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/freezer.h>
  29 #include <linux/slab.h>
  30 #include <linux/migrate.h>
  31 #include <linux/ratelimit.h>
  32 #include <linux/uuid.h>
  33 #include <linux/semaphore.h>
  34 #include <asm/unaligned.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "hash.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "volumes.h"
  41 #include "print-tree.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "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
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  64                                       struct btrfs_root *root);
  65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  67                                         struct extent_io_tree *dirty_pages,
  68                                         int mark);
  69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  70                                        struct extent_io_tree *pinned_extents);
  71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  72 static void btrfs_error_commit_super(struct btrfs_root *root);
  73
  74 /*
  75  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  76  * is complete.  This is used during reads to verify checksums, and it is used
  77  * by writes to insert metadata for new file extents after IO is complete.
  78  */
  79 struct btrfs_end_io_wq {
  80         struct bio *bio;
  81         bio_end_io_t *end_io;
  82         void *private;
  83         struct btrfs_fs_info *info;
  84         int error;
  85         enum btrfs_wq_endio_type metadata;
  86         struct list_head list;
  87         struct btrfs_work work;
  88 };
  89
  90 static struct kmem_cache *btrfs_end_io_wq_cache;
  91
  92 int __init btrfs_end_io_wq_init(void)
  93 {
  94         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  95                                         sizeof(struct btrfs_end_io_wq),
  96                                         0,
  97                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  98                                         NULL);
  99         if (!btrfs_end_io_wq_cache)
 100                 return -ENOMEM;
 101         return 0;
 102 }
 103
 104 void btrfs_end_io_wq_exit(void)
 105 {
 106         if (btrfs_end_io_wq_cache)
 107                 kmem_cache_destroy(btrfs_end_io_wq_cache);
 108 }
 109
 110 /*
 111  * async submit bios are used to offload expensive checksumming
 112  * onto the worker threads.  They checksum file and metadata bios
 113  * just before they are sent down the IO stack.
 114  */
 115 struct async_submit_bio {
 116         struct inode *inode;
 117         struct bio *bio;
 118         struct list_head list;
 119         extent_submit_bio_hook_t *submit_bio_start;
 120         extent_submit_bio_hook_t *submit_bio_done;
 121         int rw;
 122         int mirror_num;
 123         unsigned long bio_flags;
 124         /*
 125          * bio_offset is optional, can be used if the pages in the bio
 126          * can't tell us where in the file the bio should go
 127          */
 128         u64 bio_offset;
 129         struct btrfs_work work;
 130         int error;
 131 };
 132
 133 /*
 134  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 135  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 136  * the level the eb occupies in the tree.
 137  *
 138  * Different roots are used for different purposes and may nest inside each
 139  * other and they require separate keysets.  As lockdep keys should be
 140  * static, assign keysets according to the purpose of the root as indicated
 141  * by btrfs_root->objectid.  This ensures that all special purpose roots
 142  * have separate keysets.
 143  *
 144  * Lock-nesting across peer nodes is always done with the immediate parent
 145  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 146  * subclass to avoid triggering lockdep warning in such cases.
 147  *
 148  * The key is set by the readpage_end_io_hook after the buffer has passed
 149  * csum validation but before the pages are unlocked.  It is also set by
 150  * btrfs_init_new_buffer on freshly allocated blocks.
 151  *
 152  * We also add a check to make sure the highest level of the tree is the
 153  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 154  * needs update as well.
 155  */
 156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 157 # if BTRFS_MAX_LEVEL != 8
 158 #  error
 159 # endif
 160
 161 static struct btrfs_lockdep_keyset {
 162         u64                     id;             /* root objectid */
 163         const char              *name_stem;     /* lock name stem */
 164         char                    names[BTRFS_MAX_LEVEL + 1][20];
 165         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 166 } btrfs_lockdep_keysets[] = {
 167         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 168         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 169         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 170         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 171         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 172         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 173         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 174         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 175         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 176         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 177         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 178         { .id = 0,                              .name_stem = "tree"     },
 179 };
 180
 181 void __init btrfs_init_lockdep(void)
 182 {
 183         int i, j;
 184
 185         /* initialize lockdep class names */
 186         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 187                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 188
 189                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 190                         snprintf(ks->names[j], sizeof(ks->names[j]),
 191                                  "btrfs-%s-%02d", ks->name_stem, j);
 192         }
 193 }
 194
 195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 196                                     int level)
 197 {
 198         struct btrfs_lockdep_keyset *ks;
 199
 200         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 201
 202         /* find the matching keyset, id 0 is the default entry */
 203         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 204                 if (ks->id == objectid)
 205                         break;
 206
 207         lockdep_set_class_and_name(&eb->lock,
 208                                    &ks->keys[level], ks->names[level]);
 209 }
 210
 211 #endif
 212
 213 /*
 214  * extents on the btree inode are pretty simple, there's one extent
 215  * that covers the entire device
 216  */
 217 static struct extent_map *btree_get_extent(struct inode *inode,
 218                 struct page *page, size_t pg_offset, u64 start, u64 len,
 219                 int create)
 220 {
 221         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 222         struct extent_map *em;
 223         int ret;
 224
 225         read_lock(&em_tree->lock);
 226         em = lookup_extent_mapping(em_tree, start, len);
 227         if (em) {
 228                 em->bdev =
 229                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 230                 read_unlock(&em_tree->lock);
 231                 goto out;
 232         }
 233         read_unlock(&em_tree->lock);
 234
 235         em = alloc_extent_map();
 236         if (!em) {
 237                 em = ERR_PTR(-ENOMEM);
 238                 goto out;
 239         }
 240         em->start = 0;
 241         em->len = (u64)-1;
 242         em->block_len = (u64)-1;
 243         em->block_start = 0;
 244         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 245
 246         write_lock(&em_tree->lock);
 247         ret = add_extent_mapping(em_tree, em, 0);
 248         if (ret == -EEXIST) {
 249                 free_extent_map(em);
 250                 em = lookup_extent_mapping(em_tree, start, len);
 251                 if (!em)
 252                         em = ERR_PTR(-EIO);
 253         } else if (ret) {
 254                 free_extent_map(em);
 255                 em = ERR_PTR(ret);
 256         }
 257         write_unlock(&em_tree->lock);
 258
 259 out:
 260         return em;
 261 }
 262
 263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 264 {
 265         return btrfs_crc32c(seed, data, len);
 266 }
 267
 268 void btrfs_csum_final(u32 crc, char *result)
 269 {
 270         put_unaligned_le32(~crc, result);
 271 }
 272
 273 /*
 274  * compute the csum for a btree block, and either verify it or write it
 275  * into the csum field of the block.
 276  */
 277 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 278                            int verify)
 279 {
 280         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 281         char *result = NULL;
 282         unsigned long len;
 283         unsigned long cur_len;
 284         unsigned long offset = BTRFS_CSUM_SIZE;
 285         char *kaddr;
 286         unsigned long map_start;
 287         unsigned long map_len;
 288         int err;
 289         u32 crc = ~(u32)0;
 290         unsigned long inline_result;
 291
 292         len = buf->len - offset;
 293         while (len > 0) {
 294                 err = map_private_extent_buffer(buf, offset, 32,
 295                                         &kaddr, &map_start, &map_len);
 296                 if (err)
 297                         return 1;
 298                 cur_len = min(len, map_len - (offset - map_start));
 299                 crc = btrfs_csum_data(kaddr + offset - map_start,
 300                                       crc, cur_len);
 301                 len -= cur_len;
 302                 offset += cur_len;
 303         }
 304         if (csum_size > sizeof(inline_result)) {
 305                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 306                 if (!result)
 307                         return 1;
 308         } else {
 309                 result = (char *)&inline_result;
 310         }
 311
 312         btrfs_csum_final(crc, result);
 313
 314         if (verify) {
 315                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 316                         u32 val;
 317                         u32 found = 0;
 318                         memcpy(&found, result, csum_size);
 319
 320                         read_extent_buffer(buf, &val, 0, csum_size);
 321                         printk_ratelimited(KERN_INFO
 322                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 323                                 "level %d\n",
 324                                 root->fs_info->sb->s_id, buf->start,
 325                                 val, found, btrfs_header_level(buf));
 326                         if (result != (char *)&inline_result)
 327                                 kfree(result);
 328                         return 1;
 329                 }
 330         } else {
 331                 write_extent_buffer(buf, result, 0, csum_size);
 332         }
 333         if (result != (char *)&inline_result)
 334                 kfree(result);
 335         return 0;
 336 }
 337
 338 /*
 339  * we can't consider a given block up to date unless the transid of the
 340  * block matches the transid in the parent node's pointer.  This is how we
 341  * detect blocks that either didn't get written at all or got written
 342  * in the wrong place.
 343  */
 344 static int verify_parent_transid(struct extent_io_tree *io_tree,
 345                                  struct extent_buffer *eb, u64 parent_transid,
 346                                  int atomic)
 347 {
 348         struct extent_state *cached_state = NULL;
 349         int ret;
 350         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 351
 352         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 353                 return 0;
 354
 355         if (atomic)
 356                 return -EAGAIN;
 357
 358         if (need_lock) {
 359                 btrfs_tree_read_lock(eb);
 360                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 361         }
 362
 363         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 364                          0, &cached_state);
 365         if (extent_buffer_uptodate(eb) &&
 366             btrfs_header_generation(eb) == parent_transid) {
 367                 ret = 0;
 368                 goto out;
 369         }
 370         printk_ratelimited(KERN_INFO "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 371                         eb->fs_info->sb->s_id, eb->start,
 372                         parent_transid, btrfs_header_generation(eb));
 373         ret = 1;
 374
 375         /*
 376          * Things reading via commit roots that don't have normal protection,
 377          * like send, can have a really old block in cache that may point at a
 378          * block that has been free'd and re-allocated.  So don't clear uptodate
 379          * if we find an eb that is under IO (dirty/writeback) because we could
 380          * end up reading in the stale data and then writing it back out and
 381          * making everybody very sad.
 382          */
 383         if (!extent_buffer_under_io(eb))
 384                 clear_extent_buffer_uptodate(eb);
 385 out:
 386         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 387                              &cached_state, GFP_NOFS);
 388         if (need_lock)
 389                 btrfs_tree_read_unlock_blocking(eb);
 390         return ret;
 391 }
 392
 393 /*
 394  * Return 0 if the superblock checksum type matches the checksum value of that
 395  * algorithm. Pass the raw disk superblock data.
 396  */
 397 static int btrfs_check_super_csum(char *raw_disk_sb)
 398 {
 399         struct btrfs_super_block *disk_sb =
 400                 (struct btrfs_super_block *)raw_disk_sb;
 401         u16 csum_type = btrfs_super_csum_type(disk_sb);
 402         int ret = 0;
 403
 404         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 405                 u32 crc = ~(u32)0;
 406                 const int csum_size = sizeof(crc);
 407                 char result[csum_size];
 408
 409                 /*
 410                  * The super_block structure does not span the whole
 411                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 412                  * is filled with zeros and is included in the checkum.
 413                  */
 414                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 415                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 416                 btrfs_csum_final(crc, result);
 417
 418                 if (memcmp(raw_disk_sb, result, csum_size))
 419                         ret = 1;
 420
 421                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 422                         printk(KERN_WARNING
 423                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 424                         ret = 0;
 425                 }
 426         }
 427
 428         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 429                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 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_root *root,
 442                                           struct extent_buffer *eb,
 443                                           u64 start, 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(root->fs_info->btree_inode)->io_tree;
 454         while (1) {
 455                 ret = read_extent_buffer_pages(io_tree, eb, start,
 456                                                WAIT_COMPLETE,
 457                                                btree_get_extent, mirror_num);
 458                 if (!ret) {
 459                         if (!verify_parent_transid(io_tree, eb,
 460                                                    parent_transid, 0))
 461                                 break;
 462                         else
 463                                 ret = -EIO;
 464                 }
 465
 466                 /*
 467                  * This buffer's crc is fine, but its contents are corrupted, so
 468                  * there is no reason to read the other copies, they won't be
 469                  * any less wrong.
 470                  */
 471                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 472                         break;
 473
 474                 num_copies = btrfs_num_copies(root->fs_info,
 475                                               eb->start, eb->len);
 476                 if (num_copies == 1)
 477                         break;
 478
 479                 if (!failed_mirror) {
 480                         failed = 1;
 481                         failed_mirror = eb->read_mirror;
 482                 }
 483
 484                 mirror_num++;
 485                 if (mirror_num == failed_mirror)
 486                         mirror_num++;
 487
 488                 if (mirror_num > num_copies)
 489                         break;
 490         }
 491
 492         if (failed && !ret && failed_mirror)
 493                 repair_eb_io_failure(root, eb, failed_mirror);
 494
 495         return ret;
 496 }
 497
 498 /*
 499  * checksum a dirty tree block before IO.  This has extra checks to make sure
 500  * we only fill in the checksum field in the first page of a multi-page block
 501  */
 502
 503 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 504 {
 505         u64 start = page_offset(page);
 506         u64 found_start;
 507         struct extent_buffer *eb;
 508
 509         eb = (struct extent_buffer *)page->private;
 510         if (page != eb->pages[0])
 511                 return 0;
 512         found_start = btrfs_header_bytenr(eb);
 513         if (WARN_ON(found_start != start || !PageUptodate(page)))
 514                 return 0;
 515         csum_tree_block(root, eb, 0);
 516         return 0;
 517 }
 518
 519 static int check_tree_block_fsid(struct btrfs_root *root,
 520                                  struct extent_buffer *eb)
 521 {
 522         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 523         u8 fsid[BTRFS_UUID_SIZE];
 524         int ret = 1;
 525
 526         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 527         while (fs_devices) {
 528                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 529                         ret = 0;
 530                         break;
 531                 }
 532                 fs_devices = fs_devices->seed;
 533         }
 534         return ret;
 535 }
 536
 537 #define CORRUPT(reason, eb, root, slot)                         \
 538         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 539                    "root=%llu, slot=%d", reason,                        \
 540                btrfs_header_bytenr(eb), root->objectid, slot)
 541
 542 static noinline int check_leaf(struct btrfs_root *root,
 543                                struct extent_buffer *leaf)
 544 {
 545         struct btrfs_key key;
 546         struct btrfs_key leaf_key;
 547         u32 nritems = btrfs_header_nritems(leaf);
 548         int slot;
 549
 550         if (nritems == 0)
 551                 return 0;
 552
 553         /* Check the 0 item */
 554         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 555             BTRFS_LEAF_DATA_SIZE(root)) {
 556                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 557                 return -EIO;
 558         }
 559
 560         /*
 561          * Check to make sure each items keys are in the correct order and their
 562          * offsets make sense.  We only have to loop through nritems-1 because
 563          * we check the current slot against the next slot, which verifies the
 564          * next slot's offset+size makes sense and that the current's slot
 565          * offset is correct.
 566          */
 567         for (slot = 0; slot < nritems - 1; slot++) {
 568                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 569                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 570
 571                 /* Make sure the keys are in the right order */
 572                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 573                         CORRUPT("bad key order", leaf, root, slot);
 574                         return -EIO;
 575                 }
 576
 577                 /*
 578                  * Make sure the offset and ends are right, remember that the
 579                  * item data starts at the end of the leaf and grows towards the
 580                  * front.
 581                  */
 582                 if (btrfs_item_offset_nr(leaf, slot) !=
 583                         btrfs_item_end_nr(leaf, slot + 1)) {
 584                         CORRUPT("slot offset bad", leaf, root, slot);
 585                         return -EIO;
 586                 }
 587
 588                 /*
 589                  * Check to make sure that we don't point outside of the leaf,
 590                  * just incase all the items are consistent to eachother, but
 591                  * all point outside of the leaf.
