1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/blkdev.h>
4 #include <linux/iversion.h>
5 #include "compression.h"
7 #include "delalloc-space.h"
9 #include "transaction.h"
11 #define BTRFS_MAX_DEDUPE_LEN SZ_16M
13 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
20 struct btrfs_root *root = BTRFS_I(inode)->root;
23 inode_inc_iversion(inode);
25 inode->i_mtime = inode->i_ctime = current_time(inode);
27 * We round up to the block size at eof when determining which
28 * extents to clone above, but shouldn't round up the file size.
30 if (endoff > destoff + olen)
31 endoff = destoff + olen;
32 if (endoff > inode->i_size) {
33 i_size_write(inode, endoff);
34 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
37 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
39 btrfs_abort_transaction(trans, ret);
40 btrfs_end_transaction(trans);
43 ret = btrfs_end_transaction(trans);
48 static int copy_inline_to_page(struct btrfs_inode *inode,
49 const u64 file_offset,
55 const u64 block_size = btrfs_inode_sectorsize(inode);
56 const u64 range_end = file_offset + block_size - 1;
57 const size_t inline_size = size - btrfs_file_extent_calc_inline_size(0);
58 char *data_start = inline_data + btrfs_file_extent_calc_inline_size(0);
59 struct extent_changeset *data_reserved = NULL;
60 struct page *page = NULL;
61 struct address_space *mapping = inode->vfs_inode.i_mapping;
64 ASSERT(IS_ALIGNED(file_offset, block_size));
67 * We have flushed and locked the ranges of the source and destination
68 * inodes, we also have locked the inodes, so we are safe to do a
69 * reservation here. Also we must not do the reservation while holding
70 * a transaction open, otherwise we would deadlock.
72 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, file_offset,
77 page = find_or_create_page(mapping, file_offset >> PAGE_SHIFT,
78 btrfs_alloc_write_mask(mapping));
84 ret = set_page_extent_mapped(page);
88 clear_extent_bit(&inode->io_tree, file_offset, range_end,
89 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
91 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL);
96 * After dirtying the page our caller will need to start a transaction,
97 * and if we are low on metadata free space, that can cause flushing of
98 * delalloc for all inodes in order to get metadata space released.
99 * However we are holding the range locked for the whole duration of
100 * the clone/dedupe operation, so we may deadlock if that happens and no
101 * other task releases enough space. So mark this inode as not being
102 * possible to flush to avoid such deadlock. We will clear that flag
103 * when we finish cloning all extents, since a transaction is started
104 * after finding each extent to clone.
106 set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &inode->runtime_flags);
108 if (comp_type == BTRFS_COMPRESS_NONE) {
109 memcpy_to_page(page, 0, data_start, datal);
110 flush_dcache_page(page);
112 ret = btrfs_decompress(comp_type, data_start, page, 0,
116 flush_dcache_page(page);
120 * If our inline data is smaller then the block/page size, then the
121 * remaining of the block/page is equivalent to zeroes. We had something
122 * like the following done:
124 * $ xfs_io -f -c "pwrite -S 0xab 0 500" file
125 * $ sync # (or fsync)
126 * $ xfs_io -c "falloc 0 4K" file
127 * $ xfs_io -c "pwrite -S 0xcd 4K 4K"
129 * So what's in the range [500, 4095] corresponds to zeroes.
131 if (datal < block_size) {
132 memzero_page(page, datal, block_size - datal);
133 flush_dcache_page(page);
136 SetPageUptodate(page);
137 ClearPageChecked(page);
138 set_page_dirty(page);
145 btrfs_delalloc_release_space(inode, data_reserved, file_offset,
147 btrfs_delalloc_release_extents(inode, block_size);
149 extent_changeset_free(data_reserved);
155 * Deal with cloning of inline extents. We try to copy the inline extent from
156 * the source inode to destination inode when possible. When not possible we
157 * copy the inline extent's data into the respective page of the inode.
