1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #include <linux/fs.h> 7 #include <linux/blkdev.h> 8 #include <linux/radix-tree.h> 9 #include <linux/writeback.h> 10 #include <linux/workqueue.h> 11 #include <linux/kthread.h> 12 #include <linux/slab.h> 13 #include <linux/migrate.h> 14 #include <linux/ratelimit.h> 15 #include <linux/uuid.h> 16 #include <linux/semaphore.h> 17 #include <linux/error-injection.h> 18 #include <linux/crc32c.h> 19 #include <linux/sched/mm.h> 20 #include <asm/unaligned.h> 21 #include <crypto/hash.h> 22 #include "ctree.h" 23 #include "disk-io.h" 24 #include "transaction.h" 25 #include "btrfs_inode.h" 26 #include "bio.h" 27 #include "print-tree.h" 28 #include "locking.h" 29 #include "tree-log.h" 30 #include "free-space-cache.h" 31 #include "free-space-tree.h" 32 #include "check-integrity.h" 33 #include "rcu-string.h" 34 #include "dev-replace.h" 35 #include "raid56.h" 36 #include "sysfs.h" 37 #include "qgroup.h" 38 #include "compression.h" 39 #include "tree-checker.h" 40 #include "ref-verify.h" 41 #include "block-group.h" 42 #include "discard.h" 43 #include "space-info.h" 44 #include "zoned.h" 45 #include "subpage.h" 46 #include "fs.h" 47 #include "accessors.h" 48 #include "extent-tree.h" 49 #include "root-tree.h" 50 #include "defrag.h" 51 #include "uuid-tree.h" 52 #include "relocation.h" 53 #include "scrub.h" 54 #include "super.h" 55 56 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 57 BTRFS_HEADER_FLAG_RELOC |\ 58 BTRFS_SUPER_FLAG_ERROR |\ 59 BTRFS_SUPER_FLAG_SEEDING |\ 60 BTRFS_SUPER_FLAG_METADUMP |\ 61 BTRFS_SUPER_FLAG_METADUMP_V2) 62 63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 65 66 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 67 { 68 if (fs_info->csum_shash) 69 crypto_free_shash(fs_info->csum_shash); 70 } 71 72 /* 73 * Compute the csum of a btree block and store the result to provided buffer. 74 */ 75 static void csum_tree_block(struct extent_buffer *buf, u8 *result) 76 { 77 struct btrfs_fs_info *fs_info = buf->fs_info; 78 const int num_pages = num_extent_pages(buf); 79 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 80 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 81 char *kaddr; 82 int i; 83 84 shash->tfm = fs_info->csum_shash; 85 crypto_shash_init(shash); 86 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 87 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 88 first_page_part - BTRFS_CSUM_SIZE); 89 90 for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) { 91 kaddr = page_address(buf->pages[i]); 92 crypto_shash_update(shash, kaddr, PAGE_SIZE); 93 } 94 memset(result, 0, BTRFS_CSUM_SIZE); 95 crypto_shash_final(shash, result); 96 } 97 98 /* 99 * we can't consider a given block up to date unless the transid of the 100 * block matches the transid in the parent node's pointer. This is how we 101 * detect blocks that either didn't get written at all or got written 102 * in the wrong place. 103 */ 104 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic) 105 { 106 if (!extent_buffer_uptodate(eb)) 107 return 0; 108 109 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 110 return 1; 111 112 if (atomic) 113 return -EAGAIN; 114 115 if (!extent_buffer_uptodate(eb) || 116 btrfs_header_generation(eb) != parent_transid) { 117 btrfs_err_rl(eb->fs_info, 118 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 119 eb->start, eb->read_mirror, 120 parent_transid, btrfs_header_generation(eb)); 121 clear_extent_buffer_uptodate(eb); 122 return 0; 123 } 124 return 1; 125 } 126 127 static bool btrfs_supported_super_csum(u16 csum_type) 128 { 129 switch (csum_type) { 130 case BTRFS_CSUM_TYPE_CRC32: 131 case BTRFS_CSUM_TYPE_XXHASH: 132 case BTRFS_CSUM_TYPE_SHA256: 133 case BTRFS_CSUM_TYPE_BLAKE2: 134 return true; 135 default: 136 return false; 137 } 138 } 139 140 /* 141 * Return 0 if the superblock checksum type matches the checksum value of that 142 * algorithm. Pass the raw disk superblock data. 143 */ 144 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 145 const struct btrfs_super_block *disk_sb) 146 { 147 char result[BTRFS_CSUM_SIZE]; 148 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 149 150 shash->tfm = fs_info->csum_shash; 151 152 /* 153 * The super_block structure does not span the whole 154 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 155 * filled with zeros and is included in the checksum. 156 */ 157 crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE, 158 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 159 160 if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 161 return 1; 162 163 return 0; 164 } 165 166 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, 167 int mirror_num) 168 { 169 struct btrfs_fs_info *fs_info = eb->fs_info; 170 int i, num_pages = num_extent_pages(eb); 171 int ret = 0; 172 173 if (sb_rdonly(fs_info->sb)) 174 return -EROFS; 175 176 for (i = 0; i < num_pages; i++) { 177 struct page *p = eb->pages[i]; 178 u64 start = max_t(u64, eb->start, page_offset(p)); 179 u64 end = min_t(u64, eb->start + eb->len, page_offset(p) + PAGE_SIZE); 180 u32 len = end - start; 181 182 ret = btrfs_repair_io_failure(fs_info, 0, start, len, 183 start, p, offset_in_page(start), mirror_num); 184 if (ret) 185 break; 186 } 187 188 return ret; 189 } 190 191 /* 192 * helper to read a given tree block, doing retries as required when 193 * the checksums don't match and we have alternate mirrors to try. 194 * 195 * @check: expected tree parentness check, see the comments of the 196 * structure for details. 197 */ 198 int btrfs_read_extent_buffer(struct extent_buffer *eb, 199 struct btrfs_tree_parent_check *check) 200 { 201 struct btrfs_fs_info *fs_info = eb->fs_info; 202 int failed = 0; 203 int ret; 204 int num_copies = 0; 205 int mirror_num = 0; 206 int failed_mirror = 0; 207 208 ASSERT(check); 209 210 while (1) { 211 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 212 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check); 213 if (!ret) 214 break; 215 216 num_copies = btrfs_num_copies(fs_info, 217 eb->start, eb->len); 218 if (num_copies == 1) 219 break; 220 221 if (!failed_mirror) { 222 failed = 1; 223 failed_mirror = eb->read_mirror; 224 } 225 226 mirror_num++; 227 if (mirror_num == failed_mirror) 228 mirror_num++; 229 230 if (mirror_num > num_copies) 231 break; 232 } 233 234 if (failed && !ret && failed_mirror) 235 btrfs_repair_eb_io_failure(eb, failed_mirror); 236 237 return ret; 238 } 239 240 /* 241 * Checksum a dirty tree block before IO. 242 */ 243 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio) 244 { 245 struct extent_buffer *eb = bbio->private; 246 struct btrfs_fs_info *fs_info = eb->fs_info; 247 u64 found_start = btrfs_header_bytenr(eb); 248 u8 result[BTRFS_CSUM_SIZE]; 249 int ret; 250 251 /* Btree blocks are always contiguous on disk. */ 252 if (WARN_ON_ONCE(bbio->file_offset != eb->start)) 253 return BLK_STS_IOERR; 254 if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len)) 255 return BLK_STS_IOERR; 256 257 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 258 WARN_ON_ONCE(found_start != 0); 259 return BLK_STS_OK; 260 } 261 262 if (WARN_ON_ONCE(found_start != eb->start)) 263 return BLK_STS_IOERR; 264 if (WARN_ON(!btrfs_page_test_uptodate(fs_info, eb->pages[0], eb->start, 265 eb->len))) 266 return BLK_STS_IOERR; 267 268 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 269 offsetof(struct btrfs_header, fsid), 270 BTRFS_FSID_SIZE) == 0); 271 csum_tree_block(eb, result); 272 273 if (btrfs_header_level(eb)) 274 ret = btrfs_check_node(eb); 275 else 276 ret = btrfs_check_leaf(eb); 277 278 if (ret < 0) 279 goto error; 280 281 /* 282 * Also check the generation, the eb reached here must be newer than 283 * last committed. Or something seriously wrong happened. 284 */ 285 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 286 ret = -EUCLEAN; 287 btrfs_err(fs_info, 288 "block=%llu bad generation, have %llu expect > %llu", 289 eb->start, btrfs_header_generation(eb), 290 fs_info->last_trans_committed); 291 goto error; 292 } 293 write_extent_buffer(eb, result, 0, fs_info->csum_size); 294 return BLK_STS_OK; 295 296 error: 297 btrfs_print_tree(eb, 0); 298 btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 299 eb->start); 300 /* 301 * Be noisy if this is an extent buffer from a log tree. We don't abort 302 * a transaction in case there's a bad log tree extent buffer, we just 303 * fallback to a transaction commit. Still we want to know when there is 304 * a bad log tree extent buffer, as that may signal a bug somewhere. 305 */ 306 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) || 307 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID); 308 return errno_to_blk_status(ret); 309 } 310 311 static bool check_tree_block_fsid(struct extent_buffer *eb) 312 { 313 struct btrfs_fs_info *fs_info = eb->fs_info; 314 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 315 u8 fsid[BTRFS_FSID_SIZE]; 316 u8 *metadata_uuid; 317 318 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 319 BTRFS_FSID_SIZE); 320 /* 321 * Checking the incompat flag is only valid for the current fs. For 322 * seed devices it's forbidden to have their uuid changed so reading 323 * ->fsid in this case is fine 324 */ 325 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 326 metadata_uuid = fs_devices->metadata_uuid; 327 else 328 metadata_uuid = fs_devices->fsid; 329 330 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 331 return false; 332 333 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 334 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 335 return false; 336 337 return true; 338 } 339 340 /* Do basic extent buffer checks at read time */ 341 int btrfs_validate_extent_buffer(struct extent_buffer *eb, 342 struct btrfs_tree_parent_check *check) 343 { 344 struct btrfs_fs_info *fs_info = eb->fs_info; 345 u64 found_start; 346 const u32 csum_size = fs_info->csum_size; 347 u8 found_level; 348 u8 result[BTRFS_CSUM_SIZE]; 349 const u8 *header_csum; 350 int ret = 0; 351 352 ASSERT(check); 353 354 found_start = btrfs_header_bytenr(eb); 355 if (found_start != eb->start) { 356 btrfs_err_rl(fs_info, 357 "bad tree block start, mirror %u want %llu have %llu", 358 eb->read_mirror, eb->start, found_start); 359 ret = -EIO; 360 goto out; 361 } 362 if (check_tree_block_fsid(eb)) { 363 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u", 364 eb->start, eb->read_mirror); 365 ret = -EIO; 366 goto out; 367 } 368 found_level = btrfs_header_level(eb); 369 if (found_level >= BTRFS_MAX_LEVEL) { 370 btrfs_err(fs_info, 371 "bad tree block level, mirror %u level %d on logical %llu", 372 eb->read_mirror, btrfs_header_level(eb), eb->start); 373 ret = -EIO; 374 goto out; 375 } 376 377 csum_tree_block(eb, result); 378 header_csum = page_address(eb->pages[0]) + 379 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 380 381 if (memcmp(result, header_csum, csum_size) != 0) { 382 btrfs_warn_rl(fs_info, 383 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d", 384 eb->start, eb->read_mirror, 385 CSUM_FMT_VALUE(csum_size, header_csum), 386 CSUM_FMT_VALUE(csum_size, result), 387 btrfs_header_level(eb)); 388 ret = -EUCLEAN; 389 goto out; 390 } 391 392 if (found_level != check->level) { 393 btrfs_err(fs_info, 394 "level verify failed on logical %llu mirror %u wanted %u found %u", 395 eb->start, eb->read_mirror, check->level, found_level); 396 ret = -EIO; 397 goto out; 398 } 399 if (unlikely(check->transid && 400 btrfs_header_generation(eb) != check->transid)) { 401 btrfs_err_rl(eb->fs_info, 402 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu", 403 eb->start, eb->read_mirror, check->transid, 404 btrfs_header_generation(eb)); 405 ret = -EIO; 406 goto out; 407 } 408 if (check->has_first_key) { 409 struct btrfs_key *expect_key = &check->first_key; 410 struct btrfs_key found_key; 411 412 if (found_level) 413 btrfs_node_key_to_cpu(eb, &found_key, 0); 414 else 415 btrfs_item_key_to_cpu(eb, &found_key, 0); 416 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) { 417 btrfs_err(fs_info, 418 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 419 eb->start, check->transid, 420 expect_key->objectid, 421 expect_key->type, expect_key->offset, 422 found_key.objectid, found_key.type, 423 found_key.offset); 424 ret = -EUCLEAN; 425 goto out; 426 } 427 } 428 if (check->owner_root) { 429 ret = btrfs_check_eb_owner(eb, check->owner_root); 430 if (ret < 0) 431 goto out; 432 } 433 434 /* 435 * If this is a leaf block and it is corrupt, set the corrupt bit so 436 * that we don't try and read the other copies of this block, just 437 * return -EIO. 438 */ 439 if (found_level == 0 && btrfs_check_leaf(eb)) { 440 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 441 ret = -EIO; 442 } 443 444 if (found_level > 0 && btrfs_check_node(eb)) 445 ret = -EIO; 446 447 if (ret) 448 btrfs_err(fs_info, 449 "read time tree block corruption detected on logical %llu mirror %u", 450 eb->start, eb->read_mirror); 451 out: 452 return ret; 453 } 454 455 #ifdef CONFIG_MIGRATION 456 static int btree_migrate_folio(struct address_space *mapping, 457 struct folio *dst, struct folio *src, enum migrate_mode mode) 458 { 459 /* 460 * we can't safely write a btree page from here, 461 * we haven't done the locking hook 462 */ 463 if (folio_test_dirty(src)) 464 return -EAGAIN; 465 /* 466 * Buffers may be managed in a filesystem specific way. 467 * We must have no buffers or drop them. 468 */ 469 if (folio_get_private(src) && 470 !filemap_release_folio(src, GFP_KERNEL)) 471 return -EAGAIN; 472 return migrate_folio(mapping, dst, src, mode); 473 } 474 #else 475 #define btree_migrate_folio NULL 476 #endif 477 478 static int btree_writepages(struct address_space *mapping, 479 struct writeback_control *wbc) 480 { 481 struct btrfs_fs_info *fs_info; 482 int ret; 483 484 if (wbc->sync_mode == WB_SYNC_NONE) { 485 486 if (wbc->for_kupdate) 487 return 0; 488 489 fs_info = BTRFS_I(mapping->host)->root->fs_info; 490 /* this is a bit racy, but that's ok */ 491 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 492 BTRFS_DIRTY_METADATA_THRESH, 493 fs_info->dirty_metadata_batch); 494 if (ret < 0) 495 return 0; 496 } 497 return btree_write_cache_pages(mapping, wbc); 498 } 499 500 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags) 501 { 502 if (folio_test_writeback(folio) || folio_test_dirty(folio)) 503 return false; 504 505 return try_release_extent_buffer(&folio->page); 506 } 507 508 static void btree_invalidate_folio(struct folio *folio, size_t offset, 509 size_t length) 510 { 511 struct extent_io_tree *tree; 512 tree = &BTRFS_I(folio->mapping->host)->io_tree; 513 extent_invalidate_folio(tree, folio, offset); 514 btree_release_folio(folio, GFP_NOFS); 515 if (folio_get_private(folio)) { 516 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 517 "folio private not zero on folio %llu", 518 (unsigned long long)folio_pos(folio)); 519 folio_detach_private(folio); 520 } 521 } 522 523 #ifdef DEBUG 524 static bool btree_dirty_folio(struct address_space *mapping, 525 struct folio *folio) 526 { 527 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 528 struct btrfs_subpage *subpage; 529 struct extent_buffer *eb; 530 int cur_bit = 0; 531 u64 page_start = folio_pos(folio); 532 533 if (fs_info->sectorsize == PAGE_SIZE) { 534 eb = folio_get_private(folio); 535 BUG_ON(!eb); 536 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 537 BUG_ON(!atomic_read(&eb->refs)); 538 btrfs_assert_tree_write_locked(eb); 539 return filemap_dirty_folio(mapping, folio); 540 } 541 subpage = folio_get_private(folio); 542 543 ASSERT(subpage->dirty_bitmap); 544 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 545 unsigned long flags; 546 u64 cur; 547 u16 tmp = (1 << cur_bit); 548 549 spin_lock_irqsave(&subpage->lock, flags); 550 if (!(tmp & subpage->dirty_bitmap)) { 551 spin_unlock_irqrestore(&subpage->lock, flags); 552 cur_bit++; 553 continue; 554 } 555 spin_unlock_irqrestore(&subpage->lock, flags); 556 cur = page_start + cur_bit * fs_info->sectorsize; 557 558 eb = find_extent_buffer(fs_info, cur); 559 ASSERT(eb); 560 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 561 ASSERT(atomic_read(&eb->refs)); 562 btrfs_assert_tree_write_locked(eb); 563 free_extent_buffer(eb); 564 565 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 566 } 567 return filemap_dirty_folio(mapping, folio); 568 } 569 #else 570 #define btree_dirty_folio filemap_dirty_folio 571 #endif 572 573 static const struct address_space_operations btree_aops = { 574 .writepages = btree_writepages, 575 .release_folio = btree_release_folio, 576 .invalidate_folio = btree_invalidate_folio, 577 .migrate_folio = btree_migrate_folio, 578 .dirty_folio = btree_dirty_folio, 579 }; 580 581 struct extent_buffer *btrfs_find_create_tree_block( 582 struct btrfs_fs_info *fs_info, 583 u64 bytenr, u64 owner_root, 584 int level) 585 { 586 if (btrfs_is_testing(fs_info)) 587 return alloc_test_extent_buffer(fs_info, bytenr); 588 return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 589 } 590 591 /* 592 * Read tree block at logical address @bytenr and do variant basic but critical 593 * verification. 594 * 595 * @check: expected tree parentness check, see comments of the 596 * structure for details. 