fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/slab.h>
  30 #include <linux/migrate.h>
  31 #include <linux/ratelimit.h>
  32 #include <linux/uuid.h>
  33 #include <linux/semaphore.h>
  34 #include <asm/unaligned.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "hash.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "volumes.h"
  41 #include "print-tree.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  64                                       struct btrfs_root *root);
  65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  67                                         struct extent_io_tree *dirty_pages,
  68                                         int mark);
  69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  70                                        struct extent_io_tree *pinned_extents);
  71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  72 static void btrfs_error_commit_super(struct btrfs_root *root);
  73
  74 /*
  75  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  76  * is complete.  This is used during reads to verify checksums, and it is used
  77  * by writes to insert metadata for new file extents after IO is complete.
  78  */
  79 struct btrfs_end_io_wq {
  80         struct bio *bio;
  81         bio_end_io_t *end_io;
  82         void *private;
  83         struct btrfs_fs_info *info;
  84         int error;
  85         enum btrfs_wq_endio_type metadata;
  86         struct list_head list;
  87         struct btrfs_work work;
  88 };
  89
  90 static struct kmem_cache *btrfs_end_io_wq_cache;
  91
  92 int __init btrfs_end_io_wq_init(void)
  93 {
  94         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  95                                         sizeof(struct btrfs_end_io_wq),
  96                                         0,
  97                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  98                                         NULL);
  99         if (!btrfs_end_io_wq_cache)
 100                 return -ENOMEM;
 101         return 0;
 102 }
 103
 104 void btrfs_end_io_wq_exit(void)
 105 {
 106         if (btrfs_end_io_wq_cache)
 107                 kmem_cache_destroy(btrfs_end_io_wq_cache);
 108 }
 109
 110 /*
 111  * async submit bios are used to offload expensive checksumming
 112  * onto the worker threads.  They checksum file and metadata bios
 113  * just before they are sent down the IO stack.
 114  */
 115 struct async_submit_bio {
 116         struct inode *inode;
 117         struct bio *bio;
 118         struct list_head list;
 119         extent_submit_bio_hook_t *submit_bio_start;
 120         extent_submit_bio_hook_t *submit_bio_done;
 121         int rw;
 122         int mirror_num;
 123         unsigned long bio_flags;
 124         /*
 125          * bio_offset is optional, can be used if the pages in the bio
 126          * can't tell us where in the file the bio should go
 127          */
 128         u64 bio_offset;
 129         struct btrfs_work work;
 130         int error;
 131 };
 132
 133 /*
 134  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 135  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 136  * the level the eb occupies in the tree.
 137  *
 138  * Different roots are used for different purposes and may nest inside each
 139  * other and they require separate keysets.  As lockdep keys should be
 140  * static, assign keysets according to the purpose of the root as indicated
 141  * by btrfs_root->objectid.  This ensures that all special purpose roots
 142  * have separate keysets.
 143  *
 144  * Lock-nesting across peer nodes is always done with the immediate parent
 145  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 146  * subclass to avoid triggering lockdep warning in such cases.
 147  *
 148  * The key is set by the readpage_end_io_hook after the buffer has passed
 149  * csum validation but before the pages are unlocked.  It is also set by
 150  * btrfs_init_new_buffer on freshly allocated blocks.
 151  *
 152  * We also add a check to make sure the highest level of the tree is the
 153  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 154  * needs update as well.
 155  */
 156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 157 # if BTRFS_MAX_LEVEL != 8
 158 #  error
 159 # endif
 160
 161 static struct btrfs_lockdep_keyset {
 162         u64                     id;             /* root objectid */
 163         const char              *name_stem;     /* lock name stem */
 164         char                    names[BTRFS_MAX_LEVEL + 1][20];
 165         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 166 } btrfs_lockdep_keysets[] = {
 167         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 168         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 169         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 170         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 171         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 172         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 173         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 174         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 175         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 176         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 177         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 178         { .id = 0,                              .name_stem = "tree"     },
 179 };
 180
 181 void __init btrfs_init_lockdep(void)
 182 {
 183         int i, j;
 184
 185         /* initialize lockdep class names */
 186         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 187                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 188
 189                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 190                         snprintf(ks->names[j], sizeof(ks->names[j]),
 191                                  "btrfs-%s-%02d", ks->name_stem, j);
 192         }
 193 }
 194
 195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 196                                     int level)
 197 {
 198         struct btrfs_lockdep_keyset *ks;
 199
 200         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 201
 202         /* find the matching keyset, id 0 is the default entry */
 203         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 204                 if (ks->id == objectid)
 205                         break;
 206
 207         lockdep_set_class_and_name(&eb->lock,
 208                                    &ks->keys[level], ks->names[level]);
 209 }
 210
 211 #endif
 212
 213 /*
 214  * extents on the btree inode are pretty simple, there's one extent
 215  * that covers the entire device
 216  */
 217 static struct extent_map *btree_get_extent(struct inode *inode,
 218                 struct page *page, size_t pg_offset, u64 start, u64 len,
 219                 int create)
 220 {
 221         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 222         struct extent_map *em;
 223         int ret;
 224
 225         read_lock(&em_tree->lock);
 226         em = lookup_extent_mapping(em_tree, start, len);
 227         if (em) {
 228                 em->bdev =
 229                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 230                 read_unlock(&em_tree->lock);
 231                 goto out;
 232         }
 233         read_unlock(&em_tree->lock);
 234
 235         em = alloc_extent_map();
 236         if (!em) {
 237                 em = ERR_PTR(-ENOMEM);
 238                 goto out;
 239         }
 240         em->start = 0;
 241         em->len = (u64)-1;
 242         em->block_len = (u64)-1;
 243         em->block_start = 0;
 244         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 245
 246         write_lock(&em_tree->lock);
 247         ret = add_extent_mapping(em_tree, em, 0);
 248         if (ret == -EEXIST) {
 249                 free_extent_map(em);
 250                 em = lookup_extent_mapping(em_tree, start, len);
 251                 if (!em)
 252                         em = ERR_PTR(-EIO);
 253         } else if (ret) {
 254                 free_extent_map(em);
 255                 em = ERR_PTR(ret);
 256         }
 257         write_unlock(&em_tree->lock);
 258
 259 out:
 260         return em;
 261 }
 262
 263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 264 {
 265         return btrfs_crc32c(seed, data, len);
 266 }
 267
 268 void btrfs_csum_final(u32 crc, char *result)
 269 {
 270         put_unaligned_le32(~crc, result);
 271 }
 272
 273 /*
 274  * compute the csum for a btree block, and either verify it or write it
 275  * into the csum field of the block.
 276  */
 277 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 278                            int verify)
 279 {
 280         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 281         char *result = NULL;
 282         unsigned long len;
 283         unsigned long cur_len;
 284         unsigned long offset = BTRFS_CSUM_SIZE;
 285         char *kaddr;
 286         unsigned long map_start;
 287         unsigned long map_len;
 288         int err;
 289         u32 crc = ~(u32)0;
 290         unsigned long inline_result;
 291
 292         len = buf->len - offset;
 293         while (len > 0) {
 294                 err = map_private_extent_buffer(buf, offset, 32,
 295                                         &kaddr, &map_start, &map_len);
 296                 if (err)
 297                         return 1;
 298                 cur_len = min(len, map_len - (offset - map_start));
 299                 crc = btrfs_csum_data(kaddr + offset - map_start,
 300                                       crc, cur_len);
 301                 len -= cur_len;
 302                 offset += cur_len;
 303         }
 304         if (csum_size > sizeof(inline_result)) {
 305                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 306                 if (!result)
 307                         return 1;
 308         } else {
 309                 result = (char *)&inline_result;
 310         }
 311
 312         btrfs_csum_final(crc, result);
 313
 314         if (verify) {
 315                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 316                         u32 val;
 317                         u32 found = 0;
 318                         memcpy(&found, result, csum_size);
 319
 320                         read_extent_buffer(buf, &val, 0, csum_size);
 321                         printk_ratelimited(KERN_INFO
 322                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 323                                 "level %d\n",
 324                                 root->fs_info->sb->s_id, buf->start,
 325                                 val, found, btrfs_header_level(buf));
 326                         if (result != (char *)&inline_result)
 327                                 kfree(result);
 328                         return 1;
 329                 }
 330         } else {
 331                 write_extent_buffer(buf, result, 0, csum_size);
 332         }
 333         if (result != (char *)&inline_result)
 334                 kfree(result);
 335         return 0;
 336 }
 337
 338 /*
 339  * we can't consider a given block up to date unless the transid of the
 340  * block matches the transid in the parent node's pointer.  This is how we
 341  * detect blocks that either didn't get written at all or got written
 342  * in the wrong place.
 343  */
 344 static int verify_parent_transid(struct extent_io_tree *io_tree,
 345                                  struct extent_buffer *eb, u64 parent_transid,
 346                                  int atomic)
 347 {
 348         struct extent_state *cached_state = NULL;
 349         int ret;
 350         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 351
 352         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 353                 return 0;
 354
 355         if (atomic)
 356                 return -EAGAIN;
 357
 358         if (need_lock) {
 359                 btrfs_tree_read_lock(eb);
 360                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 361         }
 362
 363         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 364                          0, &cached_state);
 365         if (extent_buffer_uptodate(eb) &&
 366             btrfs_header_generation(eb) == parent_transid) {
 367                 ret = 0;
 368                 goto out;
 369         }
 370         printk_ratelimited(KERN_INFO "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
 371                         eb->fs_info->sb->s_id, eb->start,
 372                         parent_transid, btrfs_header_generation(eb));
 373         ret = 1;
 374
 375         /*
 376          * Things reading via commit roots that don't have normal protection,
 377          * like send, can have a really old block in cache that may point at a
 378          * block that has been free'd and re-allocated.  So don't clear uptodate
 379          * if we find an eb that is under IO (dirty/writeback) because we could
 380          * end up reading in the stale data and then writing it back out and
 381          * making everybody very sad.
 382          */
 383         if (!extent_buffer_under_io(eb))
 384                 clear_extent_buffer_uptodate(eb);
 385 out:
 386         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 387                              &cached_state, GFP_NOFS);
 388         if (need_lock)
 389                 btrfs_tree_read_unlock_blocking(eb);
 390         return ret;
 391 }
 392
 393 /*
 394  * Return 0 if the superblock checksum type matches the checksum value of that
 395  * algorithm. Pass the raw disk superblock data.
 396  */
 397 static int btrfs_check_super_csum(char *raw_disk_sb)
 398 {
 399         struct btrfs_super_block *disk_sb =
 400                 (struct btrfs_super_block *)raw_disk_sb;
 401         u16 csum_type = btrfs_super_csum_type(disk_sb);
 402         int ret = 0;
 403
 404         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 405                 u32 crc = ~(u32)0;
 406                 const int csum_size = sizeof(crc);
 407                 char result[csum_size];
 408
 409                 /*
 410                  * The super_block structure does not span the whole
 411                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 412                  * is filled with zeros and is included in the checkum.
 413                  */
 414                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 415                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 416                 btrfs_csum_final(crc, result);
 417
 418                 if (memcmp(raw_disk_sb, result, csum_size))
 419                         ret = 1;
 420
 421                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 422                         printk(KERN_WARNING
 423                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 424                         ret = 0;
 425                 }
 426         }
 427
 428         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 429                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 430                                 csum_type);
 431                 ret = 1;
 432         }
 433
 434         return ret;
 435 }
 436
 437 /*
 438  * helper to read a given tree block, doing retries as required when
 439  * the checksums don't match and we have alternate mirrors to try.
 440  */
 441 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 442                                           struct extent_buffer *eb,
 443                                           u64 start, u64 parent_transid)
 444 {
 445         struct extent_io_tree *io_tree;
 446         int failed = 0;
 447         int ret;
 448         int num_copies = 0;
 449         int mirror_num = 0;
 450         int failed_mirror = 0;
 451
 452         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 453         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 454         while (1) {
 455                 ret = read_extent_buffer_pages(io_tree, eb, start,
 456                                                WAIT_COMPLETE,
 457                                                btree_get_extent, mirror_num);
 458                 if (!ret) {
 459                         if (!verify_parent_transid(io_tree, eb,
 460                                                    parent_transid, 0))
 461                                 break;
 462                         else
 463                                 ret = -EIO;
 464                 }
 465
 466                 /*
 467                  * This buffer's crc is fine, but its contents are corrupted, so
 468                  * there is no reason to read the other copies, they won't be
 469                  * any less wrong.
 470                  */
 471                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 472                         break;
 473
 474                 num_copies = btrfs_num_copies(root->fs_info,
 475                                               eb->start, eb->len);
 476                 if (num_copies == 1)
 477                         break;
 478
 479                 if (!failed_mirror) {
 480                         failed = 1;
 481                         failed_mirror = eb->read_mirror;
 482                 }
 483
 484                 mirror_num++;
 485                 if (mirror_num == failed_mirror)
 486                         mirror_num++;
 487
 488                 if (mirror_num > num_copies)
 489                         break;
 490         }
 491
 492         if (failed && !ret && failed_mirror)
 493                 repair_eb_io_failure(root, eb, failed_mirror);
 494
 495         return ret;
 496 }
 497
 498 /*
 499  * checksum a dirty tree block before IO.  This has extra checks to make sure
 500  * we only fill in the checksum field in the first page of a multi-page block
 501  */
 502
 503 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 504 {
 505         u64 start = page_offset(page);
 506         u64 found_start;
 507         struct extent_buffer *eb;
 508
 509         eb = (struct extent_buffer *)page->private;
 510         if (page != eb->pages[0])
 511                 return 0;
 512         found_start = btrfs_header_bytenr(eb);
 513         if (WARN_ON(found_start != start || !PageUptodate(page)))
 514                 return 0;
 515         csum_tree_block(root, eb, 0);
 516         return 0;
 517 }
 518
 519 static int check_tree_block_fsid(struct btrfs_root *root,
 520                                  struct extent_buffer *eb)
 521 {
 522         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 523         u8 fsid[BTRFS_UUID_SIZE];
 524         int ret = 1;
 525
 526         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 527         while (fs_devices) {
 528                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 529                         ret = 0;
 530                         break;
 531                 }
 532                 fs_devices = fs_devices->seed;
 533         }
 534         return ret;
 535 }
 536
 537 #define CORRUPT(reason, eb, root, slot)                         \
 538         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 539                    "root=%llu, slot=%d", reason,                        \
 540                btrfs_header_bytenr(eb), root->objectid, slot)
 541
 542 static noinline int check_leaf(struct btrfs_root *root,
 543                                struct extent_buffer *leaf)
 544 {
 545         struct btrfs_key key;
 546         struct btrfs_key leaf_key;
 547         u32 nritems = btrfs_header_nritems(leaf);
 548         int slot;
 549
 550         if (nritems == 0)
 551                 return 0;
 552
 553         /* Check the 0 item */
 554         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 555             BTRFS_LEAF_DATA_SIZE(root)) {
 556                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 557                 return -EIO;
 558         }
 559
 560         /*
 561          * Check to make sure each items keys are in the correct order and their
 562          * offsets make sense.  We only have to loop through nritems-1 because
 563          * we check the current slot against the next slot, which verifies the
 564          * next slot's offset+size makes sense and that the current's slot
 565          * offset is correct.
 566          */
 567         for (slot = 0; slot < nritems - 1; slot++) {
 568                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 569                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 570
 571                 /* Make sure the keys are in the right order */
 572                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 573                         CORRUPT("bad key order", leaf, root, slot);
 574                         return -EIO;
 575                 }
 576
 577                 /*
 578                  * Make sure the offset and ends are right, remember that the
 579                  * item data starts at the end of the leaf and grows towards the
 580                  * front.
 581                  */
 582                 if (btrfs_item_offset_nr(leaf, slot) !=
 583                         btrfs_item_end_nr(leaf, slot + 1)) {
 584                         CORRUPT("slot offset bad", leaf, root, slot);
 585                         return -EIO;
 586                 }
 587
 588                 /*
 589                  * Check to make sure that we don't point outside of the leaf,
 590                  * just incase all the items are consistent to eachother, but
 591                  * all point outside of the leaf.
