fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/kthread.h>
  12 #include <linux/slab.h>
  13 #include <linux/migrate.h>
  14 #include <linux/ratelimit.h>
  15 #include <linux/uuid.h>
  16 #include <linux/semaphore.h>
  17 #include <linux/error-injection.h>
  18 #include <linux/crc32c.h>
  19 #include <linux/sched/mm.h>
  20 #include <asm/unaligned.h>
  21 #include <crypto/hash.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "bio.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "rcu-string.h"
  33 #include "dev-replace.h"
  34 #include "raid56.h"
  35 #include "sysfs.h"
  36 #include "qgroup.h"
  37 #include "compression.h"
  38 #include "tree-checker.h"
  39 #include "ref-verify.h"
  40 #include "block-group.h"
  41 #include "discard.h"
  42 #include "space-info.h"
  43 #include "zoned.h"
  44 #include "subpage.h"
  45 #include "fs.h"
  46 #include "accessors.h"
  47 #include "extent-tree.h"
  48 #include "root-tree.h"
  49 #include "defrag.h"
  50 #include "uuid-tree.h"
  51 #include "relocation.h"
  52 #include "scrub.h"
  53 #include "super.h"
  54
  55 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  56                                  BTRFS_HEADER_FLAG_RELOC |\
  57                                  BTRFS_SUPER_FLAG_ERROR |\
  58                                  BTRFS_SUPER_FLAG_SEEDING |\
  59                                  BTRFS_SUPER_FLAG_METADUMP |\
  60                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  61
  62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  64
  65 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  66 {
  67         if (fs_info->csum_shash)
  68                 crypto_free_shash(fs_info->csum_shash);
  69 }
  70
  71 /*
  72  * Compute the csum of a btree block and store the result to provided buffer.
  73  */
  74 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  75 {
  76         struct btrfs_fs_info *fs_info = buf->fs_info;
  77         int num_pages;
  78         u32 first_page_part;
  79         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  80         char *kaddr;
  81         int i;
  82
  83         shash->tfm = fs_info->csum_shash;
  84         crypto_shash_init(shash);
  85
  86         if (buf->addr) {
  87                 /* Pages are contiguous, handle them as a big one. */
  88                 kaddr = buf->addr;
  89                 first_page_part = fs_info->nodesize;
  90                 num_pages = 1;
  91         } else {
  92                 kaddr = folio_address(buf->folios[0]);
  93                 first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  94                 num_pages = num_extent_pages(buf);
  95         }
  96
  97         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
  98                             first_page_part - BTRFS_CSUM_SIZE);
  99
 100         /*
 101          * Multiple single-page folios case would reach here.
 102          *
 103          * nodesize <= PAGE_SIZE and large folio all handled by above
 104          * crypto_shash_update() already.
 105          */
 106         for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
 107                 kaddr = folio_address(buf->folios[i]);
 108                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 109         }
 110         memset(result, 0, BTRFS_CSUM_SIZE);
 111         crypto_shash_final(shash, result);
 112 }
 113
 114 /*
 115  * we can't consider a given block up to date unless the transid of the
 116  * block matches the transid in the parent node's pointer.  This is how we
 117  * detect blocks that either didn't get written at all or got written
 118  * in the wrong place.
 119  */
 120 int btrfs_buffer_uptodate(struct extent_buffer *eb, u64 parent_transid, int atomic)
 121 {
 122         if (!extent_buffer_uptodate(eb))
 123                 return 0;
 124
 125         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 126                 return 1;
 127
 128         if (atomic)
 129                 return -EAGAIN;
 130
 131         if (!extent_buffer_uptodate(eb) ||
 132             btrfs_header_generation(eb) != parent_transid) {
 133                 btrfs_err_rl(eb->fs_info,
 134 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 135                         eb->start, eb->read_mirror,
 136                         parent_transid, btrfs_header_generation(eb));
 137                 clear_extent_buffer_uptodate(eb);
 138                 return 0;
 139         }
 140         return 1;
 141 }
 142
 143 static bool btrfs_supported_super_csum(u16 csum_type)
 144 {
 145         switch (csum_type) {
 146         case BTRFS_CSUM_TYPE_CRC32:
 147         case BTRFS_CSUM_TYPE_XXHASH:
 148         case BTRFS_CSUM_TYPE_SHA256:
 149         case BTRFS_CSUM_TYPE_BLAKE2:
 150                 return true;
 151         default:
 152                 return false;
 153         }
 154 }
 155
 156 /*
 157  * Return 0 if the superblock checksum type matches the checksum value of that
 158  * algorithm. Pass the raw disk superblock data.
 159  */
 160 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 161                            const struct btrfs_super_block *disk_sb)
 162 {
 163         char result[BTRFS_CSUM_SIZE];
 164         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 165
 166         shash->tfm = fs_info->csum_shash;
 167
 168         /*
 169          * The super_block structure does not span the whole
 170          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 171          * filled with zeros and is included in the checksum.
 172          */
 173         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 174                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 175
 176         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 177                 return 1;
 178
 179         return 0;
 180 }
 181
 182 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 183                                       int mirror_num)
 184 {
 185         struct btrfs_fs_info *fs_info = eb->fs_info;
 186         int num_folios = num_extent_folios(eb);
 187         int ret = 0;
 188
 189         if (sb_rdonly(fs_info->sb))
 190                 return -EROFS;
 191
 192         for (int i = 0; i < num_folios; i++) {
 193                 struct folio *folio = eb->folios[i];
 194                 u64 start = max_t(u64, eb->start, folio_pos(folio));
 195                 u64 end = min_t(u64, eb->start + eb->len,
 196                                 folio_pos(folio) + folio_size(folio));
 197                 u32 len = end - start;
 198
 199                 ret = btrfs_repair_io_failure(fs_info, 0, start, len,
 200                                               start, folio, offset_in_folio(folio, start),
 201                                               mirror_num);
 202                 if (ret)
 203                         break;
 204         }
 205
 206         return ret;
 207 }
 208
 209 /*
 210  * helper to read a given tree block, doing retries as required when
 211  * the checksums don't match and we have alternate mirrors to try.
 212  *
 213  * @check:              expected tree parentness check, see the comments of the
 214  *                      structure for details.
 215  */
 216 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 217                              struct btrfs_tree_parent_check *check)
 218 {
 219         struct btrfs_fs_info *fs_info = eb->fs_info;
 220         int failed = 0;
 221         int ret;
 222         int num_copies = 0;
 223         int mirror_num = 0;
 224         int failed_mirror = 0;
 225
 226         ASSERT(check);
 227
 228         while (1) {
 229                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 230                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
 231                 if (!ret)
 232                         break;
 233
 234                 num_copies = btrfs_num_copies(fs_info,
 235                                               eb->start, eb->len);
 236                 if (num_copies == 1)
 237                         break;
 238
 239                 if (!failed_mirror) {
 240                         failed = 1;
 241                         failed_mirror = eb->read_mirror;
 242                 }
 243
 244                 mirror_num++;
 245                 if (mirror_num == failed_mirror)
 246                         mirror_num++;
 247
 248                 if (mirror_num > num_copies)
 249                         break;
 250         }
 251
 252         if (failed && !ret && failed_mirror)
 253                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 254
 255         return ret;
 256 }
 257
 258 /*
 259  * Checksum a dirty tree block before IO.
 260  */
 261 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 262 {
 263         struct extent_buffer *eb = bbio->private;
 264         struct btrfs_fs_info *fs_info = eb->fs_info;
 265         u64 found_start = btrfs_header_bytenr(eb);
 266         u64 last_trans;
 267         u8 result[BTRFS_CSUM_SIZE];
 268         int ret;
 269
 270         /* Btree blocks are always contiguous on disk. */
 271         if (WARN_ON_ONCE(bbio->file_offset != eb->start))
 272                 return BLK_STS_IOERR;
 273         if (WARN_ON_ONCE(bbio->bio.bi_iter.bi_size != eb->len))
 274                 return BLK_STS_IOERR;
 275
 276         /*
 277          * If an extent_buffer is marked as EXTENT_BUFFER_ZONED_ZEROOUT, don't
 278          * checksum it but zero-out its content. This is done to preserve
 279          * ordering of I/O without unnecessarily writing out data.
 280          */
 281         if (test_bit(EXTENT_BUFFER_ZONED_ZEROOUT, &eb->bflags)) {
 282                 memzero_extent_buffer(eb, 0, eb->len);
 283                 return BLK_STS_OK;
 284         }
 285
 286         if (WARN_ON_ONCE(found_start != eb->start))
 287                 return BLK_STS_IOERR;
 288         if (WARN_ON(!btrfs_folio_test_uptodate(fs_info, eb->folios[0],
 289                                                eb->start, eb->len)))
 290                 return BLK_STS_IOERR;
 291
 292         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 293                                     offsetof(struct btrfs_header, fsid),
 294                                     BTRFS_FSID_SIZE) == 0);
 295         csum_tree_block(eb, result);
 296
 297         if (btrfs_header_level(eb))
 298                 ret = btrfs_check_node(eb);
 299         else
 300                 ret = btrfs_check_leaf(eb);
 301
 302         if (ret < 0)
 303                 goto error;
 304
 305         /*
 306          * Also check the generation, the eb reached here must be newer than
 307          * last committed. Or something seriously wrong happened.
 308          */
 309         last_trans = btrfs_get_last_trans_committed(fs_info);
 310         if (unlikely(btrfs_header_generation(eb) <= last_trans)) {
 311                 ret = -EUCLEAN;
 312                 btrfs_err(fs_info,
 313                         "block=%llu bad generation, have %llu expect > %llu",
 314                           eb->start, btrfs_header_generation(eb), last_trans);
 315                 goto error;
 316         }
 317         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 318         return BLK_STS_OK;
 319
 320 error:
 321         btrfs_print_tree(eb, 0);
 322         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 323                   eb->start);
 324         /*
 325          * Be noisy if this is an extent buffer from a log tree. We don't abort
 326          * a transaction in case there's a bad log tree extent buffer, we just
 327          * fallback to a transaction commit. Still we want to know when there is
 328          * a bad log tree extent buffer, as that may signal a bug somewhere.
 329          */
 330         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 331                 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 332         return errno_to_blk_status(ret);
 333 }
 334
 335 static bool check_tree_block_fsid(struct extent_buffer *eb)
 336 {
 337         struct btrfs_fs_info *fs_info = eb->fs_info;
 338         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 339         u8 fsid[BTRFS_FSID_SIZE];
 340
 341         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 342                            BTRFS_FSID_SIZE);
 343
 344         /*
 345          * alloc_fsid_devices() copies the fsid into fs_devices::metadata_uuid.
 346          * This is then overwritten by metadata_uuid if it is present in the
 347          * device_list_add(). The same true for a seed device as well. So use of
 348          * fs_devices::metadata_uuid is appropriate here.
 349          */
 350         if (memcmp(fsid, fs_info->fs_devices->metadata_uuid, BTRFS_FSID_SIZE) == 0)
 351                 return false;
 352
 353         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 354                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 355                         return false;
 356
 357         return true;
 358 }
 359
 360 /* Do basic extent buffer checks at read time */
 361 int btrfs_validate_extent_buffer(struct extent_buffer *eb,
 362                                  struct btrfs_tree_parent_check *check)
 363 {
 364         struct btrfs_fs_info *fs_info = eb->fs_info;
 365         u64 found_start;
 366         const u32 csum_size = fs_info->csum_size;
 367         u8 found_level;
 368         u8 result[BTRFS_CSUM_SIZE];
 369         const u8 *header_csum;
 370         int ret = 0;
 371
 372         ASSERT(check);
 373
 374         found_start = btrfs_header_bytenr(eb);
 375         if (found_start != eb->start) {
 376                 btrfs_err_rl(fs_info,
 377                         "bad tree block start, mirror %u want %llu have %llu",
 378                              eb->read_mirror, eb->start, found_start);
 379                 ret = -EIO;
 380                 goto out;
 381         }
 382         if (check_tree_block_fsid(eb)) {
 383                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 384                              eb->start, eb->read_mirror);
 385                 ret = -EIO;
 386                 goto out;
 387         }
 388         found_level = btrfs_header_level(eb);
 389         if (found_level >= BTRFS_MAX_LEVEL) {
 390                 btrfs_err(fs_info,
 391                         "bad tree block level, mirror %u level %d on logical %llu",
 392                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 393                 ret = -EIO;
 394                 goto out;
 395         }
 396
 397         csum_tree_block(eb, result);
 398         header_csum = folio_address(eb->folios[0]) +
 399                 get_eb_offset_in_folio(eb, offsetof(struct btrfs_header, csum));
 400
 401         if (memcmp(result, header_csum, csum_size) != 0) {
 402                 btrfs_warn_rl(fs_info,
 403 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 404                               eb->start, eb->read_mirror,
 405                               CSUM_FMT_VALUE(csum_size, header_csum),
 406                               CSUM_FMT_VALUE(csum_size, result),
 407                               btrfs_header_level(eb));
 408                 ret = -EUCLEAN;
 409                 goto out;
 410         }
 411
 412         if (found_level != check->level) {
 413                 btrfs_err(fs_info,
 414                 "level verify failed on logical %llu mirror %u wanted %u found %u",
 415                           eb->start, eb->read_mirror, check->level, found_level);
 416                 ret = -EIO;
 417                 goto out;
 418         }
 419         if (unlikely(check->transid &&
 420                      btrfs_header_generation(eb) != check->transid)) {
 421                 btrfs_err_rl(eb->fs_info,
 422 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 423                                 eb->start, eb->read_mirror, check->transid,
 424                                 btrfs_header_generation(eb));
 425                 ret = -EIO;
 426                 goto out;
 427         }
 428         if (check->has_first_key) {
 429                 struct btrfs_key *expect_key = &check->first_key;
 430                 struct btrfs_key found_key;
 431
 432                 if (found_level)
 433                         btrfs_node_key_to_cpu(eb, &found_key, 0);
 434                 else
 435                         btrfs_item_key_to_cpu(eb, &found_key, 0);
 436                 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 437                         btrfs_err(fs_info,
 438 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 439                                   eb->start, check->transid,
 440                                   expect_key->objectid,
 441                                   expect_key->type, expect_key->offset,
 442                                   found_key.objectid, found_key.type,
 443                                   found_key.offset);
 444                         ret = -EUCLEAN;
 445                         goto out;
 446                 }
 447         }
 448         if (check->owner_root) {
 449                 ret = btrfs_check_eb_owner(eb, check->owner_root);
 450                 if (ret < 0)
 451                         goto out;
 452         }
 453
 454         /*
 455          * If this is a leaf block and it is corrupt, set the corrupt bit so
 456          * that we don't try and read the other copies of this block, just
 457          * return -EIO.
 458          */
 459         if (found_level == 0 && btrfs_check_leaf(eb)) {
 460                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 461                 ret = -EIO;
 462         }
 463
 464         if (found_level > 0 && btrfs_check_node(eb))
 465                 ret = -EIO;
 466
 467         if (ret)
 468                 btrfs_err(fs_info,
 469                 "read time tree block corruption detected on logical %llu mirror %u",
 470                           eb->start, eb->read_mirror);
 471 out:
 472         return ret;
 473 }
 474
 475 #ifdef CONFIG_MIGRATION
 476 static int btree_migrate_folio(struct address_space *mapping,
 477                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 478 {
 479         /*
 480          * we can't safely write a btree page from here,
 481          * we haven't done the locking hook
 482          */
 483         if (folio_test_dirty(src))
 484                 return -EAGAIN;
 485         /*
 486          * Buffers may be managed in a filesystem specific way.
 487          * We must have no buffers or drop them.
 488          */
 489         if (folio_get_private(src) &&
 490             !filemap_release_folio(src, GFP_KERNEL))
 491                 return -EAGAIN;
 492         return migrate_folio(mapping, dst, src, mode);
 493 }
 494 #else
 495 #define btree_migrate_folio NULL
 496 #endif
 497
 498 static int btree_writepages(struct address_space *mapping,
 499                             struct writeback_control *wbc)
 500 {
 501         struct btrfs_fs_info *fs_info;
 502         int ret;
 503
 504         if (wbc->sync_mode == WB_SYNC_NONE) {
 505
 506                 if (wbc->for_kupdate)
 507                         return 0;
 508
 509                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 510                 /* this is a bit racy, but that's ok */
 511                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 512                                              BTRFS_DIRTY_METADATA_THRESH,
 513                                              fs_info->dirty_metadata_batch);
 514                 if (ret < 0)
 515                         return 0;
 516         }
 517         return btree_write_cache_pages(mapping, wbc);
 518 }
 519
 520 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 521 {
 522         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 523                 return false;
 524
 525         return try_release_extent_buffer(&folio->page);
 526 }
 527
 528 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 529                                  size_t length)
 530 {
 531         struct extent_io_tree *tree;
 532         tree = &BTRFS_I(folio->mapping->host)->io_tree;
 533         extent_invalidate_folio(tree, folio, offset);
 534         btree_release_folio(folio, GFP_NOFS);
 535         if (folio_get_private(folio)) {
 536                 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 537                            "folio private not zero on folio %llu",
 538                            (unsigned long long)folio_pos(folio));
 539                 folio_detach_private(folio);
 540         }
 541 }
 542
 543 #ifdef DEBUG
 544 static bool btree_dirty_folio(struct address_space *mapping,
 545                 struct folio *folio)
 546 {
 547         struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 548         struct btrfs_subpage_info *spi = fs_info->subpage_info;
 549         struct btrfs_subpage *subpage;
 550         struct extent_buffer *eb;
 551         int cur_bit = 0;
 552         u64 page_start = folio_pos(folio);
 553
 554         if (fs_info->sectorsize == PAGE_SIZE) {
 555                 eb = folio_get_private(folio);
 556                 BUG_ON(!eb);
 557                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 558                 BUG_ON(!atomic_read(&eb->refs));
 559                 btrfs_assert_tree_write_locked(eb);
 560                 return filemap_dirty_folio(mapping, folio);
 561         }
 562
 563         ASSERT(spi);
 564         subpage = folio_get_private(folio);
 565
 566         for (cur_bit = spi->dirty_offset;
 567              cur_bit < spi->dirty_offset + spi->bitmap_nr_bits;
 568              cur_bit++) {
 569                 unsigned long flags;
 570                 u64 cur;
 571
 572                 spin_lock_irqsave(&subpage->lock, flags);
 573                 if (!test_bit(cur_bit, subpage->bitmaps)) {
 574                         spin_unlock_irqrestore(&subpage->lock, flags);
 575                         continue;
 576                 }
 577                 spin_unlock_irqrestore(&subpage->lock, flags);
 578                 cur = page_start + cur_bit * fs_info->sectorsize;
 579
 580                 eb = find_extent_buffer(fs_info, cur);
 581                 ASSERT(eb);
 582                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 583                 ASSERT(atomic_read(&eb->refs));
 584                 btrfs_assert_tree_write_locked(eb);
 585                 free_extent_buffer(eb);
 586
 587                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits) - 1;
 588         }
 589         return filemap_dirty_folio(mapping, folio);
 590 }
 591 #else
 592 #define btree_dirty_folio filemap_dirty_folio
 593 #endif
 594
 595 static const struct address_space_operations btree_aops = {
 596         .writepages     = btree_writepages,
 597         .release_folio  = btree_release_folio,
 598         .invalidate_folio = btree_invalidate_folio,
 599         .migrate_folio  = btree_migrate_folio,
 600         .dirty_folio    = btree_dirty_folio,
 601 };
 602
 603 struct extent_buffer *btrfs_find_create_tree_block(
 604                                                 struct btrfs_fs_info *fs_info,
 605                                                 u64 bytenr, u64 owner_root,
 606                                                 int level)
 607 {
 608         if (btrfs_is_testing(fs_info))
 609                 return alloc_test_extent_buffer(fs_info, bytenr);
 610         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 611 }
 612
 613 /*
 614  * Read tree block at logical address @bytenr and do variant basic but critical
 615  * verification.
