1 // SPDX-License-Identifier: GPL-2.0
2
3 #include <linux/blkdev.h>
4 #include <linux/iversion.h>
5 #include "compression.h"
6 #include "ctree.h"
7 #include "delalloc-space.h"
8 #include "reflink.h"
9 #include "transaction.h"
10
11 #define BTRFS_MAX_DEDUPE_LEN SZ_16M
12
clone_finish_inode_update(struct btrfs_trans_handle * trans,struct inode * inode,u64 endoff,const u64 destoff,const u64 olen,int no_time_update)13 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
14 struct inode *inode,
15 u64 endoff,
16 const u64 destoff,
17 const u64 olen,
18 int no_time_update)
19 {
20 struct btrfs_root *root = BTRFS_I(inode)->root;
21 int ret;
22
23 inode_inc_iversion(inode);
24 if (!no_time_update)
25 inode->i_mtime = inode->i_ctime = current_time(inode);
26 /*
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.
29 */
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);
35 }
36
37 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
38 if (ret) {
39 btrfs_abort_transaction(trans, ret);
40 btrfs_end_transaction(trans);
41 goto out;
42 }
43 ret = btrfs_end_transaction(trans);
44 out:
45 return ret;
46 }
47
copy_inline_to_page(struct btrfs_inode * inode,const u64 file_offset,char * inline_data,const u64 size,const u64 datal,const u8 comp_type)48 static int copy_inline_to_page(struct btrfs_inode *inode,
49 const u64 file_offset,
50 char *inline_data,
51 const u64 size,
52 const u64 datal,
53 const u8 comp_type)
54 {
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;
62 int ret;
63
64 ASSERT(IS_ALIGNED(file_offset, block_size));
65
66 /*
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.
71 */
72 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, file_offset,
73 block_size);
74 if (ret)
75 goto out;
76
77 page = find_or_create_page(mapping, file_offset >> PAGE_SHIFT,
78 btrfs_alloc_write_mask(mapping));
79 if (!page) {
80 ret = -ENOMEM;
81 goto out_unlock;
82 }
83
84 ret = set_page_extent_mapped(page);
85 if (ret < 0)
86 goto out_unlock;
87
88 clear_extent_bit(&inode->io_tree, file_offset, range_end,
89 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
90 0, 0, NULL);
91 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL);
92 if (ret)
93 goto out_unlock;
94
95 /*
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.
105 */
106 set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &inode->runtime_flags);
107
108 if (comp_type == BTRFS_COMPRESS_NONE) {
109 memcpy_to_page(page, 0, data_start, datal);
110 flush_dcache_page(page);
111 } else {
112 ret = btrfs_decompress(comp_type, data_start, page, 0,
113 inline_size, datal);
114 if (ret)
115 goto out_unlock;
116 flush_dcache_page(page);
117 }
118
119 /*
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:
123 *
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"
128 *
129 * So what's in the range [500, 4095] corresponds to zeroes.
130 */
131 if (datal < block_size) {
132 memzero_page(page, datal, block_size - datal);
133 flush_dcache_page(page);
134 }
135
136 SetPageUptodate(page);
137 ClearPageChecked(page);
138 set_page_dirty(page);
139 out_unlock:
140 if (page) {
141 unlock_page(page);
142 put_page(page);
143 }
144 if (ret)
145 btrfs_delalloc_release_space(inode, data_reserved, file_offset,
146 block_size, true);
147 btrfs_delalloc_release_extents(inode, block_size);
148 out:
149 extent_changeset_free(data_reserved);
150
151 return ret;
152 }
153
154 /*
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.
