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                         printk_ratelimited(KERN_WARNING
 323                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 324                                 "level %d\n",
 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         printk_ratelimited(KERN_ERR
 372             "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 373                         eb->fs_info->sb->s_id, 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                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
 633                                "%llu %llu\n",
 634                                eb->fs_info->sb->s_id, found_start, eb->start);
 635                 ret = -EIO;
 636                 goto err;
 637         }
 638         if (check_tree_block_fsid(root->fs_info, eb)) {
 639                 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
 640                                eb->fs_info->sb->s_id, eb->start);
 641                 ret = -EIO;
 642                 goto err;
 643         }
 644         found_level = btrfs_header_level(eb);
 645         if (found_level >= BTRFS_MAX_LEVEL) {
 646                 btrfs_err(root->fs_info, "bad tree block level %d",
 647                            (int)btrfs_header_level(eb));
 648                 ret = -EIO;
 649                 goto err;
 650         }
 651
 652         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 653                                        eb, found_level);
 654
 655         ret = csum_tree_block(root->fs_info, eb, 1);
 656         if (ret) {
 657                 ret = -EIO;
 658                 goto err;
 659         }
 660
 661         /*
 662          * If this is a leaf block and it is corrupt, set the corrupt bit so
 663          * that we don't try and read the other copies of this block, just
 664          * return -EIO.
 665          */
 666         if (found_level == 0 && check_leaf(root, eb)) {
 667                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 668                 ret = -EIO;
 669         }
 670
 671         if (!ret)
 672                 set_extent_buffer_uptodate(eb);
 673 err:
 674         if (reads_done &&
 675             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 676                 btree_readahead_hook(root, eb, eb->start, ret);
 677
 678         if (ret) {
 679                 /*
 680                  * our io error hook is going to dec the io pages
 681                  * again, we have to make sure it has something
 682                  * to decrement
 683                  */
 684                 atomic_inc(&eb->io_pages);
 685                 clear_extent_buffer_uptodate(eb);
 686         }
 687         free_extent_buffer(eb);
 688 out:
 689         return ret;
 690 }
 691
 692 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 693 {
 694         struct extent_buffer *eb;
 695         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 696
 697         eb = (struct extent_buffer *)page->private;
 698         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 699         eb->read_mirror = failed_mirror;
 700         atomic_dec(&eb->io_pages);
 701         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 702                 btree_readahead_hook(root, eb, eb->start, -EIO);
 703         return -EIO;    /* we fixed nothing */
 704 }
 705
 706 static void end_workqueue_bio(struct bio *bio, int err)
 707 {
 708         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 709         struct btrfs_fs_info *fs_info;
 710         struct btrfs_workqueue *wq;
 711         btrfs_work_func_t func;
 712
 713         fs_info = end_io_wq->info;
 714         end_io_wq->error = err;
 715
 716         if (bio->bi_rw & REQ_WRITE) {
 717                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 718                         wq = fs_info->endio_meta_write_workers;
 719                         func = btrfs_endio_meta_write_helper;
 720                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 721                         wq = fs_info->endio_freespace_worker;
 722                         func = btrfs_freespace_write_helper;
 723                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 724                         wq = fs_info->endio_raid56_workers;
 725                         func = btrfs_endio_raid56_helper;
 726                 } else {
 727                         wq = fs_info->endio_write_workers;
 728                         func = btrfs_endio_write_helper;
 729                 }
 730         } else {
 731                 if (unlikely(end_io_wq->metadata ==
 732                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 733                         wq = fs_info->endio_repair_workers;
 734                         func = btrfs_endio_repair_helper;
 735                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 736                         wq = fs_info->endio_raid56_workers;
 737                         func = btrfs_endio_raid56_helper;
 738                 } else if (end_io_wq->metadata) {
 739                         wq = fs_info->endio_meta_workers;
 740                         func = btrfs_endio_meta_helper;
 741                 } else {
 742                         wq = fs_info->endio_workers;
 743                         func = btrfs_endio_helper;
 744                 }
 745         }
 746
 747         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 748         btrfs_queue_work(wq, &end_io_wq->work);
 749 }
 750
 751 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 752                         enum btrfs_wq_endio_type metadata)
 753 {
 754         struct btrfs_end_io_wq *end_io_wq;
 755
 756         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 757         if (!end_io_wq)
 758                 return -ENOMEM;
 759
 760         end_io_wq->private = bio->bi_private;
 761         end_io_wq->end_io = bio->bi_end_io;
 762         end_io_wq->info = info;
 763         end_io_wq->error = 0;
 764         end_io_wq->bio = bio;
 765         end_io_wq->metadata = metadata;
 766
 767         bio->bi_private = end_io_wq;
 768         bio->bi_end_io = end_workqueue_bio;
 769         return 0;
 770 }
 771
 772 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 773 {
 774         unsigned long limit = min_t(unsigned long,
 775                                     info->thread_pool_size,
 776                                     info->fs_devices->open_devices);
 777         return 256 * limit;
 778 }
 779
 780 static void run_one_async_start(struct btrfs_work *work)
 781 {
 782         struct async_submit_bio *async;
 783         int ret;
 784
 785         async = container_of(work, struct  async_submit_bio, work);
 786         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 787                                       async->mirror_num, async->bio_flags,
 788                                       async->bio_offset);
 789         if (ret)
 790                 async->error = ret;
 791 }
 792
 793 static void run_one_async_done(struct btrfs_work *work)
 794 {
 795         struct btrfs_fs_info *fs_info;
 796         struct async_submit_bio *async;
 797         int limit;
 798
 799         async = container_of(work, struct  async_submit_bio, work);
 800         fs_info = BTRFS_I(async->inode)->root->fs_info;
 801
 802         limit = btrfs_async_submit_limit(fs_info);
 803         limit = limit * 2 / 3;
 804
 805         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 806             waitqueue_active(&fs_info->async_submit_wait))
 807                 wake_up(&fs_info->async_submit_wait);
 808
 809         /* If an error occured we just want to clean up the bio and move on */
 810         if (async->error) {
 811                 bio_endio(async->bio, async->error);
 812                 return;
 813         }
 814
 815         async->submit_bio_done(async->inode, async->rw, async->bio,
 816                                async->mirror_num, async->bio_flags,
 817                                async->bio_offset);
 818 }
 819
 820 static void run_one_async_free(struct btrfs_work *work)
 821 {
 822         struct async_submit_bio *async;
 823
 824         async = container_of(work, struct  async_submit_bio, work);
 825         kfree(async);
 826 }
 827
 828 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 829                         int rw, struct bio *bio, int mirror_num,
 830                         unsigned long bio_flags,
 831                         u64 bio_offset,
 832                         extent_submit_bio_hook_t *submit_bio_start,
 833                         extent_submit_bio_hook_t *submit_bio_done)
 834 {
 835         struct async_submit_bio *async;
 836
 837         async = kmalloc(sizeof(*async), GFP_NOFS);
 838         if (!async)
 839                 return -ENOMEM;
 840
 841         async->inode = inode;
 842         async->rw = rw;
 843         async->bio = bio;
 844         async->mirror_num = mirror_num;
 845         async->submit_bio_start = submit_bio_start;
 846         async->submit_bio_done = submit_bio_done;
 847
 848         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 849                         run_one_async_done, run_one_async_free);
 850
 851         async->bio_flags = bio_flags;
 852         async->bio_offset = bio_offset;
 853
 854         async->error = 0;
 855
 856         atomic_inc(&fs_info->nr_async_submits);
 857
 858         if (rw & REQ_SYNC)
 859                 btrfs_set_work_high_priority(&async->work);
 860
 861         btrfs_queue_work(fs_info->workers, &async->work);
 862
 863         while (atomic_read(&fs_info->async_submit_draining) &&
 864               atomic_read(&fs_info->nr_async_submits)) {
 865                 wait_event(fs_info->async_submit_wait,
 866                            (atomic_read(&fs_info->nr_async_submits) == 0));
 867         }
 868
 869         return 0;
 870 }
 871
 872 static int btree_csum_one_bio(struct bio *bio)
 873 {
 874         struct bio_vec *bvec;
 875         struct btrfs_root *root;
 876         int i, ret = 0;
 877
 878         bio_for_each_segment_all(bvec, bio, i) {
 879                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 880                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 881                 if (ret)
 882                         break;
 883         }
 884
 885         return ret;
 886 }
 887
 888 static int __btree_submit_bio_start(struct inode *inode, int rw,
 889                                     struct bio *bio, int mirror_num,
 890                                     unsigned long bio_flags,
 891                                     u64 bio_offset)
 892 {
 893         /*
 894          * when we're called for a write, we're already in the async
 895          * submission context.  Just jump into btrfs_map_bio
 896          */
 897         return btree_csum_one_bio(bio);
 898 }
 899
 900 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 901                                  int mirror_num, unsigned long bio_flags,
 902                                  u64 bio_offset)
 903 {
 904         int ret;
 905
 906         /*
 907          * when we're called for a write, we're already in the async
 908          * submission context.  Just jump into btrfs_map_bio
 909          */
 910         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 911         if (ret)
 912                 bio_endio(bio, ret);
 913         return ret;
 914 }
 915
 916 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 917 {
 918         if (bio_flags & EXTENT_BIO_TREE_LOG)
 919                 return 0;
 920 #ifdef CONFIG_X86
 921         if (cpu_has_xmm4_2)
 922                 return 0;
 923 #endif
 924         return 1;
 925 }
 926
 927 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 928                                  int mirror_num, unsigned long bio_flags,
 929                                  u64 bio_offset)
 930 {
 931         int async = check_async_write(inode, bio_flags);
 932         int ret;
 933
 934         if (!(rw & REQ_WRITE)) {
 935                 /*
 936                  * called for a read, do the setup so that checksum validation
 937                  * can happen in the async kernel threads
 938                  */
 939                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 940                                           bio, BTRFS_WQ_ENDIO_METADATA);
 941                 if (ret)
 942                         goto out_w_error;
 943                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 944                                     mirror_num, 0);
 945         } else if (!async) {
 946                 ret = btree_csum_one_bio(bio);
 947                 if (ret)
 948                         goto out_w_error;
 949                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 950                                     mirror_num, 0);
 951         } else {
 952                 /*
 953                  * kthread helpers are used to submit writes so that
 954                  * checksumming can happen in parallel across all CPUs
 955                  */
 956                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 957                                           inode, rw, bio, mirror_num, 0,
 958                                           bio_offset,
 959                                           __btree_submit_bio_start,
 960                                           __btree_submit_bio_done);
 961         }
 962
 963         if (ret) {
 964 out_w_error:
 965                 bio_endio(bio, ret);
 966         }
 967         return ret;
 968 }
 969
 970 #ifdef CONFIG_MIGRATION
 971 static int btree_migratepage(struct address_space *mapping,
 972                         struct page *newpage, struct page *page,
 973                         enum migrate_mode mode)
 974 {
 975         /*
 976          * we can't safely write a btree page from here,
 977          * we haven't done the locking hook
 978          */
 979         if (PageDirty(page))
 980                 return -EAGAIN;
 981         /*
 982          * Buffers may be managed in a filesystem specific way.
 983          * We must have no buffers or drop them.
