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 = num_extent_pages(buf); 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; 244 245 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 246 return 0; 247 248 if (atomic) 249 return -EAGAIN; 250 251 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 252 &cached_state); 253 if (extent_buffer_uptodate(eb) && 254 btrfs_header_generation(eb) == parent_transid) { 255 ret = 0; 256 goto out; 257 } 258 btrfs_err_rl(eb->fs_info, 259 "parent transid verify failed on %llu wanted %llu found %llu", 260 eb->start, 261 parent_transid, btrfs_header_generation(eb)); 262 ret = 1; 263 clear_extent_buffer_uptodate(eb); 264 out: 265 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 266 &cached_state); 267 return ret; 268 } 269 270 static bool btrfs_supported_super_csum(u16 csum_type) 271 { 272 switch (csum_type) { 273 case BTRFS_CSUM_TYPE_CRC32: 274 case BTRFS_CSUM_TYPE_XXHASH: 275 case BTRFS_CSUM_TYPE_SHA256: 276 case BTRFS_CSUM_TYPE_BLAKE2: 277 return true; 278 default: 279 return false; 280 } 281 } 282 283 /* 284 * Return 0 if the superblock checksum type matches the checksum value of that 285 * algorithm. Pass the raw disk superblock data. 286 */ 287 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 288 char *raw_disk_sb) 289 { 290 struct btrfs_super_block *disk_sb = 291 (struct btrfs_super_block *)raw_disk_sb; 292 char result[BTRFS_CSUM_SIZE]; 293 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 294 295 shash->tfm = fs_info->csum_shash; 296 297 /* 298 * The super_block structure does not span the whole 299 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is 300 * filled with zeros and is included in the checksum. 301 */ 302 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, 303 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); 304 305 if (memcmp(disk_sb->csum, result, fs_info->csum_size)) 306 return 1; 307 308 return 0; 309 } 310 311 int btrfs_verify_level_key(struct extent_buffer *eb, int level, 312 struct btrfs_key *first_key, u64 parent_transid) 313 { 314 struct btrfs_fs_info *fs_info = eb->fs_info; 315 int found_level; 316 struct btrfs_key found_key; 317 int ret; 318 319 found_level = btrfs_header_level(eb); 320 if (found_level != level) { 321 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 322 KERN_ERR "BTRFS: tree level check failed\n"); 323 btrfs_err(fs_info, 324 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 325 eb->start, level, found_level); 326 return -EIO; 327 } 328 329 if (!first_key) 330 return 0; 331 332 /* 333 * For live tree block (new tree blocks in current transaction), 334 * we need proper lock context to avoid race, which is impossible here. 335 * So we only checks tree blocks which is read from disk, whose 336 * generation <= fs_info->last_trans_committed. 337 */ 338 if (btrfs_header_generation(eb) > fs_info->last_trans_committed) 339 return 0; 340 341 /* We have @first_key, so this @eb must have at least one item */ 342 if (btrfs_header_nritems(eb) == 0) { 343 btrfs_err(fs_info, 344 "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 345 eb->start); 346 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 347 return -EUCLEAN; 348 } 349 350 if (found_level) 351 btrfs_node_key_to_cpu(eb, &found_key, 0); 352 else 353 btrfs_item_key_to_cpu(eb, &found_key, 0); 354 ret = btrfs_comp_cpu_keys(first_key, &found_key); 355 356 if (ret) { 357 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), 358 KERN_ERR "BTRFS: tree first key check failed\n"); 359 btrfs_err(fs_info, 360 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 361 eb->start, parent_transid, first_key->objectid, 362 first_key->type, first_key->offset, 363 found_key.objectid, found_key.type, 364 found_key.offset); 365 } 366 return ret; 367 } 368 369 /* 370 * helper to read a given tree block, doing retries as required when 371 * the checksums don't match and we have alternate mirrors to try. 372 * 373 * @parent_transid: expected transid, skip check if 0 374 * @level: expected level, mandatory check 375 * @first_key: expected key of first slot, skip check if NULL 376 */ 377 static int btree_read_extent_buffer_pages(struct extent_buffer *eb, 378 u64 parent_transid, int level, 379 struct btrfs_key *first_key) 380 { 381 struct btrfs_fs_info *fs_info = eb->fs_info; 382 struct extent_io_tree *io_tree; 383 int failed = 0; 384 int ret; 385 int num_copies = 0; 386 int mirror_num = 0; 387 int failed_mirror = 0; 388 389 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 390 while (1) { 391 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 392 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); 393 if (!ret) { 394 if (verify_parent_transid(io_tree, eb, 395 parent_transid, 0)) 396 ret = -EIO; 397 else if (btrfs_verify_level_key(eb, level, 398 first_key, parent_transid)) 399 ret = -EUCLEAN; 400 else 401 break; 402 } 403 404 num_copies = btrfs_num_copies(fs_info, 405 eb->start, eb->len); 406 if (num_copies == 1) 407 break; 408 409 if (!failed_mirror) { 410 failed = 1; 411 failed_mirror = eb->read_mirror; 412 } 413 414 mirror_num++; 415 if (mirror_num == failed_mirror) 416 mirror_num++; 417 418 if (mirror_num > num_copies) 419 break; 420 } 421 422 if (failed && !ret && failed_mirror) 423 btrfs_repair_eb_io_failure(eb, failed_mirror); 424 425 return ret; 426 } 427 428 static int csum_one_extent_buffer(struct extent_buffer *eb) 429 { 430 struct btrfs_fs_info *fs_info = eb->fs_info; 431 u8 result[BTRFS_CSUM_SIZE]; 432 int ret; 433 434 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, 435 offsetof(struct btrfs_header, fsid), 436 BTRFS_FSID_SIZE) == 0); 437 csum_tree_block(eb, result); 438 439 if (btrfs_header_level(eb)) 440 ret = btrfs_check_node(eb); 441 else 442 ret = btrfs_check_leaf_full(eb); 443 444 if (ret < 0) 445 goto error; 446 447 /* 448 * Also check the generation, the eb reached here must be newer than 449 * last committed. Or something seriously wrong happened. 450 */ 451 if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) { 452 ret = -EUCLEAN; 453 btrfs_err(fs_info, 454 "block=%llu bad generation, have %llu expect > %llu", 455 eb->start, btrfs_header_generation(eb), 456 fs_info->last_trans_committed); 457 goto error; 458 } 459 write_extent_buffer(eb, result, 0, fs_info->csum_size); 460 461 return 0; 462 463 error: 464 btrfs_print_tree(eb, 0); 465 btrfs_err(fs_info, "block=%llu write time tree block corruption detected", 466 eb->start); 467 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 468 return ret; 469 } 470 471 /* Checksum all dirty extent buffers in one bio_vec */ 472 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info, 473 struct bio_vec *bvec) 474 { 475 struct page *page = bvec->bv_page; 476 u64 bvec_start = page_offset(page) + bvec->bv_offset; 477 u64 cur; 478 int ret = 0; 479 480 for (cur = bvec_start; cur < bvec_start + bvec->bv_len; 481 cur += fs_info->nodesize) { 482 struct extent_buffer *eb; 483 bool uptodate; 484 485 eb = find_extent_buffer(fs_info, cur); 486 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur, 487 fs_info->nodesize); 488 489 /* A dirty eb shouldn't disappear from buffer_radix */ 490 if (WARN_ON(!eb)) 491 return -EUCLEAN; 492 493 if (WARN_ON(cur != btrfs_header_bytenr(eb))) { 494 free_extent_buffer(eb); 495 return -EUCLEAN; 496 } 497 if (WARN_ON(!uptodate)) { 498 free_extent_buffer(eb); 499 return -EUCLEAN; 500 } 501 502 ret = csum_one_extent_buffer(eb); 503 free_extent_buffer(eb); 504 if (ret < 0) 505 return ret; 506 } 507 return ret; 508 } 509 510 /* 511 * Checksum a dirty tree block before IO. This has extra checks to make sure 512 * we only fill in the checksum field in the first page of a multi-page block. 513 * For subpage extent buffers we need bvec to also read the offset in the page. 514 */ 515 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) 516 { 517 struct page *page = bvec->bv_page; 518 u64 start = page_offset(page); 519 u64 found_start; 520 struct extent_buffer *eb; 521 522 if (fs_info->sectorsize < PAGE_SIZE) 523 return csum_dirty_subpage_buffers(fs_info, bvec); 524 525 eb = (struct extent_buffer *)page->private; 526 if (page != eb->pages[0]) 527 return 0; 528 529 found_start = btrfs_header_bytenr(eb); 530 531 if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) { 532 WARN_ON(found_start != 0); 533 return 0; 534 } 535 536 /* 537 * Please do not consolidate these warnings into a single if. 538 * It is useful to know what went wrong. 539 */ 540 if (WARN_ON(found_start != start)) 541 return -EUCLEAN; 542 if (WARN_ON(!PageUptodate(page))) 543 return -EUCLEAN; 544 545 return csum_one_extent_buffer(eb); 546 } 547 548 static int check_tree_block_fsid(struct extent_buffer *eb) 549 { 550 struct btrfs_fs_info *fs_info = eb->fs_info; 551 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 552 u8 fsid[BTRFS_FSID_SIZE]; 553 u8 *metadata_uuid; 554 555 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), 556 BTRFS_FSID_SIZE); 557 /* 558 * Checking the incompat flag is only valid for the current fs. For 559 * seed devices it's forbidden to have their uuid changed so reading 560 * ->fsid in this case is fine 561 */ 562 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 563 metadata_uuid = fs_devices->metadata_uuid; 564 else 565 metadata_uuid = fs_devices->fsid; 566 567 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) 568 return 0; 569 570 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) 571 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) 572 return 0; 573 574 return 1; 575 } 576 577 /* Do basic extent buffer checks at read time */ 578 static int validate_extent_buffer(struct extent_buffer *eb) 579 { 580 struct btrfs_fs_info *fs_info = eb->fs_info; 581 u64 found_start; 582 const u32 csum_size = fs_info->csum_size; 583 u8 found_level; 584 u8 result[BTRFS_CSUM_SIZE]; 585 const u8 *header_csum; 586 int ret = 0; 587 588 found_start = btrfs_header_bytenr(eb); 589 if (found_start != eb->start) { 590 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", 591 eb->start, found_start); 592 ret = -EIO; 593 goto out; 594 } 595 if (check_tree_block_fsid(eb)) { 596 btrfs_err_rl(fs_info, "bad fsid on block %llu", 597 eb->start); 598 ret = -EIO; 599 goto out; 600 } 601 found_level = btrfs_header_level(eb); 602 if (found_level >= BTRFS_MAX_LEVEL) { 603 btrfs_err(fs_info, "bad tree block level %d on %llu", 604 (int)btrfs_header_level(eb), eb->start); 605 ret = -EIO; 606 goto out; 607 } 608 609 csum_tree_block(eb, result); 610 header_csum = page_address(eb->pages[0]) + 611 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum)); 612 613 if (memcmp(result, header_csum, csum_size) != 0) { 614 btrfs_warn_rl(fs_info, 615 "checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", 616 eb->start, 617 CSUM_FMT_VALUE(csum_size, header_csum), 618 CSUM_FMT_VALUE(csum_size, result), 619 btrfs_header_level(eb)); 620 ret = -EUCLEAN; 621 goto out; 622 } 623 624 /* 625 * If this is a leaf block and it is corrupt, set the corrupt bit so 626 * that we don't try and read the other copies of this block, just 627 * return -EIO. 628 */ 629 if (found_level == 0 && btrfs_check_leaf_full(eb)) { 630 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 631 ret = -EIO; 632 } 633 634 if (found_level > 0 && btrfs_check_node(eb)) 635 ret = -EIO; 636 637 if (!ret) 638 set_extent_buffer_uptodate(eb); 639 else 640 btrfs_err(fs_info, 641 "block=%llu read time tree block corruption detected", 642 eb->start); 643 out: 644 return ret; 645 } 646 647 static int validate_subpage_buffer(struct page *page, u64 start, u64 end, 648 int mirror) 649 { 650 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); 651 struct extent_buffer *eb; 652 bool reads_done; 653 int ret = 0; 654 655 /* 656 * We don't allow bio merge for subpage metadata read, so we should 657 * only get one eb for each endio hook. 658 */ 659 ASSERT(end == start + fs_info->nodesize - 1); 660 ASSERT(PagePrivate(page)); 661 662 eb = find_extent_buffer(fs_info, start); 663 /* 664 * When we are reading one tree block, eb must have been inserted into 665 * the radix tree. If not, something is wrong. 666 */ 667 ASSERT(eb); 668 669 reads_done = atomic_dec_and_test(&eb->io_pages); 670 /* Subpage read must finish in page read */ 671 ASSERT(reads_done); 672 673 eb->read_mirror = mirror; 674 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 675 ret = -EIO; 676 goto err; 677 } 678 ret = validate_extent_buffer(eb); 679 if (ret < 0) 680 goto err; 681 682 set_extent_buffer_uptodate(eb); 683 684 free_extent_buffer(eb); 685 return ret; 686 err: 687 /* 688 * end_bio_extent_readpage decrements io_pages in case of error, 689 * make sure it has something to decrement. 690 */ 691 atomic_inc(&eb->io_pages); 692 clear_extent_buffer_uptodate(eb); 693 free_extent_buffer(eb); 694 return ret; 695 } 696 697 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio, 698 struct page *page, u64 start, u64 end, 699 int mirror) 700 { 701 struct extent_buffer *eb; 702 int ret = 0; 703 int reads_done; 704 705 ASSERT(page->private); 706 707 if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) 708 return validate_subpage_buffer(page, start, end, mirror); 709 710 eb = (struct extent_buffer *)page->private; 711 712 /* 713 * The pending IO might have been the only thing that kept this buffer 714 * in memory. Make sure we have a ref for all this other checks 715 */ 716 atomic_inc(&eb->refs); 717 718 reads_done = atomic_dec_and_test(&eb->io_pages); 719 if (!reads_done) 720 goto err; 721 722 eb->read_mirror = mirror; 723 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 724 ret = -EIO; 725 goto err; 726 } 727 ret = validate_extent_buffer(eb); 728 err: 729 if (ret) { 730 /* 731 * our io error hook is going to dec the io pages 732 * again, we have to make sure it has something 733 * to decrement 734 */ 735 atomic_inc(&eb->io_pages); 736 clear_extent_buffer_uptodate(eb); 737 } 738 free_extent_buffer(eb); 739 740 return ret; 741 } 742 743 static void end_workqueue_bio(struct bio *bio) 744 { 745 struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 746 struct btrfs_fs_info *fs_info; 747 struct btrfs_workqueue *wq; 748 749 fs_info = end_io_wq->info; 750 end_io_wq->status = bio->bi_status; 751 752 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 753 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 754 wq = fs_info->endio_meta_write_workers; 755 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 756 wq = fs_info->endio_freespace_worker; 757 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 758 wq = fs_info->endio_raid56_workers; 759 else 760 wq = fs_info->endio_write_workers; 761 } else { 762 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 763 wq = fs_info->endio_raid56_workers; 764 else if (end_io_wq->metadata) 765 wq = fs_info->endio_meta_workers; 766 else 767 wq = fs_info->endio_workers; 768 } 769 770 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 771 btrfs_queue_work(wq, &end_io_wq->work); 772 } 773 774 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 775 enum btrfs_wq_endio_type metadata) 776 { 777 struct btrfs_end_io_wq *end_io_wq; 778 779 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 780 if (!end_io_wq) 781 return BLK_STS_RESOURCE; 782 783 end_io_wq->private = bio->bi_private; 784 end_io_wq->end_io = bio->bi_end_io; 785 end_io_wq->info = info; 786 end_io_wq->status = 0; 787 end_io_wq->bio = bio; 788 end_io_wq->metadata = metadata; 789 790 bio->bi_private = end_io_wq; 791 bio->bi_end_io = end_workqueue_bio; 792 return 0; 793 } 794 795 static void run_one_async_start(struct btrfs_work *work) 796 { 797 struct async_submit_bio *async; 798 blk_status_t ret; 799 800 async = container_of(work, struct async_submit_bio, work); 801 ret = async->submit_bio_start(async->inode, async->bio, 802 async->dio_file_offset); 803 if (ret) 804 async->status = ret; 805 } 806 807 /* 808 * In order to insert checksums into the metadata in large chunks, we wait 809 * until bio submission time. All the pages in the bio are checksummed and 810 * sums are attached onto the ordered extent record. 811 * 812 * At IO completion time the csums attached on the ordered extent record are 813 * inserted into the tree. 814 */ 815 static void run_one_async_done(struct btrfs_work *work) 816 { 817 struct async_submit_bio *async; 818 struct inode *inode; 819 blk_status_t ret; 820 821 async = container_of(work, struct async_submit_bio, work); 822 inode = async->inode; 823 824 /* If an error occurred we just want to clean up the bio and move on */ 825 if (async->status) { 826 async->bio->bi_status = async->status; 827 bio_endio(async->bio); 828 return; 829 } 830 831 /* 832 * All of the bios that pass through here are from async helpers. 833 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. 