytenr = start;
671 failrec->len = sectorsize;
672 failrec->failed_mirror = bbio->mirror_num;
673 failrec->this_mirror = bbio->mirror_num;
674 failrec->logical = (bbio->iter.bi_sector << SECTOR_SHIFT) + bio_offset;
677 "new io failure record logical %llu start %llu",
678 failrec->logical, start);
680 failrec->num_copies = btrfs_num_copies(fs_info, failrec->logical, sectorsize);
681 if (failrec->num_copies == 1) {
683 * We only have a single copy of the data, so don't bother with
684 * all the retry and error correction code that follows. No
685 * matter what the error is, it is very likely to persist.
688 "cannot repair logical %llu num_copies %d",
689 failrec->logical, failrec->num_copies);
691 return ERR_PTR(-EIO);
694 /* Set the bits in the private failure tree */
695 ret = insert_failrec(BTRFS_I(inode), failrec);
704 int btrfs_repair_one_sector(struct btrfs_inode *inode, struct btrfs_bio *failed_bbio,
705 u32 bio_offset, struct page *page, unsigned int pgoff,
706 bool submit_buffered)
708 u64 start = failed_bbio->file_offset + bio_offset;
709 struct io_failure_record *failrec;
710 struct btrfs_fs_info *fs_info = inode->root->fs_info;
711 struct bio *failed_bio = &failed_bbio->bio;
712 const int icsum = bio_offset >> fs_info->sectorsize_bits;
713 struct bio *repair_bio;
714 struct btrfs_bio *repair_bbio;
717 "repair read error: read error at %llu", start);
719 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
721 failrec = btrfs_get_io_failure_record(&inode->vfs_inode, failed_bbio, bio_offset);
723 return PTR_ERR(failrec);
726 * There are two premises:
727 * a) deliver good data to the caller
728 * b) correct the bad sectors on disk
730 * Since we're only doing repair for one sector, we only need to get
731 * a good copy of the failed sector and if we succeed, we have setup
732 * everything for btrfs_repair_io_failure to do the rest for us.
734 failrec->this_mirror = next_mirror(failrec, failrec->this_mirror);
735 if (failrec->this_mirror == failrec->failed_mirror) {
737 "failed to repair num_copies %d this_mirror %d failed_mirror %d",
738 failrec->num_copies, failrec->this_mirror, failrec->failed_mirror);
739 free_io_failure(inode, failrec);
743 repair_bio = btrfs_bio_alloc(1, REQ_OP_READ, failed_bbio->end_io,
744 failed_bbio->private);
745 repair_bbio = btrfs_bio(repair_bio);
746 repair_bbio->file_offset = start;
747 repair_bio->bi_iter.bi_sector = failrec->logical >> 9;
749 if (failed_bbio->csum) {
750 const u32 csum_size = fs_info->csum_size;
752 repair_bbio->csum = repair_bbio->csum_inline;
753 memcpy(repair_bbio->csum,
754 failed_bbio->csum + csum_size * icsum, csum_size);
757 bio_add_page(repair_bio, page, failrec->len, pgoff);
758 repair_bbio->iter = repair_bio->bi_iter;
761 "repair read error: submitting new read to mirror %d",
762 failrec->this_mirror);
765 * At this point we have a bio, so any errors from bio submission will
766 * be handled by the endio on the repair_bio, so we can't return an
770 btrfs_submit_data_read_bio(inode, repair_bio,
771 failrec->this_mirror, 0);
773 btrfs_submit_dio_repair_bio(inode, repair_bio, failrec->this_mirror);
778 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
780 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
782 ASSERT(page_offset(page) <= start &&
783 start + len <= page_offset(page) + PAGE_SIZE);
786 if (fsverity_active(page->mapping->host) &&
788 !PageUptodate(page) &&
789 start < i_size_read(page->mapping->host) &&
790 !fsverity_verify_page(page)) {
791 btrfs_page_set_error(fs_info, page, start, len);
793 btrfs_page_set_uptodate(fs_info, page, start, len);
796 btrfs_page_clear_uptodate(fs_info, page, start, len);
797 btrfs_page_set_error(fs_info, page, start, len);
800 if (!btrfs_is_subpage(fs_info, page))
803 btrfs_subpage_end_reader(fs_info, page, start, len);
806 static void end_sector_io(struct page *page, u64 offset, bool uptodate)
808 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
809 const u32 sectorsize = inode->root->fs_info->sectorsize;
811 end_page_read(page, uptodate, offset, sectorsize);
812 unlock_extent(&inode->io_tree, offset, offset + sectorsize - 1, NULL);
815 static void submit_data_read_repair(struct inode *inode,
816 struct btrfs_bio *failed_bbio,
817 u32 bio_offset, const struct bio_vec *bvec,
818 unsigned int error_bitmap)
820 const unsigned int pgoff = bvec->bv_offset;
821 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
822 struct page *page = bvec->bv_page;
823 const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset;
824 const u64 end = start + bvec->bv_len - 1;
825 const u32 sectorsize = fs_info->sectorsize;
826 const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits;
829 BUG_ON(bio_op(&failed_bbio->bio) == REQ_OP_WRITE);
831 /* This repair is only for data */
832 ASSERT(is_data_inode(inode));
834 /* We're here because we had some read errors or csum mismatch */
835 ASSERT(error_bitmap);
838 * We only get called on buffered IO, thus page must be mapped and bio
839 * must not be cloned.
841 ASSERT(page->mapping && !bio_flagged(&failed_bbio->bio, BIO_CLONED));
843 /* Iterate through all the sectors in the range */
844 for (i = 0; i < nr_bits; i++) {
845 const unsigned int offset = i * sectorsize;
846 bool uptodate = false;
849 if (!(error_bitmap & (1U << i))) {
851 * This sector has no error, just end the page read
852 * and unlock the range.
858 ret = btrfs_repair_one_sector(BTRFS_I(inode), failed_bbio,
859 bio_offset + offset, page, pgoff + offset,
863 * We have submitted the read repair, the page release
864 * will be handled by the endio function of the
865 * submitted repair bio.
866 * Thus we don't need to do any thing here.
871 * Continue on failed repair, otherwise the remaining sectors
872 * will not be properly unlocked.
875 end_sector_io(page, start + offset, uptodate);
879 /* lots and lots of room for performance fixes in the end_bio funcs */
881 void end_extent_writepage(struct page *page, int err, u64 start, u64 end)
883 struct btrfs_inode *inode;
884 const bool uptodate = (err == 0);
887 ASSERT(page && page->mapping);
888 inode = BTRFS_I(page->mapping->host);
889 btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate);
892 const struct btrfs_fs_info *fs_info = inode->root->fs_info;
895 ASSERT(end + 1 - start <= U32_MAX);
896 len = end + 1 - start;
898 btrfs_page_clear_uptodate(fs_info, page, start, len);
899 btrfs_page_set_error(fs_info, page, start, len);
900 ret = err < 0 ? err : -EIO;
901 mapping_set_error(page->mapping, ret);
906 * after a writepage IO is done, we need to:
907 * clear the uptodate bits on error
908 * clear the writeback bits in the extent tree for this IO
909 * end_page_writeback if the page has no more pending IO
911 * Scheduling is not allowed, so the extent state tree is expected
912 * to have one and only one object corresponding to this IO.
914 static void end_bio_extent_writepage(struct btrfs_bio *bbio)
916 struct bio *bio = &bbio->bio;
917 int error = blk_status_to_errno(bio->bi_status);
918 struct bio_vec *bvec;
921 struct bvec_iter_all iter_all;
922 bool first_bvec = true;
924 ASSERT(!bio_flagged(bio, BIO_CLONED));
925 bio_for_each_segment_all(bvec, bio, iter_all) {
926 struct page *page = bvec->bv_page;
927 struct inode *inode = page->mapping->host;
928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
929 const u32 sectorsize = fs_info->sectorsize;
931 /* Our read/write should always be sector aligned. */
932 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
934 "partial page write in btrfs with offset %u and length %u",
935 bvec->bv_offset, bvec->bv_len);
936 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
938 "incomplete page write with offset %u and length %u",
939 bvec->bv_offset, bvec->bv_len);
941 start = page_offset(page) + bvec->bv_offset;
942 end = start + bvec->bv_len - 1;
945 btrfs_record_physical_zoned(inode, start, bio);
949 end_extent_writepage(page, error, start, end);
951 btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len);
958 * Record previously processed extent range
960 * For endio_readpage_release_extent() to handle a full extent range, reducing
961 * the extent io operations.
963 struct processed_extent {
964 struct btrfs_inode *inode;
965 /* Start of the range in @inode */
967 /* End of the range in @inode */
973 * Try to release processed extent range
975 * May not release the extent range right now if the current range is
976 * contiguous to processed extent.
978 * Will release processed extent when any of @inode, @uptodate, the range is
979 * no longer contiguous to the processed range.
981 * Passing @inode == NULL will force processed extent to be released.
983 static void endio_readpage_release_extent(struct processed_extent *processed,
984 struct btrfs_inode *inode, u64 start, u64 end,
987 struct extent_state *cached = NULL;
988 struct extent_io_tree *tree;
990 /* The first extent, initialize @processed */
991 if (!processed->inode)
995 * Contiguous to processed extent, just uptodate the end.
997 * Several things to notice:
999 * - bio can be merged as long as on-disk bytenr is contiguous
1000 * This means we can have page belonging to other inodes, thus need to
1001 * check if the inode still matches.
1002 * - bvec can contain range beyond current page for multi-page bvec
1003 * Thus we need to do processed->end + 1 >= start check
1005 if (processed->inode == inode && processed->uptodate == uptodate &&
1006 processed->end + 1 >= start && end >= processed->end) {
1007 processed->end = end;
1011 tree = &processed->inode->io_tree;
1013 * Now we don't have range contiguous to the processed range, release
1014 * the processed range now.
1016 unlock_extent(tree, processed->start, processed->end, &cached);
1019 /* Update processed to current range */
1020 processed->inode = inode;
1021 processed->start = start;
1022 processed->end = end;
1023 processed->uptodate = uptodate;
1026 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
1028 ASSERT(PageLocked(page));
1029 if (!btrfs_is_subpage(fs_info, page))
1032 ASSERT(PagePrivate(page));
1033 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
1037 * Find extent buffer for a givne bytenr.
1039 * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking
1042 static struct extent_buffer *find_extent_buffer_readpage(
1043 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
1045 struct extent_buffer *eb;
1048 * For regular sectorsize, we can use page->private to grab extent
1051 if (fs_info->nodesize >= PAGE_SIZE) {
1052 ASSERT(PagePrivate(page) && page->private);
1053 return (struct extent_buffer *)page->private;
1056 /* For subpage case, we need to lookup buffer radix tree */
1058 eb = radix_tree_lookup(&fs_info->buffer_radix,
1059 bytenr >> fs_info->sectorsize_bits);
1066 * after a readpage IO is done, we need to:
1067 * clear the uptodate bits on error
1068 * set the uptodate bits if things worked
1069 * set the page up to date if all extents in the tree are uptodate
1070 * clear the lock bit in the extent tree
1071 * unlock the page if there are no other extents locked for it
1073 * Scheduling is not allowed, so the extent state tree is expected
1074 * to have one and only one object corresponding to this IO.
1076 static void end_bio_extent_readpage(struct btrfs_bio *bbio)
1078 struct bio *bio = &bbio->bio;
1079 struct bio_vec *bvec;
1080 struct processed_extent processed = { 0 };
1082 * The offset to the beginning of a bio, since one bio can never be
1083 * larger than UINT_MAX, u32 here is enough.
1087 struct bvec_iter_all iter_all;
1089 ASSERT(!bio_flagged(bio, BIO_CLONED));
1090 bio_for_each_segment_all(bvec, bio, iter_all) {
1091 bool uptodate = !bio->bi_status;
1092 struct page *page = bvec->bv_page;
1093 struct inode *inode = page->mapping->host;
1094 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1095 const u32 sectorsize = fs_info->sectorsize;
1096 unsigned int error_bitmap = (unsigned int)-1;
1097 bool repair = false;
1102 btrfs_debug(fs_info,
1103 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
1104 bio->bi_iter.bi_sector, bio->bi_status,
1108 * We always issue full-sector reads, but if some block in a
1109 * page fails to read, blk_update_request() will advance
1110 * bv_offset and adjust bv_len to compensate. Print a warning
1111 * for unaligned offsets, and an error if they don't add up to
1114 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
1116 "partial page read in btrfs with offset %u and length %u",
1117 bvec->bv_offset, bvec->bv_len);
1118 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
1121 "incomplete page read with offset %u and length %u",
1122 bvec->bv_offset, bvec->bv_len);
1124 start = page_offset(page) + bvec->bv_offset;
1125 end = start + bvec->bv_len - 1;
1128 mirror = bbio->mirror_num;
1129 if (likely(uptodate)) {
1130 if (is_data_inode(inode)) {
1131 error_bitmap = btrfs_verify_data_csum(bbio,
1132 bio_offset, page, start, end);
1136 if (btrfs_validate_metadata_buffer(bbio,
1137 page, start, end, mirror))
1142 if (likely(uptodate)) {
1143 loff_t i_size = i_size_read(inode);
1144 pgoff_t end_index = i_size >> PAGE_SHIFT;
1146 btrfs_clean_io_failure(BTRFS_I(inode), start, page, 0);
1149 * Zero out the remaining part if this range straddles
1152 * Here we should only zero the range inside the bvec,
1153 * not touch anything else.
1155 * NOTE: i_size is exclusive while end is inclusive.
1157 if (page->index == end_index && i_size <= end) {
1158 u32 zero_start = max(offset_in_page(i_size),
1159 offset_in_page(start));
1161 zero_user_segment(page, zero_start,
1162 offset_in_page(end) + 1);
1164 } else if (is_data_inode(inode)) {
1166 * Only try to repair bios that actually made it to a
1167 * device. If the bio failed to be submitted mirror
1168 * is 0 and we need to fail it without retrying.
1170 * This also includes the high level bios for compressed
1171 * extents - these never make it to a device and repair
1172 * is already handled on the lower compressed bio.
1177 struct extent_buffer *eb;
1179 eb = find_extent_buffer_readpage(fs_info, page, start);
1180 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
1181 eb->read_mirror = mirror;
1182 atomic_dec(&eb->io_pages);
1187 * submit_data_read_repair() will handle all the good
1188 * and bad sectors, we just continue to the next bvec.
1190 submit_data_read_repair(inode, bbio, bio_offset, bvec,
1193 /* Update page status and unlock */
1194 end_page_read(page, uptodate, start, len);
1195 endio_readpage_release_extent(&processed, BTRFS_I(inode),
1196 start, end, PageUptodate(page));
1199 ASSERT(bio_offset + len > bio_offset);
1203 /* Release the last extent */
1204 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
1205 btrfs_bio_free_csum(bbio);
1210 * Populate every free slot in a provided array with pages.
1212 * @nr_pages: number of pages to allocate
1213 * @page_array: the array to fill with pages; any existing non-null entries in
1214 * the array will be skipped
1216 * Return: 0 if all pages were able to be allocated;
1217 * -ENOMEM otherwise, and the caller is responsible for freeing all
1218 * non-null page pointers in the array.
1220 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
1222 unsigned int allocated;
1224 for (allocated = 0; allocated < nr_pages;) {
1225 unsigned int last = allocated;
1227 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
1229 if (allocated == nr_pages)
1233 * During this iteration, no page could be allocated, even
1234 * though alloc_pages_bulk_array() falls back to alloc_page()
1235 * if it could not bulk-allocate. So we must be out of memory.
