1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/sched/mm.h>
10 #include <linux/spinlock.h>
11 #include <linux/blkdev.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/prefetch.h>
16 #include <linux/fsverity.h>
18 #include "extent_io.h"
19 #include "extent-io-tree.h"
20 #include "extent_map.h"
22 #include "btrfs_inode.h"
25 #include "rcu-string.h"
30 #include "block-group.h"
31 #include "compression.h"
33 #include "accessors.h"
34 #include "file-item.h"
36 #include "dev-replace.h"
38 #include "transaction.h"
40 static struct kmem_cache *extent_buffer_cache;
42 #ifdef CONFIG_BTRFS_DEBUG
43 static inline void btrfs_leak_debug_add_eb(struct extent_buffer *eb)
45 struct btrfs_fs_info *fs_info = eb->fs_info;
48 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
49 list_add(&eb->leak_list, &fs_info->allocated_ebs);
50 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
53 static inline void btrfs_leak_debug_del_eb(struct extent_buffer *eb)
55 struct btrfs_fs_info *fs_info = eb->fs_info;
58 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
59 list_del(&eb->leak_list);
60 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
63 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
65 struct extent_buffer *eb;
69 * If we didn't get into open_ctree our allocated_ebs will not be
70 * initialized, so just skip this.
72 if (!fs_info->allocated_ebs.next)
75 WARN_ON(!list_empty(&fs_info->allocated_ebs));
76 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
77 while (!list_empty(&fs_info->allocated_ebs)) {
78 eb = list_first_entry(&fs_info->allocated_ebs,
79 struct extent_buffer, leak_list);
81 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
82 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
83 btrfs_header_owner(eb));
84 list_del(&eb->leak_list);
85 kmem_cache_free(extent_buffer_cache, eb);
87 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
90 #define btrfs_leak_debug_add_eb(eb) do {} while (0)
91 #define btrfs_leak_debug_del_eb(eb) do {} while (0)
95 * Structure to record info about the bio being assembled, and other info like
96 * how many bytes are there before stripe/ordered extent boundary.
98 struct btrfs_bio_ctrl {
99 struct btrfs_bio *bbio;
100 enum btrfs_compression_type compress_type;
101 u32 len_to_oe_boundary;
103 btrfs_bio_end_io_t end_io_func;
104 struct writeback_control *wbc;
107 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
109 struct btrfs_bio *bbio = bio_ctrl->bbio;
114 /* Caller should ensure the bio has at least some range added */
115 ASSERT(bbio->bio.bi_iter.bi_size);
117 if (btrfs_op(&bbio->bio) == BTRFS_MAP_READ &&
118 bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
119 btrfs_submit_compressed_read(bbio);
121 btrfs_submit_bio(bbio, 0);
123 /* The bbio is owned by the end_io handler now */
124 bio_ctrl->bbio = NULL;
128 * Submit or fail the current bio in the bio_ctrl structure.
130 static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret)
132 struct btrfs_bio *bbio = bio_ctrl->bbio;
139 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
140 /* The bio is owned by the end_io handler now */
141 bio_ctrl->bbio = NULL;
143 submit_one_bio(bio_ctrl);
147 int __init extent_buffer_init_cachep(void)
149 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
150 sizeof(struct extent_buffer), 0,
151 SLAB_MEM_SPREAD, NULL);
152 if (!extent_buffer_cache)
158 void __cold extent_buffer_free_cachep(void)
161 * Make sure all delayed rcu free are flushed before we
165 kmem_cache_destroy(extent_buffer_cache);
168 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
170 unsigned long index = start >> PAGE_SHIFT;
171 unsigned long end_index = end >> PAGE_SHIFT;
174 while (index <= end_index) {
175 page = find_get_page(inode->i_mapping, index);
176 BUG_ON(!page); /* Pages should be in the extent_io_tree */
177 clear_page_dirty_for_io(page);
183 static void process_one_page(struct btrfs_fs_info *fs_info,
184 struct page *page, struct page *locked_page,
185 unsigned long page_ops, u64 start, u64 end)
189 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
190 len = end + 1 - start;
192 if (page_ops & PAGE_SET_ORDERED)
193 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
194 if (page_ops & PAGE_START_WRITEBACK) {
195 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
196 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
198 if (page_ops & PAGE_END_WRITEBACK)
199 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
201 if (page != locked_page && (page_ops & PAGE_UNLOCK))
202 btrfs_page_end_writer_lock(fs_info, page, start, len);
205 static void __process_pages_contig(struct address_space *mapping,
206 struct page *locked_page, u64 start, u64 end,
207 unsigned long page_ops)
209 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
210 pgoff_t start_index = start >> PAGE_SHIFT;
211 pgoff_t end_index = end >> PAGE_SHIFT;
212 pgoff_t index = start_index;
213 struct folio_batch fbatch;
216 folio_batch_init(&fbatch);
217 while (index <= end_index) {
220 found_folios = filemap_get_folios_contig(mapping, &index,
222 for (i = 0; i < found_folios; i++) {
223 struct folio *folio = fbatch.folios[i];
225 process_one_page(fs_info, &folio->page, locked_page,
226 page_ops, start, end);
228 folio_batch_release(&fbatch);
233 static noinline void __unlock_for_delalloc(struct inode *inode,
234 struct page *locked_page,
237 unsigned long index = start >> PAGE_SHIFT;
238 unsigned long end_index = end >> PAGE_SHIFT;
241 if (index == locked_page->index && end_index == index)
244 __process_pages_contig(inode->i_mapping, locked_page, start, end,
248 static noinline int lock_delalloc_pages(struct inode *inode,
249 struct page *locked_page,
253 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
254 struct address_space *mapping = inode->i_mapping;
255 pgoff_t start_index = start >> PAGE_SHIFT;
256 pgoff_t end_index = end >> PAGE_SHIFT;
257 pgoff_t index = start_index;
258 u64 processed_end = start;
259 struct folio_batch fbatch;
261 if (index == locked_page->index && index == end_index)
264 folio_batch_init(&fbatch);
265 while (index <= end_index) {
266 unsigned int found_folios, i;
268 found_folios = filemap_get_folios_contig(mapping, &index,
270 if (found_folios == 0)
273 for (i = 0; i < found_folios; i++) {
274 struct page *page = &fbatch.folios[i]->page;
275 u32 len = end + 1 - start;
277 if (page == locked_page)
280 if (btrfs_page_start_writer_lock(fs_info, page, start,
284 if (!PageDirty(page) || page->mapping != mapping) {
285 btrfs_page_end_writer_lock(fs_info, page, start,
290 processed_end = page_offset(page) + PAGE_SIZE - 1;
292 folio_batch_release(&fbatch);
298 folio_batch_release(&fbatch);
299 if (processed_end > start)
300 __unlock_for_delalloc(inode, locked_page, start, processed_end);
305 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
306 * more than @max_bytes.
308 * @start: The original start bytenr to search.
309 * Will store the extent range start bytenr.
310 * @end: The original end bytenr of the search range
311 * Will store the extent range end bytenr.
313 * Return true if we find a delalloc range which starts inside the original
314 * range, and @start/@end will store the delalloc range start/end.
316 * Return false if we can't find any delalloc range which starts inside the
317 * original range, and @start/@end will be the non-delalloc range start/end.
320 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
321 struct page *locked_page, u64 *start,
324 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
325 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
326 const u64 orig_start = *start;
327 const u64 orig_end = *end;
328 /* The sanity tests may not set a valid fs_info. */
329 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
333 struct extent_state *cached_state = NULL;
337 /* Caller should pass a valid @end to indicate the search range end */
338 ASSERT(orig_end > orig_start);
340 /* The range should at least cover part of the page */
341 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
342 orig_end <= page_offset(locked_page)));
344 /* step one, find a bunch of delalloc bytes starting at start */
345 delalloc_start = *start;
347 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
348 max_bytes, &cached_state);
349 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
350 *start = delalloc_start;
352 /* @delalloc_end can be -1, never go beyond @orig_end */
353 *end = min(delalloc_end, orig_end);
354 free_extent_state(cached_state);
359 * start comes from the offset of locked_page. We have to lock
360 * pages in order, so we can't process delalloc bytes before
363 if (delalloc_start < *start)
364 delalloc_start = *start;
367 * make sure to limit the number of pages we try to lock down
369 if (delalloc_end + 1 - delalloc_start > max_bytes)
370 delalloc_end = delalloc_start + max_bytes - 1;
372 /* step two, lock all the pages after the page that has start */
373 ret = lock_delalloc_pages(inode, locked_page,
374 delalloc_start, delalloc_end);
375 ASSERT(!ret || ret == -EAGAIN);
376 if (ret == -EAGAIN) {
377 /* some of the pages are gone, lets avoid looping by
378 * shortening the size of the delalloc range we're searching
380 free_extent_state(cached_state);
383 max_bytes = PAGE_SIZE;
392 /* step three, lock the state bits for the whole range */
393 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
395 /* then test to make sure it is all still delalloc */
396 ret = test_range_bit(tree, delalloc_start, delalloc_end,
397 EXTENT_DELALLOC, cached_state);
399 unlock_extent(tree, delalloc_start, delalloc_end,
401 __unlock_for_delalloc(inode, locked_page,
402 delalloc_start, delalloc_end);
406 free_extent_state(cached_state);
407 *start = delalloc_start;
413 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
414 struct page *locked_page,
415 u32 clear_bits, unsigned long page_ops)
417 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
419 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
420 start, end, page_ops);
423 static bool btrfs_verify_page(struct page *page, u64 start)
425 if (!fsverity_active(page->mapping->host) ||
426 PageUptodate(page) ||
427 start >= i_size_read(page->mapping->host))
429 return fsverity_verify_page(page);
432 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
434 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
436 ASSERT(page_offset(page) <= start &&
437 start + len <= page_offset(page) + PAGE_SIZE);
439 if (uptodate && btrfs_verify_page(page, start))
440 btrfs_page_set_uptodate(fs_info, page, start, len);
442 btrfs_page_clear_uptodate(fs_info, page, start, len);
444 if (!btrfs_is_subpage(fs_info, page))
447 btrfs_subpage_end_reader(fs_info, page, start, len);
451 * after a writepage IO is done, we need to:
452 * clear the uptodate bits on error
453 * clear the writeback bits in the extent tree for this IO
454 * end_page_writeback if the page has no more pending IO
456 * Scheduling is not allowed, so the extent state tree is expected
457 * to have one and only one object corresponding to this IO.
459 static void end_bio_extent_writepage(struct btrfs_bio *bbio)
461 struct bio *bio = &bbio->bio;
462 int error = blk_status_to_errno(bio->bi_status);
463 struct bio_vec *bvec;
464 struct bvec_iter_all iter_all;
466 ASSERT(!bio_flagged(bio, BIO_CLONED));
467 bio_for_each_segment_all(bvec, bio, iter_all) {
468 struct page *page = bvec->bv_page;
469 struct inode *inode = page->mapping->host;
470 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
471 const u32 sectorsize = fs_info->sectorsize;
472 u64 start = page_offset(page) + bvec->bv_offset;
473 u32 len = bvec->bv_len;
475 /* Our read/write should always be sector aligned. */
476 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
478 "partial page write in btrfs with offset %u and length %u",
479 bvec->bv_offset, bvec->bv_len);
480 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
482 "incomplete page write with offset %u and length %u",
483 bvec->bv_offset, bvec->bv_len);
485 btrfs_finish_ordered_extent(bbio->ordered, page, start, len, !error);
487 mapping_set_error(page->mapping, error);
488 btrfs_page_clear_writeback(fs_info, page, start, len);
495 * Record previously processed extent range
497 * For endio_readpage_release_extent() to handle a full extent range, reducing
498 * the extent io operations.
500 struct processed_extent {
501 struct btrfs_inode *inode;
502 /* Start of the range in @inode */
504 /* End of the range in @inode */
510 * Try to release processed extent range
512 * May not release the extent range right now if the current range is
513 * contiguous to processed extent.
515 * Will release processed extent when any of @inode, @uptodate, the range is
516 * no longer contiguous to the processed range.
518 * Passing @inode == NULL will force processed extent to be released.
520 static void endio_readpage_release_extent(struct processed_extent *processed,
521 struct btrfs_inode *inode, u64 start, u64 end,
524 struct extent_state *cached = NULL;
525 struct extent_io_tree *tree;
527 /* The first extent, initialize @processed */
528 if (!processed->inode)
532 * Contiguous to processed extent, just uptodate the end.
534 * Several things to notice:
536 * - bio can be merged as long as on-disk bytenr is contiguous
537 * This means we can have page belonging to other inodes, thus need to
538 * check if the inode still matches.
539 * - bvec can contain range beyond current page for multi-page bvec
540 * Thus we need to do processed->end + 1 >= start check
542 if (processed->inode == inode && processed->uptodate == uptodate &&
543 processed->end + 1 >= start && end >= processed->end) {
544 processed->end = end;
548 tree = &processed->inode->io_tree;
550 * Now we don't have range contiguous to the processed range, release
551 * the processed range now.
553 unlock_extent(tree, processed->start, processed->end, &cached);
556 /* Update processed to current range */
557 processed->inode = inode;
558 processed->start = start;
559 processed->end = end;
560 processed->uptodate = uptodate;
563 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
565 ASSERT(PageLocked(page));
566 if (!btrfs_is_subpage(fs_info, page))
569 ASSERT(PagePrivate(page));
570 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
574 * after a readpage IO is done, we need to:
575 * clear the uptodate bits on error
576 * set the uptodate bits if things worked
577 * set the page up to date if all extents in the tree are uptodate
578 * clear the lock bit in the extent tree
579 * unlock the page if there are no other extents locked for it
581 * Scheduling is not allowed, so the extent state tree is expected
582 * to have one and only one object corresponding to this IO.
