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, and the caller is responsible for freeing all
678 * non-null page pointers in the array.
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)
700 memalloc_retry_wait(GFP_NOFS);
705 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
706 struct page *page, u64 disk_bytenr,
707 unsigned int pg_offset)
709 struct bio *bio = &bio_ctrl->bbio->bio;
710 struct bio_vec *bvec = bio_last_bvec_all(bio);
711 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
713 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
715 * For compression, all IO should have its logical bytenr set
716 * to the starting bytenr of the compressed extent.
718 return bio->bi_iter.bi_sector == sector;
722 * The contig check requires the following conditions to be met:
724 * 1) The pages are belonging to the same inode
725 * This is implied by the call chain.
727 * 2) The range has adjacent logical bytenr
729 * 3) The range has adjacent file offset
730 * This is required for the usage of btrfs_bio->file_offset.
732 return bio_end_sector(bio) == sector &&
733 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
734 page_offset(page) + pg_offset;
737 static void alloc_new_bio(struct btrfs_inode *inode,
738 struct btrfs_bio_ctrl *bio_ctrl,
739 u64 disk_bytenr, u64 file_offset)
741 struct btrfs_fs_info *fs_info = inode->root->fs_info;
742 struct btrfs_bio *bbio;
744 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
745 bio_ctrl->end_io_func, NULL);
746 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
748 bbio->file_offset = file_offset;
749 bio_ctrl->bbio = bbio;
750 bio_ctrl->len_to_oe_boundary = U32_MAX;
752 /* Limit data write bios to the ordered boundary. */
754 struct btrfs_ordered_extent *ordered;
756 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
758 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
759 ordered->file_offset +
760 ordered->disk_num_bytes - file_offset);
761 bbio->ordered = ordered;
765 * Pick the last added device to support cgroup writeback. For
766 * multi-device file systems this means blk-cgroup policies have
767 * to always be set on the last added/replaced device.
768 * This is a bit odd but has been like that for a long time.
770 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
771 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
776 * @disk_bytenr: logical bytenr where the write will be
777 * @page: page to add to the bio
778 * @size: portion of page that we want to write to
779 * @pg_offset: offset of the new bio or to check whether we are adding
780 * a contiguous page to the previous one
782 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
783 * new one in @bio_ctrl->bbio.
784 * The mirror number for this IO should already be initizlied in
785 * @bio_ctrl->mirror_num.
787 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
788 u64 disk_bytenr, struct page *page,
789 size_t size, unsigned long pg_offset)
791 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
793 ASSERT(pg_offset + size <= PAGE_SIZE);
794 ASSERT(bio_ctrl->end_io_func);
796 if (bio_ctrl->bbio &&
797 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
798 submit_one_bio(bio_ctrl);
803 /* Allocate new bio if needed */
804 if (!bio_ctrl->bbio) {
805 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
806 page_offset(page) + pg_offset);
809 /* Cap to the current ordered extent boundary if there is one. */
810 if (len > bio_ctrl->len_to_oe_boundary) {
811 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
812 ASSERT(is_data_inode(&inode->vfs_inode));
813 len = bio_ctrl->len_to_oe_boundary;
816 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
817 /* bio full: move on to a new one */
818 submit_one_bio(bio_ctrl);
823 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
830 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
831 * sector aligned. alloc_new_bio() then sets it to the end of
832 * our ordered extent for writes into zoned devices.
834 * When len_to_oe_boundary is tracking an ordered extent, we
835 * trust the ordered extent code to align things properly, and
836 * the check above to cap our write to the ordered extent
837 * boundary is correct.
839 * When len_to_oe_boundary is U32_MAX, the cap above would
840 * result in a 4095 byte IO for the last page right before
841 * we hit the bio limit of UINT_MAX. bio_add_page() has all
842 * the checks required to make sure we don't overflow the bio,
843 * and we should just ignore len_to_oe_boundary completely
844 * unless we're using it to track an ordered extent.
846 * It's pretty hard to make a bio sized U32_MAX, but it can
847 * happen when the page cache is able to feed us contiguous
848 * pages for large extents.
850 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
851 bio_ctrl->len_to_oe_boundary -= len;
853 /* Ordered extent boundary: move on to a new bio. */
854 if (bio_ctrl->len_to_oe_boundary == 0)
855 submit_one_bio(bio_ctrl);
859 static int attach_extent_buffer_page(struct extent_buffer *eb,
861 struct btrfs_subpage *prealloc)
863 struct btrfs_fs_info *fs_info = eb->fs_info;
867 * If the page is mapped to btree inode, we should hold the private
868 * lock to prevent race.
869 * For cloned or dummy extent buffers, their pages are not mapped and
870 * will not race with any other ebs.
873 lockdep_assert_held(&page->mapping->private_lock);
875 if (fs_info->nodesize >= PAGE_SIZE) {
876 if (!PagePrivate(page))
877 attach_page_private(page, eb);
879 WARN_ON(page->private != (unsigned long)eb);
883 /* Already mapped, just free prealloc */
884 if (PagePrivate(page)) {
885 btrfs_free_subpage(prealloc);
890 /* Has preallocated memory for subpage */
891 attach_page_private(page, prealloc);
893 /* Do new allocation to attach subpage */
894 ret = btrfs_attach_subpage(fs_info, page,
895 BTRFS_SUBPAGE_METADATA);
899 int set_page_extent_mapped(struct page *page)
901 struct btrfs_fs_info *fs_info;
903 ASSERT(page->mapping);
905 if (PagePrivate(page))
908 fs_info = btrfs_sb(page->mapping->host->i_sb);
910 if (btrfs_is_subpage(fs_info, page))
911 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
913 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
917 void clear_page_extent_mapped(struct page *page)
919 struct btrfs_fs_info *fs_info;
921 ASSERT(page->mapping);
923 if (!PagePrivate(page))
926 fs_info = btrfs_sb(page->mapping->host->i_sb);
927 if (btrfs_is_subpage(fs_info, page))
928 return btrfs_detach_subpage(fs_info, page);
930 detach_page_private(page);
933 static struct extent_map *
934 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
935 u64 start, u64 len, struct extent_map **em_cached)
937 struct extent_map *em;
939 if (em_cached && *em_cached) {
941 if (extent_map_in_tree(em) && start >= em->start &&
942 start < extent_map_end(em)) {
943 refcount_inc(&em->refs);
951 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
952 if (em_cached && !IS_ERR(em)) {
954 refcount_inc(&em->refs);
960 * basic readpage implementation. Locked extent state structs are inserted
961 * into the tree that are removed when the IO is done (by the end_io
963 * XXX JDM: This needs looking at to ensure proper page locking
964 * return 0 on success, otherwise return error
966 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
967 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
969 struct inode *inode = page->mapping->host;
970 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
971 u64 start = page_offset(page);
972 const u64 end = start + PAGE_SIZE - 1;
975 u64 last_byte = i_size_read(inode);
977 struct extent_map *em;
979 size_t pg_offset = 0;
981 size_t blocksize = inode->i_sb->s_blocksize;
982 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
984 ret = set_page_extent_mapped(page);
986 unlock_extent(tree, start, end, NULL);
991 if (page->index == last_byte >> PAGE_SHIFT) {
992 size_t zero_offset = offset_in_page(last_byte);
995 iosize = PAGE_SIZE - zero_offset;
996 memzero_page(page, zero_offset, iosize);
999 bio_ctrl->end_io_func = end_bio_extent_readpage;
1000 begin_page_read(fs_info, page);
1001 while (cur <= end) {
1002 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1003 bool force_bio_submit = false;
1006 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1007 if (cur >= last_byte) {
1008 iosize = PAGE_SIZE - pg_offset;
1009 memzero_page(page, pg_offset, iosize);
1010 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1011 end_page_read(page, true, cur, iosize);
1014 em = __get_extent_map(inode, page, pg_offset, cur,
1015 end - cur + 1, em_cached);
1017 unlock_extent(tree, cur, end, NULL);
1018 end_page_read(page, false, cur, end + 1 - cur);
1021 extent_offset = cur - em->start;
1022 BUG_ON(extent_map_end(em) <= cur);
1025 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1026 compress_type = em->compress_type;
1028 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1029 iosize = ALIGN(iosize, blocksize);
1030 if (compress_type != BTRFS_COMPRESS_NONE)
1031 disk_bytenr = em->block_start;
1033 disk_bytenr = em->block_start + extent_offset;
1034 block_start = em->block_start;
1035 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1036 block_start = EXTENT_MAP_HOLE;
1039 * If we have a file range that points to a compressed extent
1040 * and it's followed by a consecutive file range that points
1041 * to the same compressed extent (possibly with a different
1042 * offset and/or length, so it either points to the whole extent
1043 * or only part of it), we must make sure we do not submit a
1044 * single bio to populate the pages for the 2 ranges because
1045 * this makes the compressed extent read zero out the pages
1046 * belonging to the 2nd range. Imagine the following scenario:
1049 * [0 - 8K] [8K - 24K]
1052 * points to extent X, points to extent X,
1053 * offset 4K, length of 8K offset 0, length 16K
1055 * [extent X, compressed length = 4K uncompressed length = 16K]
1057 * If the bio to read the compressed extent covers both ranges,
1058 * it will decompress extent X into the pages belonging to the
1059 * first range and then it will stop, zeroing out the remaining
1060 * pages that belong to the other range that points to extent X.
1061 * So here we make sure we submit 2 bios, one for the first
1062 * range and another one for the third range. Both will target
1063 * the same physical extent from disk, but we can't currently
1064 * make the compressed bio endio callback populate the pages
1065 * for both ranges because each compressed bio is tightly
1066 * coupled with a single extent map, and each range can have
1067 * an extent map with a different offset value relative to the
1068 * uncompressed data of our extent and different lengths. This
1069 * is a corner case so we prioritize correctness over
1070 * non-optimal behavior (submitting 2 bios for the same extent).
1072 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
1073 prev_em_start && *prev_em_start != (u64)-1 &&
1074 *prev_em_start != em->start)
1075 force_bio_submit = true;
1078 *prev_em_start = em->start;
1080 free_extent_map(em);
1083 /* we've found a hole, just zero and go on */
1084 if (block_start == EXTENT_MAP_HOLE) {
1085 memzero_page(page, pg_offset, iosize);
1087 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1088 end_page_read(page, true, cur, iosize);
1090 pg_offset += iosize;
1093 /* the get_extent function already copied into the page */
1094 if (block_start == EXTENT_MAP_INLINE) {
1095 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1096 end_page_read(page, true, cur, iosize);
1098 pg_offset += iosize;
1102 if (bio_ctrl->compress_type != compress_type) {
1103 submit_one_bio(bio_ctrl);
1104 bio_ctrl->compress_type = compress_type;
1107 if (force_bio_submit)
1108 submit_one_bio(bio_ctrl);
1109 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1112 pg_offset += iosize;
1118 int btrfs_read_folio(struct file *file, struct folio *folio)
1120 struct page *page = &folio->page;
1121 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1122 u64 start = page_offset(page);
1123 u64 end = start + PAGE_SIZE - 1;
1124 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1127 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1129 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL);
1131 * If btrfs_do_readpage() failed we will want to submit the assembled
1132 * bio to do the cleanup.
1134 submit_one_bio(&bio_ctrl);
1138 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1140 struct extent_map **em_cached,
1141 struct btrfs_bio_ctrl *bio_ctrl,
1144 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1147 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1149 for (index = 0; index < nr_pages; index++) {
1150 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1152 put_page(pages[index]);
1157 * helper for __extent_writepage, doing all of the delayed allocation setup.
1159 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1160 * to write the page (copy into inline extent). In this case the IO has
1161 * been started and the page is already unlocked.