159 static int clone_copy_inline_extent(struct inode *dst,
160 struct btrfs_path *path,
161 struct btrfs_key *new_key,
162 const u64 drop_start,
167 struct btrfs_trans_handle **trans_out)
169 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
170 struct btrfs_root *root = BTRFS_I(dst)->root;
171 const u64 aligned_end = ALIGN(new_key->offset + datal,
172 fs_info->sectorsize);
173 struct btrfs_trans_handle *trans = NULL;
174 struct btrfs_drop_extents_args drop_args = { 0 };
176 struct btrfs_key key;
178 if (new_key->offset > 0) {
179 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
180 inline_data, size, datal, comp_type);
184 key.objectid = btrfs_ino(BTRFS_I(dst));
185 key.type = BTRFS_EXTENT_DATA_KEY;
187 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
190 } else if (ret > 0) {
191 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
192 ret = btrfs_next_leaf(root, path);
196 goto copy_inline_extent;
198 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
199 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
200 key.type == BTRFS_EXTENT_DATA_KEY) {
202 * There's an implicit hole at file offset 0, copy the
203 * inline extent's data to the page.
205 ASSERT(key.offset > 0);
206 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
207 inline_data, size, datal,
211 } else if (i_size_read(dst) <= datal) {
212 struct btrfs_file_extent_item *ei;
214 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
215 struct btrfs_file_extent_item);
217 * If it's an inline extent replace it with the source inline
218 * extent, otherwise copy the source inline extent data into
219 * the respective page at the destination inode.
221 if (btrfs_file_extent_type(path->nodes[0], ei) ==
222 BTRFS_FILE_EXTENT_INLINE)
223 goto copy_inline_extent;
225 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
226 inline_data, size, datal, comp_type);
233 * We have no extent items, or we have an extent at offset 0 which may
234 * or may not be inlined. All these cases are dealt the same way.
236 if (i_size_read(dst) > datal) {
238 * At the destination offset 0 we have either a hole, a regular
239 * extent or an inline extent larger then the one we want to
240 * clone. Deal with all these cases by copying the inline extent
241 * data into the respective page at the destination inode.
243 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
244 inline_data, size, datal, comp_type);
248 btrfs_release_path(path);
250 * If we end up here it means were copy the inline extent into a leaf
251 * of the destination inode. We know we will drop or adjust at most one
252 * extent item in the destination root.
254 * 1 unit - adjusting old extent (we may have to split it)
255 * 1 unit - add new extent
256 * 1 unit - inode update
258 trans = btrfs_start_transaction(root, 3);
260 ret = PTR_ERR(trans);
264 drop_args.path = path;
265 drop_args.start = drop_start;
266 drop_args.end = aligned_end;
267 drop_args.drop_cache = true;
268 ret = btrfs_drop_extents(trans, root, BTRFS_I(dst), &drop_args);
271 ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
275 write_extent_buffer(path->nodes[0], inline_data,
276 btrfs_item_ptr_offset(path->nodes[0],
279 btrfs_update_inode_bytes(BTRFS_I(dst), datal, drop_args.bytes_found);
280 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
281 ret = btrfs_inode_set_file_extent_range(BTRFS_I(dst), 0, aligned_end);
283 if (!ret && !trans) {
285 * Release path before starting a new transaction so we don't
286 * hold locks that would confuse lockdep.
288 btrfs_release_path(path);
290 * No transaction here means we copied the inline extent into a
291 * page of the destination inode.
