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 "check-integrity.h"
  33 #include "rcu-string.h"
  34 #include "dev-replace.h"
  35 #include "raid56.h"
  36 #include "sysfs.h"
  37 #include "qgroup.h"
  38 #include "compression.h"
  39 #include "tree-checker.h"
  40 #include "ref-verify.h"
  41 #include "block-group.h"
  42 #include "discard.h"
  43 #include "space-info.h"
  44 #include "zoned.h"
  45 #include "subpage.h"
  46 #include "fs.h"
  47 #include "accessors.h"
  48 #include "extent-tree.h"
  49 #include "root-tree.h"
  50 #include "defrag.h"
  51 #include "uuid-tree.h"
  52 #include "relocation.h"
  53 #include "scrub.h"
  54 #include "super.h"
  55
  56 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  57                                  BTRFS_HEADER_FLAG_RELOC |\
  58                                  BTRFS_SUPER_FLAG_ERROR |\
  59                                  BTRFS_SUPER_FLAG_SEEDING |\
  60                                  BTRFS_SUPER_FLAG_METADUMP |\
  61                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  62
  63 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  64 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  65                                       struct btrfs_fs_info *fs_info);
  66 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  67 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  68                                         struct extent_io_tree *dirty_pages,
  69                                         int mark);
  70 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  71                                        struct extent_io_tree *pinned_extents);
  72 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  73 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  74
  75 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
  76 {
  77         if (fs_info->csum_shash)
  78                 crypto_free_shash(fs_info->csum_shash);
  79 }
  80
  81 /*
  82  * Compute the csum of a btree block and store the result to provided buffer.
  83  */
  84 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
  85 {
  86         struct btrfs_fs_info *fs_info = buf->fs_info;
  87         const int num_pages = num_extent_pages(buf);
  88         const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize);
  89         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
  90         char *kaddr;
  91         int i;
  92
  93         shash->tfm = fs_info->csum_shash;
  94         crypto_shash_init(shash);
  95         kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start);
  96         crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
  97                             first_page_part - BTRFS_CSUM_SIZE);
  98
  99         for (i = 1; i < num_pages && INLINE_EXTENT_BUFFER_PAGES > 1; i++) {
 100                 kaddr = page_address(buf->pages[i]);
 101                 crypto_shash_update(shash, kaddr, PAGE_SIZE);
 102         }
 103         memset(result, 0, BTRFS_CSUM_SIZE);
 104         crypto_shash_final(shash, result);
 105 }
 106
 107 /*
 108  * we can't consider a given block up to date unless the transid of the
 109  * block matches the transid in the parent node's pointer.  This is how we
 110  * detect blocks that either didn't get written at all or got written
 111  * in the wrong place.
 112  */
 113 static int verify_parent_transid(struct extent_io_tree *io_tree,
 114                                  struct extent_buffer *eb, u64 parent_transid,
 115                                  int atomic)
 116 {
 117         struct extent_state *cached_state = NULL;
 118         int ret;
 119
 120         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 121                 return 0;
 122
 123         if (atomic)
 124                 return -EAGAIN;
 125
 126         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, &cached_state);
 127         if (extent_buffer_uptodate(eb) &&
 128             btrfs_header_generation(eb) == parent_transid) {
 129                 ret = 0;
 130                 goto out;
 131         }
 132         btrfs_err_rl(eb->fs_info,
 133 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 134                         eb->start, eb->read_mirror,
 135                         parent_transid, btrfs_header_generation(eb));
 136         ret = 1;
 137         clear_extent_buffer_uptodate(eb);
 138 out:
 139         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 140                       &cached_state);
 141         return ret;
 142 }
 143
 144 static bool btrfs_supported_super_csum(u16 csum_type)
 145 {
 146         switch (csum_type) {
 147         case BTRFS_CSUM_TYPE_CRC32:
 148         case BTRFS_CSUM_TYPE_XXHASH:
 149         case BTRFS_CSUM_TYPE_SHA256:
 150         case BTRFS_CSUM_TYPE_BLAKE2:
 151                 return true;
 152         default:
 153                 return false;
 154         }
 155 }
 156
 157 /*
 158  * Return 0 if the superblock checksum type matches the checksum value of that
 159  * algorithm. Pass the raw disk superblock data.
 160  */
 161 int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 162                            const struct btrfs_super_block *disk_sb)
 163 {
 164         char result[BTRFS_CSUM_SIZE];
 165         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 166
 167         shash->tfm = fs_info->csum_shash;
 168
 169         /*
 170          * The super_block structure does not span the whole
 171          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
 172          * filled with zeros and is included in the checksum.
 173          */
 174         crypto_shash_digest(shash, (const u8 *)disk_sb + BTRFS_CSUM_SIZE,
 175                             BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
 176
 177         if (memcmp(disk_sb->csum, result, fs_info->csum_size))
 178                 return 1;
 179
 180         return 0;
 181 }
 182
 183 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
 184                            struct btrfs_key *first_key, u64 parent_transid)
 185 {
 186         struct btrfs_fs_info *fs_info = eb->fs_info;
 187         int found_level;
 188         struct btrfs_key found_key;
 189         int ret;
 190
 191         found_level = btrfs_header_level(eb);
 192         if (found_level != level) {
 193                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 194                      KERN_ERR "BTRFS: tree level check failed\n");
 195                 btrfs_err(fs_info,
 196 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 197                           eb->start, level, found_level);
 198                 return -EIO;
 199         }
 200
 201         if (!first_key)
 202                 return 0;
 203
 204         /*
 205          * For live tree block (new tree blocks in current transaction),
 206          * we need proper lock context to avoid race, which is impossible here.
 207          * So we only checks tree blocks which is read from disk, whose
 208          * generation <= fs_info->last_trans_committed.
 209          */
 210         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 211                 return 0;
 212
 213         /* We have @first_key, so this @eb must have at least one item */
 214         if (btrfs_header_nritems(eb) == 0) {
 215                 btrfs_err(fs_info,
 216                 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
 217                           eb->start);
 218                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
 219                 return -EUCLEAN;
 220         }
 221
 222         if (found_level)
 223                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 224         else
 225                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 226         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 227
 228         if (ret) {
 229                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
 230                      KERN_ERR "BTRFS: tree first key check failed\n");
 231                 btrfs_err(fs_info,
 232 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 233                           eb->start, parent_transid, first_key->objectid,
 234                           first_key->type, first_key->offset,
 235                           found_key.objectid, found_key.type,
 236                           found_key.offset);
 237         }
 238         return ret;
 239 }
 240
 241 static int btrfs_repair_eb_io_failure(const struct extent_buffer *eb,
 242                                       int mirror_num)
 243 {
 244         struct btrfs_fs_info *fs_info = eb->fs_info;
 245         u64 start = eb->start;
 246         int i, num_pages = num_extent_pages(eb);
 247         int ret = 0;
 248
 249         if (sb_rdonly(fs_info->sb))
 250                 return -EROFS;
 251
 252         for (i = 0; i < num_pages; i++) {
 253                 struct page *p = eb->pages[i];
 254
 255                 ret = btrfs_repair_io_failure(fs_info, 0, start, PAGE_SIZE,
 256                                 start, p, start - page_offset(p), mirror_num);
 257                 if (ret)
 258                         break;
 259                 start += PAGE_SIZE;
 260         }
 261
 262         return ret;
 263 }
 264
 265 /*
 266  * helper to read a given tree block, doing retries as required when
 267  * the checksums don't match and we have alternate mirrors to try.
 268  *
 269  * @check:              expected tree parentness check, see the comments of the
 270  *                      structure for details.
 271  */
 272 int btrfs_read_extent_buffer(struct extent_buffer *eb,
 273                              struct btrfs_tree_parent_check *check)
 274 {
 275         struct btrfs_fs_info *fs_info = eb->fs_info;
 276         int failed = 0;
 277         int ret;
 278         int num_copies = 0;
 279         int mirror_num = 0;
 280         int failed_mirror = 0;
 281
 282         ASSERT(check);
 283
 284         while (1) {
 285                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 286                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num, check);
 287                 if (!ret)
 288                         break;
 289
 290                 num_copies = btrfs_num_copies(fs_info,
 291                                               eb->start, eb->len);
 292                 if (num_copies == 1)
 293                         break;
 294
 295                 if (!failed_mirror) {
 296                         failed = 1;
 297                         failed_mirror = eb->read_mirror;
 298                 }
 299
 300                 mirror_num++;
 301                 if (mirror_num == failed_mirror)
 302                         mirror_num++;
 303
 304                 if (mirror_num > num_copies)
 305                         break;
 306         }
 307
 308         if (failed && !ret && failed_mirror)
 309                 btrfs_repair_eb_io_failure(eb, failed_mirror);
 310
 311         return ret;
 312 }
 313
 314 static int csum_one_extent_buffer(struct extent_buffer *eb)
 315 {
 316         struct btrfs_fs_info *fs_info = eb->fs_info;
 317         u8 result[BTRFS_CSUM_SIZE];
 318         int ret;
 319
 320         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 321                                     offsetof(struct btrfs_header, fsid),
 322                                     BTRFS_FSID_SIZE) == 0);
 323         csum_tree_block(eb, result);
 324
 325         if (btrfs_header_level(eb))
 326                 ret = btrfs_check_node(eb);
 327         else
 328                 ret = btrfs_check_leaf_full(eb);
 329
 330         if (ret < 0)
 331                 goto error;
 332
 333         /*
 334          * Also check the generation, the eb reached here must be newer than
 335          * last committed. Or something seriously wrong happened.
 336          */
 337         if (unlikely(btrfs_header_generation(eb) <= fs_info->last_trans_committed)) {
 338                 ret = -EUCLEAN;
 339                 btrfs_err(fs_info,
 340                         "block=%llu bad generation, have %llu expect > %llu",
 341                           eb->start, btrfs_header_generation(eb),
 342                           fs_info->last_trans_committed);
 343                 goto error;
 344         }
 345         write_extent_buffer(eb, result, 0, fs_info->csum_size);
 346
 347         return 0;
 348
 349 error:
 350         btrfs_print_tree(eb, 0);
 351         btrfs_err(fs_info, "block=%llu write time tree block corruption detected",
 352                   eb->start);
 353         /*
 354          * Be noisy if this is an extent buffer from a log tree. We don't abort
 355          * a transaction in case there's a bad log tree extent buffer, we just
 356          * fallback to a transaction commit. Still we want to know when there is
 357          * a bad log tree extent buffer, as that may signal a bug somewhere.
 358          */
 359         WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG) ||
 360                 btrfs_header_owner(eb) == BTRFS_TREE_LOG_OBJECTID);
 361         return ret;
 362 }
 363
 364 /* Checksum all dirty extent buffers in one bio_vec */
 365 static int csum_dirty_subpage_buffers(struct btrfs_fs_info *fs_info,
 366                                       struct bio_vec *bvec)
 367 {
 368         struct page *page = bvec->bv_page;
 369         u64 bvec_start = page_offset(page) + bvec->bv_offset;
 370         u64 cur;
 371         int ret = 0;
 372
 373         for (cur = bvec_start; cur < bvec_start + bvec->bv_len;
 374              cur += fs_info->nodesize) {
 375                 struct extent_buffer *eb;
 376                 bool uptodate;
 377
 378                 eb = find_extent_buffer(fs_info, cur);
 379                 uptodate = btrfs_subpage_test_uptodate(fs_info, page, cur,
 380                                                        fs_info->nodesize);
 381
 382                 /* A dirty eb shouldn't disappear from buffer_radix */
 383                 if (WARN_ON(!eb))
 384                         return -EUCLEAN;
 385
 386                 if (WARN_ON(cur != btrfs_header_bytenr(eb))) {
 387                         free_extent_buffer(eb);
 388                         return -EUCLEAN;
 389                 }
 390                 if (WARN_ON(!uptodate)) {
 391                         free_extent_buffer(eb);
 392                         return -EUCLEAN;
 393                 }
 394
 395                 ret = csum_one_extent_buffer(eb);
 396                 free_extent_buffer(eb);
 397                 if (ret < 0)
 398                         return ret;
 399         }
 400         return ret;
 401 }
 402
 403 /*
 404  * Checksum a dirty tree block before IO.  This has extra checks to make sure
 405  * we only fill in the checksum field in the first page of a multi-page block.
 406  * For subpage extent buffers we need bvec to also read the offset in the page.
 407  */
 408 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec)
 409 {
 410         struct page *page = bvec->bv_page;
 411         u64 start = page_offset(page);
 412         u64 found_start;
 413         struct extent_buffer *eb;
 414
 415         if (fs_info->nodesize < PAGE_SIZE)
 416                 return csum_dirty_subpage_buffers(fs_info, bvec);
 417
 418         eb = (struct extent_buffer *)page->private;
 419         if (page != eb->pages[0])
 420                 return 0;
 421
 422         found_start = btrfs_header_bytenr(eb);
 423
 424         if (test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)) {
 425                 WARN_ON(found_start != 0);
 426                 return 0;
 427         }
 428
 429         /*
 430          * Please do not consolidate these warnings into a single if.
 431          * It is useful to know what went wrong.
 432          */
 433         if (WARN_ON(found_start != start))
 434                 return -EUCLEAN;
 435         if (WARN_ON(!PageUptodate(page)))
 436                 return -EUCLEAN;
 437
 438         return csum_one_extent_buffer(eb);
 439 }
 440
 441 blk_status_t btree_csum_one_bio(struct btrfs_bio *bbio)
 442 {
 443         struct btrfs_fs_info *fs_info = bbio->inode->root->fs_info;
 444         struct bvec_iter iter;
 445         struct bio_vec bv;
 446         int ret = 0;
 447
 448         bio_for_each_segment(bv, &bbio->bio, iter) {
 449                 ret = csum_dirty_buffer(fs_info, &bv);
 450                 if (ret)
 451                         break;
 452         }
 453
 454         return errno_to_blk_status(ret);
 455 }
 456
 457 static int check_tree_block_fsid(struct extent_buffer *eb)
 458 {
 459         struct btrfs_fs_info *fs_info = eb->fs_info;
 460         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
 461         u8 fsid[BTRFS_FSID_SIZE];
 462         u8 *metadata_uuid;
 463
 464         read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
 465                            BTRFS_FSID_SIZE);
 466         /*
 467          * Checking the incompat flag is only valid for the current fs. For
 468          * seed devices it's forbidden to have their uuid changed so reading
 469          * ->fsid in this case is fine
 470          */
 471         if (btrfs_fs_incompat(fs_info, METADATA_UUID))
 472                 metadata_uuid = fs_devices->metadata_uuid;
 473         else
 474                 metadata_uuid = fs_devices->fsid;
 475
 476         if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
 477                 return 0;
 478
 479         list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
 480                 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
 481                         return 0;
 482
 483         return 1;
 484 }
 485
 486 /* Do basic extent buffer checks at read time */
 487 static int validate_extent_buffer(struct extent_buffer *eb,
 488                                   struct btrfs_tree_parent_check *check)
 489 {
 490         struct btrfs_fs_info *fs_info = eb->fs_info;
 491         u64 found_start;
 492         const u32 csum_size = fs_info->csum_size;
 493         u8 found_level;
 494         u8 result[BTRFS_CSUM_SIZE];
 495         const u8 *header_csum;
 496         int ret = 0;
 497
 498         ASSERT(check);
 499
 500         found_start = btrfs_header_bytenr(eb);
 501         if (found_start != eb->start) {
 502                 btrfs_err_rl(fs_info,
 503                         "bad tree block start, mirror %u want %llu have %llu",
 504                              eb->read_mirror, eb->start, found_start);
 505                 ret = -EIO;
 506                 goto out;
 507         }
 508         if (check_tree_block_fsid(eb)) {
 509                 btrfs_err_rl(fs_info, "bad fsid on logical %llu mirror %u",
 510                              eb->start, eb->read_mirror);
 511                 ret = -EIO;
 512                 goto out;
 513         }
 514         found_level = btrfs_header_level(eb);
 515         if (found_level >= BTRFS_MAX_LEVEL) {
 516                 btrfs_err(fs_info,
 517                         "bad tree block level, mirror %u level %d on logical %llu",
 518                         eb->read_mirror, btrfs_header_level(eb), eb->start);
 519                 ret = -EIO;
 520                 goto out;
 521         }
 522
 523         csum_tree_block(eb, result);
 524         header_csum = page_address(eb->pages[0]) +
 525                 get_eb_offset_in_page(eb, offsetof(struct btrfs_header, csum));
 526
 527         if (memcmp(result, header_csum, csum_size) != 0) {
 528                 btrfs_warn_rl(fs_info,
 529 "checksum verify failed on logical %llu mirror %u wanted " CSUM_FMT " found " CSUM_FMT " level %d",
 530                               eb->start, eb->read_mirror,
 531                               CSUM_FMT_VALUE(csum_size, header_csum),
 532                               CSUM_FMT_VALUE(csum_size, result),
 533                               btrfs_header_level(eb));
 534                 ret = -EUCLEAN;
 535                 goto out;
 536         }
 537
 538         if (found_level != check->level) {
 539                 btrfs_err(fs_info,
 540                 "level verify failed on logical %llu mirror %u wanted %u found %u",
 541                           eb->start, eb->read_mirror, check->level, found_level);
 542                 ret = -EIO;
 543                 goto out;
 544         }
 545         if (unlikely(check->transid &&
 546                      btrfs_header_generation(eb) != check->transid)) {
 547                 btrfs_err_rl(eb->fs_info,
 548 "parent transid verify failed on logical %llu mirror %u wanted %llu found %llu",
 549                                 eb->start, eb->read_mirror, check->transid,
 550                                 btrfs_header_generation(eb));
 551                 ret = -EIO;
 552                 goto out;
 553         }
 554         if (check->has_first_key) {
 555                 struct btrfs_key *expect_key = &check->first_key;
 556                 struct btrfs_key found_key;
 557
 558                 if (found_level)
 559                         btrfs_node_key_to_cpu(eb, &found_key, 0);
 560                 else
 561                         btrfs_item_key_to_cpu(eb, &found_key, 0);
 562                 if (unlikely(btrfs_comp_cpu_keys(expect_key, &found_key))) {
 563                         btrfs_err(fs_info,
 564 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 565                                   eb->start, check->transid,
 566                                   expect_key->objectid,
 567                                   expect_key->type, expect_key->offset,
 568                                   found_key.objectid, found_key.type,
 569                                   found_key.offset);
 570                         ret = -EUCLEAN;
 571                         goto out;
 572                 }
 573         }
 574         if (check->owner_root) {
 575                 ret = btrfs_check_eb_owner(eb, check->owner_root);
 576                 if (ret < 0)
 577                         goto out;
 578         }
 579
 580         /*
 581          * If this is a leaf block and it is corrupt, set the corrupt bit so
 582          * that we don't try and read the other copies of this block, just
 583          * return -EIO.
 584          */
 585         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
 586                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 587                 ret = -EIO;
 588         }
 589
 590         if (found_level > 0 && btrfs_check_node(eb))
 591                 ret = -EIO;
 592
 593         if (!ret)
 594                 set_extent_buffer_uptodate(eb);
 595         else
 596                 btrfs_err(fs_info,
 597                 "read time tree block corruption detected on logical %llu mirror %u",
 598                           eb->start, eb->read_mirror);
 599 out:
 600         return ret;
 601 }
 602
 603 static int validate_subpage_buffer(struct page *page, u64 start, u64 end,
 604                                    int mirror, struct btrfs_tree_parent_check *check)
 605 {
 606         struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
 607         struct extent_buffer *eb;
 608         bool reads_done;
 609         int ret = 0;
 610
 611         ASSERT(check);
 612
 613         /*
 614          * We don't allow bio merge for subpage metadata read, so we should
 615          * only get one eb for each endio hook.
 616          */
 617         ASSERT(end == start + fs_info->nodesize - 1);
 618         ASSERT(PagePrivate(page));
 619
 620         eb = find_extent_buffer(fs_info, start);
 621         /*
 622          * When we are reading one tree block, eb must have been inserted into
 623          * the radix tree. If not, something is wrong.
 624          */
 625         ASSERT(eb);
 626
 627         reads_done = atomic_dec_and_test(&eb->io_pages);
 628         /* Subpage read must finish in page read */
 629         ASSERT(reads_done);
 630
 631         eb->read_mirror = mirror;
 632         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 633                 ret = -EIO;
 634                 goto err;
 635         }
 636         ret = validate_extent_buffer(eb, check);
 637         if (ret < 0)
 638                 goto err;
 639
 640         set_extent_buffer_uptodate(eb);
 641
 642         free_extent_buffer(eb);
 643         return ret;
 644 err:
 645         /*
 646          * end_bio_extent_readpage decrements io_pages in case of error,
 647          * make sure it has something to decrement.
 648          */
 649         atomic_inc(&eb->io_pages);
 650         clear_extent_buffer_uptodate(eb);
 651         free_extent_buffer(eb);
 652         return ret;
 653 }
 654
 655 int btrfs_validate_metadata_buffer(struct btrfs_bio *bbio,
 656                                    struct page *page, u64 start, u64 end,
 657                                    int mirror)
 658 {
 659         struct extent_buffer *eb;
 660         int ret = 0;
 661         int reads_done;
 662
 663         ASSERT(page->private);
 664
 665         if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE)
 666                 return validate_subpage_buffer(page, start, end, mirror,
 667                                                &bbio->parent_check);
 668
 669         eb = (struct extent_buffer *)page->private;
 670
 671         /*
 672          * The pending IO might have been the only thing that kept this buffer
 673          * in memory.  Make sure we have a ref for all this other checks
 674          */
 675         atomic_inc(&eb->refs);
 676
 677         reads_done = atomic_dec_and_test(&eb->io_pages);
 678         if (!reads_done)
 679                 goto err;
 680
 681         eb->read_mirror = mirror;
 682         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 683                 ret = -EIO;
 684                 goto err;
 685         }
 686         ret = validate_extent_buffer(eb, &bbio->parent_check);
 687 err:
 688         if (ret) {
 689                 /*
 690                  * our io error hook is going to dec the io pages
 691                  * again, we have to make sure it has something
 692                  * to decrement
 693                  */
 694                 atomic_inc(&eb->io_pages);
 695                 clear_extent_buffer_uptodate(eb);
 696         }
 697         free_extent_buffer(eb);
 698
 699         return ret;
 700 }
 701
 702 #ifdef CONFIG_MIGRATION
 703 static int btree_migrate_folio(struct address_space *mapping,
 704                 struct folio *dst, struct folio *src, enum migrate_mode mode)
 705 {
 706         /*
 707          * we can't safely write a btree page from here,
 708          * we haven't done the locking hook
 709          */
 710         if (folio_test_dirty(src))
 711                 return -EAGAIN;
 712         /*
 713          * Buffers may be managed in a filesystem specific way.
