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(inode, 0);
37 ret = btrfs_update_inode(trans, root, 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 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;
63 ASSERT(IS_ALIGNED(file_offset, block_size));
66 * We have flushed and locked the ranges of the source and destination
67 * inodes, we also have locked the inodes, so we are safe to do a
68 * reservation here. Also we must not do the reservation while holding
69 * a transaction open, otherwise we would deadlock.
71 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, file_offset,
76 page = find_or_create_page(inode->i_mapping, file_offset >> PAGE_SHIFT,
77 btrfs_alloc_write_mask(inode->i_mapping));
83 set_page_extent_mapped(page);
84 clear_extent_bit(&BTRFS_I(inode)->io_tree, file_offset, range_end,
85 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
87 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL);
91 if (comp_type == BTRFS_COMPRESS_NONE) {
95 memcpy(map, data_start, datal);
96 flush_dcache_page(page);
99 ret = btrfs_decompress(comp_type, data_start, page, 0,
103 flush_dcache_page(page);
107 * If our inline data is smaller then the block/page size, then the
108 * remaining of the block/page is equivalent to zeroes. We had something
109 * like the following done:
111 * $ xfs_io -f -c "pwrite -S 0xab 0 500" file
112 * $ sync # (or fsync)
113 * $ xfs_io -c "falloc 0 4K" file
114 * $ xfs_io -c "pwrite -S 0xcd 4K 4K"
116 * So what's in the range [500, 4095] corresponds to zeroes.
118 if (datal < block_size) {
122 memset(map + datal, 0, block_size - datal);
123 flush_dcache_page(page);
127 SetPageUptodate(page);
128 ClearPageChecked(page);
129 set_page_dirty(page);
136 btrfs_delalloc_release_space(inode, data_reserved, file_offset,
138 btrfs_delalloc_release_extents(BTRFS_I(inode), block_size);
140 extent_changeset_free(data_reserved);
146 * Deal with cloning of inline extents. We try to copy the inline extent from
147 * the source inode to destination inode when possible. When not possible we
148 * copy the inline extent's data into the respective page of the inode.
150 static int clone_copy_inline_extent(struct inode *dst,
151 struct btrfs_path *path,
152 struct btrfs_key *new_key,
153 const u64 drop_start,
158 struct btrfs_trans_handle **trans_out)
160 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
161 struct btrfs_root *root = BTRFS_I(dst)->root;
162 const u64 aligned_end = ALIGN(new_key->offset + datal,
163 fs_info->sectorsize);
164 struct btrfs_trans_handle *trans = NULL;
166 struct btrfs_key key;
168 if (new_key->offset > 0) {
169 ret = copy_inline_to_page(dst, new_key->offset, inline_data,
170 size, datal, comp_type);
174 key.objectid = btrfs_ino(BTRFS_I(dst));
175 key.type = BTRFS_EXTENT_DATA_KEY;
177 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
180 } else if (ret > 0) {
181 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
182 ret = btrfs_next_leaf(root, path);
186 goto copy_inline_extent;
188 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
189 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
190 key.type == BTRFS_EXTENT_DATA_KEY) {
192 * There's an implicit hole at file offset 0, copy the
193 * inline extent's data to the page.
195 ASSERT(key.offset > 0);
196 ret = copy_inline_to_page(dst, new_key->offset,
197 inline_data, size, datal,
201 } else if (i_size_read(dst) <= datal) {
202 struct btrfs_file_extent_item *ei;
204 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
205 struct btrfs_file_extent_item);
207 * If it's an inline extent replace it with the source inline
208 * extent, otherwise copy the source inline extent data into
209 * the respective page at the destination inode.
211 if (btrfs_file_extent_type(path->nodes[0], ei) ==
212 BTRFS_FILE_EXTENT_INLINE)
213 goto copy_inline_extent;
215 ret = copy_inline_to_page(dst, new_key->offset, inline_data,
216 size, datal, comp_type);
223 * We have no extent items, or we have an extent at offset 0 which may
224 * or may not be inlined. All these cases are dealt the same way.
226 if (i_size_read(dst) > datal) {
228 * At the destination offset 0 we have either a hole, a regular
229 * extent or an inline extent larger then the one we want to
230 * clone. Deal with all these cases by copying the inline extent
231 * data into the respective page at the destination inode.
233 ret = copy_inline_to_page(dst, new_key->offset, inline_data,
234 size, datal, comp_type);
238 btrfs_release_path(path);
240 * If we end up here it means were copy the inline extent into a leaf
241 * of the destination inode. We know we will drop or adjust at most one
242 * extent item in the destination root.
