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 "free-space-tree.h"
  46 #include "inode-map.h"
  47 #include "check-integrity.h"
  48 #include "rcu-string.h"
  49 #include "dev-replace.h"
  50 #include "raid56.h"
  51 #include "sysfs.h"
  52 #include "qgroup.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68                                     int read_only);
  69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71                                       struct btrfs_root *root);
  72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  74                                         struct extent_io_tree *dirty_pages,
  75                                         int mark);
  76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  77                                        struct extent_io_tree *pinned_extents);
  78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  79 static void btrfs_error_commit_super(struct btrfs_root *root);
  80
  81 /*
  82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83  * is complete.  This is used during reads to verify checksums, and it is used
  84  * by writes to insert metadata for new file extents after IO is complete.
  85  */
  86 struct btrfs_end_io_wq {
  87         struct bio *bio;
  88         bio_end_io_t *end_io;
  89         void *private;
  90         struct btrfs_fs_info *info;
  91         int error;
  92         enum btrfs_wq_endio_type metadata;
  93         struct list_head list;
  94         struct btrfs_work work;
  95 };
  96
  97 static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99 int __init btrfs_end_io_wq_init(void)
 100 {
 101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102                                         sizeof(struct btrfs_end_io_wq),
 103                                         0,
 104                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         kmem_cache_destroy(btrfs_end_io_wq_cache);
 114 }
 115
 116 /*
 117  * async submit bios are used to offload expensive checksumming
 118  * onto the worker threads.  They checksum file and metadata bios
 119  * just before they are sent down the IO stack.
 120  */
 121 struct async_submit_bio {
 122         struct inode *inode;
 123         struct bio *bio;
 124         struct list_head list;
 125         extent_submit_bio_hook_t *submit_bio_start;
 126         extent_submit_bio_hook_t *submit_bio_done;
 127         int rw;
 128         int mirror_num;
 129         unsigned long bio_flags;
 130         /*
 131          * bio_offset is optional, can be used if the pages in the bio
 132          * can't tell us where in the file the bio should go
 133          */
 134         u64 bio_offset;
 135         struct btrfs_work work;
 136         int error;
 137 };
 138
 139 /*
 140  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 141  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 142  * the level the eb occupies in the tree.
 143  *
 144  * Different roots are used for different purposes and may nest inside each
 145  * other and they require separate keysets.  As lockdep keys should be
 146  * static, assign keysets according to the purpose of the root as indicated
 147  * by btrfs_root->objectid.  This ensures that all special purpose roots
 148  * have separate keysets.
 149  *
 150  * Lock-nesting across peer nodes is always done with the immediate parent
 151  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 152  * subclass to avoid triggering lockdep warning in such cases.
 153  *
 154  * The key is set by the readpage_end_io_hook after the buffer has passed
 155  * csum validation but before the pages are unlocked.  It is also set by
 156  * btrfs_init_new_buffer on freshly allocated blocks.
 157  *
 158  * We also add a check to make sure the highest level of the tree is the
 159  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 160  * needs update as well.
 161  */
 162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 163 # if BTRFS_MAX_LEVEL != 8
 164 #  error
 165 # endif
 166
 167 static struct btrfs_lockdep_keyset {
 168         u64                     id;             /* root objectid */
 169         const char              *name_stem;     /* lock name stem */
 170         char                    names[BTRFS_MAX_LEVEL + 1][20];
 171         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 172 } btrfs_lockdep_keysets[] = {
 173         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 174         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 175         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 176         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 177         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 178         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 179         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 180         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 181         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 182         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 183         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 184         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 185         { .id = 0,                              .name_stem = "tree"     },
 186 };
 187
 188 void __init btrfs_init_lockdep(void)
 189 {
 190         int i, j;
 191
 192         /* initialize lockdep class names */
 193         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 194                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 195
 196                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 197                         snprintf(ks->names[j], sizeof(ks->names[j]),
 198                                  "btrfs-%s-%02d", ks->name_stem, j);
 199         }
 200 }
 201
 202 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 203                                     int level)
 204 {
 205         struct btrfs_lockdep_keyset *ks;
 206
 207         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 208
 209         /* find the matching keyset, id 0 is the default entry */
 210         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 211                 if (ks->id == objectid)
 212                         break;
 213
 214         lockdep_set_class_and_name(&eb->lock,
 215                                    &ks->keys[level], ks->names[level]);
 216 }
 217
 218 #endif
 219
 220 /*
 221  * extents on the btree inode are pretty simple, there's one extent
 222  * that covers the entire device
 223  */
 224 static struct extent_map *btree_get_extent(struct inode *inode,
 225                 struct page *page, size_t pg_offset, u64 start, u64 len,
 226                 int create)
 227 {
 228         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 229         struct extent_map *em;
 230         int ret;
 231
 232         read_lock(&em_tree->lock);
 233         em = lookup_extent_mapping(em_tree, start, len);
 234         if (em) {
 235                 em->bdev =
 236                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 237                 read_unlock(&em_tree->lock);
 238                 goto out;
 239         }
 240         read_unlock(&em_tree->lock);
 241
 242         em = alloc_extent_map();
 243         if (!em) {
 244                 em = ERR_PTR(-ENOMEM);
 245                 goto out;
 246         }
 247         em->start = 0;
 248         em->len = (u64)-1;
 249         em->block_len = (u64)-1;
 250         em->block_start = 0;
 251         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 252
 253         write_lock(&em_tree->lock);
 254         ret = add_extent_mapping(em_tree, em, 0);
 255         if (ret == -EEXIST) {
 256                 free_extent_map(em);
 257                 em = lookup_extent_mapping(em_tree, start, len);
 258                 if (!em)
 259                         em = ERR_PTR(-EIO);
 260         } else if (ret) {
 261                 free_extent_map(em);
 262                 em = ERR_PTR(ret);
 263         }
 264         write_unlock(&em_tree->lock);
 265
 266 out:
 267         return em;
 268 }
 269
 270 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 271 {
 272         return btrfs_crc32c(seed, data, len);
 273 }
 274
 275 void btrfs_csum_final(u32 crc, char *result)
 276 {
 277         put_unaligned_le32(~crc, result);
 278 }
 279
 280 /*
 281  * compute the csum for a btree block, and either verify it or write it
 282  * into the csum field of the block.
 283  */
 284 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 285                            struct extent_buffer *buf,
 286                            int verify)
 287 {
 288         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 289         char *result = NULL;
 290         unsigned long len;
 291         unsigned long cur_len;
 292         unsigned long offset = BTRFS_CSUM_SIZE;
 293         char *kaddr;
 294         unsigned long map_start;
 295         unsigned long map_len;
 296         int err;
 297         u32 crc = ~(u32)0;
 298         unsigned long inline_result;
 299
 300         len = buf->len - offset;
 301         while (len > 0) {
 302                 err = map_private_extent_buffer(buf, offset, 32,
 303                                         &kaddr, &map_start, &map_len);
 304                 if (err)
 305                         return 1;
 306                 cur_len = min(len, map_len - (offset - map_start));
 307                 crc = btrfs_csum_data(kaddr + offset - map_start,
 308                                       crc, cur_len);
 309                 len -= cur_len;
 310                 offset += cur_len;
 311         }
 312         if (csum_size > sizeof(inline_result)) {
 313                 result = kzalloc(csum_size, GFP_NOFS);
 314                 if (!result)
 315                         return 1;
 316         } else {
 317                 result = (char *)&inline_result;
 318         }
 319
 320         btrfs_csum_final(crc, result);
 321
 322         if (verify) {
 323                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 324                         u32 val;
 325                         u32 found = 0;
 326                         memcpy(&found, result, csum_size);
 327
 328                         read_extent_buffer(buf, &val, 0, csum_size);
 329                         btrfs_warn_rl(fs_info,
 330                                 "%s checksum verify failed on %llu wanted %X found %X "
 331                                 "level %d",
 332                                 fs_info->sb->s_id, buf->start,
 333                                 val, found, btrfs_header_level(buf));
 334                         if (result != (char *)&inline_result)
 335                                 kfree(result);
 336                         return 1;
 337                 }
 338         } else {
 339                 write_extent_buffer(buf, result, 0, csum_size);
 340         }
 341         if (result != (char *)&inline_result)
 342                 kfree(result);
 343         return 0;
 344 }
 345
 346 /*
 347  * we can't consider a given block up to date unless the transid of the
 348  * block matches the transid in the parent node's pointer.  This is how we
 349  * detect blocks that either didn't get written at all or got written
 350  * in the wrong place.
 351  */
 352 static int verify_parent_transid(struct extent_io_tree *io_tree,
 353                                  struct extent_buffer *eb, u64 parent_transid,
 354                                  int atomic)
 355 {
 356         struct extent_state *cached_state = NULL;
 357         int ret;
 358         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 359
 360         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 361                 return 0;
 362
 363         if (atomic)
 364                 return -EAGAIN;
 365
 366         if (need_lock) {
 367                 btrfs_tree_read_lock(eb);
 368                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 369         }
 370
 371         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 372                          &cached_state);
 373         if (extent_buffer_uptodate(eb) &&
 374             btrfs_header_generation(eb) == parent_transid) {
 375                 ret = 0;
 376                 goto out;
 377         }
 378         btrfs_err_rl(eb->fs_info,
 379                 "parent transid verify failed on %llu wanted %llu found %llu",
 380                         eb->start,
 381                         parent_transid, btrfs_header_generation(eb));
 382         ret = 1;
 383
 384         /*
 385          * Things reading via commit roots that don't have normal protection,
 386          * like send, can have a really old block in cache that may point at a
 387          * block that has been free'd and re-allocated.  So don't clear uptodate
 388          * if we find an eb that is under IO (dirty/writeback) because we could
 389          * end up reading in the stale data and then writing it back out and
 390          * making everybody very sad.
 391          */
 392         if (!extent_buffer_under_io(eb))
 393                 clear_extent_buffer_uptodate(eb);
 394 out:
 395         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 396                              &cached_state, GFP_NOFS);
 397         if (need_lock)
 398                 btrfs_tree_read_unlock_blocking(eb);
 399         return ret;
 400 }
 401
 402 /*
 403  * Return 0 if the superblock checksum type matches the checksum value of that
 404  * algorithm. Pass the raw disk superblock data.
 405  */
 406 static int btrfs_check_super_csum(char *raw_disk_sb)
 407 {
 408         struct btrfs_super_block *disk_sb =
 409                 (struct btrfs_super_block *)raw_disk_sb;
 410         u16 csum_type = btrfs_super_csum_type(disk_sb);
 411         int ret = 0;
 412
 413         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 414                 u32 crc = ~(u32)0;
 415                 const int csum_size = sizeof(crc);
 416                 char result[csum_size];
 417
 418                 /*
 419                  * The super_block structure does not span the whole
 420                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 421                  * is filled with zeros and is included in the checkum.
 422                  */
 423                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 424                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 425                 btrfs_csum_final(crc, result);
 426
 427                 if (memcmp(raw_disk_sb, result, csum_size))
 428                         ret = 1;
 429         }
 430
 431         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 432                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 433                                 csum_type);
 434                 ret = 1;
 435         }
 436
 437         return ret;
 438 }
 439
 440 /*
 441  * helper to read a given tree block, doing retries as required when
 442  * the checksums don't match and we have alternate mirrors to try.
 443  */
 444 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 445                                           struct extent_buffer *eb,
 446                                           u64 start, u64 parent_transid)
 447 {
 448         struct extent_io_tree *io_tree;
 449         int failed = 0;
 450         int ret;
 451         int num_copies = 0;
 452         int mirror_num = 0;
 453         int failed_mirror = 0;
 454
 455         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 456         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 457         while (1) {
 458                 ret = read_extent_buffer_pages(io_tree, eb, start,
 459                                                WAIT_COMPLETE,
 460                                                btree_get_extent, mirror_num);
 461                 if (!ret) {
 462                         if (!verify_parent_transid(io_tree, eb,
 463                                                    parent_transid, 0))
 464                                 break;
 465                         else
 466                                 ret = -EIO;
 467                 }
 468
 469                 /*
 470                  * This buffer's crc is fine, but its contents are corrupted, so
 471                  * there is no reason to read the other copies, they won't be
 472                  * any less wrong.
 473                  */
 474                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 475                         break;
 476
 477                 num_copies = btrfs_num_copies(root->fs_info,
 478                                               eb->start, eb->len);
 479                 if (num_copies == 1)
 480                         break;
 481
 482                 if (!failed_mirror) {
 483                         failed = 1;
 484                         failed_mirror = eb->read_mirror;
 485                 }
 486
 487                 mirror_num++;
 488                 if (mirror_num == failed_mirror)
 489                         mirror_num++;
 490
 491                 if (mirror_num > num_copies)
 492                         break;
 493         }
 494
 495         if (failed && !ret && failed_mirror)
 496                 repair_eb_io_failure(root, eb, failed_mirror);
 497
 498         return ret;
 499 }
 500
 501 /*
 502  * checksum a dirty tree block before IO.  This has extra checks to make sure
 503  * we only fill in the checksum field in the first page of a multi-page block
 504  */
 505
 506 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 507 {
 508         u64 start = page_offset(page);
 509         u64 found_start;
 510         struct extent_buffer *eb;
 511
 512         eb = (struct extent_buffer *)page->private;
 513         if (page != eb->pages[0])
 514                 return 0;
 515         found_start = btrfs_header_bytenr(eb);
 516         if (WARN_ON(found_start != start || !PageUptodate(page)))
 517                 return 0;
 518         csum_tree_block(fs_info, eb, 0);
 519         return 0;
 520 }
 521
 522 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 523                                  struct extent_buffer *eb)
 524 {
 525         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 526         u8 fsid[BTRFS_UUID_SIZE];
 527         int ret = 1;
 528
 529         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 530         while (fs_devices) {
 531                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 532                         ret = 0;
 533                         break;
 534                 }
 535                 fs_devices = fs_devices->seed;
 536         }
 537         return ret;
 538 }
 539
 540 #define CORRUPT(reason, eb, root, slot)                         \
 541         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 542                    "root=%llu, slot=%d", reason,                        \
 543                btrfs_header_bytenr(eb), root->objectid, slot)
 544
 545 static noinline int check_leaf(struct btrfs_root *root,
 546                                struct extent_buffer *leaf)
 547 {
 548         struct btrfs_key key;
 549         struct btrfs_key leaf_key;
 550         u32 nritems = btrfs_header_nritems(leaf);
 551         int slot;
 552
 553         if (nritems == 0)
 554                 return 0;
 555
 556         /* Check the 0 item */
 557         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 558             BTRFS_LEAF_DATA_SIZE(root)) {
 559                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 560                 return -EIO;
 561         }
 562
 563         /*
 564          * Check to make sure each items keys are in the correct order and their
 565          * offsets make sense.  We only have to loop through nritems-1 because
 566          * we check the current slot against the next slot, which verifies the
 567          * next slot's offset+size makes sense and that the current's slot
 568          * offset is correct.
 569          */
 570         for (slot = 0; slot < nritems - 1; slot++) {
 571                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 572                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 573
 574                 /* Make sure the keys are in the right order */
 575                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 576                         CORRUPT("bad key order", leaf, root, slot);
 577                         return -EIO;
 578                 }
 579
 580                 /*
 581                  * Make sure the offset and ends are right, remember that the
 582                  * item data starts at the end of the leaf and grows towards the
 583                  * front.
 584                  */
 585                 if (btrfs_item_offset_nr(leaf, slot) !=
 586                         btrfs_item_end_nr(leaf, slot + 1)) {
 587                         CORRUPT("slot offset bad", leaf, root, slot);
 588                         return -EIO;
 589                 }
 590
 591                 /*
 592                  * Check to make sure that we don't point outside of the leaf,
 593                  * just incase all the items are consistent to eachother, but
 594                  * all point outside of the leaf.
 595                  */
 596                 if (btrfs_item_end_nr(leaf, slot) >
 597                     BTRFS_LEAF_DATA_SIZE(root)) {
 598                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 599                         return -EIO;
 600                 }
 601         }
 602
 603         return 0;
 604 }
 605
 606 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 607                                       u64 phy_offset, struct page *page,
 608                                       u64 start, u64 end, int mirror)
 609 {
 610         u64 found_start;
 611         int found_level;
 612         struct extent_buffer *eb;
 613         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 614         struct btrfs_fs_info *fs_info = root->fs_info;
 615         int ret = 0;
 616         int reads_done;
 617
 618         if (!page->private)
 619                 goto out;
 620
 621         eb = (struct extent_buffer *)page->private;
 622
 623         /* the pending IO might have been the only thing that kept this buffer
 624          * in memory.  Make sure we have a ref for all this other checks
 625          */
 626         extent_buffer_get(eb);
 627
 628         reads_done = atomic_dec_and_test(&eb->io_pages);
 629         if (!reads_done)
 630                 goto err;
 631
 632         eb->read_mirror = mirror;
 633         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 634                 ret = -EIO;
 635                 goto err;
 636         }
 637
 638         found_start = btrfs_header_bytenr(eb);
 639         if (found_start != eb->start) {
 640                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 641                              found_start, eb->start);
 642                 ret = -EIO;
 643                 goto err;
 644         }
 645         if (check_tree_block_fsid(fs_info, eb)) {
 646                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 647                              eb->start);
 648                 ret = -EIO;
 649                 goto err;
 650         }
 651         found_level = btrfs_header_level(eb);
 652         if (found_level >= BTRFS_MAX_LEVEL) {
 653                 btrfs_err(fs_info, "bad tree block level %d",
 654                           (int)btrfs_header_level(eb));
 655                 ret = -EIO;
 656                 goto err;
 657         }
 658
 659         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 660                                        eb, found_level);
 661
 662         ret = csum_tree_block(fs_info, eb, 1);
 663         if (ret) {
 664                 ret = -EIO;
 665                 goto err;
 666         }
 667
 668         /*
 669          * If this is a leaf block and it is corrupt, set the corrupt bit so
 670          * that we don't try and read the other copies of this block, just
 671          * return -EIO.