1237 if (allocated == last)
1240 memalloc_retry_wait(GFP_NOFS);
1246 * Attempt to add a page to bio.
1248 * @bio_ctrl: record both the bio, and its bio_flags
1249 * @page: page to add to the bio
1250 * @disk_bytenr: offset of the new bio or to check whether we are adding
1251 * a contiguous page to the previous one
1252 * @size: portion of page that we want to write
1253 * @pg_offset: starting offset in the page
1254 * @compress_type: compression type of the current bio to see if we can merge them
1256 * Attempt to add a page to bio considering stripe alignment etc.
1258 * Return >= 0 for the number of bytes added to the bio.
1259 * Can return 0 if the current bio is already at stripe/zone boundary.
1260 * Return <0 for error.
1262 static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl,
1264 u64 disk_bytenr, unsigned int size,
1265 unsigned int pg_offset,
1266 enum btrfs_compression_type compress_type)
1268 struct bio *bio = bio_ctrl->bio;
1269 u32 bio_size = bio->bi_iter.bi_size;
1271 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
1272 bool contig = false;
1276 /* The limit should be calculated when bio_ctrl->bio is allocated */
1277 ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary);
1278 if (bio_ctrl->compress_type != compress_type)
1282 if (bio->bi_iter.bi_size == 0) {
1283 /* We can always add a page into an empty bio. */
1285 } else if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE) {
1286 struct bio_vec *bvec = bio_last_bvec_all(bio);
1289 * The contig check requires the following conditions to be met:
1290 * 1) The pages are belonging to the same inode
1291 * This is implied by the call chain.
1293 * 2) The range has adjacent logical bytenr
1295 * 3) The range has adjacent file offset
1296 * This is required for the usage of btrfs_bio->file_offset.
1298 if (bio_end_sector(bio) == sector &&
1299 page_offset(bvec->bv_page) + bvec->bv_offset +
1300 bvec->bv_len == page_offset(page) + pg_offset)
1304 * For compression, all IO should have its logical bytenr
1305 * set to the starting bytenr of the compressed extent.
1307 contig = bio->bi_iter.bi_sector == sector;
1313 real_size = min(bio_ctrl->len_to_oe_boundary,
1314 bio_ctrl->len_to_stripe_boundary) - bio_size;
1315 real_size = min(real_size, size);
1318 * If real_size is 0, never call bio_add_*_page(), as even size is 0,
1319 * bio will still execute its endio function on the page!
1324 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1325 ret = bio_add_zone_append_page(bio, page, real_size, pg_offset);
1327 ret = bio_add_page(bio, page, real_size, pg_offset);
1332 static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl,
1333 struct btrfs_inode *inode, u64 file_offset)
1335 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1336 struct btrfs_io_geometry geom;
1337 struct btrfs_ordered_extent *ordered;
1338 struct extent_map *em;
1339 u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT);
1343 * Pages for compressed extent are never submitted to disk directly,
1344 * thus it has no real boundary, just set them to U32_MAX.
1346 * The split happens for real compressed bio, which happens in
1347 * btrfs_submit_compressed_read/write().
1349 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
1350 bio_ctrl->len_to_oe_boundary = U32_MAX;
1351 bio_ctrl->len_to_stripe_boundary = U32_MAX;
1354 em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize);
1357 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio),
1359 free_extent_map(em);
1363 if (geom.len > U32_MAX)
1364 bio_ctrl->len_to_stripe_boundary = U32_MAX;
1366 bio_ctrl->len_to_stripe_boundary = (u32)geom.len;
1368 if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) {
1369 bio_ctrl->len_to_oe_boundary = U32_MAX;
1373 /* Ordered extent not yet created, so we're good */
1374 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
1376 bio_ctrl->len_to_oe_boundary = U32_MAX;
1380 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
1381 ordered->disk_bytenr + ordered->disk_num_bytes - logical);
1382 btrfs_put_ordered_extent(ordered);
1386 static int alloc_new_bio(struct btrfs_inode *inode,
1387 struct btrfs_bio_ctrl *bio_ctrl,
1388 struct writeback_control *wbc,
1390 u64 disk_bytenr, u32 offset, u64 file_offset,
1391 enum btrfs_compression_type compress_type)
1393 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1397 ASSERT(bio_ctrl->end_io_func);
1399 bio = btrfs_bio_alloc(BIO_MAX_VECS, opf, bio_ctrl->end_io_func, NULL);
1401 * For compressed page range, its disk_bytenr is always @disk_bytenr
1402 * passed in, no matter if we have added any range into previous bio.
1404 if (compress_type != BTRFS_COMPRESS_NONE)
1405 bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
1407 bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT;
1408 bio_ctrl->bio = bio;
1409 bio_ctrl->compress_type = compress_type;
1410 ret = calc_bio_boundaries(bio_ctrl, inode, file_offset);
1416 * For Zone append we need the correct block_device that we are
1417 * going to write to set in the bio to be able to respect the
1418 * hardware limitation. Look it up here:
1420 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1421 struct btrfs_device *dev;
1423 dev = btrfs_zoned_get_device(fs_info, disk_bytenr,
1424 fs_info->sectorsize);
1430 bio_set_dev(bio, dev->bdev);
1433 * Otherwise pick the last added device to support
1434 * cgroup writeback. For multi-device file systems this
1435 * means blk-cgroup policies have to always be set on the
1436 * last added/replaced device. This is a bit odd but has
1437 * been like that for a long time.
1439 bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev);
1441 wbc_init_bio(wbc, bio);
1443 ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND);
1447 bio_ctrl->bio = NULL;
1448 btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret));
1453 * @opf: bio REQ_OP_* and REQ_* flags as one value
1454 * @wbc: optional writeback control for io accounting
1455 * @disk_bytenr: logical bytenr where the write will be
1456 * @page: page to add to the bio
1457 * @size: portion of page that we want to write to
1458 * @pg_offset: offset of the new bio or to check whether we are adding
1459 * a contiguous page to the previous one
1460 * @compress_type: compress type for current bio
1462 * The will either add the page into the existing @bio_ctrl->bio, or allocate a
1463 * new one in @bio_ctrl->bio.
1464 * The mirror number for this IO should already be initizlied in
1465 * @bio_ctrl->mirror_num.
1467 static int submit_extent_page(blk_opf_t opf,
1468 struct writeback_control *wbc,
1469 struct btrfs_bio_ctrl *bio_ctrl,
1470 u64 disk_bytenr, struct page *page,
1471 size_t size, unsigned long pg_offset,
1472 enum btrfs_compression_type compress_type,
1473 bool force_bio_submit)
1476 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1477 unsigned int cur = pg_offset;
1481 ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE &&
1482 pg_offset + size <= PAGE_SIZE);
1484 ASSERT(bio_ctrl->end_io_func);
1486 if (force_bio_submit)
1487 submit_one_bio(bio_ctrl);
1489 while (cur < pg_offset + size) {
1490 u32 offset = cur - pg_offset;
1493 /* Allocate new bio if needed */
1494 if (!bio_ctrl->bio) {
1495 ret = alloc_new_bio(inode, bio_ctrl, wbc, opf,
1496 disk_bytenr, offset,
1497 page_offset(page) + cur,
1503 * We must go through btrfs_bio_add_page() to ensure each
1504 * page range won't cross various boundaries.
1506 if (compress_type != BTRFS_COMPRESS_NONE)
1507 added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr,
1508 size - offset, pg_offset + offset,
1511 added = btrfs_bio_add_page(bio_ctrl, page,
1512 disk_bytenr + offset, size - offset,
1513 pg_offset + offset, compress_type);
1515 /* Metadata page range should never be split */
1516 if (!is_data_inode(&inode->vfs_inode))
1517 ASSERT(added == 0 || added == size - offset);
1519 /* At least we added some page, update the account */
1521 wbc_account_cgroup_owner(wbc, page, added);
1523 /* We have reached boundary, submit right now */
1524 if (added < size - offset) {
1525 /* The bio should contain some page(s) */
1526 ASSERT(bio_ctrl->bio->bi_iter.bi_size);
1527 submit_one_bio(bio_ctrl);
1534 static int attach_extent_buffer_page(struct extent_buffer *eb,
1536 struct btrfs_subpage *prealloc)
1538 struct btrfs_fs_info *fs_info = eb->fs_info;
1542 * If the page is mapped to btree inode, we should hold the private
1543 * lock to prevent race.
1544 * For cloned or dummy extent buffers, their pages are not mapped and
1545 * will not race with any other ebs.
1548 lockdep_assert_held(&page->mapping->private_lock);
1550 if (fs_info->nodesize >= PAGE_SIZE) {
1551 if (!PagePrivate(page))
1552 attach_page_private(page, eb);
1554 WARN_ON(page->private != (unsigned long)eb);
1558 /* Already mapped, just free prealloc */
1559 if (PagePrivate(page)) {
1560 btrfs_free_subpage(prealloc);
1565 /* Has preallocated memory for subpage */
1566 attach_page_private(page, prealloc);
1568 /* Do new allocation to attach subpage */
1569 ret = btrfs_attach_subpage(fs_info, page,
1570 BTRFS_SUBPAGE_METADATA);
1574 int set_page_extent_mapped(struct page *page)
1576 struct btrfs_fs_info *fs_info;
1578 ASSERT(page->mapping);
1580 if (PagePrivate(page))
1583 fs_info = btrfs_sb(page->mapping->host->i_sb);
1585 if (btrfs_is_subpage(fs_info, page))
1586 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
1588 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
1592 void clear_page_extent_mapped(struct page *page)
1594 struct btrfs_fs_info *fs_info;
1596 ASSERT(page->mapping);
1598 if (!PagePrivate(page))
1601 fs_info = btrfs_sb(page->mapping->host->i_sb);
1602 if (btrfs_is_subpage(fs_info, page))
1603 return btrfs_detach_subpage(fs_info, page);
1605 detach_page_private(page);
1608 static struct extent_map *
1609 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
1610 u64 start, u64 len, struct extent_map **em_cached)
1612 struct extent_map *em;
1614 if (em_cached && *em_cached) {
1616 if (extent_map_in_tree(em) && start >= em->start &&
1617 start < extent_map_end(em)) {
1618 refcount_inc(&em->refs);
1622 free_extent_map(em);
1626 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
1627 if (em_cached && !IS_ERR(em)) {
1629 refcount_inc(&em->refs);
1635 * basic readpage implementation. Locked extent state structs are inserted
1636 * into the tree that are removed when the IO is done (by the end_io
1638 * XXX JDM: This needs looking at to ensure proper page locking
1639 * return 0 on success, otherwise return error
1641 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
1642 struct btrfs_bio_ctrl *bio_ctrl,
1643 blk_opf_t read_flags, u64 *prev_em_start)
1645 struct inode *inode = page->mapping->host;
1646 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1647 u64 start = page_offset(page);
1648 const u64 end = start + PAGE_SIZE - 1;
1651 u64 last_byte = i_size_read(inode);
1653 struct extent_map *em;
1655 size_t pg_offset = 0;
1657 size_t blocksize = inode->i_sb->s_blocksize;
1658 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
1660 ret = set_page_extent_mapped(page);
1662 unlock_extent(tree, start, end, NULL);
1663 btrfs_page_set_error(fs_info, page, start, PAGE_SIZE);
1668 if (page->index == last_byte >> PAGE_SHIFT) {
1669 size_t zero_offset = offset_in_page(last_byte);
1672 iosize = PAGE_SIZE - zero_offset;
1673 memzero_page(page, zero_offset, iosize);
1676 bio_ctrl->end_io_func = end_bio_extent_readpage;
1677 begin_page_read(fs_info, page);
1678 while (cur <= end) {
1679 unsigned long this_bio_flag = 0;
1680 bool force_bio_submit = false;
1683 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1684 if (cur >= last_byte) {
1685 iosize = PAGE_SIZE - pg_offset;
1686 memzero_page(page, pg_offset, iosize);
1687 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1688 end_page_read(page, true, cur, iosize);
1691 em = __get_extent_map(inode, page, pg_offset, cur,
1692 end - cur + 1, em_cached);
1694 unlock_extent(tree, cur, end, NULL);
1695 end_page_read(page, false, cur, end + 1 - cur);
1699 extent_offset = cur - em->start;
1700 BUG_ON(extent_map_end(em) <= cur);
1703 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1704 this_bio_flag = em->compress_type;
1706 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1707 iosize = ALIGN(iosize, blocksize);
1708 if (this_bio_flag != BTRFS_COMPRESS_NONE)
1709 disk_bytenr = em->block_start;
1711 disk_bytenr = em->block_start + extent_offset;
1712 block_start = em->block_start;
1713 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1714 block_start = EXTENT_MAP_HOLE;
1717 * If we have a file range that points to a compressed extent
1718 * and it's followed by a consecutive file range that points
1719 * to the same compressed extent (possibly with a different
1720 * offset and/or length, so it either points to the whole extent
1721 * or only part of it), we must make sure we do not submit a
1722 * single bio to populate the pages for the 2 ranges because
1723 * this makes the compressed extent read zero out the pages
1724 * belonging to the 2nd range. Imagine the following scenario:
1727 * [0 - 8K] [8K - 24K]
1730 * points to extent X, points to extent X,
1731 * offset 4K, length of 8K offset 0, length 16K
1733 * [extent X, compressed length = 4K uncompressed length = 16K]
1735 * If the bio to read the compressed extent covers both ranges,
1736 * it will decompress extent X into the pages belonging to the
1737 * first range and then it will stop, zeroing out the remaining
1738 * pages that belong to the other range that points to extent X.
1739 * So here we make sure we submit 2 bios, one for the first
1740 * range and another one for the third range. Both will target
1741 * the same physical extent from disk, but we can't currently
1742 * make the compressed bio endio callback populate the pages
1743 * for both ranges because each compressed bio is tightly
1744 * coupled with a single extent map, and each range can have
1745 * an extent map with a different offset value relative to the
1746 * uncompressed data of our extent and different lengths. This
1747 * is a corner case so we prioritize correctness over
1748 * non-optimal behavior (submitting 2 bios for the same extent).
1750 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
1751 prev_em_start && *prev_em_start != (u64)-1 &&
1752 *prev_em_start != em->start)
1753 force_bio_submit = true;
1756 *prev_em_start = em->start;
1758 free_extent_map(em);
1761 /* we've found a hole, just zero and go on */
1762 if (block_start == EXTENT_MAP_HOLE) {
1763 memzero_page(page, pg_offset, iosize);
1765 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1766 end_page_read(page, true, cur, iosize);
1768 pg_offset += iosize;
1771 /* the get_extent function already copied into the page */
1772 if (block_start == EXTENT_MAP_INLINE) {
1773 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1774 end_page_read(page, true, cur, iosize);
1776 pg_offset += iosize;
1780 ret = submit_extent_page(REQ_OP_READ | read_flags, NULL,
1781 bio_ctrl, disk_bytenr, page, iosize,
1782 pg_offset, this_bio_flag,
1786 * We have to unlock the remaining range, or the page
1787 * will never be unlocked.
1789 unlock_extent(tree, cur, end, NULL);
1790 end_page_read(page, false, cur, end + 1 - cur);
1794 pg_offset += iosize;
1800 int btrfs_read_folio(struct file *file, struct folio *folio)
1802 struct page *page = &folio->page;
1803 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1804 u64 start = page_offset(page);
1805 u64 end = start + PAGE_SIZE - 1;
1806 struct btrfs_bio_ctrl bio_ctrl = { 0 };
1809 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1811 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL);
1813 * If btrfs_do_readpage() failed we will want to submit the assembled
1814 * bio to do the cleanup.