584 static void end_bio_extent_readpage(struct btrfs_bio *bbio)
586 struct bio *bio = &bbio->bio;
587 struct bio_vec *bvec;
588 struct processed_extent processed = { 0 };
590 * The offset to the beginning of a bio, since one bio can never be
591 * larger than UINT_MAX, u32 here is enough.
594 struct bvec_iter_all iter_all;
596 ASSERT(!bio_flagged(bio, BIO_CLONED));
597 bio_for_each_segment_all(bvec, bio, iter_all) {
598 bool uptodate = !bio->bi_status;
599 struct page *page = bvec->bv_page;
600 struct inode *inode = page->mapping->host;
601 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
602 const u32 sectorsize = fs_info->sectorsize;
608 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
609 bio->bi_iter.bi_sector, bio->bi_status,
613 * We always issue full-sector reads, but if some block in a
614 * page fails to read, blk_update_request() will advance
615 * bv_offset and adjust bv_len to compensate. Print a warning
616 * for unaligned offsets, and an error if they don't add up to
619 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
621 "partial page read in btrfs with offset %u and length %u",
622 bvec->bv_offset, bvec->bv_len);
623 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
626 "incomplete page read with offset %u and length %u",
627 bvec->bv_offset, bvec->bv_len);
629 start = page_offset(page) + bvec->bv_offset;
630 end = start + bvec->bv_len - 1;
633 if (likely(uptodate)) {
634 loff_t i_size = i_size_read(inode);
635 pgoff_t end_index = i_size >> PAGE_SHIFT;
638 * Zero out the remaining part if this range straddles
641 * Here we should only zero the range inside the bvec,
642 * not touch anything else.
644 * NOTE: i_size is exclusive while end is inclusive.
646 if (page->index == end_index && i_size <= end) {
647 u32 zero_start = max(offset_in_page(i_size),
648 offset_in_page(start));
650 zero_user_segment(page, zero_start,
651 offset_in_page(end) + 1);
655 /* Update page status and unlock. */
656 end_page_read(page, uptodate, start, len);
657 endio_readpage_release_extent(&processed, BTRFS_I(inode),
658 start, end, uptodate);
660 ASSERT(bio_offset + len > bio_offset);
664 /* Release the last extent */
665 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
670 * Populate every free slot in a provided array with pages.
672 * @nr_pages: number of pages to allocate
673 * @page_array: the array to fill with pages; any existing non-null entries in
674 * the array will be skipped
676 * Return: 0 if all pages were able to be allocated;
677 * -ENOMEM otherwise, the partially allocated pages would be freed and
678 * the array slots zeroed
680 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
682 unsigned int allocated;
684 for (allocated = 0; allocated < nr_pages;) {
685 unsigned int last = allocated;
687 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
689 if (allocated == nr_pages)
693 * During this iteration, no page could be allocated, even
694 * though alloc_pages_bulk_array() falls back to alloc_page()
695 * if it could not bulk-allocate. So we must be out of memory.
697 if (allocated == last) {
698 for (int i = 0; i < allocated; i++) {
699 __free_page(page_array[i]);
700 page_array[i] = NULL;
705 memalloc_retry_wait(GFP_NOFS);
710 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
711 struct page *page, u64 disk_bytenr,
712 unsigned int pg_offset)
714 struct bio *bio = &bio_ctrl->bbio->bio;
715 struct bio_vec *bvec = bio_last_bvec_all(bio);
716 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
718 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
720 * For compression, all IO should have its logical bytenr set
721 * to the starting bytenr of the compressed extent.
723 return bio->bi_iter.bi_sector == sector;
727 * The contig check requires the following conditions to be met:
729 * 1) The pages are belonging to the same inode
730 * This is implied by the call chain.
732 * 2) The range has adjacent logical bytenr
734 * 3) The range has adjacent file offset
735 * This is required for the usage of btrfs_bio->file_offset.
737 return bio_end_sector(bio) == sector &&
738 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
739 page_offset(page) + pg_offset;
742 static void alloc_new_bio(struct btrfs_inode *inode,
743 struct btrfs_bio_ctrl *bio_ctrl,
744 u64 disk_bytenr, u64 file_offset)
746 struct btrfs_fs_info *fs_info = inode->root->fs_info;
747 struct btrfs_bio *bbio;
749 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
750 bio_ctrl->end_io_func, NULL);
751 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
753 bbio->file_offset = file_offset;
754 bio_ctrl->bbio = bbio;
755 bio_ctrl->len_to_oe_boundary = U32_MAX;
757 /* Limit data write bios to the ordered boundary. */
759 struct btrfs_ordered_extent *ordered;
761 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
763 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
764 ordered->file_offset +
765 ordered->disk_num_bytes - file_offset);
766 bbio->ordered = ordered;
770 * Pick the last added device to support cgroup writeback. For
771 * multi-device file systems this means blk-cgroup policies have
772 * to always be set on the last added/replaced device.
773 * This is a bit odd but has been like that for a long time.
775 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
776 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
781 * @disk_bytenr: logical bytenr where the write will be
782 * @page: page to add to the bio
783 * @size: portion of page that we want to write to
784 * @pg_offset: offset of the new bio or to check whether we are adding
785 * a contiguous page to the previous one
787 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
788 * new one in @bio_ctrl->bbio.
789 * The mirror number for this IO should already be initizlied in
790 * @bio_ctrl->mirror_num.
792 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
793 u64 disk_bytenr, struct page *page,
794 size_t size, unsigned long pg_offset)
796 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
798 ASSERT(pg_offset + size <= PAGE_SIZE);
799 ASSERT(bio_ctrl->end_io_func);
801 if (bio_ctrl->bbio &&
802 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
803 submit_one_bio(bio_ctrl);
808 /* Allocate new bio if needed */
809 if (!bio_ctrl->bbio) {
810 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
811 page_offset(page) + pg_offset);
814 /* Cap to the current ordered extent boundary if there is one. */
815 if (len > bio_ctrl->len_to_oe_boundary) {
816 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
817 ASSERT(is_data_inode(&inode->vfs_inode));
818 len = bio_ctrl->len_to_oe_boundary;
821 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
822 /* bio full: move on to a new one */
823 submit_one_bio(bio_ctrl);
828 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
835 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
836 * sector aligned. alloc_new_bio() then sets it to the end of
837 * our ordered extent for writes into zoned devices.
839 * When len_to_oe_boundary is tracking an ordered extent, we
840 * trust the ordered extent code to align things properly, and
841 * the check above to cap our write to the ordered extent
842 * boundary is correct.
844 * When len_to_oe_boundary is U32_MAX, the cap above would
845 * result in a 4095 byte IO for the last page right before
846 * we hit the bio limit of UINT_MAX. bio_add_page() has all
847 * the checks required to make sure we don't overflow the bio,
848 * and we should just ignore len_to_oe_boundary completely
849 * unless we're using it to track an ordered extent.
851 * It's pretty hard to make a bio sized U32_MAX, but it can
852 * happen when the page cache is able to feed us contiguous
853 * pages for large extents.
855 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
856 bio_ctrl->len_to_oe_boundary -= len;
858 /* Ordered extent boundary: move on to a new bio. */
859 if (bio_ctrl->len_to_oe_boundary == 0)
860 submit_one_bio(bio_ctrl);
864 static int attach_extent_buffer_page(struct extent_buffer *eb,
866 struct btrfs_subpage *prealloc)
868 struct btrfs_fs_info *fs_info = eb->fs_info;
872 * If the page is mapped to btree inode, we should hold the private
873 * lock to prevent race.
874 * For cloned or dummy extent buffers, their pages are not mapped and
875 * will not race with any other ebs.
878 lockdep_assert_held(&page->mapping->private_lock);
880 if (fs_info->nodesize >= PAGE_SIZE) {
881 if (!PagePrivate(page))
882 attach_page_private(page, eb);
884 WARN_ON(page->private != (unsigned long)eb);
888 /* Already mapped, just free prealloc */
889 if (PagePrivate(page)) {
890 btrfs_free_subpage(prealloc);
895 /* Has preallocated memory for subpage */
896 attach_page_private(page, prealloc);
898 /* Do new allocation to attach subpage */
899 ret = btrfs_attach_subpage(fs_info, page,
900 BTRFS_SUBPAGE_METADATA);
904 int set_page_extent_mapped(struct page *page)
906 struct btrfs_fs_info *fs_info;
908 ASSERT(page->mapping);
910 if (PagePrivate(page))
913 fs_info = btrfs_sb(page->mapping->host->i_sb);
915 if (btrfs_is_subpage(fs_info, page))
916 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
918 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
922 void clear_page_extent_mapped(struct page *page)
924 struct btrfs_fs_info *fs_info;
926 ASSERT(page->mapping);
928 if (!PagePrivate(page))
931 fs_info = btrfs_sb(page->mapping->host->i_sb);
932 if (btrfs_is_subpage(fs_info, page))
933 return btrfs_detach_subpage(fs_info, page);
935 detach_page_private(page);
938 static struct extent_map *
939 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
940 u64 start, u64 len, struct extent_map **em_cached)
942 struct extent_map *em;
944 if (em_cached && *em_cached) {
946 if (extent_map_in_tree(em) && start >= em->start &&
947 start < extent_map_end(em)) {
948 refcount_inc(&em->refs);
956 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
957 if (em_cached && !IS_ERR(em)) {
959 refcount_inc(&em->refs);
965 * basic readpage implementation. Locked extent state structs are inserted
966 * into the tree that are removed when the IO is done (by the end_io
968 * XXX JDM: This needs looking at to ensure proper page locking
969 * return 0 on success, otherwise return error
971 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
972 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
974 struct inode *inode = page->mapping->host;
975 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
976 u64 start = page_offset(page);
977 const u64 end = start + PAGE_SIZE - 1;
980 u64 last_byte = i_size_read(inode);
982 struct extent_map *em;
984 size_t pg_offset = 0;
986 size_t blocksize = inode->i_sb->s_blocksize;
987 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
989 ret = set_page_extent_mapped(page);
991 unlock_extent(tree, start, end, NULL);
996 if (page->index == last_byte >> PAGE_SHIFT) {
997 size_t zero_offset = offset_in_page(last_byte);
1000 iosize = PAGE_SIZE - zero_offset;
1001 memzero_page(page, zero_offset, iosize);
1004 bio_ctrl->end_io_func = end_bio_extent_readpage;
1005 begin_page_read(fs_info, page);
1006 while (cur <= end) {
1007 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1008 bool force_bio_submit = false;
1011 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1012 if (cur >= last_byte) {
1013 iosize = PAGE_SIZE - pg_offset;
1014 memzero_page(page, pg_offset, iosize);
1015 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1016 end_page_read(page, true, cur, iosize);
1019 em = __get_extent_map(inode, page, pg_offset, cur,
1020 end - cur + 1, em_cached);
1022 unlock_extent(tree, cur, end, NULL);
1023 end_page_read(page, false, cur, end + 1 - cur);
1026 extent_offset = cur - em->start;
1027 BUG_ON(extent_map_end(em) <= cur);
1030 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1031 compress_type = em->compress_type;
1033 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1034 iosize = ALIGN(iosize, blocksize);
1035 if (compress_type != BTRFS_COMPRESS_NONE)
1036 disk_bytenr = em->block_start;
1038 disk_bytenr = em->block_start + extent_offset;
1039 block_start = em->block_start;
1040 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1041 block_start = EXTENT_MAP_HOLE;
1044 * If we have a file range that points to a compressed extent
1045 * and it's followed by a consecutive file range that points
1046 * to the same compressed extent (possibly with a different
1047 * offset and/or length, so it either points to the whole extent
1048 * or only part of it), we must make sure we do not submit a
1049 * single bio to populate the pages for the 2 ranges because
1050 * this makes the compressed extent read zero out the pages
1051 * belonging to the 2nd range. Imagine the following scenario:
1054 * [0 - 8K] [8K - 24K]
1057 * points to extent X, points to extent X,
1058 * offset 4K, length of 8K offset 0, length 16K
1060 * [extent X, compressed length = 4K uncompressed length = 16K]
1062 * If the bio to read the compressed extent covers both ranges,
1063 * it will decompress extent X into the pages belonging to the
1064 * first range and then it will stop, zeroing out the remaining
1065 * pages that belong to the other range that points to extent X.
1066 * So here we make sure we submit 2 bios, one for the first
1067 * range and another one for the third range. Both will target
1068 * the same physical extent from disk, but we can't currently
1069 * make the compressed bio endio callback populate the pages
1070 * for both ranges because each compressed bio is tightly
1071 * coupled with a single extent map, and each range can have
1072 * an extent map with a different offset value relative to the
1073 * uncompressed data of our extent and different lengths. This
1074 * is a corner case so we prioritize correctness over
1075 * non-optimal behavior (submitting 2 bios for the same extent).
