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 set_page_extent_mapped(page);
85 clear_extent_bit(&inode->io_tree, file_offset, range_end,
86 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
88 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL);
92 if (comp_type == BTRFS_COMPRESS_NONE) {
96 memcpy(map, data_start, datal);
97 flush_dcache_page(page);
100 ret = btrfs_decompress(comp_type, data_start, page, 0,
104 flush_dcache_page(page);
108 * If our inline data is smaller then the block/page size, then the
109 * remaining of the block/page is equivalent to zeroes. We had something
110 * like the following done:
112 * $ xfs_io -f -c "pwrite -S 0xab 0 500" file
113 * $ sync # (or fsync)
114 * $ xfs_io -c "falloc 0 4K" file
115 * $ xfs_io -c "pwrite -S 0xcd 4K 4K"
117 * So what's in the range [500, 4095] corresponds to zeroes.
119 if (datal < block_size) {
123 memset(map + datal, 0, block_size - datal);
124 flush_dcache_page(page);
128 SetPageUptodate(page);
129 ClearPageChecked(page);
130 set_page_dirty(page);
137 btrfs_delalloc_release_space(inode, data_reserved, file_offset,
139 btrfs_delalloc_release_extents(inode, block_size);
141 extent_changeset_free(data_reserved);
147 * Deal with cloning of inline extents. We try to copy the inline extent from
148 * the source inode to destination inode when possible. When not possible we
149 * copy the inline extent's data into the respective page of the inode.
151 static int clone_copy_inline_extent(struct inode *dst,
152 struct btrfs_path *path,
153 struct btrfs_key *new_key,
154 const u64 drop_start,
159 struct btrfs_trans_handle **trans_out)
161 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
162 struct btrfs_root *root = BTRFS_I(dst)->root;
163 const u64 aligned_end = ALIGN(new_key->offset + datal,
164 fs_info->sectorsize);
165 struct btrfs_trans_handle *trans = NULL;
166 struct btrfs_drop_extents_args drop_args = { 0 };
168 struct btrfs_key key;
170 if (new_key->offset > 0) {
171 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
172 inline_data, size, datal, comp_type);
176 key.objectid = btrfs_ino(BTRFS_I(dst));
177 key.type = BTRFS_EXTENT_DATA_KEY;
179 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
182 } else if (ret > 0) {
183 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
184 ret = btrfs_next_leaf(root, path);
188 goto copy_inline_extent;
190 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
191 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
192 key.type == BTRFS_EXTENT_DATA_KEY) {
194 * There's an implicit hole at file offset 0, copy the
195 * inline extent's data to the page.
197 ASSERT(key.offset > 0);
198 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
199 inline_data, size, datal,
203 } else if (i_size_read(dst) <= datal) {
204 struct btrfs_file_extent_item *ei;
206 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
207 struct btrfs_file_extent_item);
209 * If it's an inline extent replace it with the source inline
210 * extent, otherwise copy the source inline extent data into
211 * the respective page at the destination inode.
213 if (btrfs_file_extent_type(path->nodes[0], ei) ==
214 BTRFS_FILE_EXTENT_INLINE)
215 goto copy_inline_extent;
217 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
218 inline_data, size, datal, comp_type);
225 * We have no extent items, or we have an extent at offset 0 which may
226 * or may not be inlined. All these cases are dealt the same way.
228 if (i_size_read(dst) > datal) {
230 * At the destination offset 0 we have either a hole, a regular
231 * extent or an inline extent larger then the one we want to
232 * clone. Deal with all these cases by copying the inline extent
233 * data into the respective page at the destination inode.
235 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
236 inline_data, size, datal, comp_type);
240 btrfs_release_path(path);
242 * If we end up here it means were copy the inline extent into a leaf
243 * of the destination inode. We know we will drop or adjust at most one
244 * extent item in the destination root.
246 * 1 unit - adjusting old extent (we may have to split it)
247 * 1 unit - add new extent
248 * 1 unit - inode update
250 trans = btrfs_start_transaction(root, 3);
252 ret = PTR_ERR(trans);
256 drop_args.path = path;
257 drop_args.start = drop_start;
258 drop_args.end = aligned_end;
259 drop_args.drop_cache = true;
260 ret = btrfs_drop_extents(trans, root, BTRFS_I(dst), &drop_args);
263 ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
267 write_extent_buffer(path->nodes[0], inline_data,
268 btrfs_item_ptr_offset(path->nodes[0],
271 btrfs_update_inode_bytes(BTRFS_I(dst), datal, drop_args.bytes_found);
272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags);
273 ret = btrfs_inode_set_file_extent_range(BTRFS_I(dst), 0, aligned_end);
275 if (!ret && !trans) {
277 * No transaction here means we copied the inline extent into a
278 * page of the destination inode.
