fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/slab.h>
  29 #include <linux/migrate.h>
  30 #include <linux/ratelimit.h>
  31 #include <linux/uuid.h>
  32 #include <linux/semaphore.h>
  33 #include <asm/unaligned.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "hash.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "free-space-tree.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52 #include "compression.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
  68 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  69 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  70                                       struct btrfs_fs_info *fs_info);
  71 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  72 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  73                                         struct extent_io_tree *dirty_pages,
  74                                         int mark);
  75 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  76                                        struct extent_io_tree *pinned_extents);
  77 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  78 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  79
  80 /*
  81  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  82  * is complete.  This is used during reads to verify checksums, and it is used
  83  * by writes to insert metadata for new file extents after IO is complete.
  84  */
  85 struct btrfs_end_io_wq {
  86         struct bio *bio;
  87         bio_end_io_t *end_io;
  88         void *private;
  89         struct btrfs_fs_info *info;
  90         blk_status_t status;
  91         enum btrfs_wq_endio_type metadata;
  92         struct btrfs_work work;
  93 };
  94
  95 static struct kmem_cache *btrfs_end_io_wq_cache;
  96
  97 int __init btrfs_end_io_wq_init(void)
  98 {
  99         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 100                                         sizeof(struct btrfs_end_io_wq),
 101                                         0,
 102                                         SLAB_MEM_SPREAD,
 103                                         NULL);
 104         if (!btrfs_end_io_wq_cache)
 105                 return -ENOMEM;
 106         return 0;
 107 }
 108
 109 void btrfs_end_io_wq_exit(void)
 110 {
 111         kmem_cache_destroy(btrfs_end_io_wq_cache);
 112 }
 113
 114 /*
 115  * async submit bios are used to offload expensive checksumming
 116  * onto the worker threads.  They checksum file and metadata bios
 117  * just before they are sent down the IO stack.
 118  */
 119 struct async_submit_bio {
 120         void *private_data;
 121         struct btrfs_fs_info *fs_info;
 122         struct bio *bio;
 123         extent_submit_bio_hook_t *submit_bio_start;
 124         extent_submit_bio_hook_t *submit_bio_done;
 125         int mirror_num;
 126         unsigned long bio_flags;
 127         /*
 128          * bio_offset is optional, can be used if the pages in the bio
 129          * can't tell us where in the file the bio should go
 130          */
 131         u64 bio_offset;
 132         struct btrfs_work work;
 133         blk_status_t status;
 134 };
 135
 136 /*
 137  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 138  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 139  * the level the eb occupies in the tree.
 140  *
 141  * Different roots are used for different purposes and may nest inside each
 142  * other and they require separate keysets.  As lockdep keys should be
 143  * static, assign keysets according to the purpose of the root as indicated
 144  * by btrfs_root->objectid.  This ensures that all special purpose roots
 145  * have separate keysets.
 146  *
 147  * Lock-nesting across peer nodes is always done with the immediate parent
 148  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 149  * subclass to avoid triggering lockdep warning in such cases.
 150  *
 151  * The key is set by the readpage_end_io_hook after the buffer has passed
 152  * csum validation but before the pages are unlocked.  It is also set by
 153  * btrfs_init_new_buffer on freshly allocated blocks.
 154  *
 155  * We also add a check to make sure the highest level of the tree is the
 156  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 157  * needs update as well.
 158  */
 159 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 160 # if BTRFS_MAX_LEVEL != 8
 161 #  error
 162 # endif
 163
 164 static struct btrfs_lockdep_keyset {
 165         u64                     id;             /* root objectid */
 166         const char              *name_stem;     /* lock name stem */
 167         char                    names[BTRFS_MAX_LEVEL + 1][20];
 168         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 169 } btrfs_lockdep_keysets[] = {
 170         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 171         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 172         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 173         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 174         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 175         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 176         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 177         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 178         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 179         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 180         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 181         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 182         { .id = 0,                              .name_stem = "tree"     },
 183 };
 184
 185 void __init btrfs_init_lockdep(void)
 186 {
 187         int i, j;
 188
 189         /* initialize lockdep class names */
 190         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 191                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 192
 193                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 194                         snprintf(ks->names[j], sizeof(ks->names[j]),
 195                                  "btrfs-%s-%02d", ks->name_stem, j);
 196         }
 197 }
 198
 199 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 200                                     int level)
 201 {
 202         struct btrfs_lockdep_keyset *ks;
 203
 204         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 205
 206         /* find the matching keyset, id 0 is the default entry */
 207         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 208                 if (ks->id == objectid)
 209                         break;
 210
 211         lockdep_set_class_and_name(&eb->lock,
 212                                    &ks->keys[level], ks->names[level]);
 213 }
 214
 215 #endif
 216
 217 /*
 218  * extents on the btree inode are pretty simple, there's one extent
 219  * that covers the entire device
 220  */
 221 static struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 222                 struct page *page, size_t pg_offset, u64 start, u64 len,
 223                 int create)
 224 {
 225         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 226         struct extent_map_tree *em_tree = &inode->extent_tree;
 227         struct extent_map *em;
 228         int ret;
 229
 230         read_lock(&em_tree->lock);
 231         em = lookup_extent_mapping(em_tree, start, len);
 232         if (em) {
 233                 em->bdev = fs_info->fs_devices->latest_bdev;
 234                 read_unlock(&em_tree->lock);
 235                 goto out;
 236         }
 237         read_unlock(&em_tree->lock);
 238
 239         em = alloc_extent_map();
 240         if (!em) {
 241                 em = ERR_PTR(-ENOMEM);
 242                 goto out;
 243         }
 244         em->start = 0;
 245         em->len = (u64)-1;
 246         em->block_len = (u64)-1;
 247         em->block_start = 0;
 248         em->bdev = fs_info->fs_devices->latest_bdev;
 249
 250         write_lock(&em_tree->lock);
 251         ret = add_extent_mapping(em_tree, em, 0);
 252         if (ret == -EEXIST) {
 253                 free_extent_map(em);
 254                 em = lookup_extent_mapping(em_tree, start, len);
 255                 if (!em)
 256                         em = ERR_PTR(-EIO);
 257         } else if (ret) {
 258                 free_extent_map(em);
 259                 em = ERR_PTR(ret);
 260         }
 261         write_unlock(&em_tree->lock);
 262
 263 out:
 264         return em;
 265 }
 266
 267 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 268 {
 269         return btrfs_crc32c(seed, data, len);
 270 }
 271
 272 void btrfs_csum_final(u32 crc, u8 *result)
 273 {
 274         put_unaligned_le32(~crc, result);
 275 }
 276
 277 /*
 278  * compute the csum for a btree block, and either verify it or write it
 279  * into the csum field of the block.
 280  */
 281 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 282                            struct extent_buffer *buf,
 283                            int verify)
 284 {
 285         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 286         char *result = NULL;
 287         unsigned long len;
 288         unsigned long cur_len;
 289         unsigned long offset = BTRFS_CSUM_SIZE;
 290         char *kaddr;
 291         unsigned long map_start;
 292         unsigned long map_len;
 293         int err;
 294         u32 crc = ~(u32)0;
 295         unsigned long inline_result;
 296
 297         len = buf->len - offset;
 298         while (len > 0) {
 299                 err = map_private_extent_buffer(buf, offset, 32,
 300                                         &kaddr, &map_start, &map_len);
 301                 if (err)
 302                         return err;
 303                 cur_len = min(len, map_len - (offset - map_start));
 304                 crc = btrfs_csum_data(kaddr + offset - map_start,
 305                                       crc, cur_len);
 306                 len -= cur_len;
 307                 offset += cur_len;
 308         }
 309         if (csum_size > sizeof(inline_result)) {
 310                 result = kzalloc(csum_size, GFP_NOFS);
 311                 if (!result)
 312                         return -ENOMEM;
 313         } else {
 314                 result = (char *)&inline_result;
 315         }
 316
 317         btrfs_csum_final(crc, result);
 318
 319         if (verify) {
 320                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 321                         u32 val;
 322                         u32 found = 0;
 323                         memcpy(&found, result, csum_size);
 324
 325                         read_extent_buffer(buf, &val, 0, csum_size);
 326                         btrfs_warn_rl(fs_info,
 327                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 328                                 fs_info->sb->s_id, buf->start,
 329                                 val, found, btrfs_header_level(buf));
 330                         if (result != (char *)&inline_result)
 331                                 kfree(result);
 332                         return -EUCLEAN;
 333                 }
 334         } else {
 335                 write_extent_buffer(buf, result, 0, csum_size);
 336         }
 337         if (result != (char *)&inline_result)
 338                 kfree(result);
 339         return 0;
 340 }
 341
 342 /*
 343  * we can't consider a given block up to date unless the transid of the
 344  * block matches the transid in the parent node's pointer.  This is how we
 345  * detect blocks that either didn't get written at all or got written
 346  * in the wrong place.
 347  */
 348 static int verify_parent_transid(struct extent_io_tree *io_tree,
 349                                  struct extent_buffer *eb, u64 parent_transid,
 350                                  int atomic)
 351 {
 352         struct extent_state *cached_state = NULL;
 353         int ret;
 354         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 355
 356         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 357                 return 0;
 358
 359         if (atomic)
 360                 return -EAGAIN;
 361
 362         if (need_lock) {
 363                 btrfs_tree_read_lock(eb);
 364                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 365         }
 366
 367         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 368                          &cached_state);
 369         if (extent_buffer_uptodate(eb) &&
 370             btrfs_header_generation(eb) == parent_transid) {
 371                 ret = 0;
 372                 goto out;
 373         }
 374         btrfs_err_rl(eb->fs_info,
 375                 "parent transid verify failed on %llu wanted %llu found %llu",
 376                         eb->start,
 377                         parent_transid, btrfs_header_generation(eb));
 378         ret = 1;
 379
 380         /*
 381          * Things reading via commit roots that don't have normal protection,
 382          * like send, can have a really old block in cache that may point at a
 383          * block that has been freed and re-allocated.  So don't clear uptodate
 384          * if we find an eb that is under IO (dirty/writeback) because we could
 385          * end up reading in the stale data and then writing it back out and
 386          * making everybody very sad.
 387          */
 388         if (!extent_buffer_under_io(eb))
 389                 clear_extent_buffer_uptodate(eb);
 390 out:
 391         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 392                              &cached_state, GFP_NOFS);
 393         if (need_lock)
 394                 btrfs_tree_read_unlock_blocking(eb);
 395         return ret;
 396 }
 397
 398 /*
 399  * Return 0 if the superblock checksum type matches the checksum value of that
 400  * algorithm. Pass the raw disk superblock data.
 401  */
 402 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 403                                   char *raw_disk_sb)
 404 {
 405         struct btrfs_super_block *disk_sb =
 406                 (struct btrfs_super_block *)raw_disk_sb;
 407         u16 csum_type = btrfs_super_csum_type(disk_sb);
 408         int ret = 0;
 409
 410         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 411                 u32 crc = ~(u32)0;
 412                 const int csum_size = sizeof(crc);
 413                 char result[csum_size];
 414
 415                 /*
 416                  * The super_block structure does not span the whole
 417                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 418                  * is filled with zeros and is included in the checksum.
 419                  */
 420                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 421                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 422                 btrfs_csum_final(crc, result);
 423
 424                 if (memcmp(raw_disk_sb, result, csum_size))
 425                         ret = 1;
 426         }
 427
 428         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 429                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 430                                 csum_type);
 431                 ret = 1;
 432         }
 433
 434         return ret;
 435 }
 436
 437 /*
 438  * helper to read a given tree block, doing retries as required when
 439  * the checksums don't match and we have alternate mirrors to try.
 440  */
 441 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 442                                           struct extent_buffer *eb,
 443                                           u64 parent_transid)
 444 {
 445         struct extent_io_tree *io_tree;
 446         int failed = 0;
 447         int ret;
 448         int num_copies = 0;
 449         int mirror_num = 0;
 450         int failed_mirror = 0;
 451
 452         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 453         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 454         while (1) {
 455                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 456                                                btree_get_extent, mirror_num);
 457                 if (!ret) {
 458                         if (!verify_parent_transid(io_tree, eb,
 459                                                    parent_transid, 0))
 460                                 break;
 461                         else
 462                                 ret = -EIO;
 463                 }
 464
 465                 /*
 466                  * This buffer's crc is fine, but its contents are corrupted, so
 467                  * there is no reason to read the other copies, they won't be
 468                  * any less wrong.
 469                  */
 470                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 471                         break;
 472
 473                 num_copies = btrfs_num_copies(fs_info,
 474                                               eb->start, eb->len);
 475                 if (num_copies == 1)
 476                         break;
 477
 478                 if (!failed_mirror) {
 479                         failed = 1;
 480                         failed_mirror = eb->read_mirror;
 481                 }
 482
 483                 mirror_num++;
 484                 if (mirror_num == failed_mirror)
 485                         mirror_num++;
 486
 487                 if (mirror_num > num_copies)
 488                         break;
 489         }
 490
 491         if (failed && !ret && failed_mirror)
 492                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 493
 494         return ret;
 495 }
 496
 497 /*
 498  * checksum a dirty tree block before IO.  This has extra checks to make sure
 499  * we only fill in the checksum field in the first page of a multi-page block
 500  */
 501
 502 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 503 {
 504         u64 start = page_offset(page);
 505         u64 found_start;
 506         struct extent_buffer *eb;
 507
 508         eb = (struct extent_buffer *)page->private;
 509         if (page != eb->pages[0])
 510                 return 0;
 511
 512         found_start = btrfs_header_bytenr(eb);
 513         /*
 514          * Please do not consolidate these warnings into a single if.
 515          * It is useful to know what went wrong.
 516          */
 517         if (WARN_ON(found_start != start))
 518                 return -EUCLEAN;
 519         if (WARN_ON(!PageUptodate(page)))
 520                 return -EUCLEAN;
 521
 522         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 523                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 524
 525         return csum_tree_block(fs_info, eb, 0);
 526 }
 527
 528 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 529                                  struct extent_buffer *eb)
 530 {
 531         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 532         u8 fsid[BTRFS_FSID_SIZE];
 533         int ret = 1;
 534
 535         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 536         while (fs_devices) {
 537                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 538                         ret = 0;
 539                         break;
 540                 }
 541                 fs_devices = fs_devices->seed;
 542         }
 543         return ret;
 544 }
 545
 546 #define CORRUPT(reason, eb, root, slot)                                 \
 547         btrfs_crit(root->fs_info,                                       \
 548                    "corrupt %s, %s: block=%llu, root=%llu, slot=%d",    \
 549                    btrfs_header_level(eb) == 0 ? "leaf" : "node",       \
 550                    reason, btrfs_header_bytenr(eb), root->objectid, slot)
 551
 552 static noinline int check_leaf(struct btrfs_root *root,
 553                                struct extent_buffer *leaf)
 554 {
 555         struct btrfs_fs_info *fs_info = root->fs_info;
 556         struct btrfs_key key;
 557         struct btrfs_key leaf_key;
 558         u32 nritems = btrfs_header_nritems(leaf);
 559         int slot;
 560
 561         /*
 562          * Extent buffers from a relocation tree have a owner field that
 563          * corresponds to the subvolume tree they are based on. So just from an
 564          * extent buffer alone we can not find out what is the id of the
 565          * corresponding subvolume tree, so we can not figure out if the extent
 566          * buffer corresponds to the root of the relocation tree or not. So skip
 567          * this check for relocation trees.
 568          */
 569         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
 570                 struct btrfs_root *check_root;
 571
 572                 key.objectid = btrfs_header_owner(leaf);
 573                 key.type = BTRFS_ROOT_ITEM_KEY;
 574                 key.offset = (u64)-1;
 575
 576                 check_root = btrfs_get_fs_root(fs_info, &key, false);
 577                 /*
 578                  * The only reason we also check NULL here is that during
 579                  * open_ctree() some roots has not yet been set up.
 580                  */
 581                 if (!IS_ERR_OR_NULL(check_root)) {
 582                         struct extent_buffer *eb;
 583
 584                         eb = btrfs_root_node(check_root);
 585                         /* if leaf is the root, then it's fine */
 586                         if (leaf != eb) {
 587                                 CORRUPT("non-root leaf's nritems is 0",
 588                                         leaf, check_root, 0);
 589                                 free_extent_buffer(eb);
 590                                 return -EIO;
 591                         }
 592                         free_extent_buffer(eb);
 593                 }
 594                 return 0;
 595         }
 596
 597         if (nritems == 0)
 598                 return 0;
 599
 600         /* Check the 0 item */
 601         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 602             BTRFS_LEAF_DATA_SIZE(fs_info)) {
 603                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 604                 return -EIO;
 605         }
 606
 607         /*
 608          * Check to make sure each items keys are in the correct order and their
 609          * offsets make sense.  We only have to loop through nritems-1 because
 610          * we check the current slot against the next slot, which verifies the
 611          * next slot's offset+size makes sense and that the current's slot
 612          * offset is correct.
 613          */
 614         for (slot = 0; slot < nritems - 1; slot++) {
 615                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 616                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 617
 618                 /* Make sure the keys are in the right order */
 619                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 620                         CORRUPT("bad key order", leaf, root, slot);
 621                         return -EIO;
 622                 }
 623
 624                 /*
 625                  * Make sure the offset and ends are right, remember that the
 626                  * item data starts at the end of the leaf and grows towards the
 627                  * front.
 628                  */
 629                 if (btrfs_item_offset_nr(leaf, slot) !=
 630                         btrfs_item_end_nr(leaf, slot + 1)) {
 631                         CORRUPT("slot offset bad", leaf, root, slot);
 632                         return -EIO;
 633                 }
 634
 635                 /*
 636                  * Check to make sure that we don't point outside of the leaf,
 637                  * just in case all the items are consistent to each other, but
 638                  * all point outside of the leaf.