834 * This changes nothing when cgroups aren't in use. 835 */ 836 async->bio->bi_opf |= REQ_CGROUP_PUNT; 837 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); 838 if (ret) { 839 async->bio->bi_status = ret; 840 bio_endio(async->bio); 841 } 842 } 843 844 static void run_one_async_free(struct btrfs_work *work) 845 { 846 struct async_submit_bio *async; 847 848 async = container_of(work, struct async_submit_bio, work); 849 kfree(async); 850 } 851 852 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, 853 int mirror_num, unsigned long bio_flags, 854 u64 dio_file_offset, 855 extent_submit_bio_start_t *submit_bio_start) 856 { 857 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 858 struct async_submit_bio *async; 859 860 async = kmalloc(sizeof(*async), GFP_NOFS); 861 if (!async) 862 return BLK_STS_RESOURCE; 863 864 async->inode = inode; 865 async->bio = bio; 866 async->mirror_num = mirror_num; 867 async->submit_bio_start = submit_bio_start; 868 869 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, 870 run_one_async_free); 871 872 async->dio_file_offset = dio_file_offset; 873 874 async->status = 0; 875 876 if (op_is_sync(bio->bi_opf)) 877 btrfs_set_work_high_priority(&async->work); 878 879 btrfs_queue_work(fs_info->workers, &async->work); 880 return 0; 881 } 882 883 static blk_status_t btree_csum_one_bio(struct bio *bio) 884 { 885 struct bio_vec *bvec; 886 struct btrfs_root *root; 887 int ret = 0; 888 struct bvec_iter_all iter_all; 889 890 ASSERT(!bio_flagged(bio, BIO_CLONED)); 891 bio_for_each_segment_all(bvec, bio, iter_all) { 892 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 893 ret = csum_dirty_buffer(root->fs_info, bvec); 894 if (ret) 895 break; 896 } 897 898 return errno_to_blk_status(ret); 899 } 900 901 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, 902 u64 dio_file_offset) 903 { 904 /* 905 * when we're called for a write, we're already in the async 906 * submission context. Just jump into btrfs_map_bio 907 */ 908 return btree_csum_one_bio(bio); 909 } 910 911 static bool should_async_write(struct btrfs_fs_info *fs_info, 912 struct btrfs_inode *bi) 913 { 914 if (btrfs_is_zoned(fs_info)) 915 return false; 916 if (atomic_read(&bi->sync_writers)) 917 return false; 918 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) 919 return false; 920 return true; 921 } 922 923 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, 924 int mirror_num, unsigned long bio_flags) 925 { 926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 927 blk_status_t ret; 928 929 if (btrfs_op(bio) != BTRFS_MAP_WRITE) { 930 /* 931 * called for a read, do the setup so that checksum validation 932 * can happen in the async kernel threads 933 */ 934 ret = btrfs_bio_wq_end_io(fs_info, bio, 935 BTRFS_WQ_ENDIO_METADATA); 936 if (ret) 937 goto out_w_error; 938 ret = btrfs_map_bio(fs_info, bio, mirror_num); 939 } else if (!should_async_write(fs_info, BTRFS_I(inode))) { 940 ret = btree_csum_one_bio(bio); 941 if (ret) 942 goto out_w_error; 943 ret = btrfs_map_bio(fs_info, bio, mirror_num); 944 } else { 945 /* 946 * kthread helpers are used to submit writes so that 947 * checksumming can happen in parallel across all CPUs 948 */ 949 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, 950 0, btree_submit_bio_start); 951 } 952 953 if (ret) 954 goto out_w_error; 955 return 0; 956 957 out_w_error: 958 bio->bi_status = ret; 959 bio_endio(bio); 960 return ret; 961 } 962 963 #ifdef CONFIG_MIGRATION 964 static int btree_migratepage(struct address_space *mapping, 965 struct page *newpage, struct page *page, 966 enum migrate_mode mode) 967 { 968 /* 969 * we can't safely write a btree page from here, 970 * we haven't done the locking hook 971 */ 972 if (PageDirty(page)) 973 return -EAGAIN; 974 /* 975 * Buffers may be managed in a filesystem specific way. 976 * We must have no buffers or drop them. 977 */ 978 if (page_has_private(page) && 979 !try_to_release_page(page, GFP_KERNEL)) 980 return -EAGAIN; 981 return migrate_page(mapping, newpage, page, mode); 982 } 983 #endif 984 985 986 static int btree_writepages(struct address_space *mapping, 987 struct writeback_control *wbc) 988 { 989 struct btrfs_fs_info *fs_info; 990 int ret; 991 992 if (wbc->sync_mode == WB_SYNC_NONE) { 993 994 if (wbc->for_kupdate) 995 return 0; 996 997 fs_info = BTRFS_I(mapping->host)->root->fs_info; 998 /* this is a bit racy, but that's ok */ 999 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 1000 BTRFS_DIRTY_METADATA_THRESH, 1001 fs_info->dirty_metadata_batch); 1002 if (ret < 0) 1003 return 0; 1004 } 1005 return btree_write_cache_pages(mapping, wbc); 1006 } 1007 1008 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 1009 { 1010 if (PageWriteback(page) || PageDirty(page)) 1011 return 0; 1012 1013 return try_release_extent_buffer(page); 1014 } 1015 1016 static void btree_invalidate_folio(struct folio *folio, size_t offset, 1017 size_t length) 1018 { 1019 struct extent_io_tree *tree; 1020 tree = &BTRFS_I(folio->mapping->host)->io_tree; 1021 extent_invalidate_folio(tree, folio, offset); 1022 btree_releasepage(&folio->page, GFP_NOFS); 1023 if (folio_get_private(folio)) { 1024 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info, 1025 "folio private not zero on folio %llu", 1026 (unsigned long long)folio_pos(folio)); 1027 folio_detach_private(folio); 1028 } 1029 } 1030 1031 #ifdef DEBUG 1032 static bool btree_dirty_folio(struct address_space *mapping, 1033 struct folio *folio) 1034 { 1035 struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); 1036 struct btrfs_subpage *subpage; 1037 struct extent_buffer *eb; 1038 int cur_bit = 0; 1039 u64 page_start = folio_pos(folio); 1040 1041 if (fs_info->sectorsize == PAGE_SIZE) { 1042 eb = folio_get_private(folio); 1043 BUG_ON(!eb); 1044 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1045 BUG_ON(!atomic_read(&eb->refs)); 1046 btrfs_assert_tree_write_locked(eb); 1047 return filemap_dirty_folio(mapping, folio); 1048 } 1049 subpage = folio_get_private(folio); 1050 1051 ASSERT(subpage->dirty_bitmap); 1052 while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) { 1053 unsigned long flags; 1054 u64 cur; 1055 u16 tmp = (1 << cur_bit); 1056 1057 spin_lock_irqsave(&subpage->lock, flags); 1058 if (!(tmp & subpage->dirty_bitmap)) { 1059 spin_unlock_irqrestore(&subpage->lock, flags); 1060 cur_bit++; 1061 continue; 1062 } 1063 spin_unlock_irqrestore(&subpage->lock, flags); 1064 cur = page_start + cur_bit * fs_info->sectorsize; 1065 1066 eb = find_extent_buffer(fs_info, cur); 1067 ASSERT(eb); 1068 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1069 ASSERT(atomic_read(&eb->refs)); 1070 btrfs_assert_tree_write_locked(eb); 1071 free_extent_buffer(eb); 1072 1073 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits); 1074 } 1075 return filemap_dirty_folio(mapping, folio); 1076 } 1077 #else 1078 #define btree_dirty_folio filemap_dirty_folio 1079 #endif 1080 1081 static const struct address_space_operations btree_aops = { 1082 .writepages = btree_writepages, 1083 .releasepage = btree_releasepage, 1084 .invalidate_folio = btree_invalidate_folio, 1085 #ifdef CONFIG_MIGRATION 1086 .migratepage = btree_migratepage, 1087 #endif 1088 .dirty_folio = btree_dirty_folio, 1089 }; 1090 1091 struct extent_buffer *btrfs_find_create_tree_block( 1092 struct btrfs_fs_info *fs_info, 1093 u64 bytenr, u64 owner_root, 1094 int level) 1095 { 1096 if (btrfs_is_testing(fs_info)) 1097 return alloc_test_extent_buffer(fs_info, bytenr); 1098 return alloc_extent_buffer(fs_info, bytenr, owner_root, level); 1099 } 1100 1101 /* 1102 * Read tree block at logical address @bytenr and do variant basic but critical 1103 * verification. 1104 * 1105 * @owner_root: the objectid of the root owner for this block. 1106 * @parent_transid: expected transid of this tree block, skip check if 0 1107 * @level: expected level, mandatory check 1108 * @first_key: expected key in slot 0, skip check if NULL 1109 */ 1110 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1111 u64 owner_root, u64 parent_transid, 1112 int level, struct btrfs_key *first_key) 1113 { 1114 struct extent_buffer *buf = NULL; 1115 int ret; 1116 1117 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); 1118 if (IS_ERR(buf)) 1119 return buf; 1120 1121 ret = btree_read_extent_buffer_pages(buf, parent_transid, 1122 level, first_key); 1123 if (ret) { 1124 free_extent_buffer_stale(buf); 1125 return ERR_PTR(ret); 1126 } 1127 return buf; 1128 1129 } 1130 1131 void btrfs_clean_tree_block(struct extent_buffer *buf) 1132 { 1133 struct btrfs_fs_info *fs_info = buf->fs_info; 1134 if (btrfs_header_generation(buf) == 1135 fs_info->running_transaction->transid) { 1136 btrfs_assert_tree_write_locked(buf); 1137 1138 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1139 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1140 -buf->len, 1141 fs_info->dirty_metadata_batch); 1142 clear_extent_buffer_dirty(buf); 1143 } 1144 } 1145 } 1146 1147 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1148 u64 objectid) 1149 { 1150 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1151 1152 memset(&root->root_key, 0, sizeof(root->root_key)); 1153 memset(&root->root_item, 0, sizeof(root->root_item)); 1154 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 1155 root->fs_info = fs_info; 1156 root->root_key.objectid = objectid; 1157 root->node = NULL; 1158 root->commit_root = NULL; 1159 root->state = 0; 1160 RB_CLEAR_NODE(&root->rb_node); 1161 1162 root->last_trans = 0; 1163 root->free_objectid = 0; 1164 root->nr_delalloc_inodes = 0; 1165 root->nr_ordered_extents = 0; 1166 root->inode_tree = RB_ROOT; 1167 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1168 1169 btrfs_init_root_block_rsv(root); 1170 1171 INIT_LIST_HEAD(&root->dirty_list); 1172 INIT_LIST_HEAD(&root->root_list); 1173 INIT_LIST_HEAD(&root->delalloc_inodes); 1174 INIT_LIST_HEAD(&root->delalloc_root); 1175 INIT_LIST_HEAD(&root->ordered_extents); 1176 INIT_LIST_HEAD(&root->ordered_root); 1177 INIT_LIST_HEAD(&root->reloc_dirty_list); 1178 INIT_LIST_HEAD(&root->logged_list[0]); 1179 INIT_LIST_HEAD(&root->logged_list[1]); 1180 spin_lock_init(&root->inode_lock); 1181 spin_lock_init(&root->delalloc_lock); 1182 spin_lock_init(&root->ordered_extent_lock); 1183 spin_lock_init(&root->accounting_lock); 1184 spin_lock_init(&root->log_extents_lock[0]); 1185 spin_lock_init(&root->log_extents_lock[1]); 1186 spin_lock_init(&root->qgroup_meta_rsv_lock); 1187 mutex_init(&root->objectid_mutex); 1188 mutex_init(&root->log_mutex); 1189 mutex_init(&root->ordered_extent_mutex); 1190 mutex_init(&root->delalloc_mutex); 1191 init_waitqueue_head(&root->qgroup_flush_wait); 1192 init_waitqueue_head(&root->log_writer_wait); 1193 init_waitqueue_head(&root->log_commit_wait[0]); 1194 init_waitqueue_head(&root->log_commit_wait[1]); 1195 INIT_LIST_HEAD(&root->log_ctxs[0]); 1196 INIT_LIST_HEAD(&root->log_ctxs[1]); 1197 atomic_set(&root->log_commit[0], 0); 1198 atomic_set(&root->log_commit[1], 0); 1199 atomic_set(&root->log_writers, 0); 1200 atomic_set(&root->log_batch, 0); 1201 refcount_set(&root->refs, 1); 1202 atomic_set(&root->snapshot_force_cow, 0); 1203 atomic_set(&root->nr_swapfiles, 0); 1204 root->log_transid = 0; 1205 root->log_transid_committed = -1; 1206 root->last_log_commit = 0; 1207 root->anon_dev = 0; 1208 if (!dummy) { 1209 extent_io_tree_init(fs_info, &root->dirty_log_pages, 1210 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); 1211 extent_io_tree_init(fs_info, &root->log_csum_range, 1212 IO_TREE_LOG_CSUM_RANGE, NULL); 1213 } 1214 1215 spin_lock_init(&root->root_item_lock); 1216 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); 1217 #ifdef CONFIG_BTRFS_DEBUG 1218 INIT_LIST_HEAD(&root->leak_list); 1219 spin_lock(&fs_info->fs_roots_radix_lock); 1220 list_add_tail(&root->leak_list, &fs_info->allocated_roots); 1221 spin_unlock(&fs_info->fs_roots_radix_lock); 1222 #endif 1223 } 1224 1225 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 1226 u64 objectid, gfp_t flags) 1227 { 1228 struct btrfs_root *root = kzalloc(sizeof(*root), flags); 1229 if (root) 1230 __setup_root(root, fs_info, objectid); 1231 return root; 1232 } 1233 1234 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1235 /* Should only be used by the testing infrastructure */ 1236 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 1237 { 1238 struct btrfs_root *root; 1239 1240 if (!fs_info) 1241 return ERR_PTR(-EINVAL); 1242 1243 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); 1244 if (!root) 1245 return ERR_PTR(-ENOMEM); 1246 1247 /* We don't use the stripesize in selftest, set it as sectorsize */ 1248 root->alloc_bytenr = 0; 1249 1250 return root; 1251 } 1252 #endif 1253 1254 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node) 1255 { 1256 const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node); 1257 const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node); 1258 1259 return btrfs_comp_cpu_keys(&a->root_key, &b->root_key); 1260 } 1261 1262 static int global_root_key_cmp(const void *k, const struct rb_node *node) 1263 { 1264 const struct btrfs_key *key = k; 1265 const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node); 1266 1267 return btrfs_comp_cpu_keys(key, &root->root_key); 1268 } 1269 1270 int btrfs_global_root_insert(struct btrfs_root *root) 1271 { 1272 struct btrfs_fs_info *fs_info = root->fs_info; 1273 struct rb_node *tmp; 1274 1275 write_lock(&fs_info->global_root_lock); 1276 tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp); 1277 write_unlock(&fs_info->global_root_lock); 1278 ASSERT(!tmp); 1279 1280 return tmp ? -EEXIST : 0; 1281 } 1282 1283 void btrfs_global_root_delete(struct btrfs_root *root) 1284 { 1285 struct btrfs_fs_info *fs_info = root->fs_info; 1286 1287 write_lock(&fs_info->global_root_lock); 1288 rb_erase(&root->rb_node, &fs_info->global_root_tree); 1289 write_unlock(&fs_info->global_root_lock); 1290 } 1291 1292 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info, 1293 struct btrfs_key *key) 1294 { 1295 struct rb_node *node; 1296 struct btrfs_root *root = NULL; 1297 1298 read_lock(&fs_info->global_root_lock); 1299 node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp); 1300 if (node) 1301 root = container_of(node, struct btrfs_root, rb_node); 1302 read_unlock(&fs_info->global_root_lock); 1303 1304 return root; 1305 } 1306 1307 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr) 1308 { 1309 struct btrfs_block_group *block_group; 1310 u64 ret; 1311 1312 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 1313 return 0; 1314 1315 if (bytenr) 1316 block_group = btrfs_lookup_block_group(fs_info, bytenr); 1317 else 1318 block_group = btrfs_lookup_first_block_group(fs_info, bytenr); 1319 ASSERT(block_group); 1320 if (!block_group) 1321 return 0; 1322 ret = block_group->global_root_id; 1323 btrfs_put_block_group(block_group); 1324 1325 return ret; 1326 } 1327 1328 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1329 { 1330 struct btrfs_key key = { 1331 .objectid = BTRFS_CSUM_TREE_OBJECTID, 1332 .type = BTRFS_ROOT_ITEM_KEY, 1333 .offset = btrfs_global_root_id(fs_info, bytenr), 1334 }; 1335 1336 return btrfs_global_root(fs_info, &key); 1337 } 1338 1339 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr) 1340 { 1341 struct btrfs_key key = { 1342 .objectid = BTRFS_EXTENT_TREE_OBJECTID, 1343 .type = BTRFS_ROOT_ITEM_KEY, 1344 .offset = btrfs_global_root_id(fs_info, bytenr), 1345 }; 1346 1347 return btrfs_global_root(fs_info, &key); 1348 } 1349 1350 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 1351 u64 objectid) 1352 { 1353 struct btrfs_fs_info *fs_info = trans->fs_info; 1354 struct extent_buffer *leaf; 1355 struct btrfs_root *tree_root = fs_info->tree_root; 1356 struct btrfs_root *root; 1357 struct btrfs_key key; 1358 unsigned int nofs_flag; 1359 int ret = 0; 1360 1361 /* 1362 * We're holding a transaction handle, so use a NOFS memory allocation 1363 * context to avoid deadlock if reclaim happens. 1364 */ 1365 nofs_flag = memalloc_nofs_save(); 1366 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); 1367 memalloc_nofs_restore(nofs_flag); 1368 if (!root) 1369 return ERR_PTR(-ENOMEM); 1370 1371 root->root_key.objectid = objectid; 1372 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1373 root->root_key.offset = 0; 1374 1375 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 1376 BTRFS_NESTING_NORMAL); 1377 if (IS_ERR(leaf)) { 1378 ret = PTR_ERR(leaf); 1379 leaf = NULL; 1380 goto fail_unlock; 1381 } 1382 1383 root->node = leaf; 1384 btrfs_mark_buffer_dirty(leaf); 1385 1386 root->commit_root = btrfs_root_node(root); 1387 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 1388 1389 btrfs_set_root_flags(&root->root_item, 0); 1390 btrfs_set_root_limit(&root->root_item, 0); 1391 btrfs_set_root_bytenr(&root->root_item, leaf->start); 1392 btrfs_set_root_generation(&root->root_item, trans->transid); 1393 btrfs_set_root_level(&root->root_item, 0); 1394 btrfs_set_root_refs(&root->root_item, 1); 1395 btrfs_set_root_used(&root->root_item, leaf->len); 1396 btrfs_set_root_last_snapshot(&root->root_item, 0); 1397 btrfs_set_root_dirid(&root->root_item, 0); 1398 if (is_fstree(objectid)) 1399 generate_random_guid(root->root_item.uuid); 1400 else 1401 export_guid(root->root_item.uuid, &guid_null); 1402 btrfs_set_root_drop_level(&root->root_item, 0); 1403 1404 btrfs_tree_unlock(leaf); 1405 1406 key.objectid = objectid; 1407 key.type = BTRFS_ROOT_ITEM_KEY; 1408 key.offset = 0; 1409 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 1410 if (ret) 1411 goto fail; 1412 1413 return root; 1414 1415 fail_unlock: 1416 if (leaf) 1417 btrfs_tree_unlock(leaf); 1418 fail: 1419 btrfs_put_root(root); 1420 1421 return ERR_PTR(ret); 1422 } 1423 1424 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1425 struct btrfs_fs_info *fs_info) 1426 { 1427 struct btrfs_root *root; 1428 1429 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); 1430 if (!root) 1431 return ERR_PTR(-ENOMEM); 1432 1433 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1434 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1435 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1436 1437 return root; 1438 } 1439 1440 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans, 1441 struct btrfs_root *root) 1442 { 1443 struct extent_buffer *leaf; 1444 1445 /* 1446 * DON'T set SHAREABLE bit for log trees. 1447 * 1448 * Log trees are not exposed to user space thus can't be snapshotted, 1449 * and they go away before a real commit is actually done. 1450 * 1451 * They do store pointers to file data extents, and those reference 1452 * counts still get updated (along with back refs to the log tree). 1453 */ 1454 1455 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 1456 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); 1457 if (IS_ERR(leaf)) 1458 return PTR_ERR(leaf); 1459 1460 root->node = leaf; 1461 1462 btrfs_mark_buffer_dirty(root->node); 1463 btrfs_tree_unlock(root->node); 1464 1465 return 0; 1466 } 1467 1468 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1469 struct btrfs_fs_info *fs_info) 1470 { 1471 struct btrfs_root *log_root; 1472 1473 log_root = alloc_log_tree(trans, fs_info); 1474 if (IS_ERR(log_root)) 1475 return PTR_ERR(log_root); 1476 1477 if (!btrfs_is_zoned(fs_info)) { 1478 int ret = btrfs_alloc_log_tree_node(trans, log_root); 1479 1480 if (ret) { 1481 btrfs_put_root(log_root); 1482 return ret; 1483 } 1484 } 1485 1486 WARN_ON(fs_info->log_root_tree); 1487 fs_info->log_root_tree = log_root; 1488 return 0; 1489 } 1490 1491 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1492 struct btrfs_root *root) 1493 { 1494 struct btrfs_fs_info *fs_info = root->fs_info; 1495 struct btrfs_root *log_root; 1496 struct btrfs_inode_item *inode_item; 1497 int ret; 1498 1499 log_root = alloc_log_tree(trans, fs_info); 1500 if (IS_ERR(log_root)) 1501 return PTR_ERR(log_root); 1502 1503 ret = btrfs_alloc_log_tree_node(trans, log_root); 1504 if (ret) { 1505 btrfs_put_root(log_root); 1506 return ret; 1507 } 1508 1509 log_root->last_trans = trans->transid; 1510 log_root->root_key.offset = root->root_key.objectid; 1511 1512 inode_item = &log_root->root_item.inode; 1513 btrfs_set_stack_inode_generation(inode_item, 1); 1514 btrfs_set_stack_inode_size(inode_item, 3); 1515 btrfs_set_stack_inode_nlink(inode_item, 1); 1516 btrfs_set_stack_inode_nbytes(inode_item, 1517 fs_info->nodesize); 1518 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1519 1520 btrfs_set_root_node(&log_root->root_item, log_root->node); 1521 1522 WARN_ON(root->log_root); 1523 root->log_root = log_root; 1524 root->log_transid = 0; 1525 root->log_transid_committed = -1; 1526 root->last_log_commit = 0; 1527 return 0; 1528 } 1529 1530 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, 1531 struct btrfs_path *path, 1532 struct btrfs_key *key) 1533 { 1534 struct btrfs_root *root; 1535 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1536 u64 generation; 1537 int ret; 1538 int level; 1539 1540 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); 1541 if (!root) 1542 return ERR_PTR(-ENOMEM); 1543 1544 ret = btrfs_find_root(tree_root, key, path, 1545 &root->root_item, &root->root_key); 1546 if (ret) { 1547 if (ret > 0) 1548 ret = -ENOENT; 1549 goto fail; 1550 } 1551 1552 generation = btrfs_root_generation(&root->root_item); 1553 level = btrfs_root_level(&root->root_item); 1554 root->node = read_tree_block(fs_info, 1555 btrfs_root_bytenr(&root->root_item), 1556 key->objectid, generation, level, NULL); 1557 if (IS_ERR(root->node)) { 1558 ret = PTR_ERR(root->node); 1559 root->node = NULL; 1560 goto fail; 1561 } 1562 if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1563 ret = -EIO; 1564 goto fail; 1565 } 1566 root->commit_root = btrfs_root_node(root); 1567 return root; 1568 fail: 1569 btrfs_put_root(root); 1570 return ERR_PTR(ret); 1571 } 1572 1573 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1574 struct btrfs_key *key) 1575 { 1576 struct btrfs_root *root; 1577 struct btrfs_path *path; 1578 1579 path = btrfs_alloc_path(); 1580 if (!path) 1581 return ERR_PTR(-ENOMEM); 1582 root = read_tree_root_path(tree_root, path, key); 1583 btrfs_free_path(path); 1584 1585 return root; 1586 } 1587 1588 /* 1589 * Initialize subvolume root in-memory structure 1590 * 1591 * @anon_dev: anonymous device to attach to the root, if zero, allocate new 1592 */ 1593 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) 1594 { 1595 int ret; 1596 unsigned int nofs_flag; 1597 1598 /* 1599 * We might be called under a transaction (e.g. indirect backref 1600 * resolution) which could deadlock if it triggers memory reclaim 1601 */ 1602 nofs_flag = memalloc_nofs_save(); 1603 ret = btrfs_drew_lock_init(&root->snapshot_lock); 1604 memalloc_nofs_restore(nofs_flag); 1605 if (ret) 1606 goto fail; 1607 1608 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && 1609 !btrfs_is_data_reloc_root(root)) { 1610 set_bit(BTRFS_ROOT_SHAREABLE, &root->state); 1611 btrfs_check_and_init_root_item(&root->root_item); 1612 } 1613 1614 /* 1615 * Don't assign anonymous block device to roots that are not exposed to 1616 * userspace, the id pool is limited to 1M 1617 */ 1618 if (is_fstree(root->root_key.objectid) && 1619 btrfs_root_refs(&root->root_item) > 0) { 1620 if (!anon_dev) { 1621 ret = get_anon_bdev(&root->anon_dev); 1622 if (ret) 1623 goto fail; 1624 } else { 1625 root->anon_dev = anon_dev; 1626 } 1627 } 1628 1629 mutex_lock(&root->objectid_mutex); 1630 ret = btrfs_init_root_free_objectid(root); 1631 if (ret) { 1632 mutex_unlock(&root->objectid_mutex); 1633 goto fail; 1634 } 1635 1636 ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 1637 1638 mutex_unlock(&root->objectid_mutex); 1639 1640 return 0; 1641 fail: 1642 /* The caller is responsible to call btrfs_free_fs_root */ 1643 return ret; 1644 } 1645 1646 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1647 u64 root_id) 1648 { 1649 struct btrfs_root *root; 1650 1651 spin_lock(&fs_info->fs_roots_radix_lock); 1652 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1653 (unsigned long)root_id); 1654 if (root) 1655 root = btrfs_grab_root(root); 1656 spin_unlock(&fs_info->fs_roots_radix_lock); 1657 return root; 1658 } 1659 1660 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, 1661 u64 objectid) 1662 { 1663 struct btrfs_key key = { 1664 .objectid = objectid, 1665 .type = BTRFS_ROOT_ITEM_KEY, 1666 .offset = 0, 1667 }; 1668 1669 if (objectid == BTRFS_ROOT_TREE_OBJECTID) 1670 return btrfs_grab_root(fs_info->tree_root); 1671 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 1672 return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 1673 if (objectid == BTRFS_CHUNK_TREE_OBJECTID) 1674 return btrfs_grab_root(fs_info->chunk_root); 1675 if (objectid == BTRFS_DEV_TREE_OBJECTID) 1676 return btrfs_grab_root(fs_info->dev_root); 1677 if (objectid == BTRFS_CSUM_TREE_OBJECTID) 1678 return btrfs_grab_root(btrfs_global_root(fs_info, &key)); 1679 if (objectid == BTRFS_QUOTA_TREE_OBJECTID) 1680 return btrfs_grab_root(fs_info->quota_root) ? 