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 "volumes.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 47 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 48 BTRFS_HEADER_FLAG_RELOC |\ 49 BTRFS_SUPER_FLAG_ERROR |\ 50 BTRFS_SUPER_FLAG_SEEDING |\ 51 BTRFS_SUPER_FLAG_METADUMP |\ 52 BTRFS_SUPER_FLAG_METADUMP_V2) 53 54 static void end_workqueue_fn(struct btrfs_work *work); 55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 57 struct btrfs_fs_info *fs_info); 58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 59 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 60 struct extent_io_tree *dirty_pages, 61 int mark); 62 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 63 struct extent_io_tree *pinned_extents); 64 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 65 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 66 67 /* 68 * btrfs_end_io_wq structs are used to do processing in task context when an IO 69 * is complete. This is used during reads to verify checksums, and it is used 70 * by writes to insert metadata for new file extents after IO is complete. 71 */ 72 struct btrfs_end_io_wq { 73 struct bio *bio; 74 bio_end_io_t *end_io; 75 void *private; 76 struct btrfs_fs_info *info; 77 blk_status_t status; 78 enum btrfs_wq_endio_type metadata; 79 struct btrfs_work work; 80 }; 81 82 static struct kmem_cache *btrfs_end_io_wq_cache; 83 84 int __init btrfs_end_io_wq_init(void) 85 { 86 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 87 sizeof(struct btrfs_end_io_wq), 88 0, 89 SLAB_MEM_SPREAD, 90 NULL); 91 if (!btrfs_end_io_wq_cache) 92 return -ENOMEM; 93 return 0; 94 } 95 96 void __cold btrfs_end_io_wq_exit(void) 97 { 98 kmem_cache_destroy(btrfs_end_io_wq_cache); 99 } 100 101 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) 102 { 103 if (fs_info->csum_shash) 104 crypto_free_shash(fs_info->csum_shash); 105 } 106 107 /* 108 * async submit bios are used to offload expensive checksumming 109 * onto the worker threads. They checksum file and metadata bios 110 * just before they are sent down the IO stack. 111 */ 112 struct async_submit_bio { 113 struct inode *inode; 114 struct bio *bio; 115 extent_submit_bio_start_t *submit_bio_start; 116 int mirror_num; 117 118 /* Optional parameter for submit_bio_start used by direct io */ 119 u64 dio_file_offset; 120 struct btrfs_work work; 121 blk_status_t status; 122 }; 123 124 /* 125 * Lockdep class keys for extent_buffer->lock's in this root. For a given 126 * eb, the lockdep key is determined by the btrfs_root it belongs to and 127 * the level the eb occupies in the tree. 128 * 129 * Different roots are used for different purposes and may nest inside each 130 * other and they require separate keysets. As lockdep keys should be 131 * static, assign keysets according to the purpose of the root as indicated 132 * by btrfs_root->root_key.objectid. This ensures that all special purpose 133 * roots have separate keysets. 134 * 135 * Lock-nesting across peer nodes is always done with the immediate parent 136 * node locked thus preventing deadlock. As lockdep doesn't know this, use 137 * subclass to avoid triggering lockdep warning in such cases. 138 * 139 * The key is set by the readpage_end_io_hook after the buffer has passed 140 * csum validation but before the pages are unlocked. It is also set by 141 * btrfs_init_new_buffer on freshly allocated blocks. 142 * 143 * We also add a check to make sure the highest level of the tree is the 144 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 145 * needs update as well. 146 */ 147 #ifdef CONFIG_DEBUG_LOCK_ALLOC 148 # if BTRFS_MAX_LEVEL != 8 149 # error 150 # endif 151 152 #define DEFINE_LEVEL(stem, level) \ 153 .names[level] = "btrfs-" stem "-0" #level, 154 155 #define DEFINE_NAME(stem) \ 156 DEFINE_LEVEL(stem, 0) \ 157 DEFINE_LEVEL(stem, 1) \ 158 DEFINE_LEVEL(stem, 2) \ 159 DEFINE_LEVEL(stem, 3) \ 160 DEFINE_LEVEL(stem, 4) \ 161 DEFINE_LEVEL(stem, 5) \ 162 DEFINE_LEVEL(stem, 6) \ 163 DEFINE_LEVEL(stem, 7) 164 165 static struct btrfs_lockdep_keyset { 166 u64 id; /* root objectid */ 167 /* Longest entry: btrfs-free-space-00 */ 168 char names[BTRFS_MAX_LEVEL][20]; 169 struct lock_class_key keys[BTRFS_MAX_LEVEL]; 170 } btrfs_lockdep_keysets[] = { 171 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, 172 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, 173 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, 174 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, 175 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, 176 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, 177 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, 178 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, 179 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, 180 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, 181 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, 182 { .id = 0, DEFINE_NAME("tree") }, 183 }; 184 185 #undef DEFINE_LEVEL 186 #undef DEFINE_NAME 187 188 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 189 int level) 190 { 191 struct btrfs_lockdep_keyset *ks; 192 193 BUG_ON(level >= ARRAY_SIZE(ks->keys)); 194 195 /* find the matching keyset, id 0 is the default entry */ 196 for (ks = btrfs_lockdep_keysets; ks->id; ks++) 197 if (ks->id == objectid) 198 break; 199 200 lockdep_set_class_and_name(&eb->lock, 201 &ks->keys[level], ks->names[level]); 202 } 203 204 #endif 205 206 /* 207 * Compute the csum of a btree block and store the result to provided buffer. 208 */ 209 static void csum_tree_block(struct extent_buffer *buf, u8 *result) 210 { 211 struct btrfs_fs_info *fs_info = buf->fs_info; 212 const int num_pages = fs_info->nodesize >> PAGE_SHIFT; 213 const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); 214 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 215 char *kaddr; 216 int i; 217 218 shash->tfm = fs_info->csum_shash; 219 crypto_shash_init(shash); 220 kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); 221 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, 222 first_page_part - BTRFS_CSUM_SIZE); 223 224 for (i = 1; i < num_pages; i++) { 225 kaddr = page_address(buf->pages[i]); 226 crypto_shash_update(shash, kaddr, PAGE_SIZE); 227 } 228 memset(result, 0, BTRFS_CSUM_SIZE); 229 crypto_shash_final(shash, result); 230 } 231 232 /* 233 * we can't consider a given block up to date unless the transid of the 234 * block matches the transid in the parent node's pointer. This is how we 235 * detect blocks that either didn't get written at all or got written 236 * in the wrong place. 237 */ 238 static int verify_parent_transid(struct extent_io_tree *io_tree, 239 struct extent_buffer *eb, u64 parent_transid, 240 int atomic) 241 { 242 struct extent_state *cached_state = NULL; 243 int ret; 244 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 245 246 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 247 return 0; 248 249 if (atomic) 250 return -EAGAIN; 251 252 if (need_lock) 253 btrfs_tree_read_lock(eb); 254 255 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 256 &cached_state); 257 if (extent_buffer_uptodate(eb) && 258 btrfs_header_generation(eb) == parent_transid) { 259 ret = 0; 260 goto out; 261 } 262 btrfs_err_rl(eb->fs_info, 263 "parent transid verify failed on %llu wanted %llu found %llu", 264 eb->start, 265 parent_transid, btrfs_header_generation(eb)); 266 ret = 1; 267 268 /* 269 * Things reading via commit roots that don't have normal protection, 270 * like send, can have a really old block in cache that may point at a 271 * block that has been freed and re-allocated. So don't clear uptodate 272 * if we find an eb that is under IO (dirty/writeback) because we could 273 * end up reading in the stale data and then writing it back out and 274 * making everybody very sad. 275 */ 276 if (!extent_buffer_under_io(eb)) 277 clear_extent_buffer_uptodate(eb); 278 out: 279 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 280 &cached_state); 281 if (need_lock) 282 btrfs_tree_read_unlock(eb); 283 return ret; 284 } 285 286 static bool btrfs_supported_super_csum(u16 csum_type) 287 { 288 switch (csum_type) { 289 case BTRFS_CSUM_TYPE_CRC32: 290 case BTRFS_CSUM_TYPE_XXHASH: 291 case BTRFS_CSUM_TYPE_SHA256: 292 case BTRFS_CSUM_TYPE_BLAKE2: 293 return true; 294 default: 295 return false; 296 } 297 } 298 299 /* 300 * Return 0 if the superblock checksum type matches the checksum value of that 301 * algorithm. Pass the raw disk superblock data. 302 */ 303 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 304 char *raw_disk_sb) 305 { 306 struct btrfs_super_block *disk_sb = 307 (struct btrfs_super_block *)raw_disk_sb; 308 char result[BTRFS_CSUM_SIZE]; 309 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 310 311 shash->tfm = fs_info->csum_shash; 312 313 /* 314 * The super_block structure does not span the whole 315 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 316 * filled with zeros and is included in the checksum. 317 */ 318 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 319 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 320 321 if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 322 return 1; 323 324 return 0; 325 } 326 327 int btrfs_verify_level_key(struct extent_buffer *eb, int level, 328 struct btrfs_key *first_key, u64 parent_transid) 329 { 330 struct btrfs_fs_info *fs_info = eb->fs_info; 331 int found_level; 332 struct btrfs_key found_key; 333 int ret; 334 335 found_level = btrfs_header_level(eb); 336 if (found_level != level) { 337 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 338 KERN_ERR "BTRFS: tree level check failed\n"); 339 btrfs_err(fs_info, 340 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 341 eb->start, level, found_level); 342 return -EIO; 343 } 344 345 if (!first_key) 346 return 0; 347 348 /* 349 * For live tree block (new tree blocks in current transaction), 350 * we need proper lock context to avoid race, which is impossible here. 351 * So we only checks tree blocks which is read from disk, whose 352 * generation <= fs_info->last_trans_committed. 353 */ 354 if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 355 return 0; 356 357 /* We have @first_key, so this @eb must have at least one item */ 358 if (btrfs_header_nritems(eb) == 0) { 359 btrfs_err(fs_info, 360 "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 361 eb->start); 362 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 363 return -EUCLEAN; 364 } 365 366 if (found_level) 367 btrfs_node_key_to_cpu(eb, &found_key, 0); 368 else 369 btrfs_item_key_to_cpu(eb, &found_key, 0); 370 ret = btrfs_comp_cpu_keys(first_key, &found_key); 371 372 if (ret) { 373 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 374 KERN_ERR "BTRFS: tree first key check failed\n"); 375 btrfs_err(fs_info, 376 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 377 eb->start, parent_transid, first_key->objectid, 378 first_key->type, first_key->offset, 379 found_key.objectid, found_key.type, 380 found_key.offset); 381 } 382 return ret; 383 } 384 385 /* 386 * helper to read a given tree block, doing retries as required when 387 * the checksums don't match and we have alternate mirrors to try. 388 * 389 * @parent_transid: expected transid, skip check if 0 390 * @level: expected level, mandatory check 391 * @first_key: expected key of first slot, skip check if NULL 392 */ 393 static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 394 u64 parent_transid, int level, 395 struct btrfs_key *first_key) 396 { 397 struct btrfs_fs_info *fs_info = eb->fs_info; 398 struct extent_io_tree *io_tree; 399 int failed = 0; 400 int ret; 401 int num_copies = 0; 402 int mirror_num = 0; 403 int failed_mirror = 0; 404 405 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 406 while (1) { 407 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 408 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 409 if (!ret) { 410 if (verify_parent_transid(io_tree, eb, 411 parent_transid, 0)) 412 ret = -EIO; 413 else if (btrfs_verify_level_key(eb, level, 414 first_key, parent_transid)) 415 ret = -EUCLEAN; 416 else 417 break; 418 } 419 420 num_copies = btrfs_num_copies(fs_info, 421 eb->start, eb->len); 422 if (num_copies == 1) 423 break; 424 425 if (!failed_mirror) { 426 failed = 1; 427 failed_mirror = eb->read_mirror; 428 } 429 430 mirror_num++; 431 if (mirror_num == failed_mirror) 432 mirror_num++; 433 434 if (mirror_num > num_copies) 435 break; 436 } 437 438 if (failed && !ret && failed_mirror) 439 btrfs_repair_eb_io_failure(eb, failed_mirror); 440 441 return ret; 442 } 443 444 static int csum_one_extent_buffer(struct extent_buffer *eb) 445 { 446 struct btrfs_fs_info *fs_info = eb->fs_info; 447 u8 result[BTRFS_CSUM_SIZE]; 448 int ret; 449 450 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 451 offsetof(struct btrfs_header, fsid), 452 BTRFS_FSID_SIZE) == 0); 453 csum_tree_block(eb, result); 454 455 if (btrfs_header_level(eb)) 456 ret = btrfs_check_node(eb); 457 else 458 ret = btrfs_check_leaf_full(eb); 459 460 if (ret < 0) { 461 btrfs_print_tree(eb, 0); 462 btrfs_err(fs_info, 463 "block=%llu write time tree block corruption detected", 464 eb->start); 465 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 466 return ret; 467 } 468 write_extent_buffer(eb, result, 0, fs_info->csum_size); 469 470 return 0; 471 } 472 473 /* Checksum all dirty extent buffers in one bio_vec */ 474 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 475 struct bio_vec *bvec) 476 { 477 struct page *page = bvec->bv_page; 478 u64 bvec_start = page_offset(page) + bvec->bv_offset; 479 u64 cur; 480 int ret = 0; 481 482 for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 483 cur += fs_info->nodesize) { 484 struct extent_buffer *eb; 485 bool uptodate; 486 487 eb = find_extent_buffer(fs_info, cur); 488 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 489 fs_info->nodesize); 490 491 /* A dirty eb shouldn't disappear from buffer_radix */ 492 if (WARN_ON(!eb)) 493 return -EUCLEAN; 494 495 if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 496 free_extent_buffer(eb); 497 return -EUCLEAN; 498 } 499 if (WARN_ON(!uptodate)) { 500 free_extent_buffer(eb); 501 return -EUCLEAN; 502 } 503 504 ret = csum_one_extent_buffer(eb); 505 free_extent_buffer(eb); 506 if (ret < 0) 507 return ret; 508 } 509 return ret; 510 } 511 512 /* 513 * Checksum a dirty tree block before IO. This has extra checks to make sure 514 * we only fill in the checksum field in the first page of a multi-page block. 515 * For subpage extent buffers we need bvec to also read the offset in the page. 516 */ 517 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 518 { 519 struct page *page = bvec->bv_page; 520 u64 start = page_offset(page); 521 u64 found_start; 522 struct extent_buffer *eb; 523 524 if (fs_info->sectorsize < PAGE_SIZE) 525 return csum_dirty_subpage_buffers(fs_info, bvec); 526 527 eb = (struct extent_buffer *)page->private; 528 if (page != eb->pages[0]) 529 return 0; 530 531 found_start = btrfs_header_bytenr(eb); 532 533 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 534 WARN_ON(found_start != 0); 535 return 0; 536 } 537 538 /* 539 * Please do not consolidate these warnings into a single if. 540 * It is useful to know what went wrong. 541 */ 542 if (WARN_ON(found_start != start)) 543 return -EUCLEAN; 544 if (WARN_ON(!PageUptodate(page))) 545 return -EUCLEAN; 546 547 return csum_one_extent_buffer(eb); 548 } 549 550 static int check_tree_block_fsid(struct extent_buffer *eb) 551 { 552 struct btrfs_fs_info *fs_info = eb->fs_info; 553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 554 u8 fsid[BTRFS_FSID_SIZE]; 555 u8 *metadata_uuid; 556 557 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 558 BTRFS_FSID_SIZE); 559 /* 560 * Checking the incompat flag is only valid for the current fs. For 561 * seed devices it's forbidden to have their uuid changed so reading 562 * ->fsid in this case is fine 563 */ 564 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 565 metadata_uuid = fs_devices->metadata_uuid; 566 else 567 metadata_uuid = fs_devices->fsid; 568 569 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 570 return 0; 571 572 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 573 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 574 return 0; 575 576 return 1; 577 } 578 579 /* Do basic extent buffer checks at read time */ 580 static int validate_extent_buffer(struct extent_buffer *eb) 581 { 582 struct btrfs_fs_info *fs_info = eb->fs_info; 583 u64 found_start; 584 const u32 csum_size = fs_info->csum_size; 585 u8 found_level; 586 u8 result[BTRFS_CSUM_SIZE]; 587 int ret = 0; 588 589 found_start = btrfs_header_bytenr(eb); 590 if (found_start != eb->start) { 591 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 592 eb->start, found_start); 593 ret = -EIO; 594 goto out; 595 } 596 if (check_tree_block_fsid(eb)) { 597 btrfs_err_rl(fs_info, "bad fsid on block %llu", 598 eb->start); 599 ret = -EIO; 600 goto out; 601 } 602 found_level = btrfs_header_level(eb); 603 if (found_level >= BTRFS_MAX_LEVEL) { 604 btrfs_err(fs_info, "bad tree block level %d on %llu", 605 (int)btrfs_header_level(eb), eb->start); 606 ret = -EIO; 607 goto out; 608 } 609 610 csum_tree_block(eb, result); 611 612 if (memcmp_extent_buffer(eb, result, 0, csum_size)) { 613 u8 val[BTRFS_CSUM_SIZE] = { 0 }; 614 615 read_extent_buffer(eb, &val, 0, csum_size); 616 btrfs_warn_rl(fs_info, 617 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 618 fs_info->sb->s_id, eb->start, 619 CSUM_FMT_VALUE(csum_size, val), 620 CSUM_FMT_VALUE(csum_size, result), 621 btrfs_header_level(eb)); 622 ret = -EUCLEAN; 623 goto out; 624 } 625 626 /* 627 * If this is a leaf block and it is corrupt, set the corrupt bit so 628 * that we don't try and read the other copies of this block, just 629 * return -EIO. 630 */ 631 if (found_level == 0 && btrfs_check_leaf_full(eb)) { 632 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 633 ret = -EIO; 634 } 635 636 if (found_level > 0 && btrfs_check_node(eb)) 637 ret = -EIO; 638 639 if (!ret) 640 set_extent_buffer_uptodate(eb); 641 else 642 btrfs_err(fs_info, 643 "block=%llu read time tree block corruption detected", 644 eb->start); 645 out: 646 return ret; 647 } 648 649 static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 650 int mirror) 651 { 652 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 653 struct extent_buffer *eb; 654 bool reads_done; 655 int ret = 0; 656 657 /* 658 * We don't allow bio merge for subpage metadata read, so we should 659 * only get one eb for each endio hook. 660 */ 661 ASSERT(end == start + fs_info->nodesize - 1); 662 ASSERT(PagePrivate(page)); 663 664 eb = find_extent_buffer(fs_info, start); 665 /* 666 * When we are reading one tree block, eb must have been inserted into 667 * the radix tree. If not, something is wrong. 668 */ 669 ASSERT(eb); 670 671 reads_done = atomic_dec_and_test(&eb->io_pages); 672 /* Subpage read must finish in page read */ 673 ASSERT(reads_done); 674 675 eb->read_mirror = mirror; 676 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 677 ret = -EIO; 678 goto err; 679 } 680 ret = validate_extent_buffer(eb); 681 if (ret < 0) 682 goto err; 683 684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 685 btree_readahead_hook(eb, ret); 686 687 set_extent_buffer_uptodate(eb); 688 689 free_extent_buffer(eb); 690 return ret; 691 err: 692 /* 693 * end_bio_extent_readpage decrements io_pages in case of error, 694 * make sure it has something to decrement. 