1 /* 2 * Copyright (C) 2007 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/fs.h> 20 #include <linux/blkdev.h> 21 #include <linux/scatterlist.h> 22 #include <linux/swap.h> 23 #include <linux/radix-tree.h> 24 #include <linux/writeback.h> 25 #include <linux/buffer_head.h> 26 #include <linux/workqueue.h> 27 #include <linux/kthread.h> 28 #include <linux/slab.h> 29 #include <linux/migrate.h> 30 #include <linux/ratelimit.h> 31 #include <linux/uuid.h> 32 #include <linux/semaphore.h> 33 #include <linux/error-injection.h> 34 #include <asm/unaligned.h> 35 #include "ctree.h" 36 #include "disk-io.h" 37 #include "hash.h" 38 #include "transaction.h" 39 #include "btrfs_inode.h" 40 #include "volumes.h" 41 #include "print-tree.h" 42 #include "locking.h" 43 #include "tree-log.h" 44 #include "free-space-cache.h" 45 #include "free-space-tree.h" 46 #include "inode-map.h" 47 #include "check-integrity.h" 48 #include "rcu-string.h" 49 #include "dev-replace.h" 50 #include "raid56.h" 51 #include "sysfs.h" 52 #include "qgroup.h" 53 #include "compression.h" 54 #include "tree-checker.h" 55 #include "ref-verify.h" 56 57 #ifdef CONFIG_X86 58 #include <asm/cpufeature.h> 59 #endif 60 61 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ 62 BTRFS_HEADER_FLAG_RELOC |\ 63 BTRFS_SUPER_FLAG_ERROR |\ 64 BTRFS_SUPER_FLAG_SEEDING |\ 65 BTRFS_SUPER_FLAG_METADUMP |\ 66 BTRFS_SUPER_FLAG_METADUMP_V2) 67 68 static const struct extent_io_ops btree_extent_io_ops; 69 static void end_workqueue_fn(struct btrfs_work *work); 70 static void free_fs_root(struct btrfs_root *root); 71 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info); 72 static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 73 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 74 struct btrfs_fs_info *fs_info); 75 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 76 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 77 struct extent_io_tree *dirty_pages, 78 int mark); 79 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 80 struct extent_io_tree *pinned_extents); 81 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); 82 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); 83 84 /* 85 * btrfs_end_io_wq structs are used to do processing in task context when an IO 86 * is complete. This is used during reads to verify checksums, and it is used 87 * by writes to insert metadata for new file extents after IO is complete. 88 */ 89 struct btrfs_end_io_wq { 90 struct bio *bio; 91 bio_end_io_t *end_io; 92 void *private; 93 struct btrfs_fs_info *info; 94 blk_status_t status; 95 enum btrfs_wq_endio_type metadata; 96 struct btrfs_work work; 97 }; 98 99 static struct kmem_cache *btrfs_end_io_wq_cache; 100 101 int __init btrfs_end_io_wq_init(void) 102 { 103 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", 104 sizeof(struct btrfs_end_io_wq), 105 0, 106 SLAB_MEM_SPREAD, 107 NULL); 108 if (!btrfs_end_io_wq_cache) 109 return -ENOMEM; 110 return 0; 111 } 112 113 void btrfs_end_io_wq_exit(void) 114 { 115 kmem_cache_destroy(btrfs_end_io_wq_cache); 116 } 117 118 /* 119 * async submit bios are used to offload expensive checksumming 120 * onto the worker threads. They checksum file and metadata bios 121 * just before they are sent down the IO stack. 122 */ 123 struct async_submit_bio { 124 void *private_data; 125 struct btrfs_fs_info *fs_info; 126 struct bio *bio; 127 extent_submit_bio_hook_t *submit_bio_start; 128 extent_submit_bio_hook_t *submit_bio_done; 129 int mirror_num; 130 unsigned long bio_flags; 131 /* 132 * bio_offset is optional, can be used if the pages in the bio 133 * can't tell us where in the file the bio should go 134 */ 135 u64 bio_offset; 136 struct btrfs_work work; 137 blk_status_t status; 138 }; 139 140 /* 141 * Lockdep class keys for extent_buffer->lock's in this root. For a given 142 * eb, the lockdep key is determined by the btrfs_root it belongs to and 143 * the level the eb occupies in the tree. 144 * 145 * Different roots are used for different purposes and may nest inside each 146 * other and they require separate keysets. As lockdep keys should be 147 * static, assign keysets according to the purpose of the root as indicated 148 * by btrfs_root->objectid. This ensures that all special purpose roots 149 * have separate keysets. 150 * 151 * Lock-nesting across peer nodes is always done with the immediate parent 152 * node locked thus preventing deadlock. As lockdep doesn't know this, use 153 * subclass to avoid triggering lockdep warning in such cases. 154 * 155 * The key is set by the readpage_end_io_hook after the buffer has passed 156 * csum validation but before the pages are unlocked. It is also set by 157 * btrfs_init_new_buffer on freshly allocated blocks. 158 * 159 * We also add a check to make sure the highest level of the tree is the 160 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 161 * needs update as well. 162 */ 163 #ifdef CONFIG_DEBUG_LOCK_ALLOC 164 # if BTRFS_MAX_LEVEL != 8 165 # error 166 # endif 167 168 static struct btrfs_lockdep_keyset { 169 u64 id; /* root objectid */ 170 const char *name_stem; /* lock name stem */ 171 char names[BTRFS_MAX_LEVEL + 1][20]; 172 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 173 } btrfs_lockdep_keysets[] = { 174 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 175 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 176 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 177 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 178 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 179 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 180 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 181 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 182 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 183 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 184 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 185 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" }, 186 { .id = 0, .name_stem = "tree" }, 187 }; 188 189 void __init btrfs_init_lockdep(void) 190 { 191 int i, j; 192 193 /* initialize lockdep class names */ 194 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 195 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 196 197 for (j = 0; j < ARRAY_SIZE(ks->names); j++) 198 snprintf(ks->names[j], sizeof(ks->names[j]), 199 "btrfs-%s-%02d", ks->name_stem, j); 200 } 201 } 202 203 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 204 int level) 205 { 206 struct btrfs_lockdep_keyset *ks; 207 208 BUG_ON(level >= ARRAY_SIZE(ks->keys)); 209 210 /* find the matching keyset, id 0 is the default entry */ 211 for (ks = btrfs_lockdep_keysets; ks->id; ks++) 212 if (ks->id == objectid) 213 break; 214 215 lockdep_set_class_and_name(&eb->lock, 216 &ks->keys[level], ks->names[level]); 217 } 218 219 #endif 220 221 /* 222 * extents on the btree inode are pretty simple, there's one extent 223 * that covers the entire device 224 */ 225 struct extent_map *btree_get_extent(struct btrfs_inode *inode, 226 struct page *page, size_t pg_offset, u64 start, u64 len, 227 int create) 228 { 229 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); 230 struct extent_map_tree *em_tree = &inode->extent_tree; 231 struct extent_map *em; 232 int ret; 233 234 read_lock(&em_tree->lock); 235 em = lookup_extent_mapping(em_tree, start, len); 236 if (em) { 237 em->bdev = fs_info->fs_devices->latest_bdev; 238 read_unlock(&em_tree->lock); 239 goto out; 240 } 241 read_unlock(&em_tree->lock); 242 243 em = alloc_extent_map(); 244 if (!em) { 245 em = ERR_PTR(-ENOMEM); 246 goto out; 247 } 248 em->start = 0; 249 em->len = (u64)-1; 250 em->block_len = (u64)-1; 251 em->block_start = 0; 252 em->bdev = fs_info->fs_devices->latest_bdev; 253 254 write_lock(&em_tree->lock); 255 ret = add_extent_mapping(em_tree, em, 0); 256 if (ret == -EEXIST) { 257 free_extent_map(em); 258 em = lookup_extent_mapping(em_tree, start, len); 259 if (!em) 260 em = ERR_PTR(-EIO); 261 } else if (ret) { 262 free_extent_map(em); 263 em = ERR_PTR(ret); 264 } 265 write_unlock(&em_tree->lock); 266 267 out: 268 return em; 269 } 270 271 u32 btrfs_csum_data(const char *data, u32 seed, size_t len) 272 { 273 return btrfs_crc32c(seed, data, len); 274 } 275 276 void btrfs_csum_final(u32 crc, u8 *result) 277 { 278 put_unaligned_le32(~crc, result); 279 } 280 281 /* 282 * compute the csum for a btree block, and either verify it or write it 283 * into the csum field of the block. 284 */ 285 static int csum_tree_block(struct btrfs_fs_info *fs_info, 286 struct extent_buffer *buf, 287 int verify) 288 { 289 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); 290 char result[BTRFS_CSUM_SIZE]; 291 unsigned long len; 292 unsigned long cur_len; 293 unsigned long offset = BTRFS_CSUM_SIZE; 294 char *kaddr; 295 unsigned long map_start; 296 unsigned long map_len; 297 int err; 298 u32 crc = ~(u32)0; 299 300 len = buf->len - offset; 301 while (len > 0) { 302 err = map_private_extent_buffer(buf, offset, 32, 303 &kaddr, &map_start, &map_len); 304 if (err) 305 return err; 306 cur_len = min(len, map_len - (offset - map_start)); 307 crc = btrfs_csum_data(kaddr + offset - map_start, 308 crc, cur_len); 309 len -= cur_len; 310 offset += cur_len; 311 } 312 memset(result, 0, BTRFS_CSUM_SIZE); 313 314 btrfs_csum_final(crc, result); 315 316 if (verify) { 317 if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 318 u32 val; 319 u32 found = 0; 320 memcpy(&found, result, csum_size); 321 322 read_extent_buffer(buf, &val, 0, csum_size); 323 btrfs_warn_rl(fs_info, 324 "%s checksum verify failed on %llu wanted %X found %X level %d", 325 fs_info->sb->s_id, buf->start, 326 val, found, btrfs_header_level(buf)); 327 return -EUCLEAN; 328 } 329 } else { 330 write_extent_buffer(buf, result, 0, csum_size); 331 } 332 333 return 0; 334 } 335 336 /* 337 * we can't consider a given block up to date unless the transid of the 338 * block matches the transid in the parent node's pointer. This is how we 339 * detect blocks that either didn't get written at all or got written 340 * in the wrong place. 341 */ 342 static int verify_parent_transid(struct extent_io_tree *io_tree, 343 struct extent_buffer *eb, u64 parent_transid, 344 int atomic) 345 { 346 struct extent_state *cached_state = NULL; 347 int ret; 348 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); 349 350 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 351 return 0; 352 353 if (atomic) 354 return -EAGAIN; 355 356 if (need_lock) { 357 btrfs_tree_read_lock(eb); 358 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 359 } 360 361 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 362 &cached_state); 363 if (extent_buffer_uptodate(eb) && 364 btrfs_header_generation(eb) == parent_transid) { 365 ret = 0; 366 goto out; 367 } 368 btrfs_err_rl(eb->fs_info, 369 "parent transid verify failed on %llu wanted %llu found %llu", 370 eb->start, 371 parent_transid, btrfs_header_generation(eb)); 372 ret = 1; 373 374 /* 375 * Things reading via commit roots that don't have normal protection, 376 * like send, can have a really old block in cache that may point at a 377 * block that has been freed and re-allocated. So don't clear uptodate 378 * if we find an eb that is under IO (dirty/writeback) because we could 379 * end up reading in the stale data and then writing it back out and 380 * making everybody very sad. 381 */ 382 if (!extent_buffer_under_io(eb)) 383 clear_extent_buffer_uptodate(eb); 384 out: 385 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 386 &cached_state); 387 if (need_lock) 388 btrfs_tree_read_unlock_blocking(eb); 389 return ret; 390 } 391 392 /* 393 * Return 0 if the superblock checksum type matches the checksum value of that 394 * algorithm. Pass the raw disk superblock data. 395 */ 396 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, 397 char *raw_disk_sb) 398 { 399 struct btrfs_super_block *disk_sb = 400 (struct btrfs_super_block *)raw_disk_sb; 401 u16 csum_type = btrfs_super_csum_type(disk_sb); 402 int ret = 0; 403 404 if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 405 u32 crc = ~(u32)0; 406 const int csum_size = sizeof(crc); 407 char result[csum_size]; 408 409 /* 410 * The super_block structure does not span the whole 411 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 412 * is filled with zeros and is included in the checksum. 413 */ 414 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 415 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 416 btrfs_csum_final(crc, result); 417 418 if (memcmp(raw_disk_sb, result, csum_size)) 419 ret = 1; 420 } 421 422 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 423 btrfs_err(fs_info, "unsupported checksum algorithm %u", 424 csum_type); 425 ret = 1; 426 } 427 428 return ret; 429 } 430 431 /* 432 * helper to read a given tree block, doing retries as required when 433 * the checksums don't match and we have alternate mirrors to try. 434 */ 435 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info, 436 struct extent_buffer *eb, 437 u64 parent_transid) 438 { 439 struct extent_io_tree *io_tree; 440 int failed = 0; 441 int ret; 442 int num_copies = 0; 443 int mirror_num = 0; 444 int failed_mirror = 0; 445 446 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 447 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; 448 while (1) { 449 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE, 450 mirror_num); 451 if (!ret) { 452 if (!verify_parent_transid(io_tree, eb, 453 parent_transid, 0)) 454 break; 455 else 456 ret = -EIO; 457 } 458 459 /* 460 * This buffer's crc is fine, but its contents are corrupted, so 461 * there is no reason to read the other copies, they won't be 462 * any less wrong. 463 */ 464 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 465 break; 466 467 num_copies = btrfs_num_copies(fs_info, 468 eb->start, eb->len); 469 if (num_copies == 1) 470 break; 471 472 if (!failed_mirror) { 473 failed = 1; 474 failed_mirror = eb->read_mirror; 475 } 476 477 mirror_num++; 478 if (mirror_num == failed_mirror) 479 mirror_num++; 480 481 if (mirror_num > num_copies) 482 break; 483 } 484 485 if (failed && !ret && failed_mirror) 486 repair_eb_io_failure(fs_info, eb, failed_mirror); 487 488 return ret; 489 } 490 491 /* 492 * checksum a dirty tree block before IO. This has extra checks to make sure 493 * we only fill in the checksum field in the first page of a multi-page block 494 */ 495 496 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) 497 { 498 u64 start = page_offset(page); 499 u64 found_start; 500 struct extent_buffer *eb; 501 502 eb = (struct extent_buffer *)page->private; 503 if (page != eb->pages[0]) 504 return 0; 505 506 found_start = btrfs_header_bytenr(eb); 507 /* 508 * Please do not consolidate these warnings into a single if. 509 * It is useful to know what went wrong. 