597 */ 598 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 599 struct btrfs_tree_parent_check *check) 600 { 601 struct extent_buffer *buf = NULL; 602 int ret; 603 604 ASSERT(check); 605 606 buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root, 607 check->level); 608 if (IS_ERR(buf)) 609 return buf; 610 611 ret = btrfs_read_extent_buffer(buf, check); 612 if (ret) { 613 free_extent_buffer_stale(buf); 614 return ERR_PTR(ret); 615 } 616 if (btrfs_check_eb_owner(buf, check->owner_root)) { 617 free_extent_buffer_stale(buf); 618 return ERR_PTR(-EUCLEAN); 619 } 620 return buf; 621 622 } 623 624 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 625 u64 objectid) 626 { 627 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 628 629 memset(&root->root_key, 0, sizeof(root->root_key)); 630 memset(&root->root_item, 0, sizeof(root->root_item)); 631 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 632 root->fs_info = fs_info; 633 root->root_key.objectid = objectid; 634 root->node = NULL; 635 root->commit_root = NULL; 636 root->state = 0; 637 RB_CLEAR_NODE(&root->rb_node); 638 639 root->last_trans = 0; 640 root->free_objectid = 0; 641 root->nr_delalloc_inodes = 0; 642 root->nr_ordered_extents = 0; 643 root->inode_tree = RB_ROOT; 644 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 645 646 btrfs_init_root_block_rsv(root); 647 648 INIT_LIST_HEAD(&root->dirty_list); 649 INIT_LIST_HEAD(&root->root_list); 650 INIT_LIST_HEAD(&root->delalloc_inodes); 651 INIT_LIST_HEAD(&root->delalloc_root); 652 INIT_LIST_HEAD(&root->ordered_extents); 653 INIT_LIST_HEAD(&root->ordered_root); 654 INIT_LIST_HEAD(&root->reloc_dirty_list); 655 INIT_LIST_HEAD(&root->logged_list[0]); 656 INIT_LIST_HEAD(&root->logged_list[1]); 657 spin_lock_init(&root->inode_lock); 658 spin_lock_init(&root->delalloc_lock); 659 spin_lock_init(&root->ordered_extent_lock); 660 spin_lock_init(&root->accounting_lock); 661 spin_lock_init(&root->log_extents_lock[0]); 662 spin_lock_init(&root->log_extents_lock[1]); 663 spin_lock_init(&root->qgroup_meta_rsv_lock); 664 mutex_init(&root->objectid_mutex); 665 mutex_init(&root->log_mutex); 666 mutex_init(&root->ordered_extent_mutex); 667 mutex_init(&root->delalloc_mutex); 668 init_waitqueue_head(&root->qgroup_flush_wait); 669 init_waitqueue_head(&root->log_writer_wait); 670 init_waitqueue_head(&root->log_commit_wait[0]); 671 init_waitqueue_head(&root->log_commit_wait[1]); 672 INIT_LIST_HEAD(&root->log_ctxs[0]); 673 INIT_LIST_HEAD(&root->log_ctxs[1]); 674 atomic_set(&root->log_commit[0], 0); 675 atomic_set(&root->log_commit[1], 0); 676 atomic_set(&root->log_writers, 0); 677 atomic_set(&root->log_batch, 0); 678 refcount_set(&root->refs, 1); 679 atomic_set(&root->snapshot_force_cow, 0); 680 atomic_set(&root->nr_swapfiles, 0); 681 root->log_transid = 0; 682 root->log_transid_committed = -1; 683 root->last_log_commit = 0; 684 root->anon_dev = 0; 685 if (!dummy) { 686 extent_io_tree_init(fs_info, &root->dirty_log_pages, 687 IO_TREE_ROOT_DIRTY_LOG_PAGES); 688 extent_io_tree_init(fs_info, &root->log_csum_range, 689 IO_TREE_LOG_CSUM_RANGE); 690 } 691 692 spin_lock_init(&root->root_item_lock); 693 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 694 #ifdef CONFIG_BTRFS_DEBUG 695 INIT_LIST_HEAD(&root->leak_list); 696 spin_lock(&fs_info->fs_roots_radix_lock); 697 list_add_tail(&root->leak_list, &fs_info->allocated_roots); 698 spin_unlock(&fs_info->fs_roots_radix_lock); 699 #endif 700 } 701 702 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 703 u64 objectid, gfp_t flags) 704 { 705 struct btrfs_root *root = kzalloc(sizeof(*root), flags); 706 if (root) 707 __setup_root(root, fs_info, objectid); 708 return root; 709 } 710 711 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 712 /* Should only be used by the testing infrastructure */ 713 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 714 { 715 struct btrfs_root *root; 716 717 if (!fs_info) 718 return ERR_PTR(-EINVAL); 719 720 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 721 if (!root) 722 return ERR_PTR(-ENOMEM); 723 724 /* We don't use the stripesize in selftest, set it as sectorsize */ 725 root->alloc_bytenr = 0; 726 727 return root; 728 } 729 #endif 730 731 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 732 { 733 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 734 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 735 736 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 737 } 738 739 static int global_root_key_cmp(const void *k, const struct rb_node *node) 740 { 741 const struct btrfs_key *key = k; 742 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 743 744 return btrfs_comp_cpu_keys(key, &root->root_key); 745 } 746 747 int btrfs_global_root_insert(struct btrfs_root *root) 748 { 749 struct btrfs_fs_info *fs_info = root->fs_info; 750 struct rb_node *tmp; 751 int ret = 0; 752 753 write_lock(&fs_info->global_root_lock); 754 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 755 write_unlock(&fs_info->global_root_lock); 756 757 if (tmp) { 758 ret = -EEXIST; 759 btrfs_warn(fs_info, "global root %llu %llu already exists", 760 root->root_key.objectid, root->root_key.offset); 761 } 762 return ret; 763 } 764 765 void btrfs_global_root_delete(struct btrfs_root *root) 766 { 767 struct btrfs_fs_info *fs_info = root->fs_info; 768 769 write_lock(&fs_info->global_root_lock); 770 rb_erase(&root->rb_node, &fs_info->global_root_tree); 771 write_unlock(&fs_info->global_root_lock); 772 } 773 774 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 775 struct btrfs_key *key) 776 { 777 struct rb_node *node; 778 struct btrfs_root *root = NULL; 779 780 read_lock(&fs_info->global_root_lock); 781 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 782 if (node) 783 root = container_of(node, struct btrfs_root, rb_node); 784 read_unlock(&fs_info->global_root_lock); 785 786 return root; 787 } 788 789 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 790 { 791 struct btrfs_block_group *block_group; 792 u64 ret; 793 794 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 795 return 0; 796 797 if (bytenr) 798 block_group = btrfs_lookup_block_group(fs_info, bytenr); 799 else 800 block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 801 ASSERT(block_group); 802 if (!block_group) 803 return 0; 804 ret = block_group->global_root_id; 805 btrfs_put_block_group(block_group); 806 807 return ret; 808 } 809 810 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 811 { 812 struct btrfs_key key = { 813 .objectid = BTRFS_CSUM_TREE_OBJECTID, 814 .type = BTRFS_ROOT_ITEM_KEY, 815 .offset = btrfs_global_root_id(fs_info, bytenr), 816 }; 817 818 return btrfs_global_root(fs_info, &key); 819 } 820 821 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 822 { 823 struct btrfs_key key = { 824 .objectid = BTRFS_EXTENT_TREE_OBJECTID, 825 .type = BTRFS_ROOT_ITEM_KEY, 826 .offset = btrfs_global_root_id(fs_info, bytenr), 827 }; 828 829 return btrfs_global_root(fs_info, &key); 830 } 831 832 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info) 833 { 834 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) 835 return fs_info->block_group_root; 836 return btrfs_extent_root(fs_info, 0); 837 } 838 839 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 840 u64 objectid) 841 { 842 struct btrfs_fs_info *fs_info = trans->fs_info; 843 struct extent_buffer *leaf; 844 struct btrfs_root *tree_root = fs_info->tree_root; 845 struct btrfs_root *root; 846 struct btrfs_key key; 847 unsigned int nofs_flag; 848 int ret = 0; 849 850 /* 851 * We're holding a transaction handle, so use a NOFS memory allocation 852 * context to avoid deadlock if reclaim happens. 853 */ 854 nofs_flag = memalloc_nofs_save(); 855 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 856 memalloc_nofs_restore(nofs_flag); 857 if (!root) 858 return ERR_PTR(-ENOMEM); 859 860 root->root_key.objectid = objectid; 861 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 862 root->root_key.offset = 0; 863 864 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 865 BTRFS_NESTING_NORMAL); 866 if (IS_ERR(leaf)) { 867 ret = PTR_ERR(leaf); 868 leaf = NULL; 869 goto fail; 870 } 871 872 root->node = leaf; 873 btrfs_mark_buffer_dirty(leaf); 874 875 root->commit_root = btrfs_root_node(root); 876 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 877 878 btrfs_set_root_flags(&root->root_item, 0); 879 btrfs_set_root_limit(&root->root_item, 0); 880 btrfs_set_root_bytenr(&root->root_item, leaf->start); 881 btrfs_set_root_generation(&root->root_item, trans->transid); 882 btrfs_set_root_level(&root->root_item, 0); 883 btrfs_set_root_refs(&root->root_item, 1); 884 btrfs_set_root_used(&root->root_item, leaf->len); 885 btrfs_set_root_last_snapshot(&root->root_item, 0); 886 btrfs_set_root_dirid(&root->root_item, 0); 887 if (is_fstree(objectid)) 888 generate_random_guid(root->root_item.uuid); 889 else 890 export_guid(root->root_item.uuid, &guid_null); 891 btrfs_set_root_drop_level(&root->root_item, 0); 892 893 btrfs_tree_unlock(leaf); 894 895 key.objectid = objectid; 896 key.type = BTRFS_ROOT_ITEM_KEY; 897 key.offset = 0; 898 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 899 if (ret) 900 goto fail; 901 902 return root; 903 904 fail: 905 btrfs_put_root(root); 906 907 return ERR_PTR(ret); 908 } 909 910 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 911 struct btrfs_fs_info *fs_info) 912 { 913 struct btrfs_root *root; 914 915 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 916 if (!root) 917 return ERR_PTR(-ENOMEM); 918 919 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 920 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 921 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 922 923 return root; 924 } 925 926 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 927 struct btrfs_root *root) 928 { 929 struct extent_buffer *leaf; 930 931 /* 932 * DON'T set SHAREABLE bit for log trees. 933 * 934 * Log trees are not exposed to user space thus can't be snapshotted, 935 * and they go away before a real commit is actually done. 936 * 937 * They do store pointers to file data extents, and those reference 938 * counts still get updated (along with back refs to the log tree). 939 */ 940 941 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 942 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 943 if (IS_ERR(leaf)) 944 return PTR_ERR(leaf); 945 946 root->node = leaf; 947 948 btrfs_mark_buffer_dirty(root->node); 949 btrfs_tree_unlock(root->node); 950 951 return 0; 952 } 953 954 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 955 struct btrfs_fs_info *fs_info) 956 { 957 struct btrfs_root *log_root; 958 959 log_root = alloc_log_tree(trans, fs_info); 960 if (IS_ERR(log_root)) 961 return PTR_ERR(log_root); 962 963 if (!btrfs_is_zoned(fs_info)) { 964 int ret = btrfs_alloc_log_tree_node(trans, log_root); 965 966 if (ret) { 967 btrfs_put_root(log_root); 968 return ret; 969 } 970 } 971 972 WARN_ON(fs_info->log_root_tree); 973 fs_info->log_root_tree = log_root; 974 return 0; 975 } 976 977 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 978 struct btrfs_root *root) 979 { 980 struct btrfs_fs_info *fs_info = root->fs_info; 981 struct btrfs_root *log_root; 982 struct btrfs_inode_item *inode_item; 983 int ret; 984 985 log_root = alloc_log_tree(trans, fs_info); 986 if (IS_ERR(log_root)) 987 return PTR_ERR(log_root); 988 989 ret = btrfs_alloc_log_tree_node(trans, log_root); 990 if (ret) { 991 btrfs_put_root(log_root); 992 return ret; 993 } 994 995 log_root->last_trans = trans->transid; 996 log_root->root_key.offset = root->root_key.objectid; 997 998 inode_item = &log_root->root_item.inode; 999 btrfs_set_stack_inode_generation(inode_item, 1); 1000 btrfs_set_stack_inode_size(inode_item, 3); 1001 btrfs_set_stack_inode_nlink(inode_item, 1); 1002 btrfs_set_stack_inode_nbytes(inode_item, 1003 fs_info->nodesize); 1004 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1005 1006 btrfs_set_root_node(&log_root->root_item, log_root->node); 1007 1008 WARN_ON(root->log_root); 1009 root->log_root = log_root; 1010 root->log_transid = 0; 1011 root->log_transid_committed = -1; 1012 root->last_log_commit = 0; 1013 return 0; 1014 } 1015 1016 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 1017 struct btrfs_path *path, 1018 struct btrfs_key *key) 1019 { 1020 struct btrfs_root *root; 1021 struct btrfs_tree_parent_check check = { 0 }; 1022 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1023 u64 generation; 1024 int ret; 1025 int level; 1026 1027 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 1028 if (!root) 1029 return ERR_PTR(-ENOMEM); 1030 1031 ret = btrfs_find_root(tree_root, key, path, 1032 &root->root_item, &root->root_key); 1033 if (ret) { 1034 if (ret > 0) 1035 ret = -ENOENT; 1036 goto fail; 1037 } 1038 1039 generation = btrfs_root_generation(&root->root_item); 1040 level = btrfs_root_level(&root->root_item); 1041 check.level = level; 1042 check.transid = generation; 1043 check.owner_root = key->objectid; 1044 root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item), 1045 &check); 1046 if (IS_ERR(root->node)) { 1047 ret = PTR_ERR(root->node); 1048 root->node = NULL; 1049 goto fail; 1050 } 1051 if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1052 ret = -EIO; 1053 goto fail; 1054 } 1055 1056 /* 1057 * For real fs, and not log/reloc trees, root owner must 1058 * match its root node owner 1059 */ 1060 if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) && 1061 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1062 root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 1063 root->root_key.objectid != btrfs_header_owner(root->node)) { 1064 btrfs_crit(fs_info, 1065 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu", 1066 root->root_key.objectid, root->node->start, 1067 btrfs_header_owner(root->node), 1068 root->root_key.objectid); 1069 ret = -EUCLEAN; 1070 goto fail; 1071 } 1072 root->commit_root = btrfs_root_node(root); 1073 return root; 1074 fail: 1075 btrfs_put_root(root); 1076 return ERR_PTR(ret); 1077 } 1078 1079 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1080 struct btrfs_key *key) 1081 { 1082 struct btrfs_root *root; 1083 struct btrfs_path *path; 1084 1085 path = btrfs_alloc_path(); 1086 if (!path) 1087 return ERR_PTR(-ENOMEM); 1088 root = read_tree_root_path(tree_root, path, key); 1089 btrfs_free_path(path); 1090 1091 return root; 1092 } 1093 1094 /* 1095 * Initialize subvolume root in-memory structure 1096 * 1097 * @anon_dev: anonymous device to attach to the root, if zero, allocate new 1098 */ 1099 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1100 { 1101 int ret; 1102 1103 btrfs_drew_lock_init(&root->snapshot_lock); 1104 1105 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1106 !btrfs_is_data_reloc_root(root) && 1107 is_fstree(root->root_key.objectid)) { 1108 set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1109 btrfs_check_and_init_root_item(&root->root_item); 1110 } 1111 1112 /* 1113 * Don't assign anonymous block device to roots that are not exposed to 1114 * userspace, the id pool is limited to 1M 1115 */ 1116 if (is_fstree(root->root_key.objectid) && 1117 btrfs_root_refs(&root->root_item) > 0) { 1118 if (!anon_dev) { 1119 ret = get_anon_bdev(&root->anon_dev); 1120 if (ret) 1121 goto fail; 1122 } else { 1123 root->anon_dev = anon_dev; 1124 } 1125 } 1126 1127 mutex_lock(&root->objectid_mutex); 1128 ret = btrfs_init_root_free_objectid(root); 1129 if (ret) { 1130 mutex_unlock(&root->objectid_mutex); 1131 goto fail; 1132 } 1133 1134 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1135 1136 mutex_unlock(&root->objectid_mutex); 1137 1138 return 0; 1139 fail: 1140 /* The caller is responsible to call btrfs_free_fs_root */ 1141 return ret; 1142 } 1143 1144 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1145 u64 root_id) 1146 { 1147 struct btrfs_root *root; 1148 1149 spin_lock(&fs_info->fs_roots_radix_lock); 1150 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1151 (unsigned long)root_id); 1152 root = btrfs_grab_root(root); 1153 spin_unlock(&fs_info->fs_roots_radix_lock); 1154 return root; 1155 } 1156 1157 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 1158 u64 objectid) 1159 { 1160 struct btrfs_key key = { 1161 .objectid = objectid, 1162 .type = BTRFS_ROOT_ITEM_KEY, 1163 .offset = 0, 1164 }; 1165 1166 switch (objectid) { 1167 case BTRFS_ROOT_TREE_OBJECTID: 1168 return btrfs_grab_root(fs_info->tree_root); 1169 case BTRFS_EXTENT_TREE_OBJECTID: 1170 return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 1171 case BTRFS_CHUNK_TREE_OBJECTID: 1172 return btrfs_grab_root(fs_info->chunk_root); 1173 case BTRFS_DEV_TREE_OBJECTID: 1174 return btrfs_grab_root(fs_info->dev_root); 1175 case BTRFS_CSUM_TREE_OBJECTID: 1176 return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 1177 case BTRFS_QUOTA_TREE_OBJECTID: 1178 return btrfs_grab_root(fs_info->quota_root); 1179 case BTRFS_UUID_TREE_OBJECTID: 1180 return btrfs_grab_root(fs_info->uuid_root); 1181 case BTRFS_BLOCK_GROUP_TREE_OBJECTID: 1182 return btrfs_grab_root(fs_info->block_group_root); 1183 case BTRFS_FREE_SPACE_TREE_OBJECTID: 1184 return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 1185 default: 1186 return NULL; 1187 } 1188 } 1189 1190 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1191 struct btrfs_root *root) 1192 { 1193 int ret; 1194 1195 ret = radix_tree_preload(GFP_NOFS); 1196 if (ret) 1197 return ret; 1198 1199 spin_lock(&fs_info->fs_roots_radix_lock); 1200 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1201 (unsigned long)root->root_key.objectid, 1202 root); 1203 if (ret == 0) { 1204 btrfs_grab_root(root); 1205 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1206 } 1207 spin_unlock(&fs_info->fs_roots_radix_lock); 1208 radix_tree_preload_end(); 1209 1210 return ret; 1211 } 1212 1213 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1214 { 1215 #ifdef CONFIG_BTRFS_DEBUG 1216 struct btrfs_root *root; 1217 1218 while (!list_empty(&fs_info->allocated_roots)) { 1219 char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1220 1221 root = list_first_entry(&fs_info->allocated_roots, 1222 struct btrfs_root, leak_list); 1223 btrfs_err(fs_info, "leaked root %s refcount %d", 1224 btrfs_root_name(&root->root_key, buf), 1225 refcount_read(&root->refs)); 1226 while (refcount_read(&root->refs) > 1) 1227 btrfs_put_root(root); 1228 btrfs_put_root(root); 1229 } 1230 #endif 1231 } 1232 1233 static void free_global_roots(struct btrfs_fs_info *fs_info) 1234 { 1235 struct btrfs_root *root; 1236 struct rb_node *node; 1237 1238 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1239 root = rb_entry(node, struct btrfs_root, rb_node); 1240 rb_erase(&root->rb_node, &fs_info->global_root_tree); 1241 btrfs_put_root(root); 1242 } 1243 } 1244 1245 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 1246 { 1247 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1248 percpu_counter_destroy(&fs_info->delalloc_bytes); 1249 percpu_counter_destroy(&fs_info->ordered_bytes); 1250 percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1251 btrfs_free_csum_hash(fs_info); 1252 btrfs_free_stripe_hash_table(fs_info); 1253 btrfs_free_ref_cache(fs_info); 1254 kfree(fs_info->balance_ctl); 1255 kfree(fs_info->delayed_root); 1256 free_global_roots(fs_info); 1257 btrfs_put_root(fs_info->tree_root); 1258 btrfs_put_root(fs_info->chunk_root); 1259 btrfs_put_root(fs_info->dev_root); 1260 btrfs_put_root(fs_info->quota_root); 1261 btrfs_put_root(fs_info->uuid_root); 1262 btrfs_put_root(fs_info->fs_root); 1263 btrfs_put_root(fs_info->data_reloc_root); 1264 btrfs_put_root(fs_info->block_group_root); 1265 btrfs_check_leaked_roots(fs_info); 1266 btrfs_extent_buffer_leak_debug_check(fs_info); 1267 kfree(fs_info->super_copy); 1268 kfree(fs_info->super_for_commit); 1269 kfree(fs_info->subpage_info); 1270 kvfree(fs_info); 1271 } 1272 1273 1274 /* 1275 * Get an in-memory reference of a root structure. 1276 * 1277 * For essential trees like root/extent tree, we grab it from fs_info directly. 1278 * For subvolume trees, we check the cached filesystem roots first. If not 1279 * found, then read it from disk and add it to cached fs roots. 1280 * 1281 * Caller should release the root by calling btrfs_put_root() after the usage. 1282 * 1283 * NOTE: Reloc and log trees can't be read by this function as they share the 1284 * same root objectid. 1285 * 1286 * @objectid: root id 1287 * @anon_dev: preallocated anonymous block device number for new roots, 1288 * pass 0 for new allocation. 