695 */ 696 atomic_inc(&eb->io_pages); 697 clear_extent_buffer_uptodate(eb); 698 free_extent_buffer(eb); 699 return ret; 700 } 701 702 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, 703 struct page *page, u64 start, u64 end, 704 int mirror) 705 { 706 struct extent_buffer *eb; 707 int ret = 0; 708 int reads_done; 709 710 ASSERT(page->private); 711 712 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) 713 return validate_subpage_buffer(page, start, end, mirror); 714 715 eb = (struct extent_buffer *)page->private; 716 717 /* 718 * The pending IO might have been the only thing that kept this buffer 719 * in memory. Make sure we have a ref for all this other checks 720 */ 721 atomic_inc(&eb->refs); 722 723 reads_done = atomic_dec_and_test(&eb->io_pages); 724 if (!reads_done) 725 goto err; 726 727 eb->read_mirror = mirror; 728 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 729 ret = -EIO; 730 goto err; 731 } 732 ret = validate_extent_buffer(eb); 733 err: 734 if (reads_done && 735 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 736 btree_readahead_hook(eb, ret); 737 738 if (ret) { 739 /* 740 * our io error hook is going to dec the io pages 741 * again, we have to make sure it has something 742 * to decrement 743 */ 744 atomic_inc(&eb->io_pages); 745 clear_extent_buffer_uptodate(eb); 746 } 747 free_extent_buffer(eb); 748 749 return ret; 750 } 751 752 static void end_workqueue_bio(struct bio *bio) 753 { 754 struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 755 struct btrfs_fs_info *fs_info; 756 struct btrfs_workqueue *wq; 757 758 fs_info = end_io_wq->info; 759 end_io_wq->status = bio->bi_status; 760 761 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 762 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 763 wq = fs_info->endio_meta_write_workers; 764 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 765 wq = fs_info->endio_freespace_worker; 766 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 767 wq = fs_info->endio_raid56_workers; 768 else 769 wq = fs_info->endio_write_workers; 770 } else { 771 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 772 wq = fs_info->endio_raid56_workers; 773 else if (end_io_wq->metadata) 774 wq = fs_info->endio_meta_workers; 775 else 776 wq = fs_info->endio_workers; 777 } 778 779 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 780 btrfs_queue_work(wq, &end_io_wq->work); 781 } 782 783 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 784 enum btrfs_wq_endio_type metadata) 785 { 786 struct btrfs_end_io_wq *end_io_wq; 787 788 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 789 if (!end_io_wq) 790 return BLK_STS_RESOURCE; 791 792 end_io_wq->private = bio->bi_private; 793 end_io_wq->end_io = bio->bi_end_io; 794 end_io_wq->info = info; 795 end_io_wq->status = 0; 796 end_io_wq->bio = bio; 797 end_io_wq->metadata = metadata; 798 799 bio->bi_private = end_io_wq; 800 bio->bi_end_io = end_workqueue_bio; 801 return 0; 802 } 803 804 static void run_one_async_start(struct btrfs_work *work) 805 { 806 struct async_submit_bio *async; 807 blk_status_t ret; 808 809 async = container_of(work, struct async_submit_bio, work); 810 ret = async->submit_bio_start(async->inode, async->bio, 811 async->dio_file_offset); 812 if (ret) 813 async->status = ret; 814 } 815 816 /* 817 * In order to insert checksums into the metadata in large chunks, we wait 818 * until bio submission time. All the pages in the bio are checksummed and 819 * sums are attached onto the ordered extent record. 820 * 821 * At IO completion time the csums attached on the ordered extent record are 822 * inserted into the tree. 823 */ 824 static void run_one_async_done(struct btrfs_work *work) 825 { 826 struct async_submit_bio *async; 827 struct inode *inode; 828 blk_status_t ret; 829 830 async = container_of(work, struct async_submit_bio, work); 831 inode = async->inode; 832 833 /* If an error occurred we just want to clean up the bio and move on */ 834 if (async->status) { 835 async->bio->bi_status = async->status; 836 bio_endio(async->bio); 837 return; 838 } 839 840 /* 841 * All of the bios that pass through here are from async helpers. 842 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 843 * This changes nothing when cgroups aren't in use. 844 */ 845 async->bio->bi_opf |= REQ_CGROUP_PUNT; 846 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 847 if (ret) { 848 async->bio->bi_status = ret; 849 bio_endio(async->bio); 850 } 851 } 852 853 static void run_one_async_free(struct btrfs_work *work) 854 { 855 struct async_submit_bio *async; 856 857 async = container_of(work, struct async_submit_bio, work); 858 kfree(async); 859 } 860 861 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, 862 int mirror_num, unsigned long bio_flags, 863 u64 dio_file_offset, 864 extent_submit_bio_start_t *submit_bio_start) 865 { 866 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 867 struct async_submit_bio *async; 868 869 async = kmalloc(sizeof(*async), GFP_NOFS); 870 if (!async) 871 return BLK_STS_RESOURCE; 872 873 async->inode = inode; 874 async->bio = bio; 875 async->mirror_num = mirror_num; 876 async->submit_bio_start = submit_bio_start; 877 878 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 879 run_one_async_free); 880 881 async->dio_file_offset = dio_file_offset; 882 883 async->status = 0; 884 885 if (op_is_sync(bio->bi_opf)) 886 btrfs_set_work_high_priority(&async->work); 887 888 btrfs_queue_work(fs_info->workers, &async->work); 889 return 0; 890 } 891 892 static blk_status_t btree_csum_one_bio(struct bio *bio) 893 { 894 struct bio_vec *bvec; 895 struct btrfs_root *root; 896 int ret = 0; 897 struct bvec_iter_all iter_all; 898 899 ASSERT(!bio_flagged(bio, BIO_CLONED)); 900 bio_for_each_segment_all(bvec, bio, iter_all) { 901 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 902 ret = csum_dirty_buffer(root->fs_info, bvec); 903 if (ret) 904 break; 905 } 906 907 return errno_to_blk_status(ret); 908 } 909 910 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 911 u64 dio_file_offset) 912 { 913 /* 914 * when we're called for a write, we're already in the async 915 * submission context. Just jump into btrfs_map_bio 916 */ 917 return btree_csum_one_bio(bio); 918 } 919 920 static int check_async_write(struct btrfs_fs_info *fs_info, 921 struct btrfs_inode *bi) 922 { 923 if (btrfs_is_zoned(fs_info)) 924 return 0; 925 if (atomic_read(&bi->sync_writers)) 926 return 0; 927 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 928 return 0; 929 return 1; 930 } 931 932 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 933 int mirror_num, unsigned long bio_flags) 934 { 935 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 936 int async = check_async_write(fs_info, BTRFS_I(inode)); 937 blk_status_t ret; 938 939 if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 940 /* 941 * called for a read, do the setup so that checksum validation 942 * can happen in the async kernel threads 943 */ 944 ret = btrfs_bio_wq_end_io(fs_info, bio, 945 BTRFS_WQ_ENDIO_METADATA); 946 if (ret) 947 goto out_w_error; 948 ret = btrfs_map_bio(fs_info, bio, mirror_num); 949 } else if (!async) { 950 ret = btree_csum_one_bio(bio); 951 if (ret) 952 goto out_w_error; 953 ret = btrfs_map_bio(fs_info, bio, mirror_num); 954 } else { 955 /* 956 * kthread helpers are used to submit writes so that 957 * checksumming can happen in parallel across all CPUs 958 */ 959 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, 960 0, btree_submit_bio_start); 961 } 962 963 if (ret) 964 goto out_w_error; 965 return 0; 966 967 out_w_error: 968 bio->bi_status = ret; 969 bio_endio(bio); 970 return ret; 971 } 972 973 #ifdef CONFIG_MIGRATION 974 static int btree_migratepage(struct address_space *mapping, 975 struct page *newpage, struct page *page, 976 enum migrate_mode mode) 977 { 978 /* 979 * we can't safely write a btree page from here, 980 * we haven't done the locking hook 981 */ 982 if (PageDirty(page)) 983 return -EAGAIN; 984 /* 985 * Buffers may be managed in a filesystem specific way. 986 * We must have no buffers or drop them. 987 */ 988 if (page_has_private(page) && 989 !try_to_release_page(page, GFP_KERNEL)) 990 return -EAGAIN; 991 return migrate_page(mapping, newpage, page, mode); 992 } 993 #endif 994 995 996 static int btree_writepages(struct address_space *mapping, 997 struct writeback_control *wbc) 998 { 999 struct btrfs_fs_info *fs_info; 1000 int ret; 1001 1002 if (wbc->sync_mode == WB_SYNC_NONE) { 1003 1004 if (wbc->for_kupdate) 1005 return 0; 1006 1007 fs_info = BTRFS_I(mapping->host)->root->fs_info; 1008 /* this is a bit racy, but that's ok */ 1009 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1010 BTRFS_DIRTY_METADATA_THRESH, 1011 fs_info->dirty_metadata_batch); 1012 if (ret < 0) 1013 return 0; 1014 } 1015 return btree_write_cache_pages(mapping, wbc); 1016 } 1017 1018 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 1019 { 1020 if (PageWriteback(page) || PageDirty(page)) 1021 return 0; 1022 1023 return try_release_extent_buffer(page); 1024 } 1025 1026 static void btree_invalidatepage(struct page *page, unsigned int offset, 1027 unsigned int length) 1028 { 1029 struct extent_io_tree *tree; 1030 tree = &BTRFS_I(page->mapping->host)->io_tree; 1031 extent_invalidatepage(tree, page, offset); 1032 btree_releasepage(page, GFP_NOFS); 1033 if (PagePrivate(page)) { 1034 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1035 "page private not zero on page %llu", 1036 (unsigned long long)page_offset(page)); 1037 detach_page_private(page); 1038 } 1039 } 1040 1041 static int btree_set_page_dirty(struct page *page) 1042 { 1043 #ifdef DEBUG 1044 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 1045 struct btrfs_subpage *subpage; 1046 struct extent_buffer *eb; 1047 int cur_bit = 0; 1048 u64 page_start = page_offset(page); 1049 1050 if (fs_info->sectorsize == PAGE_SIZE) { 1051 BUG_ON(!PagePrivate(page)); 1052 eb = (struct extent_buffer *)page->private; 1053 BUG_ON(!eb); 1054 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1055 BUG_ON(!atomic_read(&eb->refs)); 1056 btrfs_assert_tree_locked(eb); 1057 return __set_page_dirty_nobuffers(page); 1058 } 1059 ASSERT(PagePrivate(page) && page->private); 1060 subpage = (struct btrfs_subpage *)page->private; 1061 1062 ASSERT(subpage->dirty_bitmap); 1063 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 1064 unsigned long flags; 1065 u64 cur; 1066 u16 tmp = (1 << cur_bit); 1067 1068 spin_lock_irqsave(&subpage->lock, flags); 1069 if (!(tmp & subpage->dirty_bitmap)) { 1070 spin_unlock_irqrestore(&subpage->lock, flags); 1071 cur_bit++; 1072 continue; 1073 } 1074 spin_unlock_irqrestore(&subpage->lock, flags); 1075 cur = page_start + cur_bit * fs_info->sectorsize; 1076 1077 eb = find_extent_buffer(fs_info, cur); 1078 ASSERT(eb); 1079 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1080 ASSERT(atomic_read(&eb->refs)); 1081 btrfs_assert_tree_locked(eb); 1082 free_extent_buffer(eb); 1083 1084 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 1085 } 1086 #endif 1087 return __set_page_dirty_nobuffers(page); 1088 } 1089 1090 static const struct address_space_operations btree_aops = { 1091 .writepages = btree_writepages, 1092 .releasepage = btree_releasepage, 1093 .invalidatepage = btree_invalidatepage, 1094 #ifdef CONFIG_MIGRATION 1095 .migratepage = btree_migratepage, 1096 #endif 1097 .set_page_dirty = btree_set_page_dirty, 1098 }; 1099 1100 struct extent_buffer *btrfs_find_create_tree_block( 1101 struct btrfs_fs_info *fs_info, 1102 u64 bytenr, u64 owner_root, 1103 int level) 1104 { 1105 if (btrfs_is_testing(fs_info)) 1106 return alloc_test_extent_buffer(fs_info, bytenr); 1107 return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 1108 } 1109 1110 /* 1111 * Read tree block at logical address @bytenr and do variant basic but critical 1112 * verification. 1113 * 1114 * @owner_root: the objectid of the root owner for this block. 1115 * @parent_transid: expected transid of this tree block, skip check if 0 1116 * @level: expected level, mandatory check 1117 * @first_key: expected key in slot 0, skip check if NULL 1118 */ 1119 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1120 u64 owner_root, u64 parent_transid, 1121 int level, struct btrfs_key *first_key) 1122 { 1123 struct extent_buffer *buf = NULL; 1124 int ret; 1125 1126 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 1127 if (IS_ERR(buf)) 1128 return buf; 1129 1130 ret = btree_read_extent_buffer_pages(buf, parent_transid, 1131 level, first_key); 1132 if (ret) { 1133 free_extent_buffer_stale(buf); 1134 return ERR_PTR(ret); 1135 } 1136 return buf; 1137 1138 } 1139 1140 void btrfs_clean_tree_block(struct extent_buffer *buf) 1141 { 1142 struct btrfs_fs_info *fs_info = buf->fs_info; 1143 if (btrfs_header_generation(buf) == 1144 fs_info->running_transaction->transid) { 1145 btrfs_assert_tree_locked(buf); 1146 1147 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1148 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1149 -buf->len, 1150 fs_info->dirty_metadata_batch); 1151 clear_extent_buffer_dirty(buf); 1152 } 1153 } 1154 } 1155 1156 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1157 u64 objectid) 1158 { 1159 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1160 root->fs_info = fs_info; 1161 root->node = NULL; 1162 root->commit_root = NULL; 1163 root->state = 0; 1164 root->orphan_cleanup_state = 0; 1165 1166 root->last_trans = 0; 1167 root->free_objectid = 0; 1168 root->nr_delalloc_inodes = 0; 1169 root->nr_ordered_extents = 0; 1170 root->inode_tree = RB_ROOT; 1171 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1172 root->block_rsv = NULL; 1173 1174 INIT_LIST_HEAD(&root->dirty_list); 1175 INIT_LIST_HEAD(&root->root_list); 1176 INIT_LIST_HEAD(&root->delalloc_inodes); 1177 INIT_LIST_HEAD(&root->delalloc_root); 1178 INIT_LIST_HEAD(&root->ordered_extents); 1179 INIT_LIST_HEAD(&root->ordered_root); 1180 INIT_LIST_HEAD(&root->reloc_dirty_list); 1181 INIT_LIST_HEAD(&root->logged_list[0]); 1182 INIT_LIST_HEAD(&root->logged_list[1]); 1183 spin_lock_init(&root->inode_lock); 1184 spin_lock_init(&root->delalloc_lock); 1185 spin_lock_init(&root->ordered_extent_lock); 1186 spin_lock_init(&root->accounting_lock); 1187 spin_lock_init(&root->log_extents_lock[0]); 1188 spin_lock_init(&root->log_extents_lock[1]); 1189 spin_lock_init(&root->qgroup_meta_rsv_lock); 1190 mutex_init(&root->objectid_mutex); 1191 mutex_init(&root->log_mutex); 1192 mutex_init(&root->ordered_extent_mutex); 1193 mutex_init(&root->delalloc_mutex); 1194 init_waitqueue_head(&root->qgroup_flush_wait); 1195 init_waitqueue_head(&root->log_writer_wait); 1196 init_waitqueue_head(&root->log_commit_wait[0]); 1197 init_waitqueue_head(&root->log_commit_wait[1]); 1198 INIT_LIST_HEAD(&root->log_ctxs[0]); 1199 INIT_LIST_HEAD(&root->log_ctxs[1]); 1200 atomic_set(&root->log_commit[0], 0); 1201 atomic_set(&root->log_commit[1], 0); 1202 atomic_set(&root->log_writers, 0); 1203 atomic_set(&root->log_batch, 0); 1204 refcount_set(&root->refs, 1); 1205 atomic_set(&root->snapshot_force_cow, 0); 1206 atomic_set(&root->nr_swapfiles, 0); 1207 root->log_transid = 0; 1208 root->log_transid_committed = -1; 1209 root->last_log_commit = 0; 1210 if (!dummy) { 1211 extent_io_tree_init(fs_info, &root->dirty_log_pages, 1212 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1213 extent_io_tree_init(fs_info, &root->log_csum_range, 1214 IO_TREE_LOG_CSUM_RANGE, NULL); 1215 } 1216 1217 memset(&root->root_key, 0, sizeof(root->root_key)); 1218 memset(&root->root_item, 0, sizeof(root->root_item)); 1219 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 1220 root->root_key.objectid = objectid; 1221 root->anon_dev = 0; 1222 1223 spin_lock_init(&root->root_item_lock); 1224 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1225 #ifdef CONFIG_BTRFS_DEBUG 1226 INIT_LIST_HEAD(&root->leak_list); 1227 spin_lock(&fs_info->fs_roots_radix_lock); 1228 list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1229 spin_unlock(&fs_info->fs_roots_radix_lock); 1230 #endif 1231 } 1232 1233 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 1234 u64 objectid, gfp_t flags) 1235 { 1236 struct btrfs_root *root = kzalloc(sizeof(*root), flags); 1237 if (root) 1238 __setup_root(root, fs_info, objectid); 1239 return root; 1240 } 1241 1242 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1243 /* Should only be used by the testing infrastructure */ 1244 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 1245 { 1246 struct btrfs_root *root; 1247 1248 if (!fs_info) 1249 return ERR_PTR(-EINVAL); 1250 1251 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 1252 if (!root) 1253 return ERR_PTR(-ENOMEM); 1254 1255 /* We don't use the stripesize in selftest, set it as sectorsize */ 1256 root->alloc_bytenr = 0; 1257 1258 return root; 1259 } 1260 #endif 1261 1262 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 1263 u64 objectid) 1264 { 1265 struct btrfs_fs_info *fs_info = trans->fs_info; 1266 struct extent_buffer *leaf; 1267 struct btrfs_root *tree_root = fs_info->tree_root; 1268 struct btrfs_root *root; 1269 struct btrfs_key key; 1270 unsigned int nofs_flag; 1271 int ret = 0; 1272 1273 /* 1274 * We're holding a transaction handle, so use a NOFS memory allocation 1275 * context to avoid deadlock if reclaim happens. 1276 */ 1277 nofs_flag = memalloc_nofs_save(); 1278 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1279 memalloc_nofs_restore(nofs_flag); 1280 if (!root) 1281 return ERR_PTR(-ENOMEM); 1282 1283 root->root_key.objectid = objectid; 1284 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1285 root->root_key.offset = 0; 1286 1287 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 1288 BTRFS_NESTING_NORMAL); 1289 if (IS_ERR(leaf)) { 1290 ret = PTR_ERR(leaf); 1291 leaf = NULL; 1292 goto fail_unlock; 1293 } 1294 1295 root->node = leaf; 1296 btrfs_mark_buffer_dirty(leaf); 1297 1298 root->commit_root = btrfs_root_node(root); 1299 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 1300 1301 btrfs_set_root_flags(&root->root_item, 0); 1302 btrfs_set_root_limit(&root->root_item, 0); 1303 btrfs_set_root_bytenr(&root->root_item, leaf->start); 1304 btrfs_set_root_generation(&root->root_item, trans->transid); 1305 btrfs_set_root_level(&root->root_item, 0); 1306 btrfs_set_root_refs(&root->root_item, 1); 1307 btrfs_set_root_used(&root->root_item, leaf->len); 1308 btrfs_set_root_last_snapshot(&root->root_item, 0); 1309 btrfs_set_root_dirid(&root->root_item, 0); 1310 if (is_fstree(objectid)) 1311 generate_random_guid(root->root_item.uuid); 1312 else 1313 export_guid(root->root_item.uuid, &guid_null); 1314 btrfs_set_root_drop_level(&root->root_item, 0); 1315 1316 btrfs_tree_unlock(leaf); 1317 1318 key.objectid = objectid; 1319 key.type = BTRFS_ROOT_ITEM_KEY; 1320 key.offset = 0; 1321 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 1322 if (ret) 1323 goto fail; 1324 1325 return root; 1326 1327 fail_unlock: 1328 if (leaf) 1329 btrfs_tree_unlock(leaf); 1330 fail: 1331 btrfs_put_root(root); 1332 1333 return ERR_PTR(ret); 1334 } 1335 1336 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1337 struct btrfs_fs_info *fs_info) 1338 { 1339 struct btrfs_root *root; 1340 1341 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1342 if (!root) 1343 return ERR_PTR(-ENOMEM); 1344 1345 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1346 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1347 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1348 1349 return root; 1350 } 1351 1352 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 1353 struct btrfs_root *root) 1354 { 1355 struct extent_buffer *leaf; 1356 1357 /* 1358 * DON'T set SHAREABLE bit for log trees. 