1163 * This returns 0 if all went well (page still locked)
1164 * This returns < 0 if there were errors (page still locked)
1166 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1167 struct page *page, struct writeback_control *wbc)
1169 const u64 page_start = page_offset(page);
1170 const u64 page_end = page_start + PAGE_SIZE - 1;
1171 u64 delalloc_start = page_start;
1172 u64 delalloc_end = page_end;
1173 u64 delalloc_to_write = 0;
1176 while (delalloc_start < page_end) {
1177 delalloc_end = page_end;
1178 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1179 &delalloc_start, &delalloc_end)) {
1180 delalloc_start = delalloc_end + 1;
1184 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1189 delalloc_start = delalloc_end + 1;
1193 * delalloc_end is already one less than the total length, so
1194 * we don't subtract one from PAGE_SIZE
1196 delalloc_to_write +=
1197 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1200 * If btrfs_run_dealloc_range() already started I/O and unlocked
1201 * the pages, we just need to account for them here.
1204 wbc->nr_to_write -= delalloc_to_write;
1208 if (wbc->nr_to_write < delalloc_to_write) {
1211 if (delalloc_to_write < thresh * 2)
1212 thresh = delalloc_to_write;
1213 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1221 * Find the first byte we need to write.
1223 * For subpage, one page can contain several sectors, and
1224 * __extent_writepage_io() will just grab all extent maps in the page
1225 * range and try to submit all non-inline/non-compressed extents.
1227 * This is a big problem for subpage, we shouldn't re-submit already written
1229 * This function will lookup subpage dirty bit to find which range we really
1232 * Return the next dirty range in [@start, @end).
1233 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1235 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1236 struct page *page, u64 *start, u64 *end)
1238 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1239 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1240 u64 orig_start = *start;
1241 /* Declare as unsigned long so we can use bitmap ops */
1242 unsigned long flags;
1243 int range_start_bit;
1247 * For regular sector size == page size case, since one page only
1248 * contains one sector, we return the page offset directly.
1250 if (!btrfs_is_subpage(fs_info, page)) {
1251 *start = page_offset(page);
1252 *end = page_offset(page) + PAGE_SIZE;
1256 range_start_bit = spi->dirty_offset +
1257 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1259 /* We should have the page locked, but just in case */
1260 spin_lock_irqsave(&subpage->lock, flags);
1261 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1262 spi->dirty_offset + spi->bitmap_nr_bits);
1263 spin_unlock_irqrestore(&subpage->lock, flags);
1265 range_start_bit -= spi->dirty_offset;
1266 range_end_bit -= spi->dirty_offset;
1268 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1269 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1273 * helper for __extent_writepage. This calls the writepage start hooks,
1274 * and does the loop to map the page into extents and bios.
1276 * We return 1 if the IO is started and the page is unlocked,
1277 * 0 if all went well (page still locked)
1278 * < 0 if there were errors (page still locked)
1280 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1282 struct btrfs_bio_ctrl *bio_ctrl,
1286 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1287 u64 cur = page_offset(page);
1288 u64 end = cur + PAGE_SIZE - 1;
1291 struct extent_map *em;
1295 ret = btrfs_writepage_cow_fixup(page);
1297 /* Fixup worker will requeue */
1298 redirty_page_for_writepage(bio_ctrl->wbc, page);
1303 bio_ctrl->end_io_func = end_bio_extent_writepage;
1304 while (cur <= end) {
1305 u32 len = end - cur + 1;
1308 u64 dirty_range_start = cur;
1309 u64 dirty_range_end;
1312 if (cur >= i_size) {
1313 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1316 * This range is beyond i_size, thus we don't need to
1317 * bother writing back.
1318 * But we still need to clear the dirty subpage bit, or
1319 * the next time the page gets dirtied, we will try to
1320 * writeback the sectors with subpage dirty bits,
1321 * causing writeback without ordered extent.
1323 btrfs_page_clear_dirty(fs_info, page, cur, len);
1327 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1329 if (cur < dirty_range_start) {
1330 cur = dirty_range_start;
1334 em = btrfs_get_extent(inode, NULL, 0, cur, len);
1336 ret = PTR_ERR_OR_ZERO(em);
1340 extent_offset = cur - em->start;
1341 em_end = extent_map_end(em);
1342 ASSERT(cur <= em_end);
1344 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1345 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1347 block_start = em->block_start;
1348 disk_bytenr = em->block_start + extent_offset;
1350 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
1351 ASSERT(block_start != EXTENT_MAP_HOLE);
1352 ASSERT(block_start != EXTENT_MAP_INLINE);
1355 * Note that em_end from extent_map_end() and dirty_range_end from
1356 * find_next_dirty_byte() are all exclusive
1358 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1359 free_extent_map(em);
1362 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1363 if (!PageWriteback(page)) {
1364 btrfs_err(inode->root->fs_info,
1365 "page %lu not writeback, cur %llu end %llu",
1366 page->index, cur, end);
1370 * Although the PageDirty bit is cleared before entering this
1371 * function, subpage dirty bit is not cleared.
1372 * So clear subpage dirty bit here so next time we won't submit
1373 * page for range already written to disk.
1375 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
1377 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1378 cur - page_offset(page));
1383 btrfs_page_assert_not_dirty(fs_info, page);
1389 * If we finish without problem, we should not only clear page dirty,
1390 * but also empty subpage dirty bits
1397 * the writepage semantics are similar to regular writepage. extent
1398 * records are inserted to lock ranges in the tree, and as dirty areas
1399 * are found, they are marked writeback. Then the lock bits are removed
1400 * and the end_io handler clears the writeback ranges
1402 * Return 0 if everything goes well.
1403 * Return <0 for error.
1405 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1407 struct folio *folio = page_folio(page);
1408 struct inode *inode = page->mapping->host;
1409 const u64 page_start = page_offset(page);
1413 loff_t i_size = i_size_read(inode);
1414 unsigned long end_index = i_size >> PAGE_SHIFT;
1416 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1418 WARN_ON(!PageLocked(page));
1420 pg_offset = offset_in_page(i_size);
1421 if (page->index > end_index ||
1422 (page->index == end_index && !pg_offset)) {
1423 folio_invalidate(folio, 0, folio_size(folio));
1424 folio_unlock(folio);
1428 if (page->index == end_index)
1429 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1431 ret = set_page_extent_mapped(page);
1435 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1441 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1445 bio_ctrl->wbc->nr_to_write--;
1449 /* make sure the mapping tag for page dirty gets cleared */
1450 set_page_writeback(page);
1451 end_page_writeback(page);
1454 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1456 mapping_set_error(page->mapping, ret);
1463 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1465 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1466 TASK_UNINTERRUPTIBLE);
1470 * Lock extent buffer status and pages for writeback.
1472 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1473 * extent buffer is not dirty)
1474 * Return %true is the extent buffer is submitted to bio.
1476 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1477 struct writeback_control *wbc)
1479 struct btrfs_fs_info *fs_info = eb->fs_info;
1482 btrfs_tree_lock(eb);
1483 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1484 btrfs_tree_unlock(eb);
1485 if (wbc->sync_mode != WB_SYNC_ALL)
1487 wait_on_extent_buffer_writeback(eb);
1488 btrfs_tree_lock(eb);
1492 * We need to do this to prevent races in people who check if the eb is
1493 * under IO since we can end up having no IO bits set for a short period
1496 spin_lock(&eb->refs_lock);
1497 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1498 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1499 spin_unlock(&eb->refs_lock);
1500 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1501 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1503 fs_info->dirty_metadata_batch);
1506 spin_unlock(&eb->refs_lock);
1508 btrfs_tree_unlock(eb);
1512 static void set_btree_ioerr(struct extent_buffer *eb)
1514 struct btrfs_fs_info *fs_info = eb->fs_info;
1516 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1519 * A read may stumble upon this buffer later, make sure that it gets an
1520 * error and knows there was an error.
1522 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1525 * We need to set the mapping with the io error as well because a write
1526 * error will flip the file system readonly, and then syncfs() will
1527 * return a 0 because we are readonly if we don't modify the err seq for
1530 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1533 * If writeback for a btree extent that doesn't belong to a log tree
1534 * failed, increment the counter transaction->eb_write_errors.
1535 * We do this because while the transaction is running and before it's
1536 * committing (when we call filemap_fdata[write|wait]_range against
1537 * the btree inode), we might have
1538 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1539 * returns an error or an error happens during writeback, when we're
1540 * committing the transaction we wouldn't know about it, since the pages
1541 * can be no longer dirty nor marked anymore for writeback (if a
1542 * subsequent modification to the extent buffer didn't happen before the
1543 * transaction commit), which makes filemap_fdata[write|wait]_range not
1544 * able to find the pages tagged with SetPageError at transaction
1545 * commit time. So if this happens we must abort the transaction,
1546 * otherwise we commit a super block with btree roots that point to
1547 * btree nodes/leafs whose content on disk is invalid - either garbage
1548 * or the content of some node/leaf from a past generation that got
1549 * cowed or deleted and is no longer valid.
1551 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1552 * not be enough - we need to distinguish between log tree extents vs
1553 * non-log tree extents, and the next filemap_fdatawait_range() call
1554 * will catch and clear such errors in the mapping - and that call might
1555 * be from a log sync and not from a transaction commit. Also, checking
1556 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1557 * not done and would not be reliable - the eb might have been released
1558 * from memory and reading it back again means that flag would not be
1559 * set (since it's a runtime flag, not persisted on disk).
1561 * Using the flags below in the btree inode also makes us achieve the
1562 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1563 * writeback for all dirty pages and before filemap_fdatawait_range()
1564 * is called, the writeback for all dirty pages had already finished
1565 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1566 * filemap_fdatawait_range() would return success, as it could not know
1567 * that writeback errors happened (the pages were no longer tagged for
1570 switch (eb->log_index) {
1572 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1575 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1578 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1581 BUG(); /* unexpected, logic error */
1586 * The endio specific version which won't touch any unsafe spinlock in endio
1589 static struct extent_buffer *find_extent_buffer_nolock(
1590 struct btrfs_fs_info *fs_info, u64 start)
1592 struct extent_buffer *eb;
1595 eb = radix_tree_lookup(&fs_info->buffer_radix,
1596 start >> fs_info->sectorsize_bits);
1597 if (eb && atomic_inc_not_zero(&eb->refs)) {
1605 static void extent_buffer_write_end_io(struct btrfs_bio *bbio)
1607 struct extent_buffer *eb = bbio->private;
1608 struct btrfs_fs_info *fs_info = eb->fs_info;
1609 bool uptodate = !bbio->bio.bi_status;
1610 struct bvec_iter_all iter_all;
1611 struct bio_vec *bvec;
1615 set_btree_ioerr(eb);
1617 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
1618 u64 start = eb->start + bio_offset;
1619 struct page *page = bvec->bv_page;
1620 u32 len = bvec->bv_len;
1622 btrfs_page_clear_writeback(fs_info, page, start, len);
1626 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1627 smp_mb__after_atomic();
1628 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1630 bio_put(&bbio->bio);
1633 static void prepare_eb_write(struct extent_buffer *eb)
1636 unsigned long start;
1639 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1641 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1642 nritems = btrfs_header_nritems(eb);
1643 if (btrfs_header_level(eb) > 0) {
1644 end = btrfs_node_key_ptr_offset(eb, nritems);
1645 memzero_extent_buffer(eb, end, eb->len - end);
1649 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1651 start = btrfs_item_nr_offset(eb, nritems);
1652 end = btrfs_item_nr_offset(eb, 0);
1654 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1656 end += btrfs_item_offset(eb, nritems - 1);
1657 memzero_extent_buffer(eb, start, end - start);
1661 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1662 struct writeback_control *wbc)
1664 struct btrfs_fs_info *fs_info = eb->fs_info;
1665 struct btrfs_bio *bbio;
1667 prepare_eb_write(eb);
1669 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1670 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1671 eb->fs_info, extent_buffer_write_end_io, eb);
1672 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1673 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1674 wbc_init_bio(wbc, &bbio->bio);
1675 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1676 bbio->file_offset = eb->start;
1677 if (fs_info->nodesize < PAGE_SIZE) {
1678 struct page *p = eb->pages[0];
1681 btrfs_subpage_set_writeback(fs_info, p, eb->start, eb->len);
1682 if (btrfs_subpage_clear_and_test_dirty(fs_info, p, eb->start,
1684 clear_page_dirty_for_io(p);
1687 __bio_add_page(&bbio->bio, p, eb->len, eb->start - page_offset(p));
1688 wbc_account_cgroup_owner(wbc, p, eb->len);
1691 for (int i = 0; i < num_extent_pages(eb); i++) {
1692 struct page *p = eb->pages[i];
1695 clear_page_dirty_for_io(p);
1696 set_page_writeback(p);
1697 __bio_add_page(&bbio->bio, p, PAGE_SIZE, 0);
1698 wbc_account_cgroup_owner(wbc, p, PAGE_SIZE);
1703 btrfs_submit_bio(bbio, 0);
1707 * Submit one subpage btree page.