1681 fs_info->quota_root : ERR_PTR(-ENOENT); 1682 if (objectid == BTRFS_UUID_TREE_OBJECTID) 1683 return btrfs_grab_root(fs_info->uuid_root) ? 1684 fs_info->uuid_root : ERR_PTR(-ENOENT); 1685 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) { 1686 struct btrfs_root *root = btrfs_global_root(fs_info, &key); 1687 1688 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT); 1689 } 1690 return NULL; 1691 } 1692 1693 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1694 struct btrfs_root *root) 1695 { 1696 int ret; 1697 1698 ret = radix_tree_preload(GFP_NOFS); 1699 if (ret) 1700 return ret; 1701 1702 spin_lock(&fs_info->fs_roots_radix_lock); 1703 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1704 (unsigned long)root->root_key.objectid, 1705 root); 1706 if (ret == 0) { 1707 btrfs_grab_root(root); 1708 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1709 } 1710 spin_unlock(&fs_info->fs_roots_radix_lock); 1711 radix_tree_preload_end(); 1712 1713 return ret; 1714 } 1715 1716 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) 1717 { 1718 #ifdef CONFIG_BTRFS_DEBUG 1719 struct btrfs_root *root; 1720 1721 while (!list_empty(&fs_info->allocated_roots)) { 1722 char buf[BTRFS_ROOT_NAME_BUF_LEN]; 1723 1724 root = list_first_entry(&fs_info->allocated_roots, 1725 struct btrfs_root, leak_list); 1726 btrfs_err(fs_info, "leaked root %s refcount %d", 1727 btrfs_root_name(&root->root_key, buf), 1728 refcount_read(&root->refs)); 1729 while (refcount_read(&root->refs) > 1) 1730 btrfs_put_root(root); 1731 btrfs_put_root(root); 1732 } 1733 #endif 1734 } 1735 1736 static void free_global_roots(struct btrfs_fs_info *fs_info) 1737 { 1738 struct btrfs_root *root; 1739 struct rb_node *node; 1740 1741 while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) { 1742 root = rb_entry(node, struct btrfs_root, rb_node); 1743 rb_erase(&root->rb_node, &fs_info->global_root_tree); 1744 btrfs_put_root(root); 1745 } 1746 } 1747 1748 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) 1749 { 1750 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 1751 percpu_counter_destroy(&fs_info->delalloc_bytes); 1752 percpu_counter_destroy(&fs_info->ordered_bytes); 1753 percpu_counter_destroy(&fs_info->dev_replace.bio_counter); 1754 btrfs_free_csum_hash(fs_info); 1755 btrfs_free_stripe_hash_table(fs_info); 1756 btrfs_free_ref_cache(fs_info); 1757 kfree(fs_info->balance_ctl); 1758 kfree(fs_info->delayed_root); 1759 free_global_roots(fs_info); 1760 btrfs_put_root(fs_info->tree_root); 1761 btrfs_put_root(fs_info->chunk_root); 1762 btrfs_put_root(fs_info->dev_root); 1763 btrfs_put_root(fs_info->quota_root); 1764 btrfs_put_root(fs_info->uuid_root); 1765 btrfs_put_root(fs_info->fs_root); 1766 btrfs_put_root(fs_info->data_reloc_root); 1767 btrfs_put_root(fs_info->block_group_root); 1768 btrfs_check_leaked_roots(fs_info); 1769 btrfs_extent_buffer_leak_debug_check(fs_info); 1770 kfree(fs_info->super_copy); 1771 kfree(fs_info->super_for_commit); 1772 kfree(fs_info->subpage_info); 1773 kvfree(fs_info); 1774 } 1775 1776 1777 /* 1778 * Get an in-memory reference of a root structure. 1779 * 1780 * For essential trees like root/extent tree, we grab it from fs_info directly. 1781 * For subvolume trees, we check the cached filesystem roots first. If not 1782 * found, then read it from disk and add it to cached fs roots. 1783 * 1784 * Caller should release the root by calling btrfs_put_root() after the usage. 1785 * 1786 * NOTE: Reloc and log trees can't be read by this function as they share the 1787 * same root objectid. 1788 * 1789 * @objectid: root id 1790 * @anon_dev: preallocated anonymous block device number for new roots, 1791 * pass 0 for new allocation. 1792 * @check_ref: whether to check root item references, If true, return -ENOENT 1793 * for orphan roots 1794 */ 1795 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, 1796 u64 objectid, dev_t anon_dev, 1797 bool check_ref) 1798 { 1799 struct btrfs_root *root; 1800 struct btrfs_path *path; 1801 struct btrfs_key key; 1802 int ret; 1803 1804 root = btrfs_get_global_root(fs_info, objectid); 1805 if (root) 1806 return root; 1807 again: 1808 root = btrfs_lookup_fs_root(fs_info, objectid); 1809 if (root) { 1810 /* Shouldn't get preallocated anon_dev for cached roots */ 1811 ASSERT(!anon_dev); 1812 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1813 btrfs_put_root(root); 1814 return ERR_PTR(-ENOENT); 1815 } 1816 return root; 1817 } 1818 1819 key.objectid = objectid; 1820 key.type = BTRFS_ROOT_ITEM_KEY; 1821 key.offset = (u64)-1; 1822 root = btrfs_read_tree_root(fs_info->tree_root, &key); 1823 if (IS_ERR(root)) 1824 return root; 1825 1826 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1827 ret = -ENOENT; 1828 goto fail; 1829 } 1830 1831 ret = btrfs_init_fs_root(root, anon_dev); 1832 if (ret) 1833 goto fail; 1834 1835 path = btrfs_alloc_path(); 1836 if (!path) { 1837 ret = -ENOMEM; 1838 goto fail; 1839 } 1840 key.objectid = BTRFS_ORPHAN_OBJECTID; 1841 key.type = BTRFS_ORPHAN_ITEM_KEY; 1842 key.offset = objectid; 1843 1844 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1845 btrfs_free_path(path); 1846 if (ret < 0) 1847 goto fail; 1848 if (ret == 0) 1849 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1850 1851 ret = btrfs_insert_fs_root(fs_info, root); 1852 if (ret) { 1853 btrfs_put_root(root); 1854 if (ret == -EEXIST) 1855 goto again; 1856 goto fail; 1857 } 1858 return root; 1859 fail: 1860 /* 1861 * If our caller provided us an anonymous device, then it's his 1862 * responsability to free it in case we fail. So we have to set our 1863 * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root() 1864 * and once again by our caller. 1865 */ 1866 if (anon_dev) 1867 root->anon_dev = 0; 1868 btrfs_put_root(root); 1869 return ERR_PTR(ret); 1870 } 1871 1872 /* 1873 * Get in-memory reference of a root structure 1874 * 1875 * @objectid: tree objectid 1876 * @check_ref: if set, verify that the tree exists and the item has at least 1877 * one reference 1878 */ 1879 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1880 u64 objectid, bool check_ref) 1881 { 1882 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); 1883 } 1884 1885 /* 1886 * Get in-memory reference of a root structure, created as new, optionally pass 1887 * the anonymous block device id 1888 * 1889 * @objectid: tree objectid 1890 * @anon_dev: if zero, allocate a new anonymous block device or use the 1891 * parameter value 1892 */ 1893 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, 1894 u64 objectid, dev_t anon_dev) 1895 { 1896 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); 1897 } 1898 1899 /* 1900 * btrfs_get_fs_root_commit_root - return a root for the given objectid 1901 * @fs_info: the fs_info 1902 * @objectid: the objectid we need to lookup 1903 * 1904 * This is exclusively used for backref walking, and exists specifically because 1905 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref 1906 * creation time, which means we may have to read the tree_root in order to look 1907 * up a fs root that is not in memory. If the root is not in memory we will 1908 * read the tree root commit root and look up the fs root from there. This is a 1909 * temporary root, it will not be inserted into the radix tree as it doesn't 1910 * have the most uptodate information, it'll simply be discarded once the 1911 * backref code is finished using the root. 1912 */ 1913 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, 1914 struct btrfs_path *path, 1915 u64 objectid) 1916 { 1917 struct btrfs_root *root; 1918 struct btrfs_key key; 1919 1920 ASSERT(path->search_commit_root && path->skip_locking); 1921 1922 /* 1923 * This can return -ENOENT if we ask for a root that doesn't exist, but 1924 * since this is called via the backref walking code we won't be looking 1925 * up a root that doesn't exist, unless there's corruption. So if root 1926 * != NULL just return it. 1927 */ 1928 root = btrfs_get_global_root(fs_info, objectid); 1929 if (root) 1930 return root; 1931 1932 root = btrfs_lookup_fs_root(fs_info, objectid); 1933 if (root) 1934 return root; 1935 1936 key.objectid = objectid; 1937 key.type = BTRFS_ROOT_ITEM_KEY; 1938 key.offset = (u64)-1; 1939 root = read_tree_root_path(fs_info->tree_root, path, &key); 1940 btrfs_release_path(path); 1941 1942 return root; 1943 } 1944 1945 /* 1946 * called by the kthread helper functions to finally call the bio end_io 1947 * functions. This is where read checksum verification actually happens 1948 */ 1949 static void end_workqueue_fn(struct btrfs_work *work) 1950 { 1951 struct bio *bio; 1952 struct btrfs_end_io_wq *end_io_wq; 1953 1954 end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1955 bio = end_io_wq->bio; 1956 1957 bio->bi_status = end_io_wq->status; 1958 bio->bi_private = end_io_wq->private; 1959 bio->bi_end_io = end_io_wq->end_io; 1960 bio_endio(bio); 1961 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1962 } 1963 1964 static int cleaner_kthread(void *arg) 1965 { 1966 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)arg; 1967 int again; 1968 1969 while (1) { 1970 again = 0; 1971 1972 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 1973 1974 /* Make the cleaner go to sleep early. */ 1975 if (btrfs_need_cleaner_sleep(fs_info)) 1976 goto sleep; 1977 1978 /* 1979 * Do not do anything if we might cause open_ctree() to block 1980 * before we have finished mounting the filesystem. 1981 */ 1982 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1983 goto sleep; 1984 1985 if (!mutex_trylock(&fs_info->cleaner_mutex)) 1986 goto sleep; 1987 1988 /* 1989 * Avoid the problem that we change the status of the fs 1990 * during the above check and trylock. 1991 */ 1992 if (btrfs_need_cleaner_sleep(fs_info)) { 1993 mutex_unlock(&fs_info->cleaner_mutex); 1994 goto sleep; 1995 } 1996 1997 btrfs_run_delayed_iputs(fs_info); 1998 1999 again = btrfs_clean_one_deleted_snapshot(fs_info); 2000 mutex_unlock(&fs_info->cleaner_mutex); 2001 2002 /* 2003 * The defragger has dealt with the R/O remount and umount, 2004 * needn't do anything special here. 2005 */ 2006 btrfs_run_defrag_inodes(fs_info); 2007 2008 /* 2009 * Acquires fs_info->reclaim_bgs_lock to avoid racing 2010 * with relocation (btrfs_relocate_chunk) and relocation 2011 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 2012 * after acquiring fs_info->reclaim_bgs_lock. So we 2013 * can't hold, nor need to, fs_info->cleaner_mutex when deleting 2014 * unused block groups. 2015 */ 2016 btrfs_delete_unused_bgs(fs_info); 2017 2018 /* 2019 * Reclaim block groups in the reclaim_bgs list after we deleted 2020 * all unused block_groups. This possibly gives us some more free 2021 * space. 2022 */ 2023 btrfs_reclaim_bgs(fs_info); 2024 sleep: 2025 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); 2026 if (kthread_should_park()) 2027 kthread_parkme(); 2028 if (kthread_should_stop()) 2029 return 0; 2030 if (!again) { 2031 set_current_state(TASK_INTERRUPTIBLE); 2032 schedule(); 2033 __set_current_state(TASK_RUNNING); 2034 } 2035 } 2036 } 2037 2038 static int transaction_kthread(void *arg) 2039 { 2040 struct btrfs_root *root = arg; 2041 struct btrfs_fs_info *fs_info = root->fs_info; 2042 struct btrfs_trans_handle *trans; 2043 struct btrfs_transaction *cur; 2044 u64 transid; 2045 time64_t delta; 2046 unsigned long delay; 2047 bool cannot_commit; 2048 2049 do { 2050 cannot_commit = false; 2051 delay = msecs_to_jiffies(fs_info->commit_interval * 1000); 2052 mutex_lock(&fs_info->transaction_kthread_mutex); 2053 2054 spin_lock(&fs_info->trans_lock); 2055 cur = fs_info->running_transaction; 2056 if (!cur) { 2057 spin_unlock(&fs_info->trans_lock); 2058 goto sleep; 2059 } 2060 2061 delta = ktime_get_seconds() - cur->start_time; 2062 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) && 2063 cur->state < TRANS_STATE_COMMIT_START && 2064 delta < fs_info->commit_interval) { 2065 spin_unlock(&fs_info->trans_lock); 2066 delay -= msecs_to_jiffies((delta - 1) * 1000); 2067 delay = min(delay, 2068 msecs_to_jiffies(fs_info->commit_interval * 1000)); 2069 goto sleep; 2070 } 2071 transid = cur->transid; 2072 spin_unlock(&fs_info->trans_lock); 2073 2074 /* If the file system is aborted, this will always fail. */ 2075 trans = btrfs_attach_transaction(root); 2076 if (IS_ERR(trans)) { 2077 if (PTR_ERR(trans) != -ENOENT) 2078 cannot_commit = true; 2079 goto sleep; 2080 } 2081 if (transid == trans->transid) { 2082 btrfs_commit_transaction(trans); 2083 } else { 2084 btrfs_end_transaction(trans); 2085 } 2086 sleep: 2087 wake_up_process(fs_info->cleaner_kthread); 2088 mutex_unlock(&fs_info->transaction_kthread_mutex); 2089 2090 if (BTRFS_FS_ERROR(fs_info)) 2091 btrfs_cleanup_transaction(fs_info); 2092 if (!kthread_should_stop() && 2093 (!btrfs_transaction_blocked(fs_info) || 2094 cannot_commit)) 2095 schedule_timeout_interruptible(delay); 2096 } while (!kthread_should_stop()); 2097 return 0; 2098 } 2099 2100 /* 2101 * This will find the highest generation in the array of root backups. The 2102 * index of the highest array is returned, or -EINVAL if we can't find 2103 * anything. 2104 * 2105 * We check to make sure the array is valid by comparing the 2106 * generation of the latest root in the array with the generation 2107 * in the super block. If they don't match we pitch it. 2108 */ 2109 static int find_newest_super_backup(struct btrfs_fs_info *info) 2110 { 2111 const u64 newest_gen = btrfs_super_generation(info->super_copy); 2112 u64 cur; 2113 struct btrfs_root_backup *root_backup; 2114 int i; 2115 2116 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 2117 root_backup = info->super_copy->super_roots + i; 2118 cur = btrfs_backup_tree_root_gen(root_backup); 2119 if (cur == newest_gen) 2120 return i; 2121 } 2122 2123 return -EINVAL; 2124 } 2125 2126 /* 2127 * copy all the root pointers into the super backup array. 2128 * this will bump the backup pointer by one when it is 2129 * done 2130 */ 2131 static void backup_super_roots(struct btrfs_fs_info *info) 2132 { 2133 const int next_backup = info->backup_root_index; 2134 struct btrfs_root_backup *root_backup; 2135 2136 root_backup = info->super_for_commit->super_roots + next_backup; 2137 2138 /* 2139 * make sure all of our padding and empty slots get zero filled 2140 * regardless of which ones we use today 2141 */ 2142 memset(root_backup, 0, sizeof(*root_backup)); 2143 2144 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 2145 2146 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 2147 btrfs_set_backup_tree_root_gen(root_backup, 2148 btrfs_header_generation(info->tree_root->node)); 2149 2150 btrfs_set_backup_tree_root_level(root_backup, 2151 btrfs_header_level(info->tree_root->node)); 2152 2153 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 2154 btrfs_set_backup_chunk_root_gen(root_backup, 2155 btrfs_header_generation(info->chunk_root->node)); 2156 btrfs_set_backup_chunk_root_level(root_backup, 2157 btrfs_header_level(info->chunk_root->node)); 2158 2159 if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) { 2160 btrfs_set_backup_block_group_root(root_backup, 2161 info->block_group_root->node->start); 2162 btrfs_set_backup_block_group_root_gen(root_backup, 2163 btrfs_header_generation(info->block_group_root->node)); 2164 btrfs_set_backup_block_group_root_level(root_backup, 2165 btrfs_header_level(info->block_group_root->node)); 2166 } else { 2167 struct btrfs_root *extent_root = btrfs_extent_root(info, 0); 2168 struct btrfs_root *csum_root = btrfs_csum_root(info, 0); 2169 2170 btrfs_set_backup_extent_root(root_backup, 2171 extent_root->node->start); 2172 btrfs_set_backup_extent_root_gen(root_backup, 2173 btrfs_header_generation(extent_root->node)); 2174 btrfs_set_backup_extent_root_level(root_backup, 2175 btrfs_header_level(extent_root->node)); 2176 2177 btrfs_set_backup_csum_root(root_backup, csum_root->node->start); 2178 btrfs_set_backup_csum_root_gen(root_backup, 2179 btrfs_header_generation(csum_root->node)); 2180 btrfs_set_backup_csum_root_level(root_backup, 2181 btrfs_header_level(csum_root->node)); 2182 } 2183 2184 /* 2185 * we might commit during log recovery, which happens before we set 2186 * the fs_root. Make sure it is valid before we fill it in. 2187 */ 2188 if (info->fs_root && info->fs_root->node) { 2189 btrfs_set_backup_fs_root(root_backup, 2190 info->fs_root->node->start); 2191 btrfs_set_backup_fs_root_gen(root_backup, 2192 btrfs_header_generation(info->fs_root->node)); 2193 btrfs_set_backup_fs_root_level(root_backup, 2194 btrfs_header_level(info->fs_root->node)); 2195 } 2196 2197 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 2198 btrfs_set_backup_dev_root_gen(root_backup, 2199 btrfs_header_generation(info->dev_root->node)); 2200 btrfs_set_backup_dev_root_level(root_backup, 2201 btrfs_header_level(info->dev_root->node)); 2202 2203 btrfs_set_backup_total_bytes(root_backup, 2204 btrfs_super_total_bytes(info->super_copy)); 2205 btrfs_set_backup_bytes_used(root_backup, 2206 btrfs_super_bytes_used(info->super_copy)); 2207 btrfs_set_backup_num_devices(root_backup, 2208 btrfs_super_num_devices(info->super_copy)); 2209 2210 /* 2211 * if we don't copy this out to the super_copy, it won't get remembered 2212 * for the next commit 2213 */ 2214 memcpy(&info->super_copy->super_roots, 2215 &info->super_for_commit->super_roots, 2216 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2217 } 2218 2219 /* 2220 * read_backup_root - Reads a backup root based on the passed priority. Prio 0 2221 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots 2222 * 2223 * fs_info - filesystem whose backup roots need to be read 2224 * priority - priority of backup root required 2225 * 2226 * Returns backup root index on success and -EINVAL otherwise. 