 714          * We must have no buffers or drop them.
 715          */
 716         if (folio_get_private(src) &&
 717             !filemap_release_folio(src, GFP_KERNEL))
 718                 return -EAGAIN;
 719         return migrate_folio(mapping, dst, src, mode);
 720 }
 721 #else
 722 #define btree_migrate_folio NULL
 723 #endif
 724
 725 static int btree_writepages(struct address_space *mapping,
 726                             struct writeback_control *wbc)
 727 {
 728         struct btrfs_fs_info *fs_info;
 729         int ret;
 730
 731         if (wbc->sync_mode == WB_SYNC_NONE) {
 732
 733                 if (wbc->for_kupdate)
 734                         return 0;
 735
 736                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 737                 /* this is a bit racy, but that's ok */
 738                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 739                                              BTRFS_DIRTY_METADATA_THRESH,
 740                                              fs_info->dirty_metadata_batch);
 741                 if (ret < 0)
 742                         return 0;
 743         }
 744         return btree_write_cache_pages(mapping, wbc);
 745 }
 746
 747 static bool btree_release_folio(struct folio *folio, gfp_t gfp_flags)
 748 {
 749         if (folio_test_writeback(folio) || folio_test_dirty(folio))
 750                 return false;
 751
 752         return try_release_extent_buffer(&folio->page);
 753 }
 754
 755 static void btree_invalidate_folio(struct folio *folio, size_t offset,
 756                                  size_t length)
 757 {
 758         struct extent_io_tree *tree;
 759         tree = &BTRFS_I(folio->mapping->host)->io_tree;
 760         extent_invalidate_folio(tree, folio, offset);
 761         btree_release_folio(folio, GFP_NOFS);
 762         if (folio_get_private(folio)) {
 763                 btrfs_warn(BTRFS_I(folio->mapping->host)->root->fs_info,
 764                            "folio private not zero on folio %llu",
 765                            (unsigned long long)folio_pos(folio));
 766                 folio_detach_private(folio);
 767         }
 768 }
 769
 770 #ifdef DEBUG
 771 static bool btree_dirty_folio(struct address_space *mapping,
 772                 struct folio *folio)
 773 {
 774         struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb);
 775         struct btrfs_subpage *subpage;
 776         struct extent_buffer *eb;
 777         int cur_bit = 0;
 778         u64 page_start = folio_pos(folio);
 779
 780         if (fs_info->sectorsize == PAGE_SIZE) {
 781                 eb = folio_get_private(folio);
 782                 BUG_ON(!eb);
 783                 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 784                 BUG_ON(!atomic_read(&eb->refs));
 785                 btrfs_assert_tree_write_locked(eb);
 786                 return filemap_dirty_folio(mapping, folio);
 787         }
 788         subpage = folio_get_private(folio);
 789
 790         ASSERT(subpage->dirty_bitmap);
 791         while (cur_bit < BTRFS_SUBPAGE_BITMAP_SIZE) {
 792                 unsigned long flags;
 793                 u64 cur;
 794                 u16 tmp = (1 << cur_bit);
 795
 796                 spin_lock_irqsave(&subpage->lock, flags);
 797                 if (!(tmp & subpage->dirty_bitmap)) {
 798                         spin_unlock_irqrestore(&subpage->lock, flags);
 799                         cur_bit++;
 800                         continue;
 801                 }
 802                 spin_unlock_irqrestore(&subpage->lock, flags);
 803                 cur = page_start + cur_bit * fs_info->sectorsize;
 804
 805                 eb = find_extent_buffer(fs_info, cur);
 806                 ASSERT(eb);
 807                 ASSERT(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
 808                 ASSERT(atomic_read(&eb->refs));
 809                 btrfs_assert_tree_write_locked(eb);
 810                 free_extent_buffer(eb);
 811
 812                 cur_bit += (fs_info->nodesize >> fs_info->sectorsize_bits);
 813         }
 814         return filemap_dirty_folio(mapping, folio);
 815 }
 816 #else
 817 #define btree_dirty_folio filemap_dirty_folio
 818 #endif
 819
 820 static const struct address_space_operations btree_aops = {
 821         .writepages     = btree_writepages,
 822         .release_folio  = btree_release_folio,
 823         .invalidate_folio = btree_invalidate_folio,
 824         .migrate_folio  = btree_migrate_folio,
 825         .dirty_folio    = btree_dirty_folio,
 826 };
 827
 828 struct extent_buffer *btrfs_find_create_tree_block(
 829                                                 struct btrfs_fs_info *fs_info,
 830                                                 u64 bytenr, u64 owner_root,
 831                                                 int level)
 832 {
 833         if (btrfs_is_testing(fs_info))
 834                 return alloc_test_extent_buffer(fs_info, bytenr);
 835         return alloc_extent_buffer(fs_info, bytenr, owner_root, level);
 836 }
 837
 838 /*
 839  * Read tree block at logical address @bytenr and do variant basic but critical
 840  * verification.
 841  *
 842  * @check:              expected tree parentness check, see comments of the
 843  *                      structure for details.
 844  */
 845 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
 846                                       struct btrfs_tree_parent_check *check)
 847 {
 848         struct extent_buffer *buf = NULL;
 849         int ret;
 850
 851         ASSERT(check);
 852
 853         buf = btrfs_find_create_tree_block(fs_info, bytenr, check->owner_root,
 854                                            check->level);
 855         if (IS_ERR(buf))
 856                 return buf;
 857
 858         ret = btrfs_read_extent_buffer(buf, check);
 859         if (ret) {
 860                 free_extent_buffer_stale(buf);
 861                 return ERR_PTR(ret);
 862         }
 863         if (btrfs_check_eb_owner(buf, check->owner_root)) {
 864                 free_extent_buffer_stale(buf);
 865                 return ERR_PTR(-EUCLEAN);
 866         }
 867         return buf;
 868
 869 }
 870
 871 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
 872                          u64 objectid)
 873 {
 874         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
 875
 876         memset(&root->root_key, 0, sizeof(root->root_key));
 877         memset(&root->root_item, 0, sizeof(root->root_item));
 878         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 879         root->fs_info = fs_info;
 880         root->root_key.objectid = objectid;
 881         root->node = NULL;
 882         root->commit_root = NULL;
 883         root->state = 0;
 884         RB_CLEAR_NODE(&root->rb_node);
 885
 886         root->last_trans = 0;
 887         root->free_objectid = 0;
 888         root->nr_delalloc_inodes = 0;
 889         root->nr_ordered_extents = 0;
 890         root->inode_tree = RB_ROOT;
 891         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
 892
 893         btrfs_init_root_block_rsv(root);
 894
 895         INIT_LIST_HEAD(&root->dirty_list);
 896         INIT_LIST_HEAD(&root->root_list);
 897         INIT_LIST_HEAD(&root->delalloc_inodes);
 898         INIT_LIST_HEAD(&root->delalloc_root);
 899         INIT_LIST_HEAD(&root->ordered_extents);
 900         INIT_LIST_HEAD(&root->ordered_root);
 901         INIT_LIST_HEAD(&root->reloc_dirty_list);
 902         INIT_LIST_HEAD(&root->logged_list[0]);
 903         INIT_LIST_HEAD(&root->logged_list[1]);
 904         spin_lock_init(&root->inode_lock);
 905         spin_lock_init(&root->delalloc_lock);
 906         spin_lock_init(&root->ordered_extent_lock);
 907         spin_lock_init(&root->accounting_lock);
 908         spin_lock_init(&root->log_extents_lock[0]);
 909         spin_lock_init(&root->log_extents_lock[1]);
 910         spin_lock_init(&root->qgroup_meta_rsv_lock);
 911         mutex_init(&root->objectid_mutex);
 912         mutex_init(&root->log_mutex);
 913         mutex_init(&root->ordered_extent_mutex);
 914         mutex_init(&root->delalloc_mutex);
 915         init_waitqueue_head(&root->qgroup_flush_wait);
 916         init_waitqueue_head(&root->log_writer_wait);
 917         init_waitqueue_head(&root->log_commit_wait[0]);
 918         init_waitqueue_head(&root->log_commit_wait[1]);
 919         INIT_LIST_HEAD(&root->log_ctxs[0]);
 920         INIT_LIST_HEAD(&root->log_ctxs[1]);
 921         atomic_set(&root->log_commit[0], 0);
 922         atomic_set(&root->log_commit[1], 0);
 923         atomic_set(&root->log_writers, 0);
 924         atomic_set(&root->log_batch, 0);
 925         refcount_set(&root->refs, 1);
 926         atomic_set(&root->snapshot_force_cow, 0);
 927         atomic_set(&root->nr_swapfiles, 0);
 928         root->log_transid = 0;
 929         root->log_transid_committed = -1;
 930         root->last_log_commit = 0;
 931         root->anon_dev = 0;
 932         if (!dummy) {
 933                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
 934                                     IO_TREE_ROOT_DIRTY_LOG_PAGES);
 935                 extent_io_tree_init(fs_info, &root->log_csum_range,
 936                                     IO_TREE_LOG_CSUM_RANGE);
 937         }
 938
 939         spin_lock_init(&root->root_item_lock);
 940         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
 941 #ifdef CONFIG_BTRFS_DEBUG
 942         INIT_LIST_HEAD(&root->leak_list);
 943         spin_lock(&fs_info->fs_roots_radix_lock);
 944         list_add_tail(&root->leak_list, &fs_info->allocated_roots);
 945         spin_unlock(&fs_info->fs_roots_radix_lock);
 946 #endif
 947 }
 948
 949 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
 950                                            u64 objectid, gfp_t flags)
 951 {
 952         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
 953         if (root)
 954                 __setup_root(root, fs_info, objectid);
 955         return root;
 956 }
 957
 958 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 959 /* Should only be used by the testing infrastructure */
 960 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
 961 {
 962         struct btrfs_root *root;
 963
 964         if (!fs_info)
 965                 return ERR_PTR(-EINVAL);
 966
 967         root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
 968         if (!root)
 969                 return ERR_PTR(-ENOMEM);
 970
 971         /* We don't use the stripesize in selftest, set it as sectorsize */
 972         root->alloc_bytenr = 0;
 973
 974         return root;
 975 }
 976 #endif
 977
 978 static int global_root_cmp(struct rb_node *a_node, const struct rb_node *b_node)
 979 {
 980         const struct btrfs_root *a = rb_entry(a_node, struct btrfs_root, rb_node);
 981         const struct btrfs_root *b = rb_entry(b_node, struct btrfs_root, rb_node);
 982
 983         return btrfs_comp_cpu_keys(&a->root_key, &b->root_key);
 984 }
 985
 986 static int global_root_key_cmp(const void *k, const struct rb_node *node)
 987 {
 988         const struct btrfs_key *key = k;
 989         const struct btrfs_root *root = rb_entry(node, struct btrfs_root, rb_node);
 990
 991         return btrfs_comp_cpu_keys(key, &root->root_key);
 992 }
 993
 994 int btrfs_global_root_insert(struct btrfs_root *root)
 995 {
 996         struct btrfs_fs_info *fs_info = root->fs_info;
 997         struct rb_node *tmp;
 998
 999         write_lock(&fs_info->global_root_lock);
1000         tmp = rb_find_add(&root->rb_node, &fs_info->global_root_tree, global_root_cmp);
1001         write_unlock(&fs_info->global_root_lock);
1002         ASSERT(!tmp);
1003
1004         return tmp ? -EEXIST : 0;
1005 }
1006
1007 void btrfs_global_root_delete(struct btrfs_root *root)
1008 {
1009         struct btrfs_fs_info *fs_info = root->fs_info;
1010
1011         write_lock(&fs_info->global_root_lock);
1012         rb_erase(&root->rb_node, &fs_info->global_root_tree);
1013         write_unlock(&fs_info->global_root_lock);
1014 }
1015
1016 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
1017                                      struct btrfs_key *key)
1018 {
1019         struct rb_node *node;
1020         struct btrfs_root *root = NULL;
1021
1022         read_lock(&fs_info->global_root_lock);
1023         node = rb_find(key, &fs_info->global_root_tree, global_root_key_cmp);
1024         if (node)
1025                 root = container_of(node, struct btrfs_root, rb_node);
1026         read_unlock(&fs_info->global_root_lock);
1027
1028         return root;
1029 }
1030
1031 static u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
1032 {
1033         struct btrfs_block_group *block_group;
1034         u64 ret;
1035
1036         if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1037                 return 0;
1038
1039         if (bytenr)
1040                 block_group = btrfs_lookup_block_group(fs_info, bytenr);
1041         else
1042                 block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
1043         ASSERT(block_group);
1044         if (!block_group)
1045                 return 0;
1046         ret = block_group->global_root_id;
1047         btrfs_put_block_group(block_group);
1048
1049         return ret;
1050 }
1051
1052 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1053 {
1054         struct btrfs_key key = {
1055                 .objectid = BTRFS_CSUM_TREE_OBJECTID,
1056                 .type = BTRFS_ROOT_ITEM_KEY,
1057                 .offset = btrfs_global_root_id(fs_info, bytenr),
1058         };
1059
1060         return btrfs_global_root(fs_info, &key);
1061 }
1062
1063 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info, u64 bytenr)
1064 {
1065         struct btrfs_key key = {
1066                 .objectid = BTRFS_EXTENT_TREE_OBJECTID,
1067                 .type = BTRFS_ROOT_ITEM_KEY,
1068                 .offset = btrfs_global_root_id(fs_info, bytenr),
1069         };
1070
1071         return btrfs_global_root(fs_info, &key);
1072 }
1073
1074 struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info)
1075 {
1076         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE))
1077                 return fs_info->block_group_root;
1078         return btrfs_extent_root(fs_info, 0);
1079 }
1080
1081 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1082                                      u64 objectid)
1083 {
1084         struct btrfs_fs_info *fs_info = trans->fs_info;
1085         struct extent_buffer *leaf;
1086         struct btrfs_root *tree_root = fs_info->tree_root;
1087         struct btrfs_root *root;
1088         struct btrfs_key key;
1089         unsigned int nofs_flag;
1090         int ret = 0;
1091
1092         /*
1093          * We're holding a transaction handle, so use a NOFS memory allocation
1094          * context to avoid deadlock if reclaim happens.
1095          */
1096         nofs_flag = memalloc_nofs_save();
1097         root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1098         memalloc_nofs_restore(nofs_flag);
1099         if (!root)
1100                 return ERR_PTR(-ENOMEM);
1101
1102         root->root_key.objectid = objectid;
1103         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1104         root->root_key.offset = 0;
1105
1106         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1107                                       BTRFS_NESTING_NORMAL);
1108         if (IS_ERR(leaf)) {
1109                 ret = PTR_ERR(leaf);
1110                 leaf = NULL;
1111                 goto fail;
1112         }
1113
1114         root->node = leaf;
1115         btrfs_mark_buffer_dirty(leaf);
1116
1117         root->commit_root = btrfs_root_node(root);
1118         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1119
1120         btrfs_set_root_flags(&root->root_item, 0);
1121         btrfs_set_root_limit(&root->root_item, 0);
1122         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1123         btrfs_set_root_generation(&root->root_item, trans->transid);
1124         btrfs_set_root_level(&root->root_item, 0);
1125         btrfs_set_root_refs(&root->root_item, 1);
1126         btrfs_set_root_used(&root->root_item, leaf->len);
1127         btrfs_set_root_last_snapshot(&root->root_item, 0);
1128         btrfs_set_root_dirid(&root->root_item, 0);
1129         if (is_fstree(objectid))
1130                 generate_random_guid(root->root_item.uuid);
1131         else
1132                 export_guid(root->root_item.uuid, &guid_null);
1133         btrfs_set_root_drop_level(&root->root_item, 0);
1134
1135         btrfs_tree_unlock(leaf);
1136
1137         key.objectid = objectid;
1138         key.type = BTRFS_ROOT_ITEM_KEY;
1139         key.offset = 0;
1140         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1141         if (ret)
1142                 goto fail;
1143
1144         return root;
1145
1146 fail:
1147         btrfs_put_root(root);
1148
1149         return ERR_PTR(ret);
1150 }
1151
1152 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1153                                          struct btrfs_fs_info *fs_info)
1154 {
1155         struct btrfs_root *root;
1156
1157         root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1158         if (!root)
1159                 return ERR_PTR(-ENOMEM);
1160
1161         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1162         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1163         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1164
1165         return root;
1166 }
1167
1168 int btrfs_alloc_log_tree_node(struct btrfs_trans_handle *trans,
1169                               struct btrfs_root *root)
1170 {
1171         struct extent_buffer *leaf;
1172
1173         /*
1174          * DON'T set SHAREABLE bit for log trees.
1175          *
1176          * Log trees are not exposed to user space thus can't be snapshotted,
1177          * and they go away before a real commit is actually done.
1178          *
1179          * They do store pointers to file data extents, and those reference
1180          * counts still get updated (along with back refs to the log tree).
1181          */
1182
1183         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1184                         NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1185         if (IS_ERR(leaf))
1186                 return PTR_ERR(leaf);
1187
1188         root->node = leaf;
1189
1190         btrfs_mark_buffer_dirty(root->node);
1191         btrfs_tree_unlock(root->node);
1192
1193         return 0;
1194 }
1195
1196 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1197                              struct btrfs_fs_info *fs_info)
1198 {
1199         struct btrfs_root *log_root;
1200
1201         log_root = alloc_log_tree(trans, fs_info);
1202         if (IS_ERR(log_root))
1203                 return PTR_ERR(log_root);
1204
1205         if (!btrfs_is_zoned(fs_info)) {
1206                 int ret = btrfs_alloc_log_tree_node(trans, log_root);
1207
1208                 if (ret) {
1209                         btrfs_put_root(log_root);
1210                         return ret;
1211                 }
1212         }
1213
1214         WARN_ON(fs_info->log_root_tree);
1215         fs_info->log_root_tree = log_root;
1216         return 0;
1217 }
1218
1219 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1220                        struct btrfs_root *root)
1221 {
1222         struct btrfs_fs_info *fs_info = root->fs_info;
1223         struct btrfs_root *log_root;
1224         struct btrfs_inode_item *inode_item;
1225         int ret;
1226
1227         log_root = alloc_log_tree(trans, fs_info);
1228         if (IS_ERR(log_root))
1229                 return PTR_ERR(log_root);
1230
1231         ret = btrfs_alloc_log_tree_node(trans, log_root);
1232         if (ret) {
1233                 btrfs_put_root(log_root);
1234                 return ret;
1235         }
1236
1237         log_root->last_trans = trans->transid;
1238         log_root->root_key.offset = root->root_key.objectid;
1239
1240         inode_item = &log_root->root_item.inode;
1241         btrfs_set_stack_inode_generation(inode_item, 1);
1242         btrfs_set_stack_inode_size(inode_item, 3);
1243         btrfs_set_stack_inode_nlink(inode_item, 1);
1244         btrfs_set_stack_inode_nbytes(inode_item,
1245                                      fs_info->nodesize);
1246         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1247
1248         btrfs_set_root_node(&log_root->root_item, log_root->node);
1249
1250         WARN_ON(root->log_root);
1251         root->log_root = log_root;
1252         root->log_transid = 0;
1253         root->log_transid_committed = -1;
1254         root->last_log_commit = 0;
1255         return 0;
1256 }
1257
1258 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1259                                               struct btrfs_path *path,
1260                                               struct btrfs_key *key)
1261 {
1262         struct btrfs_root *root;
1263         struct btrfs_tree_parent_check check = { 0 };
1264         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1265         u64 generation;
1266         int ret;
1267         int level;
1268
1269         root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1270         if (!root)
1271                 return ERR_PTR(-ENOMEM);
1272
1273         ret = btrfs_find_root(tree_root, key, path,
1274                               &root->root_item, &root->root_key);
1275         if (ret) {
1276                 if (ret > 0)
1277                         ret = -ENOENT;
1278                 goto fail;
1279         }
1280
1281         generation = btrfs_root_generation(&root->root_item);
1282         level = btrfs_root_level(&root->root_item);
1283         check.level = level;
1284         check.transid = generation;
1285         check.owner_root = key->objectid;
1286         root->node = read_tree_block(fs_info, btrfs_root_bytenr(&root->root_item),
1287                                      &check);
1288         if (IS_ERR(root->node)) {
1289                 ret = PTR_ERR(root->node);
1290                 root->node = NULL;
1291                 goto fail;
1292         }
1293         if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1294                 ret = -EIO;
1295                 goto fail;
1296         }
1297
1298         /*
1299          * For real fs, and not log/reloc trees, root owner must
1300          * match its root node owner
1301          */
1302         if (!test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state) &&
1303             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1304             root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1305             root->root_key.objectid != btrfs_header_owner(root->node)) {
1306                 btrfs_crit(fs_info,
1307 "root=%llu block=%llu, tree root owner mismatch, have %llu expect %llu",
1308                            root->root_key.objectid, root->node->start,
1309                            btrfs_header_owner(root->node),
1310                            root->root_key.objectid);
1311                 ret = -EUCLEAN;
1312                 goto fail;
1313         }
1314         root->commit_root = btrfs_root_node(root);
1315         return root;
1316 fail:
1317         btrfs_put_root(root);
1318         return ERR_PTR(ret);
1319 }
1320
1321 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1322                                         struct btrfs_key *key)
1323 {
1324         struct btrfs_root *root;
1325         struct btrfs_path *path;
1326
1327         path = btrfs_alloc_path();
1328         if (!path)
1329                 return ERR_PTR(-ENOMEM);
1330         root = read_tree_root_path(tree_root, path, key);
1331         btrfs_free_path(path);
1332
1333         return root;
1334 }
1335
1336 /*
1337  * Initialize subvolume root in-memory structure
1338  *
1339  * @anon_dev:   anonymous device to attach to the root, if zero, allocate new
1340  */
1341 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1342 {
1343         int ret;
1344
1345         btrfs_drew_lock_init(&root->snapshot_lock);
1346
1347         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1348             !btrfs_is_data_reloc_root(root)) {
1349                 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1350                 btrfs_check_and_init_root_item(&root->root_item);
1351         }
1352
1353         /*
1354          * Don't assign anonymous block device to roots that are not exposed to
1355          * userspace, the id pool is limited to 1M
1356          */
1357         if (is_fstree(root->root_key.objectid) &&
1358             btrfs_root_refs(&root->root_item) > 0) {
1359                 if (!anon_dev) {
1360                         ret = get_anon_bdev(&root->anon_dev);
1361                         if (ret)
1362                                 goto fail;
1363                 } else {
1364                         root->anon_dev = anon_dev;
1365                 }
1366         }
1367
1368         mutex_lock(&root->objectid_mutex);
1369         ret = btrfs_init_root_free_objectid(root);
1370         if (ret) {
1371                 mutex_unlock(&root->objectid_mutex);
1372                 goto fail;
1373         }
1374
1375         ASSERT(root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
1376
1377         mutex_unlock(&root->objectid_mutex);
1378
1379         return 0;
1380 fail:
1381         /* The caller is responsible to call btrfs_free_fs_root */
1382         return ret;
1383 }
1384
1385 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1386                                                u64 root_id)
1387 {
1388         struct btrfs_root *root;
1389
1390         spin_lock(&fs_info->fs_roots_radix_lock);
1391         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1392                                  (unsigned long)root_id);
1393         if (root)
1394                 root = btrfs_grab_root(root);
1395         spin_unlock(&fs_info->fs_roots_radix_lock);
1396         return root;
1397 }
1398
1399 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1400                                                 u64 objectid)
1401 {
1402         struct btrfs_key key = {
1403                 .objectid = objectid,
1404                 .type = BTRFS_ROOT_ITEM_KEY,
1405                 .offset = 0,
1406         };
1407
1408         if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1409                 return btrfs_grab_root(fs_info->tree_root);
1410         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1411                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1412         if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1413                 return btrfs_grab_root(fs_info->chunk_root);
1414         if (objectid == BTRFS_DEV_TREE_OBJECTID)
1415                 return btrfs_grab_root(fs_info->dev_root);
1416         if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1417                 return btrfs_grab_root(btrfs_global_root(fs_info, &key));
1418         if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1419                 return btrfs_grab_root(fs_info->quota_root) ?
