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/slab.h>
  29 #include <linux/migrate.h>
  30 #include <linux/ratelimit.h>
  31 #include <linux/uuid.h>
  32 #include <linux/semaphore.h>
  33 #include <asm/unaligned.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "hash.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "free-space-tree.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52 #include "compression.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68                                     int read_only);
  69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71                                       struct btrfs_root *root);
  72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  74                                         struct extent_io_tree *dirty_pages,
  75                                         int mark);
  76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  77                                        struct extent_io_tree *pinned_extents);
  78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  79 static void btrfs_error_commit_super(struct btrfs_root *root);
  80
  81 /*
  82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83  * is complete.  This is used during reads to verify checksums, and it is used
  84  * by writes to insert metadata for new file extents after IO is complete.
  85  */
  86 struct btrfs_end_io_wq {
  87         struct bio *bio;
  88         bio_end_io_t *end_io;
  89         void *private;
  90         struct btrfs_fs_info *info;
  91         int error;
  92         enum btrfs_wq_endio_type metadata;
  93         struct list_head list;
  94         struct btrfs_work work;
  95 };
  96
  97 static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99 int __init btrfs_end_io_wq_init(void)
 100 {
 101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102                                         sizeof(struct btrfs_end_io_wq),
 103                                         0,
 104                                         SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         kmem_cache_destroy(btrfs_end_io_wq_cache);
 114 }
 115
 116 /*
 117  * async submit bios are used to offload expensive checksumming
 118  * onto the worker threads.  They checksum file and metadata bios
 119  * just before they are sent down the IO stack.
 120  */
 121 struct async_submit_bio {
 122         struct inode *inode;
 123         struct bio *bio;
 124         struct list_head list;
 125         extent_submit_bio_hook_t *submit_bio_start;
 126         extent_submit_bio_hook_t *submit_bio_done;
 127         int mirror_num;
 128         unsigned long bio_flags;
 129         /*
 130          * bio_offset is optional, can be used if the pages in the bio
 131          * can't tell us where in the file the bio should go
 132          */
 133         u64 bio_offset;
 134         struct btrfs_work work;
 135         int error;
 136 };
 137
 138 /*
 139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 141  * the level the eb occupies in the tree.
 142  *
 143  * Different roots are used for different purposes and may nest inside each
 144  * other and they require separate keysets.  As lockdep keys should be
 145  * static, assign keysets according to the purpose of the root as indicated
 146  * by btrfs_root->objectid.  This ensures that all special purpose roots
 147  * have separate keysets.
 148  *
 149  * Lock-nesting across peer nodes is always done with the immediate parent
 150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 151  * subclass to avoid triggering lockdep warning in such cases.
 152  *
 153  * The key is set by the readpage_end_io_hook after the buffer has passed
 154  * csum validation but before the pages are unlocked.  It is also set by
 155  * btrfs_init_new_buffer on freshly allocated blocks.
 156  *
 157  * We also add a check to make sure the highest level of the tree is the
 158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 159  * needs update as well.
 160  */
 161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 162 # if BTRFS_MAX_LEVEL != 8
 163 #  error
 164 # endif
 165
 166 static struct btrfs_lockdep_keyset {
 167         u64                     id;             /* root objectid */
 168         const char              *name_stem;     /* lock name stem */
 169         char                    names[BTRFS_MAX_LEVEL + 1][20];
 170         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 171 } btrfs_lockdep_keysets[] = {
 172         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 173         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 174         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 175         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 176         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 177         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 178         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 179         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 180         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 181         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 182         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 183         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 184         { .id = 0,                              .name_stem = "tree"     },
 185 };
 186
 187 void __init btrfs_init_lockdep(void)
 188 {
 189         int i, j;
 190
 191         /* initialize lockdep class names */
 192         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 193                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 194
 195                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 196                         snprintf(ks->names[j], sizeof(ks->names[j]),
 197                                  "btrfs-%s-%02d", ks->name_stem, j);
 198         }
 199 }
 200
 201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 202                                     int level)
 203 {
 204         struct btrfs_lockdep_keyset *ks;
 205
 206         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 207
 208         /* find the matching keyset, id 0 is the default entry */
 209         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 210                 if (ks->id == objectid)
 211                         break;
 212
 213         lockdep_set_class_and_name(&eb->lock,
 214                                    &ks->keys[level], ks->names[level]);
 215 }
 216
 217 #endif
 218
 219 /*
 220  * extents on the btree inode are pretty simple, there's one extent
 221  * that covers the entire device
 222  */
 223 static struct extent_map *btree_get_extent(struct inode *inode,
 224                 struct page *page, size_t pg_offset, u64 start, u64 len,
 225                 int create)
 226 {
 227         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 228         struct extent_map *em;
 229         int ret;
 230
 231         read_lock(&em_tree->lock);
 232         em = lookup_extent_mapping(em_tree, start, len);
 233         if (em) {
 234                 em->bdev =
 235                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 236                 read_unlock(&em_tree->lock);
 237                 goto out;
 238         }
 239         read_unlock(&em_tree->lock);
 240
 241         em = alloc_extent_map();
 242         if (!em) {
 243                 em = ERR_PTR(-ENOMEM);
 244                 goto out;
 245         }
 246         em->start = 0;
 247         em->len = (u64)-1;
 248         em->block_len = (u64)-1;
 249         em->block_start = 0;
 250         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 251
 252         write_lock(&em_tree->lock);
 253         ret = add_extent_mapping(em_tree, em, 0);
 254         if (ret == -EEXIST) {
 255                 free_extent_map(em);
 256                 em = lookup_extent_mapping(em_tree, start, len);
 257                 if (!em)
 258                         em = ERR_PTR(-EIO);
 259         } else if (ret) {
 260                 free_extent_map(em);
 261                 em = ERR_PTR(ret);
 262         }
 263         write_unlock(&em_tree->lock);
 264
 265 out:
 266         return em;
 267 }
 268
 269 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 270 {
 271         return btrfs_crc32c(seed, data, len);
 272 }
 273
 274 void btrfs_csum_final(u32 crc, char *result)
 275 {
 276         put_unaligned_le32(~crc, result);
 277 }
 278
 279 /*
 280  * compute the csum for a btree block, and either verify it or write it
 281  * into the csum field of the block.
 282  */
 283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 284                            struct extent_buffer *buf,
 285                            int verify)
 286 {
 287         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 288         char *result = NULL;
 289         unsigned long len;
 290         unsigned long cur_len;
 291         unsigned long offset = BTRFS_CSUM_SIZE;
 292         char *kaddr;
 293         unsigned long map_start;
 294         unsigned long map_len;
 295         int err;
 296         u32 crc = ~(u32)0;
 297         unsigned long inline_result;
 298
 299         len = buf->len - offset;
 300         while (len > 0) {
 301                 err = map_private_extent_buffer(buf, offset, 32,
 302                                         &kaddr, &map_start, &map_len);
 303                 if (err)
 304                         return err;
 305                 cur_len = min(len, map_len - (offset - map_start));
 306                 crc = btrfs_csum_data(kaddr + offset - map_start,
 307                                       crc, cur_len);
 308                 len -= cur_len;
 309                 offset += cur_len;
 310         }
 311         if (csum_size > sizeof(inline_result)) {
 312                 result = kzalloc(csum_size, GFP_NOFS);
 313                 if (!result)
 314                         return -ENOMEM;
 315         } else {
 316                 result = (char *)&inline_result;
 317         }
 318
 319         btrfs_csum_final(crc, result);
 320
 321         if (verify) {
 322                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 323                         u32 val;
 324                         u32 found = 0;
 325                         memcpy(&found, result, csum_size);
 326
 327                         read_extent_buffer(buf, &val, 0, csum_size);
 328                         btrfs_warn_rl(fs_info,
 329                                 "%s checksum verify failed on %llu wanted %X found %X "
 330                                 "level %d",
 331                                 fs_info->sb->s_id, buf->start,
 332                                 val, found, btrfs_header_level(buf));
 333                         if (result != (char *)&inline_result)
 334                                 kfree(result);
 335                         return -EUCLEAN;
 336                 }
 337         } else {
 338                 write_extent_buffer(buf, result, 0, csum_size);
 339         }
 340         if (result != (char *)&inline_result)
 341                 kfree(result);
 342         return 0;
 343 }
 344
 345 /*
 346  * we can't consider a given block up to date unless the transid of the
 347  * block matches the transid in the parent node's pointer.  This is how we
 348  * detect blocks that either didn't get written at all or got written
 349  * in the wrong place.
 350  */
 351 static int verify_parent_transid(struct extent_io_tree *io_tree,
 352                                  struct extent_buffer *eb, u64 parent_transid,
 353                                  int atomic)
 354 {
 355         struct extent_state *cached_state = NULL;
 356         int ret;
 357         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 358
 359         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 360                 return 0;
 361
 362         if (atomic)
 363                 return -EAGAIN;
 364
 365         if (need_lock) {
 366                 btrfs_tree_read_lock(eb);
 367                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 368         }
 369
 370         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 371                          &cached_state);
 372         if (extent_buffer_uptodate(eb) &&
 373             btrfs_header_generation(eb) == parent_transid) {
 374                 ret = 0;
 375                 goto out;
 376         }
 377         btrfs_err_rl(eb->fs_info,
 378                 "parent transid verify failed on %llu wanted %llu found %llu",
 379                         eb->start,
 380                         parent_transid, btrfs_header_generation(eb));
 381         ret = 1;
 382
 383         /*
 384          * Things reading via commit roots that don't have normal protection,
 385          * like send, can have a really old block in cache that may point at a
 386          * block that has been freed and re-allocated.  So don't clear uptodate
 387          * if we find an eb that is under IO (dirty/writeback) because we could
 388          * end up reading in the stale data and then writing it back out and
 389          * making everybody very sad.
 390          */
 391         if (!extent_buffer_under_io(eb))
 392                 clear_extent_buffer_uptodate(eb);
 393 out:
 394         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 395                              &cached_state, GFP_NOFS);
 396         if (need_lock)
 397                 btrfs_tree_read_unlock_blocking(eb);
 398         return ret;
 399 }
 400
 401 /*
 402  * Return 0 if the superblock checksum type matches the checksum value of that
 403  * algorithm. Pass the raw disk superblock data.
 404  */
 405 static int btrfs_check_super_csum(char *raw_disk_sb)
 406 {
 407         struct btrfs_super_block *disk_sb =
 408                 (struct btrfs_super_block *)raw_disk_sb;
 409         u16 csum_type = btrfs_super_csum_type(disk_sb);
 410         int ret = 0;
 411
 412         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 413                 u32 crc = ~(u32)0;
 414                 const int csum_size = sizeof(crc);
 415                 char result[csum_size];
 416
 417                 /*
 418                  * The super_block structure does not span the whole
 419                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 420                  * is filled with zeros and is included in the checksum.
 421                  */
 422                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 423                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 424                 btrfs_csum_final(crc, result);
 425
 426                 if (memcmp(raw_disk_sb, result, csum_size))
 427                         ret = 1;
 428         }
 429
 430         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 431                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 432                                 csum_type);
 433                 ret = 1;
 434         }
 435
 436         return ret;
 437 }
 438
 439 /*
 440  * helper to read a given tree block, doing retries as required when
 441  * the checksums don't match and we have alternate mirrors to try.
 442  */
 443 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 444                                           struct extent_buffer *eb,
 445                                           u64 start, u64 parent_transid)
 446 {
 447         struct extent_io_tree *io_tree;
 448         int failed = 0;
 449         int ret;
 450         int num_copies = 0;
 451         int mirror_num = 0;
 452         int failed_mirror = 0;
 453
 454         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 455         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 456         while (1) {
 457                 ret = read_extent_buffer_pages(io_tree, eb, start,
 458                                                WAIT_COMPLETE,
 459                                                btree_get_extent, mirror_num);
 460                 if (!ret) {
 461                         if (!verify_parent_transid(io_tree, eb,
 462                                                    parent_transid, 0))
 463                                 break;
 464                         else
 465                                 ret = -EIO;
 466                 }
 467
 468                 /*
 469                  * This buffer's crc is fine, but its contents are corrupted, so
 470                  * there is no reason to read the other copies, they won't be
 471                  * any less wrong.