2227 */ 2228 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) 2229 { 2230 int backup_index = find_newest_super_backup(fs_info); 2231 struct btrfs_super_block *super = fs_info->super_copy; 2232 struct btrfs_root_backup *root_backup; 2233 2234 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { 2235 if (priority == 0) 2236 return backup_index; 2237 2238 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; 2239 backup_index %= BTRFS_NUM_BACKUP_ROOTS; 2240 } else { 2241 return -EINVAL; 2242 } 2243 2244 root_backup = super->super_roots + backup_index; 2245 2246 btrfs_set_super_generation(super, 2247 btrfs_backup_tree_root_gen(root_backup)); 2248 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2249 btrfs_set_super_root_level(super, 2250 btrfs_backup_tree_root_level(root_backup)); 2251 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2252 2253 /* 2254 * Fixme: the total bytes and num_devices need to match or we should 2255 * need a fsck 2256 */ 2257 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2258 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2259 2260 return backup_index; 2261 } 2262 2263 /* helper to cleanup workers */ 2264 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 2265 { 2266 btrfs_destroy_workqueue(fs_info->fixup_workers); 2267 btrfs_destroy_workqueue(fs_info->delalloc_workers); 2268 btrfs_destroy_workqueue(fs_info->workers); 2269 btrfs_destroy_workqueue(fs_info->endio_workers); 2270 btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2271 btrfs_destroy_workqueue(fs_info->rmw_workers); 2272 btrfs_destroy_workqueue(fs_info->endio_write_workers); 2273 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2274 btrfs_destroy_workqueue(fs_info->delayed_workers); 2275 btrfs_destroy_workqueue(fs_info->caching_workers); 2276 btrfs_destroy_workqueue(fs_info->flush_workers); 2277 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2278 if (fs_info->discard_ctl.discard_workers) 2279 destroy_workqueue(fs_info->discard_ctl.discard_workers); 2280 /* 2281 * Now that all other work queues are destroyed, we can safely destroy 2282 * the queues used for metadata I/O, since tasks from those other work 2283 * queues can do metadata I/O operations. 2284 */ 2285 btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2286 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2287 } 2288 2289 static void free_root_extent_buffers(struct btrfs_root *root) 2290 { 2291 if (root) { 2292 free_extent_buffer(root->node); 2293 free_extent_buffer(root->commit_root); 2294 root->node = NULL; 2295 root->commit_root = NULL; 2296 } 2297 } 2298 2299 static void free_global_root_pointers(struct btrfs_fs_info *fs_info) 2300 { 2301 struct btrfs_root *root, *tmp; 2302 2303 rbtree_postorder_for_each_entry_safe(root, tmp, 2304 &fs_info->global_root_tree, 2305 rb_node) 2306 free_root_extent_buffers(root); 2307 } 2308 2309 /* helper to cleanup tree roots */ 2310 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) 2311 { 2312 free_root_extent_buffers(info->tree_root); 2313 2314 free_global_root_pointers(info); 2315 free_root_extent_buffers(info->dev_root); 2316 free_root_extent_buffers(info->quota_root); 2317 free_root_extent_buffers(info->uuid_root); 2318 free_root_extent_buffers(info->fs_root); 2319 free_root_extent_buffers(info->data_reloc_root); 2320 free_root_extent_buffers(info->block_group_root); 2321 if (free_chunk_root) 2322 free_root_extent_buffers(info->chunk_root); 2323 } 2324 2325 void btrfs_put_root(struct btrfs_root *root) 2326 { 2327 if (!root) 2328 return; 2329 2330 if (refcount_dec_and_test(&root->refs)) { 2331 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 2332 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); 2333 if (root->anon_dev) 2334 free_anon_bdev(root->anon_dev); 2335 btrfs_drew_lock_destroy(&root->snapshot_lock); 2336 free_root_extent_buffers(root); 2337 #ifdef CONFIG_BTRFS_DEBUG 2338 spin_lock(&root->fs_info->fs_roots_radix_lock); 2339 list_del_init(&root->leak_list); 2340 spin_unlock(&root->fs_info->fs_roots_radix_lock); 2341 #endif 2342 kfree(root); 2343 } 2344 } 2345 2346 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2347 { 2348 int ret; 2349 struct btrfs_root *gang[8]; 2350 int i; 2351 2352 while (!list_empty(&fs_info->dead_roots)) { 2353 gang[0] = list_entry(fs_info->dead_roots.next, 2354 struct btrfs_root, root_list); 2355 list_del(&gang[0]->root_list); 2356 2357 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) 2358 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2359 btrfs_put_root(gang[0]); 2360 } 2361 2362 while (1) { 2363 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2364 (void **)gang, 0, 2365 ARRAY_SIZE(gang)); 2366 if (!ret) 2367 break; 2368 for (i = 0; i < ret; i++) 2369 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2370 } 2371 } 2372 2373 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2374 { 2375 mutex_init(&fs_info->scrub_lock); 2376 atomic_set(&fs_info->scrubs_running, 0); 2377 atomic_set(&fs_info->scrub_pause_req, 0); 2378 atomic_set(&fs_info->scrubs_paused, 0); 2379 atomic_set(&fs_info->scrub_cancel_req, 0); 2380 init_waitqueue_head(&fs_info->scrub_pause_wait); 2381 refcount_set(&fs_info->scrub_workers_refcnt, 0); 2382 } 2383 2384 static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2385 { 2386 spin_lock_init(&fs_info->balance_lock); 2387 mutex_init(&fs_info->balance_mutex); 2388 atomic_set(&fs_info->balance_pause_req, 0); 2389 atomic_set(&fs_info->balance_cancel_req, 0); 2390 fs_info->balance_ctl = NULL; 2391 init_waitqueue_head(&fs_info->balance_wait_q); 2392 atomic_set(&fs_info->reloc_cancel_req, 0); 2393 } 2394 2395 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2396 { 2397 struct inode *inode = fs_info->btree_inode; 2398 2399 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2400 set_nlink(inode, 1); 2401 /* 2402 * we set the i_size on the btree inode to the max possible int. 2403 * the real end of the address space is determined by all of 2404 * the devices in the system 2405 */ 2406 inode->i_size = OFFSET_MAX; 2407 inode->i_mapping->a_ops = &btree_aops; 2408 2409 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 2410 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, 2411 IO_TREE_BTREE_INODE_IO, inode); 2412 BTRFS_I(inode)->io_tree.track_uptodate = false; 2413 extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2414 2415 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); 2416 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 2417 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2418 btrfs_insert_inode_hash(inode); 2419 } 2420 2421 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2422 { 2423 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2424 init_rwsem(&fs_info->dev_replace.rwsem); 2425 init_waitqueue_head(&fs_info->dev_replace.replace_wait); 2426 } 2427 2428 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2429 { 2430 spin_lock_init(&fs_info->qgroup_lock); 2431 mutex_init(&fs_info->qgroup_ioctl_lock); 2432 fs_info->qgroup_tree = RB_ROOT; 2433 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2434 fs_info->qgroup_seq = 1; 2435 fs_info->qgroup_ulist = NULL; 2436 fs_info->qgroup_rescan_running = false; 2437 mutex_init(&fs_info->qgroup_rescan_lock); 2438 } 2439 2440 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info) 2441 { 2442 u32 max_active = fs_info->thread_pool_size; 2443 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 2444 2445 fs_info->workers = 2446 btrfs_alloc_workqueue(fs_info, "worker", 2447 flags | WQ_HIGHPRI, max_active, 16); 2448 2449 fs_info->delalloc_workers = 2450 btrfs_alloc_workqueue(fs_info, "delalloc", 2451 flags, max_active, 2); 2452 2453 fs_info->flush_workers = 2454 btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2455 flags, max_active, 0); 2456 2457 fs_info->caching_workers = 2458 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 2459 2460 fs_info->fixup_workers = 2461 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 2462 2463 /* 2464 * endios are largely parallel and should have a very 2465 * low idle thresh 2466 */ 2467 fs_info->endio_workers = 2468 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 2469 fs_info->endio_meta_workers = 2470 btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2471 max_active, 4); 2472 fs_info->endio_meta_write_workers = 2473 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2474 max_active, 2); 2475 fs_info->endio_raid56_workers = 2476 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2477 max_active, 4); 2478 fs_info->rmw_workers = 2479 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 2480 fs_info->endio_write_workers = 2481 btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2482 max_active, 2); 2483 fs_info->endio_freespace_worker = 2484 btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2485 max_active, 0); 2486 fs_info->delayed_workers = 2487 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2488 max_active, 0); 2489 fs_info->qgroup_rescan_workers = 2490 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2491 fs_info->discard_ctl.discard_workers = 2492 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); 2493 2494 if (!(fs_info->workers && fs_info->delalloc_workers && 2495 fs_info->flush_workers && 2496 fs_info->endio_workers && fs_info->endio_meta_workers && 2497 fs_info->endio_meta_write_workers && 2498 fs_info->endio_write_workers && fs_info->endio_raid56_workers && 2499 fs_info->endio_freespace_worker && fs_info->rmw_workers && 2500 fs_info->caching_workers && fs_info->fixup_workers && 2501 fs_info->delayed_workers && fs_info->qgroup_rescan_workers && 2502 fs_info->discard_ctl.discard_workers)) { 2503 return -ENOMEM; 2504 } 2505 2506 return 0; 2507 } 2508 2509 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) 2510 { 2511 struct crypto_shash *csum_shash; 2512 const char *csum_driver = btrfs_super_csum_driver(csum_type); 2513 2514 csum_shash = crypto_alloc_shash(csum_driver, 0, 0); 2515 2516 if (IS_ERR(csum_shash)) { 2517 btrfs_err(fs_info, "error allocating %s hash for checksum", 2518 csum_driver); 2519 return PTR_ERR(csum_shash); 2520 } 2521 2522 fs_info->csum_shash = csum_shash; 2523 2524 return 0; 2525 } 2526 2527 static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 2528 struct btrfs_fs_devices *fs_devices) 2529 { 2530 int ret; 2531 struct btrfs_root *log_tree_root; 2532 struct btrfs_super_block *disk_super = fs_info->super_copy; 2533 u64 bytenr = btrfs_super_log_root(disk_super); 2534 int level = btrfs_super_log_root_level(disk_super); 2535 2536 if (fs_devices->rw_devices == 0) { 2537 btrfs_warn(fs_info, "log replay required on RO media"); 2538 return -EIO; 2539 } 2540 2541 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, 2542 GFP_KERNEL); 2543 if (!log_tree_root) 2544 return -ENOMEM; 2545 2546 log_tree_root->node = read_tree_block(fs_info, bytenr, 2547 BTRFS_TREE_LOG_OBJECTID, 2548 fs_info->generation + 1, level, 2549 NULL); 2550 if (IS_ERR(log_tree_root->node)) { 2551 btrfs_warn(fs_info, "failed to read log tree"); 2552 ret = PTR_ERR(log_tree_root->node); 2553 log_tree_root->node = NULL; 2554 btrfs_put_root(log_tree_root); 2555 return ret; 2556 } 2557 if (!extent_buffer_uptodate(log_tree_root->node)) { 2558 btrfs_err(fs_info, "failed to read log tree"); 2559 btrfs_put_root(log_tree_root); 2560 return -EIO; 2561 } 2562 2563 /* returns with log_tree_root freed on success */ 2564 ret = btrfs_recover_log_trees(log_tree_root); 2565 if (ret) { 2566 btrfs_handle_fs_error(fs_info, ret, 2567 "Failed to recover log tree"); 2568 btrfs_put_root(log_tree_root); 2569 return ret; 2570 } 2571 2572 if (sb_rdonly(fs_info->sb)) { 2573 ret = btrfs_commit_super(fs_info); 2574 if (ret) 2575 return ret; 2576 } 2577 2578 return 0; 2579 } 2580 2581 static int load_global_roots_objectid(struct btrfs_root *tree_root, 2582 struct btrfs_path *path, u64 objectid, 2583 const char *name) 2584 { 2585 struct btrfs_fs_info *fs_info = tree_root->fs_info; 2586 struct btrfs_root *root; 2587 u64 max_global_id = 0; 2588 int ret; 2589 struct btrfs_key key = { 2590 .objectid = objectid, 2591 .type = BTRFS_ROOT_ITEM_KEY, 2592 .offset = 0, 2593 }; 2594 bool found = false; 2595 2596 /* If we have IGNOREDATACSUMS skip loading these roots. */ 2597 if (objectid == BTRFS_CSUM_TREE_OBJECTID && 2598 btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { 2599 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2600 return 0; 2601 } 2602 2603 while (1) { 2604 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); 2605 if (ret < 0) 2606 break; 2607 2608 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2609 ret = btrfs_next_leaf(tree_root, path); 2610 if (ret) { 2611 if (ret > 0) 2612 ret = 0; 2613 break; 2614 } 2615 } 2616 ret = 0; 2617 2618 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 2619 if (key.objectid != objectid) 2620 break; 2621 btrfs_release_path(path); 2622 2623 /* 2624 * Just worry about this for extent tree, it'll be the same for 2625 * everybody. 2626 */ 2627 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2628 max_global_id = max(max_global_id, key.offset); 2629 2630 found = true; 2631 root = read_tree_root_path(tree_root, path, &key); 2632 if (IS_ERR(root)) { 2633 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2634 ret = PTR_ERR(root); 2635 break; 2636 } 2637 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2638 ret = btrfs_global_root_insert(root); 2639 if (ret) { 2640 btrfs_put_root(root); 2641 break; 2642 } 2643 key.offset++; 2644 } 2645 btrfs_release_path(path); 2646 2647 if (objectid == BTRFS_EXTENT_TREE_OBJECTID) 2648 fs_info->nr_global_roots = max_global_id + 1; 2649 2650 if (!found || ret) { 2651 if (objectid == BTRFS_CSUM_TREE_OBJECTID) 2652 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state); 2653 2654 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) 2655 ret = ret ? ret : -ENOENT; 2656 else 2657 ret = 0; 2658 btrfs_err(fs_info, "failed to load root %s", name); 2659 } 2660 return ret; 2661 } 2662 2663 static int load_global_roots(struct btrfs_root *tree_root) 2664 { 2665 struct btrfs_path *path; 2666 int ret = 0; 2667 2668 path = btrfs_alloc_path(); 2669 if (!path) 2670 return -ENOMEM; 2671 2672 ret = load_global_roots_objectid(tree_root, path, 2673 BTRFS_EXTENT_TREE_OBJECTID, "extent"); 2674 if (ret) 2675 goto out; 2676 ret = load_global_roots_objectid(tree_root, path, 2677 BTRFS_CSUM_TREE_OBJECTID, "csum"); 2678 if (ret) 2679 goto out; 2680 if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE)) 2681 goto out; 2682 ret = load_global_roots_objectid(tree_root, path, 2683 BTRFS_FREE_SPACE_TREE_OBJECTID, 2684 "free space"); 2685 out: 2686 btrfs_free_path(path); 2687 return ret; 2688 } 2689 2690 static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 2691 { 2692 struct btrfs_root *tree_root = fs_info->tree_root; 2693 struct btrfs_root *root; 2694 struct btrfs_key location; 2695 int ret; 2696 2697 BUG_ON(!fs_info->tree_root); 2698 2699 ret = load_global_roots(tree_root); 2700 if (ret) 2701 return ret; 2702 2703 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2704 location.type = BTRFS_ROOT_ITEM_KEY; 2705 location.offset = 0; 2706 2707 root = btrfs_read_tree_root(tree_root, &location); 2708 if (IS_ERR(root)) { 2709 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2710 ret = PTR_ERR(root); 2711 goto out; 2712 } 2713 } else { 2714 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2715 fs_info->dev_root = root; 2716 } 2717 /* Initialize fs_info for all devices in any case */ 2718 btrfs_init_devices_late(fs_info); 2719 2720 /* 2721 * This tree can share blocks with some other fs tree during relocation 2722 * and we need a proper setup by btrfs_get_fs_root 2723 */ 2724 root = btrfs_get_fs_root(tree_root->fs_info, 2725 BTRFS_DATA_RELOC_TREE_OBJECTID, true); 2726 if (IS_ERR(root)) { 2727 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2728 ret = PTR_ERR(root); 2729 goto out; 2730 } 2731 } else { 2732 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2733 fs_info->data_reloc_root = root; 2734 } 2735 2736 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2737 root = btrfs_read_tree_root(tree_root, &location); 2738 if (!IS_ERR(root)) { 2739 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2740 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2741 fs_info->quota_root = root; 2742 } 2743 2744 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2745 root = btrfs_read_tree_root(tree_root, &location); 2746 if (IS_ERR(root)) { 2747 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { 2748 ret = PTR_ERR(root); 2749 if (ret != -ENOENT) 2750 goto out; 2751 } 2752 } else { 2753 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2754 fs_info->uuid_root = root; 2755 } 2756 2757 return 0; 2758 out: 2759 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", 2760 location.objectid, ret); 2761 return ret; 2762 } 2763 2764 /* 2765 * Real super block validation 2766 * NOTE: super csum type and incompat features will not be checked here. 2767 * 2768 * @sb: super block to check 2769 * @mirror_num: the super block number to check its bytenr: 2770 * 0 the primary (1st) sb 2771 * 1, 2 2nd and 3rd backup copy 2772 * -1 skip bytenr check 2773 */ 2774 static int validate_super(struct btrfs_fs_info *fs_info, 2775 struct btrfs_super_block *sb, int mirror_num) 2776 { 2777 u64 nodesize = btrfs_super_nodesize(sb); 2778 u64 sectorsize = btrfs_super_sectorsize(sb); 2779 int ret = 0; 2780 2781 if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 2782 btrfs_err(fs_info, "no valid FS found"); 2783 ret = -EINVAL; 2784 } 2785 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 2786 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 2787 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 2788 ret = -EINVAL; 2789 } 2790 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 2791 btrfs_err(fs_info, "tree_root level too big: %d >= %d", 2792 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 2793 ret = -EINVAL; 2794 } 2795 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 2796 btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 2797 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 2798 ret = -EINVAL; 2799 } 2800 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 2801 btrfs_err(fs_info, "log_root level too big: %d >= %d", 2802 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 2803 ret = -EINVAL; 2804 } 2805 2806 /* 2807 * Check sectorsize and nodesize first, other check will need it. 2808 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 2809 */ 2810 if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 2811 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2812 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 2813 ret = -EINVAL; 2814 } 2815 2816 /* 2817 * For 4K page size, we only support 4K sector size. 2818 * For 64K page size, we support 64K and 4K sector sizes. 