1420                         fs_info->quota_root : ERR_PTR(-ENOENT);
1421         if (objectid == BTRFS_UUID_TREE_OBJECTID)
1422                 return btrfs_grab_root(fs_info->uuid_root) ?
1423                         fs_info->uuid_root : ERR_PTR(-ENOENT);
1424         if (objectid == BTRFS_BLOCK_GROUP_TREE_OBJECTID)
1425                 return btrfs_grab_root(fs_info->block_group_root) ?
1426                         fs_info->block_group_root : ERR_PTR(-ENOENT);
1427         if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) {
1428                 struct btrfs_root *root = btrfs_global_root(fs_info, &key);
1429
1430                 return btrfs_grab_root(root) ? root : ERR_PTR(-ENOENT);
1431         }
1432         return NULL;
1433 }
1434
1435 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1436                          struct btrfs_root *root)
1437 {
1438         int ret;
1439
1440         ret = radix_tree_preload(GFP_NOFS);
1441         if (ret)
1442                 return ret;
1443
1444         spin_lock(&fs_info->fs_roots_radix_lock);
1445         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1446                                 (unsigned long)root->root_key.objectid,
1447                                 root);
1448         if (ret == 0) {
1449                 btrfs_grab_root(root);
1450                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1451         }
1452         spin_unlock(&fs_info->fs_roots_radix_lock);
1453         radix_tree_preload_end();
1454
1455         return ret;
1456 }
1457
1458 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1459 {
1460 #ifdef CONFIG_BTRFS_DEBUG
1461         struct btrfs_root *root;
1462
1463         while (!list_empty(&fs_info->allocated_roots)) {
1464                 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1465
1466                 root = list_first_entry(&fs_info->allocated_roots,
1467                                         struct btrfs_root, leak_list);
1468                 btrfs_err(fs_info, "leaked root %s refcount %d",
1469                           btrfs_root_name(&root->root_key, buf),
1470                           refcount_read(&root->refs));
1471                 while (refcount_read(&root->refs) > 1)
1472                         btrfs_put_root(root);
1473                 btrfs_put_root(root);
1474         }
1475 #endif
1476 }
1477
1478 static void free_global_roots(struct btrfs_fs_info *fs_info)
1479 {
1480         struct btrfs_root *root;
1481         struct rb_node *node;
1482
1483         while ((node = rb_first_postorder(&fs_info->global_root_tree)) != NULL) {
1484                 root = rb_entry(node, struct btrfs_root, rb_node);
1485                 rb_erase(&root->rb_node, &fs_info->global_root_tree);
1486                 btrfs_put_root(root);
1487         }
1488 }
1489
1490 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1491 {
1492         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1493         percpu_counter_destroy(&fs_info->delalloc_bytes);
1494         percpu_counter_destroy(&fs_info->ordered_bytes);
1495         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1496         btrfs_free_csum_hash(fs_info);
1497         btrfs_free_stripe_hash_table(fs_info);
1498         btrfs_free_ref_cache(fs_info);
1499         kfree(fs_info->balance_ctl);
1500         kfree(fs_info->delayed_root);
1501         free_global_roots(fs_info);
1502         btrfs_put_root(fs_info->tree_root);
1503         btrfs_put_root(fs_info->chunk_root);
1504         btrfs_put_root(fs_info->dev_root);
1505         btrfs_put_root(fs_info->quota_root);
1506         btrfs_put_root(fs_info->uuid_root);
1507         btrfs_put_root(fs_info->fs_root);
1508         btrfs_put_root(fs_info->data_reloc_root);
1509         btrfs_put_root(fs_info->block_group_root);
1510         btrfs_check_leaked_roots(fs_info);
1511         btrfs_extent_buffer_leak_debug_check(fs_info);
1512         kfree(fs_info->super_copy);
1513         kfree(fs_info->super_for_commit);
1514         kfree(fs_info->subpage_info);
1515         kvfree(fs_info);
1516 }
1517
1518
1519 /*
1520  * Get an in-memory reference of a root structure.
1521  *
1522  * For essential trees like root/extent tree, we grab it from fs_info directly.
1523  * For subvolume trees, we check the cached filesystem roots first. If not
1524  * found, then read it from disk and add it to cached fs roots.
1525  *
1526  * Caller should release the root by calling btrfs_put_root() after the usage.
1527  *
1528  * NOTE: Reloc and log trees can't be read by this function as they share the
1529  *       same root objectid.
1530  *
1531  * @objectid:   root id
1532  * @anon_dev:   preallocated anonymous block device number for new roots,
1533  *              pass 0 for new allocation.
1534  * @check_ref:  whether to check root item references, If true, return -ENOENT
1535  *              for orphan roots
1536  */
1537 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1538                                              u64 objectid, dev_t anon_dev,
1539                                              bool check_ref)
1540 {
1541         struct btrfs_root *root;
1542         struct btrfs_path *path;
1543         struct btrfs_key key;
1544         int ret;
1545
1546         root = btrfs_get_global_root(fs_info, objectid);
1547         if (root)
1548                 return root;
1549 again:
1550         root = btrfs_lookup_fs_root(fs_info, objectid);
1551         if (root) {
1552                 /* Shouldn't get preallocated anon_dev for cached roots */
1553                 ASSERT(!anon_dev);
1554                 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1555                         btrfs_put_root(root);
1556                         return ERR_PTR(-ENOENT);
1557                 }
1558                 return root;
1559         }
1560
1561         key.objectid = objectid;
1562         key.type = BTRFS_ROOT_ITEM_KEY;
1563         key.offset = (u64)-1;
1564         root = btrfs_read_tree_root(fs_info->tree_root, &key);
1565         if (IS_ERR(root))
1566                 return root;
1567
1568         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1569                 ret = -ENOENT;
1570                 goto fail;
1571         }
1572
1573         ret = btrfs_init_fs_root(root, anon_dev);
1574         if (ret)
1575                 goto fail;
1576
1577         path = btrfs_alloc_path();
1578         if (!path) {
1579                 ret = -ENOMEM;
1580                 goto fail;
1581         }
1582         key.objectid = BTRFS_ORPHAN_OBJECTID;
1583         key.type = BTRFS_ORPHAN_ITEM_KEY;
1584         key.offset = objectid;
1585
1586         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1587         btrfs_free_path(path);
1588         if (ret < 0)
1589                 goto fail;
1590         if (ret == 0)
1591                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1592
1593         ret = btrfs_insert_fs_root(fs_info, root);
1594         if (ret) {
1595                 if (ret == -EEXIST) {
1596                         btrfs_put_root(root);
1597                         goto again;
1598                 }
1599                 goto fail;
1600         }
1601         return root;
1602 fail:
1603         /*
1604          * If our caller provided us an anonymous device, then it's his
1605          * responsibility to free it in case we fail. So we have to set our
1606          * root's anon_dev to 0 to avoid a double free, once by btrfs_put_root()
1607          * and once again by our caller.
1608          */
1609         if (anon_dev)
1610                 root->anon_dev = 0;
1611         btrfs_put_root(root);
1612         return ERR_PTR(ret);
1613 }
1614
1615 /*
1616  * Get in-memory reference of a root structure
1617  *
1618  * @objectid:   tree objectid
1619  * @check_ref:  if set, verify that the tree exists and the item has at least
1620  *              one reference
1621  */
1622 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1623                                      u64 objectid, bool check_ref)
1624 {
1625         return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1626 }
1627
1628 /*
1629  * Get in-memory reference of a root structure, created as new, optionally pass
1630  * the anonymous block device id
1631  *
1632  * @objectid:   tree objectid
1633  * @anon_dev:   if zero, allocate a new anonymous block device or use the
1634  *              parameter value
1635  */
1636 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1637                                          u64 objectid, dev_t anon_dev)
1638 {
1639         return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1640 }
1641
1642 /*
1643  * btrfs_get_fs_root_commit_root - return a root for the given objectid
1644  * @fs_info:    the fs_info
1645  * @objectid:   the objectid we need to lookup
1646  *
1647  * This is exclusively used for backref walking, and exists specifically because
1648  * of how qgroups does lookups.  Qgroups will do a backref lookup at delayed ref
1649  * creation time, which means we may have to read the tree_root in order to look
1650  * up a fs root that is not in memory.  If the root is not in memory we will
1651  * read the tree root commit root and look up the fs root from there.  This is a
1652  * temporary root, it will not be inserted into the radix tree as it doesn't
1653  * have the most uptodate information, it'll simply be discarded once the
1654  * backref code is finished using the root.
1655  */
1656 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1657                                                  struct btrfs_path *path,
1658                                                  u64 objectid)
1659 {
1660         struct btrfs_root *root;
1661         struct btrfs_key key;
1662
1663         ASSERT(path->search_commit_root && path->skip_locking);
1664
1665         /*
1666          * This can return -ENOENT if we ask for a root that doesn't exist, but
1667          * since this is called via the backref walking code we won't be looking
1668          * up a root that doesn't exist, unless there's corruption.  So if root
1669          * != NULL just return it.
1670          */
1671         root = btrfs_get_global_root(fs_info, objectid);
1672         if (root)
1673                 return root;
1674
1675         root = btrfs_lookup_fs_root(fs_info, objectid);
1676         if (root)
1677                 return root;
1678
1679         key.objectid = objectid;
1680         key.type = BTRFS_ROOT_ITEM_KEY;
1681         key.offset = (u64)-1;
1682         root = read_tree_root_path(fs_info->tree_root, path, &key);
1683         btrfs_release_path(path);
1684
1685         return root;
1686 }
1687
1688 static int cleaner_kthread(void *arg)
1689 {
1690         struct btrfs_fs_info *fs_info = arg;
1691         int again;
1692
1693         while (1) {
1694                 again = 0;
1695
1696                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1697
1698                 /* Make the cleaner go to sleep early. */
1699                 if (btrfs_need_cleaner_sleep(fs_info))
1700                         goto sleep;
1701
1702                 /*
1703                  * Do not do anything if we might cause open_ctree() to block
1704                  * before we have finished mounting the filesystem.
1705                  */
1706                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1707                         goto sleep;
1708
1709                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1710                         goto sleep;
1711
1712                 /*
1713                  * Avoid the problem that we change the status of the fs
1714                  * during the above check and trylock.
1715                  */
1716                 if (btrfs_need_cleaner_sleep(fs_info)) {
1717                         mutex_unlock(&fs_info->cleaner_mutex);
1718                         goto sleep;
1719                 }
1720
1721                 if (test_and_clear_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags))
1722                         btrfs_sysfs_feature_update(fs_info);
1723
1724                 btrfs_run_delayed_iputs(fs_info);
1725
1726                 again = btrfs_clean_one_deleted_snapshot(fs_info);
1727                 mutex_unlock(&fs_info->cleaner_mutex);
1728
1729                 /*
1730                  * The defragger has dealt with the R/O remount and umount,
1731                  * needn't do anything special here.
1732                  */
1733                 btrfs_run_defrag_inodes(fs_info);
1734
1735                 /*
1736                  * Acquires fs_info->reclaim_bgs_lock to avoid racing
1737                  * with relocation (btrfs_relocate_chunk) and relocation
1738                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1739                  * after acquiring fs_info->reclaim_bgs_lock. So we
1740                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1741                  * unused block groups.
1742                  */
1743                 btrfs_delete_unused_bgs(fs_info);
1744
1745                 /*
1746                  * Reclaim block groups in the reclaim_bgs list after we deleted
1747                  * all unused block_groups. This possibly gives us some more free
1748                  * space.
1749                  */
1750                 btrfs_reclaim_bgs(fs_info);
1751 sleep:
1752                 clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1753                 if (kthread_should_park())
1754                         kthread_parkme();
1755                 if (kthread_should_stop())
1756                         return 0;
1757                 if (!again) {
1758                         set_current_state(TASK_INTERRUPTIBLE);
1759                         schedule();
1760                         __set_current_state(TASK_RUNNING);
1761                 }
1762         }
1763 }
1764
1765 static int transaction_kthread(void *arg)
1766 {
1767         struct btrfs_root *root = arg;
1768         struct btrfs_fs_info *fs_info = root->fs_info;
1769         struct btrfs_trans_handle *trans;
1770         struct btrfs_transaction *cur;
1771         u64 transid;
1772         time64_t delta;
1773         unsigned long delay;
1774         bool cannot_commit;
1775
1776         do {
1777                 cannot_commit = false;
1778                 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1779                 mutex_lock(&fs_info->transaction_kthread_mutex);
1780
1781                 spin_lock(&fs_info->trans_lock);
1782                 cur = fs_info->running_transaction;
1783                 if (!cur) {
1784                         spin_unlock(&fs_info->trans_lock);
1785                         goto sleep;
1786                 }
1787
1788                 delta = ktime_get_seconds() - cur->start_time;
1789                 if (!test_and_clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags) &&
1790                     cur->state < TRANS_STATE_COMMIT_START &&
1791                     delta < fs_info->commit_interval) {
1792                         spin_unlock(&fs_info->trans_lock);
1793                         delay -= msecs_to_jiffies((delta - 1) * 1000);
1794                         delay = min(delay,
1795                                     msecs_to_jiffies(fs_info->commit_interval * 1000));
1796                         goto sleep;
1797                 }
1798                 transid = cur->transid;
1799                 spin_unlock(&fs_info->trans_lock);
1800
1801                 /* If the file system is aborted, this will always fail. */
1802                 trans = btrfs_attach_transaction(root);
1803                 if (IS_ERR(trans)) {
1804                         if (PTR_ERR(trans) != -ENOENT)
1805                                 cannot_commit = true;
1806                         goto sleep;
1807                 }
1808                 if (transid == trans->transid) {
1809                         btrfs_commit_transaction(trans);
1810                 } else {
1811                         btrfs_end_transaction(trans);
1812                 }
1813 sleep:
1814                 wake_up_process(fs_info->cleaner_kthread);
1815                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1816
1817                 if (BTRFS_FS_ERROR(fs_info))
1818                         btrfs_cleanup_transaction(fs_info);
1819                 if (!kthread_should_stop() &&
1820                                 (!btrfs_transaction_blocked(fs_info) ||
1821                                  cannot_commit))
1822                         schedule_timeout_interruptible(delay);
1823         } while (!kthread_should_stop());
1824         return 0;
1825 }
1826
1827 /*
1828  * This will find the highest generation in the array of root backups.  The
1829  * index of the highest array is returned, or -EINVAL if we can't find
1830  * anything.
1831  *
1832  * We check to make sure the array is valid by comparing the
1833  * generation of the latest  root in the array with the generation
1834  * in the super block.  If they don't match we pitch it.
1835  */
1836 static int find_newest_super_backup(struct btrfs_fs_info *info)
1837 {
1838         const u64 newest_gen = btrfs_super_generation(info->super_copy);
1839         u64 cur;
1840         struct btrfs_root_backup *root_backup;
1841         int i;
1842
1843         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1844                 root_backup = info->super_copy->super_roots + i;
1845                 cur = btrfs_backup_tree_root_gen(root_backup);
1846                 if (cur == newest_gen)
1847                         return i;
1848         }
1849
1850         return -EINVAL;
1851 }
1852
1853 /*
1854  * copy all the root pointers into the super backup array.
1855  * this will bump the backup pointer by one when it is
1856  * done
1857  */
1858 static void backup_super_roots(struct btrfs_fs_info *info)
1859 {
1860         const int next_backup = info->backup_root_index;
1861         struct btrfs_root_backup *root_backup;
1862
1863         root_backup = info->super_for_commit->super_roots + next_backup;
1864
1865         /*
1866          * make sure all of our padding and empty slots get zero filled
1867          * regardless of which ones we use today
1868          */
1869         memset(root_backup, 0, sizeof(*root_backup));
1870
1871         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1872
1873         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1874         btrfs_set_backup_tree_root_gen(root_backup,
1875                                btrfs_header_generation(info->tree_root->node));
1876
1877         btrfs_set_backup_tree_root_level(root_backup,
1878                                btrfs_header_level(info->tree_root->node));
1879
1880         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1881         btrfs_set_backup_chunk_root_gen(root_backup,
1882                                btrfs_header_generation(info->chunk_root->node));
1883         btrfs_set_backup_chunk_root_level(root_backup,
1884                                btrfs_header_level(info->chunk_root->node));
1885
1886         if (!btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE)) {
1887                 struct btrfs_root *extent_root = btrfs_extent_root(info, 0);
1888                 struct btrfs_root *csum_root = btrfs_csum_root(info, 0);
1889
1890                 btrfs_set_backup_extent_root(root_backup,
1891                                              extent_root->node->start);
1892                 btrfs_set_backup_extent_root_gen(root_backup,
1893                                 btrfs_header_generation(extent_root->node));
1894                 btrfs_set_backup_extent_root_level(root_backup,
1895                                         btrfs_header_level(extent_root->node));
1896
1897                 btrfs_set_backup_csum_root(root_backup, csum_root->node->start);
1898                 btrfs_set_backup_csum_root_gen(root_backup,
1899                                                btrfs_header_generation(csum_root->node));
1900                 btrfs_set_backup_csum_root_level(root_backup,
1901                                                  btrfs_header_level(csum_root->node));
1902         }
1903
1904         /*
1905          * we might commit during log recovery, which happens before we set
1906          * the fs_root.  Make sure it is valid before we fill it in.
1907          */
1908         if (info->fs_root && info->fs_root->node) {
1909                 btrfs_set_backup_fs_root(root_backup,
1910                                          info->fs_root->node->start);
1911                 btrfs_set_backup_fs_root_gen(root_backup,
1912                                btrfs_header_generation(info->fs_root->node));
1913                 btrfs_set_backup_fs_root_level(root_backup,
1914                                btrfs_header_level(info->fs_root->node));
1915         }
1916
1917         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1918         btrfs_set_backup_dev_root_gen(root_backup,
1919                                btrfs_header_generation(info->dev_root->node));
1920         btrfs_set_backup_dev_root_level(root_backup,
1921                                        btrfs_header_level(info->dev_root->node));
1922
1923         btrfs_set_backup_total_bytes(root_backup,
1924                              btrfs_super_total_bytes(info->super_copy));
1925         btrfs_set_backup_bytes_used(root_backup,
1926                              btrfs_super_bytes_used(info->super_copy));
1927         btrfs_set_backup_num_devices(root_backup,
1928                              btrfs_super_num_devices(info->super_copy));
1929
1930         /*
1931          * if we don't copy this out to the super_copy, it won't get remembered
1932          * for the next commit
1933          */
1934         memcpy(&info->super_copy->super_roots,
1935                &info->super_for_commit->super_roots,
1936                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1937 }
1938
1939 /*
1940  * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1941  * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1942  *
1943  * fs_info - filesystem whose backup roots need to be read
1944  * priority - priority of backup root required
1945  *
1946  * Returns backup root index on success and -EINVAL otherwise.