 639                  */
 640                 if (btrfs_item_end_nr(leaf, slot) >
 641                     BTRFS_LEAF_DATA_SIZE(fs_info)) {
 642                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 643                         return -EIO;
 644                 }
 645         }
 646
 647         return 0;
 648 }
 649
 650 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 651 {
 652         unsigned long nr = btrfs_header_nritems(node);
 653         struct btrfs_key key, next_key;
 654         int slot;
 655         u64 bytenr;
 656         int ret = 0;
 657
 658         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
 659                 btrfs_crit(root->fs_info,
 660                            "corrupt node: block %llu root %llu nritems %lu",
 661                            node->start, root->objectid, nr);
 662                 return -EIO;
 663         }
 664
 665         for (slot = 0; slot < nr - 1; slot++) {
 666                 bytenr = btrfs_node_blockptr(node, slot);
 667                 btrfs_node_key_to_cpu(node, &key, slot);
 668                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
 669
 670                 if (!bytenr) {
 671                         CORRUPT("invalid item slot", node, root, slot);
 672                         ret = -EIO;
 673                         goto out;
 674                 }
 675
 676                 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
 677                         CORRUPT("bad key order", node, root, slot);
 678                         ret = -EIO;
 679                         goto out;
 680                 }
 681         }
 682 out:
 683         return ret;
 684 }
 685
 686 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 687                                       u64 phy_offset, struct page *page,
 688                                       u64 start, u64 end, int mirror)
 689 {
 690         u64 found_start;
 691         int found_level;
 692         struct extent_buffer *eb;
 693         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 694         struct btrfs_fs_info *fs_info = root->fs_info;
 695         int ret = 0;
 696         int reads_done;
 697
 698         if (!page->private)
 699                 goto out;
 700
 701         eb = (struct extent_buffer *)page->private;
 702
 703         /* the pending IO might have been the only thing that kept this buffer
 704          * in memory.  Make sure we have a ref for all this other checks
 705          */
 706         extent_buffer_get(eb);
 707
 708         reads_done = atomic_dec_and_test(&eb->io_pages);
 709         if (!reads_done)
 710                 goto err;
 711
 712         eb->read_mirror = mirror;
 713         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 714                 ret = -EIO;
 715                 goto err;
 716         }
 717
 718         found_start = btrfs_header_bytenr(eb);
 719         if (found_start != eb->start) {
 720                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 721                              found_start, eb->start);
 722                 ret = -EIO;
 723                 goto err;
 724         }
 725         if (check_tree_block_fsid(fs_info, eb)) {
 726                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 727                              eb->start);
 728                 ret = -EIO;
 729                 goto err;
 730         }
 731         found_level = btrfs_header_level(eb);
 732         if (found_level >= BTRFS_MAX_LEVEL) {
 733                 btrfs_err(fs_info, "bad tree block level %d",
 734                           (int)btrfs_header_level(eb));
 735                 ret = -EIO;
 736                 goto err;
 737         }
 738
 739         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 740                                        eb, found_level);
 741
 742         ret = csum_tree_block(fs_info, eb, 1);
 743         if (ret)
 744                 goto err;
 745
 746         /*
 747          * If this is a leaf block and it is corrupt, set the corrupt bit so
 748          * that we don't try and read the other copies of this block, just
 749          * return -EIO.
 750          */
 751         if (found_level == 0 && check_leaf(root, eb)) {
 752                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 753                 ret = -EIO;
 754         }
 755
 756         if (found_level > 0 && check_node(root, eb))
 757                 ret = -EIO;
 758
 759         if (!ret)
 760                 set_extent_buffer_uptodate(eb);
 761 err:
 762         if (reads_done &&
 763             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 764                 btree_readahead_hook(eb, ret);
 765
 766         if (ret) {
 767                 /*
 768                  * our io error hook is going to dec the io pages
 769                  * again, we have to make sure it has something
 770                  * to decrement
 771                  */
 772                 atomic_inc(&eb->io_pages);
 773                 clear_extent_buffer_uptodate(eb);
 774         }
 775         free_extent_buffer(eb);
 776 out:
 777         return ret;
 778 }
 779
 780 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 781 {
 782         struct extent_buffer *eb;
 783
 784         eb = (struct extent_buffer *)page->private;
 785         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 786         eb->read_mirror = failed_mirror;
 787         atomic_dec(&eb->io_pages);
 788         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 789                 btree_readahead_hook(eb, -EIO);
 790         return -EIO;    /* we fixed nothing */
 791 }
 792
 793 static void end_workqueue_bio(struct bio *bio)
 794 {
 795         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 796         struct btrfs_fs_info *fs_info;
 797         struct btrfs_workqueue *wq;
 798         btrfs_work_func_t func;
 799
 800         fs_info = end_io_wq->info;
 801         end_io_wq->status = bio->bi_status;
 802
 803         if (bio_op(bio) == REQ_OP_WRITE) {
 804                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 805                         wq = fs_info->endio_meta_write_workers;
 806                         func = btrfs_endio_meta_write_helper;
 807                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 808                         wq = fs_info->endio_freespace_worker;
 809                         func = btrfs_freespace_write_helper;
 810                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 811                         wq = fs_info->endio_raid56_workers;
 812                         func = btrfs_endio_raid56_helper;
 813                 } else {
 814                         wq = fs_info->endio_write_workers;
 815                         func = btrfs_endio_write_helper;
 816                 }
 817         } else {
 818                 if (unlikely(end_io_wq->metadata ==
 819                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 820                         wq = fs_info->endio_repair_workers;
 821                         func = btrfs_endio_repair_helper;
 822                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 823                         wq = fs_info->endio_raid56_workers;
 824                         func = btrfs_endio_raid56_helper;
 825                 } else if (end_io_wq->metadata) {
 826                         wq = fs_info->endio_meta_workers;
 827                         func = btrfs_endio_meta_helper;
 828                 } else {
 829                         wq = fs_info->endio_workers;
 830                         func = btrfs_endio_helper;
 831                 }
 832         }
 833
 834         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 835         btrfs_queue_work(wq, &end_io_wq->work);
 836 }
 837
 838 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 839                         enum btrfs_wq_endio_type metadata)
 840 {
 841         struct btrfs_end_io_wq *end_io_wq;
 842
 843         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 844         if (!end_io_wq)
 845                 return BLK_STS_RESOURCE;
 846
 847         end_io_wq->private = bio->bi_private;
 848         end_io_wq->end_io = bio->bi_end_io;
 849         end_io_wq->info = info;
 850         end_io_wq->status = 0;
 851         end_io_wq->bio = bio;
 852         end_io_wq->metadata = metadata;
 853
 854         bio->bi_private = end_io_wq;
 855         bio->bi_end_io = end_workqueue_bio;
 856         return 0;
 857 }
 858
 859 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 860 {
 861         unsigned long limit = min_t(unsigned long,
 862                                     info->thread_pool_size,
 863                                     info->fs_devices->open_devices);
 864         return 256 * limit;
 865 }
 866
 867 static void run_one_async_start(struct btrfs_work *work)
 868 {
 869         struct async_submit_bio *async;
 870         blk_status_t ret;
 871
 872         async = container_of(work, struct  async_submit_bio, work);
 873         ret = async->submit_bio_start(async->private_data, async->bio,
 874                                       async->mirror_num, async->bio_flags,
 875                                       async->bio_offset);
 876         if (ret)
 877                 async->status = ret;
 878 }
 879
 880 static void run_one_async_done(struct btrfs_work *work)
 881 {
 882         struct btrfs_fs_info *fs_info;
 883         struct async_submit_bio *async;
 884         int limit;
 885
 886         async = container_of(work, struct  async_submit_bio, work);
 887         fs_info = async->fs_info;
 888
 889         limit = btrfs_async_submit_limit(fs_info);
 890         limit = limit * 2 / 3;
 891
 892         /*
 893          * atomic_dec_return implies a barrier for waitqueue_active
 894          */
 895         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 896             waitqueue_active(&fs_info->async_submit_wait))
 897                 wake_up(&fs_info->async_submit_wait);
 898
 899         /* If an error occurred we just want to clean up the bio and move on */
 900         if (async->status) {
 901                 async->bio->bi_status = async->status;
 902                 bio_endio(async->bio);
 903                 return;
 904         }
 905
 906         async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
 907                                async->bio_flags, async->bio_offset);
 908 }
 909
 910 static void run_one_async_free(struct btrfs_work *work)
 911 {
 912         struct async_submit_bio *async;
 913
 914         async = container_of(work, struct  async_submit_bio, work);
 915         kfree(async);
 916 }
 917
 918 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 919                                  int mirror_num, unsigned long bio_flags,
 920                                  u64 bio_offset, void *private_data,
 921                                  extent_submit_bio_hook_t *submit_bio_start,
 922                                  extent_submit_bio_hook_t *submit_bio_done)
 923 {
 924         struct async_submit_bio *async;
 925
 926         async = kmalloc(sizeof(*async), GFP_NOFS);
 927         if (!async)
 928                 return BLK_STS_RESOURCE;
 929
 930         async->private_data = private_data;
 931         async->fs_info = fs_info;
 932         async->bio = bio;
 933         async->mirror_num = mirror_num;
 934         async->submit_bio_start = submit_bio_start;
 935         async->submit_bio_done = submit_bio_done;
 936
 937         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 938                         run_one_async_done, run_one_async_free);
 939
 940         async->bio_flags = bio_flags;
 941         async->bio_offset = bio_offset;
 942
 943         async->status = 0;
 944
 945         atomic_inc(&fs_info->nr_async_submits);
 946
 947         if (op_is_sync(bio->bi_opf))
 948                 btrfs_set_work_high_priority(&async->work);
 949
 950         btrfs_queue_work(fs_info->workers, &async->work);
 951
 952         while (atomic_read(&fs_info->async_submit_draining) &&
 953               atomic_read(&fs_info->nr_async_submits)) {
 954                 wait_event(fs_info->async_submit_wait,
 955                            (atomic_read(&fs_info->nr_async_submits) == 0));
 956         }
 957
 958         return 0;
 959 }
 960
 961 static blk_status_t btree_csum_one_bio(struct bio *bio)
 962 {
 963         struct bio_vec *bvec;
 964         struct btrfs_root *root;
 965         int i, ret = 0;
 966
 967         ASSERT(!bio_flagged(bio, BIO_CLONED));
 968         bio_for_each_segment_all(bvec, bio, i) {
 969                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 970                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 971                 if (ret)
 972                         break;
 973         }
 974
 975         return errno_to_blk_status(ret);
 976 }
 977
 978 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
 979                                              int mirror_num, unsigned long bio_flags,
 980                                              u64 bio_offset)
 981 {
 982         /*
 983          * when we're called for a write, we're already in the async
 984          * submission context.  Just jump into btrfs_map_bio
 985          */
 986         return btree_csum_one_bio(bio);
 987 }
 988
 989 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
 990                                             int mirror_num, unsigned long bio_flags,
 991                                             u64 bio_offset)
 992 {
 993         struct inode *inode = private_data;
 994         blk_status_t ret;
 995
 996         /*
 997          * when we're called for a write, we're already in the async
 998          * submission context.  Just jump into btrfs_map_bio
 999          */
1000         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1001         if (ret) {
1002                 bio->bi_status = ret;
1003                 bio_endio(bio);
1004         }
1005         return ret;
1006 }
1007
1008 static int check_async_write(unsigned long bio_flags)
1009 {
1010         if (bio_flags & EXTENT_BIO_TREE_LOG)
1011                 return 0;
1012 #ifdef CONFIG_X86
1013         if (static_cpu_has(X86_FEATURE_XMM4_2))
1014                 return 0;
1015 #endif
1016         return 1;
1017 }
1018
1019 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
1020                                           int mirror_num, unsigned long bio_flags,
1021                                           u64 bio_offset)
1022 {
1023         struct inode *inode = private_data;
1024         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1025         int async = check_async_write(bio_flags);
1026         blk_status_t ret;
1027
1028         if (bio_op(bio) != REQ_OP_WRITE) {
1029                 /*
1030                  * called for a read, do the setup so that checksum validation
1031                  * can happen in the async kernel threads
1032                  */
1033                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1034                                           BTRFS_WQ_ENDIO_METADATA);
1035                 if (ret)
1036                         goto out_w_error;
1037                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1038         } else if (!async) {
1039                 ret = btree_csum_one_bio(bio);
1040                 if (ret)
1041                         goto out_w_error;
1042                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1043         } else {
1044                 /*
1045                  * kthread helpers are used to submit writes so that
1046                  * checksumming can happen in parallel across all CPUs
1047                  */
1048                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
1049                                           bio_offset, private_data,
1050                                           __btree_submit_bio_start,
1051                                           __btree_submit_bio_done);
1052         }
1053
1054         if (ret)
1055                 goto out_w_error;
1056         return 0;
1057
1058 out_w_error:
1059         bio->bi_status = ret;
1060         bio_endio(bio);
1061         return ret;
1062 }
1063
1064 #ifdef CONFIG_MIGRATION
1065 static int btree_migratepage(struct address_space *mapping,
1066                         struct page *newpage, struct page *page,
1067                         enum migrate_mode mode)
1068 {
1069         /*
1070          * we can't safely write a btree page from here,
1071          * we haven't done the locking hook
1072          */
1073         if (PageDirty(page))
1074                 return -EAGAIN;
1075         /*
1076          * Buffers may be managed in a filesystem specific way.
1077          * We must have no buffers or drop them.