2819 */ 2820 if ((PAGE_SIZE == SZ_4K && sectorsize != PAGE_SIZE) || 2821 (PAGE_SIZE == SZ_64K && (sectorsize != SZ_4K && 2822 sectorsize != SZ_64K))) { 2823 btrfs_err(fs_info, 2824 "sectorsize %llu not yet supported for page size %lu", 2825 sectorsize, PAGE_SIZE); 2826 ret = -EINVAL; 2827 } 2828 2829 if (!is_power_of_2(nodesize) || nodesize < sectorsize || 2830 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 2831 btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 2832 ret = -EINVAL; 2833 } 2834 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 2835 btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 2836 le32_to_cpu(sb->__unused_leafsize), nodesize); 2837 ret = -EINVAL; 2838 } 2839 2840 /* Root alignment check */ 2841 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 2842 btrfs_warn(fs_info, "tree_root block unaligned: %llu", 2843 btrfs_super_root(sb)); 2844 ret = -EINVAL; 2845 } 2846 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 2847 btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 2848 btrfs_super_chunk_root(sb)); 2849 ret = -EINVAL; 2850 } 2851 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 2852 btrfs_warn(fs_info, "log_root block unaligned: %llu", 2853 btrfs_super_log_root(sb)); 2854 ret = -EINVAL; 2855 } 2856 2857 if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, 2858 BTRFS_FSID_SIZE)) { 2859 btrfs_err(fs_info, 2860 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU", 2861 fs_info->super_copy->fsid, fs_info->fs_devices->fsid); 2862 ret = -EINVAL; 2863 } 2864 2865 if (btrfs_fs_incompat(fs_info, METADATA_UUID) && 2866 memcmp(fs_info->fs_devices->metadata_uuid, 2867 fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) { 2868 btrfs_err(fs_info, 2869 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU", 2870 fs_info->super_copy->metadata_uuid, 2871 fs_info->fs_devices->metadata_uuid); 2872 ret = -EINVAL; 2873 } 2874 2875 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, 2876 BTRFS_FSID_SIZE) != 0) { 2877 btrfs_err(fs_info, 2878 "dev_item UUID does not match metadata fsid: %pU != %pU", 2879 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); 2880 ret = -EINVAL; 2881 } 2882 2883 /* 2884 * Hint to catch really bogus numbers, bitflips or so, more exact checks are 2885 * done later 2886 */ 2887 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 2888 btrfs_err(fs_info, "bytes_used is too small %llu", 2889 btrfs_super_bytes_used(sb)); 2890 ret = -EINVAL; 2891 } 2892 if (!is_power_of_2(btrfs_super_stripesize(sb))) { 2893 btrfs_err(fs_info, "invalid stripesize %u", 2894 btrfs_super_stripesize(sb)); 2895 ret = -EINVAL; 2896 } 2897 if (btrfs_super_num_devices(sb) > (1UL << 31)) 2898 btrfs_warn(fs_info, "suspicious number of devices: %llu", 2899 btrfs_super_num_devices(sb)); 2900 if (btrfs_super_num_devices(sb) == 0) { 2901 btrfs_err(fs_info, "number of devices is 0"); 2902 ret = -EINVAL; 2903 } 2904 2905 if (mirror_num >= 0 && 2906 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { 2907 btrfs_err(fs_info, "super offset mismatch %llu != %u", 2908 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 2909 ret = -EINVAL; 2910 } 2911 2912 /* 2913 * Obvious sys_chunk_array corruptions, it must hold at least one key 2914 * and one chunk 2915 */ 2916 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 2917 btrfs_err(fs_info, "system chunk array too big %u > %u", 2918 btrfs_super_sys_array_size(sb), 2919 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 2920 ret = -EINVAL; 2921 } 2922 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 2923 + sizeof(struct btrfs_chunk)) { 2924 btrfs_err(fs_info, "system chunk array too small %u < %zu", 2925 btrfs_super_sys_array_size(sb), 2926 sizeof(struct btrfs_disk_key) 2927 + sizeof(struct btrfs_chunk)); 2928 ret = -EINVAL; 2929 } 2930 2931 /* 2932 * The generation is a global counter, we'll trust it more than the others 2933 * but it's still possible that it's the one that's wrong. 2934 */ 2935 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 2936 btrfs_warn(fs_info, 2937 "suspicious: generation < chunk_root_generation: %llu < %llu", 2938 btrfs_super_generation(sb), 2939 btrfs_super_chunk_root_generation(sb)); 2940 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 2941 && btrfs_super_cache_generation(sb) != (u64)-1) 2942 btrfs_warn(fs_info, 2943 "suspicious: generation < cache_generation: %llu < %llu", 2944 btrfs_super_generation(sb), 2945 btrfs_super_cache_generation(sb)); 2946 2947 return ret; 2948 } 2949 2950 /* 2951 * Validation of super block at mount time. 2952 * Some checks already done early at mount time, like csum type and incompat 2953 * flags will be skipped. 2954 */ 2955 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) 2956 { 2957 return validate_super(fs_info, fs_info->super_copy, 0); 2958 } 2959 2960 /* 2961 * Validation of super block at write time. 2962 * Some checks like bytenr check will be skipped as their values will be 2963 * overwritten soon. 2964 * Extra checks like csum type and incompat flags will be done here. 2965 */ 2966 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, 2967 struct btrfs_super_block *sb) 2968 { 2969 int ret; 2970 2971 ret = validate_super(fs_info, sb, -1); 2972 if (ret < 0) 2973 goto out; 2974 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { 2975 ret = -EUCLEAN; 2976 btrfs_err(fs_info, "invalid csum type, has %u want %u", 2977 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); 2978 goto out; 2979 } 2980 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { 2981 ret = -EUCLEAN; 2982 btrfs_err(fs_info, 2983 "invalid incompat flags, has 0x%llx valid mask 0x%llx", 2984 btrfs_super_incompat_flags(sb), 2985 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); 2986 goto out; 2987 } 2988 out: 2989 if (ret < 0) 2990 btrfs_err(fs_info, 2991 "super block corruption detected before writing it to disk"); 2992 return ret; 2993 } 2994 2995 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level) 2996 { 2997 int ret = 0; 2998 2999 root->node = read_tree_block(root->fs_info, bytenr, 3000 root->root_key.objectid, gen, level, NULL); 3001 if (IS_ERR(root->node)) { 3002 ret = PTR_ERR(root->node); 3003 root->node = NULL; 3004 return ret; 3005 } 3006 if (!extent_buffer_uptodate(root->node)) { 3007 free_extent_buffer(root->node); 3008 root->node = NULL; 3009 return -EIO; 3010 } 3011 3012 btrfs_set_root_node(&root->root_item, root->node); 3013 root->commit_root = btrfs_root_node(root); 3014 btrfs_set_root_refs(&root->root_item, 1); 3015 return ret; 3016 } 3017 3018 static int load_important_roots(struct btrfs_fs_info *fs_info) 3019 { 3020 struct btrfs_super_block *sb = fs_info->super_copy; 3021 u64 gen, bytenr; 3022 int level, ret; 3023 3024 bytenr = btrfs_super_root(sb); 3025 gen = btrfs_super_generation(sb); 3026 level = btrfs_super_root_level(sb); 3027 ret = load_super_root(fs_info->tree_root, bytenr, gen, level); 3028 if (ret) { 3029 btrfs_warn(fs_info, "couldn't read tree root"); 3030 return ret; 3031 } 3032 3033 if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 3034 return 0; 3035 3036 bytenr = btrfs_super_block_group_root(sb); 3037 gen = btrfs_super_block_group_root_generation(sb); 3038 level = btrfs_super_block_group_root_level(sb); 3039 ret = load_super_root(fs_info->block_group_root, bytenr, gen, level); 3040 if (ret) 3041 btrfs_warn(fs_info, "couldn't read block group root"); 3042 return ret; 3043 } 3044 3045 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) 3046 { 3047 int backup_index = find_newest_super_backup(fs_info); 3048 struct btrfs_super_block *sb = fs_info->super_copy; 3049 struct btrfs_root *tree_root = fs_info->tree_root; 3050 bool handle_error = false; 3051 int ret = 0; 3052 int i; 3053 3054 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3055 struct btrfs_root *root; 3056 3057 root = btrfs_alloc_root(fs_info, BTRFS_BLOCK_GROUP_TREE_OBJECTID, 3058 GFP_KERNEL); 3059 if (!root) 3060 return -ENOMEM; 3061 fs_info->block_group_root = root; 3062 } 3063 3064 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 3065 if (handle_error) { 3066 if (!IS_ERR(tree_root->node)) 3067 free_extent_buffer(tree_root->node); 3068 tree_root->node = NULL; 3069 3070 if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3071 break; 3072 3073 free_root_pointers(fs_info, 0); 3074 3075 /* 3076 * Don't use the log in recovery mode, it won't be 3077 * valid 3078 */ 3079 btrfs_set_super_log_root(sb, 0); 3080 3081 /* We can't trust the free space cache either */ 3082 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3083 3084 ret = read_backup_root(fs_info, i); 3085 backup_index = ret; 3086 if (ret < 0) 3087 return ret; 3088 } 3089 3090 ret = load_important_roots(fs_info); 3091 if (ret) { 3092 handle_error = true; 3093 continue; 3094 } 3095 3096 /* 3097 * No need to hold btrfs_root::objectid_mutex since the fs 3098 * hasn't been fully initialised and we are the only user 3099 */ 3100 ret = btrfs_init_root_free_objectid(tree_root); 3101 if (ret < 0) { 3102 handle_error = true; 3103 continue; 3104 } 3105 3106 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID); 3107 3108 ret = btrfs_read_roots(fs_info); 3109 if (ret < 0) { 3110 handle_error = true; 3111 continue; 3112 } 3113 3114 /* All successful */ 3115 fs_info->generation = btrfs_header_generation(tree_root->node); 3116 fs_info->last_trans_committed = fs_info->generation; 3117 fs_info->last_reloc_trans = 0; 3118 3119 /* Always begin writing backup roots after the one being used */ 3120 if (backup_index < 0) { 3121 fs_info->backup_root_index = 0; 3122 } else { 3123 fs_info->backup_root_index = backup_index + 1; 3124 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; 3125 } 3126 break; 3127 } 3128 3129 return ret; 3130 } 3131 3132 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) 3133 { 3134 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 3135 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 3136 INIT_LIST_HEAD(&fs_info->trans_list); 3137 INIT_LIST_HEAD(&fs_info->dead_roots); 3138 INIT_LIST_HEAD(&fs_info->delayed_iputs); 3139 INIT_LIST_HEAD(&fs_info->delalloc_roots); 3140 INIT_LIST_HEAD(&fs_info->caching_block_groups); 3141 spin_lock_init(&fs_info->delalloc_root_lock); 3142 spin_lock_init(&fs_info->trans_lock); 3143 spin_lock_init(&fs_info->fs_roots_radix_lock); 3144 spin_lock_init(&fs_info->delayed_iput_lock); 3145 spin_lock_init(&fs_info->defrag_inodes_lock); 3146 spin_lock_init(&fs_info->super_lock); 3147 spin_lock_init(&fs_info->buffer_lock); 3148 spin_lock_init(&fs_info->unused_bgs_lock); 3149 spin_lock_init(&fs_info->treelog_bg_lock); 3150 spin_lock_init(&fs_info->zone_active_bgs_lock); 3151 spin_lock_init(&fs_info->relocation_bg_lock); 3152 rwlock_init(&fs_info->tree_mod_log_lock); 3153 rwlock_init(&fs_info->global_root_lock); 3154 mutex_init(&fs_info->unused_bg_unpin_mutex); 3155 mutex_init(&fs_info->reclaim_bgs_lock); 3156 mutex_init(&fs_info->reloc_mutex); 3157 mutex_init(&fs_info->delalloc_root_mutex); 3158 mutex_init(&fs_info->zoned_meta_io_lock); 3159 seqlock_init(&fs_info->profiles_lock); 3160 3161 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 3162 INIT_LIST_HEAD(&fs_info->space_info); 3163 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 3164 INIT_LIST_HEAD(&fs_info->unused_bgs); 3165 INIT_LIST_HEAD(&fs_info->reclaim_bgs); 3166 INIT_LIST_HEAD(&fs_info->zone_active_bgs); 3167 #ifdef CONFIG_BTRFS_DEBUG 3168 INIT_LIST_HEAD(&fs_info->allocated_roots); 3169 INIT_LIST_HEAD(&fs_info->allocated_ebs); 3170 spin_lock_init(&fs_info->eb_leak_lock); 3171 #endif 3172 extent_map_tree_init(&fs_info->mapping_tree); 3173 btrfs_init_block_rsv(&fs_info->global_block_rsv, 3174 BTRFS_BLOCK_RSV_GLOBAL); 3175 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 3176 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 3177 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 3178 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 3179 BTRFS_BLOCK_RSV_DELOPS); 3180 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, 3181 BTRFS_BLOCK_RSV_DELREFS); 3182 3183 atomic_set(&fs_info->async_delalloc_pages, 0); 3184 atomic_set(&fs_info->defrag_running, 0); 3185 atomic_set(&fs_info->nr_delayed_iputs, 0); 3186 atomic64_set(&fs_info->tree_mod_seq, 0); 3187 fs_info->global_root_tree = RB_ROOT; 3188 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 3189 fs_info->metadata_ratio = 0; 3190 fs_info->defrag_inodes = RB_ROOT; 3191 atomic64_set(&fs_info->free_chunk_space, 0); 3192 fs_info->tree_mod_log = RB_ROOT; 3193 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 3194 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 3195 btrfs_init_ref_verify(fs_info); 3196 3197 fs_info->thread_pool_size = min_t(unsigned long, 3198 num_online_cpus() + 2, 8); 3199 3200 INIT_LIST_HEAD(&fs_info->ordered_roots); 3201 spin_lock_init(&fs_info->ordered_root_lock); 3202 3203 btrfs_init_scrub(fs_info); 3204 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3205 fs_info->check_integrity_print_mask = 0; 3206 #endif 3207 btrfs_init_balance(fs_info); 3208 btrfs_init_async_reclaim_work(fs_info); 3209 3210 spin_lock_init(&fs_info->block_group_cache_lock); 3211 fs_info->block_group_cache_tree = RB_ROOT; 3212 fs_info->first_logical_byte = (u64)-1; 3213 3214 extent_io_tree_init(fs_info, &fs_info->excluded_extents, 3215 IO_TREE_FS_EXCLUDED_EXTENTS, NULL); 3216 3217 mutex_init(&fs_info->ordered_operations_mutex); 3218 mutex_init(&fs_info->tree_log_mutex); 3219 mutex_init(&fs_info->chunk_mutex); 3220 mutex_init(&fs_info->transaction_kthread_mutex); 3221 mutex_init(&fs_info->cleaner_mutex); 3222 mutex_init(&fs_info->ro_block_group_mutex); 3223 init_rwsem(&fs_info->commit_root_sem); 3224 init_rwsem(&fs_info->cleanup_work_sem); 3225 init_rwsem(&fs_info->subvol_sem); 3226 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 3227 3228 btrfs_init_dev_replace_locks(fs_info); 3229 btrfs_init_qgroup(fs_info); 3230 btrfs_discard_init(fs_info); 3231 3232 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 3233 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 3234 3235 init_waitqueue_head(&fs_info->transaction_throttle); 3236 init_waitqueue_head(&fs_info->transaction_wait); 3237 init_waitqueue_head(&fs_info->transaction_blocked_wait); 3238 init_waitqueue_head(&fs_info->async_submit_wait); 3239 init_waitqueue_head(&fs_info->delayed_iputs_wait); 3240 3241 /* Usable values until the real ones are cached from the superblock */ 3242 fs_info->nodesize = 4096; 3243 fs_info->sectorsize = 4096; 3244 fs_info->sectorsize_bits = ilog2(4096); 3245 fs_info->stripesize = 4096; 3246 3247 spin_lock_init(&fs_info->swapfile_pins_lock); 3248 fs_info->swapfile_pins = RB_ROOT; 3249 3250 fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH; 3251 INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work); 3252 } 3253 3254 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) 3255 { 3256 int ret; 3257 3258 fs_info->sb = sb; 3259 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 3260 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 3261 3262 ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL); 3263 if (ret) 3264 return ret; 3265 3266 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 3267 if (ret) 3268 return ret; 3269 3270 fs_info->dirty_metadata_batch = PAGE_SIZE * 3271 (1 + ilog2(nr_cpu_ids)); 3272 3273 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 3274 if (ret) 3275 return ret; 3276 3277 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, 3278 GFP_KERNEL); 3279 if (ret) 3280 return ret; 3281 3282 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 3283 GFP_KERNEL); 3284 if (!fs_info->delayed_root) 3285 return -ENOMEM; 3286 btrfs_init_delayed_root(fs_info->delayed_root); 3287 3288 if (sb_rdonly(sb)) 3289 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); 3290 3291 return btrfs_alloc_stripe_hash_table(fs_info); 3292 } 3293 3294 static int btrfs_uuid_rescan_kthread(void *data) 3295 { 3296 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; 3297 int ret; 3298 3299 /* 3300 * 1st step is to iterate through the existing UUID tree and 3301 * to delete all entries that contain outdated data. 3302 * 2nd step is to add all missing entries to the UUID tree. 3303 */ 3304 ret = btrfs_uuid_tree_iterate(fs_info); 3305 if (ret < 0) { 3306 if (ret != -EINTR) 3307 btrfs_warn(fs_info, "iterating uuid_tree failed %d", 3308 ret); 3309 up(&fs_info->uuid_tree_rescan_sem); 3310 return ret; 3311 } 3312 return btrfs_uuid_scan_kthread(data); 3313 } 3314 3315 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) 3316 { 3317 struct task_struct *task; 3318 3319 down(&fs_info->uuid_tree_rescan_sem); 3320 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); 3321 if (IS_ERR(task)) { 3322 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 3323 btrfs_warn(fs_info, "failed to start uuid_rescan task"); 3324 up(&fs_info->uuid_tree_rescan_sem); 3325 return PTR_ERR(task); 3326 } 3327 3328 return 0; 3329 } 3330 3331 /* 3332 * Some options only have meaning at mount time and shouldn't persist across 3333 * remounts, or be displayed. Clear these at the end of mount and remount 3334 * code paths. 3335 */ 3336 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) 3337 { 3338 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 3339 btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); 3340 } 3341 3342 /* 3343 * Mounting logic specific to read-write file systems. Shared by open_ctree 3344 * and btrfs_remount when remounting from read-only to read-write. 3345 */ 3346 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) 3347 { 3348 int ret; 3349 const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); 3350 bool clear_free_space_tree = false; 3351 3352 if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 3353 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 3354 clear_free_space_tree = true; 3355 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 3356 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 3357 btrfs_warn(fs_info, "free space tree is invalid"); 3358 clear_free_space_tree = true; 3359 } 3360 3361 if (clear_free_space_tree) { 3362 btrfs_info(fs_info, "clearing free space tree"); 3363 ret = btrfs_clear_free_space_tree(fs_info); 3364 if (ret) { 3365 btrfs_warn(fs_info, 3366 "failed to clear free space tree: %d", ret); 3367 goto out; 3368 } 3369 } 3370 3371 /* 3372 * btrfs_find_orphan_roots() is responsible for finding all the dead 3373 * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load 3374 * them into the fs_info->fs_roots_radix tree. This must be done before 3375 * calling btrfs_orphan_cleanup() on the tree root. If we don't do it 3376 * first, then btrfs_orphan_cleanup() will delete a dead root's orphan 3377 * item before the root's tree is deleted - this means that if we unmount 3378 * or crash before the deletion completes, on the next mount we will not 3379 * delete what remains of the tree because the orphan item does not 3380 * exists anymore, which is what tells us we have a pending deletion. 