1 // SPDX-License-Identifier: LGPL-2.1 2 /* 3 * Copyright IBM Corporation, 2007 4 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> 5 * 6 */ 7 8 #include <linux/slab.h> 9 #include "ext4_jbd2.h" 10 #include "ext4_extents.h" 11 12 /* 13 * The contiguous blocks details which can be 14 * represented by a single extent 15 */ 16 struct migrate_struct { 17 ext4_lblk_t first_block, last_block, curr_block; 18 ext4_fsblk_t first_pblock, last_pblock; 19 }; 20 21 static int finish_range(handle_t *handle, struct inode *inode, 22 struct migrate_struct *lb) 23 24 { 25 int retval = 0, needed; 26 struct ext4_extent newext; 27 struct ext4_ext_path *path; 28 if (lb->first_pblock == 0) 29 return 0; 30 31 /* Add the extent to temp inode*/ 32 newext.ee_block = cpu_to_le32(lb->first_block); 33 newext.ee_len = cpu_to_le16(lb->last_block - lb->first_block + 1); 34 ext4_ext_store_pblock(&newext, lb->first_pblock); 35 /* Locking only for convinience since we are operating on temp inode */ 36 down_write(&EXT4_I(inode)->i_data_sem); 37 path = ext4_find_extent(inode, lb->first_block, NULL, 0); 38 if (IS_ERR(path)) { 39 retval = PTR_ERR(path); 40 path = NULL; 41 goto err_out; 42 } 43 44 /* 45 * Calculate the credit needed to inserting this extent 46 * Since we are doing this in loop we may accumalate extra 47 * credit. But below we try to not accumalate too much 48 * of them by restarting the journal. 49 */ 50 needed = ext4_ext_calc_credits_for_single_extent(inode, 51 lb->last_block - lb->first_block + 1, path); 52 53 retval = ext4_datasem_ensure_credits(handle, inode, needed, needed, 0); 54 if (retval < 0) 55 goto err_out; 56 retval = ext4_ext_insert_extent(handle, inode, &path, &newext, 0); 57 err_out: 58 up_write((&EXT4_I(inode)->i_data_sem)); 59 ext4_ext_drop_refs(path); 60 kfree(path); 61 lb->first_pblock = 0; 62 return retval; 63 } 64 65 static int update_extent_range(handle_t *handle, struct inode *inode, 66 ext4_fsblk_t pblock, struct migrate_struct *lb) 67 { 68 int retval; 69 /* 70 * See if we can add on to the existing range (if it exists) 71 */ 72 if (lb->first_pblock && 73 (lb->last_pblock+1 == pblock) && 74 (lb->last_block+1 == lb->curr_block)) { 75 lb->last_pblock = pblock; 76 lb->last_block = lb->curr_block; 77 lb->curr_block++; 78 return 0; 79 } 80 /* 81 * Start a new range. 82 */ 83 retval = finish_range(handle, inode, lb); 84 lb->first_pblock = lb->last_pblock = pblock; 85 lb->first_block = lb->last_block = lb->curr_block; 86 lb->curr_block++; 87 return retval; 88 } 89 90 static int update_ind_extent_range(handle_t *handle, struct inode *inode, 91 ext4_fsblk_t pblock, 92 struct migrate_struct *lb) 93 { 94 struct buffer_head *bh; 95 __le32 *i_data; 96 int i, retval = 0; 97 unsigned long max_entries = inode->i_sb->s_blocksize >> 2; 98 99 bh = ext4_sb_bread(inode->i_sb, pblock, 0); 100 if (IS_ERR(bh)) 101 return PTR_ERR(bh); 102 103 i_data = (__le32 *)bh->b_data; 104 for (i = 0; i < max_entries; i++) { 105 if (i_data[i]) { 106 retval = update_extent_range(handle, inode, 107 le32_to_cpu(i_data[i]), lb); 108 if (retval) 109 break; 110 } else { 111 lb->curr_block++; 112 } 113 } 114 put_bh(bh); 115 return retval; 116 117 } 118 119 static int update_dind_extent_range(handle_t *handle, struct inode *inode, 120 ext4_fsblk_t pblock, 121 struct migrate_struct *lb) 122 { 123 struct buffer_head *bh; 124 __le32 *i_data; 125 int i, retval = 0; 126 unsigned long max_entries = inode->i_sb->s_blocksize >> 2; 127 128 bh = ext4_sb_bread(inode->i_sb, pblock, 0); 129 if (IS_ERR(bh)) 130 return PTR_ERR(bh); 131 132 i_data = (__le32 *)bh->b_data; 133 for (i = 0; i < max_entries; i++) { 134 if (i_data[i]) { 135 retval = update_ind_extent_range(handle, inode, 136 le32_to_cpu(i_data[i]), lb); 