293 * 1 unit to update inode item
295 trans = btrfs_start_transaction(root, 1);
297 ret = PTR_ERR(trans);
302 btrfs_abort_transaction(trans, ret);
303 btrfs_end_transaction(trans);
312 * btrfs_clone() - clone a range from inode file to another
314 * @src: Inode to clone from
315 * @inode: Inode to clone to
316 * @off: Offset within source to start clone from
317 * @olen: Original length, passed by user, of range to clone
318 * @olen_aligned: Block-aligned value of olen
319 * @destoff: Offset within @inode to start clone
320 * @no_time_update: Whether to update mtime/ctime on the target inode
322 static int btrfs_clone(struct inode *src, struct inode *inode,
323 const u64 off, const u64 olen, const u64 olen_aligned,
324 const u64 destoff, int no_time_update)
326 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
327 struct btrfs_path *path = NULL;
328 struct extent_buffer *leaf;
329 struct btrfs_trans_handle *trans;
331 struct btrfs_key key;
335 const u64 len = olen_aligned;
336 u64 last_dest_end = destoff;
339 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
343 path = btrfs_alloc_path();
349 path->reada = READA_FORWARD;
351 key.objectid = btrfs_ino(BTRFS_I(src));
352 key.type = BTRFS_EXTENT_DATA_KEY;
356 u64 next_key_min_offset = key.offset + 1;
357 struct btrfs_file_extent_item *extent;
361 struct btrfs_key new_key;
362 u64 disko = 0, diskl = 0;
363 u64 datao = 0, datal = 0;
367 /* Note the key will change type as we walk through the tree */
368 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
373 * First search, if no extent item that starts at offset off was
374 * found but the previous item is an extent item, it's possible
375 * it might overlap our target range, therefore process it.
377 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
378 btrfs_item_key_to_cpu(path->nodes[0], &key,
380 if (key.type == BTRFS_EXTENT_DATA_KEY)
384 nritems = btrfs_header_nritems(path->nodes[0]);
386 if (path->slots[0] >= nritems) {
387 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
392 nritems = btrfs_header_nritems(path->nodes[0]);
394 leaf = path->nodes[0];
395 slot = path->slots[0];
397 btrfs_item_key_to_cpu(leaf, &key, slot);
398 if (key.type > BTRFS_EXTENT_DATA_KEY ||
399 key.objectid != btrfs_ino(BTRFS_I(src)))
402 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
404 extent = btrfs_item_ptr(leaf, slot,
405 struct btrfs_file_extent_item);
406 extent_gen = btrfs_file_extent_generation(leaf, extent);
407 comp = btrfs_file_extent_compression(leaf, extent);
408 type = btrfs_file_extent_type(leaf, extent);
409 if (type == BTRFS_FILE_EXTENT_REG ||
410 type == BTRFS_FILE_EXTENT_PREALLOC) {
411 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
412 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
413 datao = btrfs_file_extent_offset(leaf, extent);
414 datal = btrfs_file_extent_num_bytes(leaf, extent);
415 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
416 /* Take upper bound, may be compressed */
417 datal = btrfs_file_extent_ram_bytes(leaf, extent);
421 * The first search might have left us at an extent item that
422 * ends before our target range's start, can happen if we have
423 * holes and NO_HOLES feature enabled.
425 if (key.offset + datal <= off) {
428 } else if (key.offset >= off + len) {
431 next_key_min_offset = key.offset + datal;
432 size = btrfs_item_size_nr(leaf, slot);
433 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
436 btrfs_release_path(path);
438 memcpy(&new_key, &key, sizeof(new_key));
439 new_key.objectid = btrfs_ino(BTRFS_I(inode));
440 if (off <= key.offset)
441 new_key.offset = key.offset + destoff - off;
443 new_key.offset = destoff;
446 * Deal with a hole that doesn't have an extent item that
447 * represents it (NO_HOLES feature enabled).
448 * This hole is either in the middle of the cloning range or at
449 * the beginning (fully overlaps it or partially overlaps it).