3381 */ 3382 ret = btrfs_find_orphan_roots(fs_info); 3383 if (ret) 3384 goto out; 3385 3386 ret = btrfs_cleanup_fs_roots(fs_info); 3387 if (ret) 3388 goto out; 3389 3390 down_read(&fs_info->cleanup_work_sem); 3391 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 3392 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 3393 up_read(&fs_info->cleanup_work_sem); 3394 goto out; 3395 } 3396 up_read(&fs_info->cleanup_work_sem); 3397 3398 mutex_lock(&fs_info->cleaner_mutex); 3399 ret = btrfs_recover_relocation(fs_info); 3400 mutex_unlock(&fs_info->cleaner_mutex); 3401 if (ret < 0) { 3402 btrfs_warn(fs_info, "failed to recover relocation: %d", ret); 3403 goto out; 3404 } 3405 3406 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 3407 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 3408 btrfs_info(fs_info, "creating free space tree"); 3409 ret = btrfs_create_free_space_tree(fs_info); 3410 if (ret) { 3411 btrfs_warn(fs_info, 3412 "failed to create free space tree: %d", ret); 3413 goto out; 3414 } 3415 } 3416 3417 if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { 3418 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); 3419 if (ret) 3420 goto out; 3421 } 3422 3423 ret = btrfs_resume_balance_async(fs_info); 3424 if (ret) 3425 goto out; 3426 3427 ret = btrfs_resume_dev_replace_async(fs_info); 3428 if (ret) { 3429 btrfs_warn(fs_info, "failed to resume dev_replace"); 3430 goto out; 3431 } 3432 3433 btrfs_qgroup_rescan_resume(fs_info); 3434 3435 if (!fs_info->uuid_root) { 3436 btrfs_info(fs_info, "creating UUID tree"); 3437 ret = btrfs_create_uuid_tree(fs_info); 3438 if (ret) { 3439 btrfs_warn(fs_info, 3440 "failed to create the UUID tree %d", ret); 3441 goto out; 3442 } 3443 } 3444 3445 out: 3446 return ret; 3447 } 3448 3449 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, 3450 char *options) 3451 { 3452 u32 sectorsize; 3453 u32 nodesize; 3454 u32 stripesize; 3455 u64 generation; 3456 u64 features; 3457 u16 csum_type; 3458 struct btrfs_super_block *disk_super; 3459 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 3460 struct btrfs_root *tree_root; 3461 struct btrfs_root *chunk_root; 3462 int ret; 3463 int err = -EINVAL; 3464 int level; 3465 3466 ret = init_mount_fs_info(fs_info, sb); 3467 if (ret) { 3468 err = ret; 3469 goto fail; 3470 } 3471 3472 /* These need to be init'ed before we start creating inodes and such. */ 3473 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, 3474 GFP_KERNEL); 3475 fs_info->tree_root = tree_root; 3476 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, 3477 GFP_KERNEL); 3478 fs_info->chunk_root = chunk_root; 3479 if (!tree_root || !chunk_root) { 3480 err = -ENOMEM; 3481 goto fail; 3482 } 3483 3484 fs_info->btree_inode = new_inode(sb); 3485 if (!fs_info->btree_inode) { 3486 err = -ENOMEM; 3487 goto fail; 3488 } 3489 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 3490 btrfs_init_btree_inode(fs_info); 3491 3492 invalidate_bdev(fs_devices->latest_dev->bdev); 3493 3494 /* 3495 * Read super block and check the signature bytes only 3496 */ 3497 disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev); 3498 if (IS_ERR(disk_super)) { 3499 err = PTR_ERR(disk_super); 3500 goto fail_alloc; 3501 } 3502 3503 /* 3504 * Verify the type first, if that or the checksum value are 3505 * corrupted, we'll find out 3506 */ 3507 csum_type = btrfs_super_csum_type(disk_super); 3508 if (!btrfs_supported_super_csum(csum_type)) { 3509 btrfs_err(fs_info, "unsupported checksum algorithm: %u", 3510 csum_type); 3511 err = -EINVAL; 3512 btrfs_release_disk_super(disk_super); 3513 goto fail_alloc; 3514 } 3515 3516 fs_info->csum_size = btrfs_super_csum_size(disk_super); 3517 3518 ret = btrfs_init_csum_hash(fs_info, csum_type); 3519 if (ret) { 3520 err = ret; 3521 btrfs_release_disk_super(disk_super); 3522 goto fail_alloc; 3523 } 3524 3525 /* 3526 * We want to check superblock checksum, the type is stored inside. 3527 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 3528 */ 3529 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { 3530 btrfs_err(fs_info, "superblock checksum mismatch"); 3531 err = -EINVAL; 3532 btrfs_release_disk_super(disk_super); 3533 goto fail_alloc; 3534 } 3535 3536 /* 3537 * super_copy is zeroed at allocation time and we never touch the 3538 * following bytes up to INFO_SIZE, the checksum is calculated from 3539 * the whole block of INFO_SIZE 3540 */ 3541 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); 3542 btrfs_release_disk_super(disk_super); 3543 3544 disk_super = fs_info->super_copy; 3545 3546 3547 features = btrfs_super_flags(disk_super); 3548 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { 3549 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; 3550 btrfs_set_super_flags(disk_super, features); 3551 btrfs_info(fs_info, 3552 "found metadata UUID change in progress flag, clearing"); 3553 } 3554 3555 memcpy(fs_info->super_for_commit, fs_info->super_copy, 3556 sizeof(*fs_info->super_for_commit)); 3557 3558 ret = btrfs_validate_mount_super(fs_info); 3559 if (ret) { 3560 btrfs_err(fs_info, "superblock contains fatal errors"); 3561 err = -EINVAL; 3562 goto fail_alloc; 3563 } 3564 3565 if (!btrfs_super_root(disk_super)) 3566 goto fail_alloc; 3567 3568 /* check FS state, whether FS is broken. */ 3569 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 3570 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 3571 3572 /* 3573 * In the long term, we'll store the compression type in the super 3574 * block, and it'll be used for per file compression control. 3575 */ 3576 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 3577 3578 /* 3579 * Flag our filesystem as having big metadata blocks if they are bigger 3580 * than the page size. 3581 */ 3582 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 3583 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 3584 btrfs_info(fs_info, 3585 "flagging fs with big metadata feature"); 3586 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 3587 } 3588 3589 /* Set up fs_info before parsing mount options */ 3590 nodesize = btrfs_super_nodesize(disk_super); 3591 sectorsize = btrfs_super_sectorsize(disk_super); 3592 stripesize = sectorsize; 3593 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 3594 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 3595 3596 fs_info->nodesize = nodesize; 3597 fs_info->sectorsize = sectorsize; 3598 fs_info->sectorsize_bits = ilog2(sectorsize); 3599 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; 3600 fs_info->stripesize = stripesize; 3601 3602 ret = btrfs_parse_options(fs_info, options, sb->s_flags); 3603 if (ret) { 3604 err = ret; 3605 goto fail_alloc; 3606 } 3607 3608 features = btrfs_super_incompat_flags(disk_super) & 3609 ~BTRFS_FEATURE_INCOMPAT_SUPP; 3610 if (features) { 3611 btrfs_err(fs_info, 3612 "cannot mount because of unsupported optional features (%llx)", 3613 features); 3614 err = -EINVAL; 3615 goto fail_alloc; 3616 } 3617 3618 features = btrfs_super_incompat_flags(disk_super); 3619 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 3620 if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 3621 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 3622 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 3623 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 3624 3625 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 3626 btrfs_info(fs_info, "has skinny extents"); 3627 3628 /* 3629 * mixed block groups end up with duplicate but slightly offset 3630 * extent buffers for the same range. It leads to corruptions 3631 */ 3632 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 3633 (sectorsize != nodesize)) { 3634 btrfs_err(fs_info, 3635 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 3636 nodesize, sectorsize); 3637 goto fail_alloc; 3638 } 3639 3640 /* 3641 * Needn't use the lock because there is no other task which will 3642 * update the flag. 3643 */ 3644 btrfs_set_super_incompat_flags(disk_super, features); 3645 3646 features = btrfs_super_compat_ro_flags(disk_super) & 3647 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 3648 if (!sb_rdonly(sb) && features) { 3649 btrfs_err(fs_info, 3650 "cannot mount read-write because of unsupported optional features (%llx)", 3651 features); 3652 err = -EINVAL; 3653 goto fail_alloc; 3654 } 3655 3656 if (sectorsize < PAGE_SIZE) { 3657 struct btrfs_subpage_info *subpage_info; 3658 3659 btrfs_warn(fs_info, 3660 "read-write for sector size %u with page size %lu is experimental", 3661 sectorsize, PAGE_SIZE); 3662 if (btrfs_super_incompat_flags(fs_info->super_copy) & 3663 BTRFS_FEATURE_INCOMPAT_RAID56) { 3664 btrfs_err(fs_info, 3665 "RAID56 is not yet supported for sector size %u with page size %lu", 3666 sectorsize, PAGE_SIZE); 3667 err = -EINVAL; 3668 goto fail_alloc; 3669 } 3670 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL); 3671 if (!subpage_info) 3672 goto fail_alloc; 3673 btrfs_init_subpage_info(subpage_info, sectorsize); 3674 fs_info->subpage_info = subpage_info; 3675 } 3676 3677 ret = btrfs_init_workqueues(fs_info); 3678 if (ret) { 3679 err = ret; 3680 goto fail_sb_buffer; 3681 } 3682 3683 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 3684 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 3685 3686 sb->s_blocksize = sectorsize; 3687 sb->s_blocksize_bits = blksize_bits(sectorsize); 3688 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); 3689 3690 mutex_lock(&fs_info->chunk_mutex); 3691 ret = btrfs_read_sys_array(fs_info); 3692 mutex_unlock(&fs_info->chunk_mutex); 3693 if (ret) { 3694 btrfs_err(fs_info, "failed to read the system array: %d", ret); 3695 goto fail_sb_buffer; 3696 } 3697 3698 generation = btrfs_super_chunk_root_generation(disk_super); 3699 level = btrfs_super_chunk_root_level(disk_super); 3700 ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super), 3701 generation, level); 3702 if (ret) { 3703 btrfs_err(fs_info, "failed to read chunk root"); 3704 goto fail_tree_roots; 3705 } 3706 3707 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 3708 offsetof(struct btrfs_header, chunk_tree_uuid), 3709 BTRFS_UUID_SIZE); 3710 3711 ret = btrfs_read_chunk_tree(fs_info); 3712 if (ret) { 3713 btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 3714 goto fail_tree_roots; 3715 } 3716 3717 /* 3718 * At this point we know all the devices that make this filesystem, 3719 * including the seed devices but we don't know yet if the replace 3720 * target is required. So free devices that are not part of this 3721 * filesystem but skip the replace target device which is checked 3722 * below in btrfs_init_dev_replace(). 3723 */ 3724 btrfs_free_extra_devids(fs_devices); 3725 if (!fs_devices->latest_dev->bdev) { 3726 btrfs_err(fs_info, "failed to read devices"); 3727 goto fail_tree_roots; 3728 } 3729 3730 ret = init_tree_roots(fs_info); 3731 if (ret) 3732 goto fail_tree_roots; 3733 3734 /* 3735 * Get zone type information of zoned block devices. This will also 3736 * handle emulation of a zoned filesystem if a regular device has the 3737 * zoned incompat feature flag set. 3738 */ 3739 ret = btrfs_get_dev_zone_info_all_devices(fs_info); 3740 if (ret) { 3741 btrfs_err(fs_info, 3742 "zoned: failed to read device zone info: %d", 3743 ret); 3744 goto fail_block_groups; 3745 } 3746 3747 /* 3748 * If we have a uuid root and we're not being told to rescan we need to 3749 * check the generation here so we can set the 3750 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the 3751 * transaction during a balance or the log replay without updating the 3752 * uuid generation, and then if we crash we would rescan the uuid tree, 3753 * even though it was perfectly fine. 3754 */ 3755 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && 3756 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) 3757 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3758 3759 ret = btrfs_verify_dev_extents(fs_info); 3760 if (ret) { 3761 btrfs_err(fs_info, 3762 "failed to verify dev extents against chunks: %d", 3763 ret); 3764 goto fail_block_groups; 3765 } 3766 ret = btrfs_recover_balance(fs_info); 3767 if (ret) { 3768 btrfs_err(fs_info, "failed to recover balance: %d", ret); 3769 goto fail_block_groups; 3770 } 3771 3772 ret = btrfs_init_dev_stats(fs_info); 3773 if (ret) { 3774 btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 3775 goto fail_block_groups; 3776 } 3777 3778 ret = btrfs_init_dev_replace(fs_info); 3779 if (ret) { 3780 btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 3781 goto fail_block_groups; 3782 } 3783 3784 ret = btrfs_check_zoned_mode(fs_info); 3785 if (ret) { 3786 btrfs_err(fs_info, "failed to initialize zoned mode: %d", 3787 ret); 3788 goto fail_block_groups; 3789 } 3790 3791 ret = btrfs_sysfs_add_fsid(fs_devices); 3792 if (ret) { 3793 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 3794 ret); 3795 goto fail_block_groups; 3796 } 3797 3798 ret = btrfs_sysfs_add_mounted(fs_info); 3799 if (ret) { 3800 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 3801 goto fail_fsdev_sysfs; 3802 } 3803 3804 ret = btrfs_init_space_info(fs_info); 3805 if (ret) { 3806 btrfs_err(fs_info, "failed to initialize space info: %d", ret); 3807 goto fail_sysfs; 3808 } 3809 3810 ret = btrfs_read_block_groups(fs_info); 3811 if (ret) { 3812 btrfs_err(fs_info, "failed to read block groups: %d", ret); 3813 goto fail_sysfs; 3814 } 3815 3816 btrfs_free_zone_cache(fs_info); 3817 3818 if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices && 3819 !btrfs_check_rw_degradable(fs_info, NULL)) { 3820 btrfs_warn(fs_info, 3821 "writable mount is not allowed due to too many missing devices"); 3822 goto fail_sysfs; 3823 } 3824 3825 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info, 3826 "btrfs-cleaner"); 3827 if (IS_ERR(fs_info->cleaner_kthread)) 3828 goto fail_sysfs; 3829 3830 fs_info->transaction_kthread = kthread_run(transaction_kthread, 3831 tree_root, 3832 "btrfs-transaction"); 3833 if (IS_ERR(fs_info->transaction_kthread)) 3834 goto fail_cleaner; 3835 3836 if (!btrfs_test_opt(fs_info, NOSSD) && 3837 !fs_info->fs_devices->rotating) { 3838 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 3839 } 3840 3841 /* 3842 * Mount does not set all options immediately, we can do it now and do 3843 * not have to wait for transaction commit 3844 */ 3845 btrfs_apply_pending_changes(fs_info); 3846 3847 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3848 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 3849 ret = btrfsic_mount(fs_info, fs_devices, 3850 btrfs_test_opt(fs_info, 3851 CHECK_INTEGRITY_DATA) ? 1 : 0, 3852 fs_info->check_integrity_print_mask); 3853 if (ret) 3854 btrfs_warn(fs_info, 3855 "failed to initialize integrity check module: %d", 3856 ret); 3857 } 3858 #endif 3859 ret = btrfs_read_qgroup_config(fs_info); 3860 if (ret) 3861 goto fail_trans_kthread; 3862 3863 if (btrfs_build_ref_tree(fs_info)) 3864 btrfs_err(fs_info, "couldn't build ref tree"); 3865 3866 /* do not make disk changes in broken FS or nologreplay is given */ 3867 if (btrfs_super_log_root(disk_super) != 0 && 3868 !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 3869 btrfs_info(fs_info, "start tree-log replay"); 3870 ret = btrfs_replay_log(fs_info, fs_devices); 3871 if (ret) { 3872 err = ret; 3873 goto fail_qgroup; 3874 } 3875 } 3876 3877 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); 3878 if (IS_ERR(fs_info->fs_root)) { 3879 err = PTR_ERR(fs_info->fs_root); 3880 btrfs_warn(fs_info, "failed to read fs tree: %d", err); 3881 fs_info->fs_root = NULL; 3882 goto fail_qgroup; 3883 } 3884 3885 if (sb_rdonly(sb)) 3886 goto clear_oneshot; 3887 3888 ret = btrfs_start_pre_rw_mount(fs_info); 3889 if (ret) { 3890 close_ctree(fs_info); 3891 return ret; 3892 } 3893 btrfs_discard_resume(fs_info); 3894 3895 if (fs_info->uuid_root && 3896 (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 3897 fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { 3898 btrfs_info(fs_info, "checking UUID tree"); 3899 ret = btrfs_check_uuid_tree(fs_info); 3900 if (ret) { 3901 btrfs_warn(fs_info, 3902 "failed to check the UUID tree: %d", ret); 3903 close_ctree(fs_info); 3904 return ret; 3905 } 3906 } 3907 3908 set_bit(BTRFS_FS_OPEN, &fs_info->flags); 3909 3910 /* Kick the cleaner thread so it'll start deleting snapshots. */ 3911 if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags)) 3912 wake_up_process(fs_info->cleaner_kthread); 3913 3914 clear_oneshot: 3915 btrfs_clear_oneshot_options(fs_info); 3916 return 0; 3917 3918 fail_qgroup: 3919 btrfs_free_qgroup_config(fs_info); 3920 fail_trans_kthread: 3921 kthread_stop(fs_info->transaction_kthread); 3922 btrfs_cleanup_transaction(fs_info); 3923 btrfs_free_fs_roots(fs_info); 3924 fail_cleaner: 3925 kthread_stop(fs_info->cleaner_kthread); 3926 3927 /* 3928 * make sure we're done with the btree inode before we stop our 3929 * kthreads 3930 */ 3931 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 3932 3933 fail_sysfs: 3934 btrfs_sysfs_remove_mounted(fs_info); 3935 3936 fail_fsdev_sysfs: 3937 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3938 3939 fail_block_groups: 3940 btrfs_put_block_group_cache(fs_info); 3941 3942 fail_tree_roots: 3943 if (fs_info->data_reloc_root) 3944 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); 3945 free_root_pointers(fs_info, true); 3946 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3947 3948 fail_sb_buffer: 3949 btrfs_stop_all_workers(fs_info); 3950 btrfs_free_block_groups(fs_info); 3951 fail_alloc: 3952 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3953 3954 iput(fs_info->btree_inode); 3955 fail: 3956 btrfs_close_devices(fs_info->fs_devices); 3957 return err; 3958 } 3959 ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3960 3961 static void btrfs_end_super_write(struct bio *bio) 3962 { 3963 struct btrfs_device *device = bio->bi_private; 3964 struct bio_vec *bvec; 3965 struct bvec_iter_all iter_all; 3966 struct page *page; 3967 3968 bio_for_each_segment_all(bvec, bio, iter_all) { 3969 page = bvec->bv_page; 3970 3971 if (bio->bi_status) { 3972 btrfs_warn_rl_in_rcu(device->fs_info, 3973 "lost page write due to IO error on %s (%d)", 3974 rcu_str_deref(device->name), 3975 blk_status_to_errno(bio->bi_status)); 3976 ClearPageUptodate(page); 3977 SetPageError(page); 3978 btrfs_dev_stat_inc_and_print(device, 3979 BTRFS_DEV_STAT_WRITE_ERRS); 3980 } else { 3981 SetPageUptodate(page); 3982 } 3983 3984 put_page(page); 3985 unlock_page(page); 3986 } 3987 3988 bio_put(bio); 3989 } 3990 3991 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, 3992 int copy_num) 3993 { 3994 struct btrfs_super_block *super; 3995 struct page *page; 3996 u64 bytenr, bytenr_orig; 3997 struct address_space *mapping = bdev->bd_inode->i_mapping; 3998 int ret; 3999 4000 bytenr_orig = btrfs_sb_offset(copy_num); 4001 ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); 4002 if (ret == -ENOENT) 4003 return ERR_PTR(-EINVAL); 4004 else if (ret) 4005 return ERR_PTR(ret); 4006 4007 if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev)) 4008 return ERR_PTR(-EINVAL); 4009 4010 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); 4011 if (IS_ERR(page)) 4012 return ERR_CAST(page); 4013 4014 super = page_address(page); 4015 if (btrfs_super_magic(super) != BTRFS_MAGIC) { 4016 btrfs_release_disk_super(super); 4017 return ERR_PTR(-ENODATA); 4018 } 4019 4020 if (btrfs_super_bytenr(super) != bytenr_orig) { 4021 btrfs_release_disk_super(super); 4022 return ERR_PTR(-EINVAL); 4023 } 4024 4025 return super; 4026 } 4027 4028 4029 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) 4030 { 4031 struct btrfs_super_block *super, *latest = NULL; 4032 int i; 4033 u64 transid = 0; 4034 4035 /* we would like to check all the supers, but that would make 4036 * a btrfs mount succeed after a mkfs from a different FS. 4037 * So, we need to add a special mount option to scan for 4038 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 4039 */ 4040 for (i = 0; i < 1; i++) { 4041 super = btrfs_read_dev_one_super(bdev, i); 4042 if (IS_ERR(super)) 4043 continue; 4044 4045 if (!latest || btrfs_super_generation(super) > transid) { 4046 if (latest) 4047 btrfs_release_disk_super(super); 4048 4049 latest = super; 4050 transid = btrfs_super_generation(super); 4051 } 4052 } 4053 4054 return super; 4055 } 4056 4057 /* 4058 * Write superblock @sb to the @device. Do not wait for completion, all the 4059 * pages we use for writing are locked. 