1816 submit_one_bio(&bio_ctrl);
1820 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1822 struct extent_map **em_cached,
1823 struct btrfs_bio_ctrl *bio_ctrl,
1826 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1829 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1831 for (index = 0; index < nr_pages; index++) {
1832 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1833 REQ_RAHEAD, prev_em_start);
1834 put_page(pages[index]);
1839 * helper for __extent_writepage, doing all of the delayed allocation setup.
1841 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1842 * to write the page (copy into inline extent). In this case the IO has
1843 * been started and the page is already unlocked.
1845 * This returns 0 if all went well (page still locked)
1846 * This returns < 0 if there were errors (page still locked)
1848 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1849 struct page *page, struct writeback_control *wbc)
1851 const u64 page_end = page_offset(page) + PAGE_SIZE - 1;
1852 u64 delalloc_start = page_offset(page);
1853 u64 delalloc_to_write = 0;
1854 /* How many pages are started by btrfs_run_delalloc_range() */
1855 unsigned long nr_written = 0;
1857 int page_started = 0;
1859 while (delalloc_start < page_end) {
1860 u64 delalloc_end = page_end;
1863 found = find_lock_delalloc_range(&inode->vfs_inode, page,
1867 delalloc_start = delalloc_end + 1;
1870 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1871 delalloc_end, &page_started, &nr_written, wbc);
1873 btrfs_page_set_error(inode->root->fs_info, page,
1874 page_offset(page), PAGE_SIZE);
1878 * delalloc_end is already one less than the total length, so
1879 * we don't subtract one from PAGE_SIZE
1881 delalloc_to_write += (delalloc_end - delalloc_start +
1882 PAGE_SIZE) >> PAGE_SHIFT;
1883 delalloc_start = delalloc_end + 1;
1885 if (wbc->nr_to_write < delalloc_to_write) {
1888 if (delalloc_to_write < thresh * 2)
1889 thresh = delalloc_to_write;
1890 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1894 /* Did btrfs_run_dealloc_range() already unlock and start the IO? */
1897 * We've unlocked the page, so we can't update the mapping's
1898 * writeback index, just update nr_to_write.
1900 wbc->nr_to_write -= nr_written;
1908 * Find the first byte we need to write.
1910 * For subpage, one page can contain several sectors, and
1911 * __extent_writepage_io() will just grab all extent maps in the page
1912 * range and try to submit all non-inline/non-compressed extents.
1914 * This is a big problem for subpage, we shouldn't re-submit already written
1916 * This function will lookup subpage dirty bit to find which range we really
1919 * Return the next dirty range in [@start, @end).
1920 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1922 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1923 struct page *page, u64 *start, u64 *end)
1925 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1926 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1927 u64 orig_start = *start;
1928 /* Declare as unsigned long so we can use bitmap ops */
1929 unsigned long flags;
1930 int range_start_bit;
1934 * For regular sector size == page size case, since one page only
1935 * contains one sector, we return the page offset directly.
1937 if (!btrfs_is_subpage(fs_info, page)) {
1938 *start = page_offset(page);
1939 *end = page_offset(page) + PAGE_SIZE;
1943 range_start_bit = spi->dirty_offset +
1944 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1946 /* We should have the page locked, but just in case */
1947 spin_lock_irqsave(&subpage->lock, flags);
1948 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1949 spi->dirty_offset + spi->bitmap_nr_bits);
1950 spin_unlock_irqrestore(&subpage->lock, flags);
1952 range_start_bit -= spi->dirty_offset;
1953 range_end_bit -= spi->dirty_offset;
1955 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1956 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1960 * helper for __extent_writepage. This calls the writepage start hooks,
1961 * and does the loop to map the page into extents and bios.
1963 * We return 1 if the IO is started and the page is unlocked,
1964 * 0 if all went well (page still locked)
1965 * < 0 if there were errors (page still locked)
1967 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1969 struct writeback_control *wbc,
1970 struct btrfs_bio_ctrl *bio_ctrl,
1974 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1975 u64 cur = page_offset(page);
1976 u64 end = cur + PAGE_SIZE - 1;
1979 struct extent_map *em;
1983 enum req_op op = REQ_OP_WRITE;
1984 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1985 bool has_error = false;
1988 ret = btrfs_writepage_cow_fixup(page);
1990 /* Fixup worker will requeue */
1991 redirty_page_for_writepage(wbc, page);
1997 * we don't want to touch the inode after unlocking the page,
1998 * so we update the mapping writeback index now
2002 bio_ctrl->end_io_func = end_bio_extent_writepage;
2003 while (cur <= end) {
2006 u64 dirty_range_start = cur;
2007 u64 dirty_range_end;
2010 if (cur >= i_size) {
2011 btrfs_writepage_endio_finish_ordered(inode, page, cur,
2014 * This range is beyond i_size, thus we don't need to
2015 * bother writing back.
2016 * But we still need to clear the dirty subpage bit, or
2017 * the next time the page gets dirtied, we will try to
2018 * writeback the sectors with subpage dirty bits,
2019 * causing writeback without ordered extent.
2021 btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur);
2025 find_next_dirty_byte(fs_info, page, &dirty_range_start,
2027 if (cur < dirty_range_start) {
2028 cur = dirty_range_start;
2032 em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1);
2034 btrfs_page_set_error(fs_info, page, cur, end - cur + 1);
2035 ret = PTR_ERR_OR_ZERO(em);
2042 extent_offset = cur - em->start;
2043 em_end = extent_map_end(em);
2044 ASSERT(cur <= em_end);
2046 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
2047 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
2048 block_start = em->block_start;
2049 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
2050 disk_bytenr = em->block_start + extent_offset;
2053 * Note that em_end from extent_map_end() and dirty_range_end from
2054 * find_next_dirty_byte() are all exclusive
2056 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
2058 if (btrfs_use_zone_append(inode, em->block_start))
2059 op = REQ_OP_ZONE_APPEND;
2061 free_extent_map(em);
2065 * compressed and inline extents are written through other
2068 if (compressed || block_start == EXTENT_MAP_HOLE ||
2069 block_start == EXTENT_MAP_INLINE) {
2073 btrfs_writepage_endio_finish_ordered(inode,
2074 page, cur, cur + iosize - 1, true);
2075 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
2080 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
2081 if (!PageWriteback(page)) {
2082 btrfs_err(inode->root->fs_info,
2083 "page %lu not writeback, cur %llu end %llu",
2084 page->index, cur, end);
2088 * Although the PageDirty bit is cleared before entering this
2089 * function, subpage dirty bit is not cleared.
2090 * So clear subpage dirty bit here so next time we won't submit
2091 * page for range already written to disk.
2093 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
2095 ret = submit_extent_page(op | write_flags, wbc,
2096 bio_ctrl, disk_bytenr,
2098 cur - page_offset(page),
2105 btrfs_page_set_error(fs_info, page, cur, iosize);
2106 if (PageWriteback(page))
2107 btrfs_page_clear_writeback(fs_info, page, cur,
2115 * If we finish without problem, we should not only clear page dirty,
2116 * but also empty subpage dirty bits
2119 btrfs_page_assert_not_dirty(fs_info, page);
2127 * the writepage semantics are similar to regular writepage. extent
2128 * records are inserted to lock ranges in the tree, and as dirty areas
2129 * are found, they are marked writeback. Then the lock bits are removed
2130 * and the end_io handler clears the writeback ranges
2132 * Return 0 if everything goes well.
2133 * Return <0 for error.
2135 static int __extent_writepage(struct page *page, struct writeback_control *wbc,
2136 struct btrfs_bio_ctrl *bio_ctrl)
2138 struct folio *folio = page_folio(page);
2139 struct inode *inode = page->mapping->host;
2140 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2141 const u64 page_start = page_offset(page);
2142 const u64 page_end = page_start + PAGE_SIZE - 1;
2146 loff_t i_size = i_size_read(inode);
2147 unsigned long end_index = i_size >> PAGE_SHIFT;
2149 trace___extent_writepage(page, inode, wbc);
2151 WARN_ON(!PageLocked(page));
2153 btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
2154 page_offset(page), PAGE_SIZE);
2156 pg_offset = offset_in_page(i_size);
2157 if (page->index > end_index ||
2158 (page->index == end_index && !pg_offset)) {
2159 folio_invalidate(folio, 0, folio_size(folio));
2160 folio_unlock(folio);
2164 if (page->index == end_index)
2165 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
2167 ret = set_page_extent_mapped(page);
2173 if (!bio_ctrl->extent_locked) {
2174 ret = writepage_delalloc(BTRFS_I(inode), page, wbc);
2181 ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, bio_ctrl, i_size,
2188 /* make sure the mapping tag for page dirty gets cleared */
2189 set_page_writeback(page);
2190 end_page_writeback(page);
2193 * Here we used to have a check for PageError() and then set @ret and
2194 * call end_extent_writepage().
2196 * But in fact setting @ret here will cause different error paths
2197 * between subpage and regular sectorsize.
2199 * For regular page size, we never submit current page, but only add
2200 * current page to current bio.
2201 * The bio submission can only happen in next page.
2202 * Thus if we hit the PageError() branch, @ret is already set to
2203 * non-zero value and will not get updated for regular sectorsize.
2205 * But for subpage case, it's possible we submit part of current page,
2206 * thus can get PageError() set by submitted bio of the same page,
2207 * while our @ret is still 0.
2209 * So here we unify the behavior and don't set @ret.
2210 * Error can still be properly passed to higher layer as page will
2211 * be set error, here we just don't handle the IO failure.
2213 * NOTE: This is just a hotfix for subpage.
2214 * The root fix will be properly ending ordered extent when we hit
2215 * an error during writeback.
2217 * But that needs a bigger refactoring, as we not only need to grab the
2218 * submitted OE, but also need to know exactly at which bytenr we hit
2220 * Currently the full page based __extent_writepage_io() is not
2223 if (PageError(page))
2224 end_extent_writepage(page, ret, page_start, page_end);
2225 if (bio_ctrl->extent_locked) {
2227 * If bio_ctrl->extent_locked, it's from extent_write_locked_range(),
2228 * the page can either be locked by lock_page() or
2229 * process_one_page().
2230 * Let btrfs_page_unlock_writer() handle both cases.
2233 btrfs_page_unlock_writer(fs_info, page, wbc->range_start,
2234 wbc->range_end + 1 - wbc->range_start);
2242 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
2244 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
2245 TASK_UNINTERRUPTIBLE);
2248 static void end_extent_buffer_writeback(struct extent_buffer *eb)
2250 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
2251 smp_mb__after_atomic();
2252 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
2256 * Lock extent buffer status and pages for writeback.
2258 * May try to flush write bio if we can't get the lock.
2260 * Return 0 if the extent buffer doesn't need to be submitted.
2261 * (E.g. the extent buffer is not dirty)
2262 * Return >0 is the extent buffer is submitted to bio.
2263 * Return <0 if something went wrong, no page is locked.
2265 static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb,
2266 struct btrfs_bio_ctrl *bio_ctrl)
2268 struct btrfs_fs_info *fs_info = eb->fs_info;
2273 if (!btrfs_try_tree_write_lock(eb)) {
2274 submit_write_bio(bio_ctrl, 0);
2276 btrfs_tree_lock(eb);
2279 if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
2280 btrfs_tree_unlock(eb);
2281 if (!bio_ctrl->sync_io)
2284 submit_write_bio(bio_ctrl, 0);
2288 wait_on_extent_buffer_writeback(eb);
2289 btrfs_tree_lock(eb);
2290 if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags))
2292 btrfs_tree_unlock(eb);
2297 * We need to do this to prevent races in people who check if the eb is
2298 * under IO since we can end up having no IO bits set for a short period
2301 spin_lock(&eb->refs_lock);
2302 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2303 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
2304 spin_unlock(&eb->refs_lock);
2305 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2306 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
2308 fs_info->dirty_metadata_batch);
2311 spin_unlock(&eb->refs_lock);
2314 btrfs_tree_unlock(eb);
2317 * Either we don't need to submit any tree block, or we're submitting
2319 * Subpage metadata doesn't use page locking at all, so we can skip
2322 if (!ret || fs_info->nodesize < PAGE_SIZE)
2325 num_pages = num_extent_pages(eb);
2326 for (i = 0; i < num_pages; i++) {
2327 struct page *p = eb->pages[i];
2329 if (!trylock_page(p)) {
2331 submit_write_bio(bio_ctrl, 0);
2341 static void set_btree_ioerr(struct page *page, struct extent_buffer *eb)
2343 struct btrfs_fs_info *fs_info = eb->fs_info;
2345 btrfs_page_set_error(fs_info, page, eb->start, eb->len);
2346 if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
2350 * A read may stumble upon this buffer later, make sure that it gets an
2351 * error and knows there was an error.
2353 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
2356 * We need to set the mapping with the io error as well because a write
2357 * error will flip the file system readonly, and then syncfs() will
2358 * return a 0 because we are readonly if we don't modify the err seq for
2361 mapping_set_error(page->mapping, -EIO);
2364 * If we error out, we should add back the dirty_metadata_bytes
2365 * to make it consistent.
2367 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
2368 eb->len, fs_info->dirty_metadata_batch);
2371 * If writeback for a btree extent that doesn't belong to a log tree
2372 * failed, increment the counter transaction->eb_write_errors.
2373 * We do this because while the transaction is running and before it's
2374 * committing (when we call filemap_fdata[write|wait]_range against
2375 * the btree inode), we might have
2376 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
2377 * returns an error or an error happens during writeback, when we're
2378 * committing the transaction we wouldn't know about it, since the pages
2379 * can be no longer dirty nor marked anymore for writeback (if a
2380 * subsequent modification to the extent buffer didn't happen before the
2381 * transaction commit), which makes filemap_fdata[write|wait]_range not
2382 * able to find the pages tagged with SetPageError at transaction
2383 * commit time. So if this happens we must abort the transaction,
2384 * otherwise we commit a super block with btree roots that point to
2385 * btree nodes/leafs whose content on disk is invalid - either garbage
2386 * or the content of some node/leaf from a past generation that got
2387 * cowed or deleted and is no longer valid.
2389 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
2390 * not be enough - we need to distinguish between log tree extents vs
2391 * non-log tree extents, and the next filemap_fdatawait_range() call
2392 * will catch and clear such errors in the mapping - and that call might
2393 * be from a log sync and not from a transaction commit. Also, checking
2394 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
2395 * not done and would not be reliable - the eb might have been released
2396 * from memory and reading it back again means that flag would not be
2397 * set (since it's a runtime flag, not persisted on disk).
2399 * Using the flags below in the btree inode also makes us achieve the
2400 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
2401 * writeback for all dirty pages and before filemap_fdatawait_range()
2402 * is called, the writeback for all dirty pages had already finished
2403 * with errors - because we were not using AS_EIO/AS_ENOSPC,
2404 * filemap_fdatawait_range() would return success, as it could not know
2405 * that writeback errors happened (the pages were no longer tagged for
2408 switch (eb->log_index) {
2410 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
2413 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
2416 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
2419 BUG(); /* unexpected, logic error */
2424 * The endio specific version which won't touch any unsafe spinlock in endio
2427 static struct extent_buffer *find_extent_buffer_nolock(
2428 struct btrfs_fs_info *fs_info, u64 start)
2430 struct extent_buffer *eb;
2433 eb = radix_tree_lookup(&fs_info->buffer_radix,
2434 start >> fs_info->sectorsize_bits);
2435 if (eb && atomic_inc_not_zero(&eb->refs)) {
2444 * The endio function for subpage extent buffer write.