1289 * @check_ref: whether to check root item references, If true, return -ENOENT 1290 * for orphan roots 1291 */ 1292 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 1293 u64 objectid, dev_t anon_dev, 1294 bool check_ref) 1295 { 1296 struct btrfs_root *root; 1297 struct btrfs_path *path; 1298 struct btrfs_key key; 1299 int ret; 1300 1301 root = btrfs_get_global_root(fs_info, objectid); 1302 if (root) 1303 return root; 1304 1305 /* 1306 * If we're called for non-subvolume trees, and above function didn't 1307 * find one, do not try to read it from disk. 1308 * 1309 * This is namely for free-space-tree and quota tree, which can change 1310 * at runtime and should only be grabbed from fs_info. 1311 */ 1312 if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) 1313 return ERR_PTR(-ENOENT); 1314 again: 1315 root = btrfs_lookup_fs_root(fs_info, objectid); 1316 if (root) { 1317 /* Shouldn't get preallocated anon_dev for cached roots */ 1318 ASSERT(!anon_dev); 1319 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1320 btrfs_put_root(root); 1321 return ERR_PTR(-ENOENT); 1322 } 1323 return root; 1324 } 1325 1326 key.objectid = objectid; 1327 key.type = BTRFS_ROOT_ITEM_KEY; 1328 key.offset = (u64)-1; 1329 root = btrfs_read_tree_root(fs_info->tree_root, &key); 1330 if (IS_ERR(root)) 1331 return root; 1332 1333 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1334 ret = -ENOENT; 1335 goto fail; 1336 } 1337 1338 ret = btrfs_init_fs_root(root, anon_dev); 1339 if (ret) 1340 goto fail; 1341 1342 path = btrfs_alloc_path(); 1343 if (!path) { 1344 ret = -ENOMEM; 1345 goto fail; 1346 } 1347 key.objectid = BTRFS_ORPHAN_OBJECTID; 1348 key.type = BTRFS_ORPHAN_ITEM_KEY; 1349 key.offset = objectid; 1350 1351 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1352 btrfs_free_path(path); 1353 if (ret < 0) 1354 goto fail; 1355 if (ret == 0) 1356 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1357 1358 ret = btrfs_insert_fs_root(fs_info, root); 1359 if (ret) { 1360 if (ret == -EEXIST) { 1361 btrfs_put_root(root); 1362 goto again; 1363 } 1364 goto fail; 1365 } 1366 return root; 1367 fail: 1368 /* 1369 * If our caller provided us an anonymous device, then it's his 1370 * responsibility to free it in case we fail. So we have to set our 1371 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 1372 * and once again by our caller. 1373 */ 1374 if (anon_dev) 1375 root->anon_dev = 0; 1376 btrfs_put_root(root); 1377 return ERR_PTR(ret); 1378 } 1379 1380 /* 1381 * Get in-memory reference of a root structure 1382 * 1383 * @objectid: tree objectid 1384 * @check_ref: if set, verify that the tree exists and the item has at least 1385 * one reference 1386 */ 1387 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1388 u64 objectid, bool check_ref) 1389 { 1390 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 1391 } 1392 1393 /* 1394 * Get in-memory reference of a root structure, created as new, optionally pass 1395 * the anonymous block device id 1396 * 1397 * @objectid: tree objectid 1398 * @anon_dev: if zero, allocate a new anonymous block device or use the 1399 * parameter value 1400 */ 1401 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 1402 u64 objectid, dev_t anon_dev) 1403 { 1404 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 1405 } 1406 1407 /* 1408 * btrfs_get_fs_root_commit_root - return a root for the given objectid 1409 * @fs_info: the fs_info 1410 * @objectid: the objectid we need to lookup 1411 * 1412 * This is exclusively used for backref walking, and exists specifically because 1413 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 1414 * creation time, which means we may have to read the tree_root in order to look 1415 * up a fs root that is not in memory. If the root is not in memory we will 1416 * read the tree root commit root and look up the fs root from there. This is a 1417 * temporary root, it will not be inserted into the radix tree as it doesn't 1418 * have the most uptodate information, it'll simply be discarded once the 1419 * backref code is finished using the root. 1420 */ 1421 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 1422 struct btrfs_path *path, 1423 u64 objectid) 1424 { 1425 struct btrfs_root *root; 1426 struct btrfs_key key; 1427 1428 ASSERT(path->search_commit_root && path->skip_locking); 1429 1430 /* 1431 * This can return -ENOENT if we ask for a root that doesn't exist, but 1432 * since this is called via the backref walking code we won't be looking 1433 * up a root that doesn't exist, unless there's corruption. So if root 1434 * != NULL just return it. 1435 */ 1436 root = btrfs_get_global_root(fs_info, objectid); 1437 if (root) 1438 return root; 1439 1440 root = btrfs_lookup_fs_root(fs_info, objectid); 1441 if (root) 1442 return root; 1443 1444 key.objectid = objectid; 1445 key.type = BTRFS_ROOT_ITEM_KEY; 1446 key.offset = (u64)-1; 1447 root = read_tree_root_path(fs_info->tree_root, path, &key); 1448 btrfs_release_path(path); 1449 1450 return root; 1451 } 1452 1453 static int cleaner_kthread(void *arg) 1454 { 1455 struct btrfs_fs_info *fs_info = arg; 1456 int again; 1457 1458 while (1) { 1459 again = 0; 1460 1461 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1462 1463 /* Make the cleaner go to sleep early. */ 1464 if (btrfs_need_cleaner_sleep(fs_info)) 1465 goto sleep; 1466 1467 /* 1468 * Do not do anything if we might cause open_ctree() to block 1469 * before we have finished mounting the filesystem. 1470 */ 1471 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1472 goto sleep; 1473 1474 if (!mutex_trylock(&fs_info->cleaner_mutex)) 1475 goto sleep; 1476 1477 /* 1478 * Avoid the problem that we change the status of the fs 1479 * during the above check and trylock. 1480 */ 1481 if (btrfs_need_cleaner_sleep(fs_info)) { 1482 mutex_unlock(&fs_info->cleaner_mutex); 1483 goto sleep; 1484 } 1485 1486 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags)) 1487 btrfs_sysfs_feature_update(fs_info); 1488 1489 btrfs_run_delayed_iputs(fs_info); 1490 1491 again = btrfs_clean_one_deleted_snapshot(fs_info); 1492 mutex_unlock(&fs_info->cleaner_mutex); 1493 1494 /* 1495 * The defragger has dealt with the R/O remount and umount, 1496 * needn't do anything special here. 1497 */ 1498 btrfs_run_defrag_inodes(fs_info); 1499 1500 /* 1501 * Acquires fs_info->reclaim_bgs_lock to avoid racing 1502 * with relocation (btrfs_relocate_chunk) and relocation 1503 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1504 * after acquiring fs_info->reclaim_bgs_lock. So we 1505 * can't hold, nor need to, fs_info->cleaner_mutex when deleting 1506 * unused block groups. 1507 */ 1508 btrfs_delete_unused_bgs(fs_info); 1509 1510 /* 1511 * Reclaim block groups in the reclaim_bgs list after we deleted 1512 * all unused block_groups. This possibly gives us some more free 1513 * space. 1514 */ 1515 btrfs_reclaim_bgs(fs_info); 1516 sleep: 1517 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1518 if (kthread_should_park()) 1519 kthread_parkme(); 1520 if (kthread_should_stop()) 1521 return 0; 1522 if (!again) { 1523 set_current_state(TASK_INTERRUPTIBLE); 1524 schedule(); 1525 __set_current_state(TASK_RUNNING); 1526 } 1527 } 1528 } 1529 1530 static int transaction_kthread(void *arg) 1531 { 1532 struct btrfs_root *root = arg; 1533 struct btrfs_fs_info *fs_info = root->fs_info; 1534 struct btrfs_trans_handle *trans; 1535 struct btrfs_transaction *cur; 1536 u64 transid; 1537 time64_t delta; 1538 unsigned long delay; 1539 bool cannot_commit; 1540 1541 do { 1542 cannot_commit = false; 1543 delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 1544 mutex_lock(&fs_info->transaction_kthread_mutex); 1545 1546 spin_lock(&fs_info->trans_lock); 1547 cur = fs_info->running_transaction; 1548 if (!cur) { 1549 spin_unlock(&fs_info->trans_lock); 1550 goto sleep; 1551 } 1552 1553 delta = ktime_get_seconds() - cur->start_time; 1554 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 1555 cur->state < TRANS_STATE_COMMIT_START && 1556 delta < fs_info->commit_interval) { 1557 spin_unlock(&fs_info->trans_lock); 1558 delay -= msecs_to_jiffies((delta - 1) * 1000); 1559 delay = min(delay, 1560 msecs_to_jiffies(fs_info->commit_interval * 1000)); 1561 goto sleep; 1562 } 1563 transid = cur->transid; 1564 spin_unlock(&fs_info->trans_lock); 1565 1566 /* If the file system is aborted, this will always fail. */ 1567 trans = btrfs_attach_transaction(root); 1568 if (IS_ERR(trans)) { 1569 if (PTR_ERR(trans) != -ENOENT) 1570 cannot_commit = true; 1571 goto sleep; 1572 } 1573 if (transid == trans->transid) { 1574 btrfs_commit_transaction(trans); 1575 } else { 1576 btrfs_end_transaction(trans); 1577 } 1578 sleep: 1579 wake_up_process(fs_info->cleaner_kthread); 1580 mutex_unlock(&fs_info->transaction_kthread_mutex); 1581 1582 if (BTRFS_FS_ERROR(fs_info)) 1583 btrfs_cleanup_transaction(fs_info); 1584 if (!kthread_should_stop() && 1585 (!btrfs_transaction_blocked(fs_info) || 1586 cannot_commit)) 1587 schedule_timeout_interruptible(delay); 1588 } while (!kthread_should_stop()); 1589 return 0; 1590 } 1591 1592 /* 1593 * This will find the highest generation in the array of root backups. The 1594 * index of the highest array is returned, or -EINVAL if we can't find 1595 * anything. 1596 * 1597 * We check to make sure the array is valid by comparing the 1598 * generation of the latest root in the array with the generation 1599 * in the super block. If they don't match we pitch it. 1600 */ 1601 static int find_newest_super_backup(struct btrfs_fs_info *info) 1602 { 1603 const u64 newest_gen = btrfs_super_generation(info->super_copy); 1604 u64 cur; 1605 struct btrfs_root_backup *root_backup; 1606 int i; 1607 1608 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1609 root_backup = info->super_copy->super_roots + i; 1610 cur = btrfs_backup_tree_root_gen(root_backup); 1611 if (cur == newest_gen) 1612 return i; 1613 } 1614 1615 return -EINVAL; 1616 } 1617 1618 /* 1619 * copy all the root pointers into the super backup array. 1620 * this will bump the backup pointer by one when it is 1621 * done 1622 */ 1623 static void backup_super_roots(struct btrfs_fs_info *info) 1624 { 1625 const int next_backup = info->backup_root_index; 1626 struct btrfs_root_backup *root_backup; 1627 1628 root_backup = info->super_for_commit->super_roots + next_backup; 1629 1630 /* 1631 * make sure all of our padding and empty slots get zero filled 1632 * regardless of which ones we use today 1633 */ 1634 memset(root_backup, 0, sizeof(*root_backup)); 1635 1636 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1637 1638 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1639 btrfs_set_backup_tree_root_gen(root_backup, 1640 btrfs_header_generation(info->tree_root->node)); 1641 1642 btrfs_set_backup_tree_root_level(root_backup, 1643 btrfs_header_level(info->tree_root->node)); 1644 1645 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1646 btrfs_set_backup_chunk_root_gen(root_backup, 1647 btrfs_header_generation(info->chunk_root->node)); 1648 btrfs_set_backup_chunk_root_level(root_backup, 1649 btrfs_header_level(info->chunk_root->node)); 1650 1651 if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) { 1652 struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 1653 struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 1654 1655 btrfs_set_backup_extent_root(root_backup, 1656 extent_root->node->start); 1657 btrfs_set_backup_extent_root_gen(root_backup, 1658 btrfs_header_generation(extent_root->node)); 1659 btrfs_set_backup_extent_root_level(root_backup, 1660 btrfs_header_level(extent_root->node)); 1661 1662 btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 1663 btrfs_set_backup_csum_root_gen(root_backup, 1664 btrfs_header_generation(csum_root->node)); 1665 btrfs_set_backup_csum_root_level(root_backup, 1666 btrfs_header_level(csum_root->node)); 1667 } 1668 1669 /* 1670 * we might commit during log recovery, which happens before we set 1671 * the fs_root. Make sure it is valid before we fill it in. 1672 */ 1673 if (info->fs_root && info->fs_root->node) { 1674 btrfs_set_backup_fs_root(root_backup, 1675 info->fs_root->node->start); 1676 btrfs_set_backup_fs_root_gen(root_backup, 1677 btrfs_header_generation(info->fs_root->node)); 1678 btrfs_set_backup_fs_root_level(root_backup, 1679 btrfs_header_level(info->fs_root->node)); 1680 } 1681 1682 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1683 btrfs_set_backup_dev_root_gen(root_backup, 1684 btrfs_header_generation(info->dev_root->node)); 1685 btrfs_set_backup_dev_root_level(root_backup, 1686 btrfs_header_level(info->dev_root->node)); 1687 1688 btrfs_set_backup_total_bytes(root_backup, 1689 btrfs_super_total_bytes(info->super_copy)); 1690 btrfs_set_backup_bytes_used(root_backup, 1691 btrfs_super_bytes_used(info->super_copy)); 1692 btrfs_set_backup_num_devices(root_backup, 1693 btrfs_super_num_devices(info->super_copy)); 1694 1695 /* 1696 * if we don't copy this out to the super_copy, it won't get remembered 1697 * for the next commit 1698 */ 1699 memcpy(&info->super_copy->super_roots, 1700 &info->super_for_commit->super_roots, 1701 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1702 } 1703 1704 /* 1705 * read_backup_root - Reads a backup root based on the passed priority. Prio 0 1706 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 1707 * 1708 * fs_info - filesystem whose backup roots need to be read 1709 * priority - priority of backup root required 1710 * 1711 * Returns backup root index on success and -EINVAL otherwise. 1712 */ 1713 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 1714 { 1715 int backup_index = find_newest_super_backup(fs_info); 1716 struct btrfs_super_block *super = fs_info->super_copy; 1717 struct btrfs_root_backup *root_backup; 1718 1719 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 1720 if (priority == 0) 1721 return backup_index; 1722 1723 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 1724 backup_index %= BTRFS_NUM_BACKUP_ROOTS; 1725 } else { 1726 return -EINVAL; 1727 } 1728 1729 root_backup = super->super_roots + backup_index; 1730 1731 btrfs_set_super_generation(super, 1732 btrfs_backup_tree_root_gen(root_backup)); 1733 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 1734 btrfs_set_super_root_level(super, 1735 btrfs_backup_tree_root_level(root_backup)); 1736 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 1737 1738 /* 1739 * Fixme: the total bytes and num_devices need to match or we should 1740 * need a fsck 1741 */ 1742 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 1743 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 1744 1745 return backup_index; 1746 } 1747 1748 /* helper to cleanup workers */ 1749 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 1750 { 1751 btrfs_destroy_workqueue(fs_info->fixup_workers); 1752 btrfs_destroy_workqueue(fs_info->delalloc_workers); 1753 btrfs_destroy_workqueue(fs_info->workers); 1754 if (fs_info->endio_workers) 1755 destroy_workqueue(fs_info->endio_workers); 1756 if (fs_info->rmw_workers) 1757 destroy_workqueue(fs_info->rmw_workers); 1758 if (fs_info->compressed_write_workers) 1759 destroy_workqueue(fs_info->compressed_write_workers); 1760 btrfs_destroy_workqueue(fs_info->endio_write_workers); 1761 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 1762 btrfs_destroy_workqueue(fs_info->delayed_workers); 1763 btrfs_destroy_workqueue(fs_info->caching_workers); 1764 btrfs_destroy_workqueue(fs_info->flush_workers); 1765 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 1766 if (fs_info->discard_ctl.discard_workers) 1767 destroy_workqueue(fs_info->discard_ctl.discard_workers); 1768 /* 1769 * Now that all other work queues are destroyed, we can safely destroy 1770 * the queues used for metadata I/O, since tasks from those other work 1771 * queues can do metadata I/O operations. 1772 */ 1773 if (fs_info->endio_meta_workers) 1774 destroy_workqueue(fs_info->endio_meta_workers); 1775 } 1776 1777 static void free_root_extent_buffers(struct btrfs_root *root) 1778 { 1779 if (root) { 1780 free_extent_buffer(root->node); 1781 free_extent_buffer(root->commit_root); 1782 root->node = NULL; 1783 root->commit_root = NULL; 1784 } 1785 } 1786 1787 static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 1788 { 1789 struct btrfs_root *root, *tmp; 1790 1791 rbtree_postorder_for_each_entry_safe(root, tmp, 1792 &fs_info->global_root_tree, 1793 rb_node) 1794 free_root_extent_buffers(root); 1795 } 1796 1797 /* helper to cleanup tree roots */ 1798 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 1799 { 1800 free_root_extent_buffers(info->tree_root); 1801 1802 free_global_root_pointers(info); 1803 free_root_extent_buffers(info->dev_root); 1804 free_root_extent_buffers(info->quota_root); 1805 free_root_extent_buffers(info->uuid_root); 1806 free_root_extent_buffers(info->fs_root); 1807 free_root_extent_buffers(info->data_reloc_root); 1808 free_root_extent_buffers(info->block_group_root); 1809 if (free_chunk_root) 1810 free_root_extent_buffers(info->chunk_root); 1811 } 1812 1813 void btrfs_put_root(struct btrfs_root *root) 1814 { 1815 if (!root) 1816 return; 1817 1818 if (refcount_dec_and_test(&root->refs)) { 1819 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 1820 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 1821 if (root->anon_dev) 1822 free_anon_bdev(root->anon_dev); 1823 free_root_extent_buffers(root); 1824 #ifdef CONFIG_BTRFS_DEBUG 1825 spin_lock(&root->fs_info->fs_roots_radix_lock); 1826 list_del_init(&root->leak_list); 1827 spin_unlock(&root->fs_info->fs_roots_radix_lock); 1828 #endif 1829 kfree(root); 1830 } 1831 } 1832 1833 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 1834 { 1835 int ret; 1836 struct btrfs_root *gang[8]; 1837 int i; 1838 1839 while (!list_empty(&fs_info->dead_roots)) { 1840 gang[0] = list_entry(fs_info->dead_roots.next, 1841 struct btrfs_root, root_list); 1842 list_del(&gang[0]->root_list); 1843 1844 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 1845 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 1846 btrfs_put_root(gang[0]); 1847 } 1848 1849 while (1) { 1850 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 1851 (void **)gang, 0, 1852 ARRAY_SIZE(gang)); 1853 if (!ret) 1854 break; 1855 for (i = 0; i < ret; i++) 1856 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 1857 } 1858 } 1859 1860 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 1861 { 1862 mutex_init(&fs_info->scrub_lock); 1863 atomic_set(&fs_info->scrubs_running, 0); 1864 atomic_set(&fs_info->scrub_pause_req, 0); 1865 atomic_set(&fs_info->scrubs_paused, 0); 1866 atomic_set(&fs_info->scrub_cancel_req, 0); 1867 init_waitqueue_head(&fs_info->scrub_pause_wait); 1868 refcount_set(&fs_info->scrub_workers_refcnt, 0); 1869 } 1870 1871 static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 1872 { 1873 spin_lock_init(&fs_info->balance_lock); 1874 mutex_init(&fs_info->balance_mutex); 1875 atomic_set(&fs_info->balance_pause_req, 0); 1876 atomic_set(&fs_info->balance_cancel_req, 0); 1877 fs_info->balance_ctl = NULL; 1878 init_waitqueue_head(&fs_info->balance_wait_q); 1879 atomic_set(&fs_info->reloc_cancel_req, 0); 1880 } 1881 1882 static int btrfs_init_btree_inode(struct super_block *sb) 1883 { 1884 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 1885 unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID, 1886 fs_info->tree_root); 1887 struct inode *inode; 1888 1889 inode = new_inode(sb); 1890 if (!inode) 1891 return -ENOMEM; 1892 1893 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 1894 set_nlink(inode, 1); 1895 /* 1896 * we set the i_size on the btree inode to the max possible int. 1897 * the real end of the address space is determined by all of 1898 * the devices in the system 1899 */ 1900 inode->i_size = OFFSET_MAX; 1901 inode->i_mapping->a_ops = &btree_aops; 1902 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS); 1903 1904 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 1905 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 1906 IO_TREE_BTREE_INODE_IO); 1907 extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 1908 1909 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 1910 BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID; 1911 BTRFS_I(inode)->location.type = 0; 1912 BTRFS_I(inode)->location.offset = 0; 1913 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 1914 __insert_inode_hash(inode, hash); 1915 fs_info->btree_inode = inode; 1916 1917 return 0; 1918 } 1919 1920 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 1921 { 1922 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 1923 init_rwsem(&fs_info->dev_replace.rwsem); 1924 init_waitqueue_head(&fs_info->dev_replace.replace_wait); 1925 } 1926 1927 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 1928 { 1929 spin_lock_init(&fs_info->qgroup_lock); 1930 mutex_init(&fs_info->qgroup_ioctl_lock); 1931 fs_info->qgroup_tree = RB_ROOT; 1932 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 1933 fs_info->qgroup_seq = 1; 1934 fs_info->qgroup_ulist = NULL; 1935 fs_info->qgroup_rescan_running = false; 1936 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL; 1937 mutex_init(&fs_info->qgroup_rescan_lock); 1938 } 1939 1940 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 1941 { 1942 u32 max_active = fs_info->thread_pool_size; 1943 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 1944 unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE; 1945 1946 fs_info->workers = 1947 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16); 1948 1949 fs_info->delalloc_workers = 1950 btrfs_alloc_workqueue(fs_info, "delalloc", 1951 flags, max_active, 2); 1952 1953 fs_info->flush_workers = 1954 btrfs_alloc_workqueue(fs_info, "flush_delalloc", 1955 flags, max_active, 0); 1956 1957 fs_info->caching_workers = 1958 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 1959 1960 fs_info->fixup_workers = 1961 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags); 1962 1963 fs_info->endio_workers = 1964 alloc_workqueue("btrfs-endio", flags, max_active); 1965 fs_info->endio_meta_workers = 1966 alloc_workqueue("btrfs-endio-meta", flags, max_active); 1967 fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active); 1968 fs_info->endio_write_workers = 1969 btrfs_alloc_workqueue(fs_info, "endio-write", flags, 1970 max_active, 2); 1971 fs_info->compressed_write_workers = 1972 alloc_workqueue("btrfs-compressed-write", flags, max_active); 1973 fs_info->endio_freespace_worker = 1974 btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 1975 max_active, 0); 1976 fs_info->delayed_workers = 1977 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 1978 max_active, 0); 1979 fs_info->qgroup_rescan_workers = 1980 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan", 1981 ordered_flags); 1982 fs_info->discard_ctl.discard_workers = 1983 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE); 1984 1985 if (!