 672          */
 673         if (found_level == 0 && check_leaf(root, eb)) {
 674                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 675                 ret = -EIO;
 676         }
 677
 678         if (!ret)
 679                 set_extent_buffer_uptodate(eb);
 680 err:
 681         if (reads_done &&
 682             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 683                 btree_readahead_hook(fs_info, eb, eb->start, ret);
 684
 685         if (ret) {
 686                 /*
 687                  * our io error hook is going to dec the io pages
 688                  * again, we have to make sure it has something
 689                  * to decrement
 690                  */
 691                 atomic_inc(&eb->io_pages);
 692                 clear_extent_buffer_uptodate(eb);
 693         }
 694         free_extent_buffer(eb);
 695 out:
 696         return ret;
 697 }
 698
 699 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 700 {
 701         struct extent_buffer *eb;
 702
 703         eb = (struct extent_buffer *)page->private;
 704         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 705         eb->read_mirror = failed_mirror;
 706         atomic_dec(&eb->io_pages);
 707         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 708                 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
 709         return -EIO;    /* we fixed nothing */
 710 }
 711
 712 static void end_workqueue_bio(struct bio *bio)
 713 {
 714         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 715         struct btrfs_fs_info *fs_info;
 716         struct btrfs_workqueue *wq;
 717         btrfs_work_func_t func;
 718
 719         fs_info = end_io_wq->info;
 720         end_io_wq->error = bio->bi_error;
 721
 722         if (bio->bi_rw & REQ_WRITE) {
 723                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 724                         wq = fs_info->endio_meta_write_workers;
 725                         func = btrfs_endio_meta_write_helper;
 726                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 727                         wq = fs_info->endio_freespace_worker;
 728                         func = btrfs_freespace_write_helper;
 729                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 730                         wq = fs_info->endio_raid56_workers;
 731                         func = btrfs_endio_raid56_helper;
 732                 } else {
 733                         wq = fs_info->endio_write_workers;
 734                         func = btrfs_endio_write_helper;
 735                 }
 736         } else {
 737                 if (unlikely(end_io_wq->metadata ==
 738                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 739                         wq = fs_info->endio_repair_workers;
 740                         func = btrfs_endio_repair_helper;
 741                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 742                         wq = fs_info->endio_raid56_workers;
 743                         func = btrfs_endio_raid56_helper;
 744                 } else if (end_io_wq->metadata) {
 745                         wq = fs_info->endio_meta_workers;
 746                         func = btrfs_endio_meta_helper;
 747                 } else {
 748                         wq = fs_info->endio_workers;
 749                         func = btrfs_endio_helper;
 750                 }
 751         }
 752
 753         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 754         btrfs_queue_work(wq, &end_io_wq->work);
 755 }
 756
 757 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 758                         enum btrfs_wq_endio_type metadata)
 759 {
 760         struct btrfs_end_io_wq *end_io_wq;
 761
 762         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 763         if (!end_io_wq)
 764                 return -ENOMEM;
 765
 766         end_io_wq->private = bio->bi_private;
 767         end_io_wq->end_io = bio->bi_end_io;
 768         end_io_wq->info = info;
 769         end_io_wq->error = 0;
 770         end_io_wq->bio = bio;
 771         end_io_wq->metadata = metadata;
 772
 773         bio->bi_private = end_io_wq;
 774         bio->bi_end_io = end_workqueue_bio;
 775         return 0;
 776 }
 777
 778 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 779 {
 780         unsigned long limit = min_t(unsigned long,
 781                                     info->thread_pool_size,
 782                                     info->fs_devices->open_devices);
 783         return 256 * limit;
 784 }
 785
 786 static void run_one_async_start(struct btrfs_work *work)
 787 {
 788         struct async_submit_bio *async;
 789         int ret;
 790
 791         async = container_of(work, struct  async_submit_bio, work);
 792         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 793                                       async->mirror_num, async->bio_flags,
 794                                       async->bio_offset);
 795         if (ret)
 796                 async->error = ret;
 797 }
 798
 799 static void run_one_async_done(struct btrfs_work *work)
 800 {
 801         struct btrfs_fs_info *fs_info;
 802         struct async_submit_bio *async;
 803         int limit;
 804
 805         async = container_of(work, struct  async_submit_bio, work);
 806         fs_info = BTRFS_I(async->inode)->root->fs_info;
 807
 808         limit = btrfs_async_submit_limit(fs_info);
 809         limit = limit * 2 / 3;
 810
 811         /*
 812          * atomic_dec_return implies a barrier for waitqueue_active
 813          */
 814         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 815             waitqueue_active(&fs_info->async_submit_wait))
 816                 wake_up(&fs_info->async_submit_wait);
 817
 818         /* If an error occured we just want to clean up the bio and move on */
 819         if (async->error) {
 820                 async->bio->bi_error = async->error;
 821                 bio_endio(async->bio);
 822                 return;
 823         }
 824
 825         async->submit_bio_done(async->inode, async->rw, async->bio,
 826                                async->mirror_num, async->bio_flags,
 827                                async->bio_offset);
 828 }
 829
 830 static void run_one_async_free(struct btrfs_work *work)
 831 {
 832         struct async_submit_bio *async;
 833
 834         async = container_of(work, struct  async_submit_bio, work);
 835         kfree(async);
 836 }
 837
 838 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 839                         int rw, struct bio *bio, int mirror_num,
 840                         unsigned long bio_flags,
 841                         u64 bio_offset,
 842                         extent_submit_bio_hook_t *submit_bio_start,
 843                         extent_submit_bio_hook_t *submit_bio_done)
 844 {
 845         struct async_submit_bio *async;
 846
 847         async = kmalloc(sizeof(*async), GFP_NOFS);
 848         if (!async)
 849                 return -ENOMEM;
 850
 851         async->inode = inode;
 852         async->rw = rw;
 853         async->bio = bio;
 854         async->mirror_num = mirror_num;
 855         async->submit_bio_start = submit_bio_start;
 856         async->submit_bio_done = submit_bio_done;
 857
 858         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 859                         run_one_async_done, run_one_async_free);
 860
 861         async->bio_flags = bio_flags;
 862         async->bio_offset = bio_offset;
 863
 864         async->error = 0;
 865
 866         atomic_inc(&fs_info->nr_async_submits);
 867
 868         if (rw & REQ_SYNC)
 869                 btrfs_set_work_high_priority(&async->work);
 870
 871         btrfs_queue_work(fs_info->workers, &async->work);
 872
 873         while (atomic_read(&fs_info->async_submit_draining) &&
 874               atomic_read(&fs_info->nr_async_submits)) {
 875                 wait_event(fs_info->async_submit_wait,
 876                            (atomic_read(&fs_info->nr_async_submits) == 0));
 877         }
 878
 879         return 0;
 880 }
 881
 882 static int btree_csum_one_bio(struct bio *bio)
 883 {
 884         struct bio_vec *bvec;
 885         struct btrfs_root *root;
 886         int i, ret = 0;
 887
 888         bio_for_each_segment_all(bvec, bio, i) {
 889                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 890                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 891                 if (ret)
 892                         break;
 893         }
 894
 895         return ret;
 896 }
 897
 898 static int __btree_submit_bio_start(struct inode *inode, int rw,
 899                                     struct bio *bio, int mirror_num,
 900                                     unsigned long bio_flags,
 901                                     u64 bio_offset)
 902 {
 903         /*
 904          * when we're called for a write, we're already in the async
 905          * submission context.  Just jump into btrfs_map_bio
 906          */
 907         return btree_csum_one_bio(bio);
 908 }
 909
 910 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 911                                  int mirror_num, unsigned long bio_flags,
 912                                  u64 bio_offset)
 913 {
 914         int ret;
 915
 916         /*
 917          * when we're called for a write, we're already in the async
 918          * submission context.  Just jump into btrfs_map_bio
 919          */
 920         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 921         if (ret) {
 922                 bio->bi_error = ret;
 923                 bio_endio(bio);
 924         }
 925         return ret;
 926 }
 927
 928 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 929 {
 930         if (bio_flags & EXTENT_BIO_TREE_LOG)
 931                 return 0;
 932 #ifdef CONFIG_X86
 933         if (static_cpu_has_safe(X86_FEATURE_XMM4_2))
 934                 return 0;
 935 #endif
 936         return 1;
 937 }
 938
 939 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 940                                  int mirror_num, unsigned long bio_flags,
 941                                  u64 bio_offset)
 942 {
 943         int async = check_async_write(inode, bio_flags);
 944         int ret;
 945
 946         if (!(rw & REQ_WRITE)) {
 947                 /*
 948                  * called for a read, do the setup so that checksum validation
 949                  * can happen in the async kernel threads
 950                  */
 951                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 952                                           bio, BTRFS_WQ_ENDIO_METADATA);
 953                 if (ret)
 954                         goto out_w_error;
 955                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 956                                     mirror_num, 0);
 957         } else if (!async) {
 958                 ret = btree_csum_one_bio(bio);
 959                 if (ret)
 960                         goto out_w_error;
 961                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 962                                     mirror_num, 0);
 963         } else {
 964                 /*
 965                  * kthread helpers are used to submit writes so that
 966                  * checksumming can happen in parallel across all CPUs
 967                  */
 968                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 969                                           inode, rw, bio, mirror_num, 0,
 970                                           bio_offset,
 971                                           __btree_submit_bio_start,
 972                                           __btree_submit_bio_done);
 973         }
 974
 975         if (ret)
 976                 goto out_w_error;
 977         return 0;
 978
 979 out_w_error:
 980         bio->bi_error = ret;
 981         bio_endio(bio);
 982         return ret;
 983 }
 984
 985 #ifdef CONFIG_MIGRATION
 986 static int btree_migratepage(struct address_space *mapping,
 987                         struct page *newpage, struct page *page,
 988                         enum migrate_mode mode)
 989 {
 990         /*
 991          * we can't safely write a btree page from here,
 992          * we haven't done the locking hook
 993          */
 994         if (PageDirty(page))
 995                 return -EAGAIN;
 996         /*
 997          * Buffers may be managed in a filesystem specific way.
 998          * We must have no buffers or drop them.
 999          */
1000         if (page_has_private(page) &&
1001             !try_to_release_page(page, GFP_KERNEL))
1002                 return -EAGAIN;
1003         return migrate_page(mapping, newpage, page, mode);
1004 }
1005 #endif
1006
1007
1008 static int btree_writepages(struct address_space *mapping,
1009                             struct writeback_control *wbc)
1010 {
1011         struct btrfs_fs_info *fs_info;
1012         int ret;
1013
1014         if (wbc->sync_mode == WB_SYNC_NONE) {
1015
1016                 if (wbc->for_kupdate)
1017                         return 0;
1018
1019                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1020                 /* this is a bit racy, but that's ok */
1021                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1022                                              BTRFS_DIRTY_METADATA_THRESH);
1023                 if (ret < 0)
1024                         return 0;
1025         }
1026         return btree_write_cache_pages(mapping, wbc);
1027 }
1028
1029 static int btree_readpage(struct file *file, struct page *page)
1030 {
1031         struct extent_io_tree *tree;
1032         tree = &BTRFS_I(page->mapping->host)->io_tree;
1033         return extent_read_full_page(tree, page, btree_get_extent, 0);
1034 }
1035
1036 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1037 {
1038         if (PageWriteback(page) || PageDirty(page))
1039                 return 0;
1040
1041         return try_release_extent_buffer(page);
1042 }
1043
1044 static void btree_invalidatepage(struct page *page, unsigned int offset,
1045                                  unsigned int length)
1046 {
1047         struct extent_io_tree *tree;
1048         tree = &BTRFS_I(page->mapping->host)->io_tree;
1049         extent_invalidatepage(tree, page, offset);
1050         btree_releasepage(page, GFP_NOFS);
1051         if (PagePrivate(page)) {
1052                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1053                            "page private not zero on page %llu",
1054                            (unsigned long long)page_offset(page));
1055                 ClearPagePrivate(page);
1056                 set_page_private(page, 0);
1057                 page_cache_release(page);
1058         }
1059 }
1060
1061 static int btree_set_page_dirty(struct page *page)
1062 {
1063 #ifdef DEBUG
1064         struct extent_buffer *eb;
1065
1066         BUG_ON(!PagePrivate(page));
1067         eb = (struct extent_buffer *)page->private;
1068         BUG_ON(!eb);
1069         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1070         BUG_ON(!atomic_read(&eb->refs));
1071         btrfs_assert_tree_locked(eb);
1072 #endif
1073         return __set_page_dirty_nobuffers(page);
1074 }
1075
1076 static const struct address_space_operations btree_aops = {
1077         .readpage       = btree_readpage,
1078         .writepages     = btree_writepages,
1079         .releasepage    = btree_releasepage,
1080         .invalidatepage = btree_invalidatepage,
1081 #ifdef CONFIG_MIGRATION
1082         .migratepage    = btree_migratepage,
1083 #endif
1084         .set_page_dirty = btree_set_page_dirty,
1085 };
1086
1087 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1088 {
1089         struct extent_buffer *buf = NULL;
1090         struct inode *btree_inode = root->fs_info->btree_inode;
1091
1092         buf = btrfs_find_create_tree_block(root, bytenr);
1093         if (!buf)
1094                 return;
1095         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1096                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1097         free_extent_buffer(buf);
1098 }
1099
1100 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1101                          int mirror_num, struct extent_buffer **eb)
1102 {
1103         struct extent_buffer *buf = NULL;
1104         struct inode *btree_inode = root->fs_info->btree_inode;
1105         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1106         int ret;
1107
1108         buf = btrfs_find_create_tree_block(root, bytenr);
1109         if (!buf)
1110                 return 0;
1111
1112         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1113
1114         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1115                                        btree_get_extent, mirror_num);
1116         if (ret) {
1117                 free_extent_buffer(buf);
1118                 return ret;
1119         }
1120
1121         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1122                 free_extent_buffer(buf);
1123                 return -EIO;
1124         } else if (extent_buffer_uptodate(buf)) {
1125                 *eb = buf;
1126         } else {
1127                 free_extent_buffer(buf);
1128         }
1129         return 0;
1130 }
1131
1132 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1133                                             u64 bytenr)
1134 {
1135         return find_extent_buffer(fs_info, bytenr);
1136 }
1137
1138 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1139                                                  u64 bytenr)
1140 {
1141         if (btrfs_test_is_dummy_root(root))
1142                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1143         return alloc_extent_buffer(root->fs_info, bytenr);
1144 }
1145
1146
1147 int btrfs_write_tree_block(struct extent_buffer *buf)
1148 {
1149         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1150                                         buf->start + buf->len - 1);
1151 }
1152
1153 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1154 {
1155         return filemap_fdatawait_range(buf->pages[0]->mapping,
1156                                        buf->start, buf->start + buf->len - 1);
1157 }
1158
1159 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1160                                       u64 parent_transid)
1161 {
1162         struct extent_buffer *buf = NULL;
1163         int ret;
1164
1165         buf = btrfs_find_create_tree_block(root, bytenr);
1166         if (!buf)
1167                 return ERR_PTR(-ENOMEM);
1168
1169         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1170         if (ret) {
1171                 free_extent_buffer(buf);
1172                 return ERR_PTR(ret);
1173         }
1174         return buf;
1175
1176 }
1177
1178 void clean_tree_block(struct btrfs_trans_handle *trans,
1179                       struct btrfs_fs_info *fs_info,
1180                       struct extent_buffer *buf)
1181 {
1182         if (btrfs_header_generation(buf) ==
1183             fs_info->running_transaction->transid) {
1184                 btrfs_assert_tree_locked(buf);
1185
1186                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1187                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1188                                              -buf->len,
1189                                              fs_info->dirty_metadata_batch);
1190                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1191                         btrfs_set_lock_blocking(buf);
1192                         clear_extent_buffer_dirty(buf);
1193                 }
1194         }
1195 }
1196
1197 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1198 {
1199         struct btrfs_subvolume_writers *writers;
1200         int ret;
1201
1202         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1203         if (!writers)
1204                 return ERR_PTR(-ENOMEM);
1205
1206         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1207         if (ret < 0) {
1208                 kfree(writers);
1209                 return ERR_PTR(ret);
1210         }
1211
1212         init_waitqueue_head(&writers->wait);
1213         return writers;
1214 }
1215
1216 static void
1217 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1218 {
1219         percpu_counter_destroy(&writers->counter);
1220         kfree(writers);
1221 }
1222
1223 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1224                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1225                          u64 objectid)
1226 {
1227         root->node = NULL;
1228         root->commit_root = NULL;
1229         root->sectorsize = sectorsize;
1230         root->nodesize = nodesize;
1231         root->stripesize = stripesize;
1232         root->state = 0;
1233         root->orphan_cleanup_state = 0;
1234
1235         root->objectid = objectid;
1236         root->last_trans = 0;
1237         root->highest_objectid = 0;
1238         root->nr_delalloc_inodes = 0;
1239         root->nr_ordered_extents = 0;
1240         root->name = NULL;
1241         root->inode_tree = RB_ROOT;