1947  */
1948 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1949 {
1950         int backup_index = find_newest_super_backup(fs_info);
1951         struct btrfs_super_block *super = fs_info->super_copy;
1952         struct btrfs_root_backup *root_backup;
1953
1954         if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1955                 if (priority == 0)
1956                         return backup_index;
1957
1958                 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1959                 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1960         } else {
1961                 return -EINVAL;
1962         }
1963
1964         root_backup = super->super_roots + backup_index;
1965
1966         btrfs_set_super_generation(super,
1967                                    btrfs_backup_tree_root_gen(root_backup));
1968         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1969         btrfs_set_super_root_level(super,
1970                                    btrfs_backup_tree_root_level(root_backup));
1971         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1972
1973         /*
1974          * Fixme: the total bytes and num_devices need to match or we should
1975          * need a fsck
1976          */
1977         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1978         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1979
1980         return backup_index;
1981 }
1982
1983 /* helper to cleanup workers */
1984 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1985 {
1986         btrfs_destroy_workqueue(fs_info->fixup_workers);
1987         btrfs_destroy_workqueue(fs_info->delalloc_workers);
1988         btrfs_destroy_workqueue(fs_info->hipri_workers);
1989         btrfs_destroy_workqueue(fs_info->workers);
1990         if (fs_info->endio_workers)
1991                 destroy_workqueue(fs_info->endio_workers);
1992         if (fs_info->rmw_workers)
1993                 destroy_workqueue(fs_info->rmw_workers);
1994         if (fs_info->compressed_write_workers)
1995                 destroy_workqueue(fs_info->compressed_write_workers);
1996         btrfs_destroy_workqueue(fs_info->endio_write_workers);
1997         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1998         btrfs_destroy_workqueue(fs_info->delayed_workers);
1999         btrfs_destroy_workqueue(fs_info->caching_workers);
2000         btrfs_destroy_workqueue(fs_info->flush_workers);
2001         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2002         if (fs_info->discard_ctl.discard_workers)
2003                 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2004         /*
2005          * Now that all other work queues are destroyed, we can safely destroy
2006          * the queues used for metadata I/O, since tasks from those other work
2007          * queues can do metadata I/O operations.
2008          */
2009         if (fs_info->endio_meta_workers)
2010                 destroy_workqueue(fs_info->endio_meta_workers);
2011 }
2012
2013 static void free_root_extent_buffers(struct btrfs_root *root)
2014 {
2015         if (root) {
2016                 free_extent_buffer(root->node);
2017                 free_extent_buffer(root->commit_root);
2018                 root->node = NULL;
2019                 root->commit_root = NULL;
2020         }
2021 }
2022
2023 static void free_global_root_pointers(struct btrfs_fs_info *fs_info)
2024 {
2025         struct btrfs_root *root, *tmp;
2026
2027         rbtree_postorder_for_each_entry_safe(root, tmp,
2028                                              &fs_info->global_root_tree,
2029                                              rb_node)
2030                 free_root_extent_buffers(root);
2031 }
2032
2033 /* helper to cleanup tree roots */
2034 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2035 {
2036         free_root_extent_buffers(info->tree_root);
2037
2038         free_global_root_pointers(info);
2039         free_root_extent_buffers(info->dev_root);
2040         free_root_extent_buffers(info->quota_root);
2041         free_root_extent_buffers(info->uuid_root);
2042         free_root_extent_buffers(info->fs_root);
2043         free_root_extent_buffers(info->data_reloc_root);
2044         free_root_extent_buffers(info->block_group_root);
2045         if (free_chunk_root)
2046                 free_root_extent_buffers(info->chunk_root);
2047 }
2048
2049 void btrfs_put_root(struct btrfs_root *root)
2050 {
2051         if (!root)
2052                 return;
2053
2054         if (refcount_dec_and_test(&root->refs)) {
2055                 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2056                 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2057                 if (root->anon_dev)
2058                         free_anon_bdev(root->anon_dev);
2059                 free_root_extent_buffers(root);
2060 #ifdef CONFIG_BTRFS_DEBUG
2061                 spin_lock(&root->fs_info->fs_roots_radix_lock);
2062                 list_del_init(&root->leak_list);
2063                 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2064 #endif
2065                 kfree(root);
2066         }
2067 }
2068
2069 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2070 {
2071         int ret;
2072         struct btrfs_root *gang[8];
2073         int i;
2074
2075         while (!list_empty(&fs_info->dead_roots)) {
2076                 gang[0] = list_entry(fs_info->dead_roots.next,
2077                                      struct btrfs_root, root_list);
2078                 list_del(&gang[0]->root_list);
2079
2080                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2081                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2082                 btrfs_put_root(gang[0]);
2083         }
2084
2085         while (1) {
2086                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2087                                              (void **)gang, 0,
2088                                              ARRAY_SIZE(gang));
2089                 if (!ret)
2090                         break;
2091                 for (i = 0; i < ret; i++)
2092                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2093         }
2094 }
2095
2096 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2097 {
2098         mutex_init(&fs_info->scrub_lock);
2099         atomic_set(&fs_info->scrubs_running, 0);
2100         atomic_set(&fs_info->scrub_pause_req, 0);
2101         atomic_set(&fs_info->scrubs_paused, 0);
2102         atomic_set(&fs_info->scrub_cancel_req, 0);
2103         init_waitqueue_head(&fs_info->scrub_pause_wait);
2104         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2105 }
2106
2107 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2108 {
2109         spin_lock_init(&fs_info->balance_lock);
2110         mutex_init(&fs_info->balance_mutex);
2111         atomic_set(&fs_info->balance_pause_req, 0);
2112         atomic_set(&fs_info->balance_cancel_req, 0);
2113         fs_info->balance_ctl = NULL;
2114         init_waitqueue_head(&fs_info->balance_wait_q);
2115         atomic_set(&fs_info->reloc_cancel_req, 0);
2116 }
2117
2118 static int btrfs_init_btree_inode(struct super_block *sb)
2119 {
2120         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2121         unsigned long hash = btrfs_inode_hash(BTRFS_BTREE_INODE_OBJECTID,
2122                                               fs_info->tree_root);
2123         struct inode *inode;
2124
2125         inode = new_inode(sb);
2126         if (!inode)
2127                 return -ENOMEM;
2128
2129         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2130         set_nlink(inode, 1);
2131         /*
2132          * we set the i_size on the btree inode to the max possible int.
2133          * the real end of the address space is determined by all of
2134          * the devices in the system
2135          */
2136         inode->i_size = OFFSET_MAX;
2137         inode->i_mapping->a_ops = &btree_aops;
2138         mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
2139
2140         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2141         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2142                             IO_TREE_BTREE_INODE_IO);
2143         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2144
2145         BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2146         BTRFS_I(inode)->location.objectid = BTRFS_BTREE_INODE_OBJECTID;
2147         BTRFS_I(inode)->location.type = 0;
2148         BTRFS_I(inode)->location.offset = 0;
2149         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2150         __insert_inode_hash(inode, hash);
2151         fs_info->btree_inode = inode;
2152
2153         return 0;
2154 }
2155
2156 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2157 {
2158         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2159         init_rwsem(&fs_info->dev_replace.rwsem);
2160         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2161 }
2162
2163 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2164 {
2165         spin_lock_init(&fs_info->qgroup_lock);
2166         mutex_init(&fs_info->qgroup_ioctl_lock);
2167         fs_info->qgroup_tree = RB_ROOT;
2168         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2169         fs_info->qgroup_seq = 1;
2170         fs_info->qgroup_ulist = NULL;
2171         fs_info->qgroup_rescan_running = false;
2172         fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
2173         mutex_init(&fs_info->qgroup_rescan_lock);
2174 }
2175
2176 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info)
2177 {
2178         u32 max_active = fs_info->thread_pool_size;
2179         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2180
2181         fs_info->workers =
2182                 btrfs_alloc_workqueue(fs_info, "worker", flags, max_active, 16);
2183         fs_info->hipri_workers =
2184                 btrfs_alloc_workqueue(fs_info, "worker-high",
2185                                       flags | WQ_HIGHPRI, max_active, 16);
2186
2187         fs_info->delalloc_workers =
2188                 btrfs_alloc_workqueue(fs_info, "delalloc",
2189                                       flags, max_active, 2);
2190
2191         fs_info->flush_workers =
2192                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2193                                       flags, max_active, 0);
2194
2195         fs_info->caching_workers =
2196                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2197
2198         fs_info->fixup_workers =
2199                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2200
2201         fs_info->endio_workers =
2202                 alloc_workqueue("btrfs-endio", flags, max_active);
2203         fs_info->endio_meta_workers =
2204                 alloc_workqueue("btrfs-endio-meta", flags, max_active);
2205         fs_info->rmw_workers = alloc_workqueue("btrfs-rmw", flags, max_active);
2206         fs_info->endio_write_workers =
2207                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2208                                       max_active, 2);
2209         fs_info->compressed_write_workers =
2210                 alloc_workqueue("btrfs-compressed-write", flags, max_active);
2211         fs_info->endio_freespace_worker =
2212                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2213                                       max_active, 0);
2214         fs_info->delayed_workers =
2215                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2216                                       max_active, 0);
2217         fs_info->qgroup_rescan_workers =
2218                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2219         fs_info->discard_ctl.discard_workers =
2220                 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2221
2222         if (!(fs_info->workers && fs_info->hipri_workers &&
2223               fs_info->delalloc_workers && fs_info->flush_workers &&
2224               fs_info->endio_workers && fs_info->endio_meta_workers &&
2225               fs_info->compressed_write_workers &&
2226               fs_info->endio_write_workers &&
2227               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2228               fs_info->caching_workers && fs_info->fixup_workers &&
2229               fs_info->delayed_workers && fs_info->qgroup_rescan_workers &&
2230               fs_info->discard_ctl.discard_workers)) {
2231                 return -ENOMEM;
2232         }
2233
2234         return 0;
2235 }
2236
2237 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2238 {
2239         struct crypto_shash *csum_shash;
2240         const char *csum_driver = btrfs_super_csum_driver(csum_type);
2241
2242         csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2243
2244         if (IS_ERR(csum_shash)) {
2245                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2246                           csum_driver);
2247                 return PTR_ERR(csum_shash);
2248         }
2249
2250         fs_info->csum_shash = csum_shash;
2251
2252         /*
2253          * Check if the checksum implementation is a fast accelerated one.
2254          * As-is this is a bit of a hack and should be replaced once the csum
2255          * implementations provide that information themselves.
2256          */
2257         switch (csum_type) {
2258         case BTRFS_CSUM_TYPE_CRC32:
2259                 if (!strstr(crypto_shash_driver_name(csum_shash), "generic"))
2260                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
2261                 break;
2262         default:
2263                 break;
2264         }
2265
2266         btrfs_info(fs_info, "using %s (%s) checksum algorithm",
2267                         btrfs_super_csum_name(csum_type),
2268                         crypto_shash_driver_name(csum_shash));
2269         return 0;
2270 }
2271
2272 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2273                             struct btrfs_fs_devices *fs_devices)
2274 {
2275         int ret;
2276         struct btrfs_tree_parent_check check = { 0 };
2277         struct btrfs_root *log_tree_root;
2278         struct btrfs_super_block *disk_super = fs_info->super_copy;
2279         u64 bytenr = btrfs_super_log_root(disk_super);
2280         int level = btrfs_super_log_root_level(disk_super);
2281
2282         if (fs_devices->rw_devices == 0) {
2283                 btrfs_warn(fs_info, "log replay required on RO media");
2284                 return -EIO;
2285         }
2286
2287         log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2288                                          GFP_KERNEL);
2289         if (!log_tree_root)
2290                 return -ENOMEM;
2291
2292         check.level = level;
2293         check.transid = fs_info->generation + 1;
2294         check.owner_root = BTRFS_TREE_LOG_OBJECTID;
2295         log_tree_root->node = read_tree_block(fs_info, bytenr, &check);
2296         if (IS_ERR(log_tree_root->node)) {
2297                 btrfs_warn(fs_info, "failed to read log tree");
2298                 ret = PTR_ERR(log_tree_root->node);
2299                 log_tree_root->node = NULL;
2300                 btrfs_put_root(log_tree_root);
2301                 return ret;
2302         }
2303         if (!extent_buffer_uptodate(log_tree_root->node)) {
2304                 btrfs_err(fs_info, "failed to read log tree");
2305                 btrfs_put_root(log_tree_root);
2306                 return -EIO;
2307         }
2308
2309         /* returns with log_tree_root freed on success */
2310         ret = btrfs_recover_log_trees(log_tree_root);
2311         if (ret) {
2312                 btrfs_handle_fs_error(fs_info, ret,
2313                                       "Failed to recover log tree");
2314                 btrfs_put_root(log_tree_root);
2315                 return ret;
2316         }
2317
2318         if (sb_rdonly(fs_info->sb)) {
2319                 ret = btrfs_commit_super(fs_info);
2320                 if (ret)
2321                         return ret;
2322         }
2323
2324         return 0;
2325 }
2326
2327 static int load_global_roots_objectid(struct btrfs_root *tree_root,
2328                                       struct btrfs_path *path, u64 objectid,
2329                                       const char *name)
2330 {
2331         struct btrfs_fs_info *fs_info = tree_root->fs_info;
2332         struct btrfs_root *root;
2333         u64 max_global_id = 0;
2334         int ret;
2335         struct btrfs_key key = {
2336                 .objectid = objectid,
2337                 .type = BTRFS_ROOT_ITEM_KEY,
2338                 .offset = 0,
2339         };
2340         bool found = false;
2341
2342         /* If we have IGNOREDATACSUMS skip loading these roots. */
2343         if (objectid == BTRFS_CSUM_TREE_OBJECTID &&
2344             btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2345                 set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2346                 return 0;
2347         }
2348
2349         while (1) {
2350                 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2351                 if (ret < 0)
2352                         break;
2353
2354                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2355                         ret = btrfs_next_leaf(tree_root, path);
2356                         if (ret) {
2357                                 if (ret > 0)
2358                                         ret = 0;
2359                                 break;
2360                         }
2361                 }
2362                 ret = 0;
2363
2364                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2365                 if (key.objectid != objectid)
2366                         break;
2367                 btrfs_release_path(path);
2368
2369                 /*
2370                  * Just worry about this for extent tree, it'll be the same for
2371                  * everybody.
2372                  */
2373                 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2374                         max_global_id = max(max_global_id, key.offset);
2375
2376                 found = true;
2377                 root = read_tree_root_path(tree_root, path, &key);
2378                 if (IS_ERR(root)) {
2379                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2380                                 ret = PTR_ERR(root);
2381                         break;
2382                 }
2383                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2384                 ret = btrfs_global_root_insert(root);
2385                 if (ret) {
2386                         btrfs_put_root(root);
2387                         break;
2388                 }
2389                 key.offset++;
2390         }
2391         btrfs_release_path(path);
2392
2393         if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
2394                 fs_info->nr_global_roots = max_global_id + 1;
2395
2396         if (!found || ret) {
2397                 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
2398                         set_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state);
2399
2400                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS))
2401                         ret = ret ? ret : -ENOENT;
2402                 else
2403                         ret = 0;
2404                 btrfs_err(fs_info, "failed to load root %s", name);
2405         }
2406         return ret;
2407 }
2408
2409 static int load_global_roots(struct btrfs_root *tree_root)
2410 {
2411         struct btrfs_path *path;
2412         int ret = 0;
2413
2414         path = btrfs_alloc_path();
2415         if (!path)
2416                 return -ENOMEM;
2417
2418         ret = load_global_roots_objectid(tree_root, path,
2419                                          BTRFS_EXTENT_TREE_OBJECTID, "extent");
2420         if (ret)
2421                 goto out;
2422         ret = load_global_roots_objectid(tree_root, path,
2423                                          BTRFS_CSUM_TREE_OBJECTID, "csum");
2424         if (ret)
2425                 goto out;
2426         if (!btrfs_fs_compat_ro(tree_root->fs_info, FREE_SPACE_TREE))
2427                 goto out;
2428         ret = load_global_roots_objectid(tree_root, path,
2429                                          BTRFS_FREE_SPACE_TREE_OBJECTID,
2430                                          "free space");
2431 out:
2432         btrfs_free_path(path);
2433         return ret;
2434 }
2435
2436 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2437 {
2438         struct btrfs_root *tree_root = fs_info->tree_root;
2439         struct btrfs_root *root;
2440         struct btrfs_key location;
2441         int ret;
2442
2443         BUG_ON(!fs_info->tree_root);
2444
2445         ret = load_global_roots(tree_root);
2446         if (ret)
2447                 return ret;
2448
2449         location.type = BTRFS_ROOT_ITEM_KEY;
2450         location.offset = 0;
2451
2452         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE)) {
2453                 location.objectid = BTRFS_BLOCK_GROUP_TREE_OBJECTID;
2454                 root = btrfs_read_tree_root(tree_root, &location);
2455                 if (IS_ERR(root)) {
2456                         if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2457                                 ret = PTR_ERR(root);
2458                                 goto out;
2459                         }
2460                 } else {
2461                         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2462                         fs_info->block_group_root = root;
2463                 }
2464         }
2465
2466         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2467         root = btrfs_read_tree_root(tree_root, &location);
2468         if (IS_ERR(root)) {
2469                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2470                         ret = PTR_ERR(root);
2471                         goto out;
2472                 }
2473         } else {
2474                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2475                 fs_info->dev_root = root;
2476         }
2477         /* Initialize fs_info for all devices in any case */
2478         ret = btrfs_init_devices_late(fs_info);
2479         if (ret)
2480                 goto out;
2481
2482         /*
2483          * This tree can share blocks with some other fs tree during relocation
2484          * and we need a proper setup by btrfs_get_fs_root
2485          */
2486         root = btrfs_get_fs_root(tree_root->fs_info,
2487                                  BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2488         if (IS_ERR(root)) {
2489                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2490                         ret = PTR_ERR(root);
2491                         goto out;
2492                 }
2493         } else {
2494                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2495                 fs_info->data_reloc_root = root;
2496         }
2497
2498         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2499         root = btrfs_read_tree_root(tree_root, &location);
2500         if (!IS_ERR(root)) {
2501                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2502                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2503                 fs_info->quota_root = root;
2504         }
2505
2506         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2507         root = btrfs_read_tree_root(tree_root, &location);
2508         if (IS_ERR(root)) {
2509                 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2510                         ret = PTR_ERR(root);
2511                         if (ret != -ENOENT)
2512                                 goto out;
2513                 }
2514         } else {
2515                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2516                 fs_info->uuid_root = root;
2517         }
2518
2519         return 0;
2520 out:
2521         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2522                    location.objectid, ret);
2523         return ret;
2524 }
2525
2526 /*
2527  * Real super block validation
2528  * NOTE: super csum type and incompat features will not be checked here.
2529  *
2530  * @sb:         super block to check
2531  * @mirror_num: the super block number to check its bytenr:
2532  *              0       the primary (1st) sb
2533  *              1, 2    2nd and 3rd backup copy
2534  *             -1       skip bytenr check
2535  */
2536 int btrfs_validate_super(struct btrfs_fs_info *fs_info,
2537                          struct btrfs_super_block *sb, int mirror_num)
2538 {
2539         u64 nodesize = btrfs_super_nodesize(sb);
2540         u64 sectorsize = btrfs_super_sectorsize(sb);
2541         int ret = 0;
2542
2543         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2544                 btrfs_err(fs_info, "no valid FS found");
2545                 ret = -EINVAL;
2546         }
2547         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2548                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2549                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2550                 ret = -EINVAL;
2551         }
2552         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2553                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2554                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2555                 ret = -EINVAL;
2556         }
2557         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2558                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2559                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2560                 ret = -EINVAL;
2561         }
2562         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2563                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2564                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2565                 ret = -EINVAL;
2566         }
2567
2568         /*
2569          * Check sectorsize and nodesize first, other check will need it.
2570          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2571          */
2572         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2573             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2574                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2575                 ret = -EINVAL;
2576         }
2577
2578         /*
2579          * We only support at most two sectorsizes: 4K and PAGE_SIZE.
2580          *
2581          * We can support 16K sectorsize with 64K page size without problem,
2582          * but such sectorsize/pagesize combination doesn't make much sense.
2583          * 4K will be our future standard, PAGE_SIZE is supported from the very
2584          * beginning.
2585          */
2586         if (sectorsize > PAGE_SIZE || (sectorsize != SZ_4K && sectorsize != PAGE_SIZE)) {
2587                 btrfs_err(fs_info,
2588                         "sectorsize %llu not yet supported for page size %lu",
2589                         sectorsize, PAGE_SIZE);
2590                 ret = -EINVAL;
2591         }
2592
2593         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2594             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2595                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2596                 ret = -EINVAL;
2597         }
2598         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2599                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2600                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2601                 ret = -EINVAL;
2602         }
2603
2604         /* Root alignment check */
2605         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2606                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2607                            btrfs_super_root(sb));
2608                 ret = -EINVAL;
2609         }
2610         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2611                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2612                            btrfs_super_chunk_root(sb));
2613                 ret = -EINVAL;
2614         }
2615         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2616                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2617                            btrfs_super_log_root(sb));
2618                 ret = -EINVAL;
2619         }
2620
2621         if (memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2622                    BTRFS_FSID_SIZE)) {
2623                 btrfs_err(fs_info,
2624                 "superblock fsid doesn't match fsid of fs_devices: %pU != %pU",
2625                         fs_info->super_copy->fsid, fs_info->fs_devices->fsid);
2626                 ret = -EINVAL;
2627         }
2628
2629         if (btrfs_fs_incompat(fs_info, METADATA_UUID) &&
2630             memcmp(fs_info->fs_devices->metadata_uuid,
2631                    fs_info->super_copy->metadata_uuid, BTRFS_FSID_SIZE)) {
2632                 btrfs_err(fs_info,
2633 "superblock metadata_uuid doesn't match metadata uuid of fs_devices: %pU != %pU",
2634                         fs_info->super_copy->metadata_uuid,
2635                         fs_info->fs_devices->metadata_uuid);
2636                 ret = -EINVAL;
2637         }
2638
2639         /*
2640          * Artificial requirement for block-group-tree to force newer features
2641          * (free-space-tree, no-holes) so the test matrix is smaller.