158 */
clone_copy_inline_extent(struct inode * dst,struct btrfs_path * path,struct btrfs_key * new_key,const u64 drop_start,const u64 datal,const u64 size,const u8 comp_type,char * inline_data,struct btrfs_trans_handle ** trans_out)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,
163 const u64 datal,
164 const u64 size,
165 const u8 comp_type,
166 char *inline_data,
167 struct btrfs_trans_handle **trans_out)
168 {
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 };
175 int ret;
176 struct btrfs_key key;
177
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);
181 goto out;
182 }
183
184 key.objectid = btrfs_ino(BTRFS_I(dst));
185 key.type = BTRFS_EXTENT_DATA_KEY;
186 key.offset = 0;
187 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
188 if (ret < 0) {
189 return ret;
190 } else if (ret > 0) {
191 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
192 ret = btrfs_next_leaf(root, path);
193 if (ret < 0)
194 return ret;
195 else if (ret > 0)
196 goto copy_inline_extent;
197 }
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) {
201 /*
202 * There's an implicit hole at file offset 0, copy the
203 * inline extent's data to the page.
204 */
205 ASSERT(key.offset > 0);
206 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
207 inline_data, size, datal,
208 comp_type);
209 goto out;
210 }
211 } else if (i_size_read(dst) <= datal) {
212 struct btrfs_file_extent_item *ei;
213
214 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
215 struct btrfs_file_extent_item);
216 /*
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.
220 */
221 if (btrfs_file_extent_type(path->nodes[0], ei) ==
222 BTRFS_FILE_EXTENT_INLINE)
223 goto copy_inline_extent;
224
225 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
226 inline_data, size, datal, comp_type);
227 goto out;
228 }
229
230 copy_inline_extent:
231 ret = 0;
232 /*
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.
235 */
236 if (i_size_read(dst) > datal) {
237 /*
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.
242 */
243 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
244 inline_data, size, datal, comp_type);
245 goto out;
246 }
247
248 btrfs_release_path(path);
249 /*
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.
253 *
254 * 1 unit - adjusting old extent (we may have to split it)
255 * 1 unit - add new extent
256 * 1 unit - inode update
257 */
258 trans = btrfs_start_transaction(root, 3);
259 if (IS_ERR(trans)) {
260 ret = PTR_ERR(trans);
261 trans = NULL;
262 goto out;
263 }
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);
269 if (ret)
270 goto out;
271 ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
272 if (ret)
273 goto out;
274
275 write_extent_buffer(path->nodes[0], inline_data,
276 btrfs_item_ptr_offset(path->nodes[0],
277 path->slots[0]),
278 size);
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);
282 out:
283 if (!ret && !trans) {
284 /*
285 * No transaction here means we copied the inline extent into a
286 * page of the destination inode.
287 *
288 * 1 unit to update inode item
289 */
290 trans = btrfs_start_transaction(root, 1);
291 if (IS_ERR(trans)) {
292 ret = PTR_ERR(trans);
293 trans = NULL;
294 }
295 }
296 if (ret && trans) {
297 btrfs_abort_transaction(trans, ret);
298 btrfs_end_transaction(trans);
299 }
300 if (!ret)
301 *trans_out = trans;
302
303 return ret;
304 }
305
306 /**
307 * btrfs_clone() - clone a range from inode file to another
308 *
309 * @src: Inode to clone from
310 * @inode: Inode to clone to
311 * @off: Offset within source to start clone from
312 * @olen: Original length, passed by user, of range to clone
313 * @olen_aligned: Block-aligned value of olen
314 * @destoff: Offset within @inode to start clone
315 * @no_time_update: Whether to update mtime/ctime on the target inode
316 */
btrfs_clone(struct inode * src,struct inode * inode,const u64 off,const u64 olen,const u64 olen_aligned,const u64 destoff,int no_time_update)317 static int btrfs_clone(struct inode *src, struct inode *inode,