280 * 1 unit to update inode item
282 trans = btrfs_start_transaction(root, 1);
284 ret = PTR_ERR(trans);
289 btrfs_abort_transaction(trans, ret);
290 btrfs_end_transaction(trans);
299 * btrfs_clone() - clone a range from inode file to another
301 * @src: Inode to clone from
302 * @inode: Inode to clone to
303 * @off: Offset within source to start clone from
304 * @olen: Original length, passed by user, of range to clone
305 * @olen_aligned: Block-aligned value of olen
306 * @destoff: Offset within @inode to start clone
307 * @no_time_update: Whether to update mtime/ctime on the target inode
309 static int btrfs_clone(struct inode *src, struct inode *inode,
310 const u64 off, const u64 olen, const u64 olen_aligned,
311 const u64 destoff, int no_time_update)
313 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
314 struct btrfs_path *path = NULL;
315 struct extent_buffer *leaf;
316 struct btrfs_trans_handle *trans;
318 struct btrfs_key key;
322 const u64 len = olen_aligned;
323 u64 last_dest_end = destoff;
326 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
330 path = btrfs_alloc_path();
336 path->reada = READA_FORWARD;
338 key.objectid = btrfs_ino(BTRFS_I(src));
339 key.type = BTRFS_EXTENT_DATA_KEY;
343 u64 next_key_min_offset = key.offset + 1;
344 struct btrfs_file_extent_item *extent;
348 struct btrfs_key new_key;
349 u64 disko = 0, diskl = 0;
350 u64 datao = 0, datal = 0;
354 /* Note the key will change type as we walk through the tree */
355 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
360 * First search, if no extent item that starts at offset off was
361 * found but the previous item is an extent item, it's possible
362 * it might overlap our target range, therefore process it.
364 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
365 btrfs_item_key_to_cpu(path->nodes[0], &key,
367 if (key.type == BTRFS_EXTENT_DATA_KEY)
371 nritems = btrfs_header_nritems(path->nodes[0]);
373 if (path->slots[0] >= nritems) {
374 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
379 nritems = btrfs_header_nritems(path->nodes[0]);
381 leaf = path->nodes[0];
382 slot = path->slots[0];
384 btrfs_item_key_to_cpu(leaf, &key, slot);
385 if (key.type > BTRFS_EXTENT_DATA_KEY ||
386 key.objectid != btrfs_ino(BTRFS_I(src)))
389 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
391 extent = btrfs_item_ptr(leaf, slot,
392 struct btrfs_file_extent_item);
393 extent_gen = btrfs_file_extent_generation(leaf, extent);
394 comp = btrfs_file_extent_compression(leaf, extent);
395 type = btrfs_file_extent_type(leaf, extent);
396 if (type == BTRFS_FILE_EXTENT_REG ||
397 type == BTRFS_FILE_EXTENT_PREALLOC) {
398 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
399 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
400 datao = btrfs_file_extent_offset(leaf, extent);
401 datal = btrfs_file_extent_num_bytes(leaf, extent);
402 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
403 /* Take upper bound, may be compressed */
404 datal = btrfs_file_extent_ram_bytes(leaf, extent);
408 * The first search might have left us at an extent item that
409 * ends before our target range's start, can happen if we have
410 * holes and NO_HOLES feature enabled.
412 if (key.offset + datal <= off) {
415 } else if (key.offset >= off + len) {
418 next_key_min_offset = key.offset + datal;
419 size = btrfs_item_size_nr(leaf, slot);
420 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
423 btrfs_release_path(path);
425 memcpy(&new_key, &key, sizeof(new_key));
426 new_key.objectid = btrfs_ino(BTRFS_I(inode));
427 if (off <= key.offset)
428 new_key.offset = key.offset + destoff - off;
430 new_key.offset = destoff;
433 * Deal with a hole that doesn't have an extent item that
434 * represents it (NO_HOLES feature enabled).
435 * This hole is either in the middle of the cloning range or at
436 * the beginning (fully overlaps it or partially overlaps it).
