1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/iversion.h> 4 #include "ctree.h" 5 #include "reflink.h" 6 #include "transaction.h" 7 8 #define BTRFS_MAX_DEDUPE_LEN SZ_16M 9 10 static int clone_finish_inode_update(struct btrfs_trans_handle *trans, 11 struct inode *inode, 12 u64 endoff, 13 const u64 destoff, 14 const u64 olen, 15 int no_time_update) 16 { 17 struct btrfs_root *root = BTRFS_I(inode)->root; 18 int ret; 19 20 inode_inc_iversion(inode); 21 if (!no_time_update) 22 inode->i_mtime = inode->i_ctime = current_time(inode); 23 /* 24 * We round up to the block size at eof when determining which 25 * extents to clone above, but shouldn't round up the file size. 26 */ 27 if (endoff > destoff + olen) 28 endoff = destoff + olen; 29 if (endoff > inode->i_size) { 30 i_size_write(inode, endoff); 31 btrfs_inode_safe_disk_i_size_write(inode, 0); 32 } 33 34 ret = btrfs_update_inode(trans, root, inode); 35 if (ret) { 36 btrfs_abort_transaction(trans, ret); 37 btrfs_end_transaction(trans); 38 goto out; 39 } 40 ret = btrfs_end_transaction(trans); 41 out: 42 return ret; 43 } 44 45 /* 46 * Make sure we do not end up inserting an inline extent into a file that has 47 * already other (non-inline) extents. If a file has an inline extent it can 48 * not have any other extents and the (single) inline extent must start at the 49 * file offset 0. Failing to respect these rules will lead to file corruption, 50 * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc 51 * 52 * We can have extents that have been already written to disk or we can have 53 * dirty ranges still in delalloc, in which case the extent maps and items are 54 * created only when we run delalloc, and the delalloc ranges might fall outside 55 * the range we are currently locking in the inode's io tree. So we check the 56 * inode's i_size because of that (i_size updates are done while holding the 57 * i_mutex, which we are holding here). 58 * We also check to see if the inode has a size not greater than "datal" but has 59 * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are 60 * protected against such concurrent fallocate calls by the i_mutex). 61 * 62 * If the file has no extents but a size greater than datal, do not allow the 63 * copy because we would need turn the inline extent into a non-inline one (even 64 * with NO_HOLES enabled). If we find our destination inode only has one inline 65 * extent, just overwrite it with the source inline extent if its size is less 66 * than the source extent's size, or we could copy the source inline extent's 67 * data into the destination inode's inline extent if the later is greater then 68 * the former. 69 */ 70 static int clone_copy_inline_extent(struct inode *dst, 71 struct btrfs_trans_handle *trans, 72 struct btrfs_path *path, 73 struct btrfs_key *new_key, 74 const u64 drop_start, 75 const u64 datal, 76 const u64 size, 77 const char *inline_data) 78 { 79 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb); 80 struct btrfs_root *root = BTRFS_I(dst)->root; 81 const u64 aligned_end = ALIGN(new_key->offset + datal, 82 fs_info->sectorsize); 83 int ret; 84 struct btrfs_key key; 85 86 if (new_key->offset > 0) 87 return -EOPNOTSUPP; 88 89 key.objectid = btrfs_ino(BTRFS_I(dst)); 90 key.type = BTRFS_EXTENT_DATA_KEY; 91 key.offset = 0; 92 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 93 if (ret < 0) { 94 return ret; 95 } else if (ret > 0) { 96 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 97 ret = btrfs_next_leaf(root, path); 98 if (ret < 0) 99 return ret; 100 else if (ret > 0) 101 goto copy_inline_extent; 102 } 103 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 104 if (key.objectid == btrfs_ino(BTRFS_I(dst)) && 105 key.type == BTRFS_EXTENT_DATA_KEY) { 106 ASSERT(key.offset > 0); 107 return -EOPNOTSUPP; 108 } 109 } else if (i_size_read(dst) <= datal) { 110 struct btrfs_file_extent_item *ei; 111 u64 ext_len; 112 113 /* 114 * If the file size is <= datal, make sure there are no other 115 * extents following (can happen do to an fallocate call with 116 * the flag FALLOC_FL_KEEP_SIZE). 