 472                  */
 473                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 474                         break;
 475
 476                 num_copies = btrfs_num_copies(root->fs_info,
 477                                               eb->start, eb->len);
 478                 if (num_copies == 1)
 479                         break;
 480
 481                 if (!failed_mirror) {
 482                         failed = 1;
 483                         failed_mirror = eb->read_mirror;
 484                 }
 485
 486                 mirror_num++;
 487                 if (mirror_num == failed_mirror)
 488                         mirror_num++;
 489
 490                 if (mirror_num > num_copies)
 491                         break;
 492         }
 493
 494         if (failed && !ret && failed_mirror)
 495                 repair_eb_io_failure(root, eb, failed_mirror);
 496
 497         return ret;
 498 }
 499
 500 /*
 501  * checksum a dirty tree block before IO.  This has extra checks to make sure
 502  * we only fill in the checksum field in the first page of a multi-page block
 503  */
 504
 505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 506 {
 507         u64 start = page_offset(page);
 508         u64 found_start;
 509         struct extent_buffer *eb;
 510
 511         eb = (struct extent_buffer *)page->private;
 512         if (page != eb->pages[0])
 513                 return 0;
 514
 515         found_start = btrfs_header_bytenr(eb);
 516         /*
 517          * Please do not consolidate these warnings into a single if.
 518          * It is useful to know what went wrong.
 519          */
 520         if (WARN_ON(found_start != start))
 521                 return -EUCLEAN;
 522         if (WARN_ON(!PageUptodate(page)))
 523                 return -EUCLEAN;
 524
 525         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 526                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 527
 528         return csum_tree_block(fs_info, eb, 0);
 529 }
 530
 531 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 532                                  struct extent_buffer *eb)
 533 {
 534         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 535         u8 fsid[BTRFS_UUID_SIZE];
 536         int ret = 1;
 537
 538         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 539         while (fs_devices) {
 540                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 541                         ret = 0;
 542                         break;
 543                 }
 544                 fs_devices = fs_devices->seed;
 545         }
 546         return ret;
 547 }
 548
 549 #define CORRUPT(reason, eb, root, slot)                         \
 550         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 551                    "root=%llu, slot=%d", reason,                        \
 552                btrfs_header_bytenr(eb), root->objectid, slot)
 553
 554 static noinline int check_leaf(struct btrfs_root *root,
 555                                struct extent_buffer *leaf)
 556 {
 557         struct btrfs_key key;
 558         struct btrfs_key leaf_key;
 559         u32 nritems = btrfs_header_nritems(leaf);
 560         int slot;
 561
 562         if (nritems == 0) {
 563                 struct btrfs_root *check_root;
 564
 565                 key.objectid = btrfs_header_owner(leaf);
 566                 key.type = BTRFS_ROOT_ITEM_KEY;
 567                 key.offset = (u64)-1;
 568
 569                 check_root = btrfs_get_fs_root(root->fs_info, &key, false);
 570                 /*
 571                  * The only reason we also check NULL here is that during
 572                  * open_ctree() some roots has not yet been set up.
 573                  */
 574                 if (!IS_ERR_OR_NULL(check_root)) {
 575                         /* if leaf is the root, then it's fine */
 576                         if (leaf->start !=
 577                             btrfs_root_bytenr(&check_root->root_item)) {
 578                                 CORRUPT("non-root leaf's nritems is 0",
 579                                         leaf, root, 0);
 580                                 return -EIO;
 581                         }
 582                 }
 583                 return 0;
 584         }
 585
 586         /* Check the 0 item */
 587         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 588             BTRFS_LEAF_DATA_SIZE(root)) {
 589                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 590                 return -EIO;
 591         }
 592
 593         /*
 594          * Check to make sure each items keys are in the correct order and their
 595          * offsets make sense.  We only have to loop through nritems-1 because
 596          * we check the current slot against the next slot, which verifies the
 597          * next slot's offset+size makes sense and that the current's slot
 598          * offset is correct.
 599          */
 600         for (slot = 0; slot < nritems - 1; slot++) {
 601                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 602                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 603
 604                 /* Make sure the keys are in the right order */
 605                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 606                         CORRUPT("bad key order", leaf, root, slot);
 607                         return -EIO;
 608                 }
 609
 610                 /*
 611                  * Make sure the offset and ends are right, remember that the
 612                  * item data starts at the end of the leaf and grows towards the
 613                  * front.
 614                  */
 615                 if (btrfs_item_offset_nr(leaf, slot) !=
 616                         btrfs_item_end_nr(leaf, slot + 1)) {
 617                         CORRUPT("slot offset bad", leaf, root, slot);
 618                         return -EIO;
 619                 }
 620
 621                 /*
 622                  * Check to make sure that we don't point outside of the leaf,
 623                  * just in case all the items are consistent to each other, but
 624                  * all point outside of the leaf.
 625                  */
 626                 if (btrfs_item_end_nr(leaf, slot) >
 627                     BTRFS_LEAF_DATA_SIZE(root)) {
 628                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 629                         return -EIO;
 630                 }
 631         }
 632
 633         return 0;
 634 }
 635
 636 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 637 {
 638         unsigned long nr = btrfs_header_nritems(node);
 639
 640         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root)) {
 641                 btrfs_crit(root->fs_info,
 642                            "corrupt node: block %llu root %llu nritems %lu",
 643                            node->start, root->objectid, nr);
 644                 return -EIO;
 645         }
 646         return 0;
 647 }
 648
 649 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 650                                       u64 phy_offset, struct page *page,
 651                                       u64 start, u64 end, int mirror)
 652 {
 653         u64 found_start;
 654         int found_level;
 655         struct extent_buffer *eb;
 656         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 657         struct btrfs_fs_info *fs_info = root->fs_info;
 658         int ret = 0;
 659         int reads_done;
 660
 661         if (!page->private)
 662                 goto out;
 663
 664         eb = (struct extent_buffer *)page->private;
 665
 666         /* the pending IO might have been the only thing that kept this buffer
 667          * in memory.  Make sure we have a ref for all this other checks
 668          */
 669         extent_buffer_get(eb);
 670
 671         reads_done = atomic_dec_and_test(&eb->io_pages);
 672         if (!reads_done)
 673                 goto err;
 674
 675         eb->read_mirror = mirror;
 676         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 677                 ret = -EIO;
 678                 goto err;
 679         }
 680
 681         found_start = btrfs_header_bytenr(eb);
 682         if (found_start != eb->start) {
 683                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 684                              found_start, eb->start);
 685                 ret = -EIO;
 686                 goto err;
 687         }
 688         if (check_tree_block_fsid(fs_info, eb)) {
 689                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 690                              eb->start);
 691                 ret = -EIO;
 692                 goto err;
 693         }
 694         found_level = btrfs_header_level(eb);
 695         if (found_level >= BTRFS_MAX_LEVEL) {
 696                 btrfs_err(fs_info, "bad tree block level %d",
 697                           (int)btrfs_header_level(eb));
 698                 ret = -EIO;
 699                 goto err;
 700         }
 701
 702         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 703                                        eb, found_level);
 704
 705         ret = csum_tree_block(fs_info, eb, 1);
 706         if (ret)
 707                 goto err;
 708
 709         /*
 710          * If this is a leaf block and it is corrupt, set the corrupt bit so
 711          * that we don't try and read the other copies of this block, just
 712          * return -EIO.
 713          */
 714         if (found_level == 0 && check_leaf(root, eb)) {
 715                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 716                 ret = -EIO;
 717         }
 718
 719         if (found_level > 0 && check_node(root, eb))
 720                 ret = -EIO;
 721
 722         if (!ret)
 723                 set_extent_buffer_uptodate(eb);
 724 err:
 725         if (reads_done &&
 726             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 727                 btree_readahead_hook(fs_info, eb, eb->start, ret);
 728
 729         if (ret) {
 730                 /*
 731                  * our io error hook is going to dec the io pages
 732                  * again, we have to make sure it has something
 733                  * to decrement
 734                  */
 735                 atomic_inc(&eb->io_pages);
 736                 clear_extent_buffer_uptodate(eb);
 737         }
 738         free_extent_buffer(eb);
 739 out:
 740         return ret;
 741 }
 742
 743 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 744 {
 745         struct extent_buffer *eb;
 746
 747         eb = (struct extent_buffer *)page->private;
 748         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 749         eb->read_mirror = failed_mirror;
 750         atomic_dec(&eb->io_pages);
 751         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 752                 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
 753         return -EIO;    /* we fixed nothing */
 754 }
 755
 756 static void end_workqueue_bio(struct bio *bio)
 757 {
 758         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 759         struct btrfs_fs_info *fs_info;
 760         struct btrfs_workqueue *wq;
 761         btrfs_work_func_t func;
 762
 763         fs_info = end_io_wq->info;
 764         end_io_wq->error = bio->bi_error;
 765
 766         if (bio_op(bio) == REQ_OP_WRITE) {
 767                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 768                         wq = fs_info->endio_meta_write_workers;
 769                         func = btrfs_endio_meta_write_helper;
 770                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 771                         wq = fs_info->endio_freespace_worker;
 772                         func = btrfs_freespace_write_helper;
 773                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 774                         wq = fs_info->endio_raid56_workers;
 775                         func = btrfs_endio_raid56_helper;
 776                 } else {
 777                         wq = fs_info->endio_write_workers;
 778                         func = btrfs_endio_write_helper;
 779                 }
 780         } else {
 781                 if (unlikely(end_io_wq->metadata ==
 782                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 783                         wq = fs_info->endio_repair_workers;
 784                         func = btrfs_endio_repair_helper;
 785                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 786                         wq = fs_info->endio_raid56_workers;
 787                         func = btrfs_endio_raid56_helper;
 788                 } else if (end_io_wq->metadata) {
 789                         wq = fs_info->endio_meta_workers;
 790                         func = btrfs_endio_meta_helper;
 791                 } else {
 792                         wq = fs_info->endio_workers;
 793                         func = btrfs_endio_helper;
 794                 }
 795         }
 796
 797         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 798         btrfs_queue_work(wq, &end_io_wq->work);
 799 }
 800
 801 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 802                         enum btrfs_wq_endio_type metadata)
 803 {
 804         struct btrfs_end_io_wq *end_io_wq;
 805
 806         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 807         if (!end_io_wq)
 808                 return -ENOMEM;
 809
 810         end_io_wq->private = bio->bi_private;
 811         end_io_wq->end_io = bio->bi_end_io;
 812         end_io_wq->info = info;
 813         end_io_wq->error = 0;
 814         end_io_wq->bio = bio;
 815         end_io_wq->metadata = metadata;
 816
 817         bio->bi_private = end_io_wq;
 818         bio->bi_end_io = end_workqueue_bio;
 819         return 0;
 820 }
 821
 822 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 823 {
 824         unsigned long limit = min_t(unsigned long,
 825                                     info->thread_pool_size,
 826                                     info->fs_devices->open_devices);
 827         return 256 * limit;
 828 }
 829
 830 static void run_one_async_start(struct btrfs_work *work)
 831 {
 832         struct async_submit_bio *async;
 833         int ret;
 834
 835         async = container_of(work, struct  async_submit_bio, work);
 836         ret = async->submit_bio_start(async->inode, async->bio,
 837                                       async->mirror_num, async->bio_flags,
 838                                       async->bio_offset);
 839         if (ret)
 840                 async->error = ret;
 841 }
 842
 843 static void run_one_async_done(struct btrfs_work *work)
 844 {
 845         struct btrfs_fs_info *fs_info;
 846         struct async_submit_bio *async;
 847         int limit;
 848
 849         async = container_of(work, struct  async_submit_bio, work);
 850         fs_info = BTRFS_I(async->inode)->root->fs_info;
 851
 852         limit = btrfs_async_submit_limit(fs_info);
 853         limit = limit * 2 / 3;
 854
 855         /*
 856          * atomic_dec_return implies a barrier for waitqueue_active
 857          */
 858         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 859             waitqueue_active(&fs_info->async_submit_wait))
 860                 wake_up(&fs_info->async_submit_wait);
 861
 862         /* If an error occurred we just want to clean up the bio and move on */
 863         if (async->error) {
 864                 async->bio->bi_error = async->error;
 865                 bio_endio(async->bio);
 866                 return;
 867         }
 868
 869         async->submit_bio_done(async->inode, async->bio, async->mirror_num,
 870                                async->bio_flags, async->bio_offset);
 871 }
 872
 873 static void run_one_async_free(struct btrfs_work *work)
 874 {
 875         struct async_submit_bio *async;
 876
 877         async = container_of(work, struct  async_submit_bio, work);
 878         kfree(async);
 879 }
 880
 881 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 882                         struct bio *bio, int mirror_num,
 883                         unsigned long bio_flags,
 884                         u64 bio_offset,
 885                         extent_submit_bio_hook_t *submit_bio_start,
 886                         extent_submit_bio_hook_t *submit_bio_done)
 887 {
 888         struct async_submit_bio *async;
 889
 890         async = kmalloc(sizeof(*async), GFP_NOFS);
 891         if (!async)
 892                 return -ENOMEM;
 893
 894         async->inode = inode;
 895         async->bio = bio;
 896         async->mirror_num = mirror_num;
 897         async->submit_bio_start = submit_bio_start;
 898         async->submit_bio_done = submit_bio_done;
 899
 900         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 901                         run_one_async_done, run_one_async_free);
 902
 903         async->bio_flags = bio_flags;
 904         async->bio_offset = bio_offset;
 905
 906         async->error = 0;
 907
 908         atomic_inc(&fs_info->nr_async_submits);
 909
 910         if (bio->bi_opf & REQ_SYNC)
 911                 btrfs_set_work_high_priority(&async->work);
 912
 913         btrfs_queue_work(fs_info->workers, &async->work);