2446 * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback()
2447 * after all extent buffers in the page has finished their writeback.
2449 static void end_bio_subpage_eb_writepage(struct btrfs_bio *bbio)
2451 struct bio *bio = &bbio->bio;
2452 struct btrfs_fs_info *fs_info;
2453 struct bio_vec *bvec;
2454 struct bvec_iter_all iter_all;
2456 fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb);
2457 ASSERT(fs_info->nodesize < PAGE_SIZE);
2459 ASSERT(!bio_flagged(bio, BIO_CLONED));
2460 bio_for_each_segment_all(bvec, bio, iter_all) {
2461 struct page *page = bvec->bv_page;
2462 u64 bvec_start = page_offset(page) + bvec->bv_offset;
2463 u64 bvec_end = bvec_start + bvec->bv_len - 1;
2464 u64 cur_bytenr = bvec_start;
2466 ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize));
2468 /* Iterate through all extent buffers in the range */
2469 while (cur_bytenr <= bvec_end) {
2470 struct extent_buffer *eb;
2474 * Here we can't use find_extent_buffer(), as it may
2475 * try to lock eb->refs_lock, which is not safe in endio
2478 eb = find_extent_buffer_nolock(fs_info, cur_bytenr);
2481 cur_bytenr = eb->start + eb->len;
2483 ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags));
2484 done = atomic_dec_and_test(&eb->io_pages);
2487 if (bio->bi_status ||
2488 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
2489 ClearPageUptodate(page);
2490 set_btree_ioerr(page, eb);
2493 btrfs_subpage_clear_writeback(fs_info, page, eb->start,
2495 end_extent_buffer_writeback(eb);
2497 * free_extent_buffer() will grab spinlock which is not
2498 * safe in endio context. Thus here we manually dec
2501 atomic_dec(&eb->refs);
2507 static void end_bio_extent_buffer_writepage(struct btrfs_bio *bbio)
2509 struct bio *bio = &bbio->bio;
2510 struct bio_vec *bvec;
2511 struct extent_buffer *eb;
2513 struct bvec_iter_all iter_all;
2515 ASSERT(!bio_flagged(bio, BIO_CLONED));
2516 bio_for_each_segment_all(bvec, bio, iter_all) {
2517 struct page *page = bvec->bv_page;
2519 eb = (struct extent_buffer *)page->private;
2521 done = atomic_dec_and_test(&eb->io_pages);
2523 if (bio->bi_status ||
2524 test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) {
2525 ClearPageUptodate(page);
2526 set_btree_ioerr(page, eb);
2529 end_page_writeback(page);
2534 end_extent_buffer_writeback(eb);
2540 static void prepare_eb_write(struct extent_buffer *eb)
2543 unsigned long start;
2546 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
2547 atomic_set(&eb->io_pages, num_extent_pages(eb));
2549 /* Set btree blocks beyond nritems with 0 to avoid stale content */
2550 nritems = btrfs_header_nritems(eb);
2551 if (btrfs_header_level(eb) > 0) {
2552 end = btrfs_node_key_ptr_offset(eb, nritems);
2553 memzero_extent_buffer(eb, end, eb->len - end);
2557 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
2559 start = btrfs_item_nr_offset(eb, nritems);
2560 end = btrfs_item_nr_offset(eb, 0);
2562 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
2564 end += btrfs_item_offset(eb, nritems - 1);
2565 memzero_extent_buffer(eb, start, end - start);
2570 * Unlike the work in write_one_eb(), we rely completely on extent locking.
2571 * Page locking is only utilized at minimum to keep the VMM code happy.
2573 static int write_one_subpage_eb(struct extent_buffer *eb,
2574 struct writeback_control *wbc,
2575 struct btrfs_bio_ctrl *bio_ctrl)
2577 struct btrfs_fs_info *fs_info = eb->fs_info;
2578 struct page *page = eb->pages[0];
2579 blk_opf_t write_flags = wbc_to_write_flags(wbc);
2580 bool no_dirty_ebs = false;
2583 prepare_eb_write(eb);
2585 /* clear_page_dirty_for_io() in subpage helper needs page locked */
2587 btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len);
2589 /* Check if this is the last dirty bit to update nr_written */
2590 no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page,
2591 eb->start, eb->len);
2593 clear_page_dirty_for_io(page);
2595 bio_ctrl->end_io_func = end_bio_subpage_eb_writepage;
2597 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
2598 bio_ctrl, eb->start, page, eb->len,
2599 eb->start - page_offset(page), 0, false);
2601 btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len);
2602 set_btree_ioerr(page, eb);
2605 if (atomic_dec_and_test(&eb->io_pages))
2606 end_extent_buffer_writeback(eb);
2611 * Submission finished without problem, if no range of the page is
2612 * dirty anymore, we have submitted a page. Update nr_written in wbc.
2619 static noinline_for_stack int write_one_eb(struct extent_buffer *eb,
2620 struct writeback_control *wbc,
2621 struct btrfs_bio_ctrl *bio_ctrl)
2623 u64 disk_bytenr = eb->start;
2625 blk_opf_t write_flags = wbc_to_write_flags(wbc);
2628 prepare_eb_write(eb);
2630 bio_ctrl->end_io_func = end_bio_extent_buffer_writepage;
2632 num_pages = num_extent_pages(eb);
2633 for (i = 0; i < num_pages; i++) {
2634 struct page *p = eb->pages[i];
2636 clear_page_dirty_for_io(p);
2637 set_page_writeback(p);
2638 ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc,
2639 bio_ctrl, disk_bytenr, p,
2640 PAGE_SIZE, 0, 0, false);
2642 set_btree_ioerr(p, eb);
2643 if (PageWriteback(p))
2644 end_page_writeback(p);
2645 if (atomic_sub_and_test(num_pages - i, &eb->io_pages))
2646 end_extent_buffer_writeback(eb);
2650 disk_bytenr += PAGE_SIZE;
2655 if (unlikely(ret)) {
2656 for (; i < num_pages; i++) {
2657 struct page *p = eb->pages[i];
2658 clear_page_dirty_for_io(p);
2667 * Submit one subpage btree page.
2669 * The main difference to submit_eb_page() is:
2671 * For subpage, we don't rely on page locking at all.
2674 * We only flush bio if we may be unable to fit current extent buffers into
2677 * Return >=0 for the number of submitted extent buffers.
2678 * Return <0 for fatal error.
2680 static int submit_eb_subpage(struct page *page,
2681 struct writeback_control *wbc,
2682 struct btrfs_bio_ctrl *bio_ctrl)
2684 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
2686 u64 page_start = page_offset(page);
2688 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
2691 /* Lock and write each dirty extent buffers in the range */
2692 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
2693 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
2694 struct extent_buffer *eb;
2695 unsigned long flags;
2699 * Take private lock to ensure the subpage won't be detached
2702 spin_lock(&page->mapping->private_lock);
2703 if (!PagePrivate(page)) {
2704 spin_unlock(&page->mapping->private_lock);
2707 spin_lock_irqsave(&subpage->lock, flags);
2708 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
2709 subpage->bitmaps)) {
2710 spin_unlock_irqrestore(&subpage->lock, flags);
2711 spin_unlock(&page->mapping->private_lock);
2716 start = page_start + bit_start * fs_info->sectorsize;
2717 bit_start += sectors_per_node;
2720 * Here we just want to grab the eb without touching extra
2721 * spin locks, so call find_extent_buffer_nolock().
2723 eb = find_extent_buffer_nolock(fs_info, start);
2724 spin_unlock_irqrestore(&subpage->lock, flags);
2725 spin_unlock(&page->mapping->private_lock);
2728 * The eb has already reached 0 refs thus find_extent_buffer()
2729 * doesn't return it. We don't need to write back such eb
2735 ret = lock_extent_buffer_for_io(eb, bio_ctrl);
2737 free_extent_buffer(eb);
2741 free_extent_buffer(eb);
2744 ret = write_one_subpage_eb(eb, wbc, bio_ctrl);
2745 free_extent_buffer(eb);
2753 /* We hit error, end bio for the submitted extent buffers */
2754 submit_write_bio(bio_ctrl, ret);
2759 * Submit all page(s) of one extent buffer.
2761 * @page: the page of one extent buffer
2762 * @eb_context: to determine if we need to submit this page, if current page
2763 * belongs to this eb, we don't need to submit
2765 * The caller should pass each page in their bytenr order, and here we use
2766 * @eb_context to determine if we have submitted pages of one extent buffer.
2768 * If we have, we just skip until we hit a new page that doesn't belong to
2769 * current @eb_context.
2771 * If not, we submit all the page(s) of the extent buffer.
2773 * Return >0 if we have submitted the extent buffer successfully.
2774 * Return 0 if we don't need to submit the page, as it's already submitted by
2776 * Return <0 for fatal error.
2778 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
2779 struct btrfs_bio_ctrl *bio_ctrl,
2780 struct extent_buffer **eb_context)
2782 struct address_space *mapping = page->mapping;
2783 struct btrfs_block_group *cache = NULL;
2784 struct extent_buffer *eb;
2787 if (!PagePrivate(page))
2790 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
2791 return submit_eb_subpage(page, wbc, bio_ctrl);
2793 spin_lock(&mapping->private_lock);
2794 if (!PagePrivate(page)) {
2795 spin_unlock(&mapping->private_lock);
2799 eb = (struct extent_buffer *)page->private;
2802 * Shouldn't happen and normally this would be a BUG_ON but no point
2803 * crashing the machine for something we can survive anyway.
2806 spin_unlock(&mapping->private_lock);
2810 if (eb == *eb_context) {
2811 spin_unlock(&mapping->private_lock);
2814 ret = atomic_inc_not_zero(&eb->refs);
2815 spin_unlock(&mapping->private_lock);
2819 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
2821 * If for_sync, this hole will be filled with
2822 * trasnsaction commit.
2824 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
2828 free_extent_buffer(eb);
2834 ret = lock_extent_buffer_for_io(eb, bio_ctrl);
2836 btrfs_revert_meta_write_pointer(cache, eb);
2838 btrfs_put_block_group(cache);
2839 free_extent_buffer(eb);
2844 * Implies write in zoned mode. Mark the last eb in a block group.
2846 btrfs_schedule_zone_finish_bg(cache, eb);
2847 btrfs_put_block_group(cache);
2849 ret = write_one_eb(eb, wbc, bio_ctrl);
2850 free_extent_buffer(eb);
2856 int btree_write_cache_pages(struct address_space *mapping,
2857 struct writeback_control *wbc)
2859 struct extent_buffer *eb_context = NULL;
2860 struct btrfs_bio_ctrl bio_ctrl = {
2862 .sync_io = (wbc->sync_mode == WB_SYNC_ALL),
2864 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
2867 int nr_to_write_done = 0;
2868 struct pagevec pvec;
2871 pgoff_t end; /* Inclusive */
2875 pagevec_init(&pvec);
2876 if (wbc->range_cyclic) {
2877 index = mapping->writeback_index; /* Start from prev offset */
2880 * Start from the beginning does not need to cycle over the
2881 * range, mark it as scanned.
2883 scanned = (index == 0);
2885 index = wbc->range_start >> PAGE_SHIFT;
2886 end = wbc->range_end >> PAGE_SHIFT;
2889 if (wbc->sync_mode == WB_SYNC_ALL)
2890 tag = PAGECACHE_TAG_TOWRITE;
2892 tag = PAGECACHE_TAG_DIRTY;
2893 btrfs_zoned_meta_io_lock(fs_info);
2895 if (wbc->sync_mode == WB_SYNC_ALL)
2896 tag_pages_for_writeback(mapping, index, end);
2897 while (!done && !nr_to_write_done && (index <= end) &&
2898 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2902 for (i = 0; i < nr_pages; i++) {
2903 struct page *page = pvec.pages[i];
2905 ret = submit_eb_page(page, wbc, &bio_ctrl, &eb_context);
2914 * the filesystem may choose to bump up nr_to_write.
2915 * We have to make sure to honor the new nr_to_write
2918 nr_to_write_done = wbc->nr_to_write <= 0;
2920 pagevec_release(&pvec);
2923 if (!scanned && !done) {
2925 * We hit the last page and there is more work to be done: wrap
2926 * back to the start of the file
2933 * If something went wrong, don't allow any metadata write bio to be
2936 * This would prevent use-after-free if we had dirty pages not
2937 * cleaned up, which can still happen by fuzzed images.
2940 * Allowing existing tree block to be allocated for other trees.
2942 * - Log tree operations
2943 * Exiting tree blocks get allocated to log tree, bumps its
2944 * generation, then get cleaned in tree re-balance.
2945 * Such tree block will not be written back, since it's clean,
2946 * thus no WRITTEN flag set.
2947 * And after log writes back, this tree block is not traced by
2948 * any dirty extent_io_tree.
2950 * - Offending tree block gets re-dirtied from its original owner
2951 * Since it has bumped generation, no WRITTEN flag, it can be
2952 * reused without COWing. This tree block will not be traced
2953 * by btrfs_transaction::dirty_pages.
2955 * Now such dirty tree block will not be cleaned by any dirty
2956 * extent io tree. Thus we don't want to submit such wild eb
2957 * if the fs already has error.
2959 * We can get ret > 0 from submit_extent_page() indicating how many ebs
2960 * were submitted. Reset it to 0 to avoid false alerts for the caller.
2964 if (!ret && BTRFS_FS_ERROR(fs_info))
2966 submit_write_bio(&bio_ctrl, ret);
2968 btrfs_zoned_meta_io_unlock(fs_info);
2973 * Walk the list of dirty pages of the given address space and write all of them.
2975 * @mapping: address space structure to write
2976 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2977 * @bio_ctrl: holds context for the write, namely the bio
2979 * If a page is already under I/O, write_cache_pages() skips it, even
2980 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2981 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2982 * and msync() need to guarantee that all the data which was dirty at the time
2983 * the call was made get new I/O started against them. If wbc->sync_mode is
2984 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2985 * existing IO to complete.
2987 static int extent_write_cache_pages(struct address_space *mapping,
2988 struct writeback_control *wbc,
2989 struct btrfs_bio_ctrl *bio_ctrl)
2991 struct inode *inode = mapping->host;
2994 int nr_to_write_done = 0;
2995 struct pagevec pvec;
2998 pgoff_t end; /* Inclusive */
3000 int range_whole = 0;
3005 * We have to hold onto the inode so that ordered extents can do their
3006 * work when the IO finishes. The alternative to this is failing to add
3007 * an ordered extent if the igrab() fails there and that is a huge pain
3008 * to deal with, so instead just hold onto the inode throughout the
3009 * writepages operation. If it fails here we are freeing up the inode
3010 * anyway and we'd rather not waste our time writing out stuff that is
3011 * going to be truncated anyway.
3016 pagevec_init(&pvec);
3017 if (wbc->range_cyclic) {
3018 index = mapping->writeback_index; /* Start from prev offset */
3021 * Start from the beginning does not need to cycle over the
3022 * range, mark it as scanned.
3024 scanned = (index == 0);
3026 index = wbc->range_start >> PAGE_SHIFT;
3027 end = wbc->range_end >> PAGE_SHIFT;
3028 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3034 * We do the tagged writepage as long as the snapshot flush bit is set
3035 * and we are the first one who do the filemap_flush() on this inode.
3037 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
3038 * not race in and drop the bit.