 592                  */
 593                 if (btrfs_item_end_nr(leaf, slot) >
 594                     BTRFS_LEAF_DATA_SIZE(root)) {
 595                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 596                         return -EIO;
 597                 }
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 604                                       u64 phy_offset, struct page *page,
 605                                       u64 start, u64 end, int mirror)
 606 {
 607         u64 found_start;
 608         int found_level;
 609         struct extent_buffer *eb;
 610         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 611         int ret = 0;
 612         int reads_done;
 613
 614         if (!page->private)
 615                 goto out;
 616
 617         eb = (struct extent_buffer *)page->private;
 618
 619         /* the pending IO might have been the only thing that kept this buffer
 620          * in memory.  Make sure we have a ref for all this other checks
 621          */
 622         extent_buffer_get(eb);
 623
 624         reads_done = atomic_dec_and_test(&eb->io_pages);
 625         if (!reads_done)
 626                 goto err;
 627
 628         eb->read_mirror = mirror;
 629         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
 630                 ret = -EIO;
 631                 goto err;
 632         }
 633
 634         found_start = btrfs_header_bytenr(eb);
 635         if (found_start != eb->start) {
 636                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad tree block start "
 637                                "%llu %llu\n",
 638                                eb->fs_info->sb->s_id, found_start, eb->start);
 639                 ret = -EIO;
 640                 goto err;
 641         }
 642         if (check_tree_block_fsid(root, eb)) {
 643                 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad fsid on block %llu\n",
 644                                eb->fs_info->sb->s_id, eb->start);
 645                 ret = -EIO;
 646                 goto err;
 647         }
 648         found_level = btrfs_header_level(eb);
 649         if (found_level >= BTRFS_MAX_LEVEL) {
 650                 btrfs_info(root->fs_info, "bad tree block level %d",
 651                            (int)btrfs_header_level(eb));
 652                 ret = -EIO;
 653                 goto err;
 654         }
 655
 656         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 657                                        eb, found_level);
 658
 659         ret = csum_tree_block(root, eb, 1);
 660         if (ret) {
 661                 ret = -EIO;
 662                 goto err;
 663         }
 664
 665         /*
 666          * If this is a leaf block and it is corrupt, set the corrupt bit so
 667          * that we don't try and read the other copies of this block, just
 668          * return -EIO.
 669          */
 670         if (found_level == 0 && check_leaf(root, eb)) {
 671                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 672                 ret = -EIO;
 673         }
 674
 675         if (!ret)
 676                 set_extent_buffer_uptodate(eb);
 677 err:
 678         if (reads_done &&
 679             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 680                 btree_readahead_hook(root, eb, eb->start, ret);
 681
 682         if (ret) {
 683                 /*
 684                  * our io error hook is going to dec the io pages
 685                  * again, we have to make sure it has something
 686                  * to decrement
 687                  */
 688                 atomic_inc(&eb->io_pages);
 689                 clear_extent_buffer_uptodate(eb);
 690         }
 691         free_extent_buffer(eb);
 692 out:
 693         return ret;
 694 }
 695
 696 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 697 {
 698         struct extent_buffer *eb;
 699         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 700
 701         eb = (struct extent_buffer *)page->private;
 702         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
 703         eb->read_mirror = failed_mirror;
 704         atomic_dec(&eb->io_pages);
 705         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 706                 btree_readahead_hook(root, eb, eb->start, -EIO);
 707         return -EIO;    /* we fixed nothing */
 708 }
 709
 710 static void end_workqueue_bio(struct bio *bio, int err)
 711 {
 712         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 713         struct btrfs_fs_info *fs_info;
 714         struct btrfs_workqueue *wq;
 715         btrfs_work_func_t func;
 716
 717         fs_info = end_io_wq->info;
 718         end_io_wq->error = err;
 719
 720         if (bio->bi_rw & REQ_WRITE) {
 721                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 722                         wq = fs_info->endio_meta_write_workers;
 723                         func = btrfs_endio_meta_write_helper;
 724                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 725                         wq = fs_info->endio_freespace_worker;
 726                         func = btrfs_freespace_write_helper;
 727                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 728                         wq = fs_info->endio_raid56_workers;
 729                         func = btrfs_endio_raid56_helper;
 730                 } else {
 731                         wq = fs_info->endio_write_workers;
 732                         func = btrfs_endio_write_helper;
 733                 }
 734         } else {
 735                 if (unlikely(end_io_wq->metadata ==
 736                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 737                         wq = fs_info->endio_repair_workers;
 738                         func = btrfs_endio_repair_helper;
 739                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 740                         wq = fs_info->endio_raid56_workers;
 741                         func = btrfs_endio_raid56_helper;
 742                 } else if (end_io_wq->metadata) {
 743                         wq = fs_info->endio_meta_workers;
 744                         func = btrfs_endio_meta_helper;
 745                 } else {
 746                         wq = fs_info->endio_workers;
 747                         func = btrfs_endio_helper;
 748                 }
 749         }
 750
 751         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 752         btrfs_queue_work(wq, &end_io_wq->work);
 753 }
 754
 755 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 756                         enum btrfs_wq_endio_type metadata)
 757 {
 758         struct btrfs_end_io_wq *end_io_wq;
 759
 760         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 761         if (!end_io_wq)
 762                 return -ENOMEM;
 763
 764         end_io_wq->private = bio->bi_private;
 765         end_io_wq->end_io = bio->bi_end_io;
 766         end_io_wq->info = info;
 767         end_io_wq->error = 0;
 768         end_io_wq->bio = bio;
 769         end_io_wq->metadata = metadata;
 770
 771         bio->bi_private = end_io_wq;
 772         bio->bi_end_io = end_workqueue_bio;
 773         return 0;
 774 }
 775
 776 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 777 {
 778         unsigned long limit = min_t(unsigned long,
 779                                     info->thread_pool_size,
 780                                     info->fs_devices->open_devices);
 781         return 256 * limit;
 782 }
 783
 784 static void run_one_async_start(struct btrfs_work *work)
 785 {
 786         struct async_submit_bio *async;
 787         int ret;
 788
 789         async = container_of(work, struct  async_submit_bio, work);
 790         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 791                                       async->mirror_num, async->bio_flags,
 792                                       async->bio_offset);
 793         if (ret)
 794                 async->error = ret;
 795 }
 796
 797 static void run_one_async_done(struct btrfs_work *work)
 798 {
 799         struct btrfs_fs_info *fs_info;
 800         struct async_submit_bio *async;
 801         int limit;
 802
 803         async = container_of(work, struct  async_submit_bio, work);
 804         fs_info = BTRFS_I(async->inode)->root->fs_info;
 805
 806         limit = btrfs_async_submit_limit(fs_info);
 807         limit = limit * 2 / 3;
 808
 809         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 810             waitqueue_active(&fs_info->async_submit_wait))
 811                 wake_up(&fs_info->async_submit_wait);
 812
 813         /* If an error occured we just want to clean up the bio and move on */
 814         if (async->error) {
 815                 bio_endio(async->bio, async->error);
 816                 return;
 817         }
 818
 819         async->submit_bio_done(async->inode, async->rw, async->bio,
 820                                async->mirror_num, async->bio_flags,
 821                                async->bio_offset);
 822 }
 823
 824 static void run_one_async_free(struct btrfs_work *work)
 825 {
 826         struct async_submit_bio *async;
 827
 828         async = container_of(work, struct  async_submit_bio, work);
 829         kfree(async);
 830 }
 831
 832 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 833                         int rw, struct bio *bio, int mirror_num,
 834                         unsigned long bio_flags,
 835                         u64 bio_offset,
 836                         extent_submit_bio_hook_t *submit_bio_start,
 837                         extent_submit_bio_hook_t *submit_bio_done)
 838 {
 839         struct async_submit_bio *async;
 840
 841         async = kmalloc(sizeof(*async), GFP_NOFS);
 842         if (!async)
 843                 return -ENOMEM;
 844
 845         async->inode = inode;
 846         async->rw = rw;
 847         async->bio = bio;
 848         async->mirror_num = mirror_num;
 849         async->submit_bio_start = submit_bio_start;
 850         async->submit_bio_done = submit_bio_done;
 851
 852         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 853                         run_one_async_done, run_one_async_free);
 854
 855         async->bio_flags = bio_flags;
 856         async->bio_offset = bio_offset;
 857
 858         async->error = 0;
 859
 860         atomic_inc(&fs_info->nr_async_submits);
 861
 862         if (rw & REQ_SYNC)
 863                 btrfs_set_work_high_priority(&async->work);
 864
 865         btrfs_queue_work(fs_info->workers, &async->work);
 866
 867         while (atomic_read(&fs_info->async_submit_draining) &&
 868               atomic_read(&fs_info->nr_async_submits)) {
 869                 wait_event(fs_info->async_submit_wait,
 870                            (atomic_read(&fs_info->nr_async_submits) == 0));
 871         }
 872
 873         return 0;
 874 }
 875
 876 static int btree_csum_one_bio(struct bio *bio)
 877 {
 878         struct bio_vec *bvec;
 879         struct btrfs_root *root;
 880         int i, ret = 0;
 881
 882         bio_for_each_segment_all(bvec, bio, i) {
 883                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 884                 ret = csum_dirty_buffer(root, bvec->bv_page);
 885                 if (ret)
 886                         break;
 887         }
 888
 889         return ret;
 890 }
 891
 892 static int __btree_submit_bio_start(struct inode *inode, int rw,
 893                                     struct bio *bio, int mirror_num,
 894                                     unsigned long bio_flags,
 895                                     u64 bio_offset)
 896 {
 897         /*
 898          * when we're called for a write, we're already in the async
 899          * submission context.  Just jump into btrfs_map_bio
 900          */
 901         return btree_csum_one_bio(bio);
 902 }
 903
 904 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 905                                  int mirror_num, unsigned long bio_flags,
 906                                  u64 bio_offset)
 907 {
 908         int ret;
 909
 910         /*
 911          * when we're called for a write, we're already in the async
 912          * submission context.  Just jump into btrfs_map_bio
 913          */
 914         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 915         if (ret)
 916                 bio_endio(bio, ret);
 917         return ret;
 918 }
 919
 920 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 921 {
 922         if (bio_flags & EXTENT_BIO_TREE_LOG)
 923                 return 0;
 924 #ifdef CONFIG_X86
 925         if (cpu_has_xmm4_2)
 926                 return 0;
 927 #endif
 928         return 1;
 929 }
 930
 931 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 932                                  int mirror_num, unsigned long bio_flags,
 933                                  u64 bio_offset)
 934 {
 935         int async = check_async_write(inode, bio_flags);
 936         int ret;
 937
 938         if (!(rw & REQ_WRITE)) {
 939                 /*
 940                  * called for a read, do the setup so that checksum validation
 941                  * can happen in the async kernel threads
 942                  */
 943                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 944                                           bio, BTRFS_WQ_ENDIO_METADATA);
 945                 if (ret)
 946                         goto out_w_error;
 947                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 948                                     mirror_num, 0);
 949         } else if (!async) {
 950                 ret = btree_csum_one_bio(bio);
 951                 if (ret)
 952                         goto out_w_error;
 953                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 954                                     mirror_num, 0);
 955         } else {
 956                 /*
 957                  * kthread helpers are used to submit writes so that
 958                  * checksumming can happen in parallel across all CPUs
 959                  */
 960                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 961                                           inode, rw, bio, mirror_num, 0,
 962                                           bio_offset,
 963                                           __btree_submit_bio_start,
 964                                           __btree_submit_bio_done);
 965         }
 966
 967         if (ret) {
 968 out_w_error:
 969                 bio_endio(bio, ret);
 970         }
 971         return ret;
 972 }
 973
 974 #ifdef CONFIG_MIGRATION
 975 static int btree_migratepage(struct address_space *mapping,
 976                         struct page *newpage, struct page *page,
 977                         enum migrate_mode mode)
 978 {
 979         /*
 980          * we can't safely write a btree page from here,
 981          * we haven't done the locking hook
 982          */
 983         if (PageDirty(page))
 984                 return -EAGAIN;
 985         /*
 986          * Buffers may be managed in a filesystem specific way.
 987          * We must have no buffers or drop them.