451 if (new_key.offset != last_dest_end)
452 drop_start = last_dest_end;
454 drop_start = new_key.offset;
456 if (type == BTRFS_FILE_EXTENT_REG ||
457 type == BTRFS_FILE_EXTENT_PREALLOC) {
458 struct btrfs_replace_extent_info clone_info;
461 * a | --- range to clone ---| b
462 * | ------------- extent ------------- |
465 /* Subtract range b */
466 if (key.offset + datal > off + len)
467 datal = off + len - key.offset;
469 /* Subtract range a */
470 if (off > key.offset) {
471 datao += off - key.offset;
472 datal -= off - key.offset;
475 clone_info.disk_offset = disko;
476 clone_info.disk_len = diskl;
477 clone_info.data_offset = datao;
478 clone_info.data_len = datal;
479 clone_info.file_offset = new_key.offset;
480 clone_info.extent_buf = buf;
481 clone_info.is_new_extent = false;
482 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
483 drop_start, new_key.offset + datal - 1,
484 &clone_info, &trans);
487 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
489 * Inline extents always have to start at file offset 0
490 * and can never be bigger then the sector size. We can
491 * never clone only parts of an inline extent, since all
492 * reflink operations must start at a sector size aligned
493 * offset, and the length must be aligned too or end at
494 * the i_size (which implies the whole inlined data).
496 ASSERT(key.offset == 0);
497 ASSERT(datal <= fs_info->sectorsize);
498 if (key.offset != 0 || datal > fs_info->sectorsize)
501 ret = clone_copy_inline_extent(inode, path, &new_key,
502 drop_start, datal, size,
508 btrfs_release_path(path);
511 * If this is a new extent update the last_reflink_trans of both
512 * inodes. This is used by fsync to make sure it does not log
513 * multiple checksum items with overlapping ranges. For older
514 * extents we don't need to do it since inode logging skips the
515 * checksums for older extents. Also ignore holes and inline
516 * extents because they don't have checksums in the csum tree.
518 if (extent_gen == trans->transid && disko > 0) {
519 BTRFS_I(src)->last_reflink_trans = trans->transid;
520 BTRFS_I(inode)->last_reflink_trans = trans->transid;
523 last_dest_end = ALIGN(new_key.offset + datal,
524 fs_info->sectorsize);
525 ret = clone_finish_inode_update(trans, inode, last_dest_end,
526 destoff, olen, no_time_update);
529 if (new_key.offset + datal >= destoff + len)
532 btrfs_release_path(path);
533 key.offset = next_key_min_offset;
535 if (fatal_signal_pending(current)) {
544 if (last_dest_end < destoff + len) {
546 * We have an implicit hole that fully or partially overlaps our
547 * cloning range at its end. This means that we either have the
548 * NO_HOLES feature enabled or the implicit hole happened due to
549 * mixing buffered and direct IO writes against this file.
551 btrfs_release_path(path);
554 * When using NO_HOLES and we are cloning a range that covers
555 * only a hole (no extents) into a range beyond the current
556 * i_size, punching a hole in the target range will not create
557 * an extent map defining a hole, because the range starts at or
558 * beyond current i_size. If the file previously had an i_size
559 * greater than the new i_size set by this clone operation, we
560 * need to make sure the next fsync is a full fsync, so that it
561 * detects and logs a hole covering a range from the current
562 * i_size to the new i_size. If the clone range covers extents,
563 * besides a hole, then we know the full sync flag was already
564 * set by previous calls to btrfs_replace_file_extents() that
565 * replaced file extent items.
