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_WARNING
 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_ERR
 371             "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 372                         eb->fs_info->sb->s_id, eb->start,
 373                         parent_transid, btrfs_header_generation(eb));
 374         ret = 1;
 375
 376         /*
 377          * Things reading via commit roots that don't have normal protection,
 378          * like send, can have a really old block in cache that may point at a
 379          * block that has been free'd and re-allocated.  So don't clear uptodate
 380          * if we find an eb that is under IO (dirty/writeback) because we could
 381          * end up reading in the stale data and then writing it back out and
 382          * making everybody very sad.
 383          */
 384         if (!extent_buffer_under_io(eb))
 385                 clear_extent_buffer_uptodate(eb);
 386 out:
 387         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 388                              &cached_state, GFP_NOFS);
 389         if (need_lock)
 390                 btrfs_tree_read_unlock_blocking(eb);
 391         return ret;
 392 }
 393
 394 /*
 395  * Return 0 if the superblock checksum type matches the checksum value of that
 396  * algorithm. Pass the raw disk superblock data.
 397  */
 398 static int btrfs_check_super_csum(char *raw_disk_sb)
 399 {
 400         struct btrfs_super_block *disk_sb =
 401                 (struct btrfs_super_block *)raw_disk_sb;
 402         u16 csum_type = btrfs_super_csum_type(disk_sb);
 403         int ret = 0;
 404
 405         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 406                 u32 crc = ~(u32)0;
 407                 const int csum_size = sizeof(crc);
 408                 char result[csum_size];
 409
 410                 /*
 411                  * The super_block structure does not span the whole
 412                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 413                  * is filled with zeros and is included in the checkum.
 414                  */
 415                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 416                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 417                 btrfs_csum_final(crc, result);
 418
 419                 if (memcmp(raw_disk_sb, result, csum_size))
 420                         ret = 1;
 421
 422                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 423                         printk(KERN_WARNING
 424                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 425                         ret = 0;
 426                 }
 427         }
 428
 429         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 430                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 431                                 csum_type);
 432                 ret = 1;
 433         }
 434
 435         return ret;
 436 }
 437
 438 /*
 439  * helper to read a given tree block, doing retries as required when
 440  * the checksums don't match and we have alternate mirrors to try.
 441  */
 442 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 443                                           struct extent_buffer *eb,
 444                                           u64 start, u64 parent_transid)
 445 {
 446         struct extent_io_tree *io_tree;
 447         int failed = 0;
 448         int ret;
 449         int num_copies = 0;
 450         int mirror_num = 0;
 451         int failed_mirror = 0;
 452
 453         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 454         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 455         while (1) {
 456                 ret = read_extent_buffer_pages(io_tree, eb, start,
 457                                                WAIT_COMPLETE,
 458                                                btree_get_extent, mirror_num);
 459                 if (!ret) {
 460                         if (!verify_parent_transid(io_tree, eb,
 461                                                    parent_transid, 0))
 462                                 break;
 463                         else
 464                                 ret = -EIO;
 465                 }
 466
 467                 /*
 468                  * This buffer's crc is fine, but its contents are corrupted, so
 469                  * there is no reason to read the other copies, they won't be
 470                  * any less wrong.
 471                  */
 472                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 473                         break;
 474
 475                 num_copies = btrfs_num_copies(root->fs_info,
 476                                               eb->start, eb->len);
 477                 if (num_copies == 1)
 478                         break;
 479
 480                 if (!failed_mirror) {
 481                         failed = 1;
 482                         failed_mirror = eb->read_mirror;
 483                 }
 484
 485                 mirror_num++;
 486                 if (mirror_num == failed_mirror)
 487                         mirror_num++;
 488
 489                 if (mirror_num > num_copies)
 490                         break;
 491         }
 492
 493         if (failed && !ret && failed_mirror)
 494                 repair_eb_io_failure(root, eb, failed_mirror);
 495
 496         return ret;
 497 }
 498
 499 /*
 500  * checksum a dirty tree block before IO.  This has extra checks to make sure
 501  * we only fill in the checksum field in the first page of a multi-page block
 502  */
 503
 504 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 505 {
 506         u64 start = page_offset(page);
 507         u64 found_start;
 508         struct extent_buffer *eb;
 509
 510         eb = (struct extent_buffer *)page->private;
 511         if (page != eb->pages[0])
 512                 return 0;
 513         found_start = btrfs_header_bytenr(eb);
 514         if (WARN_ON(found_start != start || !PageUptodate(page)))
 515                 return 0;
 516         csum_tree_block(root, eb, 0);
 517         return 0;
 518 }
 519
 520 static int check_tree_block_fsid(struct btrfs_root *root,
 521                                  struct extent_buffer *eb)
 522 {
 523         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 524         u8 fsid[BTRFS_UUID_SIZE];
 525         int ret = 1;
 526
 527         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 528         while (fs_devices) {
 529                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 530                         ret = 0;
 531                         break;
 532                 }
 533                 fs_devices = fs_devices->seed;
 534         }
 535         return ret;
 536 }
 537
 538 #define CORRUPT(reason, eb, root, slot)                         \
 539         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 540                    "root=%llu, slot=%d", reason,                        \
 541                btrfs_header_bytenr(eb), root->objectid, slot)
 542
 543 static noinline int check_leaf(struct btrfs_root *root,
 544                                struct extent_buffer *leaf)
 545 {
 546         struct btrfs_key key;
 547         struct btrfs_key leaf_key;
 548         u32 nritems = btrfs_header_nritems(leaf);
 549         int slot;
 550
 551         if (nritems == 0)
 552                 return 0;
 553
 554         /* Check the 0 item */
 555         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 556             BTRFS_LEAF_DATA_SIZE(root)) {
 557                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 558                 return -EIO;
 559         }
 560
 561         /*
 562          * Check to make sure each items keys are in the correct order and their
 563          * offsets make sense.  We only have to loop through nritems-1 because
 564          * we check the current slot against the next slot, which verifies the
 565          * next slot's offset+size makes sense and that the current's slot
 566          * offset is correct.
 567          */
 568         for (slot = 0; slot < nritems - 1; slot++) {
 569                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 570                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 571
 572                 /* Make sure the keys are in the right order */
 573                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 574                         CORRUPT("bad key order", leaf, root, slot);
 575                         return -EIO;
 576                 }
 577
 578                 /*
 579                  * Make sure the offset and ends are right, remember that the
 580                  * item data starts at the end of the leaf and grows towards the
 581                  * front.
 582                  */
 583                 if (btrfs_item_offset_nr(leaf, slot) !=
 584                         btrfs_item_end_nr(leaf, slot + 1)) {
 585                         CORRUPT("slot offset bad", leaf, root, slot);
 586                         return -EIO;
 587                 }
 588
 589                 /*
 590                  * Check to make sure that we don't point outside of the leaf,
 591                  * just incase all the items are consistent to eachother, but
 592                  * all point outside of the leaf.
 593                  */
 594                 if (btrfs_item_end_nr(leaf, slot) >
 595                     BTRFS_LEAF_DATA_SIZE(root)) {
 596                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 597                         return -EIO;
 598                 }
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 605                                       u64 phy_offset, struct page *page,
 606                                       u64 start, u64 end, int mirror)
 607 {
 608         u64 found_start;
 609         int found_level;
 610         struct extent_buffer *eb;
 611         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 612         int ret = 0;
 613         int reads_done;
 614
 615         if (!page->private)
 616                 goto out;
 617
 618         eb = (struct extent_buffer *)page->private;
 619
 620         /* the pending IO might have been the only thing that kept this buffer
 621          * in memory.  Make sure we have a ref for all this other checks
 622          */
 623         extent_buffer_get(eb);
 624
 625         reads_done = atomic_dec_and_test(&eb->io_pages);
 626         if (!reads_done)
 627                 goto err;
 628
 629         eb->read_mirror = mirror;
 630         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 631                 ret = -EIO;
 632                 goto err;
 633         }
 634
 635         found_start = btrfs_header_bytenr(eb);
 636         if (found_start != eb->start) {
 637                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
 638                                "%llu %llu\n",
 639                                eb->fs_info->sb->s_id, found_start, eb->start);
 640                 ret = -EIO;
 641                 goto err;
 642         }
 643         if (check_tree_block_fsid(root, eb)) {
 644                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
 645                                eb->fs_info->sb->s_id, eb->start);
 646                 ret = -EIO;
 647                 goto err;
 648         }
 649         found_level = btrfs_header_level(eb);
 650         if (found_level >= BTRFS_MAX_LEVEL) {
 651                 btrfs_err(root->fs_info, "bad tree block level %d",
 652                            (int)btrfs_header_level(eb));
 653                 ret = -EIO;
 654                 goto err;
 655         }
 656
 657         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 658                                        eb, found_level);
 659
 660         ret = csum_tree_block(root, eb, 1);
 661         if (ret) {
 662                 ret = -EIO;
 663                 goto err;
 664         }
 665
 666         /*
 667          * If this is a leaf block and it is corrupt, set the corrupt bit so
 668          * that we don't try and read the other copies of this block, just
 669          * return -EIO.
 670          */
 671         if (found_level == 0 && check_leaf(root, eb)) {
 672                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 673                 ret = -EIO;
 674         }
 675
 676         if (!ret)
 677                 set_extent_buffer_uptodate(eb);
 678 err:
 679         if (reads_done &&
 680             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 681                 btree_readahead_hook(root, eb, eb->start, ret);
 682
 683         if (ret) {
 684                 /*
 685                  * our io error hook is going to dec the io pages
 686                  * again, we have to make sure it has something
 687                  * to decrement
 688                  */
 689                 atomic_inc(&eb->io_pages);
 690                 clear_extent_buffer_uptodate(eb);
 691         }
 692         free_extent_buffer(eb);
 693 out:
 694         return ret;
 695 }
 696
 697 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 698 {
 699         struct extent_buffer *eb;
 700         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 701
 702         eb = (struct extent_buffer *)page->private;
 703         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 704         eb->read_mirror = failed_mirror;
 705         atomic_dec(&eb->io_pages);
 706         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 707                 btree_readahead_hook(root, eb, eb->start, -EIO);
 708         return -EIO;    /* we fixed nothing */
 709 }
 710
 711 static void end_workqueue_bio(struct bio *bio, int err)
 712 {
 713         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 714         struct btrfs_fs_info *fs_info;
 715         struct btrfs_workqueue *wq;
 716         btrfs_work_func_t func;
 717
 718         fs_info = end_io_wq->info;
 719         end_io_wq->error = err;
 720
 721         if (bio->bi_rw & REQ_WRITE) {
 722                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 723                         wq = fs_info->endio_meta_write_workers;
 724                         func = btrfs_endio_meta_write_helper;
 725                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 726                         wq = fs_info->endio_freespace_worker;
 727                         func = btrfs_freespace_write_helper;
 728                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 729                         wq = fs_info->endio_raid56_workers;
 730                         func = btrfs_endio_raid56_helper;
 731                 } else {
 732                         wq = fs_info->endio_write_workers;
 733                         func = btrfs_endio_write_helper;
 734                 }
 735         } else {
 736                 if (unlikely(end_io_wq->metadata ==
 737                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 738                         wq = fs_info->endio_repair_workers;
 739                         func = btrfs_endio_repair_helper;
 740                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 741                         wq = fs_info->endio_raid56_workers;
 742                         func = btrfs_endio_raid56_helper;
 743                 } else if (end_io_wq->metadata) {
 744                         wq = fs_info->endio_meta_workers;
 745                         func = btrfs_endio_meta_helper;
 746                 } else {
 747                         wq = fs_info->endio_workers;
 748                         func = btrfs_endio_helper;
 749                 }
 750         }
 751
 752         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 753         btrfs_queue_work(wq, &end_io_wq->work);
 754 }
 755
 756 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 757                         enum btrfs_wq_endio_type metadata)
 758 {
 759         struct btrfs_end_io_wq *end_io_wq;
 760
 761         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 762         if (!end_io_wq)
 763                 return -ENOMEM;
 764
 765         end_io_wq->private = bio->bi_private;
 766         end_io_wq->end_io = bio->bi_end_io;
 767         end_io_wq->info = info;
 768         end_io_wq->error = 0;
 769         end_io_wq->bio = bio;
 770         end_io_wq->metadata = metadata;
 771
 772         bio->bi_private = end_io_wq;
 773         bio->bi_end_io = end_workqueue_bio;
 774         return 0;
 775 }
 776
 777 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 778 {
 779         unsigned long limit = min_t(unsigned long,
 780                                     info->thread_pool_size,
 781                                     info->fs_devices->open_devices);
 782         return 256 * limit;
 783 }
 784
 785 static void run_one_async_start(struct btrfs_work *work)
 786 {
 787         struct async_submit_bio *async;
 788         int ret;
 789
 790         async = container_of(work, struct  async_submit_bio, work);
 791         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 792                                       async->mirror_num, async->bio_flags,
 793                                       async->bio_offset);
 794         if (ret)
 795                 async->error = ret;
 796 }
 797
 798 static void run_one_async_done(struct btrfs_work *work)
 799 {
 800         struct btrfs_fs_info *fs_info;
 801         struct async_submit_bio *async;
 802         int limit;
 803
 804         async = container_of(work, struct  async_submit_bio, work);
 805         fs_info = BTRFS_I(async->inode)->root->fs_info;
 806
 807         limit = btrfs_async_submit_limit(fs_info);
 808         limit = limit * 2 / 3;
 809
 810         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 811             waitqueue_active(&fs_info->async_submit_wait))
 812                 wake_up(&fs_info->async_submit_wait);
 813
 814         /* If an error occured we just want to clean up the bio and move on */
 815         if (async->error) {
 816                 bio_endio(async->bio, async->error);
 817                 return;
 818         }
 819
 820         async->submit_bio_done(async->inode, async->rw, async->bio,
 821                                async->mirror_num, async->bio_flags,
 822                                async->bio_offset);
 823 }
 824
 825 static void run_one_async_free(struct btrfs_work *work)
 826 {
 827         struct async_submit_bio *async;
 828
 829         async = container_of(work, struct  async_submit_bio, work);
 830         kfree(async);
 831 }
 832
 833 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 834                         int rw, struct bio *bio, int mirror_num,
 835                         unsigned long bio_flags,
 836                         u64 bio_offset,
 837                         extent_submit_bio_hook_t *submit_bio_start,
 838                         extent_submit_bio_hook_t *submit_bio_done)
 839 {
 840         struct async_submit_bio *async;
 841
 842         async = kmalloc(sizeof(*async), GFP_NOFS);
 843         if (!async)
 844                 return -ENOMEM;
 845
 846         async->inode = inode;
 847         async->rw = rw;
 848         async->bio = bio;
 849         async->mirror_num = mirror_num;
 850         async->submit_bio_start = submit_bio_start;
 851         async->submit_bio_done = submit_bio_done;
 852
 853         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 854                         run_one_async_done, run_one_async_free);
 855
 856         async->bio_flags = bio_flags;
 857         async->bio_offset = bio_offset;
 858
 859         async->error = 0;
 860
 861         atomic_inc(&fs_info->nr_async_submits);
 862
 863         if (rw & REQ_SYNC)
 864                 btrfs_set_work_high_priority(&async->work);
 865
 866         btrfs_queue_work(fs_info->workers, &async->work);
 867
 868         while (atomic_read(&fs_info->async_submit_draining) &&
 869               atomic_read(&fs_info->nr_async_submits)) {
 870                 wait_event(fs_info->async_submit_wait,
 871                            (atomic_read(&fs_info->nr_async_submits) == 0));
 872         }
 873
 874         return 0;
 875 }
 876
 877 static int btree_csum_one_bio(struct bio *bio)
 878 {
 879         struct bio_vec *bvec;
 880         struct btrfs_root *root;
 881         int i, ret = 0;
 882
 883         bio_for_each_segment_all(bvec, bio, i) {
 884                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 885                 ret = csum_dirty_buffer(root, bvec->bv_page);
 886                 if (ret)
 887                         break;
 888         }
 889
 890         return ret;
 891 }
 892
 893 static int __btree_submit_bio_start(struct inode *inode, int rw,
 894                                     struct bio *bio, int mirror_num,
 895                                     unsigned long bio_flags,
 896                                     u64 bio_offset)
 897 {
 898         /*
 899          * when we're called for a write, we're already in the async
 900          * submission context.  Just jump into btrfs_map_bio
 901          */
 902         return btree_csum_one_bio(bio);
 903 }
 904
 905 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 906                                  int mirror_num, unsigned long bio_flags,
 907                                  u64 bio_offset)
 908 {
 909         int ret;
 910
 911         /*
 912          * when we're called for a write, we're already in the async
 913          * submission context.  Just jump into btrfs_map_bio
 914          */
 915         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 916         if (ret)
 917                 bio_endio(bio, ret);
 918         return ret;
 919 }
 920
 921 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 922 {
 923         if (bio_flags & EXTENT_BIO_TREE_LOG)
 924                 return 0;
 925 #ifdef CONFIG_X86
 926         if (cpu_has_xmm4_2)
 927                 return 0;
 928 #endif
 929         return 1;
 930 }
 931
 932 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 933                                  int mirror_num, unsigned long bio_flags,
 934                                  u64 bio_offset)
 935 {
 936         int async = check_async_write(inode, bio_flags);
 937         int ret;
 938
 939         if (!(rw & REQ_WRITE)) {
 940                 /*
 941                  * called for a read, do the setup so that checksum validation
 942                  * can happen in the async kernel threads
 943                  */
 944                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 945                                           bio, BTRFS_WQ_ENDIO_METADATA);
 946                 if (ret)
 947                         goto out_w_error;
 948                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 949                                     mirror_num, 0);
 950         } else if (!async) {
 951                 ret = btree_csum_one_bio(bio);
 952                 if (ret)
 953                         goto out_w_error;
 954                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 955                                     mirror_num, 0);
 956         } else {
 957                 /*
 958                  * kthread helpers are used to submit writes so that
 959                  * checksumming can happen in parallel across all CPUs
 960                  */
 961                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 962                                           inode, rw, bio, mirror_num, 0,
 963                                           bio_offset,
 964                                           __btree_submit_bio_start,
 965                                           __btree_submit_bio_done);
 966         }
 967
 968         if (ret) {
 969 out_w_error:
 970                 bio_endio(bio, ret);
 971         }
 972         return ret;
 973 }
 974
 975 #ifdef CONFIG_MIGRATION
 976 static int btree_migratepage(struct address_space *mapping,
 977                         struct page *newpage, struct page *page,
 978                         enum migrate_mode mode)
 979 {
 980         /*
 981          * we can't safely write a btree page from here,
 982          * we haven't done the locking hook
 983          */
 984         if (PageDirty(page))
 985                 return -EAGAIN;
 986         /*
 987          * Buffers may be managed in a filesystem specific way.
