1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
7 #include <linux/pagemap.h>
8 #include <linux/page-flags.h>
9 #include <linux/sched/mm.h>
10 #include <linux/spinlock.h>
11 #include <linux/blkdev.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/pagevec.h>
15 #include <linux/prefetch.h>
16 #include <linux/fsverity.h>
18 #include "extent_io.h"
19 #include "extent-io-tree.h"
20 #include "extent_map.h"
22 #include "btrfs_inode.h"
24 #include "check-integrity.h"
26 #include "rcu-string.h"
31 #include "block-group.h"
32 #include "compression.h"
34 #include "accessors.h"
35 #include "file-item.h"
37 #include "dev-replace.h"
39 #include "transaction.h"
41 static struct kmem_cache *extent_buffer_cache;
43 #ifdef CONFIG_BTRFS_DEBUG
44 static inline void btrfs_leak_debug_add_eb(struct extent_buffer *eb)
46 struct btrfs_fs_info *fs_info = eb->fs_info;
49 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
50 list_add(&eb->leak_list, &fs_info->allocated_ebs);
51 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
54 static inline void btrfs_leak_debug_del_eb(struct extent_buffer *eb)
56 struct btrfs_fs_info *fs_info = eb->fs_info;
59 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
60 list_del(&eb->leak_list);
61 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
64 void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info)
66 struct extent_buffer *eb;
70 * If we didn't get into open_ctree our allocated_ebs will not be
71 * initialized, so just skip this.
73 if (!fs_info->allocated_ebs.next)
76 WARN_ON(!list_empty(&fs_info->allocated_ebs));
77 spin_lock_irqsave(&fs_info->eb_leak_lock, flags);
78 while (!list_empty(&fs_info->allocated_ebs)) {
79 eb = list_first_entry(&fs_info->allocated_ebs,
80 struct extent_buffer, leak_list);
82 "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n",
83 eb->start, eb->len, atomic_read(&eb->refs), eb->bflags,
84 btrfs_header_owner(eb));
85 list_del(&eb->leak_list);
86 kmem_cache_free(extent_buffer_cache, eb);
88 spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags);
91 #define btrfs_leak_debug_add_eb(eb) do {} while (0)
92 #define btrfs_leak_debug_del_eb(eb) do {} while (0)
96 * Structure to record info about the bio being assembled, and other info like
97 * how many bytes are there before stripe/ordered extent boundary.
99 struct btrfs_bio_ctrl {
100 struct btrfs_bio *bbio;
101 enum btrfs_compression_type compress_type;
102 u32 len_to_oe_boundary;
104 btrfs_bio_end_io_t end_io_func;
105 struct writeback_control *wbc;
108 static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl)
110 struct btrfs_bio *bbio = bio_ctrl->bbio;
115 /* Caller should ensure the bio has at least some range added */
116 ASSERT(bbio->bio.bi_iter.bi_size);
118 if (btrfs_op(&bbio->bio) == BTRFS_MAP_READ &&
119 bio_ctrl->compress_type != BTRFS_COMPRESS_NONE)
120 btrfs_submit_compressed_read(bbio);
122 btrfs_submit_bio(bbio, 0);
124 /* The bbio is owned by the end_io handler now */
125 bio_ctrl->bbio = NULL;
129 * Submit or fail the current bio in the bio_ctrl structure.
131 static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret)
133 struct btrfs_bio *bbio = bio_ctrl->bbio;
140 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
141 /* The bio is owned by the end_io handler now */
142 bio_ctrl->bbio = NULL;
144 submit_one_bio(bio_ctrl);
148 int __init extent_buffer_init_cachep(void)
150 extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer",
151 sizeof(struct extent_buffer), 0,
152 SLAB_MEM_SPREAD, NULL);
153 if (!extent_buffer_cache)
159 void __cold extent_buffer_free_cachep(void)
162 * Make sure all delayed rcu free are flushed before we
166 kmem_cache_destroy(extent_buffer_cache);
169 void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end)
171 unsigned long index = start >> PAGE_SHIFT;
172 unsigned long end_index = end >> PAGE_SHIFT;
175 while (index <= end_index) {
176 page = find_get_page(inode->i_mapping, index);
177 BUG_ON(!page); /* Pages should be in the extent_io_tree */
178 clear_page_dirty_for_io(page);
184 static void process_one_page(struct btrfs_fs_info *fs_info,
185 struct page *page, struct page *locked_page,
186 unsigned long page_ops, u64 start, u64 end)
190 ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX);
191 len = end + 1 - start;
193 if (page_ops & PAGE_SET_ORDERED)
194 btrfs_page_clamp_set_ordered(fs_info, page, start, len);
195 if (page_ops & PAGE_START_WRITEBACK) {
196 btrfs_page_clamp_clear_dirty(fs_info, page, start, len);
197 btrfs_page_clamp_set_writeback(fs_info, page, start, len);
199 if (page_ops & PAGE_END_WRITEBACK)
200 btrfs_page_clamp_clear_writeback(fs_info, page, start, len);
202 if (page != locked_page && (page_ops & PAGE_UNLOCK))
203 btrfs_page_end_writer_lock(fs_info, page, start, len);
206 static void __process_pages_contig(struct address_space *mapping,
207 struct page *locked_page, u64 start, u64 end,
208 unsigned long page_ops)
210 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
211 pgoff_t start_index = start >> PAGE_SHIFT;
212 pgoff_t end_index = end >> PAGE_SHIFT;
213 pgoff_t index = start_index;
214 struct folio_batch fbatch;
217 folio_batch_init(&fbatch);
218 while (index <= end_index) {
221 found_folios = filemap_get_folios_contig(mapping, &index,
223 for (i = 0; i < found_folios; i++) {
224 struct folio *folio = fbatch.folios[i];
226 process_one_page(fs_info, &folio->page, locked_page,
227 page_ops, start, end);
229 folio_batch_release(&fbatch);
234 static noinline void __unlock_for_delalloc(struct inode *inode,
235 struct page *locked_page,
238 unsigned long index = start >> PAGE_SHIFT;
239 unsigned long end_index = end >> PAGE_SHIFT;
242 if (index == locked_page->index && end_index == index)
245 __process_pages_contig(inode->i_mapping, locked_page, start, end,
249 static noinline int lock_delalloc_pages(struct inode *inode,
250 struct page *locked_page,
254 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
255 struct address_space *mapping = inode->i_mapping;
256 pgoff_t start_index = start >> PAGE_SHIFT;
257 pgoff_t end_index = end >> PAGE_SHIFT;
258 pgoff_t index = start_index;
259 u64 processed_end = start;
260 struct folio_batch fbatch;
262 if (index == locked_page->index && index == end_index)
265 folio_batch_init(&fbatch);
266 while (index <= end_index) {
267 unsigned int found_folios, i;
269 found_folios = filemap_get_folios_contig(mapping, &index,
271 if (found_folios == 0)
274 for (i = 0; i < found_folios; i++) {
275 struct page *page = &fbatch.folios[i]->page;
276 u32 len = end + 1 - start;
278 if (page == locked_page)
281 if (btrfs_page_start_writer_lock(fs_info, page, start,
285 if (!PageDirty(page) || page->mapping != mapping) {
286 btrfs_page_end_writer_lock(fs_info, page, start,
291 processed_end = page_offset(page) + PAGE_SIZE - 1;
293 folio_batch_release(&fbatch);
299 folio_batch_release(&fbatch);
300 if (processed_end > start)
301 __unlock_for_delalloc(inode, locked_page, start, processed_end);
306 * Find and lock a contiguous range of bytes in the file marked as delalloc, no
307 * more than @max_bytes.
309 * @start: The original start bytenr to search.
310 * Will store the extent range start bytenr.
311 * @end: The original end bytenr of the search range
312 * Will store the extent range end bytenr.
314 * Return true if we find a delalloc range which starts inside the original
315 * range, and @start/@end will store the delalloc range start/end.
317 * Return false if we can't find any delalloc range which starts inside the
318 * original range, and @start/@end will be the non-delalloc range start/end.
321 noinline_for_stack bool find_lock_delalloc_range(struct inode *inode,
322 struct page *locked_page, u64 *start,
325 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
326 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
327 const u64 orig_start = *start;
328 const u64 orig_end = *end;
329 /* The sanity tests may not set a valid fs_info. */
330 u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE;
334 struct extent_state *cached_state = NULL;
338 /* Caller should pass a valid @end to indicate the search range end */
339 ASSERT(orig_end > orig_start);
341 /* The range should at least cover part of the page */
342 ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE ||
343 orig_end <= page_offset(locked_page)));
345 /* step one, find a bunch of delalloc bytes starting at start */
346 delalloc_start = *start;
348 found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end,
349 max_bytes, &cached_state);
350 if (!found || delalloc_end <= *start || delalloc_start > orig_end) {
351 *start = delalloc_start;
353 /* @delalloc_end can be -1, never go beyond @orig_end */
354 *end = min(delalloc_end, orig_end);
355 free_extent_state(cached_state);
360 * start comes from the offset of locked_page. We have to lock
361 * pages in order, so we can't process delalloc bytes before
364 if (delalloc_start < *start)
365 delalloc_start = *start;
368 * make sure to limit the number of pages we try to lock down
370 if (delalloc_end + 1 - delalloc_start > max_bytes)
371 delalloc_end = delalloc_start + max_bytes - 1;
373 /* step two, lock all the pages after the page that has start */
374 ret = lock_delalloc_pages(inode, locked_page,
375 delalloc_start, delalloc_end);
376 ASSERT(!ret || ret == -EAGAIN);
377 if (ret == -EAGAIN) {
378 /* some of the pages are gone, lets avoid looping by
379 * shortening the size of the delalloc range we're searching
381 free_extent_state(cached_state);
384 max_bytes = PAGE_SIZE;
393 /* step three, lock the state bits for the whole range */
394 lock_extent(tree, delalloc_start, delalloc_end, &cached_state);
396 /* then test to make sure it is all still delalloc */
397 ret = test_range_bit(tree, delalloc_start, delalloc_end,
398 EXTENT_DELALLOC, 1, cached_state);
400 unlock_extent(tree, delalloc_start, delalloc_end,
402 __unlock_for_delalloc(inode, locked_page,
403 delalloc_start, delalloc_end);
407 free_extent_state(cached_state);
408 *start = delalloc_start;
414 void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
415 struct page *locked_page,
416 u32 clear_bits, unsigned long page_ops)
418 clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL);
420 __process_pages_contig(inode->vfs_inode.i_mapping, locked_page,
421 start, end, page_ops);
424 static bool btrfs_verify_page(struct page *page, u64 start)
426 if (!fsverity_active(page->mapping->host) ||
427 PageUptodate(page) ||
428 start >= i_size_read(page->mapping->host))
430 return fsverity_verify_page(page);
433 static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len)
435 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
437 ASSERT(page_offset(page) <= start &&
438 start + len <= page_offset(page) + PAGE_SIZE);
440 if (uptodate && btrfs_verify_page(page, start))
441 btrfs_page_set_uptodate(fs_info, page, start, len);
443 btrfs_page_clear_uptodate(fs_info, page, start, len);
445 if (!btrfs_is_subpage(fs_info, page))
448 btrfs_subpage_end_reader(fs_info, page, start, len);
452 * after a writepage IO is done, we need to:
453 * clear the uptodate bits on error
454 * clear the writeback bits in the extent tree for this IO
455 * end_page_writeback if the page has no more pending IO
457 * Scheduling is not allowed, so the extent state tree is expected
458 * to have one and only one object corresponding to this IO.
460 static void end_bio_extent_writepage(struct btrfs_bio *bbio)
462 struct bio *bio = &bbio->bio;
463 int error = blk_status_to_errno(bio->bi_status);
464 struct bio_vec *bvec;
465 struct bvec_iter_all iter_all;
467 ASSERT(!bio_flagged(bio, BIO_CLONED));
468 bio_for_each_segment_all(bvec, bio, iter_all) {
469 struct page *page = bvec->bv_page;
470 struct inode *inode = page->mapping->host;
471 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
472 const u32 sectorsize = fs_info->sectorsize;
473 u64 start = page_offset(page) + bvec->bv_offset;
474 u32 len = bvec->bv_len;
476 /* Our read/write should always be sector aligned. */
477 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
479 "partial page write in btrfs with offset %u and length %u",
480 bvec->bv_offset, bvec->bv_len);
481 else if (!IS_ALIGNED(bvec->bv_len, sectorsize))
483 "incomplete page write with offset %u and length %u",
484 bvec->bv_offset, bvec->bv_len);
486 btrfs_finish_ordered_extent(bbio->ordered, page, start, len, !error);
488 btrfs_page_clear_uptodate(fs_info, page, start, len);
489 mapping_set_error(page->mapping, error);
491 btrfs_page_clear_writeback(fs_info, page, start, len);
498 * Record previously processed extent range
500 * For endio_readpage_release_extent() to handle a full extent range, reducing
501 * the extent io operations.
503 struct processed_extent {
504 struct btrfs_inode *inode;
505 /* Start of the range in @inode */
507 /* End of the range in @inode */
513 * Try to release processed extent range
515 * May not release the extent range right now if the current range is
516 * contiguous to processed extent.
518 * Will release processed extent when any of @inode, @uptodate, the range is
519 * no longer contiguous to the processed range.
521 * Passing @inode == NULL will force processed extent to be released.
523 static void endio_readpage_release_extent(struct processed_extent *processed,
524 struct btrfs_inode *inode, u64 start, u64 end,
527 struct extent_state *cached = NULL;
528 struct extent_io_tree *tree;
530 /* The first extent, initialize @processed */
531 if (!processed->inode)
535 * Contiguous to processed extent, just uptodate the end.
537 * Several things to notice:
539 * - bio can be merged as long as on-disk bytenr is contiguous
540 * This means we can have page belonging to other inodes, thus need to
541 * check if the inode still matches.
542 * - bvec can contain range beyond current page for multi-page bvec
543 * Thus we need to do processed->end + 1 >= start check
545 if (processed->inode == inode && processed->uptodate == uptodate &&
546 processed->end + 1 >= start && end >= processed->end) {
547 processed->end = end;
551 tree = &processed->inode->io_tree;
553 * Now we don't have range contiguous to the processed range, release
554 * the processed range now.
556 unlock_extent(tree, processed->start, processed->end, &cached);
559 /* Update processed to current range */
560 processed->inode = inode;
561 processed->start = start;
562 processed->end = end;
563 processed->uptodate = uptodate;
566 static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page)
568 ASSERT(PageLocked(page));
569 if (!btrfs_is_subpage(fs_info, page))
572 ASSERT(PagePrivate(page));
573 btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE);
577 * after a readpage IO is done, we need to:
578 * clear the uptodate bits on error
579 * set the uptodate bits if things worked
580 * set the page up to date if all extents in the tree are uptodate
581 * clear the lock bit in the extent tree
582 * unlock the page if there are no other extents locked for it
584 * Scheduling is not allowed, so the extent state tree is expected
585 * to have one and only one object corresponding to this IO.