510 */ 511 if (WARN_ON(found_start != start)) 512 return -EUCLEAN; 513 if (WARN_ON(!PageUptodate(page))) 514 return -EUCLEAN; 515 516 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid, 517 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0); 518 519 return csum_tree_block(fs_info, eb, 0); 520 } 521 522 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info, 523 struct extent_buffer *eb) 524 { 525 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 526 u8 fsid[BTRFS_FSID_SIZE]; 527 int ret = 1; 528 529 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 530 while (fs_devices) { 531 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 532 ret = 0; 533 break; 534 } 535 fs_devices = fs_devices->seed; 536 } 537 return ret; 538 } 539 540 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 541 u64 phy_offset, struct page *page, 542 u64 start, u64 end, int mirror) 543 { 544 u64 found_start; 545 int found_level; 546 struct extent_buffer *eb; 547 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 548 struct btrfs_fs_info *fs_info = root->fs_info; 549 int ret = 0; 550 int reads_done; 551 552 if (!page->private) 553 goto out; 554 555 eb = (struct extent_buffer *)page->private; 556 557 /* the pending IO might have been the only thing that kept this buffer 558 * in memory. Make sure we have a ref for all this other checks 559 */ 560 extent_buffer_get(eb); 561 562 reads_done = atomic_dec_and_test(&eb->io_pages); 563 if (!reads_done) 564 goto err; 565 566 eb->read_mirror = mirror; 567 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { 568 ret = -EIO; 569 goto err; 570 } 571 572 found_start = btrfs_header_bytenr(eb); 573 if (found_start != eb->start) { 574 btrfs_err_rl(fs_info, "bad tree block start %llu %llu", 575 found_start, eb->start); 576 ret = -EIO; 577 goto err; 578 } 579 if (check_tree_block_fsid(fs_info, eb)) { 580 btrfs_err_rl(fs_info, "bad fsid on block %llu", 581 eb->start); 582 ret = -EIO; 583 goto err; 584 } 585 found_level = btrfs_header_level(eb); 586 if (found_level >= BTRFS_MAX_LEVEL) { 587 btrfs_err(fs_info, "bad tree block level %d", 588 (int)btrfs_header_level(eb)); 589 ret = -EIO; 590 goto err; 591 } 592 593 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 594 eb, found_level); 595 596 ret = csum_tree_block(fs_info, eb, 1); 597 if (ret) 598 goto err; 599 600 /* 601 * If this is a leaf block and it is corrupt, set the corrupt bit so 602 * that we don't try and read the other copies of this block, just 603 * return -EIO. 604 */ 605 if (found_level == 0 && btrfs_check_leaf_full(root, eb)) { 606 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 607 ret = -EIO; 608 } 609 610 if (found_level > 0 && btrfs_check_node(root, eb)) 611 ret = -EIO; 612 613 if (!ret) 614 set_extent_buffer_uptodate(eb); 615 err: 616 if (reads_done && 617 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 618 btree_readahead_hook(eb, ret); 619 620 if (ret) { 621 /* 622 * our io error hook is going to dec the io pages 623 * again, we have to make sure it has something 624 * to decrement 625 */ 626 atomic_inc(&eb->io_pages); 627 clear_extent_buffer_uptodate(eb); 628 } 629 free_extent_buffer(eb); 630 out: 631 return ret; 632 } 633 634 static int btree_io_failed_hook(struct page *page, int failed_mirror) 635 { 636 struct extent_buffer *eb; 637 638 eb = (struct extent_buffer *)page->private; 639 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); 640 eb->read_mirror = failed_mirror; 641 atomic_dec(&eb->io_pages); 642 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 643 btree_readahead_hook(eb, -EIO); 644 return -EIO; /* we fixed nothing */ 645 } 646 647 static void end_workqueue_bio(struct bio *bio) 648 { 649 struct btrfs_end_io_wq *end_io_wq = bio->bi_private; 650 struct btrfs_fs_info *fs_info; 651 struct btrfs_workqueue *wq; 652 btrfs_work_func_t func; 653 654 fs_info = end_io_wq->info; 655 end_io_wq->status = bio->bi_status; 656 657 if (bio_op(bio) == REQ_OP_WRITE) { 658 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) { 659 wq = fs_info->endio_meta_write_workers; 660 func = btrfs_endio_meta_write_helper; 661 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) { 662 wq = fs_info->endio_freespace_worker; 663 func = btrfs_freespace_write_helper; 664 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 665 wq = fs_info->endio_raid56_workers; 666 func = btrfs_endio_raid56_helper; 667 } else { 668 wq = fs_info->endio_write_workers; 669 func = btrfs_endio_write_helper; 670 } 671 } else { 672 if (unlikely(end_io_wq->metadata == 673 BTRFS_WQ_ENDIO_DIO_REPAIR)) { 674 wq = fs_info->endio_repair_workers; 675 func = btrfs_endio_repair_helper; 676 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) { 677 wq = fs_info->endio_raid56_workers; 678 func = btrfs_endio_raid56_helper; 679 } else if (end_io_wq->metadata) { 680 wq = fs_info->endio_meta_workers; 681 func = btrfs_endio_meta_helper; 682 } else { 683 wq = fs_info->endio_workers; 684 func = btrfs_endio_helper; 685 } 686 } 687 688 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL); 689 btrfs_queue_work(wq, &end_io_wq->work); 690 } 691 692 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 693 enum btrfs_wq_endio_type metadata) 694 { 695 struct btrfs_end_io_wq *end_io_wq; 696 697 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); 698 if (!end_io_wq) 699 return BLK_STS_RESOURCE; 700 701 end_io_wq->private = bio->bi_private; 702 end_io_wq->end_io = bio->bi_end_io; 703 end_io_wq->info = info; 704 end_io_wq->status = 0; 705 end_io_wq->bio = bio; 706 end_io_wq->metadata = metadata; 707 708 bio->bi_private = end_io_wq; 709 bio->bi_end_io = end_workqueue_bio; 710 return 0; 711 } 712 713 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 714 { 715 unsigned long limit = min_t(unsigned long, 716 info->thread_pool_size, 717 info->fs_devices->open_devices); 718 return 256 * limit; 719 } 720 721 static void run_one_async_start(struct btrfs_work *work) 722 { 723 struct async_submit_bio *async; 724 blk_status_t ret; 725 726 async = container_of(work, struct async_submit_bio, work); 727 ret = async->submit_bio_start(async->private_data, async->bio, 728 async->mirror_num, async->bio_flags, 729 async->bio_offset); 730 if (ret) 731 async->status = ret; 732 } 733 734 static void run_one_async_done(struct btrfs_work *work) 735 { 736 struct async_submit_bio *async; 737 738 async = container_of(work, struct async_submit_bio, work); 739 740 /* If an error occurred we just want to clean up the bio and move on */ 741 if (async->status) { 742 async->bio->bi_status = async->status; 743 bio_endio(async->bio); 744 return; 745 } 746 747 async->submit_bio_done(async->private_data, async->bio, async->mirror_num, 748 async->bio_flags, async->bio_offset); 749 } 750 751 static void run_one_async_free(struct btrfs_work *work) 752 { 753 struct async_submit_bio *async; 754 755 async = container_of(work, struct async_submit_bio, work); 756 kfree(async); 757 } 758 759 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 760 int mirror_num, unsigned long bio_flags, 761 u64 bio_offset, void *private_data, 762 extent_submit_bio_hook_t *submit_bio_start, 763 extent_submit_bio_hook_t *submit_bio_done) 764 { 765 struct async_submit_bio *async; 766 767 async = kmalloc(sizeof(*async), GFP_NOFS); 768 if (!async) 769 return BLK_STS_RESOURCE; 770 771 async->private_data = private_data; 772 async->fs_info = fs_info; 773 async->bio = bio; 774 async->mirror_num = mirror_num; 775 async->submit_bio_start = submit_bio_start; 776 async->submit_bio_done = submit_bio_done; 777 778 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start, 779 run_one_async_done, run_one_async_free); 780 781 async->bio_flags = bio_flags; 782 async->bio_offset = bio_offset; 783 784 async->status = 0; 785 786 if (op_is_sync(bio->bi_opf)) 787 btrfs_set_work_high_priority(&async->work); 788 789 btrfs_queue_work(fs_info->workers, &async->work); 790 return 0; 791 } 792 793 static blk_status_t btree_csum_one_bio(struct bio *bio) 794 { 795 struct bio_vec *bvec; 796 struct btrfs_root *root; 797 int i, ret = 0; 798 799 ASSERT(!bio_flagged(bio, BIO_CLONED)); 800 bio_for_each_segment_all(bvec, bio, i) { 801 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 802 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); 803 if (ret) 804 break; 805 } 806 807 return errno_to_blk_status(ret); 808 } 809 810 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio, 811 int mirror_num, unsigned long bio_flags, 812 u64 bio_offset) 813 { 814 /* 815 * when we're called for a write, we're already in the async 816 * submission context. Just jump into btrfs_map_bio 817 */ 818 return btree_csum_one_bio(bio); 819 } 820 821 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio, 822 int mirror_num, unsigned long bio_flags, 823 u64 bio_offset) 824 { 825 struct inode *inode = private_data; 826 blk_status_t ret; 827 828 /* 829 * when we're called for a write, we're already in the async 830 * submission context. Just jump into btrfs_map_bio 831 */ 832 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1); 833 if (ret) { 834 bio->bi_status = ret; 835 bio_endio(bio); 836 } 837 return ret; 838 } 839 840 static int check_async_write(struct btrfs_inode *bi) 841 { 842 if (atomic_read(&bi->sync_writers)) 843 return 0; 844 #ifdef CONFIG_X86 845 if (static_cpu_has(X86_FEATURE_XMM4_2)) 846 return 0; 847 #endif 848 return 1; 849 } 850 851 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio, 852 int mirror_num, unsigned long bio_flags, 853 u64 bio_offset) 854 { 855 struct inode *inode = private_data; 856 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 857 int async = check_async_write(BTRFS_I(inode)); 858 blk_status_t ret; 859 860 if (bio_op(bio) != REQ_OP_WRITE) { 861 /* 862 * called for a read, do the setup so that checksum validation 863 * can happen in the async kernel threads 864 */ 865 ret = btrfs_bio_wq_end_io(fs_info, bio, 866 BTRFS_WQ_ENDIO_METADATA); 867 if (ret) 868 goto out_w_error; 869 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 870 } else if (!async) { 871 ret = btree_csum_one_bio(bio); 872 if (ret) 873 goto out_w_error; 874 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); 875 } else { 876 /* 877 * kthread helpers are used to submit writes so that 878 * checksumming can happen in parallel across all CPUs 879 */ 880 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0, 881 bio_offset, private_data, 882 __btree_submit_bio_start, 883 __btree_submit_bio_done); 884 } 885 886 if (ret) 887 goto out_w_error; 888 return 0; 889 890 out_w_error: 891 bio->bi_status = ret; 892 bio_endio(bio); 893 return ret; 894 } 895 896 #ifdef CONFIG_MIGRATION 897 static int btree_migratepage(struct address_space *mapping, 898 struct page *newpage, struct page *page, 899 enum migrate_mode mode) 900 { 901 /* 902 * we can't safely write a btree page from here, 903 * we haven't done the locking hook 904 */ 905 if (PageDirty(page)) 906 return -EAGAIN; 907 /* 908 * Buffers may be managed in a filesystem specific way. 909 * We must have no buffers or drop them. 910 */ 911 if (page_has_private(page) && 912 !try_to_release_page(page, GFP_KERNEL)) 913 return -EAGAIN; 914 return migrate_page(mapping, newpage, page, mode); 915 } 916 #endif 917 918 919 static int btree_writepages(struct address_space *mapping, 920 struct writeback_control *wbc) 921 { 922 struct btrfs_fs_info *fs_info; 923 int ret; 924 925 if (wbc->sync_mode == WB_SYNC_NONE) { 926 927 if (wbc->for_kupdate) 928 return 0; 929 930 fs_info = BTRFS_I(mapping->host)->root->fs_info; 931 /* this is a bit racy, but that's ok */ 932 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 933 BTRFS_DIRTY_METADATA_THRESH); 934 if (ret < 0) 935 return 0; 936 } 937 return btree_write_cache_pages(mapping, wbc); 938 } 939 940 static int btree_readpage(struct file *file, struct page *page) 941 { 942 struct extent_io_tree *tree; 943 tree = &BTRFS_I(page->mapping->host)->io_tree; 944 return extent_read_full_page(tree, page, btree_get_extent, 0); 945 } 946 947 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 948 { 949 if (PageWriteback(page) || PageDirty(page)) 950 return 0; 951 952 return try_release_extent_buffer(page); 953 } 954 955 static void btree_invalidatepage(struct page *page, unsigned int offset, 956 unsigned int length) 957 { 958 struct extent_io_tree *tree; 959 tree = &BTRFS_I(page->mapping->host)->io_tree; 960 extent_invalidatepage(tree, page, offset); 961 btree_releasepage(page, GFP_NOFS); 962 if (PagePrivate(page)) { 963 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 964 "page private not zero on page %llu", 965 (unsigned long long)page_offset(page)); 966 ClearPagePrivate(page); 967 set_page_private(page, 0); 968 put_page(page); 969 } 970 } 971 972 static int btree_set_page_dirty(struct page *page) 973 { 974 #ifdef DEBUG 975 struct extent_buffer *eb; 976 977 BUG_ON(!PagePrivate(page)); 978 eb = (struct extent_buffer *)page->private; 979 BUG_ON(!eb); 980 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 981 BUG_ON(!atomic_read(&eb->refs)); 982 btrfs_assert_tree_locked(eb); 983 #endif 984 return __set_page_dirty_nobuffers(page); 985 } 986 987 static const struct address_space_operations btree_aops = { 988 .readpage = btree_readpage, 989 .writepages = btree_writepages, 990 .releasepage = btree_releasepage, 991 .invalidatepage = btree_invalidatepage, 992 #ifdef CONFIG_MIGRATION 993 .migratepage = btree_migratepage, 994 #endif 995 .set_page_dirty = btree_set_page_dirty, 996 }; 997 998 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) 999 { 1000 struct extent_buffer *buf = NULL; 1001 struct inode *btree_inode = fs_info->btree_inode; 1002 1003 buf = btrfs_find_create_tree_block(fs_info, bytenr); 1004 if (IS_ERR(buf)) 1005 return; 1006 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1007 buf, WAIT_NONE, 0); 1008 free_extent_buffer(buf); 1009 } 1010 1011 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr, 1012 int mirror_num, struct extent_buffer **eb) 1013 { 1014 struct extent_buffer *buf = NULL; 1015 struct inode *btree_inode = fs_info->btree_inode; 1016 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1017 int ret; 1018 1019 buf = btrfs_find_create_tree_block(fs_info, bytenr); 1020 if (IS_ERR(buf)) 1021 return 0; 1022 1023 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1024 1025 ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK, 1026 mirror_num); 1027 if (ret) { 1028 free_extent_buffer(buf); 1029 return ret; 1030 } 1031 1032 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1033 free_extent_buffer(buf); 1034 return -EIO; 1035 } else if (extent_buffer_uptodate(buf)) { 1036 *eb = buf; 1037 } else { 1038 free_extent_buffer(buf); 1039 } 1040 return 0; 1041 } 1042 1043 struct extent_buffer *btrfs_find_create_tree_block( 1044 struct btrfs_fs_info *fs_info, 1045 u64 bytenr) 1046 { 1047 if (btrfs_is_testing(fs_info)) 1048 return alloc_test_extent_buffer(fs_info, bytenr); 1049 return alloc_extent_buffer(fs_info, bytenr); 1050 } 1051 1052 1053 int btrfs_write_tree_block(struct extent_buffer *buf) 1054 { 1055 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 1056 buf->start + buf->len - 1); 1057 } 1058 1059 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1060 { 1061 filemap_fdatawait_range(buf->pages[0]->mapping, 1062 buf->start, buf->start + buf->len - 1); 1063 } 1064 1065 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, 1066 u64 parent_transid) 1067 { 1068 struct extent_buffer *buf = NULL; 1069 int ret; 1070 1071 buf = btrfs_find_create_tree_block(fs_info, bytenr); 1072 if (IS_ERR(buf)) 1073 return buf; 1074 1075 ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid); 1076 if (ret) { 1077 free_extent_buffer(buf); 1078 return ERR_PTR(ret); 1079 } 1080 return buf; 1081 1082 } 1083 1084 void clean_tree_block(struct btrfs_fs_info *fs_info, 1085 struct extent_buffer *buf) 1086 { 1087 if (btrfs_header_generation(buf) == 1088 fs_info->running_transaction->transid) { 1089 btrfs_assert_tree_locked(buf); 1090 1091 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1092 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 1093 -buf->len, 1094 fs_info->dirty_metadata_batch); 1095 /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1096 btrfs_set_lock_blocking(buf); 1097 clear_extent_buffer_dirty(buf); 1098 } 1099 } 1100 } 1101 1102 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 1103 { 1104 struct btrfs_subvolume_writers *writers; 1105 int ret; 1106 1107 writers = kmalloc(sizeof(*writers), GFP_NOFS); 1108 if (!writers) 1109 return ERR_PTR(-ENOMEM); 1110 1111 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL); 1112 if (ret < 0) { 1113 kfree(writers); 1114 return ERR_PTR(ret); 1115 } 1116 1117 init_waitqueue_head(&writers->wait); 1118 return writers; 1119 } 1120 1121 static void 1122 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 1123 { 1124 percpu_counter_destroy(&writers->counter); 1125 kfree(writers); 1126 } 1127 1128 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, 1129 u64 objectid) 1130 { 1131 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); 1132 root->node = NULL; 1133 root->commit_root = NULL; 1134 root->state = 0; 1135 root->orphan_cleanup_state = 0; 1136 1137 root->objectid = objectid; 1138 root->last_trans = 0; 1139 root->highest_objectid = 0; 1140 root->nr_delalloc_inodes = 0; 1141 root->nr_ordered_extents = 0; 1142 root->name = NULL; 1143 root->inode_tree = RB_ROOT; 1144 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1145 root->block_rsv = NULL; 1146 root->orphan_block_rsv = NULL; 1147 1148 INIT_LIST_HEAD(&root->dirty_list); 1149 INIT_LIST_HEAD(&root->root_list); 1150 INIT_LIST_HEAD(&root->delalloc_inodes); 1151 INIT_LIST_HEAD(&root->delalloc_root); 1152 INIT_LIST_HEAD(&root->ordered_extents); 1153 INIT_LIST_HEAD(&root->ordered_root); 1154 INIT_LIST_HEAD(&root->logged_list[0]); 1155 INIT_LIST_HEAD(&root->logged_list[1]); 1156 spin_lock_init(&root->orphan_lock); 1157 spin_lock_init(&root->inode_lock); 1158 spin_lock_init(&root->delalloc_lock); 1159 spin_lock_init(&root->ordered_extent_lock); 1160 spin_lock_init(&root->accounting_lock); 1161 spin_lock_init(&root->log_extents_lock[0]); 1162 spin_lock_init(&root->log_extents_lock[1]); 1163 mutex_init(&root->objectid_mutex); 1164 mutex_init(&root->log_mutex); 1165 mutex_init(&root->ordered_extent_mutex); 1166 mutex_init(&root->delalloc_mutex); 1167 init_waitqueue_head(&root->log_writer_wait); 1168 init_waitqueue_head(&root->log_commit_wait[0]); 1169 init_waitqueue_head(&root->log_commit_wait[1]); 1170 INIT_LIST_HEAD(&root->log_ctxs[0]); 1171 INIT_LIST_HEAD(&root->log_ctxs[1]); 1172 atomic_set(&root->log_commit[0], 0); 1173 atomic_set(&root->log_commit[1], 0); 1174 atomic_set(&root->log_writers, 0); 1175 atomic_set(&root->log_batch, 0); 1176 atomic_set(&root->orphan_inodes, 0); 1177 refcount_set(&root->refs, 1); 1178 atomic_set(&root->will_be_snapshotted, 0); 1179 atomic64_set(&root->qgroup_meta_rsv, 0); 1180 root->log_transid = 0; 1181 root->log_transid_committed = -1; 1182 root->last_log_commit = 0; 1183 if (!dummy) 1184 extent_io_tree_init(&root->dirty_log_pages, NULL); 1185 1186 memset(&root->root_key, 0, sizeof(root->root_key)); 1187 memset(&root->root_item, 0, sizeof(root->root_item)); 1188 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 1189 if (!dummy) 1190 root->defrag_trans_start = fs_info->generation; 1191 else 1192 root->defrag_trans_start = 0; 1193 root->root_key.objectid = objectid; 1194 root->anon_dev = 0; 1195 1196 spin_lock_init(&root->root_item_lock); 1197 } 1198 1199 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, 1200 gfp_t flags) 1201 { 1202 struct btrfs_root *root = kzalloc(sizeof(*root), flags); 1203 if (root) 1204 root->fs_info = fs_info; 1205 return root; 1206 } 1207 1208 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1209 /* Should only be used by the testing infrastructure */ 1210 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) 1211 { 1212 struct btrfs_root *root; 1213 1214 if (!fs_info) 1215 return ERR_PTR(-EINVAL); 1216 1217 root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1218 if (!root) 1219 return ERR_PTR(-ENOMEM); 1220 1221 /* We don't use the stripesize in selftest, set it as sectorsize */ 1222 __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 1223 root->alloc_bytenr = 0; 1224 1225 return root; 1226 } 1227 #endif 1228 1229 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 1230 struct btrfs_fs_info *fs_info, 1231 u64 objectid) 1232 { 1233 struct extent_buffer *leaf; 1234 struct btrfs_root *tree_root = fs_info->tree_root; 1235 struct btrfs_root *root; 1236 struct btrfs_key key; 1237 int ret = 0; 1238 uuid_le uuid = NULL_UUID_LE; 1239 1240 root = btrfs_alloc_root(fs_info, GFP_KERNEL); 1241 if (!root) 1242 return ERR_PTR(-ENOMEM); 1243 1244 __setup_root(root, fs_info, objectid); 1245 root->root_key.objectid = objectid; 1246 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1247 root->root_key.offset = 0; 1248 1249 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 1250 if (IS_ERR(leaf)) { 1251 ret = PTR_ERR(leaf); 1252 leaf = NULL; 1253 goto fail; 1254 } 1255 1256 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header)); 1257 btrfs_set_header_bytenr(leaf, leaf->start); 1258 btrfs_set_header_generation(leaf, trans->transid); 1259 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1260 btrfs_set_header_owner(leaf, objectid); 1261 root->node = leaf; 1262 1263 write_extent_buffer_fsid(leaf, fs_info->fsid); 1264 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); 1265 btrfs_mark_buffer_dirty(leaf); 1266 1267 root->commit_root = btrfs_root_node(root); 1268 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 1269 1270 root->root_item.flags = 0; 1271 root->root_item.byte_limit = 0; 1272 btrfs_set_root_bytenr(&root->root_item, leaf->start); 1273 btrfs_set_root_generation(&root->root_item, trans->transid); 1274 btrfs_set_root_level(&root->root_item, 0); 1275 btrfs_set_root_refs(&root->root_item, 1); 1276 btrfs_set_root_used(&root->root_item, leaf->len); 1277 btrfs_set_root_last_snapshot(&root->root_item, 0); 1278 btrfs_set_root_dirid(&root->root_item, 0); 1279 if (is_fstree(objectid)) 1280 uuid_le_gen(&uuid); 1281 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 1282 root->root_item.drop_level = 0; 1283 1284 key.objectid = objectid; 1285 key.type = BTRFS_ROOT_ITEM_KEY; 1286 key.offset = 0; 1287 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 1288 if (ret) 1289 goto fail; 1290 1291 btrfs_tree_unlock(leaf); 1292 1293 return root; 1294 1295 fail: 1296 if (leaf) { 1297 btrfs_tree_unlock(leaf); 1298 free_extent_buffer(root->commit_root); 1299 free_extent_buffer(leaf); 1300 } 1301 kfree(root); 1302 1303 return ERR_PTR(ret); 1304 } 1305 1306 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1307 struct btrfs_fs_info *fs_info) 1308 { 1309 struct btrfs_root *root; 1310 struct extent_buffer *leaf; 1311 1312 root = btrfs_alloc_root(fs_info, GFP_NOFS); 1313 if (!root) 1314 return ERR_PTR(-ENOMEM); 1315 1316 __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1317 1318 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1319 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1320 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1321 1322 /* 1323 * DON'T set REF_COWS for log trees 1324 * 1325 * log trees do not get reference counted because they go away 1326 * before a real commit is actually done. They do store pointers 1327 * to file data extents, and those reference counts still get 1328 * updated (along with back refs to the log tree). 1329 */ 1330 1331 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, 1332 NULL, 0, 0, 0); 1333 if (IS_ERR(leaf)) { 1334 kfree(root); 1335 return ERR_CAST(leaf); 1336 } 1337 1338 memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header)); 1339 btrfs_set_header_bytenr(leaf, leaf->start); 1340 btrfs_set_header_generation(leaf, trans->transid); 1341 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1342 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 1343 root->node = leaf; 1344 1345 write_extent_buffer_fsid(root->node, fs_info->fsid); 1346 btrfs_mark_buffer_dirty(root->node); 1347 btrfs_tree_unlock(root->node); 1348 return root; 1349 } 1350 1351 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1352 struct btrfs_fs_info *fs_info) 1353 { 1354 struct btrfs_root *log_root; 1355 1356 log_root = alloc_log_tree(trans, fs_info); 1357 if (IS_ERR(log_root)) 1358 return PTR_ERR(log_root); 1359 WARN_ON(fs_info->log_root_tree); 1360 fs_info->log_root_tree = log_root; 1361 return 0; 1362 } 1363 1364 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1365 struct btrfs_root *root) 1366 { 1367 struct btrfs_fs_info *fs_info = root->fs_info; 1368 struct btrfs_root *log_root; 1369 struct btrfs_inode_item *inode_item; 1370 1371 log_root = alloc_log_tree(trans, fs_info); 1372 if (IS_ERR(log_root)) 1373 return PTR_ERR(log_root); 1374 1375 log_root->last_trans = trans->transid; 1376 log_root->root_key.offset = root->root_key.objectid; 1377 1378 inode_item = &log_root->root_item.inode; 1379 btrfs_set_stack_inode_generation(inode_item, 1); 1380 btrfs_set_stack_inode_size(inode_item, 3); 1381 btrfs_set_stack_inode_nlink(inode_item, 1); 1382 btrfs_set_stack_inode_nbytes(inode_item, 1383 fs_info->nodesize); 1384 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1385 1386 btrfs_set_root_node(&log_root->root_item, log_root->node); 1387 1388 WARN_ON(root->log_root); 1389 root->log_root = log_root; 1390 root->log_transid = 0; 1391 root->log_transid_committed = -1; 1392 root->last_log_commit = 0; 1393 return 0; 1394 } 1395 1396 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1397 struct btrfs_key *key) 1398 { 1399 struct btrfs_root *root; 1400 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1401 struct btrfs_path *path; 1402 u64 generation; 1403 int ret; 1404 1405 path = btrfs_alloc_path(); 1406 if (!path) 1407 return ERR_PTR(-ENOMEM); 1408 1409 root = btrfs_alloc_root(fs_info, GFP_NOFS); 1410 if (!root) { 1411 ret = -ENOMEM; 1412 goto alloc_fail; 1413 } 1414 1415 __setup_root(root, fs_info, key->objectid); 1416 1417 ret = btrfs_find_root(tree_root, key, path, 1418 &root->root_item, &root->root_key); 1419 if (ret) { 1420 if (ret > 0) 1421 ret = -ENOENT; 1422 goto find_fail; 1423 } 1424 1425 generation = btrfs_root_generation(&root->root_item); 1426 root->node = read_tree_block(fs_info, 1427 btrfs_root_bytenr(&root->root_item), 1428 generation); 1429 if (IS_ERR(root->node)) { 1430 ret = PTR_ERR(root->node); 1431 goto find_fail; 1432 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1433 ret = -EIO; 1434 free_extent_buffer(root->node); 1435 goto find_fail; 1436 } 1437 root->commit_root = btrfs_root_node(root); 1438 out: 1439 btrfs_free_path(path); 1440 return root; 1441 1442 find_fail: 1443 kfree(root); 1444 alloc_fail: 1445 root = ERR_PTR(ret); 1446 goto out; 1447 } 1448 1449 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1450 struct btrfs_key *location) 1451 { 1452 struct btrfs_root *root; 1453 1454 root = btrfs_read_tree_root(tree_root, location); 1455 if (IS_ERR(root)) 1456 return root; 1457 1458 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 1459 set_bit(BTRFS_ROOT_REF_COWS, &root->state); 1460 btrfs_check_and_init_root_item(&root->root_item); 1461 } 1462 1463 return root; 1464 } 1465 1466 int btrfs_init_fs_root(struct btrfs_root *root) 1467 { 1468 int ret; 1469 struct btrfs_subvolume_writers *writers; 1470 1471 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1472 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1473 GFP_NOFS); 1474 if (!root->free_ino_pinned || !root->free_ino_ctl) { 1475 ret = -ENOMEM; 1476 goto fail; 1477 } 1478 1479 writers = btrfs_alloc_subvolume_writers(); 1480 if (IS_ERR(writers)) { 1481 ret = PTR_ERR(writers); 1482 goto fail; 1483 } 1484 root->subv_writers = writers; 1485 1486 btrfs_init_free_ino_ctl(root); 1487 spin_lock_init(&root->ino_cache_lock); 1488 init_waitqueue_head(&root->ino_cache_wait); 1489 1490 ret = get_anon_bdev(&root->anon_dev); 1491 if (ret) 1492 goto fail; 1493 1494 mutex_lock(&root->objectid_mutex); 1495 ret = btrfs_find_highest_objectid(root, 1496 &root->highest_objectid); 1497 if (ret) { 1498 mutex_unlock(&root->objectid_mutex); 1499 goto fail; 1500 } 1501 1502 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 1503 1504 mutex_unlock(&root->objectid_mutex); 1505 1506 return 0; 1507 fail: 1508 /* the caller is responsible to call free_fs_root */ 1509 return ret; 1510 } 1511 1512 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1513 u64 root_id) 1514 { 1515 struct btrfs_root *root; 1516 1517 spin_lock(&fs_info->fs_roots_radix_lock); 1518 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1519 (unsigned long)root_id); 1520 spin_unlock(&fs_info->fs_roots_radix_lock); 1521 return root; 1522 } 1523 1524 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1525 struct btrfs_root *root) 1526 { 1527 int ret; 1528 1529 ret = radix_tree_preload(GFP_NOFS); 1530 if (ret) 1531 return ret; 1532 1533 spin_lock(&fs_info->fs_roots_radix_lock); 1534 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1535 (unsigned long)root->root_key.objectid, 1536 root); 1537 if (ret == 0) 1538 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1539 spin_unlock(&fs_info->fs_roots_radix_lock); 1540 radix_tree_preload_end(); 1541 1542 return ret; 1543 } 1544 1545 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1546 struct btrfs_key *location, 1547 bool check_ref) 1548 { 1549 struct btrfs_root *root; 1550 struct btrfs_path *path; 1551 struct btrfs_key key; 1552 int ret; 1553 1554 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1555 return fs_info->tree_root; 1556 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1557 return fs_info->extent_root; 1558 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 1559 return fs_info->chunk_root; 1560 if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 1561 return fs_info->dev_root; 1562 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 1563 return fs_info->csum_root; 1564 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1565 return fs_info->quota_root ? fs_info->quota_root : 1566 ERR_PTR(-ENOENT); 1567 if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1568 return fs_info->uuid_root ? fs_info->uuid_root : 1569 ERR_PTR(-ENOENT); 1570 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 1571 return fs_info->free_space_root ? fs_info->free_space_root : 1572 ERR_PTR(-ENOENT); 1573 again: 1574 root = btrfs_lookup_fs_root(fs_info, location->objectid); 1575 if (root) { 1576 if (check_ref && btrfs_root_refs(&root->root_item) == 0) 1577 return ERR_PTR(-ENOENT); 1578 return root; 1579 } 1580 1581 root = btrfs_read_fs_root(fs_info->tree_root, location); 1582 if (IS_ERR(root)) 1583 return root; 1584 1585 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1586 ret = -ENOENT; 1587 goto fail; 1588 } 1589 1590 ret = btrfs_init_fs_root(root); 1591 if (ret) 1592 goto fail; 1593 1594 path = btrfs_alloc_path(); 1595 if (!path) { 1596 ret = -ENOMEM; 1597 goto fail; 1598 } 1599 key.objectid = BTRFS_ORPHAN_OBJECTID; 1600 key.type = BTRFS_ORPHAN_ITEM_KEY; 1601 key.offset = location->objectid; 1602 1603 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 1604 btrfs_free_path(path); 1605 if (ret < 0) 1606 goto fail; 1607 if (ret == 0) 1608 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1609 1610 ret = btrfs_insert_fs_root(fs_info, root); 1611 if (ret) { 1612 if (ret == -EEXIST) { 1613 free_fs_root(root); 1614 goto again; 1615 } 1616 goto fail; 1617 } 1618 return root; 1619 fail: 1620 free_fs_root(root); 1621 return ERR_PTR(ret); 1622 } 1623 1624 static int btrfs_congested_fn(void *congested_data, int bdi_bits) 1625 { 1626 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 1627 int ret = 0; 1628 struct btrfs_device *device; 1629 struct backing_dev_info *bdi; 1630 1631 rcu_read_lock(); 1632 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1633 if (!device->bdev) 1634 continue; 1635 bdi = device->bdev->bd_bdi; 1636 if (bdi_congested(bdi, bdi_bits)) { 1637 ret = 1; 1638 break; 1639 } 1640 } 1641 rcu_read_unlock(); 1642 return ret; 1643 } 1644 1645 /* 1646 * called by the kthread helper functions to finally call the bio end_io 1647 * functions. This is where read checksum verification actually happens 1648 */ 1649 static void end_workqueue_fn(struct btrfs_work *work) 1650 { 1651 struct bio *bio; 1652 struct btrfs_end_io_wq *end_io_wq; 1653 1654 end_io_wq = container_of(work, struct btrfs_end_io_wq, work); 1655 bio = end_io_wq->bio; 1656 1657 bio->bi_status = end_io_wq->status; 1658 bio->bi_private = end_io_wq->private; 1659 bio->bi_end_io = end_io_wq->end_io; 1660 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); 1661 bio_endio(bio); 1662 } 1663 1664 static int cleaner_kthread(void *arg) 1665 { 1666 struct btrfs_root *root = arg; 1667 struct btrfs_fs_info *fs_info = root->fs_info; 1668 int again; 1669 struct btrfs_trans_handle *trans; 1670 1671 do { 1672 again = 0; 1673 1674 /* Make the cleaner go to sleep early. */ 1675 if (btrfs_need_cleaner_sleep(fs_info)) 1676 goto sleep; 1677 1678 /* 1679 * Do not do anything if we might cause open_ctree() to block 1680 * before we have finished mounting the filesystem. 1681 */ 1682 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) 1683 goto sleep; 1684 1685 if (!mutex_trylock(&fs_info->cleaner_mutex)) 1686 goto sleep; 1687 1688 /* 1689 * Avoid the problem that we change the status of the fs 1690 * during the above check and trylock. 1691 */ 1692 if (btrfs_need_cleaner_sleep(fs_info)) { 1693 mutex_unlock(&fs_info->cleaner_mutex); 1694 goto sleep; 1695 } 1696 1697 mutex_lock(&fs_info->cleaner_delayed_iput_mutex); 1698 btrfs_run_delayed_iputs(fs_info); 1699 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); 1700 1701 again = btrfs_clean_one_deleted_snapshot(root); 1702 mutex_unlock(&fs_info->cleaner_mutex); 1703 1704 /* 1705 * The defragger has dealt with the R/O remount and umount, 1706 * needn't do anything special here. 1707 */ 1708 btrfs_run_defrag_inodes(fs_info); 1709 1710 /* 1711 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing 1712 * with relocation (btrfs_relocate_chunk) and relocation 1713 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) 1714 * after acquiring fs_info->delete_unused_bgs_mutex. So we 1715 * can't hold, nor need to, fs_info->cleaner_mutex when deleting 1716 * unused block groups. 