 592                  */
 593                 if (btrfs_item_end_nr(leaf, slot) >
 594                     BTRFS_LEAF_DATA_SIZE(root)) {
 595                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 596                         return -EIO;
 597                 }
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 604                                       u64 phy_offset, struct page *page,
 605                                       u64 start, u64 end, int mirror)
 606 {
 607         u64 found_start;
 608         int found_level;
 609         struct extent_buffer *eb;
 610         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 611         int ret = 0;
 612         int reads_done;
 613
 614         if (!page->private)
 615                 goto out;
 616
 617         eb = (struct extent_buffer *)page->private;
 618
 619         /* the pending IO might have been the only thing that kept this buffer
 620          * in memory.  Make sure we have a ref for all this other checks
 621          */
 622         extent_buffer_get(eb);
 623
 624         reads_done = atomic_dec_and_test(&eb->io_pages);
 625         if (!reads_done)
 626                 goto err;
 627
 628         eb->read_mirror = mirror;
 629         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633
 634         found_start = btrfs_header_bytenr(eb);
 635         if (found_start != eb->start) {
 636                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad tree block start "
 637                                "%llu %llu\n",
 638                                eb->fs_info->sb->s_id, found_start, eb->start);
 639                 ret = -EIO;
 640                 goto err;
 641         }
 642         if (check_tree_block_fsid(root, eb)) {
 643                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad fsid on block %llu\n",
 644                                eb->fs_info->sb->s_id, eb->start);
 645                 ret = -EIO;
 646                 goto err;
 647         }
 648         found_level = btrfs_header_level(eb);
 649         if (found_level >= BTRFS_MAX_LEVEL) {
 650                 btrfs_info(root->fs_info, "bad tree block level %d",
 651                            (int)btrfs_header_level(eb));
 652                 ret = -EIO;
 653                 goto err;
 654         }
 655
 656         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 657                                        eb, found_level);
 658
 659         ret = csum_tree_block(root, eb, 1);
 660         if (ret) {
 661                 ret = -EIO;
 662                 goto err;
 663         }
 664
 665         /*
 666          * If this is a leaf block and it is corrupt, set the corrupt bit so
 667          * that we don't try and read the other copies of this block, just
 668          * return -EIO.
 669          */
 670         if (found_level == 0 && check_leaf(root, eb)) {
 671                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 672                 ret = -EIO;
 673         }
 674
 675         if (!ret)
 676                 set_extent_buffer_uptodate(eb);
 677 err:
 678         if (reads_done &&
 679             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 680                 btree_readahead_hook(root, eb, eb->start, ret);
 681
 682         if (ret) {
 683                 /*
 684                  * our io error hook is going to dec the io pages
 685                  * again, we have to make sure it has something
 686                  * to decrement
 687                  */
 688                 atomic_inc(&eb->io_pages);
 689                 clear_extent_buffer_uptodate(eb);
 690         }
 691         free_extent_buffer(eb);
 692 out:
 693         return ret;
 694 }
 695
 696 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 697 {
 698         struct extent_buffer *eb;
 699         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 700
 701         eb = (struct extent_buffer *)page->private;
 702         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 703         eb->read_mirror = failed_mirror;
 704         atomic_dec(&eb->io_pages);
 705         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 706                 btree_readahead_hook(root, eb, eb->start, -EIO);
 707         return -EIO;    /* we fixed nothing */
 708 }
 709
 710 static void end_workqueue_bio(struct bio *bio, int err)
 711 {
 712         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 713         struct btrfs_fs_info *fs_info;
 714         struct btrfs_workqueue *wq;
 715         btrfs_work_func_t func;
 716
 717         fs_info = end_io_wq->info;
 718         end_io_wq->error = err;
 719
 720         if (bio->bi_rw & REQ_WRITE) {
 721                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 722                         wq = fs_info->endio_meta_write_workers;
 723                         func = btrfs_endio_meta_write_helper;
 724                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 725                         wq = fs_info->endio_freespace_worker;
 726                         func = btrfs_freespace_write_helper;
 727                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 728                         wq = fs_info->endio_raid56_workers;
 729                         func = btrfs_endio_raid56_helper;
 730                 } else {
 731                         wq = fs_info->endio_write_workers;
 732                         func = btrfs_endio_write_helper;
 733                 }
 734         } else {
 735                 if (unlikely(end_io_wq->metadata ==
 736                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 737                         wq = fs_info->endio_repair_workers;
 738                         func = btrfs_endio_repair_helper;
 739                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 740                         wq = fs_info->endio_raid56_workers;
 741                         func = btrfs_endio_raid56_helper;
 742                 } else if (end_io_wq->metadata) {
 743                         wq = fs_info->endio_meta_workers;
 744                         func = btrfs_endio_meta_helper;
 745                 } else {
 746                         wq = fs_info->endio_workers;
 747                         func = btrfs_endio_helper;
 748                 }
 749         }
 750
 751         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 752         btrfs_queue_work(wq, &end_io_wq->work);
 753 }
 754
 755 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 756                         enum btrfs_wq_endio_type metadata)
 757 {
 758         struct btrfs_end_io_wq *end_io_wq;
 759
 760         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 761         if (!end_io_wq)
 762                 return -ENOMEM;
 763
 764         end_io_wq->private = bio->bi_private;
 765         end_io_wq->end_io = bio->bi_end_io;
 766         end_io_wq->info = info;
 767         end_io_wq->error = 0;
 768         end_io_wq->bio = bio;
 769         end_io_wq->metadata = metadata;
 770
 771         bio->bi_private = end_io_wq;
 772         bio->bi_end_io = end_workqueue_bio;
 773         return 0;
 774 }
 775
 776 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 777 {
 778         unsigned long limit = min_t(unsigned long,
 779                                     info->thread_pool_size,
 780                                     info->fs_devices->open_devices);
 781         return 256 * limit;
 782 }
 783
 784 static void run_one_async_start(struct btrfs_work *work)
 785 {
 786         struct async_submit_bio *async;
 787         int ret;
 788
 789         async = container_of(work, struct  async_submit_bio, work);
 790         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 791                                       async->mirror_num, async->bio_flags,
 792                                       async->bio_offset);
 793         if (ret)
 794                 async->error = ret;
 795 }
 796
 797 static void run_one_async_done(struct btrfs_work *work)
 798 {
 799         struct btrfs_fs_info *fs_info;
 800         struct async_submit_bio *async;
 801         int limit;
 802
 803         async = container_of(work, struct  async_submit_bio, work);
 804         fs_info = BTRFS_I(async->inode)->root->fs_info;
 805
 806         limit = btrfs_async_submit_limit(fs_info);
 807         limit = limit * 2 / 3;
 808
 809         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 810             waitqueue_active(&fs_info->async_submit_wait))
 811                 wake_up(&fs_info->async_submit_wait);
 812
 813         /* If an error occured we just want to clean up the bio and move on */
 814         if (async->error) {
 815                 bio_endio(async->bio, async->error);
 816                 return;
 817         }
 818
 819         async->submit_bio_done(async->inode, async->rw, async->bio,
 820                                async->mirror_num, async->bio_flags,
 821                                async->bio_offset);
 822 }
 823
 824 static void run_one_async_free(struct btrfs_work *work)
 825 {
 826         struct async_submit_bio *async;
 827
 828         async = container_of(work, struct  async_submit_bio, work);
 829         kfree(async);
 830 }
 831
 832 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 833                         int rw, struct bio *bio, int mirror_num,
 834                         unsigned long bio_flags,
 835                         u64 bio_offset,
 836                         extent_submit_bio_hook_t *submit_bio_start,
 837                         extent_submit_bio_hook_t *submit_bio_done)
 838 {
 839         struct async_submit_bio *async;
 840
 841         async = kmalloc(sizeof(*async), GFP_NOFS);
 842         if (!async)
 843                 return -ENOMEM;
 844
 845         async->inode = inode;
 846         async->rw = rw;
 847         async->bio = bio;
 848         async->mirror_num = mirror_num;
 849         async->submit_bio_start = submit_bio_start;
 850         async->submit_bio_done = submit_bio_done;
 851
 852         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 853                         run_one_async_done, run_one_async_free);
 854
 855         async->bio_flags = bio_flags;
 856         async->bio_offset = bio_offset;
 857
 858         async->error = 0;
 859
 860         atomic_inc(&fs_info->nr_async_submits);
 861
 862         if (rw & REQ_SYNC)
 863                 btrfs_set_work_high_priority(&async->work);
 864
 865         btrfs_queue_work(fs_info->workers, &async->work);
 866
 867         while (atomic_read(&fs_info->async_submit_draining) &&
 868               atomic_read(&fs_info->nr_async_submits)) {
 869                 wait_event(fs_info->async_submit_wait,
 870                            (atomic_read(&fs_info->nr_async_submits) == 0));
 871         }
 872
 873         return 0;
 874 }
 875
 876 static int btree_csum_one_bio(struct bio *bio)
 877 {
 878         struct bio_vec *bvec;
 879         struct btrfs_root *root;
 880         int i, ret = 0;
 881
 882         bio_for_each_segment_all(bvec, bio, i) {
 883                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 884                 ret = csum_dirty_buffer(root, bvec->bv_page);
 885                 if (ret)
 886                         break;
 887         }
 888
 889         return ret;
 890 }
 891
 892 static int __btree_submit_bio_start(struct inode *inode, int rw,
 893                                     struct bio *bio, int mirror_num,
 894                                     unsigned long bio_flags,
 895                                     u64 bio_offset)
 896 {
 897         /*
 898          * when we're called for a write, we're already in the async
 899          * submission context.  Just jump into btrfs_map_bio
 900          */
 901         return btree_csum_one_bio(bio);
 902 }
 903
 904 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 905                                  int mirror_num, unsigned long bio_flags,
 906                                  u64 bio_offset)
 907 {
 908         int ret;
 909
 910         /*
 911          * when we're called for a write, we're already in the async
 912          * submission context.  Just jump into btrfs_map_bio
 913          */
 914         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 915         if (ret)
 916                 bio_endio(bio, ret);
 917         return ret;
 918 }
 919
 920 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 921 {
 922         if (bio_flags & EXTENT_BIO_TREE_LOG)
 923                 return 0;
 924 #ifdef CONFIG_X86
 925         if (cpu_has_xmm4_2)
 926                 return 0;
 927 #endif
 928         return 1;
 929 }
 930
 931 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 932                                  int mirror_num, unsigned long bio_flags,
 933                                  u64 bio_offset)
 934 {
 935         int async = check_async_write(inode, bio_flags);
 936         int ret;
 937
 938         if (!(rw & REQ_WRITE)) {
 939                 /*
 940                  * called for a read, do the setup so that checksum validation
 941                  * can happen in the async kernel threads
 942                  */
 943                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 944                                           bio, BTRFS_WQ_ENDIO_METADATA);
 945                 if (ret)
 946                         goto out_w_error;
 947                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 948                                     mirror_num, 0);
 949         } else if (!async) {
 950                 ret = btree_csum_one_bio(bio);
 951                 if (ret)
 952                         goto out_w_error;
 953                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 954                                     mirror_num, 0);
 955         } else {
 956                 /*
 957                  * kthread helpers are used to submit writes so that
 958                  * checksumming can happen in parallel across all CPUs
 959                  */
 960                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 961                                           inode, rw, bio, mirror_num, 0,
 962                                           bio_offset,
 963                                           __btree_submit_bio_start,
 964                                           __btree_submit_bio_done);
 965         }
 966
 967         if (ret) {
 968 out_w_error:
 969                 bio_endio(bio, ret);
 970         }
 971         return ret;
 972 }
 973
 974 #ifdef CONFIG_MIGRATION
 975 static int btree_migratepage(struct address_space *mapping,
 976                         struct page *newpage, struct page *page,
 977                         enum migrate_mode mode)
 978 {
 979         /*
 980          * we can't safely write a btree page from here,
 981          * we haven't done the locking hook
 982          */
 983         if (PageDirty(page))
 984                 return -EAGAIN;
 985         /*
 986          * Buffers may be managed in a filesystem specific way.
 987          * We must have no buffers or drop them.
 988          */
 989         if (page_has_private(page) &&
 990             !try_to_release_page(page, GFP_KERNEL))
 991                 return -EAGAIN;
 992         return migrate_page(mapping, newpage, page, mode);
 993 }
 994 #endif
 995
 996
 997 static int btree_writepages(struct address_space *mapping,
 998                             struct writeback_control *wbc)
 999 {
1000         struct btrfs_fs_info *fs_info;
1001         int ret;
1002
1003         if (wbc->sync_mode == WB_SYNC_NONE) {
1004
1005                 if (wbc->for_kupdate)
1006                         return 0;
1007
1008                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1009                 /* this is a bit racy, but that's ok */
1010                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1011                                              BTRFS_DIRTY_METADATA_THRESH);
1012                 if (ret < 0)
1013                         return 0;
1014         }
1015         return btree_write_cache_pages(mapping, wbc);
1016 }
1017
1018 static int btree_readpage(struct file *file, struct page *page)
1019 {
1020         struct extent_io_tree *tree;
1021         tree = &BTRFS_I(page->mapping->host)->io_tree;
1022         return extent_read_full_page(tree, page, btree_get_extent, 0);
1023 }
1024
1025 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1026 {
1027         if (PageWriteback(page) || PageDirty(page))
1028                 return 0;
1029
1030         return try_release_extent_buffer(page);
1031 }
1032
1033 static void btree_invalidatepage(struct page *page, unsigned int offset,
1034                                  unsigned int length)
1035 {
1036         struct extent_io_tree *tree;
1037         tree = &BTRFS_I(page->mapping->host)->io_tree;
1038         extent_invalidatepage(tree, page, offset);
1039         btree_releasepage(page, GFP_NOFS);
1040         if (PagePrivate(page)) {
1041                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1042                            "page private not zero on page %llu",
1043                            (unsigned long long)page_offset(page));
1044                 ClearPagePrivate(page);
1045                 set_page_private(page, 0);
1046                 page_cache_release(page);
1047         }
1048 }
1049
1050 static int btree_set_page_dirty(struct page *page)
1051 {
1052 #ifdef DEBUG
1053         struct extent_buffer *eb;
1054
1055         BUG_ON(!PagePrivate(page));
1056         eb = (struct extent_buffer *)page->private;
1057         BUG_ON(!eb);
1058         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1059         BUG_ON(!atomic_read(&eb->refs));
1060         btrfs_assert_tree_locked(eb);
1061 #endif
1062         return __set_page_dirty_nobuffers(page);
1063 }
1064
1065 static const struct address_space_operations btree_aops = {
1066         .readpage       = btree_readpage,
1067         .writepages     = btree_writepages,
1068         .releasepage    = btree_releasepage,
1069         .invalidatepage = btree_invalidatepage,
1070 #ifdef CONFIG_MIGRATION
1071         .migratepage    = btree_migratepage,
1072 #endif
1073         .set_page_dirty = btree_set_page_dirty,
1074 };
1075
1076 void readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
1077 {
1078         struct extent_buffer *buf = NULL;
1079         struct inode *btree_inode = root->fs_info->btree_inode;
1080
1081         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082         if (!buf)
1083                 return;
1084         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1085                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1086         free_extent_buffer(buf);
1087 }
1088
1089 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1090                          int mirror_num, struct extent_buffer **eb)
1091 {
1092         struct extent_buffer *buf = NULL;
1093         struct inode *btree_inode = root->fs_info->btree_inode;
1094         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1095         int ret;
1096
1097         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1098         if (!buf)
1099                 return 0;
1100
1101         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1102
1103         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1104                                        btree_get_extent, mirror_num);
1105         if (ret) {
1106                 free_extent_buffer(buf);
1107                 return ret;
1108         }
1109
1110         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1111                 free_extent_buffer(buf);
1112                 return -EIO;
1113         } else if (extent_buffer_uptodate(buf)) {
1114                 *eb = buf;
1115         } else {
1116                 free_extent_buffer(buf);
1117         }
1118         return 0;
1119 }
1120
1121 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1122                                             u64 bytenr)
1123 {
1124         return find_extent_buffer(root->fs_info, bytenr);
1125 }
1126
1127 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1128                                                  u64 bytenr, u32 blocksize)
1129 {
1130         if (btrfs_test_is_dummy_root(root))
1131                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1132                                                 blocksize);
1133         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1134 }
1135
1136
1137 int btrfs_write_tree_block(struct extent_buffer *buf)
1138 {
1139         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1140                                         buf->start + buf->len - 1);
1141 }
1142
1143 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1144 {
1145         return filemap_fdatawait_range(buf->pages[0]->mapping,
1146                                        buf->start, buf->start + buf->len - 1);
1147 }
1148
1149 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1150                                       u64 parent_transid)
1151 {
1152         struct extent_buffer *buf = NULL;
1153         int ret;
1154
1155         buf = btrfs_find_create_tree_block(root, bytenr, root->nodesize);
1156         if (!buf)
1157                 return NULL;
1158
1159         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1160         if (ret) {
1161                 free_extent_buffer(buf);
1162                 return NULL;
1163         }
1164         return buf;
1165
1166 }
1167
1168 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1169                       struct extent_buffer *buf)
1170 {
1171         struct btrfs_fs_info *fs_info = root->fs_info;
1172
1173         if (btrfs_header_generation(buf) ==
1174             fs_info->running_transaction->transid) {
1175                 btrfs_assert_tree_locked(buf);
1176
1177                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1178                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1179                                              -buf->len,
1180                                              fs_info->dirty_metadata_batch);