 914
 915         while (atomic_read(&fs_info->async_submit_draining) &&
 916               atomic_read(&fs_info->nr_async_submits)) {
 917                 wait_event(fs_info->async_submit_wait,
 918                            (atomic_read(&fs_info->nr_async_submits) == 0));
 919         }
 920
 921         return 0;
 922 }
 923
 924 static int btree_csum_one_bio(struct bio *bio)
 925 {
 926         struct bio_vec *bvec;
 927         struct btrfs_root *root;
 928         int i, ret = 0;
 929
 930         bio_for_each_segment_all(bvec, bio, i) {
 931                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 932                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 933                 if (ret)
 934                         break;
 935         }
 936
 937         return ret;
 938 }
 939
 940 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
 941                                     int mirror_num, unsigned long bio_flags,
 942                                     u64 bio_offset)
 943 {
 944         /*
 945          * when we're called for a write, we're already in the async
 946          * submission context.  Just jump into btrfs_map_bio
 947          */
 948         return btree_csum_one_bio(bio);
 949 }
 950
 951 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
 952                                  int mirror_num, unsigned long bio_flags,
 953                                  u64 bio_offset)
 954 {
 955         int ret;
 956
 957         /*
 958          * when we're called for a write, we're already in the async
 959          * submission context.  Just jump into btrfs_map_bio
 960          */
 961         ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 1);
 962         if (ret) {
 963                 bio->bi_error = ret;
 964                 bio_endio(bio);
 965         }
 966         return ret;
 967 }
 968
 969 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 970 {
 971         if (bio_flags & EXTENT_BIO_TREE_LOG)
 972                 return 0;
 973 #ifdef CONFIG_X86
 974         if (static_cpu_has(X86_FEATURE_XMM4_2))
 975                 return 0;
 976 #endif
 977         return 1;
 978 }
 979
 980 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 981                                  int mirror_num, unsigned long bio_flags,
 982                                  u64 bio_offset)
 983 {
 984         int async = check_async_write(inode, bio_flags);
 985         int ret;
 986
 987         if (bio_op(bio) != REQ_OP_WRITE) {
 988                 /*
 989                  * called for a read, do the setup so that checksum validation
 990                  * can happen in the async kernel threads
 991                  */
 992                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 993                                           bio, BTRFS_WQ_ENDIO_METADATA);
 994                 if (ret)
 995                         goto out_w_error;
 996                 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
 997         } else if (!async) {
 998                 ret = btree_csum_one_bio(bio);
 999                 if (ret)
1000                         goto out_w_error;
1001                 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
1002         } else {
1003                 /*
1004                  * kthread helpers are used to submit writes so that
1005                  * checksumming can happen in parallel across all CPUs
1006                  */
1007                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1008                                           inode, bio, mirror_num, 0,
1009                                           bio_offset,
1010                                           __btree_submit_bio_start,
1011                                           __btree_submit_bio_done);
1012         }
1013
1014         if (ret)
1015                 goto out_w_error;
1016         return 0;
1017
1018 out_w_error:
1019         bio->bi_error = ret;
1020         bio_endio(bio);
1021         return ret;
1022 }
1023
1024 #ifdef CONFIG_MIGRATION
1025 static int btree_migratepage(struct address_space *mapping,
1026                         struct page *newpage, struct page *page,
1027                         enum migrate_mode mode)
1028 {
1029         /*
1030          * we can't safely write a btree page from here,
1031          * we haven't done the locking hook
1032          */
1033         if (PageDirty(page))
1034                 return -EAGAIN;
1035         /*
1036          * Buffers may be managed in a filesystem specific way.
1037          * We must have no buffers or drop them.
1038          */
1039         if (page_has_private(page) &&
1040             !try_to_release_page(page, GFP_KERNEL))
1041                 return -EAGAIN;
1042         return migrate_page(mapping, newpage, page, mode);
1043 }
1044 #endif
1045
1046
1047 static int btree_writepages(struct address_space *mapping,
1048                             struct writeback_control *wbc)
1049 {
1050         struct btrfs_fs_info *fs_info;
1051         int ret;
1052
1053         if (wbc->sync_mode == WB_SYNC_NONE) {
1054
1055                 if (wbc->for_kupdate)
1056                         return 0;
1057
1058                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1059                 /* this is a bit racy, but that's ok */
1060                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1061                                              BTRFS_DIRTY_METADATA_THRESH);
1062                 if (ret < 0)
1063                         return 0;
1064         }
1065         return btree_write_cache_pages(mapping, wbc);
1066 }
1067
1068 static int btree_readpage(struct file *file, struct page *page)
1069 {
1070         struct extent_io_tree *tree;
1071         tree = &BTRFS_I(page->mapping->host)->io_tree;
1072         return extent_read_full_page(tree, page, btree_get_extent, 0);
1073 }
1074
1075 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1076 {
1077         if (PageWriteback(page) || PageDirty(page))
1078                 return 0;
1079
1080         return try_release_extent_buffer(page);
1081 }
1082
1083 static void btree_invalidatepage(struct page *page, unsigned int offset,
1084                                  unsigned int length)
1085 {
1086         struct extent_io_tree *tree;
1087         tree = &BTRFS_I(page->mapping->host)->io_tree;
1088         extent_invalidatepage(tree, page, offset);
1089         btree_releasepage(page, GFP_NOFS);
1090         if (PagePrivate(page)) {
1091                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1092                            "page private not zero on page %llu",
1093                            (unsigned long long)page_offset(page));
1094                 ClearPagePrivate(page);
1095                 set_page_private(page, 0);
1096                 put_page(page);
1097         }
1098 }
1099
1100 static int btree_set_page_dirty(struct page *page)
1101 {
1102 #ifdef DEBUG
1103         struct extent_buffer *eb;
1104
1105         BUG_ON(!PagePrivate(page));
1106         eb = (struct extent_buffer *)page->private;
1107         BUG_ON(!eb);
1108         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1109         BUG_ON(!atomic_read(&eb->refs));
1110         btrfs_assert_tree_locked(eb);
1111 #endif
1112         return __set_page_dirty_nobuffers(page);
1113 }
1114
1115 static const struct address_space_operations btree_aops = {
1116         .readpage       = btree_readpage,
1117         .writepages     = btree_writepages,
1118         .releasepage    = btree_releasepage,
1119         .invalidatepage = btree_invalidatepage,
1120 #ifdef CONFIG_MIGRATION
1121         .migratepage    = btree_migratepage,
1122 #endif
1123         .set_page_dirty = btree_set_page_dirty,
1124 };
1125
1126 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1127 {
1128         struct extent_buffer *buf = NULL;
1129         struct inode *btree_inode = root->fs_info->btree_inode;
1130
1131         buf = btrfs_find_create_tree_block(root, bytenr);
1132         if (IS_ERR(buf))
1133                 return;
1134         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1135                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1136         free_extent_buffer(buf);
1137 }
1138
1139 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1140                          int mirror_num, struct extent_buffer **eb)
1141 {
1142         struct extent_buffer *buf = NULL;
1143         struct inode *btree_inode = root->fs_info->btree_inode;
1144         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1145         int ret;
1146
1147         buf = btrfs_find_create_tree_block(root, bytenr);
1148         if (IS_ERR(buf))
1149                 return 0;
1150
1151         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1152
1153         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1154                                        btree_get_extent, mirror_num);
1155         if (ret) {
1156                 free_extent_buffer(buf);
1157                 return ret;
1158         }
1159
1160         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1161                 free_extent_buffer(buf);
1162                 return -EIO;
1163         } else if (extent_buffer_uptodate(buf)) {
1164                 *eb = buf;
1165         } else {
1166                 free_extent_buffer(buf);
1167         }
1168         return 0;
1169 }
1170
1171 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1172                                             u64 bytenr)
1173 {
1174         return find_extent_buffer(fs_info, bytenr);
1175 }
1176
1177 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1178                                                  u64 bytenr)
1179 {
1180         if (btrfs_is_testing(root->fs_info))
1181                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1182                                 root->nodesize);
1183         return alloc_extent_buffer(root->fs_info, bytenr);
1184 }
1185
1186
1187 int btrfs_write_tree_block(struct extent_buffer *buf)
1188 {
1189         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1190                                         buf->start + buf->len - 1);
1191 }
1192
1193 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1194 {
1195         return filemap_fdatawait_range(buf->pages[0]->mapping,
1196                                        buf->start, buf->start + buf->len - 1);
1197 }
1198
1199 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1200                                       u64 parent_transid)
1201 {
1202         struct extent_buffer *buf = NULL;
1203         int ret;
1204
1205         buf = btrfs_find_create_tree_block(root, bytenr);
1206         if (IS_ERR(buf))
1207                 return buf;
1208
1209         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1210         if (ret) {
1211                 free_extent_buffer(buf);
1212                 return ERR_PTR(ret);
1213         }
1214         return buf;
1215
1216 }
1217
1218 void clean_tree_block(struct btrfs_trans_handle *trans,
1219                       struct btrfs_fs_info *fs_info,
1220                       struct extent_buffer *buf)
1221 {
1222         if (btrfs_header_generation(buf) ==
1223             fs_info->running_transaction->transid) {
1224                 btrfs_assert_tree_locked(buf);
1225
1226                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1227                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1228                                              -buf->len,
1229                                              fs_info->dirty_metadata_batch);
1230                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1231                         btrfs_set_lock_blocking(buf);
1232                         clear_extent_buffer_dirty(buf);
1233                 }
1234         }
1235 }
1236
1237 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1238 {
1239         struct btrfs_subvolume_writers *writers;
1240         int ret;
1241
1242         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1243         if (!writers)
1244                 return ERR_PTR(-ENOMEM);
1245
1246         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1247         if (ret < 0) {
1248                 kfree(writers);
1249                 return ERR_PTR(ret);
1250         }
1251
1252         init_waitqueue_head(&writers->wait);
1253         return writers;
1254 }
1255
1256 static void
1257 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1258 {
1259         percpu_counter_destroy(&writers->counter);
1260         kfree(writers);
1261 }
1262
1263 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1264                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1265                          u64 objectid)
1266 {
1267         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1268         root->node = NULL;
1269         root->commit_root = NULL;
1270         root->sectorsize = sectorsize;
1271         root->nodesize = nodesize;
1272         root->stripesize = stripesize;
1273         root->state = 0;
1274         root->orphan_cleanup_state = 0;
1275
1276         root->objectid = objectid;
1277         root->last_trans = 0;
1278         root->highest_objectid = 0;
1279         root->nr_delalloc_inodes = 0;
1280         root->nr_ordered_extents = 0;
1281         root->name = NULL;
1282         root->inode_tree = RB_ROOT;
1283         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1284         root->block_rsv = NULL;
1285         root->orphan_block_rsv = NULL;
1286
1287         INIT_LIST_HEAD(&root->dirty_list);
1288         INIT_LIST_HEAD(&root->root_list);
1289         INIT_LIST_HEAD(&root->delalloc_inodes);
1290         INIT_LIST_HEAD(&root->delalloc_root);
1291         INIT_LIST_HEAD(&root->ordered_extents);
1292         INIT_LIST_HEAD(&root->ordered_root);
1293         INIT_LIST_HEAD(&root->logged_list[0]);
1294         INIT_LIST_HEAD(&root->logged_list[1]);
1295         spin_lock_init(&root->orphan_lock);
1296         spin_lock_init(&root->inode_lock);
1297         spin_lock_init(&root->delalloc_lock);
1298         spin_lock_init(&root->ordered_extent_lock);
1299         spin_lock_init(&root->accounting_lock);
1300         spin_lock_init(&root->log_extents_lock[0]);
1301         spin_lock_init(&root->log_extents_lock[1]);
1302         mutex_init(&root->objectid_mutex);
1303         mutex_init(&root->log_mutex);
1304         mutex_init(&root->ordered_extent_mutex);
1305         mutex_init(&root->delalloc_mutex);
1306         init_waitqueue_head(&root->log_writer_wait);
1307         init_waitqueue_head(&root->log_commit_wait[0]);
1308         init_waitqueue_head(&root->log_commit_wait[1]);
1309         INIT_LIST_HEAD(&root->log_ctxs[0]);
1310         INIT_LIST_HEAD(&root->log_ctxs[1]);
1311         atomic_set(&root->log_commit[0], 0);
1312         atomic_set(&root->log_commit[1], 0);
1313         atomic_set(&root->log_writers, 0);
1314         atomic_set(&root->log_batch, 0);
1315         atomic_set(&root->orphan_inodes, 0);
1316         atomic_set(&root->refs, 1);
1317         atomic_set(&root->will_be_snapshoted, 0);
1318         atomic_set(&root->qgroup_meta_rsv, 0);
1319         root->log_transid = 0;
1320         root->log_transid_committed = -1;
1321         root->last_log_commit = 0;
1322         if (!dummy)
1323                 extent_io_tree_init(&root->dirty_log_pages,
1324                                      fs_info->btree_inode->i_mapping);
1325
1326         memset(&root->root_key, 0, sizeof(root->root_key));
1327         memset(&root->root_item, 0, sizeof(root->root_item));
1328         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1329         if (!dummy)
1330                 root->defrag_trans_start = fs_info->generation;
1331         else
1332                 root->defrag_trans_start = 0;
1333         root->root_key.objectid = objectid;
1334         root->anon_dev = 0;
1335
1336         spin_lock_init(&root->root_item_lock);
1337 }
1338
1339 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1340                 gfp_t flags)
1341 {
1342         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1343         if (root)
1344                 root->fs_info = fs_info;
1345         return root;
1346 }
1347
1348 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1349 /* Should only be used by the testing infrastructure */
1350 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1351                                           u32 sectorsize, u32 nodesize)
1352 {
1353         struct btrfs_root *root;
1354
1355         if (!fs_info)
1356                 return ERR_PTR(-EINVAL);
1357
1358         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1359         if (!root)
1360                 return ERR_PTR(-ENOMEM);
1361         /* We don't use the stripesize in selftest, set it as sectorsize */
1362         __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1363                         BTRFS_ROOT_TREE_OBJECTID);
1364         root->alloc_bytenr = 0;
1365
1366         return root;
1367 }
1368 #endif
1369
1370 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1371                                      struct btrfs_fs_info *fs_info,
1372                                      u64 objectid)
1373 {
1374         struct extent_buffer *leaf;