1242         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1243         root->block_rsv = NULL;
1244         root->orphan_block_rsv = NULL;
1245
1246         INIT_LIST_HEAD(&root->dirty_list);
1247         INIT_LIST_HEAD(&root->root_list);
1248         INIT_LIST_HEAD(&root->delalloc_inodes);
1249         INIT_LIST_HEAD(&root->delalloc_root);
1250         INIT_LIST_HEAD(&root->ordered_extents);
1251         INIT_LIST_HEAD(&root->ordered_root);
1252         INIT_LIST_HEAD(&root->logged_list[0]);
1253         INIT_LIST_HEAD(&root->logged_list[1]);
1254         spin_lock_init(&root->orphan_lock);
1255         spin_lock_init(&root->inode_lock);
1256         spin_lock_init(&root->delalloc_lock);
1257         spin_lock_init(&root->ordered_extent_lock);
1258         spin_lock_init(&root->accounting_lock);
1259         spin_lock_init(&root->log_extents_lock[0]);
1260         spin_lock_init(&root->log_extents_lock[1]);
1261         mutex_init(&root->objectid_mutex);
1262         mutex_init(&root->log_mutex);
1263         mutex_init(&root->ordered_extent_mutex);
1264         mutex_init(&root->delalloc_mutex);
1265         init_waitqueue_head(&root->log_writer_wait);
1266         init_waitqueue_head(&root->log_commit_wait[0]);
1267         init_waitqueue_head(&root->log_commit_wait[1]);
1268         INIT_LIST_HEAD(&root->log_ctxs[0]);
1269         INIT_LIST_HEAD(&root->log_ctxs[1]);
1270         atomic_set(&root->log_commit[0], 0);
1271         atomic_set(&root->log_commit[1], 0);
1272         atomic_set(&root->log_writers, 0);
1273         atomic_set(&root->log_batch, 0);
1274         atomic_set(&root->orphan_inodes, 0);
1275         atomic_set(&root->refs, 1);
1276         atomic_set(&root->will_be_snapshoted, 0);
1277         atomic_set(&root->qgroup_meta_rsv, 0);
1278         root->log_transid = 0;
1279         root->log_transid_committed = -1;
1280         root->last_log_commit = 0;
1281         if (fs_info)
1282                 extent_io_tree_init(&root->dirty_log_pages,
1283                                      fs_info->btree_inode->i_mapping);
1284
1285         memset(&root->root_key, 0, sizeof(root->root_key));
1286         memset(&root->root_item, 0, sizeof(root->root_item));
1287         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1288         if (fs_info)
1289                 root->defrag_trans_start = fs_info->generation;
1290         else
1291                 root->defrag_trans_start = 0;
1292         root->root_key.objectid = objectid;
1293         root->anon_dev = 0;
1294
1295         spin_lock_init(&root->root_item_lock);
1296 }
1297
1298 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1299                 gfp_t flags)
1300 {
1301         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1302         if (root)
1303                 root->fs_info = fs_info;
1304         return root;
1305 }
1306
1307 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1308 /* Should only be used by the testing infrastructure */
1309 struct btrfs_root *btrfs_alloc_dummy_root(void)
1310 {
1311         struct btrfs_root *root;
1312
1313         root = btrfs_alloc_root(NULL, GFP_KERNEL);
1314         if (!root)
1315                 return ERR_PTR(-ENOMEM);
1316         __setup_root(4096, 4096, 4096, root, NULL, 1);
1317         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1318         root->alloc_bytenr = 0;
1319
1320         return root;
1321 }
1322 #endif
1323
1324 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1325                                      struct btrfs_fs_info *fs_info,
1326                                      u64 objectid)
1327 {
1328         struct extent_buffer *leaf;
1329         struct btrfs_root *tree_root = fs_info->tree_root;
1330         struct btrfs_root *root;
1331         struct btrfs_key key;
1332         int ret = 0;
1333         uuid_le uuid;
1334
1335         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1336         if (!root)
1337                 return ERR_PTR(-ENOMEM);
1338
1339         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1340                 tree_root->stripesize, root, fs_info, objectid);
1341         root->root_key.objectid = objectid;
1342         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1343         root->root_key.offset = 0;
1344
1345         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1346         if (IS_ERR(leaf)) {
1347                 ret = PTR_ERR(leaf);
1348                 leaf = NULL;
1349                 goto fail;
1350         }
1351
1352         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1353         btrfs_set_header_bytenr(leaf, leaf->start);
1354         btrfs_set_header_generation(leaf, trans->transid);
1355         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1356         btrfs_set_header_owner(leaf, objectid);
1357         root->node = leaf;
1358
1359         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1360                             BTRFS_FSID_SIZE);
1361         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1362                             btrfs_header_chunk_tree_uuid(leaf),
1363                             BTRFS_UUID_SIZE);
1364         btrfs_mark_buffer_dirty(leaf);
1365
1366         root->commit_root = btrfs_root_node(root);
1367         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1368
1369         root->root_item.flags = 0;
1370         root->root_item.byte_limit = 0;
1371         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1372         btrfs_set_root_generation(&root->root_item, trans->transid);
1373         btrfs_set_root_level(&root->root_item, 0);
1374         btrfs_set_root_refs(&root->root_item, 1);
1375         btrfs_set_root_used(&root->root_item, leaf->len);
1376         btrfs_set_root_last_snapshot(&root->root_item, 0);
1377         btrfs_set_root_dirid(&root->root_item, 0);
1378         uuid_le_gen(&uuid);
1379         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1380         root->root_item.drop_level = 0;
1381
1382         key.objectid = objectid;
1383         key.type = BTRFS_ROOT_ITEM_KEY;
1384         key.offset = 0;
1385         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1386         if (ret)
1387                 goto fail;
1388
1389         btrfs_tree_unlock(leaf);
1390
1391         return root;
1392
1393 fail:
1394         if (leaf) {
1395                 btrfs_tree_unlock(leaf);
1396                 free_extent_buffer(root->commit_root);
1397                 free_extent_buffer(leaf);
1398         }
1399         kfree(root);
1400
1401         return ERR_PTR(ret);
1402 }
1403
1404 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1405                                          struct btrfs_fs_info *fs_info)
1406 {
1407         struct btrfs_root *root;
1408         struct btrfs_root *tree_root = fs_info->tree_root;
1409         struct extent_buffer *leaf;
1410
1411         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1412         if (!root)
1413                 return ERR_PTR(-ENOMEM);
1414
1415         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1416                      tree_root->stripesize, root, fs_info,
1417                      BTRFS_TREE_LOG_OBJECTID);
1418
1419         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1420         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1421         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1422
1423         /*
1424          * DON'T set REF_COWS for log trees
1425          *
1426          * log trees do not get reference counted because they go away
1427          * before a real commit is actually done.  They do store pointers
1428          * to file data extents, and those reference counts still get
1429          * updated (along with back refs to the log tree).
1430          */
1431
1432         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1433                         NULL, 0, 0, 0);
1434         if (IS_ERR(leaf)) {
1435                 kfree(root);
1436                 return ERR_CAST(leaf);
1437         }
1438
1439         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1440         btrfs_set_header_bytenr(leaf, leaf->start);
1441         btrfs_set_header_generation(leaf, trans->transid);
1442         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1443         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1444         root->node = leaf;
1445
1446         write_extent_buffer(root->node, root->fs_info->fsid,
1447                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1448         btrfs_mark_buffer_dirty(root->node);
1449         btrfs_tree_unlock(root->node);
1450         return root;
1451 }
1452
1453 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1454                              struct btrfs_fs_info *fs_info)
1455 {
1456         struct btrfs_root *log_root;
1457
1458         log_root = alloc_log_tree(trans, fs_info);
1459         if (IS_ERR(log_root))
1460                 return PTR_ERR(log_root);
1461         WARN_ON(fs_info->log_root_tree);
1462         fs_info->log_root_tree = log_root;
1463         return 0;
1464 }
1465
1466 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1467                        struct btrfs_root *root)
1468 {
1469         struct btrfs_root *log_root;
1470         struct btrfs_inode_item *inode_item;
1471
1472         log_root = alloc_log_tree(trans, root->fs_info);
1473         if (IS_ERR(log_root))
1474                 return PTR_ERR(log_root);
1475
1476         log_root->last_trans = trans->transid;
1477         log_root->root_key.offset = root->root_key.objectid;
1478
1479         inode_item = &log_root->root_item.inode;
1480         btrfs_set_stack_inode_generation(inode_item, 1);
1481         btrfs_set_stack_inode_size(inode_item, 3);
1482         btrfs_set_stack_inode_nlink(inode_item, 1);
1483         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1484         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1485
1486         btrfs_set_root_node(&log_root->root_item, log_root->node);
1487
1488         WARN_ON(root->log_root);
1489         root->log_root = log_root;
1490         root->log_transid = 0;
1491         root->log_transid_committed = -1;
1492         root->last_log_commit = 0;
1493         return 0;
1494 }
1495
1496 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1497                                                struct btrfs_key *key)
1498 {
1499         struct btrfs_root *root;
1500         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1501         struct btrfs_path *path;
1502         u64 generation;
1503         int ret;
1504
1505         path = btrfs_alloc_path();
1506         if (!path)
1507                 return ERR_PTR(-ENOMEM);
1508
1509         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1510         if (!root) {
1511                 ret = -ENOMEM;
1512                 goto alloc_fail;
1513         }
1514
1515         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1516                 tree_root->stripesize, root, fs_info, key->objectid);
1517
1518         ret = btrfs_find_root(tree_root, key, path,
1519                               &root->root_item, &root->root_key);
1520         if (ret) {
1521                 if (ret > 0)
1522                         ret = -ENOENT;
1523                 goto find_fail;
1524         }
1525
1526         generation = btrfs_root_generation(&root->root_item);
1527         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1528                                      generation);
1529         if (IS_ERR(root->node)) {
1530                 ret = PTR_ERR(root->node);
1531                 goto find_fail;
1532         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1533                 ret = -EIO;
1534                 free_extent_buffer(root->node);
1535                 goto find_fail;
1536         }
1537         root->commit_root = btrfs_root_node(root);
1538 out:
1539         btrfs_free_path(path);
1540         return root;
1541
1542 find_fail:
1543         kfree(root);
1544 alloc_fail:
1545         root = ERR_PTR(ret);
1546         goto out;
1547 }
1548
1549 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1550                                       struct btrfs_key *location)
1551 {
1552         struct btrfs_root *root;
1553
1554         root = btrfs_read_tree_root(tree_root, location);
1555         if (IS_ERR(root))
1556                 return root;
1557
1558         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1559                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1560                 btrfs_check_and_init_root_item(&root->root_item);
1561         }
1562
1563         return root;
1564 }
1565
1566 int btrfs_init_fs_root(struct btrfs_root *root)
1567 {
1568         int ret;
1569         struct btrfs_subvolume_writers *writers;
1570
1571         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1572         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1573                                         GFP_NOFS);
1574         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1575                 ret = -ENOMEM;
1576                 goto fail;
1577         }
1578
1579         writers = btrfs_alloc_subvolume_writers();
1580         if (IS_ERR(writers)) {
1581                 ret = PTR_ERR(writers);
1582                 goto fail;
1583         }
1584         root->subv_writers = writers;
1585
1586         btrfs_init_free_ino_ctl(root);
1587         spin_lock_init(&root->ino_cache_lock);
1588         init_waitqueue_head(&root->ino_cache_wait);
1589
1590         ret = get_anon_bdev(&root->anon_dev);
1591         if (ret)
1592                 goto free_writers;
1593
1594         mutex_lock(&root->objectid_mutex);
1595         ret = btrfs_find_highest_objectid(root,
1596                                         &root->highest_objectid);
1597         if (ret) {
1598                 mutex_unlock(&root->objectid_mutex);
1599                 goto free_root_dev;
1600         }
1601
1602         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1603
1604         mutex_unlock(&root->objectid_mutex);
1605
1606         return 0;
1607
1608 free_root_dev:
1609         free_anon_bdev(root->anon_dev);
1610 free_writers:
1611         btrfs_free_subvolume_writers(root->subv_writers);
1612 fail:
1613         kfree(root->free_ino_ctl);
1614         kfree(root->free_ino_pinned);
1615         return ret;
1616 }
1617
1618 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1619                                                u64 root_id)
1620 {
1621         struct btrfs_root *root;
1622
1623         spin_lock(&fs_info->fs_roots_radix_lock);
1624         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1625                                  (unsigned long)root_id);
1626         spin_unlock(&fs_info->fs_roots_radix_lock);
1627         return root;
1628 }
1629
1630 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1631                          struct btrfs_root *root)
1632 {
1633         int ret;
1634
1635         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1636         if (ret)
1637                 return ret;
1638
1639         spin_lock(&fs_info->fs_roots_radix_lock);
1640         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1641                                 (unsigned long)root->root_key.objectid,
1642                                 root);
1643         if (ret == 0)
1644                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1645         spin_unlock(&fs_info->fs_roots_radix_lock);
1646         radix_tree_preload_end();
1647
1648         return ret;
1649 }
1650
1651 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1652                                      struct btrfs_key *location,
1653                                      bool check_ref)
1654 {
1655         struct btrfs_root *root;
1656         struct btrfs_path *path;
1657         struct btrfs_key key;
1658         int ret;
1659
1660         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1661                 return fs_info->tree_root;
1662         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1663                 return fs_info->extent_root;
1664         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1665                 return fs_info->chunk_root;
1666         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1667                 return fs_info->dev_root;
1668         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1669                 return fs_info->csum_root;
1670         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1671                 return fs_info->quota_root ? fs_info->quota_root :
1672                                              ERR_PTR(-ENOENT);
1673         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1674                 return fs_info->uuid_root ? fs_info->uuid_root :
1675                                             ERR_PTR(-ENOENT);
1676         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1677                 return fs_info->free_space_root ? fs_info->free_space_root :
1678                                                   ERR_PTR(-ENOENT);
1679 again:
1680         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1681         if (root) {
1682                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1683                         return ERR_PTR(-ENOENT);
1684                 return root;
1685         }
1686
1687         root = btrfs_read_fs_root(fs_info->tree_root, location);
1688         if (IS_ERR(root))
1689                 return root;
1690
1691         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1692                 ret = -ENOENT;
1693                 goto fail;
1694         }
1695
1696         ret = btrfs_init_fs_root(root);
1697         if (ret)
1698                 goto fail;
1699
1700         path = btrfs_alloc_path();
1701         if (!path) {
1702                 ret = -ENOMEM;
1703                 goto fail;
1704         }
1705         key.objectid = BTRFS_ORPHAN_OBJECTID;
1706         key.type = BTRFS_ORPHAN_ITEM_KEY;
1707         key.offset = location->objectid;
1708
1709         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1710         btrfs_free_path(path);
1711         if (ret < 0)
1712                 goto fail;
1713         if (ret == 0)
1714                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1715
1716         ret = btrfs_insert_fs_root(fs_info, root);
1717         if (ret) {
1718                 if (ret == -EEXIST) {
1719                         free_fs_root(root);
1720                         goto again;
1721                 }
1722                 goto fail;
1723         }
1724         return root;
1725 fail:
1726         free_fs_root(root);
1727         return ERR_PTR(ret);
1728 }
1729
1730 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1731 {
1732         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1733         int ret = 0;
1734         struct btrfs_device *device;
1735         struct backing_dev_info *bdi;
1736
1737         rcu_read_lock();
1738         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1739                 if (!device->bdev)
1740                         continue;
1741                 bdi = blk_get_backing_dev_info(device->bdev);
1742                 if (bdi_congested(bdi, bdi_bits)) {
1743                         ret = 1;
1744                         break;
1745                 }
1746         }
1747         rcu_read_unlock();
1748         return ret;
1749 }
1750
1751 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1752 {
1753         int err;
1754
1755         err = bdi_setup_and_register(bdi, "btrfs");
1756         if (err)
1757                 return err;
1758
1759         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1760         bdi->congested_fn       = btrfs_congested_fn;
1761         bdi->congested_data     = info;
1762         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1763         return 0;
1764 }
1765
1766 /*
1767  * called by the kthread helper functions to finally call the bio end_io
1768  * functions.  This is where read checksum verification actually happens
1769  */
1770 static void end_workqueue_fn(struct btrfs_work *work)
1771 {
1772         struct bio *bio;
1773         struct btrfs_end_io_wq *end_io_wq;
1774
1775         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1776         bio = end_io_wq->bio;
1777
1778         bio->bi_error = end_io_wq->error;
1779         bio->bi_private = end_io_wq->private;
1780         bio->bi_end_io = end_io_wq->end_io;
1781         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1782         bio_endio(bio);
1783 }
1784
1785 static int cleaner_kthread(void *arg)
1786 {
1787         struct btrfs_root *root = arg;
1788         int again;
1789         struct btrfs_trans_handle *trans;
1790
1791         do {
1792                 again = 0;
1793
1794                 /* Make the cleaner go to sleep early. */
1795                 if (btrfs_need_cleaner_sleep(root))
1796                         goto sleep;
1797
1798                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1799                         goto sleep;
1800
1801                 /*
1802                  * Avoid the problem that we change the status of the fs
1803                  * during the above check and trylock.