318 const u64 off, const u64 olen, const u64 olen_aligned,
319 const u64 destoff, int no_time_update)
320 {
321 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
322 struct btrfs_path *path = NULL;
323 struct extent_buffer *leaf;
324 struct btrfs_trans_handle *trans;
325 char *buf = NULL;
326 struct btrfs_key key;
327 u32 nritems;
328 int slot;
329 int ret;
330 const u64 len = olen_aligned;
331 u64 last_dest_end = destoff;
332
333 ret = -ENOMEM;
334 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
335 if (!buf)
336 return ret;
337
338 path = btrfs_alloc_path();
339 if (!path) {
340 kvfree(buf);
341 return ret;
342 }
343
344 path->reada = READA_FORWARD;
345 /* Clone data */
346 key.objectid = btrfs_ino(BTRFS_I(src));
347 key.type = BTRFS_EXTENT_DATA_KEY;
348 key.offset = off;
349
350 while (1) {
351 u64 next_key_min_offset = key.offset + 1;
352 struct btrfs_file_extent_item *extent;
353 u64 extent_gen;
354 int type;
355 u32 size;
356 struct btrfs_key new_key;
357 u64 disko = 0, diskl = 0;
358 u64 datao = 0, datal = 0;
359 u8 comp;
360 u64 drop_start;
361
362 /* Note the key will change type as we walk through the tree */
363 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
364 0, 0);
365 if (ret < 0)
366 goto out;
367 /*
368 * First search, if no extent item that starts at offset off was
369 * found but the previous item is an extent item, it's possible
370 * it might overlap our target range, therefore process it.
371 */
372 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
373 btrfs_item_key_to_cpu(path->nodes[0], &key,
374 path->slots[0] - 1);
375 if (key.type == BTRFS_EXTENT_DATA_KEY)
376 path->slots[0]--;
377 }
378
379 nritems = btrfs_header_nritems(path->nodes[0]);
380 process_slot:
381 if (path->slots[0] >= nritems) {
382 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
383 if (ret < 0)
384 goto out;
385 if (ret > 0)
386 break;
387 nritems = btrfs_header_nritems(path->nodes[0]);
388 }
389 leaf = path->nodes[0];
390 slot = path->slots[0];
391
392 btrfs_item_key_to_cpu(leaf, &key, slot);
393 if (key.type > BTRFS_EXTENT_DATA_KEY ||
394 key.objectid != btrfs_ino(BTRFS_I(src)))
395 break;
396
397 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
398
399 extent = btrfs_item_ptr(leaf, slot,
400 struct btrfs_file_extent_item);
401 extent_gen = btrfs_file_extent_generation(leaf, extent);
402 comp = btrfs_file_extent_compression(leaf, extent);
403 type = btrfs_file_extent_type(leaf, extent);
404 if (type == BTRFS_FILE_EXTENT_REG ||
405 type == BTRFS_FILE_EXTENT_PREALLOC) {
406 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
407 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
408 datao = btrfs_file_extent_offset(leaf, extent);
409 datal = btrfs_file_extent_num_bytes(leaf, extent);
410 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
411 /* Take upper bound, may be compressed */
412 datal = btrfs_file_extent_ram_bytes(leaf, extent);
413 }
414
415 /*
416 * The first search might have left us at an extent item that
417 * ends before our target range's start, can happen if we have
418 * holes and NO_HOLES feature enabled.
419 */
420 if (key.offset + datal <= off) {
421 path->slots[0]++;
422 goto process_slot;
423 } else if (key.offset >= off + len) {
424 break;
425 }
426 next_key_min_offset = key.offset + datal;
427 size = btrfs_item_size_nr(leaf, slot);
428 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
429 size);
430
431 btrfs_release_path(path);
432
433 memcpy(&new_key, &key, sizeof(new_key));
434 new_key.objectid = btrfs_ino(BTRFS_I(inode));
435 if (off <= key.offset)
436 new_key.offset = key.offset + destoff - off;
437 else
438 new_key.offset = destoff;
439
440 /*
441 * Deal with a hole that doesn't have an extent item that
442 * represents it (NO_HOLES feature enabled).
443 * This hole is either in the middle of the cloning range or at
444 * the beginning (fully overlaps it or partially overlaps it).
