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 "async-thread.h"
  43 #include "locking.h"
  44 #include "tree-log.h"
  45 #include "free-space-cache.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
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  63                                              struct btrfs_root *root);
  64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  66                                       struct btrfs_root *root);
  67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  69                                         struct extent_io_tree *dirty_pages,
  70                                         int mark);
  71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  72                                        struct extent_io_tree *pinned_extents);
  73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  74 static void btrfs_error_commit_super(struct btrfs_root *root);
  75
  76 /*
  77  * end_io_wq structs are used to do processing in task context when an IO is
  78  * complete.  This is used during reads to verify checksums, and it is used
  79  * by writes to insert metadata for new file extents after IO is complete.
  80  */
  81 struct end_io_wq {
  82         struct bio *bio;
  83         bio_end_io_t *end_io;
  84         void *private;
  85         struct btrfs_fs_info *info;
  86         int error;
  87         int metadata;
  88         struct list_head list;
  89         struct btrfs_work work;
  90 };
  91
  92 /*
  93  * async submit bios are used to offload expensive checksumming
  94  * onto the worker threads.  They checksum file and metadata bios
  95  * just before they are sent down the IO stack.
  96  */
  97 struct async_submit_bio {
  98         struct inode *inode;
  99         struct bio *bio;
 100         struct list_head list;
 101         extent_submit_bio_hook_t *submit_bio_start;
 102         extent_submit_bio_hook_t *submit_bio_done;
 103         int rw;
 104         int mirror_num;
 105         unsigned long bio_flags;
 106         /*
 107          * bio_offset is optional, can be used if the pages in the bio
 108          * can't tell us where in the file the bio should go
 109          */
 110         u64 bio_offset;
 111         struct btrfs_work work;
 112         int error;
 113 };
 114
 115 /*
 116  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 117  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 118  * the level the eb occupies in the tree.
 119  *
 120  * Different roots are used for different purposes and may nest inside each
 121  * other and they require separate keysets.  As lockdep keys should be
 122  * static, assign keysets according to the purpose of the root as indicated
 123  * by btrfs_root->objectid.  This ensures that all special purpose roots
 124  * have separate keysets.
 125  *
 126  * Lock-nesting across peer nodes is always done with the immediate parent
 127  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 128  * subclass to avoid triggering lockdep warning in such cases.
 129  *
 130  * The key is set by the readpage_end_io_hook after the buffer has passed
 131  * csum validation but before the pages are unlocked.  It is also set by
 132  * btrfs_init_new_buffer on freshly allocated blocks.
 133  *
 134  * We also add a check to make sure the highest level of the tree is the
 135  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 136  * needs update as well.
 137  */
 138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 139 # if BTRFS_MAX_LEVEL != 8
 140 #  error
 141 # endif
 142
 143 static struct btrfs_lockdep_keyset {
 144         u64                     id;             /* root objectid */
 145         const char              *name_stem;     /* lock name stem */
 146         char                    names[BTRFS_MAX_LEVEL + 1][20];
 147         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 148 } btrfs_lockdep_keysets[] = {
 149         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 150         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 151         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 152         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 153         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 154         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 155         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 156         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 157         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 158         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 159         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 160         { .id = 0,                              .name_stem = "tree"     },
 161 };
 162
 163 void __init btrfs_init_lockdep(void)
 164 {
 165         int i, j;
 166
 167         /* initialize lockdep class names */
 168         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 169                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 170
 171                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 172                         snprintf(ks->names[j], sizeof(ks->names[j]),
 173                                  "btrfs-%s-%02d", ks->name_stem, j);
 174         }
 175 }
 176
 177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 178                                     int level)
 179 {
 180         struct btrfs_lockdep_keyset *ks;
 181
 182         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 183
 184         /* find the matching keyset, id 0 is the default entry */
 185         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 186                 if (ks->id == objectid)
 187                         break;
 188
 189         lockdep_set_class_and_name(&eb->lock,
 190                                    &ks->keys[level], ks->names[level]);
 191 }
 192
 193 #endif
 194
 195 /*
 196  * extents on the btree inode are pretty simple, there's one extent
 197  * that covers the entire device
 198  */
 199 static struct extent_map *btree_get_extent(struct inode *inode,
 200                 struct page *page, size_t pg_offset, u64 start, u64 len,
 201                 int create)
 202 {
 203         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 204         struct extent_map *em;
 205         int ret;
 206
 207         read_lock(&em_tree->lock);
 208         em = lookup_extent_mapping(em_tree, start, len);
 209         if (em) {
 210                 em->bdev =
 211                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 212                 read_unlock(&em_tree->lock);
 213                 goto out;
 214         }
 215         read_unlock(&em_tree->lock);
 216
 217         em = alloc_extent_map();
 218         if (!em) {
 219                 em = ERR_PTR(-ENOMEM);
 220                 goto out;
 221         }
 222         em->start = 0;
 223         em->len = (u64)-1;
 224         em->block_len = (u64)-1;
 225         em->block_start = 0;
 226         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 227
 228         write_lock(&em_tree->lock);
 229         ret = add_extent_mapping(em_tree, em, 0);
 230         if (ret == -EEXIST) {
 231                 free_extent_map(em);
 232                 em = lookup_extent_mapping(em_tree, start, len);
 233                 if (!em)
 234                         em = ERR_PTR(-EIO);
 235         } else if (ret) {
 236                 free_extent_map(em);
 237                 em = ERR_PTR(ret);
 238         }
 239         write_unlock(&em_tree->lock);
 240
 241 out:
 242         return em;
 243 }
 244
 245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 246 {
 247         return btrfs_crc32c(seed, data, len);
 248 }
 249
 250 void btrfs_csum_final(u32 crc, char *result)
 251 {
 252         put_unaligned_le32(~crc, result);
 253 }
 254
 255 /*
 256  * compute the csum for a btree block, and either verify it or write it
 257  * into the csum field of the block.
 258  */
 259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 260                            int verify)
 261 {
 262         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 263         char *result = NULL;
 264         unsigned long len;
 265         unsigned long cur_len;
 266         unsigned long offset = BTRFS_CSUM_SIZE;
 267         char *kaddr;
 268         unsigned long map_start;
 269         unsigned long map_len;
 270         int err;
 271         u32 crc = ~(u32)0;
 272         unsigned long inline_result;
 273
 274         len = buf->len - offset;
 275         while (len > 0) {
 276                 err = map_private_extent_buffer(buf, offset, 32,
 277                                         &kaddr, &map_start, &map_len);
 278                 if (err)
 279                         return 1;
 280                 cur_len = min(len, map_len - (offset - map_start));
 281                 crc = btrfs_csum_data(kaddr + offset - map_start,
 282                                       crc, cur_len);
 283                 len -= cur_len;
 284                 offset += cur_len;
 285         }
 286         if (csum_size > sizeof(inline_result)) {
 287                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 288                 if (!result)
 289                         return 1;
 290         } else {
 291                 result = (char *)&inline_result;
 292         }
 293
 294         btrfs_csum_final(crc, result);
 295
 296         if (verify) {
 297                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 298                         u32 val;
 299                         u32 found = 0;
 300                         memcpy(&found, result, csum_size);
 301
 302                         read_extent_buffer(buf, &val, 0, csum_size);
 303                         printk_ratelimited(KERN_INFO
 304                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 305                                 "level %d\n",
 306                                 root->fs_info->sb->s_id, buf->start,
 307                                 val, found, btrfs_header_level(buf));
 308                         if (result != (char *)&inline_result)
 309                                 kfree(result);
 310                         return 1;
 311                 }
 312         } else {
 313                 write_extent_buffer(buf, result, 0, csum_size);
 314         }
 315         if (result != (char *)&inline_result)
 316                 kfree(result);
 317         return 0;
 318 }
 319
 320 /*
 321  * we can't consider a given block up to date unless the transid of the
 322  * block matches the transid in the parent node's pointer.  This is how we
 323  * detect blocks that either didn't get written at all or got written
 324  * in the wrong place.
 325  */
 326 static int verify_parent_transid(struct extent_io_tree *io_tree,
 327                                  struct extent_buffer *eb, u64 parent_transid,
 328                                  int atomic)
 329 {
 330         struct extent_state *cached_state = NULL;
 331         int ret;
 332         bool need_lock = (current->journal_info ==
 333                           (void *)BTRFS_SEND_TRANS_STUB);
 334
 335         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 336                 return 0;
 337
 338         if (atomic)
 339                 return -EAGAIN;
 340
 341         if (need_lock) {
 342                 btrfs_tree_read_lock(eb);
 343                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 344         }
 345
 346         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 347                          0, &cached_state);
 348         if (extent_buffer_uptodate(eb) &&
 349             btrfs_header_generation(eb) == parent_transid) {
 350                 ret = 0;
 351                 goto out;
 352         }
 353         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 354                        "found %llu\n",
 355                        eb->start, parent_transid, btrfs_header_generation(eb));
 356         ret = 1;
 357
 358         /*
 359          * Things reading via commit roots that don't have normal protection,
 360          * like send, can have a really old block in cache that may point at a
 361          * block that has been free'd and re-allocated.  So don't clear uptodate
 362          * if we find an eb that is under IO (dirty/writeback) because we could
 363          * end up reading in the stale data and then writing it back out and
 364          * making everybody very sad.
 365          */
 366         if (!extent_buffer_under_io(eb))
 367                 clear_extent_buffer_uptodate(eb);
 368 out:
 369         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 370                              &cached_state, GFP_NOFS);
 371         btrfs_tree_read_unlock_blocking(eb);
 372         return ret;
 373 }
 374
 375 /*
 376  * Return 0 if the superblock checksum type matches the checksum value of that
 377  * algorithm. Pass the raw disk superblock data.
 378  */
 379 static int btrfs_check_super_csum(char *raw_disk_sb)
 380 {
 381         struct btrfs_super_block *disk_sb =
 382                 (struct btrfs_super_block *)raw_disk_sb;
 383         u16 csum_type = btrfs_super_csum_type(disk_sb);
 384         int ret = 0;
 385
 386         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 387                 u32 crc = ~(u32)0;
 388                 const int csum_size = sizeof(crc);
 389                 char result[csum_size];
 390
 391                 /*
 392                  * The super_block structure does not span the whole
 393                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 394                  * is filled with zeros and is included in the checkum.
 395                  */
 396                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 397                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 398                 btrfs_csum_final(crc, result);
 399
 400                 if (memcmp(raw_disk_sb, result, csum_size))
 401                         ret = 1;
 402
 403                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 404                         printk(KERN_WARNING
 405                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 406                         ret = 0;
 407                 }
 408         }
 409
 410         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 411                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 412                                 csum_type);
 413                 ret = 1;
 414         }
 415
 416         return ret;
 417 }
 418
 419 /*
 420  * helper to read a given tree block, doing retries as required when
 421  * the checksums don't match and we have alternate mirrors to try.
 422  */
 423 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 424                                           struct extent_buffer *eb,
 425                                           u64 start, u64 parent_transid)
 426 {
 427         struct extent_io_tree *io_tree;
 428         int failed = 0;
 429         int ret;
 430         int num_copies = 0;
 431         int mirror_num = 0;
 432         int failed_mirror = 0;
 433
 434         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 435         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 436         while (1) {
 437                 ret = read_extent_buffer_pages(io_tree, eb, start,
 438                                                WAIT_COMPLETE,
 439                                                btree_get_extent, mirror_num);
 440                 if (!ret) {
 441                         if (!verify_parent_transid(io_tree, eb,
 442                                                    parent_transid, 0))
 443                                 break;
 444                         else
 445                                 ret = -EIO;
 446                 }
 447
 448                 /*
 449                  * This buffer's crc is fine, but its contents are corrupted, so
 450                  * there is no reason to read the other copies, they won't be
 451                  * any less wrong.
 452                  */
 453                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 454                         break;
 455
 456                 num_copies = btrfs_num_copies(root->fs_info,
 457                                               eb->start, eb->len);
 458                 if (num_copies == 1)
 459                         break;
 460
 461                 if (!failed_mirror) {
 462                         failed = 1;
 463                         failed_mirror = eb->read_mirror;
 464                 }
 465
 466                 mirror_num++;
 467                 if (mirror_num == failed_mirror)
 468                         mirror_num++;
 469
 470                 if (mirror_num > num_copies)
 471                         break;
 472         }
 473
 474         if (failed && !ret && failed_mirror)
 475                 repair_eb_io_failure(root, eb, failed_mirror);
 476
 477         return ret;
 478 }
 479
 480 /*
 481  * checksum a dirty tree block before IO.  This has extra checks to make sure
 482  * we only fill in the checksum field in the first page of a multi-page block
 483  */
 484
 485 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 486 {
 487         u64 start = page_offset(page);
 488         u64 found_start;
 489         struct extent_buffer *eb;
 490
 491         eb = (struct extent_buffer *)page->private;
 492         if (page != eb->pages[0])
 493                 return 0;
 494         found_start = btrfs_header_bytenr(eb);
 495         if (WARN_ON(found_start != start || !PageUptodate(page)))
 496                 return 0;
 497         csum_tree_block(root, eb, 0);
 498         return 0;
 499 }
 500
 501 static int check_tree_block_fsid(struct btrfs_root *root,
 502                                  struct extent_buffer *eb)
 503 {
 504         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 505         u8 fsid[BTRFS_UUID_SIZE];
 506         int ret = 1;
 507
 508         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 509         while (fs_devices) {
 510                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 511                         ret = 0;
 512                         break;
 513                 }
 514                 fs_devices = fs_devices->seed;
 515         }
 516         return ret;
 517 }
 518
 519 #define CORRUPT(reason, eb, root, slot)                         \
 520         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 521                    "root=%llu, slot=%d", reason,                        \
 522                btrfs_header_bytenr(eb), root->objectid, slot)
 523
 524 static noinline int check_leaf(struct btrfs_root *root,
 525                                struct extent_buffer *leaf)
 526 {
 527         struct btrfs_key key;
 528         struct btrfs_key leaf_key;
 529         u32 nritems = btrfs_header_nritems(leaf);
 530         int slot;
 531
 532         if (nritems == 0)
 533                 return 0;
 534
 535         /* Check the 0 item */
 536         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 537             BTRFS_LEAF_DATA_SIZE(root)) {
 538                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 539                 return -EIO;
 540         }
 541
 542         /*
 543          * Check to make sure each items keys are in the correct order and their
 544          * offsets make sense.  We only have to loop through nritems-1 because
 545          * we check the current slot against the next slot, which verifies the
 546          * next slot's offset+size makes sense and that the current's slot
 547          * offset is correct.
 548          */
 549         for (slot = 0; slot < nritems - 1; slot++) {
 550                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 551                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 552
 553                 /* Make sure the keys are in the right order */
 554                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 555                         CORRUPT("bad key order", leaf, root, slot);
 556                         return -EIO;
 557                 }
 558
 559                 /*
 560                  * Make sure the offset and ends are right, remember that the
 561                  * item data starts at the end of the leaf and grows towards the
 562                  * front.
 563                  */
 564                 if (btrfs_item_offset_nr(leaf, slot) !=
 565                         btrfs_item_end_nr(leaf, slot + 1)) {
 566                         CORRUPT("slot offset bad", leaf, root, slot);
 567                         return -EIO;
 568                 }
 569
 570                 /*
 571                  * Check to make sure that we don't point outside of the leaf,
 572                  * just incase all the items are consistent to eachother, but
 573                  * all point outside of the leaf.