2642          */
2643         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
2644             (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID) ||
2645              !btrfs_fs_incompat(fs_info, NO_HOLES))) {
2646                 btrfs_err(fs_info,
2647                 "block-group-tree feature requires fres-space-tree and no-holes");
2648                 ret = -EINVAL;
2649         }
2650
2651         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2652                    BTRFS_FSID_SIZE) != 0) {
2653                 btrfs_err(fs_info,
2654                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2655                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2656                 ret = -EINVAL;
2657         }
2658
2659         /*
2660          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2661          * done later
2662          */
2663         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2664                 btrfs_err(fs_info, "bytes_used is too small %llu",
2665                           btrfs_super_bytes_used(sb));
2666                 ret = -EINVAL;
2667         }
2668         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2669                 btrfs_err(fs_info, "invalid stripesize %u",
2670                           btrfs_super_stripesize(sb));
2671                 ret = -EINVAL;
2672         }
2673         if (btrfs_super_num_devices(sb) > (1UL << 31))
2674                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2675                            btrfs_super_num_devices(sb));
2676         if (btrfs_super_num_devices(sb) == 0) {
2677                 btrfs_err(fs_info, "number of devices is 0");
2678                 ret = -EINVAL;
2679         }
2680
2681         if (mirror_num >= 0 &&
2682             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2683                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2684                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2685                 ret = -EINVAL;
2686         }
2687
2688         /*
2689          * Obvious sys_chunk_array corruptions, it must hold at least one key
2690          * and one chunk
2691          */
2692         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2693                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2694                           btrfs_super_sys_array_size(sb),
2695                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2696                 ret = -EINVAL;
2697         }
2698         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2699                         + sizeof(struct btrfs_chunk)) {
2700                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2701                           btrfs_super_sys_array_size(sb),
2702                           sizeof(struct btrfs_disk_key)
2703                           + sizeof(struct btrfs_chunk));
2704                 ret = -EINVAL;
2705         }
2706
2707         /*
2708          * The generation is a global counter, we'll trust it more than the others
2709          * but it's still possible that it's the one that's wrong.
2710          */
2711         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2712                 btrfs_warn(fs_info,
2713                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2714                         btrfs_super_generation(sb),
2715                         btrfs_super_chunk_root_generation(sb));
2716         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2717             && btrfs_super_cache_generation(sb) != (u64)-1)
2718                 btrfs_warn(fs_info,
2719                         "suspicious: generation < cache_generation: %llu < %llu",
2720                         btrfs_super_generation(sb),
2721                         btrfs_super_cache_generation(sb));
2722
2723         return ret;
2724 }
2725
2726 /*
2727  * Validation of super block at mount time.
2728  * Some checks already done early at mount time, like csum type and incompat
2729  * flags will be skipped.
2730  */
2731 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2732 {
2733         return btrfs_validate_super(fs_info, fs_info->super_copy, 0);
2734 }
2735
2736 /*
2737  * Validation of super block at write time.
2738  * Some checks like bytenr check will be skipped as their values will be
2739  * overwritten soon.
2740  * Extra checks like csum type and incompat flags will be done here.
2741  */
2742 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2743                                       struct btrfs_super_block *sb)
2744 {
2745         int ret;
2746
2747         ret = btrfs_validate_super(fs_info, sb, -1);
2748         if (ret < 0)
2749                 goto out;
2750         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2751                 ret = -EUCLEAN;
2752                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2753                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2754                 goto out;
2755         }
2756         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2757                 ret = -EUCLEAN;
2758                 btrfs_err(fs_info,
2759                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2760                           btrfs_super_incompat_flags(sb),
2761                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2762                 goto out;
2763         }
2764 out:
2765         if (ret < 0)
2766                 btrfs_err(fs_info,
2767                 "super block corruption detected before writing it to disk");
2768         return ret;
2769 }
2770
2771 static int load_super_root(struct btrfs_root *root, u64 bytenr, u64 gen, int level)
2772 {
2773         struct btrfs_tree_parent_check check = {
2774                 .level = level,
2775                 .transid = gen,
2776                 .owner_root = root->root_key.objectid
2777         };
2778         int ret = 0;
2779
2780         root->node = read_tree_block(root->fs_info, bytenr, &check);
2781         if (IS_ERR(root->node)) {
2782                 ret = PTR_ERR(root->node);
2783                 root->node = NULL;
2784                 return ret;
2785         }
2786         if (!extent_buffer_uptodate(root->node)) {
2787                 free_extent_buffer(root->node);
2788                 root->node = NULL;
2789                 return -EIO;
2790         }
2791
2792         btrfs_set_root_node(&root->root_item, root->node);
2793         root->commit_root = btrfs_root_node(root);
2794         btrfs_set_root_refs(&root->root_item, 1);
2795         return ret;
2796 }
2797
2798 static int load_important_roots(struct btrfs_fs_info *fs_info)
2799 {
2800         struct btrfs_super_block *sb = fs_info->super_copy;
2801         u64 gen, bytenr;
2802         int level, ret;
2803
2804         bytenr = btrfs_super_root(sb);
2805         gen = btrfs_super_generation(sb);
2806         level = btrfs_super_root_level(sb);
2807         ret = load_super_root(fs_info->tree_root, bytenr, gen, level);
2808         if (ret) {
2809                 btrfs_warn(fs_info, "couldn't read tree root");
2810                 return ret;
2811         }
2812         return 0;
2813 }
2814
2815 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2816 {
2817         int backup_index = find_newest_super_backup(fs_info);
2818         struct btrfs_super_block *sb = fs_info->super_copy;
2819         struct btrfs_root *tree_root = fs_info->tree_root;
2820         bool handle_error = false;
2821         int ret = 0;
2822         int i;
2823
2824         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2825                 if (handle_error) {
2826                         if (!IS_ERR(tree_root->node))
2827                                 free_extent_buffer(tree_root->node);
2828                         tree_root->node = NULL;
2829
2830                         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2831                                 break;
2832
2833                         free_root_pointers(fs_info, 0);
2834
2835                         /*
2836                          * Don't use the log in recovery mode, it won't be
2837                          * valid
2838                          */
2839                         btrfs_set_super_log_root(sb, 0);
2840
2841                         /* We can't trust the free space cache either */
2842                         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2843
2844                         ret = read_backup_root(fs_info, i);
2845                         backup_index = ret;
2846                         if (ret < 0)
2847                                 return ret;
2848                 }
2849
2850                 ret = load_important_roots(fs_info);
2851                 if (ret) {
2852                         handle_error = true;
2853                         continue;
2854                 }
2855
2856                 /*
2857                  * No need to hold btrfs_root::objectid_mutex since the fs
2858                  * hasn't been fully initialised and we are the only user
2859                  */
2860                 ret = btrfs_init_root_free_objectid(tree_root);
2861                 if (ret < 0) {
2862                         handle_error = true;
2863                         continue;
2864                 }
2865
2866                 ASSERT(tree_root->free_objectid <= BTRFS_LAST_FREE_OBJECTID);
2867
2868                 ret = btrfs_read_roots(fs_info);
2869                 if (ret < 0) {
2870                         handle_error = true;
2871                         continue;
2872                 }
2873
2874                 /* All successful */
2875                 fs_info->generation = btrfs_header_generation(tree_root->node);
2876                 fs_info->last_trans_committed = fs_info->generation;
2877                 fs_info->last_reloc_trans = 0;
2878
2879                 /* Always begin writing backup roots after the one being used */
2880                 if (backup_index < 0) {
2881                         fs_info->backup_root_index = 0;
2882                 } else {
2883                         fs_info->backup_root_index = backup_index + 1;
2884                         fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2885                 }
2886                 break;
2887         }
2888
2889         return ret;
2890 }
2891
2892 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2893 {
2894         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2895         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2896         INIT_LIST_HEAD(&fs_info->trans_list);
2897         INIT_LIST_HEAD(&fs_info->dead_roots);
2898         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2899         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2900         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2901         spin_lock_init(&fs_info->delalloc_root_lock);
2902         spin_lock_init(&fs_info->trans_lock);
2903         spin_lock_init(&fs_info->fs_roots_radix_lock);
2904         spin_lock_init(&fs_info->delayed_iput_lock);
2905         spin_lock_init(&fs_info->defrag_inodes_lock);
2906         spin_lock_init(&fs_info->super_lock);
2907         spin_lock_init(&fs_info->buffer_lock);
2908         spin_lock_init(&fs_info->unused_bgs_lock);
2909         spin_lock_init(&fs_info->treelog_bg_lock);
2910         spin_lock_init(&fs_info->zone_active_bgs_lock);
2911         spin_lock_init(&fs_info->relocation_bg_lock);
2912         rwlock_init(&fs_info->tree_mod_log_lock);
2913         rwlock_init(&fs_info->global_root_lock);
2914         mutex_init(&fs_info->unused_bg_unpin_mutex);
2915         mutex_init(&fs_info->reclaim_bgs_lock);
2916         mutex_init(&fs_info->reloc_mutex);
2917         mutex_init(&fs_info->delalloc_root_mutex);
2918         mutex_init(&fs_info->zoned_meta_io_lock);
2919         mutex_init(&fs_info->zoned_data_reloc_io_lock);
2920         seqlock_init(&fs_info->profiles_lock);
2921
2922         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_writers);
2923         btrfs_lockdep_init_map(fs_info, btrfs_trans_num_extwriters);
2924         btrfs_lockdep_init_map(fs_info, btrfs_trans_pending_ordered);
2925         btrfs_lockdep_init_map(fs_info, btrfs_ordered_extent);
2926         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_commit_start,
2927                                      BTRFS_LOCKDEP_TRANS_COMMIT_START);
2928         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_unblocked,
2929                                      BTRFS_LOCKDEP_TRANS_UNBLOCKED);
2930         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_super_committed,
2931                                      BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED);
2932         btrfs_state_lockdep_init_map(fs_info, btrfs_trans_completed,
2933                                      BTRFS_LOCKDEP_TRANS_COMPLETED);
2934
2935         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2936         INIT_LIST_HEAD(&fs_info->space_info);
2937         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2938         INIT_LIST_HEAD(&fs_info->unused_bgs);
2939         INIT_LIST_HEAD(&fs_info->reclaim_bgs);
2940         INIT_LIST_HEAD(&fs_info->zone_active_bgs);
2941 #ifdef CONFIG_BTRFS_DEBUG
2942         INIT_LIST_HEAD(&fs_info->allocated_roots);
2943         INIT_LIST_HEAD(&fs_info->allocated_ebs);
2944         spin_lock_init(&fs_info->eb_leak_lock);
2945 #endif
2946         extent_map_tree_init(&fs_info->mapping_tree);
2947         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2948                              BTRFS_BLOCK_RSV_GLOBAL);
2949         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2950         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2951         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2952         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2953                              BTRFS_BLOCK_RSV_DELOPS);
2954         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2955                              BTRFS_BLOCK_RSV_DELREFS);
2956
2957         atomic_set(&fs_info->async_delalloc_pages, 0);
2958         atomic_set(&fs_info->defrag_running, 0);
2959         atomic_set(&fs_info->nr_delayed_iputs, 0);
2960         atomic64_set(&fs_info->tree_mod_seq, 0);
2961         fs_info->global_root_tree = RB_ROOT;
2962         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2963         fs_info->metadata_ratio = 0;
2964         fs_info->defrag_inodes = RB_ROOT;
2965         atomic64_set(&fs_info->free_chunk_space, 0);
2966         fs_info->tree_mod_log = RB_ROOT;
2967         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2968         btrfs_init_ref_verify(fs_info);
2969
2970         fs_info->thread_pool_size = min_t(unsigned long,
2971                                           num_online_cpus() + 2, 8);
2972
2973         INIT_LIST_HEAD(&fs_info->ordered_roots);
2974         spin_lock_init(&fs_info->ordered_root_lock);
2975
2976         btrfs_init_scrub(fs_info);
2977 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2978         fs_info->check_integrity_print_mask = 0;
2979 #endif
2980         btrfs_init_balance(fs_info);
2981         btrfs_init_async_reclaim_work(fs_info);
2982
2983         rwlock_init(&fs_info->block_group_cache_lock);
2984         fs_info->block_group_cache_tree = RB_ROOT_CACHED;
2985
2986         extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2987                             IO_TREE_FS_EXCLUDED_EXTENTS);
2988
2989         mutex_init(&fs_info->ordered_operations_mutex);
2990         mutex_init(&fs_info->tree_log_mutex);
2991         mutex_init(&fs_info->chunk_mutex);
2992         mutex_init(&fs_info->transaction_kthread_mutex);
2993         mutex_init(&fs_info->cleaner_mutex);
2994         mutex_init(&fs_info->ro_block_group_mutex);
2995         init_rwsem(&fs_info->commit_root_sem);
2996         init_rwsem(&fs_info->cleanup_work_sem);
2997         init_rwsem(&fs_info->subvol_sem);
2998         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2999
3000         btrfs_init_dev_replace_locks(fs_info);
3001         btrfs_init_qgroup(fs_info);
3002         btrfs_discard_init(fs_info);
3003
3004         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
3005         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
3006
3007         init_waitqueue_head(&fs_info->transaction_throttle);
3008         init_waitqueue_head(&fs_info->transaction_wait);
3009         init_waitqueue_head(&fs_info->transaction_blocked_wait);
3010         init_waitqueue_head(&fs_info->async_submit_wait);
3011         init_waitqueue_head(&fs_info->delayed_iputs_wait);
3012
3013         /* Usable values until the real ones are cached from the superblock */
3014         fs_info->nodesize = 4096;
3015         fs_info->sectorsize = 4096;
3016         fs_info->sectorsize_bits = ilog2(4096);
3017         fs_info->stripesize = 4096;
3018
3019         fs_info->max_extent_size = BTRFS_MAX_EXTENT_SIZE;
3020
3021         spin_lock_init(&fs_info->swapfile_pins_lock);
3022         fs_info->swapfile_pins = RB_ROOT;
3023
3024         fs_info->bg_reclaim_threshold = BTRFS_DEFAULT_RECLAIM_THRESH;
3025         INIT_WORK(&fs_info->reclaim_bgs_work, btrfs_reclaim_bgs_work);
3026 }
3027
3028 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
3029 {
3030         int ret;
3031
3032         fs_info->sb = sb;
3033         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
3034         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
3035
3036         ret = percpu_counter_init(&fs_info->ordered_bytes, 0, GFP_KERNEL);
3037         if (ret)
3038                 return ret;
3039
3040         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
3041         if (ret)
3042                 return ret;
3043
3044         fs_info->dirty_metadata_batch = PAGE_SIZE *
3045                                         (1 + ilog2(nr_cpu_ids));
3046
3047         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
3048         if (ret)
3049                 return ret;
3050
3051         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
3052                         GFP_KERNEL);
3053         if (ret)
3054                 return ret;
3055
3056         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
3057                                         GFP_KERNEL);
3058         if (!fs_info->delayed_root)
3059                 return -ENOMEM;
3060         btrfs_init_delayed_root(fs_info->delayed_root);
3061
3062         if (sb_rdonly(sb))
3063                 set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
3064
3065         return btrfs_alloc_stripe_hash_table(fs_info);
3066 }
3067
3068 static int btrfs_uuid_rescan_kthread(void *data)
3069 {
3070         struct btrfs_fs_info *fs_info = data;
3071         int ret;
3072
3073         /*
3074          * 1st step is to iterate through the existing UUID tree and
3075          * to delete all entries that contain outdated data.
3076          * 2nd step is to add all missing entries to the UUID tree.
3077          */
3078         ret = btrfs_uuid_tree_iterate(fs_info);
3079         if (ret < 0) {
3080                 if (ret != -EINTR)
3081                         btrfs_warn(fs_info, "iterating uuid_tree failed %d",
3082                                    ret);
3083                 up(&fs_info->uuid_tree_rescan_sem);
3084                 return ret;
3085         }
3086         return btrfs_uuid_scan_kthread(data);
3087 }
3088
3089 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3090 {
3091         struct task_struct *task;
3092
3093         down(&fs_info->uuid_tree_rescan_sem);
3094         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3095         if (IS_ERR(task)) {
3096                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3097                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3098                 up(&fs_info->uuid_tree_rescan_sem);
3099                 return PTR_ERR(task);
3100         }
3101
3102         return 0;
3103 }
3104
3105 /*
3106  * Some options only have meaning at mount time and shouldn't persist across
3107  * remounts, or be displayed. Clear these at the end of mount and remount
3108  * code paths.
3109  */
3110 void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info)
3111 {
3112         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3113         btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE);
3114 }
3115
3116 /*
3117  * Mounting logic specific to read-write file systems. Shared by open_ctree
3118  * and btrfs_remount when remounting from read-only to read-write.
3119  */
3120 int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info)
3121 {
3122         int ret;
3123         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
3124         bool rebuild_free_space_tree = false;
3125
3126         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3127             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3128                 rebuild_free_space_tree = true;
3129         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3130                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3131                 btrfs_warn(fs_info, "free space tree is invalid");
3132                 rebuild_free_space_tree = true;
3133         }
3134
3135         if (rebuild_free_space_tree) {
3136                 btrfs_info(fs_info, "rebuilding free space tree");
3137                 ret = btrfs_rebuild_free_space_tree(fs_info);
3138                 if (ret) {
3139                         btrfs_warn(fs_info,
3140                                    "failed to rebuild free space tree: %d", ret);
3141                         goto out;
3142                 }
3143         }
3144
3145         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3146             !btrfs_test_opt(fs_info, FREE_SPACE_TREE)) {
3147                 btrfs_info(fs_info, "disabling free space tree");
3148                 ret = btrfs_delete_free_space_tree(fs_info);
3149                 if (ret) {
3150                         btrfs_warn(fs_info,
3151                                    "failed to disable free space tree: %d", ret);
3152                         goto out;
3153                 }
3154         }
3155
3156         /*
3157          * btrfs_find_orphan_roots() is responsible for finding all the dead
3158          * roots (with 0 refs), flag them with BTRFS_ROOT_DEAD_TREE and load
3159          * them into the fs_info->fs_roots_radix tree. This must be done before
3160          * calling btrfs_orphan_cleanup() on the tree root. If we don't do it
3161          * first, then btrfs_orphan_cleanup() will delete a dead root's orphan
3162          * item before the root's tree is deleted - this means that if we unmount
3163          * or crash before the deletion completes, on the next mount we will not
3164          * delete what remains of the tree because the orphan item does not
3165          * exists anymore, which is what tells us we have a pending deletion.
3166          */
3167         ret = btrfs_find_orphan_roots(fs_info);
3168         if (ret)
3169                 goto out;
3170
3171         ret = btrfs_cleanup_fs_roots(fs_info);
3172         if (ret)
3173                 goto out;
3174
3175         down_read(&fs_info->cleanup_work_sem);
3176         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3177             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3178                 up_read(&fs_info->cleanup_work_sem);
3179                 goto out;
3180         }
3181         up_read(&fs_info->cleanup_work_sem);
3182
3183         mutex_lock(&fs_info->cleaner_mutex);
3184         ret = btrfs_recover_relocation(fs_info);
3185         mutex_unlock(&fs_info->cleaner_mutex);
3186         if (ret < 0) {
3187                 btrfs_warn(fs_info, "failed to recover relocation: %d", ret);
3188                 goto out;
3189         }
3190
3191         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3192             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3193                 btrfs_info(fs_info, "creating free space tree");
3194                 ret = btrfs_create_free_space_tree(fs_info);
3195                 if (ret) {
3196                         btrfs_warn(fs_info,
3197                                 "failed to create free space tree: %d", ret);
3198                         goto out;
3199                 }
3200         }
3201
3202         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) {
3203                 ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
3204                 if (ret)
3205                         goto out;
3206         }
3207
3208         ret = btrfs_resume_balance_async(fs_info);
3209         if (ret)
3210                 goto out;
3211
3212         ret = btrfs_resume_dev_replace_async(fs_info);
3213         if (ret) {
3214                 btrfs_warn(fs_info, "failed to resume dev_replace");
3215                 goto out;
3216         }
3217
3218         btrfs_qgroup_rescan_resume(fs_info);
3219
3220         if (!fs_info->uuid_root) {
3221                 btrfs_info(fs_info, "creating UUID tree");
3222                 ret = btrfs_create_uuid_tree(fs_info);
3223                 if (ret) {
3224                         btrfs_warn(fs_info,
3225                                    "failed to create the UUID tree %d", ret);
3226                         goto out;
3227                 }
3228         }
3229
3230 out:
3231         return ret;
3232 }
3233
3234 /*
3235  * Do various sanity and dependency checks of different features.
3236  *
3237  * @is_rw_mount:        If the mount is read-write.
3238  *
3239  * This is the place for less strict checks (like for subpage or artificial
3240  * feature dependencies).
3241  *
3242  * For strict checks or possible corruption detection, see
3243  * btrfs_validate_super().
3244  *
3245  * This should be called after btrfs_parse_options(), as some mount options
3246  * (space cache related) can modify on-disk format like free space tree and
3247  * screw up certain feature dependencies.
3248  */
3249 int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount)
3250 {
3251         struct btrfs_super_block *disk_super = fs_info->super_copy;
3252         u64 incompat = btrfs_super_incompat_flags(disk_super);
3253         const u64 compat_ro = btrfs_super_compat_ro_flags(disk_super);
3254         const u64 compat_ro_unsupp = (compat_ro & ~BTRFS_FEATURE_COMPAT_RO_SUPP);
3255
3256         if (incompat & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
3257                 btrfs_err(fs_info,
3258                 "cannot mount because of unknown incompat features (0x%llx)",
3259                     incompat);
3260                 return -EINVAL;
3261         }
3262
3263         /* Runtime limitation for mixed block groups. */
3264         if ((incompat & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3265             (fs_info->sectorsize != fs_info->nodesize)) {
3266                 btrfs_err(fs_info,
3267 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3268                         fs_info->nodesize, fs_info->sectorsize);
3269                 return -EINVAL;
3270         }
3271
3272         /* Mixed backref is an always-enabled feature. */
3273         incompat |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3274
3275         /* Set compression related flags just in case. */
3276         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3277                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3278         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3279                 incompat |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3280
3281         /*
3282          * An ancient flag, which should really be marked deprecated.