1804                  */
1805                 if (btrfs_need_cleaner_sleep(root)) {
1806                         mutex_unlock(&root->fs_info->cleaner_mutex);
1807                         goto sleep;
1808                 }
1809
1810                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1811                 btrfs_run_delayed_iputs(root);
1812                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1813
1814                 again = btrfs_clean_one_deleted_snapshot(root);
1815                 mutex_unlock(&root->fs_info->cleaner_mutex);
1816
1817                 /*
1818                  * The defragger has dealt with the R/O remount and umount,
1819                  * needn't do anything special here.
1820                  */
1821                 btrfs_run_defrag_inodes(root->fs_info);
1822
1823                 /*
1824                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1825                  * with relocation (btrfs_relocate_chunk) and relocation
1826                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1827                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1828                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1829                  * unused block groups.
1830                  */
1831                 btrfs_delete_unused_bgs(root->fs_info);
1832 sleep:
1833                 if (!try_to_freeze() && !again) {
1834                         set_current_state(TASK_INTERRUPTIBLE);
1835                         if (!kthread_should_stop())
1836                                 schedule();
1837                         __set_current_state(TASK_RUNNING);
1838                 }
1839         } while (!kthread_should_stop());
1840
1841         /*
1842          * Transaction kthread is stopped before us and wakes us up.
1843          * However we might have started a new transaction and COWed some
1844          * tree blocks when deleting unused block groups for example. So
1845          * make sure we commit the transaction we started to have a clean
1846          * shutdown when evicting the btree inode - if it has dirty pages
1847          * when we do the final iput() on it, eviction will trigger a
1848          * writeback for it which will fail with null pointer dereferences
1849          * since work queues and other resources were already released and
1850          * destroyed by the time the iput/eviction/writeback is made.
1851          */
1852         trans = btrfs_attach_transaction(root);
1853         if (IS_ERR(trans)) {
1854                 if (PTR_ERR(trans) != -ENOENT)
1855                         btrfs_err(root->fs_info,
1856                                   "cleaner transaction attach returned %ld",
1857                                   PTR_ERR(trans));
1858         } else {
1859                 int ret;
1860
1861                 ret = btrfs_commit_transaction(trans, root);
1862                 if (ret)
1863                         btrfs_err(root->fs_info,
1864                                   "cleaner open transaction commit returned %d",
1865                                   ret);
1866         }
1867
1868         return 0;
1869 }
1870
1871 static int transaction_kthread(void *arg)
1872 {
1873         struct btrfs_root *root = arg;
1874         struct btrfs_trans_handle *trans;
1875         struct btrfs_transaction *cur;
1876         u64 transid;
1877         unsigned long now;
1878         unsigned long delay;
1879         bool cannot_commit;
1880
1881         do {
1882                 cannot_commit = false;
1883                 delay = HZ * root->fs_info->commit_interval;
1884                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1885
1886                 spin_lock(&root->fs_info->trans_lock);
1887                 cur = root->fs_info->running_transaction;
1888                 if (!cur) {
1889                         spin_unlock(&root->fs_info->trans_lock);
1890                         goto sleep;
1891                 }
1892
1893                 now = get_seconds();
1894                 if (cur->state < TRANS_STATE_BLOCKED &&
1895                     (now < cur->start_time ||
1896                      now - cur->start_time < root->fs_info->commit_interval)) {
1897                         spin_unlock(&root->fs_info->trans_lock);
1898                         delay = HZ * 5;
1899                         goto sleep;
1900                 }
1901                 transid = cur->transid;
1902                 spin_unlock(&root->fs_info->trans_lock);
1903
1904                 /* If the file system is aborted, this will always fail. */
1905                 trans = btrfs_attach_transaction(root);
1906                 if (IS_ERR(trans)) {
1907                         if (PTR_ERR(trans) != -ENOENT)
1908                                 cannot_commit = true;
1909                         goto sleep;
1910                 }
1911                 if (transid == trans->transid) {
1912                         btrfs_commit_transaction(trans, root);
1913                 } else {
1914                         btrfs_end_transaction(trans, root);
1915                 }
1916 sleep:
1917                 wake_up_process(root->fs_info->cleaner_kthread);
1918                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1919
1920                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1921                                       &root->fs_info->fs_state)))
1922                         btrfs_cleanup_transaction(root);
1923                 if (!try_to_freeze()) {
1924                         set_current_state(TASK_INTERRUPTIBLE);
1925                         if (!kthread_should_stop() &&
1926                             (!btrfs_transaction_blocked(root->fs_info) ||
1927                              cannot_commit))
1928                                 schedule_timeout(delay);
1929                         __set_current_state(TASK_RUNNING);
1930                 }
1931         } while (!kthread_should_stop());
1932         return 0;
1933 }
1934
1935 /*
1936  * this will find the highest generation in the array of
1937  * root backups.  The index of the highest array is returned,
1938  * or -1 if we can't find anything.
1939  *
1940  * We check to make sure the array is valid by comparing the
1941  * generation of the latest  root in the array with the generation
1942  * in the super block.  If they don't match we pitch it.
1943  */
1944 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1945 {
1946         u64 cur;
1947         int newest_index = -1;
1948         struct btrfs_root_backup *root_backup;
1949         int i;
1950
1951         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1952                 root_backup = info->super_copy->super_roots + i;
1953                 cur = btrfs_backup_tree_root_gen(root_backup);
1954                 if (cur == newest_gen)
1955                         newest_index = i;
1956         }
1957
1958         /* check to see if we actually wrapped around */
1959         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1960                 root_backup = info->super_copy->super_roots;
1961                 cur = btrfs_backup_tree_root_gen(root_backup);
1962                 if (cur == newest_gen)
1963                         newest_index = 0;
1964         }
1965         return newest_index;
1966 }
1967
1968
1969 /*
1970  * find the oldest backup so we know where to store new entries
1971  * in the backup array.  This will set the backup_root_index
1972  * field in the fs_info struct
1973  */
1974 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1975                                      u64 newest_gen)
1976 {
1977         int newest_index = -1;
1978
1979         newest_index = find_newest_super_backup(info, newest_gen);
1980         /* if there was garbage in there, just move along */
1981         if (newest_index == -1) {
1982                 info->backup_root_index = 0;
1983         } else {
1984                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1985         }
1986 }
1987
1988 /*
1989  * copy all the root pointers into the super backup array.
1990  * this will bump the backup pointer by one when it is
1991  * done
1992  */
1993 static void backup_super_roots(struct btrfs_fs_info *info)
1994 {
1995         int next_backup;
1996         struct btrfs_root_backup *root_backup;
1997         int last_backup;
1998
1999         next_backup = info->backup_root_index;
2000         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2001                 BTRFS_NUM_BACKUP_ROOTS;
2002
2003         /*
2004          * just overwrite the last backup if we're at the same generation
2005          * this happens only at umount
2006          */
2007         root_backup = info->super_for_commit->super_roots + last_backup;
2008         if (btrfs_backup_tree_root_gen(root_backup) ==
2009             btrfs_header_generation(info->tree_root->node))
2010                 next_backup = last_backup;
2011
2012         root_backup = info->super_for_commit->super_roots + next_backup;
2013
2014         /*
2015          * make sure all of our padding and empty slots get zero filled
2016          * regardless of which ones we use today
2017          */
2018         memset(root_backup, 0, sizeof(*root_backup));
2019
2020         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2021
2022         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2023         btrfs_set_backup_tree_root_gen(root_backup,
2024                                btrfs_header_generation(info->tree_root->node));
2025
2026         btrfs_set_backup_tree_root_level(root_backup,
2027                                btrfs_header_level(info->tree_root->node));
2028
2029         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2030         btrfs_set_backup_chunk_root_gen(root_backup,
2031                                btrfs_header_generation(info->chunk_root->node));
2032         btrfs_set_backup_chunk_root_level(root_backup,
2033                                btrfs_header_level(info->chunk_root->node));
2034
2035         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2036         btrfs_set_backup_extent_root_gen(root_backup,
2037                                btrfs_header_generation(info->extent_root->node));
2038         btrfs_set_backup_extent_root_level(root_backup,
2039                                btrfs_header_level(info->extent_root->node));
2040
2041         /*
2042          * we might commit during log recovery, which happens before we set
2043          * the fs_root.  Make sure it is valid before we fill it in.
2044          */
2045         if (info->fs_root && info->fs_root->node) {
2046                 btrfs_set_backup_fs_root(root_backup,
2047                                          info->fs_root->node->start);
2048                 btrfs_set_backup_fs_root_gen(root_backup,
2049                                btrfs_header_generation(info->fs_root->node));
2050                 btrfs_set_backup_fs_root_level(root_backup,
2051                                btrfs_header_level(info->fs_root->node));
2052         }
2053
2054         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2055         btrfs_set_backup_dev_root_gen(root_backup,
2056                                btrfs_header_generation(info->dev_root->node));
2057         btrfs_set_backup_dev_root_level(root_backup,
2058                                        btrfs_header_level(info->dev_root->node));
2059
2060         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2061         btrfs_set_backup_csum_root_gen(root_backup,
2062                                btrfs_header_generation(info->csum_root->node));
2063         btrfs_set_backup_csum_root_level(root_backup,
2064                                btrfs_header_level(info->csum_root->node));
2065
2066         btrfs_set_backup_total_bytes(root_backup,
2067                              btrfs_super_total_bytes(info->super_copy));
2068         btrfs_set_backup_bytes_used(root_backup,
2069                              btrfs_super_bytes_used(info->super_copy));
2070         btrfs_set_backup_num_devices(root_backup,
2071                              btrfs_super_num_devices(info->super_copy));
2072
2073         /*
2074          * if we don't copy this out to the super_copy, it won't get remembered
2075          * for the next commit
2076          */
2077         memcpy(&info->super_copy->super_roots,
2078                &info->super_for_commit->super_roots,
2079                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2080 }
2081
2082 /*
2083  * this copies info out of the root backup array and back into
2084  * the in-memory super block.  It is meant to help iterate through
2085  * the array, so you send it the number of backups you've already
2086  * tried and the last backup index you used.
2087  *
2088  * this returns -1 when it has tried all the backups
2089  */
2090 static noinline int next_root_backup(struct btrfs_fs_info *info,
2091                                      struct btrfs_super_block *super,
2092                                      int *num_backups_tried, int *backup_index)
2093 {
2094         struct btrfs_root_backup *root_backup;
2095         int newest = *backup_index;
2096
2097         if (*num_backups_tried == 0) {
2098                 u64 gen = btrfs_super_generation(super);
2099
2100                 newest = find_newest_super_backup(info, gen);
2101                 if (newest == -1)
2102                         return -1;
2103
2104                 *backup_index = newest;
2105                 *num_backups_tried = 1;
2106         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2107                 /* we've tried all the backups, all done */
2108                 return -1;
2109         } else {
2110                 /* jump to the next oldest backup */
2111                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2112                         BTRFS_NUM_BACKUP_ROOTS;
2113                 *backup_index = newest;
2114                 *num_backups_tried += 1;
2115         }
2116         root_backup = super->super_roots + newest;
2117
2118         btrfs_set_super_generation(super,
2119                                    btrfs_backup_tree_root_gen(root_backup));
2120         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2121         btrfs_set_super_root_level(super,
2122                                    btrfs_backup_tree_root_level(root_backup));
2123         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2124
2125         /*
2126          * fixme: the total bytes and num_devices need to match or we should
2127          * need a fsck
2128          */
2129         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2130         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2131         return 0;
2132 }
2133
2134 /* helper to cleanup workers */
2135 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2136 {
2137         btrfs_destroy_workqueue(fs_info->fixup_workers);
2138         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2139         btrfs_destroy_workqueue(fs_info->workers);
2140         btrfs_destroy_workqueue(fs_info->endio_workers);
2141         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2142         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2143         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2144         btrfs_destroy_workqueue(fs_info->rmw_workers);
2145         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2146         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2147         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2148         btrfs_destroy_workqueue(fs_info->submit_workers);
2149         btrfs_destroy_workqueue(fs_info->delayed_workers);
2150         btrfs_destroy_workqueue(fs_info->caching_workers);
2151         btrfs_destroy_workqueue(fs_info->readahead_workers);
2152         btrfs_destroy_workqueue(fs_info->flush_workers);
2153         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2154         btrfs_destroy_workqueue(fs_info->extent_workers);
2155 }
2156
2157 static void free_root_extent_buffers(struct btrfs_root *root)
2158 {
2159         if (root) {
2160                 free_extent_buffer(root->node);
2161                 free_extent_buffer(root->commit_root);
2162                 root->node = NULL;
2163                 root->commit_root = NULL;
2164         }
2165 }
2166
2167 /* helper to cleanup tree roots */
2168 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2169 {
2170         free_root_extent_buffers(info->tree_root);
2171
2172         free_root_extent_buffers(info->dev_root);
2173         free_root_extent_buffers(info->extent_root);
2174         free_root_extent_buffers(info->csum_root);
2175         free_root_extent_buffers(info->quota_root);
2176         free_root_extent_buffers(info->uuid_root);
2177         if (chunk_root)
2178                 free_root_extent_buffers(info->chunk_root);
2179         free_root_extent_buffers(info->free_space_root);
2180 }
2181
2182 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2183 {
2184         int ret;
2185         struct btrfs_root *gang[8];
2186         int i;
2187
2188         while (!list_empty(&fs_info->dead_roots)) {
2189                 gang[0] = list_entry(fs_info->dead_roots.next,
2190                                      struct btrfs_root, root_list);
2191                 list_del(&gang[0]->root_list);
2192
2193                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2194                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2195                 } else {
2196                         free_extent_buffer(gang[0]->node);
2197                         free_extent_buffer(gang[0]->commit_root);
2198                         btrfs_put_fs_root(gang[0]);
2199                 }
2200         }
2201
2202         while (1) {
2203                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2204                                              (void **)gang, 0,
2205                                              ARRAY_SIZE(gang));
2206                 if (!ret)
2207                         break;
2208                 for (i = 0; i < ret; i++)
2209                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2210         }
2211
2212         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2213                 btrfs_free_log_root_tree(NULL, fs_info);
2214                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2215                                             fs_info->pinned_extents);
2216         }
2217 }
2218
2219 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220 {
2221         mutex_init(&fs_info->scrub_lock);
2222         atomic_set(&fs_info->scrubs_running, 0);
2223         atomic_set(&fs_info->scrub_pause_req, 0);
2224         atomic_set(&fs_info->scrubs_paused, 0);
2225         atomic_set(&fs_info->scrub_cancel_req, 0);
2226         init_waitqueue_head(&fs_info->scrub_pause_wait);
2227         fs_info->scrub_workers_refcnt = 0;
2228 }
2229
2230 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231 {
2232         spin_lock_init(&fs_info->balance_lock);
2233         mutex_init(&fs_info->balance_mutex);
2234         atomic_set(&fs_info->balance_running, 0);
2235         atomic_set(&fs_info->balance_pause_req, 0);
2236         atomic_set(&fs_info->balance_cancel_req, 0);
2237         fs_info->balance_ctl = NULL;
2238         init_waitqueue_head(&fs_info->balance_wait_q);
2239 }
2240
2241 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2242                                    struct btrfs_root *tree_root)
2243 {
2244         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2245         set_nlink(fs_info->btree_inode, 1);
2246         /*
2247          * we set the i_size on the btree inode to the max possible int.