244 * 1 unit - adjusting old extent (we may have to split it)
245 * 1 unit - add new extent
246 * 1 unit - inode update
248 trans = btrfs_start_transaction(root, 3);
250 ret = PTR_ERR(trans);
254 ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1);
257 ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
261 write_extent_buffer(path->nodes[0], inline_data,
262 btrfs_item_ptr_offset(path->nodes[0],
265 inode_add_bytes(dst, datal);
266 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
267 ret = btrfs_inode_set_file_extent_range(BTRFS_I(dst), 0, aligned_end);
269 if (!ret && !trans) {
271 * No transaction here means we copied the inline extent into a
272 * page of the destination inode.
274 * 1 unit to update inode item
276 trans = btrfs_start_transaction(root, 1);
278 ret = PTR_ERR(trans);
283 btrfs_abort_transaction(trans, ret);
284 btrfs_end_transaction(trans);
293 * btrfs_clone() - clone a range from inode file to another
295 * @src: Inode to clone from
296 * @inode: Inode to clone to
297 * @off: Offset within source to start clone from
298 * @olen: Original length, passed by user, of range to clone
299 * @olen_aligned: Block-aligned value of olen
300 * @destoff: Offset within @inode to start clone
301 * @no_time_update: Whether to update mtime/ctime on the target inode
303 static int btrfs_clone(struct inode *src, struct inode *inode,
304 const u64 off, const u64 olen, const u64 olen_aligned,
305 const u64 destoff, int no_time_update)
307 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
308 struct btrfs_path *path = NULL;
309 struct extent_buffer *leaf;
310 struct btrfs_trans_handle *trans;
312 struct btrfs_key key;
316 const u64 len = olen_aligned;
317 u64 last_dest_end = destoff;
320 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
324 path = btrfs_alloc_path();
330 path->reada = READA_FORWARD;
332 key.objectid = btrfs_ino(BTRFS_I(src));
333 key.type = BTRFS_EXTENT_DATA_KEY;
337 u64 next_key_min_offset = key.offset + 1;
338 struct btrfs_file_extent_item *extent;
341 struct btrfs_key new_key;
342 u64 disko = 0, diskl = 0;
343 u64 datao = 0, datal = 0;
347 /* Note the key will change type as we walk through the tree */
348 path->leave_spinning = 1;
349 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
354 * First search, if no extent item that starts at offset off was
355 * found but the previous item is an extent item, it's possible
356 * it might overlap our target range, therefore process it.
358 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
359 btrfs_item_key_to_cpu(path->nodes[0], &key,
361 if (key.type == BTRFS_EXTENT_DATA_KEY)
365 nritems = btrfs_header_nritems(path->nodes[0]);
367 if (path->slots[0] >= nritems) {
368 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
373 nritems = btrfs_header_nritems(path->nodes[0]);
375 leaf = path->nodes[0];
376 slot = path->slots[0];
378 btrfs_item_key_to_cpu(leaf, &key, slot);
379 if (key.type > BTRFS_EXTENT_DATA_KEY ||
380 key.objectid != btrfs_ino(BTRFS_I(src)))
383 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
385 extent = btrfs_item_ptr(leaf, slot,
386 struct btrfs_file_extent_item);
387 comp = btrfs_file_extent_compression(leaf, extent);
388 type = btrfs_file_extent_type(leaf, extent);
389 if (type == BTRFS_FILE_EXTENT_REG ||
390 type == BTRFS_FILE_EXTENT_PREALLOC) {
391 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
392 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
393 datao = btrfs_file_extent_offset(leaf, extent);
394 datal = btrfs_file_extent_num_bytes(leaf, extent);
395 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
396 /* Take upper bound, may be compressed */
397 datal = btrfs_file_extent_ram_bytes(leaf, extent);
401 * The first search might have left us at an extent item that
402 * ends before our target range's start, can happen if we have
403 * holes and NO_HOLES feature enabled.
405 if (key.offset + datal <= off) {
408 } else if (key.offset >= off + len) {
411 next_key_min_offset = key.offset + datal;
412 size = btrfs_item_size_nr(leaf, slot);
413 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
416 btrfs_release_path(path);
417 path->leave_spinning = 0;
419 memcpy(&new_key, &key, sizeof(new_key));
420 new_key.objectid = btrfs_ino(BTRFS_I(inode));
421 if (off <= key.offset)
422 new_key.offset = key.offset + destoff - off;
424 new_key.offset = destoff;
427 * Deal with a hole that doesn't have an extent item that
428 * represents it (NO_HOLES feature enabled).