3040 if (range_whole && wbc->nr_to_write == LONG_MAX &&
3041 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
3042 &BTRFS_I(inode)->runtime_flags))
3043 wbc->tagged_writepages = 1;
3045 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3046 tag = PAGECACHE_TAG_TOWRITE;
3048 tag = PAGECACHE_TAG_DIRTY;
3050 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3051 tag_pages_for_writeback(mapping, index, end);
3053 while (!done && !nr_to_write_done && (index <= end) &&
3054 (nr_pages = pagevec_lookup_range_tag(&pvec, mapping,
3055 &index, end, tag))) {
3058 for (i = 0; i < nr_pages; i++) {
3059 struct page *page = pvec.pages[i];
3061 done_index = page->index + 1;
3063 * At this point we hold neither the i_pages lock nor
3064 * the page lock: the page may be truncated or
3065 * invalidated (changing page->mapping to NULL),
3066 * or even swizzled back from swapper_space to
3067 * tmpfs file mapping
3069 if (!trylock_page(page)) {
3070 submit_write_bio(bio_ctrl, 0);
3074 if (unlikely(page->mapping != mapping)) {
3079 if (wbc->sync_mode != WB_SYNC_NONE) {
3080 if (PageWriteback(page))
3081 submit_write_bio(bio_ctrl, 0);
3082 wait_on_page_writeback(page);
3085 if (PageWriteback(page) ||
3086 !clear_page_dirty_for_io(page)) {
3091 ret = __extent_writepage(page, wbc, bio_ctrl);
3098 * the filesystem may choose to bump up nr_to_write.
3099 * We have to make sure to honor the new nr_to_write
3102 nr_to_write_done = wbc->nr_to_write <= 0;
3104 pagevec_release(&pvec);
3107 if (!scanned && !done) {
3109 * We hit the last page and there is more work to be done: wrap
3110 * back to the start of the file
3116 * If we're looping we could run into a page that is locked by a
3117 * writer and that writer could be waiting on writeback for a
3118 * page in our current bio, and thus deadlock, so flush the
3121 submit_write_bio(bio_ctrl, 0);
3125 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
3126 mapping->writeback_index = done_index;
3128 btrfs_add_delayed_iput(BTRFS_I(inode));
3133 * Submit the pages in the range to bio for call sites which delalloc range has
3134 * already been ran (aka, ordered extent inserted) and all pages are still
3137 int extent_write_locked_range(struct inode *inode, u64 start, u64 end)
3139 bool found_error = false;
3140 int first_error = 0;
3142 struct address_space *mapping = inode->i_mapping;
3145 unsigned long nr_pages;
3146 const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize;
3147 struct btrfs_bio_ctrl bio_ctrl = {
3151 struct writeback_control wbc_writepages = {
3152 .sync_mode = WB_SYNC_ALL,
3153 .range_start = start,
3154 .range_end = end + 1,
3155 /* We're called from an async helper function */
3156 .punt_to_cgroup = 1,
3157 .no_cgroup_owner = 1,
3160 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
3161 nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >>
3163 wbc_writepages.nr_to_write = nr_pages * 2;
3165 wbc_attach_fdatawrite_inode(&wbc_writepages, inode);
3166 while (cur <= end) {
3167 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
3169 page = find_get_page(mapping, cur >> PAGE_SHIFT);
3171 * All pages in the range are locked since
3172 * btrfs_run_delalloc_range(), thus there is no way to clear
3173 * the page dirty flag.
3175 ASSERT(PageLocked(page));
3176 ASSERT(PageDirty(page));
3177 clear_page_dirty_for_io(page);
3178 ret = __extent_writepage(page, &wbc_writepages, &bio_ctrl);
3188 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
3190 wbc_detach_inode(&wbc_writepages);
3196 int extent_writepages(struct address_space *mapping,
3197 struct writeback_control *wbc)
3199 struct inode *inode = mapping->host;
3201 struct btrfs_bio_ctrl bio_ctrl = {
3203 .sync_io = (wbc->sync_mode == WB_SYNC_ALL),
3207 * Allow only a single thread to do the reloc work in zoned mode to
3208 * protect the write pointer updates.
3210 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
3211 ret = extent_write_cache_pages(mapping, wbc, &bio_ctrl);
3212 submit_write_bio(&bio_ctrl, ret);
3213 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
3217 void extent_readahead(struct readahead_control *rac)
3219 struct btrfs_bio_ctrl bio_ctrl = { 0 };
3220 struct page *pagepool[16];
3221 struct extent_map *em_cached = NULL;
3222 u64 prev_em_start = (u64)-1;
3225 while ((nr = readahead_page_batch(rac, pagepool))) {
3226 u64 contig_start = readahead_pos(rac);
3227 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
3229 contiguous_readpages(pagepool, nr, contig_start, contig_end,
3230 &em_cached, &bio_ctrl, &prev_em_start);
3234 free_extent_map(em_cached);
3235 submit_one_bio(&bio_ctrl);
3239 * basic invalidate_folio code, this waits on any locked or writeback
3240 * ranges corresponding to the folio, and then deletes any extent state
3241 * records from the tree
3243 int extent_invalidate_folio(struct extent_io_tree *tree,
3244 struct folio *folio, size_t offset)
3246 struct extent_state *cached_state = NULL;
3247 u64 start = folio_pos(folio);
3248 u64 end = start + folio_size(folio) - 1;
3249 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
3251 /* This function is only called for the btree inode */
3252 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
3254 start += ALIGN(offset, blocksize);
3258 lock_extent(tree, start, end, &cached_state);
3259 folio_wait_writeback(folio);
3262 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
3263 * so here we only need to unlock the extent range to free any
3264 * existing extent state.
3266 unlock_extent(tree, start, end, &cached_state);
3271 * a helper for release_folio, this tests for areas of the page that
3272 * are locked or under IO and drops the related state bits if it is safe
3275 static int try_release_extent_state(struct extent_io_tree *tree,
3276 struct page *page, gfp_t mask)
3278 u64 start = page_offset(page);
3279 u64 end = start + PAGE_SIZE - 1;
3282 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
3285 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
3286 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS);
3289 * At this point we can safely clear everything except the
3290 * locked bit, the nodatasum bit and the delalloc new bit.
3291 * The delalloc new bit will be cleared by ordered extent
3294 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL,
3297 /* if clear_extent_bit failed for enomem reasons,
3298 * we can't allow the release to continue.
3309 * a helper for release_folio. As long as there are no locked extents
3310 * in the range corresponding to the page, both state records and extent
3311 * map records are removed
3313 int try_release_extent_mapping(struct page *page, gfp_t mask)
3315 struct extent_map *em;
3316 u64 start = page_offset(page);
3317 u64 end = start + PAGE_SIZE - 1;
3318 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
3319 struct extent_io_tree *tree = &btrfs_inode->io_tree;
3320 struct extent_map_tree *map = &btrfs_inode->extent_tree;
3322 if (gfpflags_allow_blocking(mask) &&
3323 page->mapping->host->i_size > SZ_16M) {
3325 while (start <= end) {
3326 struct btrfs_fs_info *fs_info;
3329 len = end - start + 1;
3330 write_lock(&map->lock);
3331 em = lookup_extent_mapping(map, start, len);
3333 write_unlock(&map->lock);
3336 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
3337 em->start != start) {
3338 write_unlock(&map->lock);
3339 free_extent_map(em);
3342 if (test_range_bit(tree, em->start,
3343 extent_map_end(em) - 1,
3344 EXTENT_LOCKED, 0, NULL))
3347 * If it's not in the list of modified extents, used
3348 * by a fast fsync, we can remove it. If it's being
3349 * logged we can safely remove it since fsync took an
3350 * extra reference on the em.
3352 if (list_empty(&em->list) ||
3353 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
3356 * If it's in the list of modified extents, remove it
3357 * only if its generation is older then the current one,
3358 * in which case we don't need it for a fast fsync.
3359 * Otherwise don't remove it, we could be racing with an
3360 * ongoing fast fsync that could miss the new extent.
3362 fs_info = btrfs_inode->root->fs_info;
3363 spin_lock(&fs_info->trans_lock);
3364 cur_gen = fs_info->generation;
3365 spin_unlock(&fs_info->trans_lock);
3366 if (em->generation >= cur_gen)
3370 * We only remove extent maps that are not in the list of
3371 * modified extents or that are in the list but with a
3372 * generation lower then the current generation, so there
3373 * is no need to set the full fsync flag on the inode (it
3374 * hurts the fsync performance for workloads with a data
3375 * size that exceeds or is close to the system's memory).
3377 remove_extent_mapping(map, em);
3378 /* once for the rb tree */
3379 free_extent_map(em);
3381 start = extent_map_end(em);
3382 write_unlock(&map->lock);
3385 free_extent_map(em);
3387 cond_resched(); /* Allow large-extent preemption. */
3390 return try_release_extent_state(tree, page, mask);
3394 * To cache previous fiemap extent
3396 * Will be used for merging fiemap extent
3398 struct fiemap_cache {
3407 * Helper to submit fiemap extent.
3409 * Will try to merge current fiemap extent specified by @offset, @phys,
3410 * @len and @flags with cached one.
3411 * And only when we fails to merge, cached one will be submitted as
3414 * Return value is the same as fiemap_fill_next_extent().
3416 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
3417 struct fiemap_cache *cache,
3418 u64 offset, u64 phys, u64 len, u32 flags)
3422 /* Set at the end of extent_fiemap(). */
3423 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
3429 * Sanity check, extent_fiemap() should have ensured that new
3430 * fiemap extent won't overlap with cached one.
3433 * NOTE: Physical address can overlap, due to compression
3435 if (cache->offset + cache->len > offset) {
3441 * Only merges fiemap extents if
3442 * 1) Their logical addresses are continuous
3444 * 2) Their physical addresses are continuous
3445 * So truly compressed (physical size smaller than logical size)
3446 * extents won't get merged with each other
3448 * 3) Share same flags
3450 if (cache->offset + cache->len == offset &&
3451 cache->phys + cache->len == phys &&
3452 cache->flags == flags) {
3457 /* Not mergeable, need to submit cached one */
3458 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
3459 cache->len, cache->flags);
3460 cache->cached = false;
3464 cache->cached = true;
3465 cache->offset = offset;
3468 cache->flags = flags;
3474 * Emit last fiemap cache
3476 * The last fiemap cache may still be cached in the following case:
3478 * |<- Fiemap range ->|
3479 * |<------------ First extent ----------->|
3481 * In this case, the first extent range will be cached but not emitted.
3482 * So we must emit it before ending extent_fiemap().
3484 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
3485 struct fiemap_cache *cache)
3492 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
3493 cache->len, cache->flags);
3494 cache->cached = false;
3500 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
3502 struct extent_buffer *clone;
3503 struct btrfs_key key;
3508 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
3511 ret = btrfs_next_leaf(inode->root, path);
3516 * Don't bother with cloning if there are no more file extent items for
3519 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3520 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
3523 /* See the comment at fiemap_search_slot() about why we clone. */
3524 clone = btrfs_clone_extent_buffer(path->nodes[0]);
3528 slot = path->slots[0];
3529 btrfs_release_path(path);
3530 path->nodes[0] = clone;
3531 path->slots[0] = slot;
3537 * Search for the first file extent item that starts at a given file offset or
3538 * the one that starts immediately before that offset.
3539 * Returns: 0 on success, < 0 on error, 1 if not found.
3541 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
3544 const u64 ino = btrfs_ino(inode);
3545 struct btrfs_root *root = inode->root;
3546 struct extent_buffer *clone;
3547 struct btrfs_key key;
3552 key.type = BTRFS_EXTENT_DATA_KEY;
3553 key.offset = file_offset;
3555 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3559 if (ret > 0 && path->slots[0] > 0) {
3560 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3561 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3565 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3566 ret = btrfs_next_leaf(root, path);
3570 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3571 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3576 * We clone the leaf and use it during fiemap. This is because while
3577 * using the leaf we do expensive things like checking if an extent is
3578 * shared, which can take a long time. In order to prevent blocking
3579 * other tasks for too long, we use a clone of the leaf. We have locked
3580 * the file range in the inode's io tree, so we know none of our file
3581 * extent items can change. This way we avoid blocking other tasks that
3582 * want to insert items for other inodes in the same leaf or b+tree
3583 * rebalance operations (triggered for example when someone is trying
3584 * to push items into this leaf when trying to insert an item in a
3586 * We also need the private clone because holding a read lock on an
3587 * extent buffer of the subvolume's b+tree will make lockdep unhappy
3588 * when we call fiemap_fill_next_extent(), because that may cause a page
3589 * fault when filling the user space buffer with fiemap data.
3591 clone = btrfs_clone_extent_buffer(path->nodes[0]);
3595 slot = path->slots[0];
3596 btrfs_release_path(path);
3597 path->nodes[0] = clone;
3598 path->slots[0] = slot;
3604 * Process a range which is a hole or a prealloc extent in the inode's subvolume
3605 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
3606 * extent. The end offset (@end) is inclusive.
3608 static int fiemap_process_hole(struct btrfs_inode *inode,
3609 struct fiemap_extent_info *fieinfo,
3610 struct fiemap_cache *cache,
3611 struct extent_state **delalloc_cached_state,
3612 struct btrfs_backref_share_check_ctx *backref_ctx,
3613 u64 disk_bytenr, u64 extent_offset,
3617 const u64 i_size = i_size_read(&inode->vfs_inode);
3618 u64 cur_offset = start;
3619 u64 last_delalloc_end = 0;
3620 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
3621 bool checked_extent_shared = false;
3625 * There can be no delalloc past i_size, so don't waste time looking for
3628 while (cur_offset < end && cur_offset < i_size) {
3632 u64 prealloc_len = 0;
3635 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
3636 delalloc_cached_state,
3643 * If this is a prealloc extent we have to report every section
3644 * of it that has no delalloc.
3646 if (disk_bytenr != 0) {
3647 if (last_delalloc_end == 0) {
3648 prealloc_start = start;
3649 prealloc_len = delalloc_start - start;
3651 prealloc_start = last_delalloc_end + 1;
3652 prealloc_len = delalloc_start - prealloc_start;
3656 if (prealloc_len > 0) {
3657 if (!checked_extent_shared && fieinfo->fi_extents_max) {
3658 ret = btrfs_is_data_extent_shared(inode,
3665 prealloc_flags |= FIEMAP_EXTENT_SHARED;
3667 checked_extent_shared = true;
3669 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
3670 disk_bytenr + extent_offset,
3671 prealloc_len, prealloc_flags);
3674 extent_offset += prealloc_len;
3677 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
3678 delalloc_end + 1 - delalloc_start,
3679 FIEMAP_EXTENT_DELALLOC |
3680 FIEMAP_EXTENT_UNKNOWN);
3684 last_delalloc_end = delalloc_end;
3685 cur_offset = delalloc_end + 1;
3686 extent_offset += cur_offset - delalloc_start;
3691 * Either we found no delalloc for the whole prealloc extent or we have
3692 * a prealloc extent that spans i_size or starts at or after i_size.