1359 * 1360 * Log trees are not exposed to user space thus can't be snapshotted, 1361 * and they go away before a real commit is actually done. 1362 * 1363 * They do store pointers to file data extents, and those reference 1364 * counts still get updated (along with back refs to the log tree). 1365 */ 1366 1367 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 1368 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 1369 if (IS_ERR(leaf)) 1370 return PTR_ERR(leaf); 1371 1372 root->node = leaf; 1373 1374 btrfs_mark_buffer_dirty(root->node); 1375 btrfs_tree_unlock(root->node); 1376 1377 return 0; 1378 } 1379 1380 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1381 struct btrfs_fs_info *fs_info) 1382 { 1383 struct btrfs_root *log_root; 1384 1385 log_root = alloc_log_tree(trans, fs_info); 1386 if (IS_ERR(log_root)) 1387 return PTR_ERR(log_root); 1388 1389 if (!btrfs_is_zoned(fs_info)) { 1390 int ret = btrfs_alloc_log_tree_node(trans, log_root); 1391 1392 if (ret) { 1393 btrfs_put_root(log_root); 1394 return ret; 1395 } 1396 } 1397 1398 WARN_ON(fs_info->log_root_tree); 1399 fs_info->log_root_tree = log_root; 1400 return 0; 1401 } 1402 1403 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1404 struct btrfs_root *root) 1405 { 1406 struct btrfs_fs_info *fs_info = root->fs_info; 1407 struct btrfs_root *log_root; 1408 struct btrfs_inode_item *inode_item; 1409 int ret; 1410 1411 log_root = alloc_log_tree(trans, fs_info); 1412 if (IS_ERR(log_root)) 1413 return PTR_ERR(log_root); 1414 1415 ret = btrfs_alloc_log_tree_node(trans, log_root); 1416 if (ret) { 1417 btrfs_put_root(log_root); 1418 return ret; 1419 } 1420 1421 log_root->last_trans = trans->transid; 1422 log_root->root_key.offset = root->root_key.objectid; 1423 1424 inode_item = &log_root->root_item.inode; 1425 btrfs_set_stack_inode_generation(inode_item, 1); 1426 btrfs_set_stack_inode_size(inode_item, 3); 1427 btrfs_set_stack_inode_nlink(inode_item, 1); 1428 btrfs_set_stack_inode_nbytes(inode_item, 1429 fs_info->nodesize); 1430 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1431 1432 btrfs_set_root_node(&log_root->root_item, log_root->node); 1433 1434 WARN_ON(root->log_root); 1435 root->log_root = log_root; 1436 root->log_transid = 0; 1437 root->log_transid_committed = -1; 1438 root->last_log_commit = 0; 1439 return 0; 1440 } 1441 1442 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 1443 struct btrfs_path *path, 1444 struct btrfs_key *key) 1445 { 1446 struct btrfs_root *root; 1447 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1448 u64 generation; 1449 int ret; 1450 int level; 1451 1452 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 1453 if (!root) 1454 return ERR_PTR(-ENOMEM); 1455 1456 ret = btrfs_find_root(tree_root, key, path, 1457 &root->root_item, &root->root_key); 1458 if (ret) { 1459 if (ret > 0) 1460 ret = -ENOENT; 1461 goto fail; 1462 } 1463 1464 generation = btrfs_root_generation(&root->root_item); 1465 level = btrfs_root_level(&root->root_item); 1466 root->node = read_tree_block(fs_info, 1467 btrfs_root_bytenr(&root->root_item), 1468 key->objectid, generation, level, NULL); 1469 if (IS_ERR(root->node)) { 1470 ret = PTR_ERR(root->node); 1471 root->node = NULL; 1472 goto fail; 1473 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1474 ret = -EIO; 1475 goto fail; 1476 } 1477 root->commit_root = btrfs_root_node(root); 1478 return root; 1479 fail: 1480 btrfs_put_root(root); 1481 return ERR_PTR(ret); 1482 } 1483 1484 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1485 struct btrfs_key *key) 1486 { 1487 struct btrfs_root *root; 1488 struct btrfs_path *path; 1489 1490 path = btrfs_alloc_path(); 1491 if (!path) 1492 return ERR_PTR(-ENOMEM); 1493 root = read_tree_root_path(tree_root, path, key); 1494 btrfs_free_path(path); 1495 1496 return root; 1497 } 1498 1499 /* 1500 * Initialize subvolume root in-memory structure 1501 * 1502 * @anon_dev: anonymous device to attach to the root, if zero, allocate new 1503 */ 1504 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1505 { 1506 int ret; 1507 unsigned int nofs_flag; 1508 1509 /* 1510 * We might be called under a transaction (e.g. indirect backref 1511 * resolution) which could deadlock if it triggers memory reclaim 1512 */ 1513 nofs_flag = memalloc_nofs_save(); 1514 ret = btrfs_drew_lock_init(&root->snapshot_lock); 1515 memalloc_nofs_restore(nofs_flag); 1516 if (ret) 1517 goto fail; 1518 1519 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1520 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { 1521 set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1522 btrfs_check_and_init_root_item(&root->root_item); 1523 } 1524 1525 /* 1526 * Don't assign anonymous block device to roots that are not exposed to 1527 * userspace, the id pool is limited to 1M 1528 */ 1529 if (is_fstree(root->root_key.objectid) && 1530 btrfs_root_refs(&root->root_item) > 0) { 1531 if (!anon_dev) { 1532 ret = get_anon_bdev(&root->anon_dev); 1533 if (ret) 1534 goto fail; 1535 } else { 1536 root->anon_dev = anon_dev; 1537 } 1538 } 1539 1540 mutex_lock(&root->objectid_mutex); 1541 ret = btrfs_init_root_free_objectid(root); 1542 if (ret) { 1543 mutex_unlock(&root->objectid_mutex); 1544 goto fail; 1545 } 1546 1547 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1548 1549 mutex_unlock(&root->objectid_mutex); 1550 1551 return 0; 1552 fail: 1553 /* The caller is responsible to call btrfs_free_fs_root */ 1554 return ret; 1555 } 1556 1557 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1558 u64 root_id) 1559 { 1560 struct btrfs_root *root; 1561 1562 spin_lock(&fs_info->fs_roots_radix_lock); 1563 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1564 (unsigned long)root_id); 1565 if (root) 1566 root = btrfs_grab_root(root); 1567 spin_unlock(&fs_info->fs_roots_radix_lock); 1568 return root; 1569 } 1570 1571 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 1572 u64 objectid) 1573 { 1574 if (objectid == BTRFS_ROOT_TREE_OBJECTID) 1575 return btrfs_grab_root(fs_info->tree_root); 1576 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1577 return btrfs_grab_root(fs_info->extent_root); 1578 if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 1579 return btrfs_grab_root(fs_info->chunk_root); 1580 if (objectid == BTRFS_DEV_TREE_OBJECTID) 1581 return btrfs_grab_root(fs_info->dev_root); 1582 if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1583 return btrfs_grab_root(fs_info->csum_root); 1584 if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 1585 return btrfs_grab_root(fs_info->quota_root) ? 1586 fs_info->quota_root : ERR_PTR(-ENOENT); 1587 if (objectid == BTRFS_UUID_TREE_OBJECTID) 1588 return btrfs_grab_root(fs_info->uuid_root) ? 1589 fs_info->uuid_root : ERR_PTR(-ENOENT); 1590 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 1591 return btrfs_grab_root(fs_info->free_space_root) ? 1592 fs_info->free_space_root : ERR_PTR(-ENOENT); 1593 return NULL; 1594 } 1595 1596 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1597 struct btrfs_root *root) 1598 { 1599 int ret; 1600 1601 ret = radix_tree_preload(GFP_NOFS); 1602 if (ret) 1603 return ret; 1604 1605 spin_lock(&fs_info->fs_roots_radix_lock); 1606 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1607 (unsigned long)root->root_key.objectid, 1608 root); 1609 if (ret == 0) { 1610 btrfs_grab_root(root); 1611 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1612 } 1613 spin_unlock(&fs_info->fs_roots_radix_lock); 1614 radix_tree_preload_end(); 1615 1616 return ret; 1617 } 1618 1619 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1620 { 1621 #ifdef CONFIG_BTRFS_DEBUG 1622 struct btrfs_root *root; 1623 1624 while (!list_empty(&fs_info->allocated_roots)) { 1625 char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1626 1627 root = list_first_entry(&fs_info->allocated_roots, 1628 struct btrfs_root, leak_list); 1629 btrfs_err(fs_info, "leaked root %s refcount %d", 1630 btrfs_root_name(&root->root_key, buf), 1631 refcount_read(&root->refs)); 1632 while (refcount_read(&root->refs) > 1) 1633 btrfs_put_root(root); 1634 btrfs_put_root(root); 1635 } 1636 #endif 1637 } 1638 1639 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 1640 { 1641 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1642 percpu_counter_destroy(&fs_info->delalloc_bytes); 1643 percpu_counter_destroy(&fs_info->ordered_bytes); 1644 percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1645 btrfs_free_csum_hash(fs_info); 1646 btrfs_free_stripe_hash_table(fs_info); 1647 btrfs_free_ref_cache(fs_info); 1648 kfree(fs_info->balance_ctl); 1649 kfree(fs_info->delayed_root); 1650 btrfs_put_root(fs_info->extent_root); 1651 btrfs_put_root(fs_info->tree_root); 1652 btrfs_put_root(fs_info->chunk_root); 1653 btrfs_put_root(fs_info->dev_root); 1654 btrfs_put_root(fs_info->csum_root); 1655 btrfs_put_root(fs_info->quota_root); 1656 btrfs_put_root(fs_info->uuid_root); 1657 btrfs_put_root(fs_info->free_space_root); 1658 btrfs_put_root(fs_info->fs_root); 1659 btrfs_put_root(fs_info->data_reloc_root); 1660 btrfs_check_leaked_roots(fs_info); 1661 btrfs_extent_buffer_leak_debug_check(fs_info); 1662 kfree(fs_info->super_copy); 1663 kfree(fs_info->super_for_commit); 1664 kvfree(fs_info); 1665 } 1666 1667 1668 /* 1669 * Get an in-memory reference of a root structure. 1670 * 1671 * For essential trees like root/extent tree, we grab it from fs_info directly. 1672 * For subvolume trees, we check the cached filesystem roots first. If not 1673 * found, then read it from disk and add it to cached fs roots. 1674 * 1675 * Caller should release the root by calling btrfs_put_root() after the usage. 1676 * 1677 * NOTE: Reloc and log trees can't be read by this function as they share the 1678 * same root objectid. 1679 * 1680 * @objectid: root id 1681 * @anon_dev: preallocated anonymous block device number for new roots, 1682 * pass 0 for new allocation. 1683 * @check_ref: whether to check root item references, If true, return -ENOENT 1684 * for orphan roots 1685 */ 1686 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 1687 u64 objectid, dev_t anon_dev, 1688 bool check_ref) 1689 { 1690 struct btrfs_root *root; 1691 struct btrfs_path *path; 1692 struct btrfs_key key; 1693 int ret; 1694 1695 root = btrfs_get_global_root(fs_info, objectid); 1696 if (root) 1697 return root; 1698 again: 1699 root = btrfs_lookup_fs_root(fs_info, objectid); 1700 if (root) { 1701 /* Shouldn't get preallocated anon_dev for cached roots */ 1702 ASSERT(!anon_dev); 1703 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1704 btrfs_put_root(root); 1705 return ERR_PTR(-ENOENT); 1706 } 1707 return root; 1708 } 1709 1710 key.objectid = objectid; 1711 key.type = BTRFS_ROOT_ITEM_KEY; 1712 key.offset = (u64)-1; 1713 root = btrfs_read_tree_root(fs_info->tree_root, &key); 1714 if (IS_ERR(root)) 1715 return root; 1716 1717 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1718 ret = -ENOENT; 1719 goto fail; 1720 } 1721 1722 ret = btrfs_init_fs_root(root, anon_dev); 1723 if (ret) 1724 goto fail; 1725 1726 path = btrfs_alloc_path(); 1727 if (!path) { 1728 ret = -ENOMEM; 1729 goto fail; 1730 } 1731 key.objectid = BTRFS_ORPHAN_OBJECTID; 1732 key.type = BTRFS_ORPHAN_ITEM_KEY; 1733 key.offset = objectid; 1734 1735 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1736 btrfs_free_path(path); 1737 if (ret < 0) 1738 goto fail; 1739 if (ret == 0) 1740 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1741 1742 ret = btrfs_insert_fs_root(fs_info, root); 1743 if (ret) { 1744 btrfs_put_root(root); 1745 if (ret == -EEXIST) 1746 goto again; 1747 goto fail; 1748 } 1749 return root; 1750 fail: 1751 btrfs_put_root(root); 1752 return ERR_PTR(ret); 1753 } 1754 1755 /* 1756 * Get in-memory reference of a root structure 1757 * 1758 * @objectid: tree objectid 1759 * @check_ref: if set, verify that the tree exists and the item has at least 1760 * one reference 1761 */ 1762 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1763 u64 objectid, bool check_ref) 1764 { 1765 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 1766 } 1767 1768 /* 1769 * Get in-memory reference of a root structure, created as new, optionally pass 1770 * the anonymous block device id 1771 * 1772 * @objectid: tree objectid 1773 * @anon_dev: if zero, allocate a new anonymous block device or use the 1774 * parameter value 1775 */ 1776 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 1777 u64 objectid, dev_t anon_dev) 1778 { 1779 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 1780 } 1781 1782 /* 1783 * btrfs_get_fs_root_commit_root - return a root for the given objectid 1784 * @fs_info: the fs_info 1785 * @objectid: the objectid we need to lookup 1786 * 1787 * This is exclusively used for backref walking, and exists specifically because 1788 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 1789 * creation time, which means we may have to read the tree_root in order to look 1790 * up a fs root that is not in memory. If the root is not in memory we will 1791 * read the tree root commit root and look up the fs root from there. This is a 1792 * temporary root, it will not be inserted into the radix tree as it doesn't 1793 * have the most uptodate information, it'll simply be discarded once the 1794 * backref code is finished using the root. 1795 */ 1796 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 1797 struct btrfs_path *path, 1798 u64 objectid) 1799 { 1800 struct btrfs_root *root; 1801 struct btrfs_key key; 1802 1803 ASSERT(path->search_commit_root && path->skip_locking); 1804 1805 /* 1806 * This can return -ENOENT if we ask for a root that doesn't exist, but 1807 * since this is called via the backref walking code we won't be looking 1808 * up a root that doesn't exist, unless there's corruption. So if root 1809 * != NULL just return it. 1810 */ 1811 root = btrfs_get_global_root(fs_info, objectid); 1812 if (root) 1813 return root; 1814 1815 root = btrfs_lookup_fs_root(fs_info, objectid); 1816 if (root) 1817 return root; 1818 1819 key.objectid = objectid; 1820 key.type = BTRFS_ROOT_ITEM_KEY; 1821 key.offset = (u64)-1; 1822 root = read_tree_root_path(fs_info->tree_root, path, &key); 1823 btrfs_release_path(path); 1824 1825 return root; 1826 } 1827 1828 /* 1829 * called by the kthread helper functions to finally call the bio end_io 1830 * functions. This is where read checksum verification actually happens 1831 */ 1832 static void end_workqueue_fn(struct btrfs_work *work) 1833 { 1834 struct bio *bio; 1835 struct btrfs_end_io_wq *end_io_wq; 1836 1837 end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1838 bio = end_io_wq->bio; 1839 1840 bio->bi_status = end_io_wq->status; 1841 bio->bi_private = end_io_wq->private; 1842 bio->bi_end_io = end_io_wq->end_io; 1843 bio_endio(bio); 1844 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1845 } 1846 1847 static int cleaner_kthread(void *arg) 1848 { 1849 struct btrfs_root *root = arg; 1850 struct btrfs_fs_info *fs_info = root->fs_info; 1851 int again; 1852 1853 while (1) { 1854 again = 0; 1855 1856 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1857 1858 /* Make the cleaner go to sleep early. */ 1859 if (btrfs_need_cleaner_sleep(fs_info)) 1860 goto sleep; 1861 1862 /* 1863 * Do not do anything if we might cause open_ctree() to block 1864 * before we have finished mounting the filesystem. 1865 */ 1866 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1867 goto sleep; 1868 1869 if (!mutex_trylock(&fs_info->cleaner_mutex)) 1870 goto sleep; 1871 1872 /* 1873 * Avoid the problem that we change the status of the fs 1874 * during the above check and trylock. 1875 */ 1876 if (btrfs_need_cleaner_sleep(fs_info)) { 1877 mutex_unlock(&fs_info->cleaner_mutex); 1878 goto sleep; 1879 } 1880 1881 btrfs_run_delayed_iputs(fs_info); 1882 1883 again = btrfs_clean_one_deleted_snapshot(root); 1884 mutex_unlock(&fs_info->cleaner_mutex); 1885 1886 /* 1887 * The defragger has dealt with the R/O remount and umount, 1888 * needn't do anything special here. 1889 */ 1890 btrfs_run_defrag_inodes(fs_info); 1891 1892 /* 1893 * Acquires fs_info->reclaim_bgs_lock to avoid racing 1894 * with relocation (btrfs_relocate_chunk) and relocation 1895 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1896 * after acquiring fs_info->reclaim_bgs_lock. So we 1897 * can't hold, nor need to, fs_info->cleaner_mutex when deleting 1898 * unused block groups. 1899 */ 1900 btrfs_delete_unused_bgs(fs_info); 1901 1902 /* 1903 * Reclaim block groups in the reclaim_bgs list after we deleted 1904 * all unused block_groups. This possibly gives us some more free 1905 * space. 1906 */ 1907 btrfs_reclaim_bgs(fs_info); 1908 sleep: 1909 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1910 if (kthread_should_park()) 1911 kthread_parkme(); 1912 if (kthread_should_stop()) 1913 return 0; 1914 if (!again) { 1915 set_current_state(TASK_INTERRUPTIBLE); 1916 schedule(); 1917 __set_current_state(TASK_RUNNING); 1918 } 1919 } 1920 } 1921 1922 static int transaction_kthread(void *arg) 1923 { 1924 struct btrfs_root *root = arg; 1925 struct btrfs_fs_info *fs_info = root->fs_info; 1926 struct btrfs_trans_handle *trans; 1927 struct btrfs_transaction *cur; 1928 u64 transid; 1929 time64_t delta; 1930 unsigned long delay; 1931 bool cannot_commit; 1932 1933 do { 1934 cannot_commit = false; 1935 delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 1936 mutex_lock(&fs_info->transaction_kthread_mutex); 1937 1938 spin_lock(&fs_info->trans_lock); 1939 cur = fs_info->running_transaction; 1940 if (!cur) { 1941 spin_unlock(&fs_info->trans_lock); 1942 goto sleep; 1943 } 1944 1945 delta = ktime_get_seconds() - cur->start_time; 1946 if (cur->state < TRANS_STATE_COMMIT_START && 1947 delta < fs_info->commit_interval) { 1948 spin_unlock(&fs_info->trans_lock); 1949 delay -= msecs_to_jiffies((delta - 1) * 1000); 1950 delay = min(delay, 1951 msecs_to_jiffies(fs_info->commit_interval * 1000)); 1952 goto sleep; 1953 } 1954 transid = cur->transid; 1955 spin_unlock(&fs_info->trans_lock); 1956 1957 /* If the file system is aborted, this will always fail. */ 1958 trans = btrfs_attach_transaction(root); 1959 if (IS_ERR(trans)) { 1960 if (PTR_ERR(trans) != -ENOENT) 1961 cannot_commit = true; 1962 goto sleep; 1963 } 1964 if (transid == trans->transid) { 1965 btrfs_commit_transaction(trans); 1966 } else { 1967 btrfs_end_transaction(trans); 1968 } 1969 sleep: 1970 wake_up_process(fs_info->cleaner_kthread); 1971 mutex_unlock(&fs_info->transaction_kthread_mutex); 1972 1973 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 1974 &fs_info->fs_state))) 1975 btrfs_cleanup_transaction(fs_info); 1976 if (!kthread_should_stop() && 1977 (!btrfs_transaction_blocked(fs_info) || 1978 cannot_commit)) 1979 schedule_timeout_interruptible(delay); 1980 } while (!kthread_should_stop()); 1981 return 0; 1982 } 1983 1984 /* 1985 * This will find the highest generation in the array of root backups. The 1986 * index of the highest array is returned, or -EINVAL if we can't find 1987 * anything. 