587 static void end_bio_extent_readpage(struct btrfs_bio *bbio)
589 struct bio *bio = &bbio->bio;
590 struct bio_vec *bvec;
591 struct processed_extent processed = { 0 };
593 * The offset to the beginning of a bio, since one bio can never be
594 * larger than UINT_MAX, u32 here is enough.
597 struct bvec_iter_all iter_all;
599 ASSERT(!bio_flagged(bio, BIO_CLONED));
600 bio_for_each_segment_all(bvec, bio, iter_all) {
601 bool uptodate = !bio->bi_status;
602 struct page *page = bvec->bv_page;
603 struct inode *inode = page->mapping->host;
604 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
605 const u32 sectorsize = fs_info->sectorsize;
611 "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u",
612 bio->bi_iter.bi_sector, bio->bi_status,
616 * We always issue full-sector reads, but if some block in a
617 * page fails to read, blk_update_request() will advance
618 * bv_offset and adjust bv_len to compensate. Print a warning
619 * for unaligned offsets, and an error if they don't add up to
622 if (!IS_ALIGNED(bvec->bv_offset, sectorsize))
624 "partial page read in btrfs with offset %u and length %u",
625 bvec->bv_offset, bvec->bv_len);
626 else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len,
629 "incomplete page read with offset %u and length %u",
630 bvec->bv_offset, bvec->bv_len);
632 start = page_offset(page) + bvec->bv_offset;
633 end = start + bvec->bv_len - 1;
636 if (likely(uptodate)) {
637 loff_t i_size = i_size_read(inode);
638 pgoff_t end_index = i_size >> PAGE_SHIFT;
641 * Zero out the remaining part if this range straddles
644 * Here we should only zero the range inside the bvec,
645 * not touch anything else.
647 * NOTE: i_size is exclusive while end is inclusive.
649 if (page->index == end_index && i_size <= end) {
650 u32 zero_start = max(offset_in_page(i_size),
651 offset_in_page(start));
653 zero_user_segment(page, zero_start,
654 offset_in_page(end) + 1);
658 /* Update page status and unlock. */
659 end_page_read(page, uptodate, start, len);
660 endio_readpage_release_extent(&processed, BTRFS_I(inode),
661 start, end, uptodate);
663 ASSERT(bio_offset + len > bio_offset);
667 /* Release the last extent */
668 endio_readpage_release_extent(&processed, NULL, 0, 0, false);
673 * Populate every free slot in a provided array with pages.
675 * @nr_pages: number of pages to allocate
676 * @page_array: the array to fill with pages; any existing non-null entries in
677 * the array will be skipped
679 * Return: 0 if all pages were able to be allocated;
680 * -ENOMEM otherwise, and the caller is responsible for freeing all
681 * non-null page pointers in the array.
683 int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array)
685 unsigned int allocated;
687 for (allocated = 0; allocated < nr_pages;) {
688 unsigned int last = allocated;
690 allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array);
692 if (allocated == nr_pages)
696 * During this iteration, no page could be allocated, even
697 * though alloc_pages_bulk_array() falls back to alloc_page()
698 * if it could not bulk-allocate. So we must be out of memory.
700 if (allocated == last)
703 memalloc_retry_wait(GFP_NOFS);
708 static bool btrfs_bio_is_contig(struct btrfs_bio_ctrl *bio_ctrl,
709 struct page *page, u64 disk_bytenr,
710 unsigned int pg_offset)
712 struct bio *bio = &bio_ctrl->bbio->bio;
713 struct bio_vec *bvec = bio_last_bvec_all(bio);
714 const sector_t sector = disk_bytenr >> SECTOR_SHIFT;
716 if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) {
718 * For compression, all IO should have its logical bytenr set
719 * to the starting bytenr of the compressed extent.
721 return bio->bi_iter.bi_sector == sector;
725 * The contig check requires the following conditions to be met:
727 * 1) The pages are belonging to the same inode
728 * This is implied by the call chain.
730 * 2) The range has adjacent logical bytenr
732 * 3) The range has adjacent file offset
733 * This is required for the usage of btrfs_bio->file_offset.
735 return bio_end_sector(bio) == sector &&
736 page_offset(bvec->bv_page) + bvec->bv_offset + bvec->bv_len ==
737 page_offset(page) + pg_offset;
740 static void alloc_new_bio(struct btrfs_inode *inode,
741 struct btrfs_bio_ctrl *bio_ctrl,
742 u64 disk_bytenr, u64 file_offset)
744 struct btrfs_fs_info *fs_info = inode->root->fs_info;
745 struct btrfs_bio *bbio;
747 bbio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, fs_info,
748 bio_ctrl->end_io_func, NULL);
749 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
751 bbio->file_offset = file_offset;
752 bio_ctrl->bbio = bbio;
753 bio_ctrl->len_to_oe_boundary = U32_MAX;
755 /* Limit data write bios to the ordered boundary. */
757 struct btrfs_ordered_extent *ordered;
759 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
761 bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX,
762 ordered->file_offset +
763 ordered->disk_num_bytes - file_offset);
764 bbio->ordered = ordered;
768 * Pick the last added device to support cgroup writeback. For
769 * multi-device file systems this means blk-cgroup policies have
770 * to always be set on the last added/replaced device.
771 * This is a bit odd but has been like that for a long time.
773 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
774 wbc_init_bio(bio_ctrl->wbc, &bbio->bio);
779 * @disk_bytenr: logical bytenr where the write will be
780 * @page: page to add to the bio
781 * @size: portion of page that we want to write to
782 * @pg_offset: offset of the new bio or to check whether we are adding
783 * a contiguous page to the previous one
785 * The will either add the page into the existing @bio_ctrl->bbio, or allocate a
786 * new one in @bio_ctrl->bbio.
787 * The mirror number for this IO should already be initizlied in
788 * @bio_ctrl->mirror_num.
790 static void submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl,
791 u64 disk_bytenr, struct page *page,
792 size_t size, unsigned long pg_offset)
794 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
796 ASSERT(pg_offset + size <= PAGE_SIZE);
797 ASSERT(bio_ctrl->end_io_func);
799 if (bio_ctrl->bbio &&
800 !btrfs_bio_is_contig(bio_ctrl, page, disk_bytenr, pg_offset))
801 submit_one_bio(bio_ctrl);
806 /* Allocate new bio if needed */
807 if (!bio_ctrl->bbio) {
808 alloc_new_bio(inode, bio_ctrl, disk_bytenr,
809 page_offset(page) + pg_offset);
812 /* Cap to the current ordered extent boundary if there is one. */
813 if (len > bio_ctrl->len_to_oe_boundary) {
814 ASSERT(bio_ctrl->compress_type == BTRFS_COMPRESS_NONE);
815 ASSERT(is_data_inode(&inode->vfs_inode));
816 len = bio_ctrl->len_to_oe_boundary;
819 if (bio_add_page(&bio_ctrl->bbio->bio, page, len, pg_offset) != len) {
820 /* bio full: move on to a new one */
821 submit_one_bio(bio_ctrl);
826 wbc_account_cgroup_owner(bio_ctrl->wbc, page, len);
833 * len_to_oe_boundary defaults to U32_MAX, which isn't page or
834 * sector aligned. alloc_new_bio() then sets it to the end of
835 * our ordered extent for writes into zoned devices.
837 * When len_to_oe_boundary is tracking an ordered extent, we
838 * trust the ordered extent code to align things properly, and
839 * the check above to cap our write to the ordered extent
840 * boundary is correct.
842 * When len_to_oe_boundary is U32_MAX, the cap above would
843 * result in a 4095 byte IO for the last page right before
844 * we hit the bio limit of UINT_MAX. bio_add_page() has all
845 * the checks required to make sure we don't overflow the bio,
846 * and we should just ignore len_to_oe_boundary completely
847 * unless we're using it to track an ordered extent.
849 * It's pretty hard to make a bio sized U32_MAX, but it can
850 * happen when the page cache is able to feed us contiguous
851 * pages for large extents.
853 if (bio_ctrl->len_to_oe_boundary != U32_MAX)
854 bio_ctrl->len_to_oe_boundary -= len;
856 /* Ordered extent boundary: move on to a new bio. */
857 if (bio_ctrl->len_to_oe_boundary == 0)
858 submit_one_bio(bio_ctrl);
862 static int attach_extent_buffer_page(struct extent_buffer *eb,
864 struct btrfs_subpage *prealloc)
866 struct btrfs_fs_info *fs_info = eb->fs_info;
870 * If the page is mapped to btree inode, we should hold the private
871 * lock to prevent race.
872 * For cloned or dummy extent buffers, their pages are not mapped and
873 * will not race with any other ebs.
876 lockdep_assert_held(&page->mapping->private_lock);
878 if (fs_info->nodesize >= PAGE_SIZE) {
879 if (!PagePrivate(page))
880 attach_page_private(page, eb);
882 WARN_ON(page->private != (unsigned long)eb);
886 /* Already mapped, just free prealloc */
887 if (PagePrivate(page)) {
888 btrfs_free_subpage(prealloc);
893 /* Has preallocated memory for subpage */
894 attach_page_private(page, prealloc);
896 /* Do new allocation to attach subpage */
897 ret = btrfs_attach_subpage(fs_info, page,
898 BTRFS_SUBPAGE_METADATA);
902 int set_page_extent_mapped(struct page *page)
904 struct btrfs_fs_info *fs_info;
906 ASSERT(page->mapping);
908 if (PagePrivate(page))
911 fs_info = btrfs_sb(page->mapping->host->i_sb);
913 if (btrfs_is_subpage(fs_info, page))
914 return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA);
916 attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE);
920 void clear_page_extent_mapped(struct page *page)
922 struct btrfs_fs_info *fs_info;
924 ASSERT(page->mapping);
926 if (!PagePrivate(page))
929 fs_info = btrfs_sb(page->mapping->host->i_sb);
930 if (btrfs_is_subpage(fs_info, page))
931 return btrfs_detach_subpage(fs_info, page);
933 detach_page_private(page);
936 static struct extent_map *
937 __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset,
938 u64 start, u64 len, struct extent_map **em_cached)
940 struct extent_map *em;
942 if (em_cached && *em_cached) {
944 if (extent_map_in_tree(em) && start >= em->start &&
945 start < extent_map_end(em)) {
946 refcount_inc(&em->refs);
954 em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len);
955 if (em_cached && !IS_ERR(em)) {
957 refcount_inc(&em->refs);
963 * basic readpage implementation. Locked extent state structs are inserted
964 * into the tree that are removed when the IO is done (by the end_io
966 * XXX JDM: This needs looking at to ensure proper page locking
967 * return 0 on success, otherwise return error
969 static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached,
970 struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start)
972 struct inode *inode = page->mapping->host;
973 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
974 u64 start = page_offset(page);
975 const u64 end = start + PAGE_SIZE - 1;
978 u64 last_byte = i_size_read(inode);
980 struct extent_map *em;
982 size_t pg_offset = 0;
984 size_t blocksize = inode->i_sb->s_blocksize;
985 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
987 ret = set_page_extent_mapped(page);
989 unlock_extent(tree, start, end, NULL);
994 if (page->index == last_byte >> PAGE_SHIFT) {
995 size_t zero_offset = offset_in_page(last_byte);
998 iosize = PAGE_SIZE - zero_offset;
999 memzero_page(page, zero_offset, iosize);
1002 bio_ctrl->end_io_func = end_bio_extent_readpage;
1003 begin_page_read(fs_info, page);
1004 while (cur <= end) {
1005 enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE;
1006 bool force_bio_submit = false;
1009 ASSERT(IS_ALIGNED(cur, fs_info->sectorsize));
1010 if (cur >= last_byte) {
1011 iosize = PAGE_SIZE - pg_offset;
1012 memzero_page(page, pg_offset, iosize);
1013 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1014 end_page_read(page, true, cur, iosize);
1017 em = __get_extent_map(inode, page, pg_offset, cur,
1018 end - cur + 1, em_cached);
1020 unlock_extent(tree, cur, end, NULL);
1021 end_page_read(page, false, cur, end + 1 - cur);
1024 extent_offset = cur - em->start;
1025 BUG_ON(extent_map_end(em) <= cur);
1028 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1029 compress_type = em->compress_type;
1031 iosize = min(extent_map_end(em) - cur, end - cur + 1);
1032 iosize = ALIGN(iosize, blocksize);
1033 if (compress_type != BTRFS_COMPRESS_NONE)
1034 disk_bytenr = em->block_start;
1036 disk_bytenr = em->block_start + extent_offset;
1037 block_start = em->block_start;
1038 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
1039 block_start = EXTENT_MAP_HOLE;
1042 * If we have a file range that points to a compressed extent
1043 * and it's followed by a consecutive file range that points
1044 * to the same compressed extent (possibly with a different
1045 * offset and/or length, so it either points to the whole extent
1046 * or only part of it), we must make sure we do not submit a
1047 * single bio to populate the pages for the 2 ranges because
1048 * this makes the compressed extent read zero out the pages
1049 * belonging to the 2nd range. Imagine the following scenario:
1052 * [0 - 8K] [8K - 24K]
1055 * points to extent X, points to extent X,
1056 * offset 4K, length of 8K offset 0, length 16K
1058 * [extent X, compressed length = 4K uncompressed length = 16K]
1060 * If the bio to read the compressed extent covers both ranges,
1061 * it will decompress extent X into the pages belonging to the
1062 * first range and then it will stop, zeroing out the remaining
1063 * pages that belong to the other range that points to extent X.
1064 * So here we make sure we submit 2 bios, one for the first
1065 * range and another one for the third range. Both will target
1066 * the same physical extent from disk, but we can't currently
1067 * make the compressed bio endio callback populate the pages
1068 * for both ranges because each compressed bio is tightly
1069 * coupled with a single extent map, and each range can have
1070 * an extent map with a different offset value relative to the
1071 * uncompressed data of our extent and different lengths. This
1072 * is a corner case so we prioritize correctness over
1073 * non-optimal behavior (submitting 2 bios for the same extent).
1075 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) &&
1076 prev_em_start && *prev_em_start != (u64)-1 &&
1077 *prev_em_start != em->start)
1078 force_bio_submit = true;
1081 *prev_em_start = em->start;
1083 free_extent_map(em);
1086 /* we've found a hole, just zero and go on */
1087 if (block_start == EXTENT_MAP_HOLE) {
1088 memzero_page(page, pg_offset, iosize);
1090 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1091 end_page_read(page, true, cur, iosize);
1093 pg_offset += iosize;
1096 /* the get_extent function already copied into the page */
1097 if (block_start == EXTENT_MAP_INLINE) {
1098 unlock_extent(tree, cur, cur + iosize - 1, NULL);
1099 end_page_read(page, true, cur, iosize);
1101 pg_offset += iosize;
1105 if (bio_ctrl->compress_type != compress_type) {
1106 submit_one_bio(bio_ctrl);
1107 bio_ctrl->compress_type = compress_type;
1110 if (force_bio_submit)
1111 submit_one_bio(bio_ctrl);
1112 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1115 pg_offset += iosize;
1121 int btrfs_read_folio(struct file *file, struct folio *folio)
1123 struct page *page = &folio->page;
1124 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
1125 u64 start = page_offset(page);
1126 u64 end = start + PAGE_SIZE - 1;
1127 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ };
1130 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1132 ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL);
1134 * If btrfs_do_readpage() failed we will want to submit the assembled
1135 * bio to do the cleanup.