2248          * the real end of the address space is determined by all of
2249          * the devices in the system
2250          */
2251         fs_info->btree_inode->i_size = OFFSET_MAX;
2252         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2253
2254         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2255         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2256                              fs_info->btree_inode->i_mapping);
2257         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2258         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2259
2260         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2261
2262         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2263         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2264                sizeof(struct btrfs_key));
2265         set_bit(BTRFS_INODE_DUMMY,
2266                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2267         btrfs_insert_inode_hash(fs_info->btree_inode);
2268 }
2269
2270 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2271 {
2272         fs_info->dev_replace.lock_owner = 0;
2273         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2274         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2275         rwlock_init(&fs_info->dev_replace.lock);
2276         atomic_set(&fs_info->dev_replace.read_locks, 0);
2277         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2278         init_waitqueue_head(&fs_info->replace_wait);
2279         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2280 }
2281
2282 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2283 {
2284         spin_lock_init(&fs_info->qgroup_lock);
2285         mutex_init(&fs_info->qgroup_ioctl_lock);
2286         fs_info->qgroup_tree = RB_ROOT;
2287         fs_info->qgroup_op_tree = RB_ROOT;
2288         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2289         fs_info->qgroup_seq = 1;
2290         fs_info->quota_enabled = 0;
2291         fs_info->pending_quota_state = 0;
2292         fs_info->qgroup_ulist = NULL;
2293         mutex_init(&fs_info->qgroup_rescan_lock);
2294 }
2295
2296 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2297                 struct btrfs_fs_devices *fs_devices)
2298 {
2299         int max_active = fs_info->thread_pool_size;
2300         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2301
2302         fs_info->workers =
2303                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2304                                       max_active, 16);
2305
2306         fs_info->delalloc_workers =
2307                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2308
2309         fs_info->flush_workers =
2310                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2311
2312         fs_info->caching_workers =
2313                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2314
2315         /*
2316          * a higher idle thresh on the submit workers makes it much more
2317          * likely that bios will be send down in a sane order to the
2318          * devices
2319          */
2320         fs_info->submit_workers =
2321                 btrfs_alloc_workqueue("submit", flags,
2322                                       min_t(u64, fs_devices->num_devices,
2323                                             max_active), 64);
2324
2325         fs_info->fixup_workers =
2326                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2327
2328         /*
2329          * endios are largely parallel and should have a very
2330          * low idle thresh
2331          */
2332         fs_info->endio_workers =
2333                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2334         fs_info->endio_meta_workers =
2335                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2336         fs_info->endio_meta_write_workers =
2337                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2338         fs_info->endio_raid56_workers =
2339                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2340         fs_info->endio_repair_workers =
2341                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2342         fs_info->rmw_workers =
2343                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2344         fs_info->endio_write_workers =
2345                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2346         fs_info->endio_freespace_worker =
2347                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2348         fs_info->delayed_workers =
2349                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2350         fs_info->readahead_workers =
2351                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2352         fs_info->qgroup_rescan_workers =
2353                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2354         fs_info->extent_workers =
2355                 btrfs_alloc_workqueue("extent-refs", flags,
2356                                       min_t(u64, fs_devices->num_devices,
2357                                             max_active), 8);
2358
2359         if (!(fs_info->workers && fs_info->delalloc_workers &&
2360               fs_info->submit_workers && fs_info->flush_workers &&
2361               fs_info->endio_workers && fs_info->endio_meta_workers &&
2362               fs_info->endio_meta_write_workers &&
2363               fs_info->endio_repair_workers &&
2364               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2365               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2366               fs_info->caching_workers && fs_info->readahead_workers &&
2367               fs_info->fixup_workers && fs_info->delayed_workers &&
2368               fs_info->extent_workers &&
2369               fs_info->qgroup_rescan_workers)) {
2370                 return -ENOMEM;
2371         }
2372
2373         return 0;
2374 }
2375
2376 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2377                             struct btrfs_fs_devices *fs_devices)
2378 {
2379         int ret;
2380         struct btrfs_root *tree_root = fs_info->tree_root;
2381         struct btrfs_root *log_tree_root;
2382         struct btrfs_super_block *disk_super = fs_info->super_copy;
2383         u64 bytenr = btrfs_super_log_root(disk_super);
2384
2385         if (fs_devices->rw_devices == 0) {
2386                 btrfs_warn(fs_info, "log replay required on RO media");
2387                 return -EIO;
2388         }
2389
2390         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2391         if (!log_tree_root)
2392                 return -ENOMEM;
2393
2394         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2395                         tree_root->stripesize, log_tree_root, fs_info,
2396                         BTRFS_TREE_LOG_OBJECTID);
2397
2398         log_tree_root->node = read_tree_block(tree_root, bytenr,
2399                         fs_info->generation + 1);
2400         if (IS_ERR(log_tree_root->node)) {
2401                 btrfs_warn(fs_info, "failed to read log tree");
2402                 ret = PTR_ERR(log_tree_root->node);
2403                 kfree(log_tree_root);
2404                 return ret;
2405         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2406                 btrfs_err(fs_info, "failed to read log tree");
2407                 free_extent_buffer(log_tree_root->node);
2408                 kfree(log_tree_root);
2409                 return -EIO;
2410         }
2411         /* returns with log_tree_root freed on success */
2412         ret = btrfs_recover_log_trees(log_tree_root);
2413         if (ret) {
2414                 btrfs_std_error(tree_root->fs_info, ret,
2415                             "Failed to recover log tree");
2416                 free_extent_buffer(log_tree_root->node);
2417                 kfree(log_tree_root);
2418                 return ret;
2419         }
2420
2421         if (fs_info->sb->s_flags & MS_RDONLY) {
2422                 ret = btrfs_commit_super(tree_root);
2423                 if (ret)
2424                         return ret;
2425         }
2426
2427         return 0;
2428 }
2429
2430 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2431                             struct btrfs_root *tree_root)
2432 {
2433         struct btrfs_root *root;
2434         struct btrfs_key location;
2435         int ret;
2436
2437         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2438         location.type = BTRFS_ROOT_ITEM_KEY;
2439         location.offset = 0;
2440
2441         root = btrfs_read_tree_root(tree_root, &location);
2442         if (IS_ERR(root))
2443                 return PTR_ERR(root);
2444         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2445         fs_info->extent_root = root;
2446
2447         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2448         root = btrfs_read_tree_root(tree_root, &location);
2449         if (IS_ERR(root))
2450                 return PTR_ERR(root);
2451         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2452         fs_info->dev_root = root;
2453         btrfs_init_devices_late(fs_info);
2454
2455         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2456         root = btrfs_read_tree_root(tree_root, &location);
2457         if (IS_ERR(root))
2458                 return PTR_ERR(root);
2459         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2460         fs_info->csum_root = root;
2461
2462         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2463         root = btrfs_read_tree_root(tree_root, &location);
2464         if (!IS_ERR(root)) {
2465                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2466                 fs_info->quota_enabled = 1;
2467                 fs_info->pending_quota_state = 1;
2468                 fs_info->quota_root = root;
2469         }
2470
2471         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2472         root = btrfs_read_tree_root(tree_root, &location);
2473         if (IS_ERR(root)) {
2474                 ret = PTR_ERR(root);
2475                 if (ret != -ENOENT)
2476                         return ret;
2477         } else {
2478                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479                 fs_info->uuid_root = root;
2480         }
2481
2482         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2483                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2484                 root = btrfs_read_tree_root(tree_root, &location);
2485                 if (IS_ERR(root))
2486                         return PTR_ERR(root);
2487                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2488                 fs_info->free_space_root = root;
2489         }
2490
2491         return 0;
2492 }
2493
2494 int open_ctree(struct super_block *sb,
2495                struct btrfs_fs_devices *fs_devices,
2496                char *options)
2497 {
2498         u32 sectorsize;
2499         u32 nodesize;
2500         u32 stripesize;
2501         u64 generation;
2502         u64 features;
2503         struct btrfs_key location;
2504         struct buffer_head *bh;
2505         struct btrfs_super_block *disk_super;
2506         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2507         struct btrfs_root *tree_root;
2508         struct btrfs_root *chunk_root;
2509         int ret;
2510         int err = -EINVAL;
2511         int num_backups_tried = 0;
2512         int backup_index = 0;
2513         int max_active;
2514
2515         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2516         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2517         if (!tree_root || !chunk_root) {
2518                 err = -ENOMEM;
2519                 goto fail;
2520         }
2521
2522         ret = init_srcu_struct(&fs_info->subvol_srcu);
2523         if (ret) {
2524                 err = ret;
2525                 goto fail;
2526         }
2527
2528         ret = setup_bdi(fs_info, &fs_info->bdi);
2529         if (ret) {
2530                 err = ret;
2531                 goto fail_srcu;
2532         }
2533
2534         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2535         if (ret) {
2536                 err = ret;
2537                 goto fail_bdi;
2538         }
2539         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2540                                         (1 + ilog2(nr_cpu_ids));
2541
2542         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2543         if (ret) {
2544                 err = ret;
2545                 goto fail_dirty_metadata_bytes;
2546         }
2547
2548         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2549         if (ret) {
2550                 err = ret;
2551                 goto fail_delalloc_bytes;
2552         }
2553
2554         fs_info->btree_inode = new_inode(sb);
2555         if (!fs_info->btree_inode) {
2556                 err = -ENOMEM;
2557                 goto fail_bio_counter;
2558         }
2559
2560         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2561
2562         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2563         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2564         INIT_LIST_HEAD(&fs_info->trans_list);
2565         INIT_LIST_HEAD(&fs_info->dead_roots);
2566         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2567         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2568         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2569         spin_lock_init(&fs_info->delalloc_root_lock);
2570         spin_lock_init(&fs_info->trans_lock);
2571         spin_lock_init(&fs_info->fs_roots_radix_lock);
2572         spin_lock_init(&fs_info->delayed_iput_lock);
2573         spin_lock_init(&fs_info->defrag_inodes_lock);
2574         spin_lock_init(&fs_info->free_chunk_lock);
2575         spin_lock_init(&fs_info->tree_mod_seq_lock);
2576         spin_lock_init(&fs_info->super_lock);
2577         spin_lock_init(&fs_info->qgroup_op_lock);
2578         spin_lock_init(&fs_info->buffer_lock);
2579         spin_lock_init(&fs_info->unused_bgs_lock);
2580         rwlock_init(&fs_info->tree_mod_log_lock);
2581         mutex_init(&fs_info->unused_bg_unpin_mutex);
2582         mutex_init(&fs_info->delete_unused_bgs_mutex);
2583         mutex_init(&fs_info->reloc_mutex);
2584         mutex_init(&fs_info->delalloc_root_mutex);
2585         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2586         seqlock_init(&fs_info->profiles_lock);
2587
2588         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2589         INIT_LIST_HEAD(&fs_info->space_info);
2590         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2591         INIT_LIST_HEAD(&fs_info->unused_bgs);
2592         btrfs_mapping_init(&fs_info->mapping_tree);
2593         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2594                              BTRFS_BLOCK_RSV_GLOBAL);
2595         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2596                              BTRFS_BLOCK_RSV_DELALLOC);
2597         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2598         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2599         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2600         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2601                              BTRFS_BLOCK_RSV_DELOPS);
2602         atomic_set(&fs_info->nr_async_submits, 0);
2603         atomic_set(&fs_info->async_delalloc_pages, 0);
2604         atomic_set(&fs_info->async_submit_draining, 0);
2605         atomic_set(&fs_info->nr_async_bios, 0);
2606         atomic_set(&fs_info->defrag_running, 0);
2607         atomic_set(&fs_info->qgroup_op_seq, 0);
2608         atomic_set(&fs_info->reada_works_cnt, 0);
2609         atomic64_set(&fs_info->tree_mod_seq, 0);
2610         fs_info->sb = sb;
2611         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2612         fs_info->metadata_ratio = 0;
2613         fs_info->defrag_inodes = RB_ROOT;
2614         fs_info->free_chunk_space = 0;
2615         fs_info->tree_mod_log = RB_ROOT;
2616         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2617         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2618         /* readahead state */
2619         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2620         spin_lock_init(&fs_info->reada_lock);
2621
2622         fs_info->thread_pool_size = min_t(unsigned long,
2623                                           num_online_cpus() + 2, 8);
2624
2625         INIT_LIST_HEAD(&fs_info->ordered_roots);
2626         spin_lock_init(&fs_info->ordered_root_lock);
2627         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2628                                         GFP_KERNEL);
2629         if (!fs_info->delayed_root) {
2630                 err = -ENOMEM;
2631                 goto fail_iput;
2632         }
2633         btrfs_init_delayed_root(fs_info->delayed_root);
2634
2635         btrfs_init_scrub(fs_info);
2636 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2637         fs_info->check_integrity_print_mask = 0;
2638 #endif
2639         btrfs_init_balance(fs_info);
2640         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2641
2642         sb->s_blocksize = 4096;
2643         sb->s_blocksize_bits = blksize_bits(4096);
2644         sb->s_bdi = &fs_info->bdi;
2645
2646         btrfs_init_btree_inode(fs_info, tree_root);
2647
2648         spin_lock_init(&fs_info->block_group_cache_lock);
2649         fs_info->block_group_cache_tree = RB_ROOT;
2650         fs_info->first_logical_byte = (u64)-1;
2651
2652         extent_io_tree_init(&fs_info->freed_extents[0],
2653                              fs_info->btree_inode->i_mapping);
2654         extent_io_tree_init(&fs_info->freed_extents[1],
2655                              fs_info->btree_inode->i_mapping);
2656         fs_info->pinned_extents = &fs_info->freed_extents[0];
2657         fs_info->do_barriers = 1;
2658
2659
2660         mutex_init(&fs_info->ordered_operations_mutex);
2661         mutex_init(&fs_info->tree_log_mutex);
2662         mutex_init(&fs_info->chunk_mutex);
2663         mutex_init(&fs_info->transaction_kthread_mutex);
2664         mutex_init(&fs_info->cleaner_mutex);
2665         mutex_init(&fs_info->volume_mutex);
2666         mutex_init(&fs_info->ro_block_group_mutex);
2667         init_rwsem(&fs_info->commit_root_sem);
2668         init_rwsem(&fs_info->cleanup_work_sem);
2669         init_rwsem(&fs_info->subvol_sem);
2670         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2671
2672         btrfs_init_dev_replace_locks(fs_info);
2673         btrfs_init_qgroup(fs_info);
2674
2675         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2676         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2677
2678         init_waitqueue_head(&fs_info->transaction_throttle);
2679         init_waitqueue_head(&fs_info->transaction_wait);
2680         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2681         init_waitqueue_head(&fs_info->async_submit_wait);
2682
2683         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2684
2685         ret = btrfs_alloc_stripe_hash_table(fs_info);
2686         if (ret) {
2687                 err = ret;
2688                 goto fail_alloc;
2689         }
2690
2691         __setup_root(4096, 4096, 4096, tree_root,
2692                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2693
2694         invalidate_bdev(fs_devices->latest_bdev);
2695
2696         /*
2697          * Read super block and check the signature bytes only
2698          */
2699         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2700         if (IS_ERR(bh)) {
2701                 err = PTR_ERR(bh);
2702                 goto fail_alloc;
2703         }
2704
2705         /*
2706          * We want to check superblock checksum, the type is stored inside.
2707          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2708          */
2709         if (btrfs_check_super_csum(bh->b_data)) {
2710                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2711                 err = -EINVAL;
2712                 brelse(bh);
2713                 goto fail_alloc;
2714         }
2715
2716         /*
2717          * super_copy is zeroed at allocation time and we never touch the
2718          * following bytes up to INFO_SIZE, the checksum is calculated from
2719          * the whole block of INFO_SIZE
2720          */
2721         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2722         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2723                sizeof(*fs_info->super_for_commit));
2724         brelse(bh);
2725
2726         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2727
2728         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2729         if (ret) {
2730                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2731                 err = -EINVAL;
2732                 goto fail_alloc;
2733         }
2734
2735         disk_super = fs_info->super_copy;
2736         if (!btrfs_super_root(disk_super))
2737                 goto fail_alloc;
2738
2739         /* check FS state, whether FS is broken. */
2740         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2741                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2742
2743         /*
2744          * run through our array of backup supers and setup
2745          * our ring pointer to the oldest one
2746          */
2747         generation = btrfs_super_generation(disk_super);
2748         find_oldest_super_backup(fs_info, generation);
2749
2750         /*
2751          * In the long term, we'll store the compression type in the super
2752          * block, and it'll be used for per file compression control.
2753          */
2754         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2755
2756         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2757         if (ret) {
2758                 err = ret;
2759                 goto fail_alloc;
2760         }
2761
2762         features = btrfs_super_incompat_flags(disk_super) &
2763                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2764         if (features) {
2765                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2766                        "unsupported optional features (%Lx).\n",
2767                        features);
2768                 err = -EINVAL;
2769                 goto fail_alloc;
2770         }
2771
2772         features = btrfs_super_incompat_flags(disk_super);
2773         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2774         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2775                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2776
2777         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2778                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2779
2780         /*
2781          * flag our filesystem as having big metadata blocks if
2782          * they are bigger than the page size
2783          */
2784         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2785                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2786                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2787                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2788         }
2789
2790         nodesize = btrfs_super_nodesize(disk_super);
2791         sectorsize = btrfs_super_sectorsize(disk_super);
2792         stripesize = btrfs_super_stripesize(disk_super);
2793         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2794         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2795
2796         /*
2797          * mixed block groups end up with duplicate but slightly offset
2798          * extent buffers for the same range.  It leads to corruptions
2799          */
2800         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2801             (sectorsize != nodesize)) {
2802                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2803                                 "are not allowed for mixed block groups on %s\n",
2804                                 sb->s_id);
2805                 goto fail_alloc;
2806         }
2807
2808         /*
2809          * Needn't use the lock because there is no other task which will
2810          * update the flag.