1988 * 1989 * We check to make sure the array is valid by comparing the 1990 * generation of the latest root in the array with the generation 1991 * in the super block. If they don't match we pitch it. 1992 */ 1993 static int find_newest_super_backup(struct btrfs_fs_info *info) 1994 { 1995 const u64 newest_gen = btrfs_super_generation(info->super_copy); 1996 u64 cur; 1997 struct btrfs_root_backup *root_backup; 1998 int i; 1999 2000 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2001 root_backup = info->super_copy->super_roots + i; 2002 cur = btrfs_backup_tree_root_gen(root_backup); 2003 if (cur == newest_gen) 2004 return i; 2005 } 2006 2007 return -EINVAL; 2008 } 2009 2010 /* 2011 * copy all the root pointers into the super backup array. 2012 * this will bump the backup pointer by one when it is 2013 * done 2014 */ 2015 static void backup_super_roots(struct btrfs_fs_info *info) 2016 { 2017 const int next_backup = info->backup_root_index; 2018 struct btrfs_root_backup *root_backup; 2019 2020 root_backup = info->super_for_commit->super_roots + next_backup; 2021 2022 /* 2023 * make sure all of our padding and empty slots get zero filled 2024 * regardless of which ones we use today 2025 */ 2026 memset(root_backup, 0, sizeof(*root_backup)); 2027 2028 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 2029 2030 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 2031 btrfs_set_backup_tree_root_gen(root_backup, 2032 btrfs_header_generation(info->tree_root->node)); 2033 2034 btrfs_set_backup_tree_root_level(root_backup, 2035 btrfs_header_level(info->tree_root->node)); 2036 2037 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 2038 btrfs_set_backup_chunk_root_gen(root_backup, 2039 btrfs_header_generation(info->chunk_root->node)); 2040 btrfs_set_backup_chunk_root_level(root_backup, 2041 btrfs_header_level(info->chunk_root->node)); 2042 2043 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 2044 btrfs_set_backup_extent_root_gen(root_backup, 2045 btrfs_header_generation(info->extent_root->node)); 2046 btrfs_set_backup_extent_root_level(root_backup, 2047 btrfs_header_level(info->extent_root->node)); 2048 2049 /* 2050 * we might commit during log recovery, which happens before we set 2051 * the fs_root. Make sure it is valid before we fill it in. 2052 */ 2053 if (info->fs_root && info->fs_root->node) { 2054 btrfs_set_backup_fs_root(root_backup, 2055 info->fs_root->node->start); 2056 btrfs_set_backup_fs_root_gen(root_backup, 2057 btrfs_header_generation(info->fs_root->node)); 2058 btrfs_set_backup_fs_root_level(root_backup, 2059 btrfs_header_level(info->fs_root->node)); 2060 } 2061 2062 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2063 btrfs_set_backup_dev_root_gen(root_backup, 2064 btrfs_header_generation(info->dev_root->node)); 2065 btrfs_set_backup_dev_root_level(root_backup, 2066 btrfs_header_level(info->dev_root->node)); 2067 2068 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 2069 btrfs_set_backup_csum_root_gen(root_backup, 2070 btrfs_header_generation(info->csum_root->node)); 2071 btrfs_set_backup_csum_root_level(root_backup, 2072 btrfs_header_level(info->csum_root->node)); 2073 2074 btrfs_set_backup_total_bytes(root_backup, 2075 btrfs_super_total_bytes(info->super_copy)); 2076 btrfs_set_backup_bytes_used(root_backup, 2077 btrfs_super_bytes_used(info->super_copy)); 2078 btrfs_set_backup_num_devices(root_backup, 2079 btrfs_super_num_devices(info->super_copy)); 2080 2081 /* 2082 * if we don't copy this out to the super_copy, it won't get remembered 2083 * for the next commit 2084 */ 2085 memcpy(&info->super_copy->super_roots, 2086 &info->super_for_commit->super_roots, 2087 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2088 } 2089 2090 /* 2091 * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2092 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2093 * 2094 * fs_info - filesystem whose backup roots need to be read 2095 * priority - priority of backup root required 2096 * 2097 * Returns backup root index on success and -EINVAL otherwise. 2098 */ 2099 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2100 { 2101 int backup_index = find_newest_super_backup(fs_info); 2102 struct btrfs_super_block *super = fs_info->super_copy; 2103 struct btrfs_root_backup *root_backup; 2104 2105 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2106 if (priority == 0) 2107 return backup_index; 2108 2109 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2110 backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2111 } else { 2112 return -EINVAL; 2113 } 2114 2115 root_backup = super->super_roots + backup_index; 2116 2117 btrfs_set_super_generation(super, 2118 btrfs_backup_tree_root_gen(root_backup)); 2119 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2120 btrfs_set_super_root_level(super, 2121 btrfs_backup_tree_root_level(root_backup)); 2122 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2123 2124 /* 2125 * Fixme: the total bytes and num_devices need to match or we should 2126 * need a fsck 2127 */ 2128 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2129 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2130 2131 return backup_index; 2132 } 2133 2134 /* helper to cleanup workers */ 2135 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 2136 { 2137 btrfs_destroy_workqueue(fs_info->fixup_workers); 2138 btrfs_destroy_workqueue(fs_info->delalloc_workers); 2139 btrfs_destroy_workqueue(fs_info->workers); 2140 btrfs_destroy_workqueue(fs_info->endio_workers); 2141 btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2142 btrfs_destroy_workqueue(fs_info->rmw_workers); 2143 btrfs_destroy_workqueue(fs_info->endio_write_workers); 2144 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2145 btrfs_destroy_workqueue(fs_info->delayed_workers); 2146 btrfs_destroy_workqueue(fs_info->caching_workers); 2147 btrfs_destroy_workqueue(fs_info->readahead_workers); 2148 btrfs_destroy_workqueue(fs_info->flush_workers); 2149 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2150 if (fs_info->discard_ctl.discard_workers) 2151 destroy_workqueue(fs_info->discard_ctl.discard_workers); 2152 /* 2153 * Now that all other work queues are destroyed, we can safely destroy 2154 * the queues used for metadata I/O, since tasks from those other work 2155 * queues can do metadata I/O operations. 2156 */ 2157 btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2158 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2159 } 2160 2161 static void free_root_extent_buffers(struct btrfs_root *root) 2162 { 2163 if (root) { 2164 free_extent_buffer(root->node); 2165 free_extent_buffer(root->commit_root); 2166 root->node = NULL; 2167 root->commit_root = NULL; 2168 } 2169 } 2170 2171 /* helper to cleanup tree roots */ 2172 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2173 { 2174 free_root_extent_buffers(info->tree_root); 2175 2176 free_root_extent_buffers(info->dev_root); 2177 free_root_extent_buffers(info->extent_root); 2178 free_root_extent_buffers(info->csum_root); 2179 free_root_extent_buffers(info->quota_root); 2180 free_root_extent_buffers(info->uuid_root); 2181 free_root_extent_buffers(info->fs_root); 2182 free_root_extent_buffers(info->data_reloc_root); 2183 if (free_chunk_root) 2184 free_root_extent_buffers(info->chunk_root); 2185 free_root_extent_buffers(info->free_space_root); 2186 } 2187 2188 void btrfs_put_root(struct btrfs_root *root) 2189 { 2190 if (!root) 2191 return; 2192 2193 if (refcount_dec_and_test(&root->refs)) { 2194 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 2195 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 2196 if (root->anon_dev) 2197 free_anon_bdev(root->anon_dev); 2198 btrfs_drew_lock_destroy(&root->snapshot_lock); 2199 free_root_extent_buffers(root); 2200 #ifdef CONFIG_BTRFS_DEBUG 2201 spin_lock(&root->fs_info->fs_roots_radix_lock); 2202 list_del_init(&root->leak_list); 2203 spin_unlock(&root->fs_info->fs_roots_radix_lock); 2204 #endif 2205 kfree(root); 2206 } 2207 } 2208 2209 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2210 { 2211 int ret; 2212 struct btrfs_root *gang[8]; 2213 int i; 2214 2215 while (!list_empty(&fs_info->dead_roots)) { 2216 gang[0] = list_entry(fs_info->dead_roots.next, 2217 struct btrfs_root, root_list); 2218 list_del(&gang[0]->root_list); 2219 2220 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2221 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2222 btrfs_put_root(gang[0]); 2223 } 2224 2225 while (1) { 2226 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2227 (void **)gang, 0, 2228 ARRAY_SIZE(gang)); 2229 if (!ret) 2230 break; 2231 for (i = 0; i < ret; i++) 2232 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2233 } 2234 } 2235 2236 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2237 { 2238 mutex_init(&fs_info->scrub_lock); 2239 atomic_set(&fs_info->scrubs_running, 0); 2240 atomic_set(&fs_info->scrub_pause_req, 0); 2241 atomic_set(&fs_info->scrubs_paused, 0); 2242 atomic_set(&fs_info->scrub_cancel_req, 0); 2243 init_waitqueue_head(&fs_info->scrub_pause_wait); 2244 refcount_set(&fs_info->scrub_workers_refcnt, 0); 2245 } 2246 2247 static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2248 { 2249 spin_lock_init(&fs_info->balance_lock); 2250 mutex_init(&fs_info->balance_mutex); 2251 atomic_set(&fs_info->balance_pause_req, 0); 2252 atomic_set(&fs_info->balance_cancel_req, 0); 2253 fs_info->balance_ctl = NULL; 2254 init_waitqueue_head(&fs_info->balance_wait_q); 2255 } 2256 2257 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2258 { 2259 struct inode *inode = fs_info->btree_inode; 2260 2261 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2262 set_nlink(inode, 1); 2263 /* 2264 * we set the i_size on the btree inode to the max possible int. 2265 * the real end of the address space is determined by all of 2266 * the devices in the system 2267 */ 2268 inode->i_size = OFFSET_MAX; 2269 inode->i_mapping->a_ops = &btree_aops; 2270 2271 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 2272 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 2273 IO_TREE_BTREE_INODE_IO, inode); 2274 BTRFS_I(inode)->io_tree.track_uptodate = false; 2275 extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2276 2277 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2278 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 2279 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2280 btrfs_insert_inode_hash(inode); 2281 } 2282 2283 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2284 { 2285 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2286 init_rwsem(&fs_info->dev_replace.rwsem); 2287 init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2288 } 2289 2290 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2291 { 2292 spin_lock_init(&fs_info->qgroup_lock); 2293 mutex_init(&fs_info->qgroup_ioctl_lock); 2294 fs_info->qgroup_tree = RB_ROOT; 2295 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2296 fs_info->qgroup_seq = 1; 2297 fs_info->qgroup_ulist = NULL; 2298 fs_info->qgroup_rescan_running = false; 2299 mutex_init(&fs_info->qgroup_rescan_lock); 2300 } 2301 2302 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 2303 struct btrfs_fs_devices *fs_devices) 2304 { 2305 u32 max_active = fs_info->thread_pool_size; 2306 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 2307 2308 fs_info->workers = 2309 btrfs_alloc_workqueue(fs_info, "worker", 2310 flags | WQ_HIGHPRI, max_active, 16); 2311 2312 fs_info->delalloc_workers = 2313 btrfs_alloc_workqueue(fs_info, "delalloc", 2314 flags, max_active, 2); 2315 2316 fs_info->flush_workers = 2317 btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2318 flags, max_active, 0); 2319 2320 fs_info->caching_workers = 2321 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 2322 2323 fs_info->fixup_workers = 2324 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 2325 2326 /* 2327 * endios are largely parallel and should have a very 2328 * low idle thresh 2329 */ 2330 fs_info->endio_workers = 2331 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 2332 fs_info->endio_meta_workers = 2333 btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2334 max_active, 4); 2335 fs_info->endio_meta_write_workers = 2336 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2337 max_active, 2); 2338 fs_info->endio_raid56_workers = 2339 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2340 max_active, 4); 2341 fs_info->rmw_workers = 2342 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 2343 fs_info->endio_write_workers = 2344 btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2345 max_active, 2); 2346 fs_info->endio_freespace_worker = 2347 btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2348 max_active, 0); 2349 fs_info->delayed_workers = 2350 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2351 max_active, 0); 2352 fs_info->readahead_workers = 2353 btrfs_alloc_workqueue(fs_info, "readahead", flags, 2354 max_active, 2); 2355 fs_info->qgroup_rescan_workers = 2356 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2357 fs_info->discard_ctl.discard_workers = 2358 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 2359 2360 if (!(fs_info->workers && fs_info->delalloc_workers && 2361 fs_info->flush_workers && 2362 fs_info->endio_workers && fs_info->endio_meta_workers && 2363 fs_info->endio_meta_write_workers && 2364 fs_info->endio_write_workers && fs_info->endio_raid56_workers && 2365 fs_info->endio_freespace_worker && fs_info->rmw_workers && 2366 fs_info->caching_workers && fs_info->readahead_workers && 2367 fs_info->fixup_workers && fs_info->delayed_workers && 2368 fs_info->qgroup_rescan_workers && 2369 fs_info->discard_ctl.discard_workers)) { 2370 return -ENOMEM; 2371 } 2372 2373 return 0; 2374 } 2375 2376 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 2377 { 2378 struct crypto_shash *csum_shash; 2379 const char *csum_driver = btrfs_super_csum_driver(csum_type); 2380 2381 csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 2382 2383 if (IS_ERR(csum_shash)) { 2384 btrfs_err(fs_info, "error allocating %s hash for checksum", 2385 csum_driver); 2386 return PTR_ERR(csum_shash); 2387 } 2388 2389 fs_info->csum_shash = csum_shash; 2390 2391 return 0; 2392 } 2393 2394 static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 2395 struct btrfs_fs_devices *fs_devices) 2396 { 2397 int ret; 2398 struct btrfs_root *log_tree_root; 2399 struct btrfs_super_block *disk_super = fs_info->super_copy; 2400 u64 bytenr = btrfs_super_log_root(disk_super); 2401 int level = btrfs_super_log_root_level(disk_super); 2402 2403 if (fs_devices->rw_devices == 0) { 2404 btrfs_warn(fs_info, "log replay required on RO media"); 2405 return -EIO; 2406 } 2407 2408 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 2409 GFP_KERNEL); 2410 if (!log_tree_root) 2411 return -ENOMEM; 2412 2413 log_tree_root->node = read_tree_block(fs_info, bytenr, 2414 BTRFS_TREE_LOG_OBJECTID, 2415 fs_info->generation + 1, level, 2416 NULL); 2417 if (IS_ERR(log_tree_root->node)) { 2418 btrfs_warn(fs_info, "failed to read log tree"); 2419 ret = PTR_ERR(log_tree_root->node); 2420 log_tree_root->node = NULL; 2421 btrfs_put_root(log_tree_root); 2422 return ret; 2423 } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2424 btrfs_err(fs_info, "failed to read log tree"); 2425 btrfs_put_root(log_tree_root); 2426 return -EIO; 2427 } 2428 /* returns with log_tree_root freed on success */ 2429 ret = btrfs_recover_log_trees(log_tree_root); 2430 if (ret) { 2431 btrfs_handle_fs_error(fs_info, ret, 2432 "Failed to recover log tree"); 2433 btrfs_put_root(log_tree_root); 2434 return ret; 2435 } 2436 2437 if (sb_rdonly(fs_info->sb)) { 2438 ret = btrfs_commit_super(fs_info); 2439 if (ret) 2440 return ret; 2441 } 2442 2443 return 0; 2444 } 2445 2446 static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 2447 { 2448 struct btrfs_root *tree_root = fs_info->tree_root; 2449 struct btrfs_root *root; 2450 struct btrfs_key location; 2451 int ret; 2452 2453 BUG_ON(!fs_info->tree_root); 2454 2455 location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 2456 location.type = BTRFS_ROOT_ITEM_KEY; 2457 location.offset = 0; 2458 2459 root = btrfs_read_tree_root(tree_root, &location); 2460 if (IS_ERR(root)) { 2461 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2462 ret = PTR_ERR(root); 2463 goto out; 2464 } 2465 } else { 2466 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2467 fs_info->extent_root = root; 2468 } 2469 2470 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2471 root = btrfs_read_tree_root(tree_root, &location); 2472 if (IS_ERR(root)) { 2473 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2474 ret = PTR_ERR(root); 2475 goto out; 2476 } 2477 } else { 2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2479 fs_info->dev_root = root; 2480 } 2481 /* Initialize fs_info for all devices in any case */ 2482 btrfs_init_devices_late(fs_info); 2483 2484 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ 2485 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2486 location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2487 root = btrfs_read_tree_root(tree_root, &location); 2488 if (IS_ERR(root)) { 2489 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2490 ret = PTR_ERR(root); 2491 goto out; 2492 } 2493 } else { 2494 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2495 fs_info->csum_root = root; 2496 } 2497 } 2498 2499 /* 2500 * This tree can share blocks with some other fs tree during relocation 2501 * and we need a proper setup by btrfs_get_fs_root 2502 */ 2503 root = btrfs_get_fs_root(tree_root->fs_info, 2504 BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2505 if (IS_ERR(root)) { 2506 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2507 ret = PTR_ERR(root); 2508 goto out; 2509 } 2510 } else { 2511 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2512 fs_info->data_reloc_root = root; 2513 } 2514 2515 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2516 root = btrfs_read_tree_root(tree_root, &location); 2517 if (!IS_ERR(root)) { 2518 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2519 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2520 fs_info->quota_root = root; 2521 } 2522 2523 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2524 root = btrfs_read_tree_root(tree_root, &location); 2525 if (IS_ERR(root)) { 2526 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2527 ret = PTR_ERR(root); 2528 if (ret != -ENOENT) 2529 goto out; 2530 } 2531 } else { 2532 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2533 fs_info->uuid_root = root; 2534 } 2535 2536 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2537 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 2538 root = btrfs_read_tree_root(tree_root, &location); 