117 */ 118 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 119 struct btrfs_file_extent_item); 120 /* 121 * If it's an inline extent, it can not have other extents 122 * following it. 123 */ 124 if (btrfs_file_extent_type(path->nodes[0], ei) == 125 BTRFS_FILE_EXTENT_INLINE) 126 goto copy_inline_extent; 127 128 ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei); 129 if (ext_len > aligned_end) 130 return -EOPNOTSUPP; 131 132 ret = btrfs_next_item(root, path); 133 if (ret < 0) { 134 return ret; 135 } else if (ret == 0) { 136 btrfs_item_key_to_cpu(path->nodes[0], &key, 137 path->slots[0]); 138 if (key.objectid == btrfs_ino(BTRFS_I(dst)) && 139 key.type == BTRFS_EXTENT_DATA_KEY) 140 return -EOPNOTSUPP; 141 } 142 } 143 144 copy_inline_extent: 145 /* 146 * We have no extent items, or we have an extent at offset 0 which may 147 * or may not be inlined. All these cases are dealt the same way. 148 */ 149 if (i_size_read(dst) > datal) { 150 /* 151 * If the destination inode has an inline extent. 152 * This would require copying the data from the source inline 153 * extent into the beginning of the destination's inline extent. 154 * But this is really complex, both extents can be compressed 155 * or just one of them, which would require decompressing and 156 * re-compressing data (which could increase the new compressed 157 * size, not allowing the compressed data to fit anymore in an 158 * inline extent). 159 * So just don't support this case for now (it should be rare, 160 * we are not really saving space when cloning inline extents). 161 */ 162 return -EOPNOTSUPP; 163 } 164 165 btrfs_release_path(path); 166 ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1); 167 if (ret) 168 return ret; 169 ret = btrfs_insert_empty_item(trans, root, path, new_key, size); 170 if (ret) 171 return ret; 172 173 write_extent_buffer(path->nodes[0], inline_data, 174 btrfs_item_ptr_offset(path->nodes[0], 175 path->slots[0]), 176 size); 177 inode_add_bytes(dst, datal); 178 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(dst)->runtime_flags); 179 180 return 0; 181 } 182 183 /** 184 * btrfs_clone() - clone a range from inode file to another 185 * 186 * @src: Inode to clone from 187 * @inode: Inode to clone to 188 * @off: Offset within source to start clone from 189 * @olen: Original length, passed by user, of range to clone 190 * @olen_aligned: Block-aligned value of olen 191 * @destoff: Offset within @inode to start clone 192 * @no_time_update: Whether to update mtime/ctime on the target inode 193 */ 194 static int btrfs_clone(struct inode *src, struct inode *inode, 195 const u64 off, const u64 olen, const u64 olen_aligned, 196 const u64 destoff, int no_time_update) 197 { 198 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 199 struct btrfs_root *root = BTRFS_I(inode)->root; 200 struct btrfs_path *path = NULL; 201 struct extent_buffer *leaf; 202 struct btrfs_trans_handle *trans; 203 char *buf = NULL; 204 struct btrfs_key key; 205 u32 nritems; 206 int slot; 207 int ret; 208 const u64 len = olen_aligned; 209 u64 last_dest_end = destoff; 210 211 ret = -ENOMEM; 212 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); 213 if (!buf) 214 return ret; 215 216 path = btrfs_alloc_path(); 217 if (!path) { 218 kvfree(buf); 219 return ret; 220 } 221 222 path->reada = READA_FORWARD; 223 /* Clone data */ 224 key.objectid = btrfs_ino(BTRFS_I(src)); 225 key.type = BTRFS_EXTENT_DATA_KEY; 226 key.offset = off; 227 228 while (1) { 229 u64 next_key_min_offset = key.offset + 1; 230 struct btrfs_file_extent_item *extent; 231 int type; 232 u32 size; 233 struct btrfs_key new_key; 234 u64 disko = 0, diskl = 0; 235 u64 datao = 0, datal = 0; 236 u64 drop_start; 237 238 /* Note the key will change type as we walk through the tree */ 239 path->leave_spinning = 1; 240 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 241 0, 0); 242 if (ret < 0) 243 goto out; 244 /* 245 * First search, if no extent item that starts at offset off was 246 * found but the previous item is an extent item, it's possible 247 * it might overlap our target range, therefore process it. 248 */ 249 if (key.offset == off && ret > 0 && path->slots[0] > 0) { 250 btrfs_item_key_to_cpu(path->nodes[0], &key, 251 path->slots[0] - 1); 252 if (key.type == BTRFS_EXTENT_DATA_KEY) 253 path->slots[0]--; 254 } 255 256 nritems = btrfs_header_nritems(path->nodes[0]); 257 process_slot: 258 if (path->slots[0] >= nritems) { 259 ret = btrfs_next_leaf(BTRFS_I(src)->root, path); 260 if (ret < 0) 261 goto out; 262 if (ret > 0) 263 break; 264 nritems = btrfs_header_nritems(path->nodes[0]); 265 } 266 leaf = path->nodes[0]; 267 slot = path->slots[0]; 268 269 btrfs_item_key_to_cpu(leaf, &key, slot); 270 if (key.type > BTRFS_EXTENT_DATA_KEY || 271 key.objectid != btrfs_ino(BTRFS_I(src))) 272 break; 273 274 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY); 275 276 extent = btrfs_item_ptr(leaf, slot, 277 struct btrfs_file_extent_item); 278 type = btrfs_file_extent_type(leaf, extent); 279 if (type == BTRFS_FILE_EXTENT_REG || 280 type == BTRFS_FILE_EXTENT_PREALLOC) { 281 disko = btrfs_file_extent_disk_bytenr(leaf, extent); 282 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent); 283 datao = btrfs_file_extent_offset(leaf, extent); 284 datal = btrfs_file_extent_num_bytes(leaf, extent); 285 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 286 /* Take upper bound, may be compressed */ 287 datal = btrfs_file_extent_ram_bytes(leaf, extent); 288 } 289 290 /* 291 * The first search might have left us at an extent item that 292 * ends before our target range's start, can happen if we have 293 * holes and NO_HOLES feature enabled. 294 */ 295 if (key.offset + datal <= off) { 296 path->slots[0]++; 297 goto process_slot; 298 } else if (key.offset >= off + len) { 299 break; 300 } 301 next_key_min_offset = key.offset + datal; 302 size = btrfs_item_size_nr(leaf, slot); 303 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot), 304 size); 305 306 btrfs_release_path(path); 307 path->leave_spinning = 0; 308 309 memcpy(&new_key, &key, sizeof(new_key)); 310 new_key.objectid = btrfs_ino(BTRFS_I(inode)); 311 if (off <= key.offset) 312 new_key.offset = key.offset + destoff - off; 313 else 314 new_key.offset = destoff; 315 316 /* 317 * Deal with a hole that doesn't have an extent item that 318 * represents it (NO_HOLES feature enabled). 319 * This hole is either in the middle of the cloning range or at 320 * the beginning (fully overlaps it or partially overlaps it). 321 */ 322 if (new_key.offset != last_dest_end) 323 drop_start = last_dest_end; 324 else 325 drop_start = new_key.offset; 326 327 if (type == BTRFS_FILE_EXTENT_REG || 328 type == BTRFS_FILE_EXTENT_PREALLOC) { 329 struct btrfs_clone_extent_info clone_info; 330 331 /* 332 * a | --- range to clone ---| b 333 * | ------------- extent ------------- | 334 */ 335 336 /* Subtract range b */ 337 if (key.offset + datal > off + len) 338 datal = off + len - key.offset; 339 340 /* Subtract range a */ 341 if (off > key.offset) { 342 datao += off - key.offset; 343 datal -= off - key.offset; 344 } 345 346 clone_info.disk_offset = disko; 347 clone_info.disk_len = diskl; 348 clone_info.data_offset = datao; 349 