1077 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
1078 prev_em_start && *prev_em_start != (u64)-1 &&
1079 *prev_em_start != em->start)
1080 force_bio_submit = true;
1083 *prev_em_start = em->start;
1085 free_extent_map(em);
1088 /* we've found a hole, just zero and go on */
1089 if (block_start == EXTENT_MAP_HOLE) {
1090 memzero_page(page, pg_offset, iosize);
1092 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1093 end_page_read(page, true, cur, iosize);
1095 pg_offset += iosize;
1098 /* the get_extent function already copied into the page */
1099 if (block_start == EXTENT_MAP_INLINE) {
1100 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1101 end_page_read(page, true, cur, iosize);
1103 pg_offset += iosize;
1107 if (bio_ctrl->compress_type != compress_type) {
1108 submit_one_bio(bio_ctrl);
1109 bio_ctrl->compress_type = compress_type;
1112 if (force_bio_submit)
1113 submit_one_bio(bio_ctrl);
1114 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1117 pg_offset += iosize;
1123 int btrfs_read_folio(struct file *file, struct folio *folio)
1125 struct page *page = &folio->page;
1126 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1127 u64 start = page_offset(page);
1128 u64 end = start + PAGE_SIZE - 1;
1129 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1132 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1134 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL);
1136 * If btrfs_do_readpage() failed we will want to submit the assembled
1137 * bio to do the cleanup.
1139 submit_one_bio(&bio_ctrl);
1143 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1145 struct extent_map **em_cached,
1146 struct btrfs_bio_ctrl *bio_ctrl,
1149 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1152 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1154 for (index = 0; index < nr_pages; index++) {
1155 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1157 put_page(pages[index]);
1162 * helper for __extent_writepage, doing all of the delayed allocation setup.
1164 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1165 * to write the page (copy into inline extent). In this case the IO has
1166 * been started and the page is already unlocked.
1168 * This returns 0 if all went well (page still locked)
1169 * This returns < 0 if there were errors (page still locked)
1171 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1172 struct page *page, struct writeback_control *wbc)
1174 const u64 page_start = page_offset(page);
1175 const u64 page_end = page_start + PAGE_SIZE - 1;
1176 u64 delalloc_start = page_start;
1177 u64 delalloc_end = page_end;
1178 u64 delalloc_to_write = 0;
1181 while (delalloc_start < page_end) {
1182 delalloc_end = page_end;
1183 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1184 &delalloc_start, &delalloc_end)) {
1185 delalloc_start = delalloc_end + 1;
1189 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1194 delalloc_start = delalloc_end + 1;
1198 * delalloc_end is already one less than the total length, so
1199 * we don't subtract one from PAGE_SIZE
1201 delalloc_to_write +=
1202 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1205 * If btrfs_run_dealloc_range() already started I/O and unlocked
1206 * the pages, we just need to account for them here.
1209 wbc->nr_to_write -= delalloc_to_write;
1213 if (wbc->nr_to_write < delalloc_to_write) {
1216 if (delalloc_to_write < thresh * 2)
1217 thresh = delalloc_to_write;
1218 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1226 * Find the first byte we need to write.
1228 * For subpage, one page can contain several sectors, and
1229 * __extent_writepage_io() will just grab all extent maps in the page
1230 * range and try to submit all non-inline/non-compressed extents.
1232 * This is a big problem for subpage, we shouldn't re-submit already written
1234 * This function will lookup subpage dirty bit to find which range we really
1237 * Return the next dirty range in [@start, @end).
1238 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1240 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1241 struct page *page, u64 *start, u64 *end)
1243 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1244 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1245 u64 orig_start = *start;
1246 /* Declare as unsigned long so we can use bitmap ops */
1247 unsigned long flags;
1248 int range_start_bit;
1252 * For regular sector size == page size case, since one page only
1253 * contains one sector, we return the page offset directly.
1255 if (!btrfs_is_subpage(fs_info, page)) {
1256 *start = page_offset(page);
1257 *end = page_offset(page) + PAGE_SIZE;
1261 range_start_bit = spi->dirty_offset +
1262 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1264 /* We should have the page locked, but just in case */
1265 spin_lock_irqsave(&subpage->lock, flags);
1266 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1267 spi->dirty_offset + spi->bitmap_nr_bits);
1268 spin_unlock_irqrestore(&subpage->lock, flags);
1270 range_start_bit -= spi->dirty_offset;
1271 range_end_bit -= spi->dirty_offset;
1273 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1274 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1278 * helper for __extent_writepage. This calls the writepage start hooks,
1279 * and does the loop to map the page into extents and bios.
1281 * We return 1 if the IO is started and the page is unlocked,
1282 * 0 if all went well (page still locked)
1283 * < 0 if there were errors (page still locked)
1285 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1287 struct btrfs_bio_ctrl *bio_ctrl,
1291 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1292 u64 cur = page_offset(page);
1293 u64 end = cur + PAGE_SIZE - 1;
1296 struct extent_map *em;
1300 ret = btrfs_writepage_cow_fixup(page);
1302 /* Fixup worker will requeue */
1303 redirty_page_for_writepage(bio_ctrl->wbc, page);
1308 bio_ctrl->end_io_func = end_bio_extent_writepage;
1309 while (cur <= end) {
1310 u32 len = end - cur + 1;
1313 u64 dirty_range_start = cur;
1314 u64 dirty_range_end;
1317 if (cur >= i_size) {
1318 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1321 * This range is beyond i_size, thus we don't need to
1322 * bother writing back.
1323 * But we still need to clear the dirty subpage bit, or
1324 * the next time the page gets dirtied, we will try to
1325 * writeback the sectors with subpage dirty bits,
1326 * causing writeback without ordered extent.
1328 btrfs_page_clear_dirty(fs_info, page, cur, len);
1332 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1334 if (cur < dirty_range_start) {
1335 cur = dirty_range_start;
1339 em = btrfs_get_extent(inode, NULL, 0, cur, len);
1341 ret = PTR_ERR_OR_ZERO(em);
1345 extent_offset = cur - em->start;
1346 em_end = extent_map_end(em);
1347 ASSERT(cur <= em_end);
1349 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1350 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1352 block_start = em->block_start;
1353 disk_bytenr = em->block_start + extent_offset;
1355 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
1356 ASSERT(block_start != EXTENT_MAP_HOLE);
1357 ASSERT(block_start != EXTENT_MAP_INLINE);
1360 * Note that em_end from extent_map_end() and dirty_range_end from
1361 * find_next_dirty_byte() are all exclusive
1363 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1364 free_extent_map(em);
1367 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1368 if (!PageWriteback(page)) {
1369 btrfs_err(inode->root->fs_info,
1370 "page %lu not writeback, cur %llu end %llu",
1371 page->index, cur, end);
1375 * Although the PageDirty bit is cleared before entering this
1376 * function, subpage dirty bit is not cleared.
1377 * So clear subpage dirty bit here so next time we won't submit
1378 * page for range already written to disk.
1380 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
1382 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1383 cur - page_offset(page));
1388 btrfs_page_assert_not_dirty(fs_info, page);
1394 * If we finish without problem, we should not only clear page dirty,
1395 * but also empty subpage dirty bits
1402 * the writepage semantics are similar to regular writepage. extent
1403 * records are inserted to lock ranges in the tree, and as dirty areas
1404 * are found, they are marked writeback. Then the lock bits are removed
1405 * and the end_io handler clears the writeback ranges
1407 * Return 0 if everything goes well.
1408 * Return <0 for error.
1410 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1412 struct folio *folio = page_folio(page);
1413 struct inode *inode = page->mapping->host;
1414 const u64 page_start = page_offset(page);
1418 loff_t i_size = i_size_read(inode);
1419 unsigned long end_index = i_size >> PAGE_SHIFT;
1421 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1423 WARN_ON(!PageLocked(page));
1425 pg_offset = offset_in_page(i_size);
1426 if (page->index > end_index ||
1427 (page->index == end_index && !pg_offset)) {
1428 folio_invalidate(folio, 0, folio_size(folio));
1429 folio_unlock(folio);
1433 if (page->index == end_index)
1434 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1436 ret = set_page_extent_mapped(page);
1440 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1446 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1450 bio_ctrl->wbc->nr_to_write--;
1454 /* make sure the mapping tag for page dirty gets cleared */
1455 set_page_writeback(page);
1456 end_page_writeback(page);
1459 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1461 mapping_set_error(page->mapping, ret);
1468 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1470 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1471 TASK_UNINTERRUPTIBLE);
1475 * Lock extent buffer status and pages for writeback.
1477 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1478 * extent buffer is not dirty)
1479 * Return %true is the extent buffer is submitted to bio.
1481 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1482 struct writeback_control *wbc)
1484 struct btrfs_fs_info *fs_info = eb->fs_info;
1487 btrfs_tree_lock(eb);
1488 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1489 btrfs_tree_unlock(eb);
1490 if (wbc->sync_mode != WB_SYNC_ALL)
1492 wait_on_extent_buffer_writeback(eb);
1493 btrfs_tree_lock(eb);
1497 * We need to do this to prevent races in people who check if the eb is
1498 * under IO since we can end up having no IO bits set for a short period
1501 spin_lock(&eb->refs_lock);
1502 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1503 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1504 spin_unlock(&eb->refs_lock);
1505 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1506 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1508 fs_info->dirty_metadata_batch);
1511 spin_unlock(&eb->refs_lock);
1513 btrfs_tree_unlock(eb);
1517 static void set_btree_ioerr(struct extent_buffer *eb)
1519 struct btrfs_fs_info *fs_info = eb->fs_info;
1521 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1524 * A read may stumble upon this buffer later, make sure that it gets an
1525 * error and knows there was an error.
1527 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1530 * We need to set the mapping with the io error as well because a write
1531 * error will flip the file system readonly, and then syncfs() will
1532 * return a 0 because we are readonly if we don't modify the err seq for
1535 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1538 * If writeback for a btree extent that doesn't belong to a log tree
1539 * failed, increment the counter transaction->eb_write_errors.
1540 * We do this because while the transaction is running and before it's
1541 * committing (when we call filemap_fdata[write|wait]_range against
1542 * the btree inode), we might have
1543 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1544 * returns an error or an error happens during writeback, when we're
1545 * committing the transaction we wouldn't know about it, since the pages
1546 * can be no longer dirty nor marked anymore for writeback (if a
1547 * subsequent modification to the extent buffer didn't happen before the
1548 * transaction commit), which makes filemap_fdata[write|wait]_range not
1549 * able to find the pages tagged with SetPageError at transaction
1550 * commit time. So if this happens we must abort the transaction,
1551 * otherwise we commit a super block with btree roots that point to
1552 * btree nodes/leafs whose content on disk is invalid - either garbage
1553 * or the content of some node/leaf from a past generation that got
1554 * cowed or deleted and is no longer valid.
1556 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1557 * not be enough - we need to distinguish between log tree extents vs
1558 * non-log tree extents, and the next filemap_fdatawait_range() call
1559 * will catch and clear such errors in the mapping - and that call might
1560 * be from a log sync and not from a transaction commit. Also, checking
1561 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1562 * not done and would not be reliable - the eb might have been released
1563 * from memory and reading it back again means that flag would not be
1564 * set (since it's a runtime flag, not persisted on disk).
1566 * Using the flags below in the btree inode also makes us achieve the
1567 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1568 * writeback for all dirty pages and before filemap_fdatawait_range()
1569 * is called, the writeback for all dirty pages had already finished
1570 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1571 * filemap_fdatawait_range() would return success, as it could not know
1572 * that writeback errors happened (the pages were no longer tagged for
1575 switch (eb->log_index) {
1577 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1580 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1583 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1586 BUG(); /* unexpected, logic error */
1591 * The endio specific version which won't touch any unsafe spinlock in endio
1594 static struct extent_buffer *find_extent_buffer_nolock(
1595 struct btrfs_fs_info *fs_info, u64 start)
1597 struct extent_buffer *eb;
1600 eb = radix_tree_lookup(&fs_info->buffer_radix,
1601 start >> fs_info->sectorsize_bits);
1602 if (eb && atomic_inc_not_zero(&eb->refs)) {
1610 static void extent_buffer_write_end_io(struct btrfs_bio *bbio)
1612 struct extent_buffer *eb = bbio->private;
1613 struct btrfs_fs_info *fs_info = eb->fs_info;
1614 bool uptodate = !bbio->bio.bi_status;
1615 struct bvec_iter_all iter_all;
1616 struct bio_vec *bvec;
1620 set_btree_ioerr(eb);
1622 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
1623 u64 start = eb->start + bio_offset;
1624 struct page *page = bvec->bv_page;
1625 u32 len = bvec->bv_len;
1627 btrfs_page_clear_writeback(fs_info, page, start, len);
1631 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1632 smp_mb__after_atomic();
1633 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1635 bio_put(&bbio->bio);
1638 static void prepare_eb_write(struct extent_buffer *eb)
1641 unsigned long start;
1644 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1646 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1647 nritems = btrfs_header_nritems(eb);
1648 if (btrfs_header_level(eb) > 0) {
1649 end = btrfs_node_key_ptr_offset(eb, nritems);
1650 memzero_extent_buffer(eb, end, eb->len - end);
1654 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1656 start = btrfs_item_nr_offset(eb, nritems);
1657 end = btrfs_item_nr_offset(eb, 0);
1659 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1661 end += btrfs_item_offset(eb, nritems - 1);
1662 memzero_extent_buffer(eb, start, end - start);
1666 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1667 struct writeback_control *wbc)
1669 struct btrfs_fs_info *fs_info = eb->fs_info;
1670 struct btrfs_bio *bbio;
1672 prepare_eb_write(eb);
1674 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1675 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1676 eb->fs_info, extent_buffer_write_end_io, eb);
1677 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1678 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1679 wbc_init_bio(wbc, &bbio->bio);
1680 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1681 bbio->file_offset = eb->start;
1682 if (fs_info->nodesize < PAGE_SIZE) {
1683 struct page *p = eb->pages[0];
1686 btrfs_subpage_set_writeback(fs_info, p, eb->start, eb->len);
1687 if (btrfs_subpage_clear_and_test_dirty(fs_info, p, eb->start,
1689 clear_page_dirty_for_io(p);
1692 __bio_add_page(&bbio->bio, p, eb->len, eb->start - page_offset(p));
1693 wbc_account_cgroup_owner(wbc, p, eb->len);
1696 for (int i = 0; i < num_extent_pages(eb); i++) {
1697 struct page *p = eb->pages[i];
1700 clear_page_dirty_for_io(p);
1701 set_page_writeback(p);
1702 __bio_add_page(&bbio->bio, p, PAGE_SIZE, 0);
1703 wbc_account_cgroup_owner(wbc, p, PAGE_SIZE);
1708 btrfs_submit_bio(bbio, 0);
1712 * Submit one subpage btree page.