1137 submit_one_bio(&bio_ctrl);
1141 static inline void contiguous_readpages(struct page *pages[], int nr_pages,
1143 struct extent_map **em_cached,
1144 struct btrfs_bio_ctrl *bio_ctrl,
1147 struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host);
1150 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
1152 for (index = 0; index < nr_pages; index++) {
1153 btrfs_do_readpage(pages[index], em_cached, bio_ctrl,
1155 put_page(pages[index]);
1160 * helper for __extent_writepage, doing all of the delayed allocation setup.
1162 * This returns 1 if btrfs_run_delalloc_range function did all the work required
1163 * to write the page (copy into inline extent). In this case the IO has
1164 * been started and the page is already unlocked.
1166 * This returns 0 if all went well (page still locked)
1167 * This returns < 0 if there were errors (page still locked)
1169 static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode,
1170 struct page *page, struct writeback_control *wbc)
1172 const u64 page_start = page_offset(page);
1173 const u64 page_end = page_start + PAGE_SIZE - 1;
1174 u64 delalloc_start = page_start;
1175 u64 delalloc_end = page_end;
1176 u64 delalloc_to_write = 0;
1179 while (delalloc_start < page_end) {
1180 delalloc_end = page_end;
1181 if (!find_lock_delalloc_range(&inode->vfs_inode, page,
1182 &delalloc_start, &delalloc_end)) {
1183 delalloc_start = delalloc_end + 1;
1187 ret = btrfs_run_delalloc_range(inode, page, delalloc_start,
1192 delalloc_start = delalloc_end + 1;
1196 * delalloc_end is already one less than the total length, so
1197 * we don't subtract one from PAGE_SIZE
1199 delalloc_to_write +=
1200 DIV_ROUND_UP(delalloc_end + 1 - page_start, PAGE_SIZE);
1203 * If btrfs_run_dealloc_range() already started I/O and unlocked
1204 * the pages, we just need to account for them here.
1207 wbc->nr_to_write -= delalloc_to_write;
1211 if (wbc->nr_to_write < delalloc_to_write) {
1214 if (delalloc_to_write < thresh * 2)
1215 thresh = delalloc_to_write;
1216 wbc->nr_to_write = min_t(u64, delalloc_to_write,
1224 * Find the first byte we need to write.
1226 * For subpage, one page can contain several sectors, and
1227 * __extent_writepage_io() will just grab all extent maps in the page
1228 * range and try to submit all non-inline/non-compressed extents.
1230 * This is a big problem for subpage, we shouldn't re-submit already written
1232 * This function will lookup subpage dirty bit to find which range we really
1235 * Return the next dirty range in [@start, @end).
1236 * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE.
1238 static void find_next_dirty_byte(struct btrfs_fs_info *fs_info,
1239 struct page *page, u64 *start, u64 *end)
1241 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1242 struct btrfs_subpage_info *spi = fs_info->subpage_info;
1243 u64 orig_start = *start;
1244 /* Declare as unsigned long so we can use bitmap ops */
1245 unsigned long flags;
1246 int range_start_bit;
1250 * For regular sector size == page size case, since one page only
1251 * contains one sector, we return the page offset directly.
1253 if (!btrfs_is_subpage(fs_info, page)) {
1254 *start = page_offset(page);
1255 *end = page_offset(page) + PAGE_SIZE;
1259 range_start_bit = spi->dirty_offset +
1260 (offset_in_page(orig_start) >> fs_info->sectorsize_bits);
1262 /* We should have the page locked, but just in case */
1263 spin_lock_irqsave(&subpage->lock, flags);
1264 bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit,
1265 spi->dirty_offset + spi->bitmap_nr_bits);
1266 spin_unlock_irqrestore(&subpage->lock, flags);
1268 range_start_bit -= spi->dirty_offset;
1269 range_end_bit -= spi->dirty_offset;
1271 *start = page_offset(page) + range_start_bit * fs_info->sectorsize;
1272 *end = page_offset(page) + range_end_bit * fs_info->sectorsize;
1276 * helper for __extent_writepage. This calls the writepage start hooks,
1277 * and does the loop to map the page into extents and bios.
1279 * We return 1 if the IO is started and the page is unlocked,
1280 * 0 if all went well (page still locked)
1281 * < 0 if there were errors (page still locked)
1283 static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode,
1285 struct btrfs_bio_ctrl *bio_ctrl,
1289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1290 u64 cur = page_offset(page);
1291 u64 end = cur + PAGE_SIZE - 1;
1294 struct extent_map *em;
1298 ret = btrfs_writepage_cow_fixup(page);
1300 /* Fixup worker will requeue */
1301 redirty_page_for_writepage(bio_ctrl->wbc, page);
1306 bio_ctrl->end_io_func = end_bio_extent_writepage;
1307 while (cur <= end) {
1308 u32 len = end - cur + 1;
1311 u64 dirty_range_start = cur;
1312 u64 dirty_range_end;
1315 if (cur >= i_size) {
1316 btrfs_mark_ordered_io_finished(inode, page, cur, len,
1319 * This range is beyond i_size, thus we don't need to
1320 * bother writing back.
1321 * But we still need to clear the dirty subpage bit, or
1322 * the next time the page gets dirtied, we will try to
1323 * writeback the sectors with subpage dirty bits,
1324 * causing writeback without ordered extent.
1326 btrfs_page_clear_dirty(fs_info, page, cur, len);
1330 find_next_dirty_byte(fs_info, page, &dirty_range_start,
1332 if (cur < dirty_range_start) {
1333 cur = dirty_range_start;
1337 em = btrfs_get_extent(inode, NULL, 0, cur, len);
1339 ret = PTR_ERR_OR_ZERO(em);
1343 extent_offset = cur - em->start;
1344 em_end = extent_map_end(em);
1345 ASSERT(cur <= em_end);
1347 ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize));
1348 ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize));
1350 block_start = em->block_start;
1351 disk_bytenr = em->block_start + extent_offset;
1353 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
1354 ASSERT(block_start != EXTENT_MAP_HOLE);
1355 ASSERT(block_start != EXTENT_MAP_INLINE);
1358 * Note that em_end from extent_map_end() and dirty_range_end from
1359 * find_next_dirty_byte() are all exclusive
1361 iosize = min(min(em_end, end + 1), dirty_range_end) - cur;
1362 free_extent_map(em);
1365 btrfs_set_range_writeback(inode, cur, cur + iosize - 1);
1366 if (!PageWriteback(page)) {
1367 btrfs_err(inode->root->fs_info,
1368 "page %lu not writeback, cur %llu end %llu",
1369 page->index, cur, end);
1373 * Although the PageDirty bit is cleared before entering this
1374 * function, subpage dirty bit is not cleared.
1375 * So clear subpage dirty bit here so next time we won't submit
1376 * page for range already written to disk.
1378 btrfs_page_clear_dirty(fs_info, page, cur, iosize);
1380 submit_extent_page(bio_ctrl, disk_bytenr, page, iosize,
1381 cur - page_offset(page));
1386 btrfs_page_assert_not_dirty(fs_info, page);
1392 * If we finish without problem, we should not only clear page dirty,
1393 * but also empty subpage dirty bits
1400 * the writepage semantics are similar to regular writepage. extent
1401 * records are inserted to lock ranges in the tree, and as dirty areas
1402 * are found, they are marked writeback. Then the lock bits are removed
1403 * and the end_io handler clears the writeback ranges
1405 * Return 0 if everything goes well.
1406 * Return <0 for error.
1408 static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl)
1410 struct folio *folio = page_folio(page);
1411 struct inode *inode = page->mapping->host;
1412 const u64 page_start = page_offset(page);
1416 loff_t i_size = i_size_read(inode);
1417 unsigned long end_index = i_size >> PAGE_SHIFT;
1419 trace___extent_writepage(page, inode, bio_ctrl->wbc);
1421 WARN_ON(!PageLocked(page));
1423 pg_offset = offset_in_page(i_size);
1424 if (page->index > end_index ||
1425 (page->index == end_index && !pg_offset)) {
1426 folio_invalidate(folio, 0, folio_size(folio));
1427 folio_unlock(folio);
1431 if (page->index == end_index)
1432 memzero_page(page, pg_offset, PAGE_SIZE - pg_offset);
1434 ret = set_page_extent_mapped(page);
1438 ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc);
1444 ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr);
1448 bio_ctrl->wbc->nr_to_write--;
1452 /* make sure the mapping tag for page dirty gets cleared */
1453 set_page_writeback(page);
1454 end_page_writeback(page);
1457 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page, page_start,
1459 btrfs_page_clear_uptodate(btrfs_sb(inode->i_sb), page,
1460 page_start, PAGE_SIZE);
1461 mapping_set_error(page->mapping, ret);
1468 void wait_on_extent_buffer_writeback(struct extent_buffer *eb)
1470 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK,
1471 TASK_UNINTERRUPTIBLE);
1475 * Lock extent buffer status and pages for writeback.
1477 * Return %false if the extent buffer doesn't need to be submitted (e.g. the
1478 * extent buffer is not dirty)
1479 * Return %true is the extent buffer is submitted to bio.
1481 static noinline_for_stack bool lock_extent_buffer_for_io(struct extent_buffer *eb,
1482 struct writeback_control *wbc)
1484 struct btrfs_fs_info *fs_info = eb->fs_info;
1487 btrfs_tree_lock(eb);
1488 while (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) {
1489 btrfs_tree_unlock(eb);
1490 if (wbc->sync_mode != WB_SYNC_ALL)
1492 wait_on_extent_buffer_writeback(eb);
1493 btrfs_tree_lock(eb);
1497 * We need to do this to prevent races in people who check if the eb is
1498 * under IO since we can end up having no IO bits set for a short period
1501 spin_lock(&eb->refs_lock);
1502 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
1503 set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1504 spin_unlock(&eb->refs_lock);
1505 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
1506 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1508 fs_info->dirty_metadata_batch);
1511 spin_unlock(&eb->refs_lock);
1513 btrfs_tree_unlock(eb);
1517 static void set_btree_ioerr(struct extent_buffer *eb)
1519 struct btrfs_fs_info *fs_info = eb->fs_info;
1521 set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1524 * A read may stumble upon this buffer later, make sure that it gets an
1525 * error and knows there was an error.
1527 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
1530 * We need to set the mapping with the io error as well because a write
1531 * error will flip the file system readonly, and then syncfs() will
1532 * return a 0 because we are readonly if we don't modify the err seq for
1535 mapping_set_error(eb->fs_info->btree_inode->i_mapping, -EIO);
1538 * If writeback for a btree extent that doesn't belong to a log tree
1539 * failed, increment the counter transaction->eb_write_errors.
1540 * We do this because while the transaction is running and before it's
1541 * committing (when we call filemap_fdata[write|wait]_range against
1542 * the btree inode), we might have
1543 * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it
1544 * returns an error or an error happens during writeback, when we're
1545 * committing the transaction we wouldn't know about it, since the pages
1546 * can be no longer dirty nor marked anymore for writeback (if a
1547 * subsequent modification to the extent buffer didn't happen before the
1548 * transaction commit), which makes filemap_fdata[write|wait]_range not
1549 * able to find the pages tagged with SetPageError at transaction
1550 * commit time. So if this happens we must abort the transaction,
1551 * otherwise we commit a super block with btree roots that point to
1552 * btree nodes/leafs whose content on disk is invalid - either garbage
1553 * or the content of some node/leaf from a past generation that got
1554 * cowed or deleted and is no longer valid.
1556 * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would
1557 * not be enough - we need to distinguish between log tree extents vs
1558 * non-log tree extents, and the next filemap_fdatawait_range() call
1559 * will catch and clear such errors in the mapping - and that call might
1560 * be from a log sync and not from a transaction commit. Also, checking
1561 * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is
1562 * not done and would not be reliable - the eb might have been released
1563 * from memory and reading it back again means that flag would not be
1564 * set (since it's a runtime flag, not persisted on disk).
1566 * Using the flags below in the btree inode also makes us achieve the
1567 * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started
1568 * writeback for all dirty pages and before filemap_fdatawait_range()
1569 * is called, the writeback for all dirty pages had already finished
1570 * with errors - because we were not using AS_EIO/AS_ENOSPC,
1571 * filemap_fdatawait_range() would return success, as it could not know
1572 * that writeback errors happened (the pages were no longer tagged for
1575 switch (eb->log_index) {
1577 set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags);
1580 set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags);
1583 set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags);
1586 BUG(); /* unexpected, logic error */
1591 * The endio specific version which won't touch any unsafe spinlock in endio
1594 static struct extent_buffer *find_extent_buffer_nolock(
1595 struct btrfs_fs_info *fs_info, u64 start)
1597 struct extent_buffer *eb;
1600 eb = radix_tree_lookup(&fs_info->buffer_radix,
1601 start >> fs_info->sectorsize_bits);
1602 if (eb && atomic_inc_not_zero(&eb->refs)) {
1610 static void extent_buffer_write_end_io(struct btrfs_bio *bbio)
1612 struct extent_buffer *eb = bbio->private;
1613 struct btrfs_fs_info *fs_info = eb->fs_info;
1614 bool uptodate = !bbio->bio.bi_status;
1615 struct bvec_iter_all iter_all;
1616 struct bio_vec *bvec;
1620 set_btree_ioerr(eb);
1622 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
1623 u64 start = eb->start + bio_offset;
1624 struct page *page = bvec->bv_page;
1625 u32 len = bvec->bv_len;
1628 btrfs_page_clear_uptodate(fs_info, page, start, len);
1629 btrfs_page_clear_writeback(fs_info, page, start, len);
1633 clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags);
1634 smp_mb__after_atomic();
1635 wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK);
1637 bio_put(&bbio->bio);
1640 static void prepare_eb_write(struct extent_buffer *eb)
1643 unsigned long start;
1646 clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags);
1648 /* Set btree blocks beyond nritems with 0 to avoid stale content */
1649 nritems = btrfs_header_nritems(eb);
1650 if (btrfs_header_level(eb) > 0) {
1651 end = btrfs_node_key_ptr_offset(eb, nritems);
1652 memzero_extent_buffer(eb, end, eb->len - end);
1656 * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0
1658 start = btrfs_item_nr_offset(eb, nritems);
1659 end = btrfs_item_nr_offset(eb, 0);
1661 end += BTRFS_LEAF_DATA_SIZE(eb->fs_info);
1663 end += btrfs_item_offset(eb, nritems - 1);
1664 memzero_extent_buffer(eb, start, end - start);
1668 static noinline_for_stack void write_one_eb(struct extent_buffer *eb,
1669 struct writeback_control *wbc)
1671 struct btrfs_fs_info *fs_info = eb->fs_info;
1672 struct btrfs_bio *bbio;
1674 prepare_eb_write(eb);
1676 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
1677 REQ_OP_WRITE | REQ_META | wbc_to_write_flags(wbc),
1678 eb->fs_info, extent_buffer_write_end_io, eb);
1679 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
1680 bio_set_dev(&bbio->bio, fs_info->fs_devices->latest_dev->bdev);
1681 wbc_init_bio(wbc, &bbio->bio);
1682 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
1683 bbio->file_offset = eb->start;
1684 if (fs_info->nodesize < PAGE_SIZE) {
1685 struct page *p = eb->pages[0];
1688 btrfs_subpage_set_writeback(fs_info, p, eb->start, eb->len);
1689 if (btrfs_subpage_clear_and_test_dirty(fs_info, p, eb->start,
1691 clear_page_dirty_for_io(p);
1694 __bio_add_page(&bbio->bio, p, eb->len, eb->start - page_offset(p));
1695 wbc_account_cgroup_owner(wbc, p, eb->len);
1698 for (int i = 0; i < num_extent_pages(eb); i++) {
1699 struct page *p = eb->pages[i];
1702 clear_page_dirty_for_io(p);
1703 set_page_writeback(p);
1704 __bio_add_page(&bbio->bio, p, PAGE_SIZE, 0);
1705 wbc_account_cgroup_owner(wbc, p, PAGE_SIZE);
1710 btrfs_submit_bio(bbio, 0);
1714 * Submit one subpage btree page.