3694 if (disk_bytenr != 0 && last_delalloc_end < end) {
3698 if (last_delalloc_end == 0) {
3699 prealloc_start = start;
3700 prealloc_len = end + 1 - start;
3702 prealloc_start = last_delalloc_end + 1;
3703 prealloc_len = end + 1 - prealloc_start;
3706 if (!checked_extent_shared && fieinfo->fi_extents_max) {
3707 ret = btrfs_is_data_extent_shared(inode,
3714 prealloc_flags |= FIEMAP_EXTENT_SHARED;
3716 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
3717 disk_bytenr + extent_offset,
3718 prealloc_len, prealloc_flags);
3726 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
3727 struct btrfs_path *path,
3728 u64 *last_extent_end_ret)
3730 const u64 ino = btrfs_ino(inode);
3731 struct btrfs_root *root = inode->root;
3732 struct extent_buffer *leaf;
3733 struct btrfs_file_extent_item *ei;
3734 struct btrfs_key key;
3739 * Lookup the last file extent. We're not using i_size here because
3740 * there might be preallocation past i_size.
3742 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
3743 /* There can't be a file extent item at offset (u64)-1 */
3749 * For a non-existing key, btrfs_search_slot() always leaves us at a
3750 * slot > 0, except if the btree is empty, which is impossible because
3751 * at least it has the inode item for this inode and all the items for
3752 * the root inode 256.
3754 ASSERT(path->slots[0] > 0);
3756 leaf = path->nodes[0];
3757 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3758 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
3759 /* No file extent items in the subvolume tree. */
3760 *last_extent_end_ret = 0;
3765 * For an inline extent, the disk_bytenr is where inline data starts at,
3766 * so first check if we have an inline extent item before checking if we
3767 * have an implicit hole (disk_bytenr == 0).
3769 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
3770 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
3771 *last_extent_end_ret = btrfs_file_extent_end(path);
3776 * Find the last file extent item that is not a hole (when NO_HOLES is
3777 * not enabled). This should take at most 2 iterations in the worst
3778 * case: we have one hole file extent item at slot 0 of a leaf and
3779 * another hole file extent item as the last item in the previous leaf.
3780 * This is because we merge file extent items that represent holes.
3782 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3783 while (disk_bytenr == 0) {
3784 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
3787 } else if (ret > 0) {
3788 /* No file extent items that are not holes. */
3789 *last_extent_end_ret = 0;
3792 leaf = path->nodes[0];
3793 ei = btrfs_item_ptr(leaf, path->slots[0],
3794 struct btrfs_file_extent_item);
3795 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3798 *last_extent_end_ret = btrfs_file_extent_end(path);
3802 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
3805 const u64 ino = btrfs_ino(inode);
3806 struct extent_state *cached_state = NULL;
3807 struct extent_state *delalloc_cached_state = NULL;
3808 struct btrfs_path *path;
3809 struct fiemap_cache cache = { 0 };
3810 struct btrfs_backref_share_check_ctx *backref_ctx;
3811 u64 last_extent_end;
3812 u64 prev_extent_end;
3815 bool stopped = false;
3818 backref_ctx = btrfs_alloc_backref_share_check_ctx();
3819 path = btrfs_alloc_path();
3820 if (!backref_ctx || !path) {
3825 lockstart = round_down(start, inode->root->fs_info->sectorsize);
3826 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
3827 prev_extent_end = lockstart;
3829 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3830 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3832 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
3835 btrfs_release_path(path);
3837 path->reada = READA_FORWARD;
3838 ret = fiemap_search_slot(inode, path, lockstart);
3841 } else if (ret > 0) {
3843 * No file extent item found, but we may have delalloc between
3844 * the current offset and i_size. So check for that.
3847 goto check_eof_delalloc;
3850 while (prev_extent_end < lockend) {
3851 struct extent_buffer *leaf = path->nodes[0];
3852 struct btrfs_file_extent_item *ei;
3853 struct btrfs_key key;
3856 u64 extent_offset = 0;
3858 u64 disk_bytenr = 0;
3863 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3864 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3867 extent_end = btrfs_file_extent_end(path);
3870 * The first iteration can leave us at an extent item that ends
3871 * before our range's start. Move to the next item.
3873 if (extent_end <= lockstart)
3876 backref_ctx->curr_leaf_bytenr = leaf->start;
3878 /* We have in implicit hole (NO_HOLES feature enabled). */
3879 if (prev_extent_end < key.offset) {
3880 const u64 range_end = min(key.offset, lockend) - 1;
3882 ret = fiemap_process_hole(inode, fieinfo, &cache,
3883 &delalloc_cached_state,
3884 backref_ctx, 0, 0, 0,
3885 prev_extent_end, range_end);
3888 } else if (ret > 0) {
3889 /* fiemap_fill_next_extent() told us to stop. */
3894 /* We've reached the end of the fiemap range, stop. */
3895 if (key.offset >= lockend) {
3901 extent_len = extent_end - key.offset;
3902 ei = btrfs_item_ptr(leaf, path->slots[0],
3903 struct btrfs_file_extent_item);
3904 compression = btrfs_file_extent_compression(leaf, ei);
3905 extent_type = btrfs_file_extent_type(leaf, ei);
3906 extent_gen = btrfs_file_extent_generation(leaf, ei);
3908 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3909 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
3910 if (compression == BTRFS_COMPRESS_NONE)
3911 extent_offset = btrfs_file_extent_offset(leaf, ei);
3914 if (compression != BTRFS_COMPRESS_NONE)
3915 flags |= FIEMAP_EXTENT_ENCODED;
3917 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3918 flags |= FIEMAP_EXTENT_DATA_INLINE;
3919 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
3920 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
3922 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
3923 ret = fiemap_process_hole(inode, fieinfo, &cache,
3924 &delalloc_cached_state,
3926 disk_bytenr, extent_offset,
3927 extent_gen, key.offset,
3929 } else if (disk_bytenr == 0) {
3930 /* We have an explicit hole. */
3931 ret = fiemap_process_hole(inode, fieinfo, &cache,
3932 &delalloc_cached_state,
3933 backref_ctx, 0, 0, 0,
3934 key.offset, extent_end - 1);
3936 /* We have a regular extent. */
3937 if (fieinfo->fi_extents_max) {
3938 ret = btrfs_is_data_extent_shared(inode,
3945 flags |= FIEMAP_EXTENT_SHARED;
3948 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
3949 disk_bytenr + extent_offset,
3955 } else if (ret > 0) {
3956 /* fiemap_fill_next_extent() told us to stop. */
3961 prev_extent_end = extent_end;
3963 if (fatal_signal_pending(current)) {
3968 ret = fiemap_next_leaf_item(inode, path);
3971 } else if (ret > 0) {
3972 /* No more file extent items for this inode. */
3980 * Release (and free) the path before emitting any final entries to
3981 * fiemap_fill_next_extent() to keep lockdep happy. This is because
3982 * once we find no more file extent items exist, we may have a
3983 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
3984 * faults when copying data to the user space buffer.
3986 btrfs_free_path(path);
3989 if (!stopped && prev_extent_end < lockend) {
3990 ret = fiemap_process_hole(inode, fieinfo, &cache,
3991 &delalloc_cached_state, backref_ctx,
3992 0, 0, 0, prev_extent_end, lockend - 1);
3995 prev_extent_end = lockend;
3998 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3999 const u64 i_size = i_size_read(&inode->vfs_inode);
4001 if (prev_extent_end < i_size) {
4006 delalloc = btrfs_find_delalloc_in_range(inode,
4009 &delalloc_cached_state,
4013 cache.flags |= FIEMAP_EXTENT_LAST;
4015 cache.flags |= FIEMAP_EXTENT_LAST;
4019 ret = emit_last_fiemap_cache(fieinfo, &cache);
4022 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
4023 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
4025 free_extent_state(delalloc_cached_state);
4026 btrfs_free_backref_share_ctx(backref_ctx);
4027 btrfs_free_path(path);
4031 static void __free_extent_buffer(struct extent_buffer *eb)
4033 kmem_cache_free(extent_buffer_cache, eb);
4036 int extent_buffer_under_io(const struct extent_buffer *eb)
4038 return (atomic_read(&eb->io_pages) ||
4039 test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
4040 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4043 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
4045 struct btrfs_subpage *subpage;
4047 lockdep_assert_held(&page->mapping->private_lock);
4049 if (PagePrivate(page)) {
4050 subpage = (struct btrfs_subpage *)page->private;
4051 if (atomic_read(&subpage->eb_refs))
4054 * Even there is no eb refs here, we may still have
4055 * end_page_read() call relying on page::private.
4057 if (atomic_read(&subpage->readers))
4063 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
4065 struct btrfs_fs_info *fs_info = eb->fs_info;
4066 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4069 * For mapped eb, we're going to change the page private, which should
4070 * be done under the private_lock.
4073 spin_lock(&page->mapping->private_lock);
4075 if (!PagePrivate(page)) {
4077 spin_unlock(&page->mapping->private_lock);
4081 if (fs_info->nodesize >= PAGE_SIZE) {
4083 * We do this since we'll remove the pages after we've
4084 * removed the eb from the radix tree, so we could race
4085 * and have this page now attached to the new eb. So
4086 * only clear page_private if it's still connected to
4089 if (PagePrivate(page) &&
4090 page->private == (unsigned long)eb) {
4091 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
4092 BUG_ON(PageDirty(page));
4093 BUG_ON(PageWriteback(page));
4095 * We need to make sure we haven't be attached
4098 detach_page_private(page);
4101 spin_unlock(&page->mapping->private_lock);
4106 * For subpage, we can have dummy eb with page private. In this case,
4107 * we can directly detach the private as such page is only attached to
4108 * one dummy eb, no sharing.
4111 btrfs_detach_subpage(fs_info, page);
4115 btrfs_page_dec_eb_refs(fs_info, page);
4118 * We can only detach the page private if there are no other ebs in the
4119 * page range and no unfinished IO.
4121 if (!page_range_has_eb(fs_info, page))
4122 btrfs_detach_subpage(fs_info, page);
4124 spin_unlock(&page->mapping->private_lock);
4127 /* Release all pages attached to the extent buffer */
4128 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
4133 ASSERT(!extent_buffer_under_io(eb));
4135 num_pages = num_extent_pages(eb);
4136 for (i = 0; i < num_pages; i++) {
4137 struct page *page = eb->pages[i];
4142 detach_extent_buffer_page(eb, page);
4144 /* One for when we allocated the page */
4150 * Helper for releasing the extent buffer.
4152 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
4154 btrfs_release_extent_buffer_pages(eb);
4155 btrfs_leak_debug_del_eb(eb);
4156 __free_extent_buffer(eb);
4159 static struct extent_buffer *
4160 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
4163 struct extent_buffer *eb = NULL;
4165 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
4168 eb->fs_info = fs_info;
4169 init_rwsem(&eb->lock);
4171 btrfs_leak_debug_add_eb(eb);
4172 INIT_LIST_HEAD(&eb->release_list);
4174 spin_lock_init(&eb->refs_lock);
4175 atomic_set(&eb->refs, 1);
4176 atomic_set(&eb->io_pages, 0);
4178 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
4183 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
4186 struct extent_buffer *new;
4187 int num_pages = num_extent_pages(src);
4190 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
4195 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
4196 * btrfs_release_extent_buffer() have different behavior for
4197 * UNMAPPED subpage extent buffer.
4199 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
4201 ret = btrfs_alloc_page_array(num_pages, new->pages);
4203 btrfs_release_extent_buffer(new);
4207 for (i = 0; i < num_pages; i++) {
4209 struct page *p = new->pages[i];
4211 ret = attach_extent_buffer_page(new, p, NULL);
4213 btrfs_release_extent_buffer(new);
4216 WARN_ON(PageDirty(p));
4217 copy_page(page_address(p), page_address(src->pages[i]));
4219 set_extent_buffer_uptodate(new);
4224 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4225 u64 start, unsigned long len)
4227 struct extent_buffer *eb;
4232 eb = __alloc_extent_buffer(fs_info, start, len);
4236 num_pages = num_extent_pages(eb);
4237 ret = btrfs_alloc_page_array(num_pages, eb->pages);
4241 for (i = 0; i < num_pages; i++) {
4242 struct page *p = eb->pages[i];
4244 ret = attach_extent_buffer_page(eb, p, NULL);
4249 set_extent_buffer_uptodate(eb);
4250 btrfs_set_header_nritems(eb, 0);
4251 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4255 for (i = 0; i < num_pages; i++) {
4257 detach_extent_buffer_page(eb, eb->pages[i]);
4258 __free_page(eb->pages[i]);
4261 __free_extent_buffer(eb);
4265 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
4268 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
4271 static void check_buffer_tree_ref(struct extent_buffer *eb)
4275 * The TREE_REF bit is first set when the extent_buffer is added
4276 * to the radix tree. It is also reset, if unset, when a new reference
4277 * is created by find_extent_buffer.
4279 * It is only cleared in two cases: freeing the last non-tree
4280 * reference to the extent_buffer when its STALE bit is set or
4281 * calling release_folio when the tree reference is the only reference.
4283 * In both cases, care is taken to ensure that the extent_buffer's
4284 * pages are not under io. However, release_folio can be concurrently
4285 * called with creating new references, which is prone to race
4286 * conditions between the calls to check_buffer_tree_ref in those
4287 * codepaths and clearing TREE_REF in try_release_extent_buffer.
4289 * The actual lifetime of the extent_buffer in the radix tree is
4290 * adequately protected by the refcount, but the TREE_REF bit and
4291 * its corresponding reference are not. To protect against this
4292 * class of races, we call check_buffer_tree_ref from the codepaths
4293 * which trigger io after they set eb->io_pages. Note that once io is
4294 * initiated, TREE_REF can no longer be cleared, so that is the
4295 * moment at which any such race is best fixed.
4297 refs = atomic_read(&eb->refs);
4298 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4301 spin_lock(&eb->refs_lock);
4302 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4303 atomic_inc(&eb->refs);
4304 spin_unlock(&eb->refs_lock);
4307 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
4308 struct page *accessed)
4312 check_buffer_tree_ref(eb);
4314 num_pages = num_extent_pages(eb);
4315 for (i = 0; i < num_pages; i++) {
4316 struct page *p = eb->pages[i];
4319 mark_page_accessed(p);
4323 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
4326 struct extent_buffer *eb;
4328 eb = find_extent_buffer_nolock(fs_info, start);
4332 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
4333 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
4334 * another task running free_extent_buffer() might have seen that flag
4335 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
4336 * writeback flags not set) and it's still in the tree (flag
4337 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
4338 * decrementing the extent buffer's reference count twice. So here we
4339 * could race and increment the eb's reference count, clear its stale
4340 * flag, mark it as dirty and drop our reference before the other task
4341 * finishes executing free_extent_buffer, which would later result in
4342 * an attempt to free an extent buffer that is dirty.
4344 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
4345 spin_lock(&eb->refs_lock);
4346 spin_unlock(&eb->refs_lock);
4348 mark_extent_buffer_accessed(eb, NULL);
4352 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4353 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
4356 struct extent_buffer *eb, *exists = NULL;
4359 eb = find_extent_buffer(fs_info, start);
4362 eb = alloc_dummy_extent_buffer(fs_info, start);
4364 return ERR_PTR(-ENOMEM);
4365 eb->fs_info = fs_info;
4367 ret = radix_tree_preload(GFP_NOFS);
4369 exists = ERR_PTR(ret);
4372 spin_lock(&fs_info->buffer_lock);
4373 ret = radix_tree_insert(&fs_info->buffer_radix,
4374 start >> fs_info->sectorsize_bits, eb);
4375 spin_unlock(&fs_info->buffer_lock);
4376 radix_tree_preload_end();
4377 if (ret == -EEXIST) {
4378 exists = find_extent_buffer(fs_info, start);
4384 check_buffer_tree_ref(eb);
4385 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4389 btrfs_release_extent_buffer(eb);
4394 static struct extent_buffer *grab_extent_buffer(
4395 struct btrfs_fs_info *fs_info, struct page *page)
4397 struct extent_buffer *exists;
4400 * For subpage case, we completely rely on radix tree to ensure we
4401 * don't try to insert two ebs for the same bytenr. So here we always
4402 * return NULL and just continue.