1078          */
1079         if (page_has_private(page) &&
1080             !try_to_release_page(page, GFP_KERNEL))
1081                 return -EAGAIN;
1082         return migrate_page(mapping, newpage, page, mode);
1083 }
1084 #endif
1085
1086
1087 static int btree_writepages(struct address_space *mapping,
1088                             struct writeback_control *wbc)
1089 {
1090         struct btrfs_fs_info *fs_info;
1091         int ret;
1092
1093         if (wbc->sync_mode == WB_SYNC_NONE) {
1094
1095                 if (wbc->for_kupdate)
1096                         return 0;
1097
1098                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1099                 /* this is a bit racy, but that's ok */
1100                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1101                                              BTRFS_DIRTY_METADATA_THRESH);
1102                 if (ret < 0)
1103                         return 0;
1104         }
1105         return btree_write_cache_pages(mapping, wbc);
1106 }
1107
1108 static int btree_readpage(struct file *file, struct page *page)
1109 {
1110         struct extent_io_tree *tree;
1111         tree = &BTRFS_I(page->mapping->host)->io_tree;
1112         return extent_read_full_page(tree, page, btree_get_extent, 0);
1113 }
1114
1115 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1116 {
1117         if (PageWriteback(page) || PageDirty(page))
1118                 return 0;
1119
1120         return try_release_extent_buffer(page);
1121 }
1122
1123 static void btree_invalidatepage(struct page *page, unsigned int offset,
1124                                  unsigned int length)
1125 {
1126         struct extent_io_tree *tree;
1127         tree = &BTRFS_I(page->mapping->host)->io_tree;
1128         extent_invalidatepage(tree, page, offset);
1129         btree_releasepage(page, GFP_NOFS);
1130         if (PagePrivate(page)) {
1131                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1132                            "page private not zero on page %llu",
1133                            (unsigned long long)page_offset(page));
1134                 ClearPagePrivate(page);
1135                 set_page_private(page, 0);
1136                 put_page(page);
1137         }
1138 }
1139
1140 static int btree_set_page_dirty(struct page *page)
1141 {
1142 #ifdef DEBUG
1143         struct extent_buffer *eb;
1144
1145         BUG_ON(!PagePrivate(page));
1146         eb = (struct extent_buffer *)page->private;
1147         BUG_ON(!eb);
1148         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1149         BUG_ON(!atomic_read(&eb->refs));
1150         btrfs_assert_tree_locked(eb);
1151 #endif
1152         return __set_page_dirty_nobuffers(page);
1153 }
1154
1155 static const struct address_space_operations btree_aops = {
1156         .readpage       = btree_readpage,
1157         .writepages     = btree_writepages,
1158         .releasepage    = btree_releasepage,
1159         .invalidatepage = btree_invalidatepage,
1160 #ifdef CONFIG_MIGRATION
1161         .migratepage    = btree_migratepage,
1162 #endif
1163         .set_page_dirty = btree_set_page_dirty,
1164 };
1165
1166 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1167 {
1168         struct extent_buffer *buf = NULL;
1169         struct inode *btree_inode = fs_info->btree_inode;
1170
1171         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1172         if (IS_ERR(buf))
1173                 return;
1174         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1175                                  buf, WAIT_NONE, btree_get_extent, 0);
1176         free_extent_buffer(buf);
1177 }
1178
1179 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1180                          int mirror_num, struct extent_buffer **eb)
1181 {
1182         struct extent_buffer *buf = NULL;
1183         struct inode *btree_inode = fs_info->btree_inode;
1184         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1185         int ret;
1186
1187         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1188         if (IS_ERR(buf))
1189                 return 0;
1190
1191         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1192
1193         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1194                                        btree_get_extent, mirror_num);
1195         if (ret) {
1196                 free_extent_buffer(buf);
1197                 return ret;
1198         }
1199
1200         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1201                 free_extent_buffer(buf);
1202                 return -EIO;
1203         } else if (extent_buffer_uptodate(buf)) {
1204                 *eb = buf;
1205         } else {
1206                 free_extent_buffer(buf);
1207         }
1208         return 0;
1209 }
1210
1211 struct extent_buffer *btrfs_find_create_tree_block(
1212                                                 struct btrfs_fs_info *fs_info,
1213                                                 u64 bytenr)
1214 {
1215         if (btrfs_is_testing(fs_info))
1216                 return alloc_test_extent_buffer(fs_info, bytenr);
1217         return alloc_extent_buffer(fs_info, bytenr);
1218 }
1219
1220
1221 int btrfs_write_tree_block(struct extent_buffer *buf)
1222 {
1223         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1224                                         buf->start + buf->len - 1);
1225 }
1226
1227 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1228 {
1229         filemap_fdatawait_range(buf->pages[0]->mapping,
1230                                 buf->start, buf->start + buf->len - 1);
1231 }
1232
1233 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1234                                       u64 parent_transid)
1235 {
1236         struct extent_buffer *buf = NULL;
1237         int ret;
1238
1239         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1240         if (IS_ERR(buf))
1241                 return buf;
1242
1243         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1244         if (ret) {
1245                 free_extent_buffer(buf);
1246                 return ERR_PTR(ret);
1247         }
1248         return buf;
1249
1250 }
1251
1252 void clean_tree_block(struct btrfs_fs_info *fs_info,
1253                       struct extent_buffer *buf)
1254 {
1255         if (btrfs_header_generation(buf) ==
1256             fs_info->running_transaction->transid) {
1257                 btrfs_assert_tree_locked(buf);
1258
1259                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1260                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1261                                                  -buf->len,
1262                                                  fs_info->dirty_metadata_batch);
1263                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1264                         btrfs_set_lock_blocking(buf);
1265                         clear_extent_buffer_dirty(buf);
1266                 }
1267         }
1268 }
1269
1270 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1271 {
1272         struct btrfs_subvolume_writers *writers;
1273         int ret;
1274
1275         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1276         if (!writers)
1277                 return ERR_PTR(-ENOMEM);
1278
1279         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1280         if (ret < 0) {
1281                 kfree(writers);
1282                 return ERR_PTR(ret);
1283         }
1284
1285         init_waitqueue_head(&writers->wait);
1286         return writers;
1287 }
1288
1289 static void
1290 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1291 {
1292         percpu_counter_destroy(&writers->counter);
1293         kfree(writers);
1294 }
1295
1296 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1297                          u64 objectid)
1298 {
1299         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1300         root->node = NULL;
1301         root->commit_root = NULL;
1302         root->state = 0;
1303         root->orphan_cleanup_state = 0;
1304
1305         root->objectid = objectid;
1306         root->last_trans = 0;
1307         root->highest_objectid = 0;
1308         root->nr_delalloc_inodes = 0;
1309         root->nr_ordered_extents = 0;
1310         root->name = NULL;
1311         root->inode_tree = RB_ROOT;
1312         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1313         root->block_rsv = NULL;
1314         root->orphan_block_rsv = NULL;
1315
1316         INIT_LIST_HEAD(&root->dirty_list);
1317         INIT_LIST_HEAD(&root->root_list);
1318         INIT_LIST_HEAD(&root->delalloc_inodes);
1319         INIT_LIST_HEAD(&root->delalloc_root);
1320         INIT_LIST_HEAD(&root->ordered_extents);
1321         INIT_LIST_HEAD(&root->ordered_root);
1322         INIT_LIST_HEAD(&root->logged_list[0]);
1323         INIT_LIST_HEAD(&root->logged_list[1]);
1324         spin_lock_init(&root->orphan_lock);
1325         spin_lock_init(&root->inode_lock);
1326         spin_lock_init(&root->delalloc_lock);
1327         spin_lock_init(&root->ordered_extent_lock);
1328         spin_lock_init(&root->accounting_lock);
1329         spin_lock_init(&root->log_extents_lock[0]);
1330         spin_lock_init(&root->log_extents_lock[1]);
1331         mutex_init(&root->objectid_mutex);
1332         mutex_init(&root->log_mutex);
1333         mutex_init(&root->ordered_extent_mutex);
1334         mutex_init(&root->delalloc_mutex);
1335         init_waitqueue_head(&root->log_writer_wait);
1336         init_waitqueue_head(&root->log_commit_wait[0]);
1337         init_waitqueue_head(&root->log_commit_wait[1]);
1338         INIT_LIST_HEAD(&root->log_ctxs[0]);
1339         INIT_LIST_HEAD(&root->log_ctxs[1]);
1340         atomic_set(&root->log_commit[0], 0);
1341         atomic_set(&root->log_commit[1], 0);
1342         atomic_set(&root->log_writers, 0);
1343         atomic_set(&root->log_batch, 0);
1344         atomic_set(&root->orphan_inodes, 0);
1345         refcount_set(&root->refs, 1);
1346         atomic_set(&root->will_be_snapshotted, 0);
1347         atomic64_set(&root->qgroup_meta_rsv, 0);
1348         root->log_transid = 0;
1349         root->log_transid_committed = -1;
1350         root->last_log_commit = 0;
1351         if (!dummy)
1352                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1353
1354         memset(&root->root_key, 0, sizeof(root->root_key));
1355         memset(&root->root_item, 0, sizeof(root->root_item));
1356         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1357         if (!dummy)
1358                 root->defrag_trans_start = fs_info->generation;
1359         else
1360                 root->defrag_trans_start = 0;
1361         root->root_key.objectid = objectid;
1362         root->anon_dev = 0;
1363
1364         spin_lock_init(&root->root_item_lock);
1365 }
1366
1367 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1368                 gfp_t flags)
1369 {
1370         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1371         if (root)
1372                 root->fs_info = fs_info;
1373         return root;
1374 }
1375
1376 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1377 /* Should only be used by the testing infrastructure */
1378 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1379 {
1380         struct btrfs_root *root;
1381
1382         if (!fs_info)
1383                 return ERR_PTR(-EINVAL);
1384
1385         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1386         if (!root)
1387                 return ERR_PTR(-ENOMEM);
1388
1389         /* We don't use the stripesize in selftest, set it as sectorsize */
1390         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1391         root->alloc_bytenr = 0;
1392
1393         return root;
1394 }
1395 #endif
1396
1397 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1398                                      struct btrfs_fs_info *fs_info,
1399                                      u64 objectid)
1400 {
1401         struct extent_buffer *leaf;
1402         struct btrfs_root *tree_root = fs_info->tree_root;
1403         struct btrfs_root *root;
1404         struct btrfs_key key;
1405         int ret = 0;
1406         uuid_le uuid;
1407
1408         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1409         if (!root)
1410                 return ERR_PTR(-ENOMEM);
1411
1412         __setup_root(root, fs_info, objectid);
1413         root->root_key.objectid = objectid;
1414         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1415         root->root_key.offset = 0;
1416
1417         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1418         if (IS_ERR(leaf)) {
1419                 ret = PTR_ERR(leaf);
1420                 leaf = NULL;
1421                 goto fail;
1422         }
1423
1424         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1425         btrfs_set_header_bytenr(leaf, leaf->start);
1426         btrfs_set_header_generation(leaf, trans->transid);
1427         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1428         btrfs_set_header_owner(leaf, objectid);
1429         root->node = leaf;
1430
1431         write_extent_buffer_fsid(leaf, fs_info->fsid);
1432         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1433         btrfs_mark_buffer_dirty(leaf);
1434
1435         root->commit_root = btrfs_root_node(root);
1436         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1437
1438         root->root_item.flags = 0;
1439         root->root_item.byte_limit = 0;
1440         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1441         btrfs_set_root_generation(&root->root_item, trans->transid);
1442         btrfs_set_root_level(&root->root_item, 0);
1443         btrfs_set_root_refs(&root->root_item, 1);
1444         btrfs_set_root_used(&root->root_item, leaf->len);
1445         btrfs_set_root_last_snapshot(&root->root_item, 0);
1446         btrfs_set_root_dirid(&root->root_item, 0);
1447         uuid_le_gen(&uuid);
1448         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1449         root->root_item.drop_level = 0;
1450
1451         key.objectid = objectid;
1452         key.type = BTRFS_ROOT_ITEM_KEY;
1453         key.offset = 0;
1454         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1455         if (ret)
1456                 goto fail;
1457
1458         btrfs_tree_unlock(leaf);
1459
1460         return root;
1461
1462 fail:
1463         if (leaf) {
1464                 btrfs_tree_unlock(leaf);
1465                 free_extent_buffer(root->commit_root);
1466                 free_extent_buffer(leaf);
1467         }
1468         kfree(root);
1469
1470         return ERR_PTR(ret);
1471 }
1472
1473 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1474                                          struct btrfs_fs_info *fs_info)
1475 {
1476         struct btrfs_root *root;
1477         struct extent_buffer *leaf;
1478
1479         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1480         if (!root)
1481                 return ERR_PTR(-ENOMEM);
1482
1483         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1484
1485         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1486         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1487         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1488
1489         /*
1490          * DON'T set REF_COWS for log trees
1491          *
1492          * log trees do not get reference counted because they go away
1493          * before a real commit is actually done.  They do store pointers
1494          * to file data extents, and those reference counts still get
1495          * updated (along with back refs to the log tree).
1496          */
1497
1498         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1499                         NULL, 0, 0, 0);
1500         if (IS_ERR(leaf)) {
1501                 kfree(root);
1502                 return ERR_CAST(leaf);
1503         }
1504
1505         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1506         btrfs_set_header_bytenr(leaf, leaf->start);
1507         btrfs_set_header_generation(leaf, trans->transid);
1508         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1509         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1510         root->node = leaf;
1511
1512         write_extent_buffer_fsid(root->node, fs_info->fsid);
1513         btrfs_mark_buffer_dirty(root->node);
1514         btrfs_tree_unlock(root->node);
1515         return root;
1516 }
1517
1518 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1519                              struct btrfs_fs_info *fs_info)
1520 {
1521         struct btrfs_root *log_root;
1522
1523         log_root = alloc_log_tree(trans, fs_info);
1524         if (IS_ERR(log_root))
1525                 return PTR_ERR(log_root);
1526         WARN_ON(fs_info->log_root_tree);
1527         fs_info->log_root_tree = log_root;
1528         return 0;
1529 }
1530
1531 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1532                        struct btrfs_root *root)
1533 {
1534         struct btrfs_fs_info *fs_info = root->fs_info;
1535         struct btrfs_root *log_root;
1536         struct btrfs_inode_item *inode_item;
1537
1538         log_root = alloc_log_tree(trans, fs_info);
1539         if (IS_ERR(log_root))
1540                 return PTR_ERR(log_root);
1541
1542         log_root->last_trans = trans->transid;
1543         log_root->root_key.offset = root->root_key.objectid;
1544
1545         inode_item = &log_root->root_item.inode;
1546         btrfs_set_stack_inode_generation(inode_item, 1);
1547         btrfs_set_stack_inode_size(inode_item, 3);
1548         btrfs_set_stack_inode_nlink(inode_item, 1);
1549         btrfs_set_stack_inode_nbytes(inode_item,
1550                                      fs_info->nodesize);
1551         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1552
1553         btrfs_set_root_node(&log_root->root_item, log_root->node);
1554
1555         WARN_ON(root->log_root);
1556         root->log_root = log_root;
1557         root->log_transid = 0;
1558         root->log_transid_committed = -1;
1559         root->last_log_commit = 0;
1560         return 0;
1561 }
1562
1563 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1564                                                struct btrfs_key *key)
1565 {
1566         struct btrfs_root *root;
1567         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1568         struct btrfs_path *path;
1569         u64 generation;
1570         int ret;
1571
1572         path = btrfs_alloc_path();
1573         if (!path)
1574                 return ERR_PTR(-ENOMEM);
1575
1576         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1577         if (!root) {
1578                 ret = -ENOMEM;
1579                 goto alloc_fail;
1580         }
1581
1582         __setup_root(root, fs_info, key->objectid);
1583
1584         ret = btrfs_find_root(tree_root, key, path,
1585                               &root->root_item, &root->root_key);
1586         if (ret) {
1587                 if (ret > 0)
1588                         ret = -ENOENT;
1589                 goto find_fail;
1590         }
1591
1592         generation = btrfs_root_generation(&root->root_item);
1593         root->node = read_tree_block(fs_info,
1594                                      btrfs_root_bytenr(&root->root_item),
1595                                      generation);
1596         if (IS_ERR(root->node)) {
1597                 ret = PTR_ERR(root->node);
1598                 goto find_fail;
1599         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1600                 ret = -EIO;
1601                 free_extent_buffer(root->node);
1602                 goto find_fail;
1603         }
1604         root->commit_root = btrfs_root_node(root);
1605 out:
1606         btrfs_free_path(path);
1607         return root;
1608
1609 find_fail:
1610         kfree(root);
1611 alloc_fail:
1612         root = ERR_PTR(ret);
1613         goto out;
1614 }
1615
1616 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1617                                       struct btrfs_key *location)
1618 {
1619         struct btrfs_root *root;
1620
1621         root = btrfs_read_tree_root(tree_root, location);
1622         if (IS_ERR(root))
1623                 return root;
1624
1625         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1626                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1627                 btrfs_check_and_init_root_item(&root->root_item);
1628         }
1629
1630         return root;
1631 }
1632
1633 int btrfs_init_fs_root(struct btrfs_root *root)
1634 {
1635         int ret;
1636         struct btrfs_subvolume_writers *writers;
1637
1638         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1639         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1640                                         GFP_NOFS);
1641         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1642                 ret = -ENOMEM;
1643                 goto fail;
1644         }
1645
1646         writers = btrfs_alloc_subvolume_writers();
1647         if (IS_ERR(writers)) {
1648                 ret = PTR_ERR(writers);
1649                 goto fail;
1650         }
1651         root->subv_writers = writers;
1652
1653         btrfs_init_free_ino_ctl(root);
1654         spin_lock_init(&root->ino_cache_lock);
1655         init_waitqueue_head(&root->ino_cache_wait);
1656
1657         ret = get_anon_bdev(&root->anon_dev);
1658         if (ret)
1659                 goto fail;
1660
1661         mutex_lock(&root->objectid_mutex);
1662         ret = btrfs_find_highest_objectid(root,
1663                                         &root->highest_objectid);
1664         if (ret) {
1665                 mutex_unlock(&root->objectid_mutex);
1666                 goto fail;
1667         }
1668
1669         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1670
1671         mutex_unlock(&root->objectid_mutex);
1672
1673         return 0;
1674 fail:
1675         /* the caller is responsible to call free_fs_root */
1676         return ret;
1677 }
1678
1679 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1680                                         u64 root_id)
1681 {
1682         struct btrfs_root *root;
1683
1684         spin_lock(&fs_info->fs_roots_radix_lock);
1685         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1686                                  (unsigned long)root_id);
1687         spin_unlock(&fs_info->fs_roots_radix_lock);
1688         return root;
1689 }
1690
1691 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1692                          struct btrfs_root *root)
1693 {
1694         int ret;
1695
1696         ret = radix_tree_preload(GFP_NOFS);
1697         if (ret)
1698                 return ret;
1699
1700         spin_lock(&fs_info->fs_roots_radix_lock);
1701         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1702                                 (unsigned long)root->root_key.objectid,
1703                                 root);
1704         if (ret == 0)
1705                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1706         spin_unlock(&fs_info->fs_roots_radix_lock);
1707         radix_tree_preload_end();
1708
1709         return ret;
1710 }
1711
1712 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1713                                      struct btrfs_key *location,
1714                                      bool check_ref)
1715 {
1716         struct btrfs_root *root;
1717         struct btrfs_path *path;
1718         struct btrfs_key key;
1719         int ret;
1720
1721         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1722                 return fs_info->tree_root;
1723         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1724                 return fs_info->extent_root;
1725         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1726                 return fs_info->chunk_root;
1727         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1728                 return fs_info->dev_root;
1729         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1730                 return fs_info->csum_root;
1731         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1732                 return fs_info->quota_root ? fs_info->quota_root :
1733                                              ERR_PTR(-ENOENT);
1734         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1735                 return fs_info->uuid_root ? fs_info->uuid_root :
1736                                             ERR_PTR(-ENOENT);
1737         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1738                 return fs_info->free_space_root ? fs_info->free_space_root :
1739                                                   ERR_PTR(-ENOENT);
1740 again:
1741         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1742         if (root) {
1743                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1744                         return ERR_PTR(-ENOENT);
1745                 return root;
1746         }
1747
1748         root = btrfs_read_fs_root(fs_info->tree_root, location);
1749         if (IS_ERR(root))
1750                 return root;
1751
1752         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1753                 ret = -ENOENT;
1754                 goto fail;
1755         }
1756
1757         ret = btrfs_init_fs_root(root);
1758         if (ret)
1759                 goto fail;
1760
1761         path = btrfs_alloc_path();
1762         if (!path) {
1763                 ret = -ENOMEM;
1764                 goto fail;
1765         }
1766         key.objectid = BTRFS_ORPHAN_OBJECTID;
1767         key.type = BTRFS_ORPHAN_ITEM_KEY;
1768         key.offset = location->objectid;
1769
1770         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1771         btrfs_free_path(path);
1772         if (ret < 0)
1773                 goto fail;
1774         if (ret == 0)
1775                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1776
1777         ret = btrfs_insert_fs_root(fs_info, root);
1778         if (ret) {
1779                 if (ret == -EEXIST) {
1780                         free_fs_root(root);
1781                         goto again;
1782                 }
1783                 goto fail;
1784         }
1785         return root;
1786 fail:
1787         free_fs_root(root);
1788         return ERR_PTR(ret);
1789 }
1790
1791 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1792 {
1793         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1794         int ret = 0;
1795         struct btrfs_device *device;
1796         struct backing_dev_info *bdi;
1797
1798         rcu_read_lock();
1799         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1800                 if (!device->bdev)
1801                         continue;
1802                 bdi = device->bdev->bd_bdi;
1803                 if (bdi_congested(bdi, bdi_bits)) {
1804                         ret = 1;
1805                         break;
1806                 }
1807         }
1808         rcu_read_unlock();
1809         return ret;
1810 }
1811
1812 /*
1813  * called by the kthread helper functions to finally call the bio end_io
1814  * functions.  This is where read checksum verification actually happens
1815  */
1816 static void end_workqueue_fn(struct btrfs_work *work)
1817 {
1818         struct bio *bio;
1819         struct btrfs_end_io_wq *end_io_wq;
1820
1821         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1822         bio = end_io_wq->bio;
1823
1824         bio->bi_status = end_io_wq->status;
1825         bio->bi_private = end_io_wq->private;
1826         bio->bi_end_io = end_io_wq->end_io;
1827         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1828         bio_endio(bio);
1829 }
1830
1831 static int cleaner_kthread(void *arg)
1832 {
1833         struct btrfs_root *root = arg;
1834         struct btrfs_fs_info *fs_info = root->fs_info;
1835         int again;
1836         struct btrfs_trans_handle *trans;
1837
1838         do {
1839                 again = 0;
1840
1841                 /* Make the cleaner go to sleep early. */
1842                 if (btrfs_need_cleaner_sleep(fs_info))
1843                         goto sleep;
1844
1845                 /*
1846                  * Do not do anything if we might cause open_ctree() to block
1847                  * before we have finished mounting the filesystem.