1709 * The main difference to submit_eb_page() is:
1711 * For subpage, we don't rely on page locking at all.
1714 * We only flush bio if we may be unable to fit current extent buffers into
1717 * Return >=0 for the number of submitted extent buffers.
1718 * Return <0 for fatal error.
1720 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1722 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1724 u64 page_start = page_offset(page);
1726 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1728 /* Lock and write each dirty extent buffers in the range */
1729 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1730 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1731 struct extent_buffer *eb;
1732 unsigned long flags;
1736 * Take private lock to ensure the subpage won't be detached
1739 spin_lock(&page->mapping->private_lock);
1740 if (!PagePrivate(page)) {
1741 spin_unlock(&page->mapping->private_lock);
1744 spin_lock_irqsave(&subpage->lock, flags);
1745 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1746 subpage->bitmaps)) {
1747 spin_unlock_irqrestore(&subpage->lock, flags);
1748 spin_unlock(&page->mapping->private_lock);
1753 start = page_start + bit_start * fs_info->sectorsize;
1754 bit_start += sectors_per_node;
1757 * Here we just want to grab the eb without touching extra
1758 * spin locks, so call find_extent_buffer_nolock().
1760 eb = find_extent_buffer_nolock(fs_info, start);
1761 spin_unlock_irqrestore(&subpage->lock, flags);
1762 spin_unlock(&page->mapping->private_lock);
1765 * The eb has already reached 0 refs thus find_extent_buffer()
1766 * doesn't return it. We don't need to write back such eb
1772 if (lock_extent_buffer_for_io(eb, wbc)) {
1773 write_one_eb(eb, wbc);
1776 free_extent_buffer(eb);
1782 * Submit all page(s) of one extent buffer.
1784 * @page: the page of one extent buffer
1785 * @eb_context: to determine if we need to submit this page, if current page
1786 * belongs to this eb, we don't need to submit
1788 * The caller should pass each page in their bytenr order, and here we use
1789 * @eb_context to determine if we have submitted pages of one extent buffer.
1791 * If we have, we just skip until we hit a new page that doesn't belong to
1792 * current @eb_context.
1794 * If not, we submit all the page(s) of the extent buffer.
1796 * Return >0 if we have submitted the extent buffer successfully.
1797 * Return 0 if we don't need to submit the page, as it's already submitted by
1799 * Return <0 for fatal error.
1801 static int submit_eb_page(struct page *page, struct btrfs_eb_write_context *ctx)
1803 struct writeback_control *wbc = ctx->wbc;
1804 struct address_space *mapping = page->mapping;
1805 struct extent_buffer *eb;
1808 if (!PagePrivate(page))
1811 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
1812 return submit_eb_subpage(page, wbc);
1814 spin_lock(&mapping->private_lock);
1815 if (!PagePrivate(page)) {
1816 spin_unlock(&mapping->private_lock);
1820 eb = (struct extent_buffer *)page->private;
1823 * Shouldn't happen and normally this would be a BUG_ON but no point
1824 * crashing the machine for something we can survive anyway.
1827 spin_unlock(&mapping->private_lock);
1831 if (eb == ctx->eb) {
1832 spin_unlock(&mapping->private_lock);
1835 ret = atomic_inc_not_zero(&eb->refs);
1836 spin_unlock(&mapping->private_lock);
1842 ret = btrfs_check_meta_write_pointer(eb->fs_info, ctx);
1846 free_extent_buffer(eb);
1850 if (!lock_extent_buffer_for_io(eb, wbc)) {
1851 free_extent_buffer(eb);
1854 /* Implies write in zoned mode. */
1855 if (ctx->zoned_bg) {
1856 /* Mark the last eb in the block group. */
1857 btrfs_schedule_zone_finish_bg(ctx->zoned_bg, eb);
1858 ctx->zoned_bg->meta_write_pointer += eb->len;
1860 write_one_eb(eb, wbc);
1861 free_extent_buffer(eb);
1865 int btree_write_cache_pages(struct address_space *mapping,
1866 struct writeback_control *wbc)
1868 struct btrfs_eb_write_context ctx = { .wbc = wbc };
1869 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
1872 int nr_to_write_done = 0;
1873 struct folio_batch fbatch;
1874 unsigned int nr_folios;
1876 pgoff_t end; /* Inclusive */
1880 folio_batch_init(&fbatch);
1881 if (wbc->range_cyclic) {
1882 index = mapping->writeback_index; /* Start from prev offset */
1885 * Start from the beginning does not need to cycle over the
1886 * range, mark it as scanned.
1888 scanned = (index == 0);
1890 index = wbc->range_start >> PAGE_SHIFT;
1891 end = wbc->range_end >> PAGE_SHIFT;
1894 if (wbc->sync_mode == WB_SYNC_ALL)
1895 tag = PAGECACHE_TAG_TOWRITE;
1897 tag = PAGECACHE_TAG_DIRTY;
1898 btrfs_zoned_meta_io_lock(fs_info);
1900 if (wbc->sync_mode == WB_SYNC_ALL)
1901 tag_pages_for_writeback(mapping, index, end);
1902 while (!done && !nr_to_write_done && (index <= end) &&
1903 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1907 for (i = 0; i < nr_folios; i++) {
1908 struct folio *folio = fbatch.folios[i];
1910 ret = submit_eb_page(&folio->page, &ctx);
1919 * the filesystem may choose to bump up nr_to_write.
1920 * We have to make sure to honor the new nr_to_write
1923 nr_to_write_done = wbc->nr_to_write <= 0;
1925 folio_batch_release(&fbatch);
1928 if (!scanned && !done) {
1930 * We hit the last page and there is more work to be done: wrap
1931 * back to the start of the file
1938 * If something went wrong, don't allow any metadata write bio to be
1941 * This would prevent use-after-free if we had dirty pages not
1942 * cleaned up, which can still happen by fuzzed images.
1945 * Allowing existing tree block to be allocated for other trees.
1947 * - Log tree operations
1948 * Exiting tree blocks get allocated to log tree, bumps its
1949 * generation, then get cleaned in tree re-balance.
1950 * Such tree block will not be written back, since it's clean,
1951 * thus no WRITTEN flag set.
1952 * And after log writes back, this tree block is not traced by
1953 * any dirty extent_io_tree.
1955 * - Offending tree block gets re-dirtied from its original owner
1956 * Since it has bumped generation, no WRITTEN flag, it can be
1957 * reused without COWing. This tree block will not be traced
1958 * by btrfs_transaction::dirty_pages.
1960 * Now such dirty tree block will not be cleaned by any dirty
1961 * extent io tree. Thus we don't want to submit such wild eb
1962 * if the fs already has error.
1964 * We can get ret > 0 from submit_extent_page() indicating how many ebs
1965 * were submitted. Reset it to 0 to avoid false alerts for the caller.
1969 if (!ret && BTRFS_FS_ERROR(fs_info))
1973 btrfs_put_block_group(ctx.zoned_bg);
1974 btrfs_zoned_meta_io_unlock(fs_info);
1979 * Walk the list of dirty pages of the given address space and write all of them.
1981 * @mapping: address space structure to write
1982 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1983 * @bio_ctrl: holds context for the write, namely the bio
1985 * If a page is already under I/O, write_cache_pages() skips it, even
1986 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
1987 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
1988 * and msync() need to guarantee that all the data which was dirty at the time
1989 * the call was made get new I/O started against them. If wbc->sync_mode is
1990 * WB_SYNC_ALL then we were called for data integrity and we must wait for
1991 * existing IO to complete.
1993 static int extent_write_cache_pages(struct address_space *mapping,
1994 struct btrfs_bio_ctrl *bio_ctrl)
1996 struct writeback_control *wbc = bio_ctrl->wbc;
1997 struct inode *inode = mapping->host;
2000 int nr_to_write_done = 0;
2001 struct folio_batch fbatch;
2002 unsigned int nr_folios;
2004 pgoff_t end; /* Inclusive */
2006 int range_whole = 0;
2011 * We have to hold onto the inode so that ordered extents can do their
2012 * work when the IO finishes. The alternative to this is failing to add
2013 * an ordered extent if the igrab() fails there and that is a huge pain
2014 * to deal with, so instead just hold onto the inode throughout the
2015 * writepages operation. If it fails here we are freeing up the inode
2016 * anyway and we'd rather not waste our time writing out stuff that is
2017 * going to be truncated anyway.
2022 folio_batch_init(&fbatch);
2023 if (wbc->range_cyclic) {
2024 index = mapping->writeback_index; /* Start from prev offset */
2027 * Start from the beginning does not need to cycle over the
2028 * range, mark it as scanned.
2030 scanned = (index == 0);
2032 index = wbc->range_start >> PAGE_SHIFT;
2033 end = wbc->range_end >> PAGE_SHIFT;
2034 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2040 * We do the tagged writepage as long as the snapshot flush bit is set
2041 * and we are the first one who do the filemap_flush() on this inode.
2043 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2044 * not race in and drop the bit.
2046 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2047 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2048 &BTRFS_I(inode)->runtime_flags))
2049 wbc->tagged_writepages = 1;
2051 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2052 tag = PAGECACHE_TAG_TOWRITE;
2054 tag = PAGECACHE_TAG_DIRTY;
2056 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2057 tag_pages_for_writeback(mapping, index, end);
2059 while (!done && !nr_to_write_done && (index <= end) &&
2060 (nr_folios = filemap_get_folios_tag(mapping, &index,
2061 end, tag, &fbatch))) {
2064 for (i = 0; i < nr_folios; i++) {
2065 struct folio *folio = fbatch.folios[i];
2067 done_index = folio_next_index(folio);
2069 * At this point we hold neither the i_pages lock nor
2070 * the page lock: the page may be truncated or
2071 * invalidated (changing page->mapping to NULL),
2072 * or even swizzled back from swapper_space to
2073 * tmpfs file mapping
2075 if (!folio_trylock(folio)) {
2076 submit_write_bio(bio_ctrl, 0);
2080 if (unlikely(folio->mapping != mapping)) {
2081 folio_unlock(folio);
2085 if (!folio_test_dirty(folio)) {
2086 /* Someone wrote it for us. */
2087 folio_unlock(folio);
2091 if (wbc->sync_mode != WB_SYNC_NONE) {
2092 if (folio_test_writeback(folio))
2093 submit_write_bio(bio_ctrl, 0);
2094 folio_wait_writeback(folio);
2097 if (folio_test_writeback(folio) ||
2098 !folio_clear_dirty_for_io(folio)) {
2099 folio_unlock(folio);
2103 ret = __extent_writepage(&folio->page, bio_ctrl);
2110 * The filesystem may choose to bump up nr_to_write.