2811          */
2812         btrfs_set_super_incompat_flags(disk_super, features);
2813
2814         features = btrfs_super_compat_ro_flags(disk_super) &
2815                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2816         if (!(sb->s_flags & MS_RDONLY) && features) {
2817                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2818                        "unsupported option features (%Lx).\n",
2819                        features);
2820                 err = -EINVAL;
2821                 goto fail_alloc;
2822         }
2823
2824         max_active = fs_info->thread_pool_size;
2825
2826         ret = btrfs_init_workqueues(fs_info, fs_devices);
2827         if (ret) {
2828                 err = ret;
2829                 goto fail_sb_buffer;
2830         }
2831
2832         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2833         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2834                                     SZ_4M / PAGE_CACHE_SIZE);
2835
2836         tree_root->nodesize = nodesize;
2837         tree_root->sectorsize = sectorsize;
2838         tree_root->stripesize = stripesize;
2839
2840         sb->s_blocksize = sectorsize;
2841         sb->s_blocksize_bits = blksize_bits(sectorsize);
2842
2843         mutex_lock(&fs_info->chunk_mutex);
2844         ret = btrfs_read_sys_array(tree_root);
2845         mutex_unlock(&fs_info->chunk_mutex);
2846         if (ret) {
2847                 printk(KERN_ERR "BTRFS: failed to read the system "
2848                        "array on %s\n", sb->s_id);
2849                 goto fail_sb_buffer;
2850         }
2851
2852         generation = btrfs_super_chunk_root_generation(disk_super);
2853
2854         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2855                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2856
2857         chunk_root->node = read_tree_block(chunk_root,
2858                                            btrfs_super_chunk_root(disk_super),
2859                                            generation);
2860         if (IS_ERR(chunk_root->node) ||
2861             !extent_buffer_uptodate(chunk_root->node)) {
2862                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2863                        sb->s_id);
2864                 if (!IS_ERR(chunk_root->node))
2865                         free_extent_buffer(chunk_root->node);
2866                 chunk_root->node = NULL;
2867                 goto fail_tree_roots;
2868         }
2869         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2870         chunk_root->commit_root = btrfs_root_node(chunk_root);
2871
2872         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2873            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2874
2875         ret = btrfs_read_chunk_tree(chunk_root);
2876         if (ret) {
2877                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2878                        sb->s_id);
2879                 goto fail_tree_roots;
2880         }
2881
2882         /*
2883          * keep the device that is marked to be the target device for the
2884          * dev_replace procedure
2885          */
2886         btrfs_close_extra_devices(fs_devices, 0);
2887
2888         if (!fs_devices->latest_bdev) {
2889                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2890                        sb->s_id);
2891                 goto fail_tree_roots;
2892         }
2893
2894 retry_root_backup:
2895         generation = btrfs_super_generation(disk_super);
2896
2897         tree_root->node = read_tree_block(tree_root,
2898                                           btrfs_super_root(disk_super),
2899                                           generation);
2900         if (IS_ERR(tree_root->node) ||
2901             !extent_buffer_uptodate(tree_root->node)) {
2902                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2903                        sb->s_id);
2904                 if (!IS_ERR(tree_root->node))
2905                         free_extent_buffer(tree_root->node);
2906                 tree_root->node = NULL;
2907                 goto recovery_tree_root;
2908         }
2909
2910         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2911         tree_root->commit_root = btrfs_root_node(tree_root);
2912         btrfs_set_root_refs(&tree_root->root_item, 1);
2913
2914         mutex_lock(&tree_root->objectid_mutex);
2915         ret = btrfs_find_highest_objectid(tree_root,
2916                                         &tree_root->highest_objectid);
2917         if (ret) {
2918                 mutex_unlock(&tree_root->objectid_mutex);
2919                 goto recovery_tree_root;
2920         }
2921
2922         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2923
2924         mutex_unlock(&tree_root->objectid_mutex);
2925
2926         ret = btrfs_read_roots(fs_info, tree_root);
2927         if (ret)
2928                 goto recovery_tree_root;
2929
2930         fs_info->generation = generation;
2931         fs_info->last_trans_committed = generation;
2932
2933         ret = btrfs_recover_balance(fs_info);
2934         if (ret) {
2935                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2936                 goto fail_block_groups;
2937         }
2938
2939         ret = btrfs_init_dev_stats(fs_info);
2940         if (ret) {
2941                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2942                        ret);
2943                 goto fail_block_groups;
2944         }
2945
2946         ret = btrfs_init_dev_replace(fs_info);
2947         if (ret) {
2948                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2949                 goto fail_block_groups;
2950         }
2951
2952         btrfs_close_extra_devices(fs_devices, 1);
2953
2954         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2955         if (ret) {
2956                 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2957                 goto fail_block_groups;
2958         }
2959
2960         ret = btrfs_sysfs_add_device(fs_devices);
2961         if (ret) {
2962                 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2963                 goto fail_fsdev_sysfs;
2964         }
2965
2966         ret = btrfs_sysfs_add_mounted(fs_info);
2967         if (ret) {
2968                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2969                 goto fail_fsdev_sysfs;
2970         }
2971
2972         ret = btrfs_init_space_info(fs_info);
2973         if (ret) {
2974                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2975                 goto fail_sysfs;
2976         }
2977
2978         ret = btrfs_read_block_groups(fs_info->extent_root);
2979         if (ret) {
2980                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2981                 goto fail_sysfs;
2982         }
2983         fs_info->num_tolerated_disk_barrier_failures =
2984                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2985         if (fs_info->fs_devices->missing_devices >
2986              fs_info->num_tolerated_disk_barrier_failures &&
2987             !(sb->s_flags & MS_RDONLY)) {
2988                 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2989                         fs_info->fs_devices->missing_devices,
2990                         fs_info->num_tolerated_disk_barrier_failures);
2991                 goto fail_sysfs;
2992         }
2993
2994         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2995                                                "btrfs-cleaner");
2996         if (IS_ERR(fs_info->cleaner_kthread))
2997                 goto fail_sysfs;
2998
2999         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3000                                                    tree_root,
3001                                                    "btrfs-transaction");
3002         if (IS_ERR(fs_info->transaction_kthread))
3003                 goto fail_cleaner;
3004
3005         if (!btrfs_test_opt(tree_root, SSD) &&
3006             !btrfs_test_opt(tree_root, NOSSD) &&
3007             !fs_info->fs_devices->rotating) {
3008                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3009                        "mode\n");
3010                 btrfs_set_opt(fs_info->mount_opt, SSD);
3011         }
3012
3013         /*
3014          * Mount does not set all options immediatelly, we can do it now and do
3015          * not have to wait for transaction commit
3016          */
3017         btrfs_apply_pending_changes(fs_info);
3018
3019 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3020         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3021                 ret = btrfsic_mount(tree_root, fs_devices,
3022                                     btrfs_test_opt(tree_root,
3023                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3024                                     1 : 0,
3025                                     fs_info->check_integrity_print_mask);
3026                 if (ret)
3027                         printk(KERN_WARNING "BTRFS: failed to initialize"
3028                                " integrity check module %s\n", sb->s_id);
3029         }
3030 #endif
3031         ret = btrfs_read_qgroup_config(fs_info);
3032         if (ret)
3033                 goto fail_trans_kthread;
3034
3035         /* do not make disk changes in broken FS or nologreplay is given */
3036         if (btrfs_super_log_root(disk_super) != 0 &&
3037             !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3038                 ret = btrfs_replay_log(fs_info, fs_devices);
3039                 if (ret) {
3040                         err = ret;
3041                         goto fail_qgroup;
3042                 }
3043         }
3044
3045         ret = btrfs_find_orphan_roots(tree_root);
3046         if (ret)
3047                 goto fail_qgroup;
3048
3049         if (!(sb->s_flags & MS_RDONLY)) {
3050                 ret = btrfs_cleanup_fs_roots(fs_info);
3051                 if (ret)
3052                         goto fail_qgroup;
3053
3054                 mutex_lock(&fs_info->cleaner_mutex);
3055                 ret = btrfs_recover_relocation(tree_root);
3056                 mutex_unlock(&fs_info->cleaner_mutex);
3057                 if (ret < 0) {
3058                         printk(KERN_WARNING
3059                                "BTRFS: failed to recover relocation\n");
3060                         err = -EINVAL;
3061                         goto fail_qgroup;
3062                 }
3063         }
3064
3065         location.objectid = BTRFS_FS_TREE_OBJECTID;
3066         location.type = BTRFS_ROOT_ITEM_KEY;
3067         location.offset = 0;
3068
3069         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3070         if (IS_ERR(fs_info->fs_root)) {
3071                 err = PTR_ERR(fs_info->fs_root);
3072                 goto fail_qgroup;
3073         }
3074
3075         if (sb->s_flags & MS_RDONLY)
3076                 return 0;
3077
3078         if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3079             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3080                 pr_info("BTRFS: creating free space tree\n");
3081                 ret = btrfs_create_free_space_tree(fs_info);
3082                 if (ret) {
3083                         pr_warn("BTRFS: failed to create free space tree %d\n",
3084                                 ret);
3085                         close_ctree(tree_root);
3086                         return ret;
3087                 }
3088         }
3089
3090         down_read(&fs_info->cleanup_work_sem);
3091         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3092             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3093                 up_read(&fs_info->cleanup_work_sem);
3094                 close_ctree(tree_root);
3095                 return ret;
3096         }
3097         up_read(&fs_info->cleanup_work_sem);
3098
3099         ret = btrfs_resume_balance_async(fs_info);
3100         if (ret) {
3101                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3102                 close_ctree(tree_root);
3103                 return ret;
3104         }
3105
3106         ret = btrfs_resume_dev_replace_async(fs_info);
3107         if (ret) {
3108                 pr_warn("BTRFS: failed to resume dev_replace\n");
3109                 close_ctree(tree_root);
3110                 return ret;
3111         }
3112
3113         btrfs_qgroup_rescan_resume(fs_info);
3114
3115         if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3116             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3117                 pr_info("BTRFS: clearing free space tree\n");
3118                 ret = btrfs_clear_free_space_tree(fs_info);
3119                 if (ret) {
3120                         pr_warn("BTRFS: failed to clear free space tree %d\n",
3121                                 ret);
3122                         close_ctree(tree_root);
3123                         return ret;
3124                 }
3125         }
3126
3127         if (!fs_info->uuid_root) {
3128                 pr_info("BTRFS: creating UUID tree\n");
3129                 ret = btrfs_create_uuid_tree(fs_info);
3130                 if (ret) {
3131                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3132                                 ret);
3133                         close_ctree(tree_root);
3134                         return ret;
3135                 }
3136         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3137                    fs_info->generation !=
3138                                 btrfs_super_uuid_tree_generation(disk_super)) {
3139                 pr_info("BTRFS: checking UUID tree\n");
3140                 ret = btrfs_check_uuid_tree(fs_info);
3141                 if (ret) {
3142                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3143                                 ret);
3144                         close_ctree(tree_root);
3145                         return ret;
3146                 }
3147         } else {
3148                 fs_info->update_uuid_tree_gen = 1;
3149         }
3150
3151         fs_info->open = 1;
3152
3153         /*
3154          * backuproot only affect mount behavior, and if open_ctree succeeded,
3155          * no need to keep the flag
3156          */
3157         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3158
3159         return 0;
3160
3161 fail_qgroup:
3162         btrfs_free_qgroup_config(fs_info);
3163 fail_trans_kthread:
3164         kthread_stop(fs_info->transaction_kthread);
3165         btrfs_cleanup_transaction(fs_info->tree_root);
3166         btrfs_free_fs_roots(fs_info);
3167 fail_cleaner:
3168         kthread_stop(fs_info->cleaner_kthread);
3169
3170         /*
3171          * make sure we're done with the btree inode before we stop our
3172          * kthreads
3173          */
3174         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3175
3176 fail_sysfs:
3177         btrfs_sysfs_remove_mounted(fs_info);
3178
3179 fail_fsdev_sysfs:
3180         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3181
3182 fail_block_groups:
3183         btrfs_put_block_group_cache(fs_info);
3184         btrfs_free_block_groups(fs_info);
3185
3186 fail_tree_roots:
3187         free_root_pointers(fs_info, 1);
3188         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3189
3190 fail_sb_buffer:
3191         btrfs_stop_all_workers(fs_info);
3192 fail_alloc:
3193 fail_iput:
3194         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3195
3196         iput(fs_info->btree_inode);
3197 fail_bio_counter:
3198         percpu_counter_destroy(&fs_info->bio_counter);
3199 fail_delalloc_bytes:
3200         percpu_counter_destroy(&fs_info->delalloc_bytes);
3201 fail_dirty_metadata_bytes:
3202         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3203 fail_bdi:
3204         bdi_destroy(&fs_info->bdi);
3205 fail_srcu:
3206         cleanup_srcu_struct(&fs_info->subvol_srcu);
3207 fail:
3208         btrfs_free_stripe_hash_table(fs_info);
3209         btrfs_close_devices(fs_info->fs_devices);
3210         return err;
3211
3212 recovery_tree_root:
3213         if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3214                 goto fail_tree_roots;
3215
3216         free_root_pointers(fs_info, 0);
3217
3218         /* don't use the log in recovery mode, it won't be valid */
3219         btrfs_set_super_log_root(disk_super, 0);
3220
3221         /* we can't trust the free space cache either */
3222         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3223
3224         ret = next_root_backup(fs_info, fs_info->super_copy,
3225                                &num_backups_tried, &backup_index);
3226         if (ret == -1)
3227                 goto fail_block_groups;
3228         goto retry_root_backup;
3229 }
3230
3231 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3232 {
3233         if (uptodate) {
3234                 set_buffer_uptodate(bh);
3235         } else {
3236                 struct btrfs_device *device = (struct btrfs_device *)
3237                         bh->b_private;
3238
3239                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3240                                 "lost page write due to IO error on %s",
3241                                           rcu_str_deref(device->name));
3242                 /* note, we dont' set_buffer_write_io_error because we have
3243                  * our own ways of dealing with the IO errors
3244                  */
3245                 clear_buffer_uptodate(bh);
3246                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3247         }
3248         unlock_buffer(bh);
3249         put_bh(bh);
3250 }
3251
3252 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3253                         struct buffer_head **bh_ret)
3254 {
3255         struct buffer_head *bh;
3256         struct btrfs_super_block *super;
3257         u64 bytenr;
3258
3259         bytenr = btrfs_sb_offset(copy_num);
3260         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3261                 return -EINVAL;
3262
3263         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3264         /*
3265          * If we fail to read from the underlying devices, as of now
3266          * the best option we have is to mark it EIO.
3267          */
3268         if (!bh)
3269                 return -EIO;
3270
3271         super = (struct btrfs_super_block *)bh->b_data;
3272         if (btrfs_super_bytenr(super) != bytenr ||
3273                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3274                 brelse(bh);
3275                 return -EINVAL;
3276         }
3277
3278         *bh_ret = bh;
3279         return 0;
3280 }
3281
3282
3283 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3284 {
3285         struct buffer_head *bh;
3286         struct buffer_head *latest = NULL;
3287         struct btrfs_super_block *super;
3288         int i;
3289         u64 transid = 0;
3290         int ret = -EINVAL;
3291
3292         /* we would like to check all the supers, but that would make
3293          * a btrfs mount succeed after a mkfs from a different FS.
3294          * So, we need to add a special mount option to scan for
3295          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3296          */
3297         for (i = 0; i < 1; i++) {
3298                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3299                 if (ret)
3300                         continue;
3301
3302                 super = (struct btrfs_super_block *)bh->b_data;
3303
3304                 if (!latest || btrfs_super_generation(super) > transid) {
3305                         brelse(latest);
3306                         latest = bh;
3307                         transid = btrfs_super_generation(super);
3308                 } else {
3309                         brelse(bh);
3310                 }
3311         }
3312
3313         if (!latest)
3314                 return ERR_PTR(ret);
3315
3316         return latest;
3317 }
3318
3319 /*
3320  * this should be called twice, once with wait == 0 and
3321  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3322  * we write are pinned.
3323  *
3324  * They are released when wait == 1 is done.
3325  * max_mirrors must be the same for both runs, and it indicates how
3326  * many supers on this one device should be written.
3327  *
3328  * max_mirrors == 0 means to write them all.
3329  */
3330 static int write_dev_supers(struct btrfs_device *device,
3331                             struct btrfs_super_block *sb,
3332                             int do_barriers, int wait, int max_mirrors)
3333 {
3334         struct buffer_head *bh;
3335         int i;
3336         int ret;
3337         int errors = 0;
3338         u32 crc;
3339         u64 bytenr;
3340
3341         if (max_mirrors == 0)
3342                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3343
3344         for (i = 0; i < max_mirrors; i++) {
3345                 bytenr = btrfs_sb_offset(i);
3346                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3347                     device->commit_total_bytes)
3348                         break;
3349
3350                 if (wait) {
3351                         bh = __find_get_block(device->bdev, bytenr / 4096,
3352                                               BTRFS_SUPER_INFO_SIZE);
3353                         if (!bh) {
3354                                 errors++;
3355                                 continue;
3356                         }
3357                         wait_on_buffer(bh);
3358                         if (!buffer_uptodate(bh))
3359                                 errors++;
3360
3361                         /* drop our reference */
3362                         brelse(bh);
3363
3364                         /* drop the reference from the wait == 0 run */
3365                         brelse(bh);
3366                         continue;
3367                 } else {
3368                         btrfs_set_super_bytenr(sb, bytenr);
3369
3370                         crc = ~(u32)0;
3371                         crc = btrfs_csum_data((char *)sb +
3372                                               BTRFS_CSUM_SIZE, crc,
3373                                               BTRFS_SUPER_INFO_SIZE -
3374                                               BTRFS_CSUM_SIZE);
3375                         btrfs_csum_final(crc, sb->csum);
3376
3377                         /*
3378                          * one reference for us, and we leave it for the
3379                          * caller
3380                          */
3381                         bh = __getblk(device->bdev, bytenr / 4096,
3382                                       BTRFS_SUPER_INFO_SIZE);
3383                         if (!bh) {
3384                                 btrfs_err(device->dev_root->fs_info,
3385                                     "couldn't get super buffer head for bytenr %llu",
3386                                     bytenr);
3387                                 errors++;
3388                                 continue;
3389                         }
3390
3391                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3392
3393                         /* one reference for submit_bh */
3394                         get_bh(bh);
3395
3396                         set_buffer_uptodate(bh);
3397                         lock_buffer(bh);
3398                         bh->b_end_io = btrfs_end_buffer_write_sync;
3399                         bh->b_private = device;
3400                 }
3401
3402                 /*
3403                  * we fua the first super.  The others we allow
3404                  * to go down lazy.