2539 if (IS_ERR(root)) { 2540 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2541 ret = PTR_ERR(root); 2542 goto out; 2543 } 2544 } else { 2545 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2546 fs_info->free_space_root = root; 2547 } 2548 } 2549 2550 return 0; 2551 out: 2552 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2553 location.objectid, ret); 2554 return ret; 2555 } 2556 2557 /* 2558 * Real super block validation 2559 * NOTE: super csum type and incompat features will not be checked here. 2560 * 2561 * @sb: super block to check 2562 * @mirror_num: the super block number to check its bytenr: 2563 * 0 the primary (1st) sb 2564 * 1, 2 2nd and 3rd backup copy 2565 * -1 skip bytenr check 2566 */ 2567 static int validate_super(struct btrfs_fs_info *fs_info, 2568 struct btrfs_super_block *sb, int mirror_num) 2569 { 2570 u64 nodesize = btrfs_super_nodesize(sb); 2571 u64 sectorsize = btrfs_super_sectorsize(sb); 2572 int ret = 0; 2573 2574 if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 2575 btrfs_err(fs_info, "no valid FS found"); 2576 ret = -EINVAL; 2577 } 2578 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 2579 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 2580 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 2581 ret = -EINVAL; 2582 } 2583 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 2584 btrfs_err(fs_info, "tree_root level too big: %d >= %d", 2585 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 2586 ret = -EINVAL; 2587 } 2588 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 2589 btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 2590 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 2591 ret = -EINVAL; 2592 } 2593 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 2594 btrfs_err(fs_info, "log_root level too big: %d >= %d", 2595 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 2596 ret = -EINVAL; 2597 } 2598 2599 /* 2600 * Check sectorsize and nodesize first, other check will need it. 2601 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 2602 */ 2603 if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 2604 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2605 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 2606 ret = -EINVAL; 2607 } 2608 2609 /* 2610 * For 4K page size, we only support 4K sector size. 2611 * For 64K page size, we support read-write for 64K sector size, and 2612 * read-only for 4K sector size. 2613 */ 2614 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) || 2615 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K && 2616 sectorsize != SZ_64K))) { 2617 btrfs_err(fs_info, 2618 "sectorsize %llu not yet supported for page size %lu", 2619 sectorsize, PAGE_SIZE); 2620 ret = -EINVAL; 2621 } 2622 2623 if (!is_power_of_2(nodesize) || nodesize < sectorsize || 2624 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2625 btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 2626 ret = -EINVAL; 2627 } 2628 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 2629 btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 2630 le32_to_cpu(sb->__unused_leafsize), nodesize); 2631 ret = -EINVAL; 2632 } 2633 2634 /* Root alignment check */ 2635 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 2636 btrfs_warn(fs_info, "tree_root block unaligned: %llu", 2637 btrfs_super_root(sb)); 2638 ret = -EINVAL; 2639 } 2640 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 2641 btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 2642 btrfs_super_chunk_root(sb)); 2643 ret = -EINVAL; 2644 } 2645 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 2646 btrfs_warn(fs_info, "log_root block unaligned: %llu", 2647 btrfs_super_log_root(sb)); 2648 ret = -EINVAL; 2649 } 2650 2651 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 2652 BTRFS_FSID_SIZE) != 0) { 2653 btrfs_err(fs_info, 2654 "dev_item UUID does not match metadata fsid: %pU != %pU", 2655 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 2656 ret = -EINVAL; 2657 } 2658 2659 /* 2660 * Hint to catch really bogus numbers, bitflips or so, more exact checks are 2661 * done later 2662 */ 2663 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 2664 btrfs_err(fs_info, "bytes_used is too small %llu", 2665 btrfs_super_bytes_used(sb)); 2666 ret = -EINVAL; 2667 } 2668 if (!is_power_of_2(btrfs_super_stripesize(sb))) { 2669 btrfs_err(fs_info, "invalid stripesize %u", 2670 btrfs_super_stripesize(sb)); 2671 ret = -EINVAL; 2672 } 2673 if (btrfs_super_num_devices(sb) > (1UL << 31)) 2674 btrfs_warn(fs_info, "suspicious number of devices: %llu", 2675 btrfs_super_num_devices(sb)); 2676 if (btrfs_super_num_devices(sb) == 0) { 2677 btrfs_err(fs_info, "number of devices is 0"); 2678 ret = -EINVAL; 2679 } 2680 2681 if (mirror_num >= 0 && 2682 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 2683 btrfs_err(fs_info, "super offset mismatch %llu != %u", 2684 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 2685 ret = -EINVAL; 2686 } 2687 2688 /* 2689 * Obvious sys_chunk_array corruptions, it must hold at least one key 2690 * and one chunk 2691 */ 2692 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 2693 btrfs_err(fs_info, "system chunk array too big %u > %u", 2694 btrfs_super_sys_array_size(sb), 2695 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 2696 ret = -EINVAL; 2697 } 2698 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 2699 + sizeof(struct btrfs_chunk)) { 2700 btrfs_err(fs_info, "system chunk array too small %u < %zu", 2701 btrfs_super_sys_array_size(sb), 2702 sizeof(struct btrfs_disk_key) 2703 + sizeof(struct btrfs_chunk)); 2704 ret = -EINVAL; 2705 } 2706 2707 /* 2708 * The generation is a global counter, we'll trust it more than the others 2709 * but it's still possible that it's the one that's wrong. 2710 */ 2711 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 2712 btrfs_warn(fs_info, 2713 "suspicious: generation < chunk_root_generation: %llu < %llu", 2714 btrfs_super_generation(sb), 2715 btrfs_super_chunk_root_generation(sb)); 2716 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 2717 && btrfs_super_cache_generation(sb) != (u64)-1) 2718 btrfs_warn(fs_info, 2719 "suspicious: generation < cache_generation: %llu < %llu", 2720 btrfs_super_generation(sb), 2721 btrfs_super_cache_generation(sb)); 2722 2723 return ret; 2724 } 2725 2726 /* 2727 * Validation of super block at mount time. 2728 * Some checks already done early at mount time, like csum type and incompat 2729 * flags will be skipped. 2730 */ 2731 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2732 { 2733 return validate_super(fs_info, fs_info->super_copy, 0); 2734 } 2735 2736 /* 2737 * Validation of super block at write time. 2738 * Some checks like bytenr check will be skipped as their values will be 2739 * overwritten soon. 2740 * Extra checks like csum type and incompat flags will be done here. 2741 */ 2742 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 2743 struct btrfs_super_block *sb) 2744 { 2745 int ret; 2746 2747 ret = validate_super(fs_info, sb, -1); 2748 if (ret < 0) 2749 goto out; 2750 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 2751 ret = -EUCLEAN; 2752 btrfs_err(fs_info, "invalid csum type, has %u want %u", 2753 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 2754 goto out; 2755 } 2756 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 2757 ret = -EUCLEAN; 2758 btrfs_err(fs_info, 2759 "invalid incompat flags, has 0x%llx valid mask 0x%llx", 2760 btrfs_super_incompat_flags(sb), 2761 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 2762 goto out; 2763 } 2764 out: 2765 if (ret < 0) 2766 btrfs_err(fs_info, 2767 "super block corruption detected before writing it to disk"); 2768 return ret; 2769 } 2770 2771 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 2772 { 2773 int backup_index = find_newest_super_backup(fs_info); 2774 struct btrfs_super_block *sb = fs_info->super_copy; 2775 struct btrfs_root *tree_root = fs_info->tree_root; 2776 bool handle_error = false; 2777 int ret = 0; 2778 int i; 2779 2780 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2781 u64 generation; 2782 int level; 2783 2784 if (handle_error) { 2785 if (!IS_ERR(tree_root->node)) 2786 free_extent_buffer(tree_root->node); 2787 tree_root->node = NULL; 2788 2789 if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 2790 break; 2791 2792 free_root_pointers(fs_info, 0); 2793 2794 /* 2795 * Don't use the log in recovery mode, it won't be 2796 * valid 2797 */ 2798 btrfs_set_super_log_root(sb, 0); 2799 2800 /* We can't trust the free space cache either */ 2801 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 2802 2803 ret = read_backup_root(fs_info, i); 2804 backup_index = ret; 2805 if (ret < 0) 2806 return ret; 2807 } 2808 generation = btrfs_super_generation(sb); 2809 level = btrfs_super_root_level(sb); 2810 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), 2811 BTRFS_ROOT_TREE_OBJECTID, 2812 generation, level, NULL); 2813 if (IS_ERR(tree_root->node)) { 2814 handle_error = true; 2815 ret = PTR_ERR(tree_root->node); 2816 tree_root->node = NULL; 2817 btrfs_warn(fs_info, "couldn't read tree root"); 2818 continue; 2819 2820 } else if (!extent_buffer_uptodate(tree_root->node)) { 2821 handle_error = true; 2822 ret = -EIO; 2823 btrfs_warn(fs_info, "error while reading tree root"); 2824 continue; 2825 } 2826 2827 btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2828 tree_root->commit_root = btrfs_root_node(tree_root); 2829 btrfs_set_root_refs(&tree_root->root_item, 1); 2830 2831 /* 2832 * No need to hold btrfs_root::objectid_mutex since the fs 2833 * hasn't been fully initialised and we are the only user 2834 */ 2835 ret = btrfs_init_root_free_objectid(tree_root); 2836 if (ret < 0) { 2837 handle_error = true; 2838 continue; 2839 } 2840 2841 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 2842 2843 ret = btrfs_read_roots(fs_info); 2844 if (ret < 0) { 2845 handle_error = true; 2846 continue; 2847 } 2848 2849 /* All successful */ 2850 fs_info->generation = generation; 2851 fs_info->last_trans_committed = generation; 2852 2853 /* Always begin writing backup roots after the one being used */ 2854 if (backup_index < 0) { 2855 fs_info->backup_root_index = 0; 2856 } else { 2857 fs_info->backup_root_index = backup_index + 1; 2858 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 2859 } 2860 break; 2861 } 2862 2863 return ret; 2864 } 2865 2866 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 2867 { 2868 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2869 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 2870 INIT_LIST_HEAD(&fs_info->trans_list); 2871 INIT_LIST_HEAD(&fs_info->dead_roots); 2872 INIT_LIST_HEAD(&fs_info->delayed_iputs); 2873 INIT_LIST_HEAD(&fs_info->delalloc_roots); 2874 INIT_LIST_HEAD(&fs_info->caching_block_groups); 2875 spin_lock_init(&fs_info->delalloc_root_lock); 2876 spin_lock_init(&fs_info->trans_lock); 2877 spin_lock_init(&fs_info->fs_roots_radix_lock); 2878 spin_lock_init(&fs_info->delayed_iput_lock); 2879 spin_lock_init(&fs_info->defrag_inodes_lock); 2880 spin_lock_init(&fs_info->super_lock); 2881 spin_lock_init(&fs_info->buffer_lock); 2882 spin_lock_init(&fs_info->unused_bgs_lock); 2883 spin_lock_init(&fs_info->treelog_bg_lock); 2884 rwlock_init(&fs_info->tree_mod_log_lock); 2885 mutex_init(&fs_info->unused_bg_unpin_mutex); 2886 mutex_init(&fs_info->reclaim_bgs_lock); 2887 mutex_init(&fs_info->reloc_mutex); 2888 mutex_init(&fs_info->delalloc_root_mutex); 2889 mutex_init(&fs_info->zoned_meta_io_lock); 2890 seqlock_init(&fs_info->profiles_lock); 2891 2892 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 2893 INIT_LIST_HEAD(&fs_info->space_info); 2894 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 2895 INIT_LIST_HEAD(&fs_info->unused_bgs); 2896 INIT_LIST_HEAD(&fs_info->reclaim_bgs); 2897 #ifdef CONFIG_BTRFS_DEBUG 2898 INIT_LIST_HEAD(&fs_info->allocated_roots); 2899 INIT_LIST_HEAD(&fs_info->allocated_ebs); 2900 spin_lock_init(&fs_info->eb_leak_lock); 2901 #endif 2902 extent_map_tree_init(&fs_info->mapping_tree); 2903 btrfs_init_block_rsv(&fs_info->global_block_rsv, 2904 BTRFS_BLOCK_RSV_GLOBAL); 2905 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 2906 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 2907 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 2908 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 2909 BTRFS_BLOCK_RSV_DELOPS); 2910 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 2911 BTRFS_BLOCK_RSV_DELREFS); 2912 2913 atomic_set(&fs_info->async_delalloc_pages, 0); 2914 atomic_set(&fs_info->defrag_running, 0); 2915 atomic_set(&fs_info->reada_works_cnt, 0); 2916 atomic_set(&fs_info->nr_delayed_iputs, 0); 2917 atomic64_set(&fs_info->tree_mod_seq, 0); 2918 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 2919 fs_info->metadata_ratio = 0; 2920 fs_info->defrag_inodes = RB_ROOT; 2921 atomic64_set(&fs_info->free_chunk_space, 0); 2922 fs_info->tree_mod_log = RB_ROOT; 2923 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2924 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 2925 /* readahead state */ 2926 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 2927 spin_lock_init(&fs_info->reada_lock); 2928 btrfs_init_ref_verify(fs_info); 2929 2930 fs_info->thread_pool_size = min_t(unsigned long, 2931 num_online_cpus() + 2, 8); 2932 2933 INIT_LIST_HEAD(&fs_info->ordered_roots); 2934 spin_lock_init(&fs_info->ordered_root_lock); 2935 2936 btrfs_init_scrub(fs_info); 2937 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2938 fs_info->check_integrity_print_mask = 0; 2939 #endif 2940 btrfs_init_balance(fs_info); 2941 btrfs_init_async_reclaim_work(fs_info); 2942 2943 spin_lock_init(&fs_info->block_group_cache_lock); 2944 fs_info->block_group_cache_tree = RB_ROOT; 2945 fs_info->first_logical_byte = (u64)-1; 2946 2947 extent_io_tree_init(fs_info, &fs_info->excluded_extents, 2948 IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 2949 set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2950 2951 mutex_init(&fs_info->ordered_operations_mutex); 2952 mutex_init(&fs_info->tree_log_mutex); 2953 mutex_init(&fs_info->chunk_mutex); 2954 mutex_init(&fs_info->transaction_kthread_mutex); 2955 mutex_init(&fs_info->cleaner_mutex); 2956 mutex_init(&fs_info->ro_block_group_mutex); 2957 init_rwsem(&fs_info->commit_root_sem); 2958 init_rwsem(&fs_info->cleanup_work_sem); 2959 init_rwsem(&fs_info->subvol_sem); 2960 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2961 2962 btrfs_init_dev_replace_locks(fs_info); 2963 btrfs_init_qgroup(fs_info); 2964 btrfs_discard_init(fs_info); 2965 2966 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2967 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2968 2969 init_waitqueue_head(&fs_info->transaction_throttle); 2970 init_waitqueue_head(&fs_info->transaction_wait); 2971 init_waitqueue_head(&fs_info->transaction_blocked_wait); 2972 init_waitqueue_head(&fs_info->async_submit_wait); 2973 init_waitqueue_head(&fs_info->delayed_iputs_wait); 2974 2975 /* Usable values until the real ones are cached from the superblock */ 2976 fs_info->nodesize = 4096; 2977 fs_info->sectorsize = 4096; 2978 fs_info->sectorsize_bits = ilog2(4096); 2979 fs_info->stripesize = 4096; 2980 2981 spin_lock_init(&fs_info->swapfile_pins_lock); 2982 fs_info->swapfile_pins = RB_ROOT; 2983 2984 fs_info->send_in_progress = 0; 2985 2986 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 2987 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 2988 } 2989 2990 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 2991 { 2992 int ret; 2993 2994 fs_info->sb = sb; 2995 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 2996 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2997 2998 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 2999 if (ret) 3000 return ret; 3001 3002 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3003 if (ret) 3004 return ret; 3005 3006 fs_info->dirty_metadata_batch = PAGE_SIZE * 3007 (1 + ilog2(nr_cpu_ids)); 3008 3009 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3010 if (ret) 3011 return ret; 3012 3013 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3014 GFP_KERNEL); 3015 if (ret) 3016 return ret; 3017 3018 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3019 GFP_KERNEL); 3020 if (!fs_info->delayed_root) 3021 return -ENOMEM; 3022 btrfs_init_delayed_root(fs_info->delayed_root); 3023 3024 if (sb_rdonly(sb)) 3025 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3026 3027 return btrfs_alloc_stripe_hash_table(fs_info); 3028 } 3029 3030 static int btrfs_uuid_rescan_kthread(void *data) 3031 { 3032 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 3033 int ret; 3034 3035 /* 3036 * 1st step is to iterate through the existing UUID tree and 3037 * to delete all entries that contain outdated data. 3038 * 2nd step is to add all missing entries to the UUID tree. 3039 */ 3040 ret = btrfs_uuid_tree_iterate(fs_info); 3041 if (ret < 0) { 3042 if (ret != -EINTR) 3043 btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3044 ret); 3045 up(&fs_info->uuid_tree_rescan_sem); 3046 return ret; 3047 } 3048 return btrfs_uuid_scan_kthread(data); 3049 } 3050 3051 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 3052 { 3053 struct task_struct *task; 3054 3055 down(&fs_info->uuid_tree_rescan_sem); 3056 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 3057 if (IS_ERR(task)) { 3058 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 3059 btrfs_warn(fs_info, "failed to start uuid_rescan task"); 3060 up(&fs_info->uuid_tree_rescan_sem); 3061 return PTR_ERR(task); 3062 } 3063 3064 return 0; 3065 } 3066 3067 /* 3068 * Some options only have meaning at mount time and shouldn't persist across 3069 * remounts, or be displayed. Clear these at the end of mount and remount 3070 * code paths. 3071 */ 3072 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 3073 { 3074 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 3075 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 3076 } 3077 3078 /* 3079 * Mounting logic specific to read-write file systems. Shared by open_ctree 3080 * and btrfs_remount when remounting from read-only to read-write. 3081 */ 3082 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 3083 { 3084 int ret; 3085 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 3086 bool clear_free_space_tree = false; 3087 3088 if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3089 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 3090 clear_free_space_tree = true; 3091 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 3092 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 3093 btrfs_warn(fs_info, "free space tree is invalid"); 3094 clear_free_space_tree = true; 3095 } 3096 3097 if (clear_free_space_tree) { 3098 btrfs_info(fs_info, "clearing free space tree"); 3099 ret = btrfs_clear_free_space_tree(fs_info); 3100 if (ret) { 3101 btrfs_warn(fs_info, 3102 "failed to clear free space tree: %d", ret); 3103 goto out; 3104 } 3105 } 3106 3107 /* 3108 * btrfs_find_orphan_roots() is responsible for finding all the dead 3109 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3110 * them into the fs_info->fs_roots_radix tree. This must be done before 3111 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 3112 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 3113 * item before the root's tree is deleted - this means that if we unmount 3114 * or crash before the deletion completes, on the next mount we will not 3115 * delete what remains of the tree because the orphan item does not 3116 * exists anymore, which is what tells us we have a pending deletion. 