4060 * 4061 * Write @max_mirrors copies of the superblock, where 0 means default that fit 4062 * the expected device size at commit time. Note that max_mirrors must be 4063 * same for write and wait phases. 4064 * 4065 * Return number of errors when page is not found or submission fails. 4066 */ 4067 static int write_dev_supers(struct btrfs_device *device, 4068 struct btrfs_super_block *sb, int max_mirrors) 4069 { 4070 struct btrfs_fs_info *fs_info = device->fs_info; 4071 struct address_space *mapping = device->bdev->bd_inode->i_mapping; 4072 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); 4073 int i; 4074 int errors = 0; 4075 int ret; 4076 u64 bytenr, bytenr_orig; 4077 4078 if (max_mirrors == 0) 4079 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4080 4081 shash->tfm = fs_info->csum_shash; 4082 4083 for (i = 0; i < max_mirrors; i++) { 4084 struct page *page; 4085 struct bio *bio; 4086 struct btrfs_super_block *disk_super; 4087 4088 bytenr_orig = btrfs_sb_offset(i); 4089 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); 4090 if (ret == -ENOENT) { 4091 continue; 4092 } else if (ret < 0) { 4093 btrfs_err(device->fs_info, 4094 "couldn't get super block location for mirror %d", 4095 i); 4096 errors++; 4097 continue; 4098 } 4099 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4100 device->commit_total_bytes) 4101 break; 4102 4103 btrfs_set_super_bytenr(sb, bytenr_orig); 4104 4105 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, 4106 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, 4107 sb->csum); 4108 4109 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, 4110 GFP_NOFS); 4111 if (!page) { 4112 btrfs_err(device->fs_info, 4113 "couldn't get super block page for bytenr %llu", 4114 bytenr); 4115 errors++; 4116 continue; 4117 } 4118 4119 /* Bump the refcount for wait_dev_supers() */ 4120 get_page(page); 4121 4122 disk_super = page_address(page); 4123 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); 4124 4125 /* 4126 * Directly use bios here instead of relying on the page cache 4127 * to do I/O, so we don't lose the ability to do integrity 4128 * checking. 4129 */ 4130 bio = bio_alloc(device->bdev, 1, 4131 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO, 4132 GFP_NOFS); 4133 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; 4134 bio->bi_private = device; 4135 bio->bi_end_io = btrfs_end_super_write; 4136 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, 4137 offset_in_page(bytenr)); 4138 4139 /* 4140 * We FUA only the first super block. The others we allow to 4141 * go down lazy and there's a short window where the on-disk 4142 * copies might still contain the older version. 4143 */ 4144 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 4145 bio->bi_opf |= REQ_FUA; 4146 4147 btrfsic_submit_bio(bio); 4148 4149 if (btrfs_advance_sb_log(device, i)) 4150 errors++; 4151 } 4152 return errors < i ? 0 : -1; 4153 } 4154 4155 /* 4156 * Wait for write completion of superblocks done by write_dev_supers, 4157 * @max_mirrors same for write and wait phases. 4158 * 4159 * Return number of errors when page is not found or not marked up to 4160 * date. 4161 */ 4162 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 4163 { 4164 int i; 4165 int errors = 0; 4166 bool primary_failed = false; 4167 int ret; 4168 u64 bytenr; 4169 4170 if (max_mirrors == 0) 4171 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 4172 4173 for (i = 0; i < max_mirrors; i++) { 4174 struct page *page; 4175 4176 ret = btrfs_sb_log_location(device, i, READ, &bytenr); 4177 if (ret == -ENOENT) { 4178 break; 4179 } else if (ret < 0) { 4180 errors++; 4181 if (i == 0) 4182 primary_failed = true; 4183 continue; 4184 } 4185 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 4186 device->commit_total_bytes) 4187 break; 4188 4189 page = find_get_page(device->bdev->bd_inode->i_mapping, 4190 bytenr >> PAGE_SHIFT); 4191 if (!page) { 4192 errors++; 4193 if (i == 0) 4194 primary_failed = true; 4195 continue; 4196 } 4197 /* Page is submitted locked and unlocked once the IO completes */ 4198 wait_on_page_locked(page); 4199 if (PageError(page)) { 4200 errors++; 4201 if (i == 0) 4202 primary_failed = true; 4203 } 4204 4205 /* Drop our reference */ 4206 put_page(page); 4207 4208 /* Drop the reference from the writing run */ 4209 put_page(page); 4210 } 4211 4212 /* log error, force error return */ 4213 if (primary_failed) { 4214 btrfs_err(device->fs_info, "error writing primary super block to device %llu", 4215 device->devid); 4216 return -1; 4217 } 4218 4219 return errors < i ? 0 : -1; 4220 } 4221 4222 /* 4223 * endio for the write_dev_flush, this will wake anyone waiting 4224 * for the barrier when it is done 4225 */ 4226 static void btrfs_end_empty_barrier(struct bio *bio) 4227 { 4228 complete(bio->bi_private); 4229 } 4230 4231 /* 4232 * Submit a flush request to the device if it supports it. Error handling is 4233 * done in the waiting counterpart. 4234 */ 4235 static void write_dev_flush(struct btrfs_device *device) 4236 { 4237 struct bio *bio = device->flush_bio; 4238 4239 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4240 /* 4241 * When a disk has write caching disabled, we skip submission of a bio 4242 * with flush and sync requests before writing the superblock, since 4243 * it's not needed. However when the integrity checker is enabled, this 4244 * results in reports that there are metadata blocks referred by a 4245 * superblock that were not properly flushed. So don't skip the bio 4246 * submission only when the integrity checker is enabled for the sake 4247 * of simplicity, since this is a debug tool and not meant for use in 4248 * non-debug builds. 4249 */ 4250 struct request_queue *q = bdev_get_queue(device->bdev); 4251 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 4252 return; 4253 #endif 4254 4255 bio_reset(bio, device->bdev, REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH); 4256 bio->bi_end_io = btrfs_end_empty_barrier; 4257 init_completion(&device->flush_wait); 4258 bio->bi_private = &device->flush_wait; 4259 4260 btrfsic_submit_bio(bio); 4261 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 4262 } 4263 4264 /* 4265 * If the flush bio has been submitted by write_dev_flush, wait for it. 4266 */ 4267 static blk_status_t wait_dev_flush(struct btrfs_device *device) 4268 { 4269 struct bio *bio = device->flush_bio; 4270 4271 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 4272 return BLK_STS_OK; 4273 4274 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 4275 wait_for_completion_io(&device->flush_wait); 4276 4277 return bio->bi_status; 4278 } 4279 4280 static int check_barrier_error(struct btrfs_fs_info *fs_info) 4281 { 4282 if (!btrfs_check_rw_degradable(fs_info, NULL)) 4283 return -EIO; 4284 return 0; 4285 } 4286 4287 /* 4288 * send an empty flush down to each device in parallel, 4289 * then wait for them 4290 */ 4291 static int barrier_all_devices(struct btrfs_fs_info *info) 4292 { 4293 struct list_head *head; 4294 struct btrfs_device *dev; 4295 int errors_wait = 0; 4296 blk_status_t ret; 4297 4298 lockdep_assert_held(&info->fs_devices->device_list_mutex); 4299 /* send down all the barriers */ 4300 head = &info->fs_devices->devices; 4301 list_for_each_entry(dev, head, dev_list) { 4302 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4303 continue; 4304 if (!dev->bdev) 4305 continue; 4306 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4307 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4308 continue; 4309 4310 write_dev_flush(dev); 4311 dev->last_flush_error = BLK_STS_OK; 4312 } 4313 4314 /* wait for all the barriers */ 4315 list_for_each_entry(dev, head, dev_list) { 4316 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 4317 continue; 4318 if (!dev->bdev) { 4319 errors_wait++; 4320 continue; 4321 } 4322 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4323 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4324 continue; 4325 4326 ret = wait_dev_flush(dev); 4327 if (ret) { 4328 dev->last_flush_error = ret; 4329 btrfs_dev_stat_inc_and_print(dev, 4330 BTRFS_DEV_STAT_FLUSH_ERRS); 4331 errors_wait++; 4332 } 4333 } 4334 4335 if (errors_wait) { 4336 /* 4337 * At some point we need the status of all disks 4338 * to arrive at the volume status. So error checking 4339 * is being pushed to a separate loop. 4340 */ 4341 return check_barrier_error(info); 4342 } 4343 return 0; 4344 } 4345 4346 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 4347 { 4348 int raid_type; 4349 int min_tolerated = INT_MAX; 4350 4351 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 4352 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 4353 min_tolerated = min_t(int, min_tolerated, 4354 btrfs_raid_array[BTRFS_RAID_SINGLE]. 4355 tolerated_failures); 4356 4357 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 4358 if (raid_type == BTRFS_RAID_SINGLE) 4359 continue; 4360 if (!(flags & btrfs_raid_array[raid_type].bg_flag)) 4361 continue; 4362 min_tolerated = min_t(int, min_tolerated, 4363 btrfs_raid_array[raid_type]. 4364 tolerated_failures); 4365 } 4366 4367 if (min_tolerated == INT_MAX) { 4368 pr_warn("BTRFS: unknown raid flag: %llu", flags); 4369 min_tolerated = 0; 4370 } 4371 4372 return min_tolerated; 4373 } 4374 4375 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 4376 { 4377 struct list_head *head; 4378 struct btrfs_device *dev; 4379 struct btrfs_super_block *sb; 4380 struct btrfs_dev_item *dev_item; 4381 int ret; 4382 int do_barriers; 4383 int max_errors; 4384 int total_errors = 0; 4385 u64 flags; 4386 4387 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 4388 4389 /* 4390 * max_mirrors == 0 indicates we're from commit_transaction, 4391 * not from fsync where the tree roots in fs_info have not 4392 * been consistent on disk. 4393 */ 4394 if (max_mirrors == 0) 4395 backup_super_roots(fs_info); 4396 4397 sb = fs_info->super_for_commit; 4398 dev_item = &sb->dev_item; 4399 4400 mutex_lock(&fs_info->fs_devices->device_list_mutex); 4401 head = &fs_info->fs_devices->devices; 4402 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 4403 4404 if (do_barriers) { 4405 ret = barrier_all_devices(fs_info); 4406 if (ret) { 4407 mutex_unlock( 4408 &fs_info->fs_devices->device_list_mutex); 4409 btrfs_handle_fs_error(fs_info, ret, 4410 "errors while submitting device barriers."); 4411 return ret; 4412 } 4413 } 4414 4415 list_for_each_entry(dev, head, dev_list) { 4416 if (!dev->bdev) { 4417 total_errors++; 4418 continue; 4419 } 4420 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4421 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4422 continue; 4423 4424 btrfs_set_stack_device_generation(dev_item, 0); 4425 btrfs_set_stack_device_type(dev_item, dev->type); 4426 btrfs_set_stack_device_id(dev_item, dev->devid); 4427 btrfs_set_stack_device_total_bytes(dev_item, 4428 dev->commit_total_bytes); 4429 btrfs_set_stack_device_bytes_used(dev_item, 4430 dev->commit_bytes_used); 4431 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 4432 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 4433 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 4434 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 4435 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, 4436 BTRFS_FSID_SIZE); 4437 4438 flags = btrfs_super_flags(sb); 4439 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 4440 4441 ret = btrfs_validate_write_super(fs_info, sb); 4442 if (ret < 0) { 4443 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4444 btrfs_handle_fs_error(fs_info, -EUCLEAN, 4445 "unexpected superblock corruption detected"); 4446 return -EUCLEAN; 4447 } 4448 4449 ret = write_dev_supers(dev, sb, max_mirrors); 4450 if (ret) 4451 total_errors++; 4452 } 4453 if (total_errors > max_errors) { 4454 btrfs_err(fs_info, "%d errors while writing supers", 4455 total_errors); 4456 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4457 4458 /* FUA is masked off if unsupported and can't be the reason */ 4459 btrfs_handle_fs_error(fs_info, -EIO, 4460 "%d errors while writing supers", 4461 total_errors); 4462 return -EIO; 4463 } 4464 4465 total_errors = 0; 4466 list_for_each_entry(dev, head, dev_list) { 4467 if (!dev->bdev) 4468 continue; 4469 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 4470 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 4471 continue; 4472 4473 ret = wait_dev_supers(dev, max_mirrors); 4474 if (ret) 4475 total_errors++; 4476 } 4477 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 4478 if (total_errors > max_errors) { 4479 btrfs_handle_fs_error(fs_info, -EIO, 4480 "%d errors while writing supers", 4481 total_errors); 4482 return -EIO; 4483 } 4484 return 0; 4485 } 4486 4487 /* Drop a fs root from the radix tree and free it. */ 4488 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 4489 struct btrfs_root *root) 4490 { 4491 bool drop_ref = false; 4492 4493 spin_lock(&fs_info->fs_roots_radix_lock); 4494 radix_tree_delete(&fs_info->fs_roots_radix, 4495 (unsigned long)root->root_key.objectid); 4496 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) 4497 drop_ref = true; 4498 spin_unlock(&fs_info->fs_roots_radix_lock); 4499 4500 if (BTRFS_FS_ERROR(fs_info)) { 4501 ASSERT(root->log_root == NULL); 4502 if (root->reloc_root) { 4503 btrfs_put_root(root->reloc_root); 4504 root->reloc_root = NULL; 4505 } 4506 } 4507 4508 if (drop_ref) 4509 btrfs_put_root(root); 4510 } 4511 4512 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 4513 { 4514 u64 root_objectid = 0; 4515 struct btrfs_root *gang[8]; 4516 int i = 0; 4517 int err = 0; 4518 unsigned int ret = 0; 4519 4520 while (1) { 4521 spin_lock(&fs_info->fs_roots_radix_lock); 4522 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4523 (void **)gang, root_objectid, 4524 ARRAY_SIZE(gang)); 4525 if (!ret) { 4526 spin_unlock(&fs_info->fs_roots_radix_lock); 4527 break; 4528 } 4529 root_objectid = gang[ret - 1]->root_key.objectid + 1; 4530 4531 for (i = 0; i < ret; i++) { 4532 /* Avoid to grab roots in dead_roots */ 4533 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 4534 gang[i] = NULL; 4535 continue; 4536 } 4537 /* grab all the search result for later use */ 4538 gang[i] = btrfs_grab_root(gang[i]); 4539 } 4540 spin_unlock(&fs_info->fs_roots_radix_lock); 4541 4542 for (i = 0; i < ret; i++) { 4543 if (!gang[i]) 4544 continue; 4545 root_objectid = gang[i]->root_key.objectid; 4546 err = btrfs_orphan_cleanup(gang[i]); 4547 if (err) 4548 break; 4549 btrfs_put_root(gang[i]); 4550 } 4551 root_objectid++; 4552 } 4553 4554 /* release the uncleaned roots due to error */ 4555 for (; i < ret; i++) { 4556 if (gang[i]) 4557 btrfs_put_root(gang[i]); 4558 } 4559 return err; 4560 } 4561 4562 int btrfs_commit_super(struct btrfs_fs_info *fs_info) 4563 { 4564 struct btrfs_root *root = fs_info->tree_root; 4565 struct btrfs_trans_handle *trans; 4566 4567 mutex_lock(&fs_info->cleaner_mutex); 4568 btrfs_run_delayed_iputs(fs_info); 4569 mutex_unlock(&fs_info->cleaner_mutex); 4570 wake_up_process(fs_info->cleaner_kthread); 4571 4572 /* wait until ongoing cleanup work done */ 4573 down_write(&fs_info->cleanup_work_sem); 4574 up_write(&fs_info->cleanup_work_sem); 4575 4576 trans = btrfs_join_transaction(root); 4577 if (IS_ERR(trans)) 4578 return PTR_ERR(trans); 4579 return btrfs_commit_transaction(trans); 4580 } 4581 4582 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info) 4583 { 4584 struct btrfs_transaction *trans; 4585 struct btrfs_transaction *tmp; 4586 bool found = false; 4587 4588 if (list_empty(&fs_info->trans_list)) 4589 return; 4590 4591 /* 4592 * This function is only called at the very end of close_ctree(), 4593 * thus no other running transaction, no need to take trans_lock. 4594 */ 4595 ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags)); 4596 list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) { 4597 struct extent_state *cached = NULL; 4598 u64 dirty_bytes = 0; 4599 u64 cur = 0; 4600 u64 found_start; 4601 u64 found_end; 4602 4603 found = true; 4604 while (!find_first_extent_bit(&trans->dirty_pages, cur, 4605 &found_start, &found_end, EXTENT_DIRTY, &cached)) { 4606 dirty_bytes += found_end + 1 - found_start; 4607 cur = found_end + 1; 4608 } 4609 btrfs_warn(fs_info, 4610 "transaction %llu (with %llu dirty metadata bytes) is not committed", 4611 trans->transid, dirty_bytes); 4612 btrfs_cleanup_one_transaction(trans, fs_info); 4613 4614 if (trans == fs_info->running_transaction) 4615 fs_info->running_transaction = NULL; 4616 list_del_init(&trans->list); 4617 4618 btrfs_put_transaction(trans); 4619 trace_btrfs_transaction_commit(fs_info); 4620 } 4621 ASSERT(!found); 4622 } 4623 4624 void __cold close_ctree(struct btrfs_fs_info *fs_info) 4625 { 4626 int ret; 4627 4628 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 4629 /* 4630 * We don't want the cleaner to start new transactions, add more delayed 4631 * iputs, etc. while we're closing. We can't use kthread_stop() yet 4632 * because that frees the task_struct, and the transaction kthread might 4633 * still try to wake up the cleaner. 4634 */ 4635 kthread_park(fs_info->cleaner_kthread); 4636 4637 /* 4638 * If we had UNFINISHED_DROPS we could still be processing them, so 4639 * clear that bit and wake up relocation so it can stop. 4640 */ 4641 btrfs_wake_unfinished_drop(fs_info); 4642 4643 /* wait for the qgroup rescan worker to stop */ 4644 btrfs_qgroup_wait_for_completion(fs_info, false); 4645 4646 /* wait for the uuid_scan task to finish */ 4647 down(&fs_info->uuid_tree_rescan_sem); 4648 /* avoid complains from lockdep et al., set sem back to initial state */ 4649 up(&fs_info->uuid_tree_rescan_sem); 4650 4651 /* pause restriper - we want to resume on mount */ 4652 btrfs_pause_balance(fs_info); 4653 4654 btrfs_dev_replace_suspend_for_unmount(fs_info); 4655 4656 btrfs_scrub_cancel(fs_info); 4657 4658 /* wait for any defraggers to finish */ 4659 wait_event(fs_info->transaction_wait, 4660 (atomic_read(&fs_info->defrag_running) == 0)); 4661 4662 /* clear out the rbtree of defraggable inodes */ 4663 btrfs_cleanup_defrag_inodes(fs_info); 4664 4665 cancel_work_sync(&fs_info->async_reclaim_work); 4666 cancel_work_sync(&fs_info->async_data_reclaim_work); 4667 cancel_work_sync(&fs_info->preempt_reclaim_work); 4668 4669 cancel_work_sync(&fs_info->reclaim_bgs_work); 4670 4671 /* Cancel or finish ongoing discard work */ 4672 btrfs_discard_cleanup(fs_info); 4673 4674 if (!sb_rdonly(fs_info->sb)) { 4675 /* 4676 * The cleaner kthread is stopped, so do one final pass over 4677 * unused block groups. 4678 */ 4679 btrfs_delete_unused_bgs(fs_info); 4680 4681 /* 4682 * There might be existing delayed inode workers still running 4683 * and holding an empty delayed inode item. We must wait for 4684 * them to complete first because they can create a transaction. 4685 * This happens when someone calls btrfs_balance_delayed_items() 4686 * and then a transaction commit runs the same delayed nodes 4687 * before any delayed worker has done something with the nodes. 4688 * We must wait for any worker here and not at transaction 4689 * commit time since that could cause a deadlock. 