clone_info.data_len = datal; 350 clone_info.file_offset = new_key.offset; 351 clone_info.extent_buf = buf; 352 clone_info.item_size = size; 353 ret = btrfs_punch_hole_range(inode, path, drop_start, 354 new_key.offset + datal - 1, &clone_info, 355 &trans); 356 if (ret) 357 goto out; 358 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 359 /* 360 * Inline extents always have to start at file offset 0 361 * and can never be bigger then the sector size. We can 362 * never clone only parts of an inline extent, since all 363 * reflink operations must start at a sector size aligned 364 * offset, and the length must be aligned too or end at 365 * the i_size (which implies the whole inlined data). 366 */ 367 ASSERT(key.offset == 0); 368 ASSERT(datal <= fs_info->sectorsize); 369 if (key.offset != 0 || datal > fs_info->sectorsize) 370 return -EUCLEAN; 371 372 /* 373 * If our extent is inline, we know we will drop or 374 * adjust at most 1 extent item in the destination root. 375 * 376 * 1 - adjusting old extent (we may have to split it) 377 * 1 - add new extent 378 * 1 - inode update 379 */ 380 trans = btrfs_start_transaction(root, 3); 381 if (IS_ERR(trans)) { 382 ret = PTR_ERR(trans); 383 goto out; 384 } 385 386 ret = clone_copy_inline_extent(inode, trans, path, 387 &new_key, drop_start, 388 datal, size, buf); 389 if (ret) { 390 if (ret != -EOPNOTSUPP) 391 btrfs_abort_transaction(trans, ret); 392 btrfs_end_transaction(trans); 393 goto out; 394 } 395 } 396 397 btrfs_release_path(path); 398 399 last_dest_end = ALIGN(new_key.offset + datal, 400 fs_info->sectorsize); 401 ret = clone_finish_inode_update(trans, inode, last_dest_end, 402 destoff, olen, no_time_update); 403 if (ret) 404 goto out; 405 if (new_key.offset + datal >= destoff + len) 406 break; 407 408 btrfs_release_path(path); 409 key.offset = next_key_min_offset; 410 411 if (fatal_signal_pending(current)) { 412 ret = -EINTR; 413 goto out; 414 } 415 } 416 ret = 0; 417 418 if (last_dest_end < destoff + len) { 419 /* 420 * We have an implicit hole that fully or partially overlaps our 421 * cloning range at its end. This means that we either have the 422 * NO_HOLES feature enabled or the implicit hole happened due to 423 * mixing buffered and direct IO writes against this file. 424 */ 425 btrfs_release_path(path); 426 path->leave_spinning = 0; 427 428 ret = btrfs_punch_hole_range(inode, path, last_dest_end, 429 destoff + len - 1, NULL, &trans); 430 if (ret) 431 goto out; 432 433 ret = clone_finish_inode_update(trans, inode, destoff + len, 434 destoff, olen, no_time_update); 435 } 436 437 out: 438 btrfs_free_path(path); 439 kvfree(buf); 440 return ret; 441 } 442 443 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1, 444 struct inode *inode2, u64 loff2, u64 len) 445 { 446 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 447 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 448 } 449 450 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1, 451 struct inode *inode2, u64 loff2, u64 len) 452 { 453 if (inode1 < inode2) { 454 swap(inode1, inode2); 455 swap(loff1, loff2); 456 } else if (inode1 == inode2 && loff2 < loff1) { 457 swap(loff1, loff2); 458 } 459 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 460 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 461 } 462 463 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len, 464 struct inode *dst, u64 dst_loff) 465 { 466 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize; 467 int ret; 468 469 /* 470 * Lock destination range to serialize with concurrent readpages() and 471 * source range to serialize with relocation. 