1375         struct btrfs_root *tree_root = fs_info->tree_root;
1376         struct btrfs_root *root;
1377         struct btrfs_key key;
1378         int ret = 0;
1379         uuid_le uuid;
1380
1381         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1382         if (!root)
1383                 return ERR_PTR(-ENOMEM);
1384
1385         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1386                 tree_root->stripesize, root, fs_info, objectid);
1387         root->root_key.objectid = objectid;
1388         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1389         root->root_key.offset = 0;
1390
1391         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1392         if (IS_ERR(leaf)) {
1393                 ret = PTR_ERR(leaf);
1394                 leaf = NULL;
1395                 goto fail;
1396         }
1397
1398         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1399         btrfs_set_header_bytenr(leaf, leaf->start);
1400         btrfs_set_header_generation(leaf, trans->transid);
1401         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1402         btrfs_set_header_owner(leaf, objectid);
1403         root->node = leaf;
1404
1405         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1406                             BTRFS_FSID_SIZE);
1407         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1408                             btrfs_header_chunk_tree_uuid(leaf),
1409                             BTRFS_UUID_SIZE);
1410         btrfs_mark_buffer_dirty(leaf);
1411
1412         root->commit_root = btrfs_root_node(root);
1413         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1414
1415         root->root_item.flags = 0;
1416         root->root_item.byte_limit = 0;
1417         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1418         btrfs_set_root_generation(&root->root_item, trans->transid);
1419         btrfs_set_root_level(&root->root_item, 0);
1420         btrfs_set_root_refs(&root->root_item, 1);
1421         btrfs_set_root_used(&root->root_item, leaf->len);
1422         btrfs_set_root_last_snapshot(&root->root_item, 0);
1423         btrfs_set_root_dirid(&root->root_item, 0);
1424         uuid_le_gen(&uuid);
1425         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1426         root->root_item.drop_level = 0;
1427
1428         key.objectid = objectid;
1429         key.type = BTRFS_ROOT_ITEM_KEY;
1430         key.offset = 0;
1431         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1432         if (ret)
1433                 goto fail;
1434
1435         btrfs_tree_unlock(leaf);
1436
1437         return root;
1438
1439 fail:
1440         if (leaf) {
1441                 btrfs_tree_unlock(leaf);
1442                 free_extent_buffer(root->commit_root);
1443                 free_extent_buffer(leaf);
1444         }
1445         kfree(root);
1446
1447         return ERR_PTR(ret);
1448 }
1449
1450 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1451                                          struct btrfs_fs_info *fs_info)
1452 {
1453         struct btrfs_root *root;
1454         struct btrfs_root *tree_root = fs_info->tree_root;
1455         struct extent_buffer *leaf;
1456
1457         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1458         if (!root)
1459                 return ERR_PTR(-ENOMEM);
1460
1461         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1462                      tree_root->stripesize, root, fs_info,
1463                      BTRFS_TREE_LOG_OBJECTID);
1464
1465         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1466         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1467         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1468
1469         /*
1470          * DON'T set REF_COWS for log trees
1471          *
1472          * log trees do not get reference counted because they go away
1473          * before a real commit is actually done.  They do store pointers
1474          * to file data extents, and those reference counts still get
1475          * updated (along with back refs to the log tree).
1476          */
1477
1478         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1479                         NULL, 0, 0, 0);
1480         if (IS_ERR(leaf)) {
1481                 kfree(root);
1482                 return ERR_CAST(leaf);
1483         }
1484
1485         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1486         btrfs_set_header_bytenr(leaf, leaf->start);
1487         btrfs_set_header_generation(leaf, trans->transid);
1488         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1489         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1490         root->node = leaf;
1491
1492         write_extent_buffer(root->node, root->fs_info->fsid,
1493                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1494         btrfs_mark_buffer_dirty(root->node);
1495         btrfs_tree_unlock(root->node);
1496         return root;
1497 }
1498
1499 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1500                              struct btrfs_fs_info *fs_info)
1501 {
1502         struct btrfs_root *log_root;
1503
1504         log_root = alloc_log_tree(trans, fs_info);
1505         if (IS_ERR(log_root))
1506                 return PTR_ERR(log_root);
1507         WARN_ON(fs_info->log_root_tree);
1508         fs_info->log_root_tree = log_root;
1509         return 0;
1510 }
1511
1512 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1513                        struct btrfs_root *root)
1514 {
1515         struct btrfs_root *log_root;
1516         struct btrfs_inode_item *inode_item;
1517
1518         log_root = alloc_log_tree(trans, root->fs_info);
1519         if (IS_ERR(log_root))
1520                 return PTR_ERR(log_root);
1521
1522         log_root->last_trans = trans->transid;
1523         log_root->root_key.offset = root->root_key.objectid;
1524
1525         inode_item = &log_root->root_item.inode;
1526         btrfs_set_stack_inode_generation(inode_item, 1);
1527         btrfs_set_stack_inode_size(inode_item, 3);
1528         btrfs_set_stack_inode_nlink(inode_item, 1);
1529         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1530         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1531
1532         btrfs_set_root_node(&log_root->root_item, log_root->node);
1533
1534         WARN_ON(root->log_root);
1535         root->log_root = log_root;
1536         root->log_transid = 0;
1537         root->log_transid_committed = -1;
1538         root->last_log_commit = 0;
1539         return 0;
1540 }
1541
1542 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1543                                                struct btrfs_key *key)
1544 {
1545         struct btrfs_root *root;
1546         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1547         struct btrfs_path *path;
1548         u64 generation;
1549         int ret;
1550
1551         path = btrfs_alloc_path();
1552         if (!path)
1553                 return ERR_PTR(-ENOMEM);
1554
1555         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1556         if (!root) {
1557                 ret = -ENOMEM;
1558                 goto alloc_fail;
1559         }
1560
1561         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1562                 tree_root->stripesize, root, fs_info, key->objectid);
1563
1564         ret = btrfs_find_root(tree_root, key, path,
1565                               &root->root_item, &root->root_key);
1566         if (ret) {
1567                 if (ret > 0)
1568                         ret = -ENOENT;
1569                 goto find_fail;
1570         }
1571
1572         generation = btrfs_root_generation(&root->root_item);
1573         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1574                                      generation);
1575         if (IS_ERR(root->node)) {
1576                 ret = PTR_ERR(root->node);
1577                 goto find_fail;
1578         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1579                 ret = -EIO;
1580                 free_extent_buffer(root->node);
1581                 goto find_fail;
1582         }
1583         root->commit_root = btrfs_root_node(root);
1584 out:
1585         btrfs_free_path(path);
1586         return root;
1587
1588 find_fail:
1589         kfree(root);
1590 alloc_fail:
1591         root = ERR_PTR(ret);
1592         goto out;
1593 }
1594
1595 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1596                                       struct btrfs_key *location)
1597 {
1598         struct btrfs_root *root;
1599
1600         root = btrfs_read_tree_root(tree_root, location);
1601         if (IS_ERR(root))
1602                 return root;
1603
1604         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1605                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1606                 btrfs_check_and_init_root_item(&root->root_item);
1607         }
1608
1609         return root;
1610 }
1611
1612 int btrfs_init_fs_root(struct btrfs_root *root)
1613 {
1614         int ret;
1615         struct btrfs_subvolume_writers *writers;
1616
1617         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1618         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1619                                         GFP_NOFS);
1620         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1621                 ret = -ENOMEM;
1622                 goto fail;
1623         }
1624
1625         writers = btrfs_alloc_subvolume_writers();
1626         if (IS_ERR(writers)) {
1627                 ret = PTR_ERR(writers);
1628                 goto fail;
1629         }
1630         root->subv_writers = writers;
1631
1632         btrfs_init_free_ino_ctl(root);
1633         spin_lock_init(&root->ino_cache_lock);
1634         init_waitqueue_head(&root->ino_cache_wait);
1635
1636         ret = get_anon_bdev(&root->anon_dev);
1637         if (ret)
1638                 goto fail;
1639
1640         mutex_lock(&root->objectid_mutex);
1641         ret = btrfs_find_highest_objectid(root,
1642                                         &root->highest_objectid);
1643         if (ret) {
1644                 mutex_unlock(&root->objectid_mutex);
1645                 goto fail;
1646         }
1647
1648         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1649
1650         mutex_unlock(&root->objectid_mutex);
1651
1652         return 0;
1653 fail:
1654         /* the caller is responsible to call free_fs_root */
1655         return ret;
1656 }
1657
1658 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1659                                         u64 root_id)
1660 {
1661         struct btrfs_root *root;
1662
1663         spin_lock(&fs_info->fs_roots_radix_lock);
1664         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1665                                  (unsigned long)root_id);
1666         spin_unlock(&fs_info->fs_roots_radix_lock);
1667         return root;
1668 }
1669
1670 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1671                          struct btrfs_root *root)
1672 {
1673         int ret;
1674
1675         ret = radix_tree_preload(GFP_NOFS);
1676         if (ret)
1677                 return ret;
1678
1679         spin_lock(&fs_info->fs_roots_radix_lock);
1680         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1681                                 (unsigned long)root->root_key.objectid,
1682                                 root);
1683         if (ret == 0)
1684                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1685         spin_unlock(&fs_info->fs_roots_radix_lock);
1686         radix_tree_preload_end();
1687
1688         return ret;
1689 }
1690
1691 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1692                                      struct btrfs_key *location,
1693                                      bool check_ref)
1694 {
1695         struct btrfs_root *root;
1696         struct btrfs_path *path;
1697         struct btrfs_key key;
1698         int ret;
1699
1700         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1701                 return fs_info->tree_root;
1702         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1703                 return fs_info->extent_root;
1704         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1705                 return fs_info->chunk_root;
1706         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1707                 return fs_info->dev_root;
1708         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1709                 return fs_info->csum_root;
1710         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1711                 return fs_info->quota_root ? fs_info->quota_root :
1712                                              ERR_PTR(-ENOENT);
1713         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1714                 return fs_info->uuid_root ? fs_info->uuid_root :
1715                                             ERR_PTR(-ENOENT);
1716         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1717                 return fs_info->free_space_root ? fs_info->free_space_root :
1718                                                   ERR_PTR(-ENOENT);
1719 again:
1720         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1721         if (root) {
1722                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1723                         return ERR_PTR(-ENOENT);
1724                 return root;
1725         }
1726
1727         root = btrfs_read_fs_root(fs_info->tree_root, location);
1728         if (IS_ERR(root))
1729                 return root;
1730
1731         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1732                 ret = -ENOENT;
1733                 goto fail;
1734         }
1735
1736         ret = btrfs_init_fs_root(root);
1737         if (ret)
1738                 goto fail;
1739
1740         path = btrfs_alloc_path();
1741         if (!path) {
1742                 ret = -ENOMEM;
1743                 goto fail;
1744         }
1745         key.objectid = BTRFS_ORPHAN_OBJECTID;
1746         key.type = BTRFS_ORPHAN_ITEM_KEY;
1747         key.offset = location->objectid;
1748
1749         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1750         btrfs_free_path(path);
1751         if (ret < 0)
1752                 goto fail;
1753         if (ret == 0)
1754                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1755
1756         ret = btrfs_insert_fs_root(fs_info, root);
1757         if (ret) {
1758                 if (ret == -EEXIST) {
1759                         free_fs_root(root);
1760                         goto again;
1761                 }
1762                 goto fail;
1763         }
1764         return root;
1765 fail:
1766         free_fs_root(root);
1767         return ERR_PTR(ret);
1768 }
1769
1770 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1771 {
1772         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1773         int ret = 0;
1774         struct btrfs_device *device;
1775         struct backing_dev_info *bdi;
1776
1777         rcu_read_lock();
1778         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1779                 if (!device->bdev)
1780                         continue;
1781                 bdi = blk_get_backing_dev_info(device->bdev);
1782                 if (bdi_congested(bdi, bdi_bits)) {
1783                         ret = 1;
1784                         break;
1785                 }
1786         }
1787         rcu_read_unlock();
1788         return ret;
1789 }
1790
1791 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1792 {
1793         int err;
1794
1795         err = bdi_setup_and_register(bdi, "btrfs");
1796         if (err)
1797                 return err;
1798
1799         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1800         bdi->congested_fn       = btrfs_congested_fn;
1801         bdi->congested_data     = info;
1802         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1803         return 0;
1804 }
1805
1806 /*
1807  * called by the kthread helper functions to finally call the bio end_io
1808  * functions.  This is where read checksum verification actually happens
1809  */
1810 static void end_workqueue_fn(struct btrfs_work *work)
1811 {
1812         struct bio *bio;
1813         struct btrfs_end_io_wq *end_io_wq;
1814
1815         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1816         bio = end_io_wq->bio;
1817
1818         bio->bi_error = end_io_wq->error;
1819         bio->bi_private = end_io_wq->private;
1820         bio->bi_end_io = end_io_wq->end_io;
1821         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1822         bio_endio(bio);
1823 }
1824
1825 static int cleaner_kthread(void *arg)
1826 {
1827         struct btrfs_root *root = arg;
1828         int again;
1829         struct btrfs_trans_handle *trans;
1830
1831         do {
1832                 again = 0;
1833
1834                 /* Make the cleaner go to sleep early. */
1835                 if (btrfs_need_cleaner_sleep(root))
1836                         goto sleep;
1837
1838                 /*
1839                  * Do not do anything if we might cause open_ctree() to block
1840                  * before we have finished mounting the filesystem.
