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 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(inode, 0); 35 } 36 37 ret = btrfs_update_inode(trans, root, 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 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 set_page_extent_mapped(page); 85 clear_extent_bit(&inode->io_tree, file_offset, range_end, 86 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 87 0, 0, NULL); 88 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL); 89 if (ret) 90 goto out_unlock; 91 92 /* 93 * After dirtying the page our caller will need to start a transaction, 94 * and if we are low on metadata free space, that can cause flushing of 95 * delalloc for all inodes in order to get metadata space released. 96 * However we are holding the range locked for the whole duration of 97 * the clone/dedupe operation, so we may deadlock if that happens and no 98 * other task releases enough space. So mark this inode as not being 99 * possible to flush to avoid such deadlock. We will clear that flag 100 * when we finish cloning all extents, since a transaction is started 101 * after finding each extent to clone. 102 */ 103 set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &inode->runtime_flags); 104 105 if (comp_type == BTRFS_COMPRESS_NONE) { 106 char *map; 107 108 map = kmap(page); 109 memcpy(map, data_start, datal); 110 flush_dcache_page(page); 111 kunmap(page); 112 } else { 113 ret = btrfs_decompress(comp_type, data_start, page, 0, 114 inline_size, datal); 115 if (ret) 116 goto out_unlock; 117 flush_dcache_page(page); 118 } 119 120 /* 121 * If our inline data is smaller then the block/page size, then the 122 * remaining of the block/page is equivalent to zeroes. We had something 123 * like the following done: 124 * 125 * $ xfs_io -f -c "pwrite -S 0xab 0 500" file 126 * $ sync # (or fsync) 127 * $ xfs_io -c "falloc 0 4K" file 128 * $ xfs_io -c "pwrite -S 0xcd 4K 4K" 129 * 130 * So what's in the range [500, 4095] corresponds to zeroes. 131 */ 132 if (datal < block_size) { 133 char *map; 134 135 map = kmap(page); 136 memset(map + datal, 0, block_size - datal); 137 flush_dcache_page(page); 138 kunmap(page); 139 } 140 141 SetPageUptodate(page); 142 ClearPageChecked(page); 143 set_page_dirty(page); 144 out_unlock: 145 if (page) { 146 unlock_page(page); 147 put_page(page); 148 } 149 if (ret) 150 btrfs_delalloc_release_space(inode, data_reserved, file_offset, 151 block_size, true); 152 btrfs_delalloc_release_extents(inode, block_size); 153 out: 154 extent_changeset_free(data_reserved); 155 156 return ret; 157 } 158 159 /* 160 * Deal with cloning of inline extents. We try to copy the inline extent from 161 * the source inode to destination inode when possible. When not possible we 162 * copy the inline extent's data into the respective page of the inode. 163 */ 164 static int clone_copy_inline_extent(struct inode *dst, 165 struct btrfs_path *path, 166 struct btrfs_key *new_key, 167 const u64 drop_start, 168 const u64 datal, 169 const u64 size, 170 const u8 comp_type, 171 char *inline_data, 172 struct btrfs_trans_handle **trans_out) 173 { 174 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb); 175 struct btrfs_root *root = BTRFS_I(dst)->root; 176 const u64 aligned_end = ALIGN(new_key->offset + datal, 177 fs_info->sectorsize); 178 struct btrfs_trans_handle *trans = NULL; 179 int ret; 180 struct btrfs_key key; 181 182 if (new_key->offset > 0) { 183 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset, 184 inline_data, size, datal, comp_type); 185 goto out; 186 } 187 188 key.objectid = btrfs_ino(BTRFS_I(dst)); 189 key.type = BTRFS_EXTENT_DATA_KEY; 190 key.offset = 0; 191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 192 if (ret < 0) { 193 return ret; 194 } else if (ret > 0) { 195 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 196 ret = btrfs_next_leaf(root, path); 197 if (ret < 0) 198 return ret; 199 else if (ret > 0) 200 goto copy_inline_extent; 201 } 202 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 203 if (key.objectid == btrfs_ino(BTRFS_I(dst)) && 204 key.type == BTRFS_EXTENT_DATA_KEY) { 205 /* 206 * There's an implicit hole at file offset 0, copy the 207 * inline extent's data to the page. 208 */ 209 ASSERT(key.offset > 0); 210 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset, 211 inline_data, size, datal, 212 comp_type); 213 goto out; 214 } 215 } else if (i_size_read(dst) <= datal) { 216 struct btrfs_file_extent_item *ei; 217 218 ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 219 struct btrfs_file_extent_item); 220 /* 221 * If it's an inline extent replace it with the source inline 222 * extent, otherwise copy the source inline extent data into 223 * the respective page at the destination inode. 