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