1848                  */
1849                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1850                         goto sleep;
1851
1852                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1853                         goto sleep;
1854
1855                 /*
1856                  * Avoid the problem that we change the status of the fs
1857                  * during the above check and trylock.
1858                  */
1859                 if (btrfs_need_cleaner_sleep(fs_info)) {
1860                         mutex_unlock(&fs_info->cleaner_mutex);
1861                         goto sleep;
1862                 }
1863
1864                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1865                 btrfs_run_delayed_iputs(fs_info);
1866                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1867
1868                 again = btrfs_clean_one_deleted_snapshot(root);
1869                 mutex_unlock(&fs_info->cleaner_mutex);
1870
1871                 /*
1872                  * The defragger has dealt with the R/O remount and umount,
1873                  * needn't do anything special here.
1874                  */
1875                 btrfs_run_defrag_inodes(fs_info);
1876
1877                 /*
1878                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1879                  * with relocation (btrfs_relocate_chunk) and relocation
1880                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1881                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1882                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1883                  * unused block groups.
1884                  */
1885                 btrfs_delete_unused_bgs(fs_info);
1886 sleep:
1887                 if (!again) {
1888                         set_current_state(TASK_INTERRUPTIBLE);
1889                         if (!kthread_should_stop())
1890                                 schedule();
1891                         __set_current_state(TASK_RUNNING);
1892                 }
1893         } while (!kthread_should_stop());
1894
1895         /*
1896          * Transaction kthread is stopped before us and wakes us up.
1897          * However we might have started a new transaction and COWed some
1898          * tree blocks when deleting unused block groups for example. So
1899          * make sure we commit the transaction we started to have a clean
1900          * shutdown when evicting the btree inode - if it has dirty pages
1901          * when we do the final iput() on it, eviction will trigger a
1902          * writeback for it which will fail with null pointer dereferences
1903          * since work queues and other resources were already released and
1904          * destroyed by the time the iput/eviction/writeback is made.
1905          */
1906         trans = btrfs_attach_transaction(root);
1907         if (IS_ERR(trans)) {
1908                 if (PTR_ERR(trans) != -ENOENT)
1909                         btrfs_err(fs_info,
1910                                   "cleaner transaction attach returned %ld",
1911                                   PTR_ERR(trans));
1912         } else {
1913                 int ret;
1914
1915                 ret = btrfs_commit_transaction(trans);
1916                 if (ret)
1917                         btrfs_err(fs_info,
1918                                   "cleaner open transaction commit returned %d",
1919                                   ret);
1920         }
1921
1922         return 0;
1923 }
1924
1925 static int transaction_kthread(void *arg)
1926 {
1927         struct btrfs_root *root = arg;
1928         struct btrfs_fs_info *fs_info = root->fs_info;
1929         struct btrfs_trans_handle *trans;
1930         struct btrfs_transaction *cur;
1931         u64 transid;
1932         unsigned long now;
1933         unsigned long delay;
1934         bool cannot_commit;
1935
1936         do {
1937                 cannot_commit = false;
1938                 delay = HZ * fs_info->commit_interval;
1939                 mutex_lock(&fs_info->transaction_kthread_mutex);
1940
1941                 spin_lock(&fs_info->trans_lock);
1942                 cur = fs_info->running_transaction;
1943                 if (!cur) {
1944                         spin_unlock(&fs_info->trans_lock);
1945                         goto sleep;
1946                 }
1947
1948                 now = get_seconds();
1949                 if (cur->state < TRANS_STATE_BLOCKED &&
1950                     (now < cur->start_time ||
1951                      now - cur->start_time < fs_info->commit_interval)) {
1952                         spin_unlock(&fs_info->trans_lock);
1953                         delay = HZ * 5;
1954                         goto sleep;
1955                 }
1956                 transid = cur->transid;
1957                 spin_unlock(&fs_info->trans_lock);
1958
1959                 /* If the file system is aborted, this will always fail. */
1960                 trans = btrfs_attach_transaction(root);
1961                 if (IS_ERR(trans)) {
1962                         if (PTR_ERR(trans) != -ENOENT)
1963                                 cannot_commit = true;
1964                         goto sleep;
1965                 }
1966                 if (transid == trans->transid) {
1967                         btrfs_commit_transaction(trans);
1968                 } else {
1969                         btrfs_end_transaction(trans);
1970                 }
1971 sleep:
1972                 wake_up_process(fs_info->cleaner_kthread);
1973                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1974
1975                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1976                                       &fs_info->fs_state)))
1977                         btrfs_cleanup_transaction(fs_info);
1978                 set_current_state(TASK_INTERRUPTIBLE);
1979                 if (!kthread_should_stop() &&
1980                                 (!btrfs_transaction_blocked(fs_info) ||
1981                                  cannot_commit))
1982                         schedule_timeout(delay);
1983                 __set_current_state(TASK_RUNNING);
1984         } while (!kthread_should_stop());
1985         return 0;
1986 }
1987
1988 /*
1989  * this will find the highest generation in the array of
1990  * root backups.  The index of the highest array is returned,
1991  * or -1 if we can't find anything.
1992  *
1993  * We check to make sure the array is valid by comparing the
1994  * generation of the latest  root in the array with the generation
1995  * in the super block.  If they don't match we pitch it.
1996  */
1997 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1998 {
1999         u64 cur;
2000         int newest_index = -1;
2001         struct btrfs_root_backup *root_backup;
2002         int i;
2003
2004         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2005                 root_backup = info->super_copy->super_roots + i;
2006                 cur = btrfs_backup_tree_root_gen(root_backup);
2007                 if (cur == newest_gen)
2008                         newest_index = i;
2009         }
2010
2011         /* check to see if we actually wrapped around */
2012         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2013                 root_backup = info->super_copy->super_roots;
2014                 cur = btrfs_backup_tree_root_gen(root_backup);
2015                 if (cur == newest_gen)
2016                         newest_index = 0;
2017         }
2018         return newest_index;
2019 }
2020
2021
2022 /*
2023  * find the oldest backup so we know where to store new entries
2024  * in the backup array.  This will set the backup_root_index
2025  * field in the fs_info struct
2026  */
2027 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2028                                      u64 newest_gen)
2029 {
2030         int newest_index = -1;
2031
2032         newest_index = find_newest_super_backup(info, newest_gen);
2033         /* if there was garbage in there, just move along */
2034         if (newest_index == -1) {
2035                 info->backup_root_index = 0;
2036         } else {
2037                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2038         }
2039 }
2040
2041 /*
2042  * copy all the root pointers into the super backup array.
2043  * this will bump the backup pointer by one when it is
2044  * done
2045  */
2046 static void backup_super_roots(struct btrfs_fs_info *info)
2047 {
2048         int next_backup;
2049         struct btrfs_root_backup *root_backup;
2050         int last_backup;
2051
2052         next_backup = info->backup_root_index;
2053         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2054                 BTRFS_NUM_BACKUP_ROOTS;
2055
2056         /*
2057          * just overwrite the last backup if we're at the same generation
2058          * this happens only at umount
2059          */
2060         root_backup = info->super_for_commit->super_roots + last_backup;
2061         if (btrfs_backup_tree_root_gen(root_backup) ==
2062             btrfs_header_generation(info->tree_root->node))
2063                 next_backup = last_backup;
2064
2065         root_backup = info->super_for_commit->super_roots + next_backup;
2066
2067         /*
2068          * make sure all of our padding and empty slots get zero filled
2069          * regardless of which ones we use today
2070          */
2071         memset(root_backup, 0, sizeof(*root_backup));
2072
2073         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2074
2075         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2076         btrfs_set_backup_tree_root_gen(root_backup,
2077                                btrfs_header_generation(info->tree_root->node));
2078
2079         btrfs_set_backup_tree_root_level(root_backup,
2080                                btrfs_header_level(info->tree_root->node));
2081
2082         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2083         btrfs_set_backup_chunk_root_gen(root_backup,
2084                                btrfs_header_generation(info->chunk_root->node));
2085         btrfs_set_backup_chunk_root_level(root_backup,
2086                                btrfs_header_level(info->chunk_root->node));
2087
2088         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2089         btrfs_set_backup_extent_root_gen(root_backup,
2090                                btrfs_header_generation(info->extent_root->node));
2091         btrfs_set_backup_extent_root_level(root_backup,
2092                                btrfs_header_level(info->extent_root->node));
2093
2094         /*
2095          * we might commit during log recovery, which happens before we set
2096          * the fs_root.  Make sure it is valid before we fill it in.
2097          */
2098         if (info->fs_root && info->fs_root->node) {
2099                 btrfs_set_backup_fs_root(root_backup,
2100                                          info->fs_root->node->start);
2101                 btrfs_set_backup_fs_root_gen(root_backup,
2102                                btrfs_header_generation(info->fs_root->node));
2103                 btrfs_set_backup_fs_root_level(root_backup,
2104                                btrfs_header_level(info->fs_root->node));
2105         }
2106
2107         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2108         btrfs_set_backup_dev_root_gen(root_backup,
2109                                btrfs_header_generation(info->dev_root->node));
2110         btrfs_set_backup_dev_root_level(root_backup,
2111                                        btrfs_header_level(info->dev_root->node));
2112
2113         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2114         btrfs_set_backup_csum_root_gen(root_backup,
2115                                btrfs_header_generation(info->csum_root->node));
2116         btrfs_set_backup_csum_root_level(root_backup,
2117                                btrfs_header_level(info->csum_root->node));
2118
2119         btrfs_set_backup_total_bytes(root_backup,
2120                              btrfs_super_total_bytes(info->super_copy));
2121         btrfs_set_backup_bytes_used(root_backup,
2122                              btrfs_super_bytes_used(info->super_copy));
2123         btrfs_set_backup_num_devices(root_backup,
2124                              btrfs_super_num_devices(info->super_copy));
2125
2126         /*
2127          * if we don't copy this out to the super_copy, it won't get remembered
2128          * for the next commit
2129          */
2130         memcpy(&info->super_copy->super_roots,
2131                &info->super_for_commit->super_roots,
2132                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2133 }
2134
2135 /*
2136  * this copies info out of the root backup array and back into
2137  * the in-memory super block.  It is meant to help iterate through
2138  * the array, so you send it the number of backups you've already
2139  * tried and the last backup index you used.
2140  *
2141  * this returns -1 when it has tried all the backups
2142  */
2143 static noinline int next_root_backup(struct btrfs_fs_info *info,
2144                                      struct btrfs_super_block *super,
2145                                      int *num_backups_tried, int *backup_index)
2146 {
2147         struct btrfs_root_backup *root_backup;
2148         int newest = *backup_index;
2149
2150         if (*num_backups_tried == 0) {
2151                 u64 gen = btrfs_super_generation(super);
2152
2153                 newest = find_newest_super_backup(info, gen);
2154                 if (newest == -1)
2155                         return -1;
2156
2157                 *backup_index = newest;
2158                 *num_backups_tried = 1;
2159         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2160                 /* we've tried all the backups, all done */
2161                 return -1;
2162         } else {
2163                 /* jump to the next oldest backup */
2164                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2165                         BTRFS_NUM_BACKUP_ROOTS;
2166                 *backup_index = newest;
2167                 *num_backups_tried += 1;
2168         }
2169         root_backup = super->super_roots + newest;
2170
2171         btrfs_set_super_generation(super,
2172                                    btrfs_backup_tree_root_gen(root_backup));
2173         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2174         btrfs_set_super_root_level(super,
2175                                    btrfs_backup_tree_root_level(root_backup));
2176         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2177
2178         /*
2179          * fixme: the total bytes and num_devices need to match or we should
2180          * need a fsck
2181          */
2182         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2183         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2184         return 0;
2185 }
2186
2187 /* helper to cleanup workers */
2188 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2189 {
2190         btrfs_destroy_workqueue(fs_info->fixup_workers);
2191         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2192         btrfs_destroy_workqueue(fs_info->workers);
2193         btrfs_destroy_workqueue(fs_info->endio_workers);
2194         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2195         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2196         btrfs_destroy_workqueue(fs_info->rmw_workers);
2197         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2198         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2199         btrfs_destroy_workqueue(fs_info->submit_workers);
2200         btrfs_destroy_workqueue(fs_info->delayed_workers);
2201         btrfs_destroy_workqueue(fs_info->caching_workers);
2202         btrfs_destroy_workqueue(fs_info->readahead_workers);
2203         btrfs_destroy_workqueue(fs_info->flush_workers);
2204         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2205         btrfs_destroy_workqueue(fs_info->extent_workers);
2206         /*
2207          * Now that all other work queues are destroyed, we can safely destroy
2208          * the queues used for metadata I/O, since tasks from those other work
2209          * queues can do metadata I/O operations.
2210          */
2211         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2212         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2213 }
2214
2215 static void free_root_extent_buffers(struct btrfs_root *root)
2216 {
2217         if (root) {
2218                 free_extent_buffer(root->node);
2219                 free_extent_buffer(root->commit_root);
2220                 root->node = NULL;
2221                 root->commit_root = NULL;
2222         }
2223 }
2224
2225 /* helper to cleanup tree roots */
2226 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2227 {
2228         free_root_extent_buffers(info->tree_root);
2229
2230         free_root_extent_buffers(info->dev_root);
2231         free_root_extent_buffers(info->extent_root);
2232         free_root_extent_buffers(info->csum_root);
2233         free_root_extent_buffers(info->quota_root);
2234         free_root_extent_buffers(info->uuid_root);
2235         if (chunk_root)
2236                 free_root_extent_buffers(info->chunk_root);
2237         free_root_extent_buffers(info->free_space_root);
2238 }
2239
2240 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2241 {
2242         int ret;
2243         struct btrfs_root *gang[8];
2244         int i;
2245
2246         while (!list_empty(&fs_info->dead_roots)) {
2247                 gang[0] = list_entry(fs_info->dead_roots.next,
2248                                      struct btrfs_root, root_list);
2249                 list_del(&gang[0]->root_list);
2250
2251                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2252                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2253                 } else {
2254                         free_extent_buffer(gang[0]->node);
2255                         free_extent_buffer(gang[0]->commit_root);
2256                         btrfs_put_fs_root(gang[0]);
2257                 }
2258         }
2259
2260         while (1) {
2261                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2262                                              (void **)gang, 0,
2263                                              ARRAY_SIZE(gang));
2264                 if (!ret)
2265                         break;
2266                 for (i = 0; i < ret; i++)
2267                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2268         }
2269
2270         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2271                 btrfs_free_log_root_tree(NULL, fs_info);
2272                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2273         }
2274 }
2275
2276 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2277 {
2278         mutex_init(&fs_info->scrub_lock);
2279         atomic_set(&fs_info->scrubs_running, 0);
2280         atomic_set(&fs_info->scrub_pause_req, 0);
2281         atomic_set(&fs_info->scrubs_paused, 0);
2282         atomic_set(&fs_info->scrub_cancel_req, 0);
2283         init_waitqueue_head(&fs_info->scrub_pause_wait);
2284         fs_info->scrub_workers_refcnt = 0;
2285 }
2286
2287 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2288 {
2289         spin_lock_init(&fs_info->balance_lock);
2290         mutex_init(&fs_info->balance_mutex);
2291         atomic_set(&fs_info->balance_running, 0);
2292         atomic_set(&fs_info->balance_pause_req, 0);
2293         atomic_set(&fs_info->balance_cancel_req, 0);
2294         fs_info->balance_ctl = NULL;
2295         init_waitqueue_head(&fs_info->balance_wait_q);
2296 }
2297
2298 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2299 {
2300         struct inode *inode = fs_info->btree_inode;
2301
2302         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2303         set_nlink(inode, 1);
2304         /*
2305          * we set the i_size on the btree inode to the max possible int.