1717 */ 1718 btrfs_delete_unused_bgs(fs_info); 1719 sleep: 1720 if (!again) { 1721 set_current_state(TASK_INTERRUPTIBLE); 1722 if (!kthread_should_stop()) 1723 schedule(); 1724 __set_current_state(TASK_RUNNING); 1725 } 1726 } while (!kthread_should_stop()); 1727 1728 /* 1729 * Transaction kthread is stopped before us and wakes us up. 1730 * However we might have started a new transaction and COWed some 1731 * tree blocks when deleting unused block groups for example. So 1732 * make sure we commit the transaction we started to have a clean 1733 * shutdown when evicting the btree inode - if it has dirty pages 1734 * when we do the final iput() on it, eviction will trigger a 1735 * writeback for it which will fail with null pointer dereferences 1736 * since work queues and other resources were already released and 1737 * destroyed by the time the iput/eviction/writeback is made. 1738 */ 1739 trans = btrfs_attach_transaction(root); 1740 if (IS_ERR(trans)) { 1741 if (PTR_ERR(trans) != -ENOENT) 1742 btrfs_err(fs_info, 1743 "cleaner transaction attach returned %ld", 1744 PTR_ERR(trans)); 1745 } else { 1746 int ret; 1747 1748 ret = btrfs_commit_transaction(trans); 1749 if (ret) 1750 btrfs_err(fs_info, 1751 "cleaner open transaction commit returned %d", 1752 ret); 1753 } 1754 1755 return 0; 1756 } 1757 1758 static int transaction_kthread(void *arg) 1759 { 1760 struct btrfs_root *root = arg; 1761 struct btrfs_fs_info *fs_info = root->fs_info; 1762 struct btrfs_trans_handle *trans; 1763 struct btrfs_transaction *cur; 1764 u64 transid; 1765 unsigned long now; 1766 unsigned long delay; 1767 bool cannot_commit; 1768 1769 do { 1770 cannot_commit = false; 1771 delay = HZ * fs_info->commit_interval; 1772 mutex_lock(&fs_info->transaction_kthread_mutex); 1773 1774 spin_lock(&fs_info->trans_lock); 1775 cur = fs_info->running_transaction; 1776 if (!cur) { 1777 spin_unlock(&fs_info->trans_lock); 1778 goto sleep; 1779 } 1780 1781 now = get_seconds(); 1782 if (cur->state < TRANS_STATE_BLOCKED && 1783 (now < cur->start_time || 1784 now - cur->start_time < fs_info->commit_interval)) { 1785 spin_unlock(&fs_info->trans_lock); 1786 delay = HZ * 5; 1787 goto sleep; 1788 } 1789 transid = cur->transid; 1790 spin_unlock(&fs_info->trans_lock); 1791 1792 /* If the file system is aborted, this will always fail. */ 1793 trans = btrfs_attach_transaction(root); 1794 if (IS_ERR(trans)) { 1795 if (PTR_ERR(trans) != -ENOENT) 1796 cannot_commit = true; 1797 goto sleep; 1798 } 1799 if (transid == trans->transid) { 1800 btrfs_commit_transaction(trans); 1801 } else { 1802 btrfs_end_transaction(trans); 1803 } 1804 sleep: 1805 wake_up_process(fs_info->cleaner_kthread); 1806 mutex_unlock(&fs_info->transaction_kthread_mutex); 1807 1808 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 1809 &fs_info->fs_state))) 1810 btrfs_cleanup_transaction(fs_info); 1811 set_current_state(TASK_INTERRUPTIBLE); 1812 if (!kthread_should_stop() && 1813 (!btrfs_transaction_blocked(fs_info) || 1814 cannot_commit)) 1815 schedule_timeout(delay); 1816 __set_current_state(TASK_RUNNING); 1817 } while (!kthread_should_stop()); 1818 return 0; 1819 } 1820 1821 /* 1822 * this will find the highest generation in the array of 1823 * root backups. The index of the highest array is returned, 1824 * or -1 if we can't find anything. 1825 * 1826 * We check to make sure the array is valid by comparing the 1827 * generation of the latest root in the array with the generation 1828 * in the super block. If they don't match we pitch it. 1829 */ 1830 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1831 { 1832 u64 cur; 1833 int newest_index = -1; 1834 struct btrfs_root_backup *root_backup; 1835 int i; 1836 1837 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1838 root_backup = info->super_copy->super_roots + i; 1839 cur = btrfs_backup_tree_root_gen(root_backup); 1840 if (cur == newest_gen) 1841 newest_index = i; 1842 } 1843 1844 /* check to see if we actually wrapped around */ 1845 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1846 root_backup = info->super_copy->super_roots; 1847 cur = btrfs_backup_tree_root_gen(root_backup); 1848 if (cur == newest_gen) 1849 newest_index = 0; 1850 } 1851 return newest_index; 1852 } 1853 1854 1855 /* 1856 * find the oldest backup so we know where to store new entries 1857 * in the backup array. This will set the backup_root_index 1858 * field in the fs_info struct 1859 */ 1860 static void find_oldest_super_backup(struct btrfs_fs_info *info, 1861 u64 newest_gen) 1862 { 1863 int newest_index = -1; 1864 1865 newest_index = find_newest_super_backup(info, newest_gen); 1866 /* if there was garbage in there, just move along */ 1867 if (newest_index == -1) { 1868 info->backup_root_index = 0; 1869 } else { 1870 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1871 } 1872 } 1873 1874 /* 1875 * copy all the root pointers into the super backup array. 1876 * this will bump the backup pointer by one when it is 1877 * done 1878 */ 1879 static void backup_super_roots(struct btrfs_fs_info *info) 1880 { 1881 int next_backup; 1882 struct btrfs_root_backup *root_backup; 1883 int last_backup; 1884 1885 next_backup = info->backup_root_index; 1886 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1887 BTRFS_NUM_BACKUP_ROOTS; 1888 1889 /* 1890 * just overwrite the last backup if we're at the same generation 1891 * this happens only at umount 1892 */ 1893 root_backup = info->super_for_commit->super_roots + last_backup; 1894 if (btrfs_backup_tree_root_gen(root_backup) == 1895 btrfs_header_generation(info->tree_root->node)) 1896 next_backup = last_backup; 1897 1898 root_backup = info->super_for_commit->super_roots + next_backup; 1899 1900 /* 1901 * make sure all of our padding and empty slots get zero filled 1902 * regardless of which ones we use today 1903 */ 1904 memset(root_backup, 0, sizeof(*root_backup)); 1905 1906 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1907 1908 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1909 btrfs_set_backup_tree_root_gen(root_backup, 1910 btrfs_header_generation(info->tree_root->node)); 1911 1912 btrfs_set_backup_tree_root_level(root_backup, 1913 btrfs_header_level(info->tree_root->node)); 1914 1915 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1916 btrfs_set_backup_chunk_root_gen(root_backup, 1917 btrfs_header_generation(info->chunk_root->node)); 1918 btrfs_set_backup_chunk_root_level(root_backup, 1919 btrfs_header_level(info->chunk_root->node)); 1920 1921 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1922 btrfs_set_backup_extent_root_gen(root_backup, 1923 btrfs_header_generation(info->extent_root->node)); 1924 btrfs_set_backup_extent_root_level(root_backup, 1925 btrfs_header_level(info->extent_root->node)); 1926 1927 /* 1928 * we might commit during log recovery, which happens before we set 1929 * the fs_root. Make sure it is valid before we fill it in. 1930 */ 1931 if (info->fs_root && info->fs_root->node) { 1932 btrfs_set_backup_fs_root(root_backup, 1933 info->fs_root->node->start); 1934 btrfs_set_backup_fs_root_gen(root_backup, 1935 btrfs_header_generation(info->fs_root->node)); 1936 btrfs_set_backup_fs_root_level(root_backup, 1937 btrfs_header_level(info->fs_root->node)); 1938 } 1939 1940 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1941 btrfs_set_backup_dev_root_gen(root_backup, 1942 btrfs_header_generation(info->dev_root->node)); 1943 btrfs_set_backup_dev_root_level(root_backup, 1944 btrfs_header_level(info->dev_root->node)); 1945 1946 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1947 btrfs_set_backup_csum_root_gen(root_backup, 1948 btrfs_header_generation(info->csum_root->node)); 1949 btrfs_set_backup_csum_root_level(root_backup, 1950 btrfs_header_level(info->csum_root->node)); 1951 1952 btrfs_set_backup_total_bytes(root_backup, 1953 btrfs_super_total_bytes(info->super_copy)); 1954 btrfs_set_backup_bytes_used(root_backup, 1955 btrfs_super_bytes_used(info->super_copy)); 1956 btrfs_set_backup_num_devices(root_backup, 1957 btrfs_super_num_devices(info->super_copy)); 1958 1959 /* 1960 * if we don't copy this out to the super_copy, it won't get remembered 1961 * for the next commit 1962 */ 1963 memcpy(&info->super_copy->super_roots, 1964 &info->super_for_commit->super_roots, 1965 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1966 } 1967 1968 /* 1969 * this copies info out of the root backup array and back into 1970 * the in-memory super block. It is meant to help iterate through 1971 * the array, so you send it the number of backups you've already 1972 * tried and the last backup index you used. 1973 * 1974 * this returns -1 when it has tried all the backups 1975 */ 1976 static noinline int next_root_backup(struct btrfs_fs_info *info, 1977 struct btrfs_super_block *super, 1978 int *num_backups_tried, int *backup_index) 1979 { 1980 struct btrfs_root_backup *root_backup; 1981 int newest = *backup_index; 1982 1983 if (*num_backups_tried == 0) { 1984 u64 gen = btrfs_super_generation(super); 1985 1986 newest = find_newest_super_backup(info, gen); 1987 if (newest == -1) 1988 return -1; 1989 1990 *backup_index = newest; 1991 *num_backups_tried = 1; 1992 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1993 /* we've tried all the backups, all done */ 1994 return -1; 1995 } else { 1996 /* jump to the next oldest backup */ 1997 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1998 BTRFS_NUM_BACKUP_ROOTS; 1999 *backup_index = newest; 2000 *num_backups_tried += 1; 2001 } 2002 root_backup = super->super_roots + newest; 2003 2004 btrfs_set_super_generation(super, 2005 btrfs_backup_tree_root_gen(root_backup)); 2006 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2007 btrfs_set_super_root_level(super, 2008 btrfs_backup_tree_root_level(root_backup)); 2009 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2010 2011 /* 2012 * fixme: the total bytes and num_devices need to match or we should 2013 * need a fsck 2014 */ 2015 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2016 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2017 return 0; 2018 } 2019 2020 /* helper to cleanup workers */ 2021 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 2022 { 2023 btrfs_destroy_workqueue(fs_info->fixup_workers); 2024 btrfs_destroy_workqueue(fs_info->delalloc_workers); 2025 btrfs_destroy_workqueue(fs_info->workers); 2026 btrfs_destroy_workqueue(fs_info->endio_workers); 2027 btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2028 btrfs_destroy_workqueue(fs_info->endio_repair_workers); 2029 btrfs_destroy_workqueue(fs_info->rmw_workers); 2030 btrfs_destroy_workqueue(fs_info->endio_write_workers); 2031 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2032 btrfs_destroy_workqueue(fs_info->submit_workers); 2033 btrfs_destroy_workqueue(fs_info->delayed_workers); 2034 btrfs_destroy_workqueue(fs_info->caching_workers); 2035 btrfs_destroy_workqueue(fs_info->readahead_workers); 2036 btrfs_destroy_workqueue(fs_info->flush_workers); 2037 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2038 btrfs_destroy_workqueue(fs_info->extent_workers); 2039 /* 2040 * Now that all other work queues are destroyed, we can safely destroy 2041 * the queues used for metadata I/O, since tasks from those other work 2042 * queues can do metadata I/O operations. 2043 */ 2044 btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2045 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2046 } 2047 2048 static void free_root_extent_buffers(struct btrfs_root *root) 2049 { 2050 if (root) { 2051 free_extent_buffer(root->node); 2052 free_extent_buffer(root->commit_root); 2053 root->node = NULL; 2054 root->commit_root = NULL; 2055 } 2056 } 2057 2058 /* helper to cleanup tree roots */ 2059 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2060 { 2061 free_root_extent_buffers(info->tree_root); 2062 2063 free_root_extent_buffers(info->dev_root); 2064 free_root_extent_buffers(info->extent_root); 2065 free_root_extent_buffers(info->csum_root); 2066 free_root_extent_buffers(info->quota_root); 2067 free_root_extent_buffers(info->uuid_root); 2068 if (chunk_root) 2069 free_root_extent_buffers(info->chunk_root); 2070 free_root_extent_buffers(info->free_space_root); 2071 } 2072 2073 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2074 { 2075 int ret; 2076 struct btrfs_root *gang[8]; 2077 int i; 2078 2079 while (!list_empty(&fs_info->dead_roots)) { 2080 gang[0] = list_entry(fs_info->dead_roots.next, 2081 struct btrfs_root, root_list); 2082 list_del(&gang[0]->root_list); 2083 2084 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2085 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2086 } else { 2087 free_extent_buffer(gang[0]->node); 2088 free_extent_buffer(gang[0]->commit_root); 2089 btrfs_put_fs_root(gang[0]); 2090 } 2091 } 2092 2093 while (1) { 2094 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2095 (void **)gang, 0, 2096 ARRAY_SIZE(gang)); 2097 if (!ret) 2098 break; 2099 for (i = 0; i < ret; i++) 2100 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2101 } 2102 2103 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 2104 btrfs_free_log_root_tree(NULL, fs_info); 2105 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 2106 } 2107 } 2108 2109 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) 2110 { 2111 mutex_init(&fs_info->scrub_lock); 2112 atomic_set(&fs_info->scrubs_running, 0); 2113 atomic_set(&fs_info->scrub_pause_req, 0); 2114 atomic_set(&fs_info->scrubs_paused, 0); 2115 atomic_set(&fs_info->scrub_cancel_req, 0); 2116 init_waitqueue_head(&fs_info->scrub_pause_wait); 2117 fs_info->scrub_workers_refcnt = 0; 2118 } 2119 2120 static void btrfs_init_balance(struct btrfs_fs_info *fs_info) 2121 { 2122 spin_lock_init(&fs_info->balance_lock); 2123 mutex_init(&fs_info->balance_mutex); 2124 atomic_set(&fs_info->balance_running, 0); 2125 atomic_set(&fs_info->balance_pause_req, 0); 2126 atomic_set(&fs_info->balance_cancel_req, 0); 2127 fs_info->balance_ctl = NULL; 2128 init_waitqueue_head(&fs_info->balance_wait_q); 2129 } 2130 2131 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) 2132 { 2133 struct inode *inode = fs_info->btree_inode; 2134 2135 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2136 set_nlink(inode, 1); 2137 /* 2138 * we set the i_size on the btree inode to the max possible int. 2139 * the real end of the address space is determined by all of 2140 * the devices in the system 2141 */ 2142 inode->i_size = OFFSET_MAX; 2143 inode->i_mapping->a_ops = &btree_aops; 2144 2145 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); 2146 extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode); 2147 BTRFS_I(inode)->io_tree.track_uptodate = 0; 2148 extent_map_tree_init(&BTRFS_I(inode)->extent_tree); 2149 2150 BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops; 2151 2152 BTRFS_I(inode)->root = fs_info->tree_root; 2153 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); 2154 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); 2155 btrfs_insert_inode_hash(inode); 2156 } 2157 2158 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) 2159 { 2160 fs_info->dev_replace.lock_owner = 0; 2161 atomic_set(&fs_info->dev_replace.nesting_level, 0); 2162 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2163 rwlock_init(&fs_info->dev_replace.lock); 2164 atomic_set(&fs_info->dev_replace.read_locks, 0); 2165 atomic_set(&fs_info->dev_replace.blocking_readers, 0); 2166 init_waitqueue_head(&fs_info->replace_wait); 2167 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq); 2168 } 2169 2170 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) 2171 { 2172 spin_lock_init(&fs_info->qgroup_lock); 2173 mutex_init(&fs_info->qgroup_ioctl_lock); 2174 fs_info->qgroup_tree = RB_ROOT; 2175 fs_info->qgroup_op_tree = RB_ROOT; 2176 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2177 fs_info->qgroup_seq = 1; 2178 fs_info->qgroup_ulist = NULL; 2179 fs_info->qgroup_rescan_running = false; 2180 mutex_init(&fs_info->qgroup_rescan_lock); 2181 } 2182 2183 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, 2184 struct btrfs_fs_devices *fs_devices) 2185 { 2186 int max_active = fs_info->thread_pool_size; 2187 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 2188 2189 fs_info->workers = 2190 btrfs_alloc_workqueue(fs_info, "worker", 2191 flags | WQ_HIGHPRI, max_active, 16); 2192 2193 fs_info->delalloc_workers = 2194 btrfs_alloc_workqueue(fs_info, "delalloc", 2195 flags, max_active, 2); 2196 2197 fs_info->flush_workers = 2198 btrfs_alloc_workqueue(fs_info, "flush_delalloc", 2199 flags, max_active, 0); 2200 2201 fs_info->caching_workers = 2202 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); 2203 2204 /* 2205 * a higher idle thresh on the submit workers makes it much more 2206 * likely that bios will be send down in a sane order to the 2207 * devices 2208 */ 2209 fs_info->submit_workers = 2210 btrfs_alloc_workqueue(fs_info, "submit", flags, 2211 min_t(u64, fs_devices->num_devices, 2212 max_active), 64); 2213 2214 fs_info->fixup_workers = 2215 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); 2216 2217 /* 2218 * endios are largely parallel and should have a very 2219 * low idle thresh 2220 */ 2221 fs_info->endio_workers = 2222 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); 2223 fs_info->endio_meta_workers = 2224 btrfs_alloc_workqueue(fs_info, "endio-meta", flags, 2225 max_active, 4); 2226 fs_info->endio_meta_write_workers = 2227 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, 2228 max_active, 2); 2229 fs_info->endio_raid56_workers = 2230 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, 2231 max_active, 4); 2232 fs_info->endio_repair_workers = 2233 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0); 2234 fs_info->rmw_workers = 2235 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); 2236 fs_info->endio_write_workers = 2237 btrfs_alloc_workqueue(fs_info, "endio-write", flags, 2238 max_active, 2); 2239 fs_info->endio_freespace_worker = 2240 btrfs_alloc_workqueue(fs_info, "freespace-write", flags, 2241 max_active, 0); 2242 fs_info->delayed_workers = 2243 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, 2244 max_active, 0); 2245 fs_info->readahead_workers = 2246 btrfs_alloc_workqueue(fs_info, "readahead", flags, 2247 max_active, 2); 2248 fs_info->qgroup_rescan_workers = 2249 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); 2250 fs_info->extent_workers = 2251 btrfs_alloc_workqueue(fs_info, "extent-refs", flags, 2252 min_t(u64, fs_devices->num_devices, 2253 max_active), 8); 2254 2255 if (!