3117 */ 3118 ret = btrfs_find_orphan_roots(fs_info); 3119 if (ret) 3120 goto out; 3121 3122 ret = btrfs_cleanup_fs_roots(fs_info); 3123 if (ret) 3124 goto out; 3125 3126 down_read(&fs_info->cleanup_work_sem); 3127 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 3128 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 3129 up_read(&fs_info->cleanup_work_sem); 3130 goto out; 3131 } 3132 up_read(&fs_info->cleanup_work_sem); 3133 3134 mutex_lock(&fs_info->cleaner_mutex); 3135 ret = btrfs_recover_relocation(fs_info->tree_root); 3136 mutex_unlock(&fs_info->cleaner_mutex); 3137 if (ret < 0) { 3138 btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 3139 goto out; 3140 } 3141 3142 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3143 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 3144 btrfs_info(fs_info, "creating free space tree"); 3145 ret = btrfs_create_free_space_tree(fs_info); 3146 if (ret) { 3147 btrfs_warn(fs_info, 3148 "failed to create free space tree: %d", ret); 3149 goto out; 3150 } 3151 } 3152 3153 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 3154 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 3155 if (ret) 3156 goto out; 3157 } 3158 3159 ret = btrfs_resume_balance_async(fs_info); 3160 if (ret) 3161 goto out; 3162 3163 ret = btrfs_resume_dev_replace_async(fs_info); 3164 if (ret) { 3165 btrfs_warn(fs_info, "failed to resume dev_replace"); 3166 goto out; 3167 } 3168 3169 btrfs_qgroup_rescan_resume(fs_info); 3170 3171 if (!fs_info->uuid_root) { 3172 btrfs_info(fs_info, "creating UUID tree"); 3173 ret = btrfs_create_uuid_tree(fs_info); 3174 if (ret) { 3175 btrfs_warn(fs_info, 3176 "failed to create the UUID tree %d", ret); 3177 goto out; 3178 } 3179 } 3180 3181 out: 3182 return ret; 3183 } 3184 3185 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3186 char *options) 3187 { 3188 u32 sectorsize; 3189 u32 nodesize; 3190 u32 stripesize; 3191 u64 generation; 3192 u64 features; 3193 u16 csum_type; 3194 struct btrfs_super_block *disk_super; 3195 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3196 struct btrfs_root *tree_root; 3197 struct btrfs_root *chunk_root; 3198 int ret; 3199 int err = -EINVAL; 3200 int level; 3201 3202 ret = init_mount_fs_info(fs_info, sb); 3203 if (ret) { 3204 err = ret; 3205 goto fail; 3206 } 3207 3208 /* These need to be init'ed before we start creating inodes and such. */ 3209 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3210 GFP_KERNEL); 3211 fs_info->tree_root = tree_root; 3212 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3213 GFP_KERNEL); 3214 fs_info->chunk_root = chunk_root; 3215 if (!tree_root || !chunk_root) { 3216 err = -ENOMEM; 3217 goto fail; 3218 } 3219 3220 fs_info->btree_inode = new_inode(sb); 3221 if (!fs_info->btree_inode) { 3222 err = -ENOMEM; 3223 goto fail; 3224 } 3225 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3226 btrfs_init_btree_inode(fs_info); 3227 3228 invalidate_bdev(fs_devices->latest_bdev); 3229 3230 /* 3231 * Read super block and check the signature bytes only 3232 */ 3233 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); 3234 if (IS_ERR(disk_super)) { 3235 err = PTR_ERR(disk_super); 3236 goto fail_alloc; 3237 } 3238 3239 /* 3240 * Verify the type first, if that or the checksum value are 3241 * corrupted, we'll find out 3242 */ 3243 csum_type = btrfs_super_csum_type(disk_super); 3244 if (!btrfs_supported_super_csum(csum_type)) { 3245 btrfs_err(fs_info, "unsupported checksum algorithm: %u", 3246 csum_type); 3247 err = -EINVAL; 3248 btrfs_release_disk_super(disk_super); 3249 goto fail_alloc; 3250 } 3251 3252 fs_info->csum_size = btrfs_super_csum_size(disk_super); 3253 3254 ret = btrfs_init_csum_hash(fs_info, csum_type); 3255 if (ret) { 3256 err = ret; 3257 btrfs_release_disk_super(disk_super); 3258 goto fail_alloc; 3259 } 3260 3261 /* 3262 * We want to check superblock checksum, the type is stored inside. 3263 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 3264 */ 3265 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 3266 btrfs_err(fs_info, "superblock checksum mismatch"); 3267 err = -EINVAL; 3268 btrfs_release_disk_super(disk_super); 3269 goto fail_alloc; 3270 } 3271 3272 /* 3273 * super_copy is zeroed at allocation time and we never touch the 3274 * following bytes up to INFO_SIZE, the checksum is calculated from 3275 * the whole block of INFO_SIZE 3276 */ 3277 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 3278 btrfs_release_disk_super(disk_super); 3279 3280 disk_super = fs_info->super_copy; 3281 3282 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 3283 BTRFS_FSID_SIZE)); 3284 3285 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { 3286 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, 3287 fs_info->super_copy->metadata_uuid, 3288 BTRFS_FSID_SIZE)); 3289 } 3290 3291 features = btrfs_super_flags(disk_super); 3292 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3293 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3294 btrfs_set_super_flags(disk_super, features); 3295 btrfs_info(fs_info, 3296 "found metadata UUID change in progress flag, clearing"); 3297 } 3298 3299 memcpy(fs_info->super_for_commit, fs_info->super_copy, 3300 sizeof(*fs_info->super_for_commit)); 3301 3302 ret = btrfs_validate_mount_super(fs_info); 3303 if (ret) { 3304 btrfs_err(fs_info, "superblock contains fatal errors"); 3305 err = -EINVAL; 3306 goto fail_alloc; 3307 } 3308 3309 if (!btrfs_super_root(disk_super)) 3310 goto fail_alloc; 3311 3312 /* check FS state, whether FS is broken. */ 3313 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 3314 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3315 3316 /* 3317 * In the long term, we'll store the compression type in the super 3318 * block, and it'll be used for per file compression control. 3319 */ 3320 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 3321 3322 ret = btrfs_parse_options(fs_info, options, sb->s_flags); 3323 if (ret) { 3324 err = ret; 3325 goto fail_alloc; 3326 } 3327 3328 features = btrfs_super_incompat_flags(disk_super) & 3329 ~BTRFS_FEATURE_INCOMPAT_SUPP; 3330 if (features) { 3331 btrfs_err(fs_info, 3332 "cannot mount because of unsupported optional features (%llx)", 3333 features); 3334 err = -EINVAL; 3335 goto fail_alloc; 3336 } 3337 3338 features = btrfs_super_incompat_flags(disk_super); 3339 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3340 if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3341 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3342 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3343 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3344 3345 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 3346 btrfs_info(fs_info, "has skinny extents"); 3347 3348 /* 3349 * flag our filesystem as having big metadata blocks if 3350 * they are bigger than the page size 3351 */ 3352 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3353 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 3354 btrfs_info(fs_info, 3355 "flagging fs with big metadata feature"); 3356 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3357 } 3358 3359 nodesize = btrfs_super_nodesize(disk_super); 3360 sectorsize = btrfs_super_sectorsize(disk_super); 3361 stripesize = sectorsize; 3362 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3363 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3364 3365 /* Cache block sizes */ 3366 fs_info->nodesize = nodesize; 3367 fs_info->sectorsize = sectorsize; 3368 fs_info->sectorsize_bits = ilog2(sectorsize); 3369 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 3370 fs_info->stripesize = stripesize; 3371 3372 /* 3373 * mixed block groups end up with duplicate but slightly offset 3374 * extent buffers for the same range. It leads to corruptions 3375 */ 3376 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3377 (sectorsize != nodesize)) { 3378 btrfs_err(fs_info, 3379 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3380 nodesize, sectorsize); 3381 goto fail_alloc; 3382 } 3383 3384 /* 3385 * Needn't use the lock because there is no other task which will 3386 * update the flag. 3387 */ 3388 btrfs_set_super_incompat_flags(disk_super, features); 3389 3390 features = btrfs_super_compat_ro_flags(disk_super) & 3391 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3392 if (!sb_rdonly(sb) && features) { 3393 btrfs_err(fs_info, 3394 "cannot mount read-write because of unsupported optional features (%llx)", 3395 features); 3396 err = -EINVAL; 3397 goto fail_alloc; 3398 } 3399 3400 /* For 4K sector size support, it's only read-only */ 3401 if (PAGE_SIZE == SZ_64K && sectorsize == SZ_4K) { 3402 if (!sb_rdonly(sb) || btrfs_super_log_root(disk_super)) { 3403 btrfs_err(fs_info, 3404 "subpage sectorsize %u only supported read-only for page size %lu", 3405 sectorsize, PAGE_SIZE); 3406 err = -EINVAL; 3407 goto fail_alloc; 3408 } 3409 } 3410 3411 ret = btrfs_init_workqueues(fs_info, fs_devices); 3412 if (ret) { 3413 err = ret; 3414 goto fail_sb_buffer; 3415 } 3416 3417 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 3418 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 3419 3420 sb->s_blocksize = sectorsize; 3421 sb->s_blocksize_bits = blksize_bits(sectorsize); 3422 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3423 3424 mutex_lock(&fs_info->chunk_mutex); 3425 ret = btrfs_read_sys_array(fs_info); 3426 mutex_unlock(&fs_info->chunk_mutex); 3427 if (ret) { 3428 btrfs_err(fs_info, "failed to read the system array: %d", ret); 3429 goto fail_sb_buffer; 3430 } 3431 3432 generation = btrfs_super_chunk_root_generation(disk_super); 3433 level = btrfs_super_chunk_root_level(disk_super); 3434 3435 chunk_root->node = read_tree_block(fs_info, 3436 btrfs_super_chunk_root(disk_super), 3437 BTRFS_CHUNK_TREE_OBJECTID, 3438 generation, level, NULL); 3439 if (IS_ERR(chunk_root->node) || 3440 !extent_buffer_uptodate(chunk_root->node)) { 3441 btrfs_err(fs_info, "failed to read chunk root"); 3442 if (!IS_ERR(chunk_root->node)) 3443 free_extent_buffer(chunk_root->node); 3444 chunk_root->node = NULL; 3445 goto fail_tree_roots; 3446 } 3447 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 3448 chunk_root->commit_root = btrfs_root_node(chunk_root); 3449 3450 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3451 offsetof(struct btrfs_header, chunk_tree_uuid), 3452 BTRFS_UUID_SIZE); 3453 3454 ret = btrfs_read_chunk_tree(fs_info); 3455 if (ret) { 3456 btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3457 goto fail_tree_roots; 3458 } 3459 3460 /* 3461 * At this point we know all the devices that make this filesystem, 3462 * including the seed devices but we don't know yet if the replace 3463 * target is required. So free devices that are not part of this 3464 * filesystem but skip the replace traget device which is checked 3465 * below in btrfs_init_dev_replace(). 3466 */ 3467 btrfs_free_extra_devids(fs_devices); 3468 if (!fs_devices->latest_bdev) { 3469 btrfs_err(fs_info, "failed to read devices"); 3470 goto fail_tree_roots; 3471 } 3472 3473 ret = init_tree_roots(fs_info); 3474 if (ret) 3475 goto fail_tree_roots; 3476 3477 /* 3478 * Get zone type information of zoned block devices. This will also 3479 * handle emulation of a zoned filesystem if a regular device has the 3480 * zoned incompat feature flag set. 3481 */ 3482 ret = btrfs_get_dev_zone_info_all_devices(fs_info); 3483 if (ret) { 3484 btrfs_err(fs_info, 3485 "zoned: failed to read device zone info: %d", 3486 ret); 3487 goto fail_block_groups; 3488 } 3489 3490 /* 3491 * If we have a uuid root and we're not being told to rescan we need to 3492 * check the generation here so we can set the 3493 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 3494 * transaction during a balance or the log replay without updating the 3495 * uuid generation, and then if we crash we would rescan the uuid tree, 3496 * even though it was perfectly fine. 3497 */ 3498 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 3499 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 3500 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3501 3502 ret = btrfs_verify_dev_extents(fs_info); 3503 if (ret) { 3504 btrfs_err(fs_info, 3505 "failed to verify dev extents against chunks: %d", 3506 ret); 3507 goto fail_block_groups; 3508 } 3509 ret = btrfs_recover_balance(fs_info); 3510 if (ret) { 3511 btrfs_err(fs_info, "failed to recover balance: %d", ret); 3512 goto fail_block_groups; 3513 } 3514 3515 ret = btrfs_init_dev_stats(fs_info); 3516 if (ret) { 3517 btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3518 goto fail_block_groups; 3519 } 3520 3521 ret = btrfs_init_dev_replace(fs_info); 3522 if (ret) { 3523 btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 3524 goto fail_block_groups; 3525 } 3526 3527 ret = btrfs_check_zoned_mode(fs_info); 3528 if (ret) { 3529 btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3530 ret); 3531 goto fail_block_groups; 3532 } 3533 3534 ret = btrfs_sysfs_add_fsid(fs_devices); 3535 if (ret) { 3536 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 3537 ret); 3538 goto fail_block_groups; 3539 } 3540 3541 ret = btrfs_sysfs_add_mounted(fs_info); 3542 if (ret) { 3543 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3544 goto fail_fsdev_sysfs; 3545 } 3546 3547 ret = btrfs_init_space_info(fs_info); 3548 if (ret) { 3549 btrfs_err(fs_info, "failed to initialize space info: %d", ret); 3550 goto fail_sysfs; 3551 } 3552 3553 ret = btrfs_read_block_groups(fs_info); 3554 if (ret) { 3555 btrfs_err(fs_info, "failed to read block groups: %d", ret); 3556 goto fail_sysfs; 3557 } 3558 3559 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 3560 btrfs_warn(fs_info, 3561 "writable mount is not allowed due to too many missing devices"); 3562 goto fail_sysfs; 3563 } 3564 3565 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 3566 "btrfs-cleaner"); 3567 if (IS_ERR(fs_info->cleaner_kthread)) 3568 goto fail_sysfs; 3569 3570 fs_info->transaction_kthread = kthread_run(transaction_kthread, 3571 tree_root, 3572 "btrfs-transaction"); 3573 if (IS_ERR(fs_info->transaction_kthread)) 3574 goto fail_cleaner; 3575 3576 if (!btrfs_test_opt(fs_info, NOSSD) && 3577 !fs_info->fs_devices->rotating) { 3578 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3579 } 3580 3581 /* 3582 * Mount does not set all options immediately, we can do it now and do 3583 * not have to wait for transaction commit 3584 */ 3585 btrfs_apply_pending_changes(fs_info); 3586 3587 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3588 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 3589 ret = btrfsic_mount(fs_info, fs_devices, 3590 btrfs_test_opt(fs_info, 3591 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 3592 1 : 0, 3593 fs_info->check_integrity_print_mask); 3594 if (ret) 3595 btrfs_warn(fs_info, 3596 "failed to initialize integrity check module: %d", 3597 ret); 3598 } 3599 #endif 3600 ret = btrfs_read_qgroup_config(fs_info); 3601 if (ret) 3602 goto fail_trans_kthread; 3603 3604 if (btrfs_build_ref_tree(fs_info)) 3605 btrfs_err(fs_info, "couldn't build ref tree"); 3606 3607 /* do not make disk changes in broken FS or nologreplay is given */ 3608 if (btrfs_super_log_root(disk_super) != 0 && 3609 !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3610 btrfs_info(fs_info, "start tree-log replay"); 3611 ret = btrfs_replay_log(fs_info, fs_devices); 3612 if (ret) { 3613 err = ret; 3614 goto fail_qgroup; 3615 } 3616 } 3617 3618 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 3619 if (IS_ERR(fs_info->fs_root)) { 3620 err = PTR_ERR(fs_info->fs_root); 3621 btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3622 fs_info->fs_root = NULL; 3623 goto fail_qgroup; 3624 } 3625 3626 if (sb_rdonly(sb)) 3627 goto clear_oneshot; 3628 3629 ret = btrfs_start_pre_rw_mount(fs_info); 3630 if (ret) { 3631 close_ctree(fs_info); 3632 return ret; 3633 } 3634 btrfs_discard_resume(fs_info); 3635 3636 if (fs_info->uuid_root && 3637 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 3638 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 3639 btrfs_info(fs_info, "checking UUID tree"); 3640 ret = btrfs_check_uuid_tree(fs_info); 3641 if (ret) { 3642 btrfs_warn(fs_info, 3643 "failed to check the UUID tree: %d", ret); 3644 close_ctree(fs_info); 3645 return ret; 3646 } 3647 } 3648 3649 set_bit(BTRFS_FS_OPEN, &fs_info->flags); 3650 3651 clear_oneshot: 3652 btrfs_clear_oneshot_options(fs_info); 3653 return 0; 3654 3655 fail_qgroup: 3656 btrfs_free_qgroup_config(fs_info); 3657 fail_trans_kthread: 3658 kthread_stop(fs_info->transaction_kthread); 3659 btrfs_cleanup_transaction(fs_info); 3660 btrfs_free_fs_roots(fs_info); 3661 fail_cleaner: 3662 kthread_stop(fs_info->cleaner_kthread); 3663 3664 /* 3665 * make sure we're done with the btree inode before we stop our 3666 * kthreads 3667 */ 3668 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 3669 3670 fail_sysfs: 3671 btrfs_sysfs_remove_mounted(fs_info); 3672 3673 fail_fsdev_sysfs: 3674 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3675 3676 fail_block_groups: 3677 btrfs_put_block_group_cache(fs_info); 3678 3679 fail_tree_roots: 3680 if (fs_info->data_reloc_root) 3681 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 3682 free_root_pointers(fs_info, true); 3683 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3684 3685 fail_sb_buffer: 3686 btrfs_stop_all_workers(fs_info); 3687 btrfs_free_block_groups(fs_info); 3688 fail_alloc: 3689 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3690 3691 iput(fs_info->btree_inode); 3692 fail: 3693 btrfs_close_devices(fs_info->fs_devices); 3694 return err; 3695 } 3696 ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3697 3698 static void btrfs_end_super_write(struct bio *bio) 3699 { 3700 struct btrfs_device *device = bio->bi_private; 3701 struct bio_vec *bvec; 3702 struct bvec_iter_all iter_all; 3703 struct page *page; 3704 3705 bio_for_each_segment_all(bvec, bio, iter_all) { 3706 page = bvec->bv_page; 3707 3708 if (bio->bi_status) { 3709 btrfs_warn_rl_in_rcu(device->fs_info, 3710 "lost page write due to IO error on %s (%d)", 3711 rcu_str_deref(device->name), 3712 blk_status_to_errno(bio->bi_status)); 3713 ClearPageUptodate(page); 3714 SetPageError(page); 3715 btrfs_dev_stat_inc_and_print(device, 3716 BTRFS_DEV_STAT_WRITE_ERRS); 3717 } else { 3718 SetPageUptodate(page); 3719 } 3720 3721 put_page(page); 3722 unlock_page(page); 3723 } 3724 3725 bio_put(bio); 3726 } 3727 3728 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3729 int copy_num) 3730 { 3731 struct btrfs_super_block *super; 3732 struct page *page; 3733 u64 bytenr, bytenr_orig; 3734 struct address_space *mapping = bdev->bd_inode->i_mapping; 3735 int ret; 3736 3737 bytenr_orig = btrfs_sb_offset(copy_num); 3738 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 3739 if (ret == -ENOENT) 3740 return ERR_PTR(-EINVAL); 3741 else if (ret) 3742 return ERR_PTR(ret); 3743 3744 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 3745 return ERR_PTR(-EINVAL); 3746 3747 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 3748 if (IS_ERR(page)) 3749 return ERR_CAST(page); 3750 3751 super = page_address(page); 3752 if (btrfs_super_magic(super) != BTRFS_MAGIC) { 3753 btrfs_release_disk_super(super); 3754 return ERR_PTR(-ENODATA); 3755 } 3756 3757 if (btrfs_super_bytenr(super) != bytenr_orig) { 3758 btrfs_release_disk_super(super); 3759 return ERR_PTR(-EINVAL); 3760 } 3761 3762 return super; 3763 } 3764 3765 3766 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 3767 { 3768 struct btrfs_super_block *super, *latest = NULL; 3769 int i; 3770 u64 transid = 0; 3771 3772 /* we would like to check all the supers, but that would make 3773 * a btrfs mount succeed after a mkfs from a different FS. 3774 * So, we need to add a special mount option to scan for 3775 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3776 */ 3777 for (i = 0; i < 1; i++) { 3778 super = btrfs_read_dev_one_super(bdev, i); 3779 if (IS_ERR(super)) 3780 continue; 3781 3782 if (!latest || btrfs_super_generation(super) > transid) { 3783 if (latest) 3784 btrfs_release_disk_super(super); 3785 3786 latest = super; 3787 transid = btrfs_super_generation(super); 3788 } 3789 } 3790 3791 return super; 3792 } 3793 3794 /* 3795 * Write superblock @sb to the @device. Do not wait for completion, all the 3796 * pages we use for writing are locked. 