 574                  */
 575                 if (btrfs_item_end_nr(leaf, slot) >
 576                     BTRFS_LEAF_DATA_SIZE(root)) {
 577                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 578                         return -EIO;
 579                 }
 580         }
 581
 582         return 0;
 583 }
 584
 585 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 586                                       u64 phy_offset, struct page *page,
 587                                       u64 start, u64 end, int mirror)
 588 {
 589         u64 found_start;
 590         int found_level;
 591         struct extent_buffer *eb;
 592         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 593         int ret = 0;
 594         int reads_done;
 595
 596         if (!page->private)
 597                 goto out;
 598
 599         eb = (struct extent_buffer *)page->private;
 600
 601         /* the pending IO might have been the only thing that kept this buffer
 602          * in memory.  Make sure we have a ref for all this other checks
 603          */
 604         extent_buffer_get(eb);
 605
 606         reads_done = atomic_dec_and_test(&eb->io_pages);
 607         if (!reads_done)
 608                 goto err;
 609
 610         eb->read_mirror = mirror;
 611         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
 612                 ret = -EIO;
 613                 goto err;
 614         }
 615
 616         found_start = btrfs_header_bytenr(eb);
 617         if (found_start != eb->start) {
 618                 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
 619                                "%llu %llu\n",
 620                                found_start, eb->start);
 621                 ret = -EIO;
 622                 goto err;
 623         }
 624         if (check_tree_block_fsid(root, eb)) {
 625                 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
 626                                eb->start);
 627                 ret = -EIO;
 628                 goto err;
 629         }
 630         found_level = btrfs_header_level(eb);
 631         if (found_level >= BTRFS_MAX_LEVEL) {
 632                 btrfs_info(root->fs_info, "bad tree block level %d",
 633                            (int)btrfs_header_level(eb));
 634                 ret = -EIO;
 635                 goto err;
 636         }
 637
 638         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 639                                        eb, found_level);
 640
 641         ret = csum_tree_block(root, eb, 1);
 642         if (ret) {
 643                 ret = -EIO;
 644                 goto err;
 645         }
 646
 647         /*
 648          * If this is a leaf block and it is corrupt, set the corrupt bit so
 649          * that we don't try and read the other copies of this block, just
 650          * return -EIO.
 651          */
 652         if (found_level == 0 && check_leaf(root, eb)) {
 653                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 654                 ret = -EIO;
 655         }
 656
 657         if (!ret)
 658                 set_extent_buffer_uptodate(eb);
 659 err:
 660         if (reads_done &&
 661             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 662                 btree_readahead_hook(root, eb, eb->start, ret);
 663
 664         if (ret) {
 665                 /*
 666                  * our io error hook is going to dec the io pages
 667                  * again, we have to make sure it has something
 668                  * to decrement
 669                  */
 670                 atomic_inc(&eb->io_pages);
 671                 clear_extent_buffer_uptodate(eb);
 672         }
 673         free_extent_buffer(eb);
 674 out:
 675         return ret;
 676 }
 677
 678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 679 {
 680         struct extent_buffer *eb;
 681         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 682
 683         eb = (struct extent_buffer *)page->private;
 684         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
 685         eb->read_mirror = failed_mirror;
 686         atomic_dec(&eb->io_pages);
 687         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 688                 btree_readahead_hook(root, eb, eb->start, -EIO);
 689         return -EIO;    /* we fixed nothing */
 690 }
 691
 692 static void end_workqueue_bio(struct bio *bio, int err)
 693 {
 694         struct end_io_wq *end_io_wq = bio->bi_private;
 695         struct btrfs_fs_info *fs_info;
 696
 697         fs_info = end_io_wq->info;
 698         end_io_wq->error = err;
 699         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 700
 701         if (bio->bi_rw & REQ_WRITE) {
 702                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 703                         btrfs_queue_work(fs_info->endio_meta_write_workers,
 704                                          &end_io_wq->work);
 705                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 706                         btrfs_queue_work(fs_info->endio_freespace_worker,
 707                                          &end_io_wq->work);
 708                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 709                         btrfs_queue_work(fs_info->endio_raid56_workers,
 710                                          &end_io_wq->work);
 711                 else
 712                         btrfs_queue_work(fs_info->endio_write_workers,
 713                                          &end_io_wq->work);
 714         } else {
 715                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 716                         btrfs_queue_work(fs_info->endio_raid56_workers,
 717                                          &end_io_wq->work);
 718                 else if (end_io_wq->metadata)
 719                         btrfs_queue_work(fs_info->endio_meta_workers,
 720                                          &end_io_wq->work);
 721                 else
 722                         btrfs_queue_work(fs_info->endio_workers,
 723                                          &end_io_wq->work);
 724         }
 725 }
 726
 727 /*
 728  * For the metadata arg you want
 729  *
 730  * 0 - if data
 731  * 1 - if normal metadta
 732  * 2 - if writing to the free space cache area
 733  * 3 - raid parity work
 734  */
 735 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 736                         int metadata)
 737 {
 738         struct end_io_wq *end_io_wq;
 739         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 740         if (!end_io_wq)
 741                 return -ENOMEM;
 742
 743         end_io_wq->private = bio->bi_private;
 744         end_io_wq->end_io = bio->bi_end_io;
 745         end_io_wq->info = info;
 746         end_io_wq->error = 0;
 747         end_io_wq->bio = bio;
 748         end_io_wq->metadata = metadata;
 749
 750         bio->bi_private = end_io_wq;
 751         bio->bi_end_io = end_workqueue_bio;
 752         return 0;
 753 }
 754
 755 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 756 {
 757         unsigned long limit = min_t(unsigned long,
 758                                     info->thread_pool_size,
 759                                     info->fs_devices->open_devices);
 760         return 256 * limit;
 761 }
 762
 763 static void run_one_async_start(struct btrfs_work *work)
 764 {
 765         struct async_submit_bio *async;
 766         int ret;
 767
 768         async = container_of(work, struct  async_submit_bio, work);
 769         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 770                                       async->mirror_num, async->bio_flags,
 771                                       async->bio_offset);
 772         if (ret)
 773                 async->error = ret;
 774 }
 775
 776 static void run_one_async_done(struct btrfs_work *work)
 777 {
 778         struct btrfs_fs_info *fs_info;
 779         struct async_submit_bio *async;
 780         int limit;
 781
 782         async = container_of(work, struct  async_submit_bio, work);
 783         fs_info = BTRFS_I(async->inode)->root->fs_info;
 784
 785         limit = btrfs_async_submit_limit(fs_info);
 786         limit = limit * 2 / 3;
 787
 788         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 789             waitqueue_active(&fs_info->async_submit_wait))
 790                 wake_up(&fs_info->async_submit_wait);
 791
 792         /* If an error occured we just want to clean up the bio and move on */
 793         if (async->error) {
 794                 bio_endio(async->bio, async->error);
 795                 return;
 796         }
 797
 798         async->submit_bio_done(async->inode, async->rw, async->bio,
 799                                async->mirror_num, async->bio_flags,
 800                                async->bio_offset);
 801 }
 802
 803 static void run_one_async_free(struct btrfs_work *work)
 804 {
 805         struct async_submit_bio *async;
 806
 807         async = container_of(work, struct  async_submit_bio, work);
 808         kfree(async);
 809 }
 810
 811 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 812                         int rw, struct bio *bio, int mirror_num,
 813                         unsigned long bio_flags,
 814                         u64 bio_offset,
 815                         extent_submit_bio_hook_t *submit_bio_start,
 816                         extent_submit_bio_hook_t *submit_bio_done)
 817 {
 818         struct async_submit_bio *async;
 819
 820         async = kmalloc(sizeof(*async), GFP_NOFS);
 821         if (!async)
 822                 return -ENOMEM;
 823
 824         async->inode = inode;
 825         async->rw = rw;
 826         async->bio = bio;
 827         async->mirror_num = mirror_num;
 828         async->submit_bio_start = submit_bio_start;
 829         async->submit_bio_done = submit_bio_done;
 830
 831         btrfs_init_work(&async->work, run_one_async_start,
 832                         run_one_async_done, run_one_async_free);
 833
 834         async->bio_flags = bio_flags;
 835         async->bio_offset = bio_offset;
 836
 837         async->error = 0;
 838
 839         atomic_inc(&fs_info->nr_async_submits);
 840
 841         if (rw & REQ_SYNC)
 842                 btrfs_set_work_high_priority(&async->work);
 843
 844         btrfs_queue_work(fs_info->workers, &async->work);
 845
 846         while (atomic_read(&fs_info->async_submit_draining) &&
 847               atomic_read(&fs_info->nr_async_submits)) {
 848                 wait_event(fs_info->async_submit_wait,
 849                            (atomic_read(&fs_info->nr_async_submits) == 0));
 850         }
 851
 852         return 0;
 853 }
 854
 855 static int btree_csum_one_bio(struct bio *bio)
 856 {
 857         struct bio_vec *bvec = bio->bi_io_vec;
 858         int bio_index = 0;
 859         struct btrfs_root *root;
 860         int ret = 0;
 861
 862         WARN_ON(bio->bi_vcnt <= 0);
 863         while (bio_index < bio->bi_vcnt) {
 864                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 865                 ret = csum_dirty_buffer(root, bvec->bv_page);
 866                 if (ret)
 867                         break;
 868                 bio_index++;
 869                 bvec++;
 870         }
 871         return ret;
 872 }
 873
 874 static int __btree_submit_bio_start(struct inode *inode, int rw,
 875                                     struct bio *bio, int mirror_num,
 876                                     unsigned long bio_flags,
 877                                     u64 bio_offset)
 878 {
 879         /*
 880          * when we're called for a write, we're already in the async
 881          * submission context.  Just jump into btrfs_map_bio
 882          */
 883         return btree_csum_one_bio(bio);
 884 }
 885
 886 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 887                                  int mirror_num, unsigned long bio_flags,
 888                                  u64 bio_offset)
 889 {
 890         int ret;
 891
 892         /*
 893          * when we're called for a write, we're already in the async
 894          * submission context.  Just jump into btrfs_map_bio
 895          */
 896         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 897         if (ret)
 898                 bio_endio(bio, ret);
 899         return ret;
 900 }
 901
 902 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 903 {
 904         if (bio_flags & EXTENT_BIO_TREE_LOG)
 905                 return 0;
 906 #ifdef CONFIG_X86
 907         if (cpu_has_xmm4_2)
 908                 return 0;
 909 #endif
 910         return 1;
 911 }
 912
 913 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 914                                  int mirror_num, unsigned long bio_flags,
 915                                  u64 bio_offset)
 916 {
 917         int async = check_async_write(inode, bio_flags);
 918         int ret;
 919
 920         if (!(rw & REQ_WRITE)) {
 921                 /*
 922                  * called for a read, do the setup so that checksum validation
 923                  * can happen in the async kernel threads
 924                  */
 925                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 926                                           bio, 1);
 927                 if (ret)
 928                         goto out_w_error;
 929                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 930                                     mirror_num, 0);
 931         } else if (!async) {
 932                 ret = btree_csum_one_bio(bio);
 933                 if (ret)
 934                         goto out_w_error;
 935                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 936                                     mirror_num, 0);
 937         } else {
 938                 /*
 939                  * kthread helpers are used to submit writes so that
 940                  * checksumming can happen in parallel across all CPUs
 941                  */
 942                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 943                                           inode, rw, bio, mirror_num, 0,
 944                                           bio_offset,
 945                                           __btree_submit_bio_start,
 946                                           __btree_submit_bio_done);
 947         }
 948
 949         if (ret) {
 950 out_w_error:
 951                 bio_endio(bio, ret);
 952         }
 953         return ret;
 954 }
 955
 956 #ifdef CONFIG_MIGRATION
 957 static int btree_migratepage(struct address_space *mapping,
 958                         struct page *newpage, struct page *page,
 959                         enum migrate_mode mode)
 960 {
 961         /*
 962          * we can't safely write a btree page from here,
 963          * we haven't done the locking hook
 964          */
 965         if (PageDirty(page))
 966                 return -EAGAIN;
 967         /*
 968          * Buffers may be managed in a filesystem specific way.
 969          * We must have no buffers or drop them.