445 */
446 if (new_key.offset != last_dest_end)
447 drop_start = last_dest_end;
448 else
449 drop_start = new_key.offset;
450
451 if (type == BTRFS_FILE_EXTENT_REG ||
452 type == BTRFS_FILE_EXTENT_PREALLOC) {
453 struct btrfs_replace_extent_info clone_info;
454
455 /*
456 * a | --- range to clone ---| b
457 * | ------------- extent ------------- |
458 */
459
460 /* Subtract range b */
461 if (key.offset + datal > off + len)
462 datal = off + len - key.offset;
463
464 /* Subtract range a */
465 if (off > key.offset) {
466 datao += off - key.offset;
467 datal -= off - key.offset;
468 }
469
470 clone_info.disk_offset = disko;
471 clone_info.disk_len = diskl;
472 clone_info.data_offset = datao;
473 clone_info.data_len = datal;
474 clone_info.file_offset = new_key.offset;
475 clone_info.extent_buf = buf;
476 clone_info.is_new_extent = false;
477 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
478 drop_start, new_key.offset + datal - 1,
479 &clone_info, &trans);
480 if (ret)
481 goto out;
482 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
483 /*
484 * Inline extents always have to start at file offset 0
485 * and can never be bigger then the sector size. We can
486 * never clone only parts of an inline extent, since all
487 * reflink operations must start at a sector size aligned
488 * offset, and the length must be aligned too or end at
489 * the i_size (which implies the whole inlined data).
490 */
491 ASSERT(key.offset == 0);
492 ASSERT(datal <= fs_info->sectorsize);
493 if (key.offset != 0 || datal > fs_info->sectorsize)
494 return -EUCLEAN;
495
496 ret = clone_copy_inline_extent(inode, path, &new_key,
497 drop_start, datal, size,
498 comp, buf, &trans);
499 if (ret)
500 goto out;
501 }
502
503 btrfs_release_path(path);
504
505 /*
506 * If this is a new extent update the last_reflink_trans of both
507 * inodes. This is used by fsync to make sure it does not log
508 * multiple checksum items with overlapping ranges. For older
509 * extents we don't need to do it since inode logging skips the
510 * checksums for older extents. Also ignore holes and inline
511 * extents because they don't have checksums in the csum tree.
512 */
513 if (extent_gen == trans->transid && disko > 0) {
514 BTRFS_I(src)->last_reflink_trans = trans->transid;
515 BTRFS_I(inode)->last_reflink_trans = trans->transid;
516 }
517
518 last_dest_end = ALIGN(new_key.offset + datal,
519 fs_info->sectorsize);
520 ret = clone_finish_inode_update(trans, inode, last_dest_end,
521 destoff, olen, no_time_update);
522 if (ret)
523 goto out;
524 if (new_key.offset + datal >= destoff + len)
525 break;
526
527 btrfs_release_path(path);
528 key.offset = next_key_min_offset;
529
530 if (fatal_signal_pending(current)) {
531 ret = -EINTR;
532 goto out;
533 }
534
535 cond_resched();
536 }
537 ret = 0;
538
539 if (last_dest_end < destoff + len) {
540 /*
541 * We have an implicit hole that fully or partially overlaps our
542 * cloning range at its end. This means that we either have the
543 * NO_HOLES feature enabled or the implicit hole happened due to
544 * mixing buffered and direct IO writes against this file.
545 */
546 btrfs_release_path(path);
547
548 /*
549 * When using NO_HOLES and we are cloning a range that covers
550 * only a hole (no extents) into a range beyond the current
551 * i_size, punching a hole in the target range will not create
552 * an extent map defining a hole, because the range starts at or
553 * beyond current i_size. If the file previously had an i_size
554 * greater than the new i_size set by this clone operation, we
555 * need to make sure the next fsync is a full fsync, so that it
556 * detects and logs a hole covering a range from the current
557 * i_size to the new i_size. If the clone range covers extents,
558 * besides a hole, then we know the full sync flag was already
559 * set by previous calls to btrfs_replace_file_extents() that
560 * replaced file extent items.