2111 * We have to make sure to honor the new nr_to_write
2114 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2115 wbc->nr_to_write <= 0);
2117 folio_batch_release(&fbatch);
2120 if (!scanned && !done) {
2122 * We hit the last page and there is more work to be done: wrap
2123 * back to the start of the file
2129 * If we're looping we could run into a page that is locked by a
2130 * writer and that writer could be waiting on writeback for a
2131 * page in our current bio, and thus deadlock, so flush the
2134 submit_write_bio(bio_ctrl, 0);
2138 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2139 mapping->writeback_index = done_index;
2141 btrfs_add_delayed_iput(BTRFS_I(inode));
2146 * Submit the pages in the range to bio for call sites which delalloc range has
2147 * already been ran (aka, ordered extent inserted) and all pages are still
2150 void extent_write_locked_range(struct inode *inode, struct page *locked_page,
2151 u64 start, u64 end, struct writeback_control *wbc,
2154 bool found_error = false;
2156 struct address_space *mapping = inode->i_mapping;
2157 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2158 const u32 sectorsize = fs_info->sectorsize;
2159 loff_t i_size = i_size_read(inode);
2161 struct btrfs_bio_ctrl bio_ctrl = {
2163 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2166 if (wbc->no_cgroup_owner)
2167 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2169 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2171 while (cur <= end) {
2172 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2173 u32 cur_len = cur_end + 1 - cur;
2177 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2178 ASSERT(PageLocked(page));
2179 if (pages_dirty && page != locked_page) {
2180 ASSERT(PageDirty(page));
2181 clear_page_dirty_for_io(page);
2184 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2189 /* Make sure the mapping tag for page dirty gets cleared. */
2191 set_page_writeback(page);
2192 end_page_writeback(page);
2195 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2196 cur, cur_len, !ret);
2197 mapping_set_error(page->mapping, ret);
2199 btrfs_page_unlock_writer(fs_info, page, cur, cur_len);
2207 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2210 int extent_writepages(struct address_space *mapping,
2211 struct writeback_control *wbc)
2213 struct inode *inode = mapping->host;
2215 struct btrfs_bio_ctrl bio_ctrl = {
2217 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2221 * Allow only a single thread to do the reloc work in zoned mode to
2222 * protect the write pointer updates.
2224 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2225 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2226 submit_write_bio(&bio_ctrl, ret);
2227 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2231 void extent_readahead(struct readahead_control *rac)
2233 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2234 struct page *pagepool[16];
2235 struct extent_map *em_cached = NULL;
2236 u64 prev_em_start = (u64)-1;
2239 while ((nr = readahead_page_batch(rac, pagepool))) {
2240 u64 contig_start = readahead_pos(rac);
2241 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2243 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2244 &em_cached, &bio_ctrl, &prev_em_start);
2248 free_extent_map(em_cached);
2249 submit_one_bio(&bio_ctrl);
2253 * basic invalidate_folio code, this waits on any locked or writeback
2254 * ranges corresponding to the folio, and then deletes any extent state
2255 * records from the tree
2257 int extent_invalidate_folio(struct extent_io_tree *tree,
2258 struct folio *folio, size_t offset)
2260 struct extent_state *cached_state = NULL;
2261 u64 start = folio_pos(folio);
2262 u64 end = start + folio_size(folio) - 1;
2263 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
2265 /* This function is only called for the btree inode */
2266 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2268 start += ALIGN(offset, blocksize);
2272 lock_extent(tree, start, end, &cached_state);
2273 folio_wait_writeback(folio);
2276 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2277 * so here we only need to unlock the extent range to free any
2278 * existing extent state.
2280 unlock_extent(tree, start, end, &cached_state);
2285 * a helper for release_folio, this tests for areas of the page that
2286 * are locked or under IO and drops the related state bits if it is safe
2289 static int try_release_extent_state(struct extent_io_tree *tree,
2290 struct page *page, gfp_t mask)
2292 u64 start = page_offset(page);
2293 u64 end = start + PAGE_SIZE - 1;
2296 if (test_range_bit_exists(tree, start, end, EXTENT_LOCKED)) {
2299 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2300 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS);
2303 * At this point we can safely clear everything except the
2304 * locked bit, the nodatasum bit and the delalloc new bit.
2305 * The delalloc new bit will be cleared by ordered extent
2308 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2310 /* if clear_extent_bit failed for enomem reasons,
2311 * we can't allow the release to continue.
2322 * a helper for release_folio. As long as there are no locked extents
2323 * in the range corresponding to the page, both state records and extent
2324 * map records are removed
2326 int try_release_extent_mapping(struct page *page, gfp_t mask)
2328 struct extent_map *em;
2329 u64 start = page_offset(page);
2330 u64 end = start + PAGE_SIZE - 1;
2331 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
2332 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2333 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2335 if (gfpflags_allow_blocking(mask) &&
2336 page->mapping->host->i_size > SZ_16M) {
2338 while (start <= end) {
2339 struct btrfs_fs_info *fs_info;
2342 len = end - start + 1;
2343 write_lock(&map->lock);
2344 em = lookup_extent_mapping(map, start, len);
2346 write_unlock(&map->lock);
2349 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2350 em->start != start) {
2351 write_unlock(&map->lock);
2352 free_extent_map(em);
2355 if (test_range_bit_exists(tree, em->start,
2356 extent_map_end(em) - 1,
2360 * If it's not in the list of modified extents, used
2361 * by a fast fsync, we can remove it. If it's being
2362 * logged we can safely remove it since fsync took an
2363 * extra reference on the em.
2365 if (list_empty(&em->list) ||
2366 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
2369 * If it's in the list of modified extents, remove it
2370 * only if its generation is older then the current one,
2371 * in which case we don't need it for a fast fsync.
2372 * Otherwise don't remove it, we could be racing with an
2373 * ongoing fast fsync that could miss the new extent.
2375 fs_info = btrfs_inode->root->fs_info;
2376 spin_lock(&fs_info->trans_lock);
2377 cur_gen = fs_info->generation;
2378 spin_unlock(&fs_info->trans_lock);
2379 if (em->generation >= cur_gen)
2383 * We only remove extent maps that are not in the list of
2384 * modified extents or that are in the list but with a
2385 * generation lower then the current generation, so there
2386 * is no need to set the full fsync flag on the inode (it
2387 * hurts the fsync performance for workloads with a data
2388 * size that exceeds or is close to the system's memory).
2390 remove_extent_mapping(map, em);
2391 /* once for the rb tree */
2392 free_extent_map(em);
2394 start = extent_map_end(em);
2395 write_unlock(&map->lock);
2398 free_extent_map(em);
2400 cond_resched(); /* Allow large-extent preemption. */
2403 return try_release_extent_state(tree, page, mask);
2407 * To cache previous fiemap extent
2409 * Will be used for merging fiemap extent
2411 struct fiemap_cache {
2420 * Helper to submit fiemap extent.
2422 * Will try to merge current fiemap extent specified by @offset, @phys,
2423 * @len and @flags with cached one.
2424 * And only when we fails to merge, cached one will be submitted as
2427 * Return value is the same as fiemap_fill_next_extent().
2429 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2430 struct fiemap_cache *cache,
2431 u64 offset, u64 phys, u64 len, u32 flags)
2435 /* Set at the end of extent_fiemap(). */
2436 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2442 * Sanity check, extent_fiemap() should have ensured that new
2443 * fiemap extent won't overlap with cached one.
2446 * NOTE: Physical address can overlap, due to compression
2448 if (cache->offset + cache->len > offset) {
2454 * Only merges fiemap extents if
2455 * 1) Their logical addresses are continuous
2457 * 2) Their physical addresses are continuous
2458 * So truly compressed (physical size smaller than logical size)
2459 * extents won't get merged with each other
2461 * 3) Share same flags
2463 if (cache->offset + cache->len == offset &&
2464 cache->phys + cache->len == phys &&
2465 cache->flags == flags) {
2470 /* Not mergeable, need to submit cached one */
2471 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2472 cache->len, cache->flags);
2473 cache->cached = false;
2477 cache->cached = true;
2478 cache->offset = offset;
2481 cache->flags = flags;
2487 * Emit last fiemap cache
2489 * The last fiemap cache may still be cached in the following case:
2491 * |<- Fiemap range ->|
2492 * |<------------ First extent ----------->|
2494 * In this case, the first extent range will be cached but not emitted.
2495 * So we must emit it before ending extent_fiemap().
2497 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2498 struct fiemap_cache *cache)
2505 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2506 cache->len, cache->flags);
2507 cache->cached = false;
2513 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2515 struct extent_buffer *clone;
2516 struct btrfs_key key;
2521 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2524 ret = btrfs_next_leaf(inode->root, path);
2529 * Don't bother with cloning if there are no more file extent items for
2532 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2533 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
2536 /* See the comment at fiemap_search_slot() about why we clone. */
2537 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2541 slot = path->slots[0];
2542 btrfs_release_path(path);
2543 path->nodes[0] = clone;
2544 path->slots[0] = slot;
2550 * Search for the first file extent item that starts at a given file offset or
2551 * the one that starts immediately before that offset.
2552 * Returns: 0 on success, < 0 on error, 1 if not found.
2554 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2557 const u64 ino = btrfs_ino(inode);
2558 struct btrfs_root *root = inode->root;
2559 struct extent_buffer *clone;
2560 struct btrfs_key key;
2565 key.type = BTRFS_EXTENT_DATA_KEY;
2566 key.offset = file_offset;
2568 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2572 if (ret > 0 && path->slots[0] > 0) {
2573 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2574 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2578 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2579 ret = btrfs_next_leaf(root, path);
2583 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2584 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2589 * We clone the leaf and use it during fiemap. This is because while
2590 * using the leaf we do expensive things like checking if an extent is
2591 * shared, which can take a long time. In order to prevent blocking
2592 * other tasks for too long, we use a clone of the leaf. We have locked
2593 * the file range in the inode's io tree, so we know none of our file
2594 * extent items can change. This way we avoid blocking other tasks that
2595 * want to insert items for other inodes in the same leaf or b+tree
2596 * rebalance operations (triggered for example when someone is trying
2597 * to push items into this leaf when trying to insert an item in a
2599 * We also need the private clone because holding a read lock on an
2600 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2601 * when we call fiemap_fill_next_extent(), because that may cause a page
2602 * fault when filling the user space buffer with fiemap data.
2604 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2608 slot = path->slots[0];
2609 btrfs_release_path(path);
2610 path->nodes[0] = clone;
2611 path->slots[0] = slot;
2617 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2618 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2619 * extent. The end offset (@end) is inclusive.
2621 static int fiemap_process_hole(struct btrfs_inode *inode,
2622 struct fiemap_extent_info *fieinfo,
2623 struct fiemap_cache *cache,
2624 struct extent_state **delalloc_cached_state,
2625 struct btrfs_backref_share_check_ctx *backref_ctx,
2626 u64 disk_bytenr, u64 extent_offset,
2630 const u64 i_size = i_size_read(&inode->vfs_inode);
2631 u64 cur_offset = start;
2632 u64 last_delalloc_end = 0;
2633 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2634 bool checked_extent_shared = false;
2638 * There can be no delalloc past i_size, so don't waste time looking for
2641 while (cur_offset < end && cur_offset < i_size) {
2645 u64 prealloc_len = 0;
2648 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2649 delalloc_cached_state,
2656 * If this is a prealloc extent we have to report every section
2657 * of it that has no delalloc.