 970          */
 971         if (page_has_private(page) &&
 972             !try_to_release_page(page, GFP_KERNEL))
 973                 return -EAGAIN;
 974         return migrate_page(mapping, newpage, page, mode);
 975 }
 976 #endif
 977
 978
 979 static int btree_writepages(struct address_space *mapping,
 980                             struct writeback_control *wbc)
 981 {
 982         struct btrfs_fs_info *fs_info;
 983         int ret;
 984
 985         if (wbc->sync_mode == WB_SYNC_NONE) {
 986
 987                 if (wbc->for_kupdate)
 988                         return 0;
 989
 990                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 991                 /* this is a bit racy, but that's ok */
 992                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 993                                              BTRFS_DIRTY_METADATA_THRESH);
 994                 if (ret < 0)
 995                         return 0;
 996         }
 997         return btree_write_cache_pages(mapping, wbc);
 998 }
 999
1000 static int btree_readpage(struct file *file, struct page *page)
1001 {
1002         struct extent_io_tree *tree;
1003         tree = &BTRFS_I(page->mapping->host)->io_tree;
1004         return extent_read_full_page(tree, page, btree_get_extent, 0);
1005 }
1006
1007 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1008 {
1009         if (PageWriteback(page) || PageDirty(page))
1010                 return 0;
1011
1012         return try_release_extent_buffer(page);
1013 }
1014
1015 static void btree_invalidatepage(struct page *page, unsigned int offset,
1016                                  unsigned int length)
1017 {
1018         struct extent_io_tree *tree;
1019         tree = &BTRFS_I(page->mapping->host)->io_tree;
1020         extent_invalidatepage(tree, page, offset);
1021         btree_releasepage(page, GFP_NOFS);
1022         if (PagePrivate(page)) {
1023                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1024                            "page private not zero on page %llu",
1025                            (unsigned long long)page_offset(page));
1026                 ClearPagePrivate(page);
1027                 set_page_private(page, 0);
1028                 page_cache_release(page);
1029         }
1030 }
1031
1032 static int btree_set_page_dirty(struct page *page)
1033 {
1034 #ifdef DEBUG
1035         struct extent_buffer *eb;
1036
1037         BUG_ON(!PagePrivate(page));
1038         eb = (struct extent_buffer *)page->private;
1039         BUG_ON(!eb);
1040         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1041         BUG_ON(!atomic_read(&eb->refs));
1042         btrfs_assert_tree_locked(eb);
1043 #endif
1044         return __set_page_dirty_nobuffers(page);
1045 }
1046
1047 static const struct address_space_operations btree_aops = {
1048         .readpage       = btree_readpage,
1049         .writepages     = btree_writepages,
1050         .releasepage    = btree_releasepage,
1051         .invalidatepage = btree_invalidatepage,
1052 #ifdef CONFIG_MIGRATION
1053         .migratepage    = btree_migratepage,
1054 #endif
1055         .set_page_dirty = btree_set_page_dirty,
1056 };
1057
1058 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1059                          u64 parent_transid)
1060 {
1061         struct extent_buffer *buf = NULL;
1062         struct inode *btree_inode = root->fs_info->btree_inode;
1063         int ret = 0;
1064
1065         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1066         if (!buf)
1067                 return 0;
1068         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1069                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1070         free_extent_buffer(buf);
1071         return ret;
1072 }
1073
1074 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1075                          int mirror_num, struct extent_buffer **eb)
1076 {
1077         struct extent_buffer *buf = NULL;
1078         struct inode *btree_inode = root->fs_info->btree_inode;
1079         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1080         int ret;
1081
1082         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1083         if (!buf)
1084                 return 0;
1085
1086         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1087
1088         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1089                                        btree_get_extent, mirror_num);
1090         if (ret) {
1091                 free_extent_buffer(buf);
1092                 return ret;
1093         }
1094
1095         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1096                 free_extent_buffer(buf);
1097                 return -EIO;
1098         } else if (extent_buffer_uptodate(buf)) {
1099                 *eb = buf;
1100         } else {
1101                 free_extent_buffer(buf);
1102         }
1103         return 0;
1104 }
1105
1106 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1107                                             u64 bytenr, u32 blocksize)
1108 {
1109         return find_extent_buffer(root->fs_info, bytenr);
1110 }
1111
1112 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1113                                                  u64 bytenr, u32 blocksize)
1114 {
1115         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1116 }
1117
1118
1119 int btrfs_write_tree_block(struct extent_buffer *buf)
1120 {
1121         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1122                                         buf->start + buf->len - 1);
1123 }
1124
1125 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1126 {
1127         return filemap_fdatawait_range(buf->pages[0]->mapping,
1128                                        buf->start, buf->start + buf->len - 1);
1129 }
1130
1131 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1132                                       u32 blocksize, u64 parent_transid)
1133 {
1134         struct extent_buffer *buf = NULL;
1135         int ret;
1136
1137         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1138         if (!buf)
1139                 return NULL;
1140
1141         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1142         if (ret) {
1143                 free_extent_buffer(buf);
1144                 return NULL;
1145         }
1146         return buf;
1147
1148 }
1149
1150 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1151                       struct extent_buffer *buf)
1152 {
1153         struct btrfs_fs_info *fs_info = root->fs_info;
1154
1155         if (btrfs_header_generation(buf) ==
1156             fs_info->running_transaction->transid) {
1157                 btrfs_assert_tree_locked(buf);
1158
1159                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1160                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1161                                              -buf->len,
1162                                              fs_info->dirty_metadata_batch);
1163                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1164                         btrfs_set_lock_blocking(buf);
1165                         clear_extent_buffer_dirty(buf);
1166                 }
1167         }
1168 }
1169
1170 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1171 {
1172         struct btrfs_subvolume_writers *writers;
1173         int ret;
1174
1175         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1176         if (!writers)
1177                 return ERR_PTR(-ENOMEM);
1178
1179         ret = percpu_counter_init(&writers->counter, 0);
1180         if (ret < 0) {
1181                 kfree(writers);
1182                 return ERR_PTR(ret);
1183         }
1184
1185         init_waitqueue_head(&writers->wait);
1186         return writers;
1187 }
1188
1189 static void
1190 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1191 {
1192         percpu_counter_destroy(&writers->counter);
1193         kfree(writers);
1194 }
1195
1196 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1197                          u32 stripesize, struct btrfs_root *root,
1198                          struct btrfs_fs_info *fs_info,
1199                          u64 objectid)
1200 {
1201         root->node = NULL;
1202         root->commit_root = NULL;
1203         root->sectorsize = sectorsize;
1204         root->nodesize = nodesize;
1205         root->leafsize = leafsize;
1206         root->stripesize = stripesize;
1207         root->ref_cows = 0;
1208         root->track_dirty = 0;
1209         root->in_radix = 0;
1210         root->orphan_item_inserted = 0;
1211         root->orphan_cleanup_state = 0;
1212
1213         root->objectid = objectid;
1214         root->last_trans = 0;
1215         root->highest_objectid = 0;
1216         root->nr_delalloc_inodes = 0;
1217         root->nr_ordered_extents = 0;
1218         root->name = NULL;
1219         root->inode_tree = RB_ROOT;
1220         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1221         root->block_rsv = NULL;
1222         root->orphan_block_rsv = NULL;
1223
1224         INIT_LIST_HEAD(&root->dirty_list);
1225         INIT_LIST_HEAD(&root->root_list);
1226         INIT_LIST_HEAD(&root->delalloc_inodes);
1227         INIT_LIST_HEAD(&root->delalloc_root);
1228         INIT_LIST_HEAD(&root->ordered_extents);
1229         INIT_LIST_HEAD(&root->ordered_root);
1230         INIT_LIST_HEAD(&root->logged_list[0]);
1231         INIT_LIST_HEAD(&root->logged_list[1]);
1232         spin_lock_init(&root->orphan_lock);
1233         spin_lock_init(&root->inode_lock);
1234         spin_lock_init(&root->delalloc_lock);
1235         spin_lock_init(&root->ordered_extent_lock);
1236         spin_lock_init(&root->accounting_lock);
1237         spin_lock_init(&root->log_extents_lock[0]);
1238         spin_lock_init(&root->log_extents_lock[1]);
1239         mutex_init(&root->objectid_mutex);
1240         mutex_init(&root->log_mutex);
1241         mutex_init(&root->ordered_extent_mutex);
1242         mutex_init(&root->delalloc_mutex);
1243         init_waitqueue_head(&root->log_writer_wait);
1244         init_waitqueue_head(&root->log_commit_wait[0]);
1245         init_waitqueue_head(&root->log_commit_wait[1]);
1246         INIT_LIST_HEAD(&root->log_ctxs[0]);
1247         INIT_LIST_HEAD(&root->log_ctxs[1]);
1248         atomic_set(&root->log_commit[0], 0);
1249         atomic_set(&root->log_commit[1], 0);
1250         atomic_set(&root->log_writers, 0);
1251         atomic_set(&root->log_batch, 0);
1252         atomic_set(&root->orphan_inodes, 0);
1253         atomic_set(&root->refs, 1);
1254         atomic_set(&root->will_be_snapshoted, 0);
1255         root->log_transid = 0;
1256         root->log_transid_committed = -1;
1257         root->last_log_commit = 0;
1258         if (fs_info)
1259                 extent_io_tree_init(&root->dirty_log_pages,
1260                                      fs_info->btree_inode->i_mapping);
1261
1262         memset(&root->root_key, 0, sizeof(root->root_key));
1263         memset(&root->root_item, 0, sizeof(root->root_item));
1264         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1265         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1266         if (fs_info)
1267                 root->defrag_trans_start = fs_info->generation;
1268         else
1269                 root->defrag_trans_start = 0;
1270         init_completion(&root->kobj_unregister);
1271         root->defrag_running = 0;
1272         root->root_key.objectid = objectid;
1273         root->anon_dev = 0;
1274
1275         spin_lock_init(&root->root_item_lock);
1276 }
1277
1278 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1279 {
1280         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1281         if (root)
1282                 root->fs_info = fs_info;
1283         return root;
1284 }
1285
1286 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1287 /* Should only be used by the testing infrastructure */
1288 struct btrfs_root *btrfs_alloc_dummy_root(void)
1289 {
1290         struct btrfs_root *root;
1291
1292         root = btrfs_alloc_root(NULL);
1293         if (!root)
1294                 return ERR_PTR(-ENOMEM);
1295         __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1296         root->dummy_root = 1;
1297
1298         return root;
1299 }
1300 #endif
1301
1302 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1303                                      struct btrfs_fs_info *fs_info,
1304                                      u64 objectid)
1305 {
1306         struct extent_buffer *leaf;
1307         struct btrfs_root *tree_root = fs_info->tree_root;
1308         struct btrfs_root *root;
1309         struct btrfs_key key;
1310         int ret = 0;
1311         uuid_le uuid;
1312
1313         root = btrfs_alloc_root(fs_info);
1314         if (!root)
1315                 return ERR_PTR(-ENOMEM);
1316
1317         __setup_root(tree_root->nodesize, tree_root->leafsize,
1318                      tree_root->sectorsize, tree_root->stripesize,
1319                      root, fs_info, objectid);
1320         root->root_key.objectid = objectid;
1321         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1322         root->root_key.offset = 0;
1323
1324         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1325                                       0, objectid, NULL, 0, 0, 0);
1326         if (IS_ERR(leaf)) {
1327                 ret = PTR_ERR(leaf);
1328                 leaf = NULL;
1329                 goto fail;
1330         }
1331
1332         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1333         btrfs_set_header_bytenr(leaf, leaf->start);
1334         btrfs_set_header_generation(leaf, trans->transid);
1335         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1336         btrfs_set_header_owner(leaf, objectid);
1337         root->node = leaf;
1338
1339         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1340                             BTRFS_FSID_SIZE);
1341         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1342                             btrfs_header_chunk_tree_uuid(leaf),
1343                             BTRFS_UUID_SIZE);
1344         btrfs_mark_buffer_dirty(leaf);
1345
1346         root->commit_root = btrfs_root_node(root);
1347         root->track_dirty = 1;
1348
1349
1350         root->root_item.flags = 0;
1351         root->root_item.byte_limit = 0;
1352         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1353         btrfs_set_root_generation(&root->root_item, trans->transid);
1354         btrfs_set_root_level(&root->root_item, 0);
1355         btrfs_set_root_refs(&root->root_item, 1);
1356         btrfs_set_root_used(&root->root_item, leaf->len);
1357         btrfs_set_root_last_snapshot(&root->root_item, 0);
1358         btrfs_set_root_dirid(&root->root_item, 0);
1359         uuid_le_gen(&uuid);
1360         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1361         root->root_item.drop_level = 0;
1362
1363         key.objectid = objectid;
1364         key.type = BTRFS_ROOT_ITEM_KEY;
1365         key.offset = 0;
1366         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1367         if (ret)
1368                 goto fail;
1369
1370         btrfs_tree_unlock(leaf);
1371
1372         return root;
1373
1374 fail:
1375         if (leaf) {
1376                 btrfs_tree_unlock(leaf);
1377                 free_extent_buffer(leaf);
1378         }
1379         kfree(root);
1380
1381         return ERR_PTR(ret);
1382 }
1383
1384 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1385                                          struct btrfs_fs_info *fs_info)
1386 {
1387         struct btrfs_root *root;
1388         struct btrfs_root *tree_root = fs_info->tree_root;
1389         struct extent_buffer *leaf;
1390
1391         root = btrfs_alloc_root(fs_info);
1392         if (!root)
1393                 return ERR_PTR(-ENOMEM);
1394
1395         __setup_root(tree_root->nodesize, tree_root->leafsize,
1396                      tree_root->sectorsize, tree_root->stripesize,
1397                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1398
1399         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1400         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1401         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1402         /*
1403          * log trees do not get reference counted because they go away
1404          * before a real commit is actually done.  They do store pointers
1405          * to file data extents, and those reference counts still get
1406          * updated (along with back refs to the log tree).
1407          */
1408         root->ref_cows = 0;
1409
1410         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1411                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1412                                       0, 0, 0);
1413         if (IS_ERR(leaf)) {
1414                 kfree(root);
1415                 return ERR_CAST(leaf);
1416         }
1417
1418         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1419         btrfs_set_header_bytenr(leaf, leaf->start);
1420         btrfs_set_header_generation(leaf, trans->transid);
1421         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1422         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1423         root->node = leaf;
1424
1425         write_extent_buffer(root->node, root->fs_info->fsid,
1426                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1427         btrfs_mark_buffer_dirty(root->node);
1428         btrfs_tree_unlock(root->node);
1429         return root;
1430 }
1431
1432 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1433                              struct btrfs_fs_info *fs_info)
1434 {
1435         struct btrfs_root *log_root;
1436
1437         log_root = alloc_log_tree(trans, fs_info);
1438         if (IS_ERR(log_root))
1439                 return PTR_ERR(log_root);
1440         WARN_ON(fs_info->log_root_tree);
1441         fs_info->log_root_tree = log_root;
1442         return 0;
1443 }
1444
1445 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1446                        struct btrfs_root *root)
1447 {
1448         struct btrfs_root *log_root;
1449         struct btrfs_inode_item *inode_item;
1450
1451         log_root = alloc_log_tree(trans, root->fs_info);
1452         if (IS_ERR(log_root))
1453                 return PTR_ERR(log_root);
1454
1455         log_root->last_trans = trans->transid;
1456         log_root->root_key.offset = root->root_key.objectid;
1457
1458         inode_item = &log_root->root_item.inode;
1459         btrfs_set_stack_inode_generation(inode_item, 1);
1460         btrfs_set_stack_inode_size(inode_item, 3);
1461         btrfs_set_stack_inode_nlink(inode_item, 1);
1462         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1463         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1464
1465         btrfs_set_root_node(&log_root->root_item, log_root->node);
1466
1467         WARN_ON(root->log_root);
1468         root->log_root = log_root;
1469         root->log_transid = 0;
1470         root->log_transid_committed = -1;
1471         root->last_log_commit = 0;
1472         return 0;
1473 }
1474
1475 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1476                                                struct btrfs_key *key)
1477 {
1478         struct btrfs_root *root;
1479         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1480         struct btrfs_path *path;
1481         u64 generation;
1482         u32 blocksize;
1483         int ret;
1484
1485         path = btrfs_alloc_path();
1486         if (!path)
1487                 return ERR_PTR(-ENOMEM);
1488
1489         root = btrfs_alloc_root(fs_info);
1490         if (!root) {
1491                 ret = -ENOMEM;
1492                 goto alloc_fail;
1493         }
1494
1495         __setup_root(tree_root->nodesize, tree_root->leafsize,
1496                      tree_root->sectorsize, tree_root->stripesize,
1497                      root, fs_info, key->objectid);
1498
1499         ret = btrfs_find_root(tree_root, key, path,
1500                               &root->root_item, &root->root_key);
1501         if (ret) {
1502                 if (ret > 0)
1503                         ret = -ENOENT;
1504                 goto find_fail;
1505         }
1506
1507         generation = btrfs_root_generation(&root->root_item);
1508         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1509         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1510                                      blocksize, generation);
1511         if (!root->node) {
1512                 ret = -ENOMEM;
1513                 goto find_fail;
1514         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1515                 ret = -EIO;
1516                 goto read_fail;
1517         }
1518         root->commit_root = btrfs_root_node(root);
1519 out:
1520         btrfs_free_path(path);
1521         return root;
1522
1523 read_fail:
1524         free_extent_buffer(root->node);
1525 find_fail:
1526         kfree(root);
1527 alloc_fail:
1528         root = ERR_PTR(ret);
1529         goto out;
1530 }
1531
1532 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1533                                       struct btrfs_key *location)
1534 {
1535         struct btrfs_root *root;
1536
1537         root = btrfs_read_tree_root(tree_root, location);
1538         if (IS_ERR(root))
1539                 return root;
1540
1541         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1542                 root->ref_cows = 1;
1543                 btrfs_check_and_init_root_item(&root->root_item);
1544         }
1545
1546         return root;
1547 }
1548
1549 int btrfs_init_fs_root(struct btrfs_root *root)
1550 {
1551         int ret;
1552         struct btrfs_subvolume_writers *writers;
1553
1554         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1555         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1556                                         GFP_NOFS);
1557         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1558                 ret = -ENOMEM;
1559                 goto fail;
1560         }
1561
1562         writers = btrfs_alloc_subvolume_writers();
1563         if (IS_ERR(writers)) {
1564                 ret = PTR_ERR(writers);
1565                 goto fail;
1566         }
1567         root->subv_writers = writers;
1568
1569         btrfs_init_free_ino_ctl(root);
1570         mutex_init(&root->fs_commit_mutex);
1571         spin_lock_init(&root->cache_lock);
1572         init_waitqueue_head(&root->cache_wait);
1573
1574         ret = get_anon_bdev(&root->anon_dev);
1575         if (ret)
1576                 goto free_writers;
1577         return 0;
1578
1579 free_writers:
1580         btrfs_free_subvolume_writers(root->subv_writers);
1581 fail:
1582         kfree(root->free_ino_ctl);
1583         kfree(root->free_ino_pinned);
1584         return ret;
1585 }
1586
1587 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1588                                                u64 root_id)
1589 {
1590         struct btrfs_root *root;
1591
1592         spin_lock(&fs_info->fs_roots_radix_lock);
1593         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1594                                  (unsigned long)root_id);
1595         spin_unlock(&fs_info->fs_roots_radix_lock);
1596         return root;
1597 }
1598
1599 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1600                          struct btrfs_root *root)
1601 {
1602         int ret;
1603
1604         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1605         if (ret)
1606                 return ret;
1607
1608         spin_lock(&fs_info->fs_roots_radix_lock);
1609         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1610                                 (unsigned long)root->root_key.objectid,
1611                                 root);
1612         if (ret == 0)
1613                 root->in_radix = 1;
1614         spin_unlock(&fs_info->fs_roots_radix_lock);
1615         radix_tree_preload_end();
1616
1617         return ret;
1618 }
1619
1620 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1621                                      struct btrfs_key *location,
1622                                      bool check_ref)
1623 {
1624         struct btrfs_root *root;
1625         int ret;
1626
1627         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1628                 return fs_info->tree_root;
1629         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1630                 return fs_info->extent_root;
1631         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1632                 return fs_info->chunk_root;
1633         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1634                 return fs_info->dev_root;
1635         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1636                 return fs_info->csum_root;
1637         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1638                 return fs_info->quota_root ? fs_info->quota_root :
1639                                              ERR_PTR(-ENOENT);
1640         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1641                 return fs_info->uuid_root ? fs_info->uuid_root :
1642                                             ERR_PTR(-ENOENT);
1643 again:
1644         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1645         if (root) {
1646                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1647                         return ERR_PTR(-ENOENT);
1648                 return root;
1649         }
1650
1651         root = btrfs_read_fs_root(fs_info->tree_root, location);
1652         if (IS_ERR(root))
1653                 return root;
1654
1655         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1656                 ret = -ENOENT;
1657                 goto fail;
1658         }
1659
1660         ret = btrfs_init_fs_root(root);
1661         if (ret)
1662                 goto fail;
1663
1664         ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1665                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1666         if (ret < 0)
1667                 goto fail;
1668         if (ret == 0)
1669                 root->orphan_item_inserted = 1;
1670
1671         ret = btrfs_insert_fs_root(fs_info, root);
1672         if (ret) {
1673                 if (ret == -EEXIST) {
1674                         free_fs_root(root);
1675                         goto again;
1676                 }
1677                 goto fail;
1678         }
1679         return root;
1680 fail:
1681         free_fs_root(root);
1682         return ERR_PTR(ret);
1683 }
1684
1685 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1686 {
1687         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1688         int ret = 0;
1689         struct btrfs_device *device;
1690         struct backing_dev_info *bdi;
1691
1692         rcu_read_lock();
1693         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1694                 if (!device->bdev)
1695                         continue;
1696                 bdi = blk_get_backing_dev_info(device->bdev);
1697                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1698                         ret = 1;
1699                         break;
1700                 }
1701         }
1702         rcu_read_unlock();
1703         return ret;
1704 }
1705
1706 /*
1707  * If this fails, caller must call bdi_destroy() to get rid of the
1708  * bdi again.