561 */
562 if (last_dest_end >= i_size_read(inode))
563 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
564 &BTRFS_I(inode)->runtime_flags);
565
566 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
567 last_dest_end, destoff + len - 1, NULL, &trans);
568 if (ret)
569 goto out;
570
571 ret = clone_finish_inode_update(trans, inode, destoff + len,
572 destoff, olen, no_time_update);
573 }
574
575 out:
576 btrfs_free_path(path);
577 kvfree(buf);
578 clear_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &BTRFS_I(inode)->runtime_flags);
579
580 return ret;
581 }
582
btrfs_double_extent_unlock(struct inode * inode1,u64 loff1,struct inode * inode2,u64 loff2,u64 len)583 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
584 struct inode *inode2, u64 loff2, u64 len)
585 {
586 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
587 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
588 }
589
btrfs_double_extent_lock(struct inode * inode1,u64 loff1,struct inode * inode2,u64 loff2,u64 len)590 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
591 struct inode *inode2, u64 loff2, u64 len)
592 {
593 if (inode1 < inode2) {
594 swap(inode1, inode2);
595 swap(loff1, loff2);
596 } else if (inode1 == inode2 && loff2 < loff1) {
597 swap(loff1, loff2);
598 }
599 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1);
600 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1);
601 }
602
btrfs_double_mmap_lock(struct inode * inode1,struct inode * inode2)603 static void btrfs_double_mmap_lock(struct inode *inode1, struct inode *inode2)
604 {
605 if (inode1 < inode2)
606 swap(inode1, inode2);
607 down_write(&BTRFS_I(inode1)->i_mmap_lock);
608 down_write_nested(&BTRFS_I(inode2)->i_mmap_lock, SINGLE_DEPTH_NESTING);
609 }
610
btrfs_double_mmap_unlock(struct inode * inode1,struct inode * inode2)611 static void btrfs_double_mmap_unlock(struct inode *inode1, struct inode *inode2)
612 {
613 up_write(&BTRFS_I(inode1)->i_mmap_lock);
614 up_write(&BTRFS_I(inode2)->i_mmap_lock);
615 }
616
btrfs_extent_same_range(struct inode * src,u64 loff,u64 len,struct inode * dst,u64 dst_loff)617 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
618 struct inode *dst, u64 dst_loff)
619 {
620 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize;
621 int ret;
622
623 /*
624 * Lock destination range to serialize with concurrent readpages() and
625 * source range to serialize with relocation.
626 */
627 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
628 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
629 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
630
631 return ret;
632 }
633
btrfs_extent_same(struct inode * src,u64 loff,u64 olen,struct inode * dst,u64 dst_loff)634 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
635 struct inode *dst, u64 dst_loff)
636 {
637 int ret;
638 u64 i, tail_len, chunk_count;
639 struct btrfs_root *root_dst = BTRFS_I(dst)->root;
640
641 spin_lock(&root_dst->root_item_lock);
642 if (root_dst->send_in_progress) {
643 btrfs_warn_rl(root_dst->fs_info,
644 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
645 root_dst->root_key.objectid,
646 root_dst->send_in_progress);
647 spin_unlock(&root_dst->root_item_lock);
648 return -EAGAIN;
649 }
650 root_dst->dedupe_in_progress++;
651 spin_unlock(&root_dst->root_item_lock);
652
653 tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
654 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
655
656 for (i = 0; i < chunk_count; i++) {
657 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
658 dst, dst_loff);
659 if (ret)
660 goto out;
661
662 loff += BTRFS_MAX_DEDUPE_LEN;
663 dst_loff += BTRFS_MAX_DEDUPE_LEN;
664 }
665
666 if (tail_len > 0)
667 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff);
668 out:
669 spin_lock(&root_dst->root_item_lock);
670 root_dst->dedupe_in_progress--;
671 spin_unlock(&root_dst->root_item_lock);
672
673 return ret;
674 }
675
btrfs_clone_files(struct file * file,struct file * file_src,u64 off,u64 olen,u64 destoff)676 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
677 u64 off, u64 olen, u64 destoff)
678 {
679 struct inode *inode = file_inode(file);
680 struct inode *src = file_inode(file_src);
681 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
682 int ret;
683 int wb_ret;
684 u64 len = olen;
685 u64 bs = fs_info->sb->s_blocksize;
686
687 /*
688 * VFS's generic_remap_file_range_prep() protects us from cloning the
689 * eof block into the middle of a file, which would result in corruption
690 * if the file size is not blocksize aligned. So we don't need to check
691 * for that case here.