 988          */
 989         if (page_has_private(page) &&
 990             !try_to_release_page(page, GFP_KERNEL))
 991                 return -EAGAIN;
 992         return migrate_page(mapping, newpage, page, mode);
 993 }
 994 #endif
 995
 996
 997 static int btree_writepages(struct address_space *mapping,
 998                             struct writeback_control *wbc)
 999 {
1000         struct btrfs_fs_info *fs_info;
1001         int ret;
1002
1003         if (wbc->sync_mode == WB_SYNC_NONE) {
1004
1005                 if (wbc->for_kupdate)
1006                         return 0;
1007
1008                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1009                 /* this is a bit racy, but that's ok */
1010                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1011                                              BTRFS_DIRTY_METADATA_THRESH);
1012                 if (ret < 0)
1013                         return 0;
1014         }
1015         return btree_write_cache_pages(mapping, wbc);
1016 }
1017
1018 static int btree_readpage(struct file *file, struct page *page)
1019 {
1020         struct extent_io_tree *tree;
1021         tree = &BTRFS_I(page->mapping->host)->io_tree;
1022         return extent_read_full_page(tree, page, btree_get_extent, 0);
1023 }
1024
1025 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1026 {
1027         if (PageWriteback(page) || PageDirty(page))
1028                 return 0;
1029
1030         return try_release_extent_buffer(page);
1031 }
1032
1033 static void btree_invalidatepage(struct page *page, unsigned int offset,
1034                                  unsigned int length)
1035 {
1036         struct extent_io_tree *tree;
1037         tree = &BTRFS_I(page->mapping->host)->io_tree;
1038         extent_invalidatepage(tree, page, offset);
1039         btree_releasepage(page, GFP_NOFS);
1040         if (PagePrivate(page)) {
1041                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1042                            "page private not zero on page %llu",
1043                            (unsigned long long)page_offset(page));
1044                 ClearPagePrivate(page);
1045                 set_page_private(page, 0);
1046                 page_cache_release(page);
1047         }
1048 }
1049
1050 static int btree_set_page_dirty(struct page *page)
1051 {
1052 #ifdef DEBUG
1053         struct extent_buffer *eb;
1054
1055         BUG_ON(!PagePrivate(page));
1056         eb = (struct extent_buffer *)page->private;
1057         BUG_ON(!eb);
1058         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1059         BUG_ON(!atomic_read(&eb->refs));
1060         btrfs_assert_tree_locked(eb);
1061 #endif
1062         return __set_page_dirty_nobuffers(page);
1063 }
1064
1065 static const struct address_space_operations btree_aops = {
1066         .readpage       = btree_readpage,
1067         .writepages     = btree_writepages,
1068         .releasepage    = btree_releasepage,
1069         .invalidatepage = btree_invalidatepage,
1070 #ifdef CONFIG_MIGRATION
1071         .migratepage    = btree_migratepage,
1072 #endif
1073         .set_page_dirty = btree_set_page_dirty,
1074 };
1075
1076 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077                          u64 parent_transid)
1078 {
1079         struct extent_buffer *buf = NULL;
1080         struct inode *btree_inode = root->fs_info->btree_inode;
1081         int ret = 0;
1082
1083         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1084         if (!buf)
1085                 return 0;
1086         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1087                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1088         free_extent_buffer(buf);
1089         return ret;
1090 }
1091
1092 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1093                          int mirror_num, struct extent_buffer **eb)
1094 {
1095         struct extent_buffer *buf = NULL;
1096         struct inode *btree_inode = root->fs_info->btree_inode;
1097         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1098         int ret;
1099
1100         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1101         if (!buf)
1102                 return 0;
1103
1104         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1105
1106         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1107                                        btree_get_extent, mirror_num);
1108         if (ret) {
1109                 free_extent_buffer(buf);
1110                 return ret;
1111         }
1112
1113         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1114                 free_extent_buffer(buf);
1115                 return -EIO;
1116         } else if (extent_buffer_uptodate(buf)) {
1117                 *eb = buf;
1118         } else {
1119                 free_extent_buffer(buf);
1120         }
1121         return 0;
1122 }
1123
1124 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1125                                             u64 bytenr, u32 blocksize)
1126 {
1127         return find_extent_buffer(root->fs_info, bytenr);
1128 }
1129
1130 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1131                                                  u64 bytenr, u32 blocksize)
1132 {
1133 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1134         if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1135                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1136                                                 blocksize);
1137 #endif
1138         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1139 }
1140
1141
1142 int btrfs_write_tree_block(struct extent_buffer *buf)
1143 {
1144         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1145                                         buf->start + buf->len - 1);
1146 }
1147
1148 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1149 {
1150         return filemap_fdatawait_range(buf->pages[0]->mapping,
1151                                        buf->start, buf->start + buf->len - 1);
1152 }
1153
1154 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1155                                       u32 blocksize, u64 parent_transid)
1156 {
1157         struct extent_buffer *buf = NULL;
1158         int ret;
1159
1160         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1161         if (!buf)
1162                 return NULL;
1163
1164         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1165         if (ret) {
1166                 free_extent_buffer(buf);
1167                 return NULL;
1168         }
1169         return buf;
1170
1171 }
1172
1173 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1174                       struct extent_buffer *buf)
1175 {
1176         struct btrfs_fs_info *fs_info = root->fs_info;
1177
1178         if (btrfs_header_generation(buf) ==
1179             fs_info->running_transaction->transid) {
1180                 btrfs_assert_tree_locked(buf);
1181
1182                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1183                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1184                                              -buf->len,
1185                                              fs_info->dirty_metadata_batch);
1186                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1187                         btrfs_set_lock_blocking(buf);
1188                         clear_extent_buffer_dirty(buf);
1189                 }
1190         }
1191 }
1192
1193 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1194 {
1195         struct btrfs_subvolume_writers *writers;
1196         int ret;
1197
1198         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1199         if (!writers)
1200                 return ERR_PTR(-ENOMEM);
1201
1202         ret = percpu_counter_init(&writers->counter, 0);
1203         if (ret < 0) {
1204                 kfree(writers);
1205                 return ERR_PTR(ret);
1206         }
1207
1208         init_waitqueue_head(&writers->wait);
1209         return writers;
1210 }
1211
1212 static void
1213 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1214 {
1215         percpu_counter_destroy(&writers->counter);
1216         kfree(writers);
1217 }
1218
1219 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1220                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1221                          u64 objectid)
1222 {
1223         root->node = NULL;
1224         root->commit_root = NULL;
1225         root->sectorsize = sectorsize;
1226         root->nodesize = nodesize;
1227         root->stripesize = stripesize;
1228         root->state = 0;
1229         root->orphan_cleanup_state = 0;
1230
1231         root->objectid = objectid;
1232         root->last_trans = 0;
1233         root->highest_objectid = 0;
1234         root->nr_delalloc_inodes = 0;
1235         root->nr_ordered_extents = 0;
1236         root->name = NULL;
1237         root->inode_tree = RB_ROOT;
1238         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1239         root->block_rsv = NULL;
1240         root->orphan_block_rsv = NULL;
1241
1242         INIT_LIST_HEAD(&root->dirty_list);
1243         INIT_LIST_HEAD(&root->root_list);
1244         INIT_LIST_HEAD(&root->delalloc_inodes);
1245         INIT_LIST_HEAD(&root->delalloc_root);
1246         INIT_LIST_HEAD(&root->ordered_extents);
1247         INIT_LIST_HEAD(&root->ordered_root);
1248         INIT_LIST_HEAD(&root->logged_list[0]);
1249         INIT_LIST_HEAD(&root->logged_list[1]);
1250         spin_lock_init(&root->orphan_lock);
1251         spin_lock_init(&root->inode_lock);
1252         spin_lock_init(&root->delalloc_lock);
1253         spin_lock_init(&root->ordered_extent_lock);
1254         spin_lock_init(&root->accounting_lock);
1255         spin_lock_init(&root->log_extents_lock[0]);
1256         spin_lock_init(&root->log_extents_lock[1]);
1257         mutex_init(&root->objectid_mutex);
1258         mutex_init(&root->log_mutex);
1259         mutex_init(&root->ordered_extent_mutex);
1260         mutex_init(&root->delalloc_mutex);
1261         init_waitqueue_head(&root->log_writer_wait);
1262         init_waitqueue_head(&root->log_commit_wait[0]);
1263         init_waitqueue_head(&root->log_commit_wait[1]);
1264         INIT_LIST_HEAD(&root->log_ctxs[0]);
1265         INIT_LIST_HEAD(&root->log_ctxs[1]);
1266         atomic_set(&root->log_commit[0], 0);
1267         atomic_set(&root->log_commit[1], 0);
1268         atomic_set(&root->log_writers, 0);
1269         atomic_set(&root->log_batch, 0);
1270         atomic_set(&root->orphan_inodes, 0);
1271         atomic_set(&root->refs, 1);
1272         atomic_set(&root->will_be_snapshoted, 0);
1273         root->log_transid = 0;
1274         root->log_transid_committed = -1;
1275         root->last_log_commit = 0;
1276         if (fs_info)
1277                 extent_io_tree_init(&root->dirty_log_pages,
1278                                      fs_info->btree_inode->i_mapping);
1279
1280         memset(&root->root_key, 0, sizeof(root->root_key));
1281         memset(&root->root_item, 0, sizeof(root->root_item));
1282         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1283         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1284         if (fs_info)
1285                 root->defrag_trans_start = fs_info->generation;
1286         else
1287                 root->defrag_trans_start = 0;
1288         init_completion(&root->kobj_unregister);
1289         root->root_key.objectid = objectid;
1290         root->anon_dev = 0;
1291
1292         spin_lock_init(&root->root_item_lock);
1293 }
1294
1295 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1296 {
1297         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1298         if (root)
1299                 root->fs_info = fs_info;
1300         return root;
1301 }
1302
1303 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1304 /* Should only be used by the testing infrastructure */
1305 struct btrfs_root *btrfs_alloc_dummy_root(void)
1306 {
1307         struct btrfs_root *root;
1308
1309         root = btrfs_alloc_root(NULL);
1310         if (!root)
1311                 return ERR_PTR(-ENOMEM);
1312         __setup_root(4096, 4096, 4096, root, NULL, 1);
1313         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1314         root->alloc_bytenr = 0;
1315
1316         return root;
1317 }
1318 #endif
1319
1320 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1321                                      struct btrfs_fs_info *fs_info,
1322                                      u64 objectid)
1323 {
1324         struct extent_buffer *leaf;
1325         struct btrfs_root *tree_root = fs_info->tree_root;
1326         struct btrfs_root *root;
1327         struct btrfs_key key;
1328         int ret = 0;
1329         uuid_le uuid;
1330
1331         root = btrfs_alloc_root(fs_info);
1332         if (!root)
1333                 return ERR_PTR(-ENOMEM);
1334
1335         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1336                 tree_root->stripesize, root, fs_info, objectid);
1337         root->root_key.objectid = objectid;
1338         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1339         root->root_key.offset = 0;
1340
1341         leaf = btrfs_alloc_free_block(trans, root, root->nodesize,
1342                                       0, objectid, NULL, 0, 0, 0);
1343         if (IS_ERR(leaf)) {
1344                 ret = PTR_ERR(leaf);
1345                 leaf = NULL;
1346                 goto fail;
1347         }
1348
1349         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1350         btrfs_set_header_bytenr(leaf, leaf->start);
1351         btrfs_set_header_generation(leaf, trans->transid);
1352         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1353         btrfs_set_header_owner(leaf, objectid);
1354         root->node = leaf;
1355
1356         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1357                             BTRFS_FSID_SIZE);
1358         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1359                             btrfs_header_chunk_tree_uuid(leaf),
1360                             BTRFS_UUID_SIZE);
1361         btrfs_mark_buffer_dirty(leaf);
1362
1363         root->commit_root = btrfs_root_node(root);
1364         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1365
1366         root->root_item.flags = 0;
1367         root->root_item.byte_limit = 0;
1368         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1369         btrfs_set_root_generation(&root->root_item, trans->transid);
1370         btrfs_set_root_level(&root->root_item, 0);
1371         btrfs_set_root_refs(&root->root_item, 1);
1372         btrfs_set_root_used(&root->root_item, leaf->len);
1373         btrfs_set_root_last_snapshot(&root->root_item, 0);
1374         btrfs_set_root_dirid(&root->root_item, 0);
1375         uuid_le_gen(&uuid);
1376         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1377         root->root_item.drop_level = 0;
1378
1379         key.objectid = objectid;
1380         key.type = BTRFS_ROOT_ITEM_KEY;
1381         key.offset = 0;
1382         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1383         if (ret)
1384                 goto fail;
1385
1386         btrfs_tree_unlock(leaf);
1387
1388         return root;
1389
1390 fail:
1391         if (leaf) {
1392                 btrfs_tree_unlock(leaf);
1393                 free_extent_buffer(root->commit_root);
1394                 free_extent_buffer(leaf);
1395         }
1396         kfree(root);
1397
1398         return ERR_PTR(ret);
1399 }
1400
1401 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1402                                          struct btrfs_fs_info *fs_info)
1403 {
1404         struct btrfs_root *root;
1405         struct btrfs_root *tree_root = fs_info->tree_root;
1406         struct extent_buffer *leaf;
1407
1408         root = btrfs_alloc_root(fs_info);
1409         if (!root)
1410                 return ERR_PTR(-ENOMEM);
1411
1412         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1413                      tree_root->stripesize, root, fs_info,
1414                      BTRFS_TREE_LOG_OBJECTID);
1415
1416         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1417         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1418         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1419
1420         /*
1421          * DON'T set REF_COWS for log trees
1422          *
1423          * log trees do not get reference counted because they go away
1424          * before a real commit is actually done.  They do store pointers
1425          * to file data extents, and those reference counts still get
1426          * updated (along with back refs to the log tree).
1427          */
1428
1429         leaf = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1430                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1431                                       0, 0, 0);
1432         if (IS_ERR(leaf)) {
1433                 kfree(root);
1434                 return ERR_CAST(leaf);
1435         }
1436
1437         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1438         btrfs_set_header_bytenr(leaf, leaf->start);
1439         btrfs_set_header_generation(leaf, trans->transid);
1440         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1441         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1442         root->node = leaf;
1443
1444         write_extent_buffer(root->node, root->fs_info->fsid,
1445                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1446         btrfs_mark_buffer_dirty(root->node);
1447         btrfs_tree_unlock(root->node);
1448         return root;
1449 }
1450
1451 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1452                              struct btrfs_fs_info *fs_info)
1453 {
1454         struct btrfs_root *log_root;
1455
1456         log_root = alloc_log_tree(trans, fs_info);
1457         if (IS_ERR(log_root))
1458                 return PTR_ERR(log_root);
1459         WARN_ON(fs_info->log_root_tree);
1460         fs_info->log_root_tree = log_root;
1461         return 0;
1462 }
1463
1464 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1465                        struct btrfs_root *root)
1466 {
1467         struct btrfs_root *log_root;
1468         struct btrfs_inode_item *inode_item;
1469
1470         log_root = alloc_log_tree(trans, root->fs_info);
1471         if (IS_ERR(log_root))
1472                 return PTR_ERR(log_root);
1473
1474         log_root->last_trans = trans->transid;
1475         log_root->root_key.offset = root->root_key.objectid;
1476
1477         inode_item = &log_root->root_item.inode;
1478         btrfs_set_stack_inode_generation(inode_item, 1);
1479         btrfs_set_stack_inode_size(inode_item, 3);
1480         btrfs_set_stack_inode_nlink(inode_item, 1);
1481         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1482         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1483
1484         btrfs_set_root_node(&log_root->root_item, log_root->node);
1485
1486         WARN_ON(root->log_root);
1487         root->log_root = log_root;
1488         root->log_transid = 0;
1489         root->log_transid_committed = -1;
1490         root->last_log_commit = 0;
1491         return 0;
1492 }
1493
1494 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1495                                                struct btrfs_key *key)
1496 {
1497         struct btrfs_root *root;
1498         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1499         struct btrfs_path *path;
1500         u64 generation;
1501         u32 blocksize;
1502         int ret;
1503
1504         path = btrfs_alloc_path();
1505         if (!path)
1506                 return ERR_PTR(-ENOMEM);
1507
1508         root = btrfs_alloc_root(fs_info);
1509         if (!root) {
1510                 ret = -ENOMEM;
1511                 goto alloc_fail;
1512         }
1513
1514         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1515                 tree_root->stripesize, root, fs_info, key->objectid);
1516
1517         ret = btrfs_find_root(tree_root, key, path,
1518                               &root->root_item, &root->root_key);
1519         if (ret) {
1520                 if (ret > 0)
1521                         ret = -ENOENT;
1522                 goto find_fail;
1523         }
1524
1525         generation = btrfs_root_generation(&root->root_item);
1526         blocksize = root->nodesize;
1527         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1528                                      blocksize, generation);
1529         if (!root->node) {
1530                 ret = -ENOMEM;
1531                 goto find_fail;
1532         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1533                 ret = -EIO;
1534                 goto read_fail;
1535         }
1536         root->commit_root = btrfs_root_node(root);
1537 out:
1538         btrfs_free_path(path);
1539         return root;
1540
1541 read_fail:
1542         free_extent_buffer(root->node);
1543 find_fail:
1544         kfree(root);
1545 alloc_fail:
1546         root = ERR_PTR(ret);
1547         goto out;
1548 }
1549
1550 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1551                                       struct btrfs_key *location)
1552 {
1553         struct btrfs_root *root;
1554
1555         root = btrfs_read_tree_root(tree_root, location);
1556         if (IS_ERR(root))
1557                 return root;
1558
1559         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1560                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1561                 btrfs_check_and_init_root_item(&root->root_item);
1562         }
1563
1564         return root;
1565 }
1566
1567 int btrfs_init_fs_root(struct btrfs_root *root)
1568 {
1569         int ret;
1570         struct btrfs_subvolume_writers *writers;
1571
1572         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1573         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1574                                         GFP_NOFS);
1575         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1576                 ret = -ENOMEM;
1577                 goto fail;
1578         }
1579
1580         writers = btrfs_alloc_subvolume_writers();
1581         if (IS_ERR(writers)) {
1582                 ret = PTR_ERR(writers);
1583                 goto fail;
1584         }
1585         root->subv_writers = writers;
1586
1587         btrfs_init_free_ino_ctl(root);
1588         spin_lock_init(&root->ino_cache_lock);
1589         init_waitqueue_head(&root->ino_cache_wait);
1590
1591         ret = get_anon_bdev(&root->anon_dev);
1592         if (ret)
1593                 goto free_writers;
1594         return 0;
1595
1596 free_writers:
1597         btrfs_free_subvolume_writers(root->subv_writers);
1598 fail:
1599         kfree(root->free_ino_ctl);
1600         kfree(root->free_ino_pinned);
1601         return ret;
1602 }
1603
1604 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1605                                                u64 root_id)
1606 {
1607         struct btrfs_root *root;
1608
1609         spin_lock(&fs_info->fs_roots_radix_lock);
1610         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1611                                  (unsigned long)root_id);
1612         spin_unlock(&fs_info->fs_roots_radix_lock);
1613         return root;
1614 }
1615
1616 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1617                          struct btrfs_root *root)
1618 {
1619         int ret;
1620
1621         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1622         if (ret)
1623                 return ret;
1624
1625         spin_lock(&fs_info->fs_roots_radix_lock);
1626         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1627                                 (unsigned long)root->root_key.objectid,
1628                                 root);
1629         if (ret == 0)
1630                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1631         spin_unlock(&fs_info->fs_roots_radix_lock);
1632         radix_tree_preload_end();
1633
1634         return ret;
1635 }
1636
1637 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1638                                      struct btrfs_key *location,
1639                                      bool check_ref)
1640 {
1641         struct btrfs_root *root;
1642         int ret;
1643
1644         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1645                 return fs_info->tree_root;
1646         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1647                 return fs_info->extent_root;
1648         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1649                 return fs_info->chunk_root;
1650         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1651                 return fs_info->dev_root;
1652         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1653                 return fs_info->csum_root;
1654         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1655                 return fs_info->quota_root ? fs_info->quota_root :
1656                                              ERR_PTR(-ENOENT);
1657         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1658                 return fs_info->uuid_root ? fs_info->uuid_root :
1659                                             ERR_PTR(-ENOENT);
1660 again:
1661         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1662         if (root) {
1663                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1664                         return ERR_PTR(-ENOENT);
1665                 return root;
1666         }
1667
1668         root = btrfs_read_fs_root(fs_info->tree_root, location);
1669         if (IS_ERR(root))
1670                 return root;
1671
1672         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1673                 ret = -ENOENT;
1674                 goto fail;
1675         }
1676
1677         ret = btrfs_init_fs_root(root);
1678         if (ret)
1679                 goto fail;
1680
1681         ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1682                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1683         if (ret < 0)
1684                 goto fail;
1685         if (ret == 0)
1686                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1687
1688         ret = btrfs_insert_fs_root(fs_info, root);
1689         if (ret) {
1690                 if (ret == -EEXIST) {
1691                         free_fs_root(root);
1692                         goto again;
1693                 }
1694                 goto fail;
1695         }
1696         return root;
1697 fail:
1698         free_fs_root(root);
1699         return ERR_PTR(ret);
1700 }
1701
1702 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1703 {
1704         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1705         int ret = 0;
1706         struct btrfs_device *device;
1707         struct backing_dev_info *bdi;
1708
1709         rcu_read_lock();
1710         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1711                 if (!device->bdev)
1712                         continue;
1713                 bdi = blk_get_backing_dev_info(device->bdev);
1714                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1715                         ret = 1;
1716                         break;
1717                 }
1718         }
1719         rcu_read_unlock();
1720         return ret;
1721 }
1722
1723 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1724 {
1725         int err;
1726
1727         bdi->capabilities = BDI_CAP_MAP_COPY;
1728         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1729         if (err)
1730                 return err;
1731
1732         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1733         bdi->congested_fn       = btrfs_congested_fn;
1734         bdi->congested_data     = info;
1735         return 0;
1736 }
1737
1738 /*
1739  * called by the kthread helper functions to finally call the bio end_io
1740  * functions.  This is where read checksum verification actually happens
1741  */
1742 static void end_workqueue_fn(struct btrfs_work *work)
1743 {
1744         struct bio *bio;
1745         struct btrfs_end_io_wq *end_io_wq;
1746         int error;
1747
1748         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1749         bio = end_io_wq->bio;
1750
1751         error = end_io_wq->error;
1752         bio->bi_private = end_io_wq->private;
1753         bio->bi_end_io = end_io_wq->end_io;
1754         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1755         bio_endio_nodec(bio, error);
1756 }
1757
1758 static int cleaner_kthread(void *arg)
1759 {
1760         struct btrfs_root *root = arg;
1761         int again;
1762
1763         do {
1764                 again = 0;
1765
1766                 /* Make the cleaner go to sleep early. */
1767                 if (btrfs_need_cleaner_sleep(root))
1768                         goto sleep;
1769
1770                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1771                         goto sleep;
1772
1773                 /*
1774                  * Avoid the problem that we change the status of the fs
1775                  * during the above check and trylock.