1709  */
1710 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1711 {
1712         int err;
1713
1714         bdi->capabilities = BDI_CAP_MAP_COPY;
1715         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1716         if (err)
1717                 return err;
1718
1719         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1720         bdi->congested_fn       = btrfs_congested_fn;
1721         bdi->congested_data     = info;
1722         return 0;
1723 }
1724
1725 /*
1726  * called by the kthread helper functions to finally call the bio end_io
1727  * functions.  This is where read checksum verification actually happens
1728  */
1729 static void end_workqueue_fn(struct btrfs_work *work)
1730 {
1731         struct bio *bio;
1732         struct end_io_wq *end_io_wq;
1733         int error;
1734
1735         end_io_wq = container_of(work, struct end_io_wq, work);
1736         bio = end_io_wq->bio;
1737
1738         error = end_io_wq->error;
1739         bio->bi_private = end_io_wq->private;
1740         bio->bi_end_io = end_io_wq->end_io;
1741         kfree(end_io_wq);
1742         bio_endio(bio, error);
1743 }
1744
1745 static int cleaner_kthread(void *arg)
1746 {
1747         struct btrfs_root *root = arg;
1748         int again;
1749
1750         do {
1751                 again = 0;
1752
1753                 /* Make the cleaner go to sleep early. */
1754                 if (btrfs_need_cleaner_sleep(root))
1755                         goto sleep;
1756
1757                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1758                         goto sleep;
1759
1760                 /*
1761                  * Avoid the problem that we change the status of the fs
1762                  * during the above check and trylock.
1763                  */
1764                 if (btrfs_need_cleaner_sleep(root)) {
1765                         mutex_unlock(&root->fs_info->cleaner_mutex);
1766                         goto sleep;
1767                 }
1768
1769                 btrfs_run_delayed_iputs(root);
1770                 again = btrfs_clean_one_deleted_snapshot(root);
1771                 mutex_unlock(&root->fs_info->cleaner_mutex);
1772
1773                 /*
1774                  * The defragger has dealt with the R/O remount and umount,
1775                  * needn't do anything special here.
1776                  */
1777                 btrfs_run_defrag_inodes(root->fs_info);
1778 sleep:
1779                 if (!try_to_freeze() && !again) {
1780                         set_current_state(TASK_INTERRUPTIBLE);
1781                         if (!kthread_should_stop())
1782                                 schedule();
1783                         __set_current_state(TASK_RUNNING);
1784                 }
1785         } while (!kthread_should_stop());
1786         return 0;
1787 }
1788
1789 static int transaction_kthread(void *arg)
1790 {
1791         struct btrfs_root *root = arg;
1792         struct btrfs_trans_handle *trans;
1793         struct btrfs_transaction *cur;
1794         u64 transid;
1795         unsigned long now;
1796         unsigned long delay;
1797         bool cannot_commit;
1798
1799         do {
1800                 cannot_commit = false;
1801                 delay = HZ * root->fs_info->commit_interval;
1802                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1803
1804                 spin_lock(&root->fs_info->trans_lock);
1805                 cur = root->fs_info->running_transaction;
1806                 if (!cur) {
1807                         spin_unlock(&root->fs_info->trans_lock);
1808                         goto sleep;
1809                 }
1810
1811                 now = get_seconds();
1812                 if (cur->state < TRANS_STATE_BLOCKED &&
1813                     (now < cur->start_time ||
1814                      now - cur->start_time < root->fs_info->commit_interval)) {
1815                         spin_unlock(&root->fs_info->trans_lock);
1816                         delay = HZ * 5;
1817                         goto sleep;
1818                 }
1819                 transid = cur->transid;
1820                 spin_unlock(&root->fs_info->trans_lock);
1821
1822                 /* If the file system is aborted, this will always fail. */
1823                 trans = btrfs_attach_transaction(root);
1824                 if (IS_ERR(trans)) {
1825                         if (PTR_ERR(trans) != -ENOENT)
1826                                 cannot_commit = true;
1827                         goto sleep;
1828                 }
1829                 if (transid == trans->transid) {
1830                         btrfs_commit_transaction(trans, root);
1831                 } else {
1832                         btrfs_end_transaction(trans, root);
1833                 }
1834 sleep:
1835                 wake_up_process(root->fs_info->cleaner_kthread);
1836                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1837
1838                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1839                                       &root->fs_info->fs_state)))
1840                         btrfs_cleanup_transaction(root);
1841                 if (!try_to_freeze()) {
1842                         set_current_state(TASK_INTERRUPTIBLE);
1843                         if (!kthread_should_stop() &&
1844                             (!btrfs_transaction_blocked(root->fs_info) ||
1845                              cannot_commit))
1846                                 schedule_timeout(delay);
1847                         __set_current_state(TASK_RUNNING);
1848                 }
1849         } while (!kthread_should_stop());
1850         return 0;
1851 }
1852
1853 /*
1854  * this will find the highest generation in the array of
1855  * root backups.  The index of the highest array is returned,
1856  * or -1 if we can't find anything.
1857  *
1858  * We check to make sure the array is valid by comparing the
1859  * generation of the latest  root in the array with the generation
1860  * in the super block.  If they don't match we pitch it.
1861  */
1862 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1863 {
1864         u64 cur;
1865         int newest_index = -1;
1866         struct btrfs_root_backup *root_backup;
1867         int i;
1868
1869         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1870                 root_backup = info->super_copy->super_roots + i;
1871                 cur = btrfs_backup_tree_root_gen(root_backup);
1872                 if (cur == newest_gen)
1873                         newest_index = i;
1874         }
1875
1876         /* check to see if we actually wrapped around */
1877         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1878                 root_backup = info->super_copy->super_roots;
1879                 cur = btrfs_backup_tree_root_gen(root_backup);
1880                 if (cur == newest_gen)
1881                         newest_index = 0;
1882         }
1883         return newest_index;
1884 }
1885
1886
1887 /*
1888  * find the oldest backup so we know where to store new entries
1889  * in the backup array.  This will set the backup_root_index
1890  * field in the fs_info struct
1891  */
1892 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1893                                      u64 newest_gen)
1894 {
1895         int newest_index = -1;
1896
1897         newest_index = find_newest_super_backup(info, newest_gen);
1898         /* if there was garbage in there, just move along */
1899         if (newest_index == -1) {
1900                 info->backup_root_index = 0;
1901         } else {
1902                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1903         }
1904 }
1905
1906 /*
1907  * copy all the root pointers into the super backup array.
1908  * this will bump the backup pointer by one when it is
1909  * done
1910  */
1911 static void backup_super_roots(struct btrfs_fs_info *info)
1912 {
1913         int next_backup;
1914         struct btrfs_root_backup *root_backup;
1915         int last_backup;
1916
1917         next_backup = info->backup_root_index;
1918         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1919                 BTRFS_NUM_BACKUP_ROOTS;
1920
1921         /*
1922          * just overwrite the last backup if we're at the same generation
1923          * this happens only at umount
1924          */
1925         root_backup = info->super_for_commit->super_roots + last_backup;
1926         if (btrfs_backup_tree_root_gen(root_backup) ==
1927             btrfs_header_generation(info->tree_root->node))
1928                 next_backup = last_backup;
1929
1930         root_backup = info->super_for_commit->super_roots + next_backup;
1931
1932         /*
1933          * make sure all of our padding and empty slots get zero filled
1934          * regardless of which ones we use today
1935          */
1936         memset(root_backup, 0, sizeof(*root_backup));
1937
1938         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1939
1940         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1941         btrfs_set_backup_tree_root_gen(root_backup,
1942                                btrfs_header_generation(info->tree_root->node));
1943
1944         btrfs_set_backup_tree_root_level(root_backup,
1945                                btrfs_header_level(info->tree_root->node));
1946
1947         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1948         btrfs_set_backup_chunk_root_gen(root_backup,
1949                                btrfs_header_generation(info->chunk_root->node));
1950         btrfs_set_backup_chunk_root_level(root_backup,
1951                                btrfs_header_level(info->chunk_root->node));
1952
1953         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1954         btrfs_set_backup_extent_root_gen(root_backup,
1955                                btrfs_header_generation(info->extent_root->node));
1956         btrfs_set_backup_extent_root_level(root_backup,
1957                                btrfs_header_level(info->extent_root->node));
1958
1959         /*
1960          * we might commit during log recovery, which happens before we set
1961          * the fs_root.  Make sure it is valid before we fill it in.
1962          */
1963         if (info->fs_root && info->fs_root->node) {
1964                 btrfs_set_backup_fs_root(root_backup,
1965                                          info->fs_root->node->start);
1966                 btrfs_set_backup_fs_root_gen(root_backup,
1967                                btrfs_header_generation(info->fs_root->node));
1968                 btrfs_set_backup_fs_root_level(root_backup,
1969                                btrfs_header_level(info->fs_root->node));
1970         }
1971
1972         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1973         btrfs_set_backup_dev_root_gen(root_backup,
1974                                btrfs_header_generation(info->dev_root->node));
1975         btrfs_set_backup_dev_root_level(root_backup,
1976                                        btrfs_header_level(info->dev_root->node));
1977
1978         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1979         btrfs_set_backup_csum_root_gen(root_backup,
1980                                btrfs_header_generation(info->csum_root->node));
1981         btrfs_set_backup_csum_root_level(root_backup,
1982                                btrfs_header_level(info->csum_root->node));
1983
1984         btrfs_set_backup_total_bytes(root_backup,
1985                              btrfs_super_total_bytes(info->super_copy));
1986         btrfs_set_backup_bytes_used(root_backup,
1987                              btrfs_super_bytes_used(info->super_copy));
1988         btrfs_set_backup_num_devices(root_backup,
1989                              btrfs_super_num_devices(info->super_copy));
1990
1991         /*
1992          * if we don't copy this out to the super_copy, it won't get remembered
1993          * for the next commit
1994          */
1995         memcpy(&info->super_copy->super_roots,
1996                &info->super_for_commit->super_roots,
1997                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1998 }
1999
2000 /*
2001  * this copies info out of the root backup array and back into
2002  * the in-memory super block.  It is meant to help iterate through
2003  * the array, so you send it the number of backups you've already
2004  * tried and the last backup index you used.