 984          */
 985         if (page_has_private(page) &&
 986             !try_to_release_page(page, GFP_KERNEL))
 987                 return -EAGAIN;
 988         return migrate_page(mapping, newpage, page, mode);
 989 }
 990 #endif
 991
 992
 993 static int btree_writepages(struct address_space *mapping,
 994                             struct writeback_control *wbc)
 995 {
 996         struct btrfs_fs_info *fs_info;
 997         int ret;
 998
 999         if (wbc->sync_mode == WB_SYNC_NONE) {
1000
1001                 if (wbc->for_kupdate)
1002                         return 0;
1003
1004                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1005                 /* this is a bit racy, but that's ok */
1006                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1007                                              BTRFS_DIRTY_METADATA_THRESH);
1008                 if (ret < 0)
1009                         return 0;
1010         }
1011         return btree_write_cache_pages(mapping, wbc);
1012 }
1013
1014 static int btree_readpage(struct file *file, struct page *page)
1015 {
1016         struct extent_io_tree *tree;
1017         tree = &BTRFS_I(page->mapping->host)->io_tree;
1018         return extent_read_full_page(tree, page, btree_get_extent, 0);
1019 }
1020
1021 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1022 {
1023         if (PageWriteback(page) || PageDirty(page))
1024                 return 0;
1025
1026         return try_release_extent_buffer(page);
1027 }
1028
1029 static void btree_invalidatepage(struct page *page, unsigned int offset,
1030                                  unsigned int length)
1031 {
1032         struct extent_io_tree *tree;
1033         tree = &BTRFS_I(page->mapping->host)->io_tree;
1034         extent_invalidatepage(tree, page, offset);
1035         btree_releasepage(page, GFP_NOFS);
1036         if (PagePrivate(page)) {
1037                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1038                            "page private not zero on page %llu",
1039                            (unsigned long long)page_offset(page));
1040                 ClearPagePrivate(page);
1041                 set_page_private(page, 0);
1042                 page_cache_release(page);
1043         }
1044 }
1045
1046 static int btree_set_page_dirty(struct page *page)
1047 {
1048 #ifdef DEBUG
1049         struct extent_buffer *eb;
1050
1051         BUG_ON(!PagePrivate(page));
1052         eb = (struct extent_buffer *)page->private;
1053         BUG_ON(!eb);
1054         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1055         BUG_ON(!atomic_read(&eb->refs));
1056         btrfs_assert_tree_locked(eb);
1057 #endif
1058         return __set_page_dirty_nobuffers(page);
1059 }
1060
1061 static const struct address_space_operations btree_aops = {
1062         .readpage       = btree_readpage,
1063         .writepages     = btree_writepages,
1064         .releasepage    = btree_releasepage,
1065         .invalidatepage = btree_invalidatepage,
1066 #ifdef CONFIG_MIGRATION
1067         .migratepage    = btree_migratepage,
1068 #endif
1069         .set_page_dirty = btree_set_page_dirty,
1070 };
1071
1072 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1073 {
1074         struct extent_buffer *buf = NULL;
1075         struct inode *btree_inode = root->fs_info->btree_inode;
1076
1077         buf = btrfs_find_create_tree_block(root, bytenr);
1078         if (!buf)
1079                 return;
1080         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1081                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1082         free_extent_buffer(buf);
1083 }
1084
1085 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1086                          int mirror_num, struct extent_buffer **eb)
1087 {
1088         struct extent_buffer *buf = NULL;
1089         struct inode *btree_inode = root->fs_info->btree_inode;
1090         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1091         int ret;
1092
1093         buf = btrfs_find_create_tree_block(root, bytenr);
1094         if (!buf)
1095                 return 0;
1096
1097         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1098
1099         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1100                                        btree_get_extent, mirror_num);
1101         if (ret) {
1102                 free_extent_buffer(buf);
1103                 return ret;
1104         }
1105
1106         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1107                 free_extent_buffer(buf);
1108                 return -EIO;
1109         } else if (extent_buffer_uptodate(buf)) {
1110                 *eb = buf;
1111         } else {
1112                 free_extent_buffer(buf);
1113         }
1114         return 0;
1115 }
1116
1117 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1118                                             u64 bytenr)
1119 {
1120         return find_extent_buffer(fs_info, bytenr);
1121 }
1122
1123 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1124                                                  u64 bytenr)
1125 {
1126         if (btrfs_test_is_dummy_root(root))
1127                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1128         return alloc_extent_buffer(root->fs_info, bytenr);
1129 }
1130
1131
1132 int btrfs_write_tree_block(struct extent_buffer *buf)
1133 {
1134         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1135                                         buf->start + buf->len - 1);
1136 }
1137
1138 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1139 {
1140         return filemap_fdatawait_range(buf->pages[0]->mapping,
1141                                        buf->start, buf->start + buf->len - 1);
1142 }
1143
1144 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1145                                       u64 parent_transid)
1146 {
1147         struct extent_buffer *buf = NULL;
1148         int ret;
1149
1150         buf = btrfs_find_create_tree_block(root, bytenr);
1151         if (!buf)
1152                 return ERR_PTR(-ENOMEM);
1153
1154         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1155         if (ret) {
1156                 free_extent_buffer(buf);
1157                 return ERR_PTR(ret);
1158         }
1159         return buf;
1160
1161 }
1162
1163 void clean_tree_block(struct btrfs_trans_handle *trans,
1164                       struct btrfs_fs_info *fs_info,
1165                       struct extent_buffer *buf)
1166 {
1167         if (btrfs_header_generation(buf) ==
1168             fs_info->running_transaction->transid) {
1169                 btrfs_assert_tree_locked(buf);
1170
1171                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1172                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1173                                              -buf->len,
1174                                              fs_info->dirty_metadata_batch);
1175                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1176                         btrfs_set_lock_blocking(buf);
1177                         clear_extent_buffer_dirty(buf);
1178                 }
1179         }
1180 }
1181
1182 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1183 {
1184         struct btrfs_subvolume_writers *writers;
1185         int ret;
1186
1187         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1188         if (!writers)
1189                 return ERR_PTR(-ENOMEM);
1190
1191         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1192         if (ret < 0) {
1193                 kfree(writers);
1194                 return ERR_PTR(ret);
1195         }
1196
1197         init_waitqueue_head(&writers->wait);
1198         return writers;
1199 }
1200
1201 static void
1202 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1203 {
1204         percpu_counter_destroy(&writers->counter);
1205         kfree(writers);
1206 }
1207
1208 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1209                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1210                          u64 objectid)
1211 {
1212         root->node = NULL;
1213         root->commit_root = NULL;
1214         root->sectorsize = sectorsize;
1215         root->nodesize = nodesize;
1216         root->stripesize = stripesize;
1217         root->state = 0;
1218         root->orphan_cleanup_state = 0;
1219
1220         root->objectid = objectid;
1221         root->last_trans = 0;
1222         root->highest_objectid = 0;
1223         root->nr_delalloc_inodes = 0;
1224         root->nr_ordered_extents = 0;
1225         root->name = NULL;
1226         root->inode_tree = RB_ROOT;
1227         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1228         root->block_rsv = NULL;
1229         root->orphan_block_rsv = NULL;
1230
1231         INIT_LIST_HEAD(&root->dirty_list);
1232         INIT_LIST_HEAD(&root->root_list);
1233         INIT_LIST_HEAD(&root->delalloc_inodes);
1234         INIT_LIST_HEAD(&root->delalloc_root);
1235         INIT_LIST_HEAD(&root->ordered_extents);
1236         INIT_LIST_HEAD(&root->ordered_root);
1237         INIT_LIST_HEAD(&root->logged_list[0]);
1238         INIT_LIST_HEAD(&root->logged_list[1]);
1239         spin_lock_init(&root->orphan_lock);
1240         spin_lock_init(&root->inode_lock);
1241         spin_lock_init(&root->delalloc_lock);
1242         spin_lock_init(&root->ordered_extent_lock);
1243         spin_lock_init(&root->accounting_lock);
1244         spin_lock_init(&root->log_extents_lock[0]);
1245         spin_lock_init(&root->log_extents_lock[1]);
1246         mutex_init(&root->objectid_mutex);
1247         mutex_init(&root->log_mutex);
1248         mutex_init(&root->ordered_extent_mutex);
1249         mutex_init(&root->delalloc_mutex);
1250         init_waitqueue_head(&root->log_writer_wait);
1251         init_waitqueue_head(&root->log_commit_wait[0]);
1252         init_waitqueue_head(&root->log_commit_wait[1]);
1253         INIT_LIST_HEAD(&root->log_ctxs[0]);
1254         INIT_LIST_HEAD(&root->log_ctxs[1]);
1255         atomic_set(&root->log_commit[0], 0);
1256         atomic_set(&root->log_commit[1], 0);
1257         atomic_set(&root->log_writers, 0);
1258         atomic_set(&root->log_batch, 0);
1259         atomic_set(&root->orphan_inodes, 0);
1260         atomic_set(&root->refs, 1);
1261         atomic_set(&root->will_be_snapshoted, 0);
1262         root->log_transid = 0;
1263         root->log_transid_committed = -1;
1264         root->last_log_commit = 0;
1265         if (fs_info)
1266                 extent_io_tree_init(&root->dirty_log_pages,
1267                                      fs_info->btree_inode->i_mapping);
1268
1269         memset(&root->root_key, 0, sizeof(root->root_key));
1270         memset(&root->root_item, 0, sizeof(root->root_item));
1271         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1272         if (fs_info)
1273                 root->defrag_trans_start = fs_info->generation;
1274         else
1275                 root->defrag_trans_start = 0;
1276         root->root_key.objectid = objectid;
1277         root->anon_dev = 0;
1278
1279         spin_lock_init(&root->root_item_lock);
1280 }
1281
1282 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1283 {
1284         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1285         if (root)
1286                 root->fs_info = fs_info;
1287         return root;
1288 }
1289
1290 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1291 /* Should only be used by the testing infrastructure */
1292 struct btrfs_root *btrfs_alloc_dummy_root(void)
1293 {
1294         struct btrfs_root *root;
1295
1296         root = btrfs_alloc_root(NULL);
1297         if (!root)
1298                 return ERR_PTR(-ENOMEM);
1299         __setup_root(4096, 4096, 4096, root, NULL, 1);
1300         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1301         root->alloc_bytenr = 0;
1302
1303         return root;
1304 }
1305 #endif
1306
1307 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1308                                      struct btrfs_fs_info *fs_info,
1309                                      u64 objectid)
1310 {
1311         struct extent_buffer *leaf;
1312         struct btrfs_root *tree_root = fs_info->tree_root;
1313         struct btrfs_root *root;
1314         struct btrfs_key key;
1315         int ret = 0;
1316         uuid_le uuid;
1317
1318         root = btrfs_alloc_root(fs_info);
1319         if (!root)
1320                 return ERR_PTR(-ENOMEM);
1321
1322         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1323                 tree_root->stripesize, root, fs_info, objectid);
1324         root->root_key.objectid = objectid;
1325         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1326         root->root_key.offset = 0;
1327
1328         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1329         if (IS_ERR(leaf)) {
1330                 ret = PTR_ERR(leaf);
1331                 leaf = NULL;
1332                 goto fail;
1333         }
1334
1335         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1336         btrfs_set_header_bytenr(leaf, leaf->start);
1337         btrfs_set_header_generation(leaf, trans->transid);
1338         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1339         btrfs_set_header_owner(leaf, objectid);
1340         root->node = leaf;
1341
1342         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1343                             BTRFS_FSID_SIZE);
1344         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1345                             btrfs_header_chunk_tree_uuid(leaf),
1346                             BTRFS_UUID_SIZE);
1347         btrfs_mark_buffer_dirty(leaf);
1348
1349         root->commit_root = btrfs_root_node(root);
1350         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1351
1352         root->root_item.flags = 0;
1353         root->root_item.byte_limit = 0;
1354         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1355         btrfs_set_root_generation(&root->root_item, trans->transid);
1356         btrfs_set_root_level(&root->root_item, 0);
1357         btrfs_set_root_refs(&root->root_item, 1);
1358         btrfs_set_root_used(&root->root_item, leaf->len);
1359         btrfs_set_root_last_snapshot(&root->root_item, 0);
1360         btrfs_set_root_dirid(&root->root_item, 0);
1361         uuid_le_gen(&uuid);
1362         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1363         root->root_item.drop_level = 0;
1364
1365         key.objectid = objectid;
1366         key.type = BTRFS_ROOT_ITEM_KEY;
1367         key.offset = 0;
1368         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1369         if (ret)
1370                 goto fail;
1371
1372         btrfs_tree_unlock(leaf);
1373
1374         return root;
1375
1376 fail:
1377         if (leaf) {
1378                 btrfs_tree_unlock(leaf);
1379                 free_extent_buffer(root->commit_root);
1380                 free_extent_buffer(leaf);
1381         }
1382         kfree(root);
1383
1384         return ERR_PTR(ret);
1385 }
1386
1387 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1388                                          struct btrfs_fs_info *fs_info)
1389 {
1390         struct btrfs_root *root;
1391         struct btrfs_root *tree_root = fs_info->tree_root;
1392         struct extent_buffer *leaf;
1393
1394         root = btrfs_alloc_root(fs_info);
1395         if (!root)
1396                 return ERR_PTR(-ENOMEM);
1397
1398         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1399                      tree_root->stripesize, root, fs_info,
1400                      BTRFS_TREE_LOG_OBJECTID);
1401
1402         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1403         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1404         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1405
1406         /*
1407          * DON'T set REF_COWS for log trees
1408          *
1409          * log trees do not get reference counted because they go away
1410          * before a real commit is actually done.  They do store pointers
1411          * to file data extents, and those reference counts still get
1412          * updated (along with back refs to the log tree).
1413          */
1414
1415         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1416                         NULL, 0, 0, 0);
1417         if (IS_ERR(leaf)) {
1418                 kfree(root);
1419                 return ERR_CAST(leaf);
1420         }
1421
1422         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1423         btrfs_set_header_bytenr(leaf, leaf->start);
1424         btrfs_set_header_generation(leaf, trans->transid);
1425         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1426         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1427         root->node = leaf;
1428
1429         write_extent_buffer(root->node, root->fs_info->fsid,
1430                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1431         btrfs_mark_buffer_dirty(root->node);
1432         btrfs_tree_unlock(root->node);
1433         return root;
1434 }
1435
1436 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1437                              struct btrfs_fs_info *fs_info)
1438 {
1439         struct btrfs_root *log_root;
1440
1441         log_root = alloc_log_tree(trans, fs_info);
1442         if (IS_ERR(log_root))
1443                 return PTR_ERR(log_root);
1444         WARN_ON(fs_info->log_root_tree);
1445         fs_info->log_root_tree = log_root;
1446         return 0;
1447 }
1448
1449 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1450                        struct btrfs_root *root)
1451 {
1452         struct btrfs_root *log_root;
1453         struct btrfs_inode_item *inode_item;
1454
1455         log_root = alloc_log_tree(trans, root->fs_info);
1456         if (IS_ERR(log_root))
1457                 return PTR_ERR(log_root);
1458
1459         log_root->last_trans = trans->transid;
1460         log_root->root_key.offset = root->root_key.objectid;
1461
1462         inode_item = &log_root->root_item.inode;
1463         btrfs_set_stack_inode_generation(inode_item, 1);
1464         btrfs_set_stack_inode_size(inode_item, 3);
1465         btrfs_set_stack_inode_nlink(inode_item, 1);
1466         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1467         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1468
1469         btrfs_set_root_node(&log_root->root_item, log_root->node);
1470
1471         WARN_ON(root->log_root);
1472         root->log_root = log_root;
1473         root->log_transid = 0;
1474         root->log_transid_committed = -1;
1475         root->last_log_commit = 0;
1476         return 0;
1477 }
1478
1479 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1480                                                struct btrfs_key *key)
1481 {
1482         struct btrfs_root *root;
1483         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1484         struct btrfs_path *path;
1485         u64 generation;
1486         int ret;
1487
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         root = btrfs_alloc_root(fs_info);
1493         if (!root) {
1494                 ret = -ENOMEM;
1495                 goto alloc_fail;
1496         }
1497
1498         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1499                 tree_root->stripesize, root, fs_info, key->objectid);
1500
1501         ret = btrfs_find_root(tree_root, key, path,
1502                               &root->root_item, &root->root_key);
1503         if (ret) {
1504                 if (ret > 0)
1505                         ret = -ENOENT;
1506                 goto find_fail;
1507         }
1508
1509         generation = btrfs_root_generation(&root->root_item);
1510         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1511                                      generation);
1512         if (IS_ERR(root->node)) {
1513                 ret = PTR_ERR(root->node);
1514                 goto find_fail;
1515         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1516                 ret = -EIO;
1517                 free_extent_buffer(root->node);
1518                 goto find_fail;
1519         }
1520         root->commit_root = btrfs_root_node(root);
1521 out:
1522         btrfs_free_path(path);
1523         return root;
1524
1525 find_fail:
1526         kfree(root);
1527 alloc_fail:
1528         root = ERR_PTR(ret);
1529         goto out;
1530 }
1531
1532 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1533                                       struct btrfs_key *location)
1534 {
1535         struct btrfs_root *root;
1536
1537         root = btrfs_read_tree_root(tree_root, location);