1776                  */
1777                 if (btrfs_need_cleaner_sleep(root)) {
1778                         mutex_unlock(&root->fs_info->cleaner_mutex);
1779                         goto sleep;
1780                 }
1781
1782                 btrfs_run_delayed_iputs(root);
1783                 btrfs_delete_unused_bgs(root->fs_info);
1784                 again = btrfs_clean_one_deleted_snapshot(root);
1785                 mutex_unlock(&root->fs_info->cleaner_mutex);
1786
1787                 /*
1788                  * The defragger has dealt with the R/O remount and umount,
1789                  * needn't do anything special here.
1790                  */
1791                 btrfs_run_defrag_inodes(root->fs_info);
1792 sleep:
1793                 if (!try_to_freeze() && !again) {
1794                         set_current_state(TASK_INTERRUPTIBLE);
1795                         if (!kthread_should_stop())
1796                                 schedule();
1797                         __set_current_state(TASK_RUNNING);
1798                 }
1799         } while (!kthread_should_stop());
1800         return 0;
1801 }
1802
1803 static int transaction_kthread(void *arg)
1804 {
1805         struct btrfs_root *root = arg;
1806         struct btrfs_trans_handle *trans;
1807         struct btrfs_transaction *cur;
1808         u64 transid;
1809         unsigned long now;
1810         unsigned long delay;
1811         bool cannot_commit;
1812
1813         do {
1814                 cannot_commit = false;
1815                 delay = HZ * root->fs_info->commit_interval;
1816                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1817
1818                 spin_lock(&root->fs_info->trans_lock);
1819                 cur = root->fs_info->running_transaction;
1820                 if (!cur) {
1821                         spin_unlock(&root->fs_info->trans_lock);
1822                         goto sleep;
1823                 }
1824
1825                 now = get_seconds();
1826                 if (cur->state < TRANS_STATE_BLOCKED &&
1827                     (now < cur->start_time ||
1828                      now - cur->start_time < root->fs_info->commit_interval)) {
1829                         spin_unlock(&root->fs_info->trans_lock);
1830                         delay = HZ * 5;
1831                         goto sleep;
1832                 }
1833                 transid = cur->transid;
1834                 spin_unlock(&root->fs_info->trans_lock);
1835
1836                 /* If the file system is aborted, this will always fail. */
1837                 trans = btrfs_attach_transaction(root);
1838                 if (IS_ERR(trans)) {
1839                         if (PTR_ERR(trans) != -ENOENT)
1840                                 cannot_commit = true;
1841                         goto sleep;
1842                 }
1843                 if (transid == trans->transid) {
1844                         btrfs_commit_transaction(trans, root);
1845                 } else {
1846                         btrfs_end_transaction(trans, root);
1847                 }
1848 sleep:
1849                 wake_up_process(root->fs_info->cleaner_kthread);
1850                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1851
1852                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853                                       &root->fs_info->fs_state)))
1854                         btrfs_cleanup_transaction(root);
1855                 if (!try_to_freeze()) {
1856                         set_current_state(TASK_INTERRUPTIBLE);
1857                         if (!kthread_should_stop() &&
1858                             (!btrfs_transaction_blocked(root->fs_info) ||
1859                              cannot_commit))
1860                                 schedule_timeout(delay);
1861                         __set_current_state(TASK_RUNNING);
1862                 }
1863         } while (!kthread_should_stop());
1864         return 0;
1865 }
1866
1867 /*
1868  * this will find the highest generation in the array of
1869  * root backups.  The index of the highest array is returned,
1870  * or -1 if we can't find anything.
1871  *
1872  * We check to make sure the array is valid by comparing the
1873  * generation of the latest  root in the array with the generation
1874  * in the super block.  If they don't match we pitch it.
1875  */
1876 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1877 {
1878         u64 cur;
1879         int newest_index = -1;
1880         struct btrfs_root_backup *root_backup;
1881         int i;
1882
1883         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1884                 root_backup = info->super_copy->super_roots + i;
1885                 cur = btrfs_backup_tree_root_gen(root_backup);
1886                 if (cur == newest_gen)
1887                         newest_index = i;
1888         }
1889
1890         /* check to see if we actually wrapped around */
1891         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1892                 root_backup = info->super_copy->super_roots;
1893                 cur = btrfs_backup_tree_root_gen(root_backup);
1894                 if (cur == newest_gen)
1895                         newest_index = 0;
1896         }
1897         return newest_index;
1898 }
1899
1900
1901 /*
1902  * find the oldest backup so we know where to store new entries
1903  * in the backup array.  This will set the backup_root_index
1904  * field in the fs_info struct
1905  */
1906 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1907                                      u64 newest_gen)
1908 {
1909         int newest_index = -1;
1910
1911         newest_index = find_newest_super_backup(info, newest_gen);
1912         /* if there was garbage in there, just move along */
1913         if (newest_index == -1) {
1914                 info->backup_root_index = 0;
1915         } else {
1916                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1917         }
1918 }
1919
1920 /*
1921  * copy all the root pointers into the super backup array.
1922  * this will bump the backup pointer by one when it is
1923  * done
1924  */
1925 static void backup_super_roots(struct btrfs_fs_info *info)
1926 {
1927         int next_backup;
1928         struct btrfs_root_backup *root_backup;
1929         int last_backup;
1930
1931         next_backup = info->backup_root_index;
1932         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1933                 BTRFS_NUM_BACKUP_ROOTS;
1934
1935         /*
1936          * just overwrite the last backup if we're at the same generation
1937          * this happens only at umount
1938          */
1939         root_backup = info->super_for_commit->super_roots + last_backup;
1940         if (btrfs_backup_tree_root_gen(root_backup) ==
1941             btrfs_header_generation(info->tree_root->node))
1942                 next_backup = last_backup;
1943
1944         root_backup = info->super_for_commit->super_roots + next_backup;
1945
1946         /*
1947          * make sure all of our padding and empty slots get zero filled
1948          * regardless of which ones we use today
1949          */
1950         memset(root_backup, 0, sizeof(*root_backup));
1951
1952         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1953
1954         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1955         btrfs_set_backup_tree_root_gen(root_backup,
1956                                btrfs_header_generation(info->tree_root->node));
1957
1958         btrfs_set_backup_tree_root_level(root_backup,
1959                                btrfs_header_level(info->tree_root->node));
1960
1961         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1962         btrfs_set_backup_chunk_root_gen(root_backup,
1963                                btrfs_header_generation(info->chunk_root->node));
1964         btrfs_set_backup_chunk_root_level(root_backup,
1965                                btrfs_header_level(info->chunk_root->node));
1966
1967         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1968         btrfs_set_backup_extent_root_gen(root_backup,
1969                                btrfs_header_generation(info->extent_root->node));
1970         btrfs_set_backup_extent_root_level(root_backup,
1971                                btrfs_header_level(info->extent_root->node));
1972
1973         /*
1974          * we might commit during log recovery, which happens before we set
1975          * the fs_root.  Make sure it is valid before we fill it in.
1976          */
1977         if (info->fs_root && info->fs_root->node) {
1978                 btrfs_set_backup_fs_root(root_backup,
1979                                          info->fs_root->node->start);
1980                 btrfs_set_backup_fs_root_gen(root_backup,
1981                                btrfs_header_generation(info->fs_root->node));
1982                 btrfs_set_backup_fs_root_level(root_backup,
1983                                btrfs_header_level(info->fs_root->node));
1984         }
1985
1986         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1987         btrfs_set_backup_dev_root_gen(root_backup,
1988                                btrfs_header_generation(info->dev_root->node));
1989         btrfs_set_backup_dev_root_level(root_backup,
1990                                        btrfs_header_level(info->dev_root->node));
1991
1992         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1993         btrfs_set_backup_csum_root_gen(root_backup,
1994                                btrfs_header_generation(info->csum_root->node));
1995         btrfs_set_backup_csum_root_level(root_backup,
1996                                btrfs_header_level(info->csum_root->node));
1997
1998         btrfs_set_backup_total_bytes(root_backup,
1999                              btrfs_super_total_bytes(info->super_copy));
2000         btrfs_set_backup_bytes_used(root_backup,
2001                              btrfs_super_bytes_used(info->super_copy));
2002         btrfs_set_backup_num_devices(root_backup,
2003                              btrfs_super_num_devices(info->super_copy));
2004
2005         /*
2006          * if we don't copy this out to the super_copy, it won't get remembered
2007          * for the next commit
2008          */
2009         memcpy(&info->super_copy->super_roots,
2010                &info->super_for_commit->super_roots,
2011                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2012 }
2013
2014 /*
2015  * this copies info out of the root backup array and back into
2016  * the in-memory super block.  It is meant to help iterate through
2017  * the array, so you send it the number of backups you've already
2018  * tried and the last backup index you used.