 616  *
 617  * @check:              expected tree parentness check, see comments of the
 618  *                      structure for details.
 619  */
 620 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 621                                       struct btrfs_tree_parent_check *check)
 622 {
 623         struct extent_buffer *buf = NULL;
 624         int ret;
 625
 626         ASSERT(check);
 627
 628         buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 629                                            check->level);
 630         if (IS_ERR(buf))
 631                 return buf;
 632
 633         ret = btrfs_read_extent_buffer(buf, check);
 634         if (ret) {
 635                 free_extent_buffer_stale(buf);
 636                 return ERR_PTR(ret);
 637         }
 638         if (btrfs_check_eb_owner(buf, check->owner_root)) {
 639                 free_extent_buffer_stale(buf);
 640                 return ERR_PTR(-EUCLEAN);
 641         }
 642         return buf;
 643
 644 }
 645
 646 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 647                          u64 objectid)
 648 {
 649         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 650
 651         memset(&root->root_key, 0, sizeof(root->root_key));
 652         memset(&root->root_item, 0, sizeof(root->root_item));
 653         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 654         root->fs_info = fs_info;
 655         root->root_key.objectid = objectid;
 656         root->node = NULL;
 657         root->commit_root = NULL;
 658         root->state = 0;
 659         RB_CLEAR_NODE(&root->rb_node);
 660
 661         root->last_trans = 0;
 662         root->free_objectid = 0;
 663         root->nr_delalloc_inodes = 0;
 664         root->nr_ordered_extents = 0;
 665         root->inode_tree = RB_ROOT;
 666         /* GFP flags are compatible with XA_FLAGS_*. */
 667         xa_init_flags(&root->delayed_nodes, GFP_ATOMIC);
 668
 669         btrfs_init_root_block_rsv(root);
 670
 671         INIT_LIST_HEAD(&root->dirty_list);
 672         INIT_LIST_HEAD(&root->root_list);
 673         INIT_LIST_HEAD(&root->delalloc_inodes);
 674         INIT_LIST_HEAD(&root->delalloc_root);
 675         INIT_LIST_HEAD(&root->ordered_extents);
 676         INIT_LIST_HEAD(&root->ordered_root);
 677         INIT_LIST_HEAD(&root->reloc_dirty_list);
 678         spin_lock_init(&root->inode_lock);
 679         spin_lock_init(&root->delalloc_lock);
 680         spin_lock_init(&root->ordered_extent_lock);
 681         spin_lock_init(&root->accounting_lock);
 682         spin_lock_init(&root->qgroup_meta_rsv_lock);
 683         mutex_init(&root->objectid_mutex);
 684         mutex_init(&root->log_mutex);
 685         mutex_init(&root->ordered_extent_mutex);
 686         mutex_init(&root->delalloc_mutex);
 687         init_waitqueue_head(&root->qgroup_flush_wait);
 688         init_waitqueue_head(&root->log_writer_wait);
 689         init_waitqueue_head(&root->log_commit_wait[0]);
 690         init_waitqueue_head(&root->log_commit_wait[1]);
 691         INIT_LIST_HEAD(&root->log_ctxs[0]);
 692         INIT_LIST_HEAD(&root->log_ctxs[1]);
 693         atomic_set(&root->log_commit[0], 0);
 694         atomic_set(&root->log_commit[1], 0);
 695         atomic_set(&root->log_writers, 0);
 696         atomic_set(&root->log_batch, 0);
 697         refcount_set(&root->refs, 1);
 698         atomic_set(&root->snapshot_force_cow, 0);
 699         atomic_set(&root->nr_swapfiles, 0);
 700         btrfs_set_root_log_transid(root, 0);
 701         root->log_transid_committed = -1;
 702         btrfs_set_root_last_log_commit(root, 0);
 703         root->anon_dev = 0;
 704         if (!dummy) {
 705                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
 706                                     IO_TREE_ROOT_DIRTY_LOG_PAGES);
 707                 extent_io_tree_init(fs_info, &root->log_csum_range,
 708                                     IO_TREE_LOG_CSUM_RANGE);
 709         }
 710
 711         spin_lock_init(&root->root_item_lock);
 712         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 713 #ifdef CONFIG_BTRFS_DEBUG
 714         INIT_LIST_HEAD(&root->leak_list);
 715         spin_lock(&fs_info->fs_roots_radix_lock);
 716         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 717         spin_unlock(&fs_info->fs_roots_radix_lock);
 718 #endif
 719 }
 720
 721 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 722                                            u64 objectid, gfp_t flags)
 723 {
 724         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
 725         if (root)
 726                 __setup_root(root, fs_info, objectid);
 727         return root;
 728 }
 729
 730 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 731 /* Should only be used by the testing infrastructure */
 732 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 733 {
 734         struct btrfs_root *root;
 735
 736         if (!fs_info)
 737                 return ERR_PTR(-EINVAL);
 738
 739         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 740         if (!root)
 741                 return ERR_PTR(-ENOMEM);
 742
 743         /* We don't use the stripesize in selftest, set it as sectorsize */
 744         root->alloc_bytenr = 0;
 745
 746         return root;
 747 }
 748 #endif
 749
 750 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 751 {
 752         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 753         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 754
 755         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 756 }
 757
 758 static int global_root_key_cmp(const void *k, const struct rb_node *node)
 759 {
 760         const struct btrfs_key *key = k;
 761         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 762
 763         return btrfs_comp_cpu_keys(key, &root->root_key);
 764 }
 765
 766 int btrfs_global_root_insert(struct btrfs_root *root)
 767 {
 768         struct btrfs_fs_info *fs_info = root->fs_info;
 769         struct rb_node *tmp;
 770         int ret = 0;
 771
 772         write_lock(&fs_info->global_root_lock);
 773         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
 774         write_unlock(&fs_info->global_root_lock);
 775
 776         if (tmp) {
 777                 ret = -EEXIST;
 778                 btrfs_warn(fs_info, "global root %llu %llu already exists",
 779                                 root->root_key.objectid, root->root_key.offset);
 780         }
 781         return ret;
 782 }
 783
 784 void btrfs_global_root_delete(struct btrfs_root *root)
 785 {
 786         struct btrfs_fs_info *fs_info = root->fs_info;
 787
 788         write_lock(&fs_info->global_root_lock);
 789         rb_erase(&root->rb_node, &fs_info->global_root_tree);
 790         write_unlock(&fs_info->global_root_lock);
 791 }
 792
 793 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 794                                      struct btrfs_key *key)
 795 {
 796         struct rb_node *node;
 797         struct btrfs_root *root = NULL;
 798
 799         read_lock(&fs_info->global_root_lock);
 800         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
 801         if (node)
 802                 root = container_of(node, struct btrfs_root, rb_node);
 803         read_unlock(&fs_info->global_root_lock);
 804
 805         return root;
 806 }
 807
 808 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
 809 {
 810         struct btrfs_block_group *block_group;
 811         u64 ret;
 812
 813         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
 814                 return 0;
 815
 816         if (bytenr)
 817                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
 818         else
 819                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
 820         ASSERT(block_group);
 821         if (!block_group)
 822                 return 0;
 823         ret = block_group->global_root_id;
 824         btrfs_put_block_group(block_group);
 825
 826         return ret;
 827 }
 828
 829 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 830 {
 831         struct btrfs_key key = {
 832                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
 833                 .type = BTRFS_ROOT_ITEM_KEY,
 834                 .offset = btrfs_global_root_id(fs_info, bytenr),
 835         };
 836
 837         return btrfs_global_root(fs_info, &key);
 838 }
 839
 840 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 841 {
 842         struct btrfs_key key = {
 843                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
 844                 .type = BTRFS_ROOT_ITEM_KEY,
 845                 .offset = btrfs_global_root_id(fs_info, bytenr),
 846         };
 847
 848         return btrfs_global_root(fs_info, &key);
 849 }
 850
 851 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
 852 {
 853         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
 854                 return fs_info->block_group_root;
 855         return btrfs_extent_root(fs_info, 0);
 856 }
 857
 858 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 859                                      u64 objectid)
 860 {
 861         struct btrfs_fs_info *fs_info = trans->fs_info;
 862         struct extent_buffer *leaf;
 863         struct btrfs_root *tree_root = fs_info->tree_root;
 864         struct btrfs_root *root;
 865         struct btrfs_key key;
 866         unsigned int nofs_flag;
 867         int ret = 0;
 868
 869         /*
 870          * We're holding a transaction handle, so use a NOFS memory allocation
 871          * context to avoid deadlock if reclaim happens.
 872          */
 873         nofs_flag = memalloc_nofs_save();
 874         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
 875         memalloc_nofs_restore(nofs_flag);
 876         if (!root)
 877                 return ERR_PTR(-ENOMEM);
 878
 879         root->root_key.objectid = objectid;
 880         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 881         root->root_key.offset = 0;
 882
 883         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
 884                                       0, BTRFS_NESTING_NORMAL);
 885         if (IS_ERR(leaf)) {
 886                 ret = PTR_ERR(leaf);
 887                 leaf = NULL;
 888                 goto fail;
 889         }
 890
 891         root->node = leaf;
 892         btrfs_mark_buffer_dirty(trans, leaf);
 893
 894         root->commit_root = btrfs_root_node(root);
 895         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
 896
 897         btrfs_set_root_flags(&root->root_item, 0);
 898         btrfs_set_root_limit(&root->root_item, 0);
 899         btrfs_set_root_bytenr(&root->root_item, leaf->start);
 900         btrfs_set_root_generation(&root->root_item, trans->transid);
 901         btrfs_set_root_level(&root->root_item, 0);
 902         btrfs_set_root_refs(&root->root_item, 1);
 903         btrfs_set_root_used(&root->root_item, leaf->len);
 904         btrfs_set_root_last_snapshot(&root->root_item, 0);
 905         btrfs_set_root_dirid(&root->root_item, 0);
 906         if (is_fstree(objectid))
 907                 generate_random_guid(root->root_item.uuid);
 908         else
 909                 export_guid(root->root_item.uuid, &guid_null);
 910         btrfs_set_root_drop_level(&root->root_item, 0);
 911
 912         btrfs_tree_unlock(leaf);
 913
 914         key.objectid = objectid;
 915         key.type = BTRFS_ROOT_ITEM_KEY;
 916         key.offset = 0;
 917         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
 918         if (ret)
 919                 goto fail;
 920
 921         return root;
 922
 923 fail:
 924         btrfs_put_root(root);
 925
 926         return ERR_PTR(ret);
 927 }
 928
 929 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 930                                          struct btrfs_fs_info *fs_info)
 931 {
 932         struct btrfs_root *root;
 933
 934         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
 935         if (!root)
 936                 return ERR_PTR(-ENOMEM);
 937
 938         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
 939         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
 940         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
 941
 942         return root;
 943 }
 944
 945 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
 946                               struct btrfs_root *root)
 947 {
 948         struct extent_buffer *leaf;
 949
 950         /*
 951          * DON'T set SHAREABLE bit for log trees.
 952          *
 953          * Log trees are not exposed to user space thus can't be snapshotted,
 954          * and they go away before a real commit is actually done.
 955          *
 956          * They do store pointers to file data extents, and those reference
 957          * counts still get updated (along with back refs to the log tree).
 958          */
 959
 960         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
 961                         NULL, 0, 0, 0, 0, BTRFS_NESTING_NORMAL);
 962         if (IS_ERR(leaf))
 963                 return PTR_ERR(leaf);
 964
 965         root->node = leaf;
 966
 967         btrfs_mark_buffer_dirty(trans, root->node);
 968         btrfs_tree_unlock(root->node);
 969
 970         return 0;
 971 }
 972
 973 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
 974                              struct btrfs_fs_info *fs_info)
 975 {
 976         struct btrfs_root *log_root;
 977
 978         log_root = alloc_log_tree(trans, fs_info);
 979         if (IS_ERR(log_root))
 980                 return PTR_ERR(log_root);
 981
 982         if (!btrfs_is_zoned(fs_info)) {
 983                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
 984
 985                 if (ret) {
 986                         btrfs_put_root(log_root);
 987                         return ret;
 988                 }
 989         }
 990
 991         WARN_ON(fs_info->log_root_tree);
 992         fs_info->log_root_tree = log_root;
 993         return 0;
 994 }
 995
 996 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
 997                        struct btrfs_root *root)
 998 {
 999         struct btrfs_fs_info *fs_info = root->fs_info;
1000         struct btrfs_root *log_root;
1001         struct btrfs_inode_item *inode_item;
1002         int ret;
1003
1004         log_root = alloc_log_tree(trans, fs_info);
1005         if (IS_ERR(log_root))
1006                 return PTR_ERR(log_root);
1007
1008         ret = btrfs_alloc_log_tree_node(trans, log_root);
1009         if (ret) {
1010                 btrfs_put_root(log_root);
1011                 return ret;
1012         }
1013
1014         log_root->last_trans = trans->transid;
1015         log_root->root_key.offset = root->root_key.objectid;
1016
1017         inode_item = &log_root->root_item.inode;
1018         btrfs_set_stack_inode_generation(inode_item, 1);
1019         btrfs_set_stack_inode_size(inode_item, 3);
1020         btrfs_set_stack_inode_nlink(inode_item, 1);
1021         btrfs_set_stack_inode_nbytes(inode_item,
1022                                      fs_info->nodesize);
1023         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1024
1025         btrfs_set_root_node(&log_root->root_item, log_root->node);
1026
1027         WARN_ON(root->log_root);
1028         root->log_root = log_root;
1029         btrfs_set_root_log_transid(root, 0);
1030         root->log_transid_committed = -1;
1031         btrfs_set_root_last_log_commit(root, 0);
1032         return 0;
1033 }
1034
1035 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1036                                               struct btrfs_path *path,
1037                                               struct btrfs_key *key)
1038 {
1039         struct btrfs_root *root;
1040         struct btrfs_tree_parent_check check = { 0 };
1041         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1042         u64 generation;
1043         int ret;
1044         int level;
1045
1046         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1047         if (!root)
1048                 return ERR_PTR(-ENOMEM);
1049
1050         ret = btrfs_find_root(tree_root, key, path,
1051                               &root->root_item, &root->root_key);
1052         if (ret) {
1053                 if (ret > 0)
1054                         ret = -ENOENT;
1055                 goto fail;
1056         }
1057
1058         generation = btrfs_root_generation(&root->root_item);
1059         level = btrfs_root_level(&root->root_item);
1060         check.level = level;
1061         check.transid = generation;
1062         check.owner_root = key->objectid;
1063         root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1064                                      &check);
1065         if (IS_ERR(root->node)) {
1066                 ret = PTR_ERR(root->node);
1067                 root->node = NULL;
1068                 goto fail;
1069         }
1070         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1071                 ret = -EIO;
1072                 goto fail;
1073         }
1074
1075         /*
1076          * For real fs, and not log/reloc trees, root owner must
1077          * match its root node owner
1078          */
1079         if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1080             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1081             root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1082             root->root_key.objectid != btrfs_header_owner(root->node)) {
1083                 btrfs_crit(fs_info,
1084 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1085                            root->root_key.objectid, root->node->start,
1086                            btrfs_header_owner(root->node),
1087                            root->root_key.objectid);
1088                 ret = -EUCLEAN;
1089                 goto fail;
1090         }
1091         root->commit_root = btrfs_root_node(root);
1092         return root;
1093 fail:
1094         btrfs_put_root(root);
1095         return ERR_PTR(ret);
1096 }
1097
1098 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1099                                         struct btrfs_key *key)
1100 {
1101         struct btrfs_root *root;
1102         struct btrfs_path *path;
1103
1104         path = btrfs_alloc_path();
1105         if (!path)
1106                 return ERR_PTR(-ENOMEM);
1107         root = read_tree_root_path(tree_root, path, key);
1108         btrfs_free_path(path);
1109
1110         return root;
1111 }
1112
1113 /*
1114  * Initialize subvolume root in-memory structure
1115  *
1116  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1117  */
1118 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1119 {
1120         int ret;
1121
1122         btrfs_drew_lock_init(&root->snapshot_lock);
1123
1124         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1125             !btrfs_is_data_reloc_root(root) &&
1126             is_fstree(root->root_key.objectid)) {
1127                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1128                 btrfs_check_and_init_root_item(&root->root_item);
1129         }
1130
1131         /*
1132          * Don't assign anonymous block device to roots that are not exposed to
1133          * userspace, the id pool is limited to 1M
1134          */
1135         if (is_fstree(root->root_key.objectid) &&
1136             btrfs_root_refs(&root->root_item) > 0) {
1137                 if (!anon_dev) {
1138                         ret = get_anon_bdev(&root->anon_dev);
1139                         if (ret)
1140                                 goto fail;
1141                 } else {
1142                         root->anon_dev = anon_dev;
1143                 }
1144         }
1145
1146         mutex_lock(&root->objectid_mutex);
1147         ret = btrfs_init_root_free_objectid(root);
1148         if (ret) {
1149                 mutex_unlock(&root->objectid_mutex);
1150                 goto fail;
1151         }
1152
1153         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1154
1155         mutex_unlock(&root->objectid_mutex);
1156
1157         return 0;
1158 fail:
1159         /* The caller is responsible to call btrfs_free_fs_root */
1160         return ret;
1161 }
1162
1163 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1164                                                u64 root_id)
1165 {
1166         struct btrfs_root *root;
1167
1168         spin_lock(&fs_info->fs_roots_radix_lock);
1169         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170                                  (unsigned long)root_id);
1171         root = btrfs_grab_root(root);
1172         spin_unlock(&fs_info->fs_roots_radix_lock);
1173         return root;
1174 }
1175
1176 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1177                                                 u64 objectid)
1178 {
1179         struct btrfs_key key = {
1180                 .objectid = objectid,
1181                 .type = BTRFS_ROOT_ITEM_KEY,
1182                 .offset = 0,
1183         };
1184
1185         switch (objectid) {
1186         case BTRFS_ROOT_TREE_OBJECTID:
1187                 return btrfs_grab_root(fs_info->tree_root);
1188         case BTRFS_EXTENT_TREE_OBJECTID:
1189                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1190         case BTRFS_CHUNK_TREE_OBJECTID:
1191                 return btrfs_grab_root(fs_info->chunk_root);
1192         case BTRFS_DEV_TREE_OBJECTID:
1193                 return btrfs_grab_root(fs_info->dev_root);
1194         case BTRFS_CSUM_TREE_OBJECTID:
1195                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1196         case BTRFS_QUOTA_TREE_OBJECTID:
1197                 return btrfs_grab_root(fs_info->quota_root);
1198         case BTRFS_UUID_TREE_OBJECTID:
1199                 return btrfs_grab_root(fs_info->uuid_root);
1200         case BTRFS_BLOCK_GROUP_TREE_OBJECTID:
1201                 return btrfs_grab_root(fs_info->block_group_root);
1202         case BTRFS_FREE_SPACE_TREE_OBJECTID:
1203                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1204         case BTRFS_RAID_STRIPE_TREE_OBJECTID:
1205                 return btrfs_grab_root(fs_info->stripe_root);
1206         default:
1207                 return NULL;
1208         }
1209 }
1210
1211 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1212                          struct btrfs_root *root)
1213 {
1214         int ret;
1215
1216         ret = radix_tree_preload(GFP_NOFS);
1217         if (ret)
1218                 return ret;
1219
1220         spin_lock(&fs_info->fs_roots_radix_lock);
1221         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222                                 (unsigned long)root->root_key.objectid,
1223                                 root);
1224         if (ret == 0) {
1225                 btrfs_grab_root(root);
1226                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1227         }
1228         spin_unlock(&fs_info->fs_roots_radix_lock);
1229         radix_tree_preload_end();
1230
1231         return ret;
1232 }
1233
1234 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1235 {
1236 #ifdef CONFIG_BTRFS_DEBUG
1237         struct btrfs_root *root;
1238
1239         while (!list_empty(&fs_info->allocated_roots)) {
1240                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1241
1242                 root = list_first_entry(&fs_info->allocated_roots,
1243                                         struct btrfs_root, leak_list);
1244                 btrfs_err(fs_info, "leaked root %s refcount %d",
1245                           btrfs_root_name(&root->root_key, buf),
1246                           refcount_read(&root->refs));
1247                 while (refcount_read(&root->refs) > 1)
1248                         btrfs_put_root(root);
1249                 btrfs_put_root(root);
1250         }
1251 #endif
1252 }
1253
1254 static void free_global_roots(struct btrfs_fs_info *fs_info)
1255 {
1256         struct btrfs_root *root;
1257         struct rb_node *node;
1258
1259         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1260                 root = rb_entry(node, struct btrfs_root, rb_node);
1261                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1262                 btrfs_put_root(root);
1263         }
1264 }
1265
1266 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1267 {
1268         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1269         percpu_counter_destroy(&fs_info->delalloc_bytes);
1270         percpu_counter_destroy(&fs_info->ordered_bytes);
1271         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1272         btrfs_free_csum_hash(fs_info);
1273         btrfs_free_stripe_hash_table(fs_info);
1274         btrfs_free_ref_cache(fs_info);
1275         kfree(fs_info->balance_ctl);
1276         kfree(fs_info->delayed_root);
1277         free_global_roots(fs_info);
1278         btrfs_put_root(fs_info->tree_root);
1279         btrfs_put_root(fs_info->chunk_root);
1280         btrfs_put_root(fs_info->dev_root);
1281         btrfs_put_root(fs_info->quota_root);
1282         btrfs_put_root(fs_info->uuid_root);
1283         btrfs_put_root(fs_info->fs_root);
1284         btrfs_put_root(fs_info->data_reloc_root);
1285         btrfs_put_root(fs_info->block_group_root);
1286         btrfs_put_root(fs_info->stripe_root);
1287         btrfs_check_leaked_roots(fs_info);
1288         btrfs_extent_buffer_leak_debug_check(fs_info);
1289         kfree(fs_info->super_copy);
1290         kfree(fs_info->super_for_commit);
1291         kfree(fs_info->subpage_info);
1292         kvfree(fs_info);
1293 }
1294
1295
1296 /*
1297  * Get an in-memory reference of a root structure.