2005  *
2006  * this returns -1 when it has tried all the backups
2007  */
2008 static noinline int next_root_backup(struct btrfs_fs_info *info,
2009                                      struct btrfs_super_block *super,
2010                                      int *num_backups_tried, int *backup_index)
2011 {
2012         struct btrfs_root_backup *root_backup;
2013         int newest = *backup_index;
2014
2015         if (*num_backups_tried == 0) {
2016                 u64 gen = btrfs_super_generation(super);
2017
2018                 newest = find_newest_super_backup(info, gen);
2019                 if (newest == -1)
2020                         return -1;
2021
2022                 *backup_index = newest;
2023                 *num_backups_tried = 1;
2024         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2025                 /* we've tried all the backups, all done */
2026                 return -1;
2027         } else {
2028                 /* jump to the next oldest backup */
2029                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2030                         BTRFS_NUM_BACKUP_ROOTS;
2031                 *backup_index = newest;
2032                 *num_backups_tried += 1;
2033         }
2034         root_backup = super->super_roots + newest;
2035
2036         btrfs_set_super_generation(super,
2037                                    btrfs_backup_tree_root_gen(root_backup));
2038         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2039         btrfs_set_super_root_level(super,
2040                                    btrfs_backup_tree_root_level(root_backup));
2041         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2042
2043         /*
2044          * fixme: the total bytes and num_devices need to match or we should
2045          * need a fsck
2046          */
2047         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2048         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2049         return 0;
2050 }
2051
2052 /* helper to cleanup workers */
2053 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2054 {
2055         btrfs_destroy_workqueue(fs_info->fixup_workers);
2056         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2057         btrfs_destroy_workqueue(fs_info->workers);
2058         btrfs_destroy_workqueue(fs_info->endio_workers);
2059         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2060         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2061         btrfs_destroy_workqueue(fs_info->rmw_workers);
2062         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2063         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2064         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2065         btrfs_destroy_workqueue(fs_info->submit_workers);
2066         btrfs_destroy_workqueue(fs_info->delayed_workers);
2067         btrfs_destroy_workqueue(fs_info->caching_workers);
2068         btrfs_destroy_workqueue(fs_info->readahead_workers);
2069         btrfs_destroy_workqueue(fs_info->flush_workers);
2070         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2071 }
2072
2073 static void free_root_extent_buffers(struct btrfs_root *root)
2074 {
2075         if (root) {
2076                 free_extent_buffer(root->node);
2077                 free_extent_buffer(root->commit_root);
2078                 root->node = NULL;
2079                 root->commit_root = NULL;
2080         }
2081 }
2082
2083 /* helper to cleanup tree roots */
2084 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2085 {
2086         free_root_extent_buffers(info->tree_root);
2087
2088         free_root_extent_buffers(info->dev_root);
2089         free_root_extent_buffers(info->extent_root);
2090         free_root_extent_buffers(info->csum_root);
2091         free_root_extent_buffers(info->quota_root);
2092         free_root_extent_buffers(info->uuid_root);
2093         if (chunk_root)
2094                 free_root_extent_buffers(info->chunk_root);
2095 }
2096
2097 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2098 {
2099         int ret;
2100         struct btrfs_root *gang[8];
2101         int i;
2102
2103         while (!list_empty(&fs_info->dead_roots)) {
2104                 gang[0] = list_entry(fs_info->dead_roots.next,
2105                                      struct btrfs_root, root_list);
2106                 list_del(&gang[0]->root_list);
2107
2108                 if (gang[0]->in_radix) {
2109                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2110                 } else {
2111                         free_extent_buffer(gang[0]->node);
2112                         free_extent_buffer(gang[0]->commit_root);
2113                         btrfs_put_fs_root(gang[0]);
2114                 }
2115         }
2116
2117         while (1) {
2118                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2119                                              (void **)gang, 0,
2120                                              ARRAY_SIZE(gang));
2121                 if (!ret)
2122                         break;
2123                 for (i = 0; i < ret; i++)
2124                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2125         }
2126
2127         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2128                 btrfs_free_log_root_tree(NULL, fs_info);
2129                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2130                                             fs_info->pinned_extents);
2131         }
2132 }
2133
2134 int open_ctree(struct super_block *sb,
2135                struct btrfs_fs_devices *fs_devices,
2136                char *options)
2137 {
2138         u32 sectorsize;
2139         u32 nodesize;
2140         u32 leafsize;
2141         u32 blocksize;
2142         u32 stripesize;
2143         u64 generation;
2144         u64 features;
2145         struct btrfs_key location;
2146         struct buffer_head *bh;
2147         struct btrfs_super_block *disk_super;
2148         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2149         struct btrfs_root *tree_root;
2150         struct btrfs_root *extent_root;
2151         struct btrfs_root *csum_root;
2152         struct btrfs_root *chunk_root;
2153         struct btrfs_root *dev_root;
2154         struct btrfs_root *quota_root;
2155         struct btrfs_root *uuid_root;
2156         struct btrfs_root *log_tree_root;
2157         int ret;
2158         int err = -EINVAL;
2159         int num_backups_tried = 0;
2160         int backup_index = 0;
2161         int max_active;
2162         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2163         bool create_uuid_tree;
2164         bool check_uuid_tree;
2165
2166         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2167         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2168         if (!tree_root || !chunk_root) {
2169                 err = -ENOMEM;
2170                 goto fail;
2171         }
2172
2173         ret = init_srcu_struct(&fs_info->subvol_srcu);
2174         if (ret) {
2175                 err = ret;
2176                 goto fail;
2177         }
2178
2179         ret = setup_bdi(fs_info, &fs_info->bdi);
2180         if (ret) {
2181                 err = ret;
2182                 goto fail_srcu;
2183         }
2184
2185         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2186         if (ret) {
2187                 err = ret;
2188                 goto fail_bdi;
2189         }
2190         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2191                                         (1 + ilog2(nr_cpu_ids));
2192
2193         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2194         if (ret) {
2195                 err = ret;
2196                 goto fail_dirty_metadata_bytes;
2197         }
2198
2199         ret = percpu_counter_init(&fs_info->bio_counter, 0);
2200         if (ret) {
2201                 err = ret;
2202                 goto fail_delalloc_bytes;
2203         }
2204
2205         fs_info->btree_inode = new_inode(sb);
2206         if (!fs_info->btree_inode) {
2207                 err = -ENOMEM;
2208                 goto fail_bio_counter;
2209         }
2210
2211         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2212
2213         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2214         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2215         INIT_LIST_HEAD(&fs_info->trans_list);
2216         INIT_LIST_HEAD(&fs_info->dead_roots);
2217         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2218         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2219         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2220         spin_lock_init(&fs_info->delalloc_root_lock);
2221         spin_lock_init(&fs_info->trans_lock);
2222         spin_lock_init(&fs_info->fs_roots_radix_lock);
2223         spin_lock_init(&fs_info->delayed_iput_lock);
2224         spin_lock_init(&fs_info->defrag_inodes_lock);
2225         spin_lock_init(&fs_info->free_chunk_lock);
2226         spin_lock_init(&fs_info->tree_mod_seq_lock);
2227         spin_lock_init(&fs_info->super_lock);
2228         spin_lock_init(&fs_info->buffer_lock);
2229         rwlock_init(&fs_info->tree_mod_log_lock);
2230         mutex_init(&fs_info->reloc_mutex);
2231         mutex_init(&fs_info->delalloc_root_mutex);
2232         seqlock_init(&fs_info->profiles_lock);
2233
2234         init_completion(&fs_info->kobj_unregister);
2235         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2236         INIT_LIST_HEAD(&fs_info->space_info);
2237         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2238         btrfs_mapping_init(&fs_info->mapping_tree);
2239         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2240                              BTRFS_BLOCK_RSV_GLOBAL);
2241         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2242                              BTRFS_BLOCK_RSV_DELALLOC);
2243         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2244         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2245         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2246         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2247                              BTRFS_BLOCK_RSV_DELOPS);
2248         atomic_set(&fs_info->nr_async_submits, 0);
2249         atomic_set(&fs_info->async_delalloc_pages, 0);
2250         atomic_set(&fs_info->async_submit_draining, 0);
2251         atomic_set(&fs_info->nr_async_bios, 0);
2252         atomic_set(&fs_info->defrag_running, 0);
2253         atomic64_set(&fs_info->tree_mod_seq, 0);
2254         fs_info->sb = sb;
2255         fs_info->max_inline = 8192 * 1024;
2256         fs_info->metadata_ratio = 0;
2257         fs_info->defrag_inodes = RB_ROOT;
2258         fs_info->free_chunk_space = 0;
2259         fs_info->tree_mod_log = RB_ROOT;
2260         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2261         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2262         /* readahead state */
2263         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2264         spin_lock_init(&fs_info->reada_lock);
2265
2266         fs_info->thread_pool_size = min_t(unsigned long,
2267                                           num_online_cpus() + 2, 8);
2268
2269         INIT_LIST_HEAD(&fs_info->ordered_roots);
2270         spin_lock_init(&fs_info->ordered_root_lock);
2271         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2272                                         GFP_NOFS);
2273         if (!fs_info->delayed_root) {
2274                 err = -ENOMEM;
2275                 goto fail_iput;
2276         }
2277         btrfs_init_delayed_root(fs_info->delayed_root);
2278
2279         mutex_init(&fs_info->scrub_lock);
2280         atomic_set(&fs_info->scrubs_running, 0);
2281         atomic_set(&fs_info->scrub_pause_req, 0);
2282         atomic_set(&fs_info->scrubs_paused, 0);
2283         atomic_set(&fs_info->scrub_cancel_req, 0);
2284         init_waitqueue_head(&fs_info->replace_wait);
2285         init_waitqueue_head(&fs_info->scrub_pause_wait);
2286         fs_info->scrub_workers_refcnt = 0;
2287 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2288         fs_info->check_integrity_print_mask = 0;
2289 #endif
2290
2291         spin_lock_init(&fs_info->balance_lock);
2292         mutex_init(&fs_info->balance_mutex);
2293         atomic_set(&fs_info->balance_running, 0);
2294         atomic_set(&fs_info->balance_pause_req, 0);
2295         atomic_set(&fs_info->balance_cancel_req, 0);
2296         fs_info->balance_ctl = NULL;
2297         init_waitqueue_head(&fs_info->balance_wait_q);
2298
2299         sb->s_blocksize = 4096;
2300         sb->s_blocksize_bits = blksize_bits(4096);
2301         sb->s_bdi = &fs_info->bdi;
2302
2303         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2304         set_nlink(fs_info->btree_inode, 1);
2305         /*
2306          * we set the i_size on the btree inode to the max possible int.
2307          * the real end of the address space is determined by all of
2308          * the devices in the system
2309          */
2310         fs_info->btree_inode->i_size = OFFSET_MAX;
2311         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2312         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2313
2314         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2315         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2316                              fs_info->btree_inode->i_mapping);
2317         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2318         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2319
2320         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2321
2322         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2323         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2324                sizeof(struct btrfs_key));
2325         set_bit(BTRFS_INODE_DUMMY,
2326                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2327         btrfs_insert_inode_hash(fs_info->btree_inode);
2328
2329         spin_lock_init(&fs_info->block_group_cache_lock);
2330         fs_info->block_group_cache_tree = RB_ROOT;
2331         fs_info->first_logical_byte = (u64)-1;
2332
2333         extent_io_tree_init(&fs_info->freed_extents[0],
2334                              fs_info->btree_inode->i_mapping);
2335         extent_io_tree_init(&fs_info->freed_extents[1],
2336                              fs_info->btree_inode->i_mapping);
2337         fs_info->pinned_extents = &fs_info->freed_extents[0];
2338         fs_info->do_barriers = 1;
2339
2340
2341         mutex_init(&fs_info->ordered_operations_mutex);
2342         mutex_init(&fs_info->ordered_extent_flush_mutex);
2343         mutex_init(&fs_info->tree_log_mutex);
2344         mutex_init(&fs_info->chunk_mutex);
2345         mutex_init(&fs_info->transaction_kthread_mutex);
2346         mutex_init(&fs_info->cleaner_mutex);
2347         mutex_init(&fs_info->volume_mutex);
2348         init_rwsem(&fs_info->extent_commit_sem);
2349         init_rwsem(&fs_info->cleanup_work_sem);
2350         init_rwsem(&fs_info->subvol_sem);
2351         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2352         fs_info->dev_replace.lock_owner = 0;
2353         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2354         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2355         mutex_init(&fs_info->dev_replace.lock_management_lock);
2356         mutex_init(&fs_info->dev_replace.lock);
2357
2358         spin_lock_init(&fs_info->qgroup_lock);
2359         mutex_init(&fs_info->qgroup_ioctl_lock);
2360         fs_info->qgroup_tree = RB_ROOT;
2361         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2362         fs_info->qgroup_seq = 1;
2363         fs_info->quota_enabled = 0;
2364         fs_info->pending_quota_state = 0;
2365         fs_info->qgroup_ulist = NULL;
2366         mutex_init(&fs_info->qgroup_rescan_lock);
2367
2368         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2369         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2370
2371         init_waitqueue_head(&fs_info->transaction_throttle);
2372         init_waitqueue_head(&fs_info->transaction_wait);
2373         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2374         init_waitqueue_head(&fs_info->async_submit_wait);
2375
2376         ret = btrfs_alloc_stripe_hash_table(fs_info);
2377         if (ret) {
2378                 err = ret;
2379                 goto fail_alloc;
2380         }
2381
2382         __setup_root(4096, 4096, 4096, 4096, tree_root,
2383                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2384
2385         invalidate_bdev(fs_devices->latest_bdev);
2386
2387         /*
2388          * Read super block and check the signature bytes only
2389          */
2390         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2391         if (!bh) {
2392                 err = -EINVAL;
2393                 goto fail_alloc;
2394         }
2395
2396         /*
2397          * We want to check superblock checksum, the type is stored inside.
2398          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2399          */
2400         if (btrfs_check_super_csum(bh->b_data)) {
2401                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2402                 err = -EINVAL;
2403                 goto fail_alloc;
2404         }
2405
2406         /*
2407          * super_copy is zeroed at allocation time and we never touch the
2408          * following bytes up to INFO_SIZE, the checksum is calculated from
2409          * the whole block of INFO_SIZE
2410          */
2411         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2412         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2413                sizeof(*fs_info->super_for_commit));
2414         brelse(bh);
2415
2416         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2417
2418         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2419         if (ret) {
2420                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2421                 err = -EINVAL;
2422                 goto fail_alloc;
2423         }
2424
2425         disk_super = fs_info->super_copy;
2426         if (!btrfs_super_root(disk_super))
2427                 goto fail_alloc;
2428
2429         /* check FS state, whether FS is broken. */
2430         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2431                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2432
2433         /*
2434          * run through our array of backup supers and setup
2435          * our ring pointer to the oldest one
2436          */
2437         generation = btrfs_super_generation(disk_super);
2438         find_oldest_super_backup(fs_info, generation);
2439
2440         /*
2441          * In the long term, we'll store the compression type in the super
2442          * block, and it'll be used for per file compression control.
2443          */
2444         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2445
2446         ret = btrfs_parse_options(tree_root, options);
2447         if (ret) {
2448                 err = ret;
2449                 goto fail_alloc;
2450         }
2451
2452         features = btrfs_super_incompat_flags(disk_super) &
2453                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2454         if (features) {
2455                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2456                        "unsupported optional features (%Lx).\n",
2457                        features);
2458                 err = -EINVAL;
2459                 goto fail_alloc;
2460         }
2461
2462         if (btrfs_super_leafsize(disk_super) !=
2463             btrfs_super_nodesize(disk_super)) {
2464                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2465                        "blocksizes don't match.  node %d leaf %d\n",
2466                        btrfs_super_nodesize(disk_super),
2467                        btrfs_super_leafsize(disk_super));
2468                 err = -EINVAL;
2469                 goto fail_alloc;
2470         }
2471         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2472                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2473                        "blocksize (%d) was too large\n",
2474                        btrfs_super_leafsize(disk_super));
2475                 err = -EINVAL;
2476                 goto fail_alloc;
2477         }
2478
2479         features = btrfs_super_incompat_flags(disk_super);
2480         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2481         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2482                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2483
2484         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2485                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2486
2487         /*
2488          * flag our filesystem as having big metadata blocks if
2489          * they are bigger than the page size
2490          */
2491         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2492                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2493                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2494                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2495         }
2496
2497         nodesize = btrfs_super_nodesize(disk_super);
2498         leafsize = btrfs_super_leafsize(disk_super);
2499         sectorsize = btrfs_super_sectorsize(disk_super);
2500         stripesize = btrfs_super_stripesize(disk_super);
2501         fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2502         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2503
2504         /*
2505          * mixed block groups end up with duplicate but slightly offset
2506          * extent buffers for the same range.  It leads to corruptions
2507          */
2508         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2509             (sectorsize != leafsize)) {
2510                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2511                                 "are not allowed for mixed block groups on %s\n",
2512                                 sb->s_id);
2513                 goto fail_alloc;
2514         }
2515
2516         /*
2517          * Needn't use the lock because there is no other task which will
2518          * update the flag.