2019  *
2020  * this returns -1 when it has tried all the backups
2021  */
2022 static noinline int next_root_backup(struct btrfs_fs_info *info,
2023                                      struct btrfs_super_block *super,
2024                                      int *num_backups_tried, int *backup_index)
2025 {
2026         struct btrfs_root_backup *root_backup;
2027         int newest = *backup_index;
2028
2029         if (*num_backups_tried == 0) {
2030                 u64 gen = btrfs_super_generation(super);
2031
2032                 newest = find_newest_super_backup(info, gen);
2033                 if (newest == -1)
2034                         return -1;
2035
2036                 *backup_index = newest;
2037                 *num_backups_tried = 1;
2038         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2039                 /* we've tried all the backups, all done */
2040                 return -1;
2041         } else {
2042                 /* jump to the next oldest backup */
2043                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2044                         BTRFS_NUM_BACKUP_ROOTS;
2045                 *backup_index = newest;
2046                 *num_backups_tried += 1;
2047         }
2048         root_backup = super->super_roots + newest;
2049
2050         btrfs_set_super_generation(super,
2051                                    btrfs_backup_tree_root_gen(root_backup));
2052         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2053         btrfs_set_super_root_level(super,
2054                                    btrfs_backup_tree_root_level(root_backup));
2055         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2056
2057         /*
2058          * fixme: the total bytes and num_devices need to match or we should
2059          * need a fsck
2060          */
2061         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2062         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2063         return 0;
2064 }
2065
2066 /* helper to cleanup workers */
2067 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2068 {
2069         btrfs_destroy_workqueue(fs_info->fixup_workers);
2070         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2071         btrfs_destroy_workqueue(fs_info->workers);
2072         btrfs_destroy_workqueue(fs_info->endio_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2074         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2075         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2076         btrfs_destroy_workqueue(fs_info->rmw_workers);
2077         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2078         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2079         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2080         btrfs_destroy_workqueue(fs_info->submit_workers);
2081         btrfs_destroy_workqueue(fs_info->delayed_workers);
2082         btrfs_destroy_workqueue(fs_info->caching_workers);
2083         btrfs_destroy_workqueue(fs_info->readahead_workers);
2084         btrfs_destroy_workqueue(fs_info->flush_workers);
2085         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2086         btrfs_destroy_workqueue(fs_info->extent_workers);
2087 }
2088
2089 static void free_root_extent_buffers(struct btrfs_root *root)
2090 {
2091         if (root) {
2092                 free_extent_buffer(root->node);
2093                 free_extent_buffer(root->commit_root);
2094                 root->node = NULL;
2095                 root->commit_root = NULL;
2096         }
2097 }
2098
2099 /* helper to cleanup tree roots */
2100 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2101 {
2102         free_root_extent_buffers(info->tree_root);
2103
2104         free_root_extent_buffers(info->dev_root);
2105         free_root_extent_buffers(info->extent_root);
2106         free_root_extent_buffers(info->csum_root);
2107         free_root_extent_buffers(info->quota_root);
2108         free_root_extent_buffers(info->uuid_root);
2109         if (chunk_root)
2110                 free_root_extent_buffers(info->chunk_root);
2111 }
2112
2113 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2114 {
2115         int ret;
2116         struct btrfs_root *gang[8];
2117         int i;
2118
2119         while (!list_empty(&fs_info->dead_roots)) {
2120                 gang[0] = list_entry(fs_info->dead_roots.next,
2121                                      struct btrfs_root, root_list);
2122                 list_del(&gang[0]->root_list);
2123
2124                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2125                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2126                 } else {
2127                         free_extent_buffer(gang[0]->node);
2128                         free_extent_buffer(gang[0]->commit_root);
2129                         btrfs_put_fs_root(gang[0]);
2130                 }
2131         }
2132
2133         while (1) {
2134                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2135                                              (void **)gang, 0,
2136                                              ARRAY_SIZE(gang));
2137                 if (!ret)
2138                         break;
2139                 for (i = 0; i < ret; i++)
2140                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2141         }
2142
2143         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2144                 btrfs_free_log_root_tree(NULL, fs_info);
2145                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2146                                             fs_info->pinned_extents);
2147         }
2148 }
2149
2150 int open_ctree(struct super_block *sb,
2151                struct btrfs_fs_devices *fs_devices,
2152                char *options)
2153 {
2154         u32 sectorsize;
2155         u32 nodesize;
2156         u32 blocksize;
2157         u32 stripesize;
2158         u64 generation;
2159         u64 features;
2160         struct btrfs_key location;
2161         struct buffer_head *bh;
2162         struct btrfs_super_block *disk_super;
2163         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2164         struct btrfs_root *tree_root;
2165         struct btrfs_root *extent_root;
2166         struct btrfs_root *csum_root;
2167         struct btrfs_root *chunk_root;
2168         struct btrfs_root *dev_root;
2169         struct btrfs_root *quota_root;
2170         struct btrfs_root *uuid_root;
2171         struct btrfs_root *log_tree_root;
2172         int ret;
2173         int err = -EINVAL;
2174         int num_backups_tried = 0;
2175         int backup_index = 0;
2176         int max_active;
2177         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2178         bool create_uuid_tree;
2179         bool check_uuid_tree;
2180
2181         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2182         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2183         if (!tree_root || !chunk_root) {
2184                 err = -ENOMEM;
2185                 goto fail;
2186         }
2187
2188         ret = init_srcu_struct(&fs_info->subvol_srcu);
2189         if (ret) {
2190                 err = ret;
2191                 goto fail;
2192         }
2193
2194         ret = setup_bdi(fs_info, &fs_info->bdi);
2195         if (ret) {
2196                 err = ret;
2197                 goto fail_srcu;
2198         }
2199
2200         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2201         if (ret) {
2202                 err = ret;
2203                 goto fail_bdi;
2204         }
2205         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2206                                         (1 + ilog2(nr_cpu_ids));
2207
2208         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2209         if (ret) {
2210                 err = ret;
2211                 goto fail_dirty_metadata_bytes;
2212         }
2213
2214         ret = percpu_counter_init(&fs_info->bio_counter, 0);
2215         if (ret) {
2216                 err = ret;
2217                 goto fail_delalloc_bytes;
2218         }
2219
2220         fs_info->btree_inode = new_inode(sb);
2221         if (!fs_info->btree_inode) {
2222                 err = -ENOMEM;
2223                 goto fail_bio_counter;
2224         }
2225
2226         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2227
2228         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2229         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2230         INIT_LIST_HEAD(&fs_info->trans_list);
2231         INIT_LIST_HEAD(&fs_info->dead_roots);
2232         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2233         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2234         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2235         spin_lock_init(&fs_info->delalloc_root_lock);
2236         spin_lock_init(&fs_info->trans_lock);
2237         spin_lock_init(&fs_info->fs_roots_radix_lock);
2238         spin_lock_init(&fs_info->delayed_iput_lock);
2239         spin_lock_init(&fs_info->defrag_inodes_lock);
2240         spin_lock_init(&fs_info->free_chunk_lock);
2241         spin_lock_init(&fs_info->tree_mod_seq_lock);
2242         spin_lock_init(&fs_info->super_lock);
2243         spin_lock_init(&fs_info->qgroup_op_lock);
2244         spin_lock_init(&fs_info->buffer_lock);
2245         spin_lock_init(&fs_info->unused_bgs_lock);
2246         rwlock_init(&fs_info->tree_mod_log_lock);
2247         mutex_init(&fs_info->reloc_mutex);
2248         mutex_init(&fs_info->delalloc_root_mutex);
2249         seqlock_init(&fs_info->profiles_lock);
2250
2251         init_completion(&fs_info->kobj_unregister);
2252         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2253         INIT_LIST_HEAD(&fs_info->space_info);
2254         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2255         INIT_LIST_HEAD(&fs_info->unused_bgs);
2256         btrfs_mapping_init(&fs_info->mapping_tree);
2257         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2258                              BTRFS_BLOCK_RSV_GLOBAL);
2259         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2260                              BTRFS_BLOCK_RSV_DELALLOC);
2261         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2262         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2263         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2264         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2265                              BTRFS_BLOCK_RSV_DELOPS);
2266         atomic_set(&fs_info->nr_async_submits, 0);
2267         atomic_set(&fs_info->async_delalloc_pages, 0);
2268         atomic_set(&fs_info->async_submit_draining, 0);
2269         atomic_set(&fs_info->nr_async_bios, 0);
2270         atomic_set(&fs_info->defrag_running, 0);
2271         atomic_set(&fs_info->qgroup_op_seq, 0);
2272         atomic64_set(&fs_info->tree_mod_seq, 0);
2273         fs_info->sb = sb;
2274         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2275         fs_info->metadata_ratio = 0;
2276         fs_info->defrag_inodes = RB_ROOT;
2277         fs_info->free_chunk_space = 0;
2278         fs_info->tree_mod_log = RB_ROOT;
2279         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2280         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2281         /* readahead state */
2282         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2283         spin_lock_init(&fs_info->reada_lock);
2284
2285         fs_info->thread_pool_size = min_t(unsigned long,
2286                                           num_online_cpus() + 2, 8);
2287
2288         INIT_LIST_HEAD(&fs_info->ordered_roots);
2289         spin_lock_init(&fs_info->ordered_root_lock);
2290         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2291                                         GFP_NOFS);
2292         if (!fs_info->delayed_root) {
2293                 err = -ENOMEM;
2294                 goto fail_iput;
2295         }
2296         btrfs_init_delayed_root(fs_info->delayed_root);
2297
2298         mutex_init(&fs_info->scrub_lock);
2299         atomic_set(&fs_info->scrubs_running, 0);
2300         atomic_set(&fs_info->scrub_pause_req, 0);
2301         atomic_set(&fs_info->scrubs_paused, 0);
2302         atomic_set(&fs_info->scrub_cancel_req, 0);
2303         init_waitqueue_head(&fs_info->replace_wait);
2304         init_waitqueue_head(&fs_info->scrub_pause_wait);
2305         fs_info->scrub_workers_refcnt = 0;
2306 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2307         fs_info->check_integrity_print_mask = 0;
2308 #endif
2309
2310         spin_lock_init(&fs_info->balance_lock);
2311         mutex_init(&fs_info->balance_mutex);
2312         atomic_set(&fs_info->balance_running, 0);
2313         atomic_set(&fs_info->balance_pause_req, 0);
2314         atomic_set(&fs_info->balance_cancel_req, 0);
2315         fs_info->balance_ctl = NULL;
2316         init_waitqueue_head(&fs_info->balance_wait_q);
2317         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2318
2319         sb->s_blocksize = 4096;
2320         sb->s_blocksize_bits = blksize_bits(4096);
2321         sb->s_bdi = &fs_info->bdi;
2322
2323         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2324         set_nlink(fs_info->btree_inode, 1);
2325         /*
2326          * we set the i_size on the btree inode to the max possible int.
2327          * the real end of the address space is determined by all of
2328          * the devices in the system
2329          */
2330         fs_info->btree_inode->i_size = OFFSET_MAX;
2331         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2332         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2333
2334         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2335         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2336                              fs_info->btree_inode->i_mapping);
2337         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2338         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2339
2340         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2341
2342         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2343         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2344                sizeof(struct btrfs_key));
2345         set_bit(BTRFS_INODE_DUMMY,
2346                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2347         btrfs_insert_inode_hash(fs_info->btree_inode);
2348
2349         spin_lock_init(&fs_info->block_group_cache_lock);
2350         fs_info->block_group_cache_tree = RB_ROOT;
2351         fs_info->first_logical_byte = (u64)-1;
2352
2353         extent_io_tree_init(&fs_info->freed_extents[0],
2354                              fs_info->btree_inode->i_mapping);
2355         extent_io_tree_init(&fs_info->freed_extents[1],
2356                              fs_info->btree_inode->i_mapping);
2357         fs_info->pinned_extents = &fs_info->freed_extents[0];
2358         fs_info->do_barriers = 1;
2359
2360
2361         mutex_init(&fs_info->ordered_operations_mutex);
2362         mutex_init(&fs_info->ordered_extent_flush_mutex);
2363         mutex_init(&fs_info->tree_log_mutex);
2364         mutex_init(&fs_info->chunk_mutex);
2365         mutex_init(&fs_info->transaction_kthread_mutex);
2366         mutex_init(&fs_info->cleaner_mutex);
2367         mutex_init(&fs_info->volume_mutex);
2368         init_rwsem(&fs_info->commit_root_sem);
2369         init_rwsem(&fs_info->cleanup_work_sem);
2370         init_rwsem(&fs_info->subvol_sem);
2371         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2372         fs_info->dev_replace.lock_owner = 0;
2373         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2374         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2375         mutex_init(&fs_info->dev_replace.lock_management_lock);
2376         mutex_init(&fs_info->dev_replace.lock);
2377
2378         spin_lock_init(&fs_info->qgroup_lock);
2379         mutex_init(&fs_info->qgroup_ioctl_lock);
2380         fs_info->qgroup_tree = RB_ROOT;
2381         fs_info->qgroup_op_tree = RB_ROOT;
2382         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2383         fs_info->qgroup_seq = 1;
2384         fs_info->quota_enabled = 0;
2385         fs_info->pending_quota_state = 0;
2386         fs_info->qgroup_ulist = NULL;
2387         mutex_init(&fs_info->qgroup_rescan_lock);
2388
2389         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2390         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2391
2392         init_waitqueue_head(&fs_info->transaction_throttle);
2393         init_waitqueue_head(&fs_info->transaction_wait);
2394         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2395         init_waitqueue_head(&fs_info->async_submit_wait);
2396
2397         ret = btrfs_alloc_stripe_hash_table(fs_info);
2398         if (ret) {
2399                 err = ret;
2400                 goto fail_alloc;
2401         }
2402
2403         __setup_root(4096, 4096, 4096, tree_root,
2404                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2405
2406         invalidate_bdev(fs_devices->latest_bdev);
2407
2408         /*
2409          * Read super block and check the signature bytes only
2410          */
2411         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2412         if (!bh) {
2413                 err = -EINVAL;
2414                 goto fail_alloc;
2415         }
2416
2417         /*
2418          * We want to check superblock checksum, the type is stored inside.
2419          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2420          */
2421         if (btrfs_check_super_csum(bh->b_data)) {
2422                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2423                 err = -EINVAL;
2424                 goto fail_alloc;
2425         }
2426
2427         /*
2428          * super_copy is zeroed at allocation time and we never touch the
2429          * following bytes up to INFO_SIZE, the checksum is calculated from
2430          * the whole block of INFO_SIZE
2431          */
2432         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2433         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2434                sizeof(*fs_info->super_for_commit));
2435         brelse(bh);
2436
2437         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2438
2439         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2440         if (ret) {
2441                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2442                 err = -EINVAL;
2443                 goto fail_alloc;
2444         }
2445
2446         disk_super = fs_info->super_copy;
2447         if (!btrfs_super_root(disk_super))
2448                 goto fail_alloc;
2449
2450         /* check FS state, whether FS is broken. */
2451         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2452                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2453
2454         /*
2455          * run through our array of backup supers and setup
2456          * our ring pointer to the oldest one
2457          */
2458         generation = btrfs_super_generation(disk_super);
2459         find_oldest_super_backup(fs_info, generation);
2460
2461         /*
2462          * In the long term, we'll store the compression type in the super
2463          * block, and it'll be used for per file compression control.
2464          */
2465         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2466
2467         ret = btrfs_parse_options(tree_root, options);
2468         if (ret) {
2469                 err = ret;
2470                 goto fail_alloc;
2471         }
2472
2473         features = btrfs_super_incompat_flags(disk_super) &
2474                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2475         if (features) {
2476                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2477                        "unsupported optional features (%Lx).\n",
2478                        features);
2479                 err = -EINVAL;
2480                 goto fail_alloc;
2481         }
2482
2483         /*
2484          * Leafsize and nodesize were always equal, this is only a sanity check.
2485          */
2486         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2487             btrfs_super_nodesize(disk_super)) {
2488                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2489                        "blocksizes don't match.  node %d leaf %d\n",
2490                        btrfs_super_nodesize(disk_super),
2491                        le32_to_cpu(disk_super->__unused_leafsize));
2492                 err = -EINVAL;
2493                 goto fail_alloc;
2494         }
2495         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2496                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2497                        "blocksize (%d) was too large\n",
2498                        btrfs_super_nodesize(disk_super));
2499                 err = -EINVAL;
2500                 goto fail_alloc;
2501         }
2502
2503         features = btrfs_super_incompat_flags(disk_super);
2504         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2505         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2506                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2507
2508         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2509                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2510
2511         /*
2512          * flag our filesystem as having big metadata blocks if
2513          * they are bigger than the page size
2514          */
2515         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2516                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2517                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2518                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2519         }
2520
2521         nodesize = btrfs_super_nodesize(disk_super);
2522         sectorsize = btrfs_super_sectorsize(disk_super);
2523         stripesize = btrfs_super_stripesize(disk_super);
2524         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2525         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2526
2527         /*
2528          * mixed block groups end up with duplicate but slightly offset
2529          * extent buffers for the same range.  It leads to corruptions
2530          */
2531         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2532             (sectorsize != nodesize)) {
2533                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2534                                 "are not allowed for mixed block groups on %s\n",
2535                                 sb->s_id);
2536                 goto fail_alloc;
2537         }
2538
2539         /*
2540          * Needn't use the lock because there is no other task which will
2541          * update the flag.