1716 * The main difference to submit_eb_page() is:
1718 * For subpage, we don't rely on page locking at all.
1721 * We only flush bio if we may be unable to fit current extent buffers into
1724 * Return >=0 for the number of submitted extent buffers.
1725 * Return <0 for fatal error.
1727 static int submit_eb_subpage(struct page *page, struct writeback_control *wbc)
1729 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
1731 u64 page_start = page_offset(page);
1733 int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits;
1735 /* Lock and write each dirty extent buffers in the range */
1736 while (bit_start < fs_info->subpage_info->bitmap_nr_bits) {
1737 struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private;
1738 struct extent_buffer *eb;
1739 unsigned long flags;
1743 * Take private lock to ensure the subpage won't be detached
1746 spin_lock(&page->mapping->private_lock);
1747 if (!PagePrivate(page)) {
1748 spin_unlock(&page->mapping->private_lock);
1751 spin_lock_irqsave(&subpage->lock, flags);
1752 if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset,
1753 subpage->bitmaps)) {
1754 spin_unlock_irqrestore(&subpage->lock, flags);
1755 spin_unlock(&page->mapping->private_lock);
1760 start = page_start + bit_start * fs_info->sectorsize;
1761 bit_start += sectors_per_node;
1764 * Here we just want to grab the eb without touching extra
1765 * spin locks, so call find_extent_buffer_nolock().
1767 eb = find_extent_buffer_nolock(fs_info, start);
1768 spin_unlock_irqrestore(&subpage->lock, flags);
1769 spin_unlock(&page->mapping->private_lock);
1772 * The eb has already reached 0 refs thus find_extent_buffer()
1773 * doesn't return it. We don't need to write back such eb
1779 if (lock_extent_buffer_for_io(eb, wbc)) {
1780 write_one_eb(eb, wbc);
1783 free_extent_buffer(eb);
1789 * Submit all page(s) of one extent buffer.
1791 * @page: the page of one extent buffer
1792 * @eb_context: to determine if we need to submit this page, if current page
1793 * belongs to this eb, we don't need to submit
1795 * The caller should pass each page in their bytenr order, and here we use
1796 * @eb_context to determine if we have submitted pages of one extent buffer.
1798 * If we have, we just skip until we hit a new page that doesn't belong to
1799 * current @eb_context.
1801 * If not, we submit all the page(s) of the extent buffer.
1803 * Return >0 if we have submitted the extent buffer successfully.
1804 * Return 0 if we don't need to submit the page, as it's already submitted by
1806 * Return <0 for fatal error.
1808 static int submit_eb_page(struct page *page, struct writeback_control *wbc,
1809 struct extent_buffer **eb_context)
1811 struct address_space *mapping = page->mapping;
1812 struct btrfs_block_group *cache = NULL;
1813 struct extent_buffer *eb;
1816 if (!PagePrivate(page))
1819 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
1820 return submit_eb_subpage(page, wbc);
1822 spin_lock(&mapping->private_lock);
1823 if (!PagePrivate(page)) {
1824 spin_unlock(&mapping->private_lock);
1828 eb = (struct extent_buffer *)page->private;
1831 * Shouldn't happen and normally this would be a BUG_ON but no point
1832 * crashing the machine for something we can survive anyway.
1835 spin_unlock(&mapping->private_lock);
1839 if (eb == *eb_context) {
1840 spin_unlock(&mapping->private_lock);
1843 ret = atomic_inc_not_zero(&eb->refs);
1844 spin_unlock(&mapping->private_lock);
1848 if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) {
1850 * If for_sync, this hole will be filled with
1851 * trasnsaction commit.
1853 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
1857 free_extent_buffer(eb);
1863 if (!lock_extent_buffer_for_io(eb, wbc)) {
1864 btrfs_revert_meta_write_pointer(cache, eb);
1866 btrfs_put_block_group(cache);
1867 free_extent_buffer(eb);
1872 * Implies write in zoned mode. Mark the last eb in a block group.
1874 btrfs_schedule_zone_finish_bg(cache, eb);
1875 btrfs_put_block_group(cache);
1877 write_one_eb(eb, wbc);
1878 free_extent_buffer(eb);
1882 int btree_write_cache_pages(struct address_space *mapping,
1883 struct writeback_control *wbc)
1885 struct extent_buffer *eb_context = NULL;
1886 struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info;
1889 int nr_to_write_done = 0;
1890 struct folio_batch fbatch;
1891 unsigned int nr_folios;
1893 pgoff_t end; /* Inclusive */
1897 folio_batch_init(&fbatch);
1898 if (wbc->range_cyclic) {
1899 index = mapping->writeback_index; /* Start from prev offset */
1902 * Start from the beginning does not need to cycle over the
1903 * range, mark it as scanned.
1905 scanned = (index == 0);
1907 index = wbc->range_start >> PAGE_SHIFT;
1908 end = wbc->range_end >> PAGE_SHIFT;
1911 if (wbc->sync_mode == WB_SYNC_ALL)
1912 tag = PAGECACHE_TAG_TOWRITE;
1914 tag = PAGECACHE_TAG_DIRTY;
1915 btrfs_zoned_meta_io_lock(fs_info);
1917 if (wbc->sync_mode == WB_SYNC_ALL)
1918 tag_pages_for_writeback(mapping, index, end);
1919 while (!done && !nr_to_write_done && (index <= end) &&
1920 (nr_folios = filemap_get_folios_tag(mapping, &index, end,
1924 for (i = 0; i < nr_folios; i++) {
1925 struct folio *folio = fbatch.folios[i];
1927 ret = submit_eb_page(&folio->page, wbc, &eb_context);
1936 * the filesystem may choose to bump up nr_to_write.
1937 * We have to make sure to honor the new nr_to_write
1940 nr_to_write_done = wbc->nr_to_write <= 0;
1942 folio_batch_release(&fbatch);
1945 if (!scanned && !done) {
1947 * We hit the last page and there is more work to be done: wrap
1948 * back to the start of the file
1955 * If something went wrong, don't allow any metadata write bio to be
1958 * This would prevent use-after-free if we had dirty pages not
1959 * cleaned up, which can still happen by fuzzed images.
1962 * Allowing existing tree block to be allocated for other trees.
1964 * - Log tree operations
1965 * Exiting tree blocks get allocated to log tree, bumps its
1966 * generation, then get cleaned in tree re-balance.
1967 * Such tree block will not be written back, since it's clean,
1968 * thus no WRITTEN flag set.
1969 * And after log writes back, this tree block is not traced by
1970 * any dirty extent_io_tree.
1972 * - Offending tree block gets re-dirtied from its original owner
1973 * Since it has bumped generation, no WRITTEN flag, it can be
1974 * reused without COWing. This tree block will not be traced
1975 * by btrfs_transaction::dirty_pages.
1977 * Now such dirty tree block will not be cleaned by any dirty
1978 * extent io tree. Thus we don't want to submit such wild eb
1979 * if the fs already has error.
1981 * We can get ret > 0 from submit_extent_page() indicating how many ebs
1982 * were submitted. Reset it to 0 to avoid false alerts for the caller.
1986 if (!ret && BTRFS_FS_ERROR(fs_info))
1988 btrfs_zoned_meta_io_unlock(fs_info);
1993 * Walk the list of dirty pages of the given address space and write all of them.
1995 * @mapping: address space structure to write
1996 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1997 * @bio_ctrl: holds context for the write, namely the bio
1999 * If a page is already under I/O, write_cache_pages() skips it, even
2000 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
2001 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
2002 * and msync() need to guarantee that all the data which was dirty at the time
2003 * the call was made get new I/O started against them. If wbc->sync_mode is
2004 * WB_SYNC_ALL then we were called for data integrity and we must wait for
2005 * existing IO to complete.
2007 static int extent_write_cache_pages(struct address_space *mapping,
2008 struct btrfs_bio_ctrl *bio_ctrl)
2010 struct writeback_control *wbc = bio_ctrl->wbc;
2011 struct inode *inode = mapping->host;
2014 int nr_to_write_done = 0;
2015 struct folio_batch fbatch;
2016 unsigned int nr_folios;
2018 pgoff_t end; /* Inclusive */
2020 int range_whole = 0;
2025 * We have to hold onto the inode so that ordered extents can do their
2026 * work when the IO finishes. The alternative to this is failing to add
2027 * an ordered extent if the igrab() fails there and that is a huge pain
2028 * to deal with, so instead just hold onto the inode throughout the
2029 * writepages operation. If it fails here we are freeing up the inode
2030 * anyway and we'd rather not waste our time writing out stuff that is
2031 * going to be truncated anyway.
2036 folio_batch_init(&fbatch);
2037 if (wbc->range_cyclic) {
2038 index = mapping->writeback_index; /* Start from prev offset */
2041 * Start from the beginning does not need to cycle over the
2042 * range, mark it as scanned.
2044 scanned = (index == 0);
2046 index = wbc->range_start >> PAGE_SHIFT;
2047 end = wbc->range_end >> PAGE_SHIFT;
2048 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2054 * We do the tagged writepage as long as the snapshot flush bit is set
2055 * and we are the first one who do the filemap_flush() on this inode.
2057 * The nr_to_write == LONG_MAX is needed to make sure other flushers do
2058 * not race in and drop the bit.
2060 if (range_whole && wbc->nr_to_write == LONG_MAX &&
2061 test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
2062 &BTRFS_I(inode)->runtime_flags))
2063 wbc->tagged_writepages = 1;
2065 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2066 tag = PAGECACHE_TAG_TOWRITE;
2068 tag = PAGECACHE_TAG_DIRTY;
2070 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2071 tag_pages_for_writeback(mapping, index, end);
2073 while (!done && !nr_to_write_done && (index <= end) &&
2074 (nr_folios = filemap_get_folios_tag(mapping, &index,
2075 end, tag, &fbatch))) {
2078 for (i = 0; i < nr_folios; i++) {
2079 struct folio *folio = fbatch.folios[i];
2081 done_index = folio_next_index(folio);
2083 * At this point we hold neither the i_pages lock nor
2084 * the page lock: the page may be truncated or
2085 * invalidated (changing page->mapping to NULL),
2086 * or even swizzled back from swapper_space to
2087 * tmpfs file mapping
2089 if (!folio_trylock(folio)) {
2090 submit_write_bio(bio_ctrl, 0);
2094 if (unlikely(folio->mapping != mapping)) {
2095 folio_unlock(folio);
2099 if (!folio_test_dirty(folio)) {
2100 /* Someone wrote it for us. */
2101 folio_unlock(folio);
2105 if (wbc->sync_mode != WB_SYNC_NONE) {
2106 if (folio_test_writeback(folio))
2107 submit_write_bio(bio_ctrl, 0);
2108 folio_wait_writeback(folio);
2111 if (folio_test_writeback(folio) ||
2112 !folio_clear_dirty_for_io(folio)) {
2113 folio_unlock(folio);
2117 ret = __extent_writepage(&folio->page, bio_ctrl);
2124 * The filesystem may choose to bump up nr_to_write.
2125 * We have to make sure to honor the new nr_to_write
2128 nr_to_write_done = (wbc->sync_mode == WB_SYNC_NONE &&
2129 wbc->nr_to_write <= 0);
2131 folio_batch_release(&fbatch);
2134 if (!scanned && !done) {
2136 * We hit the last page and there is more work to be done: wrap
2137 * back to the start of the file
2143 * If we're looping we could run into a page that is locked by a
2144 * writer and that writer could be waiting on writeback for a
2145 * page in our current bio, and thus deadlock, so flush the
2148 submit_write_bio(bio_ctrl, 0);
2152 if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole))
2153 mapping->writeback_index = done_index;
2155 btrfs_add_delayed_iput(BTRFS_I(inode));
2160 * Submit the pages in the range to bio for call sites which delalloc range has
2161 * already been ran (aka, ordered extent inserted) and all pages are still
2164 void extent_write_locked_range(struct inode *inode, u64 start, u64 end,
2165 struct writeback_control *wbc, bool pages_dirty)
2167 bool found_error = false;
2169 struct address_space *mapping = inode->i_mapping;
2170 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2171 const u32 sectorsize = fs_info->sectorsize;
2172 loff_t i_size = i_size_read(inode);
2174 struct btrfs_bio_ctrl bio_ctrl = {
2176 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2179 if (wbc->no_cgroup_owner)
2180 bio_ctrl.opf |= REQ_BTRFS_CGROUP_PUNT;
2182 ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
2184 while (cur <= end) {
2185 u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
2186 u32 cur_len = cur_end + 1 - cur;
2190 page = find_get_page(mapping, cur >> PAGE_SHIFT);
2191 ASSERT(PageLocked(page));
2193 ASSERT(PageDirty(page));
2194 clear_page_dirty_for_io(page);
2197 ret = __extent_writepage_io(BTRFS_I(inode), page, &bio_ctrl,
2202 /* Make sure the mapping tag for page dirty gets cleared. */
2204 set_page_writeback(page);
2205 end_page_writeback(page);
2208 btrfs_mark_ordered_io_finished(BTRFS_I(inode), page,
2209 cur, cur_len, !ret);
2210 btrfs_page_clear_uptodate(fs_info, page, cur, cur_len);
2211 mapping_set_error(page->mapping, ret);
2213 btrfs_page_unlock_writer(fs_info, page, cur, cur_len);
2221 submit_write_bio(&bio_ctrl, found_error ? ret : 0);
2224 int extent_writepages(struct address_space *mapping,
2225 struct writeback_control *wbc)
2227 struct inode *inode = mapping->host;
2229 struct btrfs_bio_ctrl bio_ctrl = {
2231 .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc),
2235 * Allow only a single thread to do the reloc work in zoned mode to
2236 * protect the write pointer updates.