 988          * We must have no buffers or drop them.
 989          */
 990         if (page_has_private(page) &&
 991             !try_to_release_page(page, GFP_KERNEL))
 992                 return -EAGAIN;
 993         return migrate_page(mapping, newpage, page, mode);
 994 }
 995 #endif
 996
 997
 998 static int btree_writepages(struct address_space *mapping,
 999                             struct writeback_control *wbc)
1000 {
1001         struct btrfs_fs_info *fs_info;
1002         int ret;
1003
1004         if (wbc->sync_mode == WB_SYNC_NONE) {
1005
1006                 if (wbc->for_kupdate)
1007                         return 0;
1008
1009                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1010                 /* this is a bit racy, but that's ok */
1011                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1012                                              BTRFS_DIRTY_METADATA_THRESH);
1013                 if (ret < 0)
1014                         return 0;
1015         }
1016         return btree_write_cache_pages(mapping, wbc);
1017 }
1018
1019 static int btree_readpage(struct file *file, struct page *page)
1020 {
1021         struct extent_io_tree *tree;
1022         tree = &BTRFS_I(page->mapping->host)->io_tree;
1023         return extent_read_full_page(tree, page, btree_get_extent, 0);
1024 }
1025
1026 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1027 {
1028         if (PageWriteback(page) || PageDirty(page))
1029                 return 0;
1030
1031         return try_release_extent_buffer(page);
1032 }
1033
1034 static void btree_invalidatepage(struct page *page, unsigned int offset,
1035                                  unsigned int length)
1036 {
1037         struct extent_io_tree *tree;
1038         tree = &BTRFS_I(page->mapping->host)->io_tree;
1039         extent_invalidatepage(tree, page, offset);
1040         btree_releasepage(page, GFP_NOFS);
1041         if (PagePrivate(page)) {
1042                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1043                            "page private not zero on page %llu",
1044                            (unsigned long long)page_offset(page));
1045                 ClearPagePrivate(page);
1046                 set_page_private(page, 0);
1047                 page_cache_release(page);
1048         }
1049 }
1050
1051 static int btree_set_page_dirty(struct page *page)
1052 {
1053 #ifdef DEBUG
1054         struct extent_buffer *eb;
1055
1056         BUG_ON(!PagePrivate(page));
1057         eb = (struct extent_buffer *)page->private;
1058         BUG_ON(!eb);
1059         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1060         BUG_ON(!atomic_read(&eb->refs));
1061         btrfs_assert_tree_locked(eb);
1062 #endif
1063         return __set_page_dirty_nobuffers(page);
1064 }
1065
1066 static const struct address_space_operations btree_aops = {
1067         .readpage       = btree_readpage,
1068         .writepages     = btree_writepages,
1069         .releasepage    = btree_releasepage,
1070         .invalidatepage = btree_invalidatepage,
1071 #ifdef CONFIG_MIGRATION
1072         .migratepage    = btree_migratepage,
1073 #endif
1074         .set_page_dirty = btree_set_page_dirty,
1075 };
1076
1077 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1078 {
1079         struct extent_buffer *buf = NULL;
1080         struct inode *btree_inode = root->fs_info->btree_inode;
1081
1082         buf = btrfs_find_create_tree_block(root, bytenr);
1083         if (!buf)
1084                 return;
1085         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1086                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1087         free_extent_buffer(buf);
1088 }
1089
1090 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1091                          int mirror_num, struct extent_buffer **eb)
1092 {
1093         struct extent_buffer *buf = NULL;
1094         struct inode *btree_inode = root->fs_info->btree_inode;
1095         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1096         int ret;
1097
1098         buf = btrfs_find_create_tree_block(root, bytenr);
1099         if (!buf)
1100                 return 0;
1101
1102         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1103
1104         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1105                                        btree_get_extent, mirror_num);
1106         if (ret) {
1107                 free_extent_buffer(buf);
1108                 return ret;
1109         }
1110
1111         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1112                 free_extent_buffer(buf);
1113                 return -EIO;
1114         } else if (extent_buffer_uptodate(buf)) {
1115                 *eb = buf;
1116         } else {
1117                 free_extent_buffer(buf);
1118         }
1119         return 0;
1120 }
1121
1122 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1123                                             u64 bytenr)
1124 {
1125         return find_extent_buffer(root->fs_info, bytenr);
1126 }
1127
1128 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1129                                                  u64 bytenr)
1130 {
1131         if (btrfs_test_is_dummy_root(root))
1132                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1133         return alloc_extent_buffer(root->fs_info, bytenr);
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);
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         if (fs_info)
1279                 root->defrag_trans_start = fs_info->generation;
1280         else
1281                 root->defrag_trans_start = 0;
1282         root->root_key.objectid = objectid;
1283         root->anon_dev = 0;
1284
1285         spin_lock_init(&root->root_item_lock);
1286 }
1287
1288 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1289 {
1290         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1291         if (root)
1292                 root->fs_info = fs_info;
1293         return root;
1294 }
1295
1296 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1297 /* Should only be used by the testing infrastructure */
1298 struct btrfs_root *btrfs_alloc_dummy_root(void)
1299 {
1300         struct btrfs_root *root;
1301
1302         root = btrfs_alloc_root(NULL);
1303         if (!root)
1304                 return ERR_PTR(-ENOMEM);
1305         __setup_root(4096, 4096, 4096, root, NULL, 1);
1306         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1307         root->alloc_bytenr = 0;
1308
1309         return root;
1310 }
1311 #endif
1312
1313 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1314                                      struct btrfs_fs_info *fs_info,
1315                                      u64 objectid)
1316 {
1317         struct extent_buffer *leaf;
1318         struct btrfs_root *tree_root = fs_info->tree_root;
1319         struct btrfs_root *root;
1320         struct btrfs_key key;
1321         int ret = 0;
1322         uuid_le uuid;
1323
1324         root = btrfs_alloc_root(fs_info);
1325         if (!root)
1326                 return ERR_PTR(-ENOMEM);
1327
1328         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1329                 tree_root->stripesize, root, fs_info, objectid);
1330         root->root_key.objectid = objectid;
1331         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1332         root->root_key.offset = 0;
1333
1334         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1335         if (IS_ERR(leaf)) {
1336                 ret = PTR_ERR(leaf);
1337                 leaf = NULL;
1338                 goto fail;
1339         }
1340
1341         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1342         btrfs_set_header_bytenr(leaf, leaf->start);
1343         btrfs_set_header_generation(leaf, trans->transid);
1344         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1345         btrfs_set_header_owner(leaf, objectid);
1346         root->node = leaf;
1347
1348         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1349                             BTRFS_FSID_SIZE);
1350         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1351                             btrfs_header_chunk_tree_uuid(leaf),
1352                             BTRFS_UUID_SIZE);
1353         btrfs_mark_buffer_dirty(leaf);
1354
1355         root->commit_root = btrfs_root_node(root);
1356         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1357
1358         root->root_item.flags = 0;
1359         root->root_item.byte_limit = 0;
1360         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1361         btrfs_set_root_generation(&root->root_item, trans->transid);
1362         btrfs_set_root_level(&root->root_item, 0);
1363         btrfs_set_root_refs(&root->root_item, 1);
1364         btrfs_set_root_used(&root->root_item, leaf->len);
1365         btrfs_set_root_last_snapshot(&root->root_item, 0);
1366         btrfs_set_root_dirid(&root->root_item, 0);
1367         uuid_le_gen(&uuid);
1368         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1369         root->root_item.drop_level = 0;
1370
1371         key.objectid = objectid;
1372         key.type = BTRFS_ROOT_ITEM_KEY;
1373         key.offset = 0;
1374         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1375         if (ret)
1376                 goto fail;
1377
1378         btrfs_tree_unlock(leaf);
1379
1380         return root;
1381
1382 fail:
1383         if (leaf) {
1384                 btrfs_tree_unlock(leaf);
1385                 free_extent_buffer(root->commit_root);
1386                 free_extent_buffer(leaf);
1387         }
1388         kfree(root);
1389
1390         return ERR_PTR(ret);
1391 }
1392
1393 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1394                                          struct btrfs_fs_info *fs_info)
1395 {
1396         struct btrfs_root *root;
1397         struct btrfs_root *tree_root = fs_info->tree_root;
1398         struct extent_buffer *leaf;
1399
1400         root = btrfs_alloc_root(fs_info);
1401         if (!root)
1402                 return ERR_PTR(-ENOMEM);
1403
1404         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1405                      tree_root->stripesize, root, fs_info,
1406                      BTRFS_TREE_LOG_OBJECTID);
1407
1408         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1409         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1410         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1411
1412         /*
1413          * DON'T set REF_COWS for log trees
1414          *
1415          * log trees do not get reference counted because they go away
1416          * before a real commit is actually done.  They do store pointers
1417          * to file data extents, and those reference counts still get
1418          * updated (along with back refs to the log tree).