2542          */
2543         btrfs_set_super_incompat_flags(disk_super, features);
2544
2545         features = btrfs_super_compat_ro_flags(disk_super) &
2546                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2547         if (!(sb->s_flags & MS_RDONLY) && features) {
2548                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2549                        "unsupported option features (%Lx).\n",
2550                        features);
2551                 err = -EINVAL;
2552                 goto fail_alloc;
2553         }
2554
2555         max_active = fs_info->thread_pool_size;
2556
2557         fs_info->workers =
2558                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2559                                       max_active, 16);
2560
2561         fs_info->delalloc_workers =
2562                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2563
2564         fs_info->flush_workers =
2565                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2566
2567         fs_info->caching_workers =
2568                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2569
2570         /*
2571          * a higher idle thresh on the submit workers makes it much more
2572          * likely that bios will be send down in a sane order to the
2573          * devices
2574          */
2575         fs_info->submit_workers =
2576                 btrfs_alloc_workqueue("submit", flags,
2577                                       min_t(u64, fs_devices->num_devices,
2578                                             max_active), 64);
2579
2580         fs_info->fixup_workers =
2581                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2582
2583         /*
2584          * endios are largely parallel and should have a very
2585          * low idle thresh
2586          */
2587         fs_info->endio_workers =
2588                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2589         fs_info->endio_meta_workers =
2590                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2591         fs_info->endio_meta_write_workers =
2592                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2593         fs_info->endio_raid56_workers =
2594                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2595         fs_info->endio_repair_workers =
2596                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2597         fs_info->rmw_workers =
2598                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2599         fs_info->endio_write_workers =
2600                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2601         fs_info->endio_freespace_worker =
2602                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2603         fs_info->delayed_workers =
2604                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2605         fs_info->readahead_workers =
2606                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2607         fs_info->qgroup_rescan_workers =
2608                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2609         fs_info->extent_workers =
2610                 btrfs_alloc_workqueue("extent-refs", flags,
2611                                       min_t(u64, fs_devices->num_devices,
2612                                             max_active), 8);
2613
2614         if (!(fs_info->workers && fs_info->delalloc_workers &&
2615               fs_info->submit_workers && fs_info->flush_workers &&
2616               fs_info->endio_workers && fs_info->endio_meta_workers &&
2617               fs_info->endio_meta_write_workers &&
2618               fs_info->endio_repair_workers &&
2619               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2620               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2621               fs_info->caching_workers && fs_info->readahead_workers &&
2622               fs_info->fixup_workers && fs_info->delayed_workers &&
2623               fs_info->extent_workers &&
2624               fs_info->qgroup_rescan_workers)) {
2625                 err = -ENOMEM;
2626                 goto fail_sb_buffer;
2627         }
2628
2629         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2630         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2631                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2632
2633         tree_root->nodesize = nodesize;
2634         tree_root->sectorsize = sectorsize;
2635         tree_root->stripesize = stripesize;
2636
2637         sb->s_blocksize = sectorsize;
2638         sb->s_blocksize_bits = blksize_bits(sectorsize);
2639
2640         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2641                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2642                 goto fail_sb_buffer;
2643         }
2644
2645         if (sectorsize != PAGE_SIZE) {
2646                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2647                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2648                 goto fail_sb_buffer;
2649         }
2650
2651         mutex_lock(&fs_info->chunk_mutex);
2652         ret = btrfs_read_sys_array(tree_root);
2653         mutex_unlock(&fs_info->chunk_mutex);
2654         if (ret) {
2655                 printk(KERN_WARNING "BTRFS: failed to read the system "
2656                        "array on %s\n", sb->s_id);
2657                 goto fail_sb_buffer;
2658         }
2659
2660         blocksize = tree_root->nodesize;
2661         generation = btrfs_super_chunk_root_generation(disk_super);
2662
2663         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2664                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2665
2666         chunk_root->node = read_tree_block(chunk_root,
2667                                            btrfs_super_chunk_root(disk_super),
2668                                            blocksize, generation);
2669         if (!chunk_root->node ||
2670             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2671                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2672                        sb->s_id);
2673                 goto fail_tree_roots;
2674         }
2675         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2676         chunk_root->commit_root = btrfs_root_node(chunk_root);
2677
2678         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2679            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2680
2681         ret = btrfs_read_chunk_tree(chunk_root);
2682         if (ret) {
2683                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2684                        sb->s_id);
2685                 goto fail_tree_roots;
2686         }
2687
2688         /*
2689          * keep the device that is marked to be the target device for the
2690          * dev_replace procedure
2691          */
2692         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2693
2694         if (!fs_devices->latest_bdev) {
2695                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2696                        sb->s_id);
2697                 goto fail_tree_roots;
2698         }
2699
2700 retry_root_backup:
2701         blocksize = tree_root->nodesize;
2702         generation = btrfs_super_generation(disk_super);
2703
2704         tree_root->node = read_tree_block(tree_root,
2705                                           btrfs_super_root(disk_super),
2706                                           blocksize, generation);
2707         if (!tree_root->node ||
2708             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2709                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2710                        sb->s_id);
2711
2712                 goto recovery_tree_root;
2713         }
2714
2715         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2716         tree_root->commit_root = btrfs_root_node(tree_root);
2717         btrfs_set_root_refs(&tree_root->root_item, 1);
2718
2719         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2720         location.type = BTRFS_ROOT_ITEM_KEY;
2721         location.offset = 0;
2722
2723         extent_root = btrfs_read_tree_root(tree_root, &location);
2724         if (IS_ERR(extent_root)) {
2725                 ret = PTR_ERR(extent_root);
2726                 goto recovery_tree_root;
2727         }
2728         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2729         fs_info->extent_root = extent_root;
2730
2731         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2732         dev_root = btrfs_read_tree_root(tree_root, &location);
2733         if (IS_ERR(dev_root)) {
2734                 ret = PTR_ERR(dev_root);
2735                 goto recovery_tree_root;
2736         }
2737         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2738         fs_info->dev_root = dev_root;
2739         btrfs_init_devices_late(fs_info);
2740
2741         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2742         csum_root = btrfs_read_tree_root(tree_root, &location);
2743         if (IS_ERR(csum_root)) {
2744                 ret = PTR_ERR(csum_root);
2745                 goto recovery_tree_root;
2746         }
2747         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2748         fs_info->csum_root = csum_root;
2749
2750         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2751         quota_root = btrfs_read_tree_root(tree_root, &location);
2752         if (!IS_ERR(quota_root)) {
2753                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2754                 fs_info->quota_enabled = 1;
2755                 fs_info->pending_quota_state = 1;
2756                 fs_info->quota_root = quota_root;
2757         }
2758
2759         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2760         uuid_root = btrfs_read_tree_root(tree_root, &location);
2761         if (IS_ERR(uuid_root)) {
2762                 ret = PTR_ERR(uuid_root);
2763                 if (ret != -ENOENT)
2764                         goto recovery_tree_root;
2765                 create_uuid_tree = true;
2766                 check_uuid_tree = false;
2767         } else {
2768                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2769                 fs_info->uuid_root = uuid_root;
2770                 create_uuid_tree = false;
2771                 check_uuid_tree =
2772                     generation != btrfs_super_uuid_tree_generation(disk_super);
2773         }
2774
2775         fs_info->generation = generation;
2776         fs_info->last_trans_committed = generation;
2777
2778         ret = btrfs_recover_balance(fs_info);
2779         if (ret) {
2780                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2781                 goto fail_block_groups;
2782         }
2783
2784         ret = btrfs_init_dev_stats(fs_info);
2785         if (ret) {
2786                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2787                        ret);
2788                 goto fail_block_groups;
2789         }
2790
2791         ret = btrfs_init_dev_replace(fs_info);
2792         if (ret) {
2793                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2794                 goto fail_block_groups;
2795         }
2796
2797         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2798
2799         ret = btrfs_sysfs_add_one(fs_info);
2800         if (ret) {
2801                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2802                 goto fail_block_groups;
2803         }
2804
2805         ret = btrfs_init_space_info(fs_info);
2806         if (ret) {
2807                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2808                 goto fail_sysfs;
2809         }
2810
2811         ret = btrfs_read_block_groups(extent_root);
2812         if (ret) {
2813                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2814                 goto fail_sysfs;
2815         }
2816         fs_info->num_tolerated_disk_barrier_failures =
2817                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2818         if (fs_info->fs_devices->missing_devices >
2819              fs_info->num_tolerated_disk_barrier_failures &&
2820             !(sb->s_flags & MS_RDONLY)) {
2821                 printk(KERN_WARNING "BTRFS: "
2822                         "too many missing devices, writeable mount is not allowed\n");
2823                 goto fail_sysfs;
2824         }
2825
2826         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2827                                                "btrfs-cleaner");
2828         if (IS_ERR(fs_info->cleaner_kthread))
2829                 goto fail_sysfs;
2830
2831         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2832                                                    tree_root,
2833                                                    "btrfs-transaction");
2834         if (IS_ERR(fs_info->transaction_kthread))
2835                 goto fail_cleaner;
2836
2837         if (!btrfs_test_opt(tree_root, SSD) &&
2838             !btrfs_test_opt(tree_root, NOSSD) &&
2839             !fs_info->fs_devices->rotating) {
2840                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2841                        "mode\n");
2842                 btrfs_set_opt(fs_info->mount_opt, SSD);
2843         }
2844
2845         /* Set the real inode map cache flag */
2846         if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2847                 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2848
2849 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2850         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2851                 ret = btrfsic_mount(tree_root, fs_devices,
2852                                     btrfs_test_opt(tree_root,
2853                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2854                                     1 : 0,
2855                                     fs_info->check_integrity_print_mask);
2856                 if (ret)
2857                         printk(KERN_WARNING "BTRFS: failed to initialize"
2858                                " integrity check module %s\n", sb->s_id);
2859         }
2860 #endif
2861         ret = btrfs_read_qgroup_config(fs_info);
2862         if (ret)
2863                 goto fail_trans_kthread;
2864
2865         /* do not make disk changes in broken FS */
2866         if (btrfs_super_log_root(disk_super) != 0) {
2867                 u64 bytenr = btrfs_super_log_root(disk_super);
2868
2869                 if (fs_devices->rw_devices == 0) {
2870                         printk(KERN_WARNING "BTRFS: log replay required "
2871                                "on RO media\n");
2872                         err = -EIO;
2873                         goto fail_qgroup;
2874                 }
2875                 blocksize = tree_root->nodesize;
2876
2877                 log_tree_root = btrfs_alloc_root(fs_info);
2878                 if (!log_tree_root) {
2879                         err = -ENOMEM;
2880                         goto fail_qgroup;
2881                 }
2882
2883                 __setup_root(nodesize, sectorsize, stripesize,
2884                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2885
2886                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2887                                                       blocksize,
2888                                                       generation + 1);
2889                 if (!log_tree_root->node ||
2890                     !extent_buffer_uptodate(log_tree_root->node)) {
2891                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2892                         free_extent_buffer(log_tree_root->node);
2893                         kfree(log_tree_root);
2894                         goto fail_qgroup;
2895                 }
2896                 /* returns with log_tree_root freed on success */
2897                 ret = btrfs_recover_log_trees(log_tree_root);
2898                 if (ret) {
2899                         btrfs_error(tree_root->fs_info, ret,
2900                                     "Failed to recover log tree");
2901                         free_extent_buffer(log_tree_root->node);
2902                         kfree(log_tree_root);
2903                         goto fail_qgroup;
2904                 }
2905
2906                 if (sb->s_flags & MS_RDONLY) {
2907                         ret = btrfs_commit_super(tree_root);
2908                         if (ret)
2909                                 goto fail_qgroup;
2910                 }
2911         }
2912
2913         ret = btrfs_find_orphan_roots(tree_root);
2914         if (ret)
2915                 goto fail_qgroup;
2916
2917         if (!(sb->s_flags & MS_RDONLY)) {
2918                 ret = btrfs_cleanup_fs_roots(fs_info);
2919                 if (ret)
2920                         goto fail_qgroup;
2921
2922                 mutex_lock(&fs_info->cleaner_mutex);
2923                 ret = btrfs_recover_relocation(tree_root);
2924                 mutex_unlock(&fs_info->cleaner_mutex);
2925                 if (ret < 0) {
2926                         printk(KERN_WARNING
2927                                "BTRFS: failed to recover relocation\n");
2928                         err = -EINVAL;
2929                         goto fail_qgroup;
2930                 }
2931         }
2932
2933         location.objectid = BTRFS_FS_TREE_OBJECTID;
2934         location.type = BTRFS_ROOT_ITEM_KEY;
2935         location.offset = 0;
2936
2937         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2938         if (IS_ERR(fs_info->fs_root)) {
2939                 err = PTR_ERR(fs_info->fs_root);
2940                 goto fail_qgroup;
2941         }
2942
2943         if (sb->s_flags & MS_RDONLY)
2944                 return 0;
2945
2946         down_read(&fs_info->cleanup_work_sem);
2947         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2948             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2949                 up_read(&fs_info->cleanup_work_sem);
2950                 close_ctree(tree_root);
2951                 return ret;
2952         }
2953         up_read(&fs_info->cleanup_work_sem);
2954
2955         ret = btrfs_resume_balance_async(fs_info);
2956         if (ret) {
2957                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2958                 close_ctree(tree_root);
2959                 return ret;
2960         }
2961
2962         ret = btrfs_resume_dev_replace_async(fs_info);
2963         if (ret) {
2964                 pr_warn("BTRFS: failed to resume dev_replace\n");
2965                 close_ctree(tree_root);
2966                 return ret;
2967         }
2968
2969         btrfs_qgroup_rescan_resume(fs_info);
2970
2971         if (create_uuid_tree) {
2972                 pr_info("BTRFS: creating UUID tree\n");
2973                 ret = btrfs_create_uuid_tree(fs_info);
2974                 if (ret) {
2975                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2976                                 ret);
2977                         close_ctree(tree_root);
2978                         return ret;
2979                 }
2980         } else if (check_uuid_tree ||
2981                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2982                 pr_info("BTRFS: checking UUID tree\n");
2983                 ret = btrfs_check_uuid_tree(fs_info);
2984                 if (ret) {
2985                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2986                                 ret);
2987                         close_ctree(tree_root);
2988                         return ret;
2989                 }
2990         } else {
2991                 fs_info->update_uuid_tree_gen = 1;
2992         }
2993
2994         fs_info->open = 1;
2995
2996         return 0;
2997
2998 fail_qgroup:
2999         btrfs_free_qgroup_config(fs_info);
3000 fail_trans_kthread:
3001         kthread_stop(fs_info->transaction_kthread);
3002         btrfs_cleanup_transaction(fs_info->tree_root);
3003         btrfs_free_fs_roots(fs_info);
3004 fail_cleaner:
3005         kthread_stop(fs_info->cleaner_kthread);
3006
3007         /*
3008          * make sure we're done with the btree inode before we stop our
3009          * kthreads
3010          */
3011         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3012
3013 fail_sysfs:
3014         btrfs_sysfs_remove_one(fs_info);
3015
3016 fail_block_groups:
3017         btrfs_put_block_group_cache(fs_info);
3018         btrfs_free_block_groups(fs_info);
3019
3020 fail_tree_roots:
3021         free_root_pointers(fs_info, 1);
3022         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3023
3024 fail_sb_buffer:
3025         btrfs_stop_all_workers(fs_info);
3026 fail_alloc:
3027 fail_iput:
3028         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3029
3030         iput(fs_info->btree_inode);
3031 fail_bio_counter:
3032         percpu_counter_destroy(&fs_info->bio_counter);
3033 fail_delalloc_bytes:
3034         percpu_counter_destroy(&fs_info->delalloc_bytes);
3035 fail_dirty_metadata_bytes:
3036         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3037 fail_bdi:
3038         bdi_destroy(&fs_info->bdi);
3039 fail_srcu:
3040         cleanup_srcu_struct(&fs_info->subvol_srcu);
3041 fail:
3042         btrfs_free_stripe_hash_table(fs_info);
3043         btrfs_close_devices(fs_info->fs_devices);
3044         return err;
3045
3046 recovery_tree_root:
3047         if (!btrfs_test_opt(tree_root, RECOVERY))
3048                 goto fail_tree_roots;
3049
3050         free_root_pointers(fs_info, 0);
3051
3052         /* don't use the log in recovery mode, it won't be valid */
3053         btrfs_set_super_log_root(disk_super, 0);
3054
3055         /* we can't trust the free space cache either */
3056         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3057
3058         ret = next_root_backup(fs_info, fs_info->super_copy,
3059                                &num_backups_tried, &backup_index);
3060         if (ret == -1)
3061                 goto fail_block_groups;
3062         goto retry_root_backup;
3063 }
3064
3065 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3066 {
3067         if (uptodate) {
3068                 set_buffer_uptodate(bh);
3069         } else {
3070                 struct btrfs_device *device = (struct btrfs_device *)
3071                         bh->b_private;
3072
3073                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3074                                           "I/O error on %s\n",
3075                                           rcu_str_deref(device->name));
3076                 /* note, we dont' set_buffer_write_io_error because we have
3077                  * our own ways of dealing with the IO errors
3078                  */
3079                 clear_buffer_uptodate(bh);
3080                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3081         }
3082         unlock_buffer(bh);
3083         put_bh(bh);
3084 }
3085
3086 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3087 {
3088         struct buffer_head *bh;
3089         struct buffer_head *latest = NULL;
3090         struct btrfs_super_block *super;
3091         int i;
3092         u64 transid = 0;
3093         u64 bytenr;
3094
3095         /* we would like to check all the supers, but that would make
3096          * a btrfs mount succeed after a mkfs from a different FS.
3097          * So, we need to add a special mount option to scan for
3098          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3099          */
3100         for (i = 0; i < 1; i++) {
3101                 bytenr = btrfs_sb_offset(i);
3102                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3103                                         i_size_read(bdev->bd_inode))
3104                         break;
3105                 bh = __bread(bdev, bytenr / 4096,
3106                                         BTRFS_SUPER_INFO_SIZE);
3107                 if (!bh)
3108                         continue;
3109
3110                 super = (struct btrfs_super_block *)bh->b_data;
3111                 if (btrfs_super_bytenr(super) != bytenr ||
3112                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3113                         brelse(bh);
3114                         continue;
3115                 }
3116
3117                 if (!latest || btrfs_super_generation(super) > transid) {
3118                         brelse(latest);
3119                         latest = bh;
3120                         transid = btrfs_super_generation(super);
3121                 } else {
3122                         brelse(bh);
3123                 }
3124         }
3125         return latest;
3126 }
3127
3128 /*
3129  * this should be called twice, once with wait == 0 and
3130  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3131  * we write are pinned.
3132  *
3133  * They are released when wait == 1 is done.
3134  * max_mirrors must be the same for both runs, and it indicates how
3135  * many supers on this one device should be written.
3136  *
3137  * max_mirrors == 0 means to write them all.