3797 * 3798 * Write @max_mirrors copies of the superblock, where 0 means default that fit 3799 * the expected device size at commit time. Note that max_mirrors must be 3800 * same for write and wait phases. 3801 * 3802 * Return number of errors when page is not found or submission fails. 3803 */ 3804 static int write_dev_supers(struct btrfs_device *device, 3805 struct btrfs_super_block *sb, int max_mirrors) 3806 { 3807 struct btrfs_fs_info *fs_info = device->fs_info; 3808 struct address_space *mapping = device->bdev->bd_inode->i_mapping; 3809 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 3810 int i; 3811 int errors = 0; 3812 int ret; 3813 u64 bytenr, bytenr_orig; 3814 3815 if (max_mirrors == 0) 3816 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3817 3818 shash->tfm = fs_info->csum_shash; 3819 3820 for (i = 0; i < max_mirrors; i++) { 3821 struct page *page; 3822 struct bio *bio; 3823 struct btrfs_super_block *disk_super; 3824 3825 bytenr_orig = btrfs_sb_offset(i); 3826 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 3827 if (ret == -ENOENT) { 3828 continue; 3829 } else if (ret < 0) { 3830 btrfs_err(device->fs_info, 3831 "couldn't get super block location for mirror %d", 3832 i); 3833 errors++; 3834 continue; 3835 } 3836 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3837 device->commit_total_bytes) 3838 break; 3839 3840 btrfs_set_super_bytenr(sb, bytenr_orig); 3841 3842 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 3843 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 3844 sb->csum); 3845 3846 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 3847 GFP_NOFS); 3848 if (!page) { 3849 btrfs_err(device->fs_info, 3850 "couldn't get super block page for bytenr %llu", 3851 bytenr); 3852 errors++; 3853 continue; 3854 } 3855 3856 /* Bump the refcount for wait_dev_supers() */ 3857 get_page(page); 3858 3859 disk_super = page_address(page); 3860 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 3861 3862 /* 3863 * Directly use bios here instead of relying on the page cache 3864 * to do I/O, so we don't lose the ability to do integrity 3865 * checking. 3866 */ 3867 bio = bio_alloc(GFP_NOFS, 1); 3868 bio_set_dev(bio, device->bdev); 3869 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 3870 bio->bi_private = device; 3871 bio->bi_end_io = btrfs_end_super_write; 3872 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 3873 offset_in_page(bytenr)); 3874 3875 /* 3876 * We FUA only the first super block. The others we allow to 3877 * go down lazy and there's a short window where the on-disk 3878 * copies might still contain the older version. 3879 */ 3880 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; 3881 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3882 bio->bi_opf |= REQ_FUA; 3883 3884 btrfsic_submit_bio(bio); 3885 btrfs_advance_sb_log(device, i); 3886 } 3887 return errors < i ? 0 : -1; 3888 } 3889 3890 /* 3891 * Wait for write completion of superblocks done by write_dev_supers, 3892 * @max_mirrors same for write and wait phases. 3893 * 3894 * Return number of errors when page is not found or not marked up to 3895 * date. 3896 */ 3897 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3898 { 3899 int i; 3900 int errors = 0; 3901 bool primary_failed = false; 3902 int ret; 3903 u64 bytenr; 3904 3905 if (max_mirrors == 0) 3906 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3907 3908 for (i = 0; i < max_mirrors; i++) { 3909 struct page *page; 3910 3911 ret = btrfs_sb_log_location(device, i, READ, &bytenr); 3912 if (ret == -ENOENT) { 3913 break; 3914 } else if (ret < 0) { 3915 errors++; 3916 if (i == 0) 3917 primary_failed = true; 3918 continue; 3919 } 3920 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3921 device->commit_total_bytes) 3922 break; 3923 3924 page = find_get_page(device->bdev->bd_inode->i_mapping, 3925 bytenr >> PAGE_SHIFT); 3926 if (!page) { 3927 errors++; 3928 if (i == 0) 3929 primary_failed = true; 3930 continue; 3931 } 3932 /* Page is submitted locked and unlocked once the IO completes */ 3933 wait_on_page_locked(page); 3934 if (PageError(page)) { 3935 errors++; 3936 if (i == 0) 3937 primary_failed = true; 3938 } 3939 3940 /* Drop our reference */ 3941 put_page(page); 3942 3943 /* Drop the reference from the writing run */ 3944 put_page(page); 3945 } 3946 3947 /* log error, force error return */ 3948 if (primary_failed) { 3949 btrfs_err(device->fs_info, "error writing primary super block to device %llu", 3950 device->devid); 3951 return -1; 3952 } 3953 3954 return errors < i ? 0 : -1; 3955 } 3956 3957 /* 3958 * endio for the write_dev_flush, this will wake anyone waiting 3959 * for the barrier when it is done 3960 */ 3961 static void btrfs_end_empty_barrier(struct bio *bio) 3962 { 3963 complete(bio->bi_private); 3964 } 3965 3966 /* 3967 * Submit a flush request to the device if it supports it. Error handling is 3968 * done in the waiting counterpart. 3969 */ 3970 static void write_dev_flush(struct btrfs_device *device) 3971 { 3972 struct request_queue *q = bdev_get_queue(device->bdev); 3973 struct bio *bio = device->flush_bio; 3974 3975 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 3976 return; 3977 3978 bio_reset(bio); 3979 bio->bi_end_io = btrfs_end_empty_barrier; 3980 bio_set_dev(bio, device->bdev); 3981 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3982 init_completion(&device->flush_wait); 3983 bio->bi_private = &device->flush_wait; 3984 3985 btrfsic_submit_bio(bio); 3986 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 3987 } 3988 3989 /* 3990 * If the flush bio has been submitted by write_dev_flush, wait for it. 3991 */ 3992 static blk_status_t wait_dev_flush(struct btrfs_device *device) 3993 { 3994 struct bio *bio = device->flush_bio; 3995 3996 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 3997 return BLK_STS_OK; 3998 3999 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 4000 wait_for_completion_io(&device->flush_wait); 4001 4002 return bio->bi_status; 4003 } 4004 4005 static int check_barrier_error(struct btrfs_fs_info *fs_info) 4006 { 4007 if (!btrfs_check_rw_degradable(fs_info, NULL)) 4008 return -EIO; 4009 return 0; 4010 } 4011 4012 /* 4013 * send an empty flush down to each device in parallel, 4014 * then wait for them 4015 */ 4016 static int barrier_all_devices(struct btrfs_fs_info *info) 4017 { 4018 struct list_head *head; 4019 struct btrfs_device *dev; 4020 int errors_wait = 0; 4021 blk_status_t ret; 4022 4023 lockdep_assert_held(&info->fs_devices->device_list_mutex); 4024 /* send down all the barriers */ 4025 head = &info->fs_devices->devices; 4026 list_for_each_entry(dev, head, dev_list) { 4027 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4028 continue; 4029 if (!dev->bdev) 4030 continue; 4031 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4032 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4033 continue; 4034 4035 write_dev_flush(dev); 4036 dev->last_flush_error = BLK_STS_OK; 4037 } 4038 4039 /* wait for all the barriers */ 4040 list_for_each_entry(dev, head, dev_list) { 4041 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4042 continue; 4043 if (!dev->bdev) { 4044 errors_wait++; 4045 continue; 4046 } 4047 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4048 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4049 continue; 4050 4051 ret = wait_dev_flush(dev); 4052 if (ret) { 4053 dev->last_flush_error = ret; 4054 btrfs_dev_stat_inc_and_print(dev, 4055 BTRFS_DEV_STAT_FLUSH_ERRS); 4056 errors_wait++; 4057 } 4058 } 4059 4060 if (errors_wait) { 4061 /* 4062 * At some point we need the status of all disks 4063 * to arrive at the volume status. So error checking 4064 * is being pushed to a separate loop. 4065 */ 4066 return check_barrier_error(info); 4067 } 4068 return 0; 4069 } 4070 4071 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4072 { 4073 int raid_type; 4074 int min_tolerated = INT_MAX; 4075 4076 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 4077 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 4078 min_tolerated = min_t(int, min_tolerated, 4079 btrfs_raid_array[BTRFS_RAID_SINGLE]. 4080 tolerated_failures); 4081 4082 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 4083 if (raid_type == BTRFS_RAID_SINGLE) 4084 continue; 4085 if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 4086 continue; 4087 min_tolerated = min_t(int, min_tolerated, 4088 btrfs_raid_array[raid_type]. 4089 tolerated_failures); 4090 } 4091 4092 if (min_tolerated == INT_MAX) { 4093 pr_warn("BTRFS: unknown raid flag: %llu", flags); 4094 min_tolerated = 0; 4095 } 4096 4097 return min_tolerated; 4098 } 4099 4100 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4101 { 4102 struct list_head *head; 4103 struct btrfs_device *dev; 4104 struct btrfs_super_block *sb; 4105 struct btrfs_dev_item *dev_item; 4106 int ret; 4107 int do_barriers; 4108 int max_errors; 4109 int total_errors = 0; 4110 u64 flags; 4111 4112 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4113 4114 /* 4115 * max_mirrors == 0 indicates we're from commit_transaction, 4116 * not from fsync where the tree roots in fs_info have not 4117 * been consistent on disk. 4118 */ 4119 if (max_mirrors == 0) 4120 backup_super_roots(fs_info); 4121 4122 sb = fs_info->super_for_commit; 4123 dev_item = &sb->dev_item; 4124 4125 mutex_lock(&fs_info->fs_devices->device_list_mutex); 4126 head = &fs_info->fs_devices->devices; 4127 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4128 4129 if (do_barriers) { 4130 ret = barrier_all_devices(fs_info); 4131 if (ret) { 4132 mutex_unlock( 4133 &fs_info->fs_devices->device_list_mutex); 4134 btrfs_handle_fs_error(fs_info, ret, 4135 "errors while submitting device barriers."); 4136 return ret; 4137 } 4138 } 4139 4140 list_for_each_entry(dev, head, dev_list) { 4141 if (!dev->bdev) { 4142 total_errors++; 4143 continue; 4144 } 4145 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4146 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4147 continue; 4148 4149 btrfs_set_stack_device_generation(dev_item, 0); 4150 btrfs_set_stack_device_type(dev_item, dev->type); 4151 btrfs_set_stack_device_id(dev_item, dev->devid); 4152 btrfs_set_stack_device_total_bytes(dev_item, 4153 dev->commit_total_bytes); 4154 btrfs_set_stack_device_bytes_used(dev_item, 4155 dev->commit_bytes_used); 4156 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4157 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4158 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4159 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 4160 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 4161 BTRFS_FSID_SIZE); 4162 4163 flags = btrfs_super_flags(sb); 4164 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4165 4166 ret = btrfs_validate_write_super(fs_info, sb); 4167 if (ret < 0) { 4168 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4169 btrfs_handle_fs_error(fs_info, -EUCLEAN, 4170 "unexpected superblock corruption detected"); 4171 return -EUCLEAN; 4172 } 4173 4174 ret = write_dev_supers(dev, sb, max_mirrors); 4175 if (ret) 4176 total_errors++; 4177 } 4178 if (total_errors > max_errors) { 4179 btrfs_err(fs_info, "%d errors while writing supers", 4180 total_errors); 4181 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4182 4183 /* FUA is masked off if unsupported and can't be the reason */ 4184 btrfs_handle_fs_error(fs_info, -EIO, 4185 "%d errors while writing supers", 4186 total_errors); 4187 return -EIO; 4188 } 4189 4190 total_errors = 0; 4191 list_for_each_entry(dev, head, dev_list) { 4192 if (!dev->bdev) 4193 continue; 4194 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4195 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4196 continue; 4197 4198 ret = wait_dev_supers(dev, max_mirrors); 4199 if (ret) 4200 total_errors++; 4201 } 4202 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4203 if (total_errors > max_errors) { 4204 btrfs_handle_fs_error(fs_info, -EIO, 4205 "%d errors while writing supers", 4206 total_errors); 4207 return -EIO; 4208 } 4209 return 0; 4210 } 4211 4212 /* Drop a fs root from the radix tree and free it. */ 4213 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4214 struct btrfs_root *root) 4215 { 4216 bool drop_ref = false; 4217 4218 spin_lock(&fs_info->fs_roots_radix_lock); 4219 radix_tree_delete(&fs_info->fs_roots_radix, 4220 (unsigned long)root->root_key.objectid); 4221 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 4222 drop_ref = true; 4223 spin_unlock(&fs_info->fs_roots_radix_lock); 4224 4225 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 4226 ASSERT(root->log_root == NULL); 4227 if (root->reloc_root) { 4228 btrfs_put_root(root->reloc_root); 4229 root->reloc_root = NULL; 4230 } 4231 } 4232 4233 if (drop_ref) 4234 btrfs_put_root(root); 4235 } 4236 4237 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4238 { 4239 u64 root_objectid = 0; 4240 struct btrfs_root *gang[8]; 4241 int i = 0; 4242 int err = 0; 4243 unsigned int ret = 0; 4244 4245 while (1) { 4246 spin_lock(&fs_info->fs_roots_radix_lock); 4247 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4248 (void **)gang, root_objectid, 4249 ARRAY_SIZE(gang)); 4250 if (!ret) { 4251 spin_unlock(&fs_info->fs_roots_radix_lock); 4252 break; 4253 } 4254 root_objectid = gang[ret - 1]->root_key.objectid + 1; 4255 4256 for (i = 0; i < ret; i++) { 4257 /* Avoid to grab roots in dead_roots */ 4258 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4259 gang[i] = NULL; 4260 continue; 4261 } 4262 /* grab all the search result for later use */ 4263 gang[i] = btrfs_grab_root(gang[i]); 4264 } 4265 spin_unlock(&fs_info->fs_roots_radix_lock); 4266 4267 for (i = 0; i < ret; i++) { 4268 if (!gang[i]) 4269 continue; 4270 root_objectid = gang[i]->root_key.objectid; 4271 err = btrfs_orphan_cleanup(gang[i]); 4272 if (err) 4273 break; 4274 btrfs_put_root(gang[i]); 4275 } 4276 root_objectid++; 4277 } 4278 4279 /* release the uncleaned roots due to error */ 4280 for (; i < ret; i++) { 4281 if (gang[i]) 4282 btrfs_put_root(gang[i]); 4283 } 4284 return err; 4285 } 4286 4287 int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4288 { 4289 struct btrfs_root *root = fs_info->tree_root; 4290 struct btrfs_trans_handle *trans; 4291 4292 mutex_lock(&fs_info->cleaner_mutex); 4293 btrfs_run_delayed_iputs(fs_info); 4294 mutex_unlock(&fs_info->cleaner_mutex); 4295 wake_up_process(fs_info->cleaner_kthread); 4296 4297 /* wait until ongoing cleanup work done */ 4298 down_write(&fs_info->cleanup_work_sem); 4299 up_write(&fs_info->cleanup_work_sem); 4300 4301 trans = btrfs_join_transaction(root); 4302 if (IS_ERR(trans)) 4303 return PTR_ERR(trans); 4304 return btrfs_commit_transaction(trans); 4305 } 4306 4307 void __cold close_ctree(struct btrfs_fs_info *fs_info) 4308 { 4309 int ret; 4310 4311 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4312 /* 4313 * We don't want the cleaner to start new transactions, add more delayed 4314 * iputs, etc. while we're closing. We can't use kthread_stop() yet 4315 * because that frees the task_struct, and the transaction kthread might 4316 * still try to wake up the cleaner. 4317 */ 4318 kthread_park(fs_info->cleaner_kthread); 4319 4320 /* wait for the qgroup rescan worker to stop */ 4321 btrfs_qgroup_wait_for_completion(fs_info, false); 4322 4323 /* wait for the uuid_scan task to finish */ 4324 down(&fs_info->uuid_tree_rescan_sem); 4325 /* avoid complains from lockdep et al., set sem back to initial state */ 4326 up(&fs_info->uuid_tree_rescan_sem); 4327 4328 /* pause restriper - we want to resume on mount */ 4329 btrfs_pause_balance(fs_info); 4330 4331 btrfs_dev_replace_suspend_for_unmount(fs_info); 4332 4333 btrfs_scrub_cancel(fs_info); 4334 4335 /* wait for any defraggers to finish */ 4336 wait_event(fs_info->transaction_wait, 4337 (atomic_read(&fs_info->defrag_running) == 0)); 4338 4339 /* clear out the rbtree of defraggable inodes */ 4340 btrfs_cleanup_defrag_inodes(fs_info); 4341 4342 cancel_work_sync(&fs_info->async_reclaim_work); 4343 cancel_work_sync(&fs_info->async_data_reclaim_work); 4344 cancel_work_sync(&fs_info->preempt_reclaim_work); 4345 4346 cancel_work_sync(&fs_info->reclaim_bgs_work); 4347 4348 /* Cancel or finish ongoing discard work */ 4349 btrfs_discard_cleanup(fs_info); 4350 4351 if (!sb_rdonly(fs_info->sb)) { 4352 /* 4353 * The cleaner kthread is stopped, so do one final pass over 4354 * unused block groups. 4355 */ 4356 btrfs_delete_unused_bgs(fs_info); 4357 4358 /* 4359 * There might be existing delayed inode workers still running 4360 * and holding an empty delayed inode item. We must wait for 4361 * them to complete first because they can create a transaction. 4362 * This happens when someone calls btrfs_balance_delayed_items() 4363 * and then a transaction commit runs the same delayed nodes 4364 * before any delayed worker has done something with the nodes. 4365 * We must wait for any worker here and not at transaction 4366 * commit time since that could cause a deadlock. 4367 * This is a very rare case. 4368 */ 4369 btrfs_flush_workqueue(fs_info->delayed_workers); 4370 4371 ret = btrfs_commit_super(fs_info); 4372 if (ret) 4373 btrfs_err(fs_info, "commit super ret %d", ret); 4374 } 4375 4376 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || 4377 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) 4378 btrfs_error_commit_super(fs_info); 4379 4380 kthread_stop(fs_info->transaction_kthread); 4381 kthread_stop(fs_info->cleaner_kthread); 4382 4383 ASSERT(list_empty(&fs_info->delayed_iputs)); 4384 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4385 4386 if (btrfs_check_quota_leak(fs_info)) { 4387 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4388 btrfs_err(fs_info, "qgroup reserved space leaked"); 4389 } 4390 4391 btrfs_free_qgroup_config(fs_info); 4392 ASSERT(list_empty(&fs_info->delalloc_roots)); 4393 4394 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 4395 btrfs_info(fs_info, "at unmount delalloc count %lld", 4396 percpu_counter_sum(&fs_info->delalloc_bytes)); 4397 } 4398 4399 if (percpu_counter_sum(&fs_info->ordered_bytes)) 4400 btrfs_info(fs_info, "at unmount dio bytes count %lld", 4401 percpu_counter_sum(&fs_info->ordered_bytes)); 4402 4403 btrfs_sysfs_remove_mounted(fs_info); 4404 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 4405 4406 btrfs_put_block_group_cache(fs_info); 4407 4408 /* 4409 * we must make sure there is not any read request to 4410 * submit after we stopping all workers. 