1419          */
1420
1421         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1422                         NULL, 0, 0, 0);
1423         if (IS_ERR(leaf)) {
1424                 kfree(root);
1425                 return ERR_CAST(leaf);
1426         }
1427
1428         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1429         btrfs_set_header_bytenr(leaf, leaf->start);
1430         btrfs_set_header_generation(leaf, trans->transid);
1431         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1432         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1433         root->node = leaf;
1434
1435         write_extent_buffer(root->node, root->fs_info->fsid,
1436                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1437         btrfs_mark_buffer_dirty(root->node);
1438         btrfs_tree_unlock(root->node);
1439         return root;
1440 }
1441
1442 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1443                              struct btrfs_fs_info *fs_info)
1444 {
1445         struct btrfs_root *log_root;
1446
1447         log_root = alloc_log_tree(trans, fs_info);
1448         if (IS_ERR(log_root))
1449                 return PTR_ERR(log_root);
1450         WARN_ON(fs_info->log_root_tree);
1451         fs_info->log_root_tree = log_root;
1452         return 0;
1453 }
1454
1455 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1456                        struct btrfs_root *root)
1457 {
1458         struct btrfs_root *log_root;
1459         struct btrfs_inode_item *inode_item;
1460
1461         log_root = alloc_log_tree(trans, root->fs_info);
1462         if (IS_ERR(log_root))
1463                 return PTR_ERR(log_root);
1464
1465         log_root->last_trans = trans->transid;
1466         log_root->root_key.offset = root->root_key.objectid;
1467
1468         inode_item = &log_root->root_item.inode;
1469         btrfs_set_stack_inode_generation(inode_item, 1);
1470         btrfs_set_stack_inode_size(inode_item, 3);
1471         btrfs_set_stack_inode_nlink(inode_item, 1);
1472         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1473         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1474
1475         btrfs_set_root_node(&log_root->root_item, log_root->node);
1476
1477         WARN_ON(root->log_root);
1478         root->log_root = log_root;
1479         root->log_transid = 0;
1480         root->log_transid_committed = -1;
1481         root->last_log_commit = 0;
1482         return 0;
1483 }
1484
1485 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1486                                                struct btrfs_key *key)
1487 {
1488         struct btrfs_root *root;
1489         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1490         struct btrfs_path *path;
1491         u64 generation;
1492         int ret;
1493
1494         path = btrfs_alloc_path();
1495         if (!path)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         root = btrfs_alloc_root(fs_info);
1499         if (!root) {
1500                 ret = -ENOMEM;
1501                 goto alloc_fail;
1502         }
1503
1504         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1505                 tree_root->stripesize, root, fs_info, key->objectid);
1506
1507         ret = btrfs_find_root(tree_root, key, path,
1508                               &root->root_item, &root->root_key);
1509         if (ret) {
1510                 if (ret > 0)
1511                         ret = -ENOENT;
1512                 goto find_fail;
1513         }
1514
1515         generation = btrfs_root_generation(&root->root_item);
1516         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1517                                      generation);
1518         if (!root->node) {
1519                 ret = -ENOMEM;
1520                 goto find_fail;
1521         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1522                 ret = -EIO;
1523                 goto read_fail;
1524         }
1525         root->commit_root = btrfs_root_node(root);
1526 out:
1527         btrfs_free_path(path);
1528         return root;
1529
1530 read_fail:
1531         free_extent_buffer(root->node);
1532 find_fail:
1533         kfree(root);
1534 alloc_fail:
1535         root = ERR_PTR(ret);
1536         goto out;
1537 }
1538
1539 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1540                                       struct btrfs_key *location)
1541 {
1542         struct btrfs_root *root;
1543
1544         root = btrfs_read_tree_root(tree_root, location);
1545         if (IS_ERR(root))
1546                 return root;
1547
1548         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1549                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1550                 btrfs_check_and_init_root_item(&root->root_item);
1551         }
1552
1553         return root;
1554 }
1555
1556 int btrfs_init_fs_root(struct btrfs_root *root)
1557 {
1558         int ret;
1559         struct btrfs_subvolume_writers *writers;
1560
1561         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1562         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1563                                         GFP_NOFS);
1564         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1565                 ret = -ENOMEM;
1566                 goto fail;
1567         }
1568
1569         writers = btrfs_alloc_subvolume_writers();
1570         if (IS_ERR(writers)) {
1571                 ret = PTR_ERR(writers);
1572                 goto fail;
1573         }
1574         root->subv_writers = writers;
1575
1576         btrfs_init_free_ino_ctl(root);
1577         spin_lock_init(&root->ino_cache_lock);
1578         init_waitqueue_head(&root->ino_cache_wait);
1579
1580         ret = get_anon_bdev(&root->anon_dev);
1581         if (ret)
1582                 goto free_writers;
1583         return 0;
1584
1585 free_writers:
1586         btrfs_free_subvolume_writers(root->subv_writers);
1587 fail:
1588         kfree(root->free_ino_ctl);
1589         kfree(root->free_ino_pinned);
1590         return ret;
1591 }
1592
1593 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1594                                                u64 root_id)
1595 {
1596         struct btrfs_root *root;
1597
1598         spin_lock(&fs_info->fs_roots_radix_lock);
1599         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1600                                  (unsigned long)root_id);
1601         spin_unlock(&fs_info->fs_roots_radix_lock);
1602         return root;
1603 }
1604
1605 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1606                          struct btrfs_root *root)
1607 {
1608         int ret;
1609
1610         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1611         if (ret)
1612                 return ret;
1613
1614         spin_lock(&fs_info->fs_roots_radix_lock);
1615         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1616                                 (unsigned long)root->root_key.objectid,
1617                                 root);
1618         if (ret == 0)
1619                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1620         spin_unlock(&fs_info->fs_roots_radix_lock);
1621         radix_tree_preload_end();
1622
1623         return ret;
1624 }
1625
1626 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1627                                      struct btrfs_key *location,
1628                                      bool check_ref)
1629 {
1630         struct btrfs_root *root;
1631         struct btrfs_path *path;
1632         struct btrfs_key key;
1633         int ret;
1634
1635         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1636                 return fs_info->tree_root;
1637         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1638                 return fs_info->extent_root;
1639         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1640                 return fs_info->chunk_root;
1641         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1642                 return fs_info->dev_root;
1643         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1644                 return fs_info->csum_root;
1645         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1646                 return fs_info->quota_root ? fs_info->quota_root :
1647                                              ERR_PTR(-ENOENT);
1648         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1649                 return fs_info->uuid_root ? fs_info->uuid_root :
1650                                             ERR_PTR(-ENOENT);
1651 again:
1652         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1653         if (root) {
1654                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1655                         return ERR_PTR(-ENOENT);
1656                 return root;
1657         }
1658
1659         root = btrfs_read_fs_root(fs_info->tree_root, location);
1660         if (IS_ERR(root))
1661                 return root;
1662
1663         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1664                 ret = -ENOENT;
1665                 goto fail;
1666         }
1667
1668         ret = btrfs_init_fs_root(root);
1669         if (ret)
1670                 goto fail;
1671
1672         path = btrfs_alloc_path();
1673         if (!path) {
1674                 ret = -ENOMEM;
1675                 goto fail;
1676         }
1677         key.objectid = BTRFS_ORPHAN_OBJECTID;
1678         key.type = BTRFS_ORPHAN_ITEM_KEY;
1679         key.offset = location->objectid;
1680
1681         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1682         btrfs_free_path(path);
1683         if (ret < 0)
1684                 goto fail;
1685         if (ret == 0)
1686                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1687
1688         ret = btrfs_insert_fs_root(fs_info, root);
1689         if (ret) {
1690                 if (ret == -EEXIST) {
1691                         free_fs_root(root);
1692                         goto again;
1693                 }
1694                 goto fail;
1695         }
1696         return root;
1697 fail:
1698         free_fs_root(root);
1699         return ERR_PTR(ret);
1700 }
1701
1702 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1703 {
1704         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1705         int ret = 0;
1706         struct btrfs_device *device;
1707         struct backing_dev_info *bdi;
1708
1709         rcu_read_lock();
1710         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1711                 if (!device->bdev)
1712                         continue;
1713                 bdi = blk_get_backing_dev_info(device->bdev);
1714                 if (bdi_congested(bdi, bdi_bits)) {
1715                         ret = 1;
1716                         break;
1717                 }
1718         }
1719         rcu_read_unlock();
1720         return ret;
1721 }
1722
1723 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1724 {
1725         int err;
1726
1727         err = bdi_setup_and_register(bdi, "btrfs");
1728         if (err)
1729                 return err;
1730
1731         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1732         bdi->congested_fn       = btrfs_congested_fn;
1733         bdi->congested_data     = info;
1734         return 0;
1735 }
1736
1737 /*
1738  * called by the kthread helper functions to finally call the bio end_io
1739  * functions.  This is where read checksum verification actually happens
1740  */
1741 static void end_workqueue_fn(struct btrfs_work *work)
1742 {
1743         struct bio *bio;
1744         struct btrfs_end_io_wq *end_io_wq;
1745         int error;
1746
1747         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1748         bio = end_io_wq->bio;
1749
1750         error = end_io_wq->error;
1751         bio->bi_private = end_io_wq->private;
1752         bio->bi_end_io = end_io_wq->end_io;
1753         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1754         bio_endio_nodec(bio, error);
1755 }
1756
1757 static int cleaner_kthread(void *arg)
1758 {
1759         struct btrfs_root *root = arg;
1760         int again;
1761
1762         do {
1763                 again = 0;
1764
1765                 /* Make the cleaner go to sleep early. */
1766                 if (btrfs_need_cleaner_sleep(root))
1767                         goto sleep;
1768
1769                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1770                         goto sleep;
1771
1772                 /*
1773                  * Avoid the problem that we change the status of the fs
1774                  * during the above check and trylock.
1775                  */
1776                 if (btrfs_need_cleaner_sleep(root)) {
1777                         mutex_unlock(&root->fs_info->cleaner_mutex);
1778                         goto sleep;
1779                 }
1780
1781                 btrfs_run_delayed_iputs(root);
1782                 btrfs_delete_unused_bgs(root->fs_info);
1783                 again = btrfs_clean_one_deleted_snapshot(root);
1784                 mutex_unlock(&root->fs_info->cleaner_mutex);
1785
1786                 /*
1787                  * The defragger has dealt with the R/O remount and umount,
1788                  * needn't do anything special here.
1789                  */
1790                 btrfs_run_defrag_inodes(root->fs_info);
1791 sleep:
1792                 if (!try_to_freeze() && !again) {
1793                         set_current_state(TASK_INTERRUPTIBLE);
1794                         if (!kthread_should_stop())
1795                                 schedule();
1796                         __set_current_state(TASK_RUNNING);
1797                 }
1798         } while (!kthread_should_stop());
1799         return 0;
1800 }
1801
1802 static int transaction_kthread(void *arg)
1803 {
1804         struct btrfs_root *root = arg;
1805         struct btrfs_trans_handle *trans;
1806         struct btrfs_transaction *cur;
1807         u64 transid;
1808         unsigned long now;
1809         unsigned long delay;
1810         bool cannot_commit;
1811
1812         do {
1813                 cannot_commit = false;
1814                 delay = HZ * root->fs_info->commit_interval;
1815                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1816
1817                 spin_lock(&root->fs_info->trans_lock);
1818                 cur = root->fs_info->running_transaction;
1819                 if (!cur) {
1820                         spin_unlock(&root->fs_info->trans_lock);
1821                         goto sleep;
1822                 }
1823
1824                 now = get_seconds();
1825                 if (cur->state < TRANS_STATE_BLOCKED &&
1826                     (now < cur->start_time ||
1827                      now - cur->start_time < root->fs_info->commit_interval)) {
1828                         spin_unlock(&root->fs_info->trans_lock);
1829                         delay = HZ * 5;
1830                         goto sleep;
1831                 }
1832                 transid = cur->transid;
1833                 spin_unlock(&root->fs_info->trans_lock);
1834
1835                 /* If the file system is aborted, this will always fail. */
1836                 trans = btrfs_attach_transaction(root);
1837                 if (IS_ERR(trans)) {
1838                         if (PTR_ERR(trans) != -ENOENT)
1839                                 cannot_commit = true;
1840                         goto sleep;
1841                 }
1842                 if (transid == trans->transid) {
1843                         btrfs_commit_transaction(trans, root);
1844                 } else {
1845                         btrfs_end_transaction(trans, root);
1846                 }
1847 sleep:
1848                 wake_up_process(root->fs_info->cleaner_kthread);
1849                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1850
1851                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1852                                       &root->fs_info->fs_state)))
1853                         btrfs_cleanup_transaction(root);
1854                 if (!try_to_freeze()) {
1855                         set_current_state(TASK_INTERRUPTIBLE);
1856                         if (!kthread_should_stop() &&
1857                             (!btrfs_transaction_blocked(root->fs_info) ||
1858                              cannot_commit))
1859                                 schedule_timeout(delay);
1860                         __set_current_state(TASK_RUNNING);
1861                 }
1862         } while (!kthread_should_stop());
1863         return 0;
1864 }
1865
1866 /*
1867  * this will find the highest generation in the array of
1868  * root backups.  The index of the highest array is returned,
1869  * or -1 if we can't find anything.
1870  *
1871  * We check to make sure the array is valid by comparing the
1872  * generation of the latest  root in the array with the generation
1873  * in the super block.  If they don't match we pitch it.
1874  */
1875 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1876 {
1877         u64 cur;
1878         int newest_index = -1;
1879         struct btrfs_root_backup *root_backup;
1880         int i;
1881
1882         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1883                 root_backup = info->super_copy->super_roots + i;
1884                 cur = btrfs_backup_tree_root_gen(root_backup);
1885                 if (cur == newest_gen)
1886                         newest_index = i;
1887         }
1888
1889         /* check to see if we actually wrapped around */
1890         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1891                 root_backup = info->super_copy->super_roots;
1892                 cur = btrfs_backup_tree_root_gen(root_backup);
1893                 if (cur == newest_gen)
1894                         newest_index = 0;
1895         }
1896         return newest_index;
1897 }
1898
1899
1900 /*
1901  * find the oldest backup so we know where to store new entries
1902  * in the backup array.  This will set the backup_root_index
1903  * field in the fs_info struct
1904  */
1905 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1906                                      u64 newest_gen)
1907 {
1908         int newest_index = -1;
1909
1910         newest_index = find_newest_super_backup(info, newest_gen);
1911         /* if there was garbage in there, just move along */
1912         if (newest_index == -1) {
1913                 info->backup_root_index = 0;
1914         } else {
1915                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1916         }
1917 }
1918
1919 /*
1920  * copy all the root pointers into the super backup array.
1921  * this will bump the backup pointer by one when it is
1922  * done
1923  */
1924 static void backup_super_roots(struct btrfs_fs_info *info)
1925 {
1926         int next_backup;
1927         struct btrfs_root_backup *root_backup;
1928         int last_backup;
1929
1930         next_backup = info->backup_root_index;
1931         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1932                 BTRFS_NUM_BACKUP_ROOTS;
1933
1934         /*
1935          * just overwrite the last backup if we're at the same generation
1936          * this happens only at umount
1937          */
1938         root_backup = info->super_for_commit->super_roots + last_backup;
1939         if (btrfs_backup_tree_root_gen(root_backup) ==
1940             btrfs_header_generation(info->tree_root->node))
1941                 next_backup = last_backup;
1942
1943         root_backup = info->super_for_commit->super_roots + next_backup;
1944
1945         /*
1946          * make sure all of our padding and empty slots get zero filled
1947          * regardless of which ones we use today
1948          */
1949         memset(root_backup, 0, sizeof(*root_backup));
1950
1951         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1952
1953         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1954         btrfs_set_backup_tree_root_gen(root_backup,
1955                                btrfs_header_generation(info->tree_root->node));
1956
1957         btrfs_set_backup_tree_root_level(root_backup,
1958                                btrfs_header_level(info->tree_root->node));
1959
1960         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1961         btrfs_set_backup_chunk_root_gen(root_backup,
1962                                btrfs_header_generation(info->chunk_root->node));
1963         btrfs_set_backup_chunk_root_level(root_backup,
1964                                btrfs_header_level(info->chunk_root->node));
1965
1966         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1967         btrfs_set_backup_extent_root_gen(root_backup,
1968                                btrfs_header_generation(info->extent_root->node));
1969         btrfs_set_backup_extent_root_level(root_backup,
1970                                btrfs_header_level(info->extent_root->node));
1971
1972         /*
1973          * we might commit during log recovery, which happens before we set
1974          * the fs_root.  Make sure it is valid before we fill it in.