438 if (new_key.offset != last_dest_end)
439 drop_start = last_dest_end;
441 drop_start = new_key.offset;
443 if (type == BTRFS_FILE_EXTENT_REG ||
444 type == BTRFS_FILE_EXTENT_PREALLOC) {
445 struct btrfs_replace_extent_info clone_info;
448 * a | --- range to clone ---| b
449 * | ------------- extent ------------- |
452 /* Subtract range b */
453 if (key.offset + datal > off + len)
454 datal = off + len - key.offset;
456 /* Subtract range a */
457 if (off > key.offset) {
458 datao += off - key.offset;
459 datal -= off - key.offset;
462 clone_info.disk_offset = disko;
463 clone_info.disk_len = diskl;
464 clone_info.data_offset = datao;
465 clone_info.data_len = datal;
466 clone_info.file_offset = new_key.offset;
467 clone_info.extent_buf = buf;
468 clone_info.is_new_extent = false;
469 ret = btrfs_replace_file_extents(inode, path, drop_start,
470 new_key.offset + datal - 1, &clone_info,
474 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
476 * Inline extents always have to start at file offset 0
477 * and can never be bigger then the sector size. We can
478 * never clone only parts of an inline extent, since all
479 * reflink operations must start at a sector size aligned
480 * offset, and the length must be aligned too or end at
481 * the i_size (which implies the whole inlined data).
483 ASSERT(key.offset == 0);
484 ASSERT(datal <= fs_info->sectorsize);
485 if (key.offset != 0 || datal > fs_info->sectorsize)
488 ret = clone_copy_inline_extent(inode, path, &new_key,
489 drop_start, datal, size,
495 btrfs_release_path(path);
498 * If this is a new extent update the last_reflink_trans of both
499 * inodes. This is used by fsync to make sure it does not log
500 * multiple checksum items with overlapping ranges. For older
501 * extents we don't need to do it since inode logging skips the
502 * checksums for older extents. Also ignore holes and inline
503 * extents because they don't have checksums in the csum tree.
505 if (extent_gen == trans->transid && disko > 0) {
506 BTRFS_I(src)->last_reflink_trans = trans->transid;
507 BTRFS_I(inode)->last_reflink_trans = trans->transid;
510 last_dest_end = ALIGN(new_key.offset + datal,
511 fs_info->sectorsize);
512 ret = clone_finish_inode_update(trans, inode, last_dest_end,
513 destoff, olen, no_time_update);
516 if (new_key.offset + datal >= destoff + len)
519 btrfs_release_path(path);
520 key.offset = next_key_min_offset;
522 if (fatal_signal_pending(current)) {
531 if (last_dest_end < destoff + len) {
533 * We have an implicit hole that fully or partially overlaps our
534 * cloning range at its end. This means that we either have the
535 * NO_HOLES feature enabled or the implicit hole happened due to
536 * mixing buffered and direct IO writes against this file.
538 btrfs_release_path(path);
540 ret = btrfs_replace_file_extents(inode, path, last_dest_end,
541 destoff + len - 1, NULL, &trans);
545 ret = clone_finish_inode_update(trans, inode, destoff + len,
546 destoff, olen, no_time_update);
550 btrfs_free_path(path);
555 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
556 struct inode *inode2, u64 loff2, u64 len)
558 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
559 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
562 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
563 struct inode *inode2, u64 loff2, u64 len)
565 if (inode1 < inode2) {
566 swap(inode1, inode2);
568 } else if (inode1 == inode2 && loff2 < loff1) {
571 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
572 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
575 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
576 struct inode *dst, u64 dst_loff)
578 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
582 * Lock destination range to serialize with concurrent readpages() and
583 * source range to serialize with relocation.
585 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
586 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
587 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
592 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
593 struct inode *dst, u64 dst_loff)
596 u64 i, tail_len, chunk_count;
597 struct btrfs_root *root_dst = BTRFS_I(dst)->root;
599 spin_lock(&root_dst->root_item_lock);
600 if (root_dst->send_in_progress) {
601 btrfs_warn_rl(root_dst->fs_info,
602 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
603 root_dst->root_key.objectid,
604 root_dst->send_in_progress);
605 spin_unlock(&root_dst->root_item_lock);
608 root_dst->dedupe_in_progress++;
609 spin_unlock(&root_dst->root_item_lock);
611 tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
612 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
614 for (i = 0; i < chunk_count; i++) {
615 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
620 loff += BTRFS_MAX_DEDUPE_LEN;
621 dst_loff += BTRFS_MAX_DEDUPE_LEN;
625 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff);
627 spin_lock(&root_dst->root_item_lock);
628 root_dst->dedupe_in_progress--;
629 spin_unlock(&root_dst->root_item_lock);
634 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
635 u64 off, u64 olen, u64 destoff)
637 struct inode *inode = file_inode(file);
638 struct inode *src = file_inode(file_src);
639 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
643 u64 bs = fs_info->sb->s_blocksize;
646 * VFS's generic_remap_file_range_prep() protects us from cloning the
647 * eof block into the middle of a file, which would result in corruption
648 * if the file size is not blocksize aligned. So we don't need to check
649 * for that case here.