1538         if (IS_ERR(root))
1539                 return root;
1540
1541         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1542                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1543                 btrfs_check_and_init_root_item(&root->root_item);
1544         }
1545
1546         return root;
1547 }
1548
1549 int btrfs_init_fs_root(struct btrfs_root *root)
1550 {
1551         int ret;
1552         struct btrfs_subvolume_writers *writers;
1553
1554         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1555         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1556                                         GFP_NOFS);
1557         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1558                 ret = -ENOMEM;
1559                 goto fail;
1560         }
1561
1562         writers = btrfs_alloc_subvolume_writers();
1563         if (IS_ERR(writers)) {
1564                 ret = PTR_ERR(writers);
1565                 goto fail;
1566         }
1567         root->subv_writers = writers;
1568
1569         btrfs_init_free_ino_ctl(root);
1570         spin_lock_init(&root->ino_cache_lock);
1571         init_waitqueue_head(&root->ino_cache_wait);
1572
1573         ret = get_anon_bdev(&root->anon_dev);
1574         if (ret)
1575                 goto free_writers;
1576         return 0;
1577
1578 free_writers:
1579         btrfs_free_subvolume_writers(root->subv_writers);
1580 fail:
1581         kfree(root->free_ino_ctl);
1582         kfree(root->free_ino_pinned);
1583         return ret;
1584 }
1585
1586 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1587                                                u64 root_id)
1588 {
1589         struct btrfs_root *root;
1590
1591         spin_lock(&fs_info->fs_roots_radix_lock);
1592         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1593                                  (unsigned long)root_id);
1594         spin_unlock(&fs_info->fs_roots_radix_lock);
1595         return root;
1596 }
1597
1598 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1599                          struct btrfs_root *root)
1600 {
1601         int ret;
1602
1603         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1604         if (ret)
1605                 return ret;
1606
1607         spin_lock(&fs_info->fs_roots_radix_lock);
1608         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1609                                 (unsigned long)root->root_key.objectid,
1610                                 root);
1611         if (ret == 0)
1612                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1613         spin_unlock(&fs_info->fs_roots_radix_lock);
1614         radix_tree_preload_end();
1615
1616         return ret;
1617 }
1618
1619 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1620                                      struct btrfs_key *location,
1621                                      bool check_ref)
1622 {
1623         struct btrfs_root *root;
1624         struct btrfs_path *path;
1625         struct btrfs_key key;
1626         int ret;
1627
1628         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1629                 return fs_info->tree_root;
1630         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1631                 return fs_info->extent_root;
1632         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1633                 return fs_info->chunk_root;
1634         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1635                 return fs_info->dev_root;
1636         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1637                 return fs_info->csum_root;
1638         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1639                 return fs_info->quota_root ? fs_info->quota_root :
1640                                              ERR_PTR(-ENOENT);
1641         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1642                 return fs_info->uuid_root ? fs_info->uuid_root :
1643                                             ERR_PTR(-ENOENT);
1644 again:
1645         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1646         if (root) {
1647                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1648                         return ERR_PTR(-ENOENT);
1649                 return root;
1650         }
1651
1652         root = btrfs_read_fs_root(fs_info->tree_root, location);
1653         if (IS_ERR(root))
1654                 return root;
1655
1656         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1657                 ret = -ENOENT;
1658                 goto fail;
1659         }
1660
1661         ret = btrfs_init_fs_root(root);
1662         if (ret)
1663                 goto fail;
1664
1665         path = btrfs_alloc_path();
1666         if (!path) {
1667                 ret = -ENOMEM;
1668                 goto fail;
1669         }
1670         key.objectid = BTRFS_ORPHAN_OBJECTID;
1671         key.type = BTRFS_ORPHAN_ITEM_KEY;
1672         key.offset = location->objectid;
1673
1674         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1675         btrfs_free_path(path);
1676         if (ret < 0)
1677                 goto fail;
1678         if (ret == 0)
1679                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1680
1681         ret = btrfs_insert_fs_root(fs_info, root);
1682         if (ret) {
1683                 if (ret == -EEXIST) {
1684                         free_fs_root(root);
1685                         goto again;
1686                 }
1687                 goto fail;
1688         }
1689         return root;
1690 fail:
1691         free_fs_root(root);
1692         return ERR_PTR(ret);
1693 }
1694
1695 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1696 {
1697         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1698         int ret = 0;
1699         struct btrfs_device *device;
1700         struct backing_dev_info *bdi;
1701
1702         rcu_read_lock();
1703         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1704                 if (!device->bdev)
1705                         continue;
1706                 bdi = blk_get_backing_dev_info(device->bdev);
1707                 if (bdi_congested(bdi, bdi_bits)) {
1708                         ret = 1;
1709                         break;
1710                 }
1711         }
1712         rcu_read_unlock();
1713         return ret;
1714 }
1715
1716 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1717 {
1718         int err;
1719
1720         err = bdi_setup_and_register(bdi, "btrfs");
1721         if (err)
1722                 return err;
1723
1724         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1725         bdi->congested_fn       = btrfs_congested_fn;
1726         bdi->congested_data     = info;
1727         return 0;
1728 }
1729
1730 /*
1731  * called by the kthread helper functions to finally call the bio end_io
1732  * functions.  This is where read checksum verification actually happens
1733  */
1734 static void end_workqueue_fn(struct btrfs_work *work)
1735 {
1736         struct bio *bio;
1737         struct btrfs_end_io_wq *end_io_wq;
1738         int error;
1739
1740         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1741         bio = end_io_wq->bio;
1742
1743         error = end_io_wq->error;
1744         bio->bi_private = end_io_wq->private;
1745         bio->bi_end_io = end_io_wq->end_io;
1746         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1747         bio_endio(bio, error);
1748 }
1749
1750 static int cleaner_kthread(void *arg)
1751 {
1752         struct btrfs_root *root = arg;
1753         int again;
1754         struct btrfs_trans_handle *trans;
1755
1756         do {
1757                 again = 0;
1758
1759                 /* Make the cleaner go to sleep early. */
1760                 if (btrfs_need_cleaner_sleep(root))
1761                         goto sleep;
1762
1763                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1764                         goto sleep;
1765
1766                 /*
1767                  * Avoid the problem that we change the status of the fs
1768                  * during the above check and trylock.
1769                  */
1770                 if (btrfs_need_cleaner_sleep(root)) {
1771                         mutex_unlock(&root->fs_info->cleaner_mutex);
1772                         goto sleep;
1773                 }
1774
1775                 btrfs_run_delayed_iputs(root);
1776                 again = btrfs_clean_one_deleted_snapshot(root);
1777                 mutex_unlock(&root->fs_info->cleaner_mutex);
1778
1779                 /*
1780                  * The defragger has dealt with the R/O remount and umount,
1781                  * needn't do anything special here.
1782                  */
1783                 btrfs_run_defrag_inodes(root->fs_info);
1784
1785                 /*
1786                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1787                  * with relocation (btrfs_relocate_chunk) and relocation
1788                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1789                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1790                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1791                  * unused block groups.
1792                  */
1793                 btrfs_delete_unused_bgs(root->fs_info);
1794 sleep:
1795                 if (!try_to_freeze() && !again) {
1796                         set_current_state(TASK_INTERRUPTIBLE);
1797                         if (!kthread_should_stop())
1798                                 schedule();
1799                         __set_current_state(TASK_RUNNING);
1800                 }
1801         } while (!kthread_should_stop());
1802
1803         /*
1804          * Transaction kthread is stopped before us and wakes us up.
1805          * However we might have started a new transaction and COWed some
1806          * tree blocks when deleting unused block groups for example. So
1807          * make sure we commit the transaction we started to have a clean
1808          * shutdown when evicting the btree inode - if it has dirty pages
1809          * when we do the final iput() on it, eviction will trigger a
1810          * writeback for it which will fail with null pointer dereferences
1811          * since work queues and other resources were already released and
1812          * destroyed by the time the iput/eviction/writeback is made.
1813          */
1814         trans = btrfs_attach_transaction(root);
1815         if (IS_ERR(trans)) {
1816                 if (PTR_ERR(trans) != -ENOENT)
1817                         btrfs_err(root->fs_info,
1818                                   "cleaner transaction attach returned %ld",
1819                                   PTR_ERR(trans));
1820         } else {
1821                 int ret;
1822
1823                 ret = btrfs_commit_transaction(trans, root);
1824                 if (ret)
1825                         btrfs_err(root->fs_info,
1826                                   "cleaner open transaction commit returned %d",
1827                                   ret);
1828         }
1829
1830         return 0;
1831 }
1832
1833 static int transaction_kthread(void *arg)
1834 {
1835         struct btrfs_root *root = arg;
1836         struct btrfs_trans_handle *trans;
1837         struct btrfs_transaction *cur;
1838         u64 transid;
1839         unsigned long now;
1840         unsigned long delay;
1841         bool cannot_commit;
1842
1843         do {
1844                 cannot_commit = false;
1845                 delay = HZ * root->fs_info->commit_interval;
1846                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1847
1848                 spin_lock(&root->fs_info->trans_lock);
1849                 cur = root->fs_info->running_transaction;
1850                 if (!cur) {
1851                         spin_unlock(&root->fs_info->trans_lock);
1852                         goto sleep;
1853                 }
1854
1855                 now = get_seconds();
1856                 if (cur->state < TRANS_STATE_BLOCKED &&
1857                     (now < cur->start_time ||
1858                      now - cur->start_time < root->fs_info->commit_interval)) {
1859                         spin_unlock(&root->fs_info->trans_lock);
1860                         delay = HZ * 5;
1861                         goto sleep;
1862                 }
1863                 transid = cur->transid;
1864                 spin_unlock(&root->fs_info->trans_lock);
1865
1866                 /* If the file system is aborted, this will always fail. */
1867                 trans = btrfs_attach_transaction(root);
1868                 if (IS_ERR(trans)) {
1869                         if (PTR_ERR(trans) != -ENOENT)
1870                                 cannot_commit = true;
1871                         goto sleep;
1872                 }
1873                 if (transid == trans->transid) {
1874                         btrfs_commit_transaction(trans, root);
1875                 } else {
1876                         btrfs_end_transaction(trans, root);
1877                 }
1878 sleep:
1879                 wake_up_process(root->fs_info->cleaner_kthread);
1880                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1881
1882                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1883                                       &root->fs_info->fs_state)))
1884                         btrfs_cleanup_transaction(root);
1885                 if (!try_to_freeze()) {
1886                         set_current_state(TASK_INTERRUPTIBLE);
1887                         if (!kthread_should_stop() &&
1888                             (!btrfs_transaction_blocked(root->fs_info) ||
1889                              cannot_commit))
1890                                 schedule_timeout(delay);
1891                         __set_current_state(TASK_RUNNING);
1892                 }
1893         } while (!kthread_should_stop());
1894         return 0;
1895 }
1896
1897 /*
1898  * this will find the highest generation in the array of
1899  * root backups.  The index of the highest array is returned,
1900  * or -1 if we can't find anything.
1901  *
1902  * We check to make sure the array is valid by comparing the
1903  * generation of the latest  root in the array with the generation
1904  * in the super block.  If they don't match we pitch it.
1905  */
1906 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1907 {
1908         u64 cur;
1909         int newest_index = -1;
1910         struct btrfs_root_backup *root_backup;
1911         int i;
1912
1913         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1914                 root_backup = info->super_copy->super_roots + i;
1915                 cur = btrfs_backup_tree_root_gen(root_backup);
1916                 if (cur == newest_gen)
1917                         newest_index = i;
1918         }
1919
1920         /* check to see if we actually wrapped around */
1921         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1922                 root_backup = info->super_copy->super_roots;
1923                 cur = btrfs_backup_tree_root_gen(root_backup);
1924                 if (cur == newest_gen)
1925                         newest_index = 0;
1926         }
1927         return newest_index;
1928 }
1929
1930
1931 /*
1932  * find the oldest backup so we know where to store new entries
1933  * in the backup array.  This will set the backup_root_index
1934  * field in the fs_info struct
1935  */
1936 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1937                                      u64 newest_gen)
1938 {
1939         int newest_index = -1;
1940
1941         newest_index = find_newest_super_backup(info, newest_gen);
1942         /* if there was garbage in there, just move along */
1943         if (newest_index == -1) {
1944                 info->backup_root_index = 0;
1945         } else {
1946                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1947         }
1948 }
1949
1950 /*
1951  * copy all the root pointers into the super backup array.
1952  * this will bump the backup pointer by one when it is
1953  * done
1954  */
1955 static void backup_super_roots(struct btrfs_fs_info *info)
1956 {
1957         int next_backup;
1958         struct btrfs_root_backup *root_backup;
1959         int last_backup;
1960
1961         next_backup = info->backup_root_index;
1962         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1963                 BTRFS_NUM_BACKUP_ROOTS;
1964
1965         /*
1966          * just overwrite the last backup if we're at the same generation
1967          * this happens only at umount
1968          */
1969         root_backup = info->super_for_commit->super_roots + last_backup;
1970         if (btrfs_backup_tree_root_gen(root_backup) ==
1971             btrfs_header_generation(info->tree_root->node))
1972                 next_backup = last_backup;
1973
1974         root_backup = info->super_for_commit->super_roots + next_backup;
1975
1976         /*
1977          * make sure all of our padding and empty slots get zero filled
1978          * regardless of which ones we use today
1979          */
1980         memset(root_backup, 0, sizeof(*root_backup));
1981
1982         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1983
1984         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1985         btrfs_set_backup_tree_root_gen(root_backup,
1986                                btrfs_header_generation(info->tree_root->node));
1987
1988         btrfs_set_backup_tree_root_level(root_backup,
1989                                btrfs_header_level(info->tree_root->node));
1990
1991         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1992         btrfs_set_backup_chunk_root_gen(root_backup,
1993                                btrfs_header_generation(info->chunk_root->node));
1994         btrfs_set_backup_chunk_root_level(root_backup,
1995                                btrfs_header_level(info->chunk_root->node));
1996
1997         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1998         btrfs_set_backup_extent_root_gen(root_backup,
1999                                btrfs_header_generation(info->extent_root->node));
2000         btrfs_set_backup_extent_root_level(root_backup,
2001                                btrfs_header_level(info->extent_root->node));
2002
2003         /*
2004          * we might commit during log recovery, which happens before we set
2005          * the fs_root.  Make sure it is valid before we fill it in.
2006          */
2007         if (info->fs_root && info->fs_root->node) {
2008                 btrfs_set_backup_fs_root(root_backup,
2009                                          info->fs_root->node->start);
2010                 btrfs_set_backup_fs_root_gen(root_backup,
2011                                btrfs_header_generation(info->fs_root->node));
2012                 btrfs_set_backup_fs_root_level(root_backup,
2013                                btrfs_header_level(info->fs_root->node));
2014         }
2015
2016         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2017         btrfs_set_backup_dev_root_gen(root_backup,
2018                                btrfs_header_generation(info->dev_root->node));
2019         btrfs_set_backup_dev_root_level(root_backup,
2020                                        btrfs_header_level(info->dev_root->node));
2021
2022         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2023         btrfs_set_backup_csum_root_gen(root_backup,
2024                                btrfs_header_generation(info->csum_root->node));
2025         btrfs_set_backup_csum_root_level(root_backup,
2026                                btrfs_header_level(info->csum_root->node));
2027
2028         btrfs_set_backup_total_bytes(root_backup,
2029                              btrfs_super_total_bytes(info->super_copy));
2030         btrfs_set_backup_bytes_used(root_backup,
2031                              btrfs_super_bytes_used(info->super_copy));
2032         btrfs_set_backup_num_devices(root_backup,
2033                              btrfs_super_num_devices(info->super_copy));
2034
2035         /*
2036          * if we don't copy this out to the super_copy, it won't get remembered
2037          * for the next commit
2038          */
2039         memcpy(&info->super_copy->super_roots,
2040                &info->super_for_commit->super_roots,
2041                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2042 }
2043
2044 /*
2045  * this copies info out of the root backup array and back into
2046  * the in-memory super block.  It is meant to help iterate through
2047  * the array, so you send it the number of backups you've already
2048  * tried and the last backup index you used.