429 * This hole is either in the middle of the cloning range or at
430 * the beginning (fully overlaps it or partially overlaps it).
432 if (new_key.offset != last_dest_end)
433 drop_start = last_dest_end;
435 drop_start = new_key.offset;
437 if (type == BTRFS_FILE_EXTENT_REG ||
438 type == BTRFS_FILE_EXTENT_PREALLOC) {
439 struct btrfs_clone_extent_info clone_info;
442 * a | --- range to clone ---| b
443 * | ------------- extent ------------- |
446 /* Subtract range b */
447 if (key.offset + datal > off + len)
448 datal = off + len - key.offset;
450 /* Subtract range a */
451 if (off > key.offset) {
452 datao += off - key.offset;
453 datal -= off - key.offset;
456 clone_info.disk_offset = disko;
457 clone_info.disk_len = diskl;
458 clone_info.data_offset = datao;
459 clone_info.data_len = datal;
460 clone_info.file_offset = new_key.offset;
461 clone_info.extent_buf = buf;
462 clone_info.item_size = size;
463 ret = btrfs_punch_hole_range(inode, path, drop_start,
464 new_key.offset + datal - 1, &clone_info,
468 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
470 * Inline extents always have to start at file offset 0
471 * and can never be bigger then the sector size. We can
472 * never clone only parts of an inline extent, since all
473 * reflink operations must start at a sector size aligned
474 * offset, and the length must be aligned too or end at
475 * the i_size (which implies the whole inlined data).
477 ASSERT(key.offset == 0);
478 ASSERT(datal <= fs_info->sectorsize);
479 if (key.offset != 0 || datal > fs_info->sectorsize)
482 ret = clone_copy_inline_extent(inode, path, &new_key,
483 drop_start, datal, size,
489 btrfs_release_path(path);
491 last_dest_end = ALIGN(new_key.offset + datal,
492 fs_info->sectorsize);
493 ret = clone_finish_inode_update(trans, inode, last_dest_end,
494 destoff, olen, no_time_update);
497 if (new_key.offset + datal >= destoff + len)
500 btrfs_release_path(path);
501 key.offset = next_key_min_offset;
503 if (fatal_signal_pending(current)) {
510 if (last_dest_end < destoff + len) {
512 * We have an implicit hole that fully or partially overlaps our
513 * cloning range at its end. This means that we either have the
514 * NO_HOLES feature enabled or the implicit hole happened due to
515 * mixing buffered and direct IO writes against this file.
517 btrfs_release_path(path);
518 path->leave_spinning = 0;
520 ret = btrfs_punch_hole_range(inode, path, last_dest_end,
521 destoff + len - 1, NULL, &trans);
525 ret = clone_finish_inode_update(trans, inode, destoff + len,
526 destoff, olen, no_time_update);
530 btrfs_free_path(path);
535 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
536 struct inode *inode2, u64 loff2, u64 len)
538 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
539 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
542 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
543 struct inode *inode2, u64 loff2, u64 len)
545 if (inode1 < inode2) {
546 swap(inode1, inode2);
548 } else if (inode1 == inode2 && loff2 < loff1) {
551 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
552 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
555 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
556 struct inode *dst, u64 dst_loff)
558 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
562 * Lock destination range to serialize with concurrent readpages() and
563 * source range to serialize with relocation.
565 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
566 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
567 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
572 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
573 struct inode *dst, u64 dst_loff)
576 u64 i, tail_len, chunk_count;
577 struct btrfs_root *root_dst = BTRFS_I(dst)->root;
579 spin_lock(&root_dst->root_item_lock);
580 if (root_dst->send_in_progress) {
581 btrfs_warn_rl(root_dst->fs_info,
582 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
583 root_dst->root_key.objectid,
584 root_dst->send_in_progress);
585 spin_unlock(&root_dst->root_item_lock);
588 root_dst->dedupe_in_progress++;
589 spin_unlock(&root_dst->root_item_lock);
591 tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
592 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
594 for (i = 0; i < chunk_count; i++) {
595 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
600 loff += BTRFS_MAX_DEDUPE_LEN;
601 dst_loff += BTRFS_MAX_DEDUPE_LEN;
605 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff);
607 spin_lock(&root_dst->root_item_lock);
608 root_dst->dedupe_in_progress--;
609 spin_unlock(&root_dst->root_item_lock);
614 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
615 u64 off, u64 olen, u64 destoff)
617 struct inode *inode = file_inode(file);
618 struct inode *src = file_inode(file_src);
619 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
623 u64 bs = fs_info->sb->s_blocksize;
626 * VFS's generic_remap_file_range_prep() protects us from cloning the
627 * eof block into the middle of a file, which would result in corruption
628 * if the file size is not blocksize aligned. So we don't need to check
629 * for that case here.