1181                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182                         btrfs_set_lock_blocking(buf);
1183                         clear_extent_buffer_dirty(buf);
1184                 }
1185         }
1186 }
1187
1188 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1189 {
1190         struct btrfs_subvolume_writers *writers;
1191         int ret;
1192
1193         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1194         if (!writers)
1195                 return ERR_PTR(-ENOMEM);
1196
1197         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1198         if (ret < 0) {
1199                 kfree(writers);
1200                 return ERR_PTR(ret);
1201         }
1202
1203         init_waitqueue_head(&writers->wait);
1204         return writers;
1205 }
1206
1207 static void
1208 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1209 {
1210         percpu_counter_destroy(&writers->counter);
1211         kfree(writers);
1212 }
1213
1214 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1215                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1216                          u64 objectid)
1217 {
1218         root->node = NULL;
1219         root->commit_root = NULL;
1220         root->sectorsize = sectorsize;
1221         root->nodesize = nodesize;
1222         root->stripesize = stripesize;
1223         root->state = 0;
1224         root->orphan_cleanup_state = 0;
1225
1226         root->objectid = objectid;
1227         root->last_trans = 0;
1228         root->highest_objectid = 0;
1229         root->nr_delalloc_inodes = 0;
1230         root->nr_ordered_extents = 0;
1231         root->name = NULL;
1232         root->inode_tree = RB_ROOT;
1233         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1234         root->block_rsv = NULL;
1235         root->orphan_block_rsv = NULL;
1236
1237         INIT_LIST_HEAD(&root->dirty_list);
1238         INIT_LIST_HEAD(&root->root_list);
1239         INIT_LIST_HEAD(&root->delalloc_inodes);
1240         INIT_LIST_HEAD(&root->delalloc_root);
1241         INIT_LIST_HEAD(&root->ordered_extents);
1242         INIT_LIST_HEAD(&root->ordered_root);
1243         INIT_LIST_HEAD(&root->logged_list[0]);
1244         INIT_LIST_HEAD(&root->logged_list[1]);
1245         spin_lock_init(&root->orphan_lock);
1246         spin_lock_init(&root->inode_lock);
1247         spin_lock_init(&root->delalloc_lock);
1248         spin_lock_init(&root->ordered_extent_lock);
1249         spin_lock_init(&root->accounting_lock);
1250         spin_lock_init(&root->log_extents_lock[0]);
1251         spin_lock_init(&root->log_extents_lock[1]);
1252         mutex_init(&root->objectid_mutex);
1253         mutex_init(&root->log_mutex);
1254         mutex_init(&root->ordered_extent_mutex);
1255         mutex_init(&root->delalloc_mutex);
1256         init_waitqueue_head(&root->log_writer_wait);
1257         init_waitqueue_head(&root->log_commit_wait[0]);
1258         init_waitqueue_head(&root->log_commit_wait[1]);
1259         INIT_LIST_HEAD(&root->log_ctxs[0]);
1260         INIT_LIST_HEAD(&root->log_ctxs[1]);
1261         atomic_set(&root->log_commit[0], 0);
1262         atomic_set(&root->log_commit[1], 0);
1263         atomic_set(&root->log_writers, 0);
1264         atomic_set(&root->log_batch, 0);
1265         atomic_set(&root->orphan_inodes, 0);
1266         atomic_set(&root->refs, 1);
1267         atomic_set(&root->will_be_snapshoted, 0);
1268         root->log_transid = 0;
1269         root->log_transid_committed = -1;
1270         root->last_log_commit = 0;
1271         if (fs_info)
1272                 extent_io_tree_init(&root->dirty_log_pages,
1273                                      fs_info->btree_inode->i_mapping);
1274
1275         memset(&root->root_key, 0, sizeof(root->root_key));
1276         memset(&root->root_item, 0, sizeof(root->root_item));
1277         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1278         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1279         if (fs_info)
1280                 root->defrag_trans_start = fs_info->generation;
1281         else
1282                 root->defrag_trans_start = 0;
1283         init_completion(&root->kobj_unregister);
1284         root->root_key.objectid = objectid;
1285         root->anon_dev = 0;
1286
1287         spin_lock_init(&root->root_item_lock);
1288 }
1289
1290 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1291 {
1292         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1293         if (root)
1294                 root->fs_info = fs_info;
1295         return root;
1296 }
1297
1298 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1299 /* Should only be used by the testing infrastructure */
1300 struct btrfs_root *btrfs_alloc_dummy_root(void)
1301 {
1302         struct btrfs_root *root;
1303
1304         root = btrfs_alloc_root(NULL);
1305         if (!root)
1306                 return ERR_PTR(-ENOMEM);
1307         __setup_root(4096, 4096, 4096, root, NULL, 1);
1308         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1309         root->alloc_bytenr = 0;
1310
1311         return root;
1312 }
1313 #endif
1314
1315 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1316                                      struct btrfs_fs_info *fs_info,
1317                                      u64 objectid)
1318 {
1319         struct extent_buffer *leaf;
1320         struct btrfs_root *tree_root = fs_info->tree_root;
1321         struct btrfs_root *root;
1322         struct btrfs_key key;
1323         int ret = 0;
1324         uuid_le uuid;
1325
1326         root = btrfs_alloc_root(fs_info);
1327         if (!root)
1328                 return ERR_PTR(-ENOMEM);
1329
1330         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1331                 tree_root->stripesize, root, fs_info, objectid);
1332         root->root_key.objectid = objectid;
1333         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1334         root->root_key.offset = 0;
1335
1336         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1337         if (IS_ERR(leaf)) {
1338                 ret = PTR_ERR(leaf);
1339                 leaf = NULL;
1340                 goto fail;
1341         }
1342
1343         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1344         btrfs_set_header_bytenr(leaf, leaf->start);
1345         btrfs_set_header_generation(leaf, trans->transid);
1346         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1347         btrfs_set_header_owner(leaf, objectid);
1348         root->node = leaf;
1349
1350         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1351                             BTRFS_FSID_SIZE);
1352         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1353                             btrfs_header_chunk_tree_uuid(leaf),
1354                             BTRFS_UUID_SIZE);
1355         btrfs_mark_buffer_dirty(leaf);
1356
1357         root->commit_root = btrfs_root_node(root);
1358         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1359
1360         root->root_item.flags = 0;
1361         root->root_item.byte_limit = 0;
1362         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1363         btrfs_set_root_generation(&root->root_item, trans->transid);
1364         btrfs_set_root_level(&root->root_item, 0);
1365         btrfs_set_root_refs(&root->root_item, 1);
1366         btrfs_set_root_used(&root->root_item, leaf->len);
1367         btrfs_set_root_last_snapshot(&root->root_item, 0);
1368         btrfs_set_root_dirid(&root->root_item, 0);
1369         uuid_le_gen(&uuid);
1370         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1371         root->root_item.drop_level = 0;
1372
1373         key.objectid = objectid;
1374         key.type = BTRFS_ROOT_ITEM_KEY;
1375         key.offset = 0;
1376         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1377         if (ret)
1378                 goto fail;
1379
1380         btrfs_tree_unlock(leaf);
1381
1382         return root;
1383
1384 fail:
1385         if (leaf) {
1386                 btrfs_tree_unlock(leaf);
1387                 free_extent_buffer(root->commit_root);
1388                 free_extent_buffer(leaf);
1389         }
1390         kfree(root);
1391
1392         return ERR_PTR(ret);
1393 }
1394
1395 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1396                                          struct btrfs_fs_info *fs_info)
1397 {
1398         struct btrfs_root *root;
1399         struct btrfs_root *tree_root = fs_info->tree_root;
1400         struct extent_buffer *leaf;
1401
1402         root = btrfs_alloc_root(fs_info);
1403         if (!root)
1404                 return ERR_PTR(-ENOMEM);
1405
1406         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1407                      tree_root->stripesize, root, fs_info,
1408                      BTRFS_TREE_LOG_OBJECTID);
1409
1410         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1411         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1412         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1413
1414         /*
1415          * DON'T set REF_COWS for log trees
1416          *
1417          * log trees do not get reference counted because they go away
1418          * before a real commit is actually done.  They do store pointers
1419          * to file data extents, and those reference counts still get
1420          * updated (along with back refs to the log tree).
1421          */
1422
1423         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1424                         NULL, 0, 0, 0);
1425         if (IS_ERR(leaf)) {
1426                 kfree(root);
1427                 return ERR_CAST(leaf);
1428         }
1429
1430         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1431         btrfs_set_header_bytenr(leaf, leaf->start);
1432         btrfs_set_header_generation(leaf, trans->transid);
1433         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1434         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1435         root->node = leaf;
1436
1437         write_extent_buffer(root->node, root->fs_info->fsid,
1438                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1439         btrfs_mark_buffer_dirty(root->node);
1440         btrfs_tree_unlock(root->node);
1441         return root;
1442 }
1443
1444 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1445                              struct btrfs_fs_info *fs_info)
1446 {
1447         struct btrfs_root *log_root;
1448
1449         log_root = alloc_log_tree(trans, fs_info);
1450         if (IS_ERR(log_root))
1451                 return PTR_ERR(log_root);
1452         WARN_ON(fs_info->log_root_tree);
1453         fs_info->log_root_tree = log_root;
1454         return 0;
1455 }
1456
1457 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1458                        struct btrfs_root *root)
1459 {
1460         struct btrfs_root *log_root;
1461         struct btrfs_inode_item *inode_item;
1462
1463         log_root = alloc_log_tree(trans, root->fs_info);
1464         if (IS_ERR(log_root))
1465                 return PTR_ERR(log_root);
1466
1467         log_root->last_trans = trans->transid;
1468         log_root->root_key.offset = root->root_key.objectid;
1469
1470         inode_item = &log_root->root_item.inode;
1471         btrfs_set_stack_inode_generation(inode_item, 1);
1472         btrfs_set_stack_inode_size(inode_item, 3);
1473         btrfs_set_stack_inode_nlink(inode_item, 1);
1474         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1475         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1476
1477         btrfs_set_root_node(&log_root->root_item, log_root->node);
1478
1479         WARN_ON(root->log_root);
1480         root->log_root = log_root;
1481         root->log_transid = 0;
1482         root->log_transid_committed = -1;
1483         root->last_log_commit = 0;
1484         return 0;
1485 }
1486
1487 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1488                                                struct btrfs_key *key)
1489 {
1490         struct btrfs_root *root;
1491         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1492         struct btrfs_path *path;
1493         u64 generation;
1494         int ret;
1495
1496         path = btrfs_alloc_path();
1497         if (!path)
1498                 return ERR_PTR(-ENOMEM);
1499
1500         root = btrfs_alloc_root(fs_info);
1501         if (!root) {
1502                 ret = -ENOMEM;
1503                 goto alloc_fail;
1504         }
1505
1506         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1507                 tree_root->stripesize, root, fs_info, key->objectid);
1508
1509         ret = btrfs_find_root(tree_root, key, path,
1510                               &root->root_item, &root->root_key);
1511         if (ret) {
1512                 if (ret > 0)
1513                         ret = -ENOENT;
1514                 goto find_fail;
1515         }
1516
1517         generation = btrfs_root_generation(&root->root_item);
1518         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1519                                      generation);
1520         if (!root->node) {
1521                 ret = -ENOMEM;
1522                 goto find_fail;
1523         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1524                 ret = -EIO;
1525                 goto read_fail;
1526         }
1527         root->commit_root = btrfs_root_node(root);
1528 out:
1529         btrfs_free_path(path);
1530         return root;
1531
1532 read_fail:
1533         free_extent_buffer(root->node);
1534 find_fail:
1535         kfree(root);
1536 alloc_fail:
1537         root = ERR_PTR(ret);
1538         goto out;
1539 }
1540
1541 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1542                                       struct btrfs_key *location)
1543 {
1544         struct btrfs_root *root;
1545
1546         root = btrfs_read_tree_root(tree_root, location);
1547         if (IS_ERR(root))
1548                 return root;
1549
1550         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1551                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1552                 btrfs_check_and_init_root_item(&root->root_item);
1553         }
1554
1555         return root;
1556 }
1557
1558 int btrfs_init_fs_root(struct btrfs_root *root)
1559 {
1560         int ret;
1561         struct btrfs_subvolume_writers *writers;
1562
1563         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1564         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1565                                         GFP_NOFS);
1566         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1567                 ret = -ENOMEM;
1568                 goto fail;
1569         }
1570
1571         writers = btrfs_alloc_subvolume_writers();
1572         if (IS_ERR(writers)) {
1573                 ret = PTR_ERR(writers);
1574                 goto fail;
1575         }
1576         root->subv_writers = writers;
1577
1578         btrfs_init_free_ino_ctl(root);
1579         spin_lock_init(&root->ino_cache_lock);
1580         init_waitqueue_head(&root->ino_cache_wait);
1581
1582         ret = get_anon_bdev(&root->anon_dev);
1583         if (ret)
1584                 goto free_writers;
1585         return 0;
1586
1587 free_writers:
1588         btrfs_free_subvolume_writers(root->subv_writers);
1589 fail:
1590         kfree(root->free_ino_ctl);
1591         kfree(root->free_ino_pinned);
1592         return ret;
1593 }
1594
1595 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1596                                                u64 root_id)
1597 {
1598         struct btrfs_root *root;
1599
1600         spin_lock(&fs_info->fs_roots_radix_lock);
1601         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1602                                  (unsigned long)root_id);
1603         spin_unlock(&fs_info->fs_roots_radix_lock);
1604         return root;
1605 }
1606
1607 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1608                          struct btrfs_root *root)
1609 {
1610         int ret;
1611
1612         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1613         if (ret)
1614                 return ret;
1615
1616         spin_lock(&fs_info->fs_roots_radix_lock);
1617         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1618                                 (unsigned long)root->root_key.objectid,
1619                                 root);
1620         if (ret == 0)
1621                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1622         spin_unlock(&fs_info->fs_roots_radix_lock);
1623         radix_tree_preload_end();
1624
1625         return ret;
1626 }
1627
1628 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1629                                      struct btrfs_key *location,
1630                                      bool check_ref)
1631 {
1632         struct btrfs_root *root;
1633         struct btrfs_path *path;
1634         struct btrfs_key key;
1635         int ret;
1636
1637         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1638                 return fs_info->tree_root;
1639         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1640                 return fs_info->extent_root;
1641         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1642                 return fs_info->chunk_root;
1643         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1644                 return fs_info->dev_root;
1645         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1646                 return fs_info->csum_root;
1647         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1648                 return fs_info->quota_root ? fs_info->quota_root :
1649                                              ERR_PTR(-ENOENT);
1650         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1651                 return fs_info->uuid_root ? fs_info->uuid_root :
1652                                             ERR_PTR(-ENOENT);
1653 again:
1654         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1655         if (root) {
1656                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1657                         return ERR_PTR(-ENOENT);
1658                 return root;
1659         }
1660
1661         root = btrfs_read_fs_root(fs_info->tree_root, location);
1662         if (IS_ERR(root))
1663                 return root;
1664
1665         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1666                 ret = -ENOENT;
1667                 goto fail;
1668         }
1669
1670         ret = btrfs_init_fs_root(root);
1671         if (ret)
1672                 goto fail;
1673
1674         path = btrfs_alloc_path();
1675         if (!path) {
1676                 ret = -ENOMEM;
1677                 goto fail;
1678         }
1679         key.objectid = BTRFS_ORPHAN_OBJECTID;
1680         key.type = BTRFS_ORPHAN_ITEM_KEY;
1681         key.offset = location->objectid;
1682
1683         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1684         btrfs_free_path(path);
1685         if (ret < 0)
1686                 goto fail;
1687         if (ret == 0)
1688                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1689
1690         ret = btrfs_insert_fs_root(fs_info, root);
1691         if (ret) {
1692                 if (ret == -EEXIST) {
1693                         free_fs_root(root);
1694                         goto again;
1695                 }
1696                 goto fail;
1697         }
1698         return root;
1699 fail:
1700         free_fs_root(root);
1701         return ERR_PTR(ret);
1702 }
1703
1704 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1705 {
1706         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1707         int ret = 0;
1708         struct btrfs_device *device;
1709         struct backing_dev_info *bdi;
1710
1711         rcu_read_lock();
1712         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1713                 if (!device->bdev)
1714                         continue;
1715                 bdi = blk_get_backing_dev_info(device->bdev);
1716                 if (bdi_congested(bdi, bdi_bits)) {
1717                         ret = 1;
1718                         break;
1719                 }
1720         }
1721         rcu_read_unlock();
1722         return ret;
1723 }
1724
1725 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1726 {
1727         int err;
1728
1729         bdi->capabilities = BDI_CAP_MAP_COPY;
1730         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1731         if (err)
1732                 return err;
1733
1734         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1735         bdi->congested_fn       = btrfs_congested_fn;
1736         bdi->congested_data     = info;
1737         return 0;
1738 }
1739
1740 /*
1741  * called by the kthread helper functions to finally call the bio end_io
1742  * functions.  This is where read checksum verification actually happens
1743  */
1744 static void end_workqueue_fn(struct btrfs_work *work)
1745 {
1746         struct bio *bio;
1747         struct btrfs_end_io_wq *end_io_wq;
1748         int error;
1749
1750         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1751         bio = end_io_wq->bio;
1752
1753         error = end_io_wq->error;
1754         bio->bi_private = end_io_wq->private;
1755         bio->bi_end_io = end_io_wq->end_io;
1756         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1757         bio_endio_nodec(bio, error);
1758 }
1759
1760 static int cleaner_kthread(void *arg)
1761 {
1762         struct btrfs_root *root = arg;
1763         int again;
1764
1765         do {
1766                 again = 0;
1767
1768                 /* Make the cleaner go to sleep early. */
1769                 if (btrfs_need_cleaner_sleep(root))
1770                         goto sleep;
1771
1772                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1773                         goto sleep;
1774
1775                 /*
1776                  * Avoid the problem that we change the status of the fs
1777                  * during the above check and trylock.