472 */ 473 btrfs_double_extent_lock(src, loff, dst, dst_loff, len); 474 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1); 475 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len); 476 477 return ret; 478 } 479 480 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen, 481 struct inode *dst, u64 dst_loff) 482 { 483 int ret; 484 u64 i, tail_len, chunk_count; 485 struct btrfs_root *root_dst = BTRFS_I(dst)->root; 486 487 spin_lock(&root_dst->root_item_lock); 488 if (root_dst->send_in_progress) { 489 btrfs_warn_rl(root_dst->fs_info, 490 "cannot deduplicate to root %llu while send operations are using it (%d in progress)", 491 root_dst->root_key.objectid, 492 root_dst->send_in_progress); 493 spin_unlock(&root_dst->root_item_lock); 494 return -EAGAIN; 495 } 496 root_dst->dedupe_in_progress++; 497 spin_unlock(&root_dst->root_item_lock); 498 499 tail_len = olen % BTRFS_MAX_DEDUPE_LEN; 500 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN); 501 502 for (i = 0; i < chunk_count; i++) { 503 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN, 504 dst, dst_loff); 505 if (ret) 506 goto out; 507 508 loff += BTRFS_MAX_DEDUPE_LEN; 509 dst_loff += BTRFS_MAX_DEDUPE_LEN; 510 } 511 512 if (tail_len > 0) 513 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff); 514 out: 515 spin_lock(&root_dst->root_item_lock); 516 root_dst->dedupe_in_progress--; 517 spin_unlock(&root_dst->root_item_lock); 518 519 return ret; 520 } 521 522 static noinline int btrfs_clone_files(struct file *file, struct file *file_src, 523 u64 off, u64 olen, u64 destoff) 524 { 525 struct inode *inode = file_inode(file); 526 struct inode *src = file_inode(file_src); 527 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 528 int ret; 529 u64 len = olen; 530 u64 bs = fs_info->sb->s_blocksize; 531 532 /* 533 * VFS's generic_remap_file_range_prep() protects us from cloning the 534 * eof block into the middle of a file, which would result in corruption 535 * if the file size is not blocksize aligned. So we don't need to check 536 * for that case here. 537 */ 538 if (off + len == src->i_size) 539 len = ALIGN(src->i_size, bs) - off; 540 541 if (destoff > inode->i_size) { 542 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs); 543 544 ret = btrfs_cont_expand(inode, inode->i_size, destoff); 545 if (ret) 546 return ret; 547 /* 548 * We may have truncated the last block if the inode's size is 549 * not sector size aligned, so we need to wait for writeback to 550 * complete before proceeding further, otherwise we can race 551 * with cloning and attempt to increment a reference to an 552 * extent that no longer exists (writeback completed right after 553 * we found the previous extent covering eof and before we 554 * attempted to increment its reference count). 555 */ 556 ret = btrfs_wait_ordered_range(inode, wb_start, 557 destoff - wb_start); 558 if (ret) 559 return ret; 560 } 561 562 /* 563 * Lock destination range to serialize with concurrent readpages() and 564 * source range to serialize with relocation. 565 */ 566 btrfs_double_extent_lock(src, off, inode, destoff, len); 567 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0); 568 btrfs_double_extent_unlock(src, off, inode, destoff, len); 569 /* 570 * Truncate page cache pages so that future reads will see the cloned 571 * data immediately and not the previous data. 572 */ 573 truncate_inode_pages_range(&inode->i_data, 574 round_down(destoff, PAGE_SIZE), 575 round_up(destoff + len, PAGE_SIZE) - 1); 576 577 return ret; 578 } 579 580 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in, 581 struct file *file_out, loff_t pos_out, 582 loff_t *len, unsigned int remap_flags) 583 { 584 struct inode *inode_in = file_inode(file_in); 585 struct inode *inode_out = file_inode(file_out); 586 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize; 587 bool same_inode = inode_out == inode_in; 588 u64 wb_len; 589 int ret; 590 591 if (!