1298  *
1299  * For essential trees like root/extent tree, we grab it from fs_info directly.
1300  * For subvolume trees, we check the cached filesystem roots first. If not
1301  * found, then read it from disk and add it to cached fs roots.
1302  *
1303  * Caller should release the root by calling btrfs_put_root() after the usage.
1304  *
1305  * NOTE: Reloc and log trees can't be read by this function as they share the
1306  *       same root objectid.
1307  *
1308  * @objectid:   root id
1309  * @anon_dev:   preallocated anonymous block device number for new roots,
1310  *              pass 0 for new allocation.
1311  * @check_ref:  whether to check root item references, If true, return -ENOENT
1312  *              for orphan roots
1313  */
1314 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1315                                              u64 objectid, dev_t anon_dev,
1316                                              bool check_ref)
1317 {
1318         struct btrfs_root *root;
1319         struct btrfs_path *path;
1320         struct btrfs_key key;
1321         int ret;
1322
1323         root = btrfs_get_global_root(fs_info, objectid);
1324         if (root)
1325                 return root;
1326
1327         /*
1328          * If we're called for non-subvolume trees, and above function didn't
1329          * find one, do not try to read it from disk.
1330          *
1331          * This is namely for free-space-tree and quota tree, which can change
1332          * at runtime and should only be grabbed from fs_info.
1333          */
1334         if (!is_fstree(objectid) && objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1335                 return ERR_PTR(-ENOENT);
1336 again:
1337         root = btrfs_lookup_fs_root(fs_info, objectid);
1338         if (root) {
1339                 /* Shouldn't get preallocated anon_dev for cached roots */
1340                 ASSERT(!anon_dev);
1341                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1342                         btrfs_put_root(root);
1343                         return ERR_PTR(-ENOENT);
1344                 }
1345                 return root;
1346         }
1347
1348         key.objectid = objectid;
1349         key.type = BTRFS_ROOT_ITEM_KEY;
1350         key.offset = (u64)-1;
1351         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1352         if (IS_ERR(root))
1353                 return root;
1354
1355         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1356                 ret = -ENOENT;
1357                 goto fail;
1358         }
1359
1360         ret = btrfs_init_fs_root(root, anon_dev);
1361         if (ret)
1362                 goto fail;
1363
1364         path = btrfs_alloc_path();
1365         if (!path) {
1366                 ret = -ENOMEM;
1367                 goto fail;
1368         }
1369         key.objectid = BTRFS_ORPHAN_OBJECTID;
1370         key.type = BTRFS_ORPHAN_ITEM_KEY;
1371         key.offset = objectid;
1372
1373         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1374         btrfs_free_path(path);
1375         if (ret < 0)
1376                 goto fail;
1377         if (ret == 0)
1378                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1379
1380         ret = btrfs_insert_fs_root(fs_info, root);
1381         if (ret) {
1382                 if (ret == -EEXIST) {
1383                         btrfs_put_root(root);
1384                         goto again;
1385                 }
1386                 goto fail;
1387         }
1388         return root;
1389 fail:
1390         /*
1391          * If our caller provided us an anonymous device, then it's his
1392          * responsibility to free it in case we fail. So we have to set our
1393          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1394          * and once again by our caller.
1395          */
1396         if (anon_dev)
1397                 root->anon_dev = 0;
1398         btrfs_put_root(root);
1399         return ERR_PTR(ret);
1400 }
1401
1402 /*
1403  * Get in-memory reference of a root structure
1404  *
1405  * @objectid:   tree objectid
1406  * @check_ref:  if set, verify that the tree exists and the item has at least
1407  *              one reference
1408  */
1409 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1410                                      u64 objectid, bool check_ref)
1411 {
1412         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1413 }
1414
1415 /*
1416  * Get in-memory reference of a root structure, created as new, optionally pass
1417  * the anonymous block device id
1418  *
1419  * @objectid:   tree objectid
1420  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1421  *              parameter value
1422  */
1423 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1424                                          u64 objectid, dev_t anon_dev)
1425 {
1426         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1427 }
1428
1429 /*
1430  * Return a root for the given objectid.
1431  *
1432  * @fs_info:    the fs_info
1433  * @objectid:   the objectid we need to lookup
1434  *
1435  * This is exclusively used for backref walking, and exists specifically because
1436  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1437  * creation time, which means we may have to read the tree_root in order to look
1438  * up a fs root that is not in memory.  If the root is not in memory we will
1439  * read the tree root commit root and look up the fs root from there.  This is a
1440  * temporary root, it will not be inserted into the radix tree as it doesn't
1441  * have the most uptodate information, it'll simply be discarded once the
1442  * backref code is finished using the root.
1443  */
1444 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1445                                                  struct btrfs_path *path,
1446                                                  u64 objectid)
1447 {
1448         struct btrfs_root *root;
1449         struct btrfs_key key;
1450
1451         ASSERT(path->search_commit_root && path->skip_locking);
1452
1453         /*
1454          * This can return -ENOENT if we ask for a root that doesn't exist, but
1455          * since this is called via the backref walking code we won't be looking
1456          * up a root that doesn't exist, unless there's corruption.  So if root
1457          * != NULL just return it.
1458          */
1459         root = btrfs_get_global_root(fs_info, objectid);
1460         if (root)
1461                 return root;
1462
1463         root = btrfs_lookup_fs_root(fs_info, objectid);
1464         if (root)
1465                 return root;
1466
1467         key.objectid = objectid;
1468         key.type = BTRFS_ROOT_ITEM_KEY;
1469         key.offset = (u64)-1;
1470         root = read_tree_root_path(fs_info->tree_root, path, &key);
1471         btrfs_release_path(path);
1472
1473         return root;
1474 }
1475
1476 static int cleaner_kthread(void *arg)
1477 {
1478         struct btrfs_fs_info *fs_info = arg;
1479         int again;
1480
1481         while (1) {
1482                 again = 0;
1483
1484                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1485
1486                 /* Make the cleaner go to sleep early. */
1487                 if (btrfs_need_cleaner_sleep(fs_info))
1488                         goto sleep;
1489
1490                 /*
1491                  * Do not do anything if we might cause open_ctree() to block
1492                  * before we have finished mounting the filesystem.
1493                  */
1494                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1495                         goto sleep;
1496
1497                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1498                         goto sleep;
1499
1500                 /*
1501                  * Avoid the problem that we change the status of the fs
1502                  * during the above check and trylock.
1503                  */
1504                 if (btrfs_need_cleaner_sleep(fs_info)) {
1505                         mutex_unlock(&fs_info->cleaner_mutex);
1506                         goto sleep;
1507                 }
1508
1509                 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1510                         btrfs_sysfs_feature_update(fs_info);
1511
1512                 btrfs_run_delayed_iputs(fs_info);
1513
1514                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1515                 mutex_unlock(&fs_info->cleaner_mutex);
1516
1517                 /*
1518                  * The defragger has dealt with the R/O remount and umount,
1519                  * needn't do anything special here.
1520                  */
1521                 btrfs_run_defrag_inodes(fs_info);
1522
1523                 /*
1524                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1525                  * with relocation (btrfs_relocate_chunk) and relocation
1526                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1527                  * after acquiring fs_info->reclaim_bgs_lock. So we
1528                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1529                  * unused block groups.
1530                  */
1531                 btrfs_delete_unused_bgs(fs_info);
1532
1533                 /*
1534                  * Reclaim block groups in the reclaim_bgs list after we deleted
1535                  * all unused block_groups. This possibly gives us some more free
1536                  * space.
1537                  */
1538                 btrfs_reclaim_bgs(fs_info);
1539 sleep:
1540                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1541                 if (kthread_should_park())
1542                         kthread_parkme();
1543                 if (kthread_should_stop())
1544                         return 0;
1545                 if (!again) {
1546                         set_current_state(TASK_INTERRUPTIBLE);
1547                         schedule();
1548                         __set_current_state(TASK_RUNNING);
1549                 }
1550         }
1551 }
1552
1553 static int transaction_kthread(void *arg)
1554 {
1555         struct btrfs_root *root = arg;
1556         struct btrfs_fs_info *fs_info = root->fs_info;
1557         struct btrfs_trans_handle *trans;
1558         struct btrfs_transaction *cur;
1559         u64 transid;
1560         time64_t delta;
1561         unsigned long delay;
1562         bool cannot_commit;
1563
1564         do {
1565                 cannot_commit = false;
1566                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1567                 mutex_lock(&fs_info->transaction_kthread_mutex);
1568
1569                 spin_lock(&fs_info->trans_lock);
1570                 cur = fs_info->running_transaction;
1571                 if (!cur) {
1572                         spin_unlock(&fs_info->trans_lock);
1573                         goto sleep;
1574                 }
1575
1576                 delta = ktime_get_seconds() - cur->start_time;
1577                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1578                     cur->state < TRANS_STATE_COMMIT_PREP &&
1579                     delta < fs_info->commit_interval) {
1580                         spin_unlock(&fs_info->trans_lock);
1581                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1582                         delay = min(delay,
1583                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1584                         goto sleep;
1585                 }
1586                 transid = cur->transid;
1587                 spin_unlock(&fs_info->trans_lock);
1588
1589                 /* If the file system is aborted, this will always fail. */
1590                 trans = btrfs_attach_transaction(root);
1591                 if (IS_ERR(trans)) {
1592                         if (PTR_ERR(trans) != -ENOENT)
1593                                 cannot_commit = true;
1594                         goto sleep;
1595                 }
1596                 if (transid == trans->transid) {
1597                         btrfs_commit_transaction(trans);
1598                 } else {
1599                         btrfs_end_transaction(trans);
1600                 }
1601 sleep:
1602                 wake_up_process(fs_info->cleaner_kthread);
1603                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1604
1605                 if (BTRFS_FS_ERROR(fs_info))
1606                         btrfs_cleanup_transaction(fs_info);
1607                 if (!kthread_should_stop() &&
1608                                 (!btrfs_transaction_blocked(fs_info) ||
1609                                  cannot_commit))
1610                         schedule_timeout_interruptible(delay);
1611         } while (!kthread_should_stop());
1612         return 0;
1613 }
1614
1615 /*
1616  * This will find the highest generation in the array of root backups.  The
1617  * index of the highest array is returned, or -EINVAL if we can't find
1618  * anything.
1619  *
1620  * We check to make sure the array is valid by comparing the
1621  * generation of the latest  root in the array with the generation
1622  * in the super block.  If they don't match we pitch it.
1623  */
1624 static int find_newest_super_backup(struct btrfs_fs_info *info)
1625 {
1626         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1627         u64 cur;
1628         struct btrfs_root_backup *root_backup;
1629         int i;
1630
1631         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1632                 root_backup = info->super_copy->super_roots + i;
1633                 cur = btrfs_backup_tree_root_gen(root_backup);
1634                 if (cur == newest_gen)
1635                         return i;
1636         }
1637
1638         return -EINVAL;
1639 }
1640
1641 /*
1642  * copy all the root pointers into the super backup array.
1643  * this will bump the backup pointer by one when it is
1644  * done
1645  */
1646 static void backup_super_roots(struct btrfs_fs_info *info)
1647 {
1648         const int next_backup = info->backup_root_index;
1649         struct btrfs_root_backup *root_backup;
1650
1651         root_backup = info->super_for_commit->super_roots + next_backup;
1652
1653         /*
1654          * make sure all of our padding and empty slots get zero filled
1655          * regardless of which ones we use today
1656          */
1657         memset(root_backup, 0, sizeof(*root_backup));
1658
1659         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1660
1661         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1662         btrfs_set_backup_tree_root_gen(root_backup,
1663                                btrfs_header_generation(info->tree_root->node));
1664
1665         btrfs_set_backup_tree_root_level(root_backup,
1666                                btrfs_header_level(info->tree_root->node));
1667
1668         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1669         btrfs_set_backup_chunk_root_gen(root_backup,
1670                                btrfs_header_generation(info->chunk_root->node));
1671         btrfs_set_backup_chunk_root_level(root_backup,
1672                                btrfs_header_level(info->chunk_root->node));
1673
1674         if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1675                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1676                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1677
1678                 btrfs_set_backup_extent_root(root_backup,
1679                                              extent_root->node->start);
1680                 btrfs_set_backup_extent_root_gen(root_backup,
1681                                 btrfs_header_generation(extent_root->node));
1682                 btrfs_set_backup_extent_root_level(root_backup,
1683                                         btrfs_header_level(extent_root->node));
1684
1685                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1686                 btrfs_set_backup_csum_root_gen(root_backup,
1687                                                btrfs_header_generation(csum_root->node));
1688                 btrfs_set_backup_csum_root_level(root_backup,
1689                                                  btrfs_header_level(csum_root->node));
1690         }
1691
1692         /*
1693          * we might commit during log recovery, which happens before we set
1694          * the fs_root.  Make sure it is valid before we fill it in.
1695          */
1696         if (info->fs_root && info->fs_root->node) {
1697                 btrfs_set_backup_fs_root(root_backup,
1698                                          info->fs_root->node->start);
1699                 btrfs_set_backup_fs_root_gen(root_backup,
1700                                btrfs_header_generation(info->fs_root->node));
1701                 btrfs_set_backup_fs_root_level(root_backup,
1702                                btrfs_header_level(info->fs_root->node));
1703         }
1704
1705         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1706         btrfs_set_backup_dev_root_gen(root_backup,
1707                                btrfs_header_generation(info->dev_root->node));
1708         btrfs_set_backup_dev_root_level(root_backup,
1709                                        btrfs_header_level(info->dev_root->node));
1710
1711         btrfs_set_backup_total_bytes(root_backup,
1712                              btrfs_super_total_bytes(info->super_copy));
1713         btrfs_set_backup_bytes_used(root_backup,
1714                              btrfs_super_bytes_used(info->super_copy));
1715         btrfs_set_backup_num_devices(root_backup,
1716                              btrfs_super_num_devices(info->super_copy));
1717
1718         /*
1719          * if we don't copy this out to the super_copy, it won't get remembered
1720          * for the next commit
1721          */
1722         memcpy(&info->super_copy->super_roots,
1723                &info->super_for_commit->super_roots,
1724                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1725 }
1726
1727 /*
1728  * Reads a backup root based on the passed priority. Prio 0 is the newest, prio
1729  * 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1730  *
1731  * @fs_info:  filesystem whose backup roots need to be read
1732  * @priority: priority of backup root required
1733  *
1734  * Returns backup root index on success and -EINVAL otherwise.
1735  */
1736 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1737 {
1738         int backup_index = find_newest_super_backup(fs_info);
1739         struct btrfs_super_block *super = fs_info->super_copy;
1740         struct btrfs_root_backup *root_backup;
1741
1742         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1743                 if (priority == 0)
1744                         return backup_index;
1745
1746                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1747                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1748         } else {
1749                 return -EINVAL;
1750         }
1751
1752         root_backup = super->super_roots + backup_index;
1753
1754         btrfs_set_super_generation(super,
1755                                    btrfs_backup_tree_root_gen(root_backup));
1756         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1757         btrfs_set_super_root_level(super,
1758                                    btrfs_backup_tree_root_level(root_backup));
1759         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1760
1761         /*
1762          * Fixme: the total bytes and num_devices need to match or we should
1763          * need a fsck
1764          */
1765         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1766         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1767
1768         return backup_index;
1769 }
1770
1771 /* helper to cleanup workers */
1772 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1773 {
1774         btrfs_destroy_workqueue(fs_info->fixup_workers);
1775         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1776         btrfs_destroy_workqueue(fs_info->workers);
1777         if (fs_info->endio_workers)
1778                 destroy_workqueue(fs_info->endio_workers);
1779         if (fs_info->rmw_workers)
1780                 destroy_workqueue(fs_info->rmw_workers);
1781         if (fs_info->compressed_write_workers)
1782                 destroy_workqueue(fs_info->compressed_write_workers);
1783         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1784         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1785         btrfs_destroy_workqueue(fs_info->delayed_workers);
1786         btrfs_destroy_workqueue(fs_info->caching_workers);
1787         btrfs_destroy_workqueue(fs_info->flush_workers);
1788         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
1789         if (fs_info->discard_ctl.discard_workers)
1790                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
1791         /*
1792          * Now that all other work queues are destroyed, we can safely destroy
1793          * the queues used for metadata I/O, since tasks from those other work
1794          * queues can do metadata I/O operations.