1778                  */
1779                 if (btrfs_need_cleaner_sleep(root)) {
1780                         mutex_unlock(&root->fs_info->cleaner_mutex);
1781                         goto sleep;
1782                 }
1783
1784                 btrfs_run_delayed_iputs(root);
1785                 btrfs_delete_unused_bgs(root->fs_info);
1786                 again = btrfs_clean_one_deleted_snapshot(root);
1787                 mutex_unlock(&root->fs_info->cleaner_mutex);
1788
1789                 /*
1790                  * The defragger has dealt with the R/O remount and umount,
1791                  * needn't do anything special here.
1792                  */
1793                 btrfs_run_defrag_inodes(root->fs_info);
1794 sleep:
1795                 if (!try_to_freeze() && !again) {
1796                         set_current_state(TASK_INTERRUPTIBLE);
1797                         if (!kthread_should_stop())
1798                                 schedule();
1799                         __set_current_state(TASK_RUNNING);
1800                 }
1801         } while (!kthread_should_stop());
1802         return 0;
1803 }
1804
1805 static int transaction_kthread(void *arg)
1806 {
1807         struct btrfs_root *root = arg;
1808         struct btrfs_trans_handle *trans;
1809         struct btrfs_transaction *cur;
1810         u64 transid;
1811         unsigned long now;
1812         unsigned long delay;
1813         bool cannot_commit;
1814
1815         do {
1816                 cannot_commit = false;
1817                 delay = HZ * root->fs_info->commit_interval;
1818                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1819
1820                 spin_lock(&root->fs_info->trans_lock);
1821                 cur = root->fs_info->running_transaction;
1822                 if (!cur) {
1823                         spin_unlock(&root->fs_info->trans_lock);
1824                         goto sleep;
1825                 }
1826
1827                 now = get_seconds();
1828                 if (cur->state < TRANS_STATE_BLOCKED &&
1829                     (now < cur->start_time ||
1830                      now - cur->start_time < root->fs_info->commit_interval)) {
1831                         spin_unlock(&root->fs_info->trans_lock);
1832                         delay = HZ * 5;
1833                         goto sleep;
1834                 }
1835                 transid = cur->transid;
1836                 spin_unlock(&root->fs_info->trans_lock);
1837
1838                 /* If the file system is aborted, this will always fail. */
1839                 trans = btrfs_attach_transaction(root);
1840                 if (IS_ERR(trans)) {
1841                         if (PTR_ERR(trans) != -ENOENT)
1842                                 cannot_commit = true;
1843                         goto sleep;
1844                 }
1845                 if (transid == trans->transid) {
1846                         btrfs_commit_transaction(trans, root);
1847                 } else {
1848                         btrfs_end_transaction(trans, root);
1849                 }
1850 sleep:
1851                 wake_up_process(root->fs_info->cleaner_kthread);
1852                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1853
1854                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1855                                       &root->fs_info->fs_state)))
1856                         btrfs_cleanup_transaction(root);
1857                 if (!try_to_freeze()) {
1858                         set_current_state(TASK_INTERRUPTIBLE);
1859                         if (!kthread_should_stop() &&
1860                             (!btrfs_transaction_blocked(root->fs_info) ||
1861                              cannot_commit))
1862                                 schedule_timeout(delay);
1863                         __set_current_state(TASK_RUNNING);
1864                 }
1865         } while (!kthread_should_stop());
1866         return 0;
1867 }
1868
1869 /*
1870  * this will find the highest generation in the array of
1871  * root backups.  The index of the highest array is returned,
1872  * or -1 if we can't find anything.
1873  *
1874  * We check to make sure the array is valid by comparing the
1875  * generation of the latest  root in the array with the generation
1876  * in the super block.  If they don't match we pitch it.
1877  */
1878 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1879 {
1880         u64 cur;
1881         int newest_index = -1;
1882         struct btrfs_root_backup *root_backup;
1883         int i;
1884
1885         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1886                 root_backup = info->super_copy->super_roots + i;
1887                 cur = btrfs_backup_tree_root_gen(root_backup);
1888                 if (cur == newest_gen)
1889                         newest_index = i;
1890         }
1891
1892         /* check to see if we actually wrapped around */
1893         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1894                 root_backup = info->super_copy->super_roots;
1895                 cur = btrfs_backup_tree_root_gen(root_backup);
1896                 if (cur == newest_gen)
1897                         newest_index = 0;
1898         }
1899         return newest_index;
1900 }
1901
1902
1903 /*
1904  * find the oldest backup so we know where to store new entries
1905  * in the backup array.  This will set the backup_root_index
1906  * field in the fs_info struct
1907  */
1908 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1909                                      u64 newest_gen)
1910 {
1911         int newest_index = -1;
1912
1913         newest_index = find_newest_super_backup(info, newest_gen);
1914         /* if there was garbage in there, just move along */
1915         if (newest_index == -1) {
1916                 info->backup_root_index = 0;
1917         } else {
1918                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1919         }
1920 }
1921
1922 /*
1923  * copy all the root pointers into the super backup array.
1924  * this will bump the backup pointer by one when it is
1925  * done
1926  */
1927 static void backup_super_roots(struct btrfs_fs_info *info)
1928 {
1929         int next_backup;
1930         struct btrfs_root_backup *root_backup;
1931         int last_backup;
1932
1933         next_backup = info->backup_root_index;
1934         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1935                 BTRFS_NUM_BACKUP_ROOTS;
1936
1937         /*
1938          * just overwrite the last backup if we're at the same generation
1939          * this happens only at umount
1940          */
1941         root_backup = info->super_for_commit->super_roots + last_backup;
1942         if (btrfs_backup_tree_root_gen(root_backup) ==
1943             btrfs_header_generation(info->tree_root->node))
1944                 next_backup = last_backup;
1945
1946         root_backup = info->super_for_commit->super_roots + next_backup;
1947
1948         /*
1949          * make sure all of our padding and empty slots get zero filled
1950          * regardless of which ones we use today
1951          */
1952         memset(root_backup, 0, sizeof(*root_backup));
1953
1954         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1955
1956         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1957         btrfs_set_backup_tree_root_gen(root_backup,
1958                                btrfs_header_generation(info->tree_root->node));
1959
1960         btrfs_set_backup_tree_root_level(root_backup,
1961                                btrfs_header_level(info->tree_root->node));
1962
1963         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1964         btrfs_set_backup_chunk_root_gen(root_backup,
1965                                btrfs_header_generation(info->chunk_root->node));
1966         btrfs_set_backup_chunk_root_level(root_backup,
1967                                btrfs_header_level(info->chunk_root->node));
1968
1969         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1970         btrfs_set_backup_extent_root_gen(root_backup,
1971                                btrfs_header_generation(info->extent_root->node));
1972         btrfs_set_backup_extent_root_level(root_backup,
1973                                btrfs_header_level(info->extent_root->node));
1974
1975         /*
1976          * we might commit during log recovery, which happens before we set
1977          * the fs_root.  Make sure it is valid before we fill it in.
1978          */
1979         if (info->fs_root && info->fs_root->node) {
1980                 btrfs_set_backup_fs_root(root_backup,
1981                                          info->fs_root->node->start);
1982                 btrfs_set_backup_fs_root_gen(root_backup,
1983                                btrfs_header_generation(info->fs_root->node));
1984                 btrfs_set_backup_fs_root_level(root_backup,
1985                                btrfs_header_level(info->fs_root->node));
1986         }
1987
1988         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1989         btrfs_set_backup_dev_root_gen(root_backup,
1990                                btrfs_header_generation(info->dev_root->node));
1991         btrfs_set_backup_dev_root_level(root_backup,
1992                                        btrfs_header_level(info->dev_root->node));
1993
1994         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1995         btrfs_set_backup_csum_root_gen(root_backup,
1996                                btrfs_header_generation(info->csum_root->node));
1997         btrfs_set_backup_csum_root_level(root_backup,
1998                                btrfs_header_level(info->csum_root->node));
1999
2000         btrfs_set_backup_total_bytes(root_backup,
2001                              btrfs_super_total_bytes(info->super_copy));
2002         btrfs_set_backup_bytes_used(root_backup,
2003                              btrfs_super_bytes_used(info->super_copy));
2004         btrfs_set_backup_num_devices(root_backup,
2005                              btrfs_super_num_devices(info->super_copy));
2006
2007         /*
2008          * if we don't copy this out to the super_copy, it won't get remembered
2009          * for the next commit
2010          */
2011         memcpy(&info->super_copy->super_roots,
2012                &info->super_for_commit->super_roots,
2013                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2014 }
2015
2016 /*
2017  * this copies info out of the root backup array and back into
2018  * the in-memory super block.  It is meant to help iterate through
2019  * the array, so you send it the number of backups you've already
2020  * tried and the last backup index you used.
2021  *
2022  * this returns -1 when it has tried all the backups
2023  */
2024 static noinline int next_root_backup(struct btrfs_fs_info *info,
2025                                      struct btrfs_super_block *super,
2026                                      int *num_backups_tried, int *backup_index)
2027 {
2028         struct btrfs_root_backup *root_backup;
2029         int newest = *backup_index;
2030
2031         if (*num_backups_tried == 0) {
2032                 u64 gen = btrfs_super_generation(super);
2033
2034                 newest = find_newest_super_backup(info, gen);
2035                 if (newest == -1)
2036                         return -1;
2037
2038                 *backup_index = newest;
2039                 *num_backups_tried = 1;
2040         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2041                 /* we've tried all the backups, all done */
2042                 return -1;
2043         } else {
2044                 /* jump to the next oldest backup */
2045                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2046                         BTRFS_NUM_BACKUP_ROOTS;
2047                 *backup_index = newest;
2048                 *num_backups_tried += 1;
2049         }
2050         root_backup = super->super_roots + newest;
2051
2052         btrfs_set_super_generation(super,
2053                                    btrfs_backup_tree_root_gen(root_backup));
2054         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2055         btrfs_set_super_root_level(super,
2056                                    btrfs_backup_tree_root_level(root_backup));
2057         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2058
2059         /*
2060          * fixme: the total bytes and num_devices need to match or we should
2061          * need a fsck
2062          */
2063         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2064         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2065         return 0;
2066 }
2067
2068 /* helper to cleanup workers */
2069 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2070 {
2071         btrfs_destroy_workqueue(fs_info->fixup_workers);
2072         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2073         btrfs_destroy_workqueue(fs_info->workers);
2074         btrfs_destroy_workqueue(fs_info->endio_workers);
2075         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2076         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2077         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2078         btrfs_destroy_workqueue(fs_info->rmw_workers);
2079         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2080         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2081         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2082         btrfs_destroy_workqueue(fs_info->submit_workers);
2083         btrfs_destroy_workqueue(fs_info->delayed_workers);
2084         btrfs_destroy_workqueue(fs_info->caching_workers);
2085         btrfs_destroy_workqueue(fs_info->readahead_workers);
2086         btrfs_destroy_workqueue(fs_info->flush_workers);
2087         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2088         btrfs_destroy_workqueue(fs_info->extent_workers);
2089 }
2090
2091 static void free_root_extent_buffers(struct btrfs_root *root)
2092 {
2093         if (root) {
2094                 free_extent_buffer(root->node);
2095                 free_extent_buffer(root->commit_root);
2096                 root->node = NULL;
2097                 root->commit_root = NULL;
2098         }
2099 }
2100
2101 /* helper to cleanup tree roots */
2102 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103 {
2104         free_root_extent_buffers(info->tree_root);
2105
2106         free_root_extent_buffers(info->dev_root);
2107         free_root_extent_buffers(info->extent_root);
2108         free_root_extent_buffers(info->csum_root);
2109         free_root_extent_buffers(info->quota_root);
2110         free_root_extent_buffers(info->uuid_root);
2111         if (chunk_root)
2112                 free_root_extent_buffers(info->chunk_root);
2113 }
2114
2115 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2116 {
2117         int ret;
2118         struct btrfs_root *gang[8];
2119         int i;
2120
2121         while (!list_empty(&fs_info->dead_roots)) {
2122                 gang[0] = list_entry(fs_info->dead_roots.next,
2123                                      struct btrfs_root, root_list);
2124                 list_del(&gang[0]->root_list);
2125
2126                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2127                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2128                 } else {
2129                         free_extent_buffer(gang[0]->node);
2130                         free_extent_buffer(gang[0]->commit_root);
2131                         btrfs_put_fs_root(gang[0]);
2132                 }
2133         }
2134
2135         while (1) {
2136                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2137                                              (void **)gang, 0,
2138                                              ARRAY_SIZE(gang));
2139                 if (!ret)
2140                         break;
2141                 for (i = 0; i < ret; i++)
2142                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2143         }
2144
2145         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2146                 btrfs_free_log_root_tree(NULL, fs_info);
2147                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2148                                             fs_info->pinned_extents);
2149         }
2150 }
2151
2152 int open_ctree(struct super_block *sb,
2153                struct btrfs_fs_devices *fs_devices,
2154                char *options)
2155 {
2156         u32 sectorsize;
2157         u32 nodesize;
2158         u32 stripesize;
2159         u64 generation;
2160         u64 features;
2161         struct btrfs_key location;
2162         struct buffer_head *bh;
2163         struct btrfs_super_block *disk_super;