2519          */
2520         btrfs_set_super_incompat_flags(disk_super, features);
2521
2522         features = btrfs_super_compat_ro_flags(disk_super) &
2523                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2524         if (!(sb->s_flags & MS_RDONLY) && features) {
2525                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2526                        "unsupported option features (%Lx).\n",
2527                        features);
2528                 err = -EINVAL;
2529                 goto fail_alloc;
2530         }
2531
2532         max_active = fs_info->thread_pool_size;
2533
2534         fs_info->workers =
2535                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2536                                       max_active, 16);
2537
2538         fs_info->delalloc_workers =
2539                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2540
2541         fs_info->flush_workers =
2542                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2543
2544         fs_info->caching_workers =
2545                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2546
2547         /*
2548          * a higher idle thresh on the submit workers makes it much more
2549          * likely that bios will be send down in a sane order to the
2550          * devices
2551          */
2552         fs_info->submit_workers =
2553                 btrfs_alloc_workqueue("submit", flags,
2554                                       min_t(u64, fs_devices->num_devices,
2555                                             max_active), 64);
2556
2557         fs_info->fixup_workers =
2558                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2559
2560         /*
2561          * endios are largely parallel and should have a very
2562          * low idle thresh
2563          */
2564         fs_info->endio_workers =
2565                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2566         fs_info->endio_meta_workers =
2567                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2568         fs_info->endio_meta_write_workers =
2569                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2570         fs_info->endio_raid56_workers =
2571                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2572         fs_info->rmw_workers =
2573                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2574         fs_info->endio_write_workers =
2575                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2576         fs_info->endio_freespace_worker =
2577                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2578         fs_info->delayed_workers =
2579                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2580         fs_info->readahead_workers =
2581                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2582         fs_info->qgroup_rescan_workers =
2583                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2584
2585         if (!(fs_info->workers && fs_info->delalloc_workers &&
2586               fs_info->submit_workers && fs_info->flush_workers &&
2587               fs_info->endio_workers && fs_info->endio_meta_workers &&
2588               fs_info->endio_meta_write_workers &&
2589               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2590               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2591               fs_info->caching_workers && fs_info->readahead_workers &&
2592               fs_info->fixup_workers && fs_info->delayed_workers &&
2593               fs_info->qgroup_rescan_workers)) {
2594                 err = -ENOMEM;
2595                 goto fail_sb_buffer;
2596         }
2597
2598         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2599         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2600                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2601
2602         tree_root->nodesize = nodesize;
2603         tree_root->leafsize = leafsize;
2604         tree_root->sectorsize = sectorsize;
2605         tree_root->stripesize = stripesize;
2606
2607         sb->s_blocksize = sectorsize;
2608         sb->s_blocksize_bits = blksize_bits(sectorsize);
2609
2610         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2611                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2612                 goto fail_sb_buffer;
2613         }
2614
2615         if (sectorsize != PAGE_SIZE) {
2616                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2617                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2618                 goto fail_sb_buffer;
2619         }
2620
2621         mutex_lock(&fs_info->chunk_mutex);
2622         ret = btrfs_read_sys_array(tree_root);
2623         mutex_unlock(&fs_info->chunk_mutex);
2624         if (ret) {
2625                 printk(KERN_WARNING "BTRFS: failed to read the system "
2626                        "array on %s\n", sb->s_id);
2627                 goto fail_sb_buffer;
2628         }
2629
2630         blocksize = btrfs_level_size(tree_root,
2631                                      btrfs_super_chunk_root_level(disk_super));
2632         generation = btrfs_super_chunk_root_generation(disk_super);
2633
2634         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2635                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2636
2637         chunk_root->node = read_tree_block(chunk_root,
2638                                            btrfs_super_chunk_root(disk_super),
2639                                            blocksize, generation);
2640         if (!chunk_root->node ||
2641             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2642                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2643                        sb->s_id);
2644                 goto fail_tree_roots;
2645         }
2646         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2647         chunk_root->commit_root = btrfs_root_node(chunk_root);
2648
2649         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2650            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2651
2652         ret = btrfs_read_chunk_tree(chunk_root);
2653         if (ret) {
2654                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2655                        sb->s_id);
2656                 goto fail_tree_roots;
2657         }
2658
2659         /*
2660          * keep the device that is marked to be the target device for the
2661          * dev_replace procedure
2662          */
2663         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2664
2665         if (!fs_devices->latest_bdev) {
2666                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2667                        sb->s_id);
2668                 goto fail_tree_roots;
2669         }
2670
2671 retry_root_backup:
2672         blocksize = btrfs_level_size(tree_root,
2673                                      btrfs_super_root_level(disk_super));
2674         generation = btrfs_super_generation(disk_super);
2675
2676         tree_root->node = read_tree_block(tree_root,
2677                                           btrfs_super_root(disk_super),
2678                                           blocksize, generation);
2679         if (!tree_root->node ||
2680             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2681                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2682                        sb->s_id);
2683
2684                 goto recovery_tree_root;
2685         }
2686
2687         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2688         tree_root->commit_root = btrfs_root_node(tree_root);
2689         btrfs_set_root_refs(&tree_root->root_item, 1);
2690
2691         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2692         location.type = BTRFS_ROOT_ITEM_KEY;
2693         location.offset = 0;
2694
2695         extent_root = btrfs_read_tree_root(tree_root, &location);
2696         if (IS_ERR(extent_root)) {
2697                 ret = PTR_ERR(extent_root);
2698                 goto recovery_tree_root;
2699         }
2700         extent_root->track_dirty = 1;
2701         fs_info->extent_root = extent_root;
2702
2703         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2704         dev_root = btrfs_read_tree_root(tree_root, &location);
2705         if (IS_ERR(dev_root)) {
2706                 ret = PTR_ERR(dev_root);
2707                 goto recovery_tree_root;
2708         }
2709         dev_root->track_dirty = 1;
2710         fs_info->dev_root = dev_root;
2711         btrfs_init_devices_late(fs_info);
2712
2713         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2714         csum_root = btrfs_read_tree_root(tree_root, &location);
2715         if (IS_ERR(csum_root)) {
2716                 ret = PTR_ERR(csum_root);
2717                 goto recovery_tree_root;
2718         }
2719         csum_root->track_dirty = 1;
2720         fs_info->csum_root = csum_root;
2721
2722         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2723         quota_root = btrfs_read_tree_root(tree_root, &location);
2724         if (!IS_ERR(quota_root)) {
2725                 quota_root->track_dirty = 1;
2726                 fs_info->quota_enabled = 1;
2727                 fs_info->pending_quota_state = 1;
2728                 fs_info->quota_root = quota_root;
2729         }
2730
2731         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2732         uuid_root = btrfs_read_tree_root(tree_root, &location);
2733         if (IS_ERR(uuid_root)) {
2734                 ret = PTR_ERR(uuid_root);
2735                 if (ret != -ENOENT)
2736                         goto recovery_tree_root;
2737                 create_uuid_tree = true;
2738                 check_uuid_tree = false;
2739         } else {
2740                 uuid_root->track_dirty = 1;
2741                 fs_info->uuid_root = uuid_root;
2742                 create_uuid_tree = false;
2743                 check_uuid_tree =
2744                     generation != btrfs_super_uuid_tree_generation(disk_super);
2745         }
2746
2747         fs_info->generation = generation;
2748         fs_info->last_trans_committed = generation;
2749
2750         ret = btrfs_recover_balance(fs_info);
2751         if (ret) {
2752                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2753                 goto fail_block_groups;
2754         }
2755
2756         ret = btrfs_init_dev_stats(fs_info);
2757         if (ret) {
2758                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2759                        ret);
2760                 goto fail_block_groups;
2761         }
2762
2763         ret = btrfs_init_dev_replace(fs_info);
2764         if (ret) {
2765                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2766                 goto fail_block_groups;
2767         }
2768
2769         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2770
2771         ret = btrfs_sysfs_add_one(fs_info);
2772         if (ret) {
2773                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2774                 goto fail_block_groups;
2775         }
2776
2777         ret = btrfs_init_space_info(fs_info);
2778         if (ret) {
2779                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2780                 goto fail_sysfs;
2781         }
2782
2783         ret = btrfs_read_block_groups(extent_root);
2784         if (ret) {
2785                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2786                 goto fail_sysfs;
2787         }
2788         fs_info->num_tolerated_disk_barrier_failures =
2789                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2790         if (fs_info->fs_devices->missing_devices >
2791              fs_info->num_tolerated_disk_barrier_failures &&
2792             !(sb->s_flags & MS_RDONLY)) {
2793                 printk(KERN_WARNING "BTRFS: "
2794                         "too many missing devices, writeable mount is not allowed\n");
2795                 goto fail_sysfs;
2796         }
2797
2798         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2799                                                "btrfs-cleaner");
2800         if (IS_ERR(fs_info->cleaner_kthread))
2801                 goto fail_sysfs;
2802
2803         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2804                                                    tree_root,
2805                                                    "btrfs-transaction");
2806         if (IS_ERR(fs_info->transaction_kthread))
2807                 goto fail_cleaner;
2808
2809         if (!btrfs_test_opt(tree_root, SSD) &&
2810             !btrfs_test_opt(tree_root, NOSSD) &&
2811             !fs_info->fs_devices->rotating) {
2812                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2813                        "mode\n");
2814                 btrfs_set_opt(fs_info->mount_opt, SSD);
2815         }
2816
2817         /* Set the real inode map cache flag */
2818         if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2819                 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2820
2821 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2822         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2823                 ret = btrfsic_mount(tree_root, fs_devices,
2824                                     btrfs_test_opt(tree_root,
2825                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2826                                     1 : 0,
2827                                     fs_info->check_integrity_print_mask);
2828                 if (ret)
2829                         printk(KERN_WARNING "BTRFS: failed to initialize"
2830                                " integrity check module %s\n", sb->s_id);
2831         }
2832 #endif
2833         ret = btrfs_read_qgroup_config(fs_info);
2834         if (ret)
2835                 goto fail_trans_kthread;
2836
2837         /* do not make disk changes in broken FS */
2838         if (btrfs_super_log_root(disk_super) != 0) {
2839                 u64 bytenr = btrfs_super_log_root(disk_super);
2840
2841                 if (fs_devices->rw_devices == 0) {
2842                         printk(KERN_WARNING "BTRFS: log replay required "
2843                                "on RO media\n");
2844                         err = -EIO;
2845                         goto fail_qgroup;
2846                 }
2847                 blocksize =
2848                      btrfs_level_size(tree_root,
2849                                       btrfs_super_log_root_level(disk_super));
2850
2851                 log_tree_root = btrfs_alloc_root(fs_info);
2852                 if (!log_tree_root) {
2853                         err = -ENOMEM;
2854                         goto fail_qgroup;
2855                 }
2856
2857                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2858                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2859
2860                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2861                                                       blocksize,
2862                                                       generation + 1);
2863                 if (!log_tree_root->node ||
2864                     !extent_buffer_uptodate(log_tree_root->node)) {
2865                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2866                         free_extent_buffer(log_tree_root->node);
2867                         kfree(log_tree_root);
2868                         goto fail_trans_kthread;
2869                 }
2870                 /* returns with log_tree_root freed on success */
2871                 ret = btrfs_recover_log_trees(log_tree_root);
2872                 if (ret) {
2873                         btrfs_error(tree_root->fs_info, ret,
2874                                     "Failed to recover log tree");
2875                         free_extent_buffer(log_tree_root->node);
2876                         kfree(log_tree_root);
2877                         goto fail_trans_kthread;
2878                 }
2879
2880                 if (sb->s_flags & MS_RDONLY) {
2881                         ret = btrfs_commit_super(tree_root);
2882                         if (ret)
2883                                 goto fail_trans_kthread;
2884                 }
2885         }
2886
2887         ret = btrfs_find_orphan_roots(tree_root);
2888         if (ret)
2889                 goto fail_trans_kthread;
2890
2891         if (!(sb->s_flags & MS_RDONLY)) {
2892                 ret = btrfs_cleanup_fs_roots(fs_info);
2893                 if (ret)
2894                         goto fail_trans_kthread;
2895
2896                 ret = btrfs_recover_relocation(tree_root);
2897                 if (ret < 0) {
2898                         printk(KERN_WARNING
2899                                "BTRFS: failed to recover relocation\n");
2900                         err = -EINVAL;
2901                         goto fail_qgroup;
2902                 }
2903         }
2904
2905         location.objectid = BTRFS_FS_TREE_OBJECTID;
2906         location.type = BTRFS_ROOT_ITEM_KEY;
2907         location.offset = 0;
2908
2909         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2910         if (IS_ERR(fs_info->fs_root)) {
2911                 err = PTR_ERR(fs_info->fs_root);
2912                 goto fail_qgroup;
2913         }
2914
2915         if (sb->s_flags & MS_RDONLY)
2916                 return 0;
2917
2918         down_read(&fs_info->cleanup_work_sem);
2919         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2920             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2921                 up_read(&fs_info->cleanup_work_sem);
2922                 close_ctree(tree_root);
2923                 return ret;
2924         }
2925         up_read(&fs_info->cleanup_work_sem);
2926
2927         ret = btrfs_resume_balance_async(fs_info);
2928         if (ret) {
2929                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2930                 close_ctree(tree_root);
2931                 return ret;
2932         }
2933
2934         ret = btrfs_resume_dev_replace_async(fs_info);
2935         if (ret) {
2936                 pr_warn("BTRFS: failed to resume dev_replace\n");
2937                 close_ctree(tree_root);
2938                 return ret;
2939         }
2940
2941         btrfs_qgroup_rescan_resume(fs_info);
2942
2943         if (create_uuid_tree) {
2944                 pr_info("BTRFS: creating UUID tree\n");
2945                 ret = btrfs_create_uuid_tree(fs_info);
2946                 if (ret) {
2947                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2948                                 ret);
2949                         close_ctree(tree_root);
2950                         return ret;
2951                 }
2952         } else if (check_uuid_tree ||
2953                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2954                 pr_info("BTRFS: checking UUID tree\n");
2955                 ret = btrfs_check_uuid_tree(fs_info);
2956                 if (ret) {
2957                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2958                                 ret);
2959                         close_ctree(tree_root);
2960                         return ret;
2961                 }
2962         } else {
2963                 fs_info->update_uuid_tree_gen = 1;
2964         }
2965
2966         return 0;
2967
2968 fail_qgroup:
2969         btrfs_free_qgroup_config(fs_info);
2970 fail_trans_kthread:
2971         kthread_stop(fs_info->transaction_kthread);
2972         btrfs_cleanup_transaction(fs_info->tree_root);
2973         del_fs_roots(fs_info);
2974 fail_cleaner:
2975         kthread_stop(fs_info->cleaner_kthread);
2976
2977         /*
2978          * make sure we're done with the btree inode before we stop our
2979          * kthreads
2980          */
2981         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2982
2983 fail_sysfs:
2984         btrfs_sysfs_remove_one(fs_info);
2985
2986 fail_block_groups:
2987         btrfs_put_block_group_cache(fs_info);
2988         btrfs_free_block_groups(fs_info);
2989
2990 fail_tree_roots:
2991         free_root_pointers(fs_info, 1);
2992         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2993
2994 fail_sb_buffer:
2995         btrfs_stop_all_workers(fs_info);
2996 fail_alloc:
2997 fail_iput:
2998         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2999
3000         iput(fs_info->btree_inode);
3001 fail_bio_counter:
3002         percpu_counter_destroy(&fs_info->bio_counter);
3003 fail_delalloc_bytes:
3004         percpu_counter_destroy(&fs_info->delalloc_bytes);
3005 fail_dirty_metadata_bytes:
3006         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3007 fail_bdi:
3008         bdi_destroy(&fs_info->bdi);
3009 fail_srcu:
3010         cleanup_srcu_struct(&fs_info->subvol_srcu);
3011 fail:
3012         btrfs_free_stripe_hash_table(fs_info);
3013         btrfs_close_devices(fs_info->fs_devices);
3014         return err;
3015
3016 recovery_tree_root:
3017         if (!btrfs_test_opt(tree_root, RECOVERY))
3018                 goto fail_tree_roots;
3019
3020         free_root_pointers(fs_info, 0);
3021
3022         /* don't use the log in recovery mode, it won't be valid */
3023         btrfs_set_super_log_root(disk_super, 0);
3024
3025         /* we can't trust the free space cache either */
3026         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3027
3028         ret = next_root_backup(fs_info, fs_info->super_copy,
3029                                &num_backups_tried, &backup_index);
3030         if (ret == -1)
3031                 goto fail_block_groups;
3032         goto retry_root_backup;
3033 }
3034
3035 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3036 {
3037         if (uptodate) {
3038                 set_buffer_uptodate(bh);
3039         } else {
3040                 struct btrfs_device *device = (struct btrfs_device *)
3041                         bh->b_private;
3042
3043                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3044                                           "I/O error on %s\n",
3045                                           rcu_str_deref(device->name));
3046                 /* note, we dont' set_buffer_write_io_error because we have
3047                  * our own ways of dealing with the IO errors
3048                  */
3049                 clear_buffer_uptodate(bh);
3050                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3051         }
3052         unlock_buffer(bh);
3053         put_bh(bh);
3054 }
3055
3056 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3057 {
3058         struct buffer_head *bh;
3059         struct buffer_head *latest = NULL;
3060         struct btrfs_super_block *super;
3061         int i;
3062         u64 transid = 0;
3063         u64 bytenr;
3064
3065         /* we would like to check all the supers, but that would make
3066          * a btrfs mount succeed after a mkfs from a different FS.
3067          * So, we need to add a special mount option to scan for
3068          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3069          */
3070         for (i = 0; i < 1; i++) {
3071                 bytenr = btrfs_sb_offset(i);
3072                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3073                                         i_size_read(bdev->bd_inode))
3074                         break;
3075                 bh = __bread(bdev, bytenr / 4096,
3076                                         BTRFS_SUPER_INFO_SIZE);
3077                 if (!bh)
3078                         continue;
3079
3080                 super = (struct btrfs_super_block *)bh->b_data;
3081                 if (btrfs_super_bytenr(super) != bytenr ||
3082                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3083                         brelse(bh);
3084                         continue;
3085                 }
3086
3087                 if (!latest || btrfs_super_generation(super) > transid) {
3088                         brelse(latest);
3089                         latest = bh;
3090                         transid = btrfs_super_generation(super);
3091                 } else {
3092                         brelse(bh);
3093                 }
3094         }
3095         return latest;
3096 }
3097
3098 /*
3099  * this should be called twice, once with wait == 0 and
3100  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3101  * we write are pinned.
3102  *
3103  * They are released when wait == 1 is done.
3104  * max_mirrors must be the same for both runs, and it indicates how
3105  * many supers on this one device should be written.
3106  *
3107  * max_mirrors == 0 means to write them all.