2049  *
2050  * this returns -1 when it has tried all the backups
2051  */
2052 static noinline int next_root_backup(struct btrfs_fs_info *info,
2053                                      struct btrfs_super_block *super,
2054                                      int *num_backups_tried, int *backup_index)
2055 {
2056         struct btrfs_root_backup *root_backup;
2057         int newest = *backup_index;
2058
2059         if (*num_backups_tried == 0) {
2060                 u64 gen = btrfs_super_generation(super);
2061
2062                 newest = find_newest_super_backup(info, gen);
2063                 if (newest == -1)
2064                         return -1;
2065
2066                 *backup_index = newest;
2067                 *num_backups_tried = 1;
2068         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2069                 /* we've tried all the backups, all done */
2070                 return -1;
2071         } else {
2072                 /* jump to the next oldest backup */
2073                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2074                         BTRFS_NUM_BACKUP_ROOTS;
2075                 *backup_index = newest;
2076                 *num_backups_tried += 1;
2077         }
2078         root_backup = super->super_roots + newest;
2079
2080         btrfs_set_super_generation(super,
2081                                    btrfs_backup_tree_root_gen(root_backup));
2082         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2083         btrfs_set_super_root_level(super,
2084                                    btrfs_backup_tree_root_level(root_backup));
2085         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2086
2087         /*
2088          * fixme: the total bytes and num_devices need to match or we should
2089          * need a fsck
2090          */
2091         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2092         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2093         return 0;
2094 }
2095
2096 /* helper to cleanup workers */
2097 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2098 {
2099         btrfs_destroy_workqueue(fs_info->fixup_workers);
2100         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2101         btrfs_destroy_workqueue(fs_info->workers);
2102         btrfs_destroy_workqueue(fs_info->endio_workers);
2103         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2104         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2105         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2106         btrfs_destroy_workqueue(fs_info->rmw_workers);
2107         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2108         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2109         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2110         btrfs_destroy_workqueue(fs_info->submit_workers);
2111         btrfs_destroy_workqueue(fs_info->delayed_workers);
2112         btrfs_destroy_workqueue(fs_info->caching_workers);
2113         btrfs_destroy_workqueue(fs_info->readahead_workers);
2114         btrfs_destroy_workqueue(fs_info->flush_workers);
2115         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2116         btrfs_destroy_workqueue(fs_info->extent_workers);
2117 }
2118
2119 static void free_root_extent_buffers(struct btrfs_root *root)
2120 {
2121         if (root) {
2122                 free_extent_buffer(root->node);
2123                 free_extent_buffer(root->commit_root);
2124                 root->node = NULL;
2125                 root->commit_root = NULL;
2126         }
2127 }
2128
2129 /* helper to cleanup tree roots */
2130 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2131 {
2132         free_root_extent_buffers(info->tree_root);
2133
2134         free_root_extent_buffers(info->dev_root);
2135         free_root_extent_buffers(info->extent_root);
2136         free_root_extent_buffers(info->csum_root);
2137         free_root_extent_buffers(info->quota_root);
2138         free_root_extent_buffers(info->uuid_root);
2139         if (chunk_root)
2140                 free_root_extent_buffers(info->chunk_root);
2141 }
2142
2143 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2144 {
2145         int ret;
2146         struct btrfs_root *gang[8];
2147         int i;
2148
2149         while (!list_empty(&fs_info->dead_roots)) {
2150                 gang[0] = list_entry(fs_info->dead_roots.next,
2151                                      struct btrfs_root, root_list);
2152                 list_del(&gang[0]->root_list);
2153
2154                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2155                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2156                 } else {
2157                         free_extent_buffer(gang[0]->node);
2158                         free_extent_buffer(gang[0]->commit_root);
2159                         btrfs_put_fs_root(gang[0]);
2160                 }
2161         }
2162
2163         while (1) {
2164                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2165                                              (void **)gang, 0,
2166                                              ARRAY_SIZE(gang));
2167                 if (!ret)
2168                         break;
2169                 for (i = 0; i < ret; i++)
2170                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2171         }
2172
2173         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2174                 btrfs_free_log_root_tree(NULL, fs_info);
2175                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2176                                             fs_info->pinned_extents);
2177         }
2178 }
2179
2180 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2181 {
2182         mutex_init(&fs_info->scrub_lock);
2183         atomic_set(&fs_info->scrubs_running, 0);
2184         atomic_set(&fs_info->scrub_pause_req, 0);
2185         atomic_set(&fs_info->scrubs_paused, 0);
2186         atomic_set(&fs_info->scrub_cancel_req, 0);
2187         init_waitqueue_head(&fs_info->scrub_pause_wait);
2188         fs_info->scrub_workers_refcnt = 0;
2189 }
2190
2191 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2192 {
2193         spin_lock_init(&fs_info->balance_lock);
2194         mutex_init(&fs_info->balance_mutex);
2195         atomic_set(&fs_info->balance_running, 0);
2196         atomic_set(&fs_info->balance_pause_req, 0);
2197         atomic_set(&fs_info->balance_cancel_req, 0);
2198         fs_info->balance_ctl = NULL;
2199         init_waitqueue_head(&fs_info->balance_wait_q);
2200 }
2201
2202 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2203                                    struct btrfs_root *tree_root)
2204 {
2205         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2206         set_nlink(fs_info->btree_inode, 1);
2207         /*
2208          * we set the i_size on the btree inode to the max possible int.
2209          * the real end of the address space is determined by all of
2210          * the devices in the system
2211          */
2212         fs_info->btree_inode->i_size = OFFSET_MAX;
2213         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2214
2215         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2216         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2217                              fs_info->btree_inode->i_mapping);
2218         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2219         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2220
2221         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2222
2223         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2224         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2225                sizeof(struct btrfs_key));
2226         set_bit(BTRFS_INODE_DUMMY,
2227                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2228         btrfs_insert_inode_hash(fs_info->btree_inode);
2229 }
2230
2231 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2232 {
2233         fs_info->dev_replace.lock_owner = 0;
2234         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2235         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2236         mutex_init(&fs_info->dev_replace.lock_management_lock);
2237         mutex_init(&fs_info->dev_replace.lock);
2238         init_waitqueue_head(&fs_info->replace_wait);
2239 }
2240
2241 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2242 {
2243         spin_lock_init(&fs_info->qgroup_lock);
2244         mutex_init(&fs_info->qgroup_ioctl_lock);
2245         fs_info->qgroup_tree = RB_ROOT;
2246         fs_info->qgroup_op_tree = RB_ROOT;
2247         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2248         fs_info->qgroup_seq = 1;
2249         fs_info->quota_enabled = 0;
2250         fs_info->pending_quota_state = 0;
2251         fs_info->qgroup_ulist = NULL;
2252         mutex_init(&fs_info->qgroup_rescan_lock);
2253 }
2254
2255 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2256                 struct btrfs_fs_devices *fs_devices)
2257 {
2258         int max_active = fs_info->thread_pool_size;
2259         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2260
2261         fs_info->workers =
2262                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2263                                       max_active, 16);
2264
2265         fs_info->delalloc_workers =
2266                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2267
2268         fs_info->flush_workers =
2269                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2270
2271         fs_info->caching_workers =
2272                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2273
2274         /*
2275          * a higher idle thresh on the submit workers makes it much more
2276          * likely that bios will be send down in a sane order to the
2277          * devices
2278          */
2279         fs_info->submit_workers =
2280                 btrfs_alloc_workqueue("submit", flags,
2281                                       min_t(u64, fs_devices->num_devices,
2282                                             max_active), 64);
2283
2284         fs_info->fixup_workers =
2285                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2286
2287         /*
2288          * endios are largely parallel and should have a very
2289          * low idle thresh
2290          */
2291         fs_info->endio_workers =
2292                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2293         fs_info->endio_meta_workers =
2294                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2295         fs_info->endio_meta_write_workers =
2296                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2297         fs_info->endio_raid56_workers =
2298                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2299         fs_info->endio_repair_workers =
2300                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2301         fs_info->rmw_workers =
2302                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2303         fs_info->endio_write_workers =
2304                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2305         fs_info->endio_freespace_worker =
2306                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2307         fs_info->delayed_workers =
2308                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2309         fs_info->readahead_workers =
2310                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2311         fs_info->qgroup_rescan_workers =
2312                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2313         fs_info->extent_workers =
2314                 btrfs_alloc_workqueue("extent-refs", flags,
2315                                       min_t(u64, fs_devices->num_devices,
2316                                             max_active), 8);
2317
2318         if (!(fs_info->workers && fs_info->delalloc_workers &&
2319               fs_info->submit_workers && fs_info->flush_workers &&
2320               fs_info->endio_workers && fs_info->endio_meta_workers &&
2321               fs_info->endio_meta_write_workers &&
2322               fs_info->endio_repair_workers &&
2323               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2324               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2325               fs_info->caching_workers && fs_info->readahead_workers &&
2326               fs_info->fixup_workers && fs_info->delayed_workers &&
2327               fs_info->extent_workers &&
2328               fs_info->qgroup_rescan_workers)) {
2329                 return -ENOMEM;
2330         }
2331
2332         return 0;
2333 }
2334
2335 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2336                             struct btrfs_fs_devices *fs_devices)
2337 {
2338         int ret;
2339         struct btrfs_root *tree_root = fs_info->tree_root;
2340         struct btrfs_root *log_tree_root;
2341         struct btrfs_super_block *disk_super = fs_info->super_copy;
2342         u64 bytenr = btrfs_super_log_root(disk_super);
2343
2344         if (fs_devices->rw_devices == 0) {
2345                 printk(KERN_WARNING "BTRFS: log replay required "
2346                        "on RO media\n");
2347                 return -EIO;
2348         }
2349
2350         log_tree_root = btrfs_alloc_root(fs_info);
2351         if (!log_tree_root)
2352                 return -ENOMEM;
2353
2354         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2355                         tree_root->stripesize, log_tree_root, fs_info,
2356                         BTRFS_TREE_LOG_OBJECTID);
2357
2358         log_tree_root->node = read_tree_block(tree_root, bytenr,
2359                         fs_info->generation + 1);
2360         if (IS_ERR(log_tree_root->node)) {
2361                 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2362                 ret = PTR_ERR(log_tree_root->node);
2363                 kfree(log_tree_root);
2364                 return ret;
2365         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2366                 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2367                 free_extent_buffer(log_tree_root->node);
2368                 kfree(log_tree_root);
2369                 return -EIO;
2370         }
2371         /* returns with log_tree_root freed on success */
2372         ret = btrfs_recover_log_trees(log_tree_root);
2373         if (ret) {
2374                 btrfs_error(tree_root->fs_info, ret,
2375                             "Failed to recover log tree");
2376                 free_extent_buffer(log_tree_root->node);
2377                 kfree(log_tree_root);
2378                 return ret;
2379         }
2380
2381         if (fs_info->sb->s_flags & MS_RDONLY) {
2382                 ret = btrfs_commit_super(tree_root);
2383                 if (ret)
2384                         return ret;
2385         }
2386
2387         return 0;
2388 }
2389
2390 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2391                             struct btrfs_root *tree_root)
2392 {
2393         struct btrfs_root *root;
2394         struct btrfs_key location;
2395         int ret;
2396
2397         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2398         location.type = BTRFS_ROOT_ITEM_KEY;
2399         location.offset = 0;
2400
2401         root = btrfs_read_tree_root(tree_root, &location);
2402         if (IS_ERR(root))
2403                 return PTR_ERR(root);
2404         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2405         fs_info->extent_root = root;
2406
2407         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2408         root = btrfs_read_tree_root(tree_root, &location);
2409         if (IS_ERR(root))
2410                 return PTR_ERR(root);
2411         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2412         fs_info->dev_root = root;
2413         btrfs_init_devices_late(fs_info);
2414
2415         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2416         root = btrfs_read_tree_root(tree_root, &location);
2417         if (IS_ERR(root))
2418                 return PTR_ERR(root);
2419         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2420         fs_info->csum_root = root;
2421
2422         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2423         root = btrfs_read_tree_root(tree_root, &location);
2424         if (!IS_ERR(root)) {
2425                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2426                 fs_info->quota_enabled = 1;
2427                 fs_info->pending_quota_state = 1;
2428                 fs_info->quota_root = root;
2429         }
2430
2431         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2432         root = btrfs_read_tree_root(tree_root, &location);
2433         if (IS_ERR(root)) {
2434                 ret = PTR_ERR(root);
2435                 if (ret != -ENOENT)
2436                         return ret;
2437         } else {
2438                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2439                 fs_info->uuid_root = root;
2440         }
2441
2442         return 0;
2443 }
2444
2445 int open_ctree(struct super_block *sb,
2446                struct btrfs_fs_devices *fs_devices,
2447                char *options)
2448 {
2449         u32 sectorsize;
2450         u32 nodesize;
2451         u32 stripesize;
2452         u64 generation;
2453         u64 features;
2454         struct btrfs_key location;
2455         struct buffer_head *bh;
2456         struct btrfs_super_block *disk_super;
2457         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2458         struct btrfs_root *tree_root;
2459         struct btrfs_root *chunk_root;
2460         int ret;
2461         int err = -EINVAL;
2462         int num_backups_tried = 0;
2463         int backup_index = 0;
2464         int max_active;
2465
2466         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2467         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2468         if (!tree_root || !chunk_root) {
2469                 err = -ENOMEM;
2470                 goto fail;
2471         }
2472
2473         ret = init_srcu_struct(&fs_info->subvol_srcu);
2474         if (ret) {
2475                 err = ret;
2476                 goto fail;
2477         }
2478
2479         ret = setup_bdi(fs_info, &fs_info->bdi);
2480         if (ret) {
2481                 err = ret;
2482                 goto fail_srcu;
2483         }
2484
2485         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2486         if (ret) {
2487                 err = ret;
2488                 goto fail_bdi;
2489         }
2490         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2491                                         (1 + ilog2(nr_cpu_ids));
2492
2493         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2494         if (ret) {
2495                 err = ret;
2496                 goto fail_dirty_metadata_bytes;
2497         }
2498
2499         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2500         if (ret) {
2501                 err = ret;
2502                 goto fail_delalloc_bytes;
2503         }
2504
2505         fs_info->btree_inode = new_inode(sb);
2506         if (!fs_info->btree_inode) {
2507                 err = -ENOMEM;
2508                 goto fail_bio_counter;
2509         }
2510
2511         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2512
2513         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2514         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2515         INIT_LIST_HEAD(&fs_info->trans_list);
2516         INIT_LIST_HEAD(&fs_info->dead_roots);
2517         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2518         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2519         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2520         spin_lock_init(&fs_info->delalloc_root_lock);