2306          * the real end of the address space is determined by all of
2307          * the devices in the system
2308          */
2309         inode->i_size = OFFSET_MAX;
2310         inode->i_mapping->a_ops = &btree_aops;
2311
2312         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2313         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2314         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2315         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2316
2317         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2318
2319         BTRFS_I(inode)->root = fs_info->tree_root;
2320         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2321         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2322         btrfs_insert_inode_hash(inode);
2323 }
2324
2325 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2326 {
2327         fs_info->dev_replace.lock_owner = 0;
2328         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2329         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2330         rwlock_init(&fs_info->dev_replace.lock);
2331         atomic_set(&fs_info->dev_replace.read_locks, 0);
2332         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2333         init_waitqueue_head(&fs_info->replace_wait);
2334         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2335 }
2336
2337 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2338 {
2339         spin_lock_init(&fs_info->qgroup_lock);
2340         mutex_init(&fs_info->qgroup_ioctl_lock);
2341         fs_info->qgroup_tree = RB_ROOT;
2342         fs_info->qgroup_op_tree = RB_ROOT;
2343         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2344         fs_info->qgroup_seq = 1;
2345         fs_info->qgroup_ulist = NULL;
2346         fs_info->qgroup_rescan_running = false;
2347         mutex_init(&fs_info->qgroup_rescan_lock);
2348 }
2349
2350 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2351                 struct btrfs_fs_devices *fs_devices)
2352 {
2353         int max_active = fs_info->thread_pool_size;
2354         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2355
2356         fs_info->workers =
2357                 btrfs_alloc_workqueue(fs_info, "worker",
2358                                       flags | WQ_HIGHPRI, max_active, 16);
2359
2360         fs_info->delalloc_workers =
2361                 btrfs_alloc_workqueue(fs_info, "delalloc",
2362                                       flags, max_active, 2);
2363
2364         fs_info->flush_workers =
2365                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2366                                       flags, max_active, 0);
2367
2368         fs_info->caching_workers =
2369                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2370
2371         /*
2372          * a higher idle thresh on the submit workers makes it much more
2373          * likely that bios will be send down in a sane order to the
2374          * devices
2375          */
2376         fs_info->submit_workers =
2377                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2378                                       min_t(u64, fs_devices->num_devices,
2379                                             max_active), 64);
2380
2381         fs_info->fixup_workers =
2382                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2383
2384         /*
2385          * endios are largely parallel and should have a very
2386          * low idle thresh
2387          */
2388         fs_info->endio_workers =
2389                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2390         fs_info->endio_meta_workers =
2391                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2392                                       max_active, 4);
2393         fs_info->endio_meta_write_workers =
2394                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2395                                       max_active, 2);
2396         fs_info->endio_raid56_workers =
2397                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2398                                       max_active, 4);
2399         fs_info->endio_repair_workers =
2400                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2401         fs_info->rmw_workers =
2402                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2403         fs_info->endio_write_workers =
2404                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2405                                       max_active, 2);
2406         fs_info->endio_freespace_worker =
2407                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2408                                       max_active, 0);
2409         fs_info->delayed_workers =
2410                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2411                                       max_active, 0);
2412         fs_info->readahead_workers =
2413                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2414                                       max_active, 2);
2415         fs_info->qgroup_rescan_workers =
2416                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2417         fs_info->extent_workers =
2418                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2419                                       min_t(u64, fs_devices->num_devices,
2420                                             max_active), 8);
2421
2422         if (!(fs_info->workers && fs_info->delalloc_workers &&
2423               fs_info->submit_workers && fs_info->flush_workers &&
2424               fs_info->endio_workers && fs_info->endio_meta_workers &&
2425               fs_info->endio_meta_write_workers &&
2426               fs_info->endio_repair_workers &&
2427               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2428               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2429               fs_info->caching_workers && fs_info->readahead_workers &&
2430               fs_info->fixup_workers && fs_info->delayed_workers &&
2431               fs_info->extent_workers &&
2432               fs_info->qgroup_rescan_workers)) {
2433                 return -ENOMEM;
2434         }
2435
2436         return 0;
2437 }
2438
2439 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2440                             struct btrfs_fs_devices *fs_devices)
2441 {
2442         int ret;
2443         struct btrfs_root *log_tree_root;
2444         struct btrfs_super_block *disk_super = fs_info->super_copy;
2445         u64 bytenr = btrfs_super_log_root(disk_super);
2446
2447         if (fs_devices->rw_devices == 0) {
2448                 btrfs_warn(fs_info, "log replay required on RO media");
2449                 return -EIO;
2450         }
2451
2452         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2453         if (!log_tree_root)
2454                 return -ENOMEM;
2455
2456         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2457
2458         log_tree_root->node = read_tree_block(fs_info, bytenr,
2459                                               fs_info->generation + 1);
2460         if (IS_ERR(log_tree_root->node)) {
2461                 btrfs_warn(fs_info, "failed to read log tree");
2462                 ret = PTR_ERR(log_tree_root->node);
2463                 kfree(log_tree_root);
2464                 return ret;
2465         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2466                 btrfs_err(fs_info, "failed to read log tree");
2467                 free_extent_buffer(log_tree_root->node);
2468                 kfree(log_tree_root);
2469                 return -EIO;
2470         }
2471         /* returns with log_tree_root freed on success */
2472         ret = btrfs_recover_log_trees(log_tree_root);
2473         if (ret) {
2474                 btrfs_handle_fs_error(fs_info, ret,
2475                                       "Failed to recover log tree");
2476                 free_extent_buffer(log_tree_root->node);
2477                 kfree(log_tree_root);
2478                 return ret;
2479         }
2480
2481         if (fs_info->sb->s_flags & MS_RDONLY) {
2482                 ret = btrfs_commit_super(fs_info);
2483                 if (ret)
2484                         return ret;
2485         }
2486
2487         return 0;
2488 }
2489
2490 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2491 {
2492         struct btrfs_root *tree_root = fs_info->tree_root;
2493         struct btrfs_root *root;
2494         struct btrfs_key location;
2495         int ret;
2496
2497         BUG_ON(!fs_info->tree_root);
2498
2499         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2500         location.type = BTRFS_ROOT_ITEM_KEY;
2501         location.offset = 0;
2502
2503         root = btrfs_read_tree_root(tree_root, &location);
2504         if (IS_ERR(root))
2505                 return PTR_ERR(root);
2506         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2507         fs_info->extent_root = root;
2508
2509         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2510         root = btrfs_read_tree_root(tree_root, &location);
2511         if (IS_ERR(root))
2512                 return PTR_ERR(root);
2513         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2514         fs_info->dev_root = root;
2515         btrfs_init_devices_late(fs_info);
2516
2517         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2518         root = btrfs_read_tree_root(tree_root, &location);
2519         if (IS_ERR(root))
2520                 return PTR_ERR(root);
2521         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2522         fs_info->csum_root = root;
2523
2524         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2525         root = btrfs_read_tree_root(tree_root, &location);
2526         if (!IS_ERR(root)) {
2527                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2528                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2529                 fs_info->quota_root = root;
2530         }
2531
2532         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2533         root = btrfs_read_tree_root(tree_root, &location);
2534         if (IS_ERR(root)) {
2535                 ret = PTR_ERR(root);
2536                 if (ret != -ENOENT)
2537                         return ret;
2538         } else {
2539                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2540                 fs_info->uuid_root = root;
2541         }
2542
2543         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2544                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2545                 root = btrfs_read_tree_root(tree_root, &location);
2546                 if (IS_ERR(root))
2547                         return PTR_ERR(root);
2548                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2549                 fs_info->free_space_root = root;
2550         }
2551
2552         return 0;
2553 }
2554
2555 int open_ctree(struct super_block *sb,
2556                struct btrfs_fs_devices *fs_devices,
2557                char *options)
2558 {
2559         u32 sectorsize;
2560         u32 nodesize;
2561         u32 stripesize;
2562         u64 generation;
2563         u64 features;
2564         struct btrfs_key location;
2565         struct buffer_head *bh;
2566         struct btrfs_super_block *disk_super;
2567         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2568         struct btrfs_root *tree_root;
2569         struct btrfs_root *chunk_root;
2570         int ret;
2571         int err = -EINVAL;
2572         int num_backups_tried = 0;
2573         int backup_index = 0;
2574         int max_active;
2575         int clear_free_space_tree = 0;
2576
2577         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2578         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2579         if (!tree_root || !chunk_root) {
2580                 err = -ENOMEM;
2581                 goto fail;
2582         }
2583
2584         ret = init_srcu_struct(&fs_info->subvol_srcu);
2585         if (ret) {
2586                 err = ret;
2587                 goto fail;
2588         }
2589
2590         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2591         if (ret) {
2592                 err = ret;
2593                 goto fail_srcu;
2594         }
2595         fs_info->dirty_metadata_batch = PAGE_SIZE *
2596                                         (1 + ilog2(nr_cpu_ids));
2597
2598         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2599         if (ret) {
2600                 err = ret;
2601                 goto fail_dirty_metadata_bytes;
2602         }
2603
2604         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2605         if (ret) {
2606                 err = ret;
2607                 goto fail_delalloc_bytes;
2608         }
2609
2610         fs_info->btree_inode = new_inode(sb);
2611         if (!fs_info->btree_inode) {
2612                 err = -ENOMEM;
2613                 goto fail_bio_counter;
2614         }
2615
2616         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2617
2618         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2619         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2620         INIT_LIST_HEAD(&fs_info->trans_list);
2621         INIT_LIST_HEAD(&fs_info->dead_roots);
2622         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2623         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2624         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2625         spin_lock_init(&fs_info->delalloc_root_lock);
2626         spin_lock_init(&fs_info->trans_lock);
2627         spin_lock_init(&fs_info->fs_roots_radix_lock);
2628         spin_lock_init(&fs_info->delayed_iput_lock);
2629         spin_lock_init(&fs_info->defrag_inodes_lock);
2630         spin_lock_init(&fs_info->tree_mod_seq_lock);
2631         spin_lock_init(&fs_info->super_lock);
2632         spin_lock_init(&fs_info->qgroup_op_lock);
2633         spin_lock_init(&fs_info->buffer_lock);
2634         spin_lock_init(&fs_info->unused_bgs_lock);
2635         rwlock_init(&fs_info->tree_mod_log_lock);
2636         mutex_init(&fs_info->unused_bg_unpin_mutex);
2637         mutex_init(&fs_info->delete_unused_bgs_mutex);
2638         mutex_init(&fs_info->reloc_mutex);
2639         mutex_init(&fs_info->delalloc_root_mutex);
2640         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2641         seqlock_init(&fs_info->profiles_lock);
2642
2643         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2644         INIT_LIST_HEAD(&fs_info->space_info);
2645         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2646         INIT_LIST_HEAD(&fs_info->unused_bgs);
2647         btrfs_mapping_init(&fs_info->mapping_tree);
2648         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2649                              BTRFS_BLOCK_RSV_GLOBAL);
2650         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2651                              BTRFS_BLOCK_RSV_DELALLOC);
2652         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2653         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2654         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2655         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2656                              BTRFS_BLOCK_RSV_DELOPS);
2657         atomic_set(&fs_info->nr_async_submits, 0);
2658         atomic_set(&fs_info->async_delalloc_pages, 0);
2659         atomic_set(&fs_info->async_submit_draining, 0);
2660         atomic_set(&fs_info->nr_async_bios, 0);
2661         atomic_set(&fs_info->defrag_running, 0);
2662         atomic_set(&fs_info->qgroup_op_seq, 0);
2663         atomic_set(&fs_info->reada_works_cnt, 0);
2664         atomic64_set(&fs_info->tree_mod_seq, 0);
2665         fs_info->sb = sb;
2666         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2667         fs_info->metadata_ratio = 0;
2668         fs_info->defrag_inodes = RB_ROOT;
2669         atomic64_set(&fs_info->free_chunk_space, 0);
2670         fs_info->tree_mod_log = RB_ROOT;
2671         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2672         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2673         /* readahead state */
2674         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2675         spin_lock_init(&fs_info->reada_lock);
2676
2677         fs_info->thread_pool_size = min_t(unsigned long,
2678                                           num_online_cpus() + 2, 8);
2679
2680         INIT_LIST_HEAD(&fs_info->ordered_roots);
2681         spin_lock_init(&fs_info->ordered_root_lock);
2682         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2683                                         GFP_KERNEL);
2684         if (!fs_info->delayed_root) {
2685                 err = -ENOMEM;
2686                 goto fail_iput;
2687         }
2688         btrfs_init_delayed_root(fs_info->delayed_root);
2689
2690         btrfs_init_scrub(fs_info);
2691 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2692         fs_info->check_integrity_print_mask = 0;
2693 #endif
2694         btrfs_init_balance(fs_info);
2695         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2696
2697         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2698         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2699
2700         btrfs_init_btree_inode(fs_info);
2701
2702         spin_lock_init(&fs_info->block_group_cache_lock);
2703         fs_info->block_group_cache_tree = RB_ROOT;
2704         fs_info->first_logical_byte = (u64)-1;
2705
2706         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2707         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2708         fs_info->pinned_extents = &fs_info->freed_extents[0];
2709         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2710
2711         mutex_init(&fs_info->ordered_operations_mutex);
2712         mutex_init(&fs_info->tree_log_mutex);
2713         mutex_init(&fs_info->chunk_mutex);
2714         mutex_init(&fs_info->transaction_kthread_mutex);
2715         mutex_init(&fs_info->cleaner_mutex);
2716         mutex_init(&fs_info->volume_mutex);
2717         mutex_init(&fs_info->ro_block_group_mutex);
2718         init_rwsem(&fs_info->commit_root_sem);
2719         init_rwsem(&fs_info->cleanup_work_sem);
2720         init_rwsem(&fs_info->subvol_sem);
2721         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2722
2723         btrfs_init_dev_replace_locks(fs_info);
2724         btrfs_init_qgroup(fs_info);
2725
2726         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2727         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2728
2729         init_waitqueue_head(&fs_info->transaction_throttle);
2730         init_waitqueue_head(&fs_info->transaction_wait);
2731         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2732         init_waitqueue_head(&fs_info->async_submit_wait);
2733
2734         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2735
2736         /* Usable values until the real ones are cached from the superblock */
2737         fs_info->nodesize = 4096;
2738         fs_info->sectorsize = 4096;
2739         fs_info->stripesize = 4096;
2740
2741         ret = btrfs_alloc_stripe_hash_table(fs_info);
2742         if (ret) {
2743                 err = ret;
2744                 goto fail_alloc;
2745         }
2746
2747         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2748
2749         invalidate_bdev(fs_devices->latest_bdev);
2750
2751         /*
2752          * Read super block and check the signature bytes only
2753          */
2754         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2755         if (IS_ERR(bh)) {
2756                 err = PTR_ERR(bh);
2757                 goto fail_alloc;
2758         }
2759
2760         /*
2761          * We want to check superblock checksum, the type is stored inside.
2762          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2763          */
2764         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2765                 btrfs_err(fs_info, "superblock checksum mismatch");
2766                 err = -EINVAL;
2767                 brelse(bh);
2768                 goto fail_alloc;
2769         }
2770
2771         /*
2772          * super_copy is zeroed at allocation time and we never touch the
2773          * following bytes up to INFO_SIZE, the checksum is calculated from
2774          * the whole block of INFO_SIZE
2775          */
2776         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2777         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2778                sizeof(*fs_info->super_for_commit));
2779         brelse(bh);
2780
2781         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2782
2783         ret = btrfs_check_super_valid(fs_info);
2784         if (ret) {
2785                 btrfs_err(fs_info, "superblock contains fatal errors");
2786                 err = -EINVAL;
2787                 goto fail_alloc;
2788         }
2789
2790         disk_super = fs_info->super_copy;
2791         if (!btrfs_super_root(disk_super))
2792                 goto fail_alloc;
2793
2794         /* check FS state, whether FS is broken. */
2795         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2796                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2797
2798         /*
2799          * run through our array of backup supers and setup
2800          * our ring pointer to the oldest one
2801          */
2802         generation = btrfs_super_generation(disk_super);
2803         find_oldest_super_backup(fs_info, generation);
2804
2805         /*
2806          * In the long term, we'll store the compression type in the super
2807          * block, and it'll be used for per file compression control.
2808          */
2809         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2810
2811         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2812         if (ret) {
2813                 err = ret;
2814                 goto fail_alloc;
2815         }
2816
2817         features = btrfs_super_incompat_flags(disk_super) &
2818                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2819         if (features) {
2820                 btrfs_err(fs_info,
2821                     "cannot mount because of unsupported optional features (%llx)",
2822                     features);
2823                 err = -EINVAL;
2824                 goto fail_alloc;
2825         }
2826
2827         features = btrfs_super_incompat_flags(disk_super);
2828         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2829         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2830                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2831         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2832                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2833
2834         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2835                 btrfs_info(fs_info, "has skinny extents");
2836
2837         /*
2838          * flag our filesystem as having big metadata blocks if
2839          * they are bigger than the page size
2840          */
2841         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2842                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2843                         btrfs_info(fs_info,
2844                                 "flagging fs with big metadata feature");
2845                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2846         }
2847
2848         nodesize = btrfs_super_nodesize(disk_super);
2849         sectorsize = btrfs_super_sectorsize(disk_super);
2850         stripesize = sectorsize;
2851         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2852         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2853
2854         /* Cache block sizes */
2855         fs_info->nodesize = nodesize;
2856         fs_info->sectorsize = sectorsize;
2857         fs_info->stripesize = stripesize;
2858
2859         /*
2860          * mixed block groups end up with duplicate but slightly offset
2861          * extent buffers for the same range.  It leads to corruptions
2862          */
2863         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2864             (sectorsize != nodesize)) {
2865                 btrfs_err(fs_info,
2866 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2867                         nodesize, sectorsize);
2868                 goto fail_alloc;
2869         }
2870
2871         /*
2872          * Needn't use the lock because there is no other task which will
2873          * update the flag.