3138  */
3139 static int write_dev_supers(struct btrfs_device *device,
3140                             struct btrfs_super_block *sb,
3141                             int do_barriers, int wait, int max_mirrors)
3142 {
3143         struct buffer_head *bh;
3144         int i;
3145         int ret;
3146         int errors = 0;
3147         u32 crc;
3148         u64 bytenr;
3149
3150         if (max_mirrors == 0)
3151                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3152
3153         for (i = 0; i < max_mirrors; i++) {
3154                 bytenr = btrfs_sb_offset(i);
3155                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3156                     device->commit_total_bytes)
3157                         break;
3158
3159                 if (wait) {
3160                         bh = __find_get_block(device->bdev, bytenr / 4096,
3161                                               BTRFS_SUPER_INFO_SIZE);
3162                         if (!bh) {
3163                                 errors++;
3164                                 continue;
3165                         }
3166                         wait_on_buffer(bh);
3167                         if (!buffer_uptodate(bh))
3168                                 errors++;
3169
3170                         /* drop our reference */
3171                         brelse(bh);
3172
3173                         /* drop the reference from the wait == 0 run */
3174                         brelse(bh);
3175                         continue;
3176                 } else {
3177                         btrfs_set_super_bytenr(sb, bytenr);
3178
3179                         crc = ~(u32)0;
3180                         crc = btrfs_csum_data((char *)sb +
3181                                               BTRFS_CSUM_SIZE, crc,
3182                                               BTRFS_SUPER_INFO_SIZE -
3183                                               BTRFS_CSUM_SIZE);
3184                         btrfs_csum_final(crc, sb->csum);
3185
3186                         /*
3187                          * one reference for us, and we leave it for the
3188                          * caller
3189                          */
3190                         bh = __getblk(device->bdev, bytenr / 4096,
3191                                       BTRFS_SUPER_INFO_SIZE);
3192                         if (!bh) {
3193                                 printk(KERN_ERR "BTRFS: couldn't get super "
3194                                        "buffer head for bytenr %Lu\n", bytenr);
3195                                 errors++;
3196                                 continue;
3197                         }
3198
3199                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3200
3201                         /* one reference for submit_bh */
3202                         get_bh(bh);
3203
3204                         set_buffer_uptodate(bh);
3205                         lock_buffer(bh);
3206                         bh->b_end_io = btrfs_end_buffer_write_sync;
3207                         bh->b_private = device;
3208                 }
3209
3210                 /*
3211                  * we fua the first super.  The others we allow
3212                  * to go down lazy.
3213                  */
3214                 if (i == 0)
3215                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3216                 else
3217                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3218                 if (ret)
3219                         errors++;
3220         }
3221         return errors < i ? 0 : -1;
3222 }
3223
3224 /*
3225  * endio for the write_dev_flush, this will wake anyone waiting
3226  * for the barrier when it is done
3227  */
3228 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3229 {
3230         if (err) {
3231                 if (err == -EOPNOTSUPP)
3232                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3233                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3234         }
3235         if (bio->bi_private)
3236                 complete(bio->bi_private);
3237         bio_put(bio);
3238 }
3239
3240 /*
3241  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3242  * sent down.  With wait == 1, it waits for the previous flush.
3243  *
3244  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3245  * capable
3246  */
3247 static int write_dev_flush(struct btrfs_device *device, int wait)
3248 {
3249         struct bio *bio;
3250         int ret = 0;
3251
3252         if (device->nobarriers)
3253                 return 0;
3254
3255         if (wait) {
3256                 bio = device->flush_bio;
3257                 if (!bio)
3258                         return 0;
3259
3260                 wait_for_completion(&device->flush_wait);
3261
3262                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3263                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3264                                       rcu_str_deref(device->name));
3265                         device->nobarriers = 1;
3266                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3267                         ret = -EIO;
3268                         btrfs_dev_stat_inc_and_print(device,
3269                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3270                 }
3271
3272                 /* drop the reference from the wait == 0 run */
3273                 bio_put(bio);
3274                 device->flush_bio = NULL;
3275
3276                 return ret;
3277         }
3278
3279         /*
3280          * one reference for us, and we leave it for the
3281          * caller
3282          */
3283         device->flush_bio = NULL;
3284         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3285         if (!bio)
3286                 return -ENOMEM;
3287
3288         bio->bi_end_io = btrfs_end_empty_barrier;
3289         bio->bi_bdev = device->bdev;
3290         init_completion(&device->flush_wait);
3291         bio->bi_private = &device->flush_wait;
3292         device->flush_bio = bio;
3293
3294         bio_get(bio);
3295         btrfsic_submit_bio(WRITE_FLUSH, bio);
3296
3297         return 0;
3298 }
3299
3300 /*
3301  * send an empty flush down to each device in parallel,
3302  * then wait for them
3303  */
3304 static int barrier_all_devices(struct btrfs_fs_info *info)
3305 {
3306         struct list_head *head;
3307         struct btrfs_device *dev;
3308         int errors_send = 0;
3309         int errors_wait = 0;
3310         int ret;
3311
3312         /* send down all the barriers */
3313         head = &info->fs_devices->devices;
3314         list_for_each_entry_rcu(dev, head, dev_list) {
3315                 if (dev->missing)
3316                         continue;
3317                 if (!dev->bdev) {
3318                         errors_send++;
3319                         continue;
3320                 }
3321                 if (!dev->in_fs_metadata || !dev->writeable)
3322                         continue;
3323
3324                 ret = write_dev_flush(dev, 0);
3325                 if (ret)
3326                         errors_send++;
3327         }
3328
3329         /* wait for all the barriers */
3330         list_for_each_entry_rcu(dev, head, dev_list) {
3331                 if (dev->missing)
3332                         continue;
3333                 if (!dev->bdev) {
3334                         errors_wait++;
3335                         continue;
3336                 }
3337                 if (!dev->in_fs_metadata || !dev->writeable)
3338                         continue;
3339
3340                 ret = write_dev_flush(dev, 1);
3341                 if (ret)
3342                         errors_wait++;
3343         }
3344         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3345             errors_wait > info->num_tolerated_disk_barrier_failures)
3346                 return -EIO;
3347         return 0;
3348 }
3349
3350 int btrfs_calc_num_tolerated_disk_barrier_failures(
3351         struct btrfs_fs_info *fs_info)
3352 {
3353         struct btrfs_ioctl_space_info space;
3354         struct btrfs_space_info *sinfo;
3355         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3356                        BTRFS_BLOCK_GROUP_SYSTEM,
3357                        BTRFS_BLOCK_GROUP_METADATA,
3358                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3359         int num_types = 4;
3360         int i;
3361         int c;
3362         int num_tolerated_disk_barrier_failures =
3363                 (int)fs_info->fs_devices->num_devices;
3364
3365         for (i = 0; i < num_types; i++) {
3366                 struct btrfs_space_info *tmp;
3367
3368                 sinfo = NULL;
3369                 rcu_read_lock();
3370                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3371                         if (tmp->flags == types[i]) {
3372                                 sinfo = tmp;
3373                                 break;
3374                         }
3375                 }
3376                 rcu_read_unlock();
3377
3378                 if (!sinfo)
3379                         continue;
3380
3381                 down_read(&sinfo->groups_sem);
3382                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3383                         if (!list_empty(&sinfo->block_groups[c])) {
3384                                 u64 flags;
3385
3386                                 btrfs_get_block_group_info(
3387                                         &sinfo->block_groups[c], &space);
3388                                 if (space.total_bytes == 0 ||
3389                                     space.used_bytes == 0)
3390                                         continue;
3391                                 flags = space.flags;
3392                                 /*
3393                                  * return
3394                                  * 0: if dup, single or RAID0 is configured for
3395                                  *    any of metadata, system or data, else
3396                                  * 1: if RAID5 is configured, or if RAID1 or
3397                                  *    RAID10 is configured and only two mirrors
3398                                  *    are used, else
3399                                  * 2: if RAID6 is configured, else
3400                                  * num_mirrors - 1: if RAID1 or RAID10 is
3401                                  *                  configured and more than
3402                                  *                  2 mirrors are used.
3403                                  */
3404                                 if (num_tolerated_disk_barrier_failures > 0 &&
3405                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3406                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3407                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3408                                       == 0)))
3409                                         num_tolerated_disk_barrier_failures = 0;
3410                                 else if (num_tolerated_disk_barrier_failures > 1) {
3411                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3412                                             BTRFS_BLOCK_GROUP_RAID5 |
3413                                             BTRFS_BLOCK_GROUP_RAID10)) {
3414                                                 num_tolerated_disk_barrier_failures = 1;
3415                                         } else if (flags &
3416                                                    BTRFS_BLOCK_GROUP_RAID6) {
3417                                                 num_tolerated_disk_barrier_failures = 2;
3418                                         }
3419                                 }
3420                         }
3421                 }
3422                 up_read(&sinfo->groups_sem);
3423         }
3424
3425         return num_tolerated_disk_barrier_failures;
3426 }
3427
3428 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3429 {
3430         struct list_head *head;
3431         struct btrfs_device *dev;
3432         struct btrfs_super_block *sb;
3433         struct btrfs_dev_item *dev_item;
3434         int ret;
3435         int do_barriers;
3436         int max_errors;
3437         int total_errors = 0;
3438         u64 flags;
3439
3440         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3441         backup_super_roots(root->fs_info);
3442
3443         sb = root->fs_info->super_for_commit;
3444         dev_item = &sb->dev_item;
3445
3446         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3447         head = &root->fs_info->fs_devices->devices;
3448         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3449
3450         if (do_barriers) {
3451                 ret = barrier_all_devices(root->fs_info);
3452                 if (ret) {
3453                         mutex_unlock(
3454                                 &root->fs_info->fs_devices->device_list_mutex);
3455                         btrfs_error(root->fs_info, ret,
3456                                     "errors while submitting device barriers.");
3457                         return ret;
3458                 }
3459         }
3460
3461         list_for_each_entry_rcu(dev, head, dev_list) {
3462                 if (!dev->bdev) {
3463                         total_errors++;
3464                         continue;
3465                 }
3466                 if (!dev->in_fs_metadata || !dev->writeable)
3467                         continue;
3468
3469                 btrfs_set_stack_device_generation(dev_item, 0);
3470                 btrfs_set_stack_device_type(dev_item, dev->type);
3471                 btrfs_set_stack_device_id(dev_item, dev->devid);
3472                 btrfs_set_stack_device_total_bytes(dev_item,
3473                                                    dev->commit_total_bytes);
3474                 btrfs_set_stack_device_bytes_used(dev_item,
3475                                                   dev->commit_bytes_used);
3476                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3477                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3478                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3479                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3480                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3481
3482                 flags = btrfs_super_flags(sb);
3483                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3484
3485                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3486                 if (ret)
3487                         total_errors++;
3488         }
3489         if (total_errors > max_errors) {
3490                 btrfs_err(root->fs_info, "%d errors while writing supers",
3491                        total_errors);
3492                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3493
3494                 /* FUA is masked off if unsupported and can't be the reason */
3495                 btrfs_error(root->fs_info, -EIO,
3496                             "%d errors while writing supers", total_errors);
3497                 return -EIO;
3498         }
3499
3500         total_errors = 0;
3501         list_for_each_entry_rcu(dev, head, dev_list) {
3502                 if (!dev->bdev)
3503                         continue;
3504                 if (!dev->in_fs_metadata || !dev->writeable)
3505                         continue;
3506
3507                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3508                 if (ret)
3509                         total_errors++;
3510         }
3511         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3512         if (total_errors > max_errors) {
3513                 btrfs_error(root->fs_info, -EIO,
3514                             "%d errors while writing supers", total_errors);
3515                 return -EIO;
3516         }
3517         return 0;
3518 }
3519
3520 int write_ctree_super(struct btrfs_trans_handle *trans,
3521                       struct btrfs_root *root, int max_mirrors)
3522 {
3523         return write_all_supers(root, max_mirrors);
3524 }
3525
3526 /* Drop a fs root from the radix tree and free it. */
3527 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3528                                   struct btrfs_root *root)
3529 {
3530         spin_lock(&fs_info->fs_roots_radix_lock);
3531         radix_tree_delete(&fs_info->fs_roots_radix,
3532                           (unsigned long)root->root_key.objectid);
3533         spin_unlock(&fs_info->fs_roots_radix_lock);
3534
3535         if (btrfs_root_refs(&root->root_item) == 0)
3536                 synchronize_srcu(&fs_info->subvol_srcu);
3537
3538         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3539                 btrfs_free_log(NULL, root);
3540
3541         if (root->free_ino_pinned)
3542                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3543         if (root->free_ino_ctl)
3544                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3545         free_fs_root(root);
3546 }
3547
3548 static void free_fs_root(struct btrfs_root *root)
3549 {
3550         iput(root->ino_cache_inode);
3551         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3552         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3553         root->orphan_block_rsv = NULL;
3554         if (root->anon_dev)
3555                 free_anon_bdev(root->anon_dev);
3556         if (root->subv_writers)
3557                 btrfs_free_subvolume_writers(root->subv_writers);
3558         free_extent_buffer(root->node);
3559         free_extent_buffer(root->commit_root);
3560         kfree(root->free_ino_ctl);
3561         kfree(root->free_ino_pinned);
3562         kfree(root->name);
3563         btrfs_put_fs_root(root);
3564 }
3565
3566 void btrfs_free_fs_root(struct btrfs_root *root)
3567 {
3568         free_fs_root(root);
3569 }
3570
3571 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3572 {
3573         u64 root_objectid = 0;
3574         struct btrfs_root *gang[8];
3575         int i = 0;
3576         int err = 0;
3577         unsigned int ret = 0;
3578         int index;
3579
3580         while (1) {
3581                 index = srcu_read_lock(&fs_info->subvol_srcu);
3582                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3583                                              (void **)gang, root_objectid,
3584                                              ARRAY_SIZE(gang));
3585                 if (!ret) {
3586                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3587                         break;
3588                 }
3589                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3590
3591                 for (i = 0; i < ret; i++) {
3592                         /* Avoid to grab roots in dead_roots */
3593                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3594                                 gang[i] = NULL;
3595                                 continue;
3596                         }
3597                         /* grab all the search result for later use */
3598                         gang[i] = btrfs_grab_fs_root(gang[i]);
3599                 }
3600                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3601
3602                 for (i = 0; i < ret; i++) {
3603                         if (!gang[i])
3604                                 continue;
3605                         root_objectid = gang[i]->root_key.objectid;
3606                         err = btrfs_orphan_cleanup(gang[i]);
3607                         if (err)
3608                                 break;
3609                         btrfs_put_fs_root(gang[i]);
3610                 }
3611                 root_objectid++;
3612         }
3613
3614         /* release the uncleaned roots due to error */
3615         for (; i < ret; i++) {
3616                 if (gang[i])
3617                         btrfs_put_fs_root(gang[i]);
3618         }
3619         return err;
3620 }
3621
3622 int btrfs_commit_super(struct btrfs_root *root)
3623 {
3624         struct btrfs_trans_handle *trans;
3625
3626         mutex_lock(&root->fs_info->cleaner_mutex);
3627         btrfs_run_delayed_iputs(root);
3628         mutex_unlock(&root->fs_info->cleaner_mutex);
3629         wake_up_process(root->fs_info->cleaner_kthread);
3630
3631         /* wait until ongoing cleanup work done */
3632         down_write(&root->fs_info->cleanup_work_sem);
3633         up_write(&root->fs_info->cleanup_work_sem);
3634
3635         trans = btrfs_join_transaction(root);
3636         if (IS_ERR(trans))
3637                 return PTR_ERR(trans);
3638         return btrfs_commit_transaction(trans, root);
3639 }
3640
3641 void close_ctree(struct btrfs_root *root)
3642 {
3643         struct btrfs_fs_info *fs_info = root->fs_info;
3644         int ret;
3645
3646         fs_info->closing = 1;
3647         smp_mb();
3648
3649         /* wait for the uuid_scan task to finish */
3650         down(&fs_info->uuid_tree_rescan_sem);
3651         /* avoid complains from lockdep et al., set sem back to initial state */
3652         up(&fs_info->uuid_tree_rescan_sem);
3653
3654         /* pause restriper - we want to resume on mount */
3655         btrfs_pause_balance(fs_info);
3656
3657         btrfs_dev_replace_suspend_for_unmount(fs_info);
3658
3659         btrfs_scrub_cancel(fs_info);
3660
3661         /* wait for any defraggers to finish */
3662         wait_event(fs_info->transaction_wait,
3663                    (atomic_read(&fs_info->defrag_running) == 0));
3664
3665         /* clear out the rbtree of defraggable inodes */
3666         btrfs_cleanup_defrag_inodes(fs_info);
3667
3668         cancel_work_sync(&fs_info->async_reclaim_work);
3669
3670         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3671                 ret = btrfs_commit_super(root);
3672                 if (ret)
3673                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3674         }
3675
3676         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3677                 btrfs_error_commit_super(root);
3678
3679         kthread_stop(fs_info->transaction_kthread);
3680         kthread_stop(fs_info->cleaner_kthread);
3681
3682         fs_info->closing = 2;
3683         smp_mb();
3684
3685         btrfs_free_qgroup_config(root->fs_info);
3686
3687         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3688                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3689                        percpu_counter_sum(&fs_info->delalloc_bytes));
3690         }
3691
3692         btrfs_sysfs_remove_one(fs_info);
3693
3694         btrfs_free_fs_roots(fs_info);
3695
3696         btrfs_put_block_group_cache(fs_info);
3697
3698         btrfs_free_block_groups(fs_info);
3699
3700         /*
3701          * we must make sure there is not any read request to
3702          * submit after we stopping all workers.