567 if (last_dest_end >= i_size_read(inode))
568 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
569 &BTRFS_I(inode)->runtime_flags);
571 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
572 last_dest_end, destoff + len - 1, NULL, &trans);
576 ret = clone_finish_inode_update(trans, inode, destoff + len,
577 destoff, olen, no_time_update);
581 btrfs_free_path(path);
583 clear_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &BTRFS_I(inode)->runtime_flags);
588 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
589 struct inode *inode2, u64 loff2, u64 len)
591 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
592 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
595 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
596 struct inode *inode2, u64 loff2, u64 len)
598 if (inode1 < inode2) {
599 swap(inode1, inode2);
601 } else if (inode1 == inode2 && loff2 < loff1) {
604 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
605 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
608 static void btrfs_double_mmap_lock(struct inode *inode1, struct inode *inode2)
611 swap(inode1, inode2);
612 down_write(&BTRFS_I(inode1)->i_mmap_lock);
613 down_write_nested(&BTRFS_I(inode2)->i_mmap_lock, SINGLE_DEPTH_NESTING);
616 static void btrfs_double_mmap_unlock(struct inode *inode1, struct inode *inode2)
618 up_write(&BTRFS_I(inode1)->i_mmap_lock);
619 up_write(&BTRFS_I(inode2)->i_mmap_lock);
622 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
623 struct inode *dst, u64 dst_loff)
625 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
629 * Lock destination range to serialize with concurrent readpages() and
630 * source range to serialize with relocation.
632 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
633 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
634 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
639 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
640 struct inode *dst, u64 dst_loff)
643 u64 i, tail_len, chunk_count;
644 struct btrfs_root *root_dst = BTRFS_I(dst)->root;
646 spin_lock(&root_dst->root_item_lock);
647 if (root_dst->send_in_progress) {
648 btrfs_warn_rl(root_dst->fs_info,
649 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
650 root_dst->root_key.objectid,
651 root_dst->send_in_progress);
652 spin_unlock(&root_dst->root_item_lock);
655 root_dst->dedupe_in_progress++;
656 spin_unlock(&root_dst->root_item_lock);
658 tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
659 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
661 for (i = 0; i < chunk_count; i++) {
662 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
667 loff += BTRFS_MAX_DEDUPE_LEN;
668 dst_loff += BTRFS_MAX_DEDUPE_LEN;
672 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff);
674 spin_lock(&root_dst->root_item_lock);
675 root_dst->dedupe_in_progress--;
676 spin_unlock(&root_dst->root_item_lock);
681 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
682 u64 off, u64 olen, u64 destoff)
684 struct inode *inode = file_inode(file);
685 struct inode *src = file_inode(file_src);
686 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
690 u64 bs = fs_info->sb->s_blocksize;
693 * VFS's generic_remap_file_range_prep() protects us from cloning the
694 * eof block into the middle of a file, which would result in corruption
695 * if the file size is not blocksize aligned. So we don't need to check
696 * for that case here.
698 if (off + len == src->i_size)
699 len = ALIGN(src->i_size, bs) - off;
701 if (destoff > inode->i_size) {
702 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
704 ret = btrfs_cont_expand(BTRFS_I(inode), inode->i_size, destoff);
708 * We may have truncated the last block if the inode's size is
709 * not sector size aligned, so we need to wait for writeback to
710 * complete before proceeding further, otherwise we can race
711 * with cloning and attempt to increment a reference to an
712 * extent that no longer exists (writeback completed right after
713 * we found the previous extent covering eof and before we
714 * attempted to increment its reference count).
716 ret = btrfs_wait_ordered_range(inode, wb_start,
723 * Lock destination range to serialize with concurrent readpages() and
724 * source range to serialize with relocation.
726 btrfs_double_extent_lock(src, off, inode, destoff, len);
727 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
728 btrfs_double_extent_unlock(src, off, inode, destoff, len);
731 * We may have copied an inline extent into a page of the destination
732 * range, so wait for writeback to complete before truncating pages
733 * from the page cache. This is a rare case.
735 wb_ret = btrfs_wait_ordered_range(inode, destoff, len);
736 ret = ret ? ret : wb_ret;
738 * Truncate page cache pages so that future reads will see the cloned
739 * data immediately and not the previous data.
