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