1841                  */
1842                 if (!root->fs_info->open)
1843                         goto sleep;
1844
1845                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1846                         goto sleep;
1847
1848                 /*
1849                  * Avoid the problem that we change the status of the fs
1850                  * during the above check and trylock.
1851                  */
1852                 if (btrfs_need_cleaner_sleep(root)) {
1853                         mutex_unlock(&root->fs_info->cleaner_mutex);
1854                         goto sleep;
1855                 }
1856
1857                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1858                 btrfs_run_delayed_iputs(root);
1859                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1860
1861                 again = btrfs_clean_one_deleted_snapshot(root);
1862                 mutex_unlock(&root->fs_info->cleaner_mutex);
1863
1864                 /*
1865                  * The defragger has dealt with the R/O remount and umount,
1866                  * needn't do anything special here.
1867                  */
1868                 btrfs_run_defrag_inodes(root->fs_info);
1869
1870                 /*
1871                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1872                  * with relocation (btrfs_relocate_chunk) and relocation
1873                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1874                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1875                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1876                  * unused block groups.
1877                  */
1878                 btrfs_delete_unused_bgs(root->fs_info);
1879 sleep:
1880                 if (!again) {
1881                         set_current_state(TASK_INTERRUPTIBLE);
1882                         if (!kthread_should_stop())
1883                                 schedule();
1884                         __set_current_state(TASK_RUNNING);
1885                 }
1886         } while (!kthread_should_stop());
1887
1888         /*
1889          * Transaction kthread is stopped before us and wakes us up.
1890          * However we might have started a new transaction and COWed some
1891          * tree blocks when deleting unused block groups for example. So
1892          * make sure we commit the transaction we started to have a clean
1893          * shutdown when evicting the btree inode - if it has dirty pages
1894          * when we do the final iput() on it, eviction will trigger a
1895          * writeback for it which will fail with null pointer dereferences
1896          * since work queues and other resources were already released and
1897          * destroyed by the time the iput/eviction/writeback is made.
1898          */
1899         trans = btrfs_attach_transaction(root);
1900         if (IS_ERR(trans)) {
1901                 if (PTR_ERR(trans) != -ENOENT)
1902                         btrfs_err(root->fs_info,
1903                                   "cleaner transaction attach returned %ld",
1904                                   PTR_ERR(trans));
1905         } else {
1906                 int ret;
1907
1908                 ret = btrfs_commit_transaction(trans, root);
1909                 if (ret)
1910                         btrfs_err(root->fs_info,
1911                                   "cleaner open transaction commit returned %d",
1912                                   ret);
1913         }
1914
1915         return 0;
1916 }
1917
1918 static int transaction_kthread(void *arg)
1919 {
1920         struct btrfs_root *root = arg;
1921         struct btrfs_trans_handle *trans;
1922         struct btrfs_transaction *cur;
1923         u64 transid;
1924         unsigned long now;
1925         unsigned long delay;
1926         bool cannot_commit;
1927
1928         do {
1929                 cannot_commit = false;
1930                 delay = HZ * root->fs_info->commit_interval;
1931                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1932
1933                 spin_lock(&root->fs_info->trans_lock);
1934                 cur = root->fs_info->running_transaction;
1935                 if (!cur) {
1936                         spin_unlock(&root->fs_info->trans_lock);
1937                         goto sleep;
1938                 }
1939
1940                 now = get_seconds();
1941                 if (cur->state < TRANS_STATE_BLOCKED &&
1942                     (now < cur->start_time ||
1943                      now - cur->start_time < root->fs_info->commit_interval)) {
1944                         spin_unlock(&root->fs_info->trans_lock);
1945                         delay = HZ * 5;
1946                         goto sleep;
1947                 }
1948                 transid = cur->transid;
1949                 spin_unlock(&root->fs_info->trans_lock);
1950
1951                 /* If the file system is aborted, this will always fail. */
1952                 trans = btrfs_attach_transaction(root);
1953                 if (IS_ERR(trans)) {
1954                         if (PTR_ERR(trans) != -ENOENT)
1955                                 cannot_commit = true;
1956                         goto sleep;
1957                 }
1958                 if (transid == trans->transid) {
1959                         btrfs_commit_transaction(trans, root);
1960                 } else {
1961                         btrfs_end_transaction(trans, root);
1962                 }
1963 sleep:
1964                 wake_up_process(root->fs_info->cleaner_kthread);
1965                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1966
1967                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1968                                       &root->fs_info->fs_state)))
1969                         btrfs_cleanup_transaction(root);
1970                 set_current_state(TASK_INTERRUPTIBLE);
1971                 if (!kthread_should_stop() &&
1972                                 (!btrfs_transaction_blocked(root->fs_info) ||
1973                                  cannot_commit))
1974                         schedule_timeout(delay);
1975                 __set_current_state(TASK_RUNNING);
1976         } while (!kthread_should_stop());
1977         return 0;
1978 }
1979
1980 /*
1981  * this will find the highest generation in the array of
1982  * root backups.  The index of the highest array is returned,
1983  * or -1 if we can't find anything.
1984  *
1985  * We check to make sure the array is valid by comparing the
1986  * generation of the latest  root in the array with the generation
1987  * in the super block.  If they don't match we pitch it.
1988  */
1989 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1990 {
1991         u64 cur;
1992         int newest_index = -1;
1993         struct btrfs_root_backup *root_backup;
1994         int i;
1995
1996         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1997                 root_backup = info->super_copy->super_roots + i;
1998                 cur = btrfs_backup_tree_root_gen(root_backup);
1999                 if (cur == newest_gen)
2000                         newest_index = i;
2001         }
2002
2003         /* check to see if we actually wrapped around */
2004         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2005                 root_backup = info->super_copy->super_roots;
2006                 cur = btrfs_backup_tree_root_gen(root_backup);
2007                 if (cur == newest_gen)
2008                         newest_index = 0;
2009         }
2010         return newest_index;
2011 }
2012
2013
2014 /*
2015  * find the oldest backup so we know where to store new entries
2016  * in the backup array.  This will set the backup_root_index
2017  * field in the fs_info struct
2018  */
2019 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2020                                      u64 newest_gen)
2021 {
2022         int newest_index = -1;
2023
2024         newest_index = find_newest_super_backup(info, newest_gen);
2025         /* if there was garbage in there, just move along */
2026         if (newest_index == -1) {
2027                 info->backup_root_index = 0;
2028         } else {
2029                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2030         }
2031 }
2032
2033 /*
2034  * copy all the root pointers into the super backup array.
2035  * this will bump the backup pointer by one when it is
2036  * done
2037  */
2038 static void backup_super_roots(struct btrfs_fs_info *info)
2039 {
2040         int next_backup;
2041         struct btrfs_root_backup *root_backup;
2042         int last_backup;
2043
2044         next_backup = info->backup_root_index;
2045         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2046                 BTRFS_NUM_BACKUP_ROOTS;
2047
2048         /*
2049          * just overwrite the last backup if we're at the same generation
2050          * this happens only at umount
2051          */
2052         root_backup = info->super_for_commit->super_roots + last_backup;
2053         if (btrfs_backup_tree_root_gen(root_backup) ==
2054             btrfs_header_generation(info->tree_root->node))
2055                 next_backup = last_backup;
2056
2057         root_backup = info->super_for_commit->super_roots + next_backup;
2058
2059         /*
2060          * make sure all of our padding and empty slots get zero filled
2061          * regardless of which ones we use today
2062          */
2063         memset(root_backup, 0, sizeof(*root_backup));
2064
2065         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2066
2067         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2068         btrfs_set_backup_tree_root_gen(root_backup,
2069                                btrfs_header_generation(info->tree_root->node));
2070
2071         btrfs_set_backup_tree_root_level(root_backup,
2072                                btrfs_header_level(info->tree_root->node));
2073
2074         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2075         btrfs_set_backup_chunk_root_gen(root_backup,
2076                                btrfs_header_generation(info->chunk_root->node));
2077         btrfs_set_backup_chunk_root_level(root_backup,
2078                                btrfs_header_level(info->chunk_root->node));
2079
2080         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2081         btrfs_set_backup_extent_root_gen(root_backup,
2082                                btrfs_header_generation(info->extent_root->node));
2083         btrfs_set_backup_extent_root_level(root_backup,
2084                                btrfs_header_level(info->extent_root->node));
2085
2086         /*
2087          * we might commit during log recovery, which happens before we set
2088          * the fs_root.  Make sure it is valid before we fill it in.