4404 if (fs_info->nodesize < PAGE_SIZE)
4407 /* Page not yet attached to an extent buffer */
4408 if (!PagePrivate(page))
4412 * We could have already allocated an eb for this page and attached one
4413 * so lets see if we can get a ref on the existing eb, and if we can we
4414 * know it's good and we can just return that one, else we know we can
4415 * just overwrite page->private.
4417 exists = (struct extent_buffer *)page->private;
4418 if (atomic_inc_not_zero(&exists->refs))
4421 WARN_ON(PageDirty(page));
4422 detach_page_private(page);
4426 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
4428 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
4429 btrfs_err(fs_info, "bad tree block start %llu", start);
4433 if (fs_info->nodesize < PAGE_SIZE &&
4434 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
4436 "tree block crosses page boundary, start %llu nodesize %u",
4437 start, fs_info->nodesize);
4440 if (fs_info->nodesize >= PAGE_SIZE &&
4441 !PAGE_ALIGNED(start)) {
4443 "tree block is not page aligned, start %llu nodesize %u",
4444 start, fs_info->nodesize);
4450 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
4451 u64 start, u64 owner_root, int level)
4453 unsigned long len = fs_info->nodesize;
4456 unsigned long index = start >> PAGE_SHIFT;
4457 struct extent_buffer *eb;
4458 struct extent_buffer *exists = NULL;
4460 struct address_space *mapping = fs_info->btree_inode->i_mapping;
4461 u64 lockdep_owner = owner_root;
4465 if (check_eb_alignment(fs_info, start))
4466 return ERR_PTR(-EINVAL);
4468 #if BITS_PER_LONG == 32
4469 if (start >= MAX_LFS_FILESIZE) {
4470 btrfs_err_rl(fs_info,
4471 "extent buffer %llu is beyond 32bit page cache limit", start);
4472 btrfs_err_32bit_limit(fs_info);
4473 return ERR_PTR(-EOVERFLOW);
4475 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
4476 btrfs_warn_32bit_limit(fs_info);
4479 eb = find_extent_buffer(fs_info, start);
4483 eb = __alloc_extent_buffer(fs_info, start, len);
4485 return ERR_PTR(-ENOMEM);
4488 * The reloc trees are just snapshots, so we need them to appear to be
4489 * just like any other fs tree WRT lockdep.
4491 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
4492 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4494 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
4496 num_pages = num_extent_pages(eb);
4497 for (i = 0; i < num_pages; i++, index++) {
4498 struct btrfs_subpage *prealloc = NULL;
4500 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
4502 exists = ERR_PTR(-ENOMEM);
4507 * Preallocate page->private for subpage case, so that we won't
4508 * allocate memory with private_lock hold. The memory will be
4509 * freed by attach_extent_buffer_page() or freed manually if
4512 * Although we have ensured one subpage eb can only have one
4513 * page, but it may change in the future for 16K page size
4514 * support, so we still preallocate the memory in the loop.
4516 if (fs_info->nodesize < PAGE_SIZE) {
4517 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
4518 if (IS_ERR(prealloc)) {
4519 ret = PTR_ERR(prealloc);
4522 exists = ERR_PTR(ret);
4527 spin_lock(&mapping->private_lock);
4528 exists = grab_extent_buffer(fs_info, p);
4530 spin_unlock(&mapping->private_lock);
4533 mark_extent_buffer_accessed(exists, p);
4534 btrfs_free_subpage(prealloc);
4537 /* Should not fail, as we have preallocated the memory */
4538 ret = attach_extent_buffer_page(eb, p, prealloc);
4541 * To inform we have extra eb under allocation, so that
4542 * detach_extent_buffer_page() won't release the page private
4543 * when the eb hasn't yet been inserted into radix tree.
4545 * The ref will be decreased when the eb released the page, in
4546 * detach_extent_buffer_page().
4547 * Thus needs no special handling in error path.
4549 btrfs_page_inc_eb_refs(fs_info, p);
4550 spin_unlock(&mapping->private_lock);
4552 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
4554 if (!PageUptodate(p))
4558 * We can't unlock the pages just yet since the extent buffer
4559 * hasn't been properly inserted in the radix tree, this
4560 * opens a race with btree_release_folio which can free a page
4561 * while we are still filling in all pages for the buffer and
4566 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4568 ret = radix_tree_preload(GFP_NOFS);
4570 exists = ERR_PTR(ret);
4574 spin_lock(&fs_info->buffer_lock);
4575 ret = radix_tree_insert(&fs_info->buffer_radix,
4576 start >> fs_info->sectorsize_bits, eb);
4577 spin_unlock(&fs_info->buffer_lock);
4578 radix_tree_preload_end();
4579 if (ret == -EEXIST) {
4580 exists = find_extent_buffer(fs_info, start);
4586 /* add one reference for the tree */
4587 check_buffer_tree_ref(eb);
4588 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
4591 * Now it's safe to unlock the pages because any calls to
4592 * btree_release_folio will correctly detect that a page belongs to a
4593 * live buffer and won't free them prematurely.
4595 for (i = 0; i < num_pages; i++)
4596 unlock_page(eb->pages[i]);
4600 WARN_ON(!atomic_dec_and_test(&eb->refs));
4601 for (i = 0; i < num_pages; i++) {
4603 unlock_page(eb->pages[i]);
4606 btrfs_release_extent_buffer(eb);
4610 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
4612 struct extent_buffer *eb =
4613 container_of(head, struct extent_buffer, rcu_head);
4615 __free_extent_buffer(eb);
4618 static int release_extent_buffer(struct extent_buffer *eb)
4619 __releases(&eb->refs_lock)
4621 lockdep_assert_held(&eb->refs_lock);
4623 WARN_ON(atomic_read(&eb->refs) == 0);
4624 if (atomic_dec_and_test(&eb->refs)) {
4625 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
4626 struct btrfs_fs_info *fs_info = eb->fs_info;
4628 spin_unlock(&eb->refs_lock);
4630 spin_lock(&fs_info->buffer_lock);
4631 radix_tree_delete(&fs_info->buffer_radix,
4632 eb->start >> fs_info->sectorsize_bits);
4633 spin_unlock(&fs_info->buffer_lock);
4635 spin_unlock(&eb->refs_lock);
4638 btrfs_leak_debug_del_eb(eb);
4639 /* Should be safe to release our pages at this point */
4640 btrfs_release_extent_buffer_pages(eb);
4641 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4642 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
4643 __free_extent_buffer(eb);
4647 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
4650 spin_unlock(&eb->refs_lock);
4655 void free_extent_buffer(struct extent_buffer *eb)
4661 refs = atomic_read(&eb->refs);
4663 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
4664 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
4667 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
4671 spin_lock(&eb->refs_lock);
4672 if (atomic_read(&eb->refs) == 2 &&
4673 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
4674 !extent_buffer_under_io(eb) &&
4675 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4676 atomic_dec(&eb->refs);
4679 * I know this is terrible, but it's temporary until we stop tracking
4680 * the uptodate bits and such for the extent buffers.
4682 release_extent_buffer(eb);
4685 void free_extent_buffer_stale(struct extent_buffer *eb)
4690 spin_lock(&eb->refs_lock);
4691 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
4693 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
4694 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
4695 atomic_dec(&eb->refs);
4696 release_extent_buffer(eb);
4699 static void btree_clear_page_dirty(struct page *page)
4701 ASSERT(PageDirty(page));
4702 ASSERT(PageLocked(page));
4703 clear_page_dirty_for_io(page);
4704 xa_lock_irq(&page->mapping->i_pages);
4705 if (!PageDirty(page))
4706 __xa_clear_mark(&page->mapping->i_pages,
4707 page_index(page), PAGECACHE_TAG_DIRTY);
4708 xa_unlock_irq(&page->mapping->i_pages);
4711 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
4713 struct btrfs_fs_info *fs_info = eb->fs_info;
4714 struct page *page = eb->pages[0];
4717 /* btree_clear_page_dirty() needs page locked */
4719 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
4722 btree_clear_page_dirty(page);
4724 WARN_ON(atomic_read(&eb->refs) == 0);
4727 void clear_extent_buffer_dirty(const struct extent_buffer *eb)
4733 if (eb->fs_info->nodesize < PAGE_SIZE)
4734 return clear_subpage_extent_buffer_dirty(eb);
4736 num_pages = num_extent_pages(eb);
4738 for (i = 0; i < num_pages; i++) {
4739 page = eb->pages[i];
4740 if (!PageDirty(page))
4743 btree_clear_page_dirty(page);
4744 ClearPageError(page);
4747 WARN_ON(atomic_read(&eb->refs) == 0);
4750 bool set_extent_buffer_dirty(struct extent_buffer *eb)
4756 check_buffer_tree_ref(eb);
4758 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
4760 num_pages = num_extent_pages(eb);
4761 WARN_ON(atomic_read(&eb->refs) == 0);
4762 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
4765 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
4768 * For subpage case, we can have other extent buffers in the
4769 * same page, and in clear_subpage_extent_buffer_dirty() we
4770 * have to clear page dirty without subpage lock held.
4771 * This can cause race where our page gets dirty cleared after
4774 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
4775 * its page for other reasons, we can use page lock to prevent
4779 lock_page(eb->pages[0]);
4780 for (i = 0; i < num_pages; i++)
4781 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
4782 eb->start, eb->len);
4784 unlock_page(eb->pages[0]);
4786 #ifdef CONFIG_BTRFS_DEBUG
4787 for (i = 0; i < num_pages; i++)
4788 ASSERT(PageDirty(eb->pages[i]));
4794 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
4796 struct btrfs_fs_info *fs_info = eb->fs_info;
4801 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4802 num_pages = num_extent_pages(eb);
4803 for (i = 0; i < num_pages; i++) {
4804 page = eb->pages[i];
4809 * This is special handling for metadata subpage, as regular
4810 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4812 if (fs_info->nodesize >= PAGE_SIZE)
4813 ClearPageUptodate(page);
4815 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
4820 void set_extent_buffer_uptodate(struct extent_buffer *eb)
4822 struct btrfs_fs_info *fs_info = eb->fs_info;
4827 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4828 num_pages = num_extent_pages(eb);
4829 for (i = 0; i < num_pages; i++) {
4830 page = eb->pages[i];
4833 * This is special handling for metadata subpage, as regular
4834 * btrfs_is_subpage() can not handle cloned/dummy metadata.
4836 if (fs_info->nodesize >= PAGE_SIZE)
4837 SetPageUptodate(page);
4839 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
4844 static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait,
4846 struct btrfs_tree_parent_check *check)
4848 struct btrfs_fs_info *fs_info = eb->fs_info;
4849 struct extent_io_tree *io_tree;
4850 struct page *page = eb->pages[0];
4851 struct extent_state *cached_state = NULL;
4852 struct btrfs_bio_ctrl bio_ctrl = {
4853 .mirror_num = mirror_num,
4854 .parent_check = check,
4858 ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags));
4859 ASSERT(PagePrivate(page));
4861 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
4863 if (wait == WAIT_NONE) {
4864 if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1,
4868 ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1,
4875 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) ||
4876 PageUptodate(page) ||
4877 btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) {
4878 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4879 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
4884 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4885 eb->read_mirror = 0;
4886 atomic_set(&eb->io_pages, 1);
4887 check_buffer_tree_ref(eb);
4888 bio_ctrl.end_io_func = end_bio_extent_readpage;
4890 btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len);
4892 btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len);
4893 ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl,
4894 eb->start, page, eb->len,
4895 eb->start - page_offset(page), 0, true);
4898 * In the endio function, if we hit something wrong we will
4899 * increase the io_pages, so here we need to decrease it for
4902 atomic_dec(&eb->io_pages);
4904 submit_one_bio(&bio_ctrl);
4905 if (ret || wait != WAIT_COMPLETE) {
4906 free_extent_state(cached_state);
4910 wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1,
4911 EXTENT_LOCKED, &cached_state);
4912 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4917 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
4918 struct btrfs_tree_parent_check *check)
4924 int locked_pages = 0;
4925 int all_uptodate = 1;
4927 unsigned long num_reads = 0;
4928 struct btrfs_bio_ctrl bio_ctrl = {
4929 .mirror_num = mirror_num,
4930 .parent_check = check,
4933 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
4937 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
4938 * operation, which could potentially still be in flight. In this case
4939 * we simply want to return an error.
4941 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
4944 if (eb->fs_info->nodesize < PAGE_SIZE)
4945 return read_extent_buffer_subpage(eb, wait, mirror_num, check);
4947 num_pages = num_extent_pages(eb);
4948 for (i = 0; i < num_pages; i++) {
4949 page = eb->pages[i];
4950 if (wait == WAIT_NONE) {
4952 * WAIT_NONE is only utilized by readahead. If we can't
4953 * acquire the lock atomically it means either the eb
4954 * is being read out or under modification.
4955 * Either way the eb will be or has been cached,
4956 * readahead can exit safely.
4958 if (!trylock_page(page))
4966 * We need to firstly lock all pages to make sure that
4967 * the uptodate bit of our pages won't be affected by
4968 * clear_extent_buffer_uptodate().
4970 for (i = 0; i < num_pages; i++) {
4971 page = eb->pages[i];
4972 if (!PageUptodate(page)) {
4979 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
4983 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
4984 eb->read_mirror = 0;
4985 atomic_set(&eb->io_pages, num_reads);
4987 * It is possible for release_folio to clear the TREE_REF bit before we
4988 * set io_pages. See check_buffer_tree_ref for a more detailed comment.
4990 check_buffer_tree_ref(eb);
4991 bio_ctrl.end_io_func = end_bio_extent_readpage;
4992 for (i = 0; i < num_pages; i++) {
4993 page = eb->pages[i];
4995 if (!PageUptodate(page)) {
4997 atomic_dec(&eb->io_pages);
5002 ClearPageError(page);
5003 err = submit_extent_page(REQ_OP_READ, NULL,
5004 &bio_ctrl, page_offset(page), page,
5005 PAGE_SIZE, 0, 0, false);
5008 * We failed to submit the bio so it's the
5009 * caller's responsibility to perform cleanup
5010 * i.e unlock page/set error bit.
5015 atomic_dec(&eb->io_pages);
5022 submit_one_bio(&bio_ctrl);
5024 if (ret || wait != WAIT_COMPLETE)
5027 for (i = 0; i < num_pages; i++) {
5028 page = eb->pages[i];
5029 wait_on_page_locked(page);
5030 if (!PageUptodate(page))
5037 while (locked_pages > 0) {
5039 page = eb->pages[locked_pages];
5045 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
5048 btrfs_warn(eb->fs_info,
5049 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
5050 eb->start, eb->len, start, len);
5051 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5057 * Check if the [start, start + len) range is valid before reading/writing
5059 * NOTE: @start and @len are offset inside the eb, not logical address.
5061 * Caller should not touch the dst/src memory if this function returns error.