224 */ 225 if (btrfs_file_extent_type(path->nodes[0], ei) == 226 BTRFS_FILE_EXTENT_INLINE) 227 goto copy_inline_extent; 228 229 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset, 230 inline_data, size, datal, comp_type); 231 goto out; 232 } 233 234 copy_inline_extent: 235 ret = 0; 236 /* 237 * We have no extent items, or we have an extent at offset 0 which may 238 * or may not be inlined. All these cases are dealt the same way. 239 */ 240 if (i_size_read(dst) > datal) { 241 /* 242 * At the destination offset 0 we have either a hole, a regular 243 * extent or an inline extent larger then the one we want to 244 * clone. Deal with all these cases by copying the inline extent 245 * data into the respective page at the destination inode. 246 */ 247 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset, 248 inline_data, size, datal, comp_type); 249 goto out; 250 } 251 252 btrfs_release_path(path); 253 /* 254 * If we end up here it means were copy the inline extent into a leaf 255 * of the destination inode. We know we will drop or adjust at most one 256 * extent item in the destination root. 257 * 258 * 1 unit - adjusting old extent (we may have to split it) 259 * 1 unit - add new extent 260 * 1 unit - inode update 261 */ 262 trans = btrfs_start_transaction(root, 3); 263 if (IS_ERR(trans)) { 264 ret = PTR_ERR(trans); 265 trans = NULL; 266 goto out; 267 } 268 ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1); 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 inode_add_bytes(dst, datal); 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 * Release path before starting a new transaction so we don't 286 * hold locks that would confuse lockdep. 287 */ 288 btrfs_release_path(path); 289 /* 290 * No transaction here means we copied the inline extent into a 291 * page of the destination inode. 292 * 293 * 1 unit to update inode item 294 */ 295 trans = btrfs_start_transaction(root, 1); 296 if (IS_ERR(trans)) { 297 ret = PTR_ERR(trans); 298 trans = NULL; 299 } 300 } 301 if (ret && trans) { 302 btrfs_abort_transaction(trans, ret); 303 btrfs_end_transaction(trans); 304 } 305 if (!ret) 306 *trans_out = trans; 307 308 return ret; 309 } 310 311 /** 312 * btrfs_clone() - clone a range from inode file to another 313 * 314 * @src: Inode to clone from 315 * @inode: Inode to clone to 316 * @off: Offset within source to start clone from 317 * @olen: Original length, passed by user, of range to clone 318 * @olen_aligned: Block-aligned value of olen 319 * @destoff: Offset within @inode to start clone 320 * @no_time_update: Whether to update mtime/ctime on the target inode 321 */ 322 static int btrfs_clone(struct inode *src, struct inode *inode, 323 const u64 off, const u64 olen, const u64 olen_aligned, 324 const u64 destoff, int no_time_update) 325 { 326 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 327 struct btrfs_path *path = NULL; 328 struct extent_buffer *leaf; 329 struct btrfs_trans_handle *trans; 330 char *buf = NULL; 331 struct btrfs_key key; 332 u32 nritems; 333 int slot; 334 int ret; 335 const u64 len = olen_aligned; 336 u64 last_dest_end = destoff; 337 338 ret = -ENOMEM; 339 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); 340 if (!buf) 341 return ret; 342 343 path = btrfs_alloc_path(); 344 if (!path) { 345 kvfree(buf); 346 return ret; 347 } 348 349 path->reada = READA_FORWARD; 350 /* Clone data */ 351 key.objectid = btrfs_ino(BTRFS_I(src)); 352 key.type = BTRFS_EXTENT_DATA_KEY; 353 key.offset = off; 354 355 while (1) { 356 u64 next_key_min_offset = key.offset + 1; 357 struct btrfs_file_extent_item *extent; 358 u64 extent_gen; 359 int type; 360 u32 size; 361 struct btrfs_key new_key; 362 u64 disko = 0, diskl = 0; 363 u64 datao = 0, datal = 0; 364 u8 comp; 365 u64 drop_start; 366 367 /* Note the key will change type as we walk through the tree */ 368 path->leave_spinning = 1; 369 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 370 0, 0); 371 if (ret < 0) 372 goto out; 373 /* 374 * First search, if no extent item that starts at offset off was 375 * found but the previous item is an extent item, it's possible 376 * it might overlap our target range, therefore process it. 377 */ 378 if (key.offset == off && ret > 0 && path->slots[0] > 0) { 379 btrfs_item_key_to_cpu(path->nodes[0], &key, 380 path->slots[0] - 1); 381 if (key.type == BTRFS_EXTENT_DATA_KEY) 382 path->slots[0]--; 383 } 384 385 nritems = btrfs_header_nritems(path->nodes[0]); 386 process_slot: 387 if (path->slots[0] >= nritems) { 388 ret = btrfs_next_leaf(BTRFS_I(src)->root, path); 389 if (ret < 0) 390 goto out; 391 if (ret > 0) 392 break; 393 nritems = btrfs_header_nritems(path->nodes[0]); 394 } 395 leaf = path->nodes[0]; 396 slot = path->slots[0]; 397 398 btrfs_item_key_to_cpu(leaf, &key, slot); 399 if (key.type > BTRFS_EXTENT_DATA_KEY || 400 key.objectid != btrfs_ino(BTRFS_I(src))) 401 break; 402 403 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY); 404 405 extent = btrfs_item_ptr(leaf, slot, 406 struct btrfs_file_extent_item); 407 extent_gen = btrfs_file_extent_generation(leaf, extent); 408 comp = btrfs_file_extent_compression(leaf, extent); 409 type = btrfs_file_extent_type(leaf, extent); 410 if (type == BTRFS_FILE_EXTENT_REG || 411 type == BTRFS_FILE_EXTENT_PREALLOC) { 412 disko = btrfs_file_extent_disk_bytenr(leaf, extent); 413 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent); 414 datao = btrfs_file_extent_offset(leaf, extent); 415 datal = btrfs_file_extent_num_bytes(leaf, extent); 416 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 417 /* Take upper bound, may be compressed */ 418 datal = btrfs_file_extent_ram_bytes(leaf, extent); 419 } 420 421 /* 422 * The first search might have left us at an extent item that 423 * ends before our target range's start, can happen if we have 424 * holes and NO_HOLES feature enabled. 425 */ 426 if (key.offset + datal <= off) { 427 path->slots[0]++; 428 goto process_slot; 429 } else if (key.offset >= off + len) { 430 break; 431 } 432 next_key_min_offset = key.offset + datal; 433 size = btrfs_item_size_nr(leaf, slot); 434 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot), 435 size); 436 437 btrfs_release_path(path); 438 path->leave_spinning = 0; 439 440 memcpy(&new_key, &key, sizeof(new_key)); 441 new_key.objectid = btrfs_ino(BTRFS_I(inode)); 442 if (off <= key.offset) 443 new_key.offset = key.offset + destoff - off; 444 else 445 new_key.offset = destoff; 446 447 /* 448 * Deal with a hole that doesn't have an extent item that 449 * represents it (NO_HOLES feature enabled). 450 * This hole is either in the middle of the cloning range or at 451 * the beginning (fully overlaps it or partially overlaps it). 452 */ 453 if (new_key.offset != last_dest_end) 454 drop_start = last_dest_end; 455 else 456 drop_start = new_key.offset; 457 458 if (type == BTRFS_FILE_EXTENT_REG || 459 type == BTRFS_FILE_EXTENT_PREALLOC) { 460 struct btrfs_replace_extent_info clone_info; 461 462 /* 463 * a | --- range to clone ---| b 464 * | ------------- extent ------------- | 465 */ 466 467 /* Subtract range b */ 468 if (key.offset + datal > off + len) 469 datal = off + len - key.offset; 470 471 /* Subtract range a */ 472 if (off > key.offset) { 473 datao += off - key.offset; 474 datal -= off - key.offset; 475 } 476 477 clone_info.disk_offset = disko; 478 clone_info.disk_len = diskl; 479 clone_info.data_offset = datao; 480 clone_info.data_len = datal; 481 clone_info.file_offset = new_key.offset; 482 clone_info.extent_buf = buf; 483 clone_info.is_new_extent = false; 484 ret = btrfs_replace_file_extents(inode, path, drop_start, 485 new_key.offset + datal - 1, &clone_info, 486 &trans); 487 if (ret) 488 goto out; 489 } else if (type == BTRFS_FILE_EXTENT_INLINE) { 490 /* 491 * Inline extents always have to start at file offset 0 492 * and can never be bigger then the sector size. We can 493 * never clone only parts of an inline extent, since all 494 * reflink operations must start at a sector size aligned 495 * offset, and the length must be aligned too or end at 496 * the i_size (which implies the whole inlined data). 