2874          */
2875         btrfs_set_super_incompat_flags(disk_super, features);
2876
2877         features = btrfs_super_compat_ro_flags(disk_super) &
2878                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2879         if (!(sb->s_flags & MS_RDONLY) && features) {
2880                 btrfs_err(fs_info,
2881         "cannot mount read-write because of unsupported optional features (%llx)",
2882                        features);
2883                 err = -EINVAL;
2884                 goto fail_alloc;
2885         }
2886
2887         max_active = fs_info->thread_pool_size;
2888
2889         ret = btrfs_init_workqueues(fs_info, fs_devices);
2890         if (ret) {
2891                 err = ret;
2892                 goto fail_sb_buffer;
2893         }
2894
2895         sb->s_bdi->congested_fn = btrfs_congested_fn;
2896         sb->s_bdi->congested_data = fs_info;
2897         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2898         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2899         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2900         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2901
2902         sb->s_blocksize = sectorsize;
2903         sb->s_blocksize_bits = blksize_bits(sectorsize);
2904
2905         mutex_lock(&fs_info->chunk_mutex);
2906         ret = btrfs_read_sys_array(fs_info);
2907         mutex_unlock(&fs_info->chunk_mutex);
2908         if (ret) {
2909                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2910                 goto fail_sb_buffer;
2911         }
2912
2913         generation = btrfs_super_chunk_root_generation(disk_super);
2914
2915         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2916
2917         chunk_root->node = read_tree_block(fs_info,
2918                                            btrfs_super_chunk_root(disk_super),
2919                                            generation);
2920         if (IS_ERR(chunk_root->node) ||
2921             !extent_buffer_uptodate(chunk_root->node)) {
2922                 btrfs_err(fs_info, "failed to read chunk root");
2923                 if (!IS_ERR(chunk_root->node))
2924                         free_extent_buffer(chunk_root->node);
2925                 chunk_root->node = NULL;
2926                 goto fail_tree_roots;
2927         }
2928         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2929         chunk_root->commit_root = btrfs_root_node(chunk_root);
2930
2931         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2932            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2933
2934         ret = btrfs_read_chunk_tree(fs_info);
2935         if (ret) {
2936                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2937                 goto fail_tree_roots;
2938         }
2939
2940         /*
2941          * keep the device that is marked to be the target device for the
2942          * dev_replace procedure
2943          */
2944         btrfs_close_extra_devices(fs_devices, 0);
2945
2946         if (!fs_devices->latest_bdev) {
2947                 btrfs_err(fs_info, "failed to read devices");
2948                 goto fail_tree_roots;
2949         }
2950
2951 retry_root_backup:
2952         generation = btrfs_super_generation(disk_super);
2953
2954         tree_root->node = read_tree_block(fs_info,
2955                                           btrfs_super_root(disk_super),
2956                                           generation);
2957         if (IS_ERR(tree_root->node) ||
2958             !extent_buffer_uptodate(tree_root->node)) {
2959                 btrfs_warn(fs_info, "failed to read tree root");
2960                 if (!IS_ERR(tree_root->node))
2961                         free_extent_buffer(tree_root->node);
2962                 tree_root->node = NULL;
2963                 goto recovery_tree_root;
2964         }
2965
2966         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2967         tree_root->commit_root = btrfs_root_node(tree_root);
2968         btrfs_set_root_refs(&tree_root->root_item, 1);
2969
2970         mutex_lock(&tree_root->objectid_mutex);
2971         ret = btrfs_find_highest_objectid(tree_root,
2972                                         &tree_root->highest_objectid);
2973         if (ret) {
2974                 mutex_unlock(&tree_root->objectid_mutex);
2975                 goto recovery_tree_root;
2976         }
2977
2978         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2979
2980         mutex_unlock(&tree_root->objectid_mutex);
2981
2982         ret = btrfs_read_roots(fs_info);
2983         if (ret)
2984                 goto recovery_tree_root;
2985
2986         fs_info->generation = generation;
2987         fs_info->last_trans_committed = generation;
2988
2989         ret = btrfs_recover_balance(fs_info);
2990         if (ret) {
2991                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2992                 goto fail_block_groups;
2993         }
2994
2995         ret = btrfs_init_dev_stats(fs_info);
2996         if (ret) {
2997                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2998                 goto fail_block_groups;
2999         }
3000
3001         ret = btrfs_init_dev_replace(fs_info);
3002         if (ret) {
3003                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3004                 goto fail_block_groups;
3005         }
3006
3007         btrfs_close_extra_devices(fs_devices, 1);
3008
3009         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3010         if (ret) {
3011                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3012                                 ret);
3013                 goto fail_block_groups;
3014         }
3015
3016         ret = btrfs_sysfs_add_device(fs_devices);
3017         if (ret) {
3018                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3019                                 ret);
3020                 goto fail_fsdev_sysfs;
3021         }
3022
3023         ret = btrfs_sysfs_add_mounted(fs_info);
3024         if (ret) {
3025                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3026                 goto fail_fsdev_sysfs;
3027         }
3028
3029         ret = btrfs_init_space_info(fs_info);
3030         if (ret) {
3031                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3032                 goto fail_sysfs;
3033         }
3034
3035         ret = btrfs_read_block_groups(fs_info);
3036         if (ret) {
3037                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3038                 goto fail_sysfs;
3039         }
3040
3041         if (!(sb->s_flags & MS_RDONLY) && !btrfs_check_rw_degradable(fs_info)) {
3042                 btrfs_warn(fs_info,
3043                 "writeable mount is not allowed due to too many missing devices");
3044                 goto fail_sysfs;
3045         }
3046
3047         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3048                                                "btrfs-cleaner");
3049         if (IS_ERR(fs_info->cleaner_kthread))
3050                 goto fail_sysfs;
3051
3052         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3053                                                    tree_root,
3054                                                    "btrfs-transaction");
3055         if (IS_ERR(fs_info->transaction_kthread))
3056                 goto fail_cleaner;
3057
3058         if (!btrfs_test_opt(fs_info, NOSSD) &&
3059             !fs_info->fs_devices->rotating) {
3060                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3061         }
3062
3063         /*
3064          * Mount does not set all options immediately, we can do it now and do
3065          * not have to wait for transaction commit
3066          */
3067         btrfs_apply_pending_changes(fs_info);
3068
3069 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3070         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3071                 ret = btrfsic_mount(fs_info, fs_devices,
3072                                     btrfs_test_opt(fs_info,
3073                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3074                                     1 : 0,
3075                                     fs_info->check_integrity_print_mask);
3076                 if (ret)
3077                         btrfs_warn(fs_info,
3078                                 "failed to initialize integrity check module: %d",
3079                                 ret);
3080         }
3081 #endif
3082         ret = btrfs_read_qgroup_config(fs_info);
3083         if (ret)
3084                 goto fail_trans_kthread;
3085
3086         /* do not make disk changes in broken FS or nologreplay is given */
3087         if (btrfs_super_log_root(disk_super) != 0 &&
3088             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3089                 ret = btrfs_replay_log(fs_info, fs_devices);
3090                 if (ret) {
3091                         err = ret;
3092                         goto fail_qgroup;
3093                 }
3094         }
3095
3096         ret = btrfs_find_orphan_roots(fs_info);
3097         if (ret)
3098                 goto fail_qgroup;
3099
3100         if (!(sb->s_flags & MS_RDONLY)) {
3101                 ret = btrfs_cleanup_fs_roots(fs_info);
3102                 if (ret)
3103                         goto fail_qgroup;
3104
3105                 mutex_lock(&fs_info->cleaner_mutex);
3106                 ret = btrfs_recover_relocation(tree_root);
3107                 mutex_unlock(&fs_info->cleaner_mutex);
3108                 if (ret < 0) {
3109                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3110                                         ret);
3111                         err = -EINVAL;
3112                         goto fail_qgroup;
3113                 }
3114         }
3115
3116         location.objectid = BTRFS_FS_TREE_OBJECTID;
3117         location.type = BTRFS_ROOT_ITEM_KEY;
3118         location.offset = 0;
3119
3120         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3121         if (IS_ERR(fs_info->fs_root)) {
3122                 err = PTR_ERR(fs_info->fs_root);
3123                 goto fail_qgroup;
3124         }
3125
3126         if (sb->s_flags & MS_RDONLY)
3127                 return 0;
3128
3129         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3130             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3131                 clear_free_space_tree = 1;
3132         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3133                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3134                 btrfs_warn(fs_info, "free space tree is invalid");
3135                 clear_free_space_tree = 1;
3136         }
3137
3138         if (clear_free_space_tree) {
3139                 btrfs_info(fs_info, "clearing free space tree");
3140                 ret = btrfs_clear_free_space_tree(fs_info);
3141                 if (ret) {
3142                         btrfs_warn(fs_info,
3143                                    "failed to clear free space tree: %d", ret);
3144                         close_ctree(fs_info);
3145                         return ret;
3146                 }
3147         }
3148
3149         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3150             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3151                 btrfs_info(fs_info, "creating free space tree");
3152                 ret = btrfs_create_free_space_tree(fs_info);
3153                 if (ret) {
3154                         btrfs_warn(fs_info,
3155                                 "failed to create free space tree: %d", ret);
3156                         close_ctree(fs_info);
3157                         return ret;
3158                 }
3159         }
3160
3161         down_read(&fs_info->cleanup_work_sem);
3162         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3163             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3164                 up_read(&fs_info->cleanup_work_sem);
3165                 close_ctree(fs_info);
3166                 return ret;
3167         }
3168         up_read(&fs_info->cleanup_work_sem);
3169
3170         ret = btrfs_resume_balance_async(fs_info);
3171         if (ret) {
3172                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3173                 close_ctree(fs_info);
3174                 return ret;
3175         }
3176
3177         ret = btrfs_resume_dev_replace_async(fs_info);
3178         if (ret) {
3179                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3180                 close_ctree(fs_info);
3181                 return ret;
3182         }
3183
3184         btrfs_qgroup_rescan_resume(fs_info);
3185
3186         if (!fs_info->uuid_root) {
3187                 btrfs_info(fs_info, "creating UUID tree");
3188                 ret = btrfs_create_uuid_tree(fs_info);
3189                 if (ret) {
3190                         btrfs_warn(fs_info,
3191                                 "failed to create the UUID tree: %d", ret);
3192                         close_ctree(fs_info);
3193                         return ret;
3194                 }
3195         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3196                    fs_info->generation !=
3197                                 btrfs_super_uuid_tree_generation(disk_super)) {
3198                 btrfs_info(fs_info, "checking UUID tree");
3199                 ret = btrfs_check_uuid_tree(fs_info);
3200                 if (ret) {
3201                         btrfs_warn(fs_info,
3202                                 "failed to check the UUID tree: %d", ret);
3203                         close_ctree(fs_info);
3204                         return ret;
3205                 }
3206         } else {
3207                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3208         }
3209         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3210
3211         /*
3212          * backuproot only affect mount behavior, and if open_ctree succeeded,
3213          * no need to keep the flag
3214          */
3215         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3216
3217         return 0;
3218
3219 fail_qgroup:
3220         btrfs_free_qgroup_config(fs_info);
3221 fail_trans_kthread:
3222         kthread_stop(fs_info->transaction_kthread);
3223         btrfs_cleanup_transaction(fs_info);
3224         btrfs_free_fs_roots(fs_info);
3225 fail_cleaner:
3226         kthread_stop(fs_info->cleaner_kthread);
3227
3228         /*
3229          * make sure we're done with the btree inode before we stop our
3230          * kthreads
3231          */
3232         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3233
3234 fail_sysfs:
3235         btrfs_sysfs_remove_mounted(fs_info);
3236
3237 fail_fsdev_sysfs:
3238         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3239
3240 fail_block_groups:
3241         btrfs_put_block_group_cache(fs_info);
3242
3243 fail_tree_roots:
3244         free_root_pointers(fs_info, 1);
3245         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3246
3247 fail_sb_buffer:
3248         btrfs_stop_all_workers(fs_info);
3249         btrfs_free_block_groups(fs_info);
3250 fail_alloc:
3251 fail_iput:
3252         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3253
3254         iput(fs_info->btree_inode);
3255 fail_bio_counter:
3256         percpu_counter_destroy(&fs_info->bio_counter);
3257 fail_delalloc_bytes:
3258         percpu_counter_destroy(&fs_info->delalloc_bytes);
3259 fail_dirty_metadata_bytes:
3260         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3261 fail_srcu:
3262         cleanup_srcu_struct(&fs_info->subvol_srcu);
3263 fail:
3264         btrfs_free_stripe_hash_table(fs_info);
3265         btrfs_close_devices(fs_info->fs_devices);
3266         return err;
3267
3268 recovery_tree_root:
3269         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3270                 goto fail_tree_roots;
3271
3272         free_root_pointers(fs_info, 0);
3273
3274         /* don't use the log in recovery mode, it won't be valid */
3275         btrfs_set_super_log_root(disk_super, 0);
3276
3277         /* we can't trust the free space cache either */
3278         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3279
3280         ret = next_root_backup(fs_info, fs_info->super_copy,
3281                                &num_backups_tried, &backup_index);
3282         if (ret == -1)
3283                 goto fail_block_groups;
3284         goto retry_root_backup;
3285 }
3286
3287 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3288 {
3289         if (uptodate) {
3290                 set_buffer_uptodate(bh);
3291         } else {
3292                 struct btrfs_device *device = (struct btrfs_device *)
3293                         bh->b_private;
3294
3295                 btrfs_warn_rl_in_rcu(device->fs_info,
3296                                 "lost page write due to IO error on %s",
3297                                           rcu_str_deref(device->name));
3298                 /* note, we don't set_buffer_write_io_error because we have
3299                  * our own ways of dealing with the IO errors
3300                  */
3301                 clear_buffer_uptodate(bh);
3302                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3303         }
3304         unlock_buffer(bh);
3305         put_bh(bh);
3306 }
3307
3308 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3309                         struct buffer_head **bh_ret)
3310 {
3311         struct buffer_head *bh;
3312         struct btrfs_super_block *super;
3313         u64 bytenr;
3314
3315         bytenr = btrfs_sb_offset(copy_num);
3316         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3317                 return -EINVAL;
3318
3319         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3320         /*
3321          * If we fail to read from the underlying devices, as of now
3322          * the best option we have is to mark it EIO.
3323          */
3324         if (!bh)
3325                 return -EIO;
3326
3327         super = (struct btrfs_super_block *)bh->b_data;
3328         if (btrfs_super_bytenr(super) != bytenr ||
3329                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3330                 brelse(bh);
3331                 return -EINVAL;
3332         }
3333
3334         *bh_ret = bh;
3335         return 0;
3336 }
3337
3338
3339 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3340 {
3341         struct buffer_head *bh;
3342         struct buffer_head *latest = NULL;
3343         struct btrfs_super_block *super;
3344         int i;
3345         u64 transid = 0;
3346         int ret = -EINVAL;
3347
3348         /* we would like to check all the supers, but that would make
3349          * a btrfs mount succeed after a mkfs from a different FS.
3350          * So, we need to add a special mount option to scan for
3351          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3352          */
3353         for (i = 0; i < 1; i++) {
3354                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3355                 if (ret)
3356                         continue;
3357
3358                 super = (struct btrfs_super_block *)bh->b_data;
3359
3360                 if (!latest || btrfs_super_generation(super) > transid) {
3361                         brelse(latest);
3362                         latest = bh;
3363                         transid = btrfs_super_generation(super);
3364                 } else {
3365                         brelse(bh);
3366                 }
3367         }
3368
3369         if (!latest)
3370                 return ERR_PTR(ret);
3371
3372         return latest;
3373 }
3374
3375 /*
3376  * Write superblock @sb to the @device. Do not wait for completion, all the
3377  * buffer heads we write are pinned.
3378  *
3379  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3380  * the expected device size at commit time. Note that max_mirrors must be
3381  * same for write and wait phases.
3382  *
3383  * Return number of errors when buffer head is not found or submission fails.
3384  */
3385 static int write_dev_supers(struct btrfs_device *device,
3386                             struct btrfs_super_block *sb, int max_mirrors)
3387 {
3388         struct buffer_head *bh;
3389         int i;
3390         int ret;
3391         int errors = 0;
3392         u32 crc;
3393         u64 bytenr;
3394
3395         if (max_mirrors == 0)
3396                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3397
3398         for (i = 0; i < max_mirrors; i++) {
3399                 bytenr = btrfs_sb_offset(i);
3400                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3401                     device->commit_total_bytes)
3402                         break;
3403
3404                 btrfs_set_super_bytenr(sb, bytenr);
3405
3406                 crc = ~(u32)0;
3407                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3408                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3409                 btrfs_csum_final(crc, sb->csum);
3410
3411                 /* One reference for us, and we leave it for the caller */
3412                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3413                               BTRFS_SUPER_INFO_SIZE);
3414                 if (!bh) {
3415                         btrfs_err(device->fs_info,
3416                             "couldn't get super buffer head for bytenr %llu",
3417                             bytenr);
3418                         errors++;
3419                         continue;
3420                 }
3421
3422                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3423
3424                 /* one reference for submit_bh */
3425                 get_bh(bh);
3426
3427                 set_buffer_uptodate(bh);
3428                 lock_buffer(bh);
3429                 bh->b_end_io = btrfs_end_buffer_write_sync;
3430                 bh->b_private = device;
3431
3432                 /*
3433                  * we fua the first super.  The others we allow
3434                  * to go down lazy.
3435                  */
3436                 if (i == 0) {
3437                         ret = btrfsic_submit_bh(REQ_OP_WRITE,
3438                                 REQ_SYNC | REQ_FUA | REQ_META | REQ_PRIO, bh);
3439                 } else {
3440                         ret = btrfsic_submit_bh(REQ_OP_WRITE,
3441                                 REQ_SYNC | REQ_META | REQ_PRIO, bh);
3442                 }
3443                 if (ret)
3444                         errors++;
3445         }
3446         return errors < i ? 0 : -1;
3447 }
3448
3449 /*
3450  * Wait for write completion of superblocks done by write_dev_supers,
3451  * @max_mirrors same for write and wait phases.
3452  *
3453  * Return number of errors when buffer head is not found or not marked up to
3454  * date.