2521         spin_lock_init(&fs_info->trans_lock);
2522         spin_lock_init(&fs_info->fs_roots_radix_lock);
2523         spin_lock_init(&fs_info->delayed_iput_lock);
2524         spin_lock_init(&fs_info->defrag_inodes_lock);
2525         spin_lock_init(&fs_info->free_chunk_lock);
2526         spin_lock_init(&fs_info->tree_mod_seq_lock);
2527         spin_lock_init(&fs_info->super_lock);
2528         spin_lock_init(&fs_info->qgroup_op_lock);
2529         spin_lock_init(&fs_info->buffer_lock);
2530         spin_lock_init(&fs_info->unused_bgs_lock);
2531         rwlock_init(&fs_info->tree_mod_log_lock);
2532         mutex_init(&fs_info->unused_bg_unpin_mutex);
2533         mutex_init(&fs_info->delete_unused_bgs_mutex);
2534         mutex_init(&fs_info->reloc_mutex);
2535         mutex_init(&fs_info->delalloc_root_mutex);
2536         seqlock_init(&fs_info->profiles_lock);
2537         init_rwsem(&fs_info->delayed_iput_sem);
2538
2539         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2540         INIT_LIST_HEAD(&fs_info->space_info);
2541         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2542         INIT_LIST_HEAD(&fs_info->unused_bgs);
2543         btrfs_mapping_init(&fs_info->mapping_tree);
2544         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2545                              BTRFS_BLOCK_RSV_GLOBAL);
2546         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2547                              BTRFS_BLOCK_RSV_DELALLOC);
2548         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2549         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2550         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2551         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2552                              BTRFS_BLOCK_RSV_DELOPS);
2553         atomic_set(&fs_info->nr_async_submits, 0);
2554         atomic_set(&fs_info->async_delalloc_pages, 0);
2555         atomic_set(&fs_info->async_submit_draining, 0);
2556         atomic_set(&fs_info->nr_async_bios, 0);
2557         atomic_set(&fs_info->defrag_running, 0);
2558         atomic_set(&fs_info->qgroup_op_seq, 0);
2559         atomic64_set(&fs_info->tree_mod_seq, 0);
2560         fs_info->sb = sb;
2561         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2562         fs_info->metadata_ratio = 0;
2563         fs_info->defrag_inodes = RB_ROOT;
2564         fs_info->free_chunk_space = 0;
2565         fs_info->tree_mod_log = RB_ROOT;
2566         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2567         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2568         /* readahead state */
2569         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2570         spin_lock_init(&fs_info->reada_lock);
2571
2572         fs_info->thread_pool_size = min_t(unsigned long,
2573                                           num_online_cpus() + 2, 8);
2574
2575         INIT_LIST_HEAD(&fs_info->ordered_roots);
2576         spin_lock_init(&fs_info->ordered_root_lock);
2577         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2578                                         GFP_NOFS);
2579         if (!fs_info->delayed_root) {
2580                 err = -ENOMEM;
2581                 goto fail_iput;
2582         }
2583         btrfs_init_delayed_root(fs_info->delayed_root);
2584
2585         btrfs_init_scrub(fs_info);
2586 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2587         fs_info->check_integrity_print_mask = 0;
2588 #endif
2589         btrfs_init_balance(fs_info);
2590         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2591
2592         sb->s_blocksize = 4096;
2593         sb->s_blocksize_bits = blksize_bits(4096);
2594         sb->s_bdi = &fs_info->bdi;
2595
2596         btrfs_init_btree_inode(fs_info, tree_root);
2597
2598         spin_lock_init(&fs_info->block_group_cache_lock);
2599         fs_info->block_group_cache_tree = RB_ROOT;
2600         fs_info->first_logical_byte = (u64)-1;
2601
2602         extent_io_tree_init(&fs_info->freed_extents[0],
2603                              fs_info->btree_inode->i_mapping);
2604         extent_io_tree_init(&fs_info->freed_extents[1],
2605                              fs_info->btree_inode->i_mapping);
2606         fs_info->pinned_extents = &fs_info->freed_extents[0];
2607         fs_info->do_barriers = 1;
2608
2609
2610         mutex_init(&fs_info->ordered_operations_mutex);
2611         mutex_init(&fs_info->ordered_extent_flush_mutex);
2612         mutex_init(&fs_info->tree_log_mutex);
2613         mutex_init(&fs_info->chunk_mutex);
2614         mutex_init(&fs_info->transaction_kthread_mutex);
2615         mutex_init(&fs_info->cleaner_mutex);
2616         mutex_init(&fs_info->volume_mutex);
2617         mutex_init(&fs_info->ro_block_group_mutex);
2618         init_rwsem(&fs_info->commit_root_sem);
2619         init_rwsem(&fs_info->cleanup_work_sem);
2620         init_rwsem(&fs_info->subvol_sem);
2621         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2622
2623         btrfs_init_dev_replace_locks(fs_info);
2624         btrfs_init_qgroup(fs_info);
2625
2626         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2627         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2628
2629         init_waitqueue_head(&fs_info->transaction_throttle);
2630         init_waitqueue_head(&fs_info->transaction_wait);
2631         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2632         init_waitqueue_head(&fs_info->async_submit_wait);
2633
2634         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2635
2636         ret = btrfs_alloc_stripe_hash_table(fs_info);
2637         if (ret) {
2638                 err = ret;
2639                 goto fail_alloc;
2640         }
2641
2642         __setup_root(4096, 4096, 4096, tree_root,
2643                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2644
2645         invalidate_bdev(fs_devices->latest_bdev);
2646
2647         /*
2648          * Read super block and check the signature bytes only
2649          */
2650         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2651         if (!bh) {
2652                 err = -EINVAL;
2653                 goto fail_alloc;
2654         }
2655
2656         /*
2657          * We want to check superblock checksum, the type is stored inside.
2658          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2659          */
2660         if (btrfs_check_super_csum(bh->b_data)) {
2661                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2662                 err = -EINVAL;
2663                 goto fail_alloc;
2664         }
2665
2666         /*
2667          * super_copy is zeroed at allocation time and we never touch the
2668          * following bytes up to INFO_SIZE, the checksum is calculated from
2669          * the whole block of INFO_SIZE
2670          */
2671         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2672         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2673                sizeof(*fs_info->super_for_commit));
2674         brelse(bh);
2675
2676         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2677
2678         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2679         if (ret) {
2680                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2681                 err = -EINVAL;
2682                 goto fail_alloc;
2683         }
2684
2685         disk_super = fs_info->super_copy;
2686         if (!btrfs_super_root(disk_super))
2687                 goto fail_alloc;
2688
2689         /* check FS state, whether FS is broken. */
2690         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2691                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2692
2693         /*
2694          * run through our array of backup supers and setup
2695          * our ring pointer to the oldest one
2696          */
2697         generation = btrfs_super_generation(disk_super);
2698         find_oldest_super_backup(fs_info, generation);
2699
2700         /*
2701          * In the long term, we'll store the compression type in the super
2702          * block, and it'll be used for per file compression control.
2703          */
2704         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2705
2706         ret = btrfs_parse_options(tree_root, options);
2707         if (ret) {
2708                 err = ret;
2709                 goto fail_alloc;
2710         }
2711
2712         features = btrfs_super_incompat_flags(disk_super) &
2713                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2714         if (features) {
2715                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2716                        "unsupported optional features (%Lx).\n",
2717                        features);
2718                 err = -EINVAL;
2719                 goto fail_alloc;
2720         }
2721
2722         /*
2723          * Leafsize and nodesize were always equal, this is only a sanity check.
2724          */
2725         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2726             btrfs_super_nodesize(disk_super)) {
2727                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2728                        "blocksizes don't match.  node %d leaf %d\n",
2729                        btrfs_super_nodesize(disk_super),
2730                        le32_to_cpu(disk_super->__unused_leafsize));
2731                 err = -EINVAL;
2732                 goto fail_alloc;
2733         }
2734         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2735                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2736                        "blocksize (%d) was too large\n",
2737                        btrfs_super_nodesize(disk_super));
2738                 err = -EINVAL;
2739                 goto fail_alloc;
2740         }
2741
2742         features = btrfs_super_incompat_flags(disk_super);
2743         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2744         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2745                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2746
2747         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2748                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2749
2750         /*
2751          * flag our filesystem as having big metadata blocks if
2752          * they are bigger than the page size
2753          */
2754         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2755                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2756                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2757                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2758         }
2759
2760         nodesize = btrfs_super_nodesize(disk_super);
2761         sectorsize = btrfs_super_sectorsize(disk_super);
2762         stripesize = btrfs_super_stripesize(disk_super);
2763         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2764         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2765
2766         /*
2767          * mixed block groups end up with duplicate but slightly offset
2768          * extent buffers for the same range.  It leads to corruptions
2769          */
2770         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2771             (sectorsize != nodesize)) {
2772                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2773                                 "are not allowed for mixed block groups on %s\n",
2774                                 sb->s_id);
2775                 goto fail_alloc;
2776         }
2777
2778         /*
2779          * Needn't use the lock because there is no other task which will
2780          * update the flag.
2781          */
2782         btrfs_set_super_incompat_flags(disk_super, features);
2783
2784         features = btrfs_super_compat_ro_flags(disk_super) &
2785                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2786         if (!(sb->s_flags & MS_RDONLY) && features) {
2787                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2788                        "unsupported option features (%Lx).\n",
2789                        features);
2790                 err = -EINVAL;
2791                 goto fail_alloc;
2792         }
2793
2794         max_active = fs_info->thread_pool_size;
2795
2796         ret = btrfs_init_workqueues(fs_info, fs_devices);
2797         if (ret) {
2798                 err = ret;
2799                 goto fail_sb_buffer;
2800         }
2801
2802         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2803         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2804                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2805
2806         tree_root->nodesize = nodesize;
2807         tree_root->sectorsize = sectorsize;
2808         tree_root->stripesize = stripesize;
2809
2810         sb->s_blocksize = sectorsize;
2811         sb->s_blocksize_bits = blksize_bits(sectorsize);
2812
2813         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2814                 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2815                 goto fail_sb_buffer;
2816         }
2817
2818         if (sectorsize != PAGE_SIZE) {
2819                 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2820                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2821                 goto fail_sb_buffer;
2822         }
2823
2824         mutex_lock(&fs_info->chunk_mutex);
2825         ret = btrfs_read_sys_array(tree_root);
2826         mutex_unlock(&fs_info->chunk_mutex);
2827         if (ret) {
2828                 printk(KERN_ERR "BTRFS: failed to read the system "
2829                        "array on %s\n", sb->s_id);
2830                 goto fail_sb_buffer;
2831         }
2832
2833         generation = btrfs_super_chunk_root_generation(disk_super);
2834
2835         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2836                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2837
2838         chunk_root->node = read_tree_block(chunk_root,
2839                                            btrfs_super_chunk_root(disk_super),
2840                                            generation);
2841         if (IS_ERR(chunk_root->node) ||
2842             !extent_buffer_uptodate(chunk_root->node)) {
2843                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2844                        sb->s_id);
2845                 chunk_root->node = NULL;
2846                 goto fail_tree_roots;
2847         }
2848         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2849         chunk_root->commit_root = btrfs_root_node(chunk_root);
2850
2851         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2852            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2853
2854         ret = btrfs_read_chunk_tree(chunk_root);
2855         if (ret) {
2856                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2857                        sb->s_id);
2858                 goto fail_tree_roots;
2859         }
2860
2861         /*
2862          * keep the device that is marked to be the target device for the
2863          * dev_replace procedure
2864          */
2865         btrfs_close_extra_devices(fs_devices, 0);
2866
2867         if (!fs_devices->latest_bdev) {
2868                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2869                        sb->s_id);
2870                 goto fail_tree_roots;
2871         }
2872
2873 retry_root_backup:
2874         generation = btrfs_super_generation(disk_super);
2875
2876         tree_root->node = read_tree_block(tree_root,
2877                                           btrfs_super_root(disk_super),
2878                                           generation);
2879         if (IS_ERR(tree_root->node) ||
2880             !extent_buffer_uptodate(tree_root->node)) {
2881                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2882                        sb->s_id);
2883                 tree_root->node = NULL;
2884                 goto recovery_tree_root;
2885         }
2886
2887         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2888         tree_root->commit_root = btrfs_root_node(tree_root);
2889         btrfs_set_root_refs(&tree_root->root_item, 1);
2890
2891         ret = btrfs_read_roots(fs_info, tree_root);
2892         if (ret)
2893                 goto recovery_tree_root;
2894
2895         fs_info->generation = generation;
2896         fs_info->last_trans_committed = generation;
2897
2898         ret = btrfs_recover_balance(fs_info);
2899         if (ret) {
2900                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2901                 goto fail_block_groups;
2902         }
2903
2904         ret = btrfs_init_dev_stats(fs_info);
2905         if (ret) {
2906                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2907                        ret);
2908                 goto fail_block_groups;
2909         }
2910
2911         ret = btrfs_init_dev_replace(fs_info);
2912         if (ret) {
2913                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2914                 goto fail_block_groups;
2915         }
2916
2917         btrfs_close_extra_devices(fs_devices, 1);
2918
2919         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2920         if (ret) {
2921                 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2922                 goto fail_block_groups;
2923         }
2924
2925         ret = btrfs_sysfs_add_device(fs_devices);
2926         if (ret) {
2927                 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2928                 goto fail_fsdev_sysfs;
2929         }
2930
2931         ret = btrfs_sysfs_add_one(fs_info);
2932         if (ret) {
2933                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2934                 goto fail_fsdev_sysfs;
2935         }
2936
2937         ret = btrfs_init_space_info(fs_info);
2938         if (ret) {
2939                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2940                 goto fail_sysfs;
2941         }
2942
2943         ret = btrfs_read_block_groups(fs_info->extent_root);
2944         if (ret) {
2945                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2946                 goto fail_sysfs;
2947         }
2948         fs_info->num_tolerated_disk_barrier_failures =
2949                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2950         if (fs_info->fs_devices->missing_devices >
2951              fs_info->num_tolerated_disk_barrier_failures &&
2952             !(sb->s_flags & MS_RDONLY)) {
2953                 printk(KERN_WARNING "BTRFS: "
2954                         "too many missing devices, writeable mount is not allowed\n");
2955                 goto fail_sysfs;
2956         }
2957
2958         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2959                                                "btrfs-cleaner");
2960         if (IS_ERR(fs_info->cleaner_kthread))
2961                 goto fail_sysfs;
2962
2963         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2964                                                    tree_root,
2965                                                    "btrfs-transaction");
2966         if (IS_ERR(fs_info->transaction_kthread))
2967                 goto fail_cleaner;
2968
2969         if (!btrfs_test_opt(tree_root, SSD) &&
2970             !btrfs_test_opt(tree_root, NOSSD) &&
2971             !fs_info->fs_devices->rotating) {
2972                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2973                        "mode\n");
2974                 btrfs_set_opt(fs_info->mount_opt, SSD);
2975         }
2976
2977         /*
2978          * Mount does not set all options immediatelly, we can do it now and do
2979          * not have to wait for transaction commit
2980          */
2981         btrfs_apply_pending_changes(fs_info);
2982