3283          * Such runtime limitation doesn't really need a incompat flag.
3284          */
3285         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE)
3286                 incompat |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3287
3288         if (compat_ro_unsupp && is_rw_mount) {
3289                 btrfs_err(fs_info,
3290         "cannot mount read-write because of unknown compat_ro features (0x%llx)",
3291                        compat_ro);
3292                 return -EINVAL;
3293         }
3294
3295         /*
3296          * We have unsupported RO compat features, although RO mounted, we
3297          * should not cause any metadata writes, including log replay.
3298          * Or we could screw up whatever the new feature requires.
3299          */
3300         if (compat_ro_unsupp && btrfs_super_log_root(disk_super) &&
3301             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3302                 btrfs_err(fs_info,
3303 "cannot replay dirty log with unsupported compat_ro features (0x%llx), try rescue=nologreplay",
3304                           compat_ro);
3305                 return -EINVAL;
3306         }
3307
3308         /*
3309          * Artificial limitations for block group tree, to force
3310          * block-group-tree to rely on no-holes and free-space-tree.
3311          */
3312         if (btrfs_fs_compat_ro(fs_info, BLOCK_GROUP_TREE) &&
3313             (!btrfs_fs_incompat(fs_info, NO_HOLES) ||
3314              !btrfs_test_opt(fs_info, FREE_SPACE_TREE))) {
3315                 btrfs_err(fs_info,
3316 "block-group-tree feature requires no-holes and free-space-tree features");
3317                 return -EINVAL;
3318         }
3319
3320         /*
3321          * Subpage runtime limitation on v1 cache.
3322          *
3323          * V1 space cache still has some hard codeed PAGE_SIZE usage, while
3324          * we're already defaulting to v2 cache, no need to bother v1 as it's
3325          * going to be deprecated anyway.
3326          */
3327         if (fs_info->sectorsize < PAGE_SIZE && btrfs_test_opt(fs_info, SPACE_CACHE)) {
3328                 btrfs_warn(fs_info,
3329         "v1 space cache is not supported for page size %lu with sectorsize %u",
3330                            PAGE_SIZE, fs_info->sectorsize);
3331                 return -EINVAL;
3332         }
3333
3334         /* This can be called by remount, we need to protect the super block. */
3335         spin_lock(&fs_info->super_lock);
3336         btrfs_set_super_incompat_flags(disk_super, incompat);
3337         spin_unlock(&fs_info->super_lock);
3338
3339         return 0;
3340 }
3341
3342 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
3343                       char *options)
3344 {
3345         u32 sectorsize;
3346         u32 nodesize;
3347         u32 stripesize;
3348         u64 generation;
3349         u64 features;
3350         u16 csum_type;
3351         struct btrfs_super_block *disk_super;
3352         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3353         struct btrfs_root *tree_root;
3354         struct btrfs_root *chunk_root;
3355         int ret;
3356         int level;
3357
3358         ret = init_mount_fs_info(fs_info, sb);
3359         if (ret)
3360                 goto fail;
3361
3362         /* These need to be init'ed before we start creating inodes and such. */
3363         tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
3364                                      GFP_KERNEL);
3365         fs_info->tree_root = tree_root;
3366         chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
3367                                       GFP_KERNEL);
3368         fs_info->chunk_root = chunk_root;
3369         if (!tree_root || !chunk_root) {
3370                 ret = -ENOMEM;
3371                 goto fail;
3372         }
3373
3374         ret = btrfs_init_btree_inode(sb);
3375         if (ret)
3376                 goto fail;
3377
3378         invalidate_bdev(fs_devices->latest_dev->bdev);
3379
3380         /*
3381          * Read super block and check the signature bytes only
3382          */
3383         disk_super = btrfs_read_dev_super(fs_devices->latest_dev->bdev);
3384         if (IS_ERR(disk_super)) {
3385                 ret = PTR_ERR(disk_super);
3386                 goto fail_alloc;
3387         }
3388
3389         /*
3390          * Verify the type first, if that or the checksum value are
3391          * corrupted, we'll find out
3392          */
3393         csum_type = btrfs_super_csum_type(disk_super);
3394         if (!btrfs_supported_super_csum(csum_type)) {
3395                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
3396                           csum_type);
3397                 ret = -EINVAL;
3398                 btrfs_release_disk_super(disk_super);
3399                 goto fail_alloc;
3400         }
3401
3402         fs_info->csum_size = btrfs_super_csum_size(disk_super);
3403
3404         ret = btrfs_init_csum_hash(fs_info, csum_type);
3405         if (ret) {
3406                 btrfs_release_disk_super(disk_super);
3407                 goto fail_alloc;
3408         }
3409
3410         /*
3411          * We want to check superblock checksum, the type is stored inside.
3412          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
3413          */
3414         if (btrfs_check_super_csum(fs_info, disk_super)) {
3415                 btrfs_err(fs_info, "superblock checksum mismatch");
3416                 ret = -EINVAL;
3417                 btrfs_release_disk_super(disk_super);
3418                 goto fail_alloc;
3419         }
3420
3421         /*
3422          * super_copy is zeroed at allocation time and we never touch the
3423          * following bytes up to INFO_SIZE, the checksum is calculated from
3424          * the whole block of INFO_SIZE
3425          */
3426         memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
3427         btrfs_release_disk_super(disk_super);
3428
3429         disk_super = fs_info->super_copy;
3430
3431
3432         features = btrfs_super_flags(disk_super);
3433         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3434                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3435                 btrfs_set_super_flags(disk_super, features);
3436                 btrfs_info(fs_info,
3437                         "found metadata UUID change in progress flag, clearing");
3438         }
3439
3440         memcpy(fs_info->super_for_commit, fs_info->super_copy,
3441                sizeof(*fs_info->super_for_commit));
3442
3443         ret = btrfs_validate_mount_super(fs_info);
3444         if (ret) {
3445                 btrfs_err(fs_info, "superblock contains fatal errors");
3446                 ret = -EINVAL;
3447                 goto fail_alloc;
3448         }
3449
3450         if (!btrfs_super_root(disk_super)) {
3451                 btrfs_err(fs_info, "invalid superblock tree root bytenr");
3452                 ret = -EINVAL;
3453                 goto fail_alloc;
3454         }
3455
3456         /* check FS state, whether FS is broken. */
3457         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3458                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3459
3460         /*
3461          * In the long term, we'll store the compression type in the super
3462          * block, and it'll be used for per file compression control.
3463          */
3464         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3465
3466
3467         /* Set up fs_info before parsing mount options */
3468         nodesize = btrfs_super_nodesize(disk_super);
3469         sectorsize = btrfs_super_sectorsize(disk_super);
3470         stripesize = sectorsize;
3471         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3472         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3473
3474         fs_info->nodesize = nodesize;
3475         fs_info->sectorsize = sectorsize;
3476         fs_info->sectorsize_bits = ilog2(sectorsize);
3477         fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3478         fs_info->stripesize = stripesize;
3479
3480         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3481         if (ret)
3482                 goto fail_alloc;
3483
3484         ret = btrfs_check_features(fs_info, !sb_rdonly(sb));
3485         if (ret < 0)
3486                 goto fail_alloc;
3487
3488         if (sectorsize < PAGE_SIZE) {
3489                 struct btrfs_subpage_info *subpage_info;
3490
3491                 /*
3492                  * V1 space cache has some hardcoded PAGE_SIZE usage, and is
3493                  * going to be deprecated.
3494                  *
3495                  * Force to use v2 cache for subpage case.
3496                  */
3497                 btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
3498                 btrfs_set_and_info(fs_info, FREE_SPACE_TREE,
3499                         "forcing free space tree for sector size %u with page size %lu",
3500                         sectorsize, PAGE_SIZE);
3501
3502                 btrfs_warn(fs_info,
3503                 "read-write for sector size %u with page size %lu is experimental",
3504                            sectorsize, PAGE_SIZE);
3505                 subpage_info = kzalloc(sizeof(*subpage_info), GFP_KERNEL);
3506                 if (!subpage_info) {
3507                         ret = -ENOMEM;
3508                         goto fail_alloc;
3509                 }
3510                 btrfs_init_subpage_info(subpage_info, sectorsize);
3511                 fs_info->subpage_info = subpage_info;
3512         }
3513
3514         ret = btrfs_init_workqueues(fs_info);
3515         if (ret)
3516                 goto fail_sb_buffer;
3517
3518         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3519         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3520
3521         sb->s_blocksize = sectorsize;
3522         sb->s_blocksize_bits = blksize_bits(sectorsize);
3523         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3524
3525         mutex_lock(&fs_info->chunk_mutex);
3526         ret = btrfs_read_sys_array(fs_info);
3527         mutex_unlock(&fs_info->chunk_mutex);
3528         if (ret) {
3529                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3530                 goto fail_sb_buffer;
3531         }
3532
3533         generation = btrfs_super_chunk_root_generation(disk_super);
3534         level = btrfs_super_chunk_root_level(disk_super);
3535         ret = load_super_root(chunk_root, btrfs_super_chunk_root(disk_super),
3536                               generation, level);
3537         if (ret) {
3538                 btrfs_err(fs_info, "failed to read chunk root");
3539                 goto fail_tree_roots;
3540         }
3541
3542         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3543                            offsetof(struct btrfs_header, chunk_tree_uuid),
3544                            BTRFS_UUID_SIZE);
3545
3546         ret = btrfs_read_chunk_tree(fs_info);
3547         if (ret) {
3548                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3549                 goto fail_tree_roots;
3550         }
3551
3552         /*
3553          * At this point we know all the devices that make this filesystem,
3554          * including the seed devices but we don't know yet if the replace
3555          * target is required. So free devices that are not part of this
3556          * filesystem but skip the replace target device which is checked
3557          * below in btrfs_init_dev_replace().
3558          */
3559         btrfs_free_extra_devids(fs_devices);
3560         if (!fs_devices->latest_dev->bdev) {
3561                 btrfs_err(fs_info, "failed to read devices");
3562                 ret = -EIO;
3563                 goto fail_tree_roots;
3564         }
3565
3566         ret = init_tree_roots(fs_info);
3567         if (ret)
3568                 goto fail_tree_roots;
3569
3570         /*
3571          * Get zone type information of zoned block devices. This will also
3572          * handle emulation of a zoned filesystem if a regular device has the
3573          * zoned incompat feature flag set.
3574          */
3575         ret = btrfs_get_dev_zone_info_all_devices(fs_info);
3576         if (ret) {
3577                 btrfs_err(fs_info,
3578                           "zoned: failed to read device zone info: %d", ret);
3579                 goto fail_block_groups;
3580         }
3581
3582         /*
3583          * If we have a uuid root and we're not being told to rescan we need to
3584          * check the generation here so we can set the
3585          * BTRFS_FS_UPDATE_UUID_TREE_GEN bit.  Otherwise we could commit the
3586          * transaction during a balance or the log replay without updating the
3587          * uuid generation, and then if we crash we would rescan the uuid tree,
3588          * even though it was perfectly fine.
3589          */
3590         if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3591             fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3592                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3593
3594         ret = btrfs_verify_dev_extents(fs_info);
3595         if (ret) {
3596                 btrfs_err(fs_info,
3597                           "failed to verify dev extents against chunks: %d",
3598                           ret);
3599                 goto fail_block_groups;
3600         }
3601         ret = btrfs_recover_balance(fs_info);
3602         if (ret) {
3603                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3604                 goto fail_block_groups;
3605         }
3606
3607         ret = btrfs_init_dev_stats(fs_info);
3608         if (ret) {
3609                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3610                 goto fail_block_groups;
3611         }
3612
3613         ret = btrfs_init_dev_replace(fs_info);
3614         if (ret) {
3615                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3616                 goto fail_block_groups;
3617         }
3618
3619         ret = btrfs_check_zoned_mode(fs_info);
3620         if (ret) {
3621                 btrfs_err(fs_info, "failed to initialize zoned mode: %d",
3622                           ret);
3623                 goto fail_block_groups;
3624         }
3625
3626         ret = btrfs_sysfs_add_fsid(fs_devices);
3627         if (ret) {
3628                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3629                                 ret);
3630                 goto fail_block_groups;
3631         }
3632
3633         ret = btrfs_sysfs_add_mounted(fs_info);
3634         if (ret) {
3635                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3636                 goto fail_fsdev_sysfs;
3637         }
3638
3639         ret = btrfs_init_space_info(fs_info);
3640         if (ret) {
3641                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3642                 goto fail_sysfs;
3643         }
3644
3645         ret = btrfs_read_block_groups(fs_info);
3646         if (ret) {
3647                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3648                 goto fail_sysfs;
3649         }
3650
3651         btrfs_free_zone_cache(fs_info);
3652
3653         if (!sb_rdonly(sb) && fs_info->fs_devices->missing_devices &&
3654             !btrfs_check_rw_degradable(fs_info, NULL)) {
3655                 btrfs_warn(fs_info,
3656                 "writable mount is not allowed due to too many missing devices");
3657                 ret = -EINVAL;
3658                 goto fail_sysfs;
3659         }
3660
3661         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, fs_info,
3662                                                "btrfs-cleaner");
3663         if (IS_ERR(fs_info->cleaner_kthread)) {
3664                 ret = PTR_ERR(fs_info->cleaner_kthread);
3665                 goto fail_sysfs;
3666         }
3667
3668         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3669                                                    tree_root,
3670                                                    "btrfs-transaction");
3671         if (IS_ERR(fs_info->transaction_kthread)) {
3672                 ret = PTR_ERR(fs_info->transaction_kthread);
3673                 goto fail_cleaner;
3674         }
3675
3676         if (!btrfs_test_opt(fs_info, NOSSD) &&
3677             !fs_info->fs_devices->rotating) {
3678                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3679         }
3680
3681         /*
3682          * For devices supporting discard turn on discard=async automatically,
3683          * unless it's already set or disabled. This could be turned off by
3684          * nodiscard for the same mount.
3685          */
3686         if (!(btrfs_test_opt(fs_info, DISCARD_SYNC) ||
3687               btrfs_test_opt(fs_info, DISCARD_ASYNC) ||
3688               btrfs_test_opt(fs_info, NODISCARD)) &&
3689             fs_info->fs_devices->discardable) {
3690                 btrfs_set_and_info(fs_info, DISCARD_ASYNC,
3691                                    "auto enabling async discard");
3692         }
3693
3694 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3695         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3696                 ret = btrfsic_mount(fs_info, fs_devices,
3697                                     btrfs_test_opt(fs_info,
3698                                         CHECK_INTEGRITY_DATA) ? 1 : 0,
3699                                     fs_info->check_integrity_print_mask);
3700                 if (ret)
3701                         btrfs_warn(fs_info,
3702                                 "failed to initialize integrity check module: %d",
3703                                 ret);
3704         }
3705 #endif
3706         ret = btrfs_read_qgroup_config(fs_info);
3707         if (ret)
3708                 goto fail_trans_kthread;
3709
3710         if (btrfs_build_ref_tree(fs_info))
3711                 btrfs_err(fs_info, "couldn't build ref tree");
3712
3713         /* do not make disk changes in broken FS or nologreplay is given */
3714         if (btrfs_super_log_root(disk_super) != 0 &&
3715             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3716                 btrfs_info(fs_info, "start tree-log replay");
3717                 ret = btrfs_replay_log(fs_info, fs_devices);
3718                 if (ret)
3719                         goto fail_qgroup;
3720         }
3721
3722         fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3723         if (IS_ERR(fs_info->fs_root)) {
3724                 ret = PTR_ERR(fs_info->fs_root);
3725                 btrfs_warn(fs_info, "failed to read fs tree: %d", ret);
3726                 fs_info->fs_root = NULL;
3727                 goto fail_qgroup;
3728         }
3729
3730         if (sb_rdonly(sb))
3731                 goto clear_oneshot;
3732
3733         ret = btrfs_start_pre_rw_mount(fs_info);
3734         if (ret) {
3735                 close_ctree(fs_info);
3736                 return ret;
3737         }
3738         btrfs_discard_resume(fs_info);
3739
3740         if (fs_info->uuid_root &&
3741             (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3742              fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) {
3743                 btrfs_info(fs_info, "checking UUID tree");
3744                 ret = btrfs_check_uuid_tree(fs_info);
3745                 if (ret) {
3746                         btrfs_warn(fs_info,
3747                                 "failed to check the UUID tree: %d", ret);
3748                         close_ctree(fs_info);
3749                         return ret;
3750                 }
3751         }
3752
3753         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3754
3755         /* Kick the cleaner thread so it'll start deleting snapshots. */
3756         if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
3757                 wake_up_process(fs_info->cleaner_kthread);
3758
3759 clear_oneshot:
3760         btrfs_clear_oneshot_options(fs_info);
3761         return 0;
3762
3763 fail_qgroup:
3764         btrfs_free_qgroup_config(fs_info);
3765 fail_trans_kthread:
3766         kthread_stop(fs_info->transaction_kthread);
3767         btrfs_cleanup_transaction(fs_info);
3768         btrfs_free_fs_roots(fs_info);
3769 fail_cleaner:
3770         kthread_stop(fs_info->cleaner_kthread);
3771
3772         /*
3773          * make sure we're done with the btree inode before we stop our
3774          * kthreads
3775          */
3776         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3777
3778 fail_sysfs:
3779         btrfs_sysfs_remove_mounted(fs_info);
3780
3781 fail_fsdev_sysfs:
3782         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3783
3784 fail_block_groups:
3785         btrfs_put_block_group_cache(fs_info);
3786
3787 fail_tree_roots:
3788         if (fs_info->data_reloc_root)
3789                 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3790         free_root_pointers(fs_info, true);
3791         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3792
3793 fail_sb_buffer:
3794         btrfs_stop_all_workers(fs_info);
3795         btrfs_free_block_groups(fs_info);
3796 fail_alloc:
3797         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3798
3799         iput(fs_info->btree_inode);
3800 fail:
3801         btrfs_close_devices(fs_info->fs_devices);
3802         ASSERT(ret < 0);
3803         return ret;
3804 }
3805 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3806
3807 static void btrfs_end_super_write(struct bio *bio)
3808 {
3809         struct btrfs_device *device = bio->bi_private;
3810         struct bio_vec *bvec;
3811         struct bvec_iter_all iter_all;
3812         struct page *page;
3813
3814         bio_for_each_segment_all(bvec, bio, iter_all) {
3815                 page = bvec->bv_page;
3816
3817                 if (bio->bi_status) {
3818                         btrfs_warn_rl_in_rcu(device->fs_info,
3819                                 "lost page write due to IO error on %s (%d)",
3820                                 btrfs_dev_name(device),
3821                                 blk_status_to_errno(bio->bi_status));
3822                         ClearPageUptodate(page);
3823                         SetPageError(page);
3824                         btrfs_dev_stat_inc_and_print(device,
3825                                                      BTRFS_DEV_STAT_WRITE_ERRS);
3826                 } else {
3827                         SetPageUptodate(page);
3828                 }
3829
3830                 put_page(page);
3831                 unlock_page(page);
3832         }
3833
3834         bio_put(bio);
3835 }
3836
3837 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3838                                                    int copy_num, bool drop_cache)
3839 {
3840         struct btrfs_super_block *super;
3841         struct page *page;
3842         u64 bytenr, bytenr_orig;
3843         struct address_space *mapping = bdev->bd_inode->i_mapping;
3844         int ret;
3845
3846         bytenr_orig = btrfs_sb_offset(copy_num);
3847         ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr);
3848         if (ret == -ENOENT)
3849                 return ERR_PTR(-EINVAL);
3850         else if (ret)
3851                 return ERR_PTR(ret);
3852
3853         if (bytenr + BTRFS_SUPER_INFO_SIZE >= bdev_nr_bytes(bdev))
3854                 return ERR_PTR(-EINVAL);
3855
3856         if (drop_cache) {
3857                 /* This should only be called with the primary sb. */
3858                 ASSERT(copy_num == 0);
3859
3860                 /*
3861                  * Drop the page of the primary superblock, so later read will
3862                  * always read from the device.
3863                  */
3864                 invalidate_inode_pages2_range(mapping,
3865                                 bytenr >> PAGE_SHIFT,
3866                                 (bytenr + BTRFS_SUPER_INFO_SIZE) >> PAGE_SHIFT);
3867         }
3868
3869         page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3870         if (IS_ERR(page))
3871                 return ERR_CAST(page);
3872
3873         super = page_address(page);
3874         if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3875                 btrfs_release_disk_super(super);
3876                 return ERR_PTR(-ENODATA);
3877         }
3878
3879         if (btrfs_super_bytenr(super) != bytenr_orig) {
3880                 btrfs_release_disk_super(super);
3881                 return ERR_PTR(-EINVAL);
3882         }
3883
3884         return super;
3885 }
3886
3887
3888 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3889 {
3890         struct btrfs_super_block *super, *latest = NULL;
3891         int i;
3892         u64 transid = 0;
3893
3894         /* we would like to check all the supers, but that would make
3895          * a btrfs mount succeed after a mkfs from a different FS.
3896          * So, we need to add a special mount option to scan for
3897          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3898          */
3899         for (i = 0; i < 1; i++) {
3900                 super = btrfs_read_dev_one_super(bdev, i, false);
3901                 if (IS_ERR(super))
3902                         continue;
3903
3904                 if (!latest || btrfs_super_generation(super) > transid) {
3905                         if (latest)
3906                                 btrfs_release_disk_super(super);
3907
3908                         latest = super;
3909                         transid = btrfs_super_generation(super);
3910                 }
3911         }
3912
3913         return super;
3914 }
3915
3916 /*
3917  * Write superblock @sb to the @device. Do not wait for completion, all the
3918  * pages we use for writing are locked.