(fs_info->workers && 1986 fs_info->delalloc_workers && fs_info->flush_workers && 1987 fs_info->endio_workers && fs_info->endio_meta_workers && 1988 fs_info->compressed_write_workers && 1989 fs_info->endio_write_workers && 1990 fs_info->endio_freespace_worker && fs_info->rmw_workers && 1991 fs_info->caching_workers && fs_info->fixup_workers && 1992 fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 1993 fs_info->discard_ctl.discard_workers)) { 1994 return -ENOMEM; 1995 } 1996 1997 return 0; 1998 } 1999 2000 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 2001 { 2002 struct crypto_shash *csum_shash; 2003 const char *csum_driver = btrfs_super_csum_driver(csum_type); 2004 2005 csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 2006 2007 if (IS_ERR(csum_shash)) { 2008 btrfs_err(fs_info, "error allocating %s hash for checksum", 2009 csum_driver); 2010 return PTR_ERR(csum_shash); 2011 } 2012 2013 fs_info->csum_shash = csum_shash; 2014 2015 /* 2016 * Check if the checksum implementation is a fast accelerated one. 2017 * As-is this is a bit of a hack and should be replaced once the csum 2018 * implementations provide that information themselves. 2019 */ 2020 switch (csum_type) { 2021 case BTRFS_CSUM_TYPE_CRC32: 2022 if (!strstr(crypto_shash_driver_name(csum_shash), "generic")) 2023 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags); 2024 break; 2025 case BTRFS_CSUM_TYPE_XXHASH: 2026 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags); 2027 break; 2028 default: 2029 break; 2030 } 2031 2032 btrfs_info(fs_info, "using %s (%s) checksum algorithm", 2033 btrfs_super_csum_name(csum_type), 2034 crypto_shash_driver_name(csum_shash)); 2035 return 0; 2036 } 2037 2038 static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 2039 struct btrfs_fs_devices *fs_devices) 2040 { 2041 int ret; 2042 struct btrfs_tree_parent_check check = { 0 }; 2043 struct btrfs_root *log_tree_root; 2044 struct btrfs_super_block *disk_super = fs_info->super_copy; 2045 u64 bytenr = btrfs_super_log_root(disk_super); 2046 int level = btrfs_super_log_root_level(disk_super); 2047 2048 if (fs_devices->rw_devices == 0) { 2049 btrfs_warn(fs_info, "log replay required on RO media"); 2050 return -EIO; 2051 } 2052 2053 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 2054 GFP_KERNEL); 2055 if (!log_tree_root) 2056 return -ENOMEM; 2057 2058 check.level = level; 2059 check.transid = fs_info->generation + 1; 2060 check.owner_root = BTRFS_TREE_LOG_OBJECTID; 2061 log_tree_root->node = read_tree_block(fs_info, bytenr, &check); 2062 if (IS_ERR(log_tree_root->node)) { 2063 btrfs_warn(fs_info, "failed to read log tree"); 2064 ret = PTR_ERR(log_tree_root->node); 2065 log_tree_root->node = NULL; 2066 btrfs_put_root(log_tree_root); 2067 return ret; 2068 } 2069 if (!extent_buffer_uptodate(log_tree_root->node)) { 2070 btrfs_err(fs_info, "failed to read log tree"); 2071 btrfs_put_root(log_tree_root); 2072 return -EIO; 2073 } 2074 2075 /* returns with log_tree_root freed on success */ 2076 ret = btrfs_recover_log_trees(log_tree_root); 2077 if (ret) { 2078 btrfs_handle_fs_error(fs_info, ret, 2079 "Failed to recover log tree"); 2080 btrfs_put_root(log_tree_root); 2081 return ret; 2082 } 2083 2084 if (sb_rdonly(fs_info->sb)) { 2085 ret = btrfs_commit_super(fs_info); 2086 if (ret) 2087 return ret; 2088 } 2089 2090 return 0; 2091 } 2092 2093 static int load_global_roots_objectid(struct btrfs_root *tree_root, 2094 struct btrfs_path *path, u64 objectid, 2095 const char *name) 2096 { 2097 struct btrfs_fs_info *fs_info = tree_root->fs_info; 2098 struct btrfs_root *root; 2099 u64 max_global_id = 0; 2100 int ret; 2101 struct btrfs_key key = { 2102 .objectid = objectid, 2103 .type = BTRFS_ROOT_ITEM_KEY, 2104 .offset = 0, 2105 }; 2106 bool found = false; 2107 2108 /* If we have IGNOREDATACSUMS skip loading these roots. */ 2109 if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2110 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2111 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2112 return 0; 2113 } 2114 2115 while (1) { 2116 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2117 if (ret < 0) 2118 break; 2119 2120 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2121 ret = btrfs_next_leaf(tree_root, path); 2122 if (ret) { 2123 if (ret > 0) 2124 ret = 0; 2125 break; 2126 } 2127 } 2128 ret = 0; 2129 2130 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2131 if (key.objectid != objectid) 2132 break; 2133 btrfs_release_path(path); 2134 2135 /* 2136 * Just worry about this for extent tree, it'll be the same for 2137 * everybody. 2138 */ 2139 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2140 max_global_id = max(max_global_id, key.offset); 2141 2142 found = true; 2143 root = read_tree_root_path(tree_root, path, &key); 2144 if (IS_ERR(root)) { 2145 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2146 ret = PTR_ERR(root); 2147 break; 2148 } 2149 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2150 ret = btrfs_global_root_insert(root); 2151 if (ret) { 2152 btrfs_put_root(root); 2153 break; 2154 } 2155 key.offset++; 2156 } 2157 btrfs_release_path(path); 2158 2159 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2160 fs_info->nr_global_roots = max_global_id + 1; 2161 2162 if (!found || ret) { 2163 if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2164 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2165 2166 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2167 ret = ret ? ret : -ENOENT; 2168 else 2169 ret = 0; 2170 btrfs_err(fs_info, "failed to load root %s", name); 2171 } 2172 return ret; 2173 } 2174 2175 static int load_global_roots(struct btrfs_root *tree_root) 2176 { 2177 struct btrfs_path *path; 2178 int ret = 0; 2179 2180 path = btrfs_alloc_path(); 2181 if (!path) 2182 return -ENOMEM; 2183 2184 ret = load_global_roots_objectid(tree_root, path, 2185 BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2186 if (ret) 2187 goto out; 2188 ret = load_global_roots_objectid(tree_root, path, 2189 BTRFS_CSUM_TREE_OBJECTID, "csum"); 2190 if (ret) 2191 goto out; 2192 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2193 goto out; 2194 ret = load_global_roots_objectid(tree_root, path, 2195 BTRFS_FREE_SPACE_TREE_OBJECTID, 2196 "free space"); 2197 out: 2198 btrfs_free_path(path); 2199 return ret; 2200 } 2201 2202 static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 2203 { 2204 struct btrfs_root *tree_root = fs_info->tree_root; 2205 struct btrfs_root *root; 2206 struct btrfs_key location; 2207 int ret; 2208 2209 BUG_ON(!fs_info->tree_root); 2210 2211 ret = load_global_roots(tree_root); 2212 if (ret) 2213 return ret; 2214 2215 location.type = BTRFS_ROOT_ITEM_KEY; 2216 location.offset = 0; 2217 2218 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) { 2219 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID; 2220 root = btrfs_read_tree_root(tree_root, &location); 2221 if (IS_ERR(root)) { 2222 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2223 ret = PTR_ERR(root); 2224 goto out; 2225 } 2226 } else { 2227 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2228 fs_info->block_group_root = root; 2229 } 2230 } 2231 2232 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2233 root = btrfs_read_tree_root(tree_root, &location); 2234 if (IS_ERR(root)) { 2235 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2236 ret = PTR_ERR(root); 2237 goto out; 2238 } 2239 } else { 2240 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2241 fs_info->dev_root = root; 2242 } 2243 /* Initialize fs_info for all devices in any case */ 2244 ret = btrfs_init_devices_late(fs_info); 2245 if (ret) 2246 goto out; 2247 2248 /* 2249 * This tree can share blocks with some other fs tree during relocation 2250 * and we need a proper setup by btrfs_get_fs_root 2251 */ 2252 root = btrfs_get_fs_root(tree_root->fs_info, 2253 BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2254 if (IS_ERR(root)) { 2255 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2256 ret = PTR_ERR(root); 2257 goto out; 2258 } 2259 } else { 2260 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2261 fs_info->data_reloc_root = root; 2262 } 2263 2264 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2265 root = btrfs_read_tree_root(tree_root, &location); 2266 if (!IS_ERR(root)) { 2267 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2268 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2269 fs_info->quota_root = root; 2270 } 2271 2272 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2273 root = btrfs_read_tree_root(tree_root, &location); 2274 if (IS_ERR(root)) { 2275 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2276 ret = PTR_ERR(root); 2277 if (ret != -ENOENT) 2278 goto out; 2279 } 2280 } else { 2281 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2282 fs_info->uuid_root = root; 2283 } 2284 2285 return 0; 2286 out: 2287 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2288 location.objectid, ret); 2289 return ret; 2290 } 2291 2292 /* 2293 * Real super block validation 2294 * NOTE: super csum type and incompat features will not be checked here. 2295 * 2296 * @sb: super block to check 2297 * @mirror_num: the super block number to check its bytenr: 2298 * 0 the primary (1st) sb 2299 * 1, 2 2nd and 3rd backup copy 2300 * -1 skip bytenr check 2301 */ 2302 int btrfs_validate_super(struct btrfs_fs_info *fs_info, 2303 struct btrfs_super_block *sb, int mirror_num) 2304 { 2305 u64 nodesize = btrfs_super_nodesize(sb); 2306 u64 sectorsize = btrfs_super_sectorsize(sb); 2307 int ret = 0; 2308 2309 if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 2310 btrfs_err(fs_info, "no valid FS found"); 2311 ret = -EINVAL; 2312 } 2313 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 2314 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 2315 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 2316 ret = -EINVAL; 2317 } 2318 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 2319 btrfs_err(fs_info, "tree_root level too big: %d >= %d", 2320 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 2321 ret = -EINVAL; 2322 } 2323 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 2324 btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 2325 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 2326 ret = -EINVAL; 2327 } 2328 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 2329 btrfs_err(fs_info, "log_root level too big: %d >= %d", 2330 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 2331 ret = -EINVAL; 2332 } 2333 2334 /* 2335 * Check sectorsize and nodesize first, other check will need it. 2336 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 2337 */ 2338 if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 2339 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2340 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 2341 ret = -EINVAL; 2342 } 2343 2344 /* 2345 * We only support at most two sectorsizes: 4K and PAGE_SIZE. 2346 * 2347 * We can support 16K sectorsize with 64K page size without problem, 2348 * but such sectorsize/pagesize combination doesn't make much sense. 2349 * 4K will be our future standard, PAGE_SIZE is supported from the very 2350 * beginning. 2351 */ 2352 if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) { 2353 btrfs_err(fs_info, 2354 "sectorsize %llu not yet supported for page size %lu", 2355 sectorsize, PAGE_SIZE); 2356 ret = -EINVAL; 2357 } 2358 2359 if (!is_power_of_2(nodesize) || nodesize < sectorsize || 2360 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2361 btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 2362 ret = -EINVAL; 2363 } 2364 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 2365 btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 2366 le32_to_cpu(sb->__unused_leafsize), nodesize); 2367 ret = -EINVAL; 2368 } 2369 2370 /* Root alignment check */ 2371 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 2372 btrfs_warn(fs_info, "tree_root block unaligned: %llu", 2373 btrfs_super_root(sb)); 2374 ret = -EINVAL; 2375 } 2376 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 2377 btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 2378 btrfs_super_chunk_root(sb)); 2379 ret = -EINVAL; 2380 } 2381 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 2382 btrfs_warn(fs_info, "log_root block unaligned: %llu", 2383 btrfs_super_log_root(sb)); 2384 ret = -EINVAL; 2385 } 2386 2387 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2388 BTRFS_FSID_SIZE)) { 2389 btrfs_err(fs_info, 2390 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2391 fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2392 ret = -EINVAL; 2393 } 2394 2395 if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2396 memcmp(fs_info->fs_devices->metadata_uuid, 2397 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2398 btrfs_err(fs_info, 2399 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2400 fs_info->super_copy->metadata_uuid, 2401 fs_info->fs_devices->metadata_uuid); 2402 ret = -EINVAL; 2403 } 2404 2405 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 2406 BTRFS_FSID_SIZE) != 0) { 2407 btrfs_err(fs_info, 2408 "dev_item UUID does not match metadata fsid: %pU != %pU", 2409 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 2410 ret = -EINVAL; 2411 } 2412 2413 /* 2414 * Artificial requirement for block-group-tree to force newer features 2415 * (free-space-tree, no-holes) so the test matrix is smaller. 2416 */ 2417 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 2418 (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) || 2419 !btrfs_fs_incompat(fs_info, NO_HOLES))) { 2420 btrfs_err(fs_info, 2421 "block-group-tree feature requires fres-space-tree and no-holes"); 2422 ret = -EINVAL; 2423 } 2424 2425 /* 2426 * Hint to catch really bogus numbers, bitflips or so, more exact checks are 2427 * done later 2428 */ 2429 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 2430 btrfs_err(fs_info, "bytes_used is too small %llu", 2431 btrfs_super_bytes_used(sb)); 2432 ret = -EINVAL; 2433 } 2434 if (!is_power_of_2(btrfs_super_stripesize(sb))) { 2435 btrfs_err(fs_info, "invalid stripesize %u", 2436 btrfs_super_stripesize(sb)); 2437 ret = -EINVAL; 2438 } 2439 if (btrfs_super_num_devices(sb) > (1UL << 31)) 2440 btrfs_warn(fs_info, "suspicious number of devices: %llu", 2441 btrfs_super_num_devices(sb)); 2442 if (btrfs_super_num_devices(sb) == 0) { 2443 btrfs_err(fs_info, "number of devices is 0"); 2444 ret = -EINVAL; 2445 } 2446 2447 if (mirror_num >= 0 && 2448 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 2449 btrfs_err(fs_info, "super offset mismatch %llu != %u", 2450 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 2451 ret = -EINVAL; 2452 } 2453 2454 /* 2455 * Obvious sys_chunk_array corruptions, it must hold at least one key 2456 * and one chunk 2457 */ 2458 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 2459 btrfs_err(fs_info, "system chunk array too big %u > %u", 2460 btrfs_super_sys_array_size(sb), 2461 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 2462 ret = -EINVAL; 2463 } 2464 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 2465 + sizeof(struct btrfs_chunk)) { 2466 btrfs_err(fs_info, "system chunk array too small %u < %zu", 2467 btrfs_super_sys_array_size(sb), 2468 sizeof(struct btrfs_disk_key) 2469 + sizeof(struct btrfs_chunk)); 2470 ret = -EINVAL; 2471 } 2472 2473 /* 2474 * The generation is a global counter, we'll trust it more than the others 2475 * but it's still possible that it's the one that's wrong. 2476 */ 2477 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 2478 btrfs_warn(fs_info, 2479 "suspicious: generation < chunk_root_generation: %llu < %llu", 2480 btrfs_super_generation(sb), 2481 btrfs_super_chunk_root_generation(sb)); 2482 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 2483 && btrfs_super_cache_generation(sb) != (u64)-1) 2484 btrfs_warn(fs_info, 2485 "suspicious: generation < cache_generation: %llu < %llu", 2486 btrfs_super_generation(sb), 2487 btrfs_super_cache_generation(sb)); 2488 2489 return ret; 2490 } 2491 2492 /* 2493 * Validation of super block at mount time. 2494 * Some checks already done early at mount time, like csum type and incompat 2495 * flags will be skipped. 