3455  */
3456 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3457 {
3458         struct buffer_head *bh;
3459         int i;
3460         int errors = 0;
3461         u64 bytenr;
3462
3463         if (max_mirrors == 0)
3464                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3465
3466         for (i = 0; i < max_mirrors; i++) {
3467                 bytenr = btrfs_sb_offset(i);
3468                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3469                     device->commit_total_bytes)
3470                         break;
3471
3472                 bh = __find_get_block(device->bdev,
3473                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3474                                       BTRFS_SUPER_INFO_SIZE);
3475                 if (!bh) {
3476                         errors++;
3477                         continue;
3478                 }
3479                 wait_on_buffer(bh);
3480                 if (!buffer_uptodate(bh))
3481                         errors++;
3482
3483                 /* drop our reference */
3484                 brelse(bh);
3485
3486                 /* drop the reference from the writing run */
3487                 brelse(bh);
3488         }
3489
3490         return errors < i ? 0 : -1;
3491 }
3492
3493 /*
3494  * endio for the write_dev_flush, this will wake anyone waiting
3495  * for the barrier when it is done
3496  */
3497 static void btrfs_end_empty_barrier(struct bio *bio)
3498 {
3499         complete(bio->bi_private);
3500 }
3501
3502 /*
3503  * Submit a flush request to the device if it supports it. Error handling is
3504  * done in the waiting counterpart.
3505  */
3506 static void write_dev_flush(struct btrfs_device *device)
3507 {
3508         struct request_queue *q = bdev_get_queue(device->bdev);
3509         struct bio *bio = device->flush_bio;
3510
3511         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3512                 return;
3513
3514         bio_reset(bio);
3515         bio->bi_end_io = btrfs_end_empty_barrier;
3516         bio_set_dev(bio, device->bdev);
3517         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3518         init_completion(&device->flush_wait);
3519         bio->bi_private = &device->flush_wait;
3520
3521         btrfsic_submit_bio(bio);
3522         device->flush_bio_sent = 1;
3523 }
3524
3525 /*
3526  * If the flush bio has been submitted by write_dev_flush, wait for it.
3527  */
3528 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3529 {
3530         struct bio *bio = device->flush_bio;
3531
3532         if (!device->flush_bio_sent)
3533                 return BLK_STS_OK;
3534
3535         device->flush_bio_sent = 0;
3536         wait_for_completion_io(&device->flush_wait);
3537
3538         return bio->bi_status;
3539 }
3540
3541 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3542 {
3543         if (!btrfs_check_rw_degradable(fs_info))
3544                 return -EIO;
3545         return 0;
3546 }
3547
3548 /*
3549  * send an empty flush down to each device in parallel,
3550  * then wait for them
3551  */
3552 static int barrier_all_devices(struct btrfs_fs_info *info)
3553 {
3554         struct list_head *head;
3555         struct btrfs_device *dev;
3556         int errors_wait = 0;
3557         blk_status_t ret;
3558
3559         /* send down all the barriers */
3560         head = &info->fs_devices->devices;
3561         list_for_each_entry_rcu(dev, head, dev_list) {
3562                 if (dev->missing)
3563                         continue;
3564                 if (!dev->bdev)
3565                         continue;
3566                 if (!dev->in_fs_metadata || !dev->writeable)
3567                         continue;
3568
3569                 write_dev_flush(dev);
3570                 dev->last_flush_error = BLK_STS_OK;
3571         }
3572
3573         /* wait for all the barriers */
3574         list_for_each_entry_rcu(dev, head, dev_list) {
3575                 if (dev->missing)
3576                         continue;
3577                 if (!dev->bdev) {
3578                         errors_wait++;
3579                         continue;
3580                 }
3581                 if (!dev->in_fs_metadata || !dev->writeable)
3582                         continue;
3583
3584                 ret = wait_dev_flush(dev);
3585                 if (ret) {
3586                         dev->last_flush_error = ret;
3587                         btrfs_dev_stat_inc_and_print(dev,
3588                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3589                         errors_wait++;
3590                 }
3591         }
3592
3593         if (errors_wait) {
3594                 /*
3595                  * At some point we need the status of all disks
3596                  * to arrive at the volume status. So error checking
3597                  * is being pushed to a separate loop.
3598                  */
3599                 return check_barrier_error(info);
3600         }
3601         return 0;
3602 }
3603
3604 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3605 {
3606         int raid_type;
3607         int min_tolerated = INT_MAX;
3608
3609         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3610             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3611                 min_tolerated = min(min_tolerated,
3612                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3613                                     tolerated_failures);
3614
3615         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3616                 if (raid_type == BTRFS_RAID_SINGLE)
3617                         continue;
3618                 if (!(flags & btrfs_raid_group[raid_type]))
3619                         continue;
3620                 min_tolerated = min(min_tolerated,
3621                                     btrfs_raid_array[raid_type].
3622                                     tolerated_failures);
3623         }
3624
3625         if (min_tolerated == INT_MAX) {
3626                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3627                 min_tolerated = 0;
3628         }
3629
3630         return min_tolerated;
3631 }
3632
3633 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3634 {
3635         struct list_head *head;
3636         struct btrfs_device *dev;
3637         struct btrfs_super_block *sb;
3638         struct btrfs_dev_item *dev_item;
3639         int ret;
3640         int do_barriers;
3641         int max_errors;
3642         int total_errors = 0;
3643         u64 flags;
3644
3645         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3646         backup_super_roots(fs_info);
3647
3648         sb = fs_info->super_for_commit;
3649         dev_item = &sb->dev_item;
3650
3651         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3652         head = &fs_info->fs_devices->devices;
3653         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3654
3655         if (do_barriers) {
3656                 ret = barrier_all_devices(fs_info);
3657                 if (ret) {
3658                         mutex_unlock(
3659                                 &fs_info->fs_devices->device_list_mutex);
3660                         btrfs_handle_fs_error(fs_info, ret,
3661                                               "errors while submitting device barriers.");
3662                         return ret;
3663                 }
3664         }
3665
3666         list_for_each_entry_rcu(dev, head, dev_list) {
3667                 if (!dev->bdev) {
3668                         total_errors++;
3669                         continue;
3670                 }
3671                 if (!dev->in_fs_metadata || !dev->writeable)
3672                         continue;
3673
3674                 btrfs_set_stack_device_generation(dev_item, 0);
3675                 btrfs_set_stack_device_type(dev_item, dev->type);
3676                 btrfs_set_stack_device_id(dev_item, dev->devid);
3677                 btrfs_set_stack_device_total_bytes(dev_item,
3678                                                    dev->commit_total_bytes);
3679                 btrfs_set_stack_device_bytes_used(dev_item,
3680                                                   dev->commit_bytes_used);
3681                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3682                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3683                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3684                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3685                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3686
3687                 flags = btrfs_super_flags(sb);
3688                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3689
3690                 ret = write_dev_supers(dev, sb, max_mirrors);
3691                 if (ret)
3692                         total_errors++;
3693         }
3694         if (total_errors > max_errors) {
3695                 btrfs_err(fs_info, "%d errors while writing supers",
3696                           total_errors);
3697                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3698
3699                 /* FUA is masked off if unsupported and can't be the reason */
3700                 btrfs_handle_fs_error(fs_info, -EIO,
3701                                       "%d errors while writing supers",
3702                                       total_errors);
3703                 return -EIO;
3704         }
3705
3706         total_errors = 0;
3707         list_for_each_entry_rcu(dev, head, dev_list) {
3708                 if (!dev->bdev)
3709                         continue;
3710                 if (!dev->in_fs_metadata || !dev->writeable)
3711                         continue;
3712
3713                 ret = wait_dev_supers(dev, max_mirrors);
3714                 if (ret)
3715                         total_errors++;
3716         }
3717         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3718         if (total_errors > max_errors) {
3719                 btrfs_handle_fs_error(fs_info, -EIO,
3720                                       "%d errors while writing supers",
3721                                       total_errors);
3722                 return -EIO;
3723         }
3724         return 0;
3725 }
3726
3727 /* Drop a fs root from the radix tree and free it. */
3728 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3729                                   struct btrfs_root *root)
3730 {
3731         spin_lock(&fs_info->fs_roots_radix_lock);
3732         radix_tree_delete(&fs_info->fs_roots_radix,
3733                           (unsigned long)root->root_key.objectid);
3734         spin_unlock(&fs_info->fs_roots_radix_lock);
3735
3736         if (btrfs_root_refs(&root->root_item) == 0)
3737                 synchronize_srcu(&fs_info->subvol_srcu);
3738
3739         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3740                 btrfs_free_log(NULL, root);
3741                 if (root->reloc_root) {
3742                         free_extent_buffer(root->reloc_root->node);
3743                         free_extent_buffer(root->reloc_root->commit_root);
3744                         btrfs_put_fs_root(root->reloc_root);
3745                         root->reloc_root = NULL;
3746                 }
3747         }
3748
3749         if (root->free_ino_pinned)
3750                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3751         if (root->free_ino_ctl)
3752                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3753         free_fs_root(root);
3754 }
3755
3756 static void free_fs_root(struct btrfs_root *root)
3757 {
3758         iput(root->ino_cache_inode);
3759         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3760         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3761         root->orphan_block_rsv = NULL;
3762         if (root->anon_dev)
3763                 free_anon_bdev(root->anon_dev);
3764         if (root->subv_writers)
3765                 btrfs_free_subvolume_writers(root->subv_writers);
3766         free_extent_buffer(root->node);
3767         free_extent_buffer(root->commit_root);
3768         kfree(root->free_ino_ctl);
3769         kfree(root->free_ino_pinned);
3770         kfree(root->name);
3771         btrfs_put_fs_root(root);
3772 }
3773
3774 void btrfs_free_fs_root(struct btrfs_root *root)
3775 {
3776         free_fs_root(root);
3777 }
3778
3779 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3780 {
3781         u64 root_objectid = 0;
3782         struct btrfs_root *gang[8];
3783         int i = 0;
3784         int err = 0;
3785         unsigned int ret = 0;
3786         int index;
3787
3788         while (1) {
3789                 index = srcu_read_lock(&fs_info->subvol_srcu);
3790                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3791                                              (void **)gang, root_objectid,
3792                                              ARRAY_SIZE(gang));
3793                 if (!ret) {
3794                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3795                         break;
3796                 }
3797                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3798
3799                 for (i = 0; i < ret; i++) {
3800                         /* Avoid to grab roots in dead_roots */
3801                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3802                                 gang[i] = NULL;
3803                                 continue;
3804                         }
3805                         /* grab all the search result for later use */
3806                         gang[i] = btrfs_grab_fs_root(gang[i]);
3807                 }
3808                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3809
3810                 for (i = 0; i < ret; i++) {
3811                         if (!gang[i])
3812                                 continue;
3813                         root_objectid = gang[i]->root_key.objectid;
3814                         err = btrfs_orphan_cleanup(gang[i]);
3815                         if (err)
3816                                 break;
3817                         btrfs_put_fs_root(gang[i]);
3818                 }
3819                 root_objectid++;
3820         }
3821
3822         /* release the uncleaned roots due to error */
3823         for (; i < ret; i++) {
3824                 if (gang[i])
3825                         btrfs_put_fs_root(gang[i]);
3826         }
3827         return err;
3828 }
3829
3830 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3831 {
3832         struct btrfs_root *root = fs_info->tree_root;
3833         struct btrfs_trans_handle *trans;
3834
3835         mutex_lock(&fs_info->cleaner_mutex);
3836         btrfs_run_delayed_iputs(fs_info);
3837         mutex_unlock(&fs_info->cleaner_mutex);
3838         wake_up_process(fs_info->cleaner_kthread);
3839
3840         /* wait until ongoing cleanup work done */
3841         down_write(&fs_info->cleanup_work_sem);
3842         up_write(&fs_info->cleanup_work_sem);
3843
3844         trans = btrfs_join_transaction(root);
3845         if (IS_ERR(trans))
3846                 return PTR_ERR(trans);
3847         return btrfs_commit_transaction(trans);
3848 }
3849
3850 void close_ctree(struct btrfs_fs_info *fs_info)
3851 {
3852         struct btrfs_root *root = fs_info->tree_root;
3853         int ret;
3854
3855         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3856
3857         /* wait for the qgroup rescan worker to stop */
3858         btrfs_qgroup_wait_for_completion(fs_info, false);
3859
3860         /* wait for the uuid_scan task to finish */
3861         down(&fs_info->uuid_tree_rescan_sem);
3862         /* avoid complains from lockdep et al., set sem back to initial state */
3863         up(&fs_info->uuid_tree_rescan_sem);
3864
3865         /* pause restriper - we want to resume on mount */
3866         btrfs_pause_balance(fs_info);
3867
3868         btrfs_dev_replace_suspend_for_unmount(fs_info);
3869
3870         btrfs_scrub_cancel(fs_info);
3871
3872         /* wait for any defraggers to finish */
3873         wait_event(fs_info->transaction_wait,
3874                    (atomic_read(&fs_info->defrag_running) == 0));
3875
3876         /* clear out the rbtree of defraggable inodes */
3877         btrfs_cleanup_defrag_inodes(fs_info);
3878
3879         cancel_work_sync(&fs_info->async_reclaim_work);
3880
3881         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3882                 /*
3883                  * If the cleaner thread is stopped and there are
3884                  * block groups queued for removal, the deletion will be
3885                  * skipped when we quit the cleaner thread.
3886                  */
3887                 btrfs_delete_unused_bgs(fs_info);
3888
3889                 ret = btrfs_commit_super(fs_info);
3890                 if (ret)
3891                         btrfs_err(fs_info, "commit super ret %d", ret);
3892         }
3893
3894         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3895                 btrfs_error_commit_super(fs_info);
3896
3897         kthread_stop(fs_info->transaction_kthread);
3898         kthread_stop(fs_info->cleaner_kthread);
3899
3900         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3901
3902         btrfs_free_qgroup_config(fs_info);
3903
3904         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3905                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3906                        percpu_counter_sum(&fs_info->delalloc_bytes));
3907         }
3908
3909         btrfs_sysfs_remove_mounted(fs_info);
3910         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3911
3912         btrfs_free_fs_roots(fs_info);
3913
3914         btrfs_put_block_group_cache(fs_info);
3915
3916         /*
3917          * we must make sure there is not any read request to
3918          * submit after we stopping all workers.
3919          */
3920         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3921         btrfs_stop_all_workers(fs_info);
3922
3923         btrfs_free_block_groups(fs_info);
3924
3925         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3926         free_root_pointers(fs_info, 1);
3927
3928         iput(fs_info->btree_inode);
3929
3930 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3931         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3932                 btrfsic_unmount(fs_info->fs_devices);
3933 #endif
3934
3935         btrfs_close_devices(fs_info->fs_devices);
3936         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3937
3938         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3939         percpu_counter_destroy(&fs_info->delalloc_bytes);
3940         percpu_counter_destroy(&fs_info->bio_counter);
3941         cleanup_srcu_struct(&fs_info->subvol_srcu);
3942
3943         btrfs_free_stripe_hash_table(fs_info);
3944
3945         __btrfs_free_block_rsv(root->orphan_block_rsv);
3946         root->orphan_block_rsv = NULL;
3947
3948         while (!list_empty(&fs_info->pinned_chunks)) {
3949                 struct extent_map *em;
3950
3951                 em = list_first_entry(&fs_info->pinned_chunks,
3952                                       struct extent_map, list);
3953                 list_del_init(&em->list);
3954                 free_extent_map(em);
3955         }
3956 }
3957
3958 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3959                           int atomic)
3960 {
3961         int ret;
3962         struct inode *btree_inode = buf->pages[0]->mapping->host;
3963
3964         ret = extent_buffer_uptodate(buf);
3965         if (!ret)
3966                 return ret;
3967
3968         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3969                                     parent_transid, atomic);
3970         if (ret == -EAGAIN)
3971                 return ret;
3972         return !ret;
3973 }
3974
3975 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3976 {
3977         struct btrfs_fs_info *fs_info;
3978         struct btrfs_root *root;
3979         u64 transid = btrfs_header_generation(buf);
3980         int was_dirty;
3981
3982 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3983         /*
3984          * This is a fast path so only do this check if we have sanity tests
3985          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3986          * outside of the sanity tests.
3987          */
3988         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3989                 return;
3990 #endif
3991         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3992         fs_info = root->fs_info;
3993         btrfs_assert_tree_locked(buf);
3994         if (transid != fs_info->generation)
3995                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
3996                         buf->start, transid, fs_info->generation);
3997         was_dirty = set_extent_buffer_dirty(buf);
3998         if (!was_dirty)
3999                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4000                                          buf->len,
4001                                          fs_info->dirty_metadata_batch);
4002 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4003         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4004                 btrfs_print_leaf(buf);
4005                 ASSERT(0);
4006         }
4007 #endif
4008 }
4009
4010 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4011                                         int flush_delayed)
4012 {
4013         /*
4014          * looks as though older kernels can get into trouble with
4015          * this code, they end up stuck in balance_dirty_pages forever
4016          */
4017         int ret;
4018
4019         if (current->flags & PF_MEMALLOC)
4020                 return;
4021
4022         if (flush_delayed)
4023                 btrfs_balance_delayed_items(fs_info);
4024
4025         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4026                                      BTRFS_DIRTY_METADATA_THRESH);
4027         if (ret > 0) {
4028                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4029         }
4030 }
4031
4032 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4033 {
4034         __btrfs_btree_balance_dirty(fs_info, 1);
4035 }
4036
4037 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4038 {
4039         __btrfs_btree_balance_dirty(fs_info, 0);
4040 }
4041
4042 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4043 {
4044         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4045         struct btrfs_fs_info *fs_info = root->fs_info;
4046
4047         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4048 }
4049
4050 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
4051 {
4052         struct btrfs_super_block *sb = fs_info->super_copy;
4053         u64 nodesize = btrfs_super_nodesize(sb);
4054         u64 sectorsize = btrfs_super_sectorsize(sb);
4055         int ret = 0;
4056
4057         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4058                 btrfs_err(fs_info, "no valid FS found");
4059                 ret = -EINVAL;
4060         }
4061         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4062                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4063                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4064         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4065                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4066                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4067                 ret = -EINVAL;
4068         }
4069         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4070                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4071                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4072                 ret = -EINVAL;
4073         }
4074         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4075                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4076                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4077                 ret = -EINVAL;
4078         }
4079
4080         /*
4081          * Check sectorsize and nodesize first, other check will need it.