2089          */
2090         if (info->fs_root && info->fs_root->node) {
2091                 btrfs_set_backup_fs_root(root_backup,
2092                                          info->fs_root->node->start);
2093                 btrfs_set_backup_fs_root_gen(root_backup,
2094                                btrfs_header_generation(info->fs_root->node));
2095                 btrfs_set_backup_fs_root_level(root_backup,
2096                                btrfs_header_level(info->fs_root->node));
2097         }
2098
2099         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2100         btrfs_set_backup_dev_root_gen(root_backup,
2101                                btrfs_header_generation(info->dev_root->node));
2102         btrfs_set_backup_dev_root_level(root_backup,
2103                                        btrfs_header_level(info->dev_root->node));
2104
2105         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2106         btrfs_set_backup_csum_root_gen(root_backup,
2107                                btrfs_header_generation(info->csum_root->node));
2108         btrfs_set_backup_csum_root_level(root_backup,
2109                                btrfs_header_level(info->csum_root->node));
2110
2111         btrfs_set_backup_total_bytes(root_backup,
2112                              btrfs_super_total_bytes(info->super_copy));
2113         btrfs_set_backup_bytes_used(root_backup,
2114                              btrfs_super_bytes_used(info->super_copy));
2115         btrfs_set_backup_num_devices(root_backup,
2116                              btrfs_super_num_devices(info->super_copy));
2117
2118         /*
2119          * if we don't copy this out to the super_copy, it won't get remembered
2120          * for the next commit
2121          */
2122         memcpy(&info->super_copy->super_roots,
2123                &info->super_for_commit->super_roots,
2124                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2125 }
2126
2127 /*
2128  * this copies info out of the root backup array and back into
2129  * the in-memory super block.  It is meant to help iterate through
2130  * the array, so you send it the number of backups you've already
2131  * tried and the last backup index you used.
2132  *
2133  * this returns -1 when it has tried all the backups
2134  */
2135 static noinline int next_root_backup(struct btrfs_fs_info *info,
2136                                      struct btrfs_super_block *super,
2137                                      int *num_backups_tried, int *backup_index)
2138 {
2139         struct btrfs_root_backup *root_backup;
2140         int newest = *backup_index;
2141
2142         if (*num_backups_tried == 0) {
2143                 u64 gen = btrfs_super_generation(super);
2144
2145                 newest = find_newest_super_backup(info, gen);
2146                 if (newest == -1)
2147                         return -1;
2148
2149                 *backup_index = newest;
2150                 *num_backups_tried = 1;
2151         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2152                 /* we've tried all the backups, all done */
2153                 return -1;
2154         } else {
2155                 /* jump to the next oldest backup */
2156                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2157                         BTRFS_NUM_BACKUP_ROOTS;
2158                 *backup_index = newest;
2159                 *num_backups_tried += 1;
2160         }
2161         root_backup = super->super_roots + newest;
2162
2163         btrfs_set_super_generation(super,
2164                                    btrfs_backup_tree_root_gen(root_backup));
2165         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2166         btrfs_set_super_root_level(super,
2167                                    btrfs_backup_tree_root_level(root_backup));
2168         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2169
2170         /*
2171          * fixme: the total bytes and num_devices need to match or we should
2172          * need a fsck
2173          */
2174         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2175         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2176         return 0;
2177 }
2178
2179 /* helper to cleanup workers */
2180 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2181 {
2182         btrfs_destroy_workqueue(fs_info->fixup_workers);
2183         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2184         btrfs_destroy_workqueue(fs_info->workers);
2185         btrfs_destroy_workqueue(fs_info->endio_workers);
2186         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2187         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2188         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2189         btrfs_destroy_workqueue(fs_info->rmw_workers);
2190         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2191         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2192         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2193         btrfs_destroy_workqueue(fs_info->submit_workers);
2194         btrfs_destroy_workqueue(fs_info->delayed_workers);
2195         btrfs_destroy_workqueue(fs_info->caching_workers);
2196         btrfs_destroy_workqueue(fs_info->readahead_workers);
2197         btrfs_destroy_workqueue(fs_info->flush_workers);
2198         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2199         btrfs_destroy_workqueue(fs_info->extent_workers);
2200 }
2201
2202 static void free_root_extent_buffers(struct btrfs_root *root)
2203 {
2204         if (root) {
2205                 free_extent_buffer(root->node);
2206                 free_extent_buffer(root->commit_root);
2207                 root->node = NULL;
2208                 root->commit_root = NULL;
2209         }
2210 }
2211
2212 /* helper to cleanup tree roots */
2213 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2214 {
2215         free_root_extent_buffers(info->tree_root);
2216
2217         free_root_extent_buffers(info->dev_root);
2218         free_root_extent_buffers(info->extent_root);
2219         free_root_extent_buffers(info->csum_root);
2220         free_root_extent_buffers(info->quota_root);
2221         free_root_extent_buffers(info->uuid_root);
2222         if (chunk_root)
2223                 free_root_extent_buffers(info->chunk_root);
2224         free_root_extent_buffers(info->free_space_root);
2225 }
2226
2227 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2228 {
2229         int ret;
2230         struct btrfs_root *gang[8];
2231         int i;
2232
2233         while (!list_empty(&fs_info->dead_roots)) {
2234                 gang[0] = list_entry(fs_info->dead_roots.next,
2235                                      struct btrfs_root, root_list);
2236                 list_del(&gang[0]->root_list);
2237
2238                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2239                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2240                 } else {
2241                         free_extent_buffer(gang[0]->node);
2242                         free_extent_buffer(gang[0]->commit_root);
2243                         btrfs_put_fs_root(gang[0]);
2244                 }
2245         }
2246
2247         while (1) {
2248                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2249                                              (void **)gang, 0,
2250                                              ARRAY_SIZE(gang));
2251                 if (!ret)
2252                         break;
2253                 for (i = 0; i < ret; i++)
2254                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2255         }
2256
2257         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2258                 btrfs_free_log_root_tree(NULL, fs_info);
2259                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2260                                             fs_info->pinned_extents);
2261         }
2262 }
2263
2264 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2265 {
2266         mutex_init(&fs_info->scrub_lock);
2267         atomic_set(&fs_info->scrubs_running, 0);
2268         atomic_set(&fs_info->scrub_pause_req, 0);
2269         atomic_set(&fs_info->scrubs_paused, 0);
2270         atomic_set(&fs_info->scrub_cancel_req, 0);
2271         init_waitqueue_head(&fs_info->scrub_pause_wait);
2272         fs_info->scrub_workers_refcnt = 0;
2273 }
2274
2275 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2276 {
2277         spin_lock_init(&fs_info->balance_lock);
2278         mutex_init(&fs_info->balance_mutex);
2279         atomic_set(&fs_info->balance_running, 0);
2280         atomic_set(&fs_info->balance_pause_req, 0);
2281         atomic_set(&fs_info->balance_cancel_req, 0);
2282         fs_info->balance_ctl = NULL;
2283         init_waitqueue_head(&fs_info->balance_wait_q);
2284 }
2285
2286 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2287                                    struct btrfs_root *tree_root)
2288 {
2289         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2290         set_nlink(fs_info->btree_inode, 1);
2291         /*
2292          * we set the i_size on the btree inode to the max possible int.
2293          * the real end of the address space is determined by all of
2294          * the devices in the system
2295          */
2296         fs_info->btree_inode->i_size = OFFSET_MAX;
2297         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2298
2299         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2300         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2301                              fs_info->btree_inode->i_mapping);
2302         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2303         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2304
2305         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2306
2307         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2308         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2309                sizeof(struct btrfs_key));
2310         set_bit(BTRFS_INODE_DUMMY,
2311                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2312         btrfs_insert_inode_hash(fs_info->btree_inode);
2313 }
2314
2315 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2316 {
2317         fs_info->dev_replace.lock_owner = 0;
2318         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2319         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2320         rwlock_init(&fs_info->dev_replace.lock);
2321         atomic_set(&fs_info->dev_replace.read_locks, 0);
2322         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2323         init_waitqueue_head(&fs_info->replace_wait);
2324         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2325 }
2326
2327 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2328 {
2329         spin_lock_init(&fs_info->qgroup_lock);
2330         mutex_init(&fs_info->qgroup_ioctl_lock);
2331         fs_info->qgroup_tree = RB_ROOT;
2332         fs_info->qgroup_op_tree = RB_ROOT;
2333         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2334         fs_info->qgroup_seq = 1;
2335         fs_info->quota_enabled = 0;
2336         fs_info->pending_quota_state = 0;
2337         fs_info->qgroup_ulist = NULL;
2338         fs_info->qgroup_rescan_running = false;
2339         mutex_init(&fs_info->qgroup_rescan_lock);
2340 }
2341
2342 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2343                 struct btrfs_fs_devices *fs_devices)
2344 {
2345         int max_active = fs_info->thread_pool_size;
2346         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2347
2348         fs_info->workers =
2349                 btrfs_alloc_workqueue(fs_info, "worker",
2350                                       flags | WQ_HIGHPRI, max_active, 16);
2351
2352         fs_info->delalloc_workers =
2353                 btrfs_alloc_workqueue(fs_info, "delalloc",
2354                                       flags, max_active, 2);
2355
2356         fs_info->flush_workers =
2357                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2358                                       flags, max_active, 0);
2359
2360         fs_info->caching_workers =
2361                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2362
2363         /*
2364          * a higher idle thresh on the submit workers makes it much more
2365          * likely that bios will be send down in a sane order to the
2366          * devices
2367          */
2368         fs_info->submit_workers =
2369                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2370                                       min_t(u64, fs_devices->num_devices,
2371                                             max_active), 64);
2372
2373         fs_info->fixup_workers =
2374                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2375
2376         /*
2377          * endios are largely parallel and should have a very
2378          * low idle thresh
2379          */
2380         fs_info->endio_workers =
2381                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2382         fs_info->endio_meta_workers =
2383                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2384                                       max_active, 4);
2385         fs_info->endio_meta_write_workers =
2386                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2387                                       max_active, 2);
2388         fs_info->endio_raid56_workers =
2389                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2390                                       max_active, 4);
2391         fs_info->endio_repair_workers =
2392                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2393         fs_info->rmw_workers =
2394                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2395         fs_info->endio_write_workers =
2396                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2397                                       max_active, 2);
2398         fs_info->endio_freespace_worker =
2399                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2400                                       max_active, 0);
2401         fs_info->delayed_workers =
2402                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2403                                       max_active, 0);
2404         fs_info->readahead_workers =
2405                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2406                                       max_active, 2);
2407         fs_info->qgroup_rescan_workers =
2408                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2409         fs_info->extent_workers =
2410                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2411                                       min_t(u64, fs_devices->num_devices,
2412                                             max_active), 8);
2413
2414         if (!(fs_info->workers && fs_info->delalloc_workers &&
2415               fs_info->submit_workers && fs_info->flush_workers &&
2416               fs_info->endio_workers && fs_info->endio_meta_workers &&
2417               fs_info->endio_meta_write_workers &&
2418               fs_info->endio_repair_workers &&
2419               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2420               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2421               fs_info->caching_workers && fs_info->readahead_workers &&
2422               fs_info->fixup_workers && fs_info->delayed_workers &&
2423               fs_info->extent_workers &&
2424               fs_info->qgroup_rescan_workers)) {
2425                 return -ENOMEM;
2426         }
2427
2428         return 0;
2429 }
2430
2431 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2432                             struct btrfs_fs_devices *fs_devices)
2433 {
2434         int ret;
2435         struct btrfs_root *tree_root = fs_info->tree_root;
2436         struct btrfs_root *log_tree_root;
2437         struct btrfs_super_block *disk_super = fs_info->super_copy;
2438         u64 bytenr = btrfs_super_log_root(disk_super);
2439
2440         if (fs_devices->rw_devices == 0) {
2441                 btrfs_warn(fs_info, "log replay required on RO media");
2442                 return -EIO;
2443         }
2444
2445         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2446         if (!log_tree_root)
2447                 return -ENOMEM;
2448
2449         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2450                         tree_root->stripesize, log_tree_root, fs_info,
2451                         BTRFS_TREE_LOG_OBJECTID);
2452
2453         log_tree_root->node = read_tree_block(tree_root, bytenr,
2454                         fs_info->generation + 1);
2455         if (IS_ERR(log_tree_root->node)) {
2456                 btrfs_warn(fs_info, "failed to read log tree");
2457                 ret = PTR_ERR(log_tree_root->node);
2458                 kfree(log_tree_root);
2459                 return ret;