2496 */ 2497 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2498 { 2499 return btrfs_validate_super(fs_info, fs_info->super_copy, 0); 2500 } 2501 2502 /* 2503 * Validation of super block at write time. 2504 * Some checks like bytenr check will be skipped as their values will be 2505 * overwritten soon. 2506 * Extra checks like csum type and incompat flags will be done here. 2507 */ 2508 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 2509 struct btrfs_super_block *sb) 2510 { 2511 int ret; 2512 2513 ret = btrfs_validate_super(fs_info, sb, -1); 2514 if (ret < 0) 2515 goto out; 2516 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 2517 ret = -EUCLEAN; 2518 btrfs_err(fs_info, "invalid csum type, has %u want %u", 2519 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 2520 goto out; 2521 } 2522 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 2523 ret = -EUCLEAN; 2524 btrfs_err(fs_info, 2525 "invalid incompat flags, has 0x%llx valid mask 0x%llx", 2526 btrfs_super_incompat_flags(sb), 2527 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 2528 goto out; 2529 } 2530 out: 2531 if (ret < 0) 2532 btrfs_err(fs_info, 2533 "super block corruption detected before writing it to disk"); 2534 return ret; 2535 } 2536 2537 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2538 { 2539 struct btrfs_tree_parent_check check = { 2540 .level = level, 2541 .transid = gen, 2542 .owner_root = root->root_key.objectid 2543 }; 2544 int ret = 0; 2545 2546 root->node = read_tree_block(root->fs_info, bytenr, &check); 2547 if (IS_ERR(root->node)) { 2548 ret = PTR_ERR(root->node); 2549 root->node = NULL; 2550 return ret; 2551 } 2552 if (!extent_buffer_uptodate(root->node)) { 2553 free_extent_buffer(root->node); 2554 root->node = NULL; 2555 return -EIO; 2556 } 2557 2558 btrfs_set_root_node(&root->root_item, root->node); 2559 root->commit_root = btrfs_root_node(root); 2560 btrfs_set_root_refs(&root->root_item, 1); 2561 return ret; 2562 } 2563 2564 static int load_important_roots(struct btrfs_fs_info *fs_info) 2565 { 2566 struct btrfs_super_block *sb = fs_info->super_copy; 2567 u64 gen, bytenr; 2568 int level, ret; 2569 2570 bytenr = btrfs_super_root(sb); 2571 gen = btrfs_super_generation(sb); 2572 level = btrfs_super_root_level(sb); 2573 ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 2574 if (ret) { 2575 btrfs_warn(fs_info, "couldn't read tree root"); 2576 return ret; 2577 } 2578 return 0; 2579 } 2580 2581 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2582 { 2583 int backup_index = find_newest_super_backup(fs_info); 2584 struct btrfs_super_block *sb = fs_info->super_copy; 2585 struct btrfs_root *tree_root = fs_info->tree_root; 2586 bool handle_error = false; 2587 int ret = 0; 2588 int i; 2589 2590 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2591 if (handle_error) { 2592 if (!IS_ERR(tree_root->node)) 2593 free_extent_buffer(tree_root->node); 2594 tree_root->node = NULL; 2595 2596 if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2597 break; 2598 2599 free_root_pointers(fs_info, 0); 2600 2601 /* 2602 * Don't use the log in recovery mode, it won't be 2603 * valid 2604 */ 2605 btrfs_set_super_log_root(sb, 0); 2606 2607 /* We can't trust the free space cache either */ 2608 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2609 2610 btrfs_warn(fs_info, "try to load backup roots slot %d", i); 2611 ret = read_backup_root(fs_info, i); 2612 backup_index = ret; 2613 if (ret < 0) 2614 return ret; 2615 } 2616 2617 ret = load_important_roots(fs_info); 2618 if (ret) { 2619 handle_error = true; 2620 continue; 2621 } 2622 2623 /* 2624 * No need to hold btrfs_root::objectid_mutex since the fs 2625 * hasn't been fully initialised and we are the only user 2626 */ 2627 ret = btrfs_init_root_free_objectid(tree_root); 2628 if (ret < 0) { 2629 handle_error = true; 2630 continue; 2631 } 2632 2633 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2634 2635 ret = btrfs_read_roots(fs_info); 2636 if (ret < 0) { 2637 handle_error = true; 2638 continue; 2639 } 2640 2641 /* All successful */ 2642 fs_info->generation = btrfs_header_generation(tree_root->node); 2643 fs_info->last_trans_committed = fs_info->generation; 2644 fs_info->last_reloc_trans = 0; 2645 2646 /* Always begin writing backup roots after the one being used */ 2647 if (backup_index < 0) { 2648 fs_info->backup_root_index = 0; 2649 } else { 2650 fs_info->backup_root_index = backup_index + 1; 2651 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 2652 } 2653 break; 2654 } 2655 2656 return ret; 2657 } 2658 2659 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2660 { 2661 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2662 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 2663 INIT_LIST_HEAD(&fs_info->trans_list); 2664 INIT_LIST_HEAD(&fs_info->dead_roots); 2665 INIT_LIST_HEAD(&fs_info->delayed_iputs); 2666 INIT_LIST_HEAD(&fs_info->delalloc_roots); 2667 INIT_LIST_HEAD(&fs_info->caching_block_groups); 2668 spin_lock_init(&fs_info->delalloc_root_lock); 2669 spin_lock_init(&fs_info->trans_lock); 2670 spin_lock_init(&fs_info->fs_roots_radix_lock); 2671 spin_lock_init(&fs_info->delayed_iput_lock); 2672 spin_lock_init(&fs_info->defrag_inodes_lock); 2673 spin_lock_init(&fs_info->super_lock); 2674 spin_lock_init(&fs_info->buffer_lock); 2675 spin_lock_init(&fs_info->unused_bgs_lock); 2676 spin_lock_init(&fs_info->treelog_bg_lock); 2677 spin_lock_init(&fs_info->zone_active_bgs_lock); 2678 spin_lock_init(&fs_info->relocation_bg_lock); 2679 rwlock_init(&fs_info->tree_mod_log_lock); 2680 rwlock_init(&fs_info->global_root_lock); 2681 mutex_init(&fs_info->unused_bg_unpin_mutex); 2682 mutex_init(&fs_info->reclaim_bgs_lock); 2683 mutex_init(&fs_info->reloc_mutex); 2684 mutex_init(&fs_info->delalloc_root_mutex); 2685 mutex_init(&fs_info->zoned_meta_io_lock); 2686 mutex_init(&fs_info->zoned_data_reloc_io_lock); 2687 seqlock_init(&fs_info->profiles_lock); 2688 2689 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers); 2690 btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters); 2691 btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered); 2692 btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent); 2693 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start, 2694 BTRFS_LOCKDEP_TRANS_COMMIT_START); 2695 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked, 2696 BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2697 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed, 2698 BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2699 btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed, 2700 BTRFS_LOCKDEP_TRANS_COMPLETED); 2701 2702 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 2703 INIT_LIST_HEAD(&fs_info->space_info); 2704 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 2705 INIT_LIST_HEAD(&fs_info->unused_bgs); 2706 INIT_LIST_HEAD(&fs_info->reclaim_bgs); 2707 INIT_LIST_HEAD(&fs_info->zone_active_bgs); 2708 #ifdef CONFIG_BTRFS_DEBUG 2709 INIT_LIST_HEAD(&fs_info->allocated_roots); 2710 INIT_LIST_HEAD(&fs_info->allocated_ebs); 2711 spin_lock_init(&fs_info->eb_leak_lock); 2712 #endif 2713 extent_map_tree_init(&fs_info->mapping_tree); 2714 btrfs_init_block_rsv(&fs_info->global_block_rsv, 2715 BTRFS_BLOCK_RSV_GLOBAL); 2716 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 2717 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 2718 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 2719 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 2720 BTRFS_BLOCK_RSV_DELOPS); 2721 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2722 BTRFS_BLOCK_RSV_DELREFS); 2723 2724 atomic_set(&fs_info->async_delalloc_pages, 0); 2725 atomic_set(&fs_info->defrag_running, 0); 2726 atomic_set(&fs_info->nr_delayed_iputs, 0); 2727 atomic64_set(&fs_info->tree_mod_seq, 0); 2728 fs_info->global_root_tree = RB_ROOT; 2729 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 2730 fs_info->metadata_ratio = 0; 2731 fs_info->defrag_inodes = RB_ROOT; 2732 atomic64_set(&fs_info->free_chunk_space, 0); 2733 fs_info->tree_mod_log = RB_ROOT; 2734 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2735 btrfs_init_ref_verify(fs_info); 2736 2737 fs_info->thread_pool_size = min_t(unsigned long, 2738 num_online_cpus() + 2, 8); 2739 2740 INIT_LIST_HEAD(&fs_info->ordered_roots); 2741 spin_lock_init(&fs_info->ordered_root_lock); 2742 2743 btrfs_init_scrub(fs_info); 2744 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2745 fs_info->check_integrity_print_mask = 0; 2746 #endif 2747 btrfs_init_balance(fs_info); 2748 btrfs_init_async_reclaim_work(fs_info); 2749 2750 rwlock_init(&fs_info->block_group_cache_lock); 2751 fs_info->block_group_cache_tree = RB_ROOT_CACHED; 2752 2753 extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2754 IO_TREE_FS_EXCLUDED_EXTENTS); 2755 2756 mutex_init(&fs_info->ordered_operations_mutex); 2757 mutex_init(&fs_info->tree_log_mutex); 2758 mutex_init(&fs_info->chunk_mutex); 2759 mutex_init(&fs_info->transaction_kthread_mutex); 2760 mutex_init(&fs_info->cleaner_mutex); 2761 mutex_init(&fs_info->ro_block_group_mutex); 2762 init_rwsem(&fs_info->commit_root_sem); 2763 init_rwsem(&fs_info->cleanup_work_sem); 2764 init_rwsem(&fs_info->subvol_sem); 2765 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2766 2767 btrfs_init_dev_replace_locks(fs_info); 2768 btrfs_init_qgroup(fs_info); 2769 btrfs_discard_init(fs_info); 2770 2771 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2772 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2773 2774 init_waitqueue_head(&fs_info->transaction_throttle); 2775 init_waitqueue_head(&fs_info->transaction_wait); 2776 init_waitqueue_head(&fs_info->transaction_blocked_wait); 2777 init_waitqueue_head(&fs_info->async_submit_wait); 2778 init_waitqueue_head(&fs_info->delayed_iputs_wait); 2779 2780 /* Usable values until the real ones are cached from the superblock */ 2781 fs_info->nodesize = 4096; 2782 fs_info->sectorsize = 4096; 2783 fs_info->sectorsize_bits = ilog2(4096); 2784 fs_info->stripesize = 4096; 2785 2786 fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE; 2787 2788 spin_lock_init(&fs_info->swapfile_pins_lock); 2789 fs_info->swapfile_pins = RB_ROOT; 2790 2791 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 2792 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 2793 } 2794 2795 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 2796 { 2797 int ret; 2798 2799 fs_info->sb = sb; 2800 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 2801 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2802 2803 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 2804 if (ret) 2805 return ret; 2806 2807 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2808 if (ret) 2809 return ret; 2810 2811 fs_info->dirty_metadata_batch = PAGE_SIZE * 2812 (1 + ilog2(nr_cpu_ids)); 2813 2814 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2815 if (ret) 2816 return ret; 2817 2818 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 2819 GFP_KERNEL); 2820 if (ret) 2821 return ret; 2822 2823 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2824 GFP_KERNEL); 2825 if (!fs_info->delayed_root) 2826 return -ENOMEM; 2827 btrfs_init_delayed_root(fs_info->delayed_root); 2828 2829 if (sb_rdonly(sb)) 2830 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 2831 2832 return btrfs_alloc_stripe_hash_table(fs_info); 2833 } 2834 2835 static int btrfs_uuid_rescan_kthread(void *data) 2836 { 2837 struct btrfs_fs_info *fs_info = data; 2838 int ret; 2839 2840 /* 2841 * 1st step is to iterate through the existing UUID tree and 2842 * to delete all entries that contain outdated data. 2843 * 2nd step is to add all missing entries to the UUID tree. 2844 */ 2845 ret = btrfs_uuid_tree_iterate(fs_info); 2846 if (ret < 0) { 2847 if (ret != -EINTR) 2848 btrfs_warn(fs_info, "iterating uuid_tree failed %d", 2849 ret); 2850 up(&fs_info->uuid_tree_rescan_sem); 2851 return ret; 2852 } 2853 return btrfs_uuid_scan_kthread(data); 2854 } 2855 2856 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 2857 { 2858 struct task_struct *task; 2859 2860 down(&fs_info->uuid_tree_rescan_sem); 2861 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 2862 if (IS_ERR(task)) { 2863 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 2864 btrfs_warn(fs_info, "failed to start uuid_rescan task"); 2865 up(&fs_info->uuid_tree_rescan_sem); 2866 return PTR_ERR(task); 2867 } 2868 2869 return 0; 2870 } 2871 2872 /* 2873 * Some options only have meaning at mount time and shouldn't persist across 2874 * remounts, or be displayed. Clear these at the end of mount and remount 2875 * code paths. 2876 */ 2877 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 2878 { 2879 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 2880 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 2881 } 2882 2883 /* 2884 * Mounting logic specific to read-write file systems. Shared by open_ctree 2885 * and btrfs_remount when remounting from read-only to read-write. 2886 */ 2887 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 2888 { 2889 int ret; 2890 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 2891 bool rebuild_free_space_tree = false; 2892 2893 if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 2894 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2895 rebuild_free_space_tree = true; 2896 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 2897 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 2898 btrfs_warn(fs_info, "free space tree is invalid"); 2899 rebuild_free_space_tree = true; 2900 } 2901 2902 if (rebuild_free_space_tree) { 2903 btrfs_info(fs_info, "rebuilding free space tree"); 2904 ret = btrfs_rebuild_free_space_tree(fs_info); 2905 if (ret) { 2906 btrfs_warn(fs_info, 2907 "failed to rebuild free space tree: %d", ret); 2908 goto out; 2909 } 2910 } 2911 2912 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 2913 !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) { 2914 btrfs_info(fs_info, "disabling free space tree"); 2915 ret = btrfs_delete_free_space_tree(fs_info); 2916 if (ret) { 2917 btrfs_warn(fs_info, 2918 "failed to disable free space tree: %d", ret); 2919 goto out; 2920 } 2921 } 2922 2923 /* 2924 * btrfs_find_orphan_roots() is responsible for finding all the dead 2925 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 2926 * them into the fs_info->fs_roots_radix tree. This must be done before 2927 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 2928 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 2929 * item before the root's tree is deleted - this means that if we unmount 2930 * or crash before the deletion completes, on the next mount we will not 2931 * delete what remains of the tree because the orphan item does not 2932 * exists anymore, which is what tells us we have a pending deletion. 2933 */ 2934 ret = btrfs_find_orphan_roots(fs_info); 2935 if (ret) 2936 goto out; 2937 2938 ret = btrfs_cleanup_fs_roots(fs_info); 2939 if (ret) 2940 goto out; 2941 2942 down_read(&fs_info->cleanup_work_sem); 2943 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 2944 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 2945 up_read(&fs_info->cleanup_work_sem); 2946 goto out; 2947 } 2948 up_read(&fs_info->cleanup_work_sem); 2949 2950 mutex_lock(&fs_info->cleaner_mutex); 2951 ret = btrfs_recover_relocation(fs_info); 2952 mutex_unlock(&fs_info->cleaner_mutex); 2953 if (ret < 0) { 2954 btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 2955 goto out; 2956 } 2957 2958 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 2959 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2960 btrfs_info(fs_info, "creating free space tree"); 2961 ret = btrfs_create_free_space_tree(fs_info); 2962 if (ret) { 2963 btrfs_warn(fs_info, 2964 "failed to create free space tree: %d", ret); 2965 goto out; 2966 } 2967 } 2968 2969 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 2970 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 2971 if (ret) 2972 goto out; 2973 } 2974 2975 ret = btrfs_resume_balance_async(fs_info); 2976 if (ret) 2977 goto out; 2978 2979 ret = btrfs_resume_dev_replace_async(fs_info); 2980 if (ret) { 2981 btrfs_warn(fs_info, "failed to resume dev_replace"); 2982 goto out; 2983 } 2984 2985 btrfs_qgroup_rescan_resume(fs_info); 2986 2987 if (!fs_info->uuid_root) { 2988 btrfs_info(fs_info, "creating UUID tree"); 2989 ret = btrfs_create_uuid_tree(fs_info); 2990 if (ret) { 2991 btrfs_warn(fs_info, 2992 "failed to create the UUID tree %d", ret); 2993 goto out; 2994 } 2995 } 2996 2997 out: 2998 return ret; 2999 } 3000 3001 /* 3002 * Do various sanity and dependency checks of different features. 3003 * 3004 * @is_rw_mount: If the mount is read-write. 3005 * 3006 * This is the place for less strict checks (like for subpage or artificial 3007 * feature dependencies). 3008 * 3009 * For strict checks or possible corruption detection, see 3010 * btrfs_validate_super(). 3011 * 3012 * This should be called after btrfs_parse_options(), as some mount options 3013 * (space cache related) can modify on-disk format like free space tree and 3014 * screw up certain feature dependencies. 3015 */ 3016 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount) 3017 { 3018 struct btrfs_super_block *disk_super = fs_info->super_copy; 3019 u64 incompat = btrfs_super_incompat_flags(disk_super); 3020 const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super); 3021 const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP); 3022 3023 if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 3024 btrfs_err(fs_info, 3025 "cannot mount because of unknown incompat features (0x%llx)", 3026 incompat); 3027 return -EINVAL; 3028 } 3029 3030 /* Runtime limitation for mixed block groups. */ 3031 if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3032 (fs_info->sectorsize != fs_info->nodesize)) { 3033 btrfs_err(fs_info, 3034 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3035 fs_info->nodesize, fs_info->sectorsize); 3036 return -EINVAL; 3037 } 3038 3039 /* Mixed backref is an always-enabled feature. */ 3040 incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3041 3042 /* Set compression related flags just in case. */ 3043 if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3044 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3045 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3046 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3047 3048 /* 3049 * An ancient flag, which should really be marked deprecated. 3050 * Such runtime limitation doesn't really need a incompat flag. 3051 */ 3052 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) 3053 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3054 3055 if (compat_ro_unsupp && is_rw_mount) { 3056 btrfs_err(fs_info, 3057 "cannot mount read-write because of unknown compat_ro features (0x%llx)", 3058 compat_ro); 3059 return -EINVAL; 3060 } 3061 3062 /* 3063 * We have unsupported RO compat features, although RO mounted, we 3064 * should not cause any metadata writes, including log replay. 3065 * Or we could screw up whatever the new feature requires. 3066 */ 3067 if (compat_ro_unsupp && btrfs_super_log_root(disk_super) && 3068 !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3069 btrfs_err(fs_info, 3070 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay", 3071 compat_ro); 3072 return -EINVAL; 3073 } 3074 3075 /* 3076 * Artificial limitations for block group tree, to force 3077 * block-group-tree to rely on no-holes and free-space-tree. 3078 */ 3079 if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) && 3080 (!btrfs_fs_incompat(fs_info, NO_HOLES) || 3081 !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) { 3082 btrfs_err(fs_info, 3083 "block-group-tree feature requires no-holes and free-space-tree features"); 3084 return -EINVAL; 3085 } 3086 3087 /* 3088 * Subpage runtime limitation on v1 cache. 3089 * 3090 * V1 space cache still has some hard codeed PAGE_SIZE usage, while 3091 * we're already defaulting to v2 cache, no need to bother v1 as it's 3092 * going to be deprecated anyway. 