4082          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4083          */
4084         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4085             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4086                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4087                 ret = -EINVAL;
4088         }
4089         /* Only PAGE SIZE is supported yet */
4090         if (sectorsize != PAGE_SIZE) {
4091                 btrfs_err(fs_info,
4092                         "sectorsize %llu not supported yet, only support %lu",
4093                         sectorsize, PAGE_SIZE);
4094                 ret = -EINVAL;
4095         }
4096         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4097             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4098                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4099                 ret = -EINVAL;
4100         }
4101         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4102                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4103                           le32_to_cpu(sb->__unused_leafsize), nodesize);
4104                 ret = -EINVAL;
4105         }
4106
4107         /* Root alignment check */
4108         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4109                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4110                            btrfs_super_root(sb));
4111                 ret = -EINVAL;
4112         }
4113         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4114                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4115                            btrfs_super_chunk_root(sb));
4116                 ret = -EINVAL;
4117         }
4118         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4119                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4120                            btrfs_super_log_root(sb));
4121                 ret = -EINVAL;
4122         }
4123
4124         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_FSID_SIZE) != 0) {
4125                 btrfs_err(fs_info,
4126                            "dev_item UUID does not match fsid: %pU != %pU",
4127                            fs_info->fsid, sb->dev_item.fsid);
4128                 ret = -EINVAL;
4129         }
4130
4131         /*
4132          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4133          * done later
4134          */
4135         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4136                 btrfs_err(fs_info, "bytes_used is too small %llu",
4137                           btrfs_super_bytes_used(sb));
4138                 ret = -EINVAL;
4139         }
4140         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4141                 btrfs_err(fs_info, "invalid stripesize %u",
4142                           btrfs_super_stripesize(sb));
4143                 ret = -EINVAL;
4144         }
4145         if (btrfs_super_num_devices(sb) > (1UL << 31))
4146                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4147                            btrfs_super_num_devices(sb));
4148         if (btrfs_super_num_devices(sb) == 0) {
4149                 btrfs_err(fs_info, "number of devices is 0");
4150                 ret = -EINVAL;
4151         }
4152
4153         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4154                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4155                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4156                 ret = -EINVAL;
4157         }
4158
4159         /*
4160          * Obvious sys_chunk_array corruptions, it must hold at least one key
4161          * and one chunk
4162          */
4163         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4164                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4165                           btrfs_super_sys_array_size(sb),
4166                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4167                 ret = -EINVAL;
4168         }
4169         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4170                         + sizeof(struct btrfs_chunk)) {
4171                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4172                           btrfs_super_sys_array_size(sb),
4173                           sizeof(struct btrfs_disk_key)
4174                           + sizeof(struct btrfs_chunk));
4175                 ret = -EINVAL;
4176         }
4177
4178         /*
4179          * The generation is a global counter, we'll trust it more than the others
4180          * but it's still possible that it's the one that's wrong.
4181          */
4182         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4183                 btrfs_warn(fs_info,
4184                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4185                         btrfs_super_generation(sb),
4186                         btrfs_super_chunk_root_generation(sb));
4187         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4188             && btrfs_super_cache_generation(sb) != (u64)-1)
4189                 btrfs_warn(fs_info,
4190                         "suspicious: generation < cache_generation: %llu < %llu",
4191                         btrfs_super_generation(sb),
4192                         btrfs_super_cache_generation(sb));
4193
4194         return ret;
4195 }
4196
4197 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4198 {
4199         mutex_lock(&fs_info->cleaner_mutex);
4200         btrfs_run_delayed_iputs(fs_info);
4201         mutex_unlock(&fs_info->cleaner_mutex);
4202
4203         down_write(&fs_info->cleanup_work_sem);
4204         up_write(&fs_info->cleanup_work_sem);
4205
4206         /* cleanup FS via transaction */
4207         btrfs_cleanup_transaction(fs_info);
4208 }
4209
4210 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4211 {
4212         struct btrfs_ordered_extent *ordered;
4213
4214         spin_lock(&root->ordered_extent_lock);
4215         /*
4216          * This will just short circuit the ordered completion stuff which will
4217          * make sure the ordered extent gets properly cleaned up.
4218          */
4219         list_for_each_entry(ordered, &root->ordered_extents,
4220                             root_extent_list)
4221                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4222         spin_unlock(&root->ordered_extent_lock);
4223 }
4224
4225 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4226 {
4227         struct btrfs_root *root;
4228         struct list_head splice;
4229
4230         INIT_LIST_HEAD(&splice);
4231
4232         spin_lock(&fs_info->ordered_root_lock);
4233         list_splice_init(&fs_info->ordered_roots, &splice);
4234         while (!list_empty(&splice)) {
4235                 root = list_first_entry(&splice, struct btrfs_root,
4236                                         ordered_root);
4237                 list_move_tail(&root->ordered_root,
4238                                &fs_info->ordered_roots);
4239
4240                 spin_unlock(&fs_info->ordered_root_lock);
4241                 btrfs_destroy_ordered_extents(root);
4242
4243                 cond_resched();
4244                 spin_lock(&fs_info->ordered_root_lock);
4245         }
4246         spin_unlock(&fs_info->ordered_root_lock);
4247 }
4248
4249 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4250                                       struct btrfs_fs_info *fs_info)
4251 {
4252         struct rb_node *node;
4253         struct btrfs_delayed_ref_root *delayed_refs;
4254         struct btrfs_delayed_ref_node *ref;
4255         int ret = 0;
4256
4257         delayed_refs = &trans->delayed_refs;
4258
4259         spin_lock(&delayed_refs->lock);
4260         if (atomic_read(&delayed_refs->num_entries) == 0) {
4261                 spin_unlock(&delayed_refs->lock);
4262                 btrfs_info(fs_info, "delayed_refs has NO entry");
4263                 return ret;
4264         }
4265
4266         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4267                 struct btrfs_delayed_ref_head *head;
4268                 struct btrfs_delayed_ref_node *tmp;
4269                 bool pin_bytes = false;
4270
4271                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4272                                 href_node);
4273                 if (!mutex_trylock(&head->mutex)) {
4274                         refcount_inc(&head->node.refs);
4275                         spin_unlock(&delayed_refs->lock);
4276
4277                         mutex_lock(&head->mutex);
4278                         mutex_unlock(&head->mutex);
4279                         btrfs_put_delayed_ref(&head->node);
4280                         spin_lock(&delayed_refs->lock);
4281                         continue;
4282                 }
4283                 spin_lock(&head->lock);
4284                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4285                                                  list) {
4286                         ref->in_tree = 0;
4287                         list_del(&ref->list);
4288                         if (!list_empty(&ref->add_list))
4289                                 list_del(&ref->add_list);
4290                         atomic_dec(&delayed_refs->num_entries);
4291                         btrfs_put_delayed_ref(ref);
4292                 }
4293                 if (head->must_insert_reserved)
4294                         pin_bytes = true;
4295                 btrfs_free_delayed_extent_op(head->extent_op);
4296                 delayed_refs->num_heads--;
4297                 if (head->processing == 0)
4298                         delayed_refs->num_heads_ready--;
4299                 atomic_dec(&delayed_refs->num_entries);
4300                 head->node.in_tree = 0;
4301                 rb_erase(&head->href_node, &delayed_refs->href_root);
4302                 spin_unlock(&head->lock);
4303                 spin_unlock(&delayed_refs->lock);
4304                 mutex_unlock(&head->mutex);
4305
4306                 if (pin_bytes)
4307                         btrfs_pin_extent(fs_info, head->node.bytenr,
4308                                          head->node.num_bytes, 1);
4309                 btrfs_put_delayed_ref(&head->node);
4310                 cond_resched();
4311                 spin_lock(&delayed_refs->lock);
4312         }
4313
4314         spin_unlock(&delayed_refs->lock);
4315
4316         return ret;
4317 }
4318
4319 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4320 {
4321         struct btrfs_inode *btrfs_inode;
4322         struct list_head splice;
4323
4324         INIT_LIST_HEAD(&splice);
4325
4326         spin_lock(&root->delalloc_lock);
4327         list_splice_init(&root->delalloc_inodes, &splice);
4328
4329         while (!list_empty(&splice)) {
4330                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4331                                                delalloc_inodes);
4332
4333                 list_del_init(&btrfs_inode->delalloc_inodes);
4334                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4335                           &btrfs_inode->runtime_flags);
4336                 spin_unlock(&root->delalloc_lock);
4337
4338                 btrfs_invalidate_inodes(btrfs_inode->root);
4339
4340                 spin_lock(&root->delalloc_lock);
4341         }
4342
4343         spin_unlock(&root->delalloc_lock);
4344 }
4345
4346 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4347 {
4348         struct btrfs_root *root;
4349         struct list_head splice;
4350
4351         INIT_LIST_HEAD(&splice);
4352
4353         spin_lock(&fs_info->delalloc_root_lock);
4354         list_splice_init(&fs_info->delalloc_roots, &splice);
4355         while (!list_empty(&splice)) {
4356                 root = list_first_entry(&splice, struct btrfs_root,
4357                                          delalloc_root);
4358                 list_del_init(&root->delalloc_root);
4359                 root = btrfs_grab_fs_root(root);
4360                 BUG_ON(!root);
4361                 spin_unlock(&fs_info->delalloc_root_lock);
4362
4363                 btrfs_destroy_delalloc_inodes(root);
4364                 btrfs_put_fs_root(root);
4365
4366                 spin_lock(&fs_info->delalloc_root_lock);
4367         }
4368         spin_unlock(&fs_info->delalloc_root_lock);
4369 }
4370
4371 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4372                                         struct extent_io_tree *dirty_pages,
4373                                         int mark)
4374 {
4375         int ret;
4376         struct extent_buffer *eb;
4377         u64 start = 0;
4378         u64 end;
4379
4380         while (1) {
4381                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4382                                             mark, NULL);
4383                 if (ret)
4384                         break;
4385
4386                 clear_extent_bits(dirty_pages, start, end, mark);
4387                 while (start <= end) {
4388                         eb = find_extent_buffer(fs_info, start);
4389                         start += fs_info->nodesize;
4390                         if (!eb)
4391                                 continue;
4392                         wait_on_extent_buffer_writeback(eb);
4393
4394                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4395                                                &eb->bflags))
4396                                 clear_extent_buffer_dirty(eb);
4397                         free_extent_buffer_stale(eb);
4398                 }
4399         }
4400
4401         return ret;
4402 }
4403
4404 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4405                                        struct extent_io_tree *pinned_extents)
4406 {
4407         struct extent_io_tree *unpin;
4408         u64 start;
4409         u64 end;
4410         int ret;
4411         bool loop = true;
4412
4413         unpin = pinned_extents;
4414 again:
4415         while (1) {
4416                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4417                                             EXTENT_DIRTY, NULL);
4418                 if (ret)
4419                         break;
4420
4421                 clear_extent_dirty(unpin, start, end);
4422                 btrfs_error_unpin_extent_range(fs_info, start, end);
4423                 cond_resched();
4424         }
4425
4426         if (loop) {
4427                 if (unpin == &fs_info->freed_extents[0])
4428                         unpin = &fs_info->freed_extents[1];
4429                 else
4430                         unpin = &fs_info->freed_extents[0];
4431                 loop = false;
4432                 goto again;
4433         }
4434
4435         return 0;
4436 }
4437
4438 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4439 {
4440         struct inode *inode;
4441
4442         inode = cache->io_ctl.inode;
4443         if (inode) {
4444                 invalidate_inode_pages2(inode->i_mapping);
4445                 BTRFS_I(inode)->generation = 0;
4446                 cache->io_ctl.inode = NULL;
4447                 iput(inode);
4448         }
4449         btrfs_put_block_group(cache);
4450 }
4451
4452 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4453                              struct btrfs_fs_info *fs_info)
4454 {
4455         struct btrfs_block_group_cache *cache;
4456
4457         spin_lock(&cur_trans->dirty_bgs_lock);
4458         while (!list_empty(&cur_trans->dirty_bgs)) {
4459                 cache = list_first_entry(&cur_trans->dirty_bgs,
4460                                          struct btrfs_block_group_cache,
4461                                          dirty_list);
4462                 if (!cache) {
4463                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4464                         spin_unlock(&cur_trans->dirty_bgs_lock);
4465                         return;
4466                 }
4467
4468                 if (!list_empty(&cache->io_list)) {
4469                         spin_unlock(&cur_trans->dirty_bgs_lock);
4470                         list_del_init(&cache->io_list);
4471                         btrfs_cleanup_bg_io(cache);
4472                         spin_lock(&cur_trans->dirty_bgs_lock);
4473                 }
4474
4475                 list_del_init(&cache->dirty_list);
4476                 spin_lock(&cache->lock);
4477                 cache->disk_cache_state = BTRFS_DC_ERROR;
4478                 spin_unlock(&cache->lock);
4479
4480                 spin_unlock(&cur_trans->dirty_bgs_lock);
4481                 btrfs_put_block_group(cache);
4482                 spin_lock(&cur_trans->dirty_bgs_lock);
4483         }
4484         spin_unlock(&cur_trans->dirty_bgs_lock);
4485
4486         while (!list_empty(&cur_trans->io_bgs)) {
4487                 cache = list_first_entry(&cur_trans->io_bgs,
4488                                          struct btrfs_block_group_cache,
4489                                          io_list);
4490                 if (!cache) {
4491                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4492                         return;
4493                 }
4494
4495                 list_del_init(&cache->io_list);
4496                 spin_lock(&cache->lock);
4497                 cache->disk_cache_state = BTRFS_DC_ERROR;
4498                 spin_unlock(&cache->lock);
4499                 btrfs_cleanup_bg_io(cache);
4500         }
4501 }
4502
4503 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4504                                    struct btrfs_fs_info *fs_info)
4505 {
4506         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4507         ASSERT(list_empty(&cur_trans->dirty_bgs));
4508         ASSERT(list_empty(&cur_trans->io_bgs));
4509
4510         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4511
4512         cur_trans->state = TRANS_STATE_COMMIT_START;
4513         wake_up(&fs_info->transaction_blocked_wait);
4514
4515         cur_trans->state = TRANS_STATE_UNBLOCKED;
4516         wake_up(&fs_info->transaction_wait);
4517
4518         btrfs_destroy_delayed_inodes(fs_info);
4519         btrfs_assert_delayed_root_empty(fs_info);
4520
4521         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4522                                      EXTENT_DIRTY);
4523         btrfs_destroy_pinned_extent(fs_info,
4524                                     fs_info->pinned_extents);
4525
4526         cur_trans->state =TRANS_STATE_COMPLETED;
4527         wake_up(&cur_trans->commit_wait);
4528 }
4529
4530 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4531 {
4532         struct btrfs_transaction *t;
4533
4534         mutex_lock(&fs_info->transaction_kthread_mutex);
4535
4536         spin_lock(&fs_info->trans_lock);
4537         while (!list_empty(&fs_info->trans_list)) {
4538                 t = list_first_entry(&fs_info->trans_list,
4539                                      struct btrfs_transaction, list);
4540                 if (t->state >= TRANS_STATE_COMMIT_START) {
4541                         refcount_inc(&t->use_count);
4542                         spin_unlock(&fs_info->trans_lock);
4543                         btrfs_wait_for_commit(fs_info, t->transid);
4544                         btrfs_put_transaction(t);
4545                         spin_lock(&fs_info->trans_lock);
4546                         continue;
4547                 }
4548                 if (t == fs_info->running_transaction) {
4549                         t->state = TRANS_STATE_COMMIT_DOING;
4550                         spin_unlock(&fs_info->trans_lock);
4551                         /*
4552                          * We wait for 0 num_writers since we don't hold a trans
4553                          * handle open currently for this transaction.
4554                          */
4555                         wait_event(t->writer_wait,
4556                                    atomic_read(&t->num_writers) == 0);
4557                 } else {
4558                         spin_unlock(&fs_info->trans_lock);
4559                 }
4560                 btrfs_cleanup_one_transaction(t, fs_info);
4561
4562                 spin_lock(&fs_info->trans_lock);
4563                 if (t == fs_info->running_transaction)
4564                         fs_info->running_transaction = NULL;
4565                 list_del_init(&t->list);
4566                 spin_unlock(&fs_info->trans_lock);
4567
4568                 btrfs_put_transaction(t);
4569                 trace_btrfs_transaction_commit(fs_info->tree_root);
4570                 spin_lock(&fs_info->trans_lock);
4571         }
4572         spin_unlock(&fs_info->trans_lock);
4573         btrfs_destroy_all_ordered_extents(fs_info);
4574         btrfs_destroy_delayed_inodes(fs_info);
4575         btrfs_assert_delayed_root_empty(fs_info);
4576         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4577         btrfs_destroy_all_delalloc_inodes(fs_info);
4578         mutex_unlock(&fs_info->transaction_kthread_mutex);
4579
4580         return 0;
4581 }
4582
4583 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4584 {
4585         struct inode *inode = private_data;
4586         return btrfs_sb(inode->i_sb);
4587 }
4588
4589 static const struct extent_io_ops btree_extent_io_ops = {
4590         /* mandatory callbacks */
4591         .submit_bio_hook = btree_submit_bio_hook,
4592         .readpage_end_io_hook = btree_readpage_end_io_hook,
4593         /* note we're sharing with inode.c for the merge bio hook */
4594         .merge_bio_hook = btrfs_merge_bio_hook,
4595         .readpage_io_failed_hook = btree_io_failed_hook,
4596         .set_range_writeback = btrfs_set_range_writeback,
4597         .tree_fs_info = btree_fs_info,
4598
4599         /* optional callbacks */
4600 };