2164         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2165         struct btrfs_root *tree_root;
2166         struct btrfs_root *extent_root;
2167         struct btrfs_root *csum_root;
2168         struct btrfs_root *chunk_root;
2169         struct btrfs_root *dev_root;
2170         struct btrfs_root *quota_root;
2171         struct btrfs_root *uuid_root;
2172         struct btrfs_root *log_tree_root;
2173         int ret;
2174         int err = -EINVAL;
2175         int num_backups_tried = 0;
2176         int backup_index = 0;
2177         int max_active;
2178         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2179         bool create_uuid_tree;
2180         bool check_uuid_tree;
2181
2182         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2183         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2184         if (!tree_root || !chunk_root) {
2185                 err = -ENOMEM;
2186                 goto fail;
2187         }
2188
2189         ret = init_srcu_struct(&fs_info->subvol_srcu);
2190         if (ret) {
2191                 err = ret;
2192                 goto fail;
2193         }
2194
2195         ret = setup_bdi(fs_info, &fs_info->bdi);
2196         if (ret) {
2197                 err = ret;
2198                 goto fail_srcu;
2199         }
2200
2201         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2202         if (ret) {
2203                 err = ret;
2204                 goto fail_bdi;
2205         }
2206         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2207                                         (1 + ilog2(nr_cpu_ids));
2208
2209         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2210         if (ret) {
2211                 err = ret;
2212                 goto fail_dirty_metadata_bytes;
2213         }
2214
2215         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2216         if (ret) {
2217                 err = ret;
2218                 goto fail_delalloc_bytes;
2219         }
2220
2221         fs_info->btree_inode = new_inode(sb);
2222         if (!fs_info->btree_inode) {
2223                 err = -ENOMEM;
2224                 goto fail_bio_counter;
2225         }
2226
2227         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2228
2229         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2230         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2231         INIT_LIST_HEAD(&fs_info->trans_list);
2232         INIT_LIST_HEAD(&fs_info->dead_roots);
2233         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2234         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2235         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2236         spin_lock_init(&fs_info->delalloc_root_lock);
2237         spin_lock_init(&fs_info->trans_lock);
2238         spin_lock_init(&fs_info->fs_roots_radix_lock);
2239         spin_lock_init(&fs_info->delayed_iput_lock);
2240         spin_lock_init(&fs_info->defrag_inodes_lock);
2241         spin_lock_init(&fs_info->free_chunk_lock);
2242         spin_lock_init(&fs_info->tree_mod_seq_lock);
2243         spin_lock_init(&fs_info->super_lock);
2244         spin_lock_init(&fs_info->qgroup_op_lock);
2245         spin_lock_init(&fs_info->buffer_lock);
2246         spin_lock_init(&fs_info->unused_bgs_lock);
2247         rwlock_init(&fs_info->tree_mod_log_lock);
2248         mutex_init(&fs_info->reloc_mutex);
2249         mutex_init(&fs_info->delalloc_root_mutex);
2250         seqlock_init(&fs_info->profiles_lock);
2251
2252         init_completion(&fs_info->kobj_unregister);
2253         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2254         INIT_LIST_HEAD(&fs_info->space_info);
2255         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2256         INIT_LIST_HEAD(&fs_info->unused_bgs);
2257         btrfs_mapping_init(&fs_info->mapping_tree);
2258         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2259                              BTRFS_BLOCK_RSV_GLOBAL);
2260         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2261                              BTRFS_BLOCK_RSV_DELALLOC);
2262         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2263         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2264         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2265         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2266                              BTRFS_BLOCK_RSV_DELOPS);
2267         atomic_set(&fs_info->nr_async_submits, 0);
2268         atomic_set(&fs_info->async_delalloc_pages, 0);
2269         atomic_set(&fs_info->async_submit_draining, 0);
2270         atomic_set(&fs_info->nr_async_bios, 0);
2271         atomic_set(&fs_info->defrag_running, 0);
2272         atomic_set(&fs_info->qgroup_op_seq, 0);
2273         atomic64_set(&fs_info->tree_mod_seq, 0);
2274         fs_info->sb = sb;
2275         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2276         fs_info->metadata_ratio = 0;
2277         fs_info->defrag_inodes = RB_ROOT;
2278         fs_info->free_chunk_space = 0;
2279         fs_info->tree_mod_log = RB_ROOT;
2280         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2281         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2282         /* readahead state */
2283         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2284         spin_lock_init(&fs_info->reada_lock);
2285
2286         fs_info->thread_pool_size = min_t(unsigned long,
2287                                           num_online_cpus() + 2, 8);
2288
2289         INIT_LIST_HEAD(&fs_info->ordered_roots);
2290         spin_lock_init(&fs_info->ordered_root_lock);
2291         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2292                                         GFP_NOFS);
2293         if (!fs_info->delayed_root) {
2294                 err = -ENOMEM;
2295                 goto fail_iput;
2296         }
2297         btrfs_init_delayed_root(fs_info->delayed_root);
2298
2299         mutex_init(&fs_info->scrub_lock);
2300         atomic_set(&fs_info->scrubs_running, 0);
2301         atomic_set(&fs_info->scrub_pause_req, 0);
2302         atomic_set(&fs_info->scrubs_paused, 0);
2303         atomic_set(&fs_info->scrub_cancel_req, 0);
2304         init_waitqueue_head(&fs_info->replace_wait);
2305         init_waitqueue_head(&fs_info->scrub_pause_wait);
2306         fs_info->scrub_workers_refcnt = 0;
2307 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2308         fs_info->check_integrity_print_mask = 0;
2309 #endif
2310
2311         spin_lock_init(&fs_info->balance_lock);
2312         mutex_init(&fs_info->balance_mutex);
2313         atomic_set(&fs_info->balance_running, 0);
2314         atomic_set(&fs_info->balance_pause_req, 0);
2315         atomic_set(&fs_info->balance_cancel_req, 0);
2316         fs_info->balance_ctl = NULL;
2317         init_waitqueue_head(&fs_info->balance_wait_q);
2318         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2319
2320         sb->s_blocksize = 4096;
2321         sb->s_blocksize_bits = blksize_bits(4096);
2322         sb->s_bdi = &fs_info->bdi;
2323
2324         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2325         set_nlink(fs_info->btree_inode, 1);
2326         /*
2327          * we set the i_size on the btree inode to the max possible int.
2328          * the real end of the address space is determined by all of
2329          * the devices in the system
2330          */
2331         fs_info->btree_inode->i_size = OFFSET_MAX;
2332         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2333         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2334
2335         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2336         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2337                              fs_info->btree_inode->i_mapping);
2338         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2339         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2340
2341         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2342
2343         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2344         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2345                sizeof(struct btrfs_key));
2346         set_bit(BTRFS_INODE_DUMMY,
2347                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2348         btrfs_insert_inode_hash(fs_info->btree_inode);
2349
2350         spin_lock_init(&fs_info->block_group_cache_lock);
2351         fs_info->block_group_cache_tree = RB_ROOT;
2352         fs_info->first_logical_byte = (u64)-1;
2353
2354         extent_io_tree_init(&fs_info->freed_extents[0],
2355                              fs_info->btree_inode->i_mapping);
2356         extent_io_tree_init(&fs_info->freed_extents[1],
2357                              fs_info->btree_inode->i_mapping);
2358         fs_info->pinned_extents = &fs_info->freed_extents[0];
2359         fs_info->do_barriers = 1;
2360
2361
2362         mutex_init(&fs_info->ordered_operations_mutex);
2363         mutex_init(&fs_info->ordered_extent_flush_mutex);
2364         mutex_init(&fs_info->tree_log_mutex);
2365         mutex_init(&fs_info->chunk_mutex);
2366         mutex_init(&fs_info->transaction_kthread_mutex);
2367         mutex_init(&fs_info->cleaner_mutex);
2368         mutex_init(&fs_info->volume_mutex);
2369         init_rwsem(&fs_info->commit_root_sem);
2370         init_rwsem(&fs_info->cleanup_work_sem);
2371         init_rwsem(&fs_info->subvol_sem);
2372         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2373         fs_info->dev_replace.lock_owner = 0;
2374         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2375         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2376         mutex_init(&fs_info->dev_replace.lock_management_lock);
2377         mutex_init(&fs_info->dev_replace.lock);
2378
2379         spin_lock_init(&fs_info->qgroup_lock);
2380         mutex_init(&fs_info->qgroup_ioctl_lock);
2381         fs_info->qgroup_tree = RB_ROOT;
2382         fs_info->qgroup_op_tree = RB_ROOT;
2383         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2384         fs_info->qgroup_seq = 1;
2385         fs_info->quota_enabled = 0;
2386         fs_info->pending_quota_state = 0;
2387         fs_info->qgroup_ulist = NULL;
2388         mutex_init(&fs_info->qgroup_rescan_lock);
2389
2390         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2391         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2392
2393         init_waitqueue_head(&fs_info->transaction_throttle);
2394         init_waitqueue_head(&fs_info->transaction_wait);
2395         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2396         init_waitqueue_head(&fs_info->async_submit_wait);
2397
2398         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2399
2400         ret = btrfs_alloc_stripe_hash_table(fs_info);
2401         if (ret) {
2402                 err = ret;
2403                 goto fail_alloc;
2404         }
2405
2406         __setup_root(4096, 4096, 4096, tree_root,
2407                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2408
2409         invalidate_bdev(fs_devices->latest_bdev);
2410
2411         /*
2412          * Read super block and check the signature bytes only
2413          */
2414         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2415         if (!bh) {
2416                 err = -EINVAL;
2417                 goto fail_alloc;
2418         }
2419
2420         /*
2421          * We want to check superblock checksum, the type is stored inside.
2422          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2423          */
2424         if (btrfs_check_super_csum(bh->b_data)) {
2425                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2426                 err = -EINVAL;
2427                 goto fail_alloc;
2428         }
2429
2430         /*
2431          * super_copy is zeroed at allocation time and we never touch the
2432          * following bytes up to INFO_SIZE, the checksum is calculated from
2433          * the whole block of INFO_SIZE
2434          */
2435         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2436         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2437                sizeof(*fs_info->super_for_commit));
2438         brelse(bh);
2439
2440         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2441
2442         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2443         if (ret) {
2444                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2445                 err = -EINVAL;
2446                 goto fail_alloc;
2447         }
2448
2449         disk_super = fs_info->super_copy;
2450         if (!btrfs_super_root(disk_super))
2451                 goto fail_alloc;
2452
2453         /* check FS state, whether FS is broken. */
2454         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2455                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2456
2457         /*
2458          * run through our array of backup supers and setup
2459          * our ring pointer to the oldest one
2460          */
2461         generation = btrfs_super_generation(disk_super);
2462         find_oldest_super_backup(fs_info, generation);
2463
2464         /*
2465          * In the long term, we'll store the compression type in the super
2466          * block, and it'll be used for per file compression control.
2467          */
2468         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2469
2470         ret = btrfs_parse_options(tree_root, options);
2471         if (ret) {
2472                 err = ret;
2473                 goto fail_alloc;
2474         }
2475
2476         features = btrfs_super_incompat_flags(disk_super) &
2477                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2478         if (features) {
2479                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2480                        "unsupported optional features (%Lx).\n",
2481                        features);
2482                 err = -EINVAL;
2483                 goto fail_alloc;
2484         }
2485
2486         /*
2487          * Leafsize and nodesize were always equal, this is only a sanity check.
2488          */
2489         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2490             btrfs_super_nodesize(disk_super)) {
2491                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2492                        "blocksizes don't match.  node %d leaf %d\n",
2493                        btrfs_super_nodesize(disk_super),
2494                        le32_to_cpu(disk_super->__unused_leafsize));
2495                 err = -EINVAL;
2496                 goto fail_alloc;
2497         }
2498         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2499                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2500                        "blocksize (%d) was too large\n",
2501                        btrfs_super_nodesize(disk_super));
2502                 err = -EINVAL;
2503                 goto fail_alloc;
2504         }
2505
2506         features = btrfs_super_incompat_flags(disk_super);
2507         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2508         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2509                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2510
2511         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2512                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2513
2514         /*
2515          * flag our filesystem as having big metadata blocks if
2516          * they are bigger than the page size
2517          */
2518         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2519                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2520                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2521                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2522         }
2523
2524         nodesize = btrfs_super_nodesize(disk_super);
2525         sectorsize = btrfs_super_sectorsize(disk_super);
2526         stripesize = btrfs_super_stripesize(disk_super);
2527         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2528         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2529
2530         /*
2531          * mixed block groups end up with duplicate but slightly offset
2532          * extent buffers for the same range.  It leads to corruptions
2533          */
2534         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2535             (sectorsize != nodesize)) {
2536                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2537                                 "are not allowed for mixed block groups on %s\n",
2538                                 sb->s_id);
2539                 goto fail_alloc;
2540         }
2541
2542         /*
2543          * Needn't use the lock because there is no other task which will
2544          * update the flag.