741 truncate_inode_pages_range(&inode->i_data,
742 round_down(destoff, PAGE_SIZE),
743 round_up(destoff + len, PAGE_SIZE) - 1);
748 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
749 struct file *file_out, loff_t pos_out,
750 loff_t *len, unsigned int remap_flags)
752 struct inode *inode_in = file_inode(file_in);
753 struct inode *inode_out = file_inode(file_out);
754 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
755 bool same_inode = inode_out == inode_in;
759 if (!(remap_flags & REMAP_FILE_DEDUP)) {
760 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
762 if (btrfs_root_readonly(root_out))
765 if (file_in->f_path.mnt != file_out->f_path.mnt ||
766 inode_in->i_sb != inode_out->i_sb)
770 /* Don't make the dst file partly checksummed */
771 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
772 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
777 * Now that the inodes are locked, we need to start writeback ourselves
778 * and can not rely on the writeback from the VFS's generic helper
779 * generic_remap_file_range_prep() because:
781 * 1) For compression we must call filemap_fdatawrite_range() range
782 * twice (btrfs_fdatawrite_range() does it for us), and the generic
783 * helper only calls it once;
785 * 2) filemap_fdatawrite_range(), called by the generic helper only
786 * waits for the writeback to complete, i.e. for IO to be done, and
787 * not for the ordered extents to complete. We need to wait for them
788 * to complete so that new file extent items are in the fs tree.
790 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
791 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
793 wb_len = ALIGN(*len, bs);
796 * Since we don't lock ranges, wait for ongoing lockless dio writes (as
797 * any in progress could create its ordered extents after we wait for
798 * existing ordered extents below).
800 inode_dio_wait(inode_in);
802 inode_dio_wait(inode_out);
805 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
807 * Btrfs' back references do not have a block level granularity, they
808 * work at the whole extent level.
809 * NOCOW buffered write without data space reserved may not be able
810 * to fall back to CoW due to lack of data space, thus could cause
813 * Here we take a shortcut by flushing the whole inode, so that all
814 * nocow write should reach disk as nocow before we increase the
815 * reference of the extent. We could do better by only flushing NOCOW
816 * data, but that needs extra accounting.
818 * Also we don't need to check ASYNC_EXTENT, as async extent will be
819 * CoWed anyway, not affecting nocow part.
821 ret = filemap_flush(inode_in->i_mapping);
825 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
829 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
834 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
838 static bool file_sync_write(const struct file *file)
840 if (file->f_flags & (__O_SYNC | O_DSYNC))
842 if (IS_SYNC(file_inode(file)))
848 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
849 struct file *dst_file, loff_t destoff, loff_t len,
850 unsigned int remap_flags)
852 struct inode *src_inode = file_inode(src_file);
853 struct inode *dst_inode = file_inode(dst_file);
854 bool same_inode = dst_inode == src_inode;
857 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
861 btrfs_inode_lock(src_inode, BTRFS_ILOCK_MMAP);
863 lock_two_nondirectories(src_inode, dst_inode);
864 btrfs_double_mmap_lock(src_inode, dst_inode);
867 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
869 if (ret < 0 || len == 0)
872 if (remap_flags & REMAP_FILE_DEDUP)
873 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
875 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
879 btrfs_inode_unlock(src_inode, BTRFS_ILOCK_MMAP);
881 btrfs_double_mmap_unlock(src_inode, dst_inode);
882 unlock_two_nondirectories(src_inode, dst_inode);
886 * If either the source or the destination file was opened with O_SYNC,
887 * O_DSYNC or has the S_SYNC attribute, fsync both the destination and
888 * source files/ranges, so that after a successful return (0) followed
889 * by a power failure results in the reflinked data to be readable from
892 if (ret == 0 && len > 0 &&
893 (file_sync_write(src_file) || file_sync_write(dst_file))) {
894 ret = btrfs_sync_file(src_file, off, off + len - 1, 0);
896 ret = btrfs_sync_file(dst_file, destoff,
897 destoff + len - 1, 0);
900 return ret < 0 ? ret : len;