1714 * The main difference to submit_eb_page() is:
1716 * For subpage, we don't rely on page locking at all.
1719 * We only flush bio if we may be unable to fit current extent buffers into
1722 * Return >=0 for the number of submitted extent buffers.
1723 * Return <0 for fatal error.
1725 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1727 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1729 u64 page_start = page_offset(page);
1731 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1733 /* Lock and write each dirty extent buffers in the range */
1734 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1735 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1736 struct extent_buffer *eb;
1737 unsigned long flags;
1741 * Take private lock to ensure the subpage won't be detached
1744 spin_lock(&page->mapping->private_lock);
1745 if (!PagePrivate(page)) {
1746 spin_unlock(&page->mapping->private_lock);
1749 spin_lock_irqsave(&subpage->lock, flags);
1750 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1751 subpage->bitmaps)) {
1752 spin_unlock_irqrestore(&subpage->lock, flags);
1753 spin_unlock(&page->mapping->private_lock);
1758 start = page_start + bit_start * fs_info->sectorsize;
1759 bit_start += sectors_per_node;
1762 * Here we just want to grab the eb without touching extra
1763 * spin locks, so call find_extent_buffer_nolock().
1765 eb = find_extent_buffer_nolock(fs_info, start);
1766 spin_unlock_irqrestore(&subpage->lock, flags);
1767 spin_unlock(&page->mapping->private_lock);
1770 * The eb has already reached 0 refs thus find_extent_buffer()
1771 * doesn't return it. We don't need to write back such eb
1777 if (lock_extent_buffer_for_io(eb, wbc)) {
1778 write_one_eb(eb, wbc);
1781 free_extent_buffer(eb);
1787 * Submit all page(s) of one extent buffer.
1789 * @page: the page of one extent buffer
1790 * @eb_context: to determine if we need to submit this page, if current page
1791 * belongs to this eb, we don't need to submit
1793 * The caller should pass each page in their bytenr order, and here we use
1794 * @eb_context to determine if we have submitted pages of one extent buffer.
1796 * If we have, we just skip until we hit a new page that doesn't belong to
1797 * current @eb_context.
1799 * If not, we submit all the page(s) of the extent buffer.
1801 * Return >0 if we have submitted the extent buffer successfully.
1802 * Return 0 if we don't need to submit the page, as it's already submitted by
1804 * Return <0 for fatal error.
1806 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1808 struct writeback_control *wbc = ctx->wbc;
1809 struct address_space *mapping = page->mapping;
1810 struct extent_buffer *eb;
1813 if (!PagePrivate(page))
1816 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
1817 return submit_eb_subpage(page, wbc);
1819 spin_lock(&mapping->private_lock);
1820 if (!PagePrivate(page)) {
1821 spin_unlock(&mapping->private_lock);
1825 eb = (struct extent_buffer *)page->private;
1828 * Shouldn't happen and normally this would be a BUG_ON but no point
1829 * crashing the machine for something we can survive anyway.
1832 spin_unlock(&mapping->private_lock);
1836 if (eb == ctx->eb) {
1837 spin_unlock(&mapping->private_lock);
1840 ret = atomic_inc_not_zero(&eb->refs);
1841 spin_unlock(&mapping->private_lock);
1847 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1851 free_extent_buffer(eb);
1855 if (!lock_extent_buffer_for_io(eb, wbc)) {
1856 free_extent_buffer(eb);
1859 /* Implies write in zoned mode. */
1860 if (ctx->zoned_bg) {
1861 /* Mark the last eb in the block group. */
1862 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1863 ctx->zoned_bg->meta_write_pointer += eb->len;
1865 write_one_eb(eb, wbc);
1866 free_extent_buffer(eb);
1870 int btree_write_cache_pages(struct address_space *mapping,
1871 struct writeback_control *wbc)
1873 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1874 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
1877 int nr_to_write_done = 0;
1878 struct folio_batch fbatch;
1879 unsigned int nr_folios;
1881 pgoff_t end; /* Inclusive */
1885 folio_batch_init(&fbatch);
1886 if (wbc->range_cyclic) {
1887 index = mapping->writeback_index; /* Start from prev offset */
1890 * Start from the beginning does not need to cycle over the
1891 * range, mark it as scanned.
1893 scanned = (index == 0);
1895 index = wbc->range_start >> PAGE_SHIFT;
1896 end = wbc->range_end >> PAGE_SHIFT;
1899 if (wbc->sync_mode == WB_SYNC_ALL)
1900 tag = PAGECACHE_TAG_TOWRITE;
1902 tag = PAGECACHE_TAG_DIRTY;
1903 btrfs_zoned_meta_io_lock(fs_info);
1905 if (wbc->sync_mode == WB_SYNC_ALL)
1906 tag_pages_for_writeback(mapping, index, end);
1907 while (!done && !nr_to_write_done && (index <= end) &&
1908 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1912 for (i = 0; i < nr_folios; i++) {
1913 struct folio *folio = fbatch.folios[i];
1915 ret = submit_eb_page(&folio->page, &ctx);
1924 * the filesystem may choose to bump up nr_to_write.
1925 * We have to make sure to honor the new nr_to_write
1928 nr_to_write_done = wbc->nr_to_write <= 0;
1930 folio_batch_release(&fbatch);
1933 if (!scanned && !done) {
1935 * We hit the last page and there is more work to be done: wrap
1936 * back to the start of the file
1943 * If something went wrong, don't allow any metadata write bio to be
1946 * This would prevent use-after-free if we had dirty pages not
1947 * cleaned up, which can still happen by fuzzed images.
1950 * Allowing existing tree block to be allocated for other trees.
1952 * - Log tree operations
1953 * Exiting tree blocks get allocated to log tree, bumps its
1954 * generation, then get cleaned in tree re-balance.
1955 * Such tree block will not be written back, since it's clean,
1956 * thus no WRITTEN flag set.
1957 * And after log writes back, this tree block is not traced by
1958 * any dirty extent_io_tree.
1960 * - Offending tree block gets re-dirtied from its original owner
1961 * Since it has bumped generation, no WRITTEN flag, it can be
1962 * reused without COWing. This tree block will not be traced
1963 * by btrfs_transaction::dirty_pages.
1965 * Now such dirty tree block will not be cleaned by any dirty
1966 * extent io tree. Thus we don't want to submit such wild eb
1967 * if the fs already has error.
1969 * We can get ret > 0 from submit_extent_page() indicating how many ebs
1970 * were submitted. Reset it to 0 to avoid false alerts for the caller.
1974 if (!ret && BTRFS_FS_ERROR(fs_info))
1978 btrfs_put_block_group(ctx.zoned_bg);
1979 btrfs_zoned_meta_io_unlock(fs_info);
1984 * Walk the list of dirty pages of the given address space and write all of them.
1986 * @mapping: address space structure to write
1987 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1988 * @bio_ctrl: holds context for the write, namely the bio
1990 * If a page is already under I/O, write_cache_pages() skips it, even
1991 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
1992 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
1993 * and msync() need to guarantee that all the data which was dirty at the time
1994 * the call was made get new I/O started against them. If wbc->sync_mode is
1995 * WB_SYNC_ALL then we were called for data integrity and we must wait for
1996 * existing IO to complete.
1998 static int extent_write_cache_pages(struct address_space *mapping,
1999 struct btrfs_bio_ctrl *bio_ctrl)
2001 struct writeback_control *wbc = bio_ctrl->wbc;
2002 struct inode *inode = mapping->host;
2005 int nr_to_write_done = 0;
2006 struct folio_batch fbatch;
2007 unsigned int nr_folios;
2009 pgoff_t end; /* Inclusive */
2011 int range_whole = 0;
2016 * We have to hold onto the inode so that ordered extents can do their
2017 * work when the IO finishes. The alternative to this is failing to add
2018 * an ordered extent if the igrab() fails there and that is a huge pain
2019 * to deal with, so instead just hold onto the inode throughout the
2020 * writepages operation. If it fails here we are freeing up the inode
2021 * anyway and we'd rather not waste our time writing out stuff that is
2022 * going to be truncated anyway.
2027 folio_batch_init(&fbatch);
2028 if (wbc->range_cyclic) {
2029 index = mapping->writeback_index; /* Start from prev offset */
2032 * Start from the beginning does not need to cycle over the
2033 * range, mark it as scanned.
2035 scanned = (index == 0);
2037 index = wbc->range_start >> PAGE_SHIFT;
2038 end = wbc->range_end >> PAGE_SHIFT;
2039 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2045 * We do the tagged writepage as long as the snapshot flush bit is set
2046 * and we are the first one who do the filemap_flush() on this inode.
2048 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2049 * not race in and drop the bit.
2051 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2052 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2053 &BTRFS_I(inode)->runtime_flags))
2054 wbc->tagged_writepages = 1;
2056 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2057 tag = PAGECACHE_TAG_TOWRITE;
2059 tag = PAGECACHE_TAG_DIRTY;
2061 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2062 tag_pages_for_writeback(mapping, index, end);
2064 while (!done && !nr_to_write_done && (index <= end) &&
2065 (nr_folios = filemap_get_folios_tag(mapping, &index,
2066 end, tag, &fbatch))) {
2069 for (i = 0; i < nr_folios; i++) {
2070 struct folio *folio = fbatch.folios[i];
2072 done_index = folio_next_index(folio);
2074 * At this point we hold neither the i_pages lock nor
2075 * the page lock: the page may be truncated or
2076 * invalidated (changing page->mapping to NULL),
2077 * or even swizzled back from swapper_space to
2078 * tmpfs file mapping
2080 if (!folio_trylock(folio)) {
2081 submit_write_bio(bio_ctrl, 0);
2085 if (unlikely(folio->mapping != mapping)) {
2086 folio_unlock(folio);
2090 if (!folio_test_dirty(folio)) {
2091 /* Someone wrote it for us. */
2092 folio_unlock(folio);
2096 if (wbc->sync_mode != WB_SYNC_NONE) {
2097 if (folio_test_writeback(folio))
2098 submit_write_bio(bio_ctrl, 0);
2099 folio_wait_writeback(folio);
2102 if (folio_test_writeback(folio) ||
2103 !folio_clear_dirty_for_io(folio)) {
2104 folio_unlock(folio);
2108 ret = __extent_writepage(&folio->page, bio_ctrl);
2115 * The filesystem may choose to bump up nr_to_write.
2116 * We have to make sure to honor the new nr_to_write
2119 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2120 wbc->nr_to_write <= 0);
2122 folio_batch_release(&fbatch);
2125 if (!scanned && !done) {
2127 * We hit the last page and there is more work to be done: wrap
2128 * back to the start of the file
2134 * If we're looping we could run into a page that is locked by a
2135 * writer and that writer could be waiting on writeback for a
2136 * page in our current bio, and thus deadlock, so flush the
2139 submit_write_bio(bio_ctrl, 0);
2143 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2144 mapping->writeback_index = done_index;
2146 btrfs_add_delayed_iput(BTRFS_I(inode));
2151 * Submit the pages in the range to bio for call sites which delalloc range has
2152 * already been ran (aka, ordered extent inserted) and all pages are still
2155 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2156 u64 start, u64 end, struct writeback_control *wbc,
2159 bool found_error = false;
2161 struct address_space *mapping = inode->i_mapping;
2162 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2163 const u32 sectorsize = fs_info->sectorsize;
2164 loff_t i_size = i_size_read(inode);
2166 struct btrfs_bio_ctrl bio_ctrl = {
2168 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2171 if (wbc->no_cgroup_owner)
2172 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2174 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2176 while (cur <= end) {
2177 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2178 u32 cur_len = cur_end + 1 - cur;
2182 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2183 ASSERT(PageLocked(page));
2184 if (pages_dirty && page != locked_page) {
2185 ASSERT(PageDirty(page));
2186 clear_page_dirty_for_io(page);
2189 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2194 /* Make sure the mapping tag for page dirty gets cleared. */
2196 set_page_writeback(page);
2197 end_page_writeback(page);
2200 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2201 cur, cur_len, !ret);
2202 mapping_set_error(page->mapping, ret);
2204 btrfs_page_unlock_writer(fs_info, page, cur, cur_len);
2212 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2215 int extent_writepages(struct address_space *mapping,
2216 struct writeback_control *wbc)
2218 struct inode *inode = mapping->host;
2220 struct btrfs_bio_ctrl bio_ctrl = {
2222 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2226 * Allow only a single thread to do the reloc work in zoned mode to
2227 * protect the write pointer updates.