631 if (off + len == src->i_size)
632 len = ALIGN(src->i_size, bs) - off;
634 if (destoff > inode->i_size) {
635 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
637 ret = btrfs_cont_expand(inode, inode->i_size, destoff);
641 * We may have truncated the last block if the inode's size is
642 * not sector size aligned, so we need to wait for writeback to
643 * complete before proceeding further, otherwise we can race
644 * with cloning and attempt to increment a reference to an
645 * extent that no longer exists (writeback completed right after
646 * we found the previous extent covering eof and before we
647 * attempted to increment its reference count).
649 ret = btrfs_wait_ordered_range(inode, wb_start,
656 * Lock destination range to serialize with concurrent readpages() and
657 * source range to serialize with relocation.
659 btrfs_double_extent_lock(src, off, inode, destoff, len);
660 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
661 btrfs_double_extent_unlock(src, off, inode, destoff, len);
664 * We may have copied an inline extent into a page of the destination
665 * range, so wait for writeback to complete before truncating pages
666 * from the page cache. This is a rare case.
668 wb_ret = btrfs_wait_ordered_range(inode, destoff, len);
669 ret = ret ? ret : wb_ret;
671 * Truncate page cache pages so that future reads will see the cloned
672 * data immediately and not the previous data.
674 truncate_inode_pages_range(&inode->i_data,
675 round_down(destoff, PAGE_SIZE),
676 round_up(destoff + len, PAGE_SIZE) - 1);
681 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
682 struct file *file_out, loff_t pos_out,
683 loff_t *len, unsigned int remap_flags)
685 struct inode *inode_in = file_inode(file_in);
686 struct inode *inode_out = file_inode(file_out);
687 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
688 bool same_inode = inode_out == inode_in;
692 if (!(remap_flags & REMAP_FILE_DEDUP)) {
693 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
695 if (btrfs_root_readonly(root_out))
698 if (file_in->f_path.mnt != file_out->f_path.mnt ||
699 inode_in->i_sb != inode_out->i_sb)
703 /* Don't make the dst file partly checksummed */
704 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
705 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
710 * Now that the inodes are locked, we need to start writeback ourselves
711 * and can not rely on the writeback from the VFS's generic helper
712 * generic_remap_file_range_prep() because:
714 * 1) For compression we must call filemap_fdatawrite_range() range
715 * twice (btrfs_fdatawrite_range() does it for us), and the generic
716 * helper only calls it once;
718 * 2) filemap_fdatawrite_range(), called by the generic helper only
719 * waits for the writeback to complete, i.e. for IO to be done, and
720 * not for the ordered extents to complete. We need to wait for them
721 * to complete so that new file extent items are in the fs tree.
723 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
724 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
726 wb_len = ALIGN(*len, bs);
729 * Since we don't lock ranges, wait for ongoing lockless dio writes (as
730 * any in progress could create its ordered extents after we wait for
731 * existing ordered extents below).
733 inode_dio_wait(inode_in);
735 inode_dio_wait(inode_out);
738 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
740 * Btrfs' back references do not have a block level granularity, they
741 * work at the whole extent level.
742 * NOCOW buffered write without data space reserved may not be able
743 * to fall back to CoW due to lack of data space, thus could cause
746 * Here we take a shortcut by flushing the whole inode, so that all
747 * nocow write should reach disk as nocow before we increase the
748 * reference of the extent. We could do better by only flushing NOCOW
749 * data, but that needs extra accounting.
751 * Also we don't need to check ASYNC_EXTENT, as async extent will be
752 * CoWed anyway, not affecting nocow part.
754 ret = filemap_flush(inode_in->i_mapping);
758 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
762 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
767 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
771 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
772 struct file *dst_file, loff_t destoff, loff_t len,
773 unsigned int remap_flags)
775 struct inode *src_inode = file_inode(src_file);
776 struct inode *dst_inode = file_inode(dst_file);
777 bool same_inode = dst_inode == src_inode;
780 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
784 inode_lock(src_inode);
786 lock_two_nondirectories(src_inode, dst_inode);
788 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
790 if (ret < 0 || len == 0)
793 if (remap_flags & REMAP_FILE_DEDUP)
794 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
796 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
800 inode_unlock(src_inode);
802 unlock_two_nondirectories(src_inode, dst_inode);
804 return ret < 0 ? ret : len;