497 */ 498 ASSERT(key.offset == 0); 499 ASSERT(datal <= fs_info->sectorsize); 500 if (key.offset != 0 || datal > fs_info->sectorsize) 501 return -EUCLEAN; 502 503 ret = clone_copy_inline_extent(inode, path, &new_key, 504 drop_start, datal, size, 505 comp, buf, &trans); 506 if (ret) 507 goto out; 508 } 509 510 btrfs_release_path(path); 511 512 /* 513 * If this is a new extent update the last_reflink_trans of both 514 * inodes. This is used by fsync to make sure it does not log 515 * multiple checksum items with overlapping ranges. For older 516 * extents we don't need to do it since inode logging skips the 517 * checksums for older extents. Also ignore holes and inline 518 * extents because they don't have checksums in the csum tree. 519 */ 520 if (extent_gen == trans->transid && disko > 0) { 521 BTRFS_I(src)->last_reflink_trans = trans->transid; 522 BTRFS_I(inode)->last_reflink_trans = trans->transid; 523 } 524 525 last_dest_end = ALIGN(new_key.offset + datal, 526 fs_info->sectorsize); 527 ret = clone_finish_inode_update(trans, inode, last_dest_end, 528 destoff, olen, no_time_update); 529 if (ret) 530 goto out; 531 if (new_key.offset + datal >= destoff + len) 532 break; 533 534 btrfs_release_path(path); 535 key.offset = next_key_min_offset; 536 537 if (fatal_signal_pending(current)) { 538 ret = -EINTR; 539 goto out; 540 } 541 542 cond_resched(); 543 } 544 ret = 0; 545 546 if (last_dest_end < destoff + len) { 547 /* 548 * We have an implicit hole that fully or partially overlaps our 549 * cloning range at its end. This means that we either have the 550 * NO_HOLES feature enabled or the implicit hole happened due to 551 * mixing buffered and direct IO writes against this file. 552 */ 553 btrfs_release_path(path); 554 path->leave_spinning = 0; 555 556 /* 557 * When using NO_HOLES and we are cloning a range that covers 558 * only a hole (no extents) into a range beyond the current 559 * i_size, punching a hole in the target range will not create 560 * an extent map defining a hole, because the range starts at or 561 * beyond current i_size. If the file previously had an i_size 562 * greater than the new i_size set by this clone operation, we 563 * need to make sure the next fsync is a full fsync, so that it 564 * detects and logs a hole covering a range from the current 565 * i_size to the new i_size. If the clone range covers extents, 566 * besides a hole, then we know the full sync flag was already 567 * set by previous calls to btrfs_replace_file_extents() that 568 * replaced file extent items. 569 */ 570 if (last_dest_end >= i_size_read(inode)) 571 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, 572 &BTRFS_I(inode)->runtime_flags); 573 574 ret = btrfs_replace_file_extents(inode, path, last_dest_end, 575 destoff + len - 1, NULL, &trans); 576 if (ret) 577 goto out; 578 579 ret = clone_finish_inode_update(trans, inode, destoff + len, 580 destoff, olen, no_time_update); 581 } 582 583 out: 584 btrfs_free_path(path); 585 kvfree(buf); 586 clear_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &BTRFS_I(inode)->runtime_flags); 587 588 return ret; 589 } 590 591 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1, 592 struct inode *inode2, u64 loff2, u64 len) 593 { 594 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 595 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 596 } 597 598 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1, 599 struct inode *inode2, u64 loff2, u64 len) 600 { 601 if (inode1 < inode2) { 602 swap(inode1, inode2); 603 swap(loff1, loff2); 604 } else if (inode1 == inode2 && loff2 < loff1) { 605 swap(loff1, loff2); 606 } 607 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 608 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 609 } 610 611 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len, 612 struct inode *dst, u64 dst_loff) 613 { 614 const u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize; 615 int ret; 616 617 /* 618 * Lock destination range to serialize with concurrent readpages() and 619 * source range to serialize with relocation. 