2229 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2230 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2231 submit_write_bio(&bio_ctrl, ret);
2232 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2236 void extent_readahead(struct readahead_control *rac)
2238 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2239 struct page *pagepool[16];
2240 struct extent_map *em_cached = NULL;
2241 u64 prev_em_start = (u64)-1;
2244 while ((nr = readahead_page_batch(rac, pagepool))) {
2245 u64 contig_start = readahead_pos(rac);
2246 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2248 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2249 &em_cached, &bio_ctrl, &prev_em_start);
2253 free_extent_map(em_cached);
2254 submit_one_bio(&bio_ctrl);
2258 * basic invalidate_folio code, this waits on any locked or writeback
2259 * ranges corresponding to the folio, and then deletes any extent state
2260 * records from the tree
2262 int extent_invalidate_folio(struct extent_io_tree *tree,
2263 struct folio *folio, size_t offset)
2265 struct extent_state *cached_state = NULL;
2266 u64 start = folio_pos(folio);
2267 u64 end = start + folio_size(folio) - 1;
2268 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
2270 /* This function is only called for the btree inode */
2271 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2273 start += ALIGN(offset, blocksize);
2277 lock_extent(tree, start, end, &cached_state);
2278 folio_wait_writeback(folio);
2281 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2282 * so here we only need to unlock the extent range to free any
2283 * existing extent state.
2285 unlock_extent(tree, start, end, &cached_state);
2290 * a helper for release_folio, this tests for areas of the page that
2291 * are locked or under IO and drops the related state bits if it is safe
2294 static int try_release_extent_state(struct extent_io_tree *tree,
2295 struct page *page, gfp_t mask)
2297 u64 start = page_offset(page);
2298 u64 end = start + PAGE_SIZE - 1;
2301 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2304 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2305 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS |
2306 EXTENT_QGROUP_RESERVED);
2309 * At this point we can safely clear everything except the
2310 * locked bit, the nodatasum bit and the delalloc new bit.
2311 * The delalloc new bit will be cleared by ordered extent
2314 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2316 /* if clear_extent_bit failed for enomem reasons,
2317 * we can't allow the release to continue.
2328 * a helper for release_folio. As long as there are no locked extents
2329 * in the range corresponding to the page, both state records and extent
2330 * map records are removed
2332 int try_release_extent_mapping(struct page *page, gfp_t mask)
2334 struct extent_map *em;
2335 u64 start = page_offset(page);
2336 u64 end = start + PAGE_SIZE - 1;
2337 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
2338 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2339 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2341 if (gfpflags_allow_blocking(mask) &&
2342 page->mapping->host->i_size > SZ_16M) {
2344 while (start <= end) {
2345 struct btrfs_fs_info *fs_info;
2348 len = end - start + 1;
2349 write_lock(&map->lock);
2350 em = lookup_extent_mapping(map, start, len);
2352 write_unlock(&map->lock);
2355 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2356 em->start != start) {
2357 write_unlock(&map->lock);
2358 free_extent_map(em);
2361 if (test_range_bit_exists(tree, em->start,
2362 extent_map_end(em) - 1,
2366 * If it's not in the list of modified extents, used
2367 * by a fast fsync, we can remove it. If it's being
2368 * logged we can safely remove it since fsync took an
2369 * extra reference on the em.
2371 if (list_empty(&em->list) ||
2372 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
2375 * If it's in the list of modified extents, remove it
2376 * only if its generation is older then the current one,
2377 * in which case we don't need it for a fast fsync.
2378 * Otherwise don't remove it, we could be racing with an
2379 * ongoing fast fsync that could miss the new extent.
2381 fs_info = btrfs_inode->root->fs_info;
2382 spin_lock(&fs_info->trans_lock);
2383 cur_gen = fs_info->generation;
2384 spin_unlock(&fs_info->trans_lock);
2385 if (em->generation >= cur_gen)
2389 * We only remove extent maps that are not in the list of
2390 * modified extents or that are in the list but with a
2391 * generation lower then the current generation, so there
2392 * is no need to set the full fsync flag on the inode (it
2393 * hurts the fsync performance for workloads with a data
2394 * size that exceeds or is close to the system's memory).
2396 remove_extent_mapping(map, em);
2397 /* once for the rb tree */
2398 free_extent_map(em);
2400 start = extent_map_end(em);
2401 write_unlock(&map->lock);
2404 free_extent_map(em);
2406 cond_resched(); /* Allow large-extent preemption. */
2409 return try_release_extent_state(tree, page, mask);
2413 * To cache previous fiemap extent
2415 * Will be used for merging fiemap extent
2417 struct fiemap_cache {
2426 * Helper to submit fiemap extent.
2428 * Will try to merge current fiemap extent specified by @offset, @phys,
2429 * @len and @flags with cached one.
2430 * And only when we fails to merge, cached one will be submitted as
2433 * Return value is the same as fiemap_fill_next_extent().
2435 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2436 struct fiemap_cache *cache,
2437 u64 offset, u64 phys, u64 len, u32 flags)
2441 /* Set at the end of extent_fiemap(). */
2442 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2448 * Sanity check, extent_fiemap() should have ensured that new
2449 * fiemap extent won't overlap with cached one.
2452 * NOTE: Physical address can overlap, due to compression
2454 if (cache->offset + cache->len > offset) {
2460 * Only merges fiemap extents if
2461 * 1) Their logical addresses are continuous
2463 * 2) Their physical addresses are continuous
2464 * So truly compressed (physical size smaller than logical size)
2465 * extents won't get merged with each other
2467 * 3) Share same flags
2469 if (cache->offset + cache->len == offset &&
2470 cache->phys + cache->len == phys &&
2471 cache->flags == flags) {
2476 /* Not mergeable, need to submit cached one */
2477 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2478 cache->len, cache->flags);
2479 cache->cached = false;
2483 cache->cached = true;
2484 cache->offset = offset;
2487 cache->flags = flags;
2493 * Emit last fiemap cache
2495 * The last fiemap cache may still be cached in the following case:
2497 * |<- Fiemap range ->|
2498 * |<------------ First extent ----------->|
2500 * In this case, the first extent range will be cached but not emitted.
2501 * So we must emit it before ending extent_fiemap().
2503 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2504 struct fiemap_cache *cache)
2511 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2512 cache->len, cache->flags);
2513 cache->cached = false;
2519 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2521 struct extent_buffer *clone;
2522 struct btrfs_key key;
2527 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2530 ret = btrfs_next_leaf(inode->root, path);
2535 * Don't bother with cloning if there are no more file extent items for
2538 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2539 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
2542 /* See the comment at fiemap_search_slot() about why we clone. */
2543 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2547 slot = path->slots[0];
2548 btrfs_release_path(path);
2549 path->nodes[0] = clone;
2550 path->slots[0] = slot;
2556 * Search for the first file extent item that starts at a given file offset or
2557 * the one that starts immediately before that offset.
2558 * Returns: 0 on success, < 0 on error, 1 if not found.
2560 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2563 const u64 ino = btrfs_ino(inode);
2564 struct btrfs_root *root = inode->root;
2565 struct extent_buffer *clone;
2566 struct btrfs_key key;
2571 key.type = BTRFS_EXTENT_DATA_KEY;
2572 key.offset = file_offset;
2574 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2578 if (ret > 0 && path->slots[0] > 0) {
2579 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2580 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2584 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2585 ret = btrfs_next_leaf(root, path);
2589 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2590 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2595 * We clone the leaf and use it during fiemap. This is because while
2596 * using the leaf we do expensive things like checking if an extent is
2597 * shared, which can take a long time. In order to prevent blocking
2598 * other tasks for too long, we use a clone of the leaf. We have locked
2599 * the file range in the inode's io tree, so we know none of our file
2600 * extent items can change. This way we avoid blocking other tasks that
2601 * want to insert items for other inodes in the same leaf or b+tree
2602 * rebalance operations (triggered for example when someone is trying
2603 * to push items into this leaf when trying to insert an item in a
2605 * We also need the private clone because holding a read lock on an
2606 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2607 * when we call fiemap_fill_next_extent(), because that may cause a page
2608 * fault when filling the user space buffer with fiemap data.
2610 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2614 slot = path->slots[0];
2615 btrfs_release_path(path);
2616 path->nodes[0] = clone;
2617 path->slots[0] = slot;
2623 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2624 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2625 * extent. The end offset (@end) is inclusive.
2627 static int fiemap_process_hole(struct btrfs_inode *inode,
2628 struct fiemap_extent_info *fieinfo,
2629 struct fiemap_cache *cache,
2630 struct extent_state **delalloc_cached_state,
2631 struct btrfs_backref_share_check_ctx *backref_ctx,
2632 u64 disk_bytenr, u64 extent_offset,
2636 const u64 i_size = i_size_read(&inode->vfs_inode);
2637 u64 cur_offset = start;
2638 u64 last_delalloc_end = 0;
2639 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2640 bool checked_extent_shared = false;
2644 * There can be no delalloc past i_size, so don't waste time looking for
2647 while (cur_offset < end && cur_offset < i_size) {
2651 u64 prealloc_len = 0;
2654 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2655 delalloc_cached_state,
2662 * If this is a prealloc extent we have to report every section
2663 * of it that has no delalloc.
2665 if (disk_bytenr != 0) {
2666 if (last_delalloc_end == 0) {
2667 prealloc_start = start;
2668 prealloc_len = delalloc_start - start;
2670 prealloc_start = last_delalloc_end + 1;
2671 prealloc_len = delalloc_start - prealloc_start;
2675 if (prealloc_len > 0) {
2676 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2677 ret = btrfs_is_data_extent_shared(inode,
2684 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2686 checked_extent_shared = true;
2688 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2689 disk_bytenr + extent_offset,
2690 prealloc_len, prealloc_flags);
2693 extent_offset += prealloc_len;
2696 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2697 delalloc_end + 1 - delalloc_start,
2698 FIEMAP_EXTENT_DELALLOC |
2699 FIEMAP_EXTENT_UNKNOWN);
2703 last_delalloc_end = delalloc_end;
2704 cur_offset = delalloc_end + 1;
2705 extent_offset += cur_offset - delalloc_start;
2710 * Either we found no delalloc for the whole prealloc extent or we have
2711 * a prealloc extent that spans i_size or starts at or after i_size.
2713 if (disk_bytenr != 0 && last_delalloc_end < end) {
2717 if (last_delalloc_end == 0) {
2718 prealloc_start = start;
2719 prealloc_len = end + 1 - start;
2721 prealloc_start = last_delalloc_end + 1;
2722 prealloc_len = end + 1 - prealloc_start;
2725 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2726 ret = btrfs_is_data_extent_shared(inode,
2733 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2735 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2736 disk_bytenr + extent_offset,
2737 prealloc_len, prealloc_flags);
2745 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2746 struct btrfs_path *path,
2747 u64 *last_extent_end_ret)
2749 const u64 ino = btrfs_ino(inode);
2750 struct btrfs_root *root = inode->root;
2751 struct extent_buffer *leaf;
2752 struct btrfs_file_extent_item *ei;
2753 struct btrfs_key key;
2758 * Lookup the last file extent. We're not using i_size here because
2759 * there might be preallocation past i_size.
2761 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
2762 /* There can't be a file extent item at offset (u64)-1 */
2768 * For a non-existing key, btrfs_search_slot() always leaves us at a
2769 * slot > 0, except if the btree is empty, which is impossible because
2770 * at least it has the inode item for this inode and all the items for
2771 * the root inode 256.
2773 ASSERT(path->slots[0] > 0);
2775 leaf = path->nodes[0];
2776 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2777 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
2778 /* No file extent items in the subvolume tree. */
2779 *last_extent_end_ret = 0;
2784 * For an inline extent, the disk_bytenr is where inline data starts at,
2785 * so first check if we have an inline extent item before checking if we
2786 * have an implicit hole (disk_bytenr == 0).
2788 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2789 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
2790 *last_extent_end_ret = btrfs_file_extent_end(path);
2795 * Find the last file extent item that is not a hole (when NO_HOLES is
2796 * not enabled). This should take at most 2 iterations in the worst
2797 * case: we have one hole file extent item at slot 0 of a leaf and
2798 * another hole file extent item as the last item in the previous leaf.
2799 * This is because we merge file extent items that represent holes.
2801 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2802 while (disk_bytenr == 0) {
2803 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
2806 } else if (ret > 0) {
2807 /* No file extent items that are not holes. */
2808 *last_extent_end_ret = 0;
2811 leaf = path->nodes[0];
2812 ei = btrfs_item_ptr(leaf, path->slots[0],
2813 struct btrfs_file_extent_item);
2814 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2817 *last_extent_end_ret = btrfs_file_extent_end(path);
2821 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
2824 const u64 ino = btrfs_ino(inode);
2825 struct extent_state *cached_state = NULL;
2826 struct extent_state *delalloc_cached_state = NULL;
2827 struct btrfs_path *path;
2828 struct fiemap_cache cache = { 0 };
2829 struct btrfs_backref_share_check_ctx *backref_ctx;
2830 u64 last_extent_end;
2831 u64 prev_extent_end;
2834 bool stopped = false;
2837 backref_ctx = btrfs_alloc_backref_share_check_ctx();
2838 path = btrfs_alloc_path();
2839 if (!backref_ctx || !path) {
2844 lockstart = round_down(start, inode->root->fs_info->sectorsize);
2845 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
2846 prev_extent_end = lockstart;
2848 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
2849 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
2851 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
2854 btrfs_release_path(path);
2856 path->reada = READA_FORWARD;
2857 ret = fiemap_search_slot(inode, path, lockstart);
2860 } else if (ret > 0) {
2862 * No file extent item found, but we may have delalloc between
2863 * the current offset and i_size. So check for that.