5063 static inline int check_eb_range(const struct extent_buffer *eb,
5064 unsigned long start, unsigned long len)
5066 unsigned long offset;
5068 /* start, start + len should not go beyond eb->len nor overflow */
5069 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
5070 return report_eb_range(eb, start, len);
5075 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
5076 unsigned long start, unsigned long len)
5082 char *dst = (char *)dstv;
5083 unsigned long i = get_eb_page_index(start);
5085 if (check_eb_range(eb, start, len))
5088 offset = get_eb_offset_in_page(eb, start);
5091 page = eb->pages[i];
5093 cur = min(len, (PAGE_SIZE - offset));
5094 kaddr = page_address(page);
5095 memcpy(dst, kaddr + offset, cur);
5104 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
5106 unsigned long start, unsigned long len)
5112 char __user *dst = (char __user *)dstv;
5113 unsigned long i = get_eb_page_index(start);
5116 WARN_ON(start > eb->len);
5117 WARN_ON(start + len > eb->start + eb->len);
5119 offset = get_eb_offset_in_page(eb, start);
5122 page = eb->pages[i];
5124 cur = min(len, (PAGE_SIZE - offset));
5125 kaddr = page_address(page);
5126 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
5140 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
5141 unsigned long start, unsigned long len)
5147 char *ptr = (char *)ptrv;
5148 unsigned long i = get_eb_page_index(start);
5151 if (check_eb_range(eb, start, len))
5154 offset = get_eb_offset_in_page(eb, start);
5157 page = eb->pages[i];
5159 cur = min(len, (PAGE_SIZE - offset));
5161 kaddr = page_address(page);
5162 ret = memcmp(ptr, kaddr + offset, cur);
5175 * Check that the extent buffer is uptodate.
5177 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
5178 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
5180 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
5183 struct btrfs_fs_info *fs_info = eb->fs_info;
5186 * If we are using the commit root we could potentially clear a page
5187 * Uptodate while we're using the extent buffer that we've previously
5188 * looked up. We don't want to complain in this case, as the page was
5189 * valid before, we just didn't write it out. Instead we want to catch
5190 * the case where we didn't actually read the block properly, which
5191 * would have !PageUptodate && !PageError, as we clear PageError before
5194 if (fs_info->nodesize < PAGE_SIZE) {
5195 bool uptodate, error;
5197 uptodate = btrfs_subpage_test_uptodate(fs_info, page,
5198 eb->start, eb->len);
5199 error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len);
5200 WARN_ON(!uptodate && !error);
5202 WARN_ON(!PageUptodate(page) && !PageError(page));
5206 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
5211 assert_eb_page_uptodate(eb, eb->pages[0]);
5212 kaddr = page_address(eb->pages[0]) +
5213 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
5215 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
5218 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
5222 assert_eb_page_uptodate(eb, eb->pages[0]);
5223 kaddr = page_address(eb->pages[0]) +
5224 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
5225 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
5228 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
5229 unsigned long start, unsigned long len)
5235 char *src = (char *)srcv;
5236 unsigned long i = get_eb_page_index(start);
5238 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
5240 if (check_eb_range(eb, start, len))
5243 offset = get_eb_offset_in_page(eb, start);
5246 page = eb->pages[i];
5247 assert_eb_page_uptodate(eb, page);
5249 cur = min(len, PAGE_SIZE - offset);
5250 kaddr = page_address(page);
5251 memcpy(kaddr + offset, src, cur);
5260 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
5267 unsigned long i = get_eb_page_index(start);
5269 if (check_eb_range(eb, start, len))
5272 offset = get_eb_offset_in_page(eb, start);
5275 page = eb->pages[i];
5276 assert_eb_page_uptodate(eb, page);
5278 cur = min(len, PAGE_SIZE - offset);
5279 kaddr = page_address(page);
5280 memset(kaddr + offset, 0, cur);
5288 void copy_extent_buffer_full(const struct extent_buffer *dst,
5289 const struct extent_buffer *src)
5294 ASSERT(dst->len == src->len);
5296 if (dst->fs_info->nodesize >= PAGE_SIZE) {
5297 num_pages = num_extent_pages(dst);
5298 for (i = 0; i < num_pages; i++)
5299 copy_page(page_address(dst->pages[i]),
5300 page_address(src->pages[i]));
5302 size_t src_offset = get_eb_offset_in_page(src, 0);
5303 size_t dst_offset = get_eb_offset_in_page(dst, 0);
5305 ASSERT(src->fs_info->nodesize < PAGE_SIZE);
5306 memcpy(page_address(dst->pages[0]) + dst_offset,
5307 page_address(src->pages[0]) + src_offset,
5312 void copy_extent_buffer(const struct extent_buffer *dst,
5313 const struct extent_buffer *src,
5314 unsigned long dst_offset, unsigned long src_offset,
5317 u64 dst_len = dst->len;
5322 unsigned long i = get_eb_page_index(dst_offset);
5324 if (check_eb_range(dst, dst_offset, len) ||
5325 check_eb_range(src, src_offset, len))
5328 WARN_ON(src->len != dst_len);
5330 offset = get_eb_offset_in_page(dst, dst_offset);
5333 page = dst->pages[i];
5334 assert_eb_page_uptodate(dst, page);
5336 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
5338 kaddr = page_address(page);
5339 read_extent_buffer(src, kaddr + offset, src_offset, cur);
5349 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
5351 * @eb: the extent buffer
5352 * @start: offset of the bitmap item in the extent buffer
5354 * @page_index: return index of the page in the extent buffer that contains the
5356 * @page_offset: return offset into the page given by page_index
5358 * This helper hides the ugliness of finding the byte in an extent buffer which
5359 * contains a given bit.
5361 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
5362 unsigned long start, unsigned long nr,
5363 unsigned long *page_index,
5364 size_t *page_offset)
5366 size_t byte_offset = BIT_BYTE(nr);
5370 * The byte we want is the offset of the extent buffer + the offset of
5371 * the bitmap item in the extent buffer + the offset of the byte in the
5374 offset = start + offset_in_page(eb->start) + byte_offset;
5376 *page_index = offset >> PAGE_SHIFT;
5377 *page_offset = offset_in_page(offset);
5381 * Determine whether a bit in a bitmap item is set.
5383 * @eb: the extent buffer
5384 * @start: offset of the bitmap item in the extent buffer
5385 * @nr: bit number to test
5387 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
5395 eb_bitmap_offset(eb, start, nr, &i, &offset);
5396 page = eb->pages[i];
5397 assert_eb_page_uptodate(eb, page);
5398 kaddr = page_address(page);
5399 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
5403 * Set an area of a bitmap to 1.
5405 * @eb: the extent buffer
5406 * @start: offset of the bitmap item in the extent buffer
5407 * @pos: bit number of the first bit
5408 * @len: number of bits to set
5410 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
5411 unsigned long pos, unsigned long len)
5417 const unsigned int size = pos + len;
5418 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5419 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
5421 eb_bitmap_offset(eb, start, pos, &i, &offset);
5422 page = eb->pages[i];
5423 assert_eb_page_uptodate(eb, page);
5424 kaddr = page_address(page);
5426 while (len >= bits_to_set) {
5427 kaddr[offset] |= mask_to_set;
5429 bits_to_set = BITS_PER_BYTE;
5431 if (++offset >= PAGE_SIZE && len > 0) {
5433 page = eb->pages[++i];
5434 assert_eb_page_uptodate(eb, page);
5435 kaddr = page_address(page);
5439 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
5440 kaddr[offset] |= mask_to_set;
5446 * Clear an area of a bitmap.
5448 * @eb: the extent buffer
5449 * @start: offset of the bitmap item in the extent buffer
5450 * @pos: bit number of the first bit
5451 * @len: number of bits to clear
5453 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
5454 unsigned long start, unsigned long pos,
5461 const unsigned int size = pos + len;
5462 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
5463 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
5465 eb_bitmap_offset(eb, start, pos, &i, &offset);
5466 page = eb->pages[i];
5467 assert_eb_page_uptodate(eb, page);
5468 kaddr = page_address(page);
5470 while (len >= bits_to_clear) {
5471 kaddr[offset] &= ~mask_to_clear;
5472 len -= bits_to_clear;
5473 bits_to_clear = BITS_PER_BYTE;
5475 if (++offset >= PAGE_SIZE && len > 0) {
5477 page = eb->pages[++i];
5478 assert_eb_page_uptodate(eb, page);
5479 kaddr = page_address(page);
5483 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
5484 kaddr[offset] &= ~mask_to_clear;
5488 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
5490 unsigned long distance = (src > dst) ? src - dst : dst - src;
5491 return distance < len;
5494 static void copy_pages(struct page *dst_page, struct page *src_page,
5495 unsigned long dst_off, unsigned long src_off,
5498 char *dst_kaddr = page_address(dst_page);
5500 int must_memmove = 0;
5502 if (dst_page != src_page) {
5503 src_kaddr = page_address(src_page);
5505 src_kaddr = dst_kaddr;
5506 if (areas_overlap(src_off, dst_off, len))
5511 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
5513 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
5516 void memcpy_extent_buffer(const struct extent_buffer *dst,
5517 unsigned long dst_offset, unsigned long src_offset,
5521 size_t dst_off_in_page;
5522 size_t src_off_in_page;
5523 unsigned long dst_i;
5524 unsigned long src_i;
5526 if (check_eb_range(dst, dst_offset, len) ||
5527 check_eb_range(dst, src_offset, len))
5531 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
5532 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
5534 dst_i = get_eb_page_index(dst_offset);
5535 src_i = get_eb_page_index(src_offset);
5537 cur = min(len, (unsigned long)(PAGE_SIZE -
5539 cur = min_t(unsigned long, cur,
5540 (unsigned long)(PAGE_SIZE - dst_off_in_page));
5542 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5543 dst_off_in_page, src_off_in_page, cur);
5551 void memmove_extent_buffer(const struct extent_buffer *dst,
5552 unsigned long dst_offset, unsigned long src_offset,
5556 size_t dst_off_in_page;
5557 size_t src_off_in_page;
5558 unsigned long dst_end = dst_offset + len - 1;
5559 unsigned long src_end = src_offset + len - 1;
5560 unsigned long dst_i;
5561 unsigned long src_i;
5563 if (check_eb_range(dst, dst_offset, len) ||
5564 check_eb_range(dst, src_offset, len))
5566 if (dst_offset < src_offset) {
5567 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
5571 dst_i = get_eb_page_index(dst_end);
5572 src_i = get_eb_page_index(src_end);
5574 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
5575 src_off_in_page = get_eb_offset_in_page(dst, src_end);
5577 cur = min_t(unsigned long, len, src_off_in_page + 1);
5578 cur = min(cur, dst_off_in_page + 1);
5579 copy_pages(dst->pages[dst_i], dst->pages[src_i],
5580 dst_off_in_page - cur + 1,
5581 src_off_in_page - cur + 1, cur);
5589 #define GANG_LOOKUP_SIZE 16
5590 static struct extent_buffer *get_next_extent_buffer(
5591 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
5593 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
5594 struct extent_buffer *found = NULL;
5595 u64 page_start = page_offset(page);
5596 u64 cur = page_start;
5598 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
5599 lockdep_assert_held(&fs_info->buffer_lock);
5601 while (cur < page_start + PAGE_SIZE) {
5605 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
5606 (void **)gang, cur >> fs_info->sectorsize_bits,
5607 min_t(unsigned int, GANG_LOOKUP_SIZE,
5608 PAGE_SIZE / fs_info->nodesize));
5611 for (i = 0; i < ret; i++) {
5612 /* Already beyond page end */
5613 if (gang[i]->start >= page_start + PAGE_SIZE)
5616 if (gang[i]->start >= bytenr) {
5621 cur = gang[ret - 1]->start + gang[ret - 1]->len;
5627 static int try_release_subpage_extent_buffer(struct page *page)
5629 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
5630 u64 cur = page_offset(page);
5631 const u64 end = page_offset(page) + PAGE_SIZE;
5635 struct extent_buffer *eb = NULL;
5638 * Unlike try_release_extent_buffer() which uses page->private
5639 * to grab buffer, for subpage case we rely on radix tree, thus
5640 * we need to ensure radix tree consistency.
5642 * We also want an atomic snapshot of the radix tree, thus go
5643 * with spinlock rather than RCU.
5645 spin_lock(&fs_info->buffer_lock);
5646 eb = get_next_extent_buffer(fs_info, page, cur);
5648 /* No more eb in the page range after or at cur */
5649 spin_unlock(&fs_info->buffer_lock);
5652 cur = eb->start + eb->len;
5655 * The same as try_release_extent_buffer(), to ensure the eb
5656 * won't disappear out from under us.
5658 spin_lock(&eb->refs_lock);
5659 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5660 spin_unlock(&eb->refs_lock);
5661 spin_unlock(&fs_info->buffer_lock);
5664 spin_unlock(&fs_info->buffer_lock);
5667 * If tree ref isn't set then we know the ref on this eb is a
5668 * real ref, so just return, this eb will likely be freed soon
5671 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5672 spin_unlock(&eb->refs_lock);
5677 * Here we don't care about the return value, we will always
5678 * check the page private at the end. And
5679 * release_extent_buffer() will release the refs_lock.
5681 release_extent_buffer(eb);
5684 * Finally to check if we have cleared page private, as if we have
5685 * released all ebs in the page, the page private should be cleared now.
5687 spin_lock(&page->mapping->private_lock);
5688 if (!PagePrivate(page))
5692 spin_unlock(&page->mapping->private_lock);
5697 int try_release_extent_buffer(struct page *page)
5699 struct extent_buffer *eb;
5701 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
5702 return try_release_subpage_extent_buffer(page);
5705 * We need to make sure nobody is changing page->private, as we rely on
5706 * page->private as the pointer to extent buffer.
5708 spin_lock(&page->mapping->private_lock);
5709 if (!PagePrivate(page)) {
5710 spin_unlock(&page->mapping->private_lock);
5714 eb = (struct extent_buffer *)page->private;
5718 * This is a little awful but should be ok, we need to make sure that
5719 * the eb doesn't disappear out from under us while we're looking at
5722 spin_lock(&eb->refs_lock);
5723 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
5724 spin_unlock(&eb->refs_lock);
5725 spin_unlock(&page->mapping->private_lock);
5728 spin_unlock(&page->mapping->private_lock);
5731 * If tree ref isn't set then we know the ref on this eb is a real ref,
5732 * so just return, this page will likely be freed soon anyway.
5734 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
5735 spin_unlock(&eb->refs_lock);
5739 return release_extent_buffer(eb);
5743 * btrfs_readahead_tree_block - attempt to readahead a child block
5744 * @fs_info: the fs_info
5745 * @bytenr: bytenr to read
5746 * @owner_root: objectid of the root that owns this eb
5747 * @gen: generation for the uptodate check, can be 0
5748 * @level: level for the eb
5750 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
5751 * normal uptodate check of the eb, without checking the generation. If we have
5752 * to read the block we will not block on anything.
5754 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
5755 u64 bytenr, u64 owner_root, u64 gen, int level)
5757 struct btrfs_tree_parent_check check = {
5762 struct extent_buffer *eb;
5765 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
5769 if (btrfs_buffer_uptodate(eb, gen, 1)) {
5770 free_extent_buffer(eb);
5774 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
5776 free_extent_buffer_stale(eb);
5778 free_extent_buffer(eb);
5782 * btrfs_readahead_node_child - readahead a node's child block
5783 * @node: parent node we're reading from
5784 * @slot: slot in the parent node for the child we want to read
5786 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
5787 * the slot in the node provided.
5789 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
5791 btrfs_readahead_tree_block(node->fs_info,
5792 btrfs_node_blockptr(node, slot),
5793 btrfs_header_owner(node),
5794 btrfs_node_ptr_generation(node, slot),
5795 btrfs_header_level(node) - 1);