620 */ 621 btrfs_double_extent_lock(src, loff, dst, dst_loff, len); 622 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1); 623 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len); 624 625 return ret; 626 } 627 628 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen, 629 struct inode *dst, u64 dst_loff) 630 { 631 int ret; 632 u64 i, tail_len, chunk_count; 633 struct btrfs_root *root_dst = BTRFS_I(dst)->root; 634 635 spin_lock(&root_dst->root_item_lock); 636 if (root_dst->send_in_progress) { 637 btrfs_warn_rl(root_dst->fs_info, 638 "cannot deduplicate to root %llu while send operations are using it (%d in progress)", 639 root_dst->root_key.objectid, 640 root_dst->send_in_progress); 641 spin_unlock(&root_dst->root_item_lock); 642 return -EAGAIN; 643 } 644 root_dst->dedupe_in_progress++; 645 spin_unlock(&root_dst->root_item_lock); 646 647 tail_len = olen % BTRFS_MAX_DEDUPE_LEN; 648 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN); 649 650 for (i = 0; i < chunk_count; i++) { 651 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN, 652 dst, dst_loff); 653 if (ret) 654 goto out; 655 656 loff += BTRFS_MAX_DEDUPE_LEN; 657 dst_loff += BTRFS_MAX_DEDUPE_LEN; 658 } 659 660 if (tail_len > 0) 661 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff); 662 out: 663 spin_lock(&root_dst->root_item_lock); 664 root_dst->dedupe_in_progress--; 665 spin_unlock(&root_dst->root_item_lock); 666 667 return ret; 668 } 669 670 static noinline int btrfs_clone_files(struct file *file, struct file *file_src, 671 u64 off, u64 olen, u64 destoff) 672 { 673 struct inode *inode = file_inode(file); 674 struct inode *src = file_inode(file_src); 675 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 676 int ret; 677 int wb_ret; 678 u64 len = olen; 679 u64 bs = fs_info->sb->s_blocksize; 680 681 /* 682 * VFS's generic_remap_file_range_prep() protects us from cloning the 683 * eof block into the middle of a file, which would result in corruption 684 * if the file size is not blocksize aligned. So we don't need to check 685 * for that case here. 686 */ 687 if (off + len == src->i_size) 688 len = ALIGN(src->i_size, bs) - off; 689 690 if (destoff > inode->i_size) { 691 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs); 692 693 ret = btrfs_cont_expand(inode, inode->i_size, destoff); 694 if (ret) 695 return ret; 696 /* 697 * We may have truncated the last block if the inode's size is 698 * not sector size aligned, so we need to wait for writeback to 699 * complete before proceeding further, otherwise we can race 700 * with cloning and attempt to increment a reference to an 701 * extent that no longer exists (writeback completed right after 702 * we found the previous extent covering eof and before we 703 * attempted to increment its reference count). 704 */ 705 ret = btrfs_wait_ordered_range(inode, wb_start, 706 destoff - wb_start); 707 if (ret) 708 return ret; 709 } 710 711 /* 712 * Lock destination range to serialize with concurrent readpages() and 713 * source range to serialize with relocation. 714 */ 715 btrfs_double_extent_lock(src, off, inode, destoff, len); 716 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0); 717 btrfs_double_extent_unlock(src, off, inode, destoff, len); 718 719 /* 720 * We may have copied an inline extent into a page of the destination 721 * range, so wait for writeback to complete before truncating pages 722 * from the page cache. This is a rare case. 723 */ 724 wb_ret = btrfs_wait_ordered_range(inode, destoff, len); 725 ret = ret ? ret : wb_ret; 726 /* 727 * Truncate page cache pages so that future reads will see the cloned 728 * data immediately and not the previous data. 729 */ 730 truncate_inode_pages_range(&inode->i_data, 731 round_down(destoff, PAGE_SIZE), 732 round_up(destoff + len, PAGE_SIZE) - 1); 733 734 return ret; 735 } 736 737 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in, 738 struct file *file_out, loff_t pos_out, 739 loff_t *len, unsigned int remap_flags) 740 { 741 struct inode *inode_in = file_inode(file_in); 742 struct inode *inode_out = file_inode(file_out); 743 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize; 744 bool same_inode = inode_out == inode_in; 745 u64 wb_len; 746 int ret; 747 748 if (!