(fs_info->workers && fs_info->delalloc_workers && 2256 fs_info->submit_workers && fs_info->flush_workers && 2257 fs_info->endio_workers && fs_info->endio_meta_workers && 2258 fs_info->endio_meta_write_workers && 2259 fs_info->endio_repair_workers && 2260 fs_info->endio_write_workers && fs_info->endio_raid56_workers && 2261 fs_info->endio_freespace_worker && fs_info->rmw_workers && 2262 fs_info->caching_workers && fs_info->readahead_workers && 2263 fs_info->fixup_workers && fs_info->delayed_workers && 2264 fs_info->extent_workers && 2265 fs_info->qgroup_rescan_workers)) { 2266 return -ENOMEM; 2267 } 2268 2269 return 0; 2270 } 2271 2272 static int btrfs_replay_log(struct btrfs_fs_info *fs_info, 2273 struct btrfs_fs_devices *fs_devices) 2274 { 2275 int ret; 2276 struct btrfs_root *log_tree_root; 2277 struct btrfs_super_block *disk_super = fs_info->super_copy; 2278 u64 bytenr = btrfs_super_log_root(disk_super); 2279 2280 if (fs_devices->rw_devices == 0) { 2281 btrfs_warn(fs_info, "log replay required on RO media"); 2282 return -EIO; 2283 } 2284 2285 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2286 if (!log_tree_root) 2287 return -ENOMEM; 2288 2289 __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2290 2291 log_tree_root->node = read_tree_block(fs_info, bytenr, 2292 fs_info->generation + 1); 2293 if (IS_ERR(log_tree_root->node)) { 2294 btrfs_warn(fs_info, "failed to read log tree"); 2295 ret = PTR_ERR(log_tree_root->node); 2296 kfree(log_tree_root); 2297 return ret; 2298 } else if (!extent_buffer_uptodate(log_tree_root->node)) { 2299 btrfs_err(fs_info, "failed to read log tree"); 2300 free_extent_buffer(log_tree_root->node); 2301 kfree(log_tree_root); 2302 return -EIO; 2303 } 2304 /* returns with log_tree_root freed on success */ 2305 ret = btrfs_recover_log_trees(log_tree_root); 2306 if (ret) { 2307 btrfs_handle_fs_error(fs_info, ret, 2308 "Failed to recover log tree"); 2309 free_extent_buffer(log_tree_root->node); 2310 kfree(log_tree_root); 2311 return ret; 2312 } 2313 2314 if (sb_rdonly(fs_info->sb)) { 2315 ret = btrfs_commit_super(fs_info); 2316 if (ret) 2317 return ret; 2318 } 2319 2320 return 0; 2321 } 2322 2323 static int btrfs_read_roots(struct btrfs_fs_info *fs_info) 2324 { 2325 struct btrfs_root *tree_root = fs_info->tree_root; 2326 struct btrfs_root *root; 2327 struct btrfs_key location; 2328 int ret; 2329 2330 BUG_ON(!fs_info->tree_root); 2331 2332 location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 2333 location.type = BTRFS_ROOT_ITEM_KEY; 2334 location.offset = 0; 2335 2336 root = btrfs_read_tree_root(tree_root, &location); 2337 if (IS_ERR(root)) 2338 return PTR_ERR(root); 2339 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2340 fs_info->extent_root = root; 2341 2342 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2343 root = btrfs_read_tree_root(tree_root, &location); 2344 if (IS_ERR(root)) 2345 return PTR_ERR(root); 2346 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2347 fs_info->dev_root = root; 2348 btrfs_init_devices_late(fs_info); 2349 2350 location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2351 root = btrfs_read_tree_root(tree_root, &location); 2352 if (IS_ERR(root)) 2353 return PTR_ERR(root); 2354 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2355 fs_info->csum_root = root; 2356 2357 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2358 root = btrfs_read_tree_root(tree_root, &location); 2359 if (!IS_ERR(root)) { 2360 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2361 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); 2362 fs_info->quota_root = root; 2363 } 2364 2365 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2366 root = btrfs_read_tree_root(tree_root, &location); 2367 if (IS_ERR(root)) { 2368 ret = PTR_ERR(root); 2369 if (ret != -ENOENT) 2370 return ret; 2371 } else { 2372 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2373 fs_info->uuid_root = root; 2374 } 2375 2376 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2377 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; 2378 root = btrfs_read_tree_root(tree_root, &location); 2379 if (IS_ERR(root)) 2380 return PTR_ERR(root); 2381 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 2382 fs_info->free_space_root = root; 2383 } 2384 2385 return 0; 2386 } 2387 2388 int open_ctree(struct super_block *sb, 2389 struct btrfs_fs_devices *fs_devices, 2390 char *options) 2391 { 2392 u32 sectorsize; 2393 u32 nodesize; 2394 u32 stripesize; 2395 u64 generation; 2396 u64 features; 2397 struct btrfs_key location; 2398 struct buffer_head *bh; 2399 struct btrfs_super_block *disk_super; 2400 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2401 struct btrfs_root *tree_root; 2402 struct btrfs_root *chunk_root; 2403 int ret; 2404 int err = -EINVAL; 2405 int num_backups_tried = 0; 2406 int backup_index = 0; 2407 int max_active; 2408 int clear_free_space_tree = 0; 2409 2410 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2411 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL); 2412 if (!tree_root || !chunk_root) { 2413 err = -ENOMEM; 2414 goto fail; 2415 } 2416 2417 ret = init_srcu_struct(&fs_info->subvol_srcu); 2418 if (ret) { 2419 err = ret; 2420 goto fail; 2421 } 2422 2423 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); 2424 if (ret) { 2425 err = ret; 2426 goto fail_srcu; 2427 } 2428 fs_info->dirty_metadata_batch = PAGE_SIZE * 2429 (1 + ilog2(nr_cpu_ids)); 2430 2431 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); 2432 if (ret) { 2433 err = ret; 2434 goto fail_dirty_metadata_bytes; 2435 } 2436 2437 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL); 2438 if (ret) { 2439 err = ret; 2440 goto fail_delalloc_bytes; 2441 } 2442 2443 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2444 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 2445 INIT_LIST_HEAD(&fs_info->trans_list); 2446 INIT_LIST_HEAD(&fs_info->dead_roots); 2447 INIT_LIST_HEAD(&fs_info->delayed_iputs); 2448 INIT_LIST_HEAD(&fs_info->delalloc_roots); 2449 INIT_LIST_HEAD(&fs_info->caching_block_groups); 2450 spin_lock_init(&fs_info->delalloc_root_lock); 2451 spin_lock_init(&fs_info->trans_lock); 2452 spin_lock_init(&fs_info->fs_roots_radix_lock); 2453 spin_lock_init(&fs_info->delayed_iput_lock); 2454 spin_lock_init(&fs_info->defrag_inodes_lock); 2455 spin_lock_init(&fs_info->tree_mod_seq_lock); 2456 spin_lock_init(&fs_info->super_lock); 2457 spin_lock_init(&fs_info->qgroup_op_lock); 2458 spin_lock_init(&fs_info->buffer_lock); 2459 spin_lock_init(&fs_info->unused_bgs_lock); 2460 rwlock_init(&fs_info->tree_mod_log_lock); 2461 mutex_init(&fs_info->unused_bg_unpin_mutex); 2462 mutex_init(&fs_info->delete_unused_bgs_mutex); 2463 mutex_init(&fs_info->reloc_mutex); 2464 mutex_init(&fs_info->delalloc_root_mutex); 2465 mutex_init(&fs_info->cleaner_delayed_iput_mutex); 2466 seqlock_init(&fs_info->profiles_lock); 2467 2468 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 2469 INIT_LIST_HEAD(&fs_info->space_info); 2470 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 2471 INIT_LIST_HEAD(&fs_info->unused_bgs); 2472 btrfs_mapping_init(&fs_info->mapping_tree); 2473 btrfs_init_block_rsv(&fs_info->global_block_rsv, 2474 BTRFS_BLOCK_RSV_GLOBAL); 2475 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 2476 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 2477 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 2478 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 2479 BTRFS_BLOCK_RSV_DELOPS); 2480 atomic_set(&fs_info->async_delalloc_pages, 0); 2481 atomic_set(&fs_info->defrag_running, 0); 2482 atomic_set(&fs_info->qgroup_op_seq, 0); 2483 atomic_set(&fs_info->reada_works_cnt, 0); 2484 atomic64_set(&fs_info->tree_mod_seq, 0); 2485 fs_info->sb = sb; 2486 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; 2487 fs_info->metadata_ratio = 0; 2488 fs_info->defrag_inodes = RB_ROOT; 2489 atomic64_set(&fs_info->free_chunk_space, 0); 2490 fs_info->tree_mod_log = RB_ROOT; 2491 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2492 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ 2493 /* readahead state */ 2494 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 2495 spin_lock_init(&fs_info->reada_lock); 2496 btrfs_init_ref_verify(fs_info); 2497 2498 fs_info->thread_pool_size = min_t(unsigned long, 2499 num_online_cpus() + 2, 8); 2500 2501 INIT_LIST_HEAD(&fs_info->ordered_roots); 2502 spin_lock_init(&fs_info->ordered_root_lock); 2503 2504 fs_info->btree_inode = new_inode(sb); 2505 if (!fs_info->btree_inode) { 2506 err = -ENOMEM; 2507 goto fail_bio_counter; 2508 } 2509 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2510 2511 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2512 GFP_KERNEL); 2513 if (!fs_info->delayed_root) { 2514 err = -ENOMEM; 2515 goto fail_iput; 2516 } 2517 btrfs_init_delayed_root(fs_info->delayed_root); 2518 2519 btrfs_init_scrub(fs_info); 2520 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2521 fs_info->check_integrity_print_mask = 0; 2522 #endif 2523 btrfs_init_balance(fs_info); 2524 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2525 2526 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; 2527 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); 2528 2529 btrfs_init_btree_inode(fs_info); 2530 2531 spin_lock_init(&fs_info->block_group_cache_lock); 2532 fs_info->block_group_cache_tree = RB_ROOT; 2533 fs_info->first_logical_byte = (u64)-1; 2534 2535 extent_io_tree_init(&fs_info->freed_extents[0], NULL); 2536 extent_io_tree_init(&fs_info->freed_extents[1], NULL); 2537 fs_info->pinned_extents = &fs_info->freed_extents[0]; 2538 set_bit(BTRFS_FS_BARRIER, &fs_info->flags); 2539 2540 mutex_init(&fs_info->ordered_operations_mutex); 2541 mutex_init(&fs_info->tree_log_mutex); 2542 mutex_init(&fs_info->chunk_mutex); 2543 mutex_init(&fs_info->transaction_kthread_mutex); 2544 mutex_init(&fs_info->cleaner_mutex); 2545 mutex_init(&fs_info->volume_mutex); 2546 mutex_init(&fs_info->ro_block_group_mutex); 2547 init_rwsem(&fs_info->commit_root_sem); 2548 init_rwsem(&fs_info->cleanup_work_sem); 2549 init_rwsem(&fs_info->subvol_sem); 2550 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2551 2552 btrfs_init_dev_replace_locks(fs_info); 2553 btrfs_init_qgroup(fs_info); 2554 2555 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2556 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2557 2558 init_waitqueue_head(&fs_info->transaction_throttle); 2559 init_waitqueue_head(&fs_info->transaction_wait); 2560 init_waitqueue_head(&fs_info->transaction_blocked_wait); 2561 init_waitqueue_head(&fs_info->async_submit_wait); 2562 2563 INIT_LIST_HEAD(&fs_info->pinned_chunks); 2564 2565 /* Usable values until the real ones are cached from the superblock */ 2566 fs_info->nodesize = 4096; 2567 fs_info->sectorsize = 4096; 2568 fs_info->stripesize = 4096; 2569 2570 ret = btrfs_alloc_stripe_hash_table(fs_info); 2571 if (ret) { 2572 err = ret; 2573 goto fail_alloc; 2574 } 2575 2576 __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID); 2577 2578 invalidate_bdev(fs_devices->latest_bdev); 2579 2580 /* 2581 * Read super block and check the signature bytes only 2582 */ 2583 bh = btrfs_read_dev_super(fs_devices->latest_bdev); 2584 if (IS_ERR(bh)) { 2585 err = PTR_ERR(bh); 2586 goto fail_alloc; 2587 } 2588 2589 /* 2590 * We want to check superblock checksum, the type is stored inside. 2591 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 2592 */ 2593 if (btrfs_check_super_csum(fs_info, bh->b_data)) { 2594 btrfs_err(fs_info, "superblock checksum mismatch"); 2595 err = -EINVAL; 2596 brelse(bh); 2597 goto fail_alloc; 2598 } 2599 2600 /* 2601 * super_copy is zeroed at allocation time and we never touch the 2602 * following bytes up to INFO_SIZE, the checksum is calculated from 2603 * the whole block of INFO_SIZE 2604 */ 2605 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2606 memcpy(fs_info->super_for_commit, fs_info->super_copy, 2607 sizeof(*fs_info->super_for_commit)); 2608 brelse(bh); 2609 2610 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 2611 2612 ret = btrfs_check_super_valid(fs_info); 2613 if (ret) { 2614 btrfs_err(fs_info, "superblock contains fatal errors"); 2615 err = -EINVAL; 2616 goto fail_alloc; 2617 } 2618 2619 disk_super = fs_info->super_copy; 2620 if (!btrfs_super_root(disk_super)) 2621 goto fail_alloc; 2622 2623 /* check FS state, whether FS is broken. */ 2624 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 2625 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2626 2627 /* 2628 * run through our array of backup supers and setup 2629 * our ring pointer to the oldest one 2630 */ 2631 generation = btrfs_super_generation(disk_super); 2632 find_oldest_super_backup(fs_info, generation); 2633 2634 /* 2635 * In the long term, we'll store the compression type in the super 2636 * block, and it'll be used for per file compression control. 2637 */ 2638 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 2639 2640 ret = btrfs_parse_options(fs_info, options, sb->s_flags); 2641 if (ret) { 2642 err = ret; 2643 goto fail_alloc; 2644 } 2645 2646 features = btrfs_super_incompat_flags(disk_super) & 2647 ~BTRFS_FEATURE_INCOMPAT_SUPP; 2648 if (features) { 2649 btrfs_err(fs_info, 2650 "cannot mount because of unsupported optional features (%llx)", 2651 features); 2652 err = -EINVAL; 2653 goto fail_alloc; 2654 } 2655 2656 features = btrfs_super_incompat_flags(disk_super); 2657 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 2658 if (fs_info->compress_type == BTRFS_COMPRESS_LZO) 2659 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2660 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) 2661 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; 2662 2663 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 2664 btrfs_info(fs_info, "has skinny extents"); 2665 2666 /* 2667 * flag our filesystem as having big metadata blocks if 2668 * they are bigger than the page size 2669 */ 2670 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { 2671 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2672 btrfs_info(fs_info, 2673 "flagging fs with big metadata feature"); 2674 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2675 } 2676 2677 nodesize = btrfs_super_nodesize(disk_super); 2678 sectorsize = btrfs_super_sectorsize(disk_super); 2679 stripesize = sectorsize; 2680 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); 2681 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2682 2683 /* Cache block sizes */ 2684 fs_info->nodesize = nodesize; 2685 fs_info->sectorsize = sectorsize; 2686 fs_info->stripesize = stripesize; 2687 2688 /* 2689 * mixed block groups end up with duplicate but slightly offset 2690 * extent buffers for the same range. It leads to corruptions 2691 */ 2692 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2693 (sectorsize != nodesize)) { 2694 btrfs_err(fs_info, 2695 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", 2696 nodesize, sectorsize); 2697 goto fail_alloc; 2698 } 2699 2700 /* 2701 * Needn't use the lock because there is no other task which will 2702 * update the flag. 2703 */ 2704 btrfs_set_super_incompat_flags(disk_super, features); 2705 2706 features = btrfs_super_compat_ro_flags(disk_super) & 2707 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2708 if (!sb_rdonly(sb) && features) { 2709 btrfs_err(fs_info, 2710 "cannot mount read-write because of unsupported optional features (%llx)", 2711 features); 2712 err = -EINVAL; 2713 goto fail_alloc; 2714 } 2715 2716 max_active = fs_info->thread_pool_size; 2717 2718 ret = btrfs_init_workqueues(fs_info, fs_devices); 2719 if (ret) { 2720 err = ret; 2721 goto fail_sb_buffer; 2722 } 2723 2724 sb->s_bdi->congested_fn = btrfs_congested_fn; 2725 sb->s_bdi->congested_data = fs_info; 2726 sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK; 2727 sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE; 2728 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); 2729 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); 2730 2731 sb->s_blocksize = sectorsize; 2732 sb->s_blocksize_bits = blksize_bits(sectorsize); 2733 memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE); 2734 2735 mutex_lock(&fs_info->chunk_mutex); 2736 ret = btrfs_read_sys_array(fs_info); 2737 mutex_unlock(&fs_info->chunk_mutex); 2738 if (ret) { 2739 btrfs_err(fs_info, "failed to read the system array: %d", ret); 2740 goto fail_sb_buffer; 2741 } 2742 2743 generation = btrfs_super_chunk_root_generation(disk_super); 2744 2745 __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 2746 2747 chunk_root->node = read_tree_block(fs_info, 2748 btrfs_super_chunk_root(disk_super), 2749 generation); 2750 if (IS_ERR(chunk_root->node) || 2751 !extent_buffer_uptodate(chunk_root->node)) { 2752 btrfs_err(fs_info, "failed to read chunk root"); 2753 if (!IS_ERR(chunk_root->node)) 2754 free_extent_buffer(chunk_root->node); 2755 chunk_root->node = NULL; 2756 goto fail_tree_roots; 2757 } 2758 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 2759 chunk_root->commit_root = btrfs_root_node(chunk_root); 2760 2761 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2762 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2763 2764 ret = btrfs_read_chunk_tree(fs_info); 2765 if (ret) { 2766 btrfs_err(fs_info, "failed to read chunk tree: %d", ret); 2767 goto fail_tree_roots; 2768 } 2769 2770 /* 2771 * keep the device that is marked to be the target device for the 2772 * dev_replace procedure 2773 */ 2774 btrfs_close_extra_devices(fs_devices, 0); 2775 2776 if (!fs_devices->latest_bdev) { 2777 btrfs_err(fs_info, "failed to read devices"); 2778 goto fail_tree_roots; 2779 } 2780 2781 retry_root_backup: 2782 generation = btrfs_super_generation(disk_super); 2783 2784 tree_root->node = read_tree_block(fs_info, 2785 btrfs_super_root(disk_super), 2786 generation); 2787 if (IS_ERR(tree_root->node) || 2788 !extent_buffer_uptodate(tree_root->node)) { 2789 btrfs_warn(fs_info, "failed to read tree root"); 2790 if (!IS_ERR(tree_root->node)) 2791 free_extent_buffer(tree_root->node); 2792 tree_root->node = NULL; 2793 goto recovery_tree_root; 2794 } 2795 2796 btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2797 tree_root->commit_root = btrfs_root_node(tree_root); 2798 btrfs_set_root_refs(&tree_root->root_item, 1); 2799 2800 mutex_lock(&tree_root->objectid_mutex); 2801 ret = btrfs_find_highest_objectid(tree_root, 2802 &tree_root->highest_objectid); 2803 if (ret) { 2804 mutex_unlock(&tree_root->objectid_mutex); 2805 goto recovery_tree_root; 2806 } 2807 2808 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); 2809 2810 mutex_unlock(&tree_root->objectid_mutex); 2811 2812 ret = btrfs_read_roots(fs_info); 2813 if (ret) 2814 goto recovery_tree_root; 2815 2816 fs_info->generation = generation; 2817 fs_info->last_trans_committed = generation; 2818 2819 ret = btrfs_recover_balance(fs_info); 2820 if (ret) { 2821 btrfs_err(fs_info, "failed to recover balance: %d", ret); 2822 goto fail_block_groups; 2823 } 2824 2825 ret = btrfs_init_dev_stats(fs_info); 2826 if (ret) { 2827 btrfs_err(fs_info, "failed to init dev_stats: %d", ret); 2828 goto fail_block_groups; 2829 } 2830 2831 ret = btrfs_init_dev_replace(fs_info); 2832 if (ret) { 2833 btrfs_err(fs_info, "failed to init dev_replace: %d", ret); 2834 goto fail_block_groups; 2835 } 2836 2837 btrfs_close_extra_devices(fs_devices, 1); 2838 2839 ret = btrfs_sysfs_add_fsid(fs_devices, NULL); 2840 if (ret) { 2841 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", 2842 ret); 2843 goto fail_block_groups; 2844 } 2845 2846 ret = btrfs_sysfs_add_device(fs_devices); 2847 if (ret) { 2848 btrfs_err(fs_info, "failed to init sysfs device interface: %d", 2849 ret); 2850 goto fail_fsdev_sysfs; 2851 } 2852 2853 ret = btrfs_sysfs_add_mounted(fs_info); 2854 if (ret) { 2855 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); 2856 goto fail_fsdev_sysfs; 2857 } 2858 2859 ret = btrfs_init_space_info(fs_info); 2860 if (ret) { 2861 btrfs_err(fs_info, "failed to initialize space info: %d", ret); 2862 goto fail_sysfs; 2863 } 2864 2865 ret = btrfs_read_block_groups(fs_info); 2866 if (ret) { 2867 btrfs_err(fs_info, "failed to read block groups: %d", ret); 2868 goto fail_sysfs; 2869 } 2870 2871 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { 2872 btrfs_warn(fs_info, 2873 "writeable mount is not allowed due to too many missing devices"); 2874 goto fail_sysfs; 2875 } 2876 2877 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2878 "btrfs-cleaner"); 2879 if (IS_ERR(fs_info->cleaner_kthread)) 2880 goto fail_sysfs; 2881 2882 fs_info->transaction_kthread = kthread_run(transaction_kthread, 2883 tree_root, 2884 "btrfs-transaction"); 2885 if (IS_ERR(fs_info->transaction_kthread)) 2886 goto fail_cleaner; 2887 2888 if (!btrfs_test_opt(fs_info, NOSSD) && 2889 !fs_info->fs_devices->rotating) { 2890 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); 2891 } 2892 2893 /* 2894 * Mount does not set all options immediately, we can do it now and do 2895 * not have to wait for transaction commit 2896 */ 2897 btrfs_apply_pending_changes(fs_info); 2898 2899 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2900 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { 2901 ret = btrfsic_mount(fs_info, fs_devices, 2902 btrfs_test_opt(fs_info, 2903 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 2904 1 : 0, 2905 fs_info->check_integrity_print_mask); 2906 if (ret) 2907 btrfs_warn(fs_info, 2908 "failed to initialize integrity check module: %d", 2909 ret); 2910 } 2911 #endif 2912 ret = btrfs_read_qgroup_config(fs_info); 2913 if (ret) 2914 goto fail_trans_kthread; 2915 2916 if (btrfs_build_ref_tree(fs_info)) 2917 btrfs_err(fs_info, "couldn't build ref tree"); 2918 2919 /* do not make disk changes in broken FS or nologreplay is given */ 2920 if (btrfs_super_log_root(disk_super) != 0 && 2921 !btrfs_test_opt(fs_info, NOLOGREPLAY)) { 2922 ret = btrfs_replay_log(fs_info, fs_devices); 2923 if (ret) { 2924 err = ret; 2925 goto fail_qgroup; 2926 } 2927 } 2928 2929 ret = btrfs_find_orphan_roots(fs_info); 2930 if (ret) 2931 goto fail_qgroup; 2932 2933 if (!sb_rdonly(sb)) { 2934 ret = btrfs_cleanup_fs_roots(fs_info); 2935 if (ret) 2936 goto fail_qgroup; 2937 2938 mutex_lock(&fs_info->cleaner_mutex); 2939 ret = btrfs_recover_relocation(tree_root); 2940 mutex_unlock(&fs_info->cleaner_mutex); 2941 if (ret < 0) { 2942 btrfs_warn(fs_info, "failed to recover relocation: %d", 2943 ret); 2944 err = -EINVAL; 2945 goto fail_qgroup; 2946 } 2947 } 2948 2949 location.objectid = BTRFS_FS_TREE_OBJECTID; 2950 location.type = BTRFS_ROOT_ITEM_KEY; 2951 location.offset = 0; 2952 2953 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 2954 if (IS_ERR(fs_info->fs_root)) { 2955 err = PTR_ERR(fs_info->fs_root); 2956 goto fail_qgroup; 2957 } 2958 2959 if (sb_rdonly(sb)) 2960 return 0; 2961 2962 if (btrfs_test_opt(fs_info, CLEAR_CACHE) && 2963 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2964 clear_free_space_tree = 1; 2965 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && 2966 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { 2967 btrfs_warn(fs_info, "free space tree is invalid"); 2968 clear_free_space_tree = 1; 2969 } 2970 2971 if (clear_free_space_tree) { 2972 btrfs_info(fs_info, "clearing free space tree"); 2973 ret = btrfs_clear_free_space_tree(fs_info); 2974 if (ret) { 2975 btrfs_warn(fs_info, 2976 "failed to clear free space tree: %d", ret); 2977 close_ctree(fs_info); 2978 return ret; 2979 } 2980 } 2981 2982 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && 2983 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { 2984 btrfs_info(fs_info, "creating free space tree"); 2985 ret = btrfs_create_free_space_tree(fs_info); 2986 if (ret) { 2987 btrfs_warn(fs_info, 2988 "failed to create free space tree: %d", ret); 2989 close_ctree(fs_info); 2990 return ret; 2991 } 2992 } 2993 2994 down_read(&fs_info->cleanup_work_sem); 2995 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 2996 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 2997 up_read(&fs_info->cleanup_work_sem); 2998 close_ctree(fs_info); 2999 return ret; 3000 } 3001 up_read(&fs_info->cleanup_work_sem); 3002 3003 ret = btrfs_resume_balance_async(fs_info); 3004 if (ret) { 3005 btrfs_warn(fs_info, "failed to resume balance: %d", ret); 3006 close_ctree(fs_info); 3007 return ret; 3008 } 3009 3010 ret = btrfs_resume_dev_replace_async(fs_info); 3011 if (ret) { 3012 btrfs_warn(fs_info, "failed to resume device replace: %d", ret); 3013 close_ctree(fs_info); 3014 return ret; 3015 } 3016 3017 btrfs_qgroup_rescan_resume(fs_info); 3018 3019 if (!fs_info->uuid_root) { 3020 btrfs_info(fs_info, "creating UUID tree"); 3021 ret = btrfs_create_uuid_tree(fs_info); 3022 if (ret) { 3023 btrfs_warn(fs_info, 3024 "failed to create the UUID tree: %d", ret); 3025 close_ctree(fs_info); 3026 return ret; 3027 } 3028 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || 3029 fs_info->generation != 3030 btrfs_super_uuid_tree_generation(disk_super)) { 3031 btrfs_info(fs_info, "checking UUID tree"); 3032 ret = btrfs_check_uuid_tree(fs_info); 3033 if (ret) { 3034 btrfs_warn(fs_info, 3035 "failed to check the UUID tree: %d", ret); 3036 close_ctree(fs_info); 3037 return ret; 3038 } 3039 } else { 3040 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 3041 } 3042 set_bit(BTRFS_FS_OPEN, &fs_info->flags); 3043 3044 /* 3045 * backuproot only affect mount behavior, and if open_ctree succeeded, 3046 * no need to keep the flag 3047 */ 3048 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); 3049 3050 return 0; 3051 3052 fail_qgroup: 3053 btrfs_free_qgroup_config(fs_info); 3054 fail_trans_kthread: 3055 kthread_stop(fs_info->transaction_kthread); 3056 btrfs_cleanup_transaction(fs_info); 3057 btrfs_free_fs_roots(fs_info); 3058 fail_cleaner: 3059 kthread_stop(fs_info->cleaner_kthread); 3060 3061 /* 3062 * make sure we're done with the btree inode before we stop our 3063 * kthreads 3064 */ 3065 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 3066 3067 fail_sysfs: 3068 btrfs_sysfs_remove_mounted(fs_info); 3069 3070 fail_fsdev_sysfs: 3071 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3072 3073 fail_block_groups: 3074 btrfs_put_block_group_cache(fs_info); 3075 3076 fail_tree_roots: 3077 free_root_pointers(fs_info, 1); 3078 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3079 3080 fail_sb_buffer: 3081 btrfs_stop_all_workers(fs_info); 3082 btrfs_free_block_groups(fs_info); 3083 fail_alloc: 3084 fail_iput: 3085 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3086 3087 iput(fs_info->btree_inode); 3088 fail_bio_counter: 3089 percpu_counter_destroy(&fs_info->bio_counter); 3090 fail_delalloc_bytes: 3091 percpu_counter_destroy(&fs_info->delalloc_bytes); 3092 fail_dirty_metadata_bytes: 3093 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3094 fail_srcu: 3095 cleanup_srcu_struct(&fs_info->subvol_srcu); 3096 fail: 3097 btrfs_free_stripe_hash_table(fs_info); 3098 btrfs_close_devices(fs_info->fs_devices); 3099 return err; 3100 3101 recovery_tree_root: 3102 if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) 3103 goto fail_tree_roots; 3104 3105 free_root_pointers(fs_info, 0); 3106 3107 /* don't use the log in recovery mode, it won't be valid */ 3108 btrfs_set_super_log_root(disk_super, 0); 3109 3110 /* we can't trust the free space cache either */ 3111 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3112 3113 ret = next_root_backup(fs_info, fs_info->super_copy, 3114 &num_backups_tried, &backup_index); 3115 if (ret == -1) 3116 goto fail_block_groups; 3117 goto retry_root_backup; 3118 } 3119 ALLOW_ERROR_INJECTION(open_ctree, ERRNO); 3120 3121 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3122 { 3123 if (uptodate) { 3124 set_buffer_uptodate(bh); 3125 } else { 3126 struct btrfs_device *device = (struct btrfs_device *) 3127 bh->b_private; 3128 3129 btrfs_warn_rl_in_rcu(device->fs_info, 3130 "lost page write due to IO error on %s", 3131 rcu_str_deref(device->name)); 3132 /* note, we don't set_buffer_write_io_error because we have 3133 * our own ways of dealing with the IO errors 3134 */ 3135 clear_buffer_uptodate(bh); 3136 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3137 } 3138 unlock_buffer(bh); 3139 put_bh(bh); 3140 } 3141 3142 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num, 3143 struct buffer_head **bh_ret) 3144 { 3145 struct buffer_head *bh; 3146 struct btrfs_super_block *super; 3147 u64 bytenr; 3148 3149 bytenr = btrfs_sb_offset(copy_num); 3150 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) 3151 return -EINVAL; 3152 3153 bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE); 3154 /* 3155 * If we fail to read from the underlying devices, as of now 3156 * the best option we have is to mark it EIO. 3157 */ 3158 if (!bh) 3159 return -EIO; 3160 3161 super = (struct btrfs_super_block *)bh->b_data; 3162 if (btrfs_super_bytenr(super) != bytenr || 3163 btrfs_super_magic(super) != BTRFS_MAGIC) { 3164 brelse(bh); 3165 return -EINVAL; 3166 } 3167 3168 *bh_ret = bh; 3169 return 0; 3170 } 3171 3172 3173 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3174 { 3175 struct buffer_head *bh; 3176 struct buffer_head *latest = NULL; 3177 struct btrfs_super_block *super; 3178 int i; 3179 u64 transid = 0; 3180 int ret = -EINVAL; 3181 3182 /* we would like to check all the supers, but that would make 3183 * a btrfs mount succeed after a mkfs from a different FS. 3184 * So, we need to add a special mount option to scan for 3185 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3186 */ 3187 for (i = 0; i < 1; i++) { 3188 ret = btrfs_read_dev_one_super(bdev, i, &bh); 3189 if (ret) 3190 continue; 3191 3192 super = (struct btrfs_super_block *)bh->b_data; 3193 3194 if (!latest || btrfs_super_generation(super) > transid) { 3195 brelse(latest); 3196 latest = bh; 3197 transid = btrfs_super_generation(super); 3198 } else { 3199 brelse(bh); 3200 } 3201 } 3202 3203 if (!latest) 3204 return ERR_PTR(ret); 3205 3206 return latest; 3207 } 3208 3209 /* 3210 * Write superblock @sb to the @device. Do not wait for completion, all the 3211 * buffer heads we write are pinned. 3212 * 3213 * Write @max_mirrors copies of the superblock, where 0 means default that fit 3214 * the expected device size at commit time. Note that max_mirrors must be 3215 * same for write and wait phases. 3216 * 3217 * Return number of errors when buffer head is not found or submission fails. 3218 */ 3219 static int write_dev_supers(struct btrfs_device *device, 3220 struct btrfs_super_block *sb, int max_mirrors) 3221 { 3222 struct buffer_head *bh; 3223 int i; 3224 int ret; 3225 int errors = 0; 3226 u32 crc; 3227 u64 bytenr; 3228 int op_flags; 3229 3230 if (max_mirrors == 0) 3231 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3232 3233 for (i = 0; i < max_mirrors; i++) { 3234 bytenr = btrfs_sb_offset(i); 3235 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3236 device->commit_total_bytes) 3237 break; 3238 3239 btrfs_set_super_bytenr(sb, bytenr); 3240 3241 crc = ~(u32)0; 3242 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc, 3243 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3244 btrfs_csum_final(crc, sb->csum); 3245 3246 /* One reference for us, and we leave it for the caller */ 3247 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, 3248 BTRFS_SUPER_INFO_SIZE); 3249 if (!bh) { 3250 btrfs_err(device->fs_info, 3251 "couldn't get super buffer head for bytenr %llu", 3252 bytenr); 3253 errors++; 3254 continue; 3255 } 3256 3257 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3258 3259 /* one reference for submit_bh */ 3260 get_bh(bh); 3261 3262 set_buffer_uptodate(bh); 3263 lock_buffer(bh); 3264 bh->b_end_io = btrfs_end_buffer_write_sync; 3265 bh->b_private = device; 3266 3267 /* 3268 * we fua the first super. The others we allow 3269 * to go down lazy. 3270 */ 3271 op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 3272 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) 3273 op_flags |= REQ_FUA; 3274 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh); 3275 if (ret) 3276 errors++; 3277 } 3278 return errors < i ? 0 : -1; 3279 } 3280 3281 /* 3282 * Wait for write completion of superblocks done by write_dev_supers, 3283 * @max_mirrors same for write and wait phases. 3284 * 3285 * Return number of errors when buffer head is not found or not marked up to 3286 * date. 3287 */ 3288 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) 3289 { 3290 struct buffer_head *bh; 3291 int i; 3292 int errors = 0; 3293 u64 bytenr; 3294 3295 if (max_mirrors == 0) 3296 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3297 3298 for (i = 0; i < max_mirrors; i++) { 3299 bytenr = btrfs_sb_offset(i); 3300 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3301 device->commit_total_bytes) 3302 break; 3303 3304 bh = __find_get_block(device->bdev, 3305 bytenr / BTRFS_BDEV_BLOCKSIZE, 3306 BTRFS_SUPER_INFO_SIZE); 3307 if (!bh) { 3308 errors++; 3309 continue; 3310 } 3311 wait_on_buffer(bh); 3312 if (!buffer_uptodate(bh)) 3313 errors++; 3314 3315 /* drop our reference */ 3316 brelse(bh); 3317 3318 /* drop the reference from the writing run */ 3319 brelse(bh); 3320 } 3321 3322 return errors < i ? 0 : -1; 3323 } 3324 3325 /* 3326 * endio for the write_dev_flush, this will wake anyone waiting 3327 * for the barrier when it is done 3328 */ 3329 static void btrfs_end_empty_barrier(struct bio *bio) 3330 { 3331 complete(bio->bi_private); 3332 } 3333 3334 /* 3335 * Submit a flush request to the device if it supports it. Error handling is 3336 * done in the waiting counterpart. 