1975          */
1976         if (info->fs_root && info->fs_root->node) {
1977                 btrfs_set_backup_fs_root(root_backup,
1978                                          info->fs_root->node->start);
1979                 btrfs_set_backup_fs_root_gen(root_backup,
1980                                btrfs_header_generation(info->fs_root->node));
1981                 btrfs_set_backup_fs_root_level(root_backup,
1982                                btrfs_header_level(info->fs_root->node));
1983         }
1984
1985         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1986         btrfs_set_backup_dev_root_gen(root_backup,
1987                                btrfs_header_generation(info->dev_root->node));
1988         btrfs_set_backup_dev_root_level(root_backup,
1989                                        btrfs_header_level(info->dev_root->node));
1990
1991         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1992         btrfs_set_backup_csum_root_gen(root_backup,
1993                                btrfs_header_generation(info->csum_root->node));
1994         btrfs_set_backup_csum_root_level(root_backup,
1995                                btrfs_header_level(info->csum_root->node));
1996
1997         btrfs_set_backup_total_bytes(root_backup,
1998                              btrfs_super_total_bytes(info->super_copy));
1999         btrfs_set_backup_bytes_used(root_backup,
2000                              btrfs_super_bytes_used(info->super_copy));
2001         btrfs_set_backup_num_devices(root_backup,
2002                              btrfs_super_num_devices(info->super_copy));
2003
2004         /*
2005          * if we don't copy this out to the super_copy, it won't get remembered
2006          * for the next commit
2007          */
2008         memcpy(&info->super_copy->super_roots,
2009                &info->super_for_commit->super_roots,
2010                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2011 }
2012
2013 /*
2014  * this copies info out of the root backup array and back into
2015  * the in-memory super block.  It is meant to help iterate through
2016  * the array, so you send it the number of backups you've already
2017  * tried and the last backup index you used.
2018  *
2019  * this returns -1 when it has tried all the backups
2020  */
2021 static noinline int next_root_backup(struct btrfs_fs_info *info,
2022                                      struct btrfs_super_block *super,
2023                                      int *num_backups_tried, int *backup_index)
2024 {
2025         struct btrfs_root_backup *root_backup;
2026         int newest = *backup_index;
2027
2028         if (*num_backups_tried == 0) {
2029                 u64 gen = btrfs_super_generation(super);
2030
2031                 newest = find_newest_super_backup(info, gen);
2032                 if (newest == -1)
2033                         return -1;
2034
2035                 *backup_index = newest;
2036                 *num_backups_tried = 1;
2037         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2038                 /* we've tried all the backups, all done */
2039                 return -1;
2040         } else {
2041                 /* jump to the next oldest backup */
2042                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2043                         BTRFS_NUM_BACKUP_ROOTS;
2044                 *backup_index = newest;
2045                 *num_backups_tried += 1;
2046         }
2047         root_backup = super->super_roots + newest;
2048
2049         btrfs_set_super_generation(super,
2050                                    btrfs_backup_tree_root_gen(root_backup));
2051         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2052         btrfs_set_super_root_level(super,
2053                                    btrfs_backup_tree_root_level(root_backup));
2054         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2055
2056         /*
2057          * fixme: the total bytes and num_devices need to match or we should
2058          * need a fsck
2059          */
2060         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2061         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2062         return 0;
2063 }
2064
2065 /* helper to cleanup workers */
2066 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2067 {
2068         btrfs_destroy_workqueue(fs_info->fixup_workers);
2069         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2070         btrfs_destroy_workqueue(fs_info->workers);
2071         btrfs_destroy_workqueue(fs_info->endio_workers);
2072         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2074         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2075         btrfs_destroy_workqueue(fs_info->rmw_workers);
2076         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2077         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2078         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2079         btrfs_destroy_workqueue(fs_info->submit_workers);
2080         btrfs_destroy_workqueue(fs_info->delayed_workers);
2081         btrfs_destroy_workqueue(fs_info->caching_workers);
2082         btrfs_destroy_workqueue(fs_info->readahead_workers);
2083         btrfs_destroy_workqueue(fs_info->flush_workers);
2084         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2085         btrfs_destroy_workqueue(fs_info->extent_workers);
2086 }
2087
2088 static void free_root_extent_buffers(struct btrfs_root *root)
2089 {
2090         if (root) {
2091                 free_extent_buffer(root->node);
2092                 free_extent_buffer(root->commit_root);
2093                 root->node = NULL;
2094                 root->commit_root = NULL;
2095         }
2096 }
2097
2098 /* helper to cleanup tree roots */
2099 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2100 {
2101         free_root_extent_buffers(info->tree_root);
2102
2103         free_root_extent_buffers(info->dev_root);
2104         free_root_extent_buffers(info->extent_root);
2105         free_root_extent_buffers(info->csum_root);
2106         free_root_extent_buffers(info->quota_root);
2107         free_root_extent_buffers(info->uuid_root);
2108         if (chunk_root)
2109                 free_root_extent_buffers(info->chunk_root);
2110 }
2111
2112 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2113 {
2114         int ret;
2115         struct btrfs_root *gang[8];
2116         int i;
2117
2118         while (!list_empty(&fs_info->dead_roots)) {
2119                 gang[0] = list_entry(fs_info->dead_roots.next,
2120                                      struct btrfs_root, root_list);
2121                 list_del(&gang[0]->root_list);
2122
2123                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2124                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2125                 } else {
2126                         free_extent_buffer(gang[0]->node);
2127                         free_extent_buffer(gang[0]->commit_root);
2128                         btrfs_put_fs_root(gang[0]);
2129                 }
2130         }
2131
2132         while (1) {
2133                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2134                                              (void **)gang, 0,
2135                                              ARRAY_SIZE(gang));
2136                 if (!ret)
2137                         break;
2138                 for (i = 0; i < ret; i++)
2139                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2140         }
2141
2142         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2143                 btrfs_free_log_root_tree(NULL, fs_info);
2144                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2145                                             fs_info->pinned_extents);
2146         }
2147 }
2148
2149 int open_ctree(struct super_block *sb,
2150                struct btrfs_fs_devices *fs_devices,
2151                char *options)
2152 {
2153         u32 sectorsize;
2154         u32 nodesize;
2155         u32 stripesize;
2156         u64 generation;
2157         u64 features;
2158         struct btrfs_key location;
2159         struct buffer_head *bh;
2160         struct btrfs_super_block *disk_super;
2161         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2162         struct btrfs_root *tree_root;
2163         struct btrfs_root *extent_root;
2164         struct btrfs_root *csum_root;
2165         struct btrfs_root *chunk_root;
2166         struct btrfs_root *dev_root;
2167         struct btrfs_root *quota_root;
2168         struct btrfs_root *uuid_root;
2169         struct btrfs_root *log_tree_root;
2170         int ret;
2171         int err = -EINVAL;
2172         int num_backups_tried = 0;
2173         int backup_index = 0;
2174         int max_active;
2175         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2176         bool create_uuid_tree;
2177         bool check_uuid_tree;
2178
2179         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2180         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2181         if (!tree_root || !chunk_root) {
2182                 err = -ENOMEM;
2183                 goto fail;
2184         }
2185
2186         ret = init_srcu_struct(&fs_info->subvol_srcu);
2187         if (ret) {
2188                 err = ret;
2189                 goto fail;
2190         }
2191
2192         ret = setup_bdi(fs_info, &fs_info->bdi);
2193         if (ret) {
2194                 err = ret;
2195                 goto fail_srcu;
2196         }
2197
2198         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2199         if (ret) {
2200                 err = ret;
2201                 goto fail_bdi;
2202         }
2203         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2204                                         (1 + ilog2(nr_cpu_ids));
2205
2206         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2207         if (ret) {
2208                 err = ret;
2209                 goto fail_dirty_metadata_bytes;
2210         }
2211
2212         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2213         if (ret) {
2214                 err = ret;
2215                 goto fail_delalloc_bytes;
2216         }
2217
2218         fs_info->btree_inode = new_inode(sb);
2219         if (!fs_info->btree_inode) {
2220                 err = -ENOMEM;
2221                 goto fail_bio_counter;
2222         }
2223
2224         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2225
2226         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2227         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2228         INIT_LIST_HEAD(&fs_info->trans_list);
2229         INIT_LIST_HEAD(&fs_info->dead_roots);
2230         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2231         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2232         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2233         spin_lock_init(&fs_info->delalloc_root_lock);
2234         spin_lock_init(&fs_info->trans_lock);
2235         spin_lock_init(&fs_info->fs_roots_radix_lock);
2236         spin_lock_init(&fs_info->delayed_iput_lock);
2237         spin_lock_init(&fs_info->defrag_inodes_lock);
2238         spin_lock_init(&fs_info->free_chunk_lock);
2239         spin_lock_init(&fs_info->tree_mod_seq_lock);
2240         spin_lock_init(&fs_info->super_lock);
2241         spin_lock_init(&fs_info->qgroup_op_lock);
2242         spin_lock_init(&fs_info->buffer_lock);
2243         spin_lock_init(&fs_info->unused_bgs_lock);
2244         mutex_init(&fs_info->unused_bg_unpin_mutex);
2245         rwlock_init(&fs_info->tree_mod_log_lock);
2246         mutex_init(&fs_info->reloc_mutex);
2247         mutex_init(&fs_info->delalloc_root_mutex);
2248         seqlock_init(&fs_info->profiles_lock);
2249
2250         init_completion(&fs_info->kobj_unregister);
2251         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2252         INIT_LIST_HEAD(&fs_info->space_info);
2253         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2254         INIT_LIST_HEAD(&fs_info->unused_bgs);
2255         btrfs_mapping_init(&fs_info->mapping_tree);
2256         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2257                              BTRFS_BLOCK_RSV_GLOBAL);
2258         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2259                              BTRFS_BLOCK_RSV_DELALLOC);
2260         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2261         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2262         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2263         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2264                              BTRFS_BLOCK_RSV_DELOPS);
2265         atomic_set(&fs_info->nr_async_submits, 0);
2266         atomic_set(&fs_info->async_delalloc_pages, 0);
2267         atomic_set(&fs_info->async_submit_draining, 0);
2268         atomic_set(&fs_info->nr_async_bios, 0);
2269         atomic_set(&fs_info->defrag_running, 0);
2270         atomic_set(&fs_info->qgroup_op_seq, 0);
2271         atomic64_set(&fs_info->tree_mod_seq, 0);
2272         fs_info->sb = sb;
2273         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2274         fs_info->metadata_ratio = 0;
2275         fs_info->defrag_inodes = RB_ROOT;
2276         fs_info->free_chunk_space = 0;
2277         fs_info->tree_mod_log = RB_ROOT;
2278         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2279         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2280         /* readahead state */
2281         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2282         spin_lock_init(&fs_info->reada_lock);
2283
2284         fs_info->thread_pool_size = min_t(unsigned long,
2285                                           num_online_cpus() + 2, 8);
2286
2287         INIT_LIST_HEAD(&fs_info->ordered_roots);
2288         spin_lock_init(&fs_info->ordered_root_lock);
2289         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2290                                         GFP_NOFS);
2291         if (!fs_info->delayed_root) {
2292                 err = -ENOMEM;
2293                 goto fail_iput;
2294         }
2295         btrfs_init_delayed_root(fs_info->delayed_root);
2296
2297         mutex_init(&fs_info->scrub_lock);
2298         atomic_set(&fs_info->scrubs_running, 0);
2299         atomic_set(&fs_info->scrub_pause_req, 0);
2300         atomic_set(&fs_info->scrubs_paused, 0);
2301         atomic_set(&fs_info->scrub_cancel_req, 0);
2302         init_waitqueue_head(&fs_info->replace_wait);
2303         init_waitqueue_head(&fs_info->scrub_pause_wait);
2304         fs_info->scrub_workers_refcnt = 0;
2305 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2306         fs_info->check_integrity_print_mask = 0;
2307 #endif
2308
2309         spin_lock_init(&fs_info->balance_lock);
2310         mutex_init(&fs_info->balance_mutex);
2311         atomic_set(&fs_info->balance_running, 0);
2312         atomic_set(&fs_info->balance_pause_req, 0);
2313         atomic_set(&fs_info->balance_cancel_req, 0);
2314         fs_info->balance_ctl = NULL;
2315         init_waitqueue_head(&fs_info->balance_wait_q);
2316         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2317
2318         sb->s_blocksize = 4096;
2319         sb->s_blocksize_bits = blksize_bits(4096);
2320         sb->s_bdi = &fs_info->bdi;
2321
2322         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2323         set_nlink(fs_info->btree_inode, 1);
2324         /*
2325          * we set the i_size on the btree inode to the max possible int.
2326          * the real end of the address space is determined by all of
2327          * the devices in the system
2328          */
2329         fs_info->btree_inode->i_size = OFFSET_MAX;
2330         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2331
2332         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2333         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2334                              fs_info->btree_inode->i_mapping);
2335         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2336         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2337
2338         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2339
2340         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2341         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2342                sizeof(struct btrfs_key));
2343         set_bit(BTRFS_INODE_DUMMY,
2344                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2345         btrfs_insert_inode_hash(fs_info->btree_inode);
2346
2347         spin_lock_init(&fs_info->block_group_cache_lock);
2348         fs_info->block_group_cache_tree = RB_ROOT;
2349         fs_info->first_logical_byte = (u64)-1;
2350
2351         extent_io_tree_init(&fs_info->freed_extents[0],
2352                              fs_info->btree_inode->i_mapping);
2353         extent_io_tree_init(&fs_info->freed_extents[1],
2354                              fs_info->btree_inode->i_mapping);
2355         fs_info->pinned_extents = &fs_info->freed_extents[0];
2356         fs_info->do_barriers = 1;
2357
2358
2359         mutex_init(&fs_info->ordered_operations_mutex);
2360         mutex_init(&fs_info->ordered_extent_flush_mutex);
2361         mutex_init(&fs_info->tree_log_mutex);
2362         mutex_init(&fs_info->chunk_mutex);
2363         mutex_init(&fs_info->transaction_kthread_mutex);
2364         mutex_init(&fs_info->cleaner_mutex);
2365         mutex_init(&fs_info->volume_mutex);
2366         init_rwsem(&fs_info->commit_root_sem);
2367         init_rwsem(&fs_info->cleanup_work_sem);
2368         init_rwsem(&fs_info->subvol_sem);
2369         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2370         fs_info->dev_replace.lock_owner = 0;
2371         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2372         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2373         mutex_init(&fs_info->dev_replace.lock_management_lock);
2374         mutex_init(&fs_info->dev_replace.lock);
2375
2376         spin_lock_init(&fs_info->qgroup_lock);
2377         mutex_init(&fs_info->qgroup_ioctl_lock);
2378         fs_info->qgroup_tree = RB_ROOT;
2379         fs_info->qgroup_op_tree = RB_ROOT;
2380         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2381         fs_info->qgroup_seq = 1;
2382         fs_info->quota_enabled = 0;
2383         fs_info->pending_quota_state = 0;
2384         fs_info->qgroup_ulist = NULL;
2385         mutex_init(&fs_info->qgroup_rescan_lock);
2386
2387         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2388         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2389
2390         init_waitqueue_head(&fs_info->transaction_throttle);
2391         init_waitqueue_head(&fs_info->transaction_wait);
2392         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2393         init_waitqueue_head(&fs_info->async_submit_wait);
2394
2395         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2396
2397         ret = btrfs_alloc_stripe_hash_table(fs_info);
2398         if (ret) {
2399                 err = ret;
2400                 goto fail_alloc;
2401         }
2402
2403         __setup_root(4096, 4096, 4096, tree_root,
2404                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2405
2406         invalidate_bdev(fs_devices->latest_bdev);
2407
2408         /*
2409          * Read super block and check the signature bytes only
2410          */
2411         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2412         if (!bh) {
2413                 err = -EINVAL;
2414                 goto fail_alloc;
2415         }
2416
2417         /*
2418          * We want to check superblock checksum, the type is stored inside.