1795          */
1796         if (fs_info->endio_meta_workers)
1797                 destroy_workqueue(fs_info->endio_meta_workers);
1798 }
1799
1800 static void free_root_extent_buffers(struct btrfs_root *root)
1801 {
1802         if (root) {
1803                 free_extent_buffer(root->node);
1804                 free_extent_buffer(root->commit_root);
1805                 root->node = NULL;
1806                 root->commit_root = NULL;
1807         }
1808 }
1809
1810 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
1811 {
1812         struct btrfs_root *root, *tmp;
1813
1814         rbtree_postorder_for_each_entry_safe(root, tmp,
1815                                              &fs_info->global_root_tree,
1816                                              rb_node)
1817                 free_root_extent_buffers(root);
1818 }
1819
1820 /* helper to cleanup tree roots */
1821 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
1822 {
1823         free_root_extent_buffers(info->tree_root);
1824
1825         free_global_root_pointers(info);
1826         free_root_extent_buffers(info->dev_root);
1827         free_root_extent_buffers(info->quota_root);
1828         free_root_extent_buffers(info->uuid_root);
1829         free_root_extent_buffers(info->fs_root);
1830         free_root_extent_buffers(info->data_reloc_root);
1831         free_root_extent_buffers(info->block_group_root);
1832         free_root_extent_buffers(info->stripe_root);
1833         if (free_chunk_root)
1834                 free_root_extent_buffers(info->chunk_root);
1835 }
1836
1837 void btrfs_put_root(struct btrfs_root *root)
1838 {
1839         if (!root)
1840                 return;
1841
1842         if (refcount_dec_and_test(&root->refs)) {
1843                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1844                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
1845                 if (root->anon_dev)
1846                         free_anon_bdev(root->anon_dev);
1847                 free_root_extent_buffers(root);
1848 #ifdef CONFIG_BTRFS_DEBUG
1849                 spin_lock(&root->fs_info->fs_roots_radix_lock);
1850                 list_del_init(&root->leak_list);
1851                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
1852 #endif
1853                 kfree(root);
1854         }
1855 }
1856
1857 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
1858 {
1859         int ret;
1860         struct btrfs_root *gang[8];
1861         int i;
1862
1863         while (!list_empty(&fs_info->dead_roots)) {
1864                 gang[0] = list_entry(fs_info->dead_roots.next,
1865                                      struct btrfs_root, root_list);
1866                 list_del(&gang[0]->root_list);
1867
1868                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
1869                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
1870                 btrfs_put_root(gang[0]);
1871         }
1872
1873         while (1) {
1874                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1875                                              (void **)gang, 0,
1876                                              ARRAY_SIZE(gang));
1877                 if (!ret)
1878                         break;
1879                 for (i = 0; i < ret; i++)
1880                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
1881         }
1882 }
1883
1884 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
1885 {
1886         mutex_init(&fs_info->scrub_lock);
1887         atomic_set(&fs_info->scrubs_running, 0);
1888         atomic_set(&fs_info->scrub_pause_req, 0);
1889         atomic_set(&fs_info->scrubs_paused, 0);
1890         atomic_set(&fs_info->scrub_cancel_req, 0);
1891         init_waitqueue_head(&fs_info->scrub_pause_wait);
1892         refcount_set(&fs_info->scrub_workers_refcnt, 0);
1893 }
1894
1895 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
1896 {
1897         spin_lock_init(&fs_info->balance_lock);
1898         mutex_init(&fs_info->balance_mutex);
1899         atomic_set(&fs_info->balance_pause_req, 0);
1900         atomic_set(&fs_info->balance_cancel_req, 0);
1901         fs_info->balance_ctl = NULL;
1902         init_waitqueue_head(&fs_info->balance_wait_q);
1903         atomic_set(&fs_info->reloc_cancel_req, 0);
1904 }
1905
1906 static int btrfs_init_btree_inode(struct super_block *sb)
1907 {
1908         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1909         unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
1910                                               fs_info->tree_root);
1911         struct inode *inode;
1912
1913         inode = new_inode(sb);
1914         if (!inode)
1915                 return -ENOMEM;
1916
1917         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1918         set_nlink(inode, 1);
1919         /*
1920          * we set the i_size on the btree inode to the max possible int.
1921          * the real end of the address space is determined by all of
1922          * the devices in the system
1923          */
1924         inode->i_size = OFFSET_MAX;
1925         inode->i_mapping->a_ops = &btree_aops;
1926         mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
1927
1928         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
1929         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
1930                             IO_TREE_BTREE_INODE_IO);
1931         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
1932
1933         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
1934         BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
1935         BTRFS_I(inode)->location.type = 0;
1936         BTRFS_I(inode)->location.offset = 0;
1937         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
1938         __insert_inode_hash(inode, hash);
1939         fs_info->btree_inode = inode;
1940
1941         return 0;
1942 }
1943
1944 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
1945 {
1946         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
1947         init_rwsem(&fs_info->dev_replace.rwsem);
1948         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
1949 }
1950
1951 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
1952 {
1953         spin_lock_init(&fs_info->qgroup_lock);
1954         mutex_init(&fs_info->qgroup_ioctl_lock);
1955         fs_info->qgroup_tree = RB_ROOT;
1956         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
1957         fs_info->qgroup_seq = 1;
1958         fs_info->qgroup_ulist = NULL;
1959         fs_info->qgroup_rescan_running = false;
1960         fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1961         mutex_init(&fs_info->qgroup_rescan_lock);
1962 }
1963
1964 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
1965 {
1966         u32 max_active = fs_info->thread_pool_size;
1967         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
1968         unsigned int ordered_flags = WQ_MEM_RECLAIM | WQ_FREEZABLE;
1969
1970         fs_info->workers =
1971                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
1972
1973         fs_info->delalloc_workers =
1974                 btrfs_alloc_workqueue(fs_info, "delalloc",
1975                                       flags, max_active, 2);
1976
1977         fs_info->flush_workers =
1978                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
1979                                       flags, max_active, 0);
1980
1981         fs_info->caching_workers =
1982                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
1983
1984         fs_info->fixup_workers =
1985                 btrfs_alloc_ordered_workqueue(fs_info, "fixup", ordered_flags);
1986
1987         fs_info->endio_workers =
1988                 alloc_workqueue("btrfs-endio", flags, max_active);
1989         fs_info->endio_meta_workers =
1990                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
1991         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
1992         fs_info->endio_write_workers =
1993                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
1994                                       max_active, 2);
1995         fs_info->compressed_write_workers =
1996                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
1997         fs_info->endio_freespace_worker =
1998                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
1999                                       max_active, 0);
2000         fs_info->delayed_workers =
2001                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2002                                       max_active, 0);
2003         fs_info->qgroup_rescan_workers =
2004                 btrfs_alloc_ordered_workqueue(fs_info, "qgroup-rescan",
2005                                               ordered_flags);
2006         fs_info->discard_ctl.discard_workers =
2007                 alloc_ordered_workqueue("btrfs_discard", WQ_FREEZABLE);
2008
2009         if (!(fs_info->workers &&
2010               fs_info->delalloc_workers && fs_info->flush_workers &&
2011               fs_info->endio_workers && fs_info->endio_meta_workers &&
2012               fs_info->compressed_write_workers &&
2013               fs_info->endio_write_workers &&
2014               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2015               fs_info->caching_workers && fs_info->fixup_workers &&
2016               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2017               fs_info->discard_ctl.discard_workers)) {
2018                 return -ENOMEM;
2019         }
2020
2021         return 0;
2022 }
2023
2024 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2025 {
2026         struct crypto_shash *csum_shash;
2027         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2028
2029         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2030
2031         if (IS_ERR(csum_shash)) {
2032                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2033                           csum_driver);
2034                 return PTR_ERR(csum_shash);
2035         }
2036
2037         fs_info->csum_shash = csum_shash;
2038
2039         /*
2040          * Check if the checksum implementation is a fast accelerated one.
2041          * As-is this is a bit of a hack and should be replaced once the csum
2042          * implementations provide that information themselves.
2043          */
2044         switch (csum_type) {
2045         case BTRFS_CSUM_TYPE_CRC32:
2046                 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2047                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2048                 break;
2049         case BTRFS_CSUM_TYPE_XXHASH:
2050                 set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2051                 break;
2052         default:
2053                 break;
2054         }
2055
2056         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2057                         btrfs_super_csum_name(csum_type),
2058                         crypto_shash_driver_name(csum_shash));
2059         return 0;
2060 }
2061
2062 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2063                             struct btrfs_fs_devices *fs_devices)
2064 {
2065         int ret;
2066         struct btrfs_tree_parent_check check = { 0 };
2067         struct btrfs_root *log_tree_root;
2068         struct btrfs_super_block *disk_super = fs_info->super_copy;
2069         u64 bytenr = btrfs_super_log_root(disk_super);
2070         int level = btrfs_super_log_root_level(disk_super);
2071
2072         if (fs_devices->rw_devices == 0) {
2073                 btrfs_warn(fs_info, "log replay required on RO media");
2074                 return -EIO;
2075         }
2076
2077         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2078                                          GFP_KERNEL);
2079         if (!log_tree_root)
2080                 return -ENOMEM;
2081
2082         check.level = level;
2083         check.transid = fs_info->generation + 1;
2084         check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2085         log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2086         if (IS_ERR(log_tree_root->node)) {
2087                 btrfs_warn(fs_info, "failed to read log tree");
2088                 ret = PTR_ERR(log_tree_root->node);
2089                 log_tree_root->node = NULL;
2090                 btrfs_put_root(log_tree_root);
2091                 return ret;
2092         }
2093         if (!extent_buffer_uptodate(log_tree_root->node)) {
2094                 btrfs_err(fs_info, "failed to read log tree");
2095                 btrfs_put_root(log_tree_root);
2096                 return -EIO;
2097         }
2098
2099         /* returns with log_tree_root freed on success */
2100         ret = btrfs_recover_log_trees(log_tree_root);
2101         if (ret) {
2102                 btrfs_handle_fs_error(fs_info, ret,
2103                                       "Failed to recover log tree");
2104                 btrfs_put_root(log_tree_root);
2105                 return ret;
2106         }
2107
2108         if (sb_rdonly(fs_info->sb)) {
2109                 ret = btrfs_commit_super(fs_info);
2110                 if (ret)
2111                         return ret;
2112         }
2113
2114         return 0;
2115 }
2116
2117 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2118                                       struct btrfs_path *path, u64 objectid,
2119                                       const char *name)
2120 {
2121         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2122         struct btrfs_root *root;
2123         u64 max_global_id = 0;
2124         int ret;
2125         struct btrfs_key key = {
2126                 .objectid = objectid,
2127                 .type = BTRFS_ROOT_ITEM_KEY,
2128                 .offset = 0,
2129         };
2130         bool found = false;
2131
2132         /* If we have IGNOREDATACSUMS skip loading these roots. */
2133         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2134             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2135                 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2136                 return 0;
2137         }
2138
2139         while (1) {
2140                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2141                 if (ret < 0)
2142                         break;
2143
2144                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2145                         ret = btrfs_next_leaf(tree_root, path);
2146                         if (ret) {
2147                                 if (ret > 0)
2148                                         ret = 0;
2149                                 break;
2150                         }
2151                 }
2152                 ret = 0;
2153
2154                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2155                 if (key.objectid != objectid)
2156                         break;
2157                 btrfs_release_path(path);
2158
2159                 /*
2160                  * Just worry about this for extent tree, it'll be the same for
2161                  * everybody.
2162                  */
2163                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2164                         max_global_id = max(max_global_id, key.offset);
2165
2166                 found = true;
2167                 root = read_tree_root_path(tree_root, path, &key);
2168                 if (IS_ERR(root)) {
2169                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2170                                 ret = PTR_ERR(root);
2171                         break;
2172                 }
2173                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2174                 ret = btrfs_global_root_insert(root);
2175                 if (ret) {
2176                         btrfs_put_root(root);
2177                         break;
2178                 }
2179                 key.offset++;
2180         }
2181         btrfs_release_path(path);
2182
2183         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2184                 fs_info->nr_global_roots = max_global_id + 1;
2185
2186         if (!found || ret) {
2187                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2188                         set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2189
2190                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2191                         ret = ret ? ret : -ENOENT;
2192                 else
2193                         ret = 0;
2194                 btrfs_err(fs_info, "failed to load root %s", name);
2195         }
2196         return ret;
2197 }
2198
2199 static int load_global_roots(struct btrfs_root *tree_root)
2200 {
2201         struct btrfs_path *path;
2202         int ret = 0;
2203
2204         path = btrfs_alloc_path();
2205         if (!path)
2206                 return -ENOMEM;
2207
2208         ret = load_global_roots_objectid(tree_root, path,
2209                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2210         if (ret)
2211                 goto out;
2212         ret = load_global_roots_objectid(tree_root, path,
2213                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2214         if (ret)
2215                 goto out;
2216         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2217                 goto out;
2218         ret = load_global_roots_objectid(tree_root, path,
2219                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2220                                          "free space");
2221 out:
2222         btrfs_free_path(path);
2223         return ret;
2224 }
2225
2226 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2227 {
2228         struct btrfs_root *tree_root = fs_info->tree_root;
2229         struct btrfs_root *root;
2230         struct btrfs_key location;
2231         int ret;
2232
2233         BUG_ON(!fs_info->tree_root);
2234
2235         ret = load_global_roots(tree_root);
2236         if (ret)
2237                 return ret;
2238
2239         location.type = BTRFS_ROOT_ITEM_KEY;
2240         location.offset = 0;
2241
2242         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2243                 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2244                 root = btrfs_read_tree_root(tree_root, &location);
2245                 if (IS_ERR(root)) {
2246                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2247                                 ret = PTR_ERR(root);
2248                                 goto out;
2249                         }
2250                 } else {
2251                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2252                         fs_info->block_group_root = root;
2253                 }
2254         }
2255
2256         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2257         root = btrfs_read_tree_root(tree_root, &location);
2258         if (IS_ERR(root)) {
2259                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2260                         ret = PTR_ERR(root);
2261                         goto out;
2262                 }
2263         } else {
2264                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2265                 fs_info->dev_root = root;
2266         }
2267         /* Initialize fs_info for all devices in any case */
2268         ret = btrfs_init_devices_late(fs_info);
2269         if (ret)
2270                 goto out;
2271
2272         /*
2273          * This tree can share blocks with some other fs tree during relocation
2274          * and we need a proper setup by btrfs_get_fs_root
2275          */
2276         root = btrfs_get_fs_root(tree_root->fs_info,
2277                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2278         if (IS_ERR(root)) {
2279                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2280                         ret = PTR_ERR(root);
2281                         goto out;
2282                 }
2283         } else {
2284                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2285                 fs_info->data_reloc_root = root;
2286         }
2287
2288         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2289         root = btrfs_read_tree_root(tree_root, &location);
2290         if (!IS_ERR(root)) {
2291                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2292                 fs_info->quota_root = root;
2293         }
2294
2295         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2296         root = btrfs_read_tree_root(tree_root, &location);
2297         if (IS_ERR(root)) {
2298                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2299                         ret = PTR_ERR(root);
2300                         if (ret != -ENOENT)
2301                                 goto out;
2302                 }
2303         } else {
2304                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2305                 fs_info->uuid_root = root;
2306         }
2307
2308         if (btrfs_fs_incompat(fs_info, RAID_STRIPE_TREE)) {
2309                 location.objectid = BTRFS_RAID_STRIPE_TREE_OBJECTID;
2310                 root = btrfs_read_tree_root(tree_root, &location);
2311                 if (IS_ERR(root)) {
2312                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2313                                 ret = PTR_ERR(root);
2314                                 goto out;
2315                         }
2316                 } else {
2317                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2318                         fs_info->stripe_root = root;
2319                 }
2320         }
2321
2322         return 0;
2323 out:
2324         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2325                    location.objectid, ret);
2326         return ret;
2327 }
2328
2329 /*
2330  * Real super block validation
2331  * NOTE: super csum type and incompat features will not be checked here.
2332  *
2333  * @sb:         super block to check
2334  * @mirror_num: the super block number to check its bytenr:
2335  *              0       the primary (1st) sb
2336  *              1, 2    2nd and 3rd backup copy
2337  *             -1       skip bytenr check
2338  */
2339 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2340                          struct btrfs_super_block *sb, int mirror_num)
2341 {
2342         u64 nodesize = btrfs_super_nodesize(sb);
2343         u64 sectorsize = btrfs_super_sectorsize(sb);
2344         int ret = 0;
2345
2346         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2347                 btrfs_err(fs_info, "no valid FS found");
2348                 ret = -EINVAL;
2349         }
2350         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2351                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2352                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2353                 ret = -EINVAL;
2354         }
2355         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2356                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2357                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2358                 ret = -EINVAL;
2359         }
2360         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2361                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2362                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2363                 ret = -EINVAL;
2364         }
2365         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2366                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2367                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2368                 ret = -EINVAL;
2369         }
2370
2371         /*
2372          * Check sectorsize and nodesize first, other check will need it.
2373          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2374          */
2375         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2376             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2377                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2378                 ret = -EINVAL;
2379         }
2380
2381         /*
2382          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2383          *
2384          * We can support 16K sectorsize with 64K page size without problem,
2385          * but such sectorsize/pagesize combination doesn't make much sense.
2386          * 4K will be our future standard, PAGE_SIZE is supported from the very
2387          * beginning.
2388          */
2389         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2390                 btrfs_err(fs_info,
2391                         "sectorsize %llu not yet supported for page size %lu",
2392                         sectorsize, PAGE_SIZE);
2393                 ret = -EINVAL;
2394         }
2395
2396         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2397             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2398                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2399                 ret = -EINVAL;
2400         }
2401         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2402                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2403                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2404                 ret = -EINVAL;
2405         }
2406
2407         /* Root alignment check */
2408         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2409                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2410                            btrfs_super_root(sb));
2411                 ret = -EINVAL;
2412         }
2413         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2414                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2415                            btrfs_super_chunk_root(sb));
2416                 ret = -EINVAL;
2417         }
2418         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2419                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2420                            btrfs_super_log_root(sb));
2421                 ret = -EINVAL;
2422         }
2423
2424         if (!fs_info->fs_devices->temp_fsid &&
2425             memcmp(fs_info->fs_devices->fsid, sb->fsid, BTRFS_FSID_SIZE) != 0) {
2426                 btrfs_err(fs_info,
2427                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2428                           sb->fsid, fs_info->fs_devices->fsid);
2429                 ret = -EINVAL;
2430         }
2431
2432         if (memcmp(fs_info->fs_devices->metadata_uuid, btrfs_sb_fsid_ptr(sb),
2433                    BTRFS_FSID_SIZE) != 0) {
2434                 btrfs_err(fs_info,
2435 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2436                           btrfs_sb_fsid_ptr(sb), fs_info->fs_devices->metadata_uuid);
2437                 ret = -EINVAL;
2438         }
2439
2440         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2441                    BTRFS_FSID_SIZE) != 0) {
2442                 btrfs_err(fs_info,
2443                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2444                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2445                 ret = -EINVAL;
2446         }
2447
2448         /*
2449          * Artificial requirement for block-group-tree to force newer features
2450          * (free-space-tree, no-holes) so the test matrix is smaller.
2451          */
2452         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2453             (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2454              !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2455                 btrfs_err(fs_info,
2456                 "block-group-tree feature requires fres-space-tree and no-holes");
2457                 ret = -EINVAL;
2458         }
2459
2460         /*
2461          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2462          * done later
2463          */
2464         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2465                 btrfs_err(fs_info, "bytes_used is too small %llu",
2466                           btrfs_super_bytes_used(sb));
2467                 ret = -EINVAL;
2468         }
2469         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2470                 btrfs_err(fs_info, "invalid stripesize %u",
2471                           btrfs_super_stripesize(sb));
2472                 ret = -EINVAL;
2473         }
2474         if (btrfs_super_num_devices(sb) > (1UL << 31))
2475                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2476                            btrfs_super_num_devices(sb));
2477         if (btrfs_super_num_devices(sb) == 0) {
2478                 btrfs_err(fs_info, "number of devices is 0");
2479                 ret = -EINVAL;
2480         }
2481
2482         if (mirror_num >= 0 &&
2483             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2484                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2485                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2486                 ret = -EINVAL;
2487         }
2488
2489         /*
2490          * Obvious sys_chunk_array corruptions, it must hold at least one key
2491          * and one chunk
2492          */
2493         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2494                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2495                           btrfs_super_sys_array_size(sb),
2496                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2497                 ret = -EINVAL;
2498         }
2499         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2500                         + sizeof(struct btrfs_chunk)) {
2501                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2502                           btrfs_super_sys_array_size(sb),
2503                           sizeof(struct btrfs_disk_key)
2504                           + sizeof(struct btrfs_chunk));
2505                 ret = -EINVAL;
2506         }
2507
2508         /*
2509          * The generation is a global counter, we'll trust it more than the others
2510          * but it's still possible that it's the one that's wrong.