2983 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2984         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2985                 ret = btrfsic_mount(tree_root, fs_devices,
2986                                     btrfs_test_opt(tree_root,
2987                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2988                                     1 : 0,
2989                                     fs_info->check_integrity_print_mask);
2990                 if (ret)
2991                         printk(KERN_WARNING "BTRFS: failed to initialize"
2992                                " integrity check module %s\n", sb->s_id);
2993         }
2994 #endif
2995         ret = btrfs_read_qgroup_config(fs_info);
2996         if (ret)
2997                 goto fail_trans_kthread;
2998
2999         /* do not make disk changes in broken FS */
3000         if (btrfs_super_log_root(disk_super) != 0) {
3001                 ret = btrfs_replay_log(fs_info, fs_devices);
3002                 if (ret) {
3003                         err = ret;
3004                         goto fail_qgroup;
3005                 }
3006         }
3007
3008         ret = btrfs_find_orphan_roots(tree_root);
3009         if (ret)
3010                 goto fail_qgroup;
3011
3012         if (!(sb->s_flags & MS_RDONLY)) {
3013                 ret = btrfs_cleanup_fs_roots(fs_info);
3014                 if (ret)
3015                         goto fail_qgroup;
3016
3017                 mutex_lock(&fs_info->cleaner_mutex);
3018                 ret = btrfs_recover_relocation(tree_root);
3019                 mutex_unlock(&fs_info->cleaner_mutex);
3020                 if (ret < 0) {
3021                         printk(KERN_WARNING
3022                                "BTRFS: failed to recover relocation\n");
3023                         err = -EINVAL;
3024                         goto fail_qgroup;
3025                 }
3026         }
3027
3028         location.objectid = BTRFS_FS_TREE_OBJECTID;
3029         location.type = BTRFS_ROOT_ITEM_KEY;
3030         location.offset = 0;
3031
3032         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3033         if (IS_ERR(fs_info->fs_root)) {
3034                 err = PTR_ERR(fs_info->fs_root);
3035                 goto fail_qgroup;
3036         }
3037
3038         if (sb->s_flags & MS_RDONLY)
3039                 return 0;
3040
3041         down_read(&fs_info->cleanup_work_sem);
3042         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3043             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3044                 up_read(&fs_info->cleanup_work_sem);
3045                 close_ctree(tree_root);
3046                 return ret;
3047         }
3048         up_read(&fs_info->cleanup_work_sem);
3049
3050         ret = btrfs_resume_balance_async(fs_info);
3051         if (ret) {
3052                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3053                 close_ctree(tree_root);
3054                 return ret;
3055         }
3056
3057         ret = btrfs_resume_dev_replace_async(fs_info);
3058         if (ret) {
3059                 pr_warn("BTRFS: failed to resume dev_replace\n");
3060                 close_ctree(tree_root);
3061                 return ret;
3062         }
3063
3064         btrfs_qgroup_rescan_resume(fs_info);
3065
3066         if (!fs_info->uuid_root) {
3067                 pr_info("BTRFS: creating UUID tree\n");
3068                 ret = btrfs_create_uuid_tree(fs_info);
3069                 if (ret) {
3070                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3071                                 ret);
3072                         close_ctree(tree_root);
3073                         return ret;
3074                 }
3075         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3076                    fs_info->generation !=
3077                                 btrfs_super_uuid_tree_generation(disk_super)) {
3078                 pr_info("BTRFS: checking UUID tree\n");
3079                 ret = btrfs_check_uuid_tree(fs_info);
3080                 if (ret) {
3081                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3082                                 ret);
3083                         close_ctree(tree_root);
3084                         return ret;
3085                 }
3086         } else {
3087                 fs_info->update_uuid_tree_gen = 1;
3088         }
3089
3090         fs_info->open = 1;
3091
3092         return 0;
3093
3094 fail_qgroup:
3095         btrfs_free_qgroup_config(fs_info);
3096 fail_trans_kthread:
3097         kthread_stop(fs_info->transaction_kthread);
3098         btrfs_cleanup_transaction(fs_info->tree_root);
3099         btrfs_free_fs_roots(fs_info);
3100 fail_cleaner:
3101         kthread_stop(fs_info->cleaner_kthread);
3102
3103         /*
3104          * make sure we're done with the btree inode before we stop our
3105          * kthreads
3106          */
3107         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3108
3109 fail_sysfs:
3110         btrfs_sysfs_remove_one(fs_info);
3111
3112 fail_fsdev_sysfs:
3113         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3114
3115 fail_block_groups:
3116         btrfs_put_block_group_cache(fs_info);
3117         btrfs_free_block_groups(fs_info);
3118
3119 fail_tree_roots:
3120         free_root_pointers(fs_info, 1);
3121         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3122
3123 fail_sb_buffer:
3124         btrfs_stop_all_workers(fs_info);
3125 fail_alloc:
3126 fail_iput:
3127         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3128
3129         iput(fs_info->btree_inode);
3130 fail_bio_counter:
3131         percpu_counter_destroy(&fs_info->bio_counter);
3132 fail_delalloc_bytes:
3133         percpu_counter_destroy(&fs_info->delalloc_bytes);
3134 fail_dirty_metadata_bytes:
3135         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3136 fail_bdi:
3137         bdi_destroy(&fs_info->bdi);
3138 fail_srcu:
3139         cleanup_srcu_struct(&fs_info->subvol_srcu);
3140 fail:
3141         btrfs_free_stripe_hash_table(fs_info);
3142         btrfs_close_devices(fs_info->fs_devices);
3143         return err;
3144
3145 recovery_tree_root:
3146         if (!btrfs_test_opt(tree_root, RECOVERY))
3147                 goto fail_tree_roots;
3148
3149         free_root_pointers(fs_info, 0);
3150
3151         /* don't use the log in recovery mode, it won't be valid */
3152         btrfs_set_super_log_root(disk_super, 0);
3153
3154         /* we can't trust the free space cache either */
3155         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3156
3157         ret = next_root_backup(fs_info, fs_info->super_copy,
3158                                &num_backups_tried, &backup_index);
3159         if (ret == -1)
3160                 goto fail_block_groups;
3161         goto retry_root_backup;
3162 }
3163
3164 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3165 {
3166         if (uptodate) {
3167                 set_buffer_uptodate(bh);
3168         } else {
3169                 struct btrfs_device *device = (struct btrfs_device *)
3170                         bh->b_private;
3171
3172                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3173                                           "I/O error on %s\n",
3174                                           rcu_str_deref(device->name));
3175                 /* note, we dont' set_buffer_write_io_error because we have
3176                  * our own ways of dealing with the IO errors
3177                  */
3178                 clear_buffer_uptodate(bh);
3179                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3180         }
3181         unlock_buffer(bh);
3182         put_bh(bh);
3183 }
3184
3185 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3186 {
3187         struct buffer_head *bh;
3188         struct buffer_head *latest = NULL;
3189         struct btrfs_super_block *super;
3190         int i;
3191         u64 transid = 0;
3192         u64 bytenr;
3193
3194         /* we would like to check all the supers, but that would make
3195          * a btrfs mount succeed after a mkfs from a different FS.
3196          * So, we need to add a special mount option to scan for
3197          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3198          */
3199         for (i = 0; i < 1; i++) {
3200                 bytenr = btrfs_sb_offset(i);
3201                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3202                                         i_size_read(bdev->bd_inode))
3203                         break;
3204                 bh = __bread(bdev, bytenr / 4096,
3205                                         BTRFS_SUPER_INFO_SIZE);
3206                 if (!bh)
3207                         continue;
3208
3209                 super = (struct btrfs_super_block *)bh->b_data;
3210                 if (btrfs_super_bytenr(super) != bytenr ||
3211                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3212                         brelse(bh);
3213                         continue;
3214                 }
3215
3216                 if (!latest || btrfs_super_generation(super) > transid) {
3217                         brelse(latest);
3218                         latest = bh;
3219                         transid = btrfs_super_generation(super);
3220                 } else {
3221                         brelse(bh);
3222                 }
3223         }
3224         return latest;
3225 }
3226
3227 /*
3228  * this should be called twice, once with wait == 0 and
3229  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3230  * we write are pinned.
3231  *
3232  * They are released when wait == 1 is done.
3233  * max_mirrors must be the same for both runs, and it indicates how
3234  * many supers on this one device should be written.
3235  *
3236  * max_mirrors == 0 means to write them all.
3237  */
3238 static int write_dev_supers(struct btrfs_device *device,
3239                             struct btrfs_super_block *sb,
3240                             int do_barriers, int wait, int max_mirrors)
3241 {
3242         struct buffer_head *bh;
3243         int i;
3244         int ret;
3245         int errors = 0;
3246         u32 crc;
3247         u64 bytenr;
3248
3249         if (max_mirrors == 0)
3250                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3251
3252         for (i = 0; i < max_mirrors; i++) {
3253                 bytenr = btrfs_sb_offset(i);
3254                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3255                     device->commit_total_bytes)
3256                         break;
3257
3258                 if (wait) {
3259                         bh = __find_get_block(device->bdev, bytenr / 4096,
3260                                               BTRFS_SUPER_INFO_SIZE);
3261                         if (!bh) {
3262                                 errors++;
3263                                 continue;
3264                         }
3265                         wait_on_buffer(bh);
3266                         if (!buffer_uptodate(bh))
3267                                 errors++;
3268
3269                         /* drop our reference */
3270                         brelse(bh);
3271
3272                         /* drop the reference from the wait == 0 run */
3273                         brelse(bh);
3274                         continue;
3275                 } else {
3276                         btrfs_set_super_bytenr(sb, bytenr);
3277
3278                         crc = ~(u32)0;
3279                         crc = btrfs_csum_data((char *)sb +
3280                                               BTRFS_CSUM_SIZE, crc,
3281                                               BTRFS_SUPER_INFO_SIZE -
3282                                               BTRFS_CSUM_SIZE);
3283                         btrfs_csum_final(crc, sb->csum);
3284
3285                         /*
3286                          * one reference for us, and we leave it for the
3287                          * caller
3288                          */
3289                         bh = __getblk(device->bdev, bytenr / 4096,
3290                                       BTRFS_SUPER_INFO_SIZE);
3291                         if (!bh) {
3292                                 printk(KERN_ERR "BTRFS: couldn't get super "
3293                                        "buffer head for bytenr %Lu\n", bytenr);
3294                                 errors++;
3295                                 continue;
3296                         }
3297
3298                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3299
3300                         /* one reference for submit_bh */
3301                         get_bh(bh);
3302
3303                         set_buffer_uptodate(bh);
3304                         lock_buffer(bh);
3305                         bh->b_end_io = btrfs_end_buffer_write_sync;
3306                         bh->b_private = device;
3307                 }
3308
3309                 /*
3310                  * we fua the first super.  The others we allow
3311                  * to go down lazy.
3312                  */
3313                 if (i == 0)
3314                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3315                 else
3316                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3317                 if (ret)
3318                         errors++;
3319         }
3320         return errors < i ? 0 : -1;
3321 }
3322
3323 /*
3324  * endio for the write_dev_flush, this will wake anyone waiting
3325  * for the barrier when it is done
3326  */
3327 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3328 {
3329         if (err)
3330                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3331         if (bio->bi_private)
3332                 complete(bio->bi_private);
3333         bio_put(bio);
3334 }
3335
3336 /*
3337  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3338  * sent down.  With wait == 1, it waits for the previous flush.
3339  *
3340  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3341  * capable
3342  */
3343 static int write_dev_flush(struct btrfs_device *device, int wait)
3344 {
3345         struct bio *bio;
3346         int ret = 0;
3347
3348         if (device->nobarriers)
3349                 return 0;
3350
3351         if (wait) {
3352                 bio = device->flush_bio;
3353                 if (!bio)
3354                         return 0;
3355
3356                 wait_for_completion(&device->flush_wait);
3357
3358                 if (!bio_flagged(bio, BIO_UPTODATE)) {
3359                         ret = -EIO;
3360                         btrfs_dev_stat_inc_and_print(device,
3361                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3362                 }
3363
3364                 /* drop the reference from the wait == 0 run */
3365                 bio_put(bio);
3366                 device->flush_bio = NULL;
3367
3368                 return ret;
3369         }
3370
3371         /*
3372          * one reference for us, and we leave it for the
3373          * caller
3374          */
3375         device->flush_bio = NULL;
3376         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3377         if (!bio)
3378                 return -ENOMEM;
3379
3380         bio->bi_end_io = btrfs_end_empty_barrier;
3381         bio->bi_bdev = device->bdev;
3382         init_completion(&device->flush_wait);
3383         bio->bi_private = &device->flush_wait;
3384         device->flush_bio = bio;
3385
3386         bio_get(bio);
3387         btrfsic_submit_bio(WRITE_FLUSH, bio);
3388
3389         return 0;
3390 }
3391
3392 /*
3393  * send an empty flush down to each device in parallel,
3394  * then wait for them
3395  */
3396 static int barrier_all_devices(struct btrfs_fs_info *info)
3397 {
3398         struct list_head *head;
3399         struct btrfs_device *dev;
3400         int errors_send = 0;
3401         int errors_wait = 0;
3402         int ret;
3403
3404         /* send down all the barriers */
3405         head = &info->fs_devices->devices;
3406         list_for_each_entry_rcu(dev, head, dev_list) {
3407                 if (dev->missing)
3408                         continue;
3409                 if (!dev->bdev) {
3410                         errors_send++;
3411                         continue;
3412                 }
3413                 if (!dev->in_fs_metadata || !dev->writeable)
3414                         continue;
3415
3416                 ret = write_dev_flush(dev, 0);
3417                 if (ret)
3418                         errors_send++;
3419         }
3420
3421         /* wait for all the barriers */
3422         list_for_each_entry_rcu(dev, head, dev_list) {
3423                 if (dev->missing)
3424                         continue;
3425                 if (!dev->bdev) {
3426                         errors_wait++;
3427                         continue;
3428                 }
3429                 if (!dev->in_fs_metadata || !dev->writeable)
3430                         continue;
3431
3432                 ret = write_dev_flush(dev, 1);
3433                 if (ret)
3434                         errors_wait++;
3435         }
3436         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3437             errors_wait > info->num_tolerated_disk_barrier_failures)
3438                 return -EIO;
3439         return 0;
3440 }
3441
3442 int btrfs_calc_num_tolerated_disk_barrier_failures(
3443         struct btrfs_fs_info *fs_info)
3444 {
3445         struct btrfs_ioctl_space_info space;
3446         struct btrfs_space_info *sinfo;
3447         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3448                        BTRFS_BLOCK_GROUP_SYSTEM,
3449                        BTRFS_BLOCK_GROUP_METADATA,
3450                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3451         int num_types = 4;
3452         int i;
3453         int c;
3454         int num_tolerated_disk_barrier_failures =
3455                 (int)fs_info->fs_devices->num_devices;
3456
3457         for (i = 0; i < num_types; i++) {
3458                 struct btrfs_space_info *tmp;
3459
3460                 sinfo = NULL;
3461                 rcu_read_lock();
3462                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3463                         if (tmp->flags == types[i]) {
3464                                 sinfo = tmp;
3465                                 break;
3466                         }
3467                 }
3468                 rcu_read_unlock();
3469
3470                 if (!sinfo)
3471                         continue;
3472
3473                 down_read(&sinfo->groups_sem);
3474                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3475                         if (!list_empty(&sinfo->block_groups[c])) {
3476                                 u64 flags;
3477
3478                                 btrfs_get_block_group_info(
3479                                         &sinfo->block_groups[c], &space);
3480                                 if (space.total_bytes == 0 ||
3481                                     space.used_bytes == 0)
3482                                         continue;
3483                                 flags = space.flags;
3484                                 /*
3485                                  * return
3486                                  * 0: if dup, single or RAID0 is configured for
3487                                  *    any of metadata, system or data, else
3488                                  * 1: if RAID5 is configured, or if RAID1 or
3489                                  *    RAID10 is configured and only two mirrors
3490                                  *    are used, else
3491                                  * 2: if RAID6 is configured, else
3492                                  * num_mirrors - 1: if RAID1 or RAID10 is
3493                                  *                  configured and more than
3494                                  *                  2 mirrors are used.