2419          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2420          */
2421         if (btrfs_check_super_csum(bh->b_data)) {
2422                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2423                 err = -EINVAL;
2424                 goto fail_alloc;
2425         }
2426
2427         /*
2428          * super_copy is zeroed at allocation time and we never touch the
2429          * following bytes up to INFO_SIZE, the checksum is calculated from
2430          * the whole block of INFO_SIZE
2431          */
2432         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2433         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2434                sizeof(*fs_info->super_for_commit));
2435         brelse(bh);
2436
2437         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2438
2439         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2440         if (ret) {
2441                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2442                 err = -EINVAL;
2443                 goto fail_alloc;
2444         }
2445
2446         disk_super = fs_info->super_copy;
2447         if (!btrfs_super_root(disk_super))
2448                 goto fail_alloc;
2449
2450         /* check FS state, whether FS is broken. */
2451         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2452                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2453
2454         /*
2455          * run through our array of backup supers and setup
2456          * our ring pointer to the oldest one
2457          */
2458         generation = btrfs_super_generation(disk_super);
2459         find_oldest_super_backup(fs_info, generation);
2460
2461         /*
2462          * In the long term, we'll store the compression type in the super
2463          * block, and it'll be used for per file compression control.
2464          */
2465         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2466
2467         ret = btrfs_parse_options(tree_root, options);
2468         if (ret) {
2469                 err = ret;
2470                 goto fail_alloc;
2471         }
2472
2473         features = btrfs_super_incompat_flags(disk_super) &
2474                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2475         if (features) {
2476                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2477                        "unsupported optional features (%Lx).\n",
2478                        features);
2479                 err = -EINVAL;
2480                 goto fail_alloc;
2481         }
2482
2483         /*
2484          * Leafsize and nodesize were always equal, this is only a sanity check.
2485          */
2486         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2487             btrfs_super_nodesize(disk_super)) {
2488                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2489                        "blocksizes don't match.  node %d leaf %d\n",
2490                        btrfs_super_nodesize(disk_super),
2491                        le32_to_cpu(disk_super->__unused_leafsize));
2492                 err = -EINVAL;
2493                 goto fail_alloc;
2494         }
2495         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2496                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2497                        "blocksize (%d) was too large\n",
2498                        btrfs_super_nodesize(disk_super));
2499                 err = -EINVAL;
2500                 goto fail_alloc;
2501         }
2502
2503         features = btrfs_super_incompat_flags(disk_super);
2504         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2505         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2506                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2507
2508         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2509                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2510
2511         /*
2512          * flag our filesystem as having big metadata blocks if
2513          * they are bigger than the page size
2514          */
2515         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2516                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2517                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2518                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2519         }
2520
2521         nodesize = btrfs_super_nodesize(disk_super);
2522         sectorsize = btrfs_super_sectorsize(disk_super);
2523         stripesize = btrfs_super_stripesize(disk_super);
2524         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2525         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2526
2527         /*
2528          * mixed block groups end up with duplicate but slightly offset
2529          * extent buffers for the same range.  It leads to corruptions
2530          */
2531         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2532             (sectorsize != nodesize)) {
2533                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2534                                 "are not allowed for mixed block groups on %s\n",
2535                                 sb->s_id);
2536                 goto fail_alloc;
2537         }
2538
2539         /*
2540          * Needn't use the lock because there is no other task which will
2541          * update the flag.
2542          */
2543         btrfs_set_super_incompat_flags(disk_super, features);
2544
2545         features = btrfs_super_compat_ro_flags(disk_super) &
2546                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2547         if (!(sb->s_flags & MS_RDONLY) && features) {
2548                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2549                        "unsupported option features (%Lx).\n",
2550                        features);
2551                 err = -EINVAL;
2552                 goto fail_alloc;
2553         }
2554
2555         max_active = fs_info->thread_pool_size;
2556
2557         fs_info->workers =
2558                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2559                                       max_active, 16);
2560
2561         fs_info->delalloc_workers =
2562                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2563
2564         fs_info->flush_workers =
2565                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2566
2567         fs_info->caching_workers =
2568                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2569
2570         /*
2571          * a higher idle thresh on the submit workers makes it much more
2572          * likely that bios will be send down in a sane order to the
2573          * devices
2574          */
2575         fs_info->submit_workers =
2576                 btrfs_alloc_workqueue("submit", flags,
2577                                       min_t(u64, fs_devices->num_devices,
2578                                             max_active), 64);
2579
2580         fs_info->fixup_workers =
2581                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2582
2583         /*
2584          * endios are largely parallel and should have a very
2585          * low idle thresh
2586          */
2587         fs_info->endio_workers =
2588                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2589         fs_info->endio_meta_workers =
2590                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2591         fs_info->endio_meta_write_workers =
2592                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2593         fs_info->endio_raid56_workers =
2594                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2595         fs_info->endio_repair_workers =
2596                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2597         fs_info->rmw_workers =
2598                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2599         fs_info->endio_write_workers =
2600                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2601         fs_info->endio_freespace_worker =
2602                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2603         fs_info->delayed_workers =
2604                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2605         fs_info->readahead_workers =
2606                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2607         fs_info->qgroup_rescan_workers =
2608                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2609         fs_info->extent_workers =
2610                 btrfs_alloc_workqueue("extent-refs", flags,
2611                                       min_t(u64, fs_devices->num_devices,
2612                                             max_active), 8);
2613
2614         if (!(fs_info->workers && fs_info->delalloc_workers &&
2615               fs_info->submit_workers && fs_info->flush_workers &&
2616               fs_info->endio_workers && fs_info->endio_meta_workers &&
2617               fs_info->endio_meta_write_workers &&
2618               fs_info->endio_repair_workers &&
2619               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2620               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2621               fs_info->caching_workers && fs_info->readahead_workers &&
2622               fs_info->fixup_workers && fs_info->delayed_workers &&
2623               fs_info->extent_workers &&
2624               fs_info->qgroup_rescan_workers)) {
2625                 err = -ENOMEM;
2626                 goto fail_sb_buffer;
2627         }
2628
2629         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2630         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2631                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2632
2633         tree_root->nodesize = nodesize;
2634         tree_root->sectorsize = sectorsize;
2635         tree_root->stripesize = stripesize;
2636
2637         sb->s_blocksize = sectorsize;
2638         sb->s_blocksize_bits = blksize_bits(sectorsize);
2639
2640         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2641                 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2642                 goto fail_sb_buffer;
2643         }
2644
2645         if (sectorsize != PAGE_SIZE) {
2646                 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2647                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2648                 goto fail_sb_buffer;
2649         }
2650
2651         mutex_lock(&fs_info->chunk_mutex);
2652         ret = btrfs_read_sys_array(tree_root);
2653         mutex_unlock(&fs_info->chunk_mutex);
2654         if (ret) {
2655                 printk(KERN_ERR "BTRFS: failed to read the system "
2656                        "array on %s\n", sb->s_id);
2657                 goto fail_sb_buffer;
2658         }
2659
2660         generation = btrfs_super_chunk_root_generation(disk_super);
2661
2662         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2663                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2664
2665         chunk_root->node = read_tree_block(chunk_root,
2666                                            btrfs_super_chunk_root(disk_super),
2667                                            generation);
2668         if (!chunk_root->node ||
2669             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2670                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2671                        sb->s_id);
2672                 goto fail_tree_roots;
2673         }
2674         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2675         chunk_root->commit_root = btrfs_root_node(chunk_root);
2676
2677         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2678            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2679
2680         ret = btrfs_read_chunk_tree(chunk_root);
2681         if (ret) {
2682                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2683                        sb->s_id);
2684                 goto fail_tree_roots;
2685         }
2686
2687         /*
2688          * keep the device that is marked to be the target device for the
2689          * dev_replace procedure
2690          */
2691         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2692
2693         if (!fs_devices->latest_bdev) {
2694                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2695                        sb->s_id);
2696                 goto fail_tree_roots;
2697         }
2698
2699 retry_root_backup:
2700         generation = btrfs_super_generation(disk_super);
2701
2702         tree_root->node = read_tree_block(tree_root,
2703                                           btrfs_super_root(disk_super),
2704                                           generation);
2705         if (!tree_root->node ||
2706             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2707                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2708                        sb->s_id);
2709
2710                 goto recovery_tree_root;
2711         }
2712
2713         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2714         tree_root->commit_root = btrfs_root_node(tree_root);
2715         btrfs_set_root_refs(&tree_root->root_item, 1);
2716
2717         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2718         location.type = BTRFS_ROOT_ITEM_KEY;
2719         location.offset = 0;
2720
2721         extent_root = btrfs_read_tree_root(tree_root, &location);
2722         if (IS_ERR(extent_root)) {
2723                 ret = PTR_ERR(extent_root);
2724                 goto recovery_tree_root;
2725         }
2726         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2727         fs_info->extent_root = extent_root;
2728
2729         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2730         dev_root = btrfs_read_tree_root(tree_root, &location);
2731         if (IS_ERR(dev_root)) {
2732                 ret = PTR_ERR(dev_root);
2733                 goto recovery_tree_root;
2734         }
2735         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2736         fs_info->dev_root = dev_root;
2737         btrfs_init_devices_late(fs_info);
2738
2739         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2740         csum_root = btrfs_read_tree_root(tree_root, &location);
2741         if (IS_ERR(csum_root)) {
2742                 ret = PTR_ERR(csum_root);
2743                 goto recovery_tree_root;
2744         }
2745         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2746         fs_info->csum_root = csum_root;
2747
2748         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2749         quota_root = btrfs_read_tree_root(tree_root, &location);
2750         if (!IS_ERR(quota_root)) {
2751                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2752                 fs_info->quota_enabled = 1;
2753                 fs_info->pending_quota_state = 1;
2754                 fs_info->quota_root = quota_root;
2755         }
2756
2757         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2758         uuid_root = btrfs_read_tree_root(tree_root, &location);
2759         if (IS_ERR(uuid_root)) {
2760                 ret = PTR_ERR(uuid_root);
2761                 if (ret != -ENOENT)
2762                         goto recovery_tree_root;
2763                 create_uuid_tree = true;
2764                 check_uuid_tree = false;
2765         } else {
2766                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2767                 fs_info->uuid_root = uuid_root;
2768                 create_uuid_tree = false;
2769                 check_uuid_tree =
2770                     generation != btrfs_super_uuid_tree_generation(disk_super);
2771         }
2772
2773         fs_info->generation = generation;
2774         fs_info->last_trans_committed = generation;
2775
2776         ret = btrfs_recover_balance(fs_info);
2777         if (ret) {
2778                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2779                 goto fail_block_groups;
2780         }
2781
2782         ret = btrfs_init_dev_stats(fs_info);
2783         if (ret) {
2784                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2785                        ret);
2786                 goto fail_block_groups;
2787         }
2788
2789         ret = btrfs_init_dev_replace(fs_info);
2790         if (ret) {
2791                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2792                 goto fail_block_groups;
2793         }
2794
2795         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2796
2797         ret = btrfs_sysfs_add_one(fs_info);
2798         if (ret) {
2799                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2800                 goto fail_block_groups;
2801         }
2802
2803         ret = btrfs_init_space_info(fs_info);
2804         if (ret) {
2805                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2806                 goto fail_sysfs;
2807         }
2808
2809         ret = btrfs_read_block_groups(extent_root);
2810         if (ret) {
2811                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2812                 goto fail_sysfs;
2813         }
2814         fs_info->num_tolerated_disk_barrier_failures =
2815                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2816         if (fs_info->fs_devices->missing_devices >
2817              fs_info->num_tolerated_disk_barrier_failures &&
2818             !(sb->s_flags & MS_RDONLY)) {
2819                 printk(KERN_WARNING "BTRFS: "
2820                         "too many missing devices, writeable mount is not allowed\n");
2821                 goto fail_sysfs;
2822         }
2823
2824         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2825                                                "btrfs-cleaner");
2826         if (IS_ERR(fs_info->cleaner_kthread))
2827                 goto fail_sysfs;
2828
2829         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2830                                                    tree_root,
2831                                                    "btrfs-transaction");
2832         if (IS_ERR(fs_info->transaction_kthread))
2833                 goto fail_cleaner;
2834
2835         if (!btrfs_test_opt(tree_root, SSD) &&
2836             !btrfs_test_opt(tree_root, NOSSD) &&
2837             !fs_info->fs_devices->rotating) {
2838                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2839                        "mode\n");
2840                 btrfs_set_opt(fs_info->mount_opt, SSD);
2841         }
2842
2843         /*
2844          * Mount does not set all options immediatelly, we can do it now and do
2845          * not have to wait for transaction commit
2846          */
2847         btrfs_apply_pending_changes(fs_info);
2848
2849 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2850         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2851                 ret = btrfsic_mount(tree_root, fs_devices,
2852                                     btrfs_test_opt(tree_root,
2853                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2854                                     1 : 0,
2855                                     fs_info->check_integrity_print_mask);
2856                 if (ret)
2857                         printk(KERN_WARNING "BTRFS: failed to initialize"
2858                                " integrity check module %s\n", sb->s_id);
2859         }
2860 #endif
2861         ret = btrfs_read_qgroup_config(fs_info);
2862         if (ret)
2863                 goto fail_trans_kthread;
2864
2865         /* do not make disk changes in broken FS */
2866         if (btrfs_super_log_root(disk_super) != 0) {
2867                 u64 bytenr = btrfs_super_log_root(disk_super);
2868
2869                 if (fs_devices->rw_devices == 0) {
2870                         printk(KERN_WARNING "BTRFS: log replay required "
2871                                "on RO media\n");
2872                         err = -EIO;
2873                         goto fail_qgroup;
2874                 }
2875
2876                 log_tree_root = btrfs_alloc_root(fs_info);
2877                 if (!log_tree_root) {
2878                         err = -ENOMEM;
2879                         goto fail_qgroup;
2880                 }
2881
2882                 __setup_root(nodesize, sectorsize, stripesize,
2883                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2884