2866 goto check_eof_delalloc;
2869 while (prev_extent_end < lockend) {
2870 struct extent_buffer *leaf = path->nodes[0];
2871 struct btrfs_file_extent_item *ei;
2872 struct btrfs_key key;
2875 u64 extent_offset = 0;
2877 u64 disk_bytenr = 0;
2882 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2883 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2886 extent_end = btrfs_file_extent_end(path);
2889 * The first iteration can leave us at an extent item that ends
2890 * before our range's start. Move to the next item.
2892 if (extent_end <= lockstart)
2895 backref_ctx->curr_leaf_bytenr = leaf->start;
2897 /* We have in implicit hole (NO_HOLES feature enabled). */
2898 if (prev_extent_end < key.offset) {
2899 const u64 range_end = min(key.offset, lockend) - 1;
2901 ret = fiemap_process_hole(inode, fieinfo, &cache,
2902 &delalloc_cached_state,
2903 backref_ctx, 0, 0, 0,
2904 prev_extent_end, range_end);
2907 } else if (ret > 0) {
2908 /* fiemap_fill_next_extent() told us to stop. */
2913 /* We've reached the end of the fiemap range, stop. */
2914 if (key.offset >= lockend) {
2920 extent_len = extent_end - key.offset;
2921 ei = btrfs_item_ptr(leaf, path->slots[0],
2922 struct btrfs_file_extent_item);
2923 compression = btrfs_file_extent_compression(leaf, ei);
2924 extent_type = btrfs_file_extent_type(leaf, ei);
2925 extent_gen = btrfs_file_extent_generation(leaf, ei);
2927 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2928 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2929 if (compression == BTRFS_COMPRESS_NONE)
2930 extent_offset = btrfs_file_extent_offset(leaf, ei);
2933 if (compression != BTRFS_COMPRESS_NONE)
2934 flags |= FIEMAP_EXTENT_ENCODED;
2936 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2937 flags |= FIEMAP_EXTENT_DATA_INLINE;
2938 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
2939 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
2941 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2942 ret = fiemap_process_hole(inode, fieinfo, &cache,
2943 &delalloc_cached_state,
2945 disk_bytenr, extent_offset,
2946 extent_gen, key.offset,
2948 } else if (disk_bytenr == 0) {
2949 /* We have an explicit hole. */
2950 ret = fiemap_process_hole(inode, fieinfo, &cache,
2951 &delalloc_cached_state,
2952 backref_ctx, 0, 0, 0,
2953 key.offset, extent_end - 1);
2955 /* We have a regular extent. */
2956 if (fieinfo->fi_extents_max) {
2957 ret = btrfs_is_data_extent_shared(inode,
2964 flags |= FIEMAP_EXTENT_SHARED;
2967 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
2968 disk_bytenr + extent_offset,
2974 } else if (ret > 0) {
2975 /* fiemap_fill_next_extent() told us to stop. */
2980 prev_extent_end = extent_end;
2982 if (fatal_signal_pending(current)) {
2987 ret = fiemap_next_leaf_item(inode, path);
2990 } else if (ret > 0) {
2991 /* No more file extent items for this inode. */
2999 * Release (and free) the path before emitting any final entries to
3000 * fiemap_fill_next_extent() to keep lockdep happy. This is because
3001 * once we find no more file extent items exist, we may have a
3002 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
3003 * faults when copying data to the user space buffer.
3005 btrfs_free_path(path);
3008 if (!stopped && prev_extent_end < lockend) {
3009 ret = fiemap_process_hole(inode, fieinfo, &cache,
3010 &delalloc_cached_state, backref_ctx,
3011 0, 0, 0, prev_extent_end, lockend - 1);
3014 prev_extent_end = lockend;
3017 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3018 const u64 i_size = i_size_read(&inode->vfs_inode);
3020 if (prev_extent_end < i_size) {
3025 delalloc = btrfs_find_delalloc_in_range(inode,
3028 &delalloc_cached_state,
3032 cache.flags |= FIEMAP_EXTENT_LAST;
3034 cache.flags |= FIEMAP_EXTENT_LAST;
3038 ret = emit_last_fiemap_cache(fieinfo, &cache);
3041 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3042 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3044 free_extent_state(delalloc_cached_state);
3045 btrfs_free_backref_share_ctx(backref_ctx);
3046 btrfs_free_path(path);
3050 static void __free_extent_buffer(struct extent_buffer *eb)
3052 kmem_cache_free(extent_buffer_cache, eb);
3055 static int extent_buffer_under_io(const struct extent_buffer *eb)
3057 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3058 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3061 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
3063 struct btrfs_subpage *subpage;
3065 lockdep_assert_held(&page->mapping->private_lock);
3067 if (PagePrivate(page)) {
3068 subpage = (struct btrfs_subpage *)page->private;
3069 if (atomic_read(&subpage->eb_refs))
3072 * Even there is no eb refs here, we may still have
3073 * end_page_read() call relying on page::private.
3075 if (atomic_read(&subpage->readers))
3081 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
3083 struct btrfs_fs_info *fs_info = eb->fs_info;
3084 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3087 * For mapped eb, we're going to change the page private, which should
3088 * be done under the private_lock.
3091 spin_lock(&page->mapping->private_lock);
3093 if (!PagePrivate(page)) {
3095 spin_unlock(&page->mapping->private_lock);
3099 if (fs_info->nodesize >= PAGE_SIZE) {
3101 * We do this since we'll remove the pages after we've
3102 * removed the eb from the radix tree, so we could race
3103 * and have this page now attached to the new eb. So
3104 * only clear page_private if it's still connected to
3107 if (PagePrivate(page) &&
3108 page->private == (unsigned long)eb) {
3109 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3110 BUG_ON(PageDirty(page));
3111 BUG_ON(PageWriteback(page));
3113 * We need to make sure we haven't be attached
3116 detach_page_private(page);
3119 spin_unlock(&page->mapping->private_lock);
3124 * For subpage, we can have dummy eb with page private. In this case,
3125 * we can directly detach the private as such page is only attached to
3126 * one dummy eb, no sharing.
3129 btrfs_detach_subpage(fs_info, page);
3133 btrfs_page_dec_eb_refs(fs_info, page);
3136 * We can only detach the page private if there are no other ebs in the
3137 * page range and no unfinished IO.
3139 if (!page_range_has_eb(fs_info, page))
3140 btrfs_detach_subpage(fs_info, page);
3142 spin_unlock(&page->mapping->private_lock);
3145 /* Release all pages attached to the extent buffer */
3146 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3151 ASSERT(!extent_buffer_under_io(eb));
3153 num_pages = num_extent_pages(eb);
3154 for (i = 0; i < num_pages; i++) {
3155 struct page *page = eb->pages[i];
3160 detach_extent_buffer_page(eb, page);
3162 /* One for when we allocated the page */
3168 * Helper for releasing the extent buffer.
3170 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3172 btrfs_release_extent_buffer_pages(eb);
3173 btrfs_leak_debug_del_eb(eb);
3174 __free_extent_buffer(eb);
3177 static struct extent_buffer *
3178 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3181 struct extent_buffer *eb = NULL;
3183 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3186 eb->fs_info = fs_info;
3187 init_rwsem(&eb->lock);
3189 btrfs_leak_debug_add_eb(eb);
3191 spin_lock_init(&eb->refs_lock);
3192 atomic_set(&eb->refs, 1);
3194 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3199 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3202 struct extent_buffer *new;
3203 int num_pages = num_extent_pages(src);
3206 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3211 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3212 * btrfs_release_extent_buffer() have different behavior for
3213 * UNMAPPED subpage extent buffer.
3215 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3217 ret = btrfs_alloc_page_array(num_pages, new->pages);
3219 btrfs_release_extent_buffer(new);
3223 for (i = 0; i < num_pages; i++) {
3225 struct page *p = new->pages[i];
3227 ret = attach_extent_buffer_page(new, p, NULL);
3229 btrfs_release_extent_buffer(new);
3232 WARN_ON(PageDirty(p));
3234 copy_extent_buffer_full(new, src);
3235 set_extent_buffer_uptodate(new);
3240 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3241 u64 start, unsigned long len)
3243 struct extent_buffer *eb;
3248 eb = __alloc_extent_buffer(fs_info, start, len);
3252 num_pages = num_extent_pages(eb);
3253 ret = btrfs_alloc_page_array(num_pages, eb->pages);
3257 for (i = 0; i < num_pages; i++) {
3258 struct page *p = eb->pages[i];
3260 ret = attach_extent_buffer_page(eb, p, NULL);
3265 set_extent_buffer_uptodate(eb);
3266 btrfs_set_header_nritems(eb, 0);
3267 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3271 for (i = 0; i < num_pages; i++) {
3273 detach_extent_buffer_page(eb, eb->pages[i]);
3274 __free_page(eb->pages[i]);
3277 __free_extent_buffer(eb);
3281 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3284 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3287 static void check_buffer_tree_ref(struct extent_buffer *eb)
3291 * The TREE_REF bit is first set when the extent_buffer is added
3292 * to the radix tree. It is also reset, if unset, when a new reference
3293 * is created by find_extent_buffer.
3295 * It is only cleared in two cases: freeing the last non-tree
3296 * reference to the extent_buffer when its STALE bit is set or
3297 * calling release_folio when the tree reference is the only reference.
3299 * In both cases, care is taken to ensure that the extent_buffer's
3300 * pages are not under io. However, release_folio can be concurrently
3301 * called with creating new references, which is prone to race
3302 * conditions between the calls to check_buffer_tree_ref in those
3303 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3305 * The actual lifetime of the extent_buffer in the radix tree is
3306 * adequately protected by the refcount, but the TREE_REF bit and
3307 * its corresponding reference are not. To protect against this
3308 * class of races, we call check_buffer_tree_ref from the codepaths
3309 * which trigger io. Note that once io is initiated, TREE_REF can no
3310 * longer be cleared, so that is the moment at which any such race is
3313 refs = atomic_read(&eb->refs);
3314 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3317 spin_lock(&eb->refs_lock);
3318 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3319 atomic_inc(&eb->refs);
3320 spin_unlock(&eb->refs_lock);
3323 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
3324 struct page *accessed)
3328 check_buffer_tree_ref(eb);
3330 num_pages = num_extent_pages(eb);
3331 for (i = 0; i < num_pages; i++) {
3332 struct page *p = eb->pages[i];
3335 mark_page_accessed(p);
3339 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3342 struct extent_buffer *eb;
3344 eb = find_extent_buffer_nolock(fs_info, start);
3348 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3349 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3350 * another task running free_extent_buffer() might have seen that flag
3351 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3352 * writeback flags not set) and it's still in the tree (flag
3353 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3354 * decrementing the extent buffer's reference count twice. So here we
3355 * could race and increment the eb's reference count, clear its stale
3356 * flag, mark it as dirty and drop our reference before the other task
3357 * finishes executing free_extent_buffer, which would later result in
3358 * an attempt to free an extent buffer that is dirty.
3360 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3361 spin_lock(&eb->refs_lock);
3362 spin_unlock(&eb->refs_lock);
3364 mark_extent_buffer_accessed(eb, NULL);
3368 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3369 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3372 struct extent_buffer *eb, *exists = NULL;
3375 eb = find_extent_buffer(fs_info, start);
3378 eb = alloc_dummy_extent_buffer(fs_info, start);
3380 return ERR_PTR(-ENOMEM);
3381 eb->fs_info = fs_info;
3383 ret = radix_tree_preload(GFP_NOFS);
3385 exists = ERR_PTR(ret);
3388 spin_lock(&fs_info->buffer_lock);
3389 ret = radix_tree_insert(&fs_info->buffer_radix,
3390 start >> fs_info->sectorsize_bits, eb);
3391 spin_unlock(&fs_info->buffer_lock);
3392 radix_tree_preload_end();
3393 if (ret == -EEXIST) {
3394 exists = find_extent_buffer(fs_info, start);
3400 check_buffer_tree_ref(eb);
3401 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3405 btrfs_release_extent_buffer(eb);
3410 static struct extent_buffer *grab_extent_buffer(
3411 struct btrfs_fs_info *fs_info, struct page *page)
3413 struct extent_buffer *exists;
3416 * For subpage case, we completely rely on radix tree to ensure we
3417 * don't try to insert two ebs for the same bytenr. So here we always
3418 * return NULL and just continue.
3420 if (fs_info->nodesize < PAGE_SIZE)
3423 /* Page not yet attached to an extent buffer */
3424 if (!PagePrivate(page))
3428 * We could have already allocated an eb for this page and attached one
3429 * so lets see if we can get a ref on the existing eb, and if we can we
3430 * know it's good and we can just return that one, else we know we can
3431 * just overwrite page->private.