4690 * This is a very rare case. 4691 */ 4692 btrfs_flush_workqueue(fs_info->delayed_workers); 4693 4694 ret = btrfs_commit_super(fs_info); 4695 if (ret) 4696 btrfs_err(fs_info, "commit super ret %d", ret); 4697 } 4698 4699 if (BTRFS_FS_ERROR(fs_info)) 4700 btrfs_error_commit_super(fs_info); 4701 4702 kthread_stop(fs_info->transaction_kthread); 4703 kthread_stop(fs_info->cleaner_kthread); 4704 4705 ASSERT(list_empty(&fs_info->delayed_iputs)); 4706 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 4707 4708 if (btrfs_check_quota_leak(fs_info)) { 4709 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); 4710 btrfs_err(fs_info, "qgroup reserved space leaked"); 4711 } 4712 4713 btrfs_free_qgroup_config(fs_info); 4714 ASSERT(list_empty(&fs_info->delalloc_roots)); 4715 4716 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 4717 btrfs_info(fs_info, "at unmount delalloc count %lld", 4718 percpu_counter_sum(&fs_info->delalloc_bytes)); 4719 } 4720 4721 if (percpu_counter_sum(&fs_info->ordered_bytes)) 4722 btrfs_info(fs_info, "at unmount dio bytes count %lld", 4723 percpu_counter_sum(&fs_info->ordered_bytes)); 4724 4725 btrfs_sysfs_remove_mounted(fs_info); 4726 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 4727 4728 btrfs_put_block_group_cache(fs_info); 4729 4730 /* 4731 * we must make sure there is not any read request to 4732 * submit after we stopping all workers. 4733 */ 4734 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 4735 btrfs_stop_all_workers(fs_info); 4736 4737 /* We shouldn't have any transaction open at this point */ 4738 warn_about_uncommitted_trans(fs_info); 4739 4740 clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 4741 free_root_pointers(fs_info, true); 4742 btrfs_free_fs_roots(fs_info); 4743 4744 /* 4745 * We must free the block groups after dropping the fs_roots as we could 4746 * have had an IO error and have left over tree log blocks that aren't 4747 * cleaned up until the fs roots are freed. This makes the block group 4748 * accounting appear to be wrong because there's pending reserved bytes, 4749 * so make sure we do the block group cleanup afterwards. 4750 */ 4751 btrfs_free_block_groups(fs_info); 4752 4753 iput(fs_info->btree_inode); 4754 4755 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4756 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 4757 btrfsic_unmount(fs_info->fs_devices); 4758 #endif 4759 4760 btrfs_mapping_tree_free(&fs_info->mapping_tree); 4761 btrfs_close_devices(fs_info->fs_devices); 4762 } 4763 4764 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 4765 int atomic) 4766 { 4767 int ret; 4768 struct inode *btree_inode = buf->pages[0]->mapping->host; 4769 4770 ret = extent_buffer_uptodate(buf); 4771 if (!ret) 4772 return ret; 4773 4774 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 4775 parent_transid, atomic); 4776 if (ret == -EAGAIN) 4777 return ret; 4778 return !ret; 4779 } 4780 4781 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 4782 { 4783 struct btrfs_fs_info *fs_info = buf->fs_info; 4784 u64 transid = btrfs_header_generation(buf); 4785 int was_dirty; 4786 4787 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 4788 /* 4789 * This is a fast path so only do this check if we have sanity tests 4790 * enabled. Normal people shouldn't be using unmapped buffers as dirty 4791 * outside of the sanity tests. 4792 */ 4793 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) 4794 return; 4795 #endif 4796 btrfs_assert_tree_write_locked(buf); 4797 if (transid != fs_info->generation) 4798 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 4799 buf->start, transid, fs_info->generation); 4800 was_dirty = set_extent_buffer_dirty(buf); 4801 if (!was_dirty) 4802 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 4803 buf->len, 4804 fs_info->dirty_metadata_batch); 4805 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 4806 /* 4807 * Since btrfs_mark_buffer_dirty() can be called with item pointer set 4808 * but item data not updated. 4809 * So here we should only check item pointers, not item data. 4810 */ 4811 if (btrfs_header_level(buf) == 0 && 4812 btrfs_check_leaf_relaxed(buf)) { 4813 btrfs_print_leaf(buf); 4814 ASSERT(0); 4815 } 4816 #endif 4817 } 4818 4819 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 4820 int flush_delayed) 4821 { 4822 /* 4823 * looks as though older kernels can get into trouble with 4824 * this code, they end up stuck in balance_dirty_pages forever 4825 */ 4826 int ret; 4827 4828 if (current->flags & PF_MEMALLOC) 4829 return; 4830 4831 if (flush_delayed) 4832 btrfs_balance_delayed_items(fs_info); 4833 4834 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, 4835 BTRFS_DIRTY_METADATA_THRESH, 4836 fs_info->dirty_metadata_batch); 4837 if (ret > 0) { 4838 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 4839 } 4840 } 4841 4842 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 4843 { 4844 __btrfs_btree_balance_dirty(fs_info, 1); 4845 } 4846 4847 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 4848 { 4849 __btrfs_btree_balance_dirty(fs_info, 0); 4850 } 4851 4852 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, 4853 struct btrfs_key *first_key) 4854 { 4855 return btree_read_extent_buffer_pages(buf, parent_transid, 4856 level, first_key); 4857 } 4858 4859 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4860 { 4861 /* cleanup FS via transaction */ 4862 btrfs_cleanup_transaction(fs_info); 4863 4864 mutex_lock(&fs_info->cleaner_mutex); 4865 btrfs_run_delayed_iputs(fs_info); 4866 mutex_unlock(&fs_info->cleaner_mutex); 4867 4868 down_write(&fs_info->cleanup_work_sem); 4869 up_write(&fs_info->cleanup_work_sem); 4870 } 4871 4872 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) 4873 { 4874 struct btrfs_root *gang[8]; 4875 u64 root_objectid = 0; 4876 int ret; 4877 4878 spin_lock(&fs_info->fs_roots_radix_lock); 4879 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 4880 (void **)gang, root_objectid, 4881 ARRAY_SIZE(gang))) != 0) { 4882 int i; 4883 4884 for (i = 0; i < ret; i++) 4885 gang[i] = btrfs_grab_root(gang[i]); 4886 spin_unlock(&fs_info->fs_roots_radix_lock); 4887 4888 for (i = 0; i < ret; i++) { 4889 if (!gang[i]) 4890 continue; 4891 root_objectid = gang[i]->root_key.objectid; 4892 btrfs_free_log(NULL, gang[i]); 4893 btrfs_put_root(gang[i]); 4894 } 4895 root_objectid++; 4896 spin_lock(&fs_info->fs_roots_radix_lock); 4897 } 4898 spin_unlock(&fs_info->fs_roots_radix_lock); 4899 btrfs_free_log_root_tree(NULL, fs_info); 4900 } 4901 4902 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4903 { 4904 struct btrfs_ordered_extent *ordered; 4905 4906 spin_lock(&root->ordered_extent_lock); 4907 /* 4908 * This will just short circuit the ordered completion stuff which will 4909 * make sure the ordered extent gets properly cleaned up. 4910 */ 4911 list_for_each_entry(ordered, &root->ordered_extents, 4912 root_extent_list) 4913 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4914 spin_unlock(&root->ordered_extent_lock); 4915 } 4916 4917 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4918 { 4919 struct btrfs_root *root; 4920 struct list_head splice; 4921 4922 INIT_LIST_HEAD(&splice); 4923 4924 spin_lock(&fs_info->ordered_root_lock); 4925 list_splice_init(&fs_info->ordered_roots, &splice); 4926 while (!list_empty(&splice)) { 4927 root = list_first_entry(&splice, struct btrfs_root, 4928 ordered_root); 4929 list_move_tail(&root->ordered_root, 4930 &fs_info->ordered_roots); 4931 4932 spin_unlock(&fs_info->ordered_root_lock); 4933 btrfs_destroy_ordered_extents(root); 4934 4935 cond_resched(); 4936 spin_lock(&fs_info->ordered_root_lock); 4937 } 4938 spin_unlock(&fs_info->ordered_root_lock); 4939 4940 /* 4941 * We need this here because if we've been flipped read-only we won't 4942 * get sync() from the umount, so we need to make sure any ordered 4943 * extents that haven't had their dirty pages IO start writeout yet 4944 * actually get run and error out properly. 4945 */ 4946 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 4947 } 4948 4949 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4950 struct btrfs_fs_info *fs_info) 4951 { 4952 struct rb_node *node; 4953 struct btrfs_delayed_ref_root *delayed_refs; 4954 struct btrfs_delayed_ref_node *ref; 4955 int ret = 0; 4956 4957 delayed_refs = &trans->delayed_refs; 4958 4959 spin_lock(&delayed_refs->lock); 4960 if (atomic_read(&delayed_refs->num_entries) == 0) { 4961 spin_unlock(&delayed_refs->lock); 4962 btrfs_debug(fs_info, "delayed_refs has NO entry"); 4963 return ret; 4964 } 4965 4966 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { 4967 struct btrfs_delayed_ref_head *head; 4968 struct rb_node *n; 4969 bool pin_bytes = false; 4970 4971 head = rb_entry(node, struct btrfs_delayed_ref_head, 4972 href_node); 4973 if (btrfs_delayed_ref_lock(delayed_refs, head)) 4974 continue; 4975 4976 spin_lock(&head->lock); 4977 while ((n = rb_first_cached(&head->ref_tree)) != NULL) { 4978 ref = rb_entry(n, struct btrfs_delayed_ref_node, 4979 ref_node); 4980 ref->in_tree = 0; 4981 rb_erase_cached(&ref->ref_node, &head->ref_tree); 4982 RB_CLEAR_NODE(&ref->ref_node); 4983 if (!list_empty(&ref->add_list)) 4984 list_del(&ref->add_list); 4985 atomic_dec(&delayed_refs->num_entries); 4986 btrfs_put_delayed_ref(ref); 4987 } 4988 if (head->must_insert_reserved) 4989 pin_bytes = true; 4990 btrfs_free_delayed_extent_op(head->extent_op); 4991 btrfs_delete_ref_head(delayed_refs, head); 4992 spin_unlock(&head->lock); 4993 spin_unlock(&delayed_refs->lock); 4994 mutex_unlock(&head->mutex); 4995 4996 if (pin_bytes) { 4997 struct btrfs_block_group *cache; 4998 4999 cache = btrfs_lookup_block_group(fs_info, head->bytenr); 5000 BUG_ON(!cache); 5001 5002 spin_lock(&cache->space_info->lock); 5003 spin_lock(&cache->lock); 5004 cache->pinned += head->num_bytes; 5005 btrfs_space_info_update_bytes_pinned(fs_info, 5006 cache->space_info, head->num_bytes); 5007 cache->reserved -= head->num_bytes; 5008 cache->space_info->bytes_reserved -= head->num_bytes; 5009 spin_unlock(&cache->lock); 5010 spin_unlock(&cache->space_info->lock); 5011 5012 btrfs_put_block_group(cache); 5013 5014 btrfs_error_unpin_extent_range(fs_info, head->bytenr, 5015 head->bytenr + head->num_bytes - 1); 5016 } 5017 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); 5018 btrfs_put_delayed_ref_head(head); 5019 cond_resched(); 5020 spin_lock(&delayed_refs->lock); 5021 } 5022 btrfs_qgroup_destroy_extent_records(trans); 5023 5024 spin_unlock(&delayed_refs->lock); 5025 5026 return ret; 5027 } 5028 5029 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 5030 { 5031 struct btrfs_inode *btrfs_inode; 5032 struct list_head splice; 5033 5034 INIT_LIST_HEAD(&splice); 5035 5036 spin_lock(&root->delalloc_lock); 5037 list_splice_init(&root->delalloc_inodes, &splice); 5038 5039 while (!list_empty(&splice)) { 5040 struct inode *inode = NULL; 5041 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 5042 delalloc_inodes); 5043 __btrfs_del_delalloc_inode(root, btrfs_inode); 5044 spin_unlock(&root->delalloc_lock); 5045 5046 /* 5047 * Make sure we get a live inode and that it'll not disappear 5048 * meanwhile. 5049 */ 5050 inode = igrab(&btrfs_inode->vfs_inode); 5051 if (inode) { 5052 invalidate_inode_pages2(inode->i_mapping); 5053 iput(inode); 5054 } 5055 spin_lock(&root->delalloc_lock); 5056 } 5057 spin_unlock(&root->delalloc_lock); 5058 } 5059 5060 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 5061 { 5062 struct btrfs_root *root; 5063 struct list_head splice; 5064 5065 INIT_LIST_HEAD(&splice); 5066 5067 spin_lock(&fs_info->delalloc_root_lock); 5068 list_splice_init(&fs_info->delalloc_roots, &splice); 5069 while (!list_empty(&splice)) { 5070 root = list_first_entry(&splice, struct btrfs_root, 5071 delalloc_root); 5072 root = btrfs_grab_root(root); 5073 BUG_ON(!root); 5074 spin_unlock(&fs_info->delalloc_root_lock); 5075 5076 btrfs_destroy_delalloc_inodes(root); 5077 btrfs_put_root(root); 5078 5079 spin_lock(&fs_info->delalloc_root_lock); 5080 } 5081 spin_unlock(&fs_info->delalloc_root_lock); 5082 } 5083 5084 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 5085 struct extent_io_tree *dirty_pages, 5086 int mark) 5087 { 5088 int ret; 5089 struct extent_buffer *eb; 5090 u64 start = 0; 5091 u64 end; 5092 5093 while (1) { 5094 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 5095 mark, NULL); 5096 if (ret) 5097 break; 5098 5099 clear_extent_bits(dirty_pages, start, end, mark); 5100 while (start <= end) { 5101 eb = find_extent_buffer(fs_info, start); 5102 start += fs_info->nodesize; 5103 if (!eb) 5104 continue; 5105 wait_on_extent_buffer_writeback(eb); 5106 5107 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 5108 &eb->bflags)) 5109 clear_extent_buffer_dirty(eb); 5110 free_extent_buffer_stale(eb); 5111 } 5112 } 5113 5114 return ret; 5115 } 5116 5117 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 5118 struct extent_io_tree *unpin) 5119 { 5120 u64 start; 5121 u64 end; 5122 int ret; 5123 5124 while (1) { 5125 struct extent_state *cached_state = NULL; 5126 5127 /* 5128 * The btrfs_finish_extent_commit() may get the same range as 5129 * ours between find_first_extent_bit and clear_extent_dirty. 5130 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin 5131 * the same extent range. 5132 */ 5133 mutex_lock(&fs_info->unused_bg_unpin_mutex); 5134 ret = find_first_extent_bit(unpin, 0, &start, &end, 5135 EXTENT_DIRTY, &cached_state); 5136 if (ret) { 5137 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5138 break; 5139 } 5140 5141 clear_extent_dirty(unpin, start, end, &cached_state); 5142 free_extent_state(cached_state); 5143 btrfs_error_unpin_extent_range(fs_info, start, end); 5144 mutex_unlock(&fs_info->unused_bg_unpin_mutex); 5145 cond_resched(); 5146 } 5147 5148 return 0; 5149 } 5150 5151 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) 5152 { 5153 struct inode *inode; 5154 5155 inode = cache->io_ctl.inode; 5156 if (inode) { 5157 invalidate_inode_pages2(inode->i_mapping); 5158 BTRFS_I(inode)->generation = 0; 5159 cache->io_ctl.inode = NULL; 5160 iput(inode); 5161 } 5162 ASSERT(cache->io_ctl.pages == NULL); 5163 btrfs_put_block_group(cache); 5164 } 5165 5166 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 5167 struct btrfs_fs_info *fs_info) 5168 { 5169 struct btrfs_block_group *cache; 5170 5171 spin_lock(&cur_trans->dirty_bgs_lock); 5172 while (!list_empty(&cur_trans->dirty_bgs)) { 5173 cache = list_first_entry(&cur_trans->dirty_bgs, 5174 struct btrfs_block_group, 5175 dirty_list); 5176 5177 if (!list_empty(&cache->io_list)) { 5178 spin_unlock(&cur_trans->dirty_bgs_lock); 5179 list_del_init(&cache->io_list); 5180 btrfs_cleanup_bg_io(cache); 5181 spin_lock(&cur_trans->dirty_bgs_lock); 5182 } 5183 5184 list_del_init(&cache->dirty_list); 5185 spin_lock(&cache->lock); 5186 cache->disk_cache_state = BTRFS_DC_ERROR; 5187 spin_unlock(&cache->lock); 5188 5189 spin_unlock(&cur_trans->dirty_bgs_lock); 5190 btrfs_put_block_group(cache); 5191 btrfs_delayed_refs_rsv_release(fs_info, 1); 5192 spin_lock(&cur_trans->dirty_bgs_lock); 5193 } 5194 spin_unlock(&cur_trans->dirty_bgs_lock); 5195 5196 /* 5197 * Refer to the definition of io_bgs member for details why it's safe 5198 * to use it without any locking 5199 */ 5200 while (!list_empty(&cur_trans->io_bgs)) { 5201 cache = list_first_entry(&cur_trans->io_bgs, 5202 struct btrfs_block_group, 5203 io_list); 5204 5205 list_del_init(&cache->io_list); 5206 spin_lock(&cache->lock); 5207 cache->disk_cache_state = BTRFS_DC_ERROR; 5208 spin_unlock(&cache->lock); 5209 btrfs_cleanup_bg_io(cache); 5210 } 5211 } 5212 5213 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 5214 struct btrfs_fs_info *fs_info) 5215 { 5216 struct btrfs_device *dev, *tmp; 5217 5218 btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 5219 ASSERT(list_empty(&cur_trans->dirty_bgs)); 5220 ASSERT(list_empty(&cur_trans->io_bgs)); 5221 5222 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, 5223 post_commit_list) { 5224 list_del_init(&dev->post_commit_list); 5225 } 5226 5227 btrfs_destroy_delayed_refs(cur_trans, fs_info); 5228 5229 cur_trans->state = TRANS_STATE_COMMIT_START; 5230 wake_up(&fs_info->transaction_blocked_wait); 5231 5232 cur_trans->state = TRANS_STATE_UNBLOCKED; 5233 wake_up(&fs_info->transaction_wait); 5234 5235 btrfs_destroy_delayed_inodes(fs_info); 5236 5237 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 5238 EXTENT_DIRTY); 5239 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); 5240 5241 btrfs_free_redirty_list(cur_trans); 5242 5243 cur_trans->state =TRANS_STATE_COMPLETED; 5244 wake_up(&cur_trans->commit_wait); 5245 } 5246 5247 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 5248 { 5249 struct btrfs_transaction *t; 5250 5251 mutex_lock(&fs_info->transaction_kthread_mutex); 5252 5253 spin_lock(&fs_info->trans_lock); 5254 while (!list_empty(&fs_info->trans_list)) { 5255 t = list_first_entry(&fs_info->trans_list, 5256 struct btrfs_transaction, list); 5257 if (t->state >= TRANS_STATE_COMMIT_START) { 5258 refcount_inc(&t->use_count); 5259 spin_unlock(&fs_info->trans_lock); 5260 btrfs_wait_for_commit(fs_info, t->transid); 5261 btrfs_put_transaction(t); 5262 spin_lock(&fs_info->trans_lock); 5263 continue; 5264 } 5265 if (t == fs_info->running_transaction) { 5266 t->state = TRANS_STATE_COMMIT_DOING; 5267 spin_unlock(&fs_info->trans_lock); 5268 /* 5269 * We wait for 0 num_writers since we don't hold a trans 5270 * handle open currently for this transaction. 5271 */ 5272 wait_event(t->writer_wait, 5273 atomic_read(&t->num_writers) == 0); 5274 } else { 5275 spin_unlock(&fs_info->trans_lock); 5276 } 5277 btrfs_cleanup_one_transaction(t, fs_info); 5278 5279 spin_lock(&fs_info->trans_lock); 5280 if (t == fs_info->running_transaction) 5281 fs_info->running_transaction = NULL; 5282 list_del_init(&t->list); 5283 spin_unlock(&fs_info->trans_lock); 5284 5285 btrfs_put_transaction(t); 5286 trace_btrfs_transaction_commit(fs_info); 5287 spin_lock(&fs_info->trans_lock); 5288 } 5289 spin_unlock(&fs_info->trans_lock); 5290 btrfs_destroy_all_ordered_extents(fs_info); 5291 btrfs_destroy_delayed_inodes(fs_info); 5292 btrfs_assert_delayed_root_empty(fs_info); 5293 btrfs_destroy_all_delalloc_inodes(fs_info); 5294 btrfs_drop_all_logs(fs_info); 5295 mutex_unlock(&fs_info->transaction_kthread_mutex); 5296 5297 return 0; 5298 } 5299 5300 int btrfs_init_root_free_objectid(struct btrfs_root *root) 5301 { 5302 struct btrfs_path *path; 5303 int ret; 5304 struct extent_buffer *l; 5305 struct btrfs_key search_key; 5306 struct btrfs_key found_key; 5307 int slot; 5308 5309 path = btrfs_alloc_path(); 5310 if (!path) 5311 return -ENOMEM; 5312 5313 search_key.objectid = BTRFS_LAST_FREE_OBJECTID; 5314 search_key.type = -1; 5315 search_key.offset = (u64)-1; 5316 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); 5317 if (ret < 0) 5318 goto error; 5319 BUG_ON(ret == 0); /* Corruption */ 5320 if (path->slots[0] > 0) { 5321 slot = path->slots[0] - 1; 5322 l = path->nodes[0]; 5323 btrfs_item_key_to_cpu(l, &found_key, slot); 5324 root->free_objectid = max_t(u64, found_key.objectid + 1, 5325 BTRFS_FIRST_FREE_OBJECTID); 5326 } else { 5327 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID; 5328 } 5329 ret = 0; 5330 error: 5331 btrfs_free_path(path); 5332 return ret; 5333 } 5334 5335 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid) 5336 { 5337 int ret; 5338 mutex_lock(&root->objectid_mutex); 5339 5340 if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) { 5341 btrfs_warn(root->fs_info, 5342 "the objectid of root %llu reaches its highest value", 5343 root->root_key.objectid); 5344 ret = -ENOSPC; 5345 goto out; 5346 } 5347 5348 *objectid = root->free_objectid++; 5349 ret = 0; 5350 out: 5351 mutex_unlock(&root->objectid_mutex); 5352 return ret; 5353 } 5354