2511          */
2512         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2513                 btrfs_warn(fs_info,
2514                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2515                         btrfs_super_generation(sb),
2516                         btrfs_super_chunk_root_generation(sb));
2517         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2518             && btrfs_super_cache_generation(sb) != (u64)-1)
2519                 btrfs_warn(fs_info,
2520                         "suspicious: generation < cache_generation: %llu < %llu",
2521                         btrfs_super_generation(sb),
2522                         btrfs_super_cache_generation(sb));
2523
2524         return ret;
2525 }
2526
2527 /*
2528  * Validation of super block at mount time.
2529  * Some checks already done early at mount time, like csum type and incompat
2530  * flags will be skipped.
2531  */
2532 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2533 {
2534         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2535 }
2536
2537 /*
2538  * Validation of super block at write time.
2539  * Some checks like bytenr check will be skipped as their values will be
2540  * overwritten soon.
2541  * Extra checks like csum type and incompat flags will be done here.
2542  */
2543 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2544                                       struct btrfs_super_block *sb)
2545 {
2546         int ret;
2547
2548         ret = btrfs_validate_super(fs_info, sb, -1);
2549         if (ret < 0)
2550                 goto out;
2551         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2552                 ret = -EUCLEAN;
2553                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2554                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2555                 goto out;
2556         }
2557         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2558                 ret = -EUCLEAN;
2559                 btrfs_err(fs_info,
2560                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2561                           btrfs_super_incompat_flags(sb),
2562                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2563                 goto out;
2564         }
2565 out:
2566         if (ret < 0)
2567                 btrfs_err(fs_info,
2568                 "super block corruption detected before writing it to disk");
2569         return ret;
2570 }
2571
2572 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2573 {
2574         struct btrfs_tree_parent_check check = {
2575                 .level = level,
2576                 .transid = gen,
2577                 .owner_root = root->root_key.objectid
2578         };
2579         int ret = 0;
2580
2581         root->node = read_tree_block(root->fs_info, bytenr, &check);
2582         if (IS_ERR(root->node)) {
2583                 ret = PTR_ERR(root->node);
2584                 root->node = NULL;
2585                 return ret;
2586         }
2587         if (!extent_buffer_uptodate(root->node)) {
2588                 free_extent_buffer(root->node);
2589                 root->node = NULL;
2590                 return -EIO;
2591         }
2592
2593         btrfs_set_root_node(&root->root_item, root->node);
2594         root->commit_root = btrfs_root_node(root);
2595         btrfs_set_root_refs(&root->root_item, 1);
2596         return ret;
2597 }
2598
2599 static int load_important_roots(struct btrfs_fs_info *fs_info)
2600 {
2601         struct btrfs_super_block *sb = fs_info->super_copy;
2602         u64 gen, bytenr;
2603         int level, ret;
2604
2605         bytenr = btrfs_super_root(sb);
2606         gen = btrfs_super_generation(sb);
2607         level = btrfs_super_root_level(sb);
2608         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2609         if (ret) {
2610                 btrfs_warn(fs_info, "couldn't read tree root");
2611                 return ret;
2612         }
2613         return 0;
2614 }
2615
2616 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2617 {
2618         int backup_index = find_newest_super_backup(fs_info);
2619         struct btrfs_super_block *sb = fs_info->super_copy;
2620         struct btrfs_root *tree_root = fs_info->tree_root;
2621         bool handle_error = false;
2622         int ret = 0;
2623         int i;
2624
2625         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2626                 if (handle_error) {
2627                         if (!IS_ERR(tree_root->node))
2628                                 free_extent_buffer(tree_root->node);
2629                         tree_root->node = NULL;
2630
2631                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2632                                 break;
2633
2634                         free_root_pointers(fs_info, 0);
2635
2636                         /*
2637                          * Don't use the log in recovery mode, it won't be
2638                          * valid
2639                          */
2640                         btrfs_set_super_log_root(sb, 0);
2641
2642                         btrfs_warn(fs_info, "try to load backup roots slot %d", i);
2643                         ret = read_backup_root(fs_info, i);
2644                         backup_index = ret;
2645                         if (ret < 0)
2646                                 return ret;
2647                 }
2648
2649                 ret = load_important_roots(fs_info);
2650                 if (ret) {
2651                         handle_error = true;
2652                         continue;
2653                 }
2654
2655                 /*
2656                  * No need to hold btrfs_root::objectid_mutex since the fs
2657                  * hasn't been fully initialised and we are the only user
2658                  */
2659                 ret = btrfs_init_root_free_objectid(tree_root);
2660                 if (ret < 0) {
2661                         handle_error = true;
2662                         continue;
2663                 }
2664
2665                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2666
2667                 ret = btrfs_read_roots(fs_info);
2668                 if (ret < 0) {
2669                         handle_error = true;
2670                         continue;
2671                 }
2672
2673                 /* All successful */
2674                 fs_info->generation = btrfs_header_generation(tree_root->node);
2675                 btrfs_set_last_trans_committed(fs_info, fs_info->generation);
2676                 fs_info->last_reloc_trans = 0;
2677
2678                 /* Always begin writing backup roots after the one being used */
2679                 if (backup_index < 0) {
2680                         fs_info->backup_root_index = 0;
2681                 } else {
2682                         fs_info->backup_root_index = backup_index + 1;
2683                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2684                 }
2685                 break;
2686         }
2687
2688         return ret;
2689 }
2690
2691 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2692 {
2693         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2694         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2695         INIT_LIST_HEAD(&fs_info->trans_list);
2696         INIT_LIST_HEAD(&fs_info->dead_roots);
2697         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2698         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2699         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2700         spin_lock_init(&fs_info->delalloc_root_lock);
2701         spin_lock_init(&fs_info->trans_lock);
2702         spin_lock_init(&fs_info->fs_roots_radix_lock);
2703         spin_lock_init(&fs_info->delayed_iput_lock);
2704         spin_lock_init(&fs_info->defrag_inodes_lock);
2705         spin_lock_init(&fs_info->super_lock);
2706         spin_lock_init(&fs_info->buffer_lock);
2707         spin_lock_init(&fs_info->unused_bgs_lock);
2708         spin_lock_init(&fs_info->treelog_bg_lock);
2709         spin_lock_init(&fs_info->zone_active_bgs_lock);
2710         spin_lock_init(&fs_info->relocation_bg_lock);
2711         rwlock_init(&fs_info->tree_mod_log_lock);
2712         rwlock_init(&fs_info->global_root_lock);
2713         mutex_init(&fs_info->unused_bg_unpin_mutex);
2714         mutex_init(&fs_info->reclaim_bgs_lock);
2715         mutex_init(&fs_info->reloc_mutex);
2716         mutex_init(&fs_info->delalloc_root_mutex);
2717         mutex_init(&fs_info->zoned_meta_io_lock);
2718         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2719         seqlock_init(&fs_info->profiles_lock);
2720
2721         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2722         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2723         btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2724         btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2725         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_prep,
2726                                      BTRFS_LOCKDEP_TRANS_COMMIT_PREP);
2727         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2728                                      BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2729         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2730                                      BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2731         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2732                                      BTRFS_LOCKDEP_TRANS_COMPLETED);
2733
2734         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2735         INIT_LIST_HEAD(&fs_info->space_info);
2736         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2737         INIT_LIST_HEAD(&fs_info->unused_bgs);
2738         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2739         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2740 #ifdef CONFIG_BTRFS_DEBUG
2741         INIT_LIST_HEAD(&fs_info->allocated_roots);
2742         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2743         spin_lock_init(&fs_info->eb_leak_lock);
2744 #endif
2745         fs_info->mapping_tree = RB_ROOT_CACHED;
2746         rwlock_init(&fs_info->mapping_tree_lock);
2747         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2748                              BTRFS_BLOCK_RSV_GLOBAL);
2749         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2750         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2751         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2752         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2753                              BTRFS_BLOCK_RSV_DELOPS);
2754         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2755                              BTRFS_BLOCK_RSV_DELREFS);
2756
2757         atomic_set(&fs_info->async_delalloc_pages, 0);
2758         atomic_set(&fs_info->defrag_running, 0);
2759         atomic_set(&fs_info->nr_delayed_iputs, 0);
2760         atomic64_set(&fs_info->tree_mod_seq, 0);
2761         fs_info->global_root_tree = RB_ROOT;
2762         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2763         fs_info->metadata_ratio = 0;
2764         fs_info->defrag_inodes = RB_ROOT;
2765         atomic64_set(&fs_info->free_chunk_space, 0);
2766         fs_info->tree_mod_log = RB_ROOT;
2767         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2768         btrfs_init_ref_verify(fs_info);
2769
2770         fs_info->thread_pool_size = min_t(unsigned long,
2771                                           num_online_cpus() + 2, 8);
2772
2773         INIT_LIST_HEAD(&fs_info->ordered_roots);
2774         spin_lock_init(&fs_info->ordered_root_lock);
2775
2776         btrfs_init_scrub(fs_info);
2777         btrfs_init_balance(fs_info);
2778         btrfs_init_async_reclaim_work(fs_info);
2779
2780         rwlock_init(&fs_info->block_group_cache_lock);
2781         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2782
2783         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2784                             IO_TREE_FS_EXCLUDED_EXTENTS);
2785
2786         mutex_init(&fs_info->ordered_operations_mutex);
2787         mutex_init(&fs_info->tree_log_mutex);
2788         mutex_init(&fs_info->chunk_mutex);
2789         mutex_init(&fs_info->transaction_kthread_mutex);
2790         mutex_init(&fs_info->cleaner_mutex);
2791         mutex_init(&fs_info->ro_block_group_mutex);
2792         init_rwsem(&fs_info->commit_root_sem);
2793         init_rwsem(&fs_info->cleanup_work_sem);
2794         init_rwsem(&fs_info->subvol_sem);
2795         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2796
2797         btrfs_init_dev_replace_locks(fs_info);
2798         btrfs_init_qgroup(fs_info);
2799         btrfs_discard_init(fs_info);
2800
2801         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2802         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2803
2804         init_waitqueue_head(&fs_info->transaction_throttle);
2805         init_waitqueue_head(&fs_info->transaction_wait);
2806         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2807         init_waitqueue_head(&fs_info->async_submit_wait);
2808         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2809
2810         /* Usable values until the real ones are cached from the superblock */
2811         fs_info->nodesize = 4096;
2812         fs_info->sectorsize = 4096;
2813         fs_info->sectorsize_bits = ilog2(4096);
2814         fs_info->stripesize = 4096;
2815
2816         /* Default compress algorithm when user does -o compress */
2817         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2818
2819         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
2820
2821         spin_lock_init(&fs_info->swapfile_pins_lock);
2822         fs_info->swapfile_pins = RB_ROOT;
2823
2824         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
2825         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
2826 }
2827
2828 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2829 {
2830         int ret;
2831
2832         fs_info->sb = sb;
2833         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2834         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2835
2836         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
2837         if (ret)
2838                 return ret;
2839
2840         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2841         if (ret)
2842                 return ret;
2843
2844         fs_info->dirty_metadata_batch = PAGE_SIZE *
2845                                         (1 + ilog2(nr_cpu_ids));
2846
2847         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2848         if (ret)
2849                 return ret;
2850
2851         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2852                         GFP_KERNEL);
2853         if (ret)
2854                 return ret;
2855
2856         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2857                                         GFP_KERNEL);
2858         if (!fs_info->delayed_root)
2859                 return -ENOMEM;
2860         btrfs_init_delayed_root(fs_info->delayed_root);
2861
2862         if (sb_rdonly(sb))
2863                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2864
2865         return btrfs_alloc_stripe_hash_table(fs_info);
2866 }
2867
2868 static int btrfs_uuid_rescan_kthread(void *data)
2869 {
2870         struct btrfs_fs_info *fs_info = data;
2871         int ret;
2872
2873         /*
2874          * 1st step is to iterate through the existing UUID tree and
2875          * to delete all entries that contain outdated data.
2876          * 2nd step is to add all missing entries to the UUID tree.
2877          */
2878         ret = btrfs_uuid_tree_iterate(fs_info);
2879         if (ret < 0) {
2880                 if (ret != -EINTR)
2881                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2882                                    ret);
2883                 up(&fs_info->uuid_tree_rescan_sem);
2884                 return ret;
2885         }
2886         return btrfs_uuid_scan_kthread(data);
2887 }
2888
2889 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2890 {
2891         struct task_struct *task;
2892
2893         down(&fs_info->uuid_tree_rescan_sem);
2894         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2895         if (IS_ERR(task)) {
2896                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2897                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2898                 up(&fs_info->uuid_tree_rescan_sem);
2899                 return PTR_ERR(task);
2900         }
2901
2902         return 0;
2903 }
2904
2905 static int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2906 {
2907         u64 root_objectid = 0;
2908         struct btrfs_root *gang[8];
2909         int i = 0;
2910         int err = 0;
2911         unsigned int ret = 0;
2912
2913         while (1) {
2914                 spin_lock(&fs_info->fs_roots_radix_lock);
2915                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2916                                              (void **)gang, root_objectid,
2917                                              ARRAY_SIZE(gang));
2918                 if (!ret) {
2919                         spin_unlock(&fs_info->fs_roots_radix_lock);
2920                         break;
2921                 }
2922                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2923
2924                 for (i = 0; i < ret; i++) {
2925                         /* Avoid to grab roots in dead_roots. */
2926                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
2927                                 gang[i] = NULL;
2928                                 continue;
2929                         }
2930                         /* Grab all the search result for later use. */
2931                         gang[i] = btrfs_grab_root(gang[i]);
2932                 }
2933                 spin_unlock(&fs_info->fs_roots_radix_lock);
2934
2935                 for (i = 0; i < ret; i++) {
2936                         if (!gang[i])
2937                                 continue;
2938                         root_objectid = gang[i]->root_key.objectid;
2939                         err = btrfs_orphan_cleanup(gang[i]);
2940                         if (err)
2941                                 goto out;
2942                         btrfs_put_root(gang[i]);
2943                 }
2944                 root_objectid++;
2945         }
2946 out:
2947         /* Release the uncleaned roots due to error. */
2948         for (; i < ret; i++) {
2949                 if (gang[i])
2950                         btrfs_put_root(gang[i]);
2951         }
2952         return err;
2953 }
2954
2955 /*
2956  * Mounting logic specific to read-write file systems. Shared by open_ctree
2957  * and btrfs_remount when remounting from read-only to read-write.
2958  */
2959 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
2960 {
2961         int ret;
2962         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
2963         bool rebuild_free_space_tree = false;
2964
2965         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2966             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2967                 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
2968                         btrfs_warn(fs_info,
2969                                    "'clear_cache' option is ignored with extent tree v2");
2970                 else
2971                         rebuild_free_space_tree = true;
2972         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2973                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2974                 btrfs_warn(fs_info, "free space tree is invalid");
2975                 rebuild_free_space_tree = true;
2976         }
2977
2978         if (rebuild_free_space_tree) {
2979                 btrfs_info(fs_info, "rebuilding free space tree");
2980                 ret = btrfs_rebuild_free_space_tree(fs_info);
2981                 if (ret) {
2982                         btrfs_warn(fs_info,
2983                                    "failed to rebuild free space tree: %d", ret);
2984                         goto out;
2985                 }
2986         }
2987
2988         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2989             !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
2990                 btrfs_info(fs_info, "disabling free space tree");
2991                 ret = btrfs_delete_free_space_tree(fs_info);
2992                 if (ret) {
2993                         btrfs_warn(fs_info,
2994                                    "failed to disable free space tree: %d", ret);
2995                         goto out;
2996                 }
2997         }
2998
2999         /*
3000          * btrfs_find_orphan_roots() is responsible for finding all the dead
3001          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3002          * them into the fs_info->fs_roots_radix tree. This must be done before
3003          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3004          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3005          * item before the root's tree is deleted - this means that if we unmount
3006          * or crash before the deletion completes, on the next mount we will not
3007          * delete what remains of the tree because the orphan item does not
3008          * exists anymore, which is what tells us we have a pending deletion.
3009          */
3010         ret = btrfs_find_orphan_roots(fs_info);
3011         if (ret)
3012                 goto out;
3013
3014         ret = btrfs_cleanup_fs_roots(fs_info);
3015         if (ret)
3016                 goto out;
3017
3018         down_read(&fs_info->cleanup_work_sem);
3019         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3020             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3021                 up_read(&fs_info->cleanup_work_sem);
3022                 goto out;
3023         }
3024         up_read(&fs_info->cleanup_work_sem);
3025
3026         mutex_lock(&fs_info->cleaner_mutex);
3027         ret = btrfs_recover_relocation(fs_info);
3028         mutex_unlock(&fs_info->cleaner_mutex);
3029         if (ret < 0) {
3030                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3031                 goto out;
3032         }
3033
3034         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3035             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3036                 btrfs_info(fs_info, "creating free space tree");
3037                 ret = btrfs_create_free_space_tree(fs_info);
3038                 if (ret) {
3039                         btrfs_warn(fs_info,
3040                                 "failed to create free space tree: %d", ret);
3041                         goto out;
3042                 }
3043         }
3044
3045         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3046                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3047                 if (ret)
3048                         goto out;
3049         }
3050
3051         ret = btrfs_resume_balance_async(fs_info);
3052         if (ret)
3053                 goto out;
3054
3055         ret = btrfs_resume_dev_replace_async(fs_info);
3056         if (ret) {
3057                 btrfs_warn(fs_info, "failed to resume dev_replace");
3058                 goto out;
3059         }
3060
3061         btrfs_qgroup_rescan_resume(fs_info);
3062
3063         if (!fs_info->uuid_root) {
3064                 btrfs_info(fs_info, "creating UUID tree");
3065                 ret = btrfs_create_uuid_tree(fs_info);
3066                 if (ret) {
3067                         btrfs_warn(fs_info,
3068                                    "failed to create the UUID tree %d", ret);
3069                         goto out;
3070                 }
3071         }
3072
3073 out:
3074         return ret;
3075 }
3076
3077 /*
3078  * Do various sanity and dependency checks of different features.
3079  *
3080  * @is_rw_mount:        If the mount is read-write.
3081  *
3082  * This is the place for less strict checks (like for subpage or artificial
3083  * feature dependencies).
3084  *
3085  * For strict checks or possible corruption detection, see
3086  * btrfs_validate_super().
3087  *
3088  * This should be called after btrfs_parse_options(), as some mount options
3089  * (space cache related) can modify on-disk format like free space tree and
3090  * screw up certain feature dependencies.
3091  */
3092 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3093 {
3094         struct btrfs_super_block *disk_super = fs_info->super_copy;
3095         u64 incompat = btrfs_super_incompat_flags(disk_super);
3096         const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3097         const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3098
3099         if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3100                 btrfs_err(fs_info,
3101                 "cannot mount because of unknown incompat features (0x%llx)",
3102                     incompat);
3103                 return -EINVAL;
3104         }
3105
3106         /* Runtime limitation for mixed block groups. */
3107         if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3108             (fs_info->sectorsize != fs_info->nodesize)) {
3109                 btrfs_err(fs_info,
3110 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3111                         fs_info->nodesize, fs_info->sectorsize);
3112                 return -EINVAL;
3113         }
3114
3115         /* Mixed backref is an always-enabled feature. */
3116         incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3117
3118         /* Set compression related flags just in case. */
3119         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3120                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3121         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3122                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3123
3124         /*
3125          * An ancient flag, which should really be marked deprecated.