3433 exists = (struct extent_buffer *)page->private;
3434 if (atomic_inc_not_zero(&exists->refs))
3437 WARN_ON(PageDirty(page));
3438 detach_page_private(page);
3442 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3444 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3445 btrfs_err(fs_info, "bad tree block start %llu", start);
3449 if (fs_info->nodesize < PAGE_SIZE &&
3450 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3452 "tree block crosses page boundary, start %llu nodesize %u",
3453 start, fs_info->nodesize);
3456 if (fs_info->nodesize >= PAGE_SIZE &&
3457 !PAGE_ALIGNED(start)) {
3459 "tree block is not page aligned, start %llu nodesize %u",
3460 start, fs_info->nodesize);
3463 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3464 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3466 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3467 start, fs_info->nodesize);
3472 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3473 u64 start, u64 owner_root, int level)
3475 unsigned long len = fs_info->nodesize;
3478 unsigned long index = start >> PAGE_SHIFT;
3479 struct extent_buffer *eb;
3480 struct extent_buffer *exists = NULL;
3482 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3483 struct btrfs_subpage *prealloc = NULL;
3484 u64 lockdep_owner = owner_root;
3488 if (check_eb_alignment(fs_info, start))
3489 return ERR_PTR(-EINVAL);
3491 #if BITS_PER_LONG == 32
3492 if (start >= MAX_LFS_FILESIZE) {
3493 btrfs_err_rl(fs_info,
3494 "extent buffer %llu is beyond 32bit page cache limit", start);
3495 btrfs_err_32bit_limit(fs_info);
3496 return ERR_PTR(-EOVERFLOW);
3498 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3499 btrfs_warn_32bit_limit(fs_info);
3502 eb = find_extent_buffer(fs_info, start);
3506 eb = __alloc_extent_buffer(fs_info, start, len);
3508 return ERR_PTR(-ENOMEM);
3511 * The reloc trees are just snapshots, so we need them to appear to be
3512 * just like any other fs tree WRT lockdep.
3514 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3515 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3517 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3519 num_pages = num_extent_pages(eb);
3522 * Preallocate page->private for subpage case, so that we won't
3523 * allocate memory with private_lock nor page lock hold.
3525 * The memory will be freed by attach_extent_buffer_page() or freed
3526 * manually if we exit earlier.
3528 if (fs_info->nodesize < PAGE_SIZE) {
3529 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3530 if (IS_ERR(prealloc)) {
3531 exists = ERR_CAST(prealloc);
3536 for (i = 0; i < num_pages; i++, index++) {
3537 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
3539 exists = ERR_PTR(-ENOMEM);
3540 btrfs_free_subpage(prealloc);
3544 spin_lock(&mapping->private_lock);
3545 exists = grab_extent_buffer(fs_info, p);
3547 spin_unlock(&mapping->private_lock);
3550 mark_extent_buffer_accessed(exists, p);
3551 btrfs_free_subpage(prealloc);
3554 /* Should not fail, as we have preallocated the memory */
3555 ret = attach_extent_buffer_page(eb, p, prealloc);
3558 * To inform we have extra eb under allocation, so that
3559 * detach_extent_buffer_page() won't release the page private
3560 * when the eb hasn't yet been inserted into radix tree.
3562 * The ref will be decreased when the eb released the page, in
3563 * detach_extent_buffer_page().
3564 * Thus needs no special handling in error path.
3566 btrfs_page_inc_eb_refs(fs_info, p);
3567 spin_unlock(&mapping->private_lock);
3569 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
3571 if (!btrfs_page_test_uptodate(fs_info, p, eb->start, eb->len))
3575 * We can't unlock the pages just yet since the extent buffer
3576 * hasn't been properly inserted in the radix tree, this
3577 * opens a race with btree_release_folio which can free a page
3578 * while we are still filling in all pages for the buffer and
3583 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3585 ret = radix_tree_preload(GFP_NOFS);
3587 exists = ERR_PTR(ret);
3591 spin_lock(&fs_info->buffer_lock);
3592 ret = radix_tree_insert(&fs_info->buffer_radix,
3593 start >> fs_info->sectorsize_bits, eb);
3594 spin_unlock(&fs_info->buffer_lock);
3595 radix_tree_preload_end();
3596 if (ret == -EEXIST) {
3597 exists = find_extent_buffer(fs_info, start);
3603 /* add one reference for the tree */
3604 check_buffer_tree_ref(eb);
3605 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3608 * Now it's safe to unlock the pages because any calls to
3609 * btree_release_folio will correctly detect that a page belongs to a
3610 * live buffer and won't free them prematurely.
3612 for (i = 0; i < num_pages; i++)
3613 unlock_page(eb->pages[i]);
3617 WARN_ON(!atomic_dec_and_test(&eb->refs));
3618 for (i = 0; i < num_pages; i++) {
3620 unlock_page(eb->pages[i]);
3623 btrfs_release_extent_buffer(eb);
3627 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3629 struct extent_buffer *eb =
3630 container_of(head, struct extent_buffer, rcu_head);
3632 __free_extent_buffer(eb);
3635 static int release_extent_buffer(struct extent_buffer *eb)
3636 __releases(&eb->refs_lock)
3638 lockdep_assert_held(&eb->refs_lock);
3640 WARN_ON(atomic_read(&eb->refs) == 0);
3641 if (atomic_dec_and_test(&eb->refs)) {
3642 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
3643 struct btrfs_fs_info *fs_info = eb->fs_info;
3645 spin_unlock(&eb->refs_lock);
3647 spin_lock(&fs_info->buffer_lock);
3648 radix_tree_delete(&fs_info->buffer_radix,
3649 eb->start >> fs_info->sectorsize_bits);
3650 spin_unlock(&fs_info->buffer_lock);
3652 spin_unlock(&eb->refs_lock);
3655 btrfs_leak_debug_del_eb(eb);
3656 /* Should be safe to release our pages at this point */
3657 btrfs_release_extent_buffer_pages(eb);
3658 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3659 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
3660 __free_extent_buffer(eb);
3664 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
3667 spin_unlock(&eb->refs_lock);
3672 void free_extent_buffer(struct extent_buffer *eb)
3678 refs = atomic_read(&eb->refs);
3680 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
3681 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
3684 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
3688 spin_lock(&eb->refs_lock);
3689 if (atomic_read(&eb->refs) == 2 &&
3690 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
3691 !extent_buffer_under_io(eb) &&
3692 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3693 atomic_dec(&eb->refs);
3696 * I know this is terrible, but it's temporary until we stop tracking
3697 * the uptodate bits and such for the extent buffers.
3699 release_extent_buffer(eb);
3702 void free_extent_buffer_stale(struct extent_buffer *eb)
3707 spin_lock(&eb->refs_lock);
3708 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
3710 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3711 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3712 atomic_dec(&eb->refs);
3713 release_extent_buffer(eb);
3716 static void btree_clear_page_dirty(struct page *page)
3718 ASSERT(PageDirty(page));
3719 ASSERT(PageLocked(page));
3720 clear_page_dirty_for_io(page);
3721 xa_lock_irq(&page->mapping->i_pages);
3722 if (!PageDirty(page))
3723 __xa_clear_mark(&page->mapping->i_pages,
3724 page_index(page), PAGECACHE_TAG_DIRTY);
3725 xa_unlock_irq(&page->mapping->i_pages);
3728 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
3730 struct btrfs_fs_info *fs_info = eb->fs_info;
3731 struct page *page = eb->pages[0];
3734 /* btree_clear_page_dirty() needs page locked */
3736 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
3739 btree_clear_page_dirty(page);
3741 WARN_ON(atomic_read(&eb->refs) == 0);
3744 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
3745 struct extent_buffer *eb)
3747 struct btrfs_fs_info *fs_info = eb->fs_info;
3752 btrfs_assert_tree_write_locked(eb);
3754 if (trans && btrfs_header_generation(eb) != trans->transid)
3757 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
3760 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
3761 fs_info->dirty_metadata_batch);
3763 if (eb->fs_info->nodesize < PAGE_SIZE)
3764 return clear_subpage_extent_buffer_dirty(eb);
3766 num_pages = num_extent_pages(eb);
3768 for (i = 0; i < num_pages; i++) {
3769 page = eb->pages[i];
3770 if (!PageDirty(page))
3773 btree_clear_page_dirty(page);
3776 WARN_ON(atomic_read(&eb->refs) == 0);
3779 void set_extent_buffer_dirty(struct extent_buffer *eb)
3785 check_buffer_tree_ref(eb);
3787 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3789 num_pages = num_extent_pages(eb);
3790 WARN_ON(atomic_read(&eb->refs) == 0);
3791 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
3794 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
3797 * For subpage case, we can have other extent buffers in the
3798 * same page, and in clear_subpage_extent_buffer_dirty() we
3799 * have to clear page dirty without subpage lock held.
3800 * This can cause race where our page gets dirty cleared after
3803 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
3804 * its page for other reasons, we can use page lock to prevent
3808 lock_page(eb->pages[0]);
3809 for (i = 0; i < num_pages; i++)
3810 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
3811 eb->start, eb->len);
3813 unlock_page(eb->pages[0]);
3814 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
3816 eb->fs_info->dirty_metadata_batch);
3818 #ifdef CONFIG_BTRFS_DEBUG
3819 for (i = 0; i < num_pages; i++)
3820 ASSERT(PageDirty(eb->pages[i]));
3824 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
3826 struct btrfs_fs_info *fs_info = eb->fs_info;
3831 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3832 num_pages = num_extent_pages(eb);
3833 for (i = 0; i < num_pages; i++) {
3834 page = eb->pages[i];
3839 * This is special handling for metadata subpage, as regular
3840 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3842 if (fs_info->nodesize >= PAGE_SIZE)
3843 ClearPageUptodate(page);
3845 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
3850 void set_extent_buffer_uptodate(struct extent_buffer *eb)
3852 struct btrfs_fs_info *fs_info = eb->fs_info;
3857 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3858 num_pages = num_extent_pages(eb);
3859 for (i = 0; i < num_pages; i++) {
3860 page = eb->pages[i];
3863 * This is special handling for metadata subpage, as regular
3864 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3866 if (fs_info->nodesize >= PAGE_SIZE)
3867 SetPageUptodate(page);
3869 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
3874 static void extent_buffer_read_end_io(struct btrfs_bio *bbio)
3876 struct extent_buffer *eb = bbio->private;
3877 struct btrfs_fs_info *fs_info = eb->fs_info;
3878 bool uptodate = !bbio->bio.bi_status;
3879 struct bvec_iter_all iter_all;
3880 struct bio_vec *bvec;
3883 eb->read_mirror = bbio->mirror_num;
3886 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
3890 set_extent_buffer_uptodate(eb);
3892 clear_extent_buffer_uptodate(eb);
3893 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3896 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
3897 u64 start = eb->start + bio_offset;
3898 struct page *page = bvec->bv_page;
3899 u32 len = bvec->bv_len;
3902 btrfs_page_set_uptodate(fs_info, page, start, len);
3904 btrfs_page_clear_uptodate(fs_info, page, start, len);
3909 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
3910 smp_mb__after_atomic();
3911 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
3912 free_extent_buffer(eb);
3914 bio_put(&bbio->bio);
3917 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
3918 struct btrfs_tree_parent_check *check)
3920 int num_pages = num_extent_pages(eb), i;
3921 struct btrfs_bio *bbio;
3923 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3927 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
3928 * operation, which could potentially still be in flight. In this case
3929 * we simply want to return an error.
3931 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
3934 /* Someone else is already reading the buffer, just wait for it. */
3935 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
3938 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3939 eb->read_mirror = 0;
3940 check_buffer_tree_ref(eb);
3941 atomic_inc(&eb->refs);
3943 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
3944 REQ_OP_READ | REQ_META, eb->fs_info,
3945 extent_buffer_read_end_io, eb);
3946 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
3947 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
3948 bbio->file_offset = eb->start;
3949 memcpy(&bbio->parent_check, check, sizeof(*check));
3950 if (eb->fs_info->nodesize < PAGE_SIZE) {
3951 __bio_add_page(&bbio->bio, eb->pages[0], eb->len,
3952 eb->start - page_offset(eb->pages[0]));
3954 for (i = 0; i < num_pages; i++)
3955 __bio_add_page(&bbio->bio, eb->pages[i], PAGE_SIZE, 0);
3957 btrfs_submit_bio(bbio, mirror_num);
3960 if (wait == WAIT_COMPLETE) {
3961 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
3962 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3969 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
3972 btrfs_warn(eb->fs_info,
3973 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
3974 eb->start, eb->len, start, len);
3975 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3981 * Check if the [start, start + len) range is valid before reading/writing
3983 * NOTE: @start and @len are offset inside the eb, not logical address.
3985 * Caller should not touch the dst/src memory if this function returns error.
3987 static inline int check_eb_range(const struct extent_buffer *eb,
3988 unsigned long start, unsigned long len)
3990 unsigned long offset;
3992 /* start, start + len should not go beyond eb->len nor overflow */
3993 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
3994 return report_eb_range(eb, start, len);
3999 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4000 unsigned long start, unsigned long len)
4006 char *dst = (char *)dstv;
4007 unsigned long i = get_eb_page_index(start);
4009 if (check_eb_range(eb, start, len)) {
4011 * Invalid range hit, reset the memory, so callers won't get
4012 * some random garbage for their uninitialzed memory.
4014 memset(dstv, 0, len);
4018 offset = get_eb_offset_in_page(eb, start);
4021 page = eb->pages[i];
4023 cur = min(len, (PAGE_SIZE - offset));
4024 kaddr = page_address(page);
4025 memcpy(dst, kaddr + offset, cur);
4034 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4036 unsigned long start, unsigned long len)
4042 char __user *dst = (char __user *)dstv;
4043 unsigned long i = get_eb_page_index(start);
4046 WARN_ON(start > eb->len);
4047 WARN_ON(start + len > eb->start + eb->len);
4049 offset = get_eb_offset_in_page(eb, start);
4052 page = eb->pages[i];
4054 cur = min(len, (PAGE_SIZE - offset));
4055 kaddr = page_address(page);
4056 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4070 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4071 unsigned long start, unsigned long len)
4077 char *ptr = (char *)ptrv;
4078 unsigned long i = get_eb_page_index(start);
4081 if (check_eb_range(eb, start, len))
4084 offset = get_eb_offset_in_page(eb, start);
4087 page = eb->pages[i];
4089 cur = min(len, (PAGE_SIZE - offset));
4091 kaddr = page_address(page);
4092 ret = memcmp(ptr, kaddr + offset, cur);
4105 * Check that the extent buffer is uptodate.