2885                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2886                                                       generation + 1);
2887                 if (!log_tree_root->node ||
2888                     !extent_buffer_uptodate(log_tree_root->node)) {
2889                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2890                         free_extent_buffer(log_tree_root->node);
2891                         kfree(log_tree_root);
2892                         goto fail_qgroup;
2893                 }
2894                 /* returns with log_tree_root freed on success */
2895                 ret = btrfs_recover_log_trees(log_tree_root);
2896                 if (ret) {
2897                         btrfs_error(tree_root->fs_info, ret,
2898                                     "Failed to recover log tree");
2899                         free_extent_buffer(log_tree_root->node);
2900                         kfree(log_tree_root);
2901                         goto fail_qgroup;
2902                 }
2903
2904                 if (sb->s_flags & MS_RDONLY) {
2905                         ret = btrfs_commit_super(tree_root);
2906                         if (ret)
2907                                 goto fail_qgroup;
2908                 }
2909         }
2910
2911         ret = btrfs_find_orphan_roots(tree_root);
2912         if (ret)
2913                 goto fail_qgroup;
2914
2915         if (!(sb->s_flags & MS_RDONLY)) {
2916                 ret = btrfs_cleanup_fs_roots(fs_info);
2917                 if (ret)
2918                         goto fail_qgroup;
2919
2920                 mutex_lock(&fs_info->cleaner_mutex);
2921                 ret = btrfs_recover_relocation(tree_root);
2922                 mutex_unlock(&fs_info->cleaner_mutex);
2923                 if (ret < 0) {
2924                         printk(KERN_WARNING
2925                                "BTRFS: failed to recover relocation\n");
2926                         err = -EINVAL;
2927                         goto fail_qgroup;
2928                 }
2929         }
2930
2931         location.objectid = BTRFS_FS_TREE_OBJECTID;
2932         location.type = BTRFS_ROOT_ITEM_KEY;
2933         location.offset = 0;
2934
2935         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2936         if (IS_ERR(fs_info->fs_root)) {
2937                 err = PTR_ERR(fs_info->fs_root);
2938                 goto fail_qgroup;
2939         }
2940
2941         if (sb->s_flags & MS_RDONLY)
2942                 return 0;
2943
2944         down_read(&fs_info->cleanup_work_sem);
2945         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2946             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2947                 up_read(&fs_info->cleanup_work_sem);
2948                 close_ctree(tree_root);
2949                 return ret;
2950         }
2951         up_read(&fs_info->cleanup_work_sem);
2952
2953         ret = btrfs_resume_balance_async(fs_info);
2954         if (ret) {
2955                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2956                 close_ctree(tree_root);
2957                 return ret;
2958         }
2959
2960         ret = btrfs_resume_dev_replace_async(fs_info);
2961         if (ret) {
2962                 pr_warn("BTRFS: failed to resume dev_replace\n");
2963                 close_ctree(tree_root);
2964                 return ret;
2965         }
2966
2967         btrfs_qgroup_rescan_resume(fs_info);
2968
2969         if (create_uuid_tree) {
2970                 pr_info("BTRFS: creating UUID tree\n");
2971                 ret = btrfs_create_uuid_tree(fs_info);
2972                 if (ret) {
2973                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2974                                 ret);
2975                         close_ctree(tree_root);
2976                         return ret;
2977                 }
2978         } else if (check_uuid_tree ||
2979                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2980                 pr_info("BTRFS: checking UUID tree\n");
2981                 ret = btrfs_check_uuid_tree(fs_info);
2982                 if (ret) {
2983                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2984                                 ret);
2985                         close_ctree(tree_root);
2986                         return ret;
2987                 }
2988         } else {
2989                 fs_info->update_uuid_tree_gen = 1;
2990         }
2991
2992         fs_info->open = 1;
2993
2994         return 0;
2995
2996 fail_qgroup:
2997         btrfs_free_qgroup_config(fs_info);
2998 fail_trans_kthread:
2999         kthread_stop(fs_info->transaction_kthread);
3000         btrfs_cleanup_transaction(fs_info->tree_root);
3001         btrfs_free_fs_roots(fs_info);
3002 fail_cleaner:
3003         kthread_stop(fs_info->cleaner_kthread);
3004
3005         /*
3006          * make sure we're done with the btree inode before we stop our
3007          * kthreads
3008          */
3009         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3010
3011 fail_sysfs:
3012         btrfs_sysfs_remove_one(fs_info);
3013
3014 fail_block_groups:
3015         btrfs_put_block_group_cache(fs_info);
3016         btrfs_free_block_groups(fs_info);
3017
3018 fail_tree_roots:
3019         free_root_pointers(fs_info, 1);
3020         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3021
3022 fail_sb_buffer:
3023         btrfs_stop_all_workers(fs_info);
3024 fail_alloc:
3025 fail_iput:
3026         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3027
3028         iput(fs_info->btree_inode);
3029 fail_bio_counter:
3030         percpu_counter_destroy(&fs_info->bio_counter);
3031 fail_delalloc_bytes:
3032         percpu_counter_destroy(&fs_info->delalloc_bytes);
3033 fail_dirty_metadata_bytes:
3034         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3035 fail_bdi:
3036         bdi_destroy(&fs_info->bdi);
3037 fail_srcu:
3038         cleanup_srcu_struct(&fs_info->subvol_srcu);
3039 fail:
3040         btrfs_free_stripe_hash_table(fs_info);
3041         btrfs_close_devices(fs_info->fs_devices);
3042         return err;
3043
3044 recovery_tree_root:
3045         if (!btrfs_test_opt(tree_root, RECOVERY))
3046                 goto fail_tree_roots;
3047
3048         free_root_pointers(fs_info, 0);
3049
3050         /* don't use the log in recovery mode, it won't be valid */
3051         btrfs_set_super_log_root(disk_super, 0);
3052
3053         /* we can't trust the free space cache either */
3054         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3055
3056         ret = next_root_backup(fs_info, fs_info->super_copy,
3057                                &num_backups_tried, &backup_index);
3058         if (ret == -1)
3059                 goto fail_block_groups;
3060         goto retry_root_backup;
3061 }
3062
3063 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3064 {
3065         if (uptodate) {
3066                 set_buffer_uptodate(bh);
3067         } else {
3068                 struct btrfs_device *device = (struct btrfs_device *)
3069                         bh->b_private;
3070
3071                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3072                                           "I/O error on %s\n",
3073                                           rcu_str_deref(device->name));
3074                 /* note, we dont' set_buffer_write_io_error because we have
3075                  * our own ways of dealing with the IO errors
3076                  */
3077                 clear_buffer_uptodate(bh);
3078                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3079         }
3080         unlock_buffer(bh);
3081         put_bh(bh);
3082 }
3083
3084 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3085 {
3086         struct buffer_head *bh;
3087         struct buffer_head *latest = NULL;
3088         struct btrfs_super_block *super;
3089         int i;
3090         u64 transid = 0;
3091         u64 bytenr;
3092
3093         /* we would like to check all the supers, but that would make
3094          * a btrfs mount succeed after a mkfs from a different FS.
3095          * So, we need to add a special mount option to scan for
3096          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3097          */
3098         for (i = 0; i < 1; i++) {
3099                 bytenr = btrfs_sb_offset(i);
3100                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3101                                         i_size_read(bdev->bd_inode))
3102                         break;
3103                 bh = __bread(bdev, bytenr / 4096,
3104                                         BTRFS_SUPER_INFO_SIZE);
3105                 if (!bh)
3106                         continue;
3107
3108                 super = (struct btrfs_super_block *)bh->b_data;
3109                 if (btrfs_super_bytenr(super) != bytenr ||
3110                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3111                         brelse(bh);
3112                         continue;
3113                 }
3114
3115                 if (!latest || btrfs_super_generation(super) > transid) {
3116                         brelse(latest);
3117                         latest = bh;
3118                         transid = btrfs_super_generation(super);
3119                 } else {
3120                         brelse(bh);
3121                 }
3122         }
3123         return latest;
3124 }
3125
3126 /*
3127  * this should be called twice, once with wait == 0 and
3128  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3129  * we write are pinned.
3130  *
3131  * They are released when wait == 1 is done.
3132  * max_mirrors must be the same for both runs, and it indicates how
3133  * many supers on this one device should be written.
3134  *
3135  * max_mirrors == 0 means to write them all.
3136  */
3137 static int write_dev_supers(struct btrfs_device *device,
3138                             struct btrfs_super_block *sb,
3139                             int do_barriers, int wait, int max_mirrors)
3140 {
3141         struct buffer_head *bh;
3142         int i;
3143         int ret;
3144         int errors = 0;
3145         u32 crc;
3146         u64 bytenr;
3147
3148         if (max_mirrors == 0)
3149                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3150
3151         for (i = 0; i < max_mirrors; i++) {
3152                 bytenr = btrfs_sb_offset(i);
3153                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3154                     device->commit_total_bytes)
3155                         break;
3156
3157                 if (wait) {
3158                         bh = __find_get_block(device->bdev, bytenr / 4096,
3159                                               BTRFS_SUPER_INFO_SIZE);
3160                         if (!bh) {
3161                                 errors++;
3162                                 continue;
3163                         }
3164                         wait_on_buffer(bh);
3165                         if (!buffer_uptodate(bh))
3166                                 errors++;
3167
3168                         /* drop our reference */
3169                         brelse(bh);
3170
3171                         /* drop the reference from the wait == 0 run */
3172                         brelse(bh);
3173                         continue;
3174                 } else {
3175                         btrfs_set_super_bytenr(sb, bytenr);
3176
3177                         crc = ~(u32)0;
3178                         crc = btrfs_csum_data((char *)sb +
3179                                               BTRFS_CSUM_SIZE, crc,
3180                                               BTRFS_SUPER_INFO_SIZE -
3181                                               BTRFS_CSUM_SIZE);
3182                         btrfs_csum_final(crc, sb->csum);
3183
3184                         /*
3185                          * one reference for us, and we leave it for the
3186                          * caller
3187                          */
3188                         bh = __getblk(device->bdev, bytenr / 4096,
3189                                       BTRFS_SUPER_INFO_SIZE);
3190                         if (!bh) {
3191                                 printk(KERN_ERR "BTRFS: couldn't get super "
3192                                        "buffer head for bytenr %Lu\n", bytenr);
3193                                 errors++;
3194                                 continue;
3195                         }
3196
3197                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3198
3199                         /* one reference for submit_bh */
3200                         get_bh(bh);
3201
3202                         set_buffer_uptodate(bh);
3203                         lock_buffer(bh);
3204                         bh->b_end_io = btrfs_end_buffer_write_sync;
3205                         bh->b_private = device;
3206                 }
3207
3208                 /*
3209                  * we fua the first super.  The others we allow
3210                  * to go down lazy.
3211                  */
3212                 if (i == 0)
3213                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3214                 else
3215                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3216                 if (ret)
3217                         errors++;
3218         }
3219         return errors < i ? 0 : -1;
3220 }
3221
3222 /*
3223  * endio for the write_dev_flush, this will wake anyone waiting
3224  * for the barrier when it is done
3225  */
3226 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3227 {
3228         if (err) {
3229                 if (err == -EOPNOTSUPP)
3230                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3231                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3232         }
3233         if (bio->bi_private)
3234                 complete(bio->bi_private);
3235         bio_put(bio);
3236 }
3237
3238 /*
3239  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3240  * sent down.  With wait == 1, it waits for the previous flush.
3241  *
3242  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3243  * capable
3244  */
3245 static int write_dev_flush(struct btrfs_device *device, int wait)
3246 {
3247         struct bio *bio;
3248         int ret = 0;
3249
3250         if (device->nobarriers)
3251                 return 0;
3252
3253         if (wait) {
3254                 bio = device->flush_bio;
3255                 if (!bio)
3256                         return 0;
3257
3258                 wait_for_completion(&device->flush_wait);
3259
3260                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3261                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3262                                       rcu_str_deref(device->name));
3263                         device->nobarriers = 1;
3264                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3265                         ret = -EIO;
3266                         btrfs_dev_stat_inc_and_print(device,
3267                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3268                 }
3269
3270                 /* drop the reference from the wait == 0 run */
3271                 bio_put(bio);
3272                 device->flush_bio = NULL;
3273
3274                 return ret;
3275         }
3276
3277         /*
3278          * one reference for us, and we leave it for the
3279          * caller
3280          */
3281         device->flush_bio = NULL;
3282         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3283         if (!bio)
3284                 return -ENOMEM;
3285
3286         bio->bi_end_io = btrfs_end_empty_barrier;
3287         bio->bi_bdev = device->bdev;
3288         init_completion(&device->flush_wait);
3289         bio->bi_private = &device->flush_wait;
3290         device->flush_bio = bio;
3291
3292         bio_get(bio);
3293         btrfsic_submit_bio(WRITE_FLUSH, bio);
3294
3295         return 0;
3296 }
3297
3298 /*
3299  * send an empty flush down to each device in parallel,
3300  * then wait for them
3301  */
3302 static int barrier_all_devices(struct btrfs_fs_info *info)
3303 {
3304         struct list_head *head;
3305         struct btrfs_device *dev;
3306         int errors_send = 0;
3307         int errors_wait = 0;
3308         int ret;
3309
3310         /* send down all the barriers */
3311         head = &info->fs_devices->devices;
3312         list_for_each_entry_rcu(dev, head, dev_list) {
3313                 if (dev->missing)
3314                         continue;
3315                 if (!dev->bdev) {
3316                         errors_send++;
3317                         continue;
3318                 }
3319                 if (!dev->in_fs_metadata || !dev->writeable)
3320                         continue;
3321
3322                 ret = write_dev_flush(dev, 0);
3323                 if (ret)
3324                         errors_send++;
3325         }
3326
3327         /* wait for all the barriers */
3328         list_for_each_entry_rcu(dev, head, dev_list) {
3329                 if (dev->missing)
3330                         continue;
3331                 if (!dev->bdev) {
3332                         errors_wait++;
3333                         continue;
3334                 }
3335                 if (!dev->in_fs_metadata || !dev->writeable)
3336                         continue;
3337
3338                 ret = write_dev_flush(dev, 1);
3339                 if (ret)
3340                         errors_wait++;
3341         }
3342         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3343             errors_wait > info->num_tolerated_disk_barrier_failures)
3344                 return -EIO;
3345         return 0;
3346 }
3347
3348 int btrfs_calc_num_tolerated_disk_barrier_failures(
3349         struct btrfs_fs_info *fs_info)
3350 {
3351         struct btrfs_ioctl_space_info space;
3352         struct btrfs_space_info *sinfo;
3353         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3354                        BTRFS_BLOCK_GROUP_SYSTEM,
3355                        BTRFS_BLOCK_GROUP_METADATA,
3356                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3357         int num_types = 4;
3358         int i;
3359         int c;
3360         int num_tolerated_disk_barrier_failures =
3361                 (int)fs_info->fs_devices->num_devices;
3362
3363         for (i = 0; i < num_types; i++) {
3364                 struct btrfs_space_info *tmp;
3365
3366                 sinfo = NULL;
3367                 rcu_read_lock();
3368                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3369                         if (tmp->flags == types[i]) {
3370                                 sinfo = tmp;
3371                                 break;
3372                         }
3373                 }
3374                 rcu_read_unlock();
3375
3376                 if (!sinfo)
3377                         continue;
3378
3379                 down_read(&sinfo->groups_sem);
3380                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3381                         if (!list_empty(&sinfo->block_groups[c])) {
3382                                 u64 flags;
3383
3384                                 btrfs_get_block_group_info(
3385                                         &sinfo->block_groups[c], &space);
3386                                 if (space.total_bytes == 0 ||
3387                                     space.used_bytes == 0)
3388                                         continue;
3389                                 flags = space.flags;
3390                                 /*
3391                                  * return
3392                                  * 0: if dup, single or RAID0 is configured for
3393                                  *    any of metadata, system or data, else
3394                                  * 1: if RAID5 is configured, or if RAID1 or
3395                                  *    RAID10 is configured and only two mirrors
3396                                  *    are used, else
3397                                  * 2: if RAID6 is configured, else
3398                                  * num_mirrors - 1: if RAID1 or RAID10 is
3399                                  *                  configured and more than
3400                                  *                  2 mirrors are used.