3108  */
3109 static int write_dev_supers(struct btrfs_device *device,
3110                             struct btrfs_super_block *sb,
3111                             int do_barriers, int wait, int max_mirrors)
3112 {
3113         struct buffer_head *bh;
3114         int i;
3115         int ret;
3116         int errors = 0;
3117         u32 crc;
3118         u64 bytenr;
3119
3120         if (max_mirrors == 0)
3121                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3122
3123         for (i = 0; i < max_mirrors; i++) {
3124                 bytenr = btrfs_sb_offset(i);
3125                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3126                         break;
3127
3128                 if (wait) {
3129                         bh = __find_get_block(device->bdev, bytenr / 4096,
3130                                               BTRFS_SUPER_INFO_SIZE);
3131                         if (!bh) {
3132                                 errors++;
3133                                 continue;
3134                         }
3135                         wait_on_buffer(bh);
3136                         if (!buffer_uptodate(bh))
3137                                 errors++;
3138
3139                         /* drop our reference */
3140                         brelse(bh);
3141
3142                         /* drop the reference from the wait == 0 run */
3143                         brelse(bh);
3144                         continue;
3145                 } else {
3146                         btrfs_set_super_bytenr(sb, bytenr);
3147
3148                         crc = ~(u32)0;
3149                         crc = btrfs_csum_data((char *)sb +
3150                                               BTRFS_CSUM_SIZE, crc,
3151                                               BTRFS_SUPER_INFO_SIZE -
3152                                               BTRFS_CSUM_SIZE);
3153                         btrfs_csum_final(crc, sb->csum);
3154
3155                         /*
3156                          * one reference for us, and we leave it for the
3157                          * caller
3158                          */
3159                         bh = __getblk(device->bdev, bytenr / 4096,
3160                                       BTRFS_SUPER_INFO_SIZE);
3161                         if (!bh) {
3162                                 printk(KERN_ERR "BTRFS: couldn't get super "
3163                                        "buffer head for bytenr %Lu\n", bytenr);
3164                                 errors++;
3165                                 continue;
3166                         }
3167
3168                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3169
3170                         /* one reference for submit_bh */
3171                         get_bh(bh);
3172
3173                         set_buffer_uptodate(bh);
3174                         lock_buffer(bh);
3175                         bh->b_end_io = btrfs_end_buffer_write_sync;
3176                         bh->b_private = device;
3177                 }
3178
3179                 /*
3180                  * we fua the first super.  The others we allow
3181                  * to go down lazy.
3182                  */
3183                 if (i == 0)
3184                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3185                 else
3186                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3187                 if (ret)
3188                         errors++;
3189         }
3190         return errors < i ? 0 : -1;
3191 }
3192
3193 /*
3194  * endio for the write_dev_flush, this will wake anyone waiting
3195  * for the barrier when it is done
3196  */
3197 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3198 {
3199         if (err) {
3200                 if (err == -EOPNOTSUPP)
3201                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3202                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3203         }
3204         if (bio->bi_private)
3205                 complete(bio->bi_private);
3206         bio_put(bio);
3207 }
3208
3209 /*
3210  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3211  * sent down.  With wait == 1, it waits for the previous flush.
3212  *
3213  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3214  * capable
3215  */
3216 static int write_dev_flush(struct btrfs_device *device, int wait)
3217 {
3218         struct bio *bio;
3219         int ret = 0;
3220
3221         if (device->nobarriers)
3222                 return 0;
3223
3224         if (wait) {
3225                 bio = device->flush_bio;
3226                 if (!bio)
3227                         return 0;
3228
3229                 wait_for_completion(&device->flush_wait);
3230
3231                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3232                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3233                                       rcu_str_deref(device->name));
3234                         device->nobarriers = 1;
3235                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3236                         ret = -EIO;
3237                         btrfs_dev_stat_inc_and_print(device,
3238                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3239                 }
3240
3241                 /* drop the reference from the wait == 0 run */
3242                 bio_put(bio);
3243                 device->flush_bio = NULL;
3244
3245                 return ret;
3246         }
3247
3248         /*
3249          * one reference for us, and we leave it for the
3250          * caller
3251          */
3252         device->flush_bio = NULL;
3253         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3254         if (!bio)
3255                 return -ENOMEM;
3256
3257         bio->bi_end_io = btrfs_end_empty_barrier;
3258         bio->bi_bdev = device->bdev;
3259         init_completion(&device->flush_wait);
3260         bio->bi_private = &device->flush_wait;
3261         device->flush_bio = bio;
3262
3263         bio_get(bio);
3264         btrfsic_submit_bio(WRITE_FLUSH, bio);
3265
3266         return 0;
3267 }
3268
3269 /*
3270  * send an empty flush down to each device in parallel,
3271  * then wait for them
3272  */
3273 static int barrier_all_devices(struct btrfs_fs_info *info)
3274 {
3275         struct list_head *head;
3276         struct btrfs_device *dev;
3277         int errors_send = 0;
3278         int errors_wait = 0;
3279         int ret;
3280
3281         /* send down all the barriers */
3282         head = &info->fs_devices->devices;
3283         list_for_each_entry_rcu(dev, head, dev_list) {
3284                 if (dev->missing)
3285                         continue;
3286                 if (!dev->bdev) {
3287                         errors_send++;
3288                         continue;
3289                 }
3290                 if (!dev->in_fs_metadata || !dev->writeable)
3291                         continue;
3292
3293                 ret = write_dev_flush(dev, 0);
3294                 if (ret)
3295                         errors_send++;
3296         }
3297
3298         /* wait for all the barriers */
3299         list_for_each_entry_rcu(dev, head, dev_list) {
3300                 if (dev->missing)
3301                         continue;
3302                 if (!dev->bdev) {
3303                         errors_wait++;
3304                         continue;
3305                 }
3306                 if (!dev->in_fs_metadata || !dev->writeable)
3307                         continue;
3308
3309                 ret = write_dev_flush(dev, 1);
3310                 if (ret)
3311                         errors_wait++;
3312         }
3313         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3314             errors_wait > info->num_tolerated_disk_barrier_failures)
3315                 return -EIO;
3316         return 0;
3317 }
3318
3319 int btrfs_calc_num_tolerated_disk_barrier_failures(
3320         struct btrfs_fs_info *fs_info)
3321 {
3322         struct btrfs_ioctl_space_info space;
3323         struct btrfs_space_info *sinfo;
3324         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3325                        BTRFS_BLOCK_GROUP_SYSTEM,
3326                        BTRFS_BLOCK_GROUP_METADATA,
3327                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3328         int num_types = 4;
3329         int i;
3330         int c;
3331         int num_tolerated_disk_barrier_failures =
3332                 (int)fs_info->fs_devices->num_devices;
3333
3334         for (i = 0; i < num_types; i++) {
3335                 struct btrfs_space_info *tmp;
3336
3337                 sinfo = NULL;
3338                 rcu_read_lock();
3339                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3340                         if (tmp->flags == types[i]) {
3341                                 sinfo = tmp;
3342                                 break;
3343                         }
3344                 }
3345                 rcu_read_unlock();
3346
3347                 if (!sinfo)
3348                         continue;
3349
3350                 down_read(&sinfo->groups_sem);
3351                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3352                         if (!list_empty(&sinfo->block_groups[c])) {
3353                                 u64 flags;
3354
3355                                 btrfs_get_block_group_info(
3356                                         &sinfo->block_groups[c], &space);
3357                                 if (space.total_bytes == 0 ||
3358                                     space.used_bytes == 0)
3359                                         continue;
3360                                 flags = space.flags;
3361                                 /*
3362                                  * return
3363                                  * 0: if dup, single or RAID0 is configured for
3364                                  *    any of metadata, system or data, else
3365                                  * 1: if RAID5 is configured, or if RAID1 or
3366                                  *    RAID10 is configured and only two mirrors
3367                                  *    are used, else
3368                                  * 2: if RAID6 is configured, else
3369                                  * num_mirrors - 1: if RAID1 or RAID10 is
3370                                  *                  configured and more than
3371                                  *                  2 mirrors are used.
3372                                  */
3373                                 if (num_tolerated_disk_barrier_failures > 0 &&
3374                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3375                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3376                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3377                                       == 0)))
3378                                         num_tolerated_disk_barrier_failures = 0;
3379                                 else if (num_tolerated_disk_barrier_failures > 1) {
3380                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3381                                             BTRFS_BLOCK_GROUP_RAID5 |
3382                                             BTRFS_BLOCK_GROUP_RAID10)) {
3383                                                 num_tolerated_disk_barrier_failures = 1;
3384                                         } else if (flags &
3385                                                    BTRFS_BLOCK_GROUP_RAID6) {
3386                                                 num_tolerated_disk_barrier_failures = 2;
3387                                         }
3388                                 }
3389                         }
3390                 }
3391                 up_read(&sinfo->groups_sem);
3392         }
3393
3394         return num_tolerated_disk_barrier_failures;
3395 }
3396
3397 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3398 {
3399         struct list_head *head;
3400         struct btrfs_device *dev;
3401         struct btrfs_super_block *sb;
3402         struct btrfs_dev_item *dev_item;
3403         int ret;
3404         int do_barriers;
3405         int max_errors;
3406         int total_errors = 0;
3407         u64 flags;
3408
3409         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3410         backup_super_roots(root->fs_info);
3411
3412         sb = root->fs_info->super_for_commit;
3413         dev_item = &sb->dev_item;
3414
3415         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3416         head = &root->fs_info->fs_devices->devices;
3417         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3418
3419         if (do_barriers) {
3420                 ret = barrier_all_devices(root->fs_info);
3421                 if (ret) {
3422                         mutex_unlock(
3423                                 &root->fs_info->fs_devices->device_list_mutex);
3424                         btrfs_error(root->fs_info, ret,
3425                                     "errors while submitting device barriers.");
3426                         return ret;
3427                 }
3428         }
3429
3430         list_for_each_entry_rcu(dev, head, dev_list) {
3431                 if (!dev->bdev) {
3432                         total_errors++;
3433                         continue;
3434                 }
3435                 if (!dev->in_fs_metadata || !dev->writeable)
3436                         continue;
3437
3438                 btrfs_set_stack_device_generation(dev_item, 0);
3439                 btrfs_set_stack_device_type(dev_item, dev->type);
3440                 btrfs_set_stack_device_id(dev_item, dev->devid);
3441                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3442                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3443                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3444                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3445                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3446                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3447                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3448
3449                 flags = btrfs_super_flags(sb);
3450                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3451
3452                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3453                 if (ret)
3454                         total_errors++;
3455         }
3456         if (total_errors > max_errors) {
3457                 btrfs_err(root->fs_info, "%d errors while writing supers",
3458                        total_errors);
3459                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3460
3461                 /* FUA is masked off if unsupported and can't be the reason */
3462                 btrfs_error(root->fs_info, -EIO,
3463                             "%d errors while writing supers", total_errors);
3464                 return -EIO;
3465         }
3466
3467         total_errors = 0;
3468         list_for_each_entry_rcu(dev, head, dev_list) {
3469                 if (!dev->bdev)
3470                         continue;
3471                 if (!dev->in_fs_metadata || !dev->writeable)
3472                         continue;
3473
3474                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3475                 if (ret)
3476                         total_errors++;
3477         }
3478         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3479         if (total_errors > max_errors) {
3480                 btrfs_error(root->fs_info, -EIO,
3481                             "%d errors while writing supers", total_errors);
3482                 return -EIO;
3483         }
3484         return 0;
3485 }
3486
3487 int write_ctree_super(struct btrfs_trans_handle *trans,
3488                       struct btrfs_root *root, int max_mirrors)
3489 {
3490         return write_all_supers(root, max_mirrors);
3491 }
3492
3493 /* Drop a fs root from the radix tree and free it. */
3494 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3495                                   struct btrfs_root *root)
3496 {
3497         spin_lock(&fs_info->fs_roots_radix_lock);
3498         radix_tree_delete(&fs_info->fs_roots_radix,
3499                           (unsigned long)root->root_key.objectid);
3500         spin_unlock(&fs_info->fs_roots_radix_lock);
3501
3502         if (btrfs_root_refs(&root->root_item) == 0)
3503                 synchronize_srcu(&fs_info->subvol_srcu);
3504
3505         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3506                 btrfs_free_log(NULL, root);
3507
3508         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3509         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3510         free_fs_root(root);
3511 }
3512
3513 static void free_fs_root(struct btrfs_root *root)
3514 {
3515         iput(root->cache_inode);
3516         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3517         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3518         root->orphan_block_rsv = NULL;
3519         if (root->anon_dev)
3520                 free_anon_bdev(root->anon_dev);
3521         if (root->subv_writers)
3522                 btrfs_free_subvolume_writers(root->subv_writers);
3523         free_extent_buffer(root->node);
3524         free_extent_buffer(root->commit_root);
3525         kfree(root->free_ino_ctl);
3526         kfree(root->free_ino_pinned);
3527         kfree(root->name);
3528         btrfs_put_fs_root(root);
3529 }
3530
3531 void btrfs_free_fs_root(struct btrfs_root *root)
3532 {
3533         free_fs_root(root);
3534 }
3535
3536 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3537 {
3538         u64 root_objectid = 0;
3539         struct btrfs_root *gang[8];
3540         int i;
3541         int ret;
3542
3543         while (1) {
3544                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3545                                              (void **)gang, root_objectid,
3546                                              ARRAY_SIZE(gang));
3547                 if (!ret)
3548                         break;
3549
3550                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3551                 for (i = 0; i < ret; i++) {
3552                         int err;
3553
3554                         root_objectid = gang[i]->root_key.objectid;
3555                         err = btrfs_orphan_cleanup(gang[i]);
3556                         if (err)
3557                                 return err;
3558                 }
3559                 root_objectid++;
3560         }
3561         return 0;
3562 }
3563
3564 int btrfs_commit_super(struct btrfs_root *root)
3565 {
3566         struct btrfs_trans_handle *trans;
3567
3568         mutex_lock(&root->fs_info->cleaner_mutex);
3569         btrfs_run_delayed_iputs(root);
3570         mutex_unlock(&root->fs_info->cleaner_mutex);
3571         wake_up_process(root->fs_info->cleaner_kthread);
3572
3573         /* wait until ongoing cleanup work done */
3574         down_write(&root->fs_info->cleanup_work_sem);
3575         up_write(&root->fs_info->cleanup_work_sem);
3576
3577         trans = btrfs_join_transaction(root);
3578         if (IS_ERR(trans))
3579                 return PTR_ERR(trans);
3580         return btrfs_commit_transaction(trans, root);
3581 }
3582
3583 int close_ctree(struct btrfs_root *root)
3584 {
3585         struct btrfs_fs_info *fs_info = root->fs_info;
3586         int ret;
3587
3588         fs_info->closing = 1;
3589         smp_mb();
3590
3591         /* wait for the uuid_scan task to finish */
3592         down(&fs_info->uuid_tree_rescan_sem);
3593         /* avoid complains from lockdep et al., set sem back to initial state */
3594         up(&fs_info->uuid_tree_rescan_sem);
3595
3596         /* pause restriper - we want to resume on mount */
3597         btrfs_pause_balance(fs_info);
3598
3599         btrfs_dev_replace_suspend_for_unmount(fs_info);
3600
3601         btrfs_scrub_cancel(fs_info);
3602
3603         /* wait for any defraggers to finish */
3604         wait_event(fs_info->transaction_wait,
3605                    (atomic_read(&fs_info->defrag_running) == 0));
3606
3607         /* clear out the rbtree of defraggable inodes */
3608         btrfs_cleanup_defrag_inodes(fs_info);
3609
3610         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3611                 ret = btrfs_commit_super(root);
3612                 if (ret)
3613                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3614         }
3615
3616         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3617                 btrfs_error_commit_super(root);
3618
3619         kthread_stop(fs_info->transaction_kthread);
3620         kthread_stop(fs_info->cleaner_kthread);
3621
3622         fs_info->closing = 2;
3623         smp_mb();
3624
3625         btrfs_free_qgroup_config(root->fs_info);
3626
3627         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3628                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3629                        percpu_counter_sum(&fs_info->delalloc_bytes));
3630         }
3631
3632         btrfs_sysfs_remove_one(fs_info);
3633
3634         del_fs_roots(fs_info);
3635
3636         btrfs_put_block_group_cache(fs_info);
3637
3638         btrfs_free_block_groups(fs_info);
3639
3640         btrfs_stop_all_workers(fs_info);
3641
3642         free_root_pointers(fs_info, 1);
3643
3644         iput(fs_info->btree_inode);
3645
3646 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3647         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3648                 btrfsic_unmount(root, fs_info->fs_devices);
3649 #endif
3650
3651         btrfs_close_devices(fs_info->fs_devices);
3652         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3653
3654         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3655         percpu_counter_destroy(&fs_info->delalloc_bytes);
3656         percpu_counter_destroy(&fs_info->bio_counter);
3657         bdi_destroy(&fs_info->bdi);
3658         cleanup_srcu_struct(&fs_info->subvol_srcu);
3659
3660         btrfs_free_stripe_hash_table(fs_info);
3661
3662         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3663         root->orphan_block_rsv = NULL;
3664
3665         return 0;
3666 }
3667
3668 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3669                           int atomic)
3670 {
3671         int ret;
3672         struct inode *btree_inode = buf->pages[0]->mapping->host;
3673
3674         ret = extent_buffer_uptodate(buf);
3675         if (!ret)
3676                 return ret;
3677
3678         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3679                                     parent_transid, atomic);
3680         if (ret == -EAGAIN)
3681                 return ret;
3682         return !ret;
3683 }
3684
3685 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3686 {
3687         return set_extent_buffer_uptodate(buf);
3688 }
3689
3690 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3691 {
3692         struct btrfs_root *root;
3693         u64 transid = btrfs_header_generation(buf);
3694         int was_dirty;
3695
3696 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3697         /*
3698          * This is a fast path so only do this check if we have sanity tests
3699          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3700          * outside of the sanity tests.