4107 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4108 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4110 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
4113 struct btrfs_fs_info *fs_info = eb->fs_info;
4116 * If we are using the commit root we could potentially clear a page
4117 * Uptodate while we're using the extent buffer that we've previously
4118 * looked up. We don't want to complain in this case, as the page was
4119 * valid before, we just didn't write it out. Instead we want to catch
4120 * the case where we didn't actually read the block properly, which
4121 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4123 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4126 if (fs_info->nodesize < PAGE_SIZE) {
4127 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, page,
4128 eb->start, eb->len)))
4129 btrfs_subpage_dump_bitmap(fs_info, page, eb->start, eb->len);
4131 WARN_ON(!PageUptodate(page));
4135 static void __write_extent_buffer(const struct extent_buffer *eb,
4136 const void *srcv, unsigned long start,
4137 unsigned long len, bool use_memmove)
4143 char *src = (char *)srcv;
4144 unsigned long i = get_eb_page_index(start);
4145 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4146 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4148 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
4150 if (check_eb_range(eb, start, len))
4153 offset = get_eb_offset_in_page(eb, start);
4156 page = eb->pages[i];
4158 assert_eb_page_uptodate(eb, page);
4160 cur = min(len, PAGE_SIZE - offset);
4161 kaddr = page_address(page);
4163 memmove(kaddr + offset, src, cur);
4165 memcpy(kaddr + offset, src, cur);
4174 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4175 unsigned long start, unsigned long len)
4177 return __write_extent_buffer(eb, srcv, start, len, false);
4180 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4181 unsigned long start, unsigned long len)
4183 unsigned long cur = start;
4185 while (cur < start + len) {
4186 unsigned long index = get_eb_page_index(cur);
4187 unsigned int offset = get_eb_offset_in_page(eb, cur);
4188 unsigned int cur_len = min(start + len - cur, PAGE_SIZE - offset);
4189 struct page *page = eb->pages[index];
4191 assert_eb_page_uptodate(eb, page);
4192 memset(page_address(page) + offset, c, cur_len);
4198 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4201 if (check_eb_range(eb, start, len))
4203 return memset_extent_buffer(eb, 0, start, len);
4206 void copy_extent_buffer_full(const struct extent_buffer *dst,
4207 const struct extent_buffer *src)
4209 unsigned long cur = 0;
4211 ASSERT(dst->len == src->len);
4213 while (cur < src->len) {
4214 unsigned long index = get_eb_page_index(cur);
4215 unsigned long offset = get_eb_offset_in_page(src, cur);
4216 unsigned long cur_len = min(src->len, PAGE_SIZE - offset);
4217 void *addr = page_address(src->pages[index]) + offset;
4219 write_extent_buffer(dst, addr, cur, cur_len);
4225 void copy_extent_buffer(const struct extent_buffer *dst,
4226 const struct extent_buffer *src,
4227 unsigned long dst_offset, unsigned long src_offset,
4230 u64 dst_len = dst->len;
4235 unsigned long i = get_eb_page_index(dst_offset);
4237 if (check_eb_range(dst, dst_offset, len) ||
4238 check_eb_range(src, src_offset, len))
4241 WARN_ON(src->len != dst_len);
4243 offset = get_eb_offset_in_page(dst, dst_offset);
4246 page = dst->pages[i];
4247 assert_eb_page_uptodate(dst, page);
4249 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
4251 kaddr = page_address(page);
4252 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4262 * Calculate the page and offset of the byte containing the given bit number.
4264 * @eb: the extent buffer
4265 * @start: offset of the bitmap item in the extent buffer
4267 * @page_index: return index of the page in the extent buffer that contains
4268 * the given bit number
4269 * @page_offset: return offset into the page given by page_index
4271 * This helper hides the ugliness of finding the byte in an extent buffer which
4272 * contains a given bit.
4274 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4275 unsigned long start, unsigned long nr,
4276 unsigned long *page_index,
4277 size_t *page_offset)
4279 size_t byte_offset = BIT_BYTE(nr);
4283 * The byte we want is the offset of the extent buffer + the offset of
4284 * the bitmap item in the extent buffer + the offset of the byte in the
4287 offset = start + offset_in_page(eb->start) + byte_offset;
4289 *page_index = offset >> PAGE_SHIFT;
4290 *page_offset = offset_in_page(offset);
4294 * Determine whether a bit in a bitmap item is set.
4296 * @eb: the extent buffer
4297 * @start: offset of the bitmap item in the extent buffer
4298 * @nr: bit number to test
4300 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4308 eb_bitmap_offset(eb, start, nr, &i, &offset);
4309 page = eb->pages[i];
4310 assert_eb_page_uptodate(eb, page);
4311 kaddr = page_address(page);
4312 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4315 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4317 unsigned long index = get_eb_page_index(bytenr);
4319 if (check_eb_range(eb, bytenr, 1))
4321 return page_address(eb->pages[index]) + get_eb_offset_in_page(eb, bytenr);
4325 * Set an area of a bitmap to 1.
4327 * @eb: the extent buffer
4328 * @start: offset of the bitmap item in the extent buffer
4329 * @pos: bit number of the first bit
4330 * @len: number of bits to set
4332 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4333 unsigned long pos, unsigned long len)
4335 unsigned int first_byte = start + BIT_BYTE(pos);
4336 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4337 const bool same_byte = (first_byte == last_byte);
4338 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4342 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4344 /* Handle the first byte. */
4345 kaddr = extent_buffer_get_byte(eb, first_byte);
4350 /* Handle the byte aligned part. */
4351 ASSERT(first_byte + 1 <= last_byte);
4352 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4354 /* Handle the last byte. */
4355 kaddr = extent_buffer_get_byte(eb, last_byte);
4356 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4361 * Clear an area of a bitmap.
4363 * @eb: the extent buffer
4364 * @start: offset of the bitmap item in the extent buffer
4365 * @pos: bit number of the first bit
4366 * @len: number of bits to clear
4368 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4369 unsigned long start, unsigned long pos,
4372 unsigned int first_byte = start + BIT_BYTE(pos);
4373 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4374 const bool same_byte = (first_byte == last_byte);
4375 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4379 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4381 /* Handle the first byte. */
4382 kaddr = extent_buffer_get_byte(eb, first_byte);
4387 /* Handle the byte aligned part. */
4388 ASSERT(first_byte + 1 <= last_byte);
4389 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4391 /* Handle the last byte. */
4392 kaddr = extent_buffer_get_byte(eb, last_byte);
4393 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4396 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4398 unsigned long distance = (src > dst) ? src - dst : dst - src;
4399 return distance < len;
4402 void memcpy_extent_buffer(const struct extent_buffer *dst,
4403 unsigned long dst_offset, unsigned long src_offset,
4406 unsigned long cur_off = 0;
4408 if (check_eb_range(dst, dst_offset, len) ||
4409 check_eb_range(dst, src_offset, len))
4412 while (cur_off < len) {
4413 unsigned long cur_src = cur_off + src_offset;
4414 unsigned long pg_index = get_eb_page_index(cur_src);
4415 unsigned long pg_off = get_eb_offset_in_page(dst, cur_src);
4416 unsigned long cur_len = min(src_offset + len - cur_src,
4417 PAGE_SIZE - pg_off);
4418 void *src_addr = page_address(dst->pages[pg_index]) + pg_off;
4419 const bool use_memmove = areas_overlap(src_offset + cur_off,
4420 dst_offset + cur_off, cur_len);
4422 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4428 void memmove_extent_buffer(const struct extent_buffer *dst,
4429 unsigned long dst_offset, unsigned long src_offset,
4432 unsigned long dst_end = dst_offset + len - 1;
4433 unsigned long src_end = src_offset + len - 1;
4435 if (check_eb_range(dst, dst_offset, len) ||
4436 check_eb_range(dst, src_offset, len))
4439 if (dst_offset < src_offset) {
4440 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4445 unsigned long src_i;
4447 size_t dst_off_in_page;
4448 size_t src_off_in_page;
4452 src_i = get_eb_page_index(src_end);
4454 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
4455 src_off_in_page = get_eb_offset_in_page(dst, src_end);
4457 cur = min_t(unsigned long, len, src_off_in_page + 1);
4458 cur = min(cur, dst_off_in_page + 1);
4460 src_addr = page_address(dst->pages[src_i]) + src_off_in_page -
4462 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4465 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4474 #define GANG_LOOKUP_SIZE 16
4475 static struct extent_buffer *get_next_extent_buffer(
4476 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4478 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4479 struct extent_buffer *found = NULL;
4480 u64 page_start = page_offset(page);
4481 u64 cur = page_start;
4483 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4484 lockdep_assert_held(&fs_info->buffer_lock);
4486 while (cur < page_start + PAGE_SIZE) {
4490 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4491 (void **)gang, cur >> fs_info->sectorsize_bits,
4492 min_t(unsigned int, GANG_LOOKUP_SIZE,
4493 PAGE_SIZE / fs_info->nodesize));
4496 for (i = 0; i < ret; i++) {
4497 /* Already beyond page end */
4498 if (gang[i]->start >= page_start + PAGE_SIZE)
4501 if (gang[i]->start >= bytenr) {
4506 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4512 static int try_release_subpage_extent_buffer(struct page *page)
4514 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4515 u64 cur = page_offset(page);
4516 const u64 end = page_offset(page) + PAGE_SIZE;
4520 struct extent_buffer *eb = NULL;
4523 * Unlike try_release_extent_buffer() which uses page->private
4524 * to grab buffer, for subpage case we rely on radix tree, thus
4525 * we need to ensure radix tree consistency.
4527 * We also want an atomic snapshot of the radix tree, thus go
4528 * with spinlock rather than RCU.
4530 spin_lock(&fs_info->buffer_lock);
4531 eb = get_next_extent_buffer(fs_info, page, cur);
4533 /* No more eb in the page range after or at cur */
4534 spin_unlock(&fs_info->buffer_lock);
4537 cur = eb->start + eb->len;
4540 * The same as try_release_extent_buffer(), to ensure the eb
4541 * won't disappear out from under us.
4543 spin_lock(&eb->refs_lock);
4544 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4545 spin_unlock(&eb->refs_lock);
4546 spin_unlock(&fs_info->buffer_lock);
4549 spin_unlock(&fs_info->buffer_lock);
4552 * If tree ref isn't set then we know the ref on this eb is a
4553 * real ref, so just return, this eb will likely be freed soon
4556 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4557 spin_unlock(&eb->refs_lock);
4562 * Here we don't care about the return value, we will always
4563 * check the page private at the end. And
4564 * release_extent_buffer() will release the refs_lock.
4566 release_extent_buffer(eb);
4569 * Finally to check if we have cleared page private, as if we have
4570 * released all ebs in the page, the page private should be cleared now.
4572 spin_lock(&page->mapping->private_lock);
4573 if (!PagePrivate(page))
4577 spin_unlock(&page->mapping->private_lock);
4582 int try_release_extent_buffer(struct page *page)
4584 struct extent_buffer *eb;
4586 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4587 return try_release_subpage_extent_buffer(page);
4590 * We need to make sure nobody is changing page->private, as we rely on
4591 * page->private as the pointer to extent buffer.
4593 spin_lock(&page->mapping->private_lock);
4594 if (!PagePrivate(page)) {
4595 spin_unlock(&page->mapping->private_lock);
4599 eb = (struct extent_buffer *)page->private;
4603 * This is a little awful but should be ok, we need to make sure that
4604 * the eb doesn't disappear out from under us while we're looking at
4607 spin_lock(&eb->refs_lock);
4608 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4609 spin_unlock(&eb->refs_lock);
4610 spin_unlock(&page->mapping->private_lock);
4613 spin_unlock(&page->mapping->private_lock);
4616 * If tree ref isn't set then we know the ref on this eb is a real ref,
4617 * so just return, this page will likely be freed soon anyway.
4619 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4620 spin_unlock(&eb->refs_lock);
4624 return release_extent_buffer(eb);
4628 * Attempt to readahead a child block.
4630 * @fs_info: the fs_info
4631 * @bytenr: bytenr to read
4632 * @owner_root: objectid of the root that owns this eb
4633 * @gen: generation for the uptodate check, can be 0
4634 * @level: level for the eb
4636 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
4637 * normal uptodate check of the eb, without checking the generation. If we have
4638 * to read the block we will not block on anything.
4640 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
4641 u64 bytenr, u64 owner_root, u64 gen, int level)
4643 struct btrfs_tree_parent_check check = {
4648 struct extent_buffer *eb;
4651 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
4655 if (btrfs_buffer_uptodate(eb, gen, 1)) {
4656 free_extent_buffer(eb);
4660 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
4662 free_extent_buffer_stale(eb);
4664 free_extent_buffer(eb);
4668 * Readahead a node's child block.
4670 * @node: parent node we're reading from
4671 * @slot: slot in the parent node for the child we want to read
4673 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
4674 * the slot in the node provided.
4676 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
4678 btrfs_readahead_tree_block(node->fs_info,
4679 btrfs_node_blockptr(node, slot),
4680 btrfs_header_owner(node),
4681 btrfs_node_ptr_generation(node, slot),
4682 btrfs_header_level(node) - 1);