137 if (retval) 138 break; 139 } else { 140 /* Only update the file block number */ 141 lb->curr_block += max_entries; 142 } 143 } 144 put_bh(bh); 145 return retval; 146 147 } 148 149 static int update_tind_extent_range(handle_t *handle, struct inode *inode, 150 ext4_fsblk_t pblock, 151 struct migrate_struct *lb) 152 { 153 struct buffer_head *bh; 154 __le32 *i_data; 155 int i, retval = 0; 156 unsigned long max_entries = inode->i_sb->s_blocksize >> 2; 157 158 bh = ext4_sb_bread(inode->i_sb, pblock, 0); 159 if (IS_ERR(bh)) 160 return PTR_ERR(bh); 161 162 i_data = (__le32 *)bh->b_data; 163 for (i = 0; i < max_entries; i++) { 164 if (i_data[i]) { 165 retval = update_dind_extent_range(handle, inode, 166 le32_to_cpu(i_data[i]), lb); 167 if (retval) 168 break; 169 } else { 170 /* Only update the file block number */ 171 lb->curr_block += max_entries * max_entries; 172 } 173 } 174 put_bh(bh); 175 return retval; 176 177 } 178 179 static int free_dind_blocks(handle_t *handle, 180 struct inode *inode, __le32 i_data) 181 { 182 int i; 183 __le32 *tmp_idata; 184 struct buffer_head *bh; 185 struct super_block *sb = inode->i_sb; 186 unsigned long max_entries = inode->i_sb->s_blocksize >> 2; 187 int err; 188 189 bh = ext4_sb_bread(sb, le32_to_cpu(i_data), 0); 190 if (IS_ERR(bh)) 191 return PTR_ERR(bh); 192 193 tmp_idata = (__le32 *)bh->b_data; 194 for (i = 0; i < max_entries; i++) { 195 if (tmp_idata[i]) { 196 err = ext4_journal_ensure_credits(handle, 197 EXT4_RESERVE_TRANS_BLOCKS, 198 ext4_free_metadata_revoke_credits(sb, 1)); 199 if (err < 0) { 200 put_bh(bh); 201 return err; 202 } 203 ext4_free_blocks(handle, inode, NULL, 204 le32_to_cpu(tmp_idata[i]), 1, 205 EXT4_FREE_BLOCKS_METADATA | 206 EXT4_FREE_BLOCKS_FORGET); 207 } 208 } 209 put_bh(bh); 210 err = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS, 211 ext4_free_metadata_revoke_credits(sb, 1)); 212 if (err < 0) 213 return err; 214 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1, 215 EXT4_FREE_BLOCKS_METADATA | 216 EXT4_FREE_BLOCKS_FORGET); 217 return 0; 218 } 219 220 static int free_tind_blocks(handle_t *handle, 221 struct inode *inode, __le32 i_data) 222 { 223 int i, retval = 0; 224 __le32 *tmp_idata; 225 struct buffer_head *bh; 226 unsigned long max_entries = inode->i_sb->s_blocksize >> 2; 227 228 bh = ext4_sb_bread(inode->i_sb, le32_to_cpu(i_data), 0); 229 if (IS_ERR(bh)) 230 return PTR_ERR(bh); 231 232 tmp_idata = (__le32 *)bh->b_data; 233 for (i = 0; i < max_entries; i++) { 234 if (tmp_idata[i]) { 235 retval = free_dind_blocks(handle, 236 inode, tmp_idata[i]); 237 if (retval) { 238 put_bh(bh); 239 return retval; 240 } 241 } 242 } 243 put_bh(bh); 244 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS, 245 ext4_free_metadata_revoke_credits(inode->i_sb, 1)); 246 if (retval < 0) 247 return retval; 248 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1, 249 EXT4_FREE_BLOCKS_METADATA | 250 EXT4_FREE_BLOCKS_FORGET); 251 return 0; 252 } 253 254 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data) 255 { 256 int retval; 257 258 /* ei->i_data[EXT4_IND_BLOCK] */ 259 if (i_data[0]) { 260 retval = ext4_journal_ensure_credits(handle, 261 EXT4_RESERVE_TRANS_BLOCKS, 262 ext4_free_metadata_revoke_credits(inode->i_sb, 1)); 263 if (retval < 0) 264 return retval; 265 ext4_free_blocks(handle, inode, NULL, 266 le32_to_cpu(i_data[0]), 1, 267 EXT4_FREE_BLOCKS_METADATA | 268 EXT4_FREE_BLOCKS_FORGET); 269 } 270 271 /* ei->i_data[EXT4_DIND_BLOCK] */ 272 if (i_data[1]) { 273 retval = free_dind_blocks(handle, inode, i_data[1]); 274 if (retval) 275 return retval; 276 } 277 278 /* ei->i_data[EXT4_TIND_BLOCK] */ 279 if (i_data[2]) { 280 retval = free_tind_blocks(handle, inode, i_data[2]); 281 if (retval) 