3405                  */
3406                 if (i == 0)
3407                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3408                 else
3409                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3410                 if (ret)
3411                         errors++;
3412         }
3413         return errors < i ? 0 : -1;
3414 }
3415
3416 /*
3417  * endio for the write_dev_flush, this will wake anyone waiting
3418  * for the barrier when it is done
3419  */
3420 static void btrfs_end_empty_barrier(struct bio *bio)
3421 {
3422         if (bio->bi_private)
3423                 complete(bio->bi_private);
3424         bio_put(bio);
3425 }
3426
3427 /*
3428  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3429  * sent down.  With wait == 1, it waits for the previous flush.
3430  *
3431  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3432  * capable
3433  */
3434 static int write_dev_flush(struct btrfs_device *device, int wait)
3435 {
3436         struct bio *bio;
3437         int ret = 0;
3438
3439         if (device->nobarriers)
3440                 return 0;
3441
3442         if (wait) {
3443                 bio = device->flush_bio;
3444                 if (!bio)
3445                         return 0;
3446
3447                 wait_for_completion(&device->flush_wait);
3448
3449                 if (bio->bi_error) {
3450                         ret = bio->bi_error;
3451                         btrfs_dev_stat_inc_and_print(device,
3452                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3453                 }
3454
3455                 /* drop the reference from the wait == 0 run */
3456                 bio_put(bio);
3457                 device->flush_bio = NULL;
3458
3459                 return ret;
3460         }
3461
3462         /*
3463          * one reference for us, and we leave it for the
3464          * caller
3465          */
3466         device->flush_bio = NULL;
3467         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3468         if (!bio)
3469                 return -ENOMEM;
3470
3471         bio->bi_end_io = btrfs_end_empty_barrier;
3472         bio->bi_bdev = device->bdev;
3473         init_completion(&device->flush_wait);
3474         bio->bi_private = &device->flush_wait;
3475         device->flush_bio = bio;
3476
3477         bio_get(bio);
3478         btrfsic_submit_bio(WRITE_FLUSH, bio);
3479
3480         return 0;
3481 }
3482
3483 /*
3484  * send an empty flush down to each device in parallel,
3485  * then wait for them
3486  */
3487 static int barrier_all_devices(struct btrfs_fs_info *info)
3488 {
3489         struct list_head *head;
3490         struct btrfs_device *dev;
3491         int errors_send = 0;
3492         int errors_wait = 0;
3493         int ret;
3494
3495         /* send down all the barriers */
3496         head = &info->fs_devices->devices;
3497         list_for_each_entry_rcu(dev, head, dev_list) {
3498                 if (dev->missing)
3499                         continue;
3500                 if (!dev->bdev) {
3501                         errors_send++;
3502                         continue;
3503                 }
3504                 if (!dev->in_fs_metadata || !dev->writeable)
3505                         continue;
3506
3507                 ret = write_dev_flush(dev, 0);
3508                 if (ret)
3509                         errors_send++;
3510         }
3511
3512         /* wait for all the barriers */
3513         list_for_each_entry_rcu(dev, head, dev_list) {
3514                 if (dev->missing)
3515                         continue;
3516                 if (!dev->bdev) {
3517                         errors_wait++;
3518                         continue;
3519                 }
3520                 if (!dev->in_fs_metadata || !dev->writeable)
3521                         continue;
3522
3523                 ret = write_dev_flush(dev, 1);
3524                 if (ret)
3525                         errors_wait++;
3526         }
3527         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3528             errors_wait > info->num_tolerated_disk_barrier_failures)
3529                 return -EIO;
3530         return 0;
3531 }
3532
3533 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3534 {
3535         int raid_type;
3536         int min_tolerated = INT_MAX;
3537
3538         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3539             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3540                 min_tolerated = min(min_tolerated,
3541                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3542                                     tolerated_failures);
3543
3544         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3545                 if (raid_type == BTRFS_RAID_SINGLE)
3546                         continue;
3547                 if (!(flags & btrfs_raid_group[raid_type]))
3548                         continue;
3549                 min_tolerated = min(min_tolerated,
3550                                     btrfs_raid_array[raid_type].
3551                                     tolerated_failures);
3552         }
3553
3554         if (min_tolerated == INT_MAX) {
3555                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3556                 min_tolerated = 0;
3557         }
3558
3559         return min_tolerated;
3560 }
3561
3562 int btrfs_calc_num_tolerated_disk_barrier_failures(
3563         struct btrfs_fs_info *fs_info)
3564 {
3565         struct btrfs_ioctl_space_info space;
3566         struct btrfs_space_info *sinfo;
3567         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3568                        BTRFS_BLOCK_GROUP_SYSTEM,
3569                        BTRFS_BLOCK_GROUP_METADATA,
3570                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3571         int i;
3572         int c;
3573         int num_tolerated_disk_barrier_failures =
3574                 (int)fs_info->fs_devices->num_devices;
3575
3576         for (i = 0; i < ARRAY_SIZE(types); i++) {
3577                 struct btrfs_space_info *tmp;
3578
3579                 sinfo = NULL;
3580                 rcu_read_lock();
3581                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3582                         if (tmp->flags == types[i]) {
3583                                 sinfo = tmp;
3584                                 break;
3585                         }
3586                 }
3587                 rcu_read_unlock();
3588
3589                 if (!sinfo)
3590                         continue;
3591
3592                 down_read(&sinfo->groups_sem);
3593                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3594                         u64 flags;
3595
3596                         if (list_empty(&sinfo->block_groups[c]))
3597                                 continue;
3598
3599                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3600                                                    &space);
3601                         if (space.total_bytes == 0 || space.used_bytes == 0)
3602                                 continue;
3603                         flags = space.flags;
3604
3605                         num_tolerated_disk_barrier_failures = min(
3606                                 num_tolerated_disk_barrier_failures,
3607                                 btrfs_get_num_tolerated_disk_barrier_failures(
3608                                         flags));
3609                 }
3610                 up_read(&sinfo->groups_sem);
3611         }
3612
3613         return num_tolerated_disk_barrier_failures;
3614 }
3615
3616 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3617 {
3618         struct list_head *head;
3619         struct btrfs_device *dev;
3620         struct btrfs_super_block *sb;
3621         struct btrfs_dev_item *dev_item;
3622         int ret;
3623         int do_barriers;
3624         int max_errors;
3625         int total_errors = 0;
3626         u64 flags;
3627
3628         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3629         backup_super_roots(root->fs_info);
3630
3631         sb = root->fs_info->super_for_commit;
3632         dev_item = &sb->dev_item;
3633
3634         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3635         head = &root->fs_info->fs_devices->devices;
3636         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3637
3638         if (do_barriers) {
3639                 ret = barrier_all_devices(root->fs_info);
3640                 if (ret) {
3641                         mutex_unlock(
3642                                 &root->fs_info->fs_devices->device_list_mutex);
3643                         btrfs_std_error(root->fs_info, ret,
3644                                     "errors while submitting device barriers.");
3645                         return ret;
3646                 }
3647         }
3648
3649         list_for_each_entry_rcu(dev, head, dev_list) {
3650                 if (!dev->bdev) {
3651                         total_errors++;
3652                         continue;
3653                 }
3654                 if (!dev->in_fs_metadata || !dev->writeable)
3655                         continue;
3656
3657                 btrfs_set_stack_device_generation(dev_item, 0);
3658                 btrfs_set_stack_device_type(dev_item, dev->type);
3659                 btrfs_set_stack_device_id(dev_item, dev->devid);
3660                 btrfs_set_stack_device_total_bytes(dev_item,
3661                                                    dev->commit_total_bytes);
3662                 btrfs_set_stack_device_bytes_used(dev_item,
3663                                                   dev->commit_bytes_used);
3664                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3665                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3666                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3667                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3668                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3669
3670                 flags = btrfs_super_flags(sb);
3671                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3672
3673                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3674                 if (ret)
3675                         total_errors++;
3676         }
3677         if (total_errors > max_errors) {
3678                 btrfs_err(root->fs_info, "%d errors while writing supers",
3679                        total_errors);
3680                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3681
3682                 /* FUA is masked off if unsupported and can't be the reason */
3683                 btrfs_std_error(root->fs_info, -EIO,
3684                             "%d errors while writing supers", total_errors);
3685                 return -EIO;
3686         }
3687
3688         total_errors = 0;
3689         list_for_each_entry_rcu(dev, head, dev_list) {
3690                 if (!dev->bdev)
3691                         continue;
3692                 if (!dev->in_fs_metadata || !dev->writeable)
3693                         continue;
3694
3695                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3696                 if (ret)
3697                         total_errors++;
3698         }
3699         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3700         if (total_errors > max_errors) {
3701                 btrfs_std_error(root->fs_info, -EIO,
3702                             "%d errors while writing supers", total_errors);
3703                 return -EIO;
3704         }
3705         return 0;
3706 }
3707
3708 int write_ctree_super(struct btrfs_trans_handle *trans,
3709                       struct btrfs_root *root, int max_mirrors)
3710 {
3711         return write_all_supers(root, max_mirrors);
3712 }
3713
3714 /* Drop a fs root from the radix tree and free it. */
3715 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3716                                   struct btrfs_root *root)
3717 {
3718         spin_lock(&fs_info->fs_roots_radix_lock);
3719         radix_tree_delete(&fs_info->fs_roots_radix,
3720                           (unsigned long)root->root_key.objectid);
3721         spin_unlock(&fs_info->fs_roots_radix_lock);
3722
3723         if (btrfs_root_refs(&root->root_item) == 0)
3724                 synchronize_srcu(&fs_info->subvol_srcu);
3725
3726         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3727                 btrfs_free_log(NULL, root);
3728
3729         if (root->free_ino_pinned)
3730                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3731         if (root->free_ino_ctl)
3732                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3733         free_fs_root(root);
3734 }
3735
3736 static void free_fs_root(struct btrfs_root *root)
3737 {
3738         iput(root->ino_cache_inode);
3739         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3740         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3741         root->orphan_block_rsv = NULL;
3742         if (root->anon_dev)
3743                 free_anon_bdev(root->anon_dev);
3744         if (root->subv_writers)
3745                 btrfs_free_subvolume_writers(root->subv_writers);
3746         free_extent_buffer(root->node);
3747         free_extent_buffer(root->commit_root);
3748         kfree(root->free_ino_ctl);
3749         kfree(root->free_ino_pinned);
3750         kfree(root->name);
3751         btrfs_put_fs_root(root);
3752 }
3753
3754 void btrfs_free_fs_root(struct btrfs_root *root)
3755 {
3756         free_fs_root(root);
3757 }
3758
3759 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3760 {
3761         u64 root_objectid = 0;
3762         struct btrfs_root *gang[8];
3763         int i = 0;
3764         int err = 0;
3765         unsigned int ret = 0;
3766         int index;
3767
3768         while (1) {
3769                 index = srcu_read_lock(&fs_info->subvol_srcu);
3770                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3771                                              (void **)gang, root_objectid,
3772                                              ARRAY_SIZE(gang));
3773                 if (!ret) {
3774                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3775                         break;
3776                 }
3777                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3778
3779                 for (i = 0; i < ret; i++) {
3780                         /* Avoid to grab roots in dead_roots */
3781                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3782                                 gang[i] = NULL;
3783                                 continue;
3784                         }
3785                         /* grab all the search result for later use */
3786                         gang[i] = btrfs_grab_fs_root(gang[i]);
3787                 }
3788                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3789
3790                 for (i = 0; i < ret; i++) {
3791                         if (!gang[i])
3792                                 continue;
3793                         root_objectid = gang[i]->root_key.objectid;
3794                         err = btrfs_orphan_cleanup(gang[i]);
3795                         if (err)
3796                                 break;
3797                         btrfs_put_fs_root(gang[i]);
3798                 }
3799                 root_objectid++;
3800         }
3801
3802         /* release the uncleaned roots due to error */
3803         for (; i < ret; i++) {
3804                 if (gang[i])
3805                         btrfs_put_fs_root(gang[i]);
3806         }
3807         return err;
3808 }
3809
3810 int btrfs_commit_super(struct btrfs_root *root)
3811 {
3812         struct btrfs_trans_handle *trans;
3813
3814         mutex_lock(&root->fs_info->cleaner_mutex);
3815         btrfs_run_delayed_iputs(root);
3816         mutex_unlock(&root->fs_info->cleaner_mutex);
3817         wake_up_process(root->fs_info->cleaner_kthread);
3818
3819         /* wait until ongoing cleanup work done */
3820         down_write(&root->fs_info->cleanup_work_sem);
3821         up_write(&root->fs_info->cleanup_work_sem);
3822
3823         trans = btrfs_join_transaction(root);
3824         if (IS_ERR(trans))
3825                 return PTR_ERR(trans);
3826         return btrfs_commit_transaction(trans, root);
3827 }
3828
3829 void close_ctree(struct btrfs_root *root)
3830 {
3831         struct btrfs_fs_info *fs_info = root->fs_info;
3832         int ret;
3833
3834         fs_info->closing = 1;
3835         smp_mb();
3836
3837         /* wait for the qgroup rescan worker to stop */
3838         btrfs_qgroup_wait_for_completion(fs_info);
3839
3840         /* wait for the uuid_scan task to finish */
3841         down(&fs_info->uuid_tree_rescan_sem);
3842         /* avoid complains from lockdep et al., set sem back to initial state */
3843         up(&fs_info->uuid_tree_rescan_sem);
3844
3845         /* pause restriper - we want to resume on mount */
3846         btrfs_pause_balance(fs_info);
3847
3848         btrfs_dev_replace_suspend_for_unmount(fs_info);
3849
3850         btrfs_scrub_cancel(fs_info);
3851
3852         /* wait for any defraggers to finish */
3853         wait_event(fs_info->transaction_wait,
3854                    (atomic_read(&fs_info->defrag_running) == 0));
3855
3856         /* clear out the rbtree of defraggable inodes */
3857         btrfs_cleanup_defrag_inodes(fs_info);
3858
3859         cancel_work_sync(&fs_info->async_reclaim_work);
3860
3861         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3862                 /*
3863                  * If the cleaner thread is stopped and there are
3864                  * block groups queued for removal, the deletion will be
3865                  * skipped when we quit the cleaner thread.
3866                  */
3867                 btrfs_delete_unused_bgs(root->fs_info);
3868
3869                 ret = btrfs_commit_super(root);
3870                 if (ret)
3871                         btrfs_err(fs_info, "commit super ret %d", ret);
3872         }
3873
3874         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3875                 btrfs_error_commit_super(root);
3876
3877         kthread_stop(fs_info->transaction_kthread);
3878         kthread_stop(fs_info->cleaner_kthread);
3879
3880         fs_info->closing = 2;
3881         smp_mb();
3882
3883         btrfs_free_qgroup_config(fs_info);
3884
3885         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3886                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3887                        percpu_counter_sum(&fs_info->delalloc_bytes));
3888         }
3889
3890         btrfs_sysfs_remove_mounted(fs_info);
3891         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3892
3893         btrfs_free_fs_roots(fs_info);
3894
3895         btrfs_put_block_group_cache(fs_info);
3896
3897         btrfs_free_block_groups(fs_info);
3898
3899         /*
3900          * we must make sure there is not any read request to
3901          * submit after we stopping all workers.