2460         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2461                 btrfs_err(fs_info, "failed to read log tree");
2462                 free_extent_buffer(log_tree_root->node);
2463                 kfree(log_tree_root);
2464                 return -EIO;
2465         }
2466         /* returns with log_tree_root freed on success */
2467         ret = btrfs_recover_log_trees(log_tree_root);
2468         if (ret) {
2469                 btrfs_handle_fs_error(tree_root->fs_info, ret,
2470                             "Failed to recover log tree");
2471                 free_extent_buffer(log_tree_root->node);
2472                 kfree(log_tree_root);
2473                 return ret;
2474         }
2475
2476         if (fs_info->sb->s_flags & MS_RDONLY) {
2477                 ret = btrfs_commit_super(tree_root);
2478                 if (ret)
2479                         return ret;
2480         }
2481
2482         return 0;
2483 }
2484
2485 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2486                             struct btrfs_root *tree_root)
2487 {
2488         struct btrfs_root *root;
2489         struct btrfs_key location;
2490         int ret;
2491
2492         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2493         location.type = BTRFS_ROOT_ITEM_KEY;
2494         location.offset = 0;
2495
2496         root = btrfs_read_tree_root(tree_root, &location);
2497         if (IS_ERR(root))
2498                 return PTR_ERR(root);
2499         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2500         fs_info->extent_root = root;
2501
2502         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2503         root = btrfs_read_tree_root(tree_root, &location);
2504         if (IS_ERR(root))
2505                 return PTR_ERR(root);
2506         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2507         fs_info->dev_root = root;
2508         btrfs_init_devices_late(fs_info);
2509
2510         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2511         root = btrfs_read_tree_root(tree_root, &location);
2512         if (IS_ERR(root))
2513                 return PTR_ERR(root);
2514         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2515         fs_info->csum_root = root;
2516
2517         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2518         root = btrfs_read_tree_root(tree_root, &location);
2519         if (!IS_ERR(root)) {
2520                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2521                 fs_info->quota_enabled = 1;
2522                 fs_info->pending_quota_state = 1;
2523                 fs_info->quota_root = root;
2524         }
2525
2526         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2527         root = btrfs_read_tree_root(tree_root, &location);
2528         if (IS_ERR(root)) {
2529                 ret = PTR_ERR(root);
2530                 if (ret != -ENOENT)
2531                         return ret;
2532         } else {
2533                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2534                 fs_info->uuid_root = root;
2535         }
2536
2537         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2538                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2539                 root = btrfs_read_tree_root(tree_root, &location);
2540                 if (IS_ERR(root))
2541                         return PTR_ERR(root);
2542                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2543                 fs_info->free_space_root = root;
2544         }
2545
2546         return 0;
2547 }
2548
2549 int open_ctree(struct super_block *sb,
2550                struct btrfs_fs_devices *fs_devices,
2551                char *options)
2552 {
2553         u32 sectorsize;
2554         u32 nodesize;
2555         u32 stripesize;
2556         u64 generation;
2557         u64 features;
2558         struct btrfs_key location;
2559         struct buffer_head *bh;
2560         struct btrfs_super_block *disk_super;
2561         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2562         struct btrfs_root *tree_root;
2563         struct btrfs_root *chunk_root;
2564         int ret;
2565         int err = -EINVAL;
2566         int num_backups_tried = 0;
2567         int backup_index = 0;
2568         int max_active;
2569
2570         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2571         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2572         if (!tree_root || !chunk_root) {
2573                 err = -ENOMEM;
2574                 goto fail;
2575         }
2576
2577         ret = init_srcu_struct(&fs_info->subvol_srcu);
2578         if (ret) {
2579                 err = ret;
2580                 goto fail;
2581         }
2582
2583         ret = setup_bdi(fs_info, &fs_info->bdi);
2584         if (ret) {
2585                 err = ret;
2586                 goto fail_srcu;
2587         }
2588
2589         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2590         if (ret) {
2591                 err = ret;
2592                 goto fail_bdi;
2593         }
2594         fs_info->dirty_metadata_batch = PAGE_SIZE *
2595                                         (1 + ilog2(nr_cpu_ids));
2596
2597         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2598         if (ret) {
2599                 err = ret;
2600                 goto fail_dirty_metadata_bytes;
2601         }
2602
2603         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2604         if (ret) {
2605                 err = ret;
2606                 goto fail_delalloc_bytes;
2607         }
2608
2609         fs_info->btree_inode = new_inode(sb);
2610         if (!fs_info->btree_inode) {
2611                 err = -ENOMEM;
2612                 goto fail_bio_counter;
2613         }
2614
2615         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2616
2617         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2618         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2619         INIT_LIST_HEAD(&fs_info->trans_list);
2620         INIT_LIST_HEAD(&fs_info->dead_roots);
2621         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2622         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2623         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2624         spin_lock_init(&fs_info->delalloc_root_lock);
2625         spin_lock_init(&fs_info->trans_lock);
2626         spin_lock_init(&fs_info->fs_roots_radix_lock);
2627         spin_lock_init(&fs_info->delayed_iput_lock);
2628         spin_lock_init(&fs_info->defrag_inodes_lock);
2629         spin_lock_init(&fs_info->free_chunk_lock);
2630         spin_lock_init(&fs_info->tree_mod_seq_lock);
2631         spin_lock_init(&fs_info->super_lock);
2632         spin_lock_init(&fs_info->qgroup_op_lock);
2633         spin_lock_init(&fs_info->buffer_lock);
2634         spin_lock_init(&fs_info->unused_bgs_lock);
2635         rwlock_init(&fs_info->tree_mod_log_lock);
2636         mutex_init(&fs_info->unused_bg_unpin_mutex);
2637         mutex_init(&fs_info->delete_unused_bgs_mutex);
2638         mutex_init(&fs_info->reloc_mutex);
2639         mutex_init(&fs_info->delalloc_root_mutex);
2640         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2641         seqlock_init(&fs_info->profiles_lock);
2642
2643         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2644         INIT_LIST_HEAD(&fs_info->space_info);
2645         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2646         INIT_LIST_HEAD(&fs_info->unused_bgs);
2647         btrfs_mapping_init(&fs_info->mapping_tree);
2648         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2649                              BTRFS_BLOCK_RSV_GLOBAL);
2650         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2651                              BTRFS_BLOCK_RSV_DELALLOC);
2652         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2653         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2654         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2655         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2656                              BTRFS_BLOCK_RSV_DELOPS);
2657         atomic_set(&fs_info->nr_async_submits, 0);
2658         atomic_set(&fs_info->async_delalloc_pages, 0);
2659         atomic_set(&fs_info->async_submit_draining, 0);
2660         atomic_set(&fs_info->nr_async_bios, 0);
2661         atomic_set(&fs_info->defrag_running, 0);
2662         atomic_set(&fs_info->qgroup_op_seq, 0);
2663         atomic_set(&fs_info->reada_works_cnt, 0);
2664         atomic64_set(&fs_info->tree_mod_seq, 0);
2665         fs_info->fs_frozen = 0;
2666         fs_info->sb = sb;
2667         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2668         fs_info->metadata_ratio = 0;
2669         fs_info->defrag_inodes = RB_ROOT;
2670         fs_info->free_chunk_space = 0;
2671         fs_info->tree_mod_log = RB_ROOT;
2672         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2673         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2674         /* readahead state */
2675         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2676         spin_lock_init(&fs_info->reada_lock);
2677
2678         fs_info->thread_pool_size = min_t(unsigned long,
2679                                           num_online_cpus() + 2, 8);
2680
2681         INIT_LIST_HEAD(&fs_info->ordered_roots);
2682         spin_lock_init(&fs_info->ordered_root_lock);
2683         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2684                                         GFP_KERNEL);
2685         if (!fs_info->delayed_root) {
2686                 err = -ENOMEM;
2687                 goto fail_iput;
2688         }
2689         btrfs_init_delayed_root(fs_info->delayed_root);
2690
2691         btrfs_init_scrub(fs_info);
2692 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2693         fs_info->check_integrity_print_mask = 0;
2694 #endif
2695         btrfs_init_balance(fs_info);
2696         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2697
2698         sb->s_blocksize = 4096;
2699         sb->s_blocksize_bits = blksize_bits(4096);
2700         sb->s_bdi = &fs_info->bdi;
2701
2702         btrfs_init_btree_inode(fs_info, tree_root);
2703
2704         spin_lock_init(&fs_info->block_group_cache_lock);
2705         fs_info->block_group_cache_tree = RB_ROOT;
2706         fs_info->first_logical_byte = (u64)-1;
2707
2708         extent_io_tree_init(&fs_info->freed_extents[0],
2709                              fs_info->btree_inode->i_mapping);
2710         extent_io_tree_init(&fs_info->freed_extents[1],
2711                              fs_info->btree_inode->i_mapping);
2712         fs_info->pinned_extents = &fs_info->freed_extents[0];
2713         fs_info->do_barriers = 1;
2714
2715
2716         mutex_init(&fs_info->ordered_operations_mutex);
2717         mutex_init(&fs_info->tree_log_mutex);
2718         mutex_init(&fs_info->chunk_mutex);
2719         mutex_init(&fs_info->transaction_kthread_mutex);
2720         mutex_init(&fs_info->cleaner_mutex);
2721         mutex_init(&fs_info->volume_mutex);
2722         mutex_init(&fs_info->ro_block_group_mutex);
2723         init_rwsem(&fs_info->commit_root_sem);
2724         init_rwsem(&fs_info->cleanup_work_sem);
2725         init_rwsem(&fs_info->subvol_sem);
2726         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2727
2728         btrfs_init_dev_replace_locks(fs_info);
2729         btrfs_init_qgroup(fs_info);
2730
2731         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2732         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2733
2734         init_waitqueue_head(&fs_info->transaction_throttle);
2735         init_waitqueue_head(&fs_info->transaction_wait);
2736         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2737         init_waitqueue_head(&fs_info->async_submit_wait);
2738
2739         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2740
2741         ret = btrfs_alloc_stripe_hash_table(fs_info);
2742         if (ret) {
2743                 err = ret;
2744                 goto fail_alloc;
2745         }
2746
2747         __setup_root(4096, 4096, 4096, tree_root,
2748                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2749
2750         invalidate_bdev(fs_devices->latest_bdev);
2751
2752         /*
2753          * Read super block and check the signature bytes only
2754          */
2755         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2756         if (IS_ERR(bh)) {
2757                 err = PTR_ERR(bh);
2758                 goto fail_alloc;
2759         }
2760
2761         /*
2762          * We want to check superblock checksum, the type is stored inside.
2763          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2764          */
2765         if (btrfs_check_super_csum(bh->b_data)) {
2766                 btrfs_err(fs_info, "superblock checksum mismatch");
2767                 err = -EINVAL;
2768                 brelse(bh);
2769                 goto fail_alloc;
2770         }
2771
2772         /*
2773          * super_copy is zeroed at allocation time and we never touch the
2774          * following bytes up to INFO_SIZE, the checksum is calculated from
2775          * the whole block of INFO_SIZE
2776          */
2777         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2778         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2779                sizeof(*fs_info->super_for_commit));
2780         brelse(bh);
2781
2782         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2783
2784         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2785         if (ret) {
2786                 btrfs_err(fs_info, "superblock contains fatal errors");
2787                 err = -EINVAL;
2788                 goto fail_alloc;
2789         }
2790
2791         disk_super = fs_info->super_copy;
2792         if (!btrfs_super_root(disk_super))
2793                 goto fail_alloc;
2794
2795         /* check FS state, whether FS is broken. */
2796         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2797                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2798
2799         /*
2800          * run through our array of backup supers and setup
2801          * our ring pointer to the oldest one
2802          */
2803         generation = btrfs_super_generation(disk_super);
2804         find_oldest_super_backup(fs_info, generation);
2805
2806         /*
2807          * In the long term, we'll store the compression type in the super
2808          * block, and it'll be used for per file compression control.
2809          */
2810         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2811
2812         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2813         if (ret) {
2814                 err = ret;
2815                 goto fail_alloc;
2816         }
2817
2818         features = btrfs_super_incompat_flags(disk_super) &
2819                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2820         if (features) {
2821                 btrfs_err(fs_info,
2822                     "cannot mount because of unsupported optional features (%llx)",
2823                     features);
2824                 err = -EINVAL;
2825                 goto fail_alloc;
2826         }
2827
2828         features = btrfs_super_incompat_flags(disk_super);
2829         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2830         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2831                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2832
2833         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2834                 btrfs_info(fs_info, "has skinny extents");
2835
2836         /*
2837          * flag our filesystem as having big metadata blocks if
2838          * they are bigger than the page size
2839          */
2840         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2841                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2842                         btrfs_info(fs_info,
2843                                 "flagging fs with big metadata feature");
2844                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2845         }
2846
2847         nodesize = btrfs_super_nodesize(disk_super);
2848         sectorsize = btrfs_super_sectorsize(disk_super);
2849         stripesize = sectorsize;
2850         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2851         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2852
2853         /*
2854          * mixed block groups end up with duplicate but slightly offset
2855          * extent buffers for the same range.  It leads to corruptions
2856          */
2857         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2858             (sectorsize != nodesize)) {
2859                 btrfs_err(fs_info,
2860 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2861                         nodesize, sectorsize);
2862                 goto fail_alloc;
2863         }
2864
2865         /*
2866          * Needn't use the lock because there is no other task which will
2867          * update the flag.