3126          * Such runtime limitation doesn't really need a incompat flag.
3127          */
3128         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3129                 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3130
3131         if (compat_ro_unsupp && is_rw_mount) {
3132                 btrfs_err(fs_info,
3133         "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3134                        compat_ro);
3135                 return -EINVAL;
3136         }
3137
3138         /*
3139          * We have unsupported RO compat features, although RO mounted, we
3140          * should not cause any metadata writes, including log replay.
3141          * Or we could screw up whatever the new feature requires.
3142          */
3143         if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3144             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3145                 btrfs_err(fs_info,
3146 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3147                           compat_ro);
3148                 return -EINVAL;
3149         }
3150
3151         /*
3152          * Artificial limitations for block group tree, to force
3153          * block-group-tree to rely on no-holes and free-space-tree.
3154          */
3155         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3156             (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3157              !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3158                 btrfs_err(fs_info,
3159 "block-group-tree feature requires no-holes and free-space-tree features");
3160                 return -EINVAL;
3161         }
3162
3163         /*
3164          * Subpage runtime limitation on v1 cache.
3165          *
3166          * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3167          * we're already defaulting to v2 cache, no need to bother v1 as it's
3168          * going to be deprecated anyway.
3169          */
3170         if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3171                 btrfs_warn(fs_info,
3172         "v1 space cache is not supported for page size %lu with sectorsize %u",
3173                            PAGE_SIZE, fs_info->sectorsize);
3174                 return -EINVAL;
3175         }
3176
3177         /* This can be called by remount, we need to protect the super block. */
3178         spin_lock(&fs_info->super_lock);
3179         btrfs_set_super_incompat_flags(disk_super, incompat);
3180         spin_unlock(&fs_info->super_lock);
3181
3182         return 0;
3183 }
3184
3185 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3186                       char *options)
3187 {
3188         u32 sectorsize;
3189         u32 nodesize;
3190         u32 stripesize;
3191         u64 generation;
3192         u16 csum_type;
3193         struct btrfs_super_block *disk_super;
3194         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3195         struct btrfs_root *tree_root;
3196         struct btrfs_root *chunk_root;
3197         int ret;
3198         int level;
3199
3200         ret = init_mount_fs_info(fs_info, sb);
3201         if (ret)
3202                 goto fail;
3203
3204         /* These need to be init'ed before we start creating inodes and such. */
3205         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3206                                      GFP_KERNEL);
3207         fs_info->tree_root = tree_root;
3208         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3209                                       GFP_KERNEL);
3210         fs_info->chunk_root = chunk_root;
3211         if (!tree_root || !chunk_root) {
3212                 ret = -ENOMEM;
3213                 goto fail;
3214         }
3215
3216         ret = btrfs_init_btree_inode(sb);
3217         if (ret)
3218                 goto fail;
3219
3220         invalidate_bdev(fs_devices->latest_dev->bdev);
3221
3222         /*
3223          * Read super block and check the signature bytes only
3224          */
3225         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3226         if (IS_ERR(disk_super)) {
3227                 ret = PTR_ERR(disk_super);
3228                 goto fail_alloc;
3229         }
3230
3231         btrfs_info(fs_info, "first mount of filesystem %pU", disk_super->fsid);
3232         /*
3233          * Verify the type first, if that or the checksum value are
3234          * corrupted, we'll find out
3235          */
3236         csum_type = btrfs_super_csum_type(disk_super);
3237         if (!btrfs_supported_super_csum(csum_type)) {
3238                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3239                           csum_type);
3240                 ret = -EINVAL;
3241                 btrfs_release_disk_super(disk_super);
3242                 goto fail_alloc;
3243         }
3244
3245         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3246
3247         ret = btrfs_init_csum_hash(fs_info, csum_type);
3248         if (ret) {
3249                 btrfs_release_disk_super(disk_super);
3250                 goto fail_alloc;
3251         }
3252
3253         /*
3254          * We want to check superblock checksum, the type is stored inside.
3255          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3256          */
3257         if (btrfs_check_super_csum(fs_info, disk_super)) {
3258                 btrfs_err(fs_info, "superblock checksum mismatch");
3259                 ret = -EINVAL;
3260                 btrfs_release_disk_super(disk_super);
3261                 goto fail_alloc;
3262         }
3263
3264         /*
3265          * super_copy is zeroed at allocation time and we never touch the
3266          * following bytes up to INFO_SIZE, the checksum is calculated from
3267          * the whole block of INFO_SIZE
3268          */
3269         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3270         btrfs_release_disk_super(disk_super);
3271
3272         disk_super = fs_info->super_copy;
3273
3274         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3275                sizeof(*fs_info->super_for_commit));
3276
3277         ret = btrfs_validate_mount_super(fs_info);
3278         if (ret) {
3279                 btrfs_err(fs_info, "superblock contains fatal errors");
3280                 ret = -EINVAL;
3281                 goto fail_alloc;
3282         }
3283
3284         if (!btrfs_super_root(disk_super)) {
3285                 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3286                 ret = -EINVAL;
3287                 goto fail_alloc;
3288         }
3289
3290         /* check FS state, whether FS is broken. */
3291         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3292                 WRITE_ONCE(fs_info->fs_error, -EUCLEAN);
3293
3294         /* Set up fs_info before parsing mount options */
3295         nodesize = btrfs_super_nodesize(disk_super);
3296         sectorsize = btrfs_super_sectorsize(disk_super);
3297         stripesize = sectorsize;
3298         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3299         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3300
3301         fs_info->nodesize = nodesize;
3302         fs_info->sectorsize = sectorsize;
3303         fs_info->sectorsize_bits = ilog2(sectorsize);
3304         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3305         fs_info->stripesize = stripesize;
3306
3307         /*
3308          * Handle the space caching options appropriately now that we have the
3309          * super block loaded and validated.
3310          */
3311         btrfs_set_free_space_cache_settings(fs_info);
3312
3313         if (!btrfs_check_options(fs_info, &fs_info->mount_opt, sb->s_flags)) {
3314                 ret = -EINVAL;
3315                 goto fail_alloc;
3316         }
3317
3318         ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3319         if (ret < 0)
3320                 goto fail_alloc;
3321
3322         /*
3323          * At this point our mount options are validated, if we set ->max_inline
3324          * to something non-standard make sure we truncate it to sectorsize.
3325          */
3326         fs_info->max_inline = min_t(u64, fs_info->max_inline, fs_info->sectorsize);
3327
3328         if (sectorsize < PAGE_SIZE) {
3329                 struct btrfs_subpage_info *subpage_info;
3330
3331                 btrfs_warn(fs_info,
3332                 "read-write for sector size %u with page size %lu is experimental",
3333                            sectorsize, PAGE_SIZE);
3334                 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3335                 if (!subpage_info) {
3336                         ret = -ENOMEM;
3337                         goto fail_alloc;
3338                 }
3339                 btrfs_init_subpage_info(subpage_info, sectorsize);
3340                 fs_info->subpage_info = subpage_info;
3341         }
3342
3343         ret = btrfs_init_workqueues(fs_info);
3344         if (ret)
3345                 goto fail_sb_buffer;
3346
3347         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3348         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3349
3350         sb->s_blocksize = sectorsize;
3351         sb->s_blocksize_bits = blksize_bits(sectorsize);
3352         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3353
3354         mutex_lock(&fs_info->chunk_mutex);
3355         ret = btrfs_read_sys_array(fs_info);
3356         mutex_unlock(&fs_info->chunk_mutex);
3357         if (ret) {
3358                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3359                 goto fail_sb_buffer;
3360         }
3361
3362         generation = btrfs_super_chunk_root_generation(disk_super);
3363         level = btrfs_super_chunk_root_level(disk_super);
3364         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3365                               generation, level);
3366         if (ret) {
3367                 btrfs_err(fs_info, "failed to read chunk root");
3368                 goto fail_tree_roots;
3369         }
3370
3371         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3372                            offsetof(struct btrfs_header, chunk_tree_uuid),
3373                            BTRFS_UUID_SIZE);
3374
3375         ret = btrfs_read_chunk_tree(fs_info);
3376         if (ret) {
3377                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3378                 goto fail_tree_roots;
3379         }
3380
3381         /*
3382          * At this point we know all the devices that make this filesystem,
3383          * including the seed devices but we don't know yet if the replace
3384          * target is required. So free devices that are not part of this
3385          * filesystem but skip the replace target device which is checked
3386          * below in btrfs_init_dev_replace().
3387          */
3388         btrfs_free_extra_devids(fs_devices);
3389         if (!fs_devices->latest_dev->bdev) {
3390                 btrfs_err(fs_info, "failed to read devices");
3391                 ret = -EIO;
3392                 goto fail_tree_roots;
3393         }
3394
3395         ret = init_tree_roots(fs_info);
3396         if (ret)
3397                 goto fail_tree_roots;
3398
3399         /*
3400          * Get zone type information of zoned block devices. This will also
3401          * handle emulation of a zoned filesystem if a regular device has the
3402          * zoned incompat feature flag set.
3403          */
3404         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3405         if (ret) {
3406                 btrfs_err(fs_info,
3407                           "zoned: failed to read device zone info: %d", ret);
3408                 goto fail_block_groups;
3409         }
3410
3411         /*
3412          * If we have a uuid root and we're not being told to rescan we need to
3413          * check the generation here so we can set the
3414          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3415          * transaction during a balance or the log replay without updating the
3416          * uuid generation, and then if we crash we would rescan the uuid tree,
3417          * even though it was perfectly fine.
3418          */
3419         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3420             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3421                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3422
3423         ret = btrfs_verify_dev_extents(fs_info);
3424         if (ret) {
3425                 btrfs_err(fs_info,
3426                           "failed to verify dev extents against chunks: %d",
3427                           ret);
3428                 goto fail_block_groups;
3429         }
3430         ret = btrfs_recover_balance(fs_info);
3431         if (ret) {
3432                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3433                 goto fail_block_groups;
3434         }
3435
3436         ret = btrfs_init_dev_stats(fs_info);
3437         if (ret) {
3438                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3439                 goto fail_block_groups;
3440         }
3441
3442         ret = btrfs_init_dev_replace(fs_info);
3443         if (ret) {
3444                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3445                 goto fail_block_groups;
3446         }
3447
3448         ret = btrfs_check_zoned_mode(fs_info);
3449         if (ret) {
3450                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3451                           ret);
3452                 goto fail_block_groups;
3453         }
3454
3455         ret = btrfs_sysfs_add_fsid(fs_devices);
3456         if (ret) {
3457                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3458                                 ret);
3459                 goto fail_block_groups;
3460         }
3461
3462         ret = btrfs_sysfs_add_mounted(fs_info);
3463         if (ret) {
3464                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3465                 goto fail_fsdev_sysfs;
3466         }
3467
3468         ret = btrfs_init_space_info(fs_info);
3469         if (ret) {
3470                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3471                 goto fail_sysfs;
3472         }
3473
3474         ret = btrfs_read_block_groups(fs_info);
3475         if (ret) {
3476                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3477                 goto fail_sysfs;
3478         }
3479
3480         btrfs_free_zone_cache(fs_info);
3481
3482         btrfs_check_active_zone_reservation(fs_info);
3483
3484         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3485             !btrfs_check_rw_degradable(fs_info, NULL)) {
3486                 btrfs_warn(fs_info,
3487                 "writable mount is not allowed due to too many missing devices");
3488                 ret = -EINVAL;
3489                 goto fail_sysfs;
3490         }
3491
3492         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3493                                                "btrfs-cleaner");
3494         if (IS_ERR(fs_info->cleaner_kthread)) {
3495                 ret = PTR_ERR(fs_info->cleaner_kthread);
3496                 goto fail_sysfs;
3497         }
3498
3499         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3500                                                    tree_root,
3501                                                    "btrfs-transaction");
3502         if (IS_ERR(fs_info->transaction_kthread)) {
3503                 ret = PTR_ERR(fs_info->transaction_kthread);
3504                 goto fail_cleaner;
3505         }
3506
3507         ret = btrfs_read_qgroup_config(fs_info);
3508         if (ret)
3509                 goto fail_trans_kthread;
3510
3511         if (btrfs_build_ref_tree(fs_info))
3512                 btrfs_err(fs_info, "couldn't build ref tree");
3513
3514         /* do not make disk changes in broken FS or nologreplay is given */
3515         if (btrfs_super_log_root(disk_super) != 0 &&
3516             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3517                 btrfs_info(fs_info, "start tree-log replay");
3518                 ret = btrfs_replay_log(fs_info, fs_devices);
3519                 if (ret)
3520                         goto fail_qgroup;
3521         }
3522
3523         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3524         if (IS_ERR(fs_info->fs_root)) {
3525                 ret = PTR_ERR(fs_info->fs_root);
3526                 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3527                 fs_info->fs_root = NULL;
3528                 goto fail_qgroup;
3529         }
3530
3531         if (sb_rdonly(sb))
3532                 return 0;
3533
3534         ret = btrfs_start_pre_rw_mount(fs_info);
3535         if (ret) {
3536                 close_ctree(fs_info);
3537                 return ret;
3538         }
3539         btrfs_discard_resume(fs_info);
3540
3541         if (fs_info->uuid_root &&
3542             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3543              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3544                 btrfs_info(fs_info, "checking UUID tree");
3545                 ret = btrfs_check_uuid_tree(fs_info);
3546                 if (ret) {
3547                         btrfs_warn(fs_info,
3548                                 "failed to check the UUID tree: %d", ret);
3549                         close_ctree(fs_info);
3550                         return ret;
3551                 }
3552         }
3553
3554         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3555
3556         /* Kick the cleaner thread so it'll start deleting snapshots. */
3557         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3558                 wake_up_process(fs_info->cleaner_kthread);
3559
3560         return 0;
3561
3562 fail_qgroup:
3563         btrfs_free_qgroup_config(fs_info);
3564 fail_trans_kthread:
3565         kthread_stop(fs_info->transaction_kthread);
3566         btrfs_cleanup_transaction(fs_info);
3567         btrfs_free_fs_roots(fs_info);
3568 fail_cleaner:
3569         kthread_stop(fs_info->cleaner_kthread);
3570
3571         /*
3572          * make sure we're done with the btree inode before we stop our
3573          * kthreads
3574          */
3575         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3576
3577 fail_sysfs:
3578         btrfs_sysfs_remove_mounted(fs_info);
3579
3580 fail_fsdev_sysfs:
3581         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3582
3583 fail_block_groups:
3584         btrfs_put_block_group_cache(fs_info);
3585
3586 fail_tree_roots:
3587         if (fs_info->data_reloc_root)
3588                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3589         free_root_pointers(fs_info, true);
3590         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3591
3592 fail_sb_buffer:
3593         btrfs_stop_all_workers(fs_info);
3594         btrfs_free_block_groups(fs_info);
3595 fail_alloc:
3596         btrfs_mapping_tree_free(fs_info);
3597
3598         iput(fs_info->btree_inode);
3599 fail:
3600         btrfs_close_devices(fs_info->fs_devices);
3601         ASSERT(ret < 0);
3602         return ret;
3603 }
3604 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3605
3606 static void btrfs_end_super_write(struct bio *bio)
3607 {
3608         struct btrfs_device *device = bio->bi_private;
3609         struct bio_vec *bvec;
3610         struct bvec_iter_all iter_all;
3611         struct page *page;
3612
3613         bio_for_each_segment_all(bvec, bio, iter_all) {
3614                 page = bvec->bv_page;
3615
3616                 if (bio->bi_status) {
3617                         btrfs_warn_rl_in_rcu(device->fs_info,
3618                                 "lost page write due to IO error on %s (%d)",
3619                                 btrfs_dev_name(device),
3620                                 blk_status_to_errno(bio->bi_status));
3621                         ClearPageUptodate(page);
3622                         SetPageError(page);
3623                         btrfs_dev_stat_inc_and_print(device,
3624                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3625                 } else {
3626                         SetPageUptodate(page);
3627                 }
3628
3629                 put_page(page);
3630                 unlock_page(page);
3631         }
3632
3633         bio_put(bio);
3634 }
3635
3636 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3637                                                    int copy_num, bool drop_cache)
3638 {
3639         struct btrfs_super_block *super;
3640         struct page *page;
3641         u64 bytenr, bytenr_orig;
3642         struct address_space *mapping = bdev->bd_inode->i_mapping;
3643         int ret;
3644
3645         bytenr_orig = btrfs_sb_offset(copy_num);
3646         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3647         if (ret == -ENOENT)
3648                 return ERR_PTR(-EINVAL);
3649         else if (ret)
3650                 return ERR_PTR(ret);
3651
3652         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3653                 return ERR_PTR(-EINVAL);
3654
3655         if (drop_cache) {
3656                 /* This should only be called with the primary sb. */
3657                 ASSERT(copy_num == 0);
3658
3659                 /*
3660                  * Drop the page of the primary superblock, so later read will
3661                  * always read from the device.
3662                  */
3663                 invalidate_inode_pages2_range(mapping,
3664                                 bytenr >> PAGE_SHIFT,
3665                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3666         }
3667
3668         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3669         if (IS_ERR(page))
3670                 return ERR_CAST(page);
3671
3672         super = page_address(page);
3673         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3674                 btrfs_release_disk_super(super);
3675                 return ERR_PTR(-ENODATA);
3676         }
3677
3678         if (btrfs_super_bytenr(super) != bytenr_orig) {
3679                 btrfs_release_disk_super(super);
3680                 return ERR_PTR(-EINVAL);
3681         }
3682
3683         return super;
3684 }
3685
3686
3687 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3688 {
3689         struct btrfs_super_block *super, *latest = NULL;
3690         int i;
3691         u64 transid = 0;
3692
3693         /* we would like to check all the supers, but that would make
3694          * a btrfs mount succeed after a mkfs from a different FS.
3695          * So, we need to add a special mount option to scan for
3696          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3697          */
3698         for (i = 0; i < 1; i++) {
3699                 super = btrfs_read_dev_one_super(bdev, i, false);
3700                 if (IS_ERR(super))
3701                         continue;
3702
3703                 if (!latest || btrfs_super_generation(super) > transid) {
3704                         if (latest)
3705                                 btrfs_release_disk_super(super);
3706
3707                         latest = super;
3708                         transid = btrfs_super_generation(super);
3709                 }
3710         }
3711
3712         return super;
3713 }
3714
3715 /*
3716  * Write superblock @sb to the @device. Do not wait for completion, all the
3717  * pages we use for writing are locked.
3718  *
3719  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3720  * the expected device size at commit time. Note that max_mirrors must be
3721  * same for write and wait phases.
3722  *
3723  * Return number of errors when page is not found or submission fails.