282 return retval; 283 } 284 return 0; 285 } 286 287 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode, 288 struct inode *tmp_inode) 289 { 290 int retval; 291 __le32 i_data[3]; 292 struct ext4_inode_info *ei = EXT4_I(inode); 293 struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode); 294 295 /* 296 * One credit accounted for writing the 297 * i_data field of the original inode 298 */ 299 retval = ext4_journal_ensure_credits(handle, 1, 0); 300 if (retval < 0) 301 goto err_out; 302 303 i_data[0] = ei->i_data[EXT4_IND_BLOCK]; 304 i_data[1] = ei->i_data[EXT4_DIND_BLOCK]; 305 i_data[2] = ei->i_data[EXT4_TIND_BLOCK]; 306 307 down_write(&EXT4_I(inode)->i_data_sem); 308 /* 309 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation 310 * happened after we started the migrate. We need to 311 * fail the migrate 312 */ 313 if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) { 314 retval = -EAGAIN; 315 up_write(&EXT4_I(inode)->i_data_sem); 316 goto err_out; 317 } else 318 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); 319 /* 320 * We have the extent map build with the tmp inode. 321 * Now copy the i_data across 322 */ 323 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 324 memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data)); 325 326 /* 327 * Update i_blocks with the new blocks that got 328 * allocated while adding extents for extent index 329 * blocks. 330 * 331 * While converting to extents we need not 332 * update the original inode i_blocks for extent blocks 333 * via quota APIs. The quota update happened via tmp_inode already. 334 */ 335 spin_lock(&inode->i_lock); 336 inode->i_blocks += tmp_inode->i_blocks; 337 spin_unlock(&inode->i_lock); 338 up_write(&EXT4_I(inode)->i_data_sem); 339 340 /* 341 * We mark the inode dirty after, because we decrement the 342 * i_blocks when freeing the indirect meta-data blocks 343 */ 344 retval = free_ind_block(handle, inode, i_data); 345 ext4_mark_inode_dirty(handle, inode); 346 347 err_out: 348 return retval; 349 } 350 351 static int free_ext_idx(handle_t *handle, struct inode *inode, 352 struct ext4_extent_idx *ix) 353 { 354 int i, retval = 0; 355 ext4_fsblk_t block; 356 struct buffer_head *bh; 357 struct ext4_extent_header *eh; 358 359 block = ext4_idx_pblock(ix); 360 bh = ext4_sb_bread(inode->i_sb, block, 0); 361 if (IS_ERR(bh)) 362 return PTR_ERR(bh); 363 364 eh = (struct ext4_extent_header *)bh->b_data; 365 if (eh->eh_depth != 0) { 366 ix = EXT_FIRST_INDEX(eh); 367 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) { 368 retval = free_ext_idx(handle, inode, ix); 369 if (retval) { 370 put_bh(bh); 371 return retval; 372 } 373 } 374 } 375 put_bh(bh); 376 retval = ext4_journal_ensure_credits(handle, EXT4_RESERVE_TRANS_BLOCKS, 377 ext4_free_metadata_revoke_credits(inode->i_sb, 1)); 378 if (retval < 0) 379 return retval; 380 ext4_free_blocks(handle, inode, NULL, block, 1, 381 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); 382 return 0; 383 } 384 385 /* 386 * Free the extent meta data blocks only 387 */ 388 static int free_ext_block(handle_t *handle, struct inode *inode) 389 { 390 int i, retval = 0; 391 struct ext4_inode_info *ei = EXT4_I(inode); 392 struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data; 393 struct ext4_extent_idx *ix; 394 if (eh->eh_depth == 0) 395 /* 396 * No extra blocks allocated for extent meta data 397 */ 398 return 0; 399 ix = EXT_FIRST_INDEX(eh); 400 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) { 401 retval = free_ext_idx(handle, inode, ix); 402 if (retval) 403 return retval; 404 } 405 return retval; 406 } 407 408 int ext4_ext_migrate(struct inode *inode) 409 { 410 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 411 