2659 if (disk_bytenr != 0) {
2660 if (last_delalloc_end == 0) {
2661 prealloc_start = start;
2662 prealloc_len = delalloc_start - start;
2664 prealloc_start = last_delalloc_end + 1;
2665 prealloc_len = delalloc_start - prealloc_start;
2669 if (prealloc_len > 0) {
2670 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2671 ret = btrfs_is_data_extent_shared(inode,
2678 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2680 checked_extent_shared = true;
2682 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2683 disk_bytenr + extent_offset,
2684 prealloc_len, prealloc_flags);
2687 extent_offset += prealloc_len;
2690 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2691 delalloc_end + 1 - delalloc_start,
2692 FIEMAP_EXTENT_DELALLOC |
2693 FIEMAP_EXTENT_UNKNOWN);
2697 last_delalloc_end = delalloc_end;
2698 cur_offset = delalloc_end + 1;
2699 extent_offset += cur_offset - delalloc_start;
2704 * Either we found no delalloc for the whole prealloc extent or we have
2705 * a prealloc extent that spans i_size or starts at or after i_size.
2707 if (disk_bytenr != 0 && last_delalloc_end < end) {
2711 if (last_delalloc_end == 0) {
2712 prealloc_start = start;
2713 prealloc_len = end + 1 - start;
2715 prealloc_start = last_delalloc_end + 1;
2716 prealloc_len = end + 1 - prealloc_start;
2719 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2720 ret = btrfs_is_data_extent_shared(inode,
2727 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2729 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2730 disk_bytenr + extent_offset,
2731 prealloc_len, prealloc_flags);
2739 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2740 struct btrfs_path *path,
2741 u64 *last_extent_end_ret)
2743 const u64 ino = btrfs_ino(inode);
2744 struct btrfs_root *root = inode->root;
2745 struct extent_buffer *leaf;
2746 struct btrfs_file_extent_item *ei;
2747 struct btrfs_key key;
2752 * Lookup the last file extent. We're not using i_size here because
2753 * there might be preallocation past i_size.
2755 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
2756 /* There can't be a file extent item at offset (u64)-1 */
2762 * For a non-existing key, btrfs_search_slot() always leaves us at a
2763 * slot > 0, except if the btree is empty, which is impossible because
2764 * at least it has the inode item for this inode and all the items for
2765 * the root inode 256.
2767 ASSERT(path->slots[0] > 0);
2769 leaf = path->nodes[0];
2770 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2771 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
2772 /* No file extent items in the subvolume tree. */
2773 *last_extent_end_ret = 0;
2778 * For an inline extent, the disk_bytenr is where inline data starts at,
2779 * so first check if we have an inline extent item before checking if we
2780 * have an implicit hole (disk_bytenr == 0).
2782 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2783 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
2784 *last_extent_end_ret = btrfs_file_extent_end(path);
2789 * Find the last file extent item that is not a hole (when NO_HOLES is
2790 * not enabled). This should take at most 2 iterations in the worst
2791 * case: we have one hole file extent item at slot 0 of a leaf and
2792 * another hole file extent item as the last item in the previous leaf.
2793 * This is because we merge file extent items that represent holes.
2795 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2796 while (disk_bytenr == 0) {
2797 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
2800 } else if (ret > 0) {
2801 /* No file extent items that are not holes. */
2802 *last_extent_end_ret = 0;
2805 leaf = path->nodes[0];
2806 ei = btrfs_item_ptr(leaf, path->slots[0],
2807 struct btrfs_file_extent_item);
2808 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2811 *last_extent_end_ret = btrfs_file_extent_end(path);
2815 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
2818 const u64 ino = btrfs_ino(inode);
2819 struct extent_state *cached_state = NULL;
2820 struct extent_state *delalloc_cached_state = NULL;
2821 struct btrfs_path *path;
2822 struct fiemap_cache cache = { 0 };
2823 struct btrfs_backref_share_check_ctx *backref_ctx;
2824 u64 last_extent_end;
2825 u64 prev_extent_end;
2828 bool stopped = false;
2831 backref_ctx = btrfs_alloc_backref_share_check_ctx();
2832 path = btrfs_alloc_path();
2833 if (!backref_ctx || !path) {
2838 lockstart = round_down(start, inode->root->fs_info->sectorsize);
2839 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
2840 prev_extent_end = lockstart;
2842 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
2843 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
2845 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
2848 btrfs_release_path(path);
2850 path->reada = READA_FORWARD;
2851 ret = fiemap_search_slot(inode, path, lockstart);
2854 } else if (ret > 0) {
2856 * No file extent item found, but we may have delalloc between
2857 * the current offset and i_size. So check for that.
2860 goto check_eof_delalloc;
2863 while (prev_extent_end < lockend) {
2864 struct extent_buffer *leaf = path->nodes[0];
2865 struct btrfs_file_extent_item *ei;
2866 struct btrfs_key key;
2869 u64 extent_offset = 0;
2871 u64 disk_bytenr = 0;
2876 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2877 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2880 extent_end = btrfs_file_extent_end(path);
2883 * The first iteration can leave us at an extent item that ends
2884 * before our range's start. Move to the next item.
2886 if (extent_end <= lockstart)
2889 backref_ctx->curr_leaf_bytenr = leaf->start;
2891 /* We have in implicit hole (NO_HOLES feature enabled). */
2892 if (prev_extent_end < key.offset) {
2893 const u64 range_end = min(key.offset, lockend) - 1;
2895 ret = fiemap_process_hole(inode, fieinfo, &cache,
2896 &delalloc_cached_state,
2897 backref_ctx, 0, 0, 0,
2898 prev_extent_end, range_end);
2901 } else if (ret > 0) {
2902 /* fiemap_fill_next_extent() told us to stop. */
2907 /* We've reached the end of the fiemap range, stop. */
2908 if (key.offset >= lockend) {
2914 extent_len = extent_end - key.offset;
2915 ei = btrfs_item_ptr(leaf, path->slots[0],
2916 struct btrfs_file_extent_item);
2917 compression = btrfs_file_extent_compression(leaf, ei);
2918 extent_type = btrfs_file_extent_type(leaf, ei);
2919 extent_gen = btrfs_file_extent_generation(leaf, ei);
2921 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2922 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2923 if (compression == BTRFS_COMPRESS_NONE)
2924 extent_offset = btrfs_file_extent_offset(leaf, ei);
2927 if (compression != BTRFS_COMPRESS_NONE)
2928 flags |= FIEMAP_EXTENT_ENCODED;
2930 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2931 flags |= FIEMAP_EXTENT_DATA_INLINE;
2932 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
2933 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
2935 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2936 ret = fiemap_process_hole(inode, fieinfo, &cache,
2937 &delalloc_cached_state,
2939 disk_bytenr, extent_offset,
2940 extent_gen, key.offset,
2942 } else if (disk_bytenr == 0) {
2943 /* We have an explicit hole. */
2944 ret = fiemap_process_hole(inode, fieinfo, &cache,
2945 &delalloc_cached_state,
2946 backref_ctx, 0, 0, 0,
2947 key.offset, extent_end - 1);
2949 /* We have a regular extent. */
2950 if (fieinfo->fi_extents_max) {
2951 ret = btrfs_is_data_extent_shared(inode,
2958 flags |= FIEMAP_EXTENT_SHARED;
2961 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
2962 disk_bytenr + extent_offset,
2968 } else if (ret > 0) {
2969 /* fiemap_fill_next_extent() told us to stop. */
2974 prev_extent_end = extent_end;
2976 if (fatal_signal_pending(current)) {
2981 ret = fiemap_next_leaf_item(inode, path);
2984 } else if (ret > 0) {
2985 /* No more file extent items for this inode. */
2993 * Release (and free) the path before emitting any final entries to
2994 * fiemap_fill_next_extent() to keep lockdep happy. This is because
2995 * once we find no more file extent items exist, we may have a
2996 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
2997 * faults when copying data to the user space buffer.
2999 btrfs_free_path(path);
3002 if (!stopped && prev_extent_end < lockend) {
3003 ret = fiemap_process_hole(inode, fieinfo, &cache,
3004 &delalloc_cached_state, backref_ctx,
3005 0, 0, 0, prev_extent_end, lockend - 1);
3008 prev_extent_end = lockend;
3011 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3012 const u64 i_size = i_size_read(&inode->vfs_inode);
3014 if (prev_extent_end < i_size) {
3019 delalloc = btrfs_find_delalloc_in_range(inode,
3022 &delalloc_cached_state,
3026 cache.flags |= FIEMAP_EXTENT_LAST;
3028 cache.flags |= FIEMAP_EXTENT_LAST;
3032 ret = emit_last_fiemap_cache(fieinfo, &cache);
3035 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3036 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3038 free_extent_state(delalloc_cached_state);
3039 btrfs_free_backref_share_ctx(backref_ctx);
3040 btrfs_free_path(path);
3044 static void __free_extent_buffer(struct extent_buffer *eb)
3046 kmem_cache_free(extent_buffer_cache, eb);
3049 static int extent_buffer_under_io(const struct extent_buffer *eb)
3051 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3052 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3055 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
3057 struct btrfs_subpage *subpage;
3059 lockdep_assert_held(&page->mapping->private_lock);
3061 if (PagePrivate(page)) {
3062 subpage = (struct btrfs_subpage *)page->private;
3063 if (atomic_read(&subpage->eb_refs))
3066 * Even there is no eb refs here, we may still have
3067 * end_page_read() call relying on page::private.
3069 if (atomic_read(&subpage->readers))
3075 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
3077 struct btrfs_fs_info *fs_info = eb->fs_info;
3078 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3081 * For mapped eb, we're going to change the page private, which should
3082 * be done under the private_lock.
3085 spin_lock(&page->mapping->private_lock);
3087 if (!PagePrivate(page)) {
3089 spin_unlock(&page->mapping->private_lock);
3093 if (fs_info->nodesize >= PAGE_SIZE) {
3095 * We do this since we'll remove the pages after we've
3096 * removed the eb from the radix tree, so we could race
3097 * and have this page now attached to the new eb. So
3098 * only clear page_private if it's still connected to
3101 if (PagePrivate(page) &&
3102 page->private == (unsigned long)eb) {
3103 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3104 BUG_ON(PageDirty(page));
3105 BUG_ON(PageWriteback(page));
3107 * We need to make sure we haven't be attached
3110 detach_page_private(page);
3113 spin_unlock(&page->mapping->private_lock);
3118 * For subpage, we can have dummy eb with page private. In this case,
3119 * we can directly detach the private as such page is only attached to
3120 * one dummy eb, no sharing.
3123 btrfs_detach_subpage(fs_info, page);
3127 btrfs_page_dec_eb_refs(fs_info, page);
3130 * We can only detach the page private if there are no other ebs in the
3131 * page range and no unfinished IO.
3133 if (!page_range_has_eb(fs_info, page))
3134 btrfs_detach_subpage(fs_info, page);
3136 spin_unlock(&page->mapping->private_lock);
3139 /* Release all pages attached to the extent buffer */
3140 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3145 ASSERT(!extent_buffer_under_io(eb));
3147 num_pages = num_extent_pages(eb);
3148 for (i = 0; i < num_pages; i++) {
3149 struct page *page = eb->pages[i];
3154 detach_extent_buffer_page(eb, page);
3156 /* One for when we allocated the page */
3162 * Helper for releasing the extent buffer.