651 if (off + len == src->i_size)
652 len = ALIGN(src->i_size, bs) - off;
654 if (destoff > inode->i_size) {
655 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
657 ret = btrfs_cont_expand(BTRFS_I(inode), inode->i_size, destoff);
661 * We may have truncated the last block if the inode's size is
662 * not sector size aligned, so we need to wait for writeback to
663 * complete before proceeding further, otherwise we can race
664 * with cloning and attempt to increment a reference to an
665 * extent that no longer exists (writeback completed right after
666 * we found the previous extent covering eof and before we
667 * attempted to increment its reference count).
669 ret = btrfs_wait_ordered_range(inode, wb_start,
676 * Lock destination range to serialize with concurrent readpages() and
677 * source range to serialize with relocation.
679 btrfs_double_extent_lock(src, off, inode, destoff, len);
680 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
681 btrfs_double_extent_unlock(src, off, inode, destoff, len);
684 * We may have copied an inline extent into a page of the destination
685 * range, so wait for writeback to complete before truncating pages
686 * from the page cache. This is a rare case.
688 wb_ret = btrfs_wait_ordered_range(inode, destoff, len);
689 ret = ret ? ret : wb_ret;
691 * Truncate page cache pages so that future reads will see the cloned
692 * data immediately and not the previous data.
694 truncate_inode_pages_range(&inode->i_data,
695 round_down(destoff, PAGE_SIZE),
696 round_up(destoff + len, PAGE_SIZE) - 1);
701 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
702 struct file *file_out, loff_t pos_out,
703 loff_t *len, unsigned int remap_flags)
705 struct inode *inode_in = file_inode(file_in);
706 struct inode *inode_out = file_inode(file_out);
707 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
708 bool same_inode = inode_out == inode_in;
712 if (!(remap_flags & REMAP_FILE_DEDUP)) {
713 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
715 if (btrfs_root_readonly(root_out))
718 if (file_in->f_path.mnt != file_out->f_path.mnt ||
719 inode_in->i_sb != inode_out->i_sb)
723 /* Don't make the dst file partly checksummed */
724 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
725 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
730 * Now that the inodes are locked, we need to start writeback ourselves
731 * and can not rely on the writeback from the VFS's generic helper
732 * generic_remap_file_range_prep() because:
734 * 1) For compression we must call filemap_fdatawrite_range() range
735 * twice (btrfs_fdatawrite_range() does it for us), and the generic
736 * helper only calls it once;
738 * 2) filemap_fdatawrite_range(), called by the generic helper only
739 * waits for the writeback to complete, i.e. for IO to be done, and
740 * not for the ordered extents to complete. We need to wait for them
741 * to complete so that new file extent items are in the fs tree.
743 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
744 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
746 wb_len = ALIGN(*len, bs);
749 * Since we don't lock ranges, wait for ongoing lockless dio writes (as
750 * any in progress could create its ordered extents after we wait for
751 * existing ordered extents below).
753 inode_dio_wait(inode_in);
755 inode_dio_wait(inode_out);
758 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
760 * Btrfs' back references do not have a block level granularity, they
761 * work at the whole extent level.
762 * NOCOW buffered write without data space reserved may not be able
763 * to fall back to CoW due to lack of data space, thus could cause
766 * Here we take a shortcut by flushing the whole inode, so that all
767 * nocow write should reach disk as nocow before we increase the
768 * reference of the extent. We could do better by only flushing NOCOW
769 * data, but that needs extra accounting.
771 * Also we don't need to check ASYNC_EXTENT, as async extent will be
772 * CoWed anyway, not affecting nocow part.
774 ret = filemap_flush(inode_in->i_mapping);
778 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
782 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
787 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
791 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
792 struct file *dst_file, loff_t destoff, loff_t len,
793 unsigned int remap_flags)
795 struct inode *src_inode = file_inode(src_file);
796 struct inode *dst_inode = file_inode(dst_file);
797 bool same_inode = dst_inode == src_inode;
800 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
804 inode_lock(src_inode);
806 lock_two_nondirectories(src_inode, dst_inode);
808 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
810 if (ret < 0 || len == 0)
813 if (remap_flags & REMAP_FILE_DEDUP)
814 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
816 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
820 inode_unlock(src_inode);
822 unlock_two_nondirectories(src_inode, dst_inode);
824 return ret < 0 ? ret : len;