3337 */ 3338 static void write_dev_flush(struct btrfs_device *device) 3339 { 3340 struct request_queue *q = bdev_get_queue(device->bdev); 3341 struct bio *bio = device->flush_bio; 3342 3343 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 3344 return; 3345 3346 bio_reset(bio); 3347 bio->bi_end_io = btrfs_end_empty_barrier; 3348 bio_set_dev(bio, device->bdev); 3349 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; 3350 init_completion(&device->flush_wait); 3351 bio->bi_private = &device->flush_wait; 3352 3353 btrfsic_submit_bio(bio); 3354 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 3355 } 3356 3357 /* 3358 * If the flush bio has been submitted by write_dev_flush, wait for it. 3359 */ 3360 static blk_status_t wait_dev_flush(struct btrfs_device *device) 3361 { 3362 struct bio *bio = device->flush_bio; 3363 3364 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) 3365 return BLK_STS_OK; 3366 3367 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); 3368 wait_for_completion_io(&device->flush_wait); 3369 3370 return bio->bi_status; 3371 } 3372 3373 static int check_barrier_error(struct btrfs_fs_info *fs_info) 3374 { 3375 if (!btrfs_check_rw_degradable(fs_info, NULL)) 3376 return -EIO; 3377 return 0; 3378 } 3379 3380 /* 3381 * send an empty flush down to each device in parallel, 3382 * then wait for them 3383 */ 3384 static int barrier_all_devices(struct btrfs_fs_info *info) 3385 { 3386 struct list_head *head; 3387 struct btrfs_device *dev; 3388 int errors_wait = 0; 3389 blk_status_t ret; 3390 3391 lockdep_assert_held(&info->fs_devices->device_list_mutex); 3392 /* send down all the barriers */ 3393 head = &info->fs_devices->devices; 3394 list_for_each_entry(dev, head, dev_list) { 3395 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3396 continue; 3397 if (!dev->bdev) 3398 continue; 3399 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3400 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3401 continue; 3402 3403 write_dev_flush(dev); 3404 dev->last_flush_error = BLK_STS_OK; 3405 } 3406 3407 /* wait for all the barriers */ 3408 list_for_each_entry(dev, head, dev_list) { 3409 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) 3410 continue; 3411 if (!dev->bdev) { 3412 errors_wait++; 3413 continue; 3414 } 3415 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3416 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3417 continue; 3418 3419 ret = wait_dev_flush(dev); 3420 if (ret) { 3421 dev->last_flush_error = ret; 3422 btrfs_dev_stat_inc_and_print(dev, 3423 BTRFS_DEV_STAT_FLUSH_ERRS); 3424 errors_wait++; 3425 } 3426 } 3427 3428 if (errors_wait) { 3429 /* 3430 * At some point we need the status of all disks 3431 * to arrive at the volume status. So error checking 3432 * is being pushed to a separate loop. 3433 */ 3434 return check_barrier_error(info); 3435 } 3436 return 0; 3437 } 3438 3439 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) 3440 { 3441 int raid_type; 3442 int min_tolerated = INT_MAX; 3443 3444 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || 3445 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) 3446 min_tolerated = min(min_tolerated, 3447 btrfs_raid_array[BTRFS_RAID_SINGLE]. 3448 tolerated_failures); 3449 3450 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { 3451 if (raid_type == BTRFS_RAID_SINGLE) 3452 continue; 3453 if (!(flags & btrfs_raid_group[raid_type])) 3454 continue; 3455 min_tolerated = min(min_tolerated, 3456 btrfs_raid_array[raid_type]. 3457 tolerated_failures); 3458 } 3459 3460 if (min_tolerated == INT_MAX) { 3461 pr_warn("BTRFS: unknown raid flag: %llu", flags); 3462 min_tolerated = 0; 3463 } 3464 3465 return min_tolerated; 3466 } 3467 3468 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) 3469 { 3470 struct list_head *head; 3471 struct btrfs_device *dev; 3472 struct btrfs_super_block *sb; 3473 struct btrfs_dev_item *dev_item; 3474 int ret; 3475 int do_barriers; 3476 int max_errors; 3477 int total_errors = 0; 3478 u64 flags; 3479 3480 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); 3481 3482 /* 3483 * max_mirrors == 0 indicates we're from commit_transaction, 3484 * not from fsync where the tree roots in fs_info have not 3485 * been consistent on disk. 3486 */ 3487 if (max_mirrors == 0) 3488 backup_super_roots(fs_info); 3489 3490 sb = fs_info->super_for_commit; 3491 dev_item = &sb->dev_item; 3492 3493 mutex_lock(&fs_info->fs_devices->device_list_mutex); 3494 head = &fs_info->fs_devices->devices; 3495 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; 3496 3497 if (do_barriers) { 3498 ret = barrier_all_devices(fs_info); 3499 if (ret) { 3500 mutex_unlock( 3501 &fs_info->fs_devices->device_list_mutex); 3502 btrfs_handle_fs_error(fs_info, ret, 3503 "errors while submitting device barriers."); 3504 return ret; 3505 } 3506 } 3507 3508 list_for_each_entry(dev, head, dev_list) { 3509 if (!dev->bdev) { 3510 total_errors++; 3511 continue; 3512 } 3513 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3514 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3515 continue; 3516 3517 btrfs_set_stack_device_generation(dev_item, 0); 3518 btrfs_set_stack_device_type(dev_item, dev->type); 3519 btrfs_set_stack_device_id(dev_item, dev->devid); 3520 btrfs_set_stack_device_total_bytes(dev_item, 3521 dev->commit_total_bytes); 3522 btrfs_set_stack_device_bytes_used(dev_item, 3523 dev->commit_bytes_used); 3524 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3525 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3526 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3527 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 3528 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE); 3529 3530 flags = btrfs_super_flags(sb); 3531 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3532 3533 ret = write_dev_supers(dev, sb, max_mirrors); 3534 if (ret) 3535 total_errors++; 3536 } 3537 if (total_errors > max_errors) { 3538 btrfs_err(fs_info, "%d errors while writing supers", 3539 total_errors); 3540 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3541 3542 /* FUA is masked off if unsupported and can't be the reason */ 3543 btrfs_handle_fs_error(fs_info, -EIO, 3544 "%d errors while writing supers", 3545 total_errors); 3546 return -EIO; 3547 } 3548 3549 total_errors = 0; 3550 list_for_each_entry(dev, head, dev_list) { 3551 if (!dev->bdev) 3552 continue; 3553 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || 3554 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) 3555 continue; 3556 3557 ret = wait_dev_supers(dev, max_mirrors); 3558 if (ret) 3559 total_errors++; 3560 } 3561 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3562 if (total_errors > max_errors) { 3563 btrfs_handle_fs_error(fs_info, -EIO, 3564 "%d errors while writing supers", 3565 total_errors); 3566 return -EIO; 3567 } 3568 return 0; 3569 } 3570 3571 /* Drop a fs root from the radix tree and free it. */ 3572 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3573 struct btrfs_root *root) 3574 { 3575 spin_lock(&fs_info->fs_roots_radix_lock); 3576 radix_tree_delete(&fs_info->fs_roots_radix, 3577 (unsigned long)root->root_key.objectid); 3578 spin_unlock(&fs_info->fs_roots_radix_lock); 3579 3580 if (btrfs_root_refs(&root->root_item) == 0) 3581 synchronize_srcu(&fs_info->subvol_srcu); 3582 3583 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 3584 btrfs_free_log(NULL, root); 3585 if (root->reloc_root) { 3586 free_extent_buffer(root->reloc_root->node); 3587 free_extent_buffer(root->reloc_root->commit_root); 3588 btrfs_put_fs_root(root->reloc_root); 3589 root->reloc_root = NULL; 3590 } 3591 } 3592 3593 if (root->free_ino_pinned) 3594 __btrfs_remove_free_space_cache(root->free_ino_pinned); 3595 if (root->free_ino_ctl) 3596 __btrfs_remove_free_space_cache(root->free_ino_ctl); 3597 free_fs_root(root); 3598 } 3599 3600 static void free_fs_root(struct btrfs_root *root) 3601 { 3602 iput(root->ino_cache_inode); 3603 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 3604 btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv); 3605 root->orphan_block_rsv = NULL; 3606 if (root->anon_dev) 3607 free_anon_bdev(root->anon_dev); 3608 if (root->subv_writers) 3609 btrfs_free_subvolume_writers(root->subv_writers); 3610 free_extent_buffer(root->node); 3611 free_extent_buffer(root->commit_root); 3612 kfree(root->free_ino_ctl); 3613 kfree(root->free_ino_pinned); 3614 kfree(root->name); 3615 btrfs_put_fs_root(root); 3616 } 3617 3618 void btrfs_free_fs_root(struct btrfs_root *root) 3619 { 3620 free_fs_root(root); 3621 } 3622 3623 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3624 { 3625 u64 root_objectid = 0; 3626 struct btrfs_root *gang[8]; 3627 int i = 0; 3628 int err = 0; 3629 unsigned int ret = 0; 3630 int index; 3631 3632 while (1) { 3633 index = srcu_read_lock(&fs_info->subvol_srcu); 3634 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3635 (void **)gang, root_objectid, 3636 ARRAY_SIZE(gang)); 3637 if (!ret) { 3638 srcu_read_unlock(&fs_info->subvol_srcu, index); 3639 break; 3640 } 3641 root_objectid = gang[ret - 1]->root_key.objectid + 1; 3642 3643 for (i = 0; i < ret; i++) { 3644 /* Avoid to grab roots in dead_roots */ 3645 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 3646 gang[i] = NULL; 3647 continue; 3648 } 3649 /* grab all the search result for later use */ 3650 gang[i] = btrfs_grab_fs_root(gang[i]); 3651 } 3652 srcu_read_unlock(&fs_info->subvol_srcu, index); 3653 3654 for (i = 0; i < ret; i++) { 3655 if (!gang[i]) 3656 continue; 3657 root_objectid = gang[i]->root_key.objectid; 3658 err = btrfs_orphan_cleanup(gang[i]); 3659 if (err) 3660 break; 3661 btrfs_put_fs_root(gang[i]); 3662 } 3663 root_objectid++; 3664 } 3665 3666 /* release the uncleaned roots due to error */ 3667 for (; i < ret; i++) { 3668 if (gang[i]) 3669 btrfs_put_fs_root(gang[i]); 3670 } 3671 return err; 3672 } 3673 3674 int btrfs_commit_super(struct btrfs_fs_info *fs_info) 3675 { 3676 struct btrfs_root *root = fs_info->tree_root; 3677 struct btrfs_trans_handle *trans; 3678 3679 mutex_lock(&fs_info->cleaner_mutex); 3680 btrfs_run_delayed_iputs(fs_info); 3681 mutex_unlock(&fs_info->cleaner_mutex); 3682 wake_up_process(fs_info->cleaner_kthread); 3683 3684 /* wait until ongoing cleanup work done */ 3685 down_write(&fs_info->cleanup_work_sem); 3686 up_write(&fs_info->cleanup_work_sem); 3687 3688 trans = btrfs_join_transaction(root); 3689 if (IS_ERR(trans)) 3690 return PTR_ERR(trans); 3691 return btrfs_commit_transaction(trans); 3692 } 3693 3694 void close_ctree(struct btrfs_fs_info *fs_info) 3695 { 3696 struct btrfs_root *root = fs_info->tree_root; 3697 int ret; 3698 3699 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); 3700 3701 /* wait for the qgroup rescan worker to stop */ 3702 btrfs_qgroup_wait_for_completion(fs_info, false); 3703 3704 /* wait for the uuid_scan task to finish */ 3705 down(&fs_info->uuid_tree_rescan_sem); 3706 /* avoid complains from lockdep et al., set sem back to initial state */ 3707 up(&fs_info->uuid_tree_rescan_sem); 3708 3709 /* pause restriper - we want to resume on mount */ 3710 btrfs_pause_balance(fs_info); 3711 3712 btrfs_dev_replace_suspend_for_unmount(fs_info); 3713 3714 btrfs_scrub_cancel(fs_info); 3715 3716 /* wait for any defraggers to finish */ 3717 wait_event(fs_info->transaction_wait, 3718 (atomic_read(&fs_info->defrag_running) == 0)); 3719 3720 /* clear out the rbtree of defraggable inodes */ 3721 btrfs_cleanup_defrag_inodes(fs_info); 3722 3723 cancel_work_sync(&fs_info->async_reclaim_work); 3724 3725 if (!sb_rdonly(fs_info->sb)) { 3726 /* 3727 * If the cleaner thread is stopped and there are 3728 * block groups queued for removal, the deletion will be 3729 * skipped when we quit the cleaner thread. 3730 */ 3731 btrfs_delete_unused_bgs(fs_info); 3732 3733 ret = btrfs_commit_super(fs_info); 3734 if (ret) 3735 btrfs_err(fs_info, "commit super ret %d", ret); 3736 } 3737 3738 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 3739 btrfs_error_commit_super(fs_info); 3740 3741 kthread_stop(fs_info->transaction_kthread); 3742 kthread_stop(fs_info->cleaner_kthread); 3743 3744 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); 3745 3746 btrfs_free_qgroup_config(fs_info); 3747 3748 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 3749 btrfs_info(fs_info, "at unmount delalloc count %lld", 3750 percpu_counter_sum(&fs_info->delalloc_bytes)); 3751 } 3752 3753 btrfs_sysfs_remove_mounted(fs_info); 3754 btrfs_sysfs_remove_fsid(fs_info->fs_devices); 3755 3756 btrfs_free_fs_roots(fs_info); 3757 3758 btrfs_put_block_group_cache(fs_info); 3759 3760 /* 3761 * we must make sure there is not any read request to 3762 * submit after we stopping all workers. 3763 */ 3764 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3765 btrfs_stop_all_workers(fs_info); 3766 3767 btrfs_free_block_groups(fs_info); 3768 3769 clear_bit(BTRFS_FS_OPEN, &fs_info->flags); 3770 free_root_pointers(fs_info, 1); 3771 3772 iput(fs_info->btree_inode); 3773 3774 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3775 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) 3776 btrfsic_unmount(fs_info->fs_devices); 3777 #endif 3778 3779 btrfs_close_devices(fs_info->fs_devices); 3780 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3781 3782 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3783 percpu_counter_destroy(&fs_info->delalloc_bytes); 3784 percpu_counter_destroy(&fs_info->bio_counter); 3785 cleanup_srcu_struct(&fs_info->subvol_srcu); 3786 3787 btrfs_free_stripe_hash_table(fs_info); 3788 btrfs_free_ref_cache(fs_info); 3789 3790 __btrfs_free_block_rsv(root->orphan_block_rsv); 3791 root->orphan_block_rsv = NULL; 3792 3793 while (!list_empty(&fs_info->pinned_chunks)) { 3794 struct extent_map *em; 3795 3796 em = list_first_entry(&fs_info->pinned_chunks, 3797 struct extent_map, list); 3798 list_del_init(&em->list); 3799 free_extent_map(em); 3800 } 3801 } 3802 3803 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3804 int atomic) 3805 { 3806 int ret; 3807 struct inode *btree_inode = buf->pages[0]->mapping->host; 3808 3809 ret = extent_buffer_uptodate(buf); 3810 if (!ret) 3811 return ret; 3812 3813 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3814 parent_transid, atomic); 3815 if (ret == -EAGAIN) 3816 return ret; 3817 return !ret; 3818 } 3819 3820 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 3821 { 3822 struct btrfs_fs_info *fs_info; 3823 struct btrfs_root *root; 3824 u64 transid = btrfs_header_generation(buf); 3825 int was_dirty; 3826 3827 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 3828 /* 3829 * This is a fast path so only do this check if we have sanity tests 3830 * enabled. Normal people shouldn't be marking dummy buffers as dirty 3831 * outside of the sanity tests. 3832 */ 3833 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 3834 return; 3835 #endif 3836 root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3837 fs_info = root->fs_info; 3838 btrfs_assert_tree_locked(buf); 3839 if (transid != fs_info->generation) 3840 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", 3841 buf->start, transid, fs_info->generation); 3842 was_dirty = set_extent_buffer_dirty(buf); 3843 if (!was_dirty) 3844 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, 3845 buf->len, 3846 fs_info->dirty_metadata_batch); 3847 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3848 /* 3849 * Since btrfs_mark_buffer_dirty() can be called with item pointer set 3850 * but item data not updated. 3851 * So here we should only check item pointers, not item data. 3852 */ 3853 if (btrfs_header_level(buf) == 0 && 3854 btrfs_check_leaf_relaxed(root, buf)) { 3855 btrfs_print_leaf(buf); 3856 ASSERT(0); 3857 } 3858 #endif 3859 } 3860 3861 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, 3862 int flush_delayed) 3863 { 3864 /* 3865 * looks as though older kernels can get into trouble with 3866 * this code, they end up stuck in balance_dirty_pages forever 3867 */ 3868 int ret; 3869 3870 if (current->flags & PF_MEMALLOC) 3871 return; 3872 3873 if (flush_delayed) 3874 btrfs_balance_delayed_items(fs_info); 3875 3876 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 3877 BTRFS_DIRTY_METADATA_THRESH); 3878 if (ret > 0) { 3879 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); 3880 } 3881 } 3882 3883 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) 3884 { 3885 __btrfs_btree_balance_dirty(fs_info, 1); 3886 } 3887 3888 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) 3889 { 3890 __btrfs_btree_balance_dirty(fs_info, 0); 3891 } 3892 3893 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 3894 { 3895 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3896 struct btrfs_fs_info *fs_info = root->fs_info; 3897 3898 return btree_read_extent_buffer_pages(fs_info, buf, parent_transid); 3899 } 3900 3901 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info) 3902 { 3903 struct btrfs_super_block *sb = fs_info->super_copy; 3904 u64 nodesize = btrfs_super_nodesize(sb); 3905 u64 sectorsize = btrfs_super_sectorsize(sb); 3906 int ret = 0; 3907 3908 if (btrfs_super_magic(sb) != BTRFS_MAGIC) { 3909 btrfs_err(fs_info, "no valid FS found"); 3910 ret = -EINVAL; 3911 } 3912 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { 3913 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", 3914 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); 3915 ret = -EINVAL; 3916 } 3917 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { 3918 btrfs_err(fs_info, "tree_root level too big: %d >= %d", 3919 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); 3920 ret = -EINVAL; 3921 } 3922 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { 3923 btrfs_err(fs_info, "chunk_root level too big: %d >= %d", 3924 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); 3925 ret = -EINVAL; 3926 } 3927 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { 3928 btrfs_err(fs_info, "log_root level too big: %d >= %d", 3929 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); 3930 ret = -EINVAL; 3931 } 3932 3933 /* 3934 * Check sectorsize and nodesize first, other check will need it. 3935 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. 