3093 */ 3094 if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) { 3095 btrfs_warn(fs_info, 3096 "v1 space cache is not supported for page size %lu with sectorsize %u", 3097 PAGE_SIZE, fs_info->sectorsize); 3098 return -EINVAL; 3099 } 3100 3101 /* This can be called by remount, we need to protect the super block. */ 3102 spin_lock(&fs_info->super_lock); 3103 btrfs_set_super_incompat_flags(disk_super, incompat); 3104 spin_unlock(&fs_info->super_lock); 3105 3106 return 0; 3107 } 3108 3109 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3110 char *options) 3111 { 3112 u32 sectorsize; 3113 u32 nodesize; 3114 u32 stripesize; 3115 u64 generation; 3116 u64 features; 3117 u16 csum_type; 3118 struct btrfs_super_block *disk_super; 3119 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3120 struct btrfs_root *tree_root; 3121 struct btrfs_root *chunk_root; 3122 int ret; 3123 int level; 3124 3125 ret = init_mount_fs_info(fs_info, sb); 3126 if (ret) 3127 goto fail; 3128 3129 /* These need to be init'ed before we start creating inodes and such. */ 3130 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3131 GFP_KERNEL); 3132 fs_info->tree_root = tree_root; 3133 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3134 GFP_KERNEL); 3135 fs_info->chunk_root = chunk_root; 3136 if (!tree_root || !chunk_root) { 3137 ret = -ENOMEM; 3138 goto fail; 3139 } 3140 3141 ret = btrfs_init_btree_inode(sb); 3142 if (ret) 3143 goto fail; 3144 3145 invalidate_bdev(fs_devices->latest_dev->bdev); 3146 3147 /* 3148 * Read super block and check the signature bytes only 3149 */ 3150 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 3151 if (IS_ERR(disk_super)) { 3152 ret = PTR_ERR(disk_super); 3153 goto fail_alloc; 3154 } 3155 3156 /* 3157 * Verify the type first, if that or the checksum value are 3158 * corrupted, we'll find out 3159 */ 3160 csum_type = btrfs_super_csum_type(disk_super); 3161 if (!btrfs_supported_super_csum(csum_type)) { 3162 btrfs_err(fs_info, "unsupported checksum algorithm: %u", 3163 csum_type); 3164 ret = -EINVAL; 3165 btrfs_release_disk_super(disk_super); 3166 goto fail_alloc; 3167 } 3168 3169 fs_info->csum_size = btrfs_super_csum_size(disk_super); 3170 3171 ret = btrfs_init_csum_hash(fs_info, csum_type); 3172 if (ret) { 3173 btrfs_release_disk_super(disk_super); 3174 goto fail_alloc; 3175 } 3176 3177 /* 3178 * We want to check superblock checksum, the type is stored inside. 3179 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 3180 */ 3181 if (btrfs_check_super_csum(fs_info, disk_super)) { 3182 btrfs_err(fs_info, "superblock checksum mismatch"); 3183 ret = -EINVAL; 3184 btrfs_release_disk_super(disk_super); 3185 goto fail_alloc; 3186 } 3187 3188 /* 3189 * super_copy is zeroed at allocation time and we never touch the 3190 * following bytes up to INFO_SIZE, the checksum is calculated from 3191 * the whole block of INFO_SIZE 3192 */ 3193 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 3194 btrfs_release_disk_super(disk_super); 3195 3196 disk_super = fs_info->super_copy; 3197 3198 3199 features = btrfs_super_flags(disk_super); 3200 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3201 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3202 btrfs_set_super_flags(disk_super, features); 3203 btrfs_info(fs_info, 3204 "found metadata UUID change in progress flag, clearing"); 3205 } 3206 3207 memcpy(fs_info->super_for_commit, fs_info->super_copy, 3208 sizeof(*fs_info->super_for_commit)); 3209 3210 ret = btrfs_validate_mount_super(fs_info); 3211 if (ret) { 3212 btrfs_err(fs_info, "superblock contains fatal errors"); 3213 ret = -EINVAL; 3214 goto fail_alloc; 3215 } 3216 3217 if (!btrfs_super_root(disk_super)) { 3218 btrfs_err(fs_info, "invalid superblock tree root bytenr"); 3219 ret = -EINVAL; 3220 goto fail_alloc; 3221 } 3222 3223 /* check FS state, whether FS is broken. */ 3224 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 3225 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3226 3227 /* 3228 * In the long term, we'll store the compression type in the super 3229 * block, and it'll be used for per file compression control. 3230 */ 3231 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 3232 3233 3234 /* Set up fs_info before parsing mount options */ 3235 nodesize = btrfs_super_nodesize(disk_super); 3236 sectorsize = btrfs_super_sectorsize(disk_super); 3237 stripesize = sectorsize; 3238 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3239 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3240 3241 fs_info->nodesize = nodesize; 3242 fs_info->sectorsize = sectorsize; 3243 fs_info->sectorsize_bits = ilog2(sectorsize); 3244 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 3245 fs_info->stripesize = stripesize; 3246 3247 ret = btrfs_parse_options(fs_info, options, sb->s_flags); 3248 if (ret) 3249 goto fail_alloc; 3250 3251 ret = btrfs_check_features(fs_info, !sb_rdonly(sb)); 3252 if (ret < 0) 3253 goto fail_alloc; 3254 3255 if (sectorsize < PAGE_SIZE) { 3256 struct btrfs_subpage_info *subpage_info; 3257 3258 /* 3259 * V1 space cache has some hardcoded PAGE_SIZE usage, and is 3260 * going to be deprecated. 3261 * 3262 * Force to use v2 cache for subpage case. 3263 */ 3264 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE); 3265 btrfs_set_and_info(fs_info, FREE_SPACE_TREE, 3266 "forcing free space tree for sector size %u with page size %lu", 3267 sectorsize, PAGE_SIZE); 3268 3269 btrfs_warn(fs_info, 3270 "read-write for sector size %u with page size %lu is experimental", 3271 sectorsize, PAGE_SIZE); 3272 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 3273 if (!subpage_info) { 3274 ret = -ENOMEM; 3275 goto fail_alloc; 3276 } 3277 btrfs_init_subpage_info(subpage_info, sectorsize); 3278 fs_info->subpage_info = subpage_info; 3279 } 3280 3281 ret = btrfs_init_workqueues(fs_info); 3282 if (ret) 3283 goto fail_sb_buffer; 3284 3285 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 3286 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 3287 3288 sb->s_blocksize = sectorsize; 3289 sb->s_blocksize_bits = blksize_bits(sectorsize); 3290 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3291 3292 mutex_lock(&fs_info->chunk_mutex); 3293 ret = btrfs_read_sys_array(fs_info); 3294 mutex_unlock(&fs_info->chunk_mutex); 3295 if (ret) { 3296 btrfs_err(fs_info, "failed to read the system array: %d", ret); 3297 goto fail_sb_buffer; 3298 } 3299 3300 generation = btrfs_super_chunk_root_generation(disk_super); 3301 level = btrfs_super_chunk_root_level(disk_super); 3302 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3303 generation, level); 3304 if (ret) { 3305 btrfs_err(fs_info, "failed to read chunk root"); 3306 goto fail_tree_roots; 3307 } 3308 3309 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3310 offsetof(struct btrfs_header, chunk_tree_uuid), 3311 BTRFS_UUID_SIZE); 3312 3313 ret = btrfs_read_chunk_tree(fs_info); 3314 if (ret) { 3315 btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3316 goto fail_tree_roots; 3317 } 3318 3319 /* 3320 * At this point we know all the devices that make this filesystem, 3321 * including the seed devices but we don't know yet if the replace 3322 * target is required. So free devices that are not part of this 3323 * filesystem but skip the replace target device which is checked 3324 * below in btrfs_init_dev_replace(). 3325 */ 3326 btrfs_free_extra_devids(fs_devices); 3327 if (!fs_devices->latest_dev->bdev) { 3328 btrfs_err(fs_info, "failed to read devices"); 3329 ret = -EIO; 3330 goto fail_tree_roots; 3331 } 3332 3333 ret = init_tree_roots(fs_info); 3334 if (ret) 3335 goto fail_tree_roots; 3336 3337 /* 3338 * Get zone type information of zoned block devices. This will also 3339 * handle emulation of a zoned filesystem if a regular device has the 3340 * zoned incompat feature flag set. 3341 */ 3342 ret = btrfs_get_dev_zone_info_all_devices(fs_info); 3343 if (ret) { 3344 btrfs_err(fs_info, 3345 "zoned: failed to read device zone info: %d", ret); 3346 goto fail_block_groups; 3347 } 3348 3349 /* 3350 * If we have a uuid root and we're not being told to rescan we need to 3351 * check the generation here so we can set the 3352 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 3353 * transaction during a balance or the log replay without updating the 3354 * uuid generation, and then if we crash we would rescan the uuid tree, 3355 * even though it was perfectly fine. 3356 */ 3357 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 3358 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 3359 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3360 3361 ret = btrfs_verify_dev_extents(fs_info); 3362 if (ret) { 3363 btrfs_err(fs_info, 3364 "failed to verify dev extents against chunks: %d", 3365 ret); 3366 goto fail_block_groups; 3367 } 3368 ret = btrfs_recover_balance(fs_info); 3369 if (ret) { 3370 btrfs_err(fs_info, "failed to recover balance: %d", ret); 3371 goto fail_block_groups; 3372 } 3373 3374 ret = btrfs_init_dev_stats(fs_info); 3375 if (ret) { 3376 btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3377 goto fail_block_groups; 3378 } 3379 3380 ret = btrfs_init_dev_replace(fs_info); 3381 if (ret) { 3382 btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 3383 goto fail_block_groups; 3384 } 3385 3386 ret = btrfs_check_zoned_mode(fs_info); 3387 if (ret) { 3388 btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3389 ret); 3390 goto fail_block_groups; 3391 } 3392 3393 ret = btrfs_sysfs_add_fsid(fs_devices); 3394 if (ret) { 3395 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 3396 ret); 3397 goto fail_block_groups; 3398 } 3399 3400 ret = btrfs_sysfs_add_mounted(fs_info); 3401 if (ret) { 3402 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3403 goto fail_fsdev_sysfs; 3404 } 3405 3406 ret = btrfs_init_space_info(fs_info); 3407 if (ret) { 3408 btrfs_err(fs_info, "failed to initialize space info: %d", ret); 3409 goto fail_sysfs; 3410 } 3411 3412 ret = btrfs_read_block_groups(fs_info); 3413 if (ret) { 3414 btrfs_err(fs_info, "failed to read block groups: %d", ret); 3415 goto fail_sysfs; 3416 } 3417 3418 btrfs_free_zone_cache(fs_info); 3419 3420 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 3421 !btrfs_check_rw_degradable(fs_info, NULL)) { 3422 btrfs_warn(fs_info, 3423 "writable mount is not allowed due to too many missing devices"); 3424 ret = -EINVAL; 3425 goto fail_sysfs; 3426 } 3427 3428 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3429 "btrfs-cleaner"); 3430 if (IS_ERR(fs_info->cleaner_kthread)) { 3431 ret = PTR_ERR(fs_info->cleaner_kthread); 3432 goto fail_sysfs; 3433 } 3434 3435 fs_info->transaction_kthread = kthread_run(transaction_kthread, 3436 tree_root, 3437 "btrfs-transaction"); 3438 if (IS_ERR(fs_info->transaction_kthread)) { 3439 ret = PTR_ERR(fs_info->transaction_kthread); 3440 goto fail_cleaner; 3441 } 3442 3443 if (!btrfs_test_opt(fs_info, NOSSD) && 3444 !fs_info->fs_devices->rotating) { 3445 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3446 } 3447 3448 /* 3449 * For devices supporting discard turn on discard=async automatically, 3450 * unless it's already set or disabled. This could be turned off by 3451 * nodiscard for the same mount. 3452 * 3453 * The zoned mode piggy backs on the discard functionality for 3454 * resetting a zone. There is no reason to delay the zone reset as it is 3455 * fast enough. So, do not enable async discard for zoned mode. 3456 */ 3457 if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) || 3458 btrfs_test_opt(fs_info, DISCARD_ASYNC) || 3459 btrfs_test_opt(fs_info, NODISCARD)) && 3460 fs_info->fs_devices->discardable && 3461 !btrfs_is_zoned(fs_info)) { 3462 btrfs_set_and_info(fs_info, DISCARD_ASYNC, 3463 "auto enabling async discard"); 3464 } 3465 3466 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3467 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 3468 ret = btrfsic_mount(fs_info, fs_devices, 3469 btrfs_test_opt(fs_info, 3470 CHECK_INTEGRITY_DATA) ? 1 : 0, 3471 fs_info->check_integrity_print_mask); 3472 if (ret) 3473 btrfs_warn(fs_info, 3474 "failed to initialize integrity check module: %d", 3475 ret); 3476 } 3477 #endif 3478 ret = btrfs_read_qgroup_config(fs_info); 3479 if (ret) 3480 goto fail_trans_kthread; 3481 3482 if (btrfs_build_ref_tree(fs_info)) 3483 btrfs_err(fs_info, "couldn't build ref tree"); 3484 3485 /* do not make disk changes in broken FS or nologreplay is given */ 3486 if (btrfs_super_log_root(disk_super) != 0 && 3487 !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3488 btrfs_info(fs_info, "start tree-log replay"); 3489 ret = btrfs_replay_log(fs_info, fs_devices); 3490 if (ret) 3491 goto fail_qgroup; 3492 } 3493 3494 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 3495 if (IS_ERR(fs_info->fs_root)) { 3496 ret = PTR_ERR(fs_info->fs_root); 3497 btrfs_warn(fs_info, "failed to read fs tree: %d", ret); 3498 fs_info->fs_root = NULL; 3499 goto fail_qgroup; 3500 } 3501 3502 if (sb_rdonly(sb)) 3503 goto clear_oneshot; 3504 3505 ret = btrfs_start_pre_rw_mount(fs_info); 3506 if (ret) { 3507 close_ctree(fs_info); 3508 return ret; 3509 } 3510 btrfs_discard_resume(fs_info); 3511 3512 if (fs_info->uuid_root && 3513 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 3514 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 3515 btrfs_info(fs_info, "checking UUID tree"); 3516 ret = btrfs_check_uuid_tree(fs_info); 3517 if (ret) { 3518 btrfs_warn(fs_info, 3519 "failed to check the UUID tree: %d", ret); 3520 close_ctree(fs_info); 3521 return ret; 3522 } 3523 } 3524 3525 set_bit(BTRFS_FS_OPEN, &fs_info->flags); 3526 3527 /* Kick the cleaner thread so it'll start deleting snapshots. */ 3528 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3529 wake_up_process(fs_info->cleaner_kthread); 3530 3531 clear_oneshot: 3532 btrfs_clear_oneshot_options(fs_info); 3533 return 0; 3534 3535 fail_qgroup: 3536 btrfs_free_qgroup_config(fs_info); 3537 fail_trans_kthread: 3538 kthread_stop(fs_info->transaction_kthread); 3539 btrfs_cleanup_transaction(fs_info); 3540 btrfs_free_fs_roots(fs_info); 3541 fail_cleaner: 3542 kthread_stop(fs_info->cleaner_kthread); 3543 3544 /* 3545 * make sure we're done with the btree inode before we stop our 3546 * kthreads 3547 */ 3548 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 3549 3550 fail_sysfs: 3551 btrfs_sysfs_remove_mounted(fs_info); 3552 3553 fail_fsdev_sysfs: 3554 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3555 3556 fail_block_groups: 3557 btrfs_put_block_group_cache(fs_info); 3558 3559 fail_tree_roots: 3560 if (fs_info->data_reloc_root) 3561 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 3562 free_root_pointers(fs_info, true); 3563 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3564 3565 fail_sb_buffer: 3566 btrfs_stop_all_workers(fs_info); 3567 btrfs_free_block_groups(fs_info); 3568 fail_alloc: 3569 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3570 3571 iput(fs_info->btree_inode); 3572 fail: 3573 btrfs_close_devices(fs_info->fs_devices); 3574 ASSERT(ret < 0); 3575 return ret; 3576 } 3577 ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3578 3579 static void btrfs_end_super_write(struct bio *bio) 3580 { 3581 struct btrfs_device *device = bio->bi_private; 3582 struct bio_vec *bvec; 3583 struct bvec_iter_all iter_all; 3584 struct page *page; 3585 3586 bio_for_each_segment_all(bvec, bio, iter_all) { 3587 page = bvec->bv_page; 3588 3589 if (bio->bi_status) { 3590 btrfs_warn_rl_in_rcu(device->fs_info, 3591 "lost page write due to IO error on %s (%d)", 3592 btrfs_dev_name(device), 3593 blk_status_to_errno(bio->bi_status)); 3594 ClearPageUptodate(page); 3595 SetPageError(page); 3596 btrfs_dev_stat_inc_and_print(device, 3597 BTRFS_DEV_STAT_WRITE_ERRS); 3598 } else { 3599 SetPageUptodate(page); 3600 } 3601 3602 put_page(page); 3603 unlock_page(page); 3604 } 3605 3606 bio_put(bio); 3607 } 3608 3609 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3610 int copy_num, bool drop_cache) 3611 { 3612 struct btrfs_super_block *super; 3613 struct page *page; 3614 u64 bytenr, bytenr_orig; 3615 struct address_space *mapping = bdev->bd_inode->i_mapping; 3616 int ret; 3617 3618 bytenr_orig = btrfs_sb_offset(copy_num); 3619 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 3620 if (ret == -ENOENT) 3621 return ERR_PTR(-EINVAL); 3622 else if (ret) 3623 return ERR_PTR(ret); 3624 3625 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 3626 return ERR_PTR(-EINVAL); 3627 3628 if (drop_cache) { 3629 /* This should only be called with the primary sb. */ 3630 ASSERT(copy_num == 0); 3631 3632 /* 3633 * Drop the page of the primary superblock, so later read will 3634 * always read from the device. 3635 */ 3636 invalidate_inode_pages2_range(mapping, 3637 bytenr >> PAGE_SHIFT, 3638 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT); 3639 } 3640 3641 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 3642 if (IS_ERR(page)) 3643 return ERR_CAST(page); 3644 3645 super = page_address(page); 3646 if (btrfs_super_magic(super) != BTRFS_MAGIC) { 3647 btrfs_release_disk_super(super); 3648 return ERR_PTR(-ENODATA); 3649 } 3650 3651 if (btrfs_super_bytenr(super) != bytenr_orig) { 3652 btrfs_release_disk_super(super); 3653 return ERR_PTR(-EINVAL); 3654 } 3655 3656 return super; 3657 } 3658 3659 3660 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3661 { 3662 struct btrfs_super_block *super, *latest = NULL; 3663 int i; 3664 u64 transid = 0; 3665 3666 /* we would like to check all the supers, but that would make 3667 * a btrfs mount succeed after a mkfs from a different FS. 3668 * So, we need to add a special mount option to scan for 3669 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3670 */ 3671 for (i = 0; i < 1; i++) { 3672 super = btrfs_read_dev_one_super(bdev, i, false); 3673 if (IS_ERR(super)) 3674 continue; 3675 3676 if (!latest || btrfs_super_generation(super) > transid) { 3677 if (latest) 3678 btrfs_release_disk_super(super); 3679 3680 latest = super; 3681 transid = btrfs_super_generation(super); 3682 } 3683 } 3684 3685 return super; 3686 } 3687 3688 /* 3689 * Write superblock @sb to the @device. Do not wait for completion, all the 3690 * pages we use for writing are locked. 3691 * 3692 * Write @max_mirrors copies of the superblock, where 0 means default that fit 3693 * the expected device size at commit time. Note that max_mirrors must be 3694 * same for write and wait phases. 3695 * 3696 * Return number of errors when page is not found or submission fails. 