2238 btrfs_zoned_data_reloc_lock(BTRFS_I(inode));
2239 ret = extent_write_cache_pages(mapping, &bio_ctrl);
2240 submit_write_bio(&bio_ctrl, ret);
2241 btrfs_zoned_data_reloc_unlock(BTRFS_I(inode));
2245 void extent_readahead(struct readahead_control *rac)
2247 struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD };
2248 struct page *pagepool[16];
2249 struct extent_map *em_cached = NULL;
2250 u64 prev_em_start = (u64)-1;
2253 while ((nr = readahead_page_batch(rac, pagepool))) {
2254 u64 contig_start = readahead_pos(rac);
2255 u64 contig_end = contig_start + readahead_batch_length(rac) - 1;
2257 contiguous_readpages(pagepool, nr, contig_start, contig_end,
2258 &em_cached, &bio_ctrl, &prev_em_start);
2262 free_extent_map(em_cached);
2263 submit_one_bio(&bio_ctrl);
2267 * basic invalidate_folio code, this waits on any locked or writeback
2268 * ranges corresponding to the folio, and then deletes any extent state
2269 * records from the tree
2271 int extent_invalidate_folio(struct extent_io_tree *tree,
2272 struct folio *folio, size_t offset)
2274 struct extent_state *cached_state = NULL;
2275 u64 start = folio_pos(folio);
2276 u64 end = start + folio_size(folio) - 1;
2277 size_t blocksize = folio->mapping->host->i_sb->s_blocksize;
2279 /* This function is only called for the btree inode */
2280 ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO);
2282 start += ALIGN(offset, blocksize);
2286 lock_extent(tree, start, end, &cached_state);
2287 folio_wait_writeback(folio);
2290 * Currently for btree io tree, only EXTENT_LOCKED is utilized,
2291 * so here we only need to unlock the extent range to free any
2292 * existing extent state.
2294 unlock_extent(tree, start, end, &cached_state);
2299 * a helper for release_folio, this tests for areas of the page that
2300 * are locked or under IO and drops the related state bits if it is safe
2303 static int try_release_extent_state(struct extent_io_tree *tree,
2304 struct page *page, gfp_t mask)
2306 u64 start = page_offset(page);
2307 u64 end = start + PAGE_SIZE - 1;
2310 if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) {
2313 u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM |
2314 EXTENT_DELALLOC_NEW | EXTENT_CTLBITS);
2317 * At this point we can safely clear everything except the
2318 * locked bit, the nodatasum bit and the delalloc new bit.
2319 * The delalloc new bit will be cleared by ordered extent
2322 ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, NULL);
2324 /* if clear_extent_bit failed for enomem reasons,
2325 * we can't allow the release to continue.
2336 * a helper for release_folio. As long as there are no locked extents
2337 * in the range corresponding to the page, both state records and extent
2338 * map records are removed
2340 int try_release_extent_mapping(struct page *page, gfp_t mask)
2342 struct extent_map *em;
2343 u64 start = page_offset(page);
2344 u64 end = start + PAGE_SIZE - 1;
2345 struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host);
2346 struct extent_io_tree *tree = &btrfs_inode->io_tree;
2347 struct extent_map_tree *map = &btrfs_inode->extent_tree;
2349 if (gfpflags_allow_blocking(mask) &&
2350 page->mapping->host->i_size > SZ_16M) {
2352 while (start <= end) {
2353 struct btrfs_fs_info *fs_info;
2356 len = end - start + 1;
2357 write_lock(&map->lock);
2358 em = lookup_extent_mapping(map, start, len);
2360 write_unlock(&map->lock);
2363 if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
2364 em->start != start) {
2365 write_unlock(&map->lock);
2366 free_extent_map(em);
2369 if (test_range_bit(tree, em->start,
2370 extent_map_end(em) - 1,
2371 EXTENT_LOCKED, 0, NULL))
2374 * If it's not in the list of modified extents, used
2375 * by a fast fsync, we can remove it. If it's being
2376 * logged we can safely remove it since fsync took an
2377 * extra reference on the em.
2379 if (list_empty(&em->list) ||
2380 test_bit(EXTENT_FLAG_LOGGING, &em->flags))
2383 * If it's in the list of modified extents, remove it
2384 * only if its generation is older then the current one,
2385 * in which case we don't need it for a fast fsync.
2386 * Otherwise don't remove it, we could be racing with an
2387 * ongoing fast fsync that could miss the new extent.
2389 fs_info = btrfs_inode->root->fs_info;
2390 spin_lock(&fs_info->trans_lock);
2391 cur_gen = fs_info->generation;
2392 spin_unlock(&fs_info->trans_lock);
2393 if (em->generation >= cur_gen)
2397 * We only remove extent maps that are not in the list of
2398 * modified extents or that are in the list but with a
2399 * generation lower then the current generation, so there
2400 * is no need to set the full fsync flag on the inode (it
2401 * hurts the fsync performance for workloads with a data
2402 * size that exceeds or is close to the system's memory).
2404 remove_extent_mapping(map, em);
2405 /* once for the rb tree */
2406 free_extent_map(em);
2408 start = extent_map_end(em);
2409 write_unlock(&map->lock);
2412 free_extent_map(em);
2414 cond_resched(); /* Allow large-extent preemption. */
2417 return try_release_extent_state(tree, page, mask);
2421 * To cache previous fiemap extent
2423 * Will be used for merging fiemap extent
2425 struct fiemap_cache {
2434 * Helper to submit fiemap extent.
2436 * Will try to merge current fiemap extent specified by @offset, @phys,
2437 * @len and @flags with cached one.
2438 * And only when we fails to merge, cached one will be submitted as
2441 * Return value is the same as fiemap_fill_next_extent().
2443 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
2444 struct fiemap_cache *cache,
2445 u64 offset, u64 phys, u64 len, u32 flags)
2449 /* Set at the end of extent_fiemap(). */
2450 ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
2456 * Sanity check, extent_fiemap() should have ensured that new
2457 * fiemap extent won't overlap with cached one.
2460 * NOTE: Physical address can overlap, due to compression
2462 if (cache->offset + cache->len > offset) {
2468 * Only merges fiemap extents if
2469 * 1) Their logical addresses are continuous
2471 * 2) Their physical addresses are continuous
2472 * So truly compressed (physical size smaller than logical size)
2473 * extents won't get merged with each other
2475 * 3) Share same flags
2477 if (cache->offset + cache->len == offset &&
2478 cache->phys + cache->len == phys &&
2479 cache->flags == flags) {
2484 /* Not mergeable, need to submit cached one */
2485 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2486 cache->len, cache->flags);
2487 cache->cached = false;
2491 cache->cached = true;
2492 cache->offset = offset;
2495 cache->flags = flags;
2501 * Emit last fiemap cache
2503 * The last fiemap cache may still be cached in the following case:
2505 * |<- Fiemap range ->|
2506 * |<------------ First extent ----------->|
2508 * In this case, the first extent range will be cached but not emitted.
2509 * So we must emit it before ending extent_fiemap().
2511 static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
2512 struct fiemap_cache *cache)
2519 ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
2520 cache->len, cache->flags);
2521 cache->cached = false;
2527 static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path)
2529 struct extent_buffer *clone;
2530 struct btrfs_key key;
2535 if (path->slots[0] < btrfs_header_nritems(path->nodes[0]))
2538 ret = btrfs_next_leaf(inode->root, path);
2543 * Don't bother with cloning if there are no more file extent items for
2546 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2547 if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY)
2550 /* See the comment at fiemap_search_slot() about why we clone. */
2551 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2555 slot = path->slots[0];
2556 btrfs_release_path(path);
2557 path->nodes[0] = clone;
2558 path->slots[0] = slot;
2564 * Search for the first file extent item that starts at a given file offset or
2565 * the one that starts immediately before that offset.
2566 * Returns: 0 on success, < 0 on error, 1 if not found.
2568 static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path,
2571 const u64 ino = btrfs_ino(inode);
2572 struct btrfs_root *root = inode->root;
2573 struct extent_buffer *clone;
2574 struct btrfs_key key;
2579 key.type = BTRFS_EXTENT_DATA_KEY;
2580 key.offset = file_offset;
2582 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2586 if (ret > 0 && path->slots[0] > 0) {
2587 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
2588 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
2592 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2593 ret = btrfs_next_leaf(root, path);
2597 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2598 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2603 * We clone the leaf and use it during fiemap. This is because while
2604 * using the leaf we do expensive things like checking if an extent is
2605 * shared, which can take a long time. In order to prevent blocking
2606 * other tasks for too long, we use a clone of the leaf. We have locked
2607 * the file range in the inode's io tree, so we know none of our file
2608 * extent items can change. This way we avoid blocking other tasks that
2609 * want to insert items for other inodes in the same leaf or b+tree
2610 * rebalance operations (triggered for example when someone is trying
2611 * to push items into this leaf when trying to insert an item in a
2613 * We also need the private clone because holding a read lock on an
2614 * extent buffer of the subvolume's b+tree will make lockdep unhappy
2615 * when we call fiemap_fill_next_extent(), because that may cause a page
2616 * fault when filling the user space buffer with fiemap data.
2618 clone = btrfs_clone_extent_buffer(path->nodes[0]);
2622 slot = path->slots[0];
2623 btrfs_release_path(path);
2624 path->nodes[0] = clone;
2625 path->slots[0] = slot;
2631 * Process a range which is a hole or a prealloc extent in the inode's subvolume
2632 * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc
2633 * extent. The end offset (@end) is inclusive.
2635 static int fiemap_process_hole(struct btrfs_inode *inode,
2636 struct fiemap_extent_info *fieinfo,
2637 struct fiemap_cache *cache,
2638 struct extent_state **delalloc_cached_state,
2639 struct btrfs_backref_share_check_ctx *backref_ctx,
2640 u64 disk_bytenr, u64 extent_offset,
2644 const u64 i_size = i_size_read(&inode->vfs_inode);
2645 u64 cur_offset = start;
2646 u64 last_delalloc_end = 0;
2647 u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN;
2648 bool checked_extent_shared = false;
2652 * There can be no delalloc past i_size, so don't waste time looking for
2655 while (cur_offset < end && cur_offset < i_size) {
2659 u64 prealloc_len = 0;
2662 delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end,
2663 delalloc_cached_state,
2670 * If this is a prealloc extent we have to report every section
2671 * of it that has no delalloc.
2673 if (disk_bytenr != 0) {
2674 if (last_delalloc_end == 0) {
2675 prealloc_start = start;
2676 prealloc_len = delalloc_start - start;
2678 prealloc_start = last_delalloc_end + 1;
2679 prealloc_len = delalloc_start - prealloc_start;
2683 if (prealloc_len > 0) {
2684 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2685 ret = btrfs_is_data_extent_shared(inode,
2692 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2694 checked_extent_shared = true;
2696 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2697 disk_bytenr + extent_offset,
2698 prealloc_len, prealloc_flags);
2701 extent_offset += prealloc_len;
2704 ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
2705 delalloc_end + 1 - delalloc_start,
2706 FIEMAP_EXTENT_DELALLOC |
2707 FIEMAP_EXTENT_UNKNOWN);
2711 last_delalloc_end = delalloc_end;
2712 cur_offset = delalloc_end + 1;
2713 extent_offset += cur_offset - delalloc_start;
2718 * Either we found no delalloc for the whole prealloc extent or we have
2719 * a prealloc extent that spans i_size or starts at or after i_size.
2721 if (disk_bytenr != 0 && last_delalloc_end < end) {
2725 if (last_delalloc_end == 0) {
2726 prealloc_start = start;
2727 prealloc_len = end + 1 - start;
2729 prealloc_start = last_delalloc_end + 1;
2730 prealloc_len = end + 1 - prealloc_start;
2733 if (!checked_extent_shared && fieinfo->fi_extents_max) {
2734 ret = btrfs_is_data_extent_shared(inode,
2741 prealloc_flags |= FIEMAP_EXTENT_SHARED;
2743 ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
2744 disk_bytenr + extent_offset,
2745 prealloc_len, prealloc_flags);
2753 static int fiemap_find_last_extent_offset(struct btrfs_inode *inode,
2754 struct btrfs_path *path,
2755 u64 *last_extent_end_ret)
2757 const u64 ino = btrfs_ino(inode);
2758 struct btrfs_root *root = inode->root;
2759 struct extent_buffer *leaf;
2760 struct btrfs_file_extent_item *ei;
2761 struct btrfs_key key;
2766 * Lookup the last file extent. We're not using i_size here because
2767 * there might be preallocation past i_size.
2769 ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0);
2770 /* There can't be a file extent item at offset (u64)-1 */
2776 * For a non-existing key, btrfs_search_slot() always leaves us at a
2777 * slot > 0, except if the btree is empty, which is impossible because
2778 * at least it has the inode item for this inode and all the items for
2779 * the root inode 256.
2781 ASSERT(path->slots[0] > 0);
2783 leaf = path->nodes[0];
2784 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2785 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
2786 /* No file extent items in the subvolume tree. */
2787 *last_extent_end_ret = 0;
2792 * For an inline extent, the disk_bytenr is where inline data starts at,
2793 * so first check if we have an inline extent item before checking if we
2794 * have an implicit hole (disk_bytenr == 0).
2796 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2797 if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) {
2798 *last_extent_end_ret = btrfs_file_extent_end(path);
2803 * Find the last file extent item that is not a hole (when NO_HOLES is
2804 * not enabled). This should take at most 2 iterations in the worst
2805 * case: we have one hole file extent item at slot 0 of a leaf and
2806 * another hole file extent item as the last item in the previous leaf.
2807 * This is because we merge file extent items that represent holes.
2809 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2810 while (disk_bytenr == 0) {
2811 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
2814 } else if (ret > 0) {
2815 /* No file extent items that are not holes. */
2816 *last_extent_end_ret = 0;
2819 leaf = path->nodes[0];
2820 ei = btrfs_item_ptr(leaf, path->slots[0],
2821 struct btrfs_file_extent_item);
2822 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2825 *last_extent_end_ret = btrfs_file_extent_end(path);
2829 int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
2832 const u64 ino = btrfs_ino(inode);
2833 struct extent_state *cached_state = NULL;
2834 struct extent_state *delalloc_cached_state = NULL;
2835 struct btrfs_path *path;
2836 struct fiemap_cache cache = { 0 };
2837 struct btrfs_backref_share_check_ctx *backref_ctx;
2838 u64 last_extent_end;
2839 u64 prev_extent_end;
2842 bool stopped = false;
2845 backref_ctx = btrfs_alloc_backref_share_check_ctx();
2846 path = btrfs_alloc_path();
2847 if (!backref_ctx || !path) {
2852 lockstart = round_down(start, inode->root->fs_info->sectorsize);
2853 lockend = round_up(start + len, inode->root->fs_info->sectorsize);
2854 prev_extent_end = lockstart;
2856 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
2857 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
2859 ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end);
2862 btrfs_release_path(path);
2864 path->reada = READA_FORWARD;
2865 ret = fiemap_search_slot(inode, path, lockstart);
2868 } else if (ret > 0) {
2870 * No file extent item found, but we may have delalloc between
2871 * the current offset and i_size. So check for that.
2874 goto check_eof_delalloc;
2877 while (prev_extent_end < lockend) {
2878 struct extent_buffer *leaf = path->nodes[0];
2879 struct btrfs_file_extent_item *ei;
2880 struct btrfs_key key;
2883 u64 extent_offset = 0;
2885 u64 disk_bytenr = 0;
2890 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2891 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
2894 extent_end = btrfs_file_extent_end(path);
2897 * The first iteration can leave us at an extent item that ends
2898 * before our range's start. Move to the next item.