3401                                  */
3402                                 if (num_tolerated_disk_barrier_failures > 0 &&
3403                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3404                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3405                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3406                                       == 0)))
3407                                         num_tolerated_disk_barrier_failures = 0;
3408                                 else if (num_tolerated_disk_barrier_failures > 1) {
3409                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3410                                             BTRFS_BLOCK_GROUP_RAID5 |
3411                                             BTRFS_BLOCK_GROUP_RAID10)) {
3412                                                 num_tolerated_disk_barrier_failures = 1;
3413                                         } else if (flags &
3414                                                    BTRFS_BLOCK_GROUP_RAID6) {
3415                                                 num_tolerated_disk_barrier_failures = 2;
3416                                         }
3417                                 }
3418                         }
3419                 }
3420                 up_read(&sinfo->groups_sem);
3421         }
3422
3423         return num_tolerated_disk_barrier_failures;
3424 }
3425
3426 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3427 {
3428         struct list_head *head;
3429         struct btrfs_device *dev;
3430         struct btrfs_super_block *sb;
3431         struct btrfs_dev_item *dev_item;
3432         int ret;
3433         int do_barriers;
3434         int max_errors;
3435         int total_errors = 0;
3436         u64 flags;
3437
3438         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3439         backup_super_roots(root->fs_info);
3440
3441         sb = root->fs_info->super_for_commit;
3442         dev_item = &sb->dev_item;
3443
3444         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3445         head = &root->fs_info->fs_devices->devices;
3446         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3447
3448         if (do_barriers) {
3449                 ret = barrier_all_devices(root->fs_info);
3450                 if (ret) {
3451                         mutex_unlock(
3452                                 &root->fs_info->fs_devices->device_list_mutex);
3453                         btrfs_error(root->fs_info, ret,
3454                                     "errors while submitting device barriers.");
3455                         return ret;
3456                 }
3457         }
3458
3459         list_for_each_entry_rcu(dev, head, dev_list) {
3460                 if (!dev->bdev) {
3461                         total_errors++;
3462                         continue;
3463                 }
3464                 if (!dev->in_fs_metadata || !dev->writeable)
3465                         continue;
3466
3467                 btrfs_set_stack_device_generation(dev_item, 0);
3468                 btrfs_set_stack_device_type(dev_item, dev->type);
3469                 btrfs_set_stack_device_id(dev_item, dev->devid);
3470                 btrfs_set_stack_device_total_bytes(dev_item,
3471                                                    dev->commit_total_bytes);
3472                 btrfs_set_stack_device_bytes_used(dev_item,
3473                                                   dev->commit_bytes_used);
3474                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3475                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3476                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3477                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3478                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3479
3480                 flags = btrfs_super_flags(sb);
3481                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3482
3483                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3484                 if (ret)
3485                         total_errors++;
3486         }
3487         if (total_errors > max_errors) {
3488                 btrfs_err(root->fs_info, "%d errors while writing supers",
3489                        total_errors);
3490                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3491
3492                 /* FUA is masked off if unsupported and can't be the reason */
3493                 btrfs_error(root->fs_info, -EIO,
3494                             "%d errors while writing supers", total_errors);
3495                 return -EIO;
3496         }
3497
3498         total_errors = 0;
3499         list_for_each_entry_rcu(dev, head, dev_list) {
3500                 if (!dev->bdev)
3501                         continue;
3502                 if (!dev->in_fs_metadata || !dev->writeable)
3503                         continue;
3504
3505                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3506                 if (ret)
3507                         total_errors++;
3508         }
3509         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3510         if (total_errors > max_errors) {
3511                 btrfs_error(root->fs_info, -EIO,
3512                             "%d errors while writing supers", total_errors);
3513                 return -EIO;
3514         }
3515         return 0;
3516 }
3517
3518 int write_ctree_super(struct btrfs_trans_handle *trans,
3519                       struct btrfs_root *root, int max_mirrors)
3520 {
3521         return write_all_supers(root, max_mirrors);
3522 }
3523
3524 /* Drop a fs root from the radix tree and free it. */
3525 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3526                                   struct btrfs_root *root)
3527 {
3528         spin_lock(&fs_info->fs_roots_radix_lock);
3529         radix_tree_delete(&fs_info->fs_roots_radix,
3530                           (unsigned long)root->root_key.objectid);
3531         spin_unlock(&fs_info->fs_roots_radix_lock);
3532
3533         if (btrfs_root_refs(&root->root_item) == 0)
3534                 synchronize_srcu(&fs_info->subvol_srcu);
3535
3536         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3537                 btrfs_free_log(NULL, root);
3538
3539         if (root->free_ino_pinned)
3540                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3541         if (root->free_ino_ctl)
3542                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3543         free_fs_root(root);
3544 }
3545
3546 static void free_fs_root(struct btrfs_root *root)
3547 {
3548         iput(root->ino_cache_inode);
3549         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3550         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3551         root->orphan_block_rsv = NULL;
3552         if (root->anon_dev)
3553                 free_anon_bdev(root->anon_dev);
3554         if (root->subv_writers)
3555                 btrfs_free_subvolume_writers(root->subv_writers);
3556         free_extent_buffer(root->node);
3557         free_extent_buffer(root->commit_root);
3558         kfree(root->free_ino_ctl);
3559         kfree(root->free_ino_pinned);
3560         kfree(root->name);
3561         btrfs_put_fs_root(root);
3562 }
3563
3564 void btrfs_free_fs_root(struct btrfs_root *root)
3565 {
3566         free_fs_root(root);
3567 }
3568
3569 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3570 {
3571         u64 root_objectid = 0;
3572         struct btrfs_root *gang[8];
3573         int i = 0;
3574         int err = 0;
3575         unsigned int ret = 0;
3576         int index;
3577
3578         while (1) {
3579                 index = srcu_read_lock(&fs_info->subvol_srcu);
3580                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3581                                              (void **)gang, root_objectid,
3582                                              ARRAY_SIZE(gang));
3583                 if (!ret) {
3584                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3585                         break;
3586                 }
3587                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3588
3589                 for (i = 0; i < ret; i++) {
3590                         /* Avoid to grab roots in dead_roots */
3591                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3592                                 gang[i] = NULL;
3593                                 continue;
3594                         }
3595                         /* grab all the search result for later use */
3596                         gang[i] = btrfs_grab_fs_root(gang[i]);
3597                 }
3598                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3599
3600                 for (i = 0; i < ret; i++) {
3601                         if (!gang[i])
3602                                 continue;
3603                         root_objectid = gang[i]->root_key.objectid;
3604                         err = btrfs_orphan_cleanup(gang[i]);
3605                         if (err)
3606                                 break;
3607                         btrfs_put_fs_root(gang[i]);
3608                 }
3609                 root_objectid++;
3610         }
3611
3612         /* release the uncleaned roots due to error */
3613         for (; i < ret; i++) {
3614                 if (gang[i])
3615                         btrfs_put_fs_root(gang[i]);
3616         }
3617         return err;
3618 }
3619
3620 int btrfs_commit_super(struct btrfs_root *root)
3621 {
3622         struct btrfs_trans_handle *trans;
3623
3624         mutex_lock(&root->fs_info->cleaner_mutex);
3625         btrfs_run_delayed_iputs(root);
3626         mutex_unlock(&root->fs_info->cleaner_mutex);
3627         wake_up_process(root->fs_info->cleaner_kthread);
3628
3629         /* wait until ongoing cleanup work done */
3630         down_write(&root->fs_info->cleanup_work_sem);
3631         up_write(&root->fs_info->cleanup_work_sem);
3632
3633         trans = btrfs_join_transaction(root);
3634         if (IS_ERR(trans))
3635                 return PTR_ERR(trans);
3636         return btrfs_commit_transaction(trans, root);
3637 }
3638
3639 void close_ctree(struct btrfs_root *root)
3640 {
3641         struct btrfs_fs_info *fs_info = root->fs_info;
3642         int ret;
3643
3644         fs_info->closing = 1;
3645         smp_mb();
3646
3647         /* wait for the uuid_scan task to finish */
3648         down(&fs_info->uuid_tree_rescan_sem);
3649         /* avoid complains from lockdep et al., set sem back to initial state */
3650         up(&fs_info->uuid_tree_rescan_sem);
3651
3652         /* pause restriper - we want to resume on mount */
3653         btrfs_pause_balance(fs_info);
3654
3655         btrfs_dev_replace_suspend_for_unmount(fs_info);
3656
3657         btrfs_scrub_cancel(fs_info);
3658
3659         /* wait for any defraggers to finish */
3660         wait_event(fs_info->transaction_wait,
3661                    (atomic_read(&fs_info->defrag_running) == 0));
3662
3663         /* clear out the rbtree of defraggable inodes */
3664         btrfs_cleanup_defrag_inodes(fs_info);
3665
3666         cancel_work_sync(&fs_info->async_reclaim_work);
3667
3668         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3669                 ret = btrfs_commit_super(root);
3670                 if (ret)
3671                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3672         }
3673
3674         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3675                 btrfs_error_commit_super(root);
3676
3677         kthread_stop(fs_info->transaction_kthread);
3678         kthread_stop(fs_info->cleaner_kthread);
3679
3680         fs_info->closing = 2;
3681         smp_mb();
3682
3683         btrfs_free_qgroup_config(root->fs_info);
3684
3685         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3686                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3687                        percpu_counter_sum(&fs_info->delalloc_bytes));
3688         }
3689
3690         btrfs_sysfs_remove_one(fs_info);
3691
3692         btrfs_free_fs_roots(fs_info);
3693
3694         btrfs_put_block_group_cache(fs_info);
3695
3696         btrfs_free_block_groups(fs_info);
3697
3698         /*
3699          * we must make sure there is not any read request to
3700          * submit after we stopping all workers.
3701          */
3702         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3703         btrfs_stop_all_workers(fs_info);
3704
3705         fs_info->open = 0;
3706         free_root_pointers(fs_info, 1);
3707
3708         iput(fs_info->btree_inode);
3709
3710 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3711         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3712                 btrfsic_unmount(root, fs_info->fs_devices);
3713 #endif
3714
3715         btrfs_close_devices(fs_info->fs_devices);
3716         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3717
3718         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3719         percpu_counter_destroy(&fs_info->delalloc_bytes);
3720         percpu_counter_destroy(&fs_info->bio_counter);
3721         bdi_destroy(&fs_info->bdi);
3722         cleanup_srcu_struct(&fs_info->subvol_srcu);
3723
3724         btrfs_free_stripe_hash_table(fs_info);
3725
3726         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3727         root->orphan_block_rsv = NULL;
3728
3729         lock_chunks(root);
3730         while (!list_empty(&fs_info->pinned_chunks)) {
3731                 struct extent_map *em;
3732
3733                 em = list_first_entry(&fs_info->pinned_chunks,
3734                                       struct extent_map, list);
3735                 list_del_init(&em->list);
3736                 free_extent_map(em);
3737         }
3738         unlock_chunks(root);
3739 }
3740
3741 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3742                           int atomic)
3743 {
3744         int ret;
3745         struct inode *btree_inode = buf->pages[0]->mapping->host;
3746
3747         ret = extent_buffer_uptodate(buf);
3748         if (!ret)
3749                 return ret;
3750
3751         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3752                                     parent_transid, atomic);
3753         if (ret == -EAGAIN)
3754                 return ret;
3755         return !ret;
3756 }
3757
3758 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3759 {
3760         return set_extent_buffer_uptodate(buf);
3761 }
3762
3763 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3764 {
3765         struct btrfs_root *root;
3766         u64 transid = btrfs_header_generation(buf);
3767         int was_dirty;
3768
3769 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3770         /*
3771          * This is a fast path so only do this check if we have sanity tests
3772          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3773          * outside of the sanity tests.
3774          */
3775         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3776                 return;
3777 #endif
3778         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3779         btrfs_assert_tree_locked(buf);
3780         if (transid != root->fs_info->generation)
3781                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3782                        "found %llu running %llu\n",
3783                         buf->start, transid, root->fs_info->generation);
3784         was_dirty = set_extent_buffer_dirty(buf);
3785         if (!was_dirty)
3786                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3787                                      buf->len,
3788                                      root->fs_info->dirty_metadata_batch);
3789 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3790         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3791                 btrfs_print_leaf(root, buf);
3792                 ASSERT(0);
3793         }
3794 #endif
3795 }
3796
3797 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3798                                         int flush_delayed)
3799 {
3800         /*
3801          * looks as though older kernels can get into trouble with
3802          * this code, they end up stuck in balance_dirty_pages forever
3803          */
3804         int ret;
3805
3806         if (current->flags & PF_MEMALLOC)
3807                 return;
3808
3809         if (flush_delayed)
3810                 btrfs_balance_delayed_items(root);
3811
3812         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3813                                      BTRFS_DIRTY_METADATA_THRESH);
3814         if (ret > 0) {
3815                 balance_dirty_pages_ratelimited(
3816                                    root->fs_info->btree_inode->i_mapping);
3817         }
3818         return;
3819 }
3820
3821 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3822 {
3823         __btrfs_btree_balance_dirty(root, 1);
3824 }
3825
3826 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3827 {
3828         __btrfs_btree_balance_dirty(root, 0);
3829 }
3830
3831 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3832 {
3833         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3834         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3835 }
3836
3837 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3838                               int read_only)
3839 {
3840         struct btrfs_super_block *sb = fs_info->super_copy;
3841         int ret = 0;
3842
3843         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3844                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3845                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3846                 ret = -EINVAL;
3847         }
3848         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3849                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3850                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3851                 ret = -EINVAL;
3852         }
3853         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3854                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3855                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3856                 ret = -EINVAL;
3857         }
3858
3859         /*
3860          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3861          * items yet, they'll be verified later. Issue just a warning.