3724  */
3725 static int write_dev_supers(struct btrfs_device *device,
3726                             struct btrfs_super_block *sb, int max_mirrors)
3727 {
3728         struct btrfs_fs_info *fs_info = device->fs_info;
3729         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3730         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3731         int i;
3732         int errors = 0;
3733         int ret;
3734         u64 bytenr, bytenr_orig;
3735
3736         if (max_mirrors == 0)
3737                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3738
3739         shash->tfm = fs_info->csum_shash;
3740
3741         for (i = 0; i < max_mirrors; i++) {
3742                 struct page *page;
3743                 struct bio *bio;
3744                 struct btrfs_super_block *disk_super;
3745
3746                 bytenr_orig = btrfs_sb_offset(i);
3747                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3748                 if (ret == -ENOENT) {
3749                         continue;
3750                 } else if (ret < 0) {
3751                         btrfs_err(device->fs_info,
3752                                 "couldn't get super block location for mirror %d",
3753                                 i);
3754                         errors++;
3755                         continue;
3756                 }
3757                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3758                     device->commit_total_bytes)
3759                         break;
3760
3761                 btrfs_set_super_bytenr(sb, bytenr_orig);
3762
3763                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3764                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3765                                     sb->csum);
3766
3767                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3768                                            GFP_NOFS);
3769                 if (!page) {
3770                         btrfs_err(device->fs_info,
3771                             "couldn't get super block page for bytenr %llu",
3772                             bytenr);
3773                         errors++;
3774                         continue;
3775                 }
3776
3777                 /* Bump the refcount for wait_dev_supers() */
3778                 get_page(page);
3779
3780                 disk_super = page_address(page);
3781                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3782
3783                 /*
3784                  * Directly use bios here instead of relying on the page cache
3785                  * to do I/O, so we don't lose the ability to do integrity
3786                  * checking.
3787                  */
3788                 bio = bio_alloc(device->bdev, 1,
3789                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3790                                 GFP_NOFS);
3791                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3792                 bio->bi_private = device;
3793                 bio->bi_end_io = btrfs_end_super_write;
3794                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3795                                offset_in_page(bytenr));
3796
3797                 /*
3798                  * We FUA only the first super block.  The others we allow to
3799                  * go down lazy and there's a short window where the on-disk
3800                  * copies might still contain the older version.
3801                  */
3802                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3803                         bio->bi_opf |= REQ_FUA;
3804                 submit_bio(bio);
3805
3806                 if (btrfs_advance_sb_log(device, i))
3807                         errors++;
3808         }
3809         return errors < i ? 0 : -1;
3810 }
3811
3812 /*
3813  * Wait for write completion of superblocks done by write_dev_supers,
3814  * @max_mirrors same for write and wait phases.
3815  *
3816  * Return number of errors when page is not found or not marked up to
3817  * date.
3818  */
3819 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3820 {
3821         int i;
3822         int errors = 0;
3823         bool primary_failed = false;
3824         int ret;
3825         u64 bytenr;
3826
3827         if (max_mirrors == 0)
3828                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3829
3830         for (i = 0; i < max_mirrors; i++) {
3831                 struct page *page;
3832
3833                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
3834                 if (ret == -ENOENT) {
3835                         break;
3836                 } else if (ret < 0) {
3837                         errors++;
3838                         if (i == 0)
3839                                 primary_failed = true;
3840                         continue;
3841                 }
3842                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3843                     device->commit_total_bytes)
3844                         break;
3845
3846                 page = find_get_page(device->bdev->bd_inode->i_mapping,
3847                                      bytenr >> PAGE_SHIFT);
3848                 if (!page) {
3849                         errors++;
3850                         if (i == 0)
3851                                 primary_failed = true;
3852                         continue;
3853                 }
3854                 /* Page is submitted locked and unlocked once the IO completes */
3855                 wait_on_page_locked(page);
3856                 if (PageError(page)) {
3857                         errors++;
3858                         if (i == 0)
3859                                 primary_failed = true;
3860                 }
3861
3862                 /* Drop our reference */
3863                 put_page(page);
3864
3865                 /* Drop the reference from the writing run */
3866                 put_page(page);
3867         }
3868
3869         /* log error, force error return */
3870         if (primary_failed) {
3871                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3872                           device->devid);
3873                 return -1;
3874         }
3875
3876         return errors < i ? 0 : -1;
3877 }
3878
3879 /*
3880  * endio for the write_dev_flush, this will wake anyone waiting
3881  * for the barrier when it is done
3882  */
3883 static void btrfs_end_empty_barrier(struct bio *bio)
3884 {
3885         bio_uninit(bio);
3886         complete(bio->bi_private);
3887 }
3888
3889 /*
3890  * Submit a flush request to the device if it supports it. Error handling is
3891  * done in the waiting counterpart.
3892  */
3893 static void write_dev_flush(struct btrfs_device *device)
3894 {
3895         struct bio *bio = &device->flush_bio;
3896
3897         device->last_flush_error = BLK_STS_OK;
3898
3899         bio_init(bio, device->bdev, NULL, 0,
3900                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
3901         bio->bi_end_io = btrfs_end_empty_barrier;
3902         init_completion(&device->flush_wait);
3903         bio->bi_private = &device->flush_wait;
3904         submit_bio(bio);
3905         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3906 }
3907
3908 /*
3909  * If the flush bio has been submitted by write_dev_flush, wait for it.
3910  * Return true for any error, and false otherwise.
3911  */
3912 static bool wait_dev_flush(struct btrfs_device *device)
3913 {
3914         struct bio *bio = &device->flush_bio;
3915
3916         if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3917                 return false;
3918
3919         wait_for_completion_io(&device->flush_wait);
3920
3921         if (bio->bi_status) {
3922                 device->last_flush_error = bio->bi_status;
3923                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
3924                 return true;
3925         }
3926
3927         return false;
3928 }
3929
3930 /*
3931  * send an empty flush down to each device in parallel,
3932  * then wait for them
3933  */
3934 static int barrier_all_devices(struct btrfs_fs_info *info)
3935 {
3936         struct list_head *head;
3937         struct btrfs_device *dev;
3938         int errors_wait = 0;
3939
3940         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3941         /* send down all the barriers */
3942         head = &info->fs_devices->devices;
3943         list_for_each_entry(dev, head, dev_list) {
3944                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3945                         continue;
3946                 if (!dev->bdev)
3947                         continue;
3948                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3949                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3950                         continue;
3951
3952                 write_dev_flush(dev);
3953         }
3954
3955         /* wait for all the barriers */
3956         list_for_each_entry(dev, head, dev_list) {
3957                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3958                         continue;
3959                 if (!dev->bdev) {
3960                         errors_wait++;
3961                         continue;
3962                 }
3963                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3964                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3965                         continue;
3966
3967                 if (wait_dev_flush(dev))
3968                         errors_wait++;
3969         }
3970
3971         /*
3972          * Checks last_flush_error of disks in order to determine the device
3973          * state.
3974          */
3975         if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
3976                 return -EIO;
3977
3978         return 0;
3979 }
3980
3981 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3982 {
3983         int raid_type;
3984         int min_tolerated = INT_MAX;
3985
3986         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3987             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3988                 min_tolerated = min_t(int, min_tolerated,
3989                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3990                                     tolerated_failures);
3991
3992         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3993                 if (raid_type == BTRFS_RAID_SINGLE)
3994                         continue;
3995                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3996                         continue;
3997                 min_tolerated = min_t(int, min_tolerated,
3998                                     btrfs_raid_array[raid_type].
3999                                     tolerated_failures);
4000         }
4001
4002         if (min_tolerated == INT_MAX) {
4003                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4004                 min_tolerated = 0;
4005         }
4006
4007         return min_tolerated;
4008 }
4009
4010 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4011 {
4012         struct list_head *head;
4013         struct btrfs_device *dev;
4014         struct btrfs_super_block *sb;
4015         struct btrfs_dev_item *dev_item;
4016         int ret;
4017         int do_barriers;
4018         int max_errors;
4019         int total_errors = 0;
4020         u64 flags;
4021
4022         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4023
4024         /*
4025          * max_mirrors == 0 indicates we're from commit_transaction,
4026          * not from fsync where the tree roots in fs_info have not
4027          * been consistent on disk.
4028          */
4029         if (max_mirrors == 0)
4030                 backup_super_roots(fs_info);
4031
4032         sb = fs_info->super_for_commit;
4033         dev_item = &sb->dev_item;
4034
4035         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4036         head = &fs_info->fs_devices->devices;
4037         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4038
4039         if (do_barriers) {
4040                 ret = barrier_all_devices(fs_info);
4041                 if (ret) {
4042                         mutex_unlock(
4043                                 &fs_info->fs_devices->device_list_mutex);
4044                         btrfs_handle_fs_error(fs_info, ret,
4045                                               "errors while submitting device barriers.");
4046                         return ret;
4047                 }
4048         }
4049
4050         list_for_each_entry(dev, head, dev_list) {
4051                 if (!dev->bdev) {
4052                         total_errors++;
4053                         continue;
4054                 }
4055                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4056                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4057                         continue;
4058
4059                 btrfs_set_stack_device_generation(dev_item, 0);
4060                 btrfs_set_stack_device_type(dev_item, dev->type);
4061                 btrfs_set_stack_device_id(dev_item, dev->devid);
4062                 btrfs_set_stack_device_total_bytes(dev_item,
4063                                                    dev->commit_total_bytes);
4064                 btrfs_set_stack_device_bytes_used(dev_item,
4065                                                   dev->commit_bytes_used);
4066                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4067                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4068                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4069                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4070                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4071                        BTRFS_FSID_SIZE);
4072
4073                 flags = btrfs_super_flags(sb);
4074                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4075
4076                 ret = btrfs_validate_write_super(fs_info, sb);
4077                 if (ret < 0) {
4078                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4079                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4080                                 "unexpected superblock corruption detected");
4081                         return -EUCLEAN;
4082                 }
4083
4084                 ret = write_dev_supers(dev, sb, max_mirrors);
4085                 if (ret)
4086                         total_errors++;
4087         }
4088         if (total_errors > max_errors) {
4089                 btrfs_err(fs_info, "%d errors while writing supers",
4090                           total_errors);
4091                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4092
4093                 /* FUA is masked off if unsupported and can't be the reason */
4094                 btrfs_handle_fs_error(fs_info, -EIO,
4095                                       "%d errors while writing supers",
4096                                       total_errors);
4097                 return -EIO;
4098         }
4099
4100         total_errors = 0;
4101         list_for_each_entry(dev, head, dev_list) {
4102                 if (!dev->bdev)
4103                         continue;
4104                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4105                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4106                         continue;
4107
4108                 ret = wait_dev_supers(dev, max_mirrors);
4109                 if (ret)
4110                         total_errors++;
4111         }
4112         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4113         if (total_errors > max_errors) {
4114                 btrfs_handle_fs_error(fs_info, -EIO,
4115                                       "%d errors while writing supers",
4116                                       total_errors);
4117                 return -EIO;
4118         }
4119         return 0;
4120 }
4121
4122 /* Drop a fs root from the radix tree and free it. */
4123 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4124                                   struct btrfs_root *root)
4125 {
4126         bool drop_ref = false;
4127
4128         spin_lock(&fs_info->fs_roots_radix_lock);
4129         radix_tree_delete(&fs_info->fs_roots_radix,
4130                           (unsigned long)root->root_key.objectid);
4131         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4132                 drop_ref = true;
4133         spin_unlock(&fs_info->fs_roots_radix_lock);
4134
4135         if (BTRFS_FS_ERROR(fs_info)) {
4136                 ASSERT(root->log_root == NULL);
4137                 if (root->reloc_root) {
4138                         btrfs_put_root(root->reloc_root);
4139                         root->reloc_root = NULL;
4140                 }
4141         }
4142
4143         if (drop_ref)
4144                 btrfs_put_root(root);
4145 }
4146
4147 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4148 {
4149         struct btrfs_root *root = fs_info->tree_root;
4150         struct btrfs_trans_handle *trans;
4151
4152         mutex_lock(&fs_info->cleaner_mutex);
4153         btrfs_run_delayed_iputs(fs_info);
4154         mutex_unlock(&fs_info->cleaner_mutex);
4155         wake_up_process(fs_info->cleaner_kthread);
4156
4157         /* wait until ongoing cleanup work done */
4158         down_write(&fs_info->cleanup_work_sem);
4159         up_write(&fs_info->cleanup_work_sem);
4160
4161         trans = btrfs_join_transaction(root);
4162         if (IS_ERR(trans))
4163                 return PTR_ERR(trans);
4164         return btrfs_commit_transaction(trans);
4165 }
4166
4167 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4168 {
4169         struct btrfs_transaction *trans;
4170         struct btrfs_transaction *tmp;
4171         bool found = false;
4172
4173         if (list_empty(&fs_info->trans_list))
4174                 return;
4175
4176         /*
4177          * This function is only called at the very end of close_ctree(),
4178          * thus no other running transaction, no need to take trans_lock.
4179          */
4180         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4181         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4182                 struct extent_state *cached = NULL;
4183                 u64 dirty_bytes = 0;
4184                 u64 cur = 0;
4185                 u64 found_start;
4186                 u64 found_end;
4187
4188                 found = true;
4189                 while (find_first_extent_bit(&trans->dirty_pages, cur,
4190                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4191                         dirty_bytes += found_end + 1 - found_start;
4192                         cur = found_end + 1;
4193                 }
4194                 btrfs_warn(fs_info,
4195         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4196                            trans->transid, dirty_bytes);
4197                 btrfs_cleanup_one_transaction(trans, fs_info);
4198
4199                 if (trans == fs_info->running_transaction)
4200                         fs_info->running_transaction = NULL;
4201                 list_del_init(&trans->list);
4202
4203                 btrfs_put_transaction(trans);
4204                 trace_btrfs_transaction_commit(fs_info);
4205         }
4206         ASSERT(!found);
4207 }
4208
4209 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4210 {
4211         int ret;
4212
4213         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4214
4215         /*
4216          * If we had UNFINISHED_DROPS we could still be processing them, so
4217          * clear that bit and wake up relocation so it can stop.
4218          * We must do this before stopping the block group reclaim task, because
4219          * at btrfs_relocate_block_group() we wait for this bit, and after the
4220          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4221          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4222          * return 1.
4223          */
4224         btrfs_wake_unfinished_drop(fs_info);
4225
4226         /*
4227          * We may have the reclaim task running and relocating a data block group,
4228          * in which case it may create delayed iputs. So stop it before we park
4229          * the cleaner kthread otherwise we can get new delayed iputs after
4230          * parking the cleaner, and that can make the async reclaim task to hang
4231          * if it's waiting for delayed iputs to complete, since the cleaner is
4232          * parked and can not run delayed iputs - this will make us hang when
4233          * trying to stop the async reclaim task.
4234          */
4235         cancel_work_sync(&fs_info->reclaim_bgs_work);
4236         /*
4237          * We don't want the cleaner to start new transactions, add more delayed
4238          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4239          * because that frees the task_struct, and the transaction kthread might
4240          * still try to wake up the cleaner.
4241          */
4242         kthread_park(fs_info->cleaner_kthread);
4243
4244         /* wait for the qgroup rescan worker to stop */
4245         btrfs_qgroup_wait_for_completion(fs_info, false);
4246
4247         /* wait for the uuid_scan task to finish */
4248         down(&fs_info->uuid_tree_rescan_sem);
4249         /* avoid complains from lockdep et al., set sem back to initial state */
4250         up(&fs_info->uuid_tree_rescan_sem);
4251
4252         /* pause restriper - we want to resume on mount */
4253         btrfs_pause_balance(fs_info);
4254
4255         btrfs_dev_replace_suspend_for_unmount(fs_info);
4256
4257         btrfs_scrub_cancel(fs_info);
4258
4259         /* wait for any defraggers to finish */
4260         wait_event(fs_info->transaction_wait,
4261                    (atomic_read(&fs_info->defrag_running) == 0));
4262
4263         /* clear out the rbtree of defraggable inodes */
4264         btrfs_cleanup_defrag_inodes(fs_info);
4265
4266         /*
4267          * After we parked the cleaner kthread, ordered extents may have
4268          * completed and created new delayed iputs. If one of the async reclaim
4269          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4270          * can hang forever trying to stop it, because if a delayed iput is
4271          * added after it ran btrfs_run_delayed_iputs() and before it called
4272          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4273          * no one else to run iputs.
4274          *
4275          * So wait for all ongoing ordered extents to complete and then run
4276          * delayed iputs. This works because once we reach this point no one
4277          * can either create new ordered extents nor create delayed iputs
4278          * through some other means.
4279          *
4280          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4281          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4282          * but the delayed iput for the respective inode is made only when doing
4283          * the final btrfs_put_ordered_extent() (which must happen at
4284          * btrfs_finish_ordered_io() when we are unmounting).
4285          */
4286         btrfs_flush_workqueue(fs_info->endio_write_workers);
4287         /* Ordered extents for free space inodes. */
4288         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4289         btrfs_run_delayed_iputs(fs_info);
4290
4291         cancel_work_sync(&fs_info->async_reclaim_work);
4292         cancel_work_sync(&fs_info->async_data_reclaim_work);
4293         cancel_work_sync(&fs_info->preempt_reclaim_work);
4294
4295         /* Cancel or finish ongoing discard work */
4296         btrfs_discard_cleanup(fs_info);
4297
4298         if (!sb_rdonly(fs_info->sb)) {
4299                 /*
4300                  * The cleaner kthread is stopped, so do one final pass over
4301                  * unused block groups.
4302                  */
4303                 btrfs_delete_unused_bgs(fs_info);
4304
4305                 /*
4306                  * There might be existing delayed inode workers still running
4307                  * and holding an empty delayed inode item. We must wait for
4308                  * them to complete first because they can create a transaction.
4309                  * This happens when someone calls btrfs_balance_delayed_items()
4310                  * and then a transaction commit runs the same delayed nodes
4311                  * before any delayed worker has done something with the nodes.
4312                  * We must wait for any worker here and not at transaction
4313                  * commit time since that could cause a deadlock.
4314                  * This is a very rare case.
4315                  */
4316                 btrfs_flush_workqueue(fs_info->delayed_workers);
4317
4318                 ret = btrfs_commit_super(fs_info);
4319                 if (ret)
4320                         btrfs_err(fs_info, "commit super ret %d", ret);
4321         }
4322
4323         if (BTRFS_FS_ERROR(fs_info))
4324                 btrfs_error_commit_super(fs_info);
4325
4326         kthread_stop(fs_info->transaction_kthread);
4327         kthread_stop(fs_info->cleaner_kthread);
4328
4329         ASSERT(list_empty(&fs_info->delayed_iputs));
4330         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4331
4332         if (btrfs_check_quota_leak(fs_info)) {
4333                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4334                 btrfs_err(fs_info, "qgroup reserved space leaked");
4335         }
4336
4337         btrfs_free_qgroup_config(fs_info);
4338         ASSERT(list_empty(&fs_info->delalloc_roots));
4339
4340         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4341                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4342                        percpu_counter_sum(&fs_info->delalloc_bytes));
4343         }
4344
4345         if (percpu_counter_sum(&fs_info->ordered_bytes))
4346                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4347                            percpu_counter_sum(&fs_info->ordered_bytes));
4348
4349         btrfs_sysfs_remove_mounted(fs_info);
4350         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4351
4352         btrfs_put_block_group_cache(fs_info);
4353
4354         /*
4355          * we must make sure there is not any read request to
4356          * submit after we stopping all workers.
4357          */
4358         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4359         btrfs_stop_all_workers(fs_info);
4360
4361         /* We shouldn't have any transaction open at this point */
4362         warn_about_uncommitted_trans(fs_info);
4363
4364         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4365         free_root_pointers(fs_info, true);
4366         btrfs_free_fs_roots(fs_info);
4367
4368         /*
4369          * We must free the block groups after dropping the fs_roots as we could
4370          * have had an IO error and have left over tree log blocks that aren't
4371          * cleaned up until the fs roots are freed.  This makes the block group
4372          * accounting appear to be wrong because there's pending reserved bytes,
4373          * so make sure we do the block group cleanup afterwards.
4374          */
4375         btrfs_free_block_groups(fs_info);
4376
4377         iput(fs_info->btree_inode);
4378
4379         btrfs_mapping_tree_free(fs_info);
4380         btrfs_close_devices(fs_info->fs_devices);
4381 }
4382
4383 void btrfs_mark_buffer_dirty(struct btrfs_trans_handle *trans,
4384                              struct extent_buffer *buf)
4385 {
4386         struct btrfs_fs_info *fs_info = buf->fs_info;
4387         u64 transid = btrfs_header_generation(buf);
4388
4389 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4390         /*
4391          * This is a fast path so only do this check if we have sanity tests
4392          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4393          * outside of the sanity tests.