2900 if (extent_end <= lockstart)
2903 backref_ctx->curr_leaf_bytenr = leaf->start;
2905 /* We have in implicit hole (NO_HOLES feature enabled). */
2906 if (prev_extent_end < key.offset) {
2907 const u64 range_end = min(key.offset, lockend) - 1;
2909 ret = fiemap_process_hole(inode, fieinfo, &cache,
2910 &delalloc_cached_state,
2911 backref_ctx, 0, 0, 0,
2912 prev_extent_end, range_end);
2915 } else if (ret > 0) {
2916 /* fiemap_fill_next_extent() told us to stop. */
2921 /* We've reached the end of the fiemap range, stop. */
2922 if (key.offset >= lockend) {
2928 extent_len = extent_end - key.offset;
2929 ei = btrfs_item_ptr(leaf, path->slots[0],
2930 struct btrfs_file_extent_item);
2931 compression = btrfs_file_extent_compression(leaf, ei);
2932 extent_type = btrfs_file_extent_type(leaf, ei);
2933 extent_gen = btrfs_file_extent_generation(leaf, ei);
2935 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2936 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei);
2937 if (compression == BTRFS_COMPRESS_NONE)
2938 extent_offset = btrfs_file_extent_offset(leaf, ei);
2941 if (compression != BTRFS_COMPRESS_NONE)
2942 flags |= FIEMAP_EXTENT_ENCODED;
2944 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2945 flags |= FIEMAP_EXTENT_DATA_INLINE;
2946 flags |= FIEMAP_EXTENT_NOT_ALIGNED;
2947 ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
2949 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2950 ret = fiemap_process_hole(inode, fieinfo, &cache,
2951 &delalloc_cached_state,
2953 disk_bytenr, extent_offset,
2954 extent_gen, key.offset,
2956 } else if (disk_bytenr == 0) {
2957 /* We have an explicit hole. */
2958 ret = fiemap_process_hole(inode, fieinfo, &cache,
2959 &delalloc_cached_state,
2960 backref_ctx, 0, 0, 0,
2961 key.offset, extent_end - 1);
2963 /* We have a regular extent. */
2964 if (fieinfo->fi_extents_max) {
2965 ret = btrfs_is_data_extent_shared(inode,
2972 flags |= FIEMAP_EXTENT_SHARED;
2975 ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
2976 disk_bytenr + extent_offset,
2982 } else if (ret > 0) {
2983 /* fiemap_fill_next_extent() told us to stop. */
2988 prev_extent_end = extent_end;
2990 if (fatal_signal_pending(current)) {
2995 ret = fiemap_next_leaf_item(inode, path);
2998 } else if (ret > 0) {
2999 /* No more file extent items for this inode. */
3007 * Release (and free) the path before emitting any final entries to
3008 * fiemap_fill_next_extent() to keep lockdep happy. This is because
3009 * once we find no more file extent items exist, we may have a
3010 * non-cloned leaf, and fiemap_fill_next_extent() can trigger page
3011 * faults when copying data to the user space buffer.
3013 btrfs_free_path(path);
3016 if (!stopped && prev_extent_end < lockend) {
3017 ret = fiemap_process_hole(inode, fieinfo, &cache,
3018 &delalloc_cached_state, backref_ctx,
3019 0, 0, 0, prev_extent_end, lockend - 1);
3022 prev_extent_end = lockend;
3025 if (cache.cached && cache.offset + cache.len >= last_extent_end) {
3026 const u64 i_size = i_size_read(&inode->vfs_inode);
3028 if (prev_extent_end < i_size) {
3033 delalloc = btrfs_find_delalloc_in_range(inode,
3036 &delalloc_cached_state,
3040 cache.flags |= FIEMAP_EXTENT_LAST;
3042 cache.flags |= FIEMAP_EXTENT_LAST;
3046 ret = emit_last_fiemap_cache(fieinfo, &cache);
3049 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3050 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3052 free_extent_state(delalloc_cached_state);
3053 btrfs_free_backref_share_ctx(backref_ctx);
3054 btrfs_free_path(path);
3058 static void __free_extent_buffer(struct extent_buffer *eb)
3060 kmem_cache_free(extent_buffer_cache, eb);
3063 static int extent_buffer_under_io(const struct extent_buffer *eb)
3065 return (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) ||
3066 test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3069 static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page)
3071 struct btrfs_subpage *subpage;
3073 lockdep_assert_held(&page->mapping->private_lock);
3075 if (PagePrivate(page)) {
3076 subpage = (struct btrfs_subpage *)page->private;
3077 if (atomic_read(&subpage->eb_refs))
3080 * Even there is no eb refs here, we may still have
3081 * end_page_read() call relying on page::private.
3083 if (atomic_read(&subpage->readers))
3089 static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page)
3091 struct btrfs_fs_info *fs_info = eb->fs_info;
3092 const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3095 * For mapped eb, we're going to change the page private, which should
3096 * be done under the private_lock.
3099 spin_lock(&page->mapping->private_lock);
3101 if (!PagePrivate(page)) {
3103 spin_unlock(&page->mapping->private_lock);
3107 if (fs_info->nodesize >= PAGE_SIZE) {
3109 * We do this since we'll remove the pages after we've
3110 * removed the eb from the radix tree, so we could race
3111 * and have this page now attached to the new eb. So
3112 * only clear page_private if it's still connected to
3115 if (PagePrivate(page) &&
3116 page->private == (unsigned long)eb) {
3117 BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
3118 BUG_ON(PageDirty(page));
3119 BUG_ON(PageWriteback(page));
3121 * We need to make sure we haven't be attached
3124 detach_page_private(page);
3127 spin_unlock(&page->mapping->private_lock);
3132 * For subpage, we can have dummy eb with page private. In this case,
3133 * we can directly detach the private as such page is only attached to
3134 * one dummy eb, no sharing.
3137 btrfs_detach_subpage(fs_info, page);
3141 btrfs_page_dec_eb_refs(fs_info, page);
3144 * We can only detach the page private if there are no other ebs in the
3145 * page range and no unfinished IO.
3147 if (!page_range_has_eb(fs_info, page))
3148 btrfs_detach_subpage(fs_info, page);
3150 spin_unlock(&page->mapping->private_lock);
3153 /* Release all pages attached to the extent buffer */
3154 static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb)
3159 ASSERT(!extent_buffer_under_io(eb));
3161 num_pages = num_extent_pages(eb);
3162 for (i = 0; i < num_pages; i++) {
3163 struct page *page = eb->pages[i];
3168 detach_extent_buffer_page(eb, page);
3170 /* One for when we allocated the page */
3176 * Helper for releasing the extent buffer.
3178 static inline void btrfs_release_extent_buffer(struct extent_buffer *eb)
3180 btrfs_release_extent_buffer_pages(eb);
3181 btrfs_leak_debug_del_eb(eb);
3182 __free_extent_buffer(eb);
3185 static struct extent_buffer *
3186 __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start,
3189 struct extent_buffer *eb = NULL;
3191 eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL);
3194 eb->fs_info = fs_info;
3195 init_rwsem(&eb->lock);
3197 btrfs_leak_debug_add_eb(eb);
3199 spin_lock_init(&eb->refs_lock);
3200 atomic_set(&eb->refs, 1);
3202 ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE);
3207 struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src)
3210 struct extent_buffer *new;
3211 int num_pages = num_extent_pages(src);
3214 new = __alloc_extent_buffer(src->fs_info, src->start, src->len);
3219 * Set UNMAPPED before calling btrfs_release_extent_buffer(), as
3220 * btrfs_release_extent_buffer() have different behavior for
3221 * UNMAPPED subpage extent buffer.
3223 set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags);
3225 ret = btrfs_alloc_page_array(num_pages, new->pages);
3227 btrfs_release_extent_buffer(new);
3231 for (i = 0; i < num_pages; i++) {
3233 struct page *p = new->pages[i];
3235 ret = attach_extent_buffer_page(new, p, NULL);
3237 btrfs_release_extent_buffer(new);
3240 WARN_ON(PageDirty(p));
3241 copy_page(page_address(p), page_address(src->pages[i]));
3243 set_extent_buffer_uptodate(new);
3248 struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3249 u64 start, unsigned long len)
3251 struct extent_buffer *eb;
3256 eb = __alloc_extent_buffer(fs_info, start, len);
3260 num_pages = num_extent_pages(eb);
3261 ret = btrfs_alloc_page_array(num_pages, eb->pages);
3265 for (i = 0; i < num_pages; i++) {
3266 struct page *p = eb->pages[i];
3268 ret = attach_extent_buffer_page(eb, p, NULL);
3273 set_extent_buffer_uptodate(eb);
3274 btrfs_set_header_nritems(eb, 0);
3275 set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags);
3279 for (i = 0; i < num_pages; i++) {
3281 detach_extent_buffer_page(eb, eb->pages[i]);
3282 __free_page(eb->pages[i]);
3285 __free_extent_buffer(eb);
3289 struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info,
3292 return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize);
3295 static void check_buffer_tree_ref(struct extent_buffer *eb)
3299 * The TREE_REF bit is first set when the extent_buffer is added
3300 * to the radix tree. It is also reset, if unset, when a new reference
3301 * is created by find_extent_buffer.
3303 * It is only cleared in two cases: freeing the last non-tree
3304 * reference to the extent_buffer when its STALE bit is set or
3305 * calling release_folio when the tree reference is the only reference.
3307 * In both cases, care is taken to ensure that the extent_buffer's
3308 * pages are not under io. However, release_folio can be concurrently
3309 * called with creating new references, which is prone to race
3310 * conditions between the calls to check_buffer_tree_ref in those
3311 * codepaths and clearing TREE_REF in try_release_extent_buffer.
3313 * The actual lifetime of the extent_buffer in the radix tree is
3314 * adequately protected by the refcount, but the TREE_REF bit and
3315 * its corresponding reference are not. To protect against this
3316 * class of races, we call check_buffer_tree_ref from the codepaths
3317 * which trigger io. Note that once io is initiated, TREE_REF can no
3318 * longer be cleared, so that is the moment at which any such race is
3321 refs = atomic_read(&eb->refs);
3322 if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3325 spin_lock(&eb->refs_lock);
3326 if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3327 atomic_inc(&eb->refs);
3328 spin_unlock(&eb->refs_lock);
3331 static void mark_extent_buffer_accessed(struct extent_buffer *eb,
3332 struct page *accessed)
3336 check_buffer_tree_ref(eb);
3338 num_pages = num_extent_pages(eb);
3339 for (i = 0; i < num_pages; i++) {
3340 struct page *p = eb->pages[i];
3343 mark_page_accessed(p);
3347 struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
3350 struct extent_buffer *eb;
3352 eb = find_extent_buffer_nolock(fs_info, start);
3356 * Lock our eb's refs_lock to avoid races with free_extent_buffer().
3357 * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and
3358 * another task running free_extent_buffer() might have seen that flag
3359 * set, eb->refs == 2, that the buffer isn't under IO (dirty and
3360 * writeback flags not set) and it's still in the tree (flag
3361 * EXTENT_BUFFER_TREE_REF set), therefore being in the process of
3362 * decrementing the extent buffer's reference count twice. So here we
3363 * could race and increment the eb's reference count, clear its stale
3364 * flag, mark it as dirty and drop our reference before the other task
3365 * finishes executing free_extent_buffer, which would later result in
3366 * an attempt to free an extent buffer that is dirty.
3368 if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
3369 spin_lock(&eb->refs_lock);
3370 spin_unlock(&eb->refs_lock);
3372 mark_extent_buffer_accessed(eb, NULL);
3376 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3377 struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info,
3380 struct extent_buffer *eb, *exists = NULL;
3383 eb = find_extent_buffer(fs_info, start);
3386 eb = alloc_dummy_extent_buffer(fs_info, start);
3388 return ERR_PTR(-ENOMEM);
3389 eb->fs_info = fs_info;
3391 ret = radix_tree_preload(GFP_NOFS);
3393 exists = ERR_PTR(ret);
3396 spin_lock(&fs_info->buffer_lock);
3397 ret = radix_tree_insert(&fs_info->buffer_radix,
3398 start >> fs_info->sectorsize_bits, eb);
3399 spin_unlock(&fs_info->buffer_lock);
3400 radix_tree_preload_end();
3401 if (ret == -EEXIST) {
3402 exists = find_extent_buffer(fs_info, start);
3408 check_buffer_tree_ref(eb);
3409 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3413 btrfs_release_extent_buffer(eb);
3418 static struct extent_buffer *grab_extent_buffer(
3419 struct btrfs_fs_info *fs_info, struct page *page)
3421 struct extent_buffer *exists;
3424 * For subpage case, we completely rely on radix tree to ensure we
3425 * don't try to insert two ebs for the same bytenr. So here we always
3426 * return NULL and just continue.
3428 if (fs_info->nodesize < PAGE_SIZE)
3431 /* Page not yet attached to an extent buffer */
3432 if (!PagePrivate(page))
3436 * We could have already allocated an eb for this page and attached one
3437 * so lets see if we can get a ref on the existing eb, and if we can we
3438 * know it's good and we can just return that one, else we know we can
3439 * just overwrite page->private.
3441 exists = (struct extent_buffer *)page->private;
3442 if (atomic_inc_not_zero(&exists->refs))
3445 WARN_ON(PageDirty(page));
3446 detach_page_private(page);
3450 static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start)
3452 if (!IS_ALIGNED(start, fs_info->sectorsize)) {
3453 btrfs_err(fs_info, "bad tree block start %llu", start);
3457 if (fs_info->nodesize < PAGE_SIZE &&
3458 offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
3460 "tree block crosses page boundary, start %llu nodesize %u",
3461 start, fs_info->nodesize);
3464 if (fs_info->nodesize >= PAGE_SIZE &&
3465 !PAGE_ALIGNED(start)) {
3467 "tree block is not page aligned, start %llu nodesize %u",
3468 start, fs_info->nodesize);
3474 struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info,
3475 u64 start, u64 owner_root, int level)
3477 unsigned long len = fs_info->nodesize;
3480 unsigned long index = start >> PAGE_SHIFT;
3481 struct extent_buffer *eb;
3482 struct extent_buffer *exists = NULL;
3484 struct address_space *mapping = fs_info->btree_inode->i_mapping;
3485 struct btrfs_subpage *prealloc = NULL;
3486 u64 lockdep_owner = owner_root;
3490 if (check_eb_alignment(fs_info, start))
3491 return ERR_PTR(-EINVAL);
3493 #if BITS_PER_LONG == 32
3494 if (start >= MAX_LFS_FILESIZE) {
3495 btrfs_err_rl(fs_info,
3496 "extent buffer %llu is beyond 32bit page cache limit", start);
3497 btrfs_err_32bit_limit(fs_info);
3498 return ERR_PTR(-EOVERFLOW);
3500 if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD)
3501 btrfs_warn_32bit_limit(fs_info);
3504 eb = find_extent_buffer(fs_info, start);
3508 eb = __alloc_extent_buffer(fs_info, start, len);
3510 return ERR_PTR(-ENOMEM);
3513 * The reloc trees are just snapshots, so we need them to appear to be
3514 * just like any other fs tree WRT lockdep.