handle_t *handle; 412 int retval = 0, i; 413 __le32 *i_data; 414 struct ext4_inode_info *ei; 415 struct inode *tmp_inode = NULL; 416 struct migrate_struct lb; 417 unsigned long max_entries; 418 __u32 goal; 419 uid_t owner[2]; 420 421 /* 422 * If the filesystem does not support extents, or the inode 423 * already is extent-based, error out. 424 */ 425 if (!ext4_has_feature_extents(inode->i_sb) || 426 (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 427 return -EINVAL; 428 429 if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0) 430 /* 431 * don't migrate fast symlink 432 */ 433 return retval; 434 435 percpu_down_write(&sbi->s_writepages_rwsem); 436 437 /* 438 * Worst case we can touch the allocation bitmaps, a bgd 439 * block, and a block to link in the orphan list. We do need 440 * need to worry about credits for modifying the quota inode. 441 */ 442 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 443 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb)); 444 445 if (IS_ERR(handle)) { 446 retval = PTR_ERR(handle); 447 goto out_unlock; 448 } 449 goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) * 450 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1; 451 owner[0] = i_uid_read(inode); 452 owner[1] = i_gid_read(inode); 453 tmp_inode = ext4_new_inode(handle, d_inode(inode->i_sb->s_root), 454 S_IFREG, NULL, goal, owner, 0); 455 if (IS_ERR(tmp_inode)) { 456 retval = PTR_ERR(tmp_inode); 457 ext4_journal_stop(handle); 458 goto out_unlock; 459 } 460 i_size_write(tmp_inode, i_size_read(inode)); 461 /* 462 * Set the i_nlink to zero so it will be deleted later 463 * when we drop inode reference. 464 */ 465 clear_nlink(tmp_inode); 466 467 ext4_ext_tree_init(handle, tmp_inode); 468 ext4_orphan_add(handle, tmp_inode); 469 ext4_journal_stop(handle); 470 471 /* 472 * start with one credit accounted for 473 * superblock modification. 474 * 475 * For the tmp_inode we already have committed the 476 * transaction that created the inode. Later as and 477 * when we add extents we extent the journal 478 */ 479 /* 480 * Even though we take i_mutex we can still cause block 481 * allocation via mmap write to holes. If we have allocated 482 * new blocks we fail migrate. New block allocation will 483 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated 484 * with i_data_sem held to prevent racing with block 485 * allocation. 486 */ 487 down_read(&EXT4_I(inode)->i_data_sem); 488 ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE); 489 up_read((&EXT4_I(inode)->i_data_sem)); 490 491 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1); 492 if (IS_ERR(handle)) { 493 /* 494 * It is impossible to update on-disk structures without 495 * a handle, so just rollback in-core changes and live other 496 * work to orphan_list_cleanup() 497 */ 498 ext4_orphan_del(NULL, tmp_inode); 499 retval = PTR_ERR(handle); 500 goto out_tmp_inode; 501 } 502 503 ei = EXT4_I(inode); 504 i_data = ei->i_data; 505 memset(&lb, 0, sizeof(lb)); 506 507 /* 32 bit block address 4 bytes */ 508 max_entries = inode->i_sb->s_blocksize >> 2; 509 for (i = 0; i < EXT4_NDIR_BLOCKS; i++) { 510 if (i_data[i]) { 511 retval = update_extent_range(handle, tmp_inode, 512 le32_to_cpu(i_data[i]), &lb); 513 if (retval) 514 goto err_out; 515 } else 516 lb.curr_block++; 517 } 518 if (i_data[EXT4_IND_BLOCK]) { 519 retval = update_ind_extent_range(handle, tmp_inode, 520 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb); 521 if (retval) 522 goto err_out; 523 } else 524 lb.curr_block += max_entries; 525 if (i_data[EXT4_DIND_BLOCK]) { 526 retval = update_dind_extent_range(handle, tmp_inode, 527 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb); 528 if (retval) 529 goto err_out; 530 } else 