3936 */ 3937 if (!is_power_of_2(sectorsize) || sectorsize < 4096 || 3938 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { 3939 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); 3940 ret = -EINVAL; 3941 } 3942 /* Only PAGE SIZE is supported yet */ 3943 if (sectorsize != PAGE_SIZE) { 3944 btrfs_err(fs_info, 3945 "sectorsize %llu not supported yet, only support %lu", 3946 sectorsize, PAGE_SIZE); 3947 ret = -EINVAL; 3948 } 3949 if (!is_power_of_2(nodesize) || nodesize < sectorsize || 3950 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { 3951 btrfs_err(fs_info, "invalid nodesize %llu", nodesize); 3952 ret = -EINVAL; 3953 } 3954 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { 3955 btrfs_err(fs_info, "invalid leafsize %u, should be %llu", 3956 le32_to_cpu(sb->__unused_leafsize), nodesize); 3957 ret = -EINVAL; 3958 } 3959 3960 /* Root alignment check */ 3961 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { 3962 btrfs_warn(fs_info, "tree_root block unaligned: %llu", 3963 btrfs_super_root(sb)); 3964 ret = -EINVAL; 3965 } 3966 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { 3967 btrfs_warn(fs_info, "chunk_root block unaligned: %llu", 3968 btrfs_super_chunk_root(sb)); 3969 ret = -EINVAL; 3970 } 3971 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { 3972 btrfs_warn(fs_info, "log_root block unaligned: %llu", 3973 btrfs_super_log_root(sb)); 3974 ret = -EINVAL; 3975 } 3976 3977 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) { 3978 btrfs_err(fs_info, 3979 "dev_item UUID does not match fsid: %pU != %pU", 3980 fs_info->fsid, sb->dev_item.fsid); 3981 ret = -EINVAL; 3982 } 3983 3984 /* 3985 * Hint to catch really bogus numbers, bitflips or so, more exact checks are 3986 * done later 3987 */ 3988 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { 3989 btrfs_err(fs_info, "bytes_used is too small %llu", 3990 btrfs_super_bytes_used(sb)); 3991 ret = -EINVAL; 3992 } 3993 if (!is_power_of_2(btrfs_super_stripesize(sb))) { 3994 btrfs_err(fs_info, "invalid stripesize %u", 3995 btrfs_super_stripesize(sb)); 3996 ret = -EINVAL; 3997 } 3998 if (btrfs_super_num_devices(sb) > (1UL << 31)) 3999 btrfs_warn(fs_info, "suspicious number of devices: %llu", 4000 btrfs_super_num_devices(sb)); 4001 if (btrfs_super_num_devices(sb) == 0) { 4002 btrfs_err(fs_info, "number of devices is 0"); 4003 ret = -EINVAL; 4004 } 4005 4006 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) { 4007 btrfs_err(fs_info, "super offset mismatch %llu != %u", 4008 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); 4009 ret = -EINVAL; 4010 } 4011 4012 /* 4013 * Obvious sys_chunk_array corruptions, it must hold at least one key 4014 * and one chunk 4015 */ 4016 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { 4017 btrfs_err(fs_info, "system chunk array too big %u > %u", 4018 btrfs_super_sys_array_size(sb), 4019 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); 4020 ret = -EINVAL; 4021 } 4022 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) 4023 + sizeof(struct btrfs_chunk)) { 4024 btrfs_err(fs_info, "system chunk array too small %u < %zu", 4025 btrfs_super_sys_array_size(sb), 4026 sizeof(struct btrfs_disk_key) 4027 + sizeof(struct btrfs_chunk)); 4028 ret = -EINVAL; 4029 } 4030 4031 /* 4032 * The generation is a global counter, we'll trust it more than the others 4033 * but it's still possible that it's the one that's wrong. 4034 */ 4035 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) 4036 btrfs_warn(fs_info, 4037 "suspicious: generation < chunk_root_generation: %llu < %llu", 4038 btrfs_super_generation(sb), 4039 btrfs_super_chunk_root_generation(sb)); 4040 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) 4041 && btrfs_super_cache_generation(sb) != (u64)-1) 4042 btrfs_warn(fs_info, 4043 "suspicious: generation < cache_generation: %llu < %llu", 4044 btrfs_super_generation(sb), 4045 btrfs_super_cache_generation(sb)); 4046 4047 return ret; 4048 } 4049 4050 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) 4051 { 4052 mutex_lock(&fs_info->cleaner_mutex); 4053 btrfs_run_delayed_iputs(fs_info); 4054 mutex_unlock(&fs_info->cleaner_mutex); 4055 4056 down_write(&fs_info->cleanup_work_sem); 4057 up_write(&fs_info->cleanup_work_sem); 4058 4059 /* cleanup FS via transaction */ 4060 btrfs_cleanup_transaction(fs_info); 4061 } 4062 4063 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 4064 { 4065 struct btrfs_ordered_extent *ordered; 4066 4067 spin_lock(&root->ordered_extent_lock); 4068 /* 4069 * This will just short circuit the ordered completion stuff which will 4070 * make sure the ordered extent gets properly cleaned up. 4071 */ 4072 list_for_each_entry(ordered, &root->ordered_extents, 4073 root_extent_list) 4074 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 4075 spin_unlock(&root->ordered_extent_lock); 4076 } 4077 4078 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 4079 { 4080 struct btrfs_root *root; 4081 struct list_head splice; 4082 4083 INIT_LIST_HEAD(&splice); 4084 4085 spin_lock(&fs_info->ordered_root_lock); 4086 list_splice_init(&fs_info->ordered_roots, &splice); 4087 while (!list_empty(&splice)) { 4088 root = list_first_entry(&splice, struct btrfs_root, 4089 ordered_root); 4090 list_move_tail(&root->ordered_root, 4091 &fs_info->ordered_roots); 4092 4093 spin_unlock(&fs_info->ordered_root_lock); 4094 btrfs_destroy_ordered_extents(root); 4095 4096 cond_resched(); 4097 spin_lock(&fs_info->ordered_root_lock); 4098 } 4099 spin_unlock(&fs_info->ordered_root_lock); 4100 } 4101 4102 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 4103 struct btrfs_fs_info *fs_info) 4104 { 4105 struct rb_node *node; 4106 struct btrfs_delayed_ref_root *delayed_refs; 4107 struct btrfs_delayed_ref_node *ref; 4108 int ret = 0; 4109 4110 delayed_refs = &trans->delayed_refs; 4111 4112 spin_lock(&delayed_refs->lock); 4113 if (atomic_read(&delayed_refs->num_entries) == 0) { 4114 spin_unlock(&delayed_refs->lock); 4115 btrfs_info(fs_info, "delayed_refs has NO entry"); 4116 return ret; 4117 } 4118 4119 while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 4120 struct btrfs_delayed_ref_head *head; 4121 struct rb_node *n; 4122 bool pin_bytes = false; 4123 4124 head = rb_entry(node, struct btrfs_delayed_ref_head, 4125 href_node); 4126 if (!mutex_trylock(&head->mutex)) { 4127 refcount_inc(&head->refs); 4128 spin_unlock(&delayed_refs->lock); 4129 4130 mutex_lock(&head->mutex); 4131 mutex_unlock(&head->mutex); 4132 btrfs_put_delayed_ref_head(head); 4133 spin_lock(&delayed_refs->lock); 4134 continue; 4135 } 4136 spin_lock(&head->lock); 4137 while ((n = rb_first(&head->ref_tree)) != NULL) { 4138 ref = rb_entry(n, struct btrfs_delayed_ref_node, 4139 ref_node); 4140 ref->in_tree = 0; 4141 rb_erase(&ref->ref_node, &head->ref_tree); 4142 RB_CLEAR_NODE(&ref->ref_node); 4143 if (!list_empty(&ref->add_list)) 4144 list_del(&ref->add_list); 4145 atomic_dec(&delayed_refs->num_entries); 4146 btrfs_put_delayed_ref(ref); 4147 } 4148 if (head->must_insert_reserved) 4149 pin_bytes = true; 4150 btrfs_free_delayed_extent_op(head->extent_op); 4151 delayed_refs->num_heads--; 4152 if (head->processing == 0) 4153 delayed_refs->num_heads_ready--; 4154 atomic_dec(&delayed_refs->num_entries); 4155 rb_erase(&head->href_node, &delayed_refs->href_root); 4156 RB_CLEAR_NODE(&head->href_node); 4157 spin_unlock(&head->lock); 4158 spin_unlock(&delayed_refs->lock); 4159 mutex_unlock(&head->mutex); 4160 4161 if (pin_bytes) 4162 btrfs_pin_extent(fs_info, head->bytenr, 4163 head->num_bytes, 1); 4164 btrfs_put_delayed_ref_head(head); 4165 cond_resched(); 4166 spin_lock(&delayed_refs->lock); 4167 } 4168 4169 spin_unlock(&delayed_refs->lock); 4170 4171 return ret; 4172 } 4173 4174 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 4175 { 4176 struct btrfs_inode *btrfs_inode; 4177 struct list_head splice; 4178 4179 INIT_LIST_HEAD(&splice); 4180 4181 spin_lock(&root->delalloc_lock); 4182 list_splice_init(&root->delalloc_inodes, &splice); 4183 4184 while (!list_empty(&splice)) { 4185 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 4186 delalloc_inodes); 4187 4188 list_del_init(&btrfs_inode->delalloc_inodes); 4189 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 4190 &btrfs_inode->runtime_flags); 4191 spin_unlock(&root->delalloc_lock); 4192 4193 btrfs_invalidate_inodes(btrfs_inode->root); 4194 4195 spin_lock(&root->delalloc_lock); 4196 } 4197 4198 spin_unlock(&root->delalloc_lock); 4199 } 4200 4201 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 4202 { 4203 struct btrfs_root *root; 4204 struct list_head splice; 4205 4206 INIT_LIST_HEAD(&splice); 4207 4208 spin_lock(&fs_info->delalloc_root_lock); 4209 list_splice_init(&fs_info->delalloc_roots, &splice); 4210 while (!list_empty(&splice)) { 4211 root = list_first_entry(&splice, struct btrfs_root, 4212 delalloc_root); 4213 list_del_init(&root->delalloc_root); 4214 root = btrfs_grab_fs_root(root); 4215 BUG_ON(!root); 4216 spin_unlock(&fs_info->delalloc_root_lock); 4217 4218 btrfs_destroy_delalloc_inodes(root); 4219 btrfs_put_fs_root(root); 4220 4221 spin_lock(&fs_info->delalloc_root_lock); 4222 } 4223 spin_unlock(&fs_info->delalloc_root_lock); 4224 } 4225 4226 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, 4227 struct extent_io_tree *dirty_pages, 4228 int mark) 4229 { 4230 int ret; 4231 struct extent_buffer *eb; 4232 u64 start = 0; 4233 u64 end; 4234 4235 while (1) { 4236 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4237 mark, NULL); 4238 if (ret) 4239 break; 4240 4241 clear_extent_bits(dirty_pages, start, end, mark); 4242 while (start <= end) { 4243 eb = find_extent_buffer(fs_info, start); 4244 start += fs_info->nodesize; 4245 if (!eb) 4246 continue; 4247 wait_on_extent_buffer_writeback(eb); 4248 4249 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4250 &eb->bflags)) 4251 clear_extent_buffer_dirty(eb); 4252 free_extent_buffer_stale(eb); 4253 } 4254 } 4255 4256 return ret; 4257 } 4258 4259 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, 4260 struct extent_io_tree *pinned_extents) 4261 { 4262 struct extent_io_tree *unpin; 4263 u64 start; 4264 u64 end; 4265 int ret; 4266 bool loop = true; 4267 4268 unpin = pinned_extents; 4269 again: 4270 while (1) { 4271 ret = find_first_extent_bit(unpin, 0, &start, &end, 4272 EXTENT_DIRTY, NULL); 4273 if (ret) 4274 break; 4275 4276 clear_extent_dirty(unpin, start, end); 4277 btrfs_error_unpin_extent_range(fs_info, start, end); 4278 cond_resched(); 4279 } 4280 4281 if (loop) { 4282 if (unpin == &fs_info->freed_extents[0]) 4283 unpin = &fs_info->freed_extents[1]; 4284 else 4285 unpin = &fs_info->freed_extents[0]; 4286 loop = false; 4287 goto again; 4288 } 4289 4290 return 0; 4291 } 4292 4293 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache) 4294 { 4295 struct inode *inode; 4296 4297 inode = cache->io_ctl.inode; 4298 if (inode) { 4299 invalidate_inode_pages2(inode->i_mapping); 4300 BTRFS_I(inode)->generation = 0; 4301 cache->io_ctl.inode = NULL; 4302 iput(inode); 4303 } 4304 btrfs_put_block_group(cache); 4305 } 4306 4307 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, 4308 struct btrfs_fs_info *fs_info) 4309 { 4310 struct btrfs_block_group_cache *cache; 4311 4312 spin_lock(&cur_trans->dirty_bgs_lock); 4313 while (!list_empty(&cur_trans->dirty_bgs)) { 4314 cache = list_first_entry(&cur_trans->dirty_bgs, 4315 struct btrfs_block_group_cache, 4316 dirty_list); 4317 if (!cache) { 4318 btrfs_err(fs_info, "orphan block group dirty_bgs list"); 4319 spin_unlock(&cur_trans->dirty_bgs_lock); 4320 return; 4321 } 4322 4323 if (!list_empty(&cache->io_list)) { 4324 spin_unlock(&cur_trans->dirty_bgs_lock); 4325 list_del_init(&cache->io_list); 4326 btrfs_cleanup_bg_io(cache); 4327 spin_lock(&cur_trans->dirty_bgs_lock); 4328 } 4329 4330 list_del_init(&cache->dirty_list); 4331 spin_lock(&cache->lock); 4332 cache->disk_cache_state = BTRFS_DC_ERROR; 4333 spin_unlock(&cache->lock); 4334 4335 spin_unlock(&cur_trans->dirty_bgs_lock); 4336 btrfs_put_block_group(cache); 4337 spin_lock(&cur_trans->dirty_bgs_lock); 4338 } 4339 spin_unlock(&cur_trans->dirty_bgs_lock); 4340 4341 while (!list_empty(&cur_trans->io_bgs)) { 4342 cache = list_first_entry(&cur_trans->io_bgs, 4343 struct btrfs_block_group_cache, 4344 io_list); 4345 if (!cache) { 4346 btrfs_err(fs_info, "orphan block group on io_bgs list"); 4347 return; 4348 } 4349 4350 list_del_init(&cache->io_list); 4351 spin_lock(&cache->lock); 4352 cache->disk_cache_state = BTRFS_DC_ERROR; 4353 spin_unlock(&cache->lock); 4354 btrfs_cleanup_bg_io(cache); 4355 } 4356 } 4357 4358 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 4359 struct btrfs_fs_info *fs_info) 4360 { 4361 btrfs_cleanup_dirty_bgs(cur_trans, fs_info); 4362 ASSERT(list_empty(&cur_trans->dirty_bgs)); 4363 ASSERT(list_empty(&cur_trans->io_bgs)); 4364 4365 btrfs_destroy_delayed_refs(cur_trans, fs_info); 4366 4367 cur_trans->state = TRANS_STATE_COMMIT_START; 4368 wake_up(&fs_info->transaction_blocked_wait); 4369 4370 cur_trans->state = TRANS_STATE_UNBLOCKED; 4371 wake_up(&fs_info->transaction_wait); 4372 4373 btrfs_destroy_delayed_inodes(fs_info); 4374 btrfs_assert_delayed_root_empty(fs_info); 4375 4376 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, 4377 EXTENT_DIRTY); 4378 btrfs_destroy_pinned_extent(fs_info, 4379 fs_info->pinned_extents); 4380 4381 cur_trans->state =TRANS_STATE_COMPLETED; 4382 wake_up(&cur_trans->commit_wait); 4383 } 4384 4385 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) 4386 { 4387 struct btrfs_transaction *t; 4388 4389 mutex_lock(&fs_info->transaction_kthread_mutex); 4390 4391 spin_lock(&fs_info->trans_lock); 4392 while (!list_empty(&fs_info->trans_list)) { 4393 t = list_first_entry(&fs_info->trans_list, 4394 struct btrfs_transaction, list); 4395 if (t->state >= TRANS_STATE_COMMIT_START) { 4396 refcount_inc(&t->use_count); 4397 spin_unlock(&fs_info->trans_lock); 4398 btrfs_wait_for_commit(fs_info, t->transid); 4399 btrfs_put_transaction(t); 4400 spin_lock(&fs_info->trans_lock); 4401 continue; 4402 } 4403 if (t == fs_info->running_transaction) { 4404 t->state = TRANS_STATE_COMMIT_DOING; 4405 spin_unlock(&fs_info->trans_lock); 4406 /* 4407 * We wait for 0 num_writers since we don't hold a trans 4408 * handle open currently for this transaction. 4409 */ 4410 wait_event(t->writer_wait, 4411 atomic_read(&t->num_writers) == 0); 4412 } else { 4413 spin_unlock(&fs_info->trans_lock); 4414 } 4415 btrfs_cleanup_one_transaction(t, fs_info); 4416 4417 spin_lock(&fs_info->trans_lock); 4418 if (t == fs_info->running_transaction) 4419 fs_info->running_transaction = NULL; 4420 list_del_init(&t->list); 4421 spin_unlock(&fs_info->trans_lock); 4422 4423 btrfs_put_transaction(t); 4424 trace_btrfs_transaction_commit(fs_info->tree_root); 4425 spin_lock(&fs_info->trans_lock); 4426 } 4427 spin_unlock(&fs_info->trans_lock); 4428 btrfs_destroy_all_ordered_extents(fs_info); 4429 btrfs_destroy_delayed_inodes(fs_info); 4430 btrfs_assert_delayed_root_empty(fs_info); 4431 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents); 4432 btrfs_destroy_all_delalloc_inodes(fs_info); 4433 mutex_unlock(&fs_info->transaction_kthread_mutex); 4434 4435 return 0; 4436 } 4437 4438 static struct btrfs_fs_info *btree_fs_info(void *private_data) 4439 { 4440 struct inode *inode = private_data; 4441 return btrfs_sb(inode->i_sb); 4442 } 4443 4444 static const struct extent_io_ops btree_extent_io_ops = { 4445 /* mandatory callbacks */ 4446 .submit_bio_hook = btree_submit_bio_hook, 4447 .readpage_end_io_hook = btree_readpage_end_io_hook, 4448 /* note we're sharing with inode.c for the merge bio hook */ 4449 .merge_bio_hook = btrfs_merge_bio_hook, 4450 .readpage_io_failed_hook = btree_io_failed_hook, 4451 .set_range_writeback = btrfs_set_range_writeback, 4452 .tree_fs_info = btree_fs_info, 4453 4454 /* optional callbacks */ 4455 }; 4456