2545          */
2546         btrfs_set_super_incompat_flags(disk_super, features);
2547
2548         features = btrfs_super_compat_ro_flags(disk_super) &
2549                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2550         if (!(sb->s_flags & MS_RDONLY) && features) {
2551                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2552                        "unsupported option features (%Lx).\n",
2553                        features);
2554                 err = -EINVAL;
2555                 goto fail_alloc;
2556         }
2557
2558         max_active = fs_info->thread_pool_size;
2559
2560         fs_info->workers =
2561                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2562                                       max_active, 16);
2563
2564         fs_info->delalloc_workers =
2565                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2566
2567         fs_info->flush_workers =
2568                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2569
2570         fs_info->caching_workers =
2571                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2572
2573         /*
2574          * a higher idle thresh on the submit workers makes it much more
2575          * likely that bios will be send down in a sane order to the
2576          * devices
2577          */
2578         fs_info->submit_workers =
2579                 btrfs_alloc_workqueue("submit", flags,
2580                                       min_t(u64, fs_devices->num_devices,
2581                                             max_active), 64);
2582
2583         fs_info->fixup_workers =
2584                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2585
2586         /*
2587          * endios are largely parallel and should have a very
2588          * low idle thresh
2589          */
2590         fs_info->endio_workers =
2591                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2592         fs_info->endio_meta_workers =
2593                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2594         fs_info->endio_meta_write_workers =
2595                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2596         fs_info->endio_raid56_workers =
2597                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2598         fs_info->endio_repair_workers =
2599                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2600         fs_info->rmw_workers =
2601                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2602         fs_info->endio_write_workers =
2603                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2604         fs_info->endio_freespace_worker =
2605                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2606         fs_info->delayed_workers =
2607                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2608         fs_info->readahead_workers =
2609                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2610         fs_info->qgroup_rescan_workers =
2611                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2612         fs_info->extent_workers =
2613                 btrfs_alloc_workqueue("extent-refs", flags,
2614                                       min_t(u64, fs_devices->num_devices,
2615                                             max_active), 8);
2616
2617         if (!(fs_info->workers && fs_info->delalloc_workers &&
2618               fs_info->submit_workers && fs_info->flush_workers &&
2619               fs_info->endio_workers && fs_info->endio_meta_workers &&
2620               fs_info->endio_meta_write_workers &&
2621               fs_info->endio_repair_workers &&
2622               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2623               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2624               fs_info->caching_workers && fs_info->readahead_workers &&
2625               fs_info->fixup_workers && fs_info->delayed_workers &&
2626               fs_info->extent_workers &&
2627               fs_info->qgroup_rescan_workers)) {
2628                 err = -ENOMEM;
2629                 goto fail_sb_buffer;
2630         }
2631
2632         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2633         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2634                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2635
2636         tree_root->nodesize = nodesize;
2637         tree_root->sectorsize = sectorsize;
2638         tree_root->stripesize = stripesize;
2639
2640         sb->s_blocksize = sectorsize;
2641         sb->s_blocksize_bits = blksize_bits(sectorsize);
2642
2643         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2644                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2645                 goto fail_sb_buffer;
2646         }
2647
2648         if (sectorsize != PAGE_SIZE) {
2649                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2650                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2651                 goto fail_sb_buffer;
2652         }
2653
2654         mutex_lock(&fs_info->chunk_mutex);
2655         ret = btrfs_read_sys_array(tree_root);
2656         mutex_unlock(&fs_info->chunk_mutex);
2657         if (ret) {
2658                 printk(KERN_WARNING "BTRFS: failed to read the system "
2659                        "array on %s\n", sb->s_id);
2660                 goto fail_sb_buffer;
2661         }
2662
2663         generation = btrfs_super_chunk_root_generation(disk_super);
2664
2665         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2666                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2667
2668         chunk_root->node = read_tree_block(chunk_root,
2669                                            btrfs_super_chunk_root(disk_super),
2670                                            generation);
2671         if (!chunk_root->node ||
2672             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2673                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2674                        sb->s_id);
2675                 goto fail_tree_roots;
2676         }
2677         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2678         chunk_root->commit_root = btrfs_root_node(chunk_root);
2679
2680         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2681            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2682
2683         ret = btrfs_read_chunk_tree(chunk_root);
2684         if (ret) {
2685                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2686                        sb->s_id);
2687                 goto fail_tree_roots;
2688         }
2689
2690         /*
2691          * keep the device that is marked to be the target device for the
2692          * dev_replace procedure
2693          */
2694         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2695
2696         if (!fs_devices->latest_bdev) {
2697                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2698                        sb->s_id);
2699                 goto fail_tree_roots;
2700         }
2701
2702 retry_root_backup:
2703         generation = btrfs_super_generation(disk_super);
2704
2705         tree_root->node = read_tree_block(tree_root,
2706                                           btrfs_super_root(disk_super),
2707                                           generation);
2708         if (!tree_root->node ||
2709             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2710                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2711                        sb->s_id);
2712
2713                 goto recovery_tree_root;
2714         }
2715
2716         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2717         tree_root->commit_root = btrfs_root_node(tree_root);
2718         btrfs_set_root_refs(&tree_root->root_item, 1);
2719
2720         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2721         location.type = BTRFS_ROOT_ITEM_KEY;
2722         location.offset = 0;
2723
2724         extent_root = btrfs_read_tree_root(tree_root, &location);
2725         if (IS_ERR(extent_root)) {
2726                 ret = PTR_ERR(extent_root);
2727                 goto recovery_tree_root;
2728         }
2729         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2730         fs_info->extent_root = extent_root;
2731
2732         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2733         dev_root = btrfs_read_tree_root(tree_root, &location);
2734         if (IS_ERR(dev_root)) {
2735                 ret = PTR_ERR(dev_root);
2736                 goto recovery_tree_root;
2737         }
2738         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2739         fs_info->dev_root = dev_root;
2740         btrfs_init_devices_late(fs_info);
2741
2742         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2743         csum_root = btrfs_read_tree_root(tree_root, &location);
2744         if (IS_ERR(csum_root)) {
2745                 ret = PTR_ERR(csum_root);
2746                 goto recovery_tree_root;
2747         }
2748         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2749         fs_info->csum_root = csum_root;
2750
2751         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2752         quota_root = btrfs_read_tree_root(tree_root, &location);
2753         if (!IS_ERR(quota_root)) {
2754                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2755                 fs_info->quota_enabled = 1;
2756                 fs_info->pending_quota_state = 1;
2757                 fs_info->quota_root = quota_root;
2758         }
2759
2760         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2761         uuid_root = btrfs_read_tree_root(tree_root, &location);
2762         if (IS_ERR(uuid_root)) {
2763                 ret = PTR_ERR(uuid_root);
2764                 if (ret != -ENOENT)
2765                         goto recovery_tree_root;
2766                 create_uuid_tree = true;
2767                 check_uuid_tree = false;
2768         } else {
2769                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2770                 fs_info->uuid_root = uuid_root;
2771                 create_uuid_tree = false;
2772                 check_uuid_tree =
2773                     generation != btrfs_super_uuid_tree_generation(disk_super);
2774         }
2775
2776         fs_info->generation = generation;
2777         fs_info->last_trans_committed = generation;
2778
2779         ret = btrfs_recover_balance(fs_info);
2780         if (ret) {
2781                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2782                 goto fail_block_groups;
2783         }
2784
2785         ret = btrfs_init_dev_stats(fs_info);
2786         if (ret) {
2787                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2788                        ret);
2789                 goto fail_block_groups;
2790         }
2791
2792         ret = btrfs_init_dev_replace(fs_info);
2793         if (ret) {
2794                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2795                 goto fail_block_groups;
2796         }
2797
2798         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2799
2800         ret = btrfs_sysfs_add_one(fs_info);
2801         if (ret) {
2802                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2803                 goto fail_block_groups;
2804         }
2805
2806         ret = btrfs_init_space_info(fs_info);
2807         if (ret) {
2808                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2809                 goto fail_sysfs;
2810         }
2811
2812         ret = btrfs_read_block_groups(extent_root);
2813         if (ret) {
2814                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2815                 goto fail_sysfs;
2816         }
2817         fs_info->num_tolerated_disk_barrier_failures =
2818                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2819         if (fs_info->fs_devices->missing_devices >
2820              fs_info->num_tolerated_disk_barrier_failures &&
2821             !(sb->s_flags & MS_RDONLY)) {
2822                 printk(KERN_WARNING "BTRFS: "
2823                         "too many missing devices, writeable mount is not allowed\n");
2824                 goto fail_sysfs;
2825         }
2826
2827         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2828                                                "btrfs-cleaner");
2829         if (IS_ERR(fs_info->cleaner_kthread))
2830                 goto fail_sysfs;
2831
2832         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2833                                                    tree_root,
2834                                                    "btrfs-transaction");
2835         if (IS_ERR(fs_info->transaction_kthread))
2836                 goto fail_cleaner;
2837
2838         if (!btrfs_test_opt(tree_root, SSD) &&
2839             !btrfs_test_opt(tree_root, NOSSD) &&
2840             !fs_info->fs_devices->rotating) {
2841                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2842                        "mode\n");
2843                 btrfs_set_opt(fs_info->mount_opt, SSD);
2844         }
2845
2846         /*
2847          * Mount does not set all options immediatelly, we can do it now and do
2848          * not have to wait for transaction commit
2849          */
2850         btrfs_apply_pending_changes(fs_info);
2851
2852 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2853         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2854                 ret = btrfsic_mount(tree_root, fs_devices,
2855                                     btrfs_test_opt(tree_root,
2856                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2857                                     1 : 0,
2858                                     fs_info->check_integrity_print_mask);
2859                 if (ret)
2860                         printk(KERN_WARNING "BTRFS: failed to initialize"
2861                                " integrity check module %s\n", sb->s_id);
2862         }
2863 #endif
2864         ret = btrfs_read_qgroup_config(fs_info);
2865         if (ret)
2866                 goto fail_trans_kthread;
2867
2868         /* do not make disk changes in broken FS */
2869         if (btrfs_super_log_root(disk_super) != 0) {
2870                 u64 bytenr = btrfs_super_log_root(disk_super);
2871
2872                 if (fs_devices->rw_devices == 0) {
2873                         printk(KERN_WARNING "BTRFS: log replay required "
2874                                "on RO media\n");
2875                         err = -EIO;
2876                         goto fail_qgroup;
2877                 }
2878
2879                 log_tree_root = btrfs_alloc_root(fs_info);
2880                 if (!log_tree_root) {
2881                         err = -ENOMEM;
2882                         goto fail_qgroup;
2883                 }
2884
2885                 __setup_root(nodesize, sectorsize, stripesize,
2886                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2887
2888                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2889                                                       generation + 1);
2890                 if (!log_tree_root->node ||
2891                     !extent_buffer_uptodate(log_tree_root->node)) {
2892                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2893                         free_extent_buffer(log_tree_root->node);
2894                         kfree(log_tree_root);
2895                         goto fail_qgroup;
2896                 }
2897                 /* returns with log_tree_root freed on success */
2898                 ret = btrfs_recover_log_trees(log_tree_root);
2899                 if (ret) {
2900                         btrfs_error(tree_root->fs_info, ret,
2901                                     "Failed to recover log tree");
2902                         free_extent_buffer(log_tree_root->node);
2903                         kfree(log_tree_root);
2904                         goto fail_qgroup;
2905                 }
2906
2907                 if (sb->s_flags & MS_RDONLY) {
2908                         ret = btrfs_commit_super(tree_root);
2909                         if (ret)
2910                                 goto fail_qgroup;
2911                 }
2912         }
2913
2914         ret = btrfs_find_orphan_roots(tree_root);
2915         if (ret)
2916                 goto fail_qgroup;
2917
2918         if (!(sb->s_flags & MS_RDONLY)) {
2919                 ret = btrfs_cleanup_fs_roots(fs_info);
2920                 if (ret)
2921                         goto fail_qgroup;
2922
2923                 mutex_lock(&fs_info->cleaner_mutex);
2924                 ret = btrfs_recover_relocation(tree_root);
2925                 mutex_unlock(&fs_info->cleaner_mutex);
2926                 if (ret < 0) {
2927                         printk(KERN_WARNING
2928                                "BTRFS: failed to recover relocation\n");
2929                         err = -EINVAL;
2930                         goto fail_qgroup;
2931                 }
2932         }
2933
2934         location.objectid = BTRFS_FS_TREE_OBJECTID;
2935         location.type = BTRFS_ROOT_ITEM_KEY;
2936         location.offset = 0;
2937
2938         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2939         if (IS_ERR(fs_info->fs_root)) {
2940                 err = PTR_ERR(fs_info->fs_root);
2941                 goto fail_qgroup;
2942         }
2943
2944         if (sb->s_flags & MS_RDONLY)
2945                 return 0;
2946
2947         down_read(&fs_info->cleanup_work_sem);
2948         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2949             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2950                 up_read(&fs_info->cleanup_work_sem);
2951                 close_ctree(tree_root);
2952                 return ret;
2953         }
2954         up_read(&fs_info->cleanup_work_sem);
2955
2956         ret = btrfs_resume_balance_async(fs_info);
2957         if (ret) {
2958                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2959                 close_ctree(tree_root);
2960                 return ret;
2961         }
2962
2963         ret = btrfs_resume_dev_replace_async(fs_info);
2964         if (ret) {
2965                 pr_warn("BTRFS: failed to resume dev_replace\n");
2966                 close_ctree(tree_root);
2967                 return ret;
2968         }
2969
2970         btrfs_qgroup_rescan_resume(fs_info);
2971
2972         if (create_uuid_tree) {
2973                 pr_info("BTRFS: creating UUID tree\n");
2974                 ret = btrfs_create_uuid_tree(fs_info);
2975                 if (ret) {
2976                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2977                                 ret);
2978                         close_ctree(tree_root);
2979                         return ret;
2980                 }
2981         } else if (check_uuid_tree ||
2982                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2983                 pr_info("BTRFS: checking UUID tree\n");
2984                 ret = btrfs_check_uuid_tree(fs_info);
2985                 if (ret) {
2986                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2987                                 ret);
2988                         close_ctree(tree_root);
2989                         return ret;
2990                 }
2991         } else {
2992                 fs_info->update_uuid_tree_gen = 1;
2993         }
2994
2995         fs_info->open = 1;
2996
2997         return 0;
2998
2999 fail_qgroup:
3000         btrfs_free_qgroup_config(fs_info);
3001 fail_trans_kthread:
3002         kthread_stop(fs_info->transaction_kthread);
3003         btrfs_cleanup_transaction(fs_info->tree_root);
3004         btrfs_free_fs_roots(fs_info);
3005 fail_cleaner:
3006         kthread_stop(fs_info->cleaner_kthread);
3007
3008         /*
3009          * make sure we're done with the btree inode before we stop our
3010          * kthreads
3011          */
3012         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3013
3014 fail_sysfs:
3015         btrfs_sysfs_remove_one(fs_info);
3016
3017 fail_block_groups:
3018         btrfs_put_block_group_cache(fs_info);
3019         btrfs_free_block_groups(fs_info);
3020
3021 fail_tree_roots:
3022         free_root_pointers(fs_info, 1);
3023         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3024
3025 fail_sb_buffer:
3026         btrfs_stop_all_workers(fs_info);
3027 fail_alloc:
3028 fail_iput:
3029         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3030
3031         iput(fs_info->btree_inode);
3032 fail_bio_counter:
3033         percpu_counter_destroy(&fs_info->bio_counter);
3034 fail_delalloc_bytes:
3035         percpu_counter_destroy(&fs_info->delalloc_bytes);
3036 fail_dirty_metadata_bytes:
3037         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3038 fail_bdi:
3039         bdi_destroy(&fs_info->bdi);
3040 fail_srcu:
3041         cleanup_srcu_struct(&fs_info->subvol_srcu);
3042 fail:
3043         btrfs_free_stripe_hash_table(fs_info);
3044         btrfs_close_devices(fs_info->fs_devices);
3045         return err;
3046
3047 recovery_tree_root:
3048         if (!btrfs_test_opt(tree_root, RECOVERY))
3049                 goto fail_tree_roots;
3050
3051         free_root_pointers(fs_info, 0);
3052
3053         /* don't use the log in recovery mode, it won't be valid */
3054         btrfs_set_super_log_root(disk_super, 0);
3055
3056         /* we can't trust the free space cache either */
3057         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3058
3059         ret = next_root_backup(fs_info, fs_info->super_copy,
3060                                &num_backups_tried, &backup_index);
3061         if (ret == -1)
3062                 goto fail_block_groups;
3063         goto retry_root_backup;
3064 }
3065
3066 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3067 {
3068         if (uptodate) {
3069                 set_buffer_uptodate(bh);
3070         } else {
3071                 struct btrfs_device *device = (struct btrfs_device *)
3072                         bh->b_private;
3073
3074                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3075                                           "I/O error on %s\n",
3076                                           rcu_str_deref(device->name));
3077                 /* note, we dont' set_buffer_write_io_error because we have
3078                  * our own ways of dealing with the IO errors
3079                  */
3080                 clear_buffer_uptodate(bh);
3081                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3082         }
3083         unlock_buffer(bh);
3084         put_bh(bh);
3085 }
3086
3087 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3088 {
3089         struct buffer_head *bh;
3090         struct buffer_head *latest = NULL;
3091         struct btrfs_super_block *super;
3092         int i;
3093         u64 transid = 0;
3094         u64 bytenr;
3095
3096         /* we would like to check all the supers, but that would make
3097          * a btrfs mount succeed after a mkfs from a different FS.
3098          * So, we need to add a special mount option to scan for
3099          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3100          */
3101         for (i = 0; i < 1; i++) {
3102                 bytenr = btrfs_sb_offset(i);
3103                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3104                                         i_size_read(bdev->bd_inode))
3105                         break;
3106                 bh = __bread(bdev, bytenr / 4096,
3107                                         BTRFS_SUPER_INFO_SIZE);
3108                 if (!bh)
3109                         continue;
3110
3111                 super = (struct btrfs_super_block *)bh->b_data;
3112                 if (btrfs_super_bytenr(super) != bytenr ||
3113                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3114                         brelse(bh);
3115                         continue;
3116                 }
3117
3118                 if (!latest || btrfs_super_generation(super) > transid) {
3119                         brelse(latest);
3120                         latest = bh;
3121                         transid = btrfs_super_generation(super);
3122                 } else {
3123                         brelse(bh);
3124                 }
3125         }
3126         return latest;
3127 }
3128
3129 /*
3130  * this should be called twice, once with wait == 0 and
3131  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3132  * we write are pinned.
3133  *
3134  * They are released when wait == 1 is done.
3135  * max_mirrors must be the same for both runs, and it indicates how
3136  * many supers on this one device should be written.
3137  *
3138  * max_mirrors == 0 means to write them all.
3139  */
3140 static int write_dev_supers(struct btrfs_device *device,
3141                             struct btrfs_super_block *sb,
3142                             int do_barriers, int wait, int max_mirrors)
3143 {
3144         struct buffer_head *bh;
3145         int i;
3146         int ret;
3147         int errors = 0;
3148         u32 crc;
3149         u64 bytenr;
3150
3151         if (max_mirrors == 0)
3152                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3153
3154         for (i = 0; i < max_mirrors; i++) {
3155                 bytenr = btrfs_sb_offset(i);
3156                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3157                     device->commit_total_bytes)
3158                         break;
3159
3160                 if (wait) {
3161                         bh = __find_get_block(device->bdev, bytenr / 4096,
3162                                               BTRFS_SUPER_INFO_SIZE);
3163                         if (!bh) {
3164                                 errors++;
3165                                 continue;
3166                         }
3167                         wait_on_buffer(bh);
3168                         if (!buffer_uptodate(bh))
3169                                 errors++;
3170
3171                         /* drop our reference */
3172                         brelse(bh);
3173
3174                         /* drop the reference from the wait == 0 run */
3175                         brelse(bh);
3176                         continue;
3177                 } else {
3178                         btrfs_set_super_bytenr(sb, bytenr);
3179
3180                         crc = ~(u32)0;
3181                         crc = btrfs_csum_data((char *)sb +
3182                                               BTRFS_CSUM_SIZE, crc,
3183                                               BTRFS_SUPER_INFO_SIZE -
3184                                               BTRFS_CSUM_SIZE);
3185                         btrfs_csum_final(crc, sb->csum);
3186
3187                         /*
3188                          * one reference for us, and we leave it for the
3189                          * caller
3190                          */
3191                         bh = __getblk(device->bdev, bytenr / 4096,
3192                                       BTRFS_SUPER_INFO_SIZE);
3193                         if (!bh) {
3194                                 printk(KERN_ERR "BTRFS: couldn't get super "
3195                                        "buffer head for bytenr %Lu\n", bytenr);
3196                                 errors++;
3197                                 continue;
3198                         }
3199
3200                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3201
3202                         /* one reference for submit_bh */
3203                         get_bh(bh);
3204
3205                         set_buffer_uptodate(bh);
3206                         lock_buffer(bh);
3207                         bh->b_end_io = btrfs_end_buffer_write_sync;
3208                         bh->b_private = device;
3209                 }
3210
3211                 /*
3212                  * we fua the first super.  The others we allow
3213                  * to go down lazy.