3164 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3166 btrfs_release_extent_buffer_pages(eb);
3167 btrfs_leak_debug_del_eb(eb);
3168 __free_extent_buffer(eb);
3171 static struct extent_buffer *
3172 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3175 struct extent_buffer *eb = NULL;
3177 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3180 eb->fs_info = fs_info;
3181 init_rwsem(&eb->lock);
3183 btrfs_leak_debug_add_eb(eb);
3185 spin_lock_init(&eb->refs_lock);
3186 atomic_set(&eb->refs, 1);
3188 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3193 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3196 struct extent_buffer *new;
3197 int num_pages = num_extent_pages(src);
3200 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3205 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3206 * btrfs_release_extent_buffer() have different behavior for
3207 * UNMAPPED subpage extent buffer.
3209 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3211 ret = btrfs_alloc_page_array(num_pages, new->pages);
3213 btrfs_release_extent_buffer(new);
3217 for (i = 0; i < num_pages; i++) {
3219 struct page *p = new->pages[i];
3221 ret = attach_extent_buffer_page(new, p, NULL);
3223 btrfs_release_extent_buffer(new);
3226 WARN_ON(PageDirty(p));
3228 copy_extent_buffer_full(new, src);
3229 set_extent_buffer_uptodate(new);
3234 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3235 u64 start, unsigned long len)
3237 struct extent_buffer *eb;
3242 eb = __alloc_extent_buffer(fs_info, start, len);
3246 num_pages = num_extent_pages(eb);
3247 ret = btrfs_alloc_page_array(num_pages, eb->pages);
3251 for (i = 0; i < num_pages; i++) {
3252 struct page *p = eb->pages[i];
3254 ret = attach_extent_buffer_page(eb, p, NULL);
3259 set_extent_buffer_uptodate(eb);
3260 btrfs_set_header_nritems(eb, 0);
3261 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3265 for (i = 0; i < num_pages; i++) {
3267 detach_extent_buffer_page(eb, eb->pages[i]);
3268 __free_page(eb->pages[i]);
3271 __free_extent_buffer(eb);
3275 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3278 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3281 static void check_buffer_tree_ref(struct extent_buffer *eb)
3285 * The TREE_REF bit is first set when the extent_buffer is added
3286 * to the radix tree. It is also reset, if unset, when a new reference
3287 * is created by find_extent_buffer.
3289 * It is only cleared in two cases: freeing the last non-tree
3290 * reference to the extent_buffer when its STALE bit is set or
3291 * calling release_folio when the tree reference is the only reference.
3293 * In both cases, care is taken to ensure that the extent_buffer's
3294 * pages are not under io. However, release_folio can be concurrently
3295 * called with creating new references, which is prone to race
3296 * conditions between the calls to check_buffer_tree_ref in those
3297 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3299 * The actual lifetime of the extent_buffer in the radix tree is
3300 * adequately protected by the refcount, but the TREE_REF bit and
3301 * its corresponding reference are not. To protect against this
3302 * class of races, we call check_buffer_tree_ref from the codepaths
3303 * which trigger io. Note that once io is initiated, TREE_REF can no
3304 * longer be cleared, so that is the moment at which any such race is
3307 refs = atomic_read(&eb->refs);
3308 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3311 spin_lock(&eb->refs_lock);
3312 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3313 atomic_inc(&eb->refs);
3314 spin_unlock(&eb->refs_lock);
3317 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
3318 struct page *accessed)
3322 check_buffer_tree_ref(eb);
3324 num_pages = num_extent_pages(eb);
3325 for (i = 0; i < num_pages; i++) {
3326 struct page *p = eb->pages[i];
3329 mark_page_accessed(p);
3333 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3336 struct extent_buffer *eb;
3338 eb = find_extent_buffer_nolock(fs_info, start);
3342 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3343 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3344 * another task running free_extent_buffer() might have seen that flag
3345 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3346 * writeback flags not set) and it's still in the tree (flag
3347 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3348 * decrementing the extent buffer's reference count twice. So here we
3349 * could race and increment the eb's reference count, clear its stale
3350 * flag, mark it as dirty and drop our reference before the other task
3351 * finishes executing free_extent_buffer, which would later result in
3352 * an attempt to free an extent buffer that is dirty.
3354 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3355 spin_lock(&eb->refs_lock);
3356 spin_unlock(&eb->refs_lock);
3358 mark_extent_buffer_accessed(eb, NULL);
3362 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3363 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3366 struct extent_buffer *eb, *exists = NULL;
3369 eb = find_extent_buffer(fs_info, start);
3372 eb = alloc_dummy_extent_buffer(fs_info, start);
3374 return ERR_PTR(-ENOMEM);
3375 eb->fs_info = fs_info;
3377 ret = radix_tree_preload(GFP_NOFS);
3379 exists = ERR_PTR(ret);
3382 spin_lock(&fs_info->buffer_lock);
3383 ret = radix_tree_insert(&fs_info->buffer_radix,
3384 start >> fs_info->sectorsize_bits, eb);
3385 spin_unlock(&fs_info->buffer_lock);
3386 radix_tree_preload_end();
3387 if (ret == -EEXIST) {
3388 exists = find_extent_buffer(fs_info, start);
3394 check_buffer_tree_ref(eb);
3395 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3399 btrfs_release_extent_buffer(eb);
3404 static struct extent_buffer *grab_extent_buffer(
3405 struct btrfs_fs_info *fs_info, struct page *page)
3407 struct extent_buffer *exists;
3410 * For subpage case, we completely rely on radix tree to ensure we
3411 * don't try to insert two ebs for the same bytenr. So here we always
3412 * return NULL and just continue.
3414 if (fs_info->nodesize < PAGE_SIZE)
3417 /* Page not yet attached to an extent buffer */
3418 if (!PagePrivate(page))
3422 * We could have already allocated an eb for this page and attached one
3423 * so lets see if we can get a ref on the existing eb, and if we can we
3424 * know it's good and we can just return that one, else we know we can
3425 * just overwrite page->private.
3427 exists = (struct extent_buffer *)page->private;
3428 if (atomic_inc_not_zero(&exists->refs))
3431 WARN_ON(PageDirty(page));
3432 detach_page_private(page);
3436 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3438 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3439 btrfs_err(fs_info, "bad tree block start %llu", start);
3443 if (fs_info->nodesize < PAGE_SIZE &&
3444 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3446 "tree block crosses page boundary, start %llu nodesize %u",
3447 start, fs_info->nodesize);
3450 if (fs_info->nodesize >= PAGE_SIZE &&
3451 !PAGE_ALIGNED(start)) {
3453 "tree block is not page aligned, start %llu nodesize %u",
3454 start, fs_info->nodesize);
3457 if (!IS_ALIGNED(start, fs_info->nodesize) &&
3458 !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, &fs_info->flags)) {
3460 "tree block not nodesize aligned, start %llu nodesize %u, can be resolved by a full metadata balance",
3461 start, fs_info->nodesize);
3466 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3467 u64 start, u64 owner_root, int level)
3469 unsigned long len = fs_info->nodesize;
3472 unsigned long index = start >> PAGE_SHIFT;
3473 struct extent_buffer *eb;
3474 struct extent_buffer *exists = NULL;
3476 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3477 struct btrfs_subpage *prealloc = NULL;
3478 u64 lockdep_owner = owner_root;
3482 if (check_eb_alignment(fs_info, start))
3483 return ERR_PTR(-EINVAL);
3485 #if BITS_PER_LONG == 32
3486 if (start >= MAX_LFS_FILESIZE) {
3487 btrfs_err_rl(fs_info,
3488 "extent buffer %llu is beyond 32bit page cache limit", start);
3489 btrfs_err_32bit_limit(fs_info);
3490 return ERR_PTR(-EOVERFLOW);
3492 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3493 btrfs_warn_32bit_limit(fs_info);
3496 eb = find_extent_buffer(fs_info, start);
3500 eb = __alloc_extent_buffer(fs_info, start, len);
3502 return ERR_PTR(-ENOMEM);
3505 * The reloc trees are just snapshots, so we need them to appear to be
3506 * just like any other fs tree WRT lockdep.
3508 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3509 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3511 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3513 num_pages = num_extent_pages(eb);
3516 * Preallocate page->private for subpage case, so that we won't
3517 * allocate memory with private_lock nor page lock hold.
3519 * The memory will be freed by attach_extent_buffer_page() or freed
3520 * manually if we exit earlier.
3522 if (fs_info->nodesize < PAGE_SIZE) {
3523 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3524 if (IS_ERR(prealloc)) {
3525 exists = ERR_CAST(prealloc);
3530 for (i = 0; i < num_pages; i++, index++) {
3531 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
3533 exists = ERR_PTR(-ENOMEM);
3534 btrfs_free_subpage(prealloc);
3538 spin_lock(&mapping->private_lock);
3539 exists = grab_extent_buffer(fs_info, p);
3541 spin_unlock(&mapping->private_lock);
3544 mark_extent_buffer_accessed(exists, p);
3545 btrfs_free_subpage(prealloc);
3548 /* Should not fail, as we have preallocated the memory */
3549 ret = attach_extent_buffer_page(eb, p, prealloc);
3552 * To inform we have extra eb under allocation, so that
3553 * detach_extent_buffer_page() won't release the page private
3554 * when the eb hasn't yet been inserted into radix tree.
3556 * The ref will be decreased when the eb released the page, in
3557 * detach_extent_buffer_page().
3558 * Thus needs no special handling in error path.
3560 btrfs_page_inc_eb_refs(fs_info, p);
3561 spin_unlock(&mapping->private_lock);
3563 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
3565 if (!btrfs_page_test_uptodate(fs_info, p, eb->start, eb->len))
3569 * We can't unlock the pages just yet since the extent buffer
3570 * hasn't been properly inserted in the radix tree, this
3571 * opens a race with btree_release_folio which can free a page
3572 * while we are still filling in all pages for the buffer and
3577 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3579 ret = radix_tree_preload(GFP_NOFS);
3581 exists = ERR_PTR(ret);
3585 spin_lock(&fs_info->buffer_lock);
3586 ret = radix_tree_insert(&fs_info->buffer_radix,
3587 start >> fs_info->sectorsize_bits, eb);
3588 spin_unlock(&fs_info->buffer_lock);
3589 radix_tree_preload_end();
3590 if (ret == -EEXIST) {
3591 exists = find_extent_buffer(fs_info, start);
3597 /* add one reference for the tree */
3598 check_buffer_tree_ref(eb);
3599 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3602 * Now it's safe to unlock the pages because any calls to
3603 * btree_release_folio will correctly detect that a page belongs to a
3604 * live buffer and won't free them prematurely.
3606 for (i = 0; i < num_pages; i++)
3607 unlock_page(eb->pages[i]);
3611 WARN_ON(!atomic_dec_and_test(&eb->refs));
3612 for (i = 0; i < num_pages; i++) {
3614 unlock_page(eb->pages[i]);
3617 btrfs_release_extent_buffer(eb);
3621 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3623 struct extent_buffer *eb =
3624 container_of(head, struct extent_buffer, rcu_head);
3626 __free_extent_buffer(eb);
3629 static int release_extent_buffer(struct extent_buffer *eb)
3630 __releases(&eb->refs_lock)
3632 lockdep_assert_held(&eb->refs_lock);
3634 WARN_ON(atomic_read(&eb->refs) == 0);
3635 if (atomic_dec_and_test(&eb->refs)) {
3636 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
3637 struct btrfs_fs_info *fs_info = eb->fs_info;
3639 spin_unlock(&eb->refs_lock);
3641 spin_lock(&fs_info->buffer_lock);
3642 radix_tree_delete(&fs_info->buffer_radix,
3643 eb->start >> fs_info->sectorsize_bits);
3644 spin_unlock(&fs_info->buffer_lock);
3646 spin_unlock(&eb->refs_lock);
3649 btrfs_leak_debug_del_eb(eb);
3650 /* Should be safe to release our pages at this point */
3651 btrfs_release_extent_buffer_pages(eb);
3652 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3653 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
3654 __free_extent_buffer(eb);
3658 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
3661 spin_unlock(&eb->refs_lock);
3666 void free_extent_buffer(struct extent_buffer *eb)
3672 refs = atomic_read(&eb->refs);
3674 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
3675 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
3678 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
3682 spin_lock(&eb->refs_lock);
3683 if (atomic_read(&eb->refs) == 2 &&
3684 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
3685 !extent_buffer_under_io(eb) &&
3686 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3687 atomic_dec(&eb->refs);
3690 * I know this is terrible, but it's temporary until we stop tracking
3691 * the uptodate bits and such for the extent buffers.