3902          */
3903         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3904         btrfs_stop_all_workers(fs_info);
3905
3906         fs_info->open = 0;
3907         free_root_pointers(fs_info, 1);
3908
3909         iput(fs_info->btree_inode);
3910
3911 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3912         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3913                 btrfsic_unmount(root, fs_info->fs_devices);
3914 #endif
3915
3916         btrfs_close_devices(fs_info->fs_devices);
3917         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3918
3919         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3920         percpu_counter_destroy(&fs_info->delalloc_bytes);
3921         percpu_counter_destroy(&fs_info->bio_counter);
3922         bdi_destroy(&fs_info->bdi);
3923         cleanup_srcu_struct(&fs_info->subvol_srcu);
3924
3925         btrfs_free_stripe_hash_table(fs_info);
3926
3927         __btrfs_free_block_rsv(root->orphan_block_rsv);
3928         root->orphan_block_rsv = NULL;
3929
3930         lock_chunks(root);
3931         while (!list_empty(&fs_info->pinned_chunks)) {
3932                 struct extent_map *em;
3933
3934                 em = list_first_entry(&fs_info->pinned_chunks,
3935                                       struct extent_map, list);
3936                 list_del_init(&em->list);
3937                 free_extent_map(em);
3938         }
3939         unlock_chunks(root);
3940 }
3941
3942 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3943                           int atomic)
3944 {
3945         int ret;
3946         struct inode *btree_inode = buf->pages[0]->mapping->host;
3947
3948         ret = extent_buffer_uptodate(buf);
3949         if (!ret)
3950                 return ret;
3951
3952         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3953                                     parent_transid, atomic);
3954         if (ret == -EAGAIN)
3955                 return ret;
3956         return !ret;
3957 }
3958
3959 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3960 {
3961         struct btrfs_root *root;
3962         u64 transid = btrfs_header_generation(buf);
3963         int was_dirty;
3964
3965 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3966         /*
3967          * This is a fast path so only do this check if we have sanity tests
3968          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3969          * outside of the sanity tests.
3970          */
3971         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3972                 return;
3973 #endif
3974         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3975         btrfs_assert_tree_locked(buf);
3976         if (transid != root->fs_info->generation)
3977                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3978                        "found %llu running %llu\n",
3979                         buf->start, transid, root->fs_info->generation);
3980         was_dirty = set_extent_buffer_dirty(buf);
3981         if (!was_dirty)
3982                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3983                                      buf->len,
3984                                      root->fs_info->dirty_metadata_batch);
3985 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3986         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3987                 btrfs_print_leaf(root, buf);
3988                 ASSERT(0);
3989         }
3990 #endif
3991 }
3992
3993 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3994                                         int flush_delayed)
3995 {
3996         /*
3997          * looks as though older kernels can get into trouble with
3998          * this code, they end up stuck in balance_dirty_pages forever
3999          */
4000         int ret;
4001
4002         if (current->flags & PF_MEMALLOC)
4003                 return;
4004
4005         if (flush_delayed)
4006                 btrfs_balance_delayed_items(root);
4007
4008         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4009                                      BTRFS_DIRTY_METADATA_THRESH);
4010         if (ret > 0) {
4011                 balance_dirty_pages_ratelimited(
4012                                    root->fs_info->btree_inode->i_mapping);
4013         }
4014 }
4015
4016 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4017 {
4018         __btrfs_btree_balance_dirty(root, 1);
4019 }
4020
4021 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4022 {
4023         __btrfs_btree_balance_dirty(root, 0);
4024 }
4025
4026 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4027 {
4028         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4029         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4030 }
4031
4032 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4033                               int read_only)
4034 {
4035         struct btrfs_super_block *sb = fs_info->super_copy;
4036         u64 nodesize = btrfs_super_nodesize(sb);
4037         u64 sectorsize = btrfs_super_sectorsize(sb);
4038         int ret = 0;
4039
4040         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4041                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4042                 ret = -EINVAL;
4043         }
4044         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4045                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4046                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4047         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4048                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4049                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4050                 ret = -EINVAL;
4051         }
4052         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4053                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4054                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4055                 ret = -EINVAL;
4056         }
4057         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4058                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4059                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4060                 ret = -EINVAL;
4061         }
4062
4063         /*
4064          * Check sectorsize and nodesize first, other check will need it.
4065          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4066          */
4067         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4068             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4069                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4070                 ret = -EINVAL;
4071         }
4072         /* Only PAGE SIZE is supported yet */
4073         if (sectorsize != PAGE_CACHE_SIZE) {
4074                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4075                                 sectorsize, PAGE_CACHE_SIZE);
4076                 ret = -EINVAL;
4077         }
4078         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4079             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4080                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4081                 ret = -EINVAL;
4082         }
4083         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4084                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4085                                 le32_to_cpu(sb->__unused_leafsize),
4086                                 nodesize);
4087                 ret = -EINVAL;
4088         }
4089
4090         /* Root alignment check */
4091         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4092                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4093                                 btrfs_super_root(sb));
4094                 ret = -EINVAL;
4095         }
4096         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4097                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4098                                 btrfs_super_chunk_root(sb));
4099                 ret = -EINVAL;
4100         }
4101         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4102                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4103                                 btrfs_super_log_root(sb));
4104                 ret = -EINVAL;
4105         }
4106
4107         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4108                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4109                                 fs_info->fsid, sb->dev_item.fsid);
4110                 ret = -EINVAL;
4111         }
4112
4113         /*
4114          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4115          * done later
4116          */
4117         if (btrfs_super_num_devices(sb) > (1UL << 31))
4118                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4119                                 btrfs_super_num_devices(sb));
4120         if (btrfs_super_num_devices(sb) == 0) {
4121                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4122                 ret = -EINVAL;
4123         }
4124
4125         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4126                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4127                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4128                 ret = -EINVAL;
4129         }
4130
4131         /*
4132          * Obvious sys_chunk_array corruptions, it must hold at least one key
4133          * and one chunk
4134          */
4135         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4136                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4137                                 btrfs_super_sys_array_size(sb),
4138                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4139                 ret = -EINVAL;
4140         }
4141         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4142                         + sizeof(struct btrfs_chunk)) {
4143                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4144                                 btrfs_super_sys_array_size(sb),
4145                                 sizeof(struct btrfs_disk_key)
4146                                 + sizeof(struct btrfs_chunk));
4147                 ret = -EINVAL;
4148         }
4149
4150         /*
4151          * The generation is a global counter, we'll trust it more than the others
4152          * but it's still possible that it's the one that's wrong.
4153          */
4154         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4155                 printk(KERN_WARNING
4156                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4157                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4158         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4159             && btrfs_super_cache_generation(sb) != (u64)-1)
4160                 printk(KERN_WARNING
4161                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4162                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4163
4164         return ret;
4165 }
4166
4167 static void btrfs_error_commit_super(struct btrfs_root *root)
4168 {
4169         mutex_lock(&root->fs_info->cleaner_mutex);
4170         btrfs_run_delayed_iputs(root);
4171         mutex_unlock(&root->fs_info->cleaner_mutex);
4172
4173         down_write(&root->fs_info->cleanup_work_sem);
4174         up_write(&root->fs_info->cleanup_work_sem);
4175
4176         /* cleanup FS via transaction */
4177         btrfs_cleanup_transaction(root);
4178 }
4179
4180 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4181 {
4182         struct btrfs_ordered_extent *ordered;
4183
4184         spin_lock(&root->ordered_extent_lock);
4185         /*
4186          * This will just short circuit the ordered completion stuff which will
4187          * make sure the ordered extent gets properly cleaned up.
4188          */
4189         list_for_each_entry(ordered, &root->ordered_extents,
4190                             root_extent_list)
4191                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4192         spin_unlock(&root->ordered_extent_lock);
4193 }
4194
4195 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4196 {
4197         struct btrfs_root *root;
4198         struct list_head splice;
4199
4200         INIT_LIST_HEAD(&splice);
4201
4202         spin_lock(&fs_info->ordered_root_lock);
4203         list_splice_init(&fs_info->ordered_roots, &splice);
4204         while (!list_empty(&splice)) {
4205                 root = list_first_entry(&splice, struct btrfs_root,
4206                                         ordered_root);
4207                 list_move_tail(&root->ordered_root,
4208                                &fs_info->ordered_roots);
4209
4210                 spin_unlock(&fs_info->ordered_root_lock);
4211                 btrfs_destroy_ordered_extents(root);
4212
4213                 cond_resched();
4214                 spin_lock(&fs_info->ordered_root_lock);
4215         }
4216         spin_unlock(&fs_info->ordered_root_lock);
4217 }
4218
4219 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4220                                       struct btrfs_root *root)
4221 {
4222         struct rb_node *node;
4223         struct btrfs_delayed_ref_root *delayed_refs;
4224         struct btrfs_delayed_ref_node *ref;
4225         int ret = 0;
4226
4227         delayed_refs = &trans->delayed_refs;
4228
4229         spin_lock(&delayed_refs->lock);
4230         if (atomic_read(&delayed_refs->num_entries) == 0) {
4231                 spin_unlock(&delayed_refs->lock);
4232                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4233                 return ret;
4234         }
4235
4236         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4237                 struct btrfs_delayed_ref_head *head;
4238                 struct btrfs_delayed_ref_node *tmp;
4239                 bool pin_bytes = false;
4240
4241                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4242                                 href_node);
4243                 if (!mutex_trylock(&head->mutex)) {
4244                         atomic_inc(&head->node.refs);
4245                         spin_unlock(&delayed_refs->lock);
4246
4247                         mutex_lock(&head->mutex);
4248                         mutex_unlock(&head->mutex);
4249                         btrfs_put_delayed_ref(&head->node);
4250                         spin_lock(&delayed_refs->lock);
4251                         continue;
4252                 }
4253                 spin_lock(&head->lock);
4254                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4255                                                  list) {
4256                         ref->in_tree = 0;
4257                         list_del(&ref->list);
4258                         atomic_dec(&delayed_refs->num_entries);
4259                         btrfs_put_delayed_ref(ref);
4260                 }
4261                 if (head->must_insert_reserved)
4262                         pin_bytes = true;
4263                 btrfs_free_delayed_extent_op(head->extent_op);
4264                 delayed_refs->num_heads--;
4265                 if (head->processing == 0)
4266                         delayed_refs->num_heads_ready--;
4267                 atomic_dec(&delayed_refs->num_entries);
4268                 head->node.in_tree = 0;
4269                 rb_erase(&head->href_node, &delayed_refs->href_root);
4270                 spin_unlock(&head->lock);
4271                 spin_unlock(&delayed_refs->lock);
4272                 mutex_unlock(&head->mutex);
4273
4274                 if (pin_bytes)
4275                         btrfs_pin_extent(root, head->node.bytenr,
4276                                          head->node.num_bytes, 1);
4277                 btrfs_put_delayed_ref(&head->node);
4278                 cond_resched();
4279                 spin_lock(&delayed_refs->lock);
4280         }
4281
4282         spin_unlock(&delayed_refs->lock);
4283
4284         return ret;
4285 }
4286
4287 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4288 {
4289         struct btrfs_inode *btrfs_inode;
4290         struct list_head splice;
4291
4292         INIT_LIST_HEAD(&splice);
4293
4294         spin_lock(&root->delalloc_lock);
4295         list_splice_init(&root->delalloc_inodes, &splice);
4296
4297         while (!list_empty(&splice)) {
4298                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4299                                                delalloc_inodes);
4300
4301                 list_del_init(&btrfs_inode->delalloc_inodes);
4302                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4303                           &btrfs_inode->runtime_flags);
4304                 spin_unlock(&root->delalloc_lock);
4305
4306                 btrfs_invalidate_inodes(btrfs_inode->root);
4307
4308                 spin_lock(&root->delalloc_lock);
4309         }
4310
4311         spin_unlock(&root->delalloc_lock);
4312 }
4313
4314 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4315 {
4316         struct btrfs_root *root;
4317         struct list_head splice;
4318
4319         INIT_LIST_HEAD(&splice);
4320
4321         spin_lock(&fs_info->delalloc_root_lock);
4322         list_splice_init(&fs_info->delalloc_roots, &splice);
4323         while (!list_empty(&splice)) {
4324                 root = list_first_entry(&splice, struct btrfs_root,
4325                                          delalloc_root);
4326                 list_del_init(&root->delalloc_root);
4327                 root = btrfs_grab_fs_root(root);
4328                 BUG_ON(!root);
4329                 spin_unlock(&fs_info->delalloc_root_lock);
4330
4331                 btrfs_destroy_delalloc_inodes(root);
4332                 btrfs_put_fs_root(root);
4333
4334                 spin_lock(&fs_info->delalloc_root_lock);
4335         }
4336         spin_unlock(&fs_info->delalloc_root_lock);
4337 }
4338
4339 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4340                                         struct extent_io_tree *dirty_pages,
4341                                         int mark)
4342 {
4343         int ret;
4344         struct extent_buffer *eb;
4345         u64 start = 0;
4346         u64 end;
4347
4348         while (1) {
4349                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4350                                             mark, NULL);
4351                 if (ret)
4352                         break;
4353
4354                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4355                 while (start <= end) {
4356                         eb = btrfs_find_tree_block(root->fs_info, start);
4357                         start += root->nodesize;
4358                         if (!eb)
4359                                 continue;
4360                         wait_on_extent_buffer_writeback(eb);
4361
4362                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4363                                                &eb->bflags))
4364                                 clear_extent_buffer_dirty(eb);
4365                         free_extent_buffer_stale(eb);
4366                 }
4367         }
4368
4369         return ret;
4370 }
4371
4372 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4373                                        struct extent_io_tree *pinned_extents)
4374 {
4375         struct extent_io_tree *unpin;
4376         u64 start;
4377         u64 end;
4378         int ret;
4379         bool loop = true;
4380
4381         unpin = pinned_extents;
4382 again:
4383         while (1) {
4384                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4385                                             EXTENT_DIRTY, NULL);
4386                 if (ret)
4387                         break;
4388
4389                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4390                 btrfs_error_unpin_extent_range(root, start, end);
4391                 cond_resched();
4392         }
4393
4394         if (loop) {
4395                 if (unpin == &root->fs_info->freed_extents[0])
4396                         unpin = &root->fs_info->freed_extents[1];
4397                 else
4398                         unpin = &root->fs_info->freed_extents[0];
4399                 loop = false;
4400                 goto again;
4401         }
4402
4403         return 0;
4404 }
4405
4406 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4407                                    struct btrfs_root *root)
4408 {
4409         btrfs_destroy_delayed_refs(cur_trans, root);
4410
4411         cur_trans->state = TRANS_STATE_COMMIT_START;
4412         wake_up(&root->fs_info->transaction_blocked_wait);
4413
4414         cur_trans->state = TRANS_STATE_UNBLOCKED;
4415         wake_up(&root->fs_info->transaction_wait);
4416
4417         btrfs_destroy_delayed_inodes(root);
4418         btrfs_assert_delayed_root_empty(root);
4419
4420         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4421                                      EXTENT_DIRTY);
4422         btrfs_destroy_pinned_extent(root,
4423                                     root->fs_info->pinned_extents);
4424
4425         cur_trans->state =TRANS_STATE_COMPLETED;
4426         wake_up(&cur_trans->commit_wait);
4427
4428         /*
4429         memset(cur_trans, 0, sizeof(*cur_trans));
4430         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4431         */
4432 }
4433
4434 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4435 {
4436         struct btrfs_transaction *t;
4437
4438         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4439
4440         spin_lock(&root->fs_info->trans_lock);
4441         while (!list_empty(&root->fs_info->trans_list)) {
4442                 t = list_first_entry(&root->fs_info->trans_list,
4443                                      struct btrfs_transaction, list);
4444                 if (t->state >= TRANS_STATE_COMMIT_START) {
4445                         atomic_inc(&t->use_count);
4446                         spin_unlock(&root->fs_info->trans_lock);
4447                         btrfs_wait_for_commit(root, t->transid);
4448                         btrfs_put_transaction(t);
4449                         spin_lock(&root->fs_info->trans_lock);
4450                         continue;
4451                 }
4452                 if (t == root->fs_info->running_transaction) {
4453                         t->state = TRANS_STATE_COMMIT_DOING;
4454                         spin_unlock(&root->fs_info->trans_lock);
4455                         /*
4456                          * We wait for 0 num_writers since we don't hold a trans
4457                          * handle open currently for this transaction.
4458                          */
4459                         wait_event(t->writer_wait,
4460                                    atomic_read(&t->num_writers) == 0);
4461                 } else {
4462                         spin_unlock(&root->fs_info->trans_lock);
4463                 }
4464                 btrfs_cleanup_one_transaction(t, root);
4465
4466                 spin_lock(&root->fs_info->trans_lock);
4467                 if (t == root->fs_info->running_transaction)
4468                         root->fs_info->running_transaction = NULL;
4469                 list_del_init(&t->list);
4470                 spin_unlock(&root->fs_info->trans_lock);
4471
4472                 btrfs_put_transaction(t);
4473                 trace_btrfs_transaction_commit(root);
4474                 spin_lock(&root->fs_info->trans_lock);
4475         }
4476         spin_unlock(&root->fs_info->trans_lock);
4477         btrfs_destroy_all_ordered_extents(root->fs_info);
4478         btrfs_destroy_delayed_inodes(root);
4479         btrfs_assert_delayed_root_empty(root);
4480         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4481         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4482         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4483
4484         return 0;
4485 }
4486
4487 static const struct extent_io_ops btree_extent_io_ops = {
4488         .readpage_end_io_hook = btree_readpage_end_io_hook,
4489         .readpage_io_failed_hook = btree_io_failed_hook,
4490         .submit_bio_hook = btree_submit_bio_hook,
4491         /* note we're sharing with inode.c for the merge bio hook */
4492         .merge_bio_hook = btrfs_merge_bio_hook,
4493 };