3516 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID)
3517 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
3519 btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level);
3521 num_pages = num_extent_pages(eb);
3524 * Preallocate page->private for subpage case, so that we won't
3525 * allocate memory with private_lock nor page lock hold.
3527 * The memory will be freed by attach_extent_buffer_page() or freed
3528 * manually if we exit earlier.
3530 if (fs_info->nodesize < PAGE_SIZE) {
3531 prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA);
3532 if (IS_ERR(prealloc)) {
3533 exists = ERR_CAST(prealloc);
3538 for (i = 0; i < num_pages; i++, index++) {
3539 p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL);
3541 exists = ERR_PTR(-ENOMEM);
3542 btrfs_free_subpage(prealloc);
3546 spin_lock(&mapping->private_lock);
3547 exists = grab_extent_buffer(fs_info, p);
3549 spin_unlock(&mapping->private_lock);
3552 mark_extent_buffer_accessed(exists, p);
3553 btrfs_free_subpage(prealloc);
3556 /* Should not fail, as we have preallocated the memory */
3557 ret = attach_extent_buffer_page(eb, p, prealloc);
3560 * To inform we have extra eb under allocation, so that
3561 * detach_extent_buffer_page() won't release the page private
3562 * when the eb hasn't yet been inserted into radix tree.
3564 * The ref will be decreased when the eb released the page, in
3565 * detach_extent_buffer_page().
3566 * Thus needs no special handling in error path.
3568 btrfs_page_inc_eb_refs(fs_info, p);
3569 spin_unlock(&mapping->private_lock);
3571 WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len));
3573 if (!btrfs_page_test_uptodate(fs_info, p, eb->start, eb->len))
3577 * We can't unlock the pages just yet since the extent buffer
3578 * hasn't been properly inserted in the radix tree, this
3579 * opens a race with btree_release_folio which can free a page
3580 * while we are still filling in all pages for the buffer and
3585 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3587 ret = radix_tree_preload(GFP_NOFS);
3589 exists = ERR_PTR(ret);
3593 spin_lock(&fs_info->buffer_lock);
3594 ret = radix_tree_insert(&fs_info->buffer_radix,
3595 start >> fs_info->sectorsize_bits, eb);
3596 spin_unlock(&fs_info->buffer_lock);
3597 radix_tree_preload_end();
3598 if (ret == -EEXIST) {
3599 exists = find_extent_buffer(fs_info, start);
3605 /* add one reference for the tree */
3606 check_buffer_tree_ref(eb);
3607 set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags);
3610 * Now it's safe to unlock the pages because any calls to
3611 * btree_release_folio will correctly detect that a page belongs to a
3612 * live buffer and won't free them prematurely.
3614 for (i = 0; i < num_pages; i++)
3615 unlock_page(eb->pages[i]);
3619 WARN_ON(!atomic_dec_and_test(&eb->refs));
3620 for (i = 0; i < num_pages; i++) {
3622 unlock_page(eb->pages[i]);
3625 btrfs_release_extent_buffer(eb);
3629 static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head)
3631 struct extent_buffer *eb =
3632 container_of(head, struct extent_buffer, rcu_head);
3634 __free_extent_buffer(eb);
3637 static int release_extent_buffer(struct extent_buffer *eb)
3638 __releases(&eb->refs_lock)
3640 lockdep_assert_held(&eb->refs_lock);
3642 WARN_ON(atomic_read(&eb->refs) == 0);
3643 if (atomic_dec_and_test(&eb->refs)) {
3644 if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) {
3645 struct btrfs_fs_info *fs_info = eb->fs_info;
3647 spin_unlock(&eb->refs_lock);
3649 spin_lock(&fs_info->buffer_lock);
3650 radix_tree_delete(&fs_info->buffer_radix,
3651 eb->start >> fs_info->sectorsize_bits);
3652 spin_unlock(&fs_info->buffer_lock);
3654 spin_unlock(&eb->refs_lock);
3657 btrfs_leak_debug_del_eb(eb);
3658 /* Should be safe to release our pages at this point */
3659 btrfs_release_extent_buffer_pages(eb);
3660 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3661 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) {
3662 __free_extent_buffer(eb);
3666 call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu);
3669 spin_unlock(&eb->refs_lock);
3674 void free_extent_buffer(struct extent_buffer *eb)
3680 refs = atomic_read(&eb->refs);
3682 if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3)
3683 || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) &&
3686 if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1))
3690 spin_lock(&eb->refs_lock);
3691 if (atomic_read(&eb->refs) == 2 &&
3692 test_bit(EXTENT_BUFFER_STALE, &eb->bflags) &&
3693 !extent_buffer_under_io(eb) &&
3694 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3695 atomic_dec(&eb->refs);
3698 * I know this is terrible, but it's temporary until we stop tracking
3699 * the uptodate bits and such for the extent buffers.
3701 release_extent_buffer(eb);
3704 void free_extent_buffer_stale(struct extent_buffer *eb)
3709 spin_lock(&eb->refs_lock);
3710 set_bit(EXTENT_BUFFER_STALE, &eb->bflags);
3712 if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) &&
3713 test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags))
3714 atomic_dec(&eb->refs);
3715 release_extent_buffer(eb);
3718 static void btree_clear_page_dirty(struct page *page)
3720 ASSERT(PageDirty(page));
3721 ASSERT(PageLocked(page));
3722 clear_page_dirty_for_io(page);
3723 xa_lock_irq(&page->mapping->i_pages);
3724 if (!PageDirty(page))
3725 __xa_clear_mark(&page->mapping->i_pages,
3726 page_index(page), PAGECACHE_TAG_DIRTY);
3727 xa_unlock_irq(&page->mapping->i_pages);
3730 static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb)
3732 struct btrfs_fs_info *fs_info = eb->fs_info;
3733 struct page *page = eb->pages[0];
3736 /* btree_clear_page_dirty() needs page locked */
3738 last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start,
3741 btree_clear_page_dirty(page);
3743 WARN_ON(atomic_read(&eb->refs) == 0);
3746 void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans,
3747 struct extent_buffer *eb)
3749 struct btrfs_fs_info *fs_info = eb->fs_info;
3754 btrfs_assert_tree_write_locked(eb);
3756 if (trans && btrfs_header_generation(eb) != trans->transid)
3759 if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
3762 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len,
3763 fs_info->dirty_metadata_batch);
3765 if (eb->fs_info->nodesize < PAGE_SIZE)
3766 return clear_subpage_extent_buffer_dirty(eb);
3768 num_pages = num_extent_pages(eb);
3770 for (i = 0; i < num_pages; i++) {
3771 page = eb->pages[i];
3772 if (!PageDirty(page))
3775 btree_clear_page_dirty(page);
3778 WARN_ON(atomic_read(&eb->refs) == 0);
3781 void set_extent_buffer_dirty(struct extent_buffer *eb)
3787 check_buffer_tree_ref(eb);
3789 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3791 num_pages = num_extent_pages(eb);
3792 WARN_ON(atomic_read(&eb->refs) == 0);
3793 WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags));
3796 bool subpage = eb->fs_info->nodesize < PAGE_SIZE;
3799 * For subpage case, we can have other extent buffers in the
3800 * same page, and in clear_subpage_extent_buffer_dirty() we
3801 * have to clear page dirty without subpage lock held.
3802 * This can cause race where our page gets dirty cleared after
3805 * Thankfully, clear_subpage_extent_buffer_dirty() has locked
3806 * its page for other reasons, we can use page lock to prevent
3810 lock_page(eb->pages[0]);
3811 for (i = 0; i < num_pages; i++)
3812 btrfs_page_set_dirty(eb->fs_info, eb->pages[i],
3813 eb->start, eb->len);
3815 unlock_page(eb->pages[0]);
3816 percpu_counter_add_batch(&eb->fs_info->dirty_metadata_bytes,
3818 eb->fs_info->dirty_metadata_batch);
3820 #ifdef CONFIG_BTRFS_DEBUG
3821 for (i = 0; i < num_pages; i++)
3822 ASSERT(PageDirty(eb->pages[i]));
3826 void clear_extent_buffer_uptodate(struct extent_buffer *eb)
3828 struct btrfs_fs_info *fs_info = eb->fs_info;
3833 clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3834 num_pages = num_extent_pages(eb);
3835 for (i = 0; i < num_pages; i++) {
3836 page = eb->pages[i];
3841 * This is special handling for metadata subpage, as regular
3842 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3844 if (fs_info->nodesize >= PAGE_SIZE)
3845 ClearPageUptodate(page);
3847 btrfs_subpage_clear_uptodate(fs_info, page, eb->start,
3852 void set_extent_buffer_uptodate(struct extent_buffer *eb)
3854 struct btrfs_fs_info *fs_info = eb->fs_info;
3859 set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags);
3860 num_pages = num_extent_pages(eb);
3861 for (i = 0; i < num_pages; i++) {
3862 page = eb->pages[i];
3865 * This is special handling for metadata subpage, as regular
3866 * btrfs_is_subpage() can not handle cloned/dummy metadata.
3868 if (fs_info->nodesize >= PAGE_SIZE)
3869 SetPageUptodate(page);
3871 btrfs_subpage_set_uptodate(fs_info, page, eb->start,
3876 static void extent_buffer_read_end_io(struct btrfs_bio *bbio)
3878 struct extent_buffer *eb = bbio->private;
3879 struct btrfs_fs_info *fs_info = eb->fs_info;
3880 bool uptodate = !bbio->bio.bi_status;
3881 struct bvec_iter_all iter_all;
3882 struct bio_vec *bvec;
3885 eb->read_mirror = bbio->mirror_num;
3888 btrfs_validate_extent_buffer(eb, &bbio->parent_check) < 0)
3892 set_extent_buffer_uptodate(eb);
3894 clear_extent_buffer_uptodate(eb);
3895 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3898 bio_for_each_segment_all(bvec, &bbio->bio, iter_all) {
3899 u64 start = eb->start + bio_offset;
3900 struct page *page = bvec->bv_page;
3901 u32 len = bvec->bv_len;
3904 btrfs_page_set_uptodate(fs_info, page, start, len);
3906 btrfs_page_clear_uptodate(fs_info, page, start, len);
3911 clear_bit(EXTENT_BUFFER_READING, &eb->bflags);
3912 smp_mb__after_atomic();
3913 wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING);
3914 free_extent_buffer(eb);
3916 bio_put(&bbio->bio);
3919 int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num,
3920 struct btrfs_tree_parent_check *check)
3922 int num_pages = num_extent_pages(eb), i;
3923 struct btrfs_bio *bbio;
3925 if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3929 * We could have had EXTENT_BUFFER_UPTODATE cleared by the write
3930 * operation, which could potentially still be in flight. In this case
3931 * we simply want to return an error.
3933 if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)))
3936 /* Someone else is already reading the buffer, just wait for it. */
3937 if (test_and_set_bit(EXTENT_BUFFER_READING, &eb->bflags))
3940 clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
3941 eb->read_mirror = 0;
3942 check_buffer_tree_ref(eb);
3943 atomic_inc(&eb->refs);
3945 bbio = btrfs_bio_alloc(INLINE_EXTENT_BUFFER_PAGES,
3946 REQ_OP_READ | REQ_META, eb->fs_info,
3947 extent_buffer_read_end_io, eb);
3948 bbio->bio.bi_iter.bi_sector = eb->start >> SECTOR_SHIFT;
3949 bbio->inode = BTRFS_I(eb->fs_info->btree_inode);
3950 bbio->file_offset = eb->start;
3951 memcpy(&bbio->parent_check, check, sizeof(*check));
3952 if (eb->fs_info->nodesize < PAGE_SIZE) {
3953 __bio_add_page(&bbio->bio, eb->pages[0], eb->len,
3954 eb->start - page_offset(eb->pages[0]));
3956 for (i = 0; i < num_pages; i++)
3957 __bio_add_page(&bbio->bio, eb->pages[i], PAGE_SIZE, 0);
3959 btrfs_submit_bio(bbio, mirror_num);
3962 if (wait == WAIT_COMPLETE) {
3963 wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_READING, TASK_UNINTERRUPTIBLE);
3964 if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))
3971 static bool report_eb_range(const struct extent_buffer *eb, unsigned long start,
3974 btrfs_warn(eb->fs_info,
3975 "access to eb bytenr %llu len %lu out of range start %lu len %lu",
3976 eb->start, eb->len, start, len);
3977 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3983 * Check if the [start, start + len) range is valid before reading/writing
3985 * NOTE: @start and @len are offset inside the eb, not logical address.
3987 * Caller should not touch the dst/src memory if this function returns error.
3989 static inline int check_eb_range(const struct extent_buffer *eb,
3990 unsigned long start, unsigned long len)
3992 unsigned long offset;
3994 /* start, start + len should not go beyond eb->len nor overflow */
3995 if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len))
3996 return report_eb_range(eb, start, len);
4001 void read_extent_buffer(const struct extent_buffer *eb, void *dstv,
4002 unsigned long start, unsigned long len)
4008 char *dst = (char *)dstv;
4009 unsigned long i = get_eb_page_index(start);
4011 if (check_eb_range(eb, start, len))
4014 offset = get_eb_offset_in_page(eb, start);
4017 page = eb->pages[i];
4019 cur = min(len, (PAGE_SIZE - offset));
4020 kaddr = page_address(page);
4021 memcpy(dst, kaddr + offset, cur);
4030 int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb,
4032 unsigned long start, unsigned long len)
4038 char __user *dst = (char __user *)dstv;
4039 unsigned long i = get_eb_page_index(start);
4042 WARN_ON(start > eb->len);
4043 WARN_ON(start + len > eb->start + eb->len);
4045 offset = get_eb_offset_in_page(eb, start);
4048 page = eb->pages[i];
4050 cur = min(len, (PAGE_SIZE - offset));
4051 kaddr = page_address(page);
4052 if (copy_to_user_nofault(dst, kaddr + offset, cur)) {
4066 int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv,
4067 unsigned long start, unsigned long len)
4073 char *ptr = (char *)ptrv;
4074 unsigned long i = get_eb_page_index(start);
4077 if (check_eb_range(eb, start, len))
4080 offset = get_eb_offset_in_page(eb, start);
4083 page = eb->pages[i];
4085 cur = min(len, (PAGE_SIZE - offset));
4087 kaddr = page_address(page);
4088 ret = memcmp(ptr, kaddr + offset, cur);
4101 * Check that the extent buffer is uptodate.
4103 * For regular sector size == PAGE_SIZE case, check if @page is uptodate.
4104 * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE.
4106 static void assert_eb_page_uptodate(const struct extent_buffer *eb,
4109 struct btrfs_fs_info *fs_info = eb->fs_info;
4112 * If we are using the commit root we could potentially clear a page
4113 * Uptodate while we're using the extent buffer that we've previously
4114 * looked up. We don't want to complain in this case, as the page was
4115 * valid before, we just didn't write it out. Instead we want to catch
4116 * the case where we didn't actually read the block properly, which
4117 * would have !PageUptodate and !EXTENT_BUFFER_WRITE_ERR.