4394          */
4395         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4396                 return;
4397 #endif
4398         /* This is an active transaction (its state < TRANS_STATE_UNBLOCKED). */
4399         ASSERT(trans->transid == fs_info->generation);
4400         btrfs_assert_tree_write_locked(buf);
4401         if (unlikely(transid != fs_info->generation)) {
4402                 btrfs_abort_transaction(trans, -EUCLEAN);
4403                 btrfs_crit(fs_info,
4404 "dirty buffer transid mismatch, logical %llu found transid %llu running transid %llu",
4405                            buf->start, transid, fs_info->generation);
4406         }
4407         set_extent_buffer_dirty(buf);
4408 }
4409
4410 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4411                                         int flush_delayed)
4412 {
4413         /*
4414          * looks as though older kernels can get into trouble with
4415          * this code, they end up stuck in balance_dirty_pages forever
4416          */
4417         int ret;
4418
4419         if (current->flags & PF_MEMALLOC)
4420                 return;
4421
4422         if (flush_delayed)
4423                 btrfs_balance_delayed_items(fs_info);
4424
4425         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4426                                      BTRFS_DIRTY_METADATA_THRESH,
4427                                      fs_info->dirty_metadata_batch);
4428         if (ret > 0) {
4429                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4430         }
4431 }
4432
4433 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4434 {
4435         __btrfs_btree_balance_dirty(fs_info, 1);
4436 }
4437
4438 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4439 {
4440         __btrfs_btree_balance_dirty(fs_info, 0);
4441 }
4442
4443 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4444 {
4445         /* cleanup FS via transaction */
4446         btrfs_cleanup_transaction(fs_info);
4447
4448         mutex_lock(&fs_info->cleaner_mutex);
4449         btrfs_run_delayed_iputs(fs_info);
4450         mutex_unlock(&fs_info->cleaner_mutex);
4451
4452         down_write(&fs_info->cleanup_work_sem);
4453         up_write(&fs_info->cleanup_work_sem);
4454 }
4455
4456 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4457 {
4458         struct btrfs_root *gang[8];
4459         u64 root_objectid = 0;
4460         int ret;
4461
4462         spin_lock(&fs_info->fs_roots_radix_lock);
4463         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4464                                              (void **)gang, root_objectid,
4465                                              ARRAY_SIZE(gang))) != 0) {
4466                 int i;
4467
4468                 for (i = 0; i < ret; i++)
4469                         gang[i] = btrfs_grab_root(gang[i]);
4470                 spin_unlock(&fs_info->fs_roots_radix_lock);
4471
4472                 for (i = 0; i < ret; i++) {
4473                         if (!gang[i])
4474                                 continue;
4475                         root_objectid = gang[i]->root_key.objectid;
4476                         btrfs_free_log(NULL, gang[i]);
4477                         btrfs_put_root(gang[i]);
4478                 }
4479                 root_objectid++;
4480                 spin_lock(&fs_info->fs_roots_radix_lock);
4481         }
4482         spin_unlock(&fs_info->fs_roots_radix_lock);
4483         btrfs_free_log_root_tree(NULL, fs_info);
4484 }
4485
4486 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4487 {
4488         struct btrfs_ordered_extent *ordered;
4489
4490         spin_lock(&root->ordered_extent_lock);
4491         /*
4492          * This will just short circuit the ordered completion stuff which will
4493          * make sure the ordered extent gets properly cleaned up.
4494          */
4495         list_for_each_entry(ordered, &root->ordered_extents,
4496                             root_extent_list)
4497                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4498         spin_unlock(&root->ordered_extent_lock);
4499 }
4500
4501 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4502 {
4503         struct btrfs_root *root;
4504         LIST_HEAD(splice);
4505
4506         spin_lock(&fs_info->ordered_root_lock);
4507         list_splice_init(&fs_info->ordered_roots, &splice);
4508         while (!list_empty(&splice)) {
4509                 root = list_first_entry(&splice, struct btrfs_root,
4510                                         ordered_root);
4511                 list_move_tail(&root->ordered_root,
4512                                &fs_info->ordered_roots);
4513
4514                 spin_unlock(&fs_info->ordered_root_lock);
4515                 btrfs_destroy_ordered_extents(root);
4516
4517                 cond_resched();
4518                 spin_lock(&fs_info->ordered_root_lock);
4519         }
4520         spin_unlock(&fs_info->ordered_root_lock);
4521
4522         /*
4523          * We need this here because if we've been flipped read-only we won't
4524          * get sync() from the umount, so we need to make sure any ordered
4525          * extents that haven't had their dirty pages IO start writeout yet
4526          * actually get run and error out properly.
4527          */
4528         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4529 }
4530
4531 static void btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4532                                        struct btrfs_fs_info *fs_info)
4533 {
4534         struct rb_node *node;
4535         struct btrfs_delayed_ref_root *delayed_refs;
4536         struct btrfs_delayed_ref_node *ref;
4537
4538         delayed_refs = &trans->delayed_refs;
4539
4540         spin_lock(&delayed_refs->lock);
4541         if (atomic_read(&delayed_refs->num_entries) == 0) {
4542                 spin_unlock(&delayed_refs->lock);
4543                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4544                 return;
4545         }
4546
4547         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4548                 struct btrfs_delayed_ref_head *head;
4549                 struct rb_node *n;
4550                 bool pin_bytes = false;
4551
4552                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4553                                 href_node);
4554                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4555                         continue;
4556
4557                 spin_lock(&head->lock);
4558                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4559                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4560                                        ref_node);
4561                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4562                         RB_CLEAR_NODE(&ref->ref_node);
4563                         if (!list_empty(&ref->add_list))
4564                                 list_del(&ref->add_list);
4565                         atomic_dec(&delayed_refs->num_entries);
4566                         btrfs_put_delayed_ref(ref);
4567                         btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
4568                 }
4569                 if (head->must_insert_reserved)
4570                         pin_bytes = true;
4571                 btrfs_free_delayed_extent_op(head->extent_op);
4572                 btrfs_delete_ref_head(delayed_refs, head);
4573                 spin_unlock(&head->lock);
4574                 spin_unlock(&delayed_refs->lock);
4575                 mutex_unlock(&head->mutex);
4576
4577                 if (pin_bytes) {
4578                         struct btrfs_block_group *cache;
4579
4580                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4581                         BUG_ON(!cache);
4582
4583                         spin_lock(&cache->space_info->lock);
4584                         spin_lock(&cache->lock);
4585                         cache->pinned += head->num_bytes;
4586                         btrfs_space_info_update_bytes_pinned(fs_info,
4587                                 cache->space_info, head->num_bytes);
4588                         cache->reserved -= head->num_bytes;
4589                         cache->space_info->bytes_reserved -= head->num_bytes;
4590                         spin_unlock(&cache->lock);
4591                         spin_unlock(&cache->space_info->lock);
4592
4593                         btrfs_put_block_group(cache);
4594
4595                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4596                                 head->bytenr + head->num_bytes - 1);
4597                 }
4598                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4599                 btrfs_put_delayed_ref_head(head);
4600                 cond_resched();
4601                 spin_lock(&delayed_refs->lock);
4602         }
4603         btrfs_qgroup_destroy_extent_records(trans);
4604
4605         spin_unlock(&delayed_refs->lock);
4606 }
4607
4608 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4609 {
4610         struct btrfs_inode *btrfs_inode;
4611         LIST_HEAD(splice);
4612
4613         spin_lock(&root->delalloc_lock);
4614         list_splice_init(&root->delalloc_inodes, &splice);
4615
4616         while (!list_empty(&splice)) {
4617                 struct inode *inode = NULL;
4618                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4619                                                delalloc_inodes);
4620                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4621                 spin_unlock(&root->delalloc_lock);
4622
4623                 /*
4624                  * Make sure we get a live inode and that it'll not disappear
4625                  * meanwhile.
4626                  */
4627                 inode = igrab(&btrfs_inode->vfs_inode);
4628                 if (inode) {
4629                         unsigned int nofs_flag;
4630
4631                         nofs_flag = memalloc_nofs_save();
4632                         invalidate_inode_pages2(inode->i_mapping);
4633                         memalloc_nofs_restore(nofs_flag);
4634                         iput(inode);
4635                 }
4636                 spin_lock(&root->delalloc_lock);
4637         }
4638         spin_unlock(&root->delalloc_lock);
4639 }
4640
4641 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4642 {
4643         struct btrfs_root *root;
4644         LIST_HEAD(splice);
4645
4646         spin_lock(&fs_info->delalloc_root_lock);
4647         list_splice_init(&fs_info->delalloc_roots, &splice);
4648         while (!list_empty(&splice)) {
4649                 root = list_first_entry(&splice, struct btrfs_root,
4650                                          delalloc_root);
4651                 root = btrfs_grab_root(root);
4652                 BUG_ON(!root);
4653                 spin_unlock(&fs_info->delalloc_root_lock);
4654
4655                 btrfs_destroy_delalloc_inodes(root);
4656                 btrfs_put_root(root);
4657
4658                 spin_lock(&fs_info->delalloc_root_lock);
4659         }
4660         spin_unlock(&fs_info->delalloc_root_lock);
4661 }
4662
4663 static void btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4664                                          struct extent_io_tree *dirty_pages,
4665                                          int mark)
4666 {
4667         struct extent_buffer *eb;
4668         u64 start = 0;
4669         u64 end;
4670
4671         while (find_first_extent_bit(dirty_pages, start, &start, &end,
4672                                      mark, NULL)) {
4673                 clear_extent_bits(dirty_pages, start, end, mark);
4674                 while (start <= end) {
4675                         eb = find_extent_buffer(fs_info, start);
4676                         start += fs_info->nodesize;
4677                         if (!eb)
4678                                 continue;
4679
4680                         btrfs_tree_lock(eb);
4681                         wait_on_extent_buffer_writeback(eb);
4682                         btrfs_clear_buffer_dirty(NULL, eb);
4683                         btrfs_tree_unlock(eb);
4684
4685                         free_extent_buffer_stale(eb);
4686                 }
4687         }
4688 }
4689
4690 static void btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4691                                         struct extent_io_tree *unpin)
4692 {
4693         u64 start;
4694         u64 end;
4695
4696         while (1) {
4697                 struct extent_state *cached_state = NULL;
4698
4699                 /*
4700                  * The btrfs_finish_extent_commit() may get the same range as
4701                  * ours between find_first_extent_bit and clear_extent_dirty.
4702                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4703                  * the same extent range.
4704                  */
4705                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4706                 if (!find_first_extent_bit(unpin, 0, &start, &end,
4707                                            EXTENT_DIRTY, &cached_state)) {
4708                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4709                         break;
4710                 }
4711
4712                 clear_extent_dirty(unpin, start, end, &cached_state);
4713                 free_extent_state(cached_state);
4714                 btrfs_error_unpin_extent_range(fs_info, start, end);
4715                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4716                 cond_resched();
4717         }
4718 }
4719
4720 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4721 {
4722         struct inode *inode;
4723
4724         inode = cache->io_ctl.inode;
4725         if (inode) {
4726                 unsigned int nofs_flag;
4727
4728                 nofs_flag = memalloc_nofs_save();
4729                 invalidate_inode_pages2(inode->i_mapping);
4730                 memalloc_nofs_restore(nofs_flag);
4731
4732                 BTRFS_I(inode)->generation = 0;
4733                 cache->io_ctl.inode = NULL;
4734                 iput(inode);
4735         }
4736         ASSERT(cache->io_ctl.pages == NULL);
4737         btrfs_put_block_group(cache);
4738 }
4739
4740 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4741                              struct btrfs_fs_info *fs_info)
4742 {
4743         struct btrfs_block_group *cache;
4744
4745         spin_lock(&cur_trans->dirty_bgs_lock);
4746         while (!list_empty(&cur_trans->dirty_bgs)) {
4747                 cache = list_first_entry(&cur_trans->dirty_bgs,
4748                                          struct btrfs_block_group,
4749                                          dirty_list);
4750
4751                 if (!list_empty(&cache->io_list)) {
4752                         spin_unlock(&cur_trans->dirty_bgs_lock);
4753                         list_del_init(&cache->io_list);
4754                         btrfs_cleanup_bg_io(cache);
4755                         spin_lock(&cur_trans->dirty_bgs_lock);
4756                 }
4757
4758                 list_del_init(&cache->dirty_list);
4759                 spin_lock(&cache->lock);
4760                 cache->disk_cache_state = BTRFS_DC_ERROR;
4761                 spin_unlock(&cache->lock);
4762
4763                 spin_unlock(&cur_trans->dirty_bgs_lock);
4764                 btrfs_put_block_group(cache);
4765                 btrfs_dec_delayed_refs_rsv_bg_updates(fs_info);
4766                 spin_lock(&cur_trans->dirty_bgs_lock);
4767         }
4768         spin_unlock(&cur_trans->dirty_bgs_lock);
4769
4770         /*
4771          * Refer to the definition of io_bgs member for details why it's safe
4772          * to use it without any locking
4773          */
4774         while (!list_empty(&cur_trans->io_bgs)) {
4775                 cache = list_first_entry(&cur_trans->io_bgs,
4776                                          struct btrfs_block_group,
4777                                          io_list);
4778
4779                 list_del_init(&cache->io_list);
4780                 spin_lock(&cache->lock);
4781                 cache->disk_cache_state = BTRFS_DC_ERROR;
4782                 spin_unlock(&cache->lock);
4783                 btrfs_cleanup_bg_io(cache);
4784         }
4785 }
4786
4787 static void btrfs_free_all_qgroup_pertrans(struct btrfs_fs_info *fs_info)
4788 {
4789         struct btrfs_root *gang[8];
4790         int i;
4791         int ret;
4792
4793         spin_lock(&fs_info->fs_roots_radix_lock);
4794         while (1) {
4795                 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
4796                                                  (void **)gang, 0,
4797                                                  ARRAY_SIZE(gang),
4798                                                  BTRFS_ROOT_TRANS_TAG);
4799                 if (ret == 0)
4800                         break;
4801                 for (i = 0; i < ret; i++) {
4802                         struct btrfs_root *root = gang[i];
4803
4804                         btrfs_qgroup_free_meta_all_pertrans(root);
4805                         radix_tree_tag_clear(&fs_info->fs_roots_radix,
4806                                         (unsigned long)root->root_key.objectid,
4807                                         BTRFS_ROOT_TRANS_TAG);
4808                 }
4809         }
4810         spin_unlock(&fs_info->fs_roots_radix_lock);
4811 }
4812
4813 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4814                                    struct btrfs_fs_info *fs_info)
4815 {
4816         struct btrfs_device *dev, *tmp;
4817
4818         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4819         ASSERT(list_empty(&cur_trans->dirty_bgs));
4820         ASSERT(list_empty(&cur_trans->io_bgs));
4821
4822         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4823                                  post_commit_list) {
4824                 list_del_init(&dev->post_commit_list);
4825         }
4826
4827         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4828
4829         cur_trans->state = TRANS_STATE_COMMIT_START;
4830         wake_up(&fs_info->transaction_blocked_wait);
4831
4832         cur_trans->state = TRANS_STATE_UNBLOCKED;
4833         wake_up(&fs_info->transaction_wait);
4834
4835         btrfs_destroy_delayed_inodes(fs_info);
4836
4837         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4838                                      EXTENT_DIRTY);
4839         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4840
4841         btrfs_free_all_qgroup_pertrans(fs_info);
4842
4843         cur_trans->state =TRANS_STATE_COMPLETED;
4844         wake_up(&cur_trans->commit_wait);
4845 }
4846
4847 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4848 {
4849         struct btrfs_transaction *t;
4850
4851         mutex_lock(&fs_info->transaction_kthread_mutex);
4852
4853         spin_lock(&fs_info->trans_lock);
4854         while (!list_empty(&fs_info->trans_list)) {
4855                 t = list_first_entry(&fs_info->trans_list,
4856                                      struct btrfs_transaction, list);
4857                 if (t->state >= TRANS_STATE_COMMIT_PREP) {
4858                         refcount_inc(&t->use_count);
4859                         spin_unlock(&fs_info->trans_lock);
4860                         btrfs_wait_for_commit(fs_info, t->transid);
4861                         btrfs_put_transaction(t);
4862                         spin_lock(&fs_info->trans_lock);
4863                         continue;
4864                 }
4865                 if (t == fs_info->running_transaction) {
4866                         t->state = TRANS_STATE_COMMIT_DOING;
4867                         spin_unlock(&fs_info->trans_lock);
4868                         /*
4869                          * We wait for 0 num_writers since we don't hold a trans
4870                          * handle open currently for this transaction.
4871                          */
4872                         wait_event(t->writer_wait,
4873                                    atomic_read(&t->num_writers) == 0);
4874                 } else {
4875                         spin_unlock(&fs_info->trans_lock);
4876                 }
4877                 btrfs_cleanup_one_transaction(t, fs_info);
4878
4879                 spin_lock(&fs_info->trans_lock);
4880                 if (t == fs_info->running_transaction)
4881                         fs_info->running_transaction = NULL;
4882                 list_del_init(&t->list);
4883                 spin_unlock(&fs_info->trans_lock);
4884
4885                 btrfs_put_transaction(t);
4886                 trace_btrfs_transaction_commit(fs_info);
4887                 spin_lock(&fs_info->trans_lock);
4888         }
4889         spin_unlock(&fs_info->trans_lock);
4890         btrfs_destroy_all_ordered_extents(fs_info);
4891         btrfs_destroy_delayed_inodes(fs_info);
4892         btrfs_assert_delayed_root_empty(fs_info);
4893         btrfs_destroy_all_delalloc_inodes(fs_info);
4894         btrfs_drop_all_logs(fs_info);
4895         mutex_unlock(&fs_info->transaction_kthread_mutex);
4896
4897         return 0;
4898 }
4899
4900 int btrfs_init_root_free_objectid(struct btrfs_root *root)
4901 {
4902         struct btrfs_path *path;
4903         int ret;
4904         struct extent_buffer *l;
4905         struct btrfs_key search_key;
4906         struct btrfs_key found_key;
4907         int slot;
4908
4909         path = btrfs_alloc_path();
4910         if (!path)
4911                 return -ENOMEM;
4912
4913         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
4914         search_key.type = -1;
4915         search_key.offset = (u64)-1;
4916         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
4917         if (ret < 0)
4918                 goto error;
4919         BUG_ON(ret == 0); /* Corruption */
4920         if (path->slots[0] > 0) {
4921                 slot = path->slots[0] - 1;
4922                 l = path->nodes[0];
4923                 btrfs_item_key_to_cpu(l, &found_key, slot);
4924                 root->free_objectid = max_t(u64, found_key.objectid + 1,
4925                                             BTRFS_FIRST_FREE_OBJECTID);
4926         } else {
4927                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
4928         }
4929         ret = 0;
4930 error:
4931         btrfs_free_path(path);
4932         return ret;
4933 }
4934
4935 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
4936 {
4937         int ret;
4938         mutex_lock(&root->objectid_mutex);
4939
4940         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
4941                 btrfs_warn(root->fs_info,
4942                            "the objectid of root %llu reaches its highest value",
4943                            root->root_key.objectid);
4944                 ret = -ENOSPC;
4945                 goto out;
4946         }
4947
4948         *objectid = root->free_objectid++;
4949         ret = 0;
4950 out:
4951         mutex_unlock(&root->objectid_mutex);
4952         return ret;
4953 }