3693 release_extent_buffer(eb);
3696 void free_extent_buffer_stale(struct extent_buffer *eb)
3701 spin_lock(&eb->refs_lock);
3702 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
3704 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3705 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3706 atomic_dec(&eb->refs);
3707 release_extent_buffer(eb);
3710 static void btree_clear_page_dirty(struct page *page)
3712 ASSERT(PageDirty(page));
3713 ASSERT(PageLocked(page));
3714 clear_page_dirty_for_io(page);
3715 xa_lock_irq(&page->mapping->i_pages);
3716 if (!PageDirty(page))
3717 __xa_clear_mark(&page->mapping->i_pages,
3718 page_index(page), PAGECACHE_TAG_DIRTY);
3719 xa_unlock_irq(&page->mapping->i_pages);
3722 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
3724 struct btrfs_fs_info *fs_info = eb->fs_info;
3725 struct page *page = eb->pages[0];
3728 /* btree_clear_page_dirty() needs page locked */
3730 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
3733 btree_clear_page_dirty(page);
3735 WARN_ON(atomic_read(&eb->refs) == 0);
3738 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
3739 struct extent_buffer *eb)
3741 struct btrfs_fs_info *fs_info = eb->fs_info;
3746 btrfs_assert_tree_write_locked(eb);
3748 if (trans && btrfs_header_generation(eb) != trans->transid)
3751 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
3754 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
3755 fs_info->dirty_metadata_batch);
3757 if (eb->fs_info->nodesize < PAGE_SIZE)
3758 return clear_subpage_extent_buffer_dirty(eb);
3760 num_pages = num_extent_pages(eb);
3762 for (i = 0; i < num_pages; i++) {
3763 page = eb->pages[i];
3764 if (!PageDirty(page))
3767 btree_clear_page_dirty(page);
3770 WARN_ON(atomic_read(&eb->refs) == 0);
3773 void set_extent_buffer_dirty(struct extent_buffer *eb)
3779 check_buffer_tree_ref(eb);
3781 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3783 num_pages = num_extent_pages(eb);
3784 WARN_ON(atomic_read(&eb->refs) == 0);
3785 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
3788 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
3791 * For subpage case, we can have other extent buffers in the
3792 * same page, and in clear_subpage_extent_buffer_dirty() we
3793 * have to clear page dirty without subpage lock held.
3794 * This can cause race where our page gets dirty cleared after
3797 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
3798 * its page for other reasons, we can use page lock to prevent
3802 lock_page(eb->pages[0]);
3803 for (i = 0; i < num_pages; i++)
3804 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
3805 eb->start, eb->len);
3807 unlock_page(eb->pages[0]);
3808 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
3810 eb->fs_info->dirty_metadata_batch);
3812 #ifdef CONFIG_BTRFS_DEBUG
3813 for (i = 0; i < num_pages; i++)
3814 ASSERT(PageDirty(eb->pages[i]));
3818 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
3820 struct btrfs_fs_info *fs_info = eb->fs_info;
3825 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3826 num_pages = num_extent_pages(eb);
3827 for (i = 0; i < num_pages; i++) {
3828 page = eb->pages[i];
3833 * This is special handling for metadata subpage, as regular
3834 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3836 if (fs_info->nodesize >= PAGE_SIZE)
3837 ClearPageUptodate(page);
3839 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
3844 void set_extent_buffer_uptodate(struct extent_buffer *eb)
3846 struct btrfs_fs_info *fs_info = eb->fs_info;
3851 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3852 num_pages = num_extent_pages(eb);
3853 for (i = 0; i < num_pages; i++) {
3854 page = eb->pages[i];
3857 * This is special handling for metadata subpage, as regular
3858 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3860 if (fs_info->nodesize >= PAGE_SIZE)
3861 SetPageUptodate(page);
3863 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
3868 static void extent_buffer_read_end_io(struct btrfs_bio *bbio)
3870 struct extent_buffer *eb = bbio->private;
3871 struct btrfs_fs_info *fs_info = eb->fs_info;
3872 bool uptodate = !bbio->bio.bi_status;
3873 struct bvec_iter_all iter_all;
3874 struct bio_vec *bvec;
3877 eb->read_mirror = bbio->mirror_num;
3880 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
3884 set_extent_buffer_uptodate(eb);
3886 clear_extent_buffer_uptodate(eb);
3887 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3890 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
3891 u64 start = eb->start + bio_offset;
3892 struct page *page = bvec->bv_page;
3893 u32 len = bvec->bv_len;
3896 btrfs_page_set_uptodate(fs_info, page, start, len);
3898 btrfs_page_clear_uptodate(fs_info, page, start, len);
3903 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
3904 smp_mb__after_atomic();
3905 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
3906 free_extent_buffer(eb);
3908 bio_put(&bbio->bio);
3911 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
3912 struct btrfs_tree_parent_check *check)
3914 int num_pages = num_extent_pages(eb), i;
3915 struct btrfs_bio *bbio;
3917 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3921 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
3922 * operation, which could potentially still be in flight. In this case
3923 * we simply want to return an error.
3925 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
3928 /* Someone else is already reading the buffer, just wait for it. */
3929 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
3932 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3933 eb->read_mirror = 0;
3934 check_buffer_tree_ref(eb);
3935 atomic_inc(&eb->refs);
3937 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
3938 REQ_OP_READ | REQ_META, eb->fs_info,
3939 extent_buffer_read_end_io, eb);
3940 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
3941 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
3942 bbio->file_offset = eb->start;
3943 memcpy(&bbio->parent_check, check, sizeof(*check));
3944 if (eb->fs_info->nodesize < PAGE_SIZE) {
3945 __bio_add_page(&bbio->bio, eb->pages[0], eb->len,
3946 eb->start - page_offset(eb->pages[0]));
3948 for (i = 0; i < num_pages; i++)
3949 __bio_add_page(&bbio->bio, eb->pages[i], PAGE_SIZE, 0);
3951 btrfs_submit_bio(bbio, mirror_num);
3954 if (wait == WAIT_COMPLETE) {
3955 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
3956 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3963 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
3966 btrfs_warn(eb->fs_info,
3967 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
3968 eb->start, eb->len, start, len);
3969 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3975 * Check if the [start, start + len) range is valid before reading/writing
3977 * NOTE: @start and @len are offset inside the eb, not logical address.
3979 * Caller should not touch the dst/src memory if this function returns error.
3981 static inline int check_eb_range(const struct extent_buffer *eb,
3982 unsigned long start, unsigned long len)
3984 unsigned long offset;
3986 /* start, start + len should not go beyond eb->len nor overflow */
3987 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
3988 return report_eb_range(eb, start, len);
3993 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
3994 unsigned long start, unsigned long len)
4000 char *dst = (char *)dstv;
4001 unsigned long i = get_eb_page_index(start);
4003 if (check_eb_range(eb, start, len)) {
4005 * Invalid range hit, reset the memory, so callers won't get
4006 * some random garbage for their uninitialzed memory.
4008 memset(dstv, 0, len);
4012 offset = get_eb_offset_in_page(eb, start);
4015 page = eb->pages[i];
4017 cur = min(len, (PAGE_SIZE - offset));
4018 kaddr = page_address(page);
4019 memcpy(dst, kaddr + offset, cur);
4028 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4030 unsigned long start, unsigned long len)
4036 char __user *dst = (char __user *)dstv;
4037 unsigned long i = get_eb_page_index(start);
4040 WARN_ON(start > eb->len);
4041 WARN_ON(start + len > eb->start + eb->len);
4043 offset = get_eb_offset_in_page(eb, start);
4046 page = eb->pages[i];
4048 cur = min(len, (PAGE_SIZE - offset));
4049 kaddr = page_address(page);
4050 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4064 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4065 unsigned long start, unsigned long len)
4071 char *ptr = (char *)ptrv;
4072 unsigned long i = get_eb_page_index(start);
4075 if (check_eb_range(eb, start, len))
4078 offset = get_eb_offset_in_page(eb, start);
4081 page = eb->pages[i];
4083 cur = min(len, (PAGE_SIZE - offset));
4085 kaddr = page_address(page);
4086 ret = memcmp(ptr, kaddr + offset, cur);
4099 * Check that the extent buffer is uptodate.
4101 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4102 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4104 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
4107 struct btrfs_fs_info *fs_info = eb->fs_info;
4110 * If we are using the commit root we could potentially clear a page
4111 * Uptodate while we're using the extent buffer that we've previously
4112 * looked up. We don't want to complain in this case, as the page was
4113 * valid before, we just didn't write it out. Instead we want to catch
4114 * the case where we didn't actually read the block properly, which
4115 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4117 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4120 if (fs_info->nodesize < PAGE_SIZE) {
4121 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, page,
4122 eb->start, eb->len)))
4123 btrfs_subpage_dump_bitmap(fs_info, page, eb->start, eb->len);
4125 WARN_ON(!PageUptodate(page));
4129 static void __write_extent_buffer(const struct extent_buffer *eb,
4130 const void *srcv, unsigned long start,
4131 unsigned long len, bool use_memmove)
4137 char *src = (char *)srcv;
4138 unsigned long i = get_eb_page_index(start);
4139 /* For unmapped (dummy) ebs, no need to check their uptodate status. */
4140 const bool check_uptodate = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
4142 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
4144 if (check_eb_range(eb, start, len))
4147 offset = get_eb_offset_in_page(eb, start);
4150 page = eb->pages[i];
4152 assert_eb_page_uptodate(eb, page);
4154 cur = min(len, PAGE_SIZE - offset);
4155 kaddr = page_address(page);
4157 memmove(kaddr + offset, src, cur);
4159 memcpy(kaddr + offset, src, cur);
4168 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4169 unsigned long start, unsigned long len)
4171 return __write_extent_buffer(eb, srcv, start, len, false);
4174 static void memset_extent_buffer(const struct extent_buffer *eb, int c,
4175 unsigned long start, unsigned long len)
4177 unsigned long cur = start;
4179 while (cur < start + len) {
4180 unsigned long index = get_eb_page_index(cur);
4181 unsigned int offset = get_eb_offset_in_page(eb, cur);
4182 unsigned int cur_len = min(start + len - cur, PAGE_SIZE - offset);
4183 struct page *page = eb->pages[index];
4185 assert_eb_page_uptodate(eb, page);
4186 memset(page_address(page) + offset, c, cur_len);
4192 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4195 if (check_eb_range(eb, start, len))
4197 return memset_extent_buffer(eb, 0, start, len);
4200 void copy_extent_buffer_full(const struct extent_buffer *dst,
4201 const struct extent_buffer *src)
4203 unsigned long cur = 0;
4205 ASSERT(dst->len == src->len);
4207 while (cur < src->len) {
4208 unsigned long index = get_eb_page_index(cur);
4209 unsigned long offset = get_eb_offset_in_page(src, cur);
4210 unsigned long cur_len = min(src->len, PAGE_SIZE - offset);
4211 void *addr = page_address(src->pages[index]) + offset;
4213 write_extent_buffer(dst, addr, cur, cur_len);
4219 void copy_extent_buffer(const struct extent_buffer *dst,
4220 const struct extent_buffer *src,
4221 unsigned long dst_offset, unsigned long src_offset,
4224 u64 dst_len = dst->len;
4229 unsigned long i = get_eb_page_index(dst_offset);
4231 if (check_eb_range(dst, dst_offset, len) ||
4232 check_eb_range(src, src_offset, len))
4235 WARN_ON(src->len != dst_len);
4237 offset = get_eb_offset_in_page(dst, dst_offset);
4240 page = dst->pages[i];
4241 assert_eb_page_uptodate(dst, page);
4243 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
4245 kaddr = page_address(page);
4246 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4256 * Calculate the page and offset of the byte containing the given bit number.