531 lb.curr_block += max_entries * max_entries; 532 if (i_data[EXT4_TIND_BLOCK]) { 533 retval = update_tind_extent_range(handle, tmp_inode, 534 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb); 535 if (retval) 536 goto err_out; 537 } 538 /* 539 * Build the last extent 540 */ 541 retval = finish_range(handle, tmp_inode, &lb); 542 err_out: 543 if (retval) 544 /* 545 * Failure case delete the extent information with the 546 * tmp_inode 547 */ 548 free_ext_block(handle, tmp_inode); 549 else { 550 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode); 551 if (retval) 552 /* 553 * if we fail to swap inode data free the extent 554 * details of the tmp inode 555 */ 556 free_ext_block(handle, tmp_inode); 557 } 558 559 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */ 560 retval = ext4_journal_ensure_credits(handle, 1, 0); 561 if (retval < 0) 562 goto out_stop; 563 /* 564 * Mark the tmp_inode as of size zero 565 */ 566 i_size_write(tmp_inode, 0); 567 568 /* 569 * set the i_blocks count to zero 570 * so that the ext4_evict_inode() does the 571 * right job 572 * 573 * We don't need to take the i_lock because 574 * the inode is not visible to user space. 575 */ 576 tmp_inode->i_blocks = 0; 577 578 /* Reset the extent details */ 579 ext4_ext_tree_init(handle, tmp_inode); 580 out_stop: 581 ext4_journal_stop(handle); 582 out_tmp_inode: 583 unlock_new_inode(tmp_inode); 584 iput(tmp_inode); 585 out_unlock: 586 percpu_up_write(&sbi->s_writepages_rwsem); 587 return retval; 588 } 589 590 /* 591 * Migrate a simple extent-based inode to use the i_blocks[] array 592 */ 593 int ext4_ind_migrate(struct inode *inode) 594 { 595 struct ext4_extent_header *eh; 596 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); 597 struct ext4_super_block *es = sbi->s_es; 598 struct ext4_inode_info *ei = EXT4_I(inode); 599 struct ext4_extent *ex; 600 unsigned int i, len; 601 ext4_lblk_t start, end; 602 ext4_fsblk_t blk; 603 handle_t *handle; 604 int ret; 605 606 if (!ext4_has_feature_extents(inode->i_sb) || 607 (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) 608 return -EINVAL; 609 610 if (ext4_has_feature_bigalloc(inode->i_sb)) 611 return -EOPNOTSUPP; 612 613 /* 614 * In order to get correct extent info, force all delayed allocation 615 * blocks to be allocated, otherwise delayed allocation blocks may not 616 * be reflected and bypass the checks on extent header. 617 */ 618 if (test_opt(inode->i_sb, DELALLOC)) 619 ext4_alloc_da_blocks(inode); 620 621 percpu_down_write(&sbi->s_writepages_rwsem); 622 623 handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1); 624 if (IS_ERR(handle)) { 625 ret = PTR_ERR(handle); 626 goto out_unlock; 627 } 628 629 down_write(&EXT4_I(inode)->i_data_sem); 630 ret = ext4_ext_check_inode(inode); 631 if (ret) 632 goto errout; 633 634 eh = ext_inode_hdr(inode); 635 ex = EXT_FIRST_EXTENT(eh); 636 if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS || 637 eh->eh_depth != 0 || le16_to_cpu(eh->eh_entries) > 1) { 638 ret = -EOPNOTSUPP; 639 goto errout; 640 } 641 if (eh->eh_entries == 0) 642 blk = len = start = end = 0; 643 else { 644 len = le16_to_cpu(ex->ee_len); 645 blk = ext4_ext_pblock(ex); 646 start = le32_to_cpu(ex->ee_block); 647 end = start + len - 1; 648 if (end >= EXT4_NDIR_BLOCKS) { 649 ret = -EOPNOTSUPP; 650 goto errout; 651 } 652 } 653 654 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 655 memset(ei->i_data, 0, sizeof(ei->i_data)); 656 for (i = start; i <= end; i++) 657 ei->i_data[i] = cpu_to_le32(blk++); 658 ext4_mark_inode_dirty(handle, inode); 659 errout: 660 ext4_journal_stop(handle); 661 up_write(&EXT4_I(inode)->i_data_sem); 662 out_unlock: 663 percpu_up_write(&sbi->s_writepages_rwsem); 664 return ret; 665 } 666