(remap_flags & REMAP_FILE_DEDUP)) { 592 struct btrfs_root *root_out = BTRFS_I(inode_out)->root; 593 594 if (btrfs_root_readonly(root_out)) 595 return -EROFS; 596 597 if (file_in->f_path.mnt != file_out->f_path.mnt || 598 inode_in->i_sb != inode_out->i_sb) 599 return -EXDEV; 600 } 601 602 /* Don't make the dst file partly checksummed */ 603 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) != 604 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) { 605 return -EINVAL; 606 } 607 608 /* 609 * Now that the inodes are locked, we need to start writeback ourselves 610 * and can not rely on the writeback from the VFS's generic helper 611 * generic_remap_file_range_prep() because: 612 * 613 * 1) For compression we must call filemap_fdatawrite_range() range 614 * twice (btrfs_fdatawrite_range() does it for us), and the generic 615 * helper only calls it once; 616 * 617 * 2) filemap_fdatawrite_range(), called by the generic helper only 618 * waits for the writeback to complete, i.e. for IO to be done, and 619 * not for the ordered extents to complete. We need to wait for them 620 * to complete so that new file extent items are in the fs tree. 621 */ 622 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP)) 623 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs); 624 else 625 wb_len = ALIGN(*len, bs); 626 627 /* 628 * Since we don't lock ranges, wait for ongoing lockless dio writes (as 629 * any in progress could create its ordered extents after we wait for 630 * existing ordered extents below). 631 */ 632 inode_dio_wait(inode_in); 633 if (!same_inode) 634 inode_dio_wait(inode_out); 635 636 /* 637 * Workaround to make sure NOCOW buffered write reach disk as NOCOW. 638 * 639 * Btrfs' back references do not have a block level granularity, they 640 * work at the whole extent level. 641 * NOCOW buffered write without data space reserved may not be able 642 * to fall back to CoW due to lack of data space, thus could cause 643 * data loss. 644 * 645 * Here we take a shortcut by flushing the whole inode, so that all 646 * nocow write should reach disk as nocow before we increase the 647 * reference of the extent. We could do better by only flushing NOCOW 648 * data, but that needs extra accounting. 649 * 650 * Also we don't need to check ASYNC_EXTENT, as async extent will be 651 * CoWed anyway, not affecting nocow part. 652 */ 653 ret = filemap_flush(inode_in->i_mapping); 654 if (ret < 0) 655 return ret; 656 657 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs), 658 wb_len); 659 if (ret < 0) 660 return ret; 661 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs), 662 wb_len); 663 if (ret < 0) 664 return ret; 665 666 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out, 667 len, remap_flags); 668 } 669 670 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off, 671 struct file *dst_file, loff_t destoff, loff_t len, 672 unsigned int remap_flags) 673 { 674 struct inode *src_inode = file_inode(src_file); 675 struct inode *dst_inode = file_inode(dst_file); 676 bool same_inode = dst_inode == src_inode; 677 int ret; 678 679 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) 680 return -EINVAL; 681 682 if (same_inode) 683 inode_lock(src_inode); 684 else 685 lock_two_nondirectories(src_inode, dst_inode); 686 687 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff, 688 &len, remap_flags); 689 if (ret < 0 || len == 0) 690 goto out_unlock; 691 692 if (remap_flags & REMAP_FILE_DEDUP) 693 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff); 694 else 695 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff); 696 697 out_unlock: 698 if (same_inode) 699 inode_unlock(src_inode); 700 else 701 unlock_two_nondirectories(src_inode, dst_inode); 702 703 return ret < 0 ? ret : len; 704 } 705