692 */
693 if (off + len == src->i_size)
694 len = ALIGN(src->i_size, bs) - off;
695
696 if (destoff > inode->i_size) {
697 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
698
699 ret = btrfs_cont_expand(BTRFS_I(inode), inode->i_size, destoff);
700 if (ret)
701 return ret;
702 /*
703 * We may have truncated the last block if the inode's size is
704 * not sector size aligned, so we need to wait for writeback to
705 * complete before proceeding further, otherwise we can race
706 * with cloning and attempt to increment a reference to an
707 * extent that no longer exists (writeback completed right after
708 * we found the previous extent covering eof and before we
709 * attempted to increment its reference count).
710 */
711 ret = btrfs_wait_ordered_range(inode, wb_start,
712 destoff - wb_start);
713 if (ret)
714 return ret;
715 }
716
717 /*
718 * Lock destination range to serialize with concurrent readpages() and
719 * source range to serialize with relocation.
720 */
721 btrfs_double_extent_lock(src, off, inode, destoff, len);
722 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
723 btrfs_double_extent_unlock(src, off, inode, destoff, len);
724
725 /*
726 * We may have copied an inline extent into a page of the destination
727 * range, so wait for writeback to complete before truncating pages
728 * from the page cache. This is a rare case.
729 */
730 wb_ret = btrfs_wait_ordered_range(inode, destoff, len);
731 ret = ret ? ret : wb_ret;
732 /*
733 * Truncate page cache pages so that future reads will see the cloned
734 * data immediately and not the previous data.
735 */
736 truncate_inode_pages_range(&inode->i_data,
737 round_down(destoff, PAGE_SIZE),
738 round_up(destoff + len, PAGE_SIZE) - 1);
739
740 return ret;
741 }
742
btrfs_remap_file_range_prep(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t * len,unsigned int remap_flags)743 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
744 struct file *file_out, loff_t pos_out,
745 loff_t *len, unsigned int remap_flags)
746 {
747 struct inode *inode_in = file_inode(file_in);
748 struct inode *inode_out = file_inode(file_out);
749 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
750 bool same_inode = inode_out == inode_in;
751 u64 wb_len;
752 int ret;
753
754 if (!(remap_flags & REMAP_FILE_DEDUP)) {
755 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
756
757 if (btrfs_root_readonly(root_out))
758 return -EROFS;
759
760 if (file_in->f_path.mnt != file_out->f_path.mnt ||
761 inode_in->i_sb != inode_out->i_sb)
762 return -EXDEV;
763 }
764
765 /* Don't make the dst file partly checksummed */
766 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
767 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
768 return -EINVAL;
769 }
770
771 /*
772 * Now that the inodes are locked, we need to start writeback ourselves
773 * and can not rely on the writeback from the VFS's generic helper
774 * generic_remap_file_range_prep() because:
775 *
776 * 1) For compression we must call filemap_fdatawrite_range() range
777 * twice (btrfs_fdatawrite_range() does it for us), and the generic
778 * helper only calls it once;
779 *
780 * 2) filemap_fdatawrite_range(), called by the generic helper only
781 * waits for the writeback to complete, i.e. for IO to be done, and
782 * not for the ordered extents to complete. We need to wait for them
783 * to complete so that new file extent items are in the fs tree.
