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