3919  *
3920  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3921  * the expected device size at commit time. Note that max_mirrors must be
3922  * same for write and wait phases.
3923  *
3924  * Return number of errors when page is not found or submission fails.
3925  */
3926 static int write_dev_supers(struct btrfs_device *device,
3927                             struct btrfs_super_block *sb, int max_mirrors)
3928 {
3929         struct btrfs_fs_info *fs_info = device->fs_info;
3930         struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3931         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3932         int i;
3933         int errors = 0;
3934         int ret;
3935         u64 bytenr, bytenr_orig;
3936
3937         if (max_mirrors == 0)
3938                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3939
3940         shash->tfm = fs_info->csum_shash;
3941
3942         for (i = 0; i < max_mirrors; i++) {
3943                 struct page *page;
3944                 struct bio *bio;
3945                 struct btrfs_super_block *disk_super;
3946
3947                 bytenr_orig = btrfs_sb_offset(i);
3948                 ret = btrfs_sb_log_location(device, i, WRITE, &bytenr);
3949                 if (ret == -ENOENT) {
3950                         continue;
3951                 } else if (ret < 0) {
3952                         btrfs_err(device->fs_info,
3953                                 "couldn't get super block location for mirror %d",
3954                                 i);
3955                         errors++;
3956                         continue;
3957                 }
3958                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3959                     device->commit_total_bytes)
3960                         break;
3961
3962                 btrfs_set_super_bytenr(sb, bytenr_orig);
3963
3964                 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3965                                     BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3966                                     sb->csum);
3967
3968                 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3969                                            GFP_NOFS);
3970                 if (!page) {
3971                         btrfs_err(device->fs_info,
3972                             "couldn't get super block page for bytenr %llu",
3973                             bytenr);
3974                         errors++;
3975                         continue;
3976                 }
3977
3978                 /* Bump the refcount for wait_dev_supers() */
3979                 get_page(page);
3980
3981                 disk_super = page_address(page);
3982                 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3983
3984                 /*
3985                  * Directly use bios here instead of relying on the page cache
3986                  * to do I/O, so we don't lose the ability to do integrity
3987                  * checking.
3988                  */
3989                 bio = bio_alloc(device->bdev, 1,
3990                                 REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO,
3991                                 GFP_NOFS);
3992                 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3993                 bio->bi_private = device;
3994                 bio->bi_end_io = btrfs_end_super_write;
3995                 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3996                                offset_in_page(bytenr));
3997
3998                 /*
3999                  * We FUA only the first super block.  The others we allow to
4000                  * go down lazy and there's a short window where the on-disk
4001                  * copies might still contain the older version.
4002                  */
4003                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
4004                         bio->bi_opf |= REQ_FUA;
4005
4006                 btrfsic_check_bio(bio);
4007                 submit_bio(bio);
4008
4009                 if (btrfs_advance_sb_log(device, i))
4010                         errors++;
4011         }
4012         return errors < i ? 0 : -1;
4013 }
4014
4015 /*
4016  * Wait for write completion of superblocks done by write_dev_supers,
4017  * @max_mirrors same for write and wait phases.
4018  *
4019  * Return number of errors when page is not found or not marked up to
4020  * date.
4021  */
4022 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
4023 {
4024         int i;
4025         int errors = 0;
4026         bool primary_failed = false;
4027         int ret;
4028         u64 bytenr;
4029
4030         if (max_mirrors == 0)
4031                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
4032
4033         for (i = 0; i < max_mirrors; i++) {
4034                 struct page *page;
4035
4036                 ret = btrfs_sb_log_location(device, i, READ, &bytenr);
4037                 if (ret == -ENOENT) {
4038                         break;
4039                 } else if (ret < 0) {
4040                         errors++;
4041                         if (i == 0)
4042                                 primary_failed = true;
4043                         continue;
4044                 }
4045                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
4046                     device->commit_total_bytes)
4047                         break;
4048
4049                 page = find_get_page(device->bdev->bd_inode->i_mapping,
4050                                      bytenr >> PAGE_SHIFT);
4051                 if (!page) {
4052                         errors++;
4053                         if (i == 0)
4054                                 primary_failed = true;
4055                         continue;
4056                 }
4057                 /* Page is submitted locked and unlocked once the IO completes */
4058                 wait_on_page_locked(page);
4059                 if (PageError(page)) {
4060                         errors++;
4061                         if (i == 0)
4062                                 primary_failed = true;
4063                 }
4064
4065                 /* Drop our reference */
4066                 put_page(page);
4067
4068                 /* Drop the reference from the writing run */
4069                 put_page(page);
4070         }
4071
4072         /* log error, force error return */
4073         if (primary_failed) {
4074                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
4075                           device->devid);
4076                 return -1;
4077         }
4078
4079         return errors < i ? 0 : -1;
4080 }
4081
4082 /*
4083  * endio for the write_dev_flush, this will wake anyone waiting
4084  * for the barrier when it is done
4085  */
4086 static void btrfs_end_empty_barrier(struct bio *bio)
4087 {
4088         bio_uninit(bio);
4089         complete(bio->bi_private);
4090 }
4091
4092 /*
4093  * Submit a flush request to the device if it supports it. Error handling is
4094  * done in the waiting counterpart.
4095  */
4096 static void write_dev_flush(struct btrfs_device *device)
4097 {
4098         struct bio *bio = &device->flush_bio;
4099
4100         device->last_flush_error = BLK_STS_OK;
4101
4102 #ifndef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4103         /*
4104          * When a disk has write caching disabled, we skip submission of a bio
4105          * with flush and sync requests before writing the superblock, since
4106          * it's not needed. However when the integrity checker is enabled, this
4107          * results in reports that there are metadata blocks referred by a
4108          * superblock that were not properly flushed. So don't skip the bio
4109          * submission only when the integrity checker is enabled for the sake
4110          * of simplicity, since this is a debug tool and not meant for use in
4111          * non-debug builds.
4112          */
4113         if (!bdev_write_cache(device->bdev))
4114                 return;
4115 #endif
4116
4117         bio_init(bio, device->bdev, NULL, 0,
4118                  REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH);
4119         bio->bi_end_io = btrfs_end_empty_barrier;
4120         init_completion(&device->flush_wait);
4121         bio->bi_private = &device->flush_wait;
4122
4123         btrfsic_check_bio(bio);
4124         submit_bio(bio);
4125         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
4126 }
4127
4128 /*
4129  * If the flush bio has been submitted by write_dev_flush, wait for it.
4130  * Return true for any error, and false otherwise.
4131  */
4132 static bool wait_dev_flush(struct btrfs_device *device)
4133 {
4134         struct bio *bio = &device->flush_bio;
4135
4136         if (!test_and_clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
4137                 return false;
4138
4139         wait_for_completion_io(&device->flush_wait);
4140
4141         if (bio->bi_status) {
4142                 device->last_flush_error = bio->bi_status;
4143                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_FLUSH_ERRS);
4144                 return true;
4145         }
4146
4147         return false;
4148 }
4149
4150 /*
4151  * send an empty flush down to each device in parallel,
4152  * then wait for them
4153  */
4154 static int barrier_all_devices(struct btrfs_fs_info *info)
4155 {
4156         struct list_head *head;
4157         struct btrfs_device *dev;
4158         int errors_wait = 0;
4159
4160         lockdep_assert_held(&info->fs_devices->device_list_mutex);
4161         /* send down all the barriers */
4162         head = &info->fs_devices->devices;
4163         list_for_each_entry(dev, head, dev_list) {
4164                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4165                         continue;
4166                 if (!dev->bdev)
4167                         continue;
4168                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4169                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4170                         continue;
4171
4172                 write_dev_flush(dev);
4173         }
4174
4175         /* wait for all the barriers */
4176         list_for_each_entry(dev, head, dev_list) {
4177                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
4178                         continue;
4179                 if (!dev->bdev) {
4180                         errors_wait++;
4181                         continue;
4182                 }
4183                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4184                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4185                         continue;
4186
4187                 if (wait_dev_flush(dev))
4188                         errors_wait++;
4189         }
4190
4191         /*
4192          * Checks last_flush_error of disks in order to determine the device
4193          * state.
4194          */
4195         if (errors_wait && !btrfs_check_rw_degradable(info, NULL))
4196                 return -EIO;
4197
4198         return 0;
4199 }
4200
4201 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
4202 {
4203         int raid_type;
4204         int min_tolerated = INT_MAX;
4205
4206         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
4207             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
4208                 min_tolerated = min_t(int, min_tolerated,
4209                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
4210                                     tolerated_failures);
4211
4212         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4213                 if (raid_type == BTRFS_RAID_SINGLE)
4214                         continue;
4215                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
4216                         continue;
4217                 min_tolerated = min_t(int, min_tolerated,
4218                                     btrfs_raid_array[raid_type].
4219                                     tolerated_failures);
4220         }
4221
4222         if (min_tolerated == INT_MAX) {
4223                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
4224                 min_tolerated = 0;
4225         }
4226
4227         return min_tolerated;
4228 }
4229
4230 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
4231 {
4232         struct list_head *head;
4233         struct btrfs_device *dev;
4234         struct btrfs_super_block *sb;
4235         struct btrfs_dev_item *dev_item;
4236         int ret;
4237         int do_barriers;
4238         int max_errors;
4239         int total_errors = 0;
4240         u64 flags;
4241
4242         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
4243
4244         /*
4245          * max_mirrors == 0 indicates we're from commit_transaction,
4246          * not from fsync where the tree roots in fs_info have not
4247          * been consistent on disk.
4248          */
4249         if (max_mirrors == 0)
4250                 backup_super_roots(fs_info);
4251
4252         sb = fs_info->super_for_commit;
4253         dev_item = &sb->dev_item;
4254
4255         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4256         head = &fs_info->fs_devices->devices;
4257         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
4258
4259         if (do_barriers) {
4260                 ret = barrier_all_devices(fs_info);
4261                 if (ret) {
4262                         mutex_unlock(
4263                                 &fs_info->fs_devices->device_list_mutex);
4264                         btrfs_handle_fs_error(fs_info, ret,
4265                                               "errors while submitting device barriers.");
4266                         return ret;
4267                 }
4268         }
4269
4270         list_for_each_entry(dev, head, dev_list) {
4271                 if (!dev->bdev) {
4272                         total_errors++;
4273                         continue;
4274                 }
4275                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4276                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4277                         continue;
4278
4279                 btrfs_set_stack_device_generation(dev_item, 0);
4280                 btrfs_set_stack_device_type(dev_item, dev->type);
4281                 btrfs_set_stack_device_id(dev_item, dev->devid);
4282                 btrfs_set_stack_device_total_bytes(dev_item,
4283                                                    dev->commit_total_bytes);
4284                 btrfs_set_stack_device_bytes_used(dev_item,
4285                                                   dev->commit_bytes_used);
4286                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
4287                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
4288                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
4289                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
4290                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
4291                        BTRFS_FSID_SIZE);
4292
4293                 flags = btrfs_super_flags(sb);
4294                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
4295
4296                 ret = btrfs_validate_write_super(fs_info, sb);
4297                 if (ret < 0) {
4298                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4299                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
4300                                 "unexpected superblock corruption detected");
4301                         return -EUCLEAN;
4302                 }
4303
4304                 ret = write_dev_supers(dev, sb, max_mirrors);
4305                 if (ret)
4306                         total_errors++;
4307         }
4308         if (total_errors > max_errors) {
4309                 btrfs_err(fs_info, "%d errors while writing supers",
4310                           total_errors);
4311                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4312
4313                 /* FUA is masked off if unsupported and can't be the reason */
4314                 btrfs_handle_fs_error(fs_info, -EIO,
4315                                       "%d errors while writing supers",
4316                                       total_errors);
4317                 return -EIO;
4318         }
4319
4320         total_errors = 0;
4321         list_for_each_entry(dev, head, dev_list) {
4322                 if (!dev->bdev)
4323                         continue;
4324                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
4325                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
4326                         continue;
4327
4328                 ret = wait_dev_supers(dev, max_mirrors);
4329                 if (ret)
4330                         total_errors++;
4331         }
4332         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4333         if (total_errors > max_errors) {
4334                 btrfs_handle_fs_error(fs_info, -EIO,
4335                                       "%d errors while writing supers",
4336                                       total_errors);
4337                 return -EIO;
4338         }
4339         return 0;
4340 }
4341
4342 /* Drop a fs root from the radix tree and free it. */
4343 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
4344                                   struct btrfs_root *root)
4345 {
4346         bool drop_ref = false;
4347
4348         spin_lock(&fs_info->fs_roots_radix_lock);
4349         radix_tree_delete(&fs_info->fs_roots_radix,
4350                           (unsigned long)root->root_key.objectid);
4351         if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
4352                 drop_ref = true;
4353         spin_unlock(&fs_info->fs_roots_radix_lock);
4354
4355         if (BTRFS_FS_ERROR(fs_info)) {
4356                 ASSERT(root->log_root == NULL);
4357                 if (root->reloc_root) {
4358                         btrfs_put_root(root->reloc_root);
4359                         root->reloc_root = NULL;
4360                 }
4361         }
4362
4363         if (drop_ref)
4364                 btrfs_put_root(root);
4365 }
4366
4367 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
4368 {
4369         u64 root_objectid = 0;
4370         struct btrfs_root *gang[8];
4371         int i = 0;
4372         int err = 0;
4373         unsigned int ret = 0;
4374
4375         while (1) {
4376                 spin_lock(&fs_info->fs_roots_radix_lock);
4377                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4378                                              (void **)gang, root_objectid,
4379                                              ARRAY_SIZE(gang));
4380                 if (!ret) {
4381                         spin_unlock(&fs_info->fs_roots_radix_lock);
4382                         break;
4383                 }
4384                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
4385
4386                 for (i = 0; i < ret; i++) {
4387                         /* Avoid to grab roots in dead_roots */
4388                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4389                                 gang[i] = NULL;
4390                                 continue;
4391                         }
4392                         /* grab all the search result for later use */
4393                         gang[i] = btrfs_grab_root(gang[i]);
4394                 }
4395                 spin_unlock(&fs_info->fs_roots_radix_lock);
4396
4397                 for (i = 0; i < ret; i++) {
4398                         if (!gang[i])
4399                                 continue;
4400                         root_objectid = gang[i]->root_key.objectid;
4401                         err = btrfs_orphan_cleanup(gang[i]);
4402                         if (err)
4403                                 goto out;
4404                         btrfs_put_root(gang[i]);
4405                 }
4406                 root_objectid++;
4407         }
4408 out:
4409         /* release the uncleaned roots due to error */
4410         for (; i < ret; i++) {
4411                 if (gang[i])
4412                         btrfs_put_root(gang[i]);
4413         }
4414         return err;
4415 }
4416
4417 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4418 {
4419         struct btrfs_root *root = fs_info->tree_root;
4420         struct btrfs_trans_handle *trans;
4421
4422         mutex_lock(&fs_info->cleaner_mutex);
4423         btrfs_run_delayed_iputs(fs_info);
4424         mutex_unlock(&fs_info->cleaner_mutex);
4425         wake_up_process(fs_info->cleaner_kthread);
4426
4427         /* wait until ongoing cleanup work done */
4428         down_write(&fs_info->cleanup_work_sem);
4429         up_write(&fs_info->cleanup_work_sem);
4430
4431         trans = btrfs_join_transaction(root);
4432         if (IS_ERR(trans))
4433                 return PTR_ERR(trans);
4434         return btrfs_commit_transaction(trans);
4435 }
4436
4437 static void warn_about_uncommitted_trans(struct btrfs_fs_info *fs_info)
4438 {
4439         struct btrfs_transaction *trans;
4440         struct btrfs_transaction *tmp;
4441         bool found = false;
4442
4443         if (list_empty(&fs_info->trans_list))
4444                 return;
4445
4446         /*
4447          * This function is only called at the very end of close_ctree(),
4448          * thus no other running transaction, no need to take trans_lock.
4449          */
4450         ASSERT(test_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags));
4451         list_for_each_entry_safe(trans, tmp, &fs_info->trans_list, list) {
4452                 struct extent_state *cached = NULL;
4453                 u64 dirty_bytes = 0;
4454                 u64 cur = 0;
4455                 u64 found_start;
4456                 u64 found_end;
4457
4458                 found = true;
4459                 while (!find_first_extent_bit(&trans->dirty_pages, cur,
4460                         &found_start, &found_end, EXTENT_DIRTY, &cached)) {
4461                         dirty_bytes += found_end + 1 - found_start;
4462                         cur = found_end + 1;
4463                 }
4464                 btrfs_warn(fs_info,
4465         "transaction %llu (with %llu dirty metadata bytes) is not committed",
4466                            trans->transid, dirty_bytes);
4467                 btrfs_cleanup_one_transaction(trans, fs_info);
4468
4469                 if (trans == fs_info->running_transaction)
4470                         fs_info->running_transaction = NULL;
4471                 list_del_init(&trans->list);
4472
4473                 btrfs_put_transaction(trans);
4474                 trace_btrfs_transaction_commit(fs_info);
4475         }
4476         ASSERT(!found);
4477 }
4478
4479 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4480 {
4481         int ret;
4482
4483         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4484
4485         /*
4486          * If we had UNFINISHED_DROPS we could still be processing them, so
4487          * clear that bit and wake up relocation so it can stop.
4488          * We must do this before stopping the block group reclaim task, because
4489          * at btrfs_relocate_block_group() we wait for this bit, and after the
4490          * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
4491          * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
4492          * return 1.
4493          */
4494         btrfs_wake_unfinished_drop(fs_info);
4495
4496         /*
4497          * We may have the reclaim task running and relocating a data block group,
4498          * in which case it may create delayed iputs. So stop it before we park
4499          * the cleaner kthread otherwise we can get new delayed iputs after
4500          * parking the cleaner, and that can make the async reclaim task to hang
4501          * if it's waiting for delayed iputs to complete, since the cleaner is
4502          * parked and can not run delayed iputs - this will make us hang when
4503          * trying to stop the async reclaim task.
4504          */
4505         cancel_work_sync(&fs_info->reclaim_bgs_work);
4506         /*
4507          * We don't want the cleaner to start new transactions, add more delayed
4508          * iputs, etc. while we're closing. We can't use kthread_stop() yet
4509          * because that frees the task_struct, and the transaction kthread might
4510          * still try to wake up the cleaner.
4511          */
4512         kthread_park(fs_info->cleaner_kthread);
4513
4514         /* wait for the qgroup rescan worker to stop */
4515         btrfs_qgroup_wait_for_completion(fs_info, false);
4516
4517         /* wait for the uuid_scan task to finish */
4518         down(&fs_info->uuid_tree_rescan_sem);
4519         /* avoid complains from lockdep et al., set sem back to initial state */
4520         up(&fs_info->uuid_tree_rescan_sem);
4521
4522         /* pause restriper - we want to resume on mount */
4523         btrfs_pause_balance(fs_info);
4524
4525         btrfs_dev_replace_suspend_for_unmount(fs_info);
4526
4527         btrfs_scrub_cancel(fs_info);
4528
4529         /* wait for any defraggers to finish */
4530         wait_event(fs_info->transaction_wait,
4531                    (atomic_read(&fs_info->defrag_running) == 0));
4532
4533         /* clear out the rbtree of defraggable inodes */
4534         btrfs_cleanup_defrag_inodes(fs_info);
4535
4536         /*
4537          * After we parked the cleaner kthread, ordered extents may have
4538          * completed and created new delayed iputs. If one of the async reclaim
4539          * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
4540          * can hang forever trying to stop it, because if a delayed iput is
4541          * added after it ran btrfs_run_delayed_iputs() and before it called
4542          * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
4543          * no one else to run iputs.
4544          *
4545          * So wait for all ongoing ordered extents to complete and then run
4546          * delayed iputs. This works because once we reach this point no one
4547          * can either create new ordered extents nor create delayed iputs
4548          * through some other means.
4549          *
4550          * Also note that btrfs_wait_ordered_roots() is not safe here, because
4551          * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
4552          * but the delayed iput for the respective inode is made only when doing
4553          * the final btrfs_put_ordered_extent() (which must happen at
4554          * btrfs_finish_ordered_io() when we are unmounting).
4555          */
4556         btrfs_flush_workqueue(fs_info->endio_write_workers);
4557         /* Ordered extents for free space inodes. */
4558         btrfs_flush_workqueue(fs_info->endio_freespace_worker);
4559         btrfs_run_delayed_iputs(fs_info);
4560
4561         cancel_work_sync(&fs_info->async_reclaim_work);
4562         cancel_work_sync(&fs_info->async_data_reclaim_work);
4563         cancel_work_sync(&fs_info->preempt_reclaim_work);
4564
4565         /* Cancel or finish ongoing discard work */
4566         btrfs_discard_cleanup(fs_info);
4567
4568         if (!sb_rdonly(fs_info->sb)) {
4569                 /*
4570                  * The cleaner kthread is stopped, so do one final pass over
4571                  * unused block groups.
4572                  */
4573                 btrfs_delete_unused_bgs(fs_info);
4574
4575                 /*
4576                  * There might be existing delayed inode workers still running
4577                  * and holding an empty delayed inode item. We must wait for
4578                  * them to complete first because they can create a transaction.
4579                  * This happens when someone calls btrfs_balance_delayed_items()
4580                  * and then a transaction commit runs the same delayed nodes
4581                  * before any delayed worker has done something with the nodes.
4582                  * We must wait for any worker here and not at transaction
4583                  * commit time since that could cause a deadlock.
4584                  * This is a very rare case.