3697 */ 3698 static int write_dev_supers(struct btrfs_device *device, 3699 struct btrfs_super_block *sb, int max_mirrors) 3700 { 3701 struct btrfs_fs_info *fs_info = device->fs_info; 3702 struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3703 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3704 int i; 3705 int errors = 0; 3706 int ret; 3707 u64 bytenr, bytenr_orig; 3708 3709 if (max_mirrors == 0) 3710 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3711 3712 shash->tfm = fs_info->csum_shash; 3713 3714 for (i = 0; i < max_mirrors; i++) { 3715 struct page *page; 3716 struct bio *bio; 3717 struct btrfs_super_block *disk_super; 3718 3719 bytenr_orig = btrfs_sb_offset(i); 3720 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 3721 if (ret == -ENOENT) { 3722 continue; 3723 } else if (ret < 0) { 3724 btrfs_err(device->fs_info, 3725 "couldn't get super block location for mirror %d", 3726 i); 3727 errors++; 3728 continue; 3729 } 3730 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3731 device->commit_total_bytes) 3732 break; 3733 3734 btrfs_set_super_bytenr(sb, bytenr_orig); 3735 3736 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3737 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3738 sb->csum); 3739 3740 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3741 GFP_NOFS); 3742 if (!page) { 3743 btrfs_err(device->fs_info, 3744 "couldn't get super block page for bytenr %llu", 3745 bytenr); 3746 errors++; 3747 continue; 3748 } 3749 3750 /* Bump the refcount for wait_dev_supers() */ 3751 get_page(page); 3752 3753 disk_super = page_address(page); 3754 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3755 3756 /* 3757 * Directly use bios here instead of relying on the page cache 3758 * to do I/O, so we don't lose the ability to do integrity 3759 * checking. 3760 */ 3761 bio = bio_alloc(device->bdev, 1, 3762 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 3763 GFP_NOFS); 3764 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3765 bio->bi_private = device; 3766 bio->bi_end_io = btrfs_end_super_write; 3767 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3768 offset_in_page(bytenr)); 3769 3770 /* 3771 * We FUA only the first super block. The others we allow to 3772 * go down lazy and there's a short window where the on-disk 3773 * copies might still contain the older version. 3774 */ 3775 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3776 bio->bi_opf |= REQ_FUA; 3777 3778 btrfsic_check_bio(bio); 3779 submit_bio(bio); 3780 3781 if (btrfs_advance_sb_log(device, i)) 3782 errors++; 3783 } 3784 return errors < i ? 0 : -1; 3785 } 3786 3787 /* 3788 * Wait for write completion of superblocks done by write_dev_supers, 3789 * @max_mirrors same for write and wait phases. 3790 * 3791 * Return number of errors when page is not found or not marked up to 3792 * date. 3793 */ 3794 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3795 { 3796 int i; 3797 int errors = 0; 3798 bool primary_failed = false; 3799 int ret; 3800 u64 bytenr; 3801 3802 if (max_mirrors == 0) 3803 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3804 3805 for (i = 0; i < max_mirrors; i++) { 3806 struct page *page; 3807 3808 ret = btrfs_sb_log_location(device, i, READ, &bytenr); 3809 if (ret == -ENOENT) { 3810 break; 3811 } else if (ret < 0) { 3812 errors++; 3813 if (i == 0) 3814 primary_failed = true; 3815 continue; 3816 } 3817 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3818 device->commit_total_bytes) 3819 break; 3820 3821 page = find_get_page(device->bdev->bd_inode->i_mapping, 3822 bytenr >> PAGE_SHIFT); 3823 if (!page) { 3824 errors++; 3825 if (i == 0) 3826 primary_failed = true; 3827 continue; 3828 } 3829 /* Page is submitted locked and unlocked once the IO completes */ 3830 wait_on_page_locked(page); 3831 if (PageError(page)) { 3832 errors++; 3833 if (i == 0) 3834 primary_failed = true; 3835 } 3836 3837 /* Drop our reference */ 3838 put_page(page); 3839 3840 /* Drop the reference from the writing run */ 3841 put_page(page); 3842 } 3843 3844 /* log error, force error return */ 3845 if (primary_failed) { 3846 btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3847 device->devid); 3848 return -1; 3849 } 3850 3851 return errors < i ? 0 : -1; 3852 } 3853 3854 /* 3855 * endio for the write_dev_flush, this will wake anyone waiting 3856 * for the barrier when it is done 3857 */ 3858 static void btrfs_end_empty_barrier(struct bio *bio) 3859 { 3860 bio_uninit(bio); 3861 complete(bio->bi_private); 3862 } 3863 3864 /* 3865 * Submit a flush request to the device if it supports it. Error handling is 3866 * done in the waiting counterpart. 3867 */ 3868 static void write_dev_flush(struct btrfs_device *device) 3869 { 3870 struct bio *bio = &device->flush_bio; 3871 3872 device->last_flush_error = BLK_STS_OK; 3873 3874 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3875 /* 3876 * When a disk has write caching disabled, we skip submission of a bio 3877 * with flush and sync requests before writing the superblock, since 3878 * it's not needed. However when the integrity checker is enabled, this 3879 * results in reports that there are metadata blocks referred by a 3880 * superblock that were not properly flushed. So don't skip the bio 3881 * submission only when the integrity checker is enabled for the sake 3882 * of simplicity, since this is a debug tool and not meant for use in 3883 * non-debug builds. 3884 */ 3885 if (!bdev_write_cache(device->bdev)) 3886 return; 3887 #endif 3888 3889 bio_init(bio, device->bdev, NULL, 0, 3890 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 3891 bio->bi_end_io = btrfs_end_empty_barrier; 3892 init_completion(&device->flush_wait); 3893 bio->bi_private = &device->flush_wait; 3894 3895 btrfsic_check_bio(bio); 3896 submit_bio(bio); 3897 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 3898 } 3899 3900 /* 3901 * If the flush bio has been submitted by write_dev_flush, wait for it. 3902 * Return true for any error, and false otherwise. 3903 */ 3904 static bool wait_dev_flush(struct btrfs_device *device) 3905 { 3906 struct bio *bio = &device->flush_bio; 3907 3908 if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 3909 return false; 3910 3911 wait_for_completion_io(&device->flush_wait); 3912 3913 if (bio->bi_status) { 3914 device->last_flush_error = bio->bi_status; 3915 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS); 3916 return true; 3917 } 3918 3919 return false; 3920 } 3921 3922 /* 3923 * send an empty flush down to each device in parallel, 3924 * then wait for them 3925 */ 3926 static int barrier_all_devices(struct btrfs_fs_info *info) 3927 { 3928 struct list_head *head; 3929 struct btrfs_device *dev; 3930 int errors_wait = 0; 3931 3932 lockdep_assert_held(&info->fs_devices->device_list_mutex); 3933 /* send down all the barriers */ 3934 head = &info->fs_devices->devices; 3935 list_for_each_entry(dev, head, dev_list) { 3936 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3937 continue; 3938 if (!dev->bdev) 3939 continue; 3940 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3941 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3942 continue; 3943 3944 write_dev_flush(dev); 3945 } 3946 3947 /* wait for all the barriers */ 3948 list_for_each_entry(dev, head, dev_list) { 3949 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3950 continue; 3951 if (!dev->bdev) { 3952 errors_wait++; 3953 continue; 3954 } 3955 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3956 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3957 continue; 3958 3959 if (wait_dev_flush(dev)) 3960 errors_wait++; 3961 } 3962 3963 /* 3964 * Checks last_flush_error of disks in order to determine the device 3965 * state. 3966 */ 3967 if (errors_wait && !btrfs_check_rw_degradable(info, NULL)) 3968 return -EIO; 3969 3970 return 0; 3971 } 3972 3973 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3974 { 3975 int raid_type; 3976 int min_tolerated = INT_MAX; 3977 3978 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 3979 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 3980 min_tolerated = min_t(int, min_tolerated, 3981 btrfs_raid_array[BTRFS_RAID_SINGLE]. 3982 tolerated_failures); 3983 3984 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 3985 if (raid_type == BTRFS_RAID_SINGLE) 3986 continue; 3987 if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 3988 continue; 3989 min_tolerated = min_t(int, min_tolerated, 3990 btrfs_raid_array[raid_type]. 3991 tolerated_failures); 3992 } 3993 3994 if (min_tolerated == INT_MAX) { 3995 pr_warn("BTRFS: unknown raid flag: %llu", flags); 3996 min_tolerated = 0; 3997 } 3998 3999 return min_tolerated; 4000 } 4001 4002 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4003 { 4004 struct list_head *head; 4005 struct btrfs_device *dev; 4006 struct btrfs_super_block *sb; 4007 struct btrfs_dev_item *dev_item; 4008 int ret; 4009 int do_barriers; 4010 int max_errors; 4011 int total_errors = 0; 4012 u64 flags; 4013 4014 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4015 4016 /* 4017 * max_mirrors == 0 indicates we're from commit_transaction, 4018 * not from fsync where the tree roots in fs_info have not 4019 * been consistent on disk. 4020 */ 4021 if (max_mirrors == 0) 4022 backup_super_roots(fs_info); 4023 4024 sb = fs_info->super_for_commit; 4025 dev_item = &sb->dev_item; 4026 4027 mutex_lock(&fs_info->fs_devices->device_list_mutex); 4028 head = &fs_info->fs_devices->devices; 4029 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4030 4031 if (do_barriers) { 4032 ret = barrier_all_devices(fs_info); 4033 if (ret) { 4034 mutex_unlock( 4035 &fs_info->fs_devices->device_list_mutex); 4036 btrfs_handle_fs_error(fs_info, ret, 4037 "errors while submitting device barriers."); 4038 return ret; 4039 } 4040 } 4041 4042 list_for_each_entry(dev, head, dev_list) { 4043 if (!dev->bdev) { 4044 total_errors++; 4045 continue; 4046 } 4047 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4048 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4049 continue; 4050 4051 btrfs_set_stack_device_generation(dev_item, 0); 4052 btrfs_set_stack_device_type(dev_item, dev->type); 4053 btrfs_set_stack_device_id(dev_item, dev->devid); 4054 btrfs_set_stack_device_total_bytes(dev_item, 4055 dev->commit_total_bytes); 4056 btrfs_set_stack_device_bytes_used(dev_item, 4057 dev->commit_bytes_used); 4058 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4059 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4060 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4061 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 4062 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 4063 BTRFS_FSID_SIZE); 4064 4065 flags = btrfs_super_flags(sb); 4066 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4067 4068 ret = btrfs_validate_write_super(fs_info, sb); 4069 if (ret < 0) { 4070 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4071 btrfs_handle_fs_error(fs_info, -EUCLEAN, 4072 "unexpected superblock corruption detected"); 4073 return -EUCLEAN; 4074 } 4075 4076 ret = write_dev_supers(dev, sb, max_mirrors); 4077 if (ret) 4078 total_errors++; 4079 } 4080 if (total_errors > max_errors) { 4081 btrfs_err(fs_info, "%d errors while writing supers", 4082 total_errors); 4083 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4084 4085 /* FUA is masked off if unsupported and can't be the reason */ 4086 btrfs_handle_fs_error(fs_info, -EIO, 4087 "%d errors while writing supers", 4088 total_errors); 4089 return -EIO; 4090 } 4091 4092 total_errors = 0; 4093 list_for_each_entry(dev, head, dev_list) { 4094 if (!dev->bdev) 4095 continue; 4096 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4097 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4098 continue; 4099 4100 ret = wait_dev_supers(dev, max_mirrors); 4101 if (ret) 4102 total_errors++; 4103 } 4104 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4105 if (total_errors > max_errors) { 4106 btrfs_handle_fs_error(fs_info, -EIO, 4107 "%d errors while writing supers", 4108 total_errors); 4109 return -EIO; 4110 } 4111 return 0; 4112 } 4113 4114 /* Drop a fs root from the radix tree and free it. */ 4115 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4116 struct btrfs_root *root) 4117 { 4118 bool drop_ref = false; 4119 4120 spin_lock(&fs_info->fs_roots_radix_lock); 4121 radix_tree_delete(&fs_info->fs_roots_radix, 4122 (unsigned long)root->root_key.objectid); 4123 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 4124 drop_ref = true; 4125 spin_unlock(&fs_info->fs_roots_radix_lock); 4126 4127 if (BTRFS_FS_ERROR(fs_info)) { 4128 ASSERT(root->log_root == NULL); 4129 if (root->reloc_root) { 4130 btrfs_put_root(root->reloc_root); 4131 root->reloc_root = NULL; 4132 } 4133 } 4134 4135 if (drop_ref) 4136 btrfs_put_root(root); 4137 } 4138 4139 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4140 { 4141 u64 root_objectid = 0; 4142 struct btrfs_root *gang[8]; 4143 int i = 0; 4144 int err = 0; 4145 unsigned int ret = 0; 4146 4147 while (1) { 4148 spin_lock(&fs_info->fs_roots_radix_lock); 4149 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4150 (void **)gang, root_objectid, 4151 ARRAY_SIZE(gang)); 4152 if (!ret) { 4153 spin_unlock(&fs_info->fs_roots_radix_lock); 4154 break; 4155 } 4156 root_objectid = gang[ret - 1]->root_key.objectid + 1; 4157 4158 for (i = 0; i < ret; i++) { 4159 /* Avoid to grab roots in dead_roots */ 4160 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4161 gang[i] = NULL; 4162 continue; 4163 } 4164 /* grab all the search result for later use */ 4165 gang[i] = btrfs_grab_root(gang[i]); 4166 } 4167 spin_unlock(&fs_info->fs_roots_radix_lock); 4168 4169 for (i = 0; i < ret; i++) { 4170 if (!gang[i]) 4171 continue; 4172 root_objectid = gang[i]->root_key.objectid; 4173 err = btrfs_orphan_cleanup(gang[i]); 4174 if (err) 4175 goto out; 4176 btrfs_put_root(gang[i]); 4177 } 4178 root_objectid++; 4179 } 4180 out: 4181 /* release the uncleaned roots due to error */ 4182 for (; i < ret; i++) { 4183 if (gang[i]) 4184 btrfs_put_root(gang[i]); 4185 } 4186 return err; 4187 } 4188 4189 int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4190 { 4191 struct btrfs_root *root = fs_info->tree_root; 4192 struct btrfs_trans_handle *trans; 4193 4194 mutex_lock(&fs_info->cleaner_mutex); 4195 btrfs_run_delayed_iputs(fs_info); 4196 mutex_unlock(&fs_info->cleaner_mutex); 4197 wake_up_process(fs_info->cleaner_kthread); 4198 4199 /* wait until ongoing cleanup work done */ 4200 down_write(&fs_info->cleanup_work_sem); 4201 up_write(&fs_info->cleanup_work_sem); 4202 4203 trans = btrfs_join_transaction(root); 4204 if (IS_ERR(trans)) 4205 return PTR_ERR(trans); 4206 return btrfs_commit_transaction(trans); 4207 } 4208 4209 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 4210 { 4211 struct btrfs_transaction *trans; 4212 struct btrfs_transaction *tmp; 4213 bool found = false; 4214 4215 if (list_empty(&fs_info->trans_list)) 4216 return; 4217 4218 /* 4219 * This function is only called at the very end of close_ctree(), 4220 * thus no other running transaction, no need to take trans_lock. 4221 */ 4222 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 4223 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 4224 struct extent_state *cached = NULL; 4225 u64 dirty_bytes = 0; 4226 u64 cur = 0; 4227 u64 found_start; 4228 u64 found_end; 4229 4230 found = true; 4231 while (!find_first_extent_bit(&trans->dirty_pages, cur, 4232 &found_start, &found_end, EXTENT_DIRTY, &cached)) { 4233 dirty_bytes += found_end + 1 - found_start; 4234 cur = found_end + 1; 4235 } 4236 btrfs_warn(fs_info, 4237 "transaction %llu (with %llu dirty metadata bytes) is not committed", 4238 trans->transid, dirty_bytes); 4239 btrfs_cleanup_one_transaction(trans, fs_info); 4240 4241 if (trans == fs_info->running_transaction) 4242 fs_info->running_transaction = NULL; 4243 list_del_init(&trans->list); 4244 4245 btrfs_put_transaction(trans); 4246 trace_btrfs_transaction_commit(fs_info); 4247 } 4248 ASSERT(!found); 4249 } 4250 4251 void __cold close_ctree(struct btrfs_fs_info *fs_info) 4252 { 4253 int ret; 4254 4255 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4256 4257 /* 4258 * If we had UNFINISHED_DROPS we could still be processing them, so 4259 * clear that bit and wake up relocation so it can stop. 4260 * We must do this before stopping the block group reclaim task, because 4261 * at btrfs_relocate_block_group() we wait for this bit, and after the 4262 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we 4263 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will 4264 * return 1. 4265 */ 4266 btrfs_wake_unfinished_drop(fs_info); 4267 4268 /* 4269 * We may have the reclaim task running and relocating a data block group, 4270 * in which case it may create delayed iputs. So stop it before we park 4271 * the cleaner kthread otherwise we can get new delayed iputs after 4272 * parking the cleaner, and that can make the async reclaim task to hang 4273 * if it's waiting for delayed iputs to complete, since the cleaner is 4274 * parked and can not run delayed iputs - this will make us hang when 4275 * trying to stop the async reclaim task. 4276 */ 4277 cancel_work_sync(&fs_info->reclaim_bgs_work); 4278 /* 4279 * We don't want the cleaner to start new transactions, add more delayed 4280 * iputs, etc. while we're closing. We can't use kthread_stop() yet 4281 * because that frees the task_struct, and the transaction kthread might 4282 * still try to wake up the cleaner. 4283 */ 4284 kthread_park(fs_info->cleaner_kthread); 4285 4286 /* wait for the qgroup rescan worker to stop */ 4287 btrfs_qgroup_wait_for_completion(fs_info, false); 4288 4289 /* wait for the uuid_scan task to finish */ 4290 down(&fs_info->uuid_tree_rescan_sem); 4291 /* avoid complains from lockdep et al., set sem back to initial state */ 4292 up(&fs_info->uuid_tree_rescan_sem); 4293 4294 /* pause restriper - we want to resume on mount */ 4295 btrfs_pause_balance(fs_info); 4296 4297 btrfs_dev_replace_suspend_for_unmount(fs_info); 4298 4299 btrfs_scrub_cancel(fs_info); 4300 4301 /* wait for any defraggers to finish */ 4302 wait_event(fs_info->transaction_wait, 4303 (atomic_read(&fs_info->defrag_running) == 0)); 4304 4305 /* clear out the rbtree of defraggable inodes */ 4306 btrfs_cleanup_defrag_inodes(fs_info); 4307 4308 /* 4309 * After we parked the cleaner kthread, ordered extents may have 4310 * completed and created new delayed iputs. If one of the async reclaim 4311 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we 4312 * can hang forever trying to stop it, because if a delayed iput is 4313 * added after it ran btrfs_run_delayed_iputs() and before it called 4314 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is 4315 * no one else to run iputs. 4316 * 4317 * So wait for all ongoing ordered extents to complete and then run 4318 * delayed iputs. This works because once we reach this point no one 4319 * can either create new ordered extents nor create delayed iputs 4320 * through some other means. 4321 * 4322 * Also note that btrfs_wait_ordered_roots() is not safe here, because 4323 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent, 4324 * but the delayed iput for the respective inode is made only when doing 4325 * the final btrfs_put_ordered_extent() (which must happen at 4326 * btrfs_finish_ordered_io() when we are unmounting). 4327 */ 4328 btrfs_flush_workqueue(fs_info->endio_write_workers); 4329 /* Ordered extents for free space inodes. */ 4330 btrfs_flush_workqueue(fs_info->endio_freespace_worker); 4331 btrfs_run_delayed_iputs(fs_info); 4332 4333 cancel_work_sync(&fs_info->async_reclaim_work); 4334 cancel_work_sync(&fs_info->async_data_reclaim_work); 4335 cancel_work_sync(&fs_info->preempt_reclaim_work); 4336 4337 /* Cancel or finish ongoing discard work */ 4338 btrfs_discard_cleanup(fs_info); 4339 4340 if (!sb_rdonly(fs_info->sb)) { 4341 /* 4342 * The cleaner kthread is stopped, so do one final pass over 4343 * unused block groups. 4344 */ 4345 btrfs_delete_unused_bgs(fs_info); 4346 4347 /* 4348 * There might be existing delayed inode workers still running 4349 * and holding an empty delayed inode item. We must wait for 4350 * them to complete first because they can create a transaction. 