4411 */ 4412 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 4413 btrfs_stop_all_workers(fs_info); 4414 4415 /* We shouldn't have any transaction open at this point */ 4416 ASSERT(list_empty(&fs_info->trans_list)); 4417 4418 clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 4419 free_root_pointers(fs_info, true); 4420 btrfs_free_fs_roots(fs_info); 4421 4422 /* 4423 * We must free the block groups after dropping the fs_roots as we could 4424 * have had an IO error and have left over tree log blocks that aren't 4425 * cleaned up until the fs roots are freed. This makes the block group 4426 * accounting appear to be wrong because there's pending reserved bytes, 4427 * so make sure we do the block group cleanup afterwards. 4428 */ 4429 btrfs_free_block_groups(fs_info); 4430 4431 iput(fs_info->btree_inode); 4432 4433 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4434 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 4435 btrfsic_unmount(fs_info->fs_devices); 4436 #endif 4437 4438 btrfs_mapping_tree_free(&fs_info->mapping_tree); 4439 btrfs_close_devices(fs_info->fs_devices); 4440 } 4441 4442 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4443 int atomic) 4444 { 4445 int ret; 4446 struct inode *btree_inode = buf->pages[0]->mapping->host; 4447 4448 ret = extent_buffer_uptodate(buf); 4449 if (!ret) 4450 return ret; 4451 4452 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4453 parent_transid, atomic); 4454 if (ret == -EAGAIN) 4455 return ret; 4456 return !ret; 4457 } 4458 4459 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 4460 { 4461 struct btrfs_fs_info *fs_info = buf->fs_info; 4462 u64 transid = btrfs_header_generation(buf); 4463 int was_dirty; 4464 4465 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 4466 /* 4467 * This is a fast path so only do this check if we have sanity tests 4468 * enabled. Normal people shouldn't be using unmapped buffers as dirty 4469 * outside of the sanity tests. 4470 */ 4471 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 4472 return; 4473 #endif 4474 btrfs_assert_tree_locked(buf); 4475 if (transid != fs_info->generation) 4476 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 4477 buf->start, transid, fs_info->generation); 4478 was_dirty = set_extent_buffer_dirty(buf); 4479 if (!was_dirty) 4480 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4481 buf->len, 4482 fs_info->dirty_metadata_batch); 4483 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4484 /* 4485 * Since btrfs_mark_buffer_dirty() can be called with item pointer set 4486 * but item data not updated. 4487 * So here we should only check item pointers, not item data. 4488 */ 4489 if (btrfs_header_level(buf) == 0 && 4490 btrfs_check_leaf_relaxed(buf)) { 4491 btrfs_print_leaf(buf); 4492 ASSERT(0); 4493 } 4494 #endif 4495 } 4496 4497 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4498 int flush_delayed) 4499 { 4500 /* 4501 * looks as though older kernels can get into trouble with 4502 * this code, they end up stuck in balance_dirty_pages forever 4503 */ 4504 int ret; 4505 4506 if (current->flags & PF_MEMALLOC) 4507 return; 4508 4509 if (flush_delayed) 4510 btrfs_balance_delayed_items(fs_info); 4511 4512 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4513 BTRFS_DIRTY_METADATA_THRESH, 4514 fs_info->dirty_metadata_batch); 4515 if (ret > 0) { 4516 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 4517 } 4518 } 4519 4520 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 4521 { 4522 __btrfs_btree_balance_dirty(fs_info, 1); 4523 } 4524 4525 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4526 { 4527 __btrfs_btree_balance_dirty(fs_info, 0); 4528 } 4529 4530 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4531 struct btrfs_key *first_key) 4532 { 4533 return btree_read_extent_buffer_pages(buf, parent_transid, 4534 level, first_key); 4535 } 4536 4537 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4538 { 4539 /* cleanup FS via transaction */ 4540 btrfs_cleanup_transaction(fs_info); 4541 4542 mutex_lock(&fs_info->cleaner_mutex); 4543 btrfs_run_delayed_iputs(fs_info); 4544 mutex_unlock(&fs_info->cleaner_mutex); 4545 4546 down_write(&fs_info->cleanup_work_sem); 4547 up_write(&fs_info->cleanup_work_sem); 4548 } 4549 4550 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4551 { 4552 struct btrfs_root *gang[8]; 4553 u64 root_objectid = 0; 4554 int ret; 4555 4556 spin_lock(&fs_info->fs_roots_radix_lock); 4557 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4558 (void **)gang, root_objectid, 4559 ARRAY_SIZE(gang))) != 0) { 4560 int i; 4561 4562 for (i = 0; i < ret; i++) 4563 gang[i] = btrfs_grab_root(gang[i]); 4564 spin_unlock(&fs_info->fs_roots_radix_lock); 4565 4566 for (i = 0; i < ret; i++) { 4567 if (!gang[i]) 4568 continue; 4569 root_objectid = gang[i]->root_key.objectid; 4570 btrfs_free_log(NULL, gang[i]); 4571 btrfs_put_root(gang[i]); 4572 } 4573 root_objectid++; 4574 spin_lock(&fs_info->fs_roots_radix_lock); 4575 } 4576 spin_unlock(&fs_info->fs_roots_radix_lock); 4577 btrfs_free_log_root_tree(NULL, fs_info); 4578 } 4579 4580 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4581 { 4582 struct btrfs_ordered_extent *ordered; 4583 4584 spin_lock(&root->ordered_extent_lock); 4585 /* 4586 * This will just short circuit the ordered completion stuff which will 4587 * make sure the ordered extent gets properly cleaned up. 4588 */ 4589 list_for_each_entry(ordered, &root->ordered_extents, 4590 root_extent_list) 4591 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4592 spin_unlock(&root->ordered_extent_lock); 4593 } 4594 4595 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4596 { 4597 struct btrfs_root *root; 4598 struct list_head splice; 4599 4600 INIT_LIST_HEAD(&splice); 4601 4602 spin_lock(&fs_info->ordered_root_lock); 4603 list_splice_init(&fs_info->ordered_roots, &splice); 4604 while (!list_empty(&splice)) { 4605 root = list_first_entry(&splice, struct btrfs_root, 4606 ordered_root); 4607 list_move_tail(&root->ordered_root, 4608 &fs_info->ordered_roots); 4609 4610 spin_unlock(&fs_info->ordered_root_lock); 4611 btrfs_destroy_ordered_extents(root); 4612 4613 cond_resched(); 4614 spin_lock(&fs_info->ordered_root_lock); 4615 } 4616 spin_unlock(&fs_info->ordered_root_lock); 4617 4618 /* 4619 * We need this here because if we've been flipped read-only we won't 4620 * get sync() from the umount, so we need to make sure any ordered 4621 * extents that haven't had their dirty pages IO start writeout yet 4622 * actually get run and error out properly. 4623 */ 4624 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4625 } 4626 4627 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4628 struct btrfs_fs_info *fs_info) 4629 { 4630 struct rb_node *node; 4631 struct btrfs_delayed_ref_root *delayed_refs; 4632 struct btrfs_delayed_ref_node *ref; 4633 int ret = 0; 4634 4635 delayed_refs = &trans->delayed_refs; 4636 4637 spin_lock(&delayed_refs->lock); 4638 if (atomic_read(&delayed_refs->num_entries) == 0) { 4639 spin_unlock(&delayed_refs->lock); 4640 btrfs_debug(fs_info, "delayed_refs has NO entry"); 4641 return ret; 4642 } 4643 4644 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4645 struct btrfs_delayed_ref_head *head; 4646 struct rb_node *n; 4647 bool pin_bytes = false; 4648 4649 head = rb_entry(node, struct btrfs_delayed_ref_head, 4650 href_node); 4651 if (btrfs_delayed_ref_lock(delayed_refs, head)) 4652 continue; 4653 4654 spin_lock(&head->lock); 4655 while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 4656 ref = rb_entry(n, struct btrfs_delayed_ref_node, 4657 ref_node); 4658 ref->in_tree = 0; 4659 rb_erase_cached(&ref->ref_node, &head->ref_tree); 4660 RB_CLEAR_NODE(&ref->ref_node); 4661 if (!list_empty(&ref->add_list)) 4662 list_del(&ref->add_list); 4663 atomic_dec(&delayed_refs->num_entries); 4664 btrfs_put_delayed_ref(ref); 4665 } 4666 if (head->must_insert_reserved) 4667 pin_bytes = true; 4668 btrfs_free_delayed_extent_op(head->extent_op); 4669 btrfs_delete_ref_head(delayed_refs, head); 4670 spin_unlock(&head->lock); 4671 spin_unlock(&delayed_refs->lock); 4672 mutex_unlock(&head->mutex); 4673 4674 if (pin_bytes) { 4675 struct btrfs_block_group *cache; 4676 4677 cache = btrfs_lookup_block_group(fs_info, head->bytenr); 4678 BUG_ON(!cache); 4679 4680 spin_lock(&cache->space_info->lock); 4681 spin_lock(&cache->lock); 4682 cache->pinned += head->num_bytes; 4683 btrfs_space_info_update_bytes_pinned(fs_info, 4684 cache->space_info, head->num_bytes); 4685 cache->reserved -= head->num_bytes; 4686 cache->space_info->bytes_reserved -= head->num_bytes; 4687 spin_unlock(&cache->lock); 4688 spin_unlock(&cache->space_info->lock); 4689 percpu_counter_add_batch( 4690 &cache->space_info->total_bytes_pinned, 4691 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); 4692 4693 btrfs_put_block_group(cache); 4694 4695 btrfs_error_unpin_extent_range(fs_info, head->bytenr, 4696 head->bytenr + head->num_bytes - 1); 4697 } 4698 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 4699 btrfs_put_delayed_ref_head(head); 4700 cond_resched(); 4701 spin_lock(&delayed_refs->lock); 4702 } 4703 btrfs_qgroup_destroy_extent_records(trans); 4704 4705 spin_unlock(&delayed_refs->lock); 4706 4707 return ret; 4708 } 4709 4710 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4711 { 4712 struct btrfs_inode *btrfs_inode; 4713 struct list_head splice; 4714 4715 INIT_LIST_HEAD(&splice); 4716 4717 spin_lock(&root->delalloc_lock); 4718 list_splice_init(&root->delalloc_inodes, &splice); 4719 4720 while (!list_empty(&splice)) { 4721 struct inode *inode = NULL; 4722 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4723 delalloc_inodes); 4724 __btrfs_del_delalloc_inode(root, btrfs_inode); 4725 spin_unlock(&root->delalloc_lock); 4726 4727 /* 4728 * Make sure we get a live inode and that it'll not disappear 4729 * meanwhile. 4730 */ 4731 inode = igrab(&btrfs_inode->vfs_inode); 4732 if (inode) { 4733 invalidate_inode_pages2(inode->i_mapping); 4734 iput(inode); 4735 } 4736 spin_lock(&root->delalloc_lock); 4737 } 4738 spin_unlock(&root->delalloc_lock); 4739 } 4740 4741 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4742 { 4743 struct btrfs_root *root; 4744 struct list_head splice; 4745 4746 INIT_LIST_HEAD(&splice); 4747 4748 spin_lock(&fs_info->delalloc_root_lock); 4749 list_splice_init(&fs_info->delalloc_roots, &splice); 4750 while (!list_empty(&splice)) { 4751 root = list_first_entry(&splice, struct btrfs_root, 4752 delalloc_root); 4753 root = btrfs_grab_root(root); 4754 BUG_ON(!root); 4755 spin_unlock(&fs_info->delalloc_root_lock); 4756 4757 btrfs_destroy_delalloc_inodes(root); 4758 btrfs_put_root(root); 4759 4760 spin_lock(&fs_info->delalloc_root_lock); 4761 } 4762 spin_unlock(&fs_info->delalloc_root_lock); 4763 } 4764 4765 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4766 struct extent_io_tree *dirty_pages, 4767 int mark) 4768 { 4769 int ret; 4770 struct extent_buffer *eb; 4771 u64 start = 0; 4772 u64 end; 4773 4774 while (1) { 4775 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4776 mark, NULL); 4777 if (ret) 4778 break; 4779 4780 clear_extent_bits(dirty_pages, start, end, mark); 4781 while (start <= end) { 4782 eb = find_extent_buffer(fs_info, start); 4783 start += fs_info->nodesize; 4784 if (!eb) 4785 continue; 4786 wait_on_extent_buffer_writeback(eb); 4787 4788 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4789 &eb->bflags)) 4790 clear_extent_buffer_dirty(eb); 4791 free_extent_buffer_stale(eb); 4792 } 4793 } 4794 4795 return ret; 4796 } 4797 4798 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4799 struct extent_io_tree *unpin) 4800 { 4801 u64 start; 4802 u64 end; 4803 int ret; 4804 4805 while (1) { 4806 struct extent_state *cached_state = NULL; 4807 4808 /* 4809 * The btrfs_finish_extent_commit() may get the same range as 4810 * ours between find_first_extent_bit and clear_extent_dirty. 4811 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 4812 * the same extent range. 4813 */ 4814 mutex_lock(&fs_info->unused_bg_unpin_mutex); 4815 ret = find_first_extent_bit(unpin, 0, &start, &end, 4816 EXTENT_DIRTY, &cached_state); 4817 if (ret) { 4818 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4819 break; 4820 } 4821 4822 clear_extent_dirty(unpin, start, end, &cached_state); 4823 free_extent_state(cached_state); 4824 btrfs_error_unpin_extent_range(fs_info, start, end); 4825 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 4826 cond_resched(); 4827 } 4828 4829 return 0; 4830 } 4831 4832 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 4833 { 4834 struct inode *inode; 4835 4836 inode = cache->io_ctl.inode; 4837 if (inode) { 4838 invalidate_inode_pages2(inode->i_mapping); 4839 BTRFS_I(inode)->generation = 0; 4840 cache->io_ctl.inode = NULL; 4841 iput(inode); 4842 } 4843 ASSERT(cache->io_ctl.pages == NULL); 4844 btrfs_put_block_group(cache); 4845 } 4846 4847 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 4848 struct btrfs_fs_info *fs_info) 4849 { 4850 struct btrfs_block_group *cache; 4851 4852 spin_lock(&cur_trans->dirty_bgs_lock); 4853 while (!list_empty(&cur_trans->dirty_bgs)) { 4854 cache = list_first_entry(&cur_trans->dirty_bgs, 4855 struct btrfs_block_group, 4856 dirty_list); 4857 4858 if (!list_empty(&cache->io_list)) { 4859 spin_unlock(&cur_trans->dirty_bgs_lock); 4860 list_del_init(&cache->io_list); 4861 btrfs_cleanup_bg_io(cache); 4862 spin_lock(&cur_trans->dirty_bgs_lock); 4863 } 4864 4865 list_del_init(&cache->dirty_list); 4866 spin_lock(&cache->lock); 4867 cache->disk_cache_state = BTRFS_DC_ERROR; 4868 spin_unlock(&cache->lock); 4869 4870 spin_unlock(&cur_trans->dirty_bgs_lock); 4871 btrfs_put_block_group(cache); 4872 btrfs_delayed_refs_rsv_release(fs_info, 1); 4873 spin_lock(&cur_trans->dirty_bgs_lock); 4874 } 4875 spin_unlock(&cur_trans->dirty_bgs_lock); 4876 4877 /* 4878 * Refer to the definition of io_bgs member for details why it's safe 4879 * to use it without any locking 4880 */ 4881 while (!list_empty(&cur_trans->io_bgs)) { 4882 cache = list_first_entry(&cur_trans->io_bgs, 4883 struct btrfs_block_group, 4884 io_list); 4885 4886 list_del_init(&cache->io_list); 4887 spin_lock(&cache->lock); 4888 cache->disk_cache_state = BTRFS_DC_ERROR; 4889 spin_unlock(&cache->lock); 4890 btrfs_cleanup_bg_io(cache); 4891 } 4892 } 4893 4894 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 4895 struct btrfs_fs_info *fs_info) 4896 { 4897 struct btrfs_device *dev, *tmp; 4898 4899 btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4900 ASSERT(list_empty(&cur_trans->dirty_bgs)); 4901 ASSERT(list_empty(&cur_trans->io_bgs)); 4902 4903 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 4904 post_commit_list) { 4905 list_del_init(&dev->post_commit_list); 4906 } 4907 4908 btrfs_destroy_delayed_refs(cur_trans, fs_info); 4909 4910 cur_trans->state = TRANS_STATE_COMMIT_START; 4911 wake_up(&fs_info->transaction_blocked_wait); 4912 4913 cur_trans->state = TRANS_STATE_UNBLOCKED; 4914 wake_up(&fs_info->transaction_wait); 4915 4916 btrfs_destroy_delayed_inodes(fs_info); 4917 4918 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 4919 EXTENT_DIRTY); 4920 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 4921 4922 btrfs_free_redirty_list(cur_trans); 4923 4924 cur_trans->state =TRANS_STATE_COMPLETED; 4925 wake_up(&cur_trans->commit_wait); 4926 } 4927 4928 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4929 { 4930 struct btrfs_transaction *t; 4931 4932 mutex_lock(&fs_info->transaction_kthread_mutex); 4933 4934 spin_lock(&fs_info->trans_lock); 4935 while (!list_empty(&fs_info->trans_list)) { 4936 t = list_first_entry(&fs_info->trans_list, 4937 struct btrfs_transaction, list); 4938 if (t->state >= TRANS_STATE_COMMIT_START) { 4939 refcount_inc(&t->use_count); 4940 spin_unlock(&fs_info->trans_lock); 4941 btrfs_wait_for_commit(fs_info, t->transid); 4942 btrfs_put_transaction(t); 4943 spin_lock(&fs_info->trans_lock); 4944 continue; 4945 } 4946 if (t == fs_info->running_transaction) { 4947 t->state = TRANS_STATE_COMMIT_DOING; 4948 spin_unlock(&fs_info->trans_lock); 4949 /* 4950 * We wait for 0 num_writers since we don't hold a trans 4951 * handle open currently for this transaction. 4952 */ 4953 wait_event(t->writer_wait, 4954 atomic_read(&t->num_writers) == 0); 4955 } else { 4956 spin_unlock(&fs_info->trans_lock); 4957 } 4958 btrfs_cleanup_one_transaction(t, fs_info); 4959 4960 spin_lock(&fs_info->trans_lock); 4961 if (t == fs_info->running_transaction) 4962 fs_info->running_transaction = NULL; 4963 list_del_init(&t->list); 4964 spin_unlock(&fs_info->trans_lock); 4965 4966 btrfs_put_transaction(t); 4967 trace_btrfs_transaction_commit(fs_info->tree_root); 4968 spin_lock(&fs_info->trans_lock); 4969 } 4970 spin_unlock(&fs_info->trans_lock); 4971 btrfs_destroy_all_ordered_extents(fs_info); 4972 btrfs_destroy_delayed_inodes(fs_info); 4973 btrfs_assert_delayed_root_empty(fs_info); 4974 btrfs_destroy_all_delalloc_inodes(fs_info); 4975 btrfs_drop_all_logs(fs_info); 4976 mutex_unlock(&fs_info->transaction_kthread_mutex); 4977 4978 return 0; 4979 } 4980 4981 int btrfs_init_root_free_objectid(struct btrfs_root *root) 4982 { 4983 struct btrfs_path *path; 4984 int ret; 4985 struct extent_buffer *l; 4986 struct btrfs_key search_key; 4987 struct btrfs_key found_key; 4988 int slot; 4989 4990 path = btrfs_alloc_path(); 4991 if (!path) 4992 return -ENOMEM; 4993 4994 search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 4995 search_key.type = -1; 4996 search_key.offset = (u64)-1; 4997 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 4998 if (ret < 0) 4999 goto error; 5000 BUG_ON(ret == 0); /* Corruption */ 5001 if (path->slots[0] > 0) { 5002 slot = path->slots[0] - 1; 5003 l = path->nodes[0]; 5004 btrfs_item_key_to_cpu(l, &found_key, slot); 5005 root->free_objectid = max_t(u64, found_key.objectid + 1, 5006 BTRFS_FIRST_FREE_OBJECTID); 5007 } else { 5008 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5009 } 5010 ret = 0; 5011 error: 5012 btrfs_free_path(path); 5013 return ret; 5014 } 5015 5016 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5017 { 5018 int ret; 5019 mutex_lock(&root->objectid_mutex); 5020 5021 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5022 btrfs_warn(root->fs_info, 5023 "the objectid of root %llu reaches its highest value", 5024 root->root_key.objectid); 5025 ret = -ENOSPC; 5026 goto out; 5027 } 5028 5029 *objectid = root->free_objectid++; 5030 ret = 0; 5031 out: 5032 mutex_unlock(&root->objectid_mutex); 5033 return ret; 5034 } 5035