3703          */
3704         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3705         btrfs_stop_all_workers(fs_info);
3706
3707         fs_info->open = 0;
3708         free_root_pointers(fs_info, 1);
3709
3710         iput(fs_info->btree_inode);
3711
3712 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3713         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3714                 btrfsic_unmount(root, fs_info->fs_devices);
3715 #endif
3716
3717         btrfs_close_devices(fs_info->fs_devices);
3718         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3719
3720         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3721         percpu_counter_destroy(&fs_info->delalloc_bytes);
3722         percpu_counter_destroy(&fs_info->bio_counter);
3723         bdi_destroy(&fs_info->bdi);
3724         cleanup_srcu_struct(&fs_info->subvol_srcu);
3725
3726         btrfs_free_stripe_hash_table(fs_info);
3727
3728         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3729         root->orphan_block_rsv = NULL;
3730 }
3731
3732 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3733                           int atomic)
3734 {
3735         int ret;
3736         struct inode *btree_inode = buf->pages[0]->mapping->host;
3737
3738         ret = extent_buffer_uptodate(buf);
3739         if (!ret)
3740                 return ret;
3741
3742         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3743                                     parent_transid, atomic);
3744         if (ret == -EAGAIN)
3745                 return ret;
3746         return !ret;
3747 }
3748
3749 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3750 {
3751         return set_extent_buffer_uptodate(buf);
3752 }
3753
3754 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3755 {
3756         struct btrfs_root *root;
3757         u64 transid = btrfs_header_generation(buf);
3758         int was_dirty;
3759
3760 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3761         /*
3762          * This is a fast path so only do this check if we have sanity tests
3763          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3764          * outside of the sanity tests.
3765          */
3766         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3767                 return;
3768 #endif
3769         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3770         btrfs_assert_tree_locked(buf);
3771         if (transid != root->fs_info->generation)
3772                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3773                        "found %llu running %llu\n",
3774                         buf->start, transid, root->fs_info->generation);
3775         was_dirty = set_extent_buffer_dirty(buf);
3776         if (!was_dirty)
3777                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3778                                      buf->len,
3779                                      root->fs_info->dirty_metadata_batch);
3780 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3781         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3782                 btrfs_print_leaf(root, buf);
3783                 ASSERT(0);
3784         }
3785 #endif
3786 }
3787
3788 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3789                                         int flush_delayed)
3790 {
3791         /*
3792          * looks as though older kernels can get into trouble with
3793          * this code, they end up stuck in balance_dirty_pages forever
3794          */
3795         int ret;
3796
3797         if (current->flags & PF_MEMALLOC)
3798                 return;
3799
3800         if (flush_delayed)
3801                 btrfs_balance_delayed_items(root);
3802
3803         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3804                                      BTRFS_DIRTY_METADATA_THRESH);
3805         if (ret > 0) {
3806                 balance_dirty_pages_ratelimited(
3807                                    root->fs_info->btree_inode->i_mapping);
3808         }
3809         return;
3810 }
3811
3812 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3813 {
3814         __btrfs_btree_balance_dirty(root, 1);
3815 }
3816
3817 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3818 {
3819         __btrfs_btree_balance_dirty(root, 0);
3820 }
3821
3822 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3823 {
3824         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3825         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3826 }
3827
3828 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3829                               int read_only)
3830 {
3831         /*
3832          * Placeholder for checks
3833          */
3834         return 0;
3835 }
3836
3837 static void btrfs_error_commit_super(struct btrfs_root *root)
3838 {
3839         mutex_lock(&root->fs_info->cleaner_mutex);
3840         btrfs_run_delayed_iputs(root);
3841         mutex_unlock(&root->fs_info->cleaner_mutex);
3842
3843         down_write(&root->fs_info->cleanup_work_sem);
3844         up_write(&root->fs_info->cleanup_work_sem);
3845
3846         /* cleanup FS via transaction */
3847         btrfs_cleanup_transaction(root);
3848 }
3849
3850 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3851 {
3852         struct btrfs_ordered_extent *ordered;
3853
3854         spin_lock(&root->ordered_extent_lock);
3855         /*
3856          * This will just short circuit the ordered completion stuff which will
3857          * make sure the ordered extent gets properly cleaned up.
3858          */
3859         list_for_each_entry(ordered, &root->ordered_extents,
3860                             root_extent_list)
3861                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3862         spin_unlock(&root->ordered_extent_lock);
3863 }
3864
3865 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3866 {
3867         struct btrfs_root *root;
3868         struct list_head splice;
3869
3870         INIT_LIST_HEAD(&splice);
3871
3872         spin_lock(&fs_info->ordered_root_lock);
3873         list_splice_init(&fs_info->ordered_roots, &splice);
3874         while (!list_empty(&splice)) {
3875                 root = list_first_entry(&splice, struct btrfs_root,
3876                                         ordered_root);
3877                 list_move_tail(&root->ordered_root,
3878                                &fs_info->ordered_roots);
3879
3880                 spin_unlock(&fs_info->ordered_root_lock);
3881                 btrfs_destroy_ordered_extents(root);
3882
3883                 cond_resched();
3884                 spin_lock(&fs_info->ordered_root_lock);
3885         }
3886         spin_unlock(&fs_info->ordered_root_lock);
3887 }
3888
3889 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3890                                       struct btrfs_root *root)
3891 {
3892         struct rb_node *node;
3893         struct btrfs_delayed_ref_root *delayed_refs;
3894         struct btrfs_delayed_ref_node *ref;
3895         int ret = 0;
3896
3897         delayed_refs = &trans->delayed_refs;
3898
3899         spin_lock(&delayed_refs->lock);
3900         if (atomic_read(&delayed_refs->num_entries) == 0) {
3901                 spin_unlock(&delayed_refs->lock);
3902                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3903                 return ret;
3904         }
3905
3906         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3907                 struct btrfs_delayed_ref_head *head;
3908                 bool pin_bytes = false;
3909
3910                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3911                                 href_node);
3912                 if (!mutex_trylock(&head->mutex)) {
3913                         atomic_inc(&head->node.refs);
3914                         spin_unlock(&delayed_refs->lock);
3915
3916                         mutex_lock(&head->mutex);
3917                         mutex_unlock(&head->mutex);
3918                         btrfs_put_delayed_ref(&head->node);
3919                         spin_lock(&delayed_refs->lock);
3920                         continue;
3921                 }
3922                 spin_lock(&head->lock);
3923                 while ((node = rb_first(&head->ref_root)) != NULL) {
3924                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3925                                        rb_node);
3926                         ref->in_tree = 0;
3927                         rb_erase(&ref->rb_node, &head->ref_root);
3928                         atomic_dec(&delayed_refs->num_entries);
3929                         btrfs_put_delayed_ref(ref);
3930                 }
3931                 if (head->must_insert_reserved)
3932                         pin_bytes = true;
3933                 btrfs_free_delayed_extent_op(head->extent_op);
3934                 delayed_refs->num_heads--;
3935                 if (head->processing == 0)
3936                         delayed_refs->num_heads_ready--;
3937                 atomic_dec(&delayed_refs->num_entries);
3938                 head->node.in_tree = 0;
3939                 rb_erase(&head->href_node, &delayed_refs->href_root);
3940                 spin_unlock(&head->lock);
3941                 spin_unlock(&delayed_refs->lock);
3942                 mutex_unlock(&head->mutex);
3943
3944                 if (pin_bytes)
3945                         btrfs_pin_extent(root, head->node.bytenr,
3946                                          head->node.num_bytes, 1);
3947                 btrfs_put_delayed_ref(&head->node);
3948                 cond_resched();
3949                 spin_lock(&delayed_refs->lock);
3950         }
3951
3952         spin_unlock(&delayed_refs->lock);
3953
3954         return ret;
3955 }
3956
3957 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3958 {
3959         struct btrfs_inode *btrfs_inode;
3960         struct list_head splice;
3961
3962         INIT_LIST_HEAD(&splice);
3963
3964         spin_lock(&root->delalloc_lock);
3965         list_splice_init(&root->delalloc_inodes, &splice);
3966
3967         while (!list_empty(&splice)) {
3968                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3969                                                delalloc_inodes);
3970
3971                 list_del_init(&btrfs_inode->delalloc_inodes);
3972                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3973                           &btrfs_inode->runtime_flags);
3974                 spin_unlock(&root->delalloc_lock);
3975
3976                 btrfs_invalidate_inodes(btrfs_inode->root);
3977
3978                 spin_lock(&root->delalloc_lock);
3979         }
3980
3981         spin_unlock(&root->delalloc_lock);
3982 }
3983
3984 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3985 {
3986         struct btrfs_root *root;
3987         struct list_head splice;
3988
3989         INIT_LIST_HEAD(&splice);
3990
3991         spin_lock(&fs_info->delalloc_root_lock);
3992         list_splice_init(&fs_info->delalloc_roots, &splice);
3993         while (!list_empty(&splice)) {
3994                 root = list_first_entry(&splice, struct btrfs_root,
3995                                          delalloc_root);
3996                 list_del_init(&root->delalloc_root);
3997                 root = btrfs_grab_fs_root(root);
3998                 BUG_ON(!root);
3999                 spin_unlock(&fs_info->delalloc_root_lock);
4000
4001                 btrfs_destroy_delalloc_inodes(root);
4002                 btrfs_put_fs_root(root);
4003
4004                 spin_lock(&fs_info->delalloc_root_lock);
4005         }
4006         spin_unlock(&fs_info->delalloc_root_lock);
4007 }
4008
4009 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4010                                         struct extent_io_tree *dirty_pages,
4011                                         int mark)
4012 {
4013         int ret;
4014         struct extent_buffer *eb;
4015         u64 start = 0;
4016         u64 end;
4017
4018         while (1) {
4019                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4020                                             mark, NULL);
4021                 if (ret)
4022                         break;
4023
4024                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4025                 while (start <= end) {
4026                         eb = btrfs_find_tree_block(root, start,
4027                                                    root->nodesize);
4028                         start += root->nodesize;
4029                         if (!eb)
4030                                 continue;
4031                         wait_on_extent_buffer_writeback(eb);
4032
4033                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4034                                                &eb->bflags))
4035                                 clear_extent_buffer_dirty(eb);
4036                         free_extent_buffer_stale(eb);
4037                 }
4038         }
4039
4040         return ret;
4041 }
4042
4043 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4044                                        struct extent_io_tree *pinned_extents)
4045 {
4046         struct extent_io_tree *unpin;
4047         u64 start;
4048         u64 end;
4049         int ret;
4050         bool loop = true;
4051
4052         unpin = pinned_extents;
4053 again:
4054         while (1) {
4055                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4056                                             EXTENT_DIRTY, NULL);
4057                 if (ret)
4058                         break;
4059
4060                 /* opt_discard */
4061                 if (btrfs_test_opt(root, DISCARD))
4062                         ret = btrfs_error_discard_extent(root, start,
4063                                                          end + 1 - start,
4064                                                          NULL);
4065
4066                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4067                 btrfs_error_unpin_extent_range(root, start, end);
4068                 cond_resched();
4069         }
4070
4071         if (loop) {
4072                 if (unpin == &root->fs_info->freed_extents[0])
4073                         unpin = &root->fs_info->freed_extents[1];
4074                 else
4075                         unpin = &root->fs_info->freed_extents[0];
4076                 loop = false;
4077                 goto again;
4078         }
4079
4080         return 0;
4081 }
4082
4083 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4084                                    struct btrfs_root *root)
4085 {
4086         btrfs_destroy_delayed_refs(cur_trans, root);
4087
4088         cur_trans->state = TRANS_STATE_COMMIT_START;
4089         wake_up(&root->fs_info->transaction_blocked_wait);
4090
4091         cur_trans->state = TRANS_STATE_UNBLOCKED;
4092         wake_up(&root->fs_info->transaction_wait);
4093
4094         btrfs_destroy_delayed_inodes(root);
4095         btrfs_assert_delayed_root_empty(root);
4096
4097         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4098                                      EXTENT_DIRTY);
4099         btrfs_destroy_pinned_extent(root,
4100                                     root->fs_info->pinned_extents);
4101
4102         cur_trans->state =TRANS_STATE_COMPLETED;
4103         wake_up(&cur_trans->commit_wait);
4104
4105         /*
4106         memset(cur_trans, 0, sizeof(*cur_trans));
4107         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4108         */
4109 }
4110
4111 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4112 {
4113         struct btrfs_transaction *t;
4114
4115         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4116
4117         spin_lock(&root->fs_info->trans_lock);
4118         while (!list_empty(&root->fs_info->trans_list)) {
4119                 t = list_first_entry(&root->fs_info->trans_list,
4120                                      struct btrfs_transaction, list);
4121                 if (t->state >= TRANS_STATE_COMMIT_START) {
4122                         atomic_inc(&t->use_count);
4123                         spin_unlock(&root->fs_info->trans_lock);
4124                         btrfs_wait_for_commit(root, t->transid);
4125                         btrfs_put_transaction(t);
4126                         spin_lock(&root->fs_info->trans_lock);
4127                         continue;
4128                 }
4129                 if (t == root->fs_info->running_transaction) {
4130                         t->state = TRANS_STATE_COMMIT_DOING;
4131                         spin_unlock(&root->fs_info->trans_lock);
4132                         /*
4133                          * We wait for 0 num_writers since we don't hold a trans
4134                          * handle open currently for this transaction.
4135                          */
4136                         wait_event(t->writer_wait,
4137                                    atomic_read(&t->num_writers) == 0);
4138                 } else {
4139                         spin_unlock(&root->fs_info->trans_lock);
4140                 }
4141                 btrfs_cleanup_one_transaction(t, root);
4142
4143                 spin_lock(&root->fs_info->trans_lock);
4144                 if (t == root->fs_info->running_transaction)
4145                         root->fs_info->running_transaction = NULL;
4146                 list_del_init(&t->list);
4147                 spin_unlock(&root->fs_info->trans_lock);
4148
4149                 btrfs_put_transaction(t);
4150                 trace_btrfs_transaction_commit(root);
4151                 spin_lock(&root->fs_info->trans_lock);
4152         }
4153         spin_unlock(&root->fs_info->trans_lock);
4154         btrfs_destroy_all_ordered_extents(root->fs_info);
4155         btrfs_destroy_delayed_inodes(root);
4156         btrfs_assert_delayed_root_empty(root);
4157         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4158         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4159         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4160
4161         return 0;
4162 }
4163
4164 static struct extent_io_ops btree_extent_io_ops = {
4165         .readpage_end_io_hook = btree_readpage_end_io_hook,
4166         .readpage_io_failed_hook = btree_io_failed_hook,
4167         .submit_bio_hook = btree_submit_bio_hook,
4168         /* note we're sharing with inode.c for the merge bio hook */
4169         .merge_bio_hook = btrfs_merge_bio_hook,
4170 };