3214                  */
3215                 if (i == 0)
3216                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3217                 else
3218                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3219                 if (ret)
3220                         errors++;
3221         }
3222         return errors < i ? 0 : -1;
3223 }
3224
3225 /*
3226  * endio for the write_dev_flush, this will wake anyone waiting
3227  * for the barrier when it is done
3228  */
3229 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3230 {
3231         if (err) {
3232                 if (err == -EOPNOTSUPP)
3233                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3234                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3235         }
3236         if (bio->bi_private)
3237                 complete(bio->bi_private);
3238         bio_put(bio);
3239 }
3240
3241 /*
3242  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3243  * sent down.  With wait == 1, it waits for the previous flush.
3244  *
3245  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3246  * capable
3247  */
3248 static int write_dev_flush(struct btrfs_device *device, int wait)
3249 {
3250         struct bio *bio;
3251         int ret = 0;
3252
3253         if (device->nobarriers)
3254                 return 0;
3255
3256         if (wait) {
3257                 bio = device->flush_bio;
3258                 if (!bio)
3259                         return 0;
3260
3261                 wait_for_completion(&device->flush_wait);
3262
3263                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3264                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3265                                       rcu_str_deref(device->name));
3266                         device->nobarriers = 1;
3267                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3268                         ret = -EIO;
3269                         btrfs_dev_stat_inc_and_print(device,
3270                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3271                 }
3272
3273                 /* drop the reference from the wait == 0 run */
3274                 bio_put(bio);
3275                 device->flush_bio = NULL;
3276
3277                 return ret;
3278         }
3279
3280         /*
3281          * one reference for us, and we leave it for the
3282          * caller
3283          */
3284         device->flush_bio = NULL;
3285         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3286         if (!bio)
3287                 return -ENOMEM;
3288
3289         bio->bi_end_io = btrfs_end_empty_barrier;
3290         bio->bi_bdev = device->bdev;
3291         init_completion(&device->flush_wait);
3292         bio->bi_private = &device->flush_wait;
3293         device->flush_bio = bio;
3294
3295         bio_get(bio);
3296         btrfsic_submit_bio(WRITE_FLUSH, bio);
3297
3298         return 0;
3299 }
3300
3301 /*
3302  * send an empty flush down to each device in parallel,
3303  * then wait for them
3304  */
3305 static int barrier_all_devices(struct btrfs_fs_info *info)
3306 {
3307         struct list_head *head;
3308         struct btrfs_device *dev;
3309         int errors_send = 0;
3310         int errors_wait = 0;
3311         int ret;
3312
3313         /* send down all the barriers */
3314         head = &info->fs_devices->devices;
3315         list_for_each_entry_rcu(dev, head, dev_list) {
3316                 if (dev->missing)
3317                         continue;
3318                 if (!dev->bdev) {
3319                         errors_send++;
3320                         continue;
3321                 }
3322                 if (!dev->in_fs_metadata || !dev->writeable)
3323                         continue;
3324
3325                 ret = write_dev_flush(dev, 0);
3326                 if (ret)
3327                         errors_send++;
3328         }
3329
3330         /* wait for all the barriers */
3331         list_for_each_entry_rcu(dev, head, dev_list) {
3332                 if (dev->missing)
3333                         continue;
3334                 if (!dev->bdev) {
3335                         errors_wait++;
3336                         continue;
3337                 }
3338                 if (!dev->in_fs_metadata || !dev->writeable)
3339                         continue;
3340
3341                 ret = write_dev_flush(dev, 1);
3342                 if (ret)
3343                         errors_wait++;
3344         }
3345         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3346             errors_wait > info->num_tolerated_disk_barrier_failures)
3347                 return -EIO;
3348         return 0;
3349 }
3350
3351 int btrfs_calc_num_tolerated_disk_barrier_failures(
3352         struct btrfs_fs_info *fs_info)
3353 {
3354         struct btrfs_ioctl_space_info space;
3355         struct btrfs_space_info *sinfo;
3356         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3357                        BTRFS_BLOCK_GROUP_SYSTEM,
3358                        BTRFS_BLOCK_GROUP_METADATA,
3359                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3360         int num_types = 4;
3361         int i;
3362         int c;
3363         int num_tolerated_disk_barrier_failures =
3364                 (int)fs_info->fs_devices->num_devices;
3365
3366         for (i = 0; i < num_types; i++) {
3367                 struct btrfs_space_info *tmp;
3368
3369                 sinfo = NULL;
3370                 rcu_read_lock();
3371                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3372                         if (tmp->flags == types[i]) {
3373                                 sinfo = tmp;
3374                                 break;
3375                         }
3376                 }
3377                 rcu_read_unlock();
3378
3379                 if (!sinfo)
3380                         continue;
3381
3382                 down_read(&sinfo->groups_sem);
3383                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3384                         if (!list_empty(&sinfo->block_groups[c])) {
3385                                 u64 flags;
3386
3387                                 btrfs_get_block_group_info(
3388                                         &sinfo->block_groups[c], &space);
3389                                 if (space.total_bytes == 0 ||
3390                                     space.used_bytes == 0)
3391                                         continue;
3392                                 flags = space.flags;
3393                                 /*
3394                                  * return
3395                                  * 0: if dup, single or RAID0 is configured for
3396                                  *    any of metadata, system or data, else
3397                                  * 1: if RAID5 is configured, or if RAID1 or
3398                                  *    RAID10 is configured and only two mirrors
3399                                  *    are used, else
3400                                  * 2: if RAID6 is configured, else
3401                                  * num_mirrors - 1: if RAID1 or RAID10 is
3402                                  *                  configured and more than
3403                                  *                  2 mirrors are used.
3404                                  */
3405                                 if (num_tolerated_disk_barrier_failures > 0 &&
3406                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3407                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3408                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3409                                       == 0)))
3410                                         num_tolerated_disk_barrier_failures = 0;
3411                                 else if (num_tolerated_disk_barrier_failures > 1) {
3412                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3413                                             BTRFS_BLOCK_GROUP_RAID5 |
3414                                             BTRFS_BLOCK_GROUP_RAID10)) {
3415                                                 num_tolerated_disk_barrier_failures = 1;
3416                                         } else if (flags &
3417                                                    BTRFS_BLOCK_GROUP_RAID6) {
3418                                                 num_tolerated_disk_barrier_failures = 2;
3419                                         }
3420                                 }
3421                         }
3422                 }
3423                 up_read(&sinfo->groups_sem);
3424         }
3425
3426         return num_tolerated_disk_barrier_failures;
3427 }
3428
3429 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3430 {
3431         struct list_head *head;
3432         struct btrfs_device *dev;
3433         struct btrfs_super_block *sb;
3434         struct btrfs_dev_item *dev_item;
3435         int ret;
3436         int do_barriers;
3437         int max_errors;
3438         int total_errors = 0;
3439         u64 flags;
3440
3441         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3442         backup_super_roots(root->fs_info);
3443
3444         sb = root->fs_info->super_for_commit;
3445         dev_item = &sb->dev_item;
3446
3447         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3448         head = &root->fs_info->fs_devices->devices;
3449         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3450
3451         if (do_barriers) {
3452                 ret = barrier_all_devices(root->fs_info);
3453                 if (ret) {
3454                         mutex_unlock(
3455                                 &root->fs_info->fs_devices->device_list_mutex);
3456                         btrfs_error(root->fs_info, ret,
3457                                     "errors while submitting device barriers.");
3458                         return ret;
3459                 }
3460         }
3461
3462         list_for_each_entry_rcu(dev, head, dev_list) {
3463                 if (!dev->bdev) {
3464                         total_errors++;
3465                         continue;
3466                 }
3467                 if (!dev->in_fs_metadata || !dev->writeable)
3468                         continue;
3469
3470                 btrfs_set_stack_device_generation(dev_item, 0);
3471                 btrfs_set_stack_device_type(dev_item, dev->type);
3472                 btrfs_set_stack_device_id(dev_item, dev->devid);
3473                 btrfs_set_stack_device_total_bytes(dev_item,
3474                                                    dev->commit_total_bytes);
3475                 btrfs_set_stack_device_bytes_used(dev_item,
3476                                                   dev->commit_bytes_used);
3477                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3478                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3479                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3480                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3481                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3482
3483                 flags = btrfs_super_flags(sb);
3484                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3485
3486                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3487                 if (ret)
3488                         total_errors++;
3489         }
3490         if (total_errors > max_errors) {
3491                 btrfs_err(root->fs_info, "%d errors while writing supers",
3492                        total_errors);
3493                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3494
3495                 /* FUA is masked off if unsupported and can't be the reason */
3496                 btrfs_error(root->fs_info, -EIO,
3497                             "%d errors while writing supers", total_errors);
3498                 return -EIO;
3499         }
3500
3501         total_errors = 0;
3502         list_for_each_entry_rcu(dev, head, dev_list) {
3503                 if (!dev->bdev)
3504                         continue;
3505                 if (!dev->in_fs_metadata || !dev->writeable)
3506                         continue;
3507
3508                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3509                 if (ret)
3510                         total_errors++;
3511         }
3512         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3513         if (total_errors > max_errors) {
3514                 btrfs_error(root->fs_info, -EIO,
3515                             "%d errors while writing supers", total_errors);
3516                 return -EIO;
3517         }
3518         return 0;
3519 }
3520
3521 int write_ctree_super(struct btrfs_trans_handle *trans,
3522                       struct btrfs_root *root, int max_mirrors)
3523 {
3524         return write_all_supers(root, max_mirrors);
3525 }
3526
3527 /* Drop a fs root from the radix tree and free it. */
3528 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3529                                   struct btrfs_root *root)
3530 {
3531         spin_lock(&fs_info->fs_roots_radix_lock);
3532         radix_tree_delete(&fs_info->fs_roots_radix,
3533                           (unsigned long)root->root_key.objectid);
3534         spin_unlock(&fs_info->fs_roots_radix_lock);
3535
3536         if (btrfs_root_refs(&root->root_item) == 0)
3537                 synchronize_srcu(&fs_info->subvol_srcu);
3538
3539         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3540                 btrfs_free_log(NULL, root);
3541
3542         if (root->free_ino_pinned)
3543                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3544         if (root->free_ino_ctl)
3545                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3546         free_fs_root(root);
3547 }
3548
3549 static void free_fs_root(struct btrfs_root *root)
3550 {
3551         iput(root->ino_cache_inode);
3552         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3553         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3554         root->orphan_block_rsv = NULL;
3555         if (root->anon_dev)
3556                 free_anon_bdev(root->anon_dev);
3557         if (root->subv_writers)
3558                 btrfs_free_subvolume_writers(root->subv_writers);
3559         free_extent_buffer(root->node);
3560         free_extent_buffer(root->commit_root);
3561         kfree(root->free_ino_ctl);
3562         kfree(root->free_ino_pinned);
3563         kfree(root->name);
3564         btrfs_put_fs_root(root);
3565 }
3566
3567 void btrfs_free_fs_root(struct btrfs_root *root)
3568 {
3569         free_fs_root(root);
3570 }
3571
3572 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3573 {
3574         u64 root_objectid = 0;
3575         struct btrfs_root *gang[8];
3576         int i = 0;
3577         int err = 0;
3578         unsigned int ret = 0;
3579         int index;
3580
3581         while (1) {
3582                 index = srcu_read_lock(&fs_info->subvol_srcu);
3583                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3584                                              (void **)gang, root_objectid,
3585                                              ARRAY_SIZE(gang));
3586                 if (!ret) {
3587                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3588                         break;
3589                 }
3590                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3591
3592                 for (i = 0; i < ret; i++) {
3593                         /* Avoid to grab roots in dead_roots */
3594                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3595                                 gang[i] = NULL;
3596                                 continue;
3597                         }
3598                         /* grab all the search result for later use */
3599                         gang[i] = btrfs_grab_fs_root(gang[i]);
3600                 }
3601                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3602
3603                 for (i = 0; i < ret; i++) {
3604                         if (!gang[i])
3605                                 continue;
3606                         root_objectid = gang[i]->root_key.objectid;
3607                         err = btrfs_orphan_cleanup(gang[i]);
3608                         if (err)
3609                                 break;
3610                         btrfs_put_fs_root(gang[i]);
3611                 }
3612                 root_objectid++;
3613         }
3614
3615         /* release the uncleaned roots due to error */
3616         for (; i < ret; i++) {
3617                 if (gang[i])
3618                         btrfs_put_fs_root(gang[i]);
3619         }
3620         return err;
3621 }
3622
3623 int btrfs_commit_super(struct btrfs_root *root)
3624 {
3625         struct btrfs_trans_handle *trans;
3626
3627         mutex_lock(&root->fs_info->cleaner_mutex);
3628         btrfs_run_delayed_iputs(root);
3629         mutex_unlock(&root->fs_info->cleaner_mutex);
3630         wake_up_process(root->fs_info->cleaner_kthread);
3631
3632         /* wait until ongoing cleanup work done */
3633         down_write(&root->fs_info->cleanup_work_sem);
3634         up_write(&root->fs_info->cleanup_work_sem);
3635
3636         trans = btrfs_join_transaction(root);
3637         if (IS_ERR(trans))
3638                 return PTR_ERR(trans);
3639         return btrfs_commit_transaction(trans, root);
3640 }
3641
3642 void close_ctree(struct btrfs_root *root)
3643 {
3644         struct btrfs_fs_info *fs_info = root->fs_info;
3645         int ret;
3646
3647         fs_info->closing = 1;
3648         smp_mb();
3649
3650         /* wait for the uuid_scan task to finish */
3651         down(&fs_info->uuid_tree_rescan_sem);
3652         /* avoid complains from lockdep et al., set sem back to initial state */
3653         up(&fs_info->uuid_tree_rescan_sem);
3654
3655         /* pause restriper - we want to resume on mount */
3656         btrfs_pause_balance(fs_info);
3657
3658         btrfs_dev_replace_suspend_for_unmount(fs_info);
3659
3660         btrfs_scrub_cancel(fs_info);
3661
3662         /* wait for any defraggers to finish */
3663         wait_event(fs_info->transaction_wait,
3664                    (atomic_read(&fs_info->defrag_running) == 0));
3665
3666         /* clear out the rbtree of defraggable inodes */
3667         btrfs_cleanup_defrag_inodes(fs_info);
3668
3669         cancel_work_sync(&fs_info->async_reclaim_work);
3670
3671         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3672                 ret = btrfs_commit_super(root);
3673                 if (ret)
3674                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3675         }
3676
3677         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3678                 btrfs_error_commit_super(root);
3679
3680         kthread_stop(fs_info->transaction_kthread);
3681         kthread_stop(fs_info->cleaner_kthread);
3682
3683         fs_info->closing = 2;
3684         smp_mb();
3685
3686         btrfs_free_qgroup_config(root->fs_info);
3687
3688         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3689                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3690                        percpu_counter_sum(&fs_info->delalloc_bytes));
3691         }
3692
3693         btrfs_sysfs_remove_one(fs_info);
3694
3695         btrfs_free_fs_roots(fs_info);
3696
3697         btrfs_put_block_group_cache(fs_info);
3698
3699         btrfs_free_block_groups(fs_info);
3700
3701         /*
3702          * we must make sure there is not any read request to
3703          * submit after we stopping all workers.