3701          */
3702         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3703                 return;
3704 #endif
3705         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3706         btrfs_assert_tree_locked(buf);
3707         if (transid != root->fs_info->generation)
3708                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3709                        "found %llu running %llu\n",
3710                         buf->start, transid, root->fs_info->generation);
3711         was_dirty = set_extent_buffer_dirty(buf);
3712         if (!was_dirty)
3713                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3714                                      buf->len,
3715                                      root->fs_info->dirty_metadata_batch);
3716 }
3717
3718 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3719                                         int flush_delayed)
3720 {
3721         /*
3722          * looks as though older kernels can get into trouble with
3723          * this code, they end up stuck in balance_dirty_pages forever
3724          */
3725         int ret;
3726
3727         if (current->flags & PF_MEMALLOC)
3728                 return;
3729
3730         if (flush_delayed)
3731                 btrfs_balance_delayed_items(root);
3732
3733         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3734                                      BTRFS_DIRTY_METADATA_THRESH);
3735         if (ret > 0) {
3736                 balance_dirty_pages_ratelimited(
3737                                    root->fs_info->btree_inode->i_mapping);
3738         }
3739         return;
3740 }
3741
3742 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3743 {
3744         __btrfs_btree_balance_dirty(root, 1);
3745 }
3746
3747 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3748 {
3749         __btrfs_btree_balance_dirty(root, 0);
3750 }
3751
3752 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3753 {
3754         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3755         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3756 }
3757
3758 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3759                               int read_only)
3760 {
3761         /*
3762          * Placeholder for checks
3763          */
3764         return 0;
3765 }
3766
3767 static void btrfs_error_commit_super(struct btrfs_root *root)
3768 {
3769         mutex_lock(&root->fs_info->cleaner_mutex);
3770         btrfs_run_delayed_iputs(root);
3771         mutex_unlock(&root->fs_info->cleaner_mutex);
3772
3773         down_write(&root->fs_info->cleanup_work_sem);
3774         up_write(&root->fs_info->cleanup_work_sem);
3775
3776         /* cleanup FS via transaction */
3777         btrfs_cleanup_transaction(root);
3778 }
3779
3780 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3781                                              struct btrfs_root *root)
3782 {
3783         struct btrfs_inode *btrfs_inode;
3784         struct list_head splice;
3785
3786         INIT_LIST_HEAD(&splice);
3787
3788         mutex_lock(&root->fs_info->ordered_operations_mutex);
3789         spin_lock(&root->fs_info->ordered_root_lock);
3790
3791         list_splice_init(&t->ordered_operations, &splice);
3792         while (!list_empty(&splice)) {
3793                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3794                                          ordered_operations);
3795
3796                 list_del_init(&btrfs_inode->ordered_operations);
3797                 spin_unlock(&root->fs_info->ordered_root_lock);
3798
3799                 btrfs_invalidate_inodes(btrfs_inode->root);
3800
3801                 spin_lock(&root->fs_info->ordered_root_lock);
3802         }
3803
3804         spin_unlock(&root->fs_info->ordered_root_lock);
3805         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3806 }
3807
3808 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3809 {
3810         struct btrfs_ordered_extent *ordered;
3811
3812         spin_lock(&root->ordered_extent_lock);
3813         /*
3814          * This will just short circuit the ordered completion stuff which will
3815          * make sure the ordered extent gets properly cleaned up.
3816          */
3817         list_for_each_entry(ordered, &root->ordered_extents,
3818                             root_extent_list)
3819                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3820         spin_unlock(&root->ordered_extent_lock);
3821 }
3822
3823 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3824 {
3825         struct btrfs_root *root;
3826         struct list_head splice;
3827
3828         INIT_LIST_HEAD(&splice);
3829
3830         spin_lock(&fs_info->ordered_root_lock);
3831         list_splice_init(&fs_info->ordered_roots, &splice);
3832         while (!list_empty(&splice)) {
3833                 root = list_first_entry(&splice, struct btrfs_root,
3834                                         ordered_root);
3835                 list_move_tail(&root->ordered_root,
3836                                &fs_info->ordered_roots);
3837
3838                 spin_unlock(&fs_info->ordered_root_lock);
3839                 btrfs_destroy_ordered_extents(root);
3840
3841                 cond_resched();
3842                 spin_lock(&fs_info->ordered_root_lock);
3843         }
3844         spin_unlock(&fs_info->ordered_root_lock);
3845 }
3846
3847 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3848                                       struct btrfs_root *root)
3849 {
3850         struct rb_node *node;
3851         struct btrfs_delayed_ref_root *delayed_refs;
3852         struct btrfs_delayed_ref_node *ref;
3853         int ret = 0;
3854
3855         delayed_refs = &trans->delayed_refs;
3856
3857         spin_lock(&delayed_refs->lock);
3858         if (atomic_read(&delayed_refs->num_entries) == 0) {
3859                 spin_unlock(&delayed_refs->lock);
3860                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3861                 return ret;
3862         }
3863
3864         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3865                 struct btrfs_delayed_ref_head *head;
3866                 bool pin_bytes = false;
3867
3868                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3869                                 href_node);
3870                 if (!mutex_trylock(&head->mutex)) {
3871                         atomic_inc(&head->node.refs);
3872                         spin_unlock(&delayed_refs->lock);
3873
3874                         mutex_lock(&head->mutex);
3875                         mutex_unlock(&head->mutex);
3876                         btrfs_put_delayed_ref(&head->node);
3877                         spin_lock(&delayed_refs->lock);
3878                         continue;
3879                 }
3880                 spin_lock(&head->lock);
3881                 while ((node = rb_first(&head->ref_root)) != NULL) {
3882                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3883                                        rb_node);
3884                         ref->in_tree = 0;
3885                         rb_erase(&ref->rb_node, &head->ref_root);
3886                         atomic_dec(&delayed_refs->num_entries);
3887                         btrfs_put_delayed_ref(ref);
3888                 }
3889                 if (head->must_insert_reserved)
3890                         pin_bytes = true;
3891                 btrfs_free_delayed_extent_op(head->extent_op);
3892                 delayed_refs->num_heads--;
3893                 if (head->processing == 0)
3894                         delayed_refs->num_heads_ready--;
3895                 atomic_dec(&delayed_refs->num_entries);
3896                 head->node.in_tree = 0;
3897                 rb_erase(&head->href_node, &delayed_refs->href_root);
3898                 spin_unlock(&head->lock);
3899                 spin_unlock(&delayed_refs->lock);
3900                 mutex_unlock(&head->mutex);
3901
3902                 if (pin_bytes)
3903                         btrfs_pin_extent(root, head->node.bytenr,
3904                                          head->node.num_bytes, 1);
3905                 btrfs_put_delayed_ref(&head->node);
3906                 cond_resched();
3907                 spin_lock(&delayed_refs->lock);
3908         }
3909
3910         spin_unlock(&delayed_refs->lock);
3911
3912         return ret;
3913 }
3914
3915 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3916 {
3917         struct btrfs_inode *btrfs_inode;
3918         struct list_head splice;
3919
3920         INIT_LIST_HEAD(&splice);
3921
3922         spin_lock(&root->delalloc_lock);
3923         list_splice_init(&root->delalloc_inodes, &splice);
3924
3925         while (!list_empty(&splice)) {
3926                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3927                                                delalloc_inodes);
3928
3929                 list_del_init(&btrfs_inode->delalloc_inodes);
3930                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3931                           &btrfs_inode->runtime_flags);
3932                 spin_unlock(&root->delalloc_lock);
3933
3934                 btrfs_invalidate_inodes(btrfs_inode->root);
3935
3936                 spin_lock(&root->delalloc_lock);
3937         }
3938
3939         spin_unlock(&root->delalloc_lock);
3940 }
3941
3942 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3943 {
3944         struct btrfs_root *root;
3945         struct list_head splice;
3946
3947         INIT_LIST_HEAD(&splice);
3948
3949         spin_lock(&fs_info->delalloc_root_lock);
3950         list_splice_init(&fs_info->delalloc_roots, &splice);
3951         while (!list_empty(&splice)) {
3952                 root = list_first_entry(&splice, struct btrfs_root,
3953                                          delalloc_root);
3954                 list_del_init(&root->delalloc_root);
3955                 root = btrfs_grab_fs_root(root);
3956                 BUG_ON(!root);
3957                 spin_unlock(&fs_info->delalloc_root_lock);
3958
3959                 btrfs_destroy_delalloc_inodes(root);
3960                 btrfs_put_fs_root(root);
3961
3962                 spin_lock(&fs_info->delalloc_root_lock);
3963         }
3964         spin_unlock(&fs_info->delalloc_root_lock);
3965 }
3966
3967 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3968                                         struct extent_io_tree *dirty_pages,
3969                                         int mark)
3970 {
3971         int ret;
3972         struct extent_buffer *eb;
3973         u64 start = 0;
3974         u64 end;
3975
3976         while (1) {
3977                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3978                                             mark, NULL);
3979                 if (ret)
3980                         break;
3981
3982                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3983                 while (start <= end) {
3984                         eb = btrfs_find_tree_block(root, start,
3985                                                    root->leafsize);
3986                         start += root->leafsize;
3987                         if (!eb)
3988                                 continue;
3989                         wait_on_extent_buffer_writeback(eb);
3990
3991                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3992                                                &eb->bflags))
3993                                 clear_extent_buffer_dirty(eb);
3994                         free_extent_buffer_stale(eb);
3995                 }
3996         }
3997
3998         return ret;
3999 }
4000
4001 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4002                                        struct extent_io_tree *pinned_extents)
4003 {
4004         struct extent_io_tree *unpin;
4005         u64 start;
4006         u64 end;
4007         int ret;
4008         bool loop = true;
4009
4010         unpin = pinned_extents;
4011 again:
4012         while (1) {
4013                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4014                                             EXTENT_DIRTY, NULL);
4015                 if (ret)
4016                         break;
4017
4018                 /* opt_discard */
4019                 if (btrfs_test_opt(root, DISCARD))
4020                         ret = btrfs_error_discard_extent(root, start,
4021                                                          end + 1 - start,
4022                                                          NULL);
4023
4024                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4025                 btrfs_error_unpin_extent_range(root, start, end);
4026                 cond_resched();
4027         }
4028
4029         if (loop) {
4030                 if (unpin == &root->fs_info->freed_extents[0])
4031                         unpin = &root->fs_info->freed_extents[1];
4032                 else
4033                         unpin = &root->fs_info->freed_extents[0];
4034                 loop = false;
4035                 goto again;
4036         }
4037
4038         return 0;
4039 }
4040
4041 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4042                                    struct btrfs_root *root)
4043 {
4044         btrfs_destroy_ordered_operations(cur_trans, root);
4045
4046         btrfs_destroy_delayed_refs(cur_trans, root);
4047
4048         cur_trans->state = TRANS_STATE_COMMIT_START;
4049         wake_up(&root->fs_info->transaction_blocked_wait);
4050
4051         cur_trans->state = TRANS_STATE_UNBLOCKED;
4052         wake_up(&root->fs_info->transaction_wait);
4053
4054         btrfs_destroy_delayed_inodes(root);
4055         btrfs_assert_delayed_root_empty(root);
4056
4057         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4058                                      EXTENT_DIRTY);
4059         btrfs_destroy_pinned_extent(root,
4060                                     root->fs_info->pinned_extents);
4061
4062         cur_trans->state =TRANS_STATE_COMPLETED;
4063         wake_up(&cur_trans->commit_wait);
4064
4065         /*
4066         memset(cur_trans, 0, sizeof(*cur_trans));
4067         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4068         */
4069 }
4070
4071 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4072 {
4073         struct btrfs_transaction *t;
4074
4075         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4076
4077         spin_lock(&root->fs_info->trans_lock);
4078         while (!list_empty(&root->fs_info->trans_list)) {
4079                 t = list_first_entry(&root->fs_info->trans_list,
4080                                      struct btrfs_transaction, list);
4081                 if (t->state >= TRANS_STATE_COMMIT_START) {
4082                         atomic_inc(&t->use_count);
4083                         spin_unlock(&root->fs_info->trans_lock);
4084                         btrfs_wait_for_commit(root, t->transid);
4085                         btrfs_put_transaction(t);
4086                         spin_lock(&root->fs_info->trans_lock);
4087                         continue;
4088                 }
4089                 if (t == root->fs_info->running_transaction) {
4090                         t->state = TRANS_STATE_COMMIT_DOING;
4091                         spin_unlock(&root->fs_info->trans_lock);
4092                         /*
4093                          * We wait for 0 num_writers since we don't hold a trans
4094                          * handle open currently for this transaction.
4095                          */
4096                         wait_event(t->writer_wait,
4097                                    atomic_read(&t->num_writers) == 0);
4098                 } else {
4099                         spin_unlock(&root->fs_info->trans_lock);
4100                 }
4101                 btrfs_cleanup_one_transaction(t, root);
4102
4103                 spin_lock(&root->fs_info->trans_lock);
4104                 if (t == root->fs_info->running_transaction)
4105                         root->fs_info->running_transaction = NULL;
4106                 list_del_init(&t->list);
4107                 spin_unlock(&root->fs_info->trans_lock);
4108
4109                 btrfs_put_transaction(t);
4110                 trace_btrfs_transaction_commit(root);
4111                 spin_lock(&root->fs_info->trans_lock);
4112         }
4113         spin_unlock(&root->fs_info->trans_lock);
4114         btrfs_destroy_all_ordered_extents(root->fs_info);
4115         btrfs_destroy_delayed_inodes(root);
4116         btrfs_assert_delayed_root_empty(root);
4117         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4118         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4119         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4120
4121         return 0;
4122 }
4123
4124 static struct extent_io_ops btree_extent_io_ops = {
4125         .readpage_end_io_hook = btree_readpage_end_io_hook,
4126         .readpage_io_failed_hook = btree_io_failed_hook,
4127         .submit_bio_hook = btree_submit_bio_hook,
4128         /* note we're sharing with inode.c for the merge bio hook */
4129         .merge_bio_hook = btrfs_merge_bio_hook,
4130 };