4119 if (test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))
4122 if (fs_info->nodesize < PAGE_SIZE) {
4123 if (WARN_ON(!btrfs_subpage_test_uptodate(fs_info, page,
4124 eb->start, eb->len)))
4125 btrfs_subpage_dump_bitmap(fs_info, page, eb->start, eb->len);
4127 WARN_ON(!PageUptodate(page));
4131 void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb,
4136 assert_eb_page_uptodate(eb, eb->pages[0]);
4137 kaddr = page_address(eb->pages[0]) +
4138 get_eb_offset_in_page(eb, offsetof(struct btrfs_header,
4140 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
4143 void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv)
4147 assert_eb_page_uptodate(eb, eb->pages[0]);
4148 kaddr = page_address(eb->pages[0]) +
4149 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid));
4150 memcpy(kaddr, srcv, BTRFS_FSID_SIZE);
4153 void write_extent_buffer(const struct extent_buffer *eb, const void *srcv,
4154 unsigned long start, unsigned long len)
4160 char *src = (char *)srcv;
4161 unsigned long i = get_eb_page_index(start);
4163 WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags));
4165 if (check_eb_range(eb, start, len))
4168 offset = get_eb_offset_in_page(eb, start);
4171 page = eb->pages[i];
4172 assert_eb_page_uptodate(eb, page);
4174 cur = min(len, PAGE_SIZE - offset);
4175 kaddr = page_address(page);
4176 memcpy(kaddr + offset, src, cur);
4185 void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start,
4192 unsigned long i = get_eb_page_index(start);
4194 if (check_eb_range(eb, start, len))
4197 offset = get_eb_offset_in_page(eb, start);
4200 page = eb->pages[i];
4201 assert_eb_page_uptodate(eb, page);
4203 cur = min(len, PAGE_SIZE - offset);
4204 kaddr = page_address(page);
4205 memset(kaddr + offset, 0, cur);
4213 void copy_extent_buffer_full(const struct extent_buffer *dst,
4214 const struct extent_buffer *src)
4219 ASSERT(dst->len == src->len);
4221 if (dst->fs_info->nodesize >= PAGE_SIZE) {
4222 num_pages = num_extent_pages(dst);
4223 for (i = 0; i < num_pages; i++)
4224 copy_page(page_address(dst->pages[i]),
4225 page_address(src->pages[i]));
4227 size_t src_offset = get_eb_offset_in_page(src, 0);
4228 size_t dst_offset = get_eb_offset_in_page(dst, 0);
4230 ASSERT(src->fs_info->nodesize < PAGE_SIZE);
4231 memcpy(page_address(dst->pages[0]) + dst_offset,
4232 page_address(src->pages[0]) + src_offset,
4237 void copy_extent_buffer(const struct extent_buffer *dst,
4238 const struct extent_buffer *src,
4239 unsigned long dst_offset, unsigned long src_offset,
4242 u64 dst_len = dst->len;
4247 unsigned long i = get_eb_page_index(dst_offset);
4249 if (check_eb_range(dst, dst_offset, len) ||
4250 check_eb_range(src, src_offset, len))
4253 WARN_ON(src->len != dst_len);
4255 offset = get_eb_offset_in_page(dst, dst_offset);
4258 page = dst->pages[i];
4259 assert_eb_page_uptodate(dst, page);
4261 cur = min(len, (unsigned long)(PAGE_SIZE - offset));
4263 kaddr = page_address(page);
4264 read_extent_buffer(src, kaddr + offset, src_offset, cur);
4274 * eb_bitmap_offset() - calculate the page and offset of the byte containing the
4276 * @eb: the extent buffer
4277 * @start: offset of the bitmap item in the extent buffer
4279 * @page_index: return index of the page in the extent buffer that contains the
4281 * @page_offset: return offset into the page given by page_index
4283 * This helper hides the ugliness of finding the byte in an extent buffer which
4284 * contains a given bit.
4286 static inline void eb_bitmap_offset(const struct extent_buffer *eb,
4287 unsigned long start, unsigned long nr,
4288 unsigned long *page_index,
4289 size_t *page_offset)
4291 size_t byte_offset = BIT_BYTE(nr);
4295 * The byte we want is the offset of the extent buffer + the offset of
4296 * the bitmap item in the extent buffer + the offset of the byte in the
4299 offset = start + offset_in_page(eb->start) + byte_offset;
4301 *page_index = offset >> PAGE_SHIFT;
4302 *page_offset = offset_in_page(offset);
4306 * Determine whether a bit in a bitmap item is set.
4308 * @eb: the extent buffer
4309 * @start: offset of the bitmap item in the extent buffer
4310 * @nr: bit number to test
4312 int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start,
4320 eb_bitmap_offset(eb, start, nr, &i, &offset);
4321 page = eb->pages[i];
4322 assert_eb_page_uptodate(eb, page);
4323 kaddr = page_address(page);
4324 return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1)));
4328 * Set an area of a bitmap to 1.
4330 * @eb: the extent buffer
4331 * @start: offset of the bitmap item in the extent buffer
4332 * @pos: bit number of the first bit
4333 * @len: number of bits to set
4335 void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start,
4336 unsigned long pos, unsigned long len)
4342 const unsigned int size = pos + len;
4343 int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
4344 u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);
4346 eb_bitmap_offset(eb, start, pos, &i, &offset);
4347 page = eb->pages[i];
4348 assert_eb_page_uptodate(eb, page);
4349 kaddr = page_address(page);
4351 while (len >= bits_to_set) {
4352 kaddr[offset] |= mask_to_set;
4354 bits_to_set = BITS_PER_BYTE;
4356 if (++offset >= PAGE_SIZE && len > 0) {
4358 page = eb->pages[++i];
4359 assert_eb_page_uptodate(eb, page);
4360 kaddr = page_address(page);
4364 mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
4365 kaddr[offset] |= mask_to_set;
4371 * Clear an area of a bitmap.
4373 * @eb: the extent buffer
4374 * @start: offset of the bitmap item in the extent buffer
4375 * @pos: bit number of the first bit
4376 * @len: number of bits to clear
4378 void extent_buffer_bitmap_clear(const struct extent_buffer *eb,
4379 unsigned long start, unsigned long pos,
4386 const unsigned int size = pos + len;
4387 int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
4388 u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);
4390 eb_bitmap_offset(eb, start, pos, &i, &offset);
4391 page = eb->pages[i];
4392 assert_eb_page_uptodate(eb, page);
4393 kaddr = page_address(page);
4395 while (len >= bits_to_clear) {
4396 kaddr[offset] &= ~mask_to_clear;
4397 len -= bits_to_clear;
4398 bits_to_clear = BITS_PER_BYTE;
4400 if (++offset >= PAGE_SIZE && len > 0) {
4402 page = eb->pages[++i];
4403 assert_eb_page_uptodate(eb, page);
4404 kaddr = page_address(page);
4408 mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
4409 kaddr[offset] &= ~mask_to_clear;
4413 static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len)
4415 unsigned long distance = (src > dst) ? src - dst : dst - src;
4416 return distance < len;
4419 static void copy_pages(struct page *dst_page, struct page *src_page,
4420 unsigned long dst_off, unsigned long src_off,
4423 char *dst_kaddr = page_address(dst_page);
4425 int must_memmove = 0;
4427 if (dst_page != src_page) {
4428 src_kaddr = page_address(src_page);
4430 src_kaddr = dst_kaddr;
4431 if (areas_overlap(src_off, dst_off, len))
4436 memmove(dst_kaddr + dst_off, src_kaddr + src_off, len);
4438 memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len);
4441 void memcpy_extent_buffer(const struct extent_buffer *dst,
4442 unsigned long dst_offset, unsigned long src_offset,
4446 size_t dst_off_in_page;
4447 size_t src_off_in_page;
4448 unsigned long dst_i;
4449 unsigned long src_i;
4451 if (check_eb_range(dst, dst_offset, len) ||
4452 check_eb_range(dst, src_offset, len))
4456 dst_off_in_page = get_eb_offset_in_page(dst, dst_offset);
4457 src_off_in_page = get_eb_offset_in_page(dst, src_offset);
4459 dst_i = get_eb_page_index(dst_offset);
4460 src_i = get_eb_page_index(src_offset);
4462 cur = min(len, (unsigned long)(PAGE_SIZE -
4464 cur = min_t(unsigned long, cur,
4465 (unsigned long)(PAGE_SIZE - dst_off_in_page));
4467 copy_pages(dst->pages[dst_i], dst->pages[src_i],
4468 dst_off_in_page, src_off_in_page, cur);
4476 void memmove_extent_buffer(const struct extent_buffer *dst,
4477 unsigned long dst_offset, unsigned long src_offset,
4481 size_t dst_off_in_page;
4482 size_t src_off_in_page;
4483 unsigned long dst_end = dst_offset + len - 1;
4484 unsigned long src_end = src_offset + len - 1;
4485 unsigned long dst_i;
4486 unsigned long src_i;
4488 if (check_eb_range(dst, dst_offset, len) ||
4489 check_eb_range(dst, src_offset, len))
4491 if (dst_offset < src_offset) {
4492 memcpy_extent_buffer(dst, dst_offset, src_offset, len);
4496 dst_i = get_eb_page_index(dst_end);
4497 src_i = get_eb_page_index(src_end);
4499 dst_off_in_page = get_eb_offset_in_page(dst, dst_end);
4500 src_off_in_page = get_eb_offset_in_page(dst, src_end);
4502 cur = min_t(unsigned long, len, src_off_in_page + 1);
4503 cur = min(cur, dst_off_in_page + 1);
4504 copy_pages(dst->pages[dst_i], dst->pages[src_i],
4505 dst_off_in_page - cur + 1,
4506 src_off_in_page - cur + 1, cur);
4514 #define GANG_LOOKUP_SIZE 16
4515 static struct extent_buffer *get_next_extent_buffer(
4516 struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr)
4518 struct extent_buffer *gang[GANG_LOOKUP_SIZE];
4519 struct extent_buffer *found = NULL;
4520 u64 page_start = page_offset(page);
4521 u64 cur = page_start;
4523 ASSERT(in_range(bytenr, page_start, PAGE_SIZE));
4524 lockdep_assert_held(&fs_info->buffer_lock);
4526 while (cur < page_start + PAGE_SIZE) {
4530 ret = radix_tree_gang_lookup(&fs_info->buffer_radix,
4531 (void **)gang, cur >> fs_info->sectorsize_bits,
4532 min_t(unsigned int, GANG_LOOKUP_SIZE,
4533 PAGE_SIZE / fs_info->nodesize));
4536 for (i = 0; i < ret; i++) {
4537 /* Already beyond page end */
4538 if (gang[i]->start >= page_start + PAGE_SIZE)
4541 if (gang[i]->start >= bytenr) {
4546 cur = gang[ret - 1]->start + gang[ret - 1]->len;
4552 static int try_release_subpage_extent_buffer(struct page *page)
4554 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
4555 u64 cur = page_offset(page);
4556 const u64 end = page_offset(page) + PAGE_SIZE;
4560 struct extent_buffer *eb = NULL;
4563 * Unlike try_release_extent_buffer() which uses page->private
4564 * to grab buffer, for subpage case we rely on radix tree, thus
4565 * we need to ensure radix tree consistency.
4567 * We also want an atomic snapshot of the radix tree, thus go
4568 * with spinlock rather than RCU.
4570 spin_lock(&fs_info->buffer_lock);
4571 eb = get_next_extent_buffer(fs_info, page, cur);
4573 /* No more eb in the page range after or at cur */
4574 spin_unlock(&fs_info->buffer_lock);
4577 cur = eb->start + eb->len;
4580 * The same as try_release_extent_buffer(), to ensure the eb
4581 * won't disappear out from under us.
4583 spin_lock(&eb->refs_lock);
4584 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4585 spin_unlock(&eb->refs_lock);
4586 spin_unlock(&fs_info->buffer_lock);
4589 spin_unlock(&fs_info->buffer_lock);
4592 * If tree ref isn't set then we know the ref on this eb is a
4593 * real ref, so just return, this eb will likely be freed soon
4596 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4597 spin_unlock(&eb->refs_lock);
4602 * Here we don't care about the return value, we will always
4603 * check the page private at the end. And
4604 * release_extent_buffer() will release the refs_lock.
4606 release_extent_buffer(eb);
4609 * Finally to check if we have cleared page private, as if we have
4610 * released all ebs in the page, the page private should be cleared now.
4612 spin_lock(&page->mapping->private_lock);
4613 if (!PagePrivate(page))
4617 spin_unlock(&page->mapping->private_lock);
4622 int try_release_extent_buffer(struct page *page)
4624 struct extent_buffer *eb;
4626 if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
4627 return try_release_subpage_extent_buffer(page);
4630 * We need to make sure nobody is changing page->private, as we rely on
4631 * page->private as the pointer to extent buffer.
4633 spin_lock(&page->mapping->private_lock);
4634 if (!PagePrivate(page)) {
4635 spin_unlock(&page->mapping->private_lock);
4639 eb = (struct extent_buffer *)page->private;
4643 * This is a little awful but should be ok, we need to make sure that
4644 * the eb doesn't disappear out from under us while we're looking at
4647 spin_lock(&eb->refs_lock);
4648 if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) {
4649 spin_unlock(&eb->refs_lock);
4650 spin_unlock(&page->mapping->private_lock);
4653 spin_unlock(&page->mapping->private_lock);
4656 * If tree ref isn't set then we know the ref on this eb is a real ref,
4657 * so just return, this page will likely be freed soon anyway.
4659 if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) {
4660 spin_unlock(&eb->refs_lock);
4664 return release_extent_buffer(eb);
4668 * btrfs_readahead_tree_block - attempt to readahead a child block
4669 * @fs_info: the fs_info
4670 * @bytenr: bytenr to read
4671 * @owner_root: objectid of the root that owns this eb
4672 * @gen: generation for the uptodate check, can be 0
4673 * @level: level for the eb
4675 * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a
4676 * normal uptodate check of the eb, without checking the generation. If we have
4677 * to read the block we will not block on anything.
4679 void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info,
4680 u64 bytenr, u64 owner_root, u64 gen, int level)
4682 struct btrfs_tree_parent_check check = {
4687 struct extent_buffer *eb;
4690 eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level);
4694 if (btrfs_buffer_uptodate(eb, gen, 1)) {
4695 free_extent_buffer(eb);
4699 ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check);
4701 free_extent_buffer_stale(eb);
4703 free_extent_buffer(eb);
4707 * btrfs_readahead_node_child - readahead a node's child block
4708 * @node: parent node we're reading from
4709 * @slot: slot in the parent node for the child we want to read
4711 * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at
4712 * the slot in the node provided.
4714 void btrfs_readahead_node_child(struct extent_buffer *node, int slot)
4716 btrfs_readahead_tree_block(node->fs_info,
4717 btrfs_node_blockptr(node, slot),
4718 btrfs_header_owner(node),
4719 btrfs_node_ptr_generation(node, slot),
4720 btrfs_header_level(node) - 1);