3495                                  */
3496                                 if (num_tolerated_disk_barrier_failures > 0 &&
3497                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3498                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3499                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3500                                       == 0)))
3501                                         num_tolerated_disk_barrier_failures = 0;
3502                                 else if (num_tolerated_disk_barrier_failures > 1) {
3503                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3504                                             BTRFS_BLOCK_GROUP_RAID5 |
3505                                             BTRFS_BLOCK_GROUP_RAID10)) {
3506                                                 num_tolerated_disk_barrier_failures = 1;
3507                                         } else if (flags &
3508                                                    BTRFS_BLOCK_GROUP_RAID6) {
3509                                                 num_tolerated_disk_barrier_failures = 2;
3510                                         }
3511                                 }
3512                         }
3513                 }
3514                 up_read(&sinfo->groups_sem);
3515         }
3516
3517         return num_tolerated_disk_barrier_failures;
3518 }
3519
3520 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3521 {
3522         struct list_head *head;
3523         struct btrfs_device *dev;
3524         struct btrfs_super_block *sb;
3525         struct btrfs_dev_item *dev_item;
3526         int ret;
3527         int do_barriers;
3528         int max_errors;
3529         int total_errors = 0;
3530         u64 flags;
3531
3532         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3533         backup_super_roots(root->fs_info);
3534
3535         sb = root->fs_info->super_for_commit;
3536         dev_item = &sb->dev_item;
3537
3538         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3539         head = &root->fs_info->fs_devices->devices;
3540         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3541
3542         if (do_barriers) {
3543                 ret = barrier_all_devices(root->fs_info);
3544                 if (ret) {
3545                         mutex_unlock(
3546                                 &root->fs_info->fs_devices->device_list_mutex);
3547                         btrfs_error(root->fs_info, ret,
3548                                     "errors while submitting device barriers.");
3549                         return ret;
3550                 }
3551         }
3552
3553         list_for_each_entry_rcu(dev, head, dev_list) {
3554                 if (!dev->bdev) {
3555                         total_errors++;
3556                         continue;
3557                 }
3558                 if (!dev->in_fs_metadata || !dev->writeable)
3559                         continue;
3560
3561                 btrfs_set_stack_device_generation(dev_item, 0);
3562                 btrfs_set_stack_device_type(dev_item, dev->type);
3563                 btrfs_set_stack_device_id(dev_item, dev->devid);
3564                 btrfs_set_stack_device_total_bytes(dev_item,
3565                                                    dev->commit_total_bytes);
3566                 btrfs_set_stack_device_bytes_used(dev_item,
3567                                                   dev->commit_bytes_used);
3568                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3569                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3570                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3571                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3572                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3573
3574                 flags = btrfs_super_flags(sb);
3575                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3576
3577                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3578                 if (ret)
3579                         total_errors++;
3580         }
3581         if (total_errors > max_errors) {
3582                 btrfs_err(root->fs_info, "%d errors while writing supers",
3583                        total_errors);
3584                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3585
3586                 /* FUA is masked off if unsupported and can't be the reason */
3587                 btrfs_error(root->fs_info, -EIO,
3588                             "%d errors while writing supers", total_errors);
3589                 return -EIO;
3590         }
3591
3592         total_errors = 0;
3593         list_for_each_entry_rcu(dev, head, dev_list) {
3594                 if (!dev->bdev)
3595                         continue;
3596                 if (!dev->in_fs_metadata || !dev->writeable)
3597                         continue;
3598
3599                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3600                 if (ret)
3601                         total_errors++;
3602         }
3603         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3604         if (total_errors > max_errors) {
3605                 btrfs_error(root->fs_info, -EIO,
3606                             "%d errors while writing supers", total_errors);
3607                 return -EIO;
3608         }
3609         return 0;
3610 }
3611
3612 int write_ctree_super(struct btrfs_trans_handle *trans,
3613                       struct btrfs_root *root, int max_mirrors)
3614 {
3615         return write_all_supers(root, max_mirrors);
3616 }
3617
3618 /* Drop a fs root from the radix tree and free it. */
3619 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3620                                   struct btrfs_root *root)
3621 {
3622         spin_lock(&fs_info->fs_roots_radix_lock);
3623         radix_tree_delete(&fs_info->fs_roots_radix,
3624                           (unsigned long)root->root_key.objectid);
3625         spin_unlock(&fs_info->fs_roots_radix_lock);
3626
3627         if (btrfs_root_refs(&root->root_item) == 0)
3628                 synchronize_srcu(&fs_info->subvol_srcu);
3629
3630         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3631                 btrfs_free_log(NULL, root);
3632
3633         if (root->free_ino_pinned)
3634                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3635         if (root->free_ino_ctl)
3636                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3637         free_fs_root(root);
3638 }
3639
3640 static void free_fs_root(struct btrfs_root *root)
3641 {
3642         iput(root->ino_cache_inode);
3643         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3644         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3645         root->orphan_block_rsv = NULL;
3646         if (root->anon_dev)
3647                 free_anon_bdev(root->anon_dev);
3648         if (root->subv_writers)
3649                 btrfs_free_subvolume_writers(root->subv_writers);
3650         free_extent_buffer(root->node);
3651         free_extent_buffer(root->commit_root);
3652         kfree(root->free_ino_ctl);
3653         kfree(root->free_ino_pinned);
3654         kfree(root->name);
3655         btrfs_put_fs_root(root);
3656 }
3657
3658 void btrfs_free_fs_root(struct btrfs_root *root)
3659 {
3660         free_fs_root(root);
3661 }
3662
3663 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3664 {
3665         u64 root_objectid = 0;
3666         struct btrfs_root *gang[8];
3667         int i = 0;
3668         int err = 0;
3669         unsigned int ret = 0;
3670         int index;
3671
3672         while (1) {
3673                 index = srcu_read_lock(&fs_info->subvol_srcu);
3674                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3675                                              (void **)gang, root_objectid,
3676                                              ARRAY_SIZE(gang));
3677                 if (!ret) {
3678                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3679                         break;
3680                 }
3681                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3682
3683                 for (i = 0; i < ret; i++) {
3684                         /* Avoid to grab roots in dead_roots */
3685                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3686                                 gang[i] = NULL;
3687                                 continue;
3688                         }
3689                         /* grab all the search result for later use */
3690                         gang[i] = btrfs_grab_fs_root(gang[i]);
3691                 }
3692                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3693
3694                 for (i = 0; i < ret; i++) {
3695                         if (!gang[i])
3696                                 continue;
3697                         root_objectid = gang[i]->root_key.objectid;
3698                         err = btrfs_orphan_cleanup(gang[i]);
3699                         if (err)
3700                                 break;
3701                         btrfs_put_fs_root(gang[i]);
3702                 }
3703                 root_objectid++;
3704         }
3705
3706         /* release the uncleaned roots due to error */
3707         for (; i < ret; i++) {
3708                 if (gang[i])
3709                         btrfs_put_fs_root(gang[i]);
3710         }
3711         return err;
3712 }
3713
3714 int btrfs_commit_super(struct btrfs_root *root)
3715 {
3716         struct btrfs_trans_handle *trans;
3717
3718         mutex_lock(&root->fs_info->cleaner_mutex);
3719         btrfs_run_delayed_iputs(root);
3720         mutex_unlock(&root->fs_info->cleaner_mutex);
3721         wake_up_process(root->fs_info->cleaner_kthread);
3722
3723         /* wait until ongoing cleanup work done */
3724         down_write(&root->fs_info->cleanup_work_sem);
3725         up_write(&root->fs_info->cleanup_work_sem);
3726
3727         trans = btrfs_join_transaction(root);
3728         if (IS_ERR(trans))
3729                 return PTR_ERR(trans);
3730         return btrfs_commit_transaction(trans, root);
3731 }
3732
3733 void close_ctree(struct btrfs_root *root)
3734 {
3735         struct btrfs_fs_info *fs_info = root->fs_info;
3736         int ret;
3737
3738         fs_info->closing = 1;
3739         smp_mb();
3740
3741         /* wait for the uuid_scan task to finish */
3742         down(&fs_info->uuid_tree_rescan_sem);
3743         /* avoid complains from lockdep et al., set sem back to initial state */
3744         up(&fs_info->uuid_tree_rescan_sem);
3745
3746         /* pause restriper - we want to resume on mount */
3747         btrfs_pause_balance(fs_info);
3748
3749         btrfs_dev_replace_suspend_for_unmount(fs_info);
3750
3751         btrfs_scrub_cancel(fs_info);
3752
3753         /* wait for any defraggers to finish */
3754         wait_event(fs_info->transaction_wait,
3755                    (atomic_read(&fs_info->defrag_running) == 0));
3756
3757         /* clear out the rbtree of defraggable inodes */
3758         btrfs_cleanup_defrag_inodes(fs_info);
3759
3760         cancel_work_sync(&fs_info->async_reclaim_work);
3761
3762         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3763                 ret = btrfs_commit_super(root);
3764                 if (ret)
3765                         btrfs_err(fs_info, "commit super ret %d", ret);
3766         }
3767
3768         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3769                 btrfs_error_commit_super(root);
3770
3771         kthread_stop(fs_info->transaction_kthread);
3772         kthread_stop(fs_info->cleaner_kthread);
3773
3774         fs_info->closing = 2;
3775         smp_mb();
3776
3777         btrfs_free_qgroup_config(fs_info);
3778
3779         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3780                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3781                        percpu_counter_sum(&fs_info->delalloc_bytes));
3782         }
3783
3784         btrfs_sysfs_remove_one(fs_info);
3785         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3786
3787         btrfs_free_fs_roots(fs_info);
3788
3789         btrfs_put_block_group_cache(fs_info);
3790
3791         btrfs_free_block_groups(fs_info);
3792
3793         /*
3794          * we must make sure there is not any read request to
3795          * submit after we stopping all workers.
3796          */
3797         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3798         btrfs_stop_all_workers(fs_info);
3799
3800         fs_info->open = 0;
3801         free_root_pointers(fs_info, 1);
3802
3803         iput(fs_info->btree_inode);
3804
3805 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3806         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3807                 btrfsic_unmount(root, fs_info->fs_devices);
3808 #endif
3809
3810         btrfs_close_devices(fs_info->fs_devices);
3811         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3812
3813         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3814         percpu_counter_destroy(&fs_info->delalloc_bytes);
3815         percpu_counter_destroy(&fs_info->bio_counter);
3816         bdi_destroy(&fs_info->bdi);
3817         cleanup_srcu_struct(&fs_info->subvol_srcu);
3818
3819         btrfs_free_stripe_hash_table(fs_info);
3820
3821         __btrfs_free_block_rsv(root->orphan_block_rsv);
3822         root->orphan_block_rsv = NULL;
3823
3824         lock_chunks(root);
3825         while (!list_empty(&fs_info->pinned_chunks)) {
3826                 struct extent_map *em;
3827
3828                 em = list_first_entry(&fs_info->pinned_chunks,
3829                                       struct extent_map, list);
3830                 list_del_init(&em->list);
3831                 free_extent_map(em);
3832         }
3833         unlock_chunks(root);
3834 }
3835
3836 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3837                           int atomic)
3838 {
3839         int ret;
3840         struct inode *btree_inode = buf->pages[0]->mapping->host;
3841
3842         ret = extent_buffer_uptodate(buf);
3843         if (!ret)
3844                 return ret;
3845
3846         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3847                                     parent_transid, atomic);
3848         if (ret == -EAGAIN)
3849                 return ret;
3850         return !ret;
3851 }
3852
3853 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3854 {
3855         return set_extent_buffer_uptodate(buf);
3856 }
3857
3858 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3859 {
3860         struct btrfs_root *root;
3861         u64 transid = btrfs_header_generation(buf);
3862         int was_dirty;
3863
3864 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3865         /*
3866          * This is a fast path so only do this check if we have sanity tests
3867          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3868          * outside of the sanity tests.
3869          */
3870         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3871                 return;
3872 #endif
3873         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3874         btrfs_assert_tree_locked(buf);
3875         if (transid != root->fs_info->generation)
3876                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3877                        "found %llu running %llu\n",
3878                         buf->start, transid, root->fs_info->generation);
3879         was_dirty = set_extent_buffer_dirty(buf);
3880         if (!was_dirty)
3881                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3882                                      buf->len,
3883                                      root->fs_info->dirty_metadata_batch);
3884 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3885         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3886                 btrfs_print_leaf(root, buf);
3887                 ASSERT(0);
3888         }
3889 #endif
3890 }
3891
3892 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3893                                         int flush_delayed)
3894 {
3895         /*
3896          * looks as though older kernels can get into trouble with
3897          * this code, they end up stuck in balance_dirty_pages forever
3898          */
3899         int ret;
3900
3901         if (current->flags & PF_MEMALLOC)
3902                 return;
3903
3904         if (flush_delayed)
3905                 btrfs_balance_delayed_items(root);
3906
3907         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3908                                      BTRFS_DIRTY_METADATA_THRESH);
3909         if (ret > 0) {
3910                 balance_dirty_pages_ratelimited(
3911                                    root->fs_info->btree_inode->i_mapping);
3912         }
3913         return;
3914 }
3915
3916 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3917 {
3918         __btrfs_btree_balance_dirty(root, 1);
3919 }
3920
3921 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3922 {
3923         __btrfs_btree_balance_dirty(root, 0);
3924 }
3925
3926 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3927 {
3928         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3929         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3930 }
3931
3932 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3933                               int read_only)
3934 {
3935         struct btrfs_super_block *sb = fs_info->super_copy;
3936         int ret = 0;
3937
3938         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3939                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3940                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3941                 ret = -EINVAL;
3942         }
3943         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3944                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3945                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3946                 ret = -EINVAL;
3947         }
3948         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3949                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3950                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3951                 ret = -EINVAL;
3952         }
3953
3954         /*
3955          * The common minimum, we don't know if we can trust the nodesize/sectorsize
3956          * items yet, they'll be verified later. Issue just a warning.