4351 * This happens when someone calls btrfs_balance_delayed_items() 4352 * and then a transaction commit runs the same delayed nodes 4353 * before any delayed worker has done something with the nodes. 4354 * We must wait for any worker here and not at transaction 4355 * commit time since that could cause a deadlock. 4356 * This is a very rare case. 4357 */ 4358 btrfs_flush_workqueue(fs_info->delayed_workers); 4359 4360 ret = btrfs_commit_super(fs_info); 4361 if (ret) 4362 btrfs_err(fs_info, "commit super ret %d", ret); 4363 } 4364 4365 if (BTRFS_FS_ERROR(fs_info)) 4366 btrfs_error_commit_super(fs_info); 4367 4368 kthread_stop(fs_info->transaction_kthread); 4369 kthread_stop(fs_info->cleaner_kthread); 4370 4371 ASSERT(list_empty(&fs_info->delayed_iputs)); 4372 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4373 4374 if (btrfs_check_quota_leak(fs_info)) { 4375 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4376 btrfs_err(fs_info, "qgroup reserved space leaked"); 4377 } 4378 4379 btrfs_free_qgroup_config(fs_info); 4380 ASSERT(list_empty(&fs_info->delalloc_roots)); 4381 4382 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 4383 btrfs_info(fs_info, "at unmount delalloc count %lld", 4384 percpu_counter_sum(&fs_info->delalloc_bytes)); 4385 } 4386 4387 if (percpu_counter_sum(&fs_info->ordered_bytes)) 4388 btrfs_info(fs_info, "at unmount dio bytes count %lld", 4389 percpu_counter_sum(&fs_info->ordered_bytes)); 4390 4391 btrfs_sysfs_remove_mounted(fs_info); 4392 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 4393 4394 btrfs_put_block_group_cache(fs_info); 4395 4396 /* 4397 * we must make sure there is not any read request to 4398 * submit after we stopping all workers. 4399 */ 4400 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 4401 btrfs_stop_all_workers(fs_info); 4402 4403 /* We shouldn't have any transaction open at this point */ 4404 warn_about_uncommitted_trans(fs_info); 4405 4406 clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 4407 free_root_pointers(fs_info, true); 4408 btrfs_free_fs_roots(fs_info); 4409 4410 /* 4411 * We must free the block groups after dropping the fs_roots as we could 4412 * have had an IO error and have left over tree log blocks that aren't 4413 * cleaned up until the fs roots are freed. This makes the block group 4414 * accounting appear to be wrong because there's pending reserved bytes, 4415 * so make sure we do the block group cleanup afterwards. 4416 */ 4417 btrfs_free_block_groups(fs_info); 4418 4419 iput(fs_info->btree_inode); 4420 4421 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4422 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 4423 btrfsic_unmount(fs_info->fs_devices); 4424 #endif 4425 4426 btrfs_mapping_tree_free(&fs_info->mapping_tree); 4427 btrfs_close_devices(fs_info->fs_devices); 4428 } 4429 4430 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 4431 { 4432 struct btrfs_fs_info *fs_info = buf->fs_info; 4433 u64 transid = btrfs_header_generation(buf); 4434 4435 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 4436 /* 4437 * This is a fast path so only do this check if we have sanity tests 4438 * enabled. Normal people shouldn't be using unmapped buffers as dirty 4439 * outside of the sanity tests. 4440 */ 4441 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 4442 return; 4443 #endif 4444 btrfs_assert_tree_write_locked(buf); 4445 if (transid != fs_info->generation) 4446 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 4447 buf->start, transid, fs_info->generation); 4448 set_extent_buffer_dirty(buf); 4449 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4450 /* 4451 * btrfs_check_leaf() won't check item data if we don't have WRITTEN 4452 * set, so this will only validate the basic structure of the items. 4453 */ 4454 if (btrfs_header_level(buf) == 0 && btrfs_check_leaf(buf)) { 4455 btrfs_print_leaf(buf); 4456 ASSERT(0); 4457 } 4458 #endif 4459 } 4460 4461 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4462 int flush_delayed) 4463 { 4464 /* 4465 * looks as though older kernels can get into trouble with 4466 * this code, they end up stuck in balance_dirty_pages forever 4467 */ 4468 int ret; 4469 4470 if (current->flags & PF_MEMALLOC) 4471 return; 4472 4473 if (flush_delayed) 4474 btrfs_balance_delayed_items(fs_info); 4475 4476 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4477 BTRFS_DIRTY_METADATA_THRESH, 4478 fs_info->dirty_metadata_batch); 4479 if (ret > 0) { 4480 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 4481 } 4482 } 4483 4484 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 4485 { 4486 __btrfs_btree_balance_dirty(fs_info, 1); 4487 } 4488 4489 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4490 { 4491 __btrfs_btree_balance_dirty(fs_info, 0); 4492 } 4493 4494 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4495 { 4496 /* cleanup FS via transaction */ 4497 btrfs_cleanup_transaction(fs_info); 4498 4499 mutex_lock(&fs_info->cleaner_mutex); 4500 btrfs_run_delayed_iputs(fs_info); 4501 mutex_unlock(&fs_info->cleaner_mutex); 4502 4503 down_write(&fs_info->cleanup_work_sem); 4504 up_write(&fs_info->cleanup_work_sem); 4505 } 4506 4507 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4508 { 4509 struct btrfs_root *gang[8]; 4510 u64 root_objectid = 0; 4511 int ret; 4512 4513 spin_lock(&fs_info->fs_roots_radix_lock); 4514 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4515 (void **)gang, root_objectid, 4516 ARRAY_SIZE(gang))) != 0) { 4517 int i; 4518 4519 for (i = 0; i < ret; i++) 4520 gang[i] = btrfs_grab_root(gang[i]); 4521 spin_unlock(&fs_info->fs_roots_radix_lock); 4522 4523 for (i = 0; i < ret; i++) { 4524 if (!gang[i]) 4525 continue; 4526 root_objectid = gang[i]->root_key.objectid; 4527 btrfs_free_log(NULL, gang[i]); 4528 btrfs_put_root(gang[i]); 4529 } 4530 root_objectid++; 4531 spin_lock(&fs_info->fs_roots_radix_lock); 4532 } 4533 spin_unlock(&fs_info->fs_roots_radix_lock); 4534 btrfs_free_log_root_tree(NULL, fs_info); 4535 } 4536 4537 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4538 { 4539 struct btrfs_ordered_extent *ordered; 4540 4541 spin_lock(&root->ordered_extent_lock); 4542 /* 4543 * This will just short circuit the ordered completion stuff which will 4544 * make sure the ordered extent gets properly cleaned up. 4545 */ 4546 list_for_each_entry(ordered, &root->ordered_extents, 4547 root_extent_list) 4548 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4549 spin_unlock(&root->ordered_extent_lock); 4550 } 4551 4552 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4553 { 4554 struct btrfs_root *root; 4555 struct list_head splice; 4556 4557 INIT_LIST_HEAD(&splice); 4558 4559 spin_lock(&fs_info->ordered_root_lock); 4560 list_splice_init(&fs_info->ordered_roots, &splice); 4561 while (!list_empty(&splice)) { 4562 root = list_first_entry(&splice, struct btrfs_root, 4563 ordered_root); 4564 list_move_tail(&root->ordered_root, 4565 &fs_info->ordered_roots); 4566 4567 spin_unlock(&fs_info->ordered_root_lock); 4568 btrfs_destroy_ordered_extents(root); 4569 4570 cond_resched(); 4571 spin_lock(&fs_info->ordered_root_lock); 4572 } 4573 spin_unlock(&fs_info->ordered_root_lock); 4574 4575 /* 4576 * We need this here because if we've been flipped read-only we won't 4577 * get sync() from the umount, so we need to make sure any ordered 4578 * extents that haven't had their dirty pages IO start writeout yet 4579 * actually get run and error out properly. 4580 */ 4581 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4582 } 4583 4584 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4585 struct btrfs_fs_info *fs_info) 4586 { 4587 struct rb_node *node; 4588 struct btrfs_delayed_ref_root *delayed_refs; 4589 struct btrfs_delayed_ref_node *ref; 4590 4591 delayed_refs = &trans->delayed_refs; 4592 4593 spin_lock(&delayed_refs->lock); 4594 if (atomic_read(&delayed_refs->num_entries) == 0) { 4595 spin_unlock(&delayed_refs->lock); 4596 btrfs_debug(fs_info, "delayed_refs has NO entry"); 4597 return; 4598 } 4599 4600 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4601 struct btrfs_delayed_ref_head *head; 4602 struct rb_node *n; 4603 bool pin_bytes = false; 4604 4605 head = rb_entry(node, struct btrfs_delayed_ref_head, 4606 href_node); 4607 if (btrfs_delayed_ref_lock(delayed_refs, head)) 4608 continue; 4609 4610 spin_lock(&head->lock); 4611 while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 4612 ref = rb_entry(n, struct btrfs_delayed_ref_node, 4613 ref_node); 4614 rb_erase_cached(&ref->ref_node, &head->ref_tree); 4615 RB_CLEAR_NODE(&ref->ref_node); 4616 if (!list_empty(&ref->add_list)) 4617 list_del(&ref->add_list); 4618 atomic_dec(&delayed_refs->num_entries); 4619 btrfs_put_delayed_ref(ref); 4620 } 4621 if (head->must_insert_reserved) 4622 pin_bytes = true; 4623 btrfs_free_delayed_extent_op(head->extent_op); 4624 btrfs_delete_ref_head(delayed_refs, head); 4625 spin_unlock(&head->lock); 4626 spin_unlock(&delayed_refs->lock); 4627 mutex_unlock(&head->mutex); 4628 4629 if (pin_bytes) { 4630 struct btrfs_block_group *cache; 4631 4632 cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4633 BUG_ON(!cache); 4634 4635 spin_lock(&cache->space_info->lock); 4636 spin_lock(&cache->lock); 4637 cache->pinned += head->num_bytes; 4638 btrfs_space_info_update_bytes_pinned(fs_info, 4639 cache->space_info, head->num_bytes); 4640 cache->reserved -= head->num_bytes; 4641 cache->space_info->bytes_reserved -= head->num_bytes; 4642 spin_unlock(&cache->lock); 4643 spin_unlock(&cache->space_info->lock); 4644 4645 btrfs_put_block_group(cache); 4646 4647 btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4648 head->bytenr + head->num_bytes - 1); 4649 } 4650 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4651 btrfs_put_delayed_ref_head(head); 4652 cond_resched(); 4653 spin_lock(&delayed_refs->lock); 4654 } 4655 btrfs_qgroup_destroy_extent_records(trans); 4656 4657 spin_unlock(&delayed_refs->lock); 4658 } 4659 4660 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4661 { 4662 struct btrfs_inode *btrfs_inode; 4663 struct list_head splice; 4664 4665 INIT_LIST_HEAD(&splice); 4666 4667 spin_lock(&root->delalloc_lock); 4668 list_splice_init(&root->delalloc_inodes, &splice); 4669 4670 while (!list_empty(&splice)) { 4671 struct inode *inode = NULL; 4672 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4673 delalloc_inodes); 4674 __btrfs_del_delalloc_inode(root, btrfs_inode); 4675 spin_unlock(&root->delalloc_lock); 4676 4677 /* 4678 * Make sure we get a live inode and that it'll not disappear 4679 * meanwhile. 4680 */ 4681 inode = igrab(&btrfs_inode->vfs_inode); 4682 if (inode) { 4683 unsigned int nofs_flag; 4684 4685 nofs_flag = memalloc_nofs_save(); 4686 invalidate_inode_pages2(inode->i_mapping); 4687 memalloc_nofs_restore(nofs_flag); 4688 iput(inode); 4689 } 4690 spin_lock(&root->delalloc_lock); 4691 } 4692 spin_unlock(&root->delalloc_lock); 4693 } 4694 4695 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4696 { 4697 struct btrfs_root *root; 4698 struct list_head splice; 4699 4700 INIT_LIST_HEAD(&splice); 4701 4702 spin_lock(&fs_info->delalloc_root_lock); 4703 list_splice_init(&fs_info->delalloc_roots, &splice); 4704 while (!list_empty(&splice)) { 4705 root = list_first_entry(&splice, struct btrfs_root, 4706 delalloc_root); 4707 root = btrfs_grab_root(root); 4708 BUG_ON(!root); 4709 spin_unlock(&fs_info->delalloc_root_lock); 4710 4711 btrfs_destroy_delalloc_inodes(root); 4712 btrfs_put_root(root); 4713 4714 spin_lock(&fs_info->delalloc_root_lock); 4715 } 4716 spin_unlock(&fs_info->delalloc_root_lock); 4717 } 4718 4719 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4720 struct extent_io_tree *dirty_pages, 4721 int mark) 4722 { 4723 int ret; 4724 struct extent_buffer *eb; 4725 u64 start = 0; 4726 u64 end; 4727 4728 while (1) { 4729 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4730 mark, NULL); 4731 if (ret) 4732 break; 4733 4734 clear_extent_bits(dirty_pages, start, end, mark); 4735 while (start <= end) { 4736 eb = find_extent_buffer(fs_info, start); 4737 start += fs_info->nodesize; 4738 if (!eb) 4739 continue; 4740 4741 btrfs_tree_lock(eb); 4742 wait_on_extent_buffer_writeback(eb); 4743 btrfs_clear_buffer_dirty(NULL, eb); 4744 btrfs_tree_unlock(eb); 4745 4746 free_extent_buffer_stale(eb); 4747 } 4748 } 4749 4750 return ret; 4751 } 4752 4753 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4754 struct extent_io_tree *unpin) 4755 { 4756 u64 start; 4757 u64 end; 4758 int ret; 4759 4760 while (1) { 4761 struct extent_state *cached_state = NULL; 4762 4763 /* 4764 * The btrfs_finish_extent_commit() may get the same range as 4765 * ours between find_first_extent_bit and clear_extent_dirty. 4766 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4767 * the same extent range. 4768 */ 4769 mutex_lock(&fs_info->unused_bg_unpin_mutex); 4770 ret = find_first_extent_bit(unpin, 0, &start, &end, 4771 EXTENT_DIRTY, &cached_state); 4772 if (ret) { 4773 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4774 break; 4775 } 4776 4777 clear_extent_dirty(unpin, start, end, &cached_state); 4778 free_extent_state(cached_state); 4779 btrfs_error_unpin_extent_range(fs_info, start, end); 4780 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4781 cond_resched(); 4782 } 4783 4784 return 0; 4785 } 4786 4787 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4788 { 4789 struct inode *inode; 4790 4791 inode = cache->io_ctl.inode; 4792 if (inode) { 4793 unsigned int nofs_flag; 4794 4795 nofs_flag = memalloc_nofs_save(); 4796 invalidate_inode_pages2(inode->i_mapping); 4797 memalloc_nofs_restore(nofs_flag); 4798 4799 BTRFS_I(inode)->generation = 0; 4800 cache->io_ctl.inode = NULL; 4801 iput(inode); 4802 } 4803 ASSERT(cache->io_ctl.pages == NULL); 4804 btrfs_put_block_group(cache); 4805 } 4806 4807 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 4808 struct btrfs_fs_info *fs_info) 4809 { 4810 struct btrfs_block_group *cache; 4811 4812 spin_lock(&cur_trans->dirty_bgs_lock); 4813 while (!list_empty(&cur_trans->dirty_bgs)) { 4814 cache = list_first_entry(&cur_trans->dirty_bgs, 4815 struct btrfs_block_group, 4816 dirty_list); 4817 4818 if (!list_empty(&cache->io_list)) { 4819 spin_unlock(&cur_trans->dirty_bgs_lock); 4820 list_del_init(&cache->io_list); 4821 btrfs_cleanup_bg_io(cache); 4822 spin_lock(&cur_trans->dirty_bgs_lock); 4823 } 4824 4825 list_del_init(&cache->dirty_list); 4826 spin_lock(&cache->lock); 4827 cache->disk_cache_state = BTRFS_DC_ERROR; 4828 spin_unlock(&cache->lock); 4829 4830 spin_unlock(&cur_trans->dirty_bgs_lock); 4831 btrfs_put_block_group(cache); 4832 btrfs_delayed_refs_rsv_release(fs_info, 1); 4833 spin_lock(&cur_trans->dirty_bgs_lock); 4834 } 4835 spin_unlock(&cur_trans->dirty_bgs_lock); 4836 4837 /* 4838 * Refer to the definition of io_bgs member for details why it's safe 4839 * to use it without any locking 4840 */ 4841 while (!list_empty(&cur_trans->io_bgs)) { 4842 cache = list_first_entry(&cur_trans->io_bgs, 4843 struct btrfs_block_group, 4844 io_list); 4845 4846 list_del_init(&cache->io_list); 4847 spin_lock(&cache->lock); 4848 cache->disk_cache_state = BTRFS_DC_ERROR; 4849 spin_unlock(&cache->lock); 4850 btrfs_cleanup_bg_io(cache); 4851 } 4852 } 4853 4854 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 4855 struct btrfs_fs_info *fs_info) 4856 { 4857 struct btrfs_device *dev, *tmp; 4858 4859 btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4860 ASSERT(list_empty(&cur_trans->dirty_bgs)); 4861 ASSERT(list_empty(&cur_trans->io_bgs)); 4862 4863 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4864 post_commit_list) { 4865 list_del_init(&dev->post_commit_list); 4866 } 4867 4868 btrfs_destroy_delayed_refs(cur_trans, fs_info); 4869 4870 cur_trans->state = TRANS_STATE_COMMIT_START; 4871 wake_up(&fs_info->transaction_blocked_wait); 4872 4873 cur_trans->state = TRANS_STATE_UNBLOCKED; 4874 wake_up(&fs_info->transaction_wait); 4875 4876 btrfs_destroy_delayed_inodes(fs_info); 4877 4878 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 4879 EXTENT_DIRTY); 4880 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 4881 4882 cur_trans->state =TRANS_STATE_COMPLETED; 4883 wake_up(&cur_trans->commit_wait); 4884 } 4885 4886 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4887 { 4888 struct btrfs_transaction *t; 4889 4890 mutex_lock(&fs_info->transaction_kthread_mutex); 4891 4892 spin_lock(&fs_info->trans_lock); 4893 while (!list_empty(&fs_info->trans_list)) { 4894 t = list_first_entry(&fs_info->trans_list, 4895 struct btrfs_transaction, list); 4896 if (t->state >= TRANS_STATE_COMMIT_START) { 4897 refcount_inc(&t->use_count); 4898 spin_unlock(&fs_info->trans_lock); 4899 btrfs_wait_for_commit(fs_info, t->transid); 4900 btrfs_put_transaction(t); 4901 spin_lock(&fs_info->trans_lock); 4902 continue; 4903 } 4904 if (t == fs_info->running_transaction) { 4905 t->state = TRANS_STATE_COMMIT_DOING; 4906 spin_unlock(&fs_info->trans_lock); 4907 /* 4908 * We wait for 0 num_writers since we don't hold a trans 4909 * handle open currently for this transaction. 4910 */ 4911 wait_event(t->writer_wait, 4912 atomic_read(&t->num_writers) == 0); 4913 } else { 4914 spin_unlock(&fs_info->trans_lock); 4915 } 4916 btrfs_cleanup_one_transaction(t, fs_info); 4917 4918 spin_lock(&fs_info->trans_lock); 4919 if (t == fs_info->running_transaction) 4920 fs_info->running_transaction = NULL; 4921 list_del_init(&t->list); 4922 spin_unlock(&fs_info->trans_lock); 4923 4924 btrfs_put_transaction(t); 4925 trace_btrfs_transaction_commit(fs_info); 4926 spin_lock(&fs_info->trans_lock); 4927 } 4928 spin_unlock(&fs_info->trans_lock); 4929 btrfs_destroy_all_ordered_extents(fs_info); 4930 btrfs_destroy_delayed_inodes(fs_info); 4931 btrfs_assert_delayed_root_empty(fs_info); 4932 btrfs_destroy_all_delalloc_inodes(fs_info); 4933 btrfs_drop_all_logs(fs_info); 4934 mutex_unlock(&fs_info->transaction_kthread_mutex); 4935 4936 return 0; 4937 } 4938 4939 int btrfs_init_root_free_objectid(struct btrfs_root *root) 4940 { 4941 struct btrfs_path *path; 4942 int ret; 4943 struct extent_buffer *l; 4944 struct btrfs_key search_key; 4945 struct btrfs_key found_key; 4946 int slot; 4947 4948 path = btrfs_alloc_path(); 4949 if (!path) 4950 return -ENOMEM; 4951 4952 search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 4953 search_key.type = -1; 4954 search_key.offset = (u64)-1; 4955 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 4956 if (ret < 0) 4957 goto error; 4958 BUG_ON(ret == 0); /* Corruption */ 4959 if (path->slots[0] > 0) { 4960 slot = path->slots[0] - 1; 4961 l = path->nodes[0]; 4962 btrfs_item_key_to_cpu(l, &found_key, slot); 4963 root->free_objectid = max_t(u64, found_key.objectid + 1, 4964 BTRFS_FIRST_FREE_OBJECTID); 4965 } else { 4966 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 4967 } 4968 ret = 0; 4969 error: 4970 btrfs_free_path(path); 4971 return ret; 4972 } 4973 4974 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 4975 { 4976 int ret; 4977 mutex_lock(&root->objectid_mutex); 4978 4979 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 4980 btrfs_warn(root->fs_info, 4981 "the objectid of root %llu reaches its highest value", 4982 root->root_key.objectid); 4983 ret = -ENOSPC; 4984 goto out; 4985 } 4986 4987 *objectid = root->free_objectid++; 4988 ret = 0; 4989 out: 4990 mutex_unlock(&root->objectid_mutex); 4991 return ret; 4992 } 4993