(remap_flags & REMAP_FILE_DEDUP)) { 749 struct btrfs_root *root_out = BTRFS_I(inode_out)->root; 750 751 if (btrfs_root_readonly(root_out)) 752 return -EROFS; 753 754 if (file_in->f_path.mnt != file_out->f_path.mnt || 755 inode_in->i_sb != inode_out->i_sb) 756 return -EXDEV; 757 } 758 759 /* Don't make the dst file partly checksummed */ 760 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) != 761 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) { 762 return -EINVAL; 763 } 764 765 /* 766 * Now that the inodes are locked, we need to start writeback ourselves 767 * and can not rely on the writeback from the VFS's generic helper 768 * generic_remap_file_range_prep() because: 769 * 770 * 1) For compression we must call filemap_fdatawrite_range() range 771 * twice (btrfs_fdatawrite_range() does it for us), and the generic 772 * helper only calls it once; 773 * 774 * 2) filemap_fdatawrite_range(), called by the generic helper only 775 * waits for the writeback to complete, i.e. for IO to be done, and 776 * not for the ordered extents to complete. We need to wait for them 777 * to complete so that new file extent items are in the fs tree. 778 */ 779 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP)) 780 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs); 781 else 782 wb_len = ALIGN(*len, bs); 783 784 /* 785 * Since we don't lock ranges, wait for ongoing lockless dio writes (as 786 * any in progress could create its ordered extents after we wait for 787 * existing ordered extents below). 788 */ 789 inode_dio_wait(inode_in); 790 if (!same_inode) 791 inode_dio_wait(inode_out); 792 793 /* 794 * Workaround to make sure NOCOW buffered write reach disk as NOCOW. 795 * 796 * Btrfs' back references do not have a block level granularity, they 797 * work at the whole extent level. 798 * NOCOW buffered write without data space reserved may not be able 799 * to fall back to CoW due to lack of data space, thus could cause 800 * data loss. 801 * 802 * Here we take a shortcut by flushing the whole inode, so that all 803 * nocow write should reach disk as nocow before we increase the 804 * reference of the extent. We could do better by only flushing NOCOW 805 * data, but that needs extra accounting. 806 * 807 * Also we don't need to check ASYNC_EXTENT, as async extent will be 808 * CoWed anyway, not affecting nocow part. 809 */ 810 ret = filemap_flush(inode_in->i_mapping); 811 if (ret < 0) 812 return ret; 813 814 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs), 815 wb_len); 816 if (ret < 0) 817 return ret; 818 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs), 819 wb_len); 820 if (ret < 0) 821 return ret; 822 823 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out, 824 len, remap_flags); 825 } 826 827 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off, 828 struct file *dst_file, loff_t destoff, loff_t len, 829 unsigned int remap_flags) 830 { 831 struct inode *src_inode = file_inode(src_file); 832 struct inode *dst_inode = file_inode(dst_file); 833 bool same_inode = dst_inode == src_inode; 834 int ret; 835 836 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) 837 return -EINVAL; 838 839 if (same_inode) 840 inode_lock(src_inode); 841 else 842 lock_two_nondirectories(src_inode, dst_inode); 843 844 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff, 845 &len, remap_flags); 846 if (ret < 0 || len == 0) 847 goto out_unlock; 848 849 if (remap_flags & REMAP_FILE_DEDUP) 850 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff); 851 else 852 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff); 853 854 out_unlock: 855 if (same_inode) 856 inode_unlock(src_inode); 857 else 858 unlock_two_nondirectories(src_inode, dst_inode); 859 860 return ret < 0 ? ret : len; 861 } 862