4258 * @eb: the extent buffer
4259 * @start: offset of the bitmap item in the extent buffer
4261 * @page_index: return index of the page in the extent buffer that contains
4262 * the given bit number
4263 * @page_offset: return offset into the page given by page_index
4265 * This helper hides the ugliness of finding the byte in an extent buffer which
4266 * contains a given bit.
4268 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4269 unsigned long start, unsigned long nr,
4270 unsigned long *page_index,
4271 size_t *page_offset)
4273 size_t byte_offset = BIT_BYTE(nr);
4277 * The byte we want is the offset of the extent buffer + the offset of
4278 * the bitmap item in the extent buffer + the offset of the byte in the
4281 offset = start + offset_in_page(eb->start) + byte_offset;
4283 *page_index = offset >> PAGE_SHIFT;
4284 *page_offset = offset_in_page(offset);
4288 * Determine whether a bit in a bitmap item is set.
4290 * @eb: the extent buffer
4291 * @start: offset of the bitmap item in the extent buffer
4292 * @nr: bit number to test
4294 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4302 eb_bitmap_offset(eb, start, nr, &i, &offset);
4303 page = eb->pages[i];
4304 assert_eb_page_uptodate(eb, page);
4305 kaddr = page_address(page);
4306 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4309 static u8 *extent_buffer_get_byte(const struct extent_buffer *eb, unsigned long bytenr)
4311 unsigned long index = get_eb_page_index(bytenr);
4313 if (check_eb_range(eb, bytenr, 1))
4315 return page_address(eb->pages[index]) + get_eb_offset_in_page(eb, bytenr);
4319 * Set an area of a bitmap to 1.
4321 * @eb: the extent buffer
4322 * @start: offset of the bitmap item in the extent buffer
4323 * @pos: bit number of the first bit
4324 * @len: number of bits to set
4326 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4327 unsigned long pos, unsigned long len)
4329 unsigned int first_byte = start + BIT_BYTE(pos);
4330 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4331 const bool same_byte = (first_byte == last_byte);
4332 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4336 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4338 /* Handle the first byte. */
4339 kaddr = extent_buffer_get_byte(eb, first_byte);
4344 /* Handle the byte aligned part. */
4345 ASSERT(first_byte + 1 <= last_byte);
4346 memset_extent_buffer(eb, 0xff, first_byte + 1, last_byte - first_byte - 1);
4348 /* Handle the last byte. */
4349 kaddr = extent_buffer_get_byte(eb, last_byte);
4350 *kaddr |= BITMAP_LAST_BYTE_MASK(pos + len);
4355 * Clear an area of a bitmap.
4357 * @eb: the extent buffer
4358 * @start: offset of the bitmap item in the extent buffer
4359 * @pos: bit number of the first bit
4360 * @len: number of bits to clear
4362 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4363 unsigned long start, unsigned long pos,
4366 unsigned int first_byte = start + BIT_BYTE(pos);
4367 unsigned int last_byte = start + BIT_BYTE(pos + len - 1);
4368 const bool same_byte = (first_byte == last_byte);
4369 u8 mask = BITMAP_FIRST_BYTE_MASK(pos);
4373 mask &= BITMAP_LAST_BYTE_MASK(pos + len);
4375 /* Handle the first byte. */
4376 kaddr = extent_buffer_get_byte(eb, first_byte);
4381 /* Handle the byte aligned part. */
4382 ASSERT(first_byte + 1 <= last_byte);
4383 memset_extent_buffer(eb, 0, first_byte + 1, last_byte - first_byte - 1);
4385 /* Handle the last byte. */
4386 kaddr = extent_buffer_get_byte(eb, last_byte);
4387 *kaddr &= ~BITMAP_LAST_BYTE_MASK(pos + len);
4390 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4392 unsigned long distance = (src > dst) ? src - dst : dst - src;
4393 return distance < len;
4396 void memcpy_extent_buffer(const struct extent_buffer *dst,
4397 unsigned long dst_offset, unsigned long src_offset,
4400 unsigned long cur_off = 0;
4402 if (check_eb_range(dst, dst_offset, len) ||
4403 check_eb_range(dst, src_offset, len))
4406 while (cur_off < len) {
4407 unsigned long cur_src = cur_off + src_offset;
4408 unsigned long pg_index = get_eb_page_index(cur_src);
4409 unsigned long pg_off = get_eb_offset_in_page(dst, cur_src);
4410 unsigned long cur_len = min(src_offset + len - cur_src,
4411 PAGE_SIZE - pg_off);
4412 void *src_addr = page_address(dst->pages[pg_index]) + pg_off;
4413 const bool use_memmove = areas_overlap(src_offset + cur_off,
4414 dst_offset + cur_off, cur_len);
4416 __write_extent_buffer(dst, src_addr, dst_offset + cur_off, cur_len,
4422 void memmove_extent_buffer(const struct extent_buffer *dst,
4423 unsigned long dst_offset, unsigned long src_offset,
4426 unsigned long dst_end = dst_offset + len - 1;
4427 unsigned long src_end = src_offset + len - 1;
4429 if (check_eb_range(dst, dst_offset, len) ||
4430 check_eb_range(dst, src_offset, len))
4433 if (dst_offset < src_offset) {
4434 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4439 unsigned long src_i;
4441 size_t dst_off_in_page;
4442 size_t src_off_in_page;
4446 src_i = get_eb_page_index(src_end);
4448 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
4449 src_off_in_page = get_eb_offset_in_page(dst, src_end);
4451 cur = min_t(unsigned long, len, src_off_in_page + 1);
4452 cur = min(cur, dst_off_in_page + 1);
4454 src_addr = page_address(dst->pages[src_i]) + src_off_in_page -
4456 use_memmove = areas_overlap(src_end - cur + 1, dst_end - cur + 1,
4459 __write_extent_buffer(dst, src_addr, dst_end - cur + 1, cur,
4468 #define GANG_LOOKUP_SIZE 16
4469 static struct extent_buffer *get_next_extent_buffer(
4470 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4472 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4473 struct extent_buffer *found = NULL;
4474 u64 page_start = page_offset(page);
4475 u64 cur = page_start;
4477 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4478 lockdep_assert_held(&fs_info->buffer_lock);
4480 while (cur < page_start + PAGE_SIZE) {
4484 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4485 (void **)gang, cur >> fs_info->sectorsize_bits,
4486 min_t(unsigned int, GANG_LOOKUP_SIZE,
4487 PAGE_SIZE / fs_info->nodesize));
4490 for (i = 0; i < ret; i++) {
4491 /* Already beyond page end */
4492 if (gang[i]->start >= page_start + PAGE_SIZE)
4495 if (gang[i]->start >= bytenr) {
4500 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4506 static int try_release_subpage_extent_buffer(struct page *page)
4508 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4509 u64 cur = page_offset(page);
4510 const u64 end = page_offset(page) + PAGE_SIZE;
4514 struct extent_buffer *eb = NULL;
4517 * Unlike try_release_extent_buffer() which uses page->private
4518 * to grab buffer, for subpage case we rely on radix tree, thus
4519 * we need to ensure radix tree consistency.
4521 * We also want an atomic snapshot of the radix tree, thus go
4522 * with spinlock rather than RCU.
4524 spin_lock(&fs_info->buffer_lock);
4525 eb = get_next_extent_buffer(fs_info, page, cur);
4527 /* No more eb in the page range after or at cur */
4528 spin_unlock(&fs_info->buffer_lock);
4531 cur = eb->start + eb->len;
4534 * The same as try_release_extent_buffer(), to ensure the eb
4535 * won't disappear out from under us.
4537 spin_lock(&eb->refs_lock);
4538 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4539 spin_unlock(&eb->refs_lock);
4540 spin_unlock(&fs_info->buffer_lock);
4543 spin_unlock(&fs_info->buffer_lock);
4546 * If tree ref isn't set then we know the ref on this eb is a
4547 * real ref, so just return, this eb will likely be freed soon
4550 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4551 spin_unlock(&eb->refs_lock);
4556 * Here we don't care about the return value, we will always
4557 * check the page private at the end. And
4558 * release_extent_buffer() will release the refs_lock.
4560 release_extent_buffer(eb);
4563 * Finally to check if we have cleared page private, as if we have
4564 * released all ebs in the page, the page private should be cleared now.
4566 spin_lock(&page->mapping->private_lock);
4567 if (!PagePrivate(page))
4571 spin_unlock(&page->mapping->private_lock);
4576 int try_release_extent_buffer(struct page *page)
4578 struct extent_buffer *eb;
4580 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4581 return try_release_subpage_extent_buffer(page);
4584 * We need to make sure nobody is changing page->private, as we rely on
4585 * page->private as the pointer to extent buffer.
4587 spin_lock(&page->mapping->private_lock);
4588 if (!PagePrivate(page)) {
4589 spin_unlock(&page->mapping->private_lock);
4593 eb = (struct extent_buffer *)page->private;
4597 * This is a little awful but should be ok, we need to make sure that
4598 * the eb doesn't disappear out from under us while we're looking at
4601 spin_lock(&eb->refs_lock);
4602 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4603 spin_unlock(&eb->refs_lock);
4604 spin_unlock(&page->mapping->private_lock);
4607 spin_unlock(&page->mapping->private_lock);
4610 * If tree ref isn't set then we know the ref on this eb is a real ref,
4611 * so just return, this page will likely be freed soon anyway.
4613 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4614 spin_unlock(&eb->refs_lock);
4618 return release_extent_buffer(eb);
4622 * Attempt to readahead a child block.
4624 * @fs_info: the fs_info
4625 * @bytenr: bytenr to read
4626 * @owner_root: objectid of the root that owns this eb
4627 * @gen: generation for the uptodate check, can be 0
4628 * @level: level for the eb
4630 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
4631 * normal uptodate check of the eb, without checking the generation. If we have
4632 * to read the block we will not block on anything.
4634 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
4635 u64 bytenr, u64 owner_root, u64 gen, int level)
4637 struct btrfs_tree_parent_check check = {
4642 struct extent_buffer *eb;
4645 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
4649 if (btrfs_buffer_uptodate(eb, gen, 1)) {
4650 free_extent_buffer(eb);
4654 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
4656 free_extent_buffer_stale(eb);
4658 free_extent_buffer(eb);
4662 * Readahead a node's child block.
4664 * @node: parent node we're reading from
4665 * @slot: slot in the parent node for the child we want to read
4667 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
4668 * the slot in the node provided.
4670 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
4672 btrfs_readahead_tree_block(node->fs_info,
4673 btrfs_node_blockptr(node, slot),
4674 btrfs_header_owner(node),
4675 btrfs_node_ptr_generation(node, slot),
4676 btrfs_header_level(node) - 1);