4585                  */
4586                 btrfs_flush_workqueue(fs_info->delayed_workers);
4587
4588                 ret = btrfs_commit_super(fs_info);
4589                 if (ret)
4590                         btrfs_err(fs_info, "commit super ret %d", ret);
4591         }
4592
4593         if (BTRFS_FS_ERROR(fs_info))
4594                 btrfs_error_commit_super(fs_info);
4595
4596         kthread_stop(fs_info->transaction_kthread);
4597         kthread_stop(fs_info->cleaner_kthread);
4598
4599         ASSERT(list_empty(&fs_info->delayed_iputs));
4600         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4601
4602         if (btrfs_check_quota_leak(fs_info)) {
4603                 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4604                 btrfs_err(fs_info, "qgroup reserved space leaked");
4605         }
4606
4607         btrfs_free_qgroup_config(fs_info);
4608         ASSERT(list_empty(&fs_info->delalloc_roots));
4609
4610         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4611                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4612                        percpu_counter_sum(&fs_info->delalloc_bytes));
4613         }
4614
4615         if (percpu_counter_sum(&fs_info->ordered_bytes))
4616                 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4617                            percpu_counter_sum(&fs_info->ordered_bytes));
4618
4619         btrfs_sysfs_remove_mounted(fs_info);
4620         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4621
4622         btrfs_put_block_group_cache(fs_info);
4623
4624         /*
4625          * we must make sure there is not any read request to
4626          * submit after we stopping all workers.
4627          */
4628         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4629         btrfs_stop_all_workers(fs_info);
4630
4631         /* We shouldn't have any transaction open at this point */
4632         warn_about_uncommitted_trans(fs_info);
4633
4634         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4635         free_root_pointers(fs_info, true);
4636         btrfs_free_fs_roots(fs_info);
4637
4638         /*
4639          * We must free the block groups after dropping the fs_roots as we could
4640          * have had an IO error and have left over tree log blocks that aren't
4641          * cleaned up until the fs roots are freed.  This makes the block group
4642          * accounting appear to be wrong because there's pending reserved bytes,
4643          * so make sure we do the block group cleanup afterwards.
4644          */
4645         btrfs_free_block_groups(fs_info);
4646
4647         iput(fs_info->btree_inode);
4648
4649 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4650         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4651                 btrfsic_unmount(fs_info->fs_devices);
4652 #endif
4653
4654         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4655         btrfs_close_devices(fs_info->fs_devices);
4656 }
4657
4658 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4659                           int atomic)
4660 {
4661         int ret;
4662         struct inode *btree_inode = buf->pages[0]->mapping->host;
4663
4664         ret = extent_buffer_uptodate(buf);
4665         if (!ret)
4666                 return ret;
4667
4668         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4669                                     parent_transid, atomic);
4670         if (ret == -EAGAIN)
4671                 return ret;
4672         return !ret;
4673 }
4674
4675 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4676 {
4677         struct btrfs_fs_info *fs_info = buf->fs_info;
4678         u64 transid = btrfs_header_generation(buf);
4679         int was_dirty;
4680
4681 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4682         /*
4683          * This is a fast path so only do this check if we have sanity tests
4684          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4685          * outside of the sanity tests.
4686          */
4687         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4688                 return;
4689 #endif
4690         btrfs_assert_tree_write_locked(buf);
4691         if (transid != fs_info->generation)
4692                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4693                         buf->start, transid, fs_info->generation);
4694         was_dirty = set_extent_buffer_dirty(buf);
4695         if (!was_dirty)
4696                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4697                                          buf->len,
4698                                          fs_info->dirty_metadata_batch);
4699 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4700         /*
4701          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4702          * but item data not updated.
4703          * So here we should only check item pointers, not item data.
4704          */
4705         if (btrfs_header_level(buf) == 0 &&
4706             btrfs_check_leaf_relaxed(buf)) {
4707                 btrfs_print_leaf(buf);
4708                 ASSERT(0);
4709         }
4710 #endif
4711 }
4712
4713 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4714                                         int flush_delayed)
4715 {
4716         /*
4717          * looks as though older kernels can get into trouble with
4718          * this code, they end up stuck in balance_dirty_pages forever
4719          */
4720         int ret;
4721
4722         if (current->flags & PF_MEMALLOC)
4723                 return;
4724
4725         if (flush_delayed)
4726                 btrfs_balance_delayed_items(fs_info);
4727
4728         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4729                                      BTRFS_DIRTY_METADATA_THRESH,
4730                                      fs_info->dirty_metadata_batch);
4731         if (ret > 0) {
4732                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4733         }
4734 }
4735
4736 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4737 {
4738         __btrfs_btree_balance_dirty(fs_info, 1);
4739 }
4740
4741 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4742 {
4743         __btrfs_btree_balance_dirty(fs_info, 0);
4744 }
4745
4746 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4747 {
4748         /* cleanup FS via transaction */
4749         btrfs_cleanup_transaction(fs_info);
4750
4751         mutex_lock(&fs_info->cleaner_mutex);
4752         btrfs_run_delayed_iputs(fs_info);
4753         mutex_unlock(&fs_info->cleaner_mutex);
4754
4755         down_write(&fs_info->cleanup_work_sem);
4756         up_write(&fs_info->cleanup_work_sem);
4757 }
4758
4759 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4760 {
4761         struct btrfs_root *gang[8];
4762         u64 root_objectid = 0;
4763         int ret;
4764
4765         spin_lock(&fs_info->fs_roots_radix_lock);
4766         while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4767                                              (void **)gang, root_objectid,
4768                                              ARRAY_SIZE(gang))) != 0) {
4769                 int i;
4770
4771                 for (i = 0; i < ret; i++)
4772                         gang[i] = btrfs_grab_root(gang[i]);
4773                 spin_unlock(&fs_info->fs_roots_radix_lock);
4774
4775                 for (i = 0; i < ret; i++) {
4776                         if (!gang[i])
4777                                 continue;
4778                         root_objectid = gang[i]->root_key.objectid;
4779                         btrfs_free_log(NULL, gang[i]);
4780                         btrfs_put_root(gang[i]);
4781                 }
4782                 root_objectid++;
4783                 spin_lock(&fs_info->fs_roots_radix_lock);
4784         }
4785         spin_unlock(&fs_info->fs_roots_radix_lock);
4786         btrfs_free_log_root_tree(NULL, fs_info);
4787 }
4788
4789 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4790 {
4791         struct btrfs_ordered_extent *ordered;
4792
4793         spin_lock(&root->ordered_extent_lock);
4794         /*
4795          * This will just short circuit the ordered completion stuff which will
4796          * make sure the ordered extent gets properly cleaned up.
4797          */
4798         list_for_each_entry(ordered, &root->ordered_extents,
4799                             root_extent_list)
4800                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4801         spin_unlock(&root->ordered_extent_lock);
4802 }
4803
4804 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4805 {
4806         struct btrfs_root *root;
4807         struct list_head splice;
4808
4809         INIT_LIST_HEAD(&splice);
4810
4811         spin_lock(&fs_info->ordered_root_lock);
4812         list_splice_init(&fs_info->ordered_roots, &splice);
4813         while (!list_empty(&splice)) {
4814                 root = list_first_entry(&splice, struct btrfs_root,
4815                                         ordered_root);
4816                 list_move_tail(&root->ordered_root,
4817                                &fs_info->ordered_roots);
4818
4819                 spin_unlock(&fs_info->ordered_root_lock);
4820                 btrfs_destroy_ordered_extents(root);
4821
4822                 cond_resched();
4823                 spin_lock(&fs_info->ordered_root_lock);
4824         }
4825         spin_unlock(&fs_info->ordered_root_lock);
4826
4827         /*
4828          * We need this here because if we've been flipped read-only we won't
4829          * get sync() from the umount, so we need to make sure any ordered
4830          * extents that haven't had their dirty pages IO start writeout yet
4831          * actually get run and error out properly.
4832          */
4833         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4834 }
4835
4836 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4837                                       struct btrfs_fs_info *fs_info)
4838 {
4839         struct rb_node *node;
4840         struct btrfs_delayed_ref_root *delayed_refs;
4841         struct btrfs_delayed_ref_node *ref;
4842         int ret = 0;
4843
4844         delayed_refs = &trans->delayed_refs;
4845
4846         spin_lock(&delayed_refs->lock);
4847         if (atomic_read(&delayed_refs->num_entries) == 0) {
4848                 spin_unlock(&delayed_refs->lock);
4849                 btrfs_debug(fs_info, "delayed_refs has NO entry");
4850                 return ret;
4851         }
4852
4853         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4854                 struct btrfs_delayed_ref_head *head;
4855                 struct rb_node *n;
4856                 bool pin_bytes = false;
4857
4858                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4859                                 href_node);
4860                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4861                         continue;
4862
4863                 spin_lock(&head->lock);
4864                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4865                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4866                                        ref_node);
4867                         ref->in_tree = 0;
4868                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4869                         RB_CLEAR_NODE(&ref->ref_node);
4870                         if (!list_empty(&ref->add_list))
4871                                 list_del(&ref->add_list);
4872                         atomic_dec(&delayed_refs->num_entries);
4873                         btrfs_put_delayed_ref(ref);
4874                 }
4875                 if (head->must_insert_reserved)
4876                         pin_bytes = true;
4877                 btrfs_free_delayed_extent_op(head->extent_op);
4878                 btrfs_delete_ref_head(delayed_refs, head);
4879                 spin_unlock(&head->lock);
4880                 spin_unlock(&delayed_refs->lock);
4881                 mutex_unlock(&head->mutex);
4882
4883                 if (pin_bytes) {
4884                         struct btrfs_block_group *cache;
4885
4886                         cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4887                         BUG_ON(!cache);
4888
4889                         spin_lock(&cache->space_info->lock);
4890                         spin_lock(&cache->lock);
4891                         cache->pinned += head->num_bytes;
4892                         btrfs_space_info_update_bytes_pinned(fs_info,
4893                                 cache->space_info, head->num_bytes);
4894                         cache->reserved -= head->num_bytes;
4895                         cache->space_info->bytes_reserved -= head->num_bytes;
4896                         spin_unlock(&cache->lock);
4897                         spin_unlock(&cache->space_info->lock);
4898
4899                         btrfs_put_block_group(cache);
4900
4901                         btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4902                                 head->bytenr + head->num_bytes - 1);
4903                 }
4904                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4905                 btrfs_put_delayed_ref_head(head);
4906                 cond_resched();
4907                 spin_lock(&delayed_refs->lock);
4908         }
4909         btrfs_qgroup_destroy_extent_records(trans);
4910
4911         spin_unlock(&delayed_refs->lock);
4912
4913         return ret;
4914 }
4915
4916 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4917 {
4918         struct btrfs_inode *btrfs_inode;
4919         struct list_head splice;
4920
4921         INIT_LIST_HEAD(&splice);
4922
4923         spin_lock(&root->delalloc_lock);
4924         list_splice_init(&root->delalloc_inodes, &splice);
4925
4926         while (!list_empty(&splice)) {
4927                 struct inode *inode = NULL;
4928                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4929                                                delalloc_inodes);
4930                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4931                 spin_unlock(&root->delalloc_lock);
4932
4933                 /*
4934                  * Make sure we get a live inode and that it'll not disappear
4935                  * meanwhile.
4936                  */
4937                 inode = igrab(&btrfs_inode->vfs_inode);
4938                 if (inode) {
4939                         unsigned int nofs_flag;
4940
4941                         nofs_flag = memalloc_nofs_save();
4942                         invalidate_inode_pages2(inode->i_mapping);
4943                         memalloc_nofs_restore(nofs_flag);
4944                         iput(inode);
4945                 }
4946                 spin_lock(&root->delalloc_lock);
4947         }
4948         spin_unlock(&root->delalloc_lock);
4949 }
4950
4951 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4952 {
4953         struct btrfs_root *root;
4954         struct list_head splice;
4955
4956         INIT_LIST_HEAD(&splice);
4957
4958         spin_lock(&fs_info->delalloc_root_lock);
4959         list_splice_init(&fs_info->delalloc_roots, &splice);
4960         while (!list_empty(&splice)) {
4961                 root = list_first_entry(&splice, struct btrfs_root,
4962                                          delalloc_root);
4963                 root = btrfs_grab_root(root);
4964                 BUG_ON(!root);
4965                 spin_unlock(&fs_info->delalloc_root_lock);
4966
4967                 btrfs_destroy_delalloc_inodes(root);
4968                 btrfs_put_root(root);
4969
4970                 spin_lock(&fs_info->delalloc_root_lock);
4971         }
4972         spin_unlock(&fs_info->delalloc_root_lock);
4973 }
4974
4975 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4976                                         struct extent_io_tree *dirty_pages,
4977                                         int mark)
4978 {
4979         int ret;
4980         struct extent_buffer *eb;
4981         u64 start = 0;
4982         u64 end;
4983
4984         while (1) {
4985                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4986                                             mark, NULL);
4987                 if (ret)
4988                         break;
4989
4990                 clear_extent_bits(dirty_pages, start, end, mark);
4991                 while (start <= end) {
4992                         eb = find_extent_buffer(fs_info, start);
4993                         start += fs_info->nodesize;
4994                         if (!eb)
4995                                 continue;
4996
4997                         btrfs_tree_lock(eb);
4998                         wait_on_extent_buffer_writeback(eb);
4999                         btrfs_clear_buffer_dirty(NULL, eb);
5000                         btrfs_tree_unlock(eb);
5001
5002                         free_extent_buffer_stale(eb);
5003                 }
5004         }
5005
5006         return ret;
5007 }
5008
5009 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
5010                                        struct extent_io_tree *unpin)
5011 {
5012         u64 start;
5013         u64 end;
5014         int ret;
5015
5016         while (1) {
5017                 struct extent_state *cached_state = NULL;
5018
5019                 /*
5020                  * The btrfs_finish_extent_commit() may get the same range as
5021                  * ours between find_first_extent_bit and clear_extent_dirty.
5022                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
5023                  * the same extent range.
5024                  */
5025                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5026                 ret = find_first_extent_bit(unpin, 0, &start, &end,
5027                                             EXTENT_DIRTY, &cached_state);
5028                 if (ret) {
5029                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5030                         break;
5031                 }
5032
5033                 clear_extent_dirty(unpin, start, end, &cached_state);
5034                 free_extent_state(cached_state);
5035                 btrfs_error_unpin_extent_range(fs_info, start, end);
5036                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5037                 cond_resched();
5038         }
5039
5040         return 0;
5041 }
5042
5043 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
5044 {
5045         struct inode *inode;
5046
5047         inode = cache->io_ctl.inode;
5048         if (inode) {
5049                 unsigned int nofs_flag;
5050
5051                 nofs_flag = memalloc_nofs_save();
5052                 invalidate_inode_pages2(inode->i_mapping);
5053                 memalloc_nofs_restore(nofs_flag);
5054
5055                 BTRFS_I(inode)->generation = 0;
5056                 cache->io_ctl.inode = NULL;
5057                 iput(inode);
5058         }
5059         ASSERT(cache->io_ctl.pages == NULL);
5060         btrfs_put_block_group(cache);
5061 }
5062
5063 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
5064                              struct btrfs_fs_info *fs_info)
5065 {
5066         struct btrfs_block_group *cache;
5067
5068         spin_lock(&cur_trans->dirty_bgs_lock);
5069         while (!list_empty(&cur_trans->dirty_bgs)) {
5070                 cache = list_first_entry(&cur_trans->dirty_bgs,
5071                                          struct btrfs_block_group,
5072                                          dirty_list);
5073
5074                 if (!list_empty(&cache->io_list)) {
5075                         spin_unlock(&cur_trans->dirty_bgs_lock);
5076                         list_del_init(&cache->io_list);
5077                         btrfs_cleanup_bg_io(cache);
5078                         spin_lock(&cur_trans->dirty_bgs_lock);
5079                 }
5080
5081                 list_del_init(&cache->dirty_list);
5082                 spin_lock(&cache->lock);
5083                 cache->disk_cache_state = BTRFS_DC_ERROR;
5084                 spin_unlock(&cache->lock);
5085
5086                 spin_unlock(&cur_trans->dirty_bgs_lock);
5087                 btrfs_put_block_group(cache);
5088                 btrfs_delayed_refs_rsv_release(fs_info, 1);
5089                 spin_lock(&cur_trans->dirty_bgs_lock);
5090         }
5091         spin_unlock(&cur_trans->dirty_bgs_lock);
5092
5093         /*
5094          * Refer to the definition of io_bgs member for details why it's safe
5095          * to use it without any locking
5096          */
5097         while (!list_empty(&cur_trans->io_bgs)) {
5098                 cache = list_first_entry(&cur_trans->io_bgs,
5099                                          struct btrfs_block_group,
5100                                          io_list);
5101
5102                 list_del_init(&cache->io_list);
5103                 spin_lock(&cache->lock);
5104                 cache->disk_cache_state = BTRFS_DC_ERROR;
5105                 spin_unlock(&cache->lock);
5106                 btrfs_cleanup_bg_io(cache);
5107         }
5108 }
5109
5110 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
5111                                    struct btrfs_fs_info *fs_info)
5112 {
5113         struct btrfs_device *dev, *tmp;
5114
5115         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
5116         ASSERT(list_empty(&cur_trans->dirty_bgs));
5117         ASSERT(list_empty(&cur_trans->io_bgs));
5118
5119         list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
5120                                  post_commit_list) {
5121                 list_del_init(&dev->post_commit_list);
5122         }
5123
5124         btrfs_destroy_delayed_refs(cur_trans, fs_info);
5125
5126         cur_trans->state = TRANS_STATE_COMMIT_START;
5127         wake_up(&fs_info->transaction_blocked_wait);
5128
5129         cur_trans->state = TRANS_STATE_UNBLOCKED;
5130         wake_up(&fs_info->transaction_wait);
5131
5132         btrfs_destroy_delayed_inodes(fs_info);
5133
5134         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
5135                                      EXTENT_DIRTY);
5136         btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
5137
5138         btrfs_free_redirty_list(cur_trans);
5139
5140         cur_trans->state =TRANS_STATE_COMPLETED;
5141         wake_up(&cur_trans->commit_wait);
5142 }
5143
5144 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
5145 {
5146         struct btrfs_transaction *t;
5147
5148         mutex_lock(&fs_info->transaction_kthread_mutex);
5149
5150         spin_lock(&fs_info->trans_lock);
5151         while (!list_empty(&fs_info->trans_list)) {
5152                 t = list_first_entry(&fs_info->trans_list,
5153                                      struct btrfs_transaction, list);
5154                 if (t->state >= TRANS_STATE_COMMIT_START) {
5155                         refcount_inc(&t->use_count);
5156                         spin_unlock(&fs_info->trans_lock);
5157                         btrfs_wait_for_commit(fs_info, t->transid);
5158                         btrfs_put_transaction(t);
5159                         spin_lock(&fs_info->trans_lock);
5160                         continue;
5161                 }
5162                 if (t == fs_info->running_transaction) {
5163                         t->state = TRANS_STATE_COMMIT_DOING;
5164                         spin_unlock(&fs_info->trans_lock);
5165                         /*
5166                          * We wait for 0 num_writers since we don't hold a trans
5167                          * handle open currently for this transaction.
5168                          */
5169                         wait_event(t->writer_wait,
5170                                    atomic_read(&t->num_writers) == 0);
5171                 } else {
5172                         spin_unlock(&fs_info->trans_lock);
5173                 }
5174                 btrfs_cleanup_one_transaction(t, fs_info);
5175
5176                 spin_lock(&fs_info->trans_lock);
5177                 if (t == fs_info->running_transaction)
5178                         fs_info->running_transaction = NULL;
5179                 list_del_init(&t->list);
5180                 spin_unlock(&fs_info->trans_lock);
5181
5182                 btrfs_put_transaction(t);
5183                 trace_btrfs_transaction_commit(fs_info);
5184                 spin_lock(&fs_info->trans_lock);
5185         }
5186         spin_unlock(&fs_info->trans_lock);
5187         btrfs_destroy_all_ordered_extents(fs_info);
5188         btrfs_destroy_delayed_inodes(fs_info);
5189         btrfs_assert_delayed_root_empty(fs_info);
5190         btrfs_destroy_all_delalloc_inodes(fs_info);
5191         btrfs_drop_all_logs(fs_info);
5192         mutex_unlock(&fs_info->transaction_kthread_mutex);
5193
5194         return 0;
5195 }
5196
5197 int btrfs_init_root_free_objectid(struct btrfs_root *root)
5198 {
5199         struct btrfs_path *path;
5200         int ret;
5201         struct extent_buffer *l;
5202         struct btrfs_key search_key;
5203         struct btrfs_key found_key;
5204         int slot;
5205
5206         path = btrfs_alloc_path();
5207         if (!path)
5208                 return -ENOMEM;
5209
5210         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
5211         search_key.type = -1;
5212         search_key.offset = (u64)-1;
5213         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5214         if (ret < 0)
5215                 goto error;
5216         BUG_ON(ret == 0); /* Corruption */
5217         if (path->slots[0] > 0) {
5218                 slot = path->slots[0] - 1;
5219                 l = path->nodes[0];
5220                 btrfs_item_key_to_cpu(l, &found_key, slot);
5221                 root->free_objectid = max_t(u64, found_key.objectid + 1,
5222                                             BTRFS_FIRST_FREE_OBJECTID);
5223         } else {
5224                 root->free_objectid = BTRFS_FIRST_FREE_OBJECTID;
5225         }
5226         ret = 0;
5227 error:
5228         btrfs_free_path(path);
5229         return ret;
5230 }
5231
5232 int btrfs_get_free_objectid(struct btrfs_root *root, u64 *objectid)
5233 {
5234         int ret;
5235         mutex_lock(&root->objectid_mutex);
5236
5237         if (unlikely(root->free_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
5238                 btrfs_warn(root->fs_info,
5239                            "the objectid of root %llu reaches its highest value",
5240                            root->root_key.objectid);
5241                 ret = -ENOSPC;
5242                 goto out;
5243         }
5244
5245         *objectid = root->free_objectid++;
5246         ret = 0;
5247 out:
5248         mutex_unlock(&root->objectid_mutex);
5249         return ret;
5250 }