784 */
785 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
786 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
787 else
788 wb_len = ALIGN(*len, bs);
789
790 /*
791 * Since we don't lock ranges, wait for ongoing lockless dio writes (as
792 * any in progress could create its ordered extents after we wait for
793 * existing ordered extents below).
794 */
795 inode_dio_wait(inode_in);
796 if (!same_inode)
797 inode_dio_wait(inode_out);
798
799 /*
800 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
801 *
802 * Btrfs' back references do not have a block level granularity, they
803 * work at the whole extent level.
804 * NOCOW buffered write without data space reserved may not be able
805 * to fall back to CoW due to lack of data space, thus could cause
806 * data loss.
807 *
808 * Here we take a shortcut by flushing the whole inode, so that all
809 * nocow write should reach disk as nocow before we increase the
810 * reference of the extent. We could do better by only flushing NOCOW
811 * data, but that needs extra accounting.
812 *
813 * Also we don't need to check ASYNC_EXTENT, as async extent will be
814 * CoWed anyway, not affecting nocow part.
815 */
816 ret = filemap_flush(inode_in->i_mapping);
817 if (ret < 0)
818 return ret;
819
820 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
821 wb_len);
822 if (ret < 0)
823 return ret;
824 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
825 wb_len);
826 if (ret < 0)
827 return ret;
828
829 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
830 len, remap_flags);
831 }
832
file_sync_write(const struct file * file)833 static bool file_sync_write(const struct file *file)
834 {
835 if (file->f_flags & (__O_SYNC | O_DSYNC))
836 return true;
837 if (IS_SYNC(file_inode(file)))
838 return true;
839
840 return false;
841 }
842
btrfs_remap_file_range(struct file * src_file,loff_t off,struct file * dst_file,loff_t destoff,loff_t len,unsigned int remap_flags)843 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
844 struct file *dst_file, loff_t destoff, loff_t len,
845 unsigned int remap_flags)
846 {
847 struct inode *src_inode = file_inode(src_file);
848 struct inode *dst_inode = file_inode(dst_file);
849 bool same_inode = dst_inode == src_inode;
850 int ret;
851
852 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
853 return -EINVAL;
854
855 if (same_inode) {
856 btrfs_inode_lock(src_inode, BTRFS_ILOCK_MMAP);
857 } else {
858 lock_two_nondirectories(src_inode, dst_inode);
859 btrfs_double_mmap_lock(src_inode, dst_inode);
860 }
861
862 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
863 &len, remap_flags);
864 if (ret < 0 || len == 0)
865 goto out_unlock;
866
867 if (remap_flags & REMAP_FILE_DEDUP)
868 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
869 else
870 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
871
872 out_unlock:
873 if (same_inode) {
874 btrfs_inode_unlock(src_inode, BTRFS_ILOCK_MMAP);
875 } else {
876 btrfs_double_mmap_unlock(src_inode, dst_inode);
877 unlock_two_nondirectories(src_inode, dst_inode);
878 }
879
880 /*
881 * If either the source or the destination file was opened with O_SYNC,
882 * O_DSYNC or has the S_SYNC attribute, fsync both the destination and
883 * source files/ranges, so that after a successful return (0) followed
884 * by a power failure results in the reflinked data to be readable from
885 * both files/ranges.
886 */
887 if (ret == 0 && len > 0 &&
888 (file_sync_write(src_file) || file_sync_write(dst_file))) {
889 ret = btrfs_sync_file(src_file, off, off + len - 1, 0);
890 if (ret == 0)
891 ret = btrfs_sync_file(dst_file, destoff,
892 destoff + len - 1, 0);
893 }
894
895 return ret < 0 ? ret : len;
896 }
897