3704          */
3705         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3706         btrfs_stop_all_workers(fs_info);
3707
3708         fs_info->open = 0;
3709         free_root_pointers(fs_info, 1);
3710
3711         iput(fs_info->btree_inode);
3712
3713 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3714         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3715                 btrfsic_unmount(root, fs_info->fs_devices);
3716 #endif
3717
3718         btrfs_close_devices(fs_info->fs_devices);
3719         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3720
3721         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3722         percpu_counter_destroy(&fs_info->delalloc_bytes);
3723         percpu_counter_destroy(&fs_info->bio_counter);
3724         bdi_destroy(&fs_info->bdi);
3725         cleanup_srcu_struct(&fs_info->subvol_srcu);
3726
3727         btrfs_free_stripe_hash_table(fs_info);
3728
3729         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3730         root->orphan_block_rsv = NULL;
3731
3732         lock_chunks(root);
3733         while (!list_empty(&fs_info->pinned_chunks)) {
3734                 struct extent_map *em;
3735
3736                 em = list_first_entry(&fs_info->pinned_chunks,
3737                                       struct extent_map, list);
3738                 list_del_init(&em->list);
3739                 free_extent_map(em);
3740         }
3741         unlock_chunks(root);
3742 }
3743
3744 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3745                           int atomic)
3746 {
3747         int ret;
3748         struct inode *btree_inode = buf->pages[0]->mapping->host;
3749
3750         ret = extent_buffer_uptodate(buf);
3751         if (!ret)
3752                 return ret;
3753
3754         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3755                                     parent_transid, atomic);
3756         if (ret == -EAGAIN)
3757                 return ret;
3758         return !ret;
3759 }
3760
3761 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3762 {
3763         return set_extent_buffer_uptodate(buf);
3764 }
3765
3766 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3767 {
3768         struct btrfs_root *root;
3769         u64 transid = btrfs_header_generation(buf);
3770         int was_dirty;
3771
3772 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3773         /*
3774          * This is a fast path so only do this check if we have sanity tests
3775          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3776          * outside of the sanity tests.
3777          */
3778         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3779                 return;
3780 #endif
3781         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3782         btrfs_assert_tree_locked(buf);
3783         if (transid != root->fs_info->generation)
3784                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3785                        "found %llu running %llu\n",
3786                         buf->start, transid, root->fs_info->generation);
3787         was_dirty = set_extent_buffer_dirty(buf);
3788         if (!was_dirty)
3789                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3790                                      buf->len,
3791                                      root->fs_info->dirty_metadata_batch);
3792 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3793         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3794                 btrfs_print_leaf(root, buf);
3795                 ASSERT(0);
3796         }
3797 #endif
3798 }
3799
3800 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3801                                         int flush_delayed)
3802 {
3803         /*
3804          * looks as though older kernels can get into trouble with
3805          * this code, they end up stuck in balance_dirty_pages forever
3806          */
3807         int ret;
3808
3809         if (current->flags & PF_MEMALLOC)
3810                 return;
3811
3812         if (flush_delayed)
3813                 btrfs_balance_delayed_items(root);
3814
3815         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3816                                      BTRFS_DIRTY_METADATA_THRESH);
3817         if (ret > 0) {
3818                 balance_dirty_pages_ratelimited(
3819                                    root->fs_info->btree_inode->i_mapping);
3820         }
3821         return;
3822 }
3823
3824 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3825 {
3826         __btrfs_btree_balance_dirty(root, 1);
3827 }
3828
3829 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3830 {
3831         __btrfs_btree_balance_dirty(root, 0);
3832 }
3833
3834 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3835 {
3836         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3837         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3838 }
3839
3840 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3841                               int read_only)
3842 {
3843         struct btrfs_super_block *sb = fs_info->super_copy;
3844         int ret = 0;
3845
3846         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3847                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3848                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3849                 ret = -EINVAL;
3850         }
3851         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3852                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3853                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3854                 ret = -EINVAL;
3855         }
3856         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3857                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3858                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3859                 ret = -EINVAL;
3860         }
3861
3862         /*
3863          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3864          * items yet, they'll be verified later. Issue just a warning.
3865          */
3866         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3867                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3868                                 btrfs_super_root(sb));
3869         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3870                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3871                                 btrfs_super_chunk_root(sb));
3872         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3873                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3874                                 btrfs_super_log_root(sb));
3875
3876         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3877                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3878                                 fs_info->fsid, sb->dev_item.fsid);
3879                 ret = -EINVAL;
3880         }
3881
3882         /*
3883          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3884          * done later
3885          */
3886         if (btrfs_super_num_devices(sb) > (1UL << 31))
3887                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3888                                 btrfs_super_num_devices(sb));
3889
3890         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3891                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3892                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3893                 ret = -EINVAL;
3894         }
3895
3896         /*
3897          * The generation is a global counter, we'll trust it more than the others
3898          * but it's still possible that it's the one that's wrong.
3899          */
3900         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3901                 printk(KERN_WARNING
3902                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3903                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3904         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3905             && btrfs_super_cache_generation(sb) != (u64)-1)
3906                 printk(KERN_WARNING
3907                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3908                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3909
3910         return ret;
3911 }
3912
3913 static void btrfs_error_commit_super(struct btrfs_root *root)
3914 {
3915         mutex_lock(&root->fs_info->cleaner_mutex);
3916         btrfs_run_delayed_iputs(root);
3917         mutex_unlock(&root->fs_info->cleaner_mutex);
3918
3919         down_write(&root->fs_info->cleanup_work_sem);
3920         up_write(&root->fs_info->cleanup_work_sem);
3921
3922         /* cleanup FS via transaction */
3923         btrfs_cleanup_transaction(root);
3924 }
3925
3926 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3927 {
3928         struct btrfs_ordered_extent *ordered;
3929
3930         spin_lock(&root->ordered_extent_lock);
3931         /*
3932          * This will just short circuit the ordered completion stuff which will
3933          * make sure the ordered extent gets properly cleaned up.
3934          */
3935         list_for_each_entry(ordered, &root->ordered_extents,
3936                             root_extent_list)
3937                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3938         spin_unlock(&root->ordered_extent_lock);
3939 }
3940
3941 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3942 {
3943         struct btrfs_root *root;
3944         struct list_head splice;
3945
3946         INIT_LIST_HEAD(&splice);
3947
3948         spin_lock(&fs_info->ordered_root_lock);
3949         list_splice_init(&fs_info->ordered_roots, &splice);
3950         while (!list_empty(&splice)) {
3951                 root = list_first_entry(&splice, struct btrfs_root,
3952                                         ordered_root);
3953                 list_move_tail(&root->ordered_root,
3954                                &fs_info->ordered_roots);
3955
3956                 spin_unlock(&fs_info->ordered_root_lock);
3957                 btrfs_destroy_ordered_extents(root);
3958
3959                 cond_resched();
3960                 spin_lock(&fs_info->ordered_root_lock);
3961         }
3962         spin_unlock(&fs_info->ordered_root_lock);
3963 }
3964
3965 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3966                                       struct btrfs_root *root)
3967 {
3968         struct rb_node *node;
3969         struct btrfs_delayed_ref_root *delayed_refs;
3970         struct btrfs_delayed_ref_node *ref;
3971         int ret = 0;
3972
3973         delayed_refs = &trans->delayed_refs;
3974
3975         spin_lock(&delayed_refs->lock);
3976         if (atomic_read(&delayed_refs->num_entries) == 0) {
3977                 spin_unlock(&delayed_refs->lock);
3978                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3979                 return ret;
3980         }
3981
3982         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3983                 struct btrfs_delayed_ref_head *head;
3984                 bool pin_bytes = false;
3985
3986                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3987                                 href_node);
3988                 if (!mutex_trylock(&head->mutex)) {
3989                         atomic_inc(&head->node.refs);
3990                         spin_unlock(&delayed_refs->lock);
3991
3992                         mutex_lock(&head->mutex);
3993                         mutex_unlock(&head->mutex);
3994                         btrfs_put_delayed_ref(&head->node);
3995                         spin_lock(&delayed_refs->lock);
3996                         continue;
3997                 }
3998                 spin_lock(&head->lock);
3999                 while ((node = rb_first(&head->ref_root)) != NULL) {
4000                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
4001                                        rb_node);
4002                         ref->in_tree = 0;
4003                         rb_erase(&ref->rb_node, &head->ref_root);
4004                         atomic_dec(&delayed_refs->num_entries);
4005                         btrfs_put_delayed_ref(ref);
4006                 }
4007                 if (head->must_insert_reserved)
4008                         pin_bytes = true;
4009                 btrfs_free_delayed_extent_op(head->extent_op);
4010                 delayed_refs->num_heads--;
4011                 if (head->processing == 0)
4012                         delayed_refs->num_heads_ready--;
4013                 atomic_dec(&delayed_refs->num_entries);
4014                 head->node.in_tree = 0;
4015                 rb_erase(&head->href_node, &delayed_refs->href_root);
4016                 spin_unlock(&head->lock);
4017                 spin_unlock(&delayed_refs->lock);
4018                 mutex_unlock(&head->mutex);
4019
4020                 if (pin_bytes)
4021                         btrfs_pin_extent(root, head->node.bytenr,
4022                                          head->node.num_bytes, 1);
4023                 btrfs_put_delayed_ref(&head->node);
4024                 cond_resched();
4025                 spin_lock(&delayed_refs->lock);
4026         }
4027
4028         spin_unlock(&delayed_refs->lock);
4029
4030         return ret;
4031 }
4032
4033 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4034 {
4035         struct btrfs_inode *btrfs_inode;
4036         struct list_head splice;
4037
4038         INIT_LIST_HEAD(&splice);
4039
4040         spin_lock(&root->delalloc_lock);
4041         list_splice_init(&root->delalloc_inodes, &splice);
4042
4043         while (!list_empty(&splice)) {
4044                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4045                                                delalloc_inodes);
4046
4047                 list_del_init(&btrfs_inode->delalloc_inodes);
4048                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4049                           &btrfs_inode->runtime_flags);
4050                 spin_unlock(&root->delalloc_lock);
4051
4052                 btrfs_invalidate_inodes(btrfs_inode->root);
4053
4054                 spin_lock(&root->delalloc_lock);
4055         }
4056
4057         spin_unlock(&root->delalloc_lock);
4058 }
4059
4060 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4061 {
4062         struct btrfs_root *root;
4063         struct list_head splice;
4064
4065         INIT_LIST_HEAD(&splice);
4066
4067         spin_lock(&fs_info->delalloc_root_lock);
4068         list_splice_init(&fs_info->delalloc_roots, &splice);
4069         while (!list_empty(&splice)) {
4070                 root = list_first_entry(&splice, struct btrfs_root,
4071                                          delalloc_root);
4072                 list_del_init(&root->delalloc_root);
4073                 root = btrfs_grab_fs_root(root);
4074                 BUG_ON(!root);
4075                 spin_unlock(&fs_info->delalloc_root_lock);
4076
4077                 btrfs_destroy_delalloc_inodes(root);
4078                 btrfs_put_fs_root(root);
4079
4080                 spin_lock(&fs_info->delalloc_root_lock);
4081         }
4082         spin_unlock(&fs_info->delalloc_root_lock);
4083 }
4084
4085 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4086                                         struct extent_io_tree *dirty_pages,
4087                                         int mark)
4088 {
4089         int ret;
4090         struct extent_buffer *eb;
4091         u64 start = 0;
4092         u64 end;
4093
4094         while (1) {
4095                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4096                                             mark, NULL);
4097                 if (ret)
4098                         break;
4099
4100                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4101                 while (start <= end) {
4102                         eb = btrfs_find_tree_block(root, start);
4103                         start += root->nodesize;
4104                         if (!eb)
4105                                 continue;
4106                         wait_on_extent_buffer_writeback(eb);
4107
4108                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4109                                                &eb->bflags))
4110                                 clear_extent_buffer_dirty(eb);
4111                         free_extent_buffer_stale(eb);
4112                 }
4113         }
4114
4115         return ret;
4116 }
4117
4118 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4119                                        struct extent_io_tree *pinned_extents)
4120 {
4121         struct extent_io_tree *unpin;
4122         u64 start;
4123         u64 end;
4124         int ret;
4125         bool loop = true;
4126
4127         unpin = pinned_extents;
4128 again:
4129         while (1) {
4130                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4131                                             EXTENT_DIRTY, NULL);
4132                 if (ret)
4133                         break;
4134
4135                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4136                 btrfs_error_unpin_extent_range(root, start, end);
4137                 cond_resched();
4138         }
4139
4140         if (loop) {
4141                 if (unpin == &root->fs_info->freed_extents[0])
4142                         unpin = &root->fs_info->freed_extents[1];
4143                 else
4144                         unpin = &root->fs_info->freed_extents[0];
4145                 loop = false;
4146                 goto again;
4147         }
4148
4149         return 0;
4150 }
4151
4152 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4153                                        struct btrfs_fs_info *fs_info)
4154 {
4155         struct btrfs_ordered_extent *ordered;
4156
4157         spin_lock(&fs_info->trans_lock);
4158         while (!list_empty(&cur_trans->pending_ordered)) {
4159                 ordered = list_first_entry(&cur_trans->pending_ordered,
4160                                            struct btrfs_ordered_extent,
4161                                            trans_list);
4162                 list_del_init(&ordered->trans_list);
4163                 spin_unlock(&fs_info->trans_lock);
4164
4165                 btrfs_put_ordered_extent(ordered);
4166                 spin_lock(&fs_info->trans_lock);
4167         }
4168         spin_unlock(&fs_info->trans_lock);
4169 }
4170
4171 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4172                                    struct btrfs_root *root)
4173 {
4174         btrfs_destroy_delayed_refs(cur_trans, root);
4175
4176         cur_trans->state = TRANS_STATE_COMMIT_START;
4177         wake_up(&root->fs_info->transaction_blocked_wait);
4178
4179         cur_trans->state = TRANS_STATE_UNBLOCKED;
4180         wake_up(&root->fs_info->transaction_wait);
4181
4182         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4183         btrfs_destroy_delayed_inodes(root);
4184         btrfs_assert_delayed_root_empty(root);
4185
4186         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4187                                      EXTENT_DIRTY);
4188         btrfs_destroy_pinned_extent(root,
4189                                     root->fs_info->pinned_extents);
4190
4191         cur_trans->state =TRANS_STATE_COMPLETED;
4192         wake_up(&cur_trans->commit_wait);
4193
4194         /*
4195         memset(cur_trans, 0, sizeof(*cur_trans));
4196         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4197         */
4198 }
4199
4200 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4201 {
4202         struct btrfs_transaction *t;
4203
4204         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4205
4206         spin_lock(&root->fs_info->trans_lock);
4207         while (!list_empty(&root->fs_info->trans_list)) {
4208                 t = list_first_entry(&root->fs_info->trans_list,
4209                                      struct btrfs_transaction, list);
4210                 if (t->state >= TRANS_STATE_COMMIT_START) {
4211                         atomic_inc(&t->use_count);
4212                         spin_unlock(&root->fs_info->trans_lock);
4213                         btrfs_wait_for_commit(root, t->transid);
4214                         btrfs_put_transaction(t);
4215                         spin_lock(&root->fs_info->trans_lock);
4216                         continue;
4217                 }
4218                 if (t == root->fs_info->running_transaction) {
4219                         t->state = TRANS_STATE_COMMIT_DOING;
4220                         spin_unlock(&root->fs_info->trans_lock);
4221                         /*
4222                          * We wait for 0 num_writers since we don't hold a trans
4223                          * handle open currently for this transaction.
4224                          */
4225                         wait_event(t->writer_wait,
4226                                    atomic_read(&t->num_writers) == 0);
4227                 } else {
4228                         spin_unlock(&root->fs_info->trans_lock);
4229                 }
4230                 btrfs_cleanup_one_transaction(t, root);
4231
4232                 spin_lock(&root->fs_info->trans_lock);
4233                 if (t == root->fs_info->running_transaction)
4234                         root->fs_info->running_transaction = NULL;
4235                 list_del_init(&t->list);
4236                 spin_unlock(&root->fs_info->trans_lock);
4237
4238                 btrfs_put_transaction(t);
4239                 trace_btrfs_transaction_commit(root);
4240                 spin_lock(&root->fs_info->trans_lock);
4241         }
4242         spin_unlock(&root->fs_info->trans_lock);
4243         btrfs_destroy_all_ordered_extents(root->fs_info);
4244         btrfs_destroy_delayed_inodes(root);
4245         btrfs_assert_delayed_root_empty(root);
4246         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4247         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4248         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4249
4250         return 0;
4251 }
4252
4253 static struct extent_io_ops btree_extent_io_ops = {
4254         .readpage_end_io_hook = btree_readpage_end_io_hook,
4255         .readpage_io_failed_hook = btree_io_failed_hook,
4256         .submit_bio_hook = btree_submit_bio_hook,
4257         /* note we're sharing with inode.c for the merge bio hook */
4258         .merge_bio_hook = btrfs_merge_bio_hook,
4259 };