3862          */
3863         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3864                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3865                                 btrfs_super_root(sb));
3866         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3867                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3868                                 btrfs_super_chunk_root(sb));
3869         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3870                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3871                                 btrfs_super_log_root(sb));
3872
3873         /*
3874          * Check the lower bound, the alignment and other constraints are
3875          * checked later.
3876          */
3877         if (btrfs_super_nodesize(sb) < 4096) {
3878                 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3879                                 btrfs_super_nodesize(sb));
3880                 ret = -EINVAL;
3881         }
3882         if (btrfs_super_sectorsize(sb) < 4096) {
3883                 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3884                                 btrfs_super_sectorsize(sb));
3885                 ret = -EINVAL;
3886         }
3887
3888         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3889                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3890                                 fs_info->fsid, sb->dev_item.fsid);
3891                 ret = -EINVAL;
3892         }
3893
3894         /*
3895          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3896          * done later
3897          */
3898         if (btrfs_super_num_devices(sb) > (1UL << 31))
3899                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3900                                 btrfs_super_num_devices(sb));
3901         if (btrfs_super_num_devices(sb) == 0) {
3902                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3903                 ret = -EINVAL;
3904         }
3905
3906         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3907                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3908                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3909                 ret = -EINVAL;
3910         }
3911
3912         /*
3913          * Obvious sys_chunk_array corruptions, it must hold at least one key
3914          * and one chunk
3915          */
3916         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3917                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3918                                 btrfs_super_sys_array_size(sb),
3919                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3920                 ret = -EINVAL;
3921         }
3922         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3923                         + sizeof(struct btrfs_chunk)) {
3924                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
3925                                 btrfs_super_sys_array_size(sb),
3926                                 sizeof(struct btrfs_disk_key)
3927                                 + sizeof(struct btrfs_chunk));
3928                 ret = -EINVAL;
3929         }
3930
3931         /*
3932          * The generation is a global counter, we'll trust it more than the others
3933          * but it's still possible that it's the one that's wrong.
3934          */
3935         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3936                 printk(KERN_WARNING
3937                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3938                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3939         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3940             && btrfs_super_cache_generation(sb) != (u64)-1)
3941                 printk(KERN_WARNING
3942                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3943                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3944
3945         return ret;
3946 }
3947
3948 static void btrfs_error_commit_super(struct btrfs_root *root)
3949 {
3950         mutex_lock(&root->fs_info->cleaner_mutex);
3951         btrfs_run_delayed_iputs(root);
3952         mutex_unlock(&root->fs_info->cleaner_mutex);
3953
3954         down_write(&root->fs_info->cleanup_work_sem);
3955         up_write(&root->fs_info->cleanup_work_sem);
3956
3957         /* cleanup FS via transaction */
3958         btrfs_cleanup_transaction(root);
3959 }
3960
3961 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3962 {
3963         struct btrfs_ordered_extent *ordered;
3964
3965         spin_lock(&root->ordered_extent_lock);
3966         /*
3967          * This will just short circuit the ordered completion stuff which will
3968          * make sure the ordered extent gets properly cleaned up.
3969          */
3970         list_for_each_entry(ordered, &root->ordered_extents,
3971                             root_extent_list)
3972                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3973         spin_unlock(&root->ordered_extent_lock);
3974 }
3975
3976 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3977 {
3978         struct btrfs_root *root;
3979         struct list_head splice;
3980
3981         INIT_LIST_HEAD(&splice);
3982
3983         spin_lock(&fs_info->ordered_root_lock);
3984         list_splice_init(&fs_info->ordered_roots, &splice);
3985         while (!list_empty(&splice)) {
3986                 root = list_first_entry(&splice, struct btrfs_root,
3987                                         ordered_root);
3988                 list_move_tail(&root->ordered_root,
3989                                &fs_info->ordered_roots);
3990
3991                 spin_unlock(&fs_info->ordered_root_lock);
3992                 btrfs_destroy_ordered_extents(root);
3993
3994                 cond_resched();
3995                 spin_lock(&fs_info->ordered_root_lock);
3996         }
3997         spin_unlock(&fs_info->ordered_root_lock);
3998 }
3999
4000 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4001                                       struct btrfs_root *root)
4002 {
4003         struct rb_node *node;
4004         struct btrfs_delayed_ref_root *delayed_refs;
4005         struct btrfs_delayed_ref_node *ref;
4006         int ret = 0;
4007
4008         delayed_refs = &trans->delayed_refs;
4009
4010         spin_lock(&delayed_refs->lock);
4011         if (atomic_read(&delayed_refs->num_entries) == 0) {
4012                 spin_unlock(&delayed_refs->lock);
4013                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4014                 return ret;
4015         }
4016
4017         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4018                 struct btrfs_delayed_ref_head *head;
4019                 bool pin_bytes = false;
4020
4021                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4022                                 href_node);
4023                 if (!mutex_trylock(&head->mutex)) {
4024                         atomic_inc(&head->node.refs);
4025                         spin_unlock(&delayed_refs->lock);
4026
4027                         mutex_lock(&head->mutex);
4028                         mutex_unlock(&head->mutex);
4029                         btrfs_put_delayed_ref(&head->node);
4030                         spin_lock(&delayed_refs->lock);
4031                         continue;
4032                 }
4033                 spin_lock(&head->lock);
4034                 while ((node = rb_first(&head->ref_root)) != NULL) {
4035                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
4036                                        rb_node);
4037                         ref->in_tree = 0;
4038                         rb_erase(&ref->rb_node, &head->ref_root);
4039                         atomic_dec(&delayed_refs->num_entries);
4040                         btrfs_put_delayed_ref(ref);
4041                 }
4042                 if (head->must_insert_reserved)
4043                         pin_bytes = true;
4044                 btrfs_free_delayed_extent_op(head->extent_op);
4045                 delayed_refs->num_heads--;
4046                 if (head->processing == 0)
4047                         delayed_refs->num_heads_ready--;
4048                 atomic_dec(&delayed_refs->num_entries);
4049                 head->node.in_tree = 0;
4050                 rb_erase(&head->href_node, &delayed_refs->href_root);
4051                 spin_unlock(&head->lock);
4052                 spin_unlock(&delayed_refs->lock);
4053                 mutex_unlock(&head->mutex);
4054
4055                 if (pin_bytes)
4056                         btrfs_pin_extent(root, head->node.bytenr,
4057                                          head->node.num_bytes, 1);
4058                 btrfs_put_delayed_ref(&head->node);
4059                 cond_resched();
4060                 spin_lock(&delayed_refs->lock);
4061         }
4062
4063         spin_unlock(&delayed_refs->lock);
4064
4065         return ret;
4066 }
4067
4068 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4069 {
4070         struct btrfs_inode *btrfs_inode;
4071         struct list_head splice;
4072
4073         INIT_LIST_HEAD(&splice);
4074
4075         spin_lock(&root->delalloc_lock);
4076         list_splice_init(&root->delalloc_inodes, &splice);
4077
4078         while (!list_empty(&splice)) {
4079                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4080                                                delalloc_inodes);
4081
4082                 list_del_init(&btrfs_inode->delalloc_inodes);
4083                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4084                           &btrfs_inode->runtime_flags);
4085                 spin_unlock(&root->delalloc_lock);
4086
4087                 btrfs_invalidate_inodes(btrfs_inode->root);
4088
4089                 spin_lock(&root->delalloc_lock);
4090         }
4091
4092         spin_unlock(&root->delalloc_lock);
4093 }
4094
4095 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4096 {
4097         struct btrfs_root *root;
4098         struct list_head splice;
4099
4100         INIT_LIST_HEAD(&splice);
4101
4102         spin_lock(&fs_info->delalloc_root_lock);
4103         list_splice_init(&fs_info->delalloc_roots, &splice);
4104         while (!list_empty(&splice)) {
4105                 root = list_first_entry(&splice, struct btrfs_root,
4106                                          delalloc_root);
4107                 list_del_init(&root->delalloc_root);
4108                 root = btrfs_grab_fs_root(root);
4109                 BUG_ON(!root);
4110                 spin_unlock(&fs_info->delalloc_root_lock);
4111
4112                 btrfs_destroy_delalloc_inodes(root);
4113                 btrfs_put_fs_root(root);
4114
4115                 spin_lock(&fs_info->delalloc_root_lock);
4116         }
4117         spin_unlock(&fs_info->delalloc_root_lock);
4118 }
4119
4120 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4121                                         struct extent_io_tree *dirty_pages,
4122                                         int mark)
4123 {
4124         int ret;
4125         struct extent_buffer *eb;
4126         u64 start = 0;
4127         u64 end;
4128
4129         while (1) {
4130                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4131                                             mark, NULL);
4132                 if (ret)
4133                         break;
4134
4135                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4136                 while (start <= end) {
4137                         eb = btrfs_find_tree_block(root, start);
4138                         start += root->nodesize;
4139                         if (!eb)
4140                                 continue;
4141                         wait_on_extent_buffer_writeback(eb);
4142
4143                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4144                                                &eb->bflags))
4145                                 clear_extent_buffer_dirty(eb);
4146                         free_extent_buffer_stale(eb);
4147                 }
4148         }
4149
4150         return ret;
4151 }
4152
4153 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4154                                        struct extent_io_tree *pinned_extents)
4155 {
4156         struct extent_io_tree *unpin;
4157         u64 start;
4158         u64 end;
4159         int ret;
4160         bool loop = true;
4161
4162         unpin = pinned_extents;
4163 again:
4164         while (1) {
4165                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4166                                             EXTENT_DIRTY, NULL);
4167                 if (ret)
4168                         break;
4169
4170                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4171                 btrfs_error_unpin_extent_range(root, start, end);
4172                 cond_resched();
4173         }
4174
4175         if (loop) {
4176                 if (unpin == &root->fs_info->freed_extents[0])
4177                         unpin = &root->fs_info->freed_extents[1];
4178                 else
4179                         unpin = &root->fs_info->freed_extents[0];
4180                 loop = false;
4181                 goto again;
4182         }
4183
4184         return 0;
4185 }
4186
4187 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4188                                        struct btrfs_fs_info *fs_info)
4189 {
4190         struct btrfs_ordered_extent *ordered;
4191
4192         spin_lock(&fs_info->trans_lock);
4193         while (!list_empty(&cur_trans->pending_ordered)) {
4194                 ordered = list_first_entry(&cur_trans->pending_ordered,
4195                                            struct btrfs_ordered_extent,
4196                                            trans_list);
4197                 list_del_init(&ordered->trans_list);
4198                 spin_unlock(&fs_info->trans_lock);
4199
4200                 btrfs_put_ordered_extent(ordered);
4201                 spin_lock(&fs_info->trans_lock);
4202         }
4203         spin_unlock(&fs_info->trans_lock);
4204 }
4205
4206 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4207                                    struct btrfs_root *root)
4208 {
4209         btrfs_destroy_delayed_refs(cur_trans, root);
4210
4211         cur_trans->state = TRANS_STATE_COMMIT_START;
4212         wake_up(&root->fs_info->transaction_blocked_wait);
4213
4214         cur_trans->state = TRANS_STATE_UNBLOCKED;
4215         wake_up(&root->fs_info->transaction_wait);
4216
4217         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4218         btrfs_destroy_delayed_inodes(root);
4219         btrfs_assert_delayed_root_empty(root);
4220
4221         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4222                                      EXTENT_DIRTY);
4223         btrfs_destroy_pinned_extent(root,
4224                                     root->fs_info->pinned_extents);
4225
4226         cur_trans->state =TRANS_STATE_COMPLETED;
4227         wake_up(&cur_trans->commit_wait);
4228
4229         /*
4230         memset(cur_trans, 0, sizeof(*cur_trans));
4231         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4232         */
4233 }
4234
4235 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4236 {
4237         struct btrfs_transaction *t;
4238
4239         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4240
4241         spin_lock(&root->fs_info->trans_lock);
4242         while (!list_empty(&root->fs_info->trans_list)) {
4243                 t = list_first_entry(&root->fs_info->trans_list,
4244                                      struct btrfs_transaction, list);
4245                 if (t->state >= TRANS_STATE_COMMIT_START) {
4246                         atomic_inc(&t->use_count);
4247                         spin_unlock(&root->fs_info->trans_lock);
4248                         btrfs_wait_for_commit(root, t->transid);
4249                         btrfs_put_transaction(t);
4250                         spin_lock(&root->fs_info->trans_lock);
4251                         continue;
4252                 }
4253                 if (t == root->fs_info->running_transaction) {
4254                         t->state = TRANS_STATE_COMMIT_DOING;
4255                         spin_unlock(&root->fs_info->trans_lock);
4256                         /*
4257                          * We wait for 0 num_writers since we don't hold a trans
4258                          * handle open currently for this transaction.
4259                          */
4260                         wait_event(t->writer_wait,
4261                                    atomic_read(&t->num_writers) == 0);
4262                 } else {
4263                         spin_unlock(&root->fs_info->trans_lock);
4264                 }
4265                 btrfs_cleanup_one_transaction(t, root);
4266
4267                 spin_lock(&root->fs_info->trans_lock);
4268                 if (t == root->fs_info->running_transaction)
4269                         root->fs_info->running_transaction = NULL;
4270                 list_del_init(&t->list);
4271                 spin_unlock(&root->fs_info->trans_lock);
4272
4273                 btrfs_put_transaction(t);
4274                 trace_btrfs_transaction_commit(root);
4275                 spin_lock(&root->fs_info->trans_lock);
4276         }
4277         spin_unlock(&root->fs_info->trans_lock);
4278         btrfs_destroy_all_ordered_extents(root->fs_info);
4279         btrfs_destroy_delayed_inodes(root);
4280         btrfs_assert_delayed_root_empty(root);
4281         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4282         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4283         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4284
4285         return 0;
4286 }
4287
4288 static struct extent_io_ops btree_extent_io_ops = {
4289         .readpage_end_io_hook = btree_readpage_end_io_hook,
4290         .readpage_io_failed_hook = btree_io_failed_hook,
4291         .submit_bio_hook = btree_submit_bio_hook,
4292         /* note we're sharing with inode.c for the merge bio hook */
4293         .merge_bio_hook = btrfs_merge_bio_hook,
4294 };