3957          */
3958         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3959                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3960                                 btrfs_super_root(sb));
3961         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3962                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3963                                 btrfs_super_chunk_root(sb));
3964         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3965                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3966                                 btrfs_super_log_root(sb));
3967
3968         /*
3969          * Check the lower bound, the alignment and other constraints are
3970          * checked later.
3971          */
3972         if (btrfs_super_nodesize(sb) < 4096) {
3973                 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3974                                 btrfs_super_nodesize(sb));
3975                 ret = -EINVAL;
3976         }
3977         if (btrfs_super_sectorsize(sb) < 4096) {
3978                 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3979                                 btrfs_super_sectorsize(sb));
3980                 ret = -EINVAL;
3981         }
3982
3983         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3984                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3985                                 fs_info->fsid, sb->dev_item.fsid);
3986                 ret = -EINVAL;
3987         }
3988
3989         /*
3990          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3991          * done later
3992          */
3993         if (btrfs_super_num_devices(sb) > (1UL << 31))
3994                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3995                                 btrfs_super_num_devices(sb));
3996         if (btrfs_super_num_devices(sb) == 0) {
3997                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3998                 ret = -EINVAL;
3999         }
4000
4001         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4002                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4003                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4004                 ret = -EINVAL;
4005         }
4006
4007         /*
4008          * Obvious sys_chunk_array corruptions, it must hold at least one key
4009          * and one chunk
4010          */
4011         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4012                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4013                                 btrfs_super_sys_array_size(sb),
4014                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4015                 ret = -EINVAL;
4016         }
4017         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4018                         + sizeof(struct btrfs_chunk)) {
4019                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4020                                 btrfs_super_sys_array_size(sb),
4021                                 sizeof(struct btrfs_disk_key)
4022                                 + sizeof(struct btrfs_chunk));
4023                 ret = -EINVAL;
4024         }
4025
4026         /*
4027          * The generation is a global counter, we'll trust it more than the others
4028          * but it's still possible that it's the one that's wrong.
4029          */
4030         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4031                 printk(KERN_WARNING
4032                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4033                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4034         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4035             && btrfs_super_cache_generation(sb) != (u64)-1)
4036                 printk(KERN_WARNING
4037                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4038                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4039
4040         return ret;
4041 }
4042
4043 static void btrfs_error_commit_super(struct btrfs_root *root)
4044 {
4045         mutex_lock(&root->fs_info->cleaner_mutex);
4046         btrfs_run_delayed_iputs(root);
4047         mutex_unlock(&root->fs_info->cleaner_mutex);
4048
4049         down_write(&root->fs_info->cleanup_work_sem);
4050         up_write(&root->fs_info->cleanup_work_sem);
4051
4052         /* cleanup FS via transaction */
4053         btrfs_cleanup_transaction(root);
4054 }
4055
4056 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4057 {
4058         struct btrfs_ordered_extent *ordered;
4059
4060         spin_lock(&root->ordered_extent_lock);
4061         /*
4062          * This will just short circuit the ordered completion stuff which will
4063          * make sure the ordered extent gets properly cleaned up.
4064          */
4065         list_for_each_entry(ordered, &root->ordered_extents,
4066                             root_extent_list)
4067                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4068         spin_unlock(&root->ordered_extent_lock);
4069 }
4070
4071 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4072 {
4073         struct btrfs_root *root;
4074         struct list_head splice;
4075
4076         INIT_LIST_HEAD(&splice);
4077
4078         spin_lock(&fs_info->ordered_root_lock);
4079         list_splice_init(&fs_info->ordered_roots, &splice);
4080         while (!list_empty(&splice)) {
4081                 root = list_first_entry(&splice, struct btrfs_root,
4082                                         ordered_root);
4083                 list_move_tail(&root->ordered_root,
4084                                &fs_info->ordered_roots);
4085
4086                 spin_unlock(&fs_info->ordered_root_lock);
4087                 btrfs_destroy_ordered_extents(root);
4088
4089                 cond_resched();
4090                 spin_lock(&fs_info->ordered_root_lock);
4091         }
4092         spin_unlock(&fs_info->ordered_root_lock);
4093 }
4094
4095 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4096                                       struct btrfs_root *root)
4097 {
4098         struct rb_node *node;
4099         struct btrfs_delayed_ref_root *delayed_refs;
4100         struct btrfs_delayed_ref_node *ref;
4101         int ret = 0;
4102
4103         delayed_refs = &trans->delayed_refs;
4104
4105         spin_lock(&delayed_refs->lock);
4106         if (atomic_read(&delayed_refs->num_entries) == 0) {
4107                 spin_unlock(&delayed_refs->lock);
4108                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4109                 return ret;
4110         }
4111
4112         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4113                 struct btrfs_delayed_ref_head *head;
4114                 struct btrfs_delayed_ref_node *tmp;
4115                 bool pin_bytes = false;
4116
4117                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4118                                 href_node);
4119                 if (!mutex_trylock(&head->mutex)) {
4120                         atomic_inc(&head->node.refs);
4121                         spin_unlock(&delayed_refs->lock);
4122
4123                         mutex_lock(&head->mutex);
4124                         mutex_unlock(&head->mutex);
4125                         btrfs_put_delayed_ref(&head->node);
4126                         spin_lock(&delayed_refs->lock);
4127                         continue;
4128                 }
4129                 spin_lock(&head->lock);
4130                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4131                                                  list) {
4132                         ref->in_tree = 0;
4133                         list_del(&ref->list);
4134                         atomic_dec(&delayed_refs->num_entries);
4135                         btrfs_put_delayed_ref(ref);
4136                 }
4137                 if (head->must_insert_reserved)
4138                         pin_bytes = true;
4139                 btrfs_free_delayed_extent_op(head->extent_op);
4140                 delayed_refs->num_heads--;
4141                 if (head->processing == 0)
4142                         delayed_refs->num_heads_ready--;
4143                 atomic_dec(&delayed_refs->num_entries);
4144                 head->node.in_tree = 0;
4145                 rb_erase(&head->href_node, &delayed_refs->href_root);
4146                 spin_unlock(&head->lock);
4147                 spin_unlock(&delayed_refs->lock);
4148                 mutex_unlock(&head->mutex);
4149
4150                 if (pin_bytes)
4151                         btrfs_pin_extent(root, head->node.bytenr,
4152                                          head->node.num_bytes, 1);
4153                 btrfs_put_delayed_ref(&head->node);
4154                 cond_resched();
4155                 spin_lock(&delayed_refs->lock);
4156         }
4157
4158         spin_unlock(&delayed_refs->lock);
4159
4160         return ret;
4161 }
4162
4163 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4164 {
4165         struct btrfs_inode *btrfs_inode;
4166         struct list_head splice;
4167
4168         INIT_LIST_HEAD(&splice);
4169
4170         spin_lock(&root->delalloc_lock);
4171         list_splice_init(&root->delalloc_inodes, &splice);
4172
4173         while (!list_empty(&splice)) {
4174                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4175                                                delalloc_inodes);
4176
4177                 list_del_init(&btrfs_inode->delalloc_inodes);
4178                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4179                           &btrfs_inode->runtime_flags);
4180                 spin_unlock(&root->delalloc_lock);
4181
4182                 btrfs_invalidate_inodes(btrfs_inode->root);
4183
4184                 spin_lock(&root->delalloc_lock);
4185         }
4186
4187         spin_unlock(&root->delalloc_lock);
4188 }
4189
4190 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4191 {
4192         struct btrfs_root *root;
4193         struct list_head splice;
4194
4195         INIT_LIST_HEAD(&splice);
4196
4197         spin_lock(&fs_info->delalloc_root_lock);
4198         list_splice_init(&fs_info->delalloc_roots, &splice);
4199         while (!list_empty(&splice)) {
4200                 root = list_first_entry(&splice, struct btrfs_root,
4201                                          delalloc_root);
4202                 list_del_init(&root->delalloc_root);
4203                 root = btrfs_grab_fs_root(root);
4204                 BUG_ON(!root);
4205                 spin_unlock(&fs_info->delalloc_root_lock);
4206
4207                 btrfs_destroy_delalloc_inodes(root);
4208                 btrfs_put_fs_root(root);
4209
4210                 spin_lock(&fs_info->delalloc_root_lock);
4211         }
4212         spin_unlock(&fs_info->delalloc_root_lock);
4213 }
4214
4215 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4216                                         struct extent_io_tree *dirty_pages,
4217                                         int mark)
4218 {
4219         int ret;
4220         struct extent_buffer *eb;
4221         u64 start = 0;
4222         u64 end;
4223
4224         while (1) {
4225                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4226                                             mark, NULL);
4227                 if (ret)
4228                         break;
4229
4230                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4231                 while (start <= end) {
4232                         eb = btrfs_find_tree_block(root->fs_info, start);
4233                         start += root->nodesize;
4234                         if (!eb)
4235                                 continue;
4236                         wait_on_extent_buffer_writeback(eb);
4237
4238                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4239                                                &eb->bflags))
4240                                 clear_extent_buffer_dirty(eb);
4241                         free_extent_buffer_stale(eb);
4242                 }
4243         }
4244
4245         return ret;
4246 }
4247
4248 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4249                                        struct extent_io_tree *pinned_extents)
4250 {
4251         struct extent_io_tree *unpin;
4252         u64 start;
4253         u64 end;
4254         int ret;
4255         bool loop = true;
4256
4257         unpin = pinned_extents;
4258 again:
4259         while (1) {
4260                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4261                                             EXTENT_DIRTY, NULL);
4262                 if (ret)
4263                         break;
4264
4265                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4266                 btrfs_error_unpin_extent_range(root, start, end);
4267                 cond_resched();
4268         }
4269
4270         if (loop) {
4271                 if (unpin == &root->fs_info->freed_extents[0])
4272                         unpin = &root->fs_info->freed_extents[1];
4273                 else
4274                         unpin = &root->fs_info->freed_extents[0];
4275                 loop = false;
4276                 goto again;
4277         }
4278
4279         return 0;
4280 }
4281
4282 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4283                                        struct btrfs_fs_info *fs_info)
4284 {
4285         struct btrfs_ordered_extent *ordered;
4286
4287         spin_lock(&fs_info->trans_lock);
4288         while (!list_empty(&cur_trans->pending_ordered)) {
4289                 ordered = list_first_entry(&cur_trans->pending_ordered,
4290                                            struct btrfs_ordered_extent,
4291                                            trans_list);
4292                 list_del_init(&ordered->trans_list);
4293                 spin_unlock(&fs_info->trans_lock);
4294
4295                 btrfs_put_ordered_extent(ordered);
4296                 spin_lock(&fs_info->trans_lock);
4297         }
4298         spin_unlock(&fs_info->trans_lock);
4299 }
4300
4301 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4302                                    struct btrfs_root *root)
4303 {
4304         btrfs_destroy_delayed_refs(cur_trans, root);
4305
4306         cur_trans->state = TRANS_STATE_COMMIT_START;
4307         wake_up(&root->fs_info->transaction_blocked_wait);
4308
4309         cur_trans->state = TRANS_STATE_UNBLOCKED;
4310         wake_up(&root->fs_info->transaction_wait);
4311
4312         btrfs_free_pending_ordered(cur_trans, root->fs_info);
4313         btrfs_destroy_delayed_inodes(root);
4314         btrfs_assert_delayed_root_empty(root);
4315
4316         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4317                                      EXTENT_DIRTY);
4318         btrfs_destroy_pinned_extent(root,
4319                                     root->fs_info->pinned_extents);
4320
4321         cur_trans->state =TRANS_STATE_COMPLETED;
4322         wake_up(&cur_trans->commit_wait);
4323
4324         /*
4325         memset(cur_trans, 0, sizeof(*cur_trans));
4326         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4327         */
4328 }
4329
4330 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4331 {
4332         struct btrfs_transaction *t;
4333
4334         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4335
4336         spin_lock(&root->fs_info->trans_lock);
4337         while (!list_empty(&root->fs_info->trans_list)) {
4338                 t = list_first_entry(&root->fs_info->trans_list,
4339                                      struct btrfs_transaction, list);
4340                 if (t->state >= TRANS_STATE_COMMIT_START) {
4341                         atomic_inc(&t->use_count);
4342                         spin_unlock(&root->fs_info->trans_lock);
4343                         btrfs_wait_for_commit(root, t->transid);
4344                         btrfs_put_transaction(t);
4345                         spin_lock(&root->fs_info->trans_lock);
4346                         continue;
4347                 }
4348                 if (t == root->fs_info->running_transaction) {
4349                         t->state = TRANS_STATE_COMMIT_DOING;
4350                         spin_unlock(&root->fs_info->trans_lock);
4351                         /*
4352                          * We wait for 0 num_writers since we don't hold a trans
4353                          * handle open currently for this transaction.
4354                          */
4355                         wait_event(t->writer_wait,
4356                                    atomic_read(&t->num_writers) == 0);
4357                 } else {
4358                         spin_unlock(&root->fs_info->trans_lock);
4359                 }
4360                 btrfs_cleanup_one_transaction(t, root);
4361
4362                 spin_lock(&root->fs_info->trans_lock);
4363                 if (t == root->fs_info->running_transaction)
4364                         root->fs_info->running_transaction = NULL;
4365                 list_del_init(&t->list);
4366                 spin_unlock(&root->fs_info->trans_lock);
4367
4368                 btrfs_put_transaction(t);
4369                 trace_btrfs_transaction_commit(root);
4370                 spin_lock(&root->fs_info->trans_lock);
4371         }
4372         spin_unlock(&root->fs_info->trans_lock);
4373         btrfs_destroy_all_ordered_extents(root->fs_info);
4374         btrfs_destroy_delayed_inodes(root);
4375         btrfs_assert_delayed_root_empty(root);
4376         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4377         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4378         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4379
4380         return 0;
4381 }
4382
4383 static const struct extent_io_ops btree_extent_io_ops = {
4384         .readpage_end_io_hook = btree_readpage_end_io_hook,
4385         .readpage_io_failed_hook = btree_io_failed_hook,
4386         .submit_bio_hook = btree_submit_bio_hook,
4387         /* note we're sharing with inode.c for the merge bio hook */
4388         .merge_bio_hook = btrfs_merge_bio_hook,
4389 };