2868          */
2869         btrfs_set_super_incompat_flags(disk_super, features);
2870
2871         features = btrfs_super_compat_ro_flags(disk_super) &
2872                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2873         if (!(sb->s_flags & MS_RDONLY) && features) {
2874                 btrfs_err(fs_info,
2875         "cannot mount read-write because of unsupported optional features (%llx)",
2876                        features);
2877                 err = -EINVAL;
2878                 goto fail_alloc;
2879         }
2880
2881         max_active = fs_info->thread_pool_size;
2882
2883         ret = btrfs_init_workqueues(fs_info, fs_devices);
2884         if (ret) {
2885                 err = ret;
2886                 goto fail_sb_buffer;
2887         }
2888
2889         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2890         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2891                                     SZ_4M / PAGE_SIZE);
2892
2893         tree_root->nodesize = nodesize;
2894         tree_root->sectorsize = sectorsize;
2895         tree_root->stripesize = stripesize;
2896
2897         sb->s_blocksize = sectorsize;
2898         sb->s_blocksize_bits = blksize_bits(sectorsize);
2899
2900         mutex_lock(&fs_info->chunk_mutex);
2901         ret = btrfs_read_sys_array(tree_root);
2902         mutex_unlock(&fs_info->chunk_mutex);
2903         if (ret) {
2904                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2905                 goto fail_sb_buffer;
2906         }
2907
2908         generation = btrfs_super_chunk_root_generation(disk_super);
2909
2910         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2911                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2912
2913         chunk_root->node = read_tree_block(chunk_root,
2914                                            btrfs_super_chunk_root(disk_super),
2915                                            generation);
2916         if (IS_ERR(chunk_root->node) ||
2917             !extent_buffer_uptodate(chunk_root->node)) {
2918                 btrfs_err(fs_info, "failed to read chunk root");
2919                 if (!IS_ERR(chunk_root->node))
2920                         free_extent_buffer(chunk_root->node);
2921                 chunk_root->node = NULL;
2922                 goto fail_tree_roots;
2923         }
2924         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2925         chunk_root->commit_root = btrfs_root_node(chunk_root);
2926
2927         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2928            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2929
2930         ret = btrfs_read_chunk_tree(chunk_root);
2931         if (ret) {
2932                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2933                 goto fail_tree_roots;
2934         }
2935
2936         /*
2937          * keep the device that is marked to be the target device for the
2938          * dev_replace procedure
2939          */
2940         btrfs_close_extra_devices(fs_devices, 0);
2941
2942         if (!fs_devices->latest_bdev) {
2943                 btrfs_err(fs_info, "failed to read devices");
2944                 goto fail_tree_roots;
2945         }
2946
2947 retry_root_backup:
2948         generation = btrfs_super_generation(disk_super);
2949
2950         tree_root->node = read_tree_block(tree_root,
2951                                           btrfs_super_root(disk_super),
2952                                           generation);
2953         if (IS_ERR(tree_root->node) ||
2954             !extent_buffer_uptodate(tree_root->node)) {
2955                 btrfs_warn(fs_info, "failed to read tree root");
2956                 if (!IS_ERR(tree_root->node))
2957                         free_extent_buffer(tree_root->node);
2958                 tree_root->node = NULL;
2959                 goto recovery_tree_root;
2960         }
2961
2962         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2963         tree_root->commit_root = btrfs_root_node(tree_root);
2964         btrfs_set_root_refs(&tree_root->root_item, 1);
2965
2966         mutex_lock(&tree_root->objectid_mutex);
2967         ret = btrfs_find_highest_objectid(tree_root,
2968                                         &tree_root->highest_objectid);
2969         if (ret) {
2970                 mutex_unlock(&tree_root->objectid_mutex);
2971                 goto recovery_tree_root;
2972         }
2973
2974         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2975
2976         mutex_unlock(&tree_root->objectid_mutex);
2977
2978         ret = btrfs_read_roots(fs_info, tree_root);
2979         if (ret)
2980                 goto recovery_tree_root;
2981
2982         fs_info->generation = generation;
2983         fs_info->last_trans_committed = generation;
2984
2985         ret = btrfs_recover_balance(fs_info);
2986         if (ret) {
2987                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2988                 goto fail_block_groups;
2989         }
2990
2991         ret = btrfs_init_dev_stats(fs_info);
2992         if (ret) {
2993                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2994                 goto fail_block_groups;
2995         }
2996
2997         ret = btrfs_init_dev_replace(fs_info);
2998         if (ret) {
2999                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3000                 goto fail_block_groups;
3001         }
3002
3003         btrfs_close_extra_devices(fs_devices, 1);
3004
3005         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3006         if (ret) {
3007                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3008                                 ret);
3009                 goto fail_block_groups;
3010         }
3011
3012         ret = btrfs_sysfs_add_device(fs_devices);
3013         if (ret) {
3014                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3015                                 ret);
3016                 goto fail_fsdev_sysfs;
3017         }
3018
3019         ret = btrfs_sysfs_add_mounted(fs_info);
3020         if (ret) {
3021                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3022                 goto fail_fsdev_sysfs;
3023         }
3024
3025         ret = btrfs_init_space_info(fs_info);
3026         if (ret) {
3027                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3028                 goto fail_sysfs;
3029         }
3030
3031         ret = btrfs_read_block_groups(fs_info->extent_root);
3032         if (ret) {
3033                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3034                 goto fail_sysfs;
3035         }
3036         fs_info->num_tolerated_disk_barrier_failures =
3037                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3038         if (fs_info->fs_devices->missing_devices >
3039              fs_info->num_tolerated_disk_barrier_failures &&
3040             !(sb->s_flags & MS_RDONLY)) {
3041                 btrfs_warn(fs_info,
3042 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3043                         fs_info->fs_devices->missing_devices,
3044                         fs_info->num_tolerated_disk_barrier_failures);
3045                 goto fail_sysfs;
3046         }
3047
3048         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3049                                                "btrfs-cleaner");
3050         if (IS_ERR(fs_info->cleaner_kthread))
3051                 goto fail_sysfs;
3052
3053         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3054                                                    tree_root,
3055                                                    "btrfs-transaction");
3056         if (IS_ERR(fs_info->transaction_kthread))
3057                 goto fail_cleaner;
3058
3059         if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3060             !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3061             !fs_info->fs_devices->rotating) {
3062                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3063                 btrfs_set_opt(fs_info->mount_opt, SSD);
3064         }
3065
3066         /*
3067          * Mount does not set all options immediately, we can do it now and do
3068          * not have to wait for transaction commit
3069          */
3070         btrfs_apply_pending_changes(fs_info);
3071
3072 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3073         if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3074                 ret = btrfsic_mount(tree_root, fs_devices,
3075                                     btrfs_test_opt(tree_root->fs_info,
3076                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3077                                     1 : 0,
3078                                     fs_info->check_integrity_print_mask);
3079                 if (ret)
3080                         btrfs_warn(fs_info,
3081                                 "failed to initialize integrity check module: %d",
3082                                 ret);
3083         }
3084 #endif
3085         ret = btrfs_read_qgroup_config(fs_info);
3086         if (ret)
3087                 goto fail_trans_kthread;
3088
3089         /* do not make disk changes in broken FS or nologreplay is given */
3090         if (btrfs_super_log_root(disk_super) != 0 &&
3091             !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3092                 ret = btrfs_replay_log(fs_info, fs_devices);
3093                 if (ret) {
3094                         err = ret;
3095                         goto fail_qgroup;
3096                 }
3097         }
3098
3099         ret = btrfs_find_orphan_roots(tree_root);
3100         if (ret)
3101                 goto fail_qgroup;
3102
3103         if (!(sb->s_flags & MS_RDONLY)) {
3104                 ret = btrfs_cleanup_fs_roots(fs_info);
3105                 if (ret)
3106                         goto fail_qgroup;
3107
3108                 mutex_lock(&fs_info->cleaner_mutex);
3109                 ret = btrfs_recover_relocation(tree_root);
3110                 mutex_unlock(&fs_info->cleaner_mutex);
3111                 if (ret < 0) {
3112                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3113                                         ret);
3114                         err = -EINVAL;
3115                         goto fail_qgroup;
3116                 }
3117         }
3118
3119         location.objectid = BTRFS_FS_TREE_OBJECTID;
3120         location.type = BTRFS_ROOT_ITEM_KEY;
3121         location.offset = 0;
3122
3123         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3124         if (IS_ERR(fs_info->fs_root)) {
3125                 err = PTR_ERR(fs_info->fs_root);
3126                 goto fail_qgroup;
3127         }
3128
3129         if (sb->s_flags & MS_RDONLY)
3130                 return 0;
3131
3132         if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3133             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3134                 btrfs_info(fs_info, "creating free space tree");
3135                 ret = btrfs_create_free_space_tree(fs_info);
3136                 if (ret) {
3137                         btrfs_warn(fs_info,
3138                                 "failed to create free space tree: %d", ret);
3139                         close_ctree(tree_root);
3140                         return ret;
3141                 }
3142         }
3143
3144         down_read(&fs_info->cleanup_work_sem);
3145         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3146             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3147                 up_read(&fs_info->cleanup_work_sem);
3148                 close_ctree(tree_root);
3149                 return ret;
3150         }
3151         up_read(&fs_info->cleanup_work_sem);
3152
3153         ret = btrfs_resume_balance_async(fs_info);
3154         if (ret) {
3155                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3156                 close_ctree(tree_root);
3157                 return ret;
3158         }
3159
3160         ret = btrfs_resume_dev_replace_async(fs_info);
3161         if (ret) {
3162                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3163                 close_ctree(tree_root);
3164                 return ret;
3165         }
3166
3167         btrfs_qgroup_rescan_resume(fs_info);
3168
3169         if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3170             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3171                 btrfs_info(fs_info, "clearing free space tree");
3172                 ret = btrfs_clear_free_space_tree(fs_info);
3173                 if (ret) {
3174                         btrfs_warn(fs_info,
3175                                 "failed to clear free space tree: %d", ret);
3176                         close_ctree(tree_root);
3177                         return ret;
3178                 }
3179         }
3180
3181         if (!fs_info->uuid_root) {
3182                 btrfs_info(fs_info, "creating UUID tree");
3183                 ret = btrfs_create_uuid_tree(fs_info);
3184                 if (ret) {
3185                         btrfs_warn(fs_info,
3186                                 "failed to create the UUID tree: %d", ret);
3187                         close_ctree(tree_root);
3188                         return ret;
3189                 }
3190         } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3191                    fs_info->generation !=
3192                                 btrfs_super_uuid_tree_generation(disk_super)) {
3193                 btrfs_info(fs_info, "checking UUID tree");
3194                 ret = btrfs_check_uuid_tree(fs_info);
3195                 if (ret) {
3196                         btrfs_warn(fs_info,
3197                                 "failed to check the UUID tree: %d", ret);
3198                         close_ctree(tree_root);
3199                         return ret;
3200                 }
3201         } else {
3202                 fs_info->update_uuid_tree_gen = 1;
3203         }
3204
3205         fs_info->open = 1;
3206
3207         /*
3208          * backuproot only affect mount behavior, and if open_ctree succeeded,
3209          * no need to keep the flag
3210          */
3211         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3212
3213         return 0;
3214
3215 fail_qgroup:
3216         btrfs_free_qgroup_config(fs_info);
3217 fail_trans_kthread:
3218         kthread_stop(fs_info->transaction_kthread);
3219         btrfs_cleanup_transaction(fs_info->tree_root);
3220         btrfs_free_fs_roots(fs_info);
3221 fail_cleaner:
3222         kthread_stop(fs_info->cleaner_kthread);
3223
3224         /*
3225          * make sure we're done with the btree inode before we stop our
3226          * kthreads
3227          */
3228         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3229
3230 fail_sysfs:
3231         btrfs_sysfs_remove_mounted(fs_info);
3232
3233 fail_fsdev_sysfs:
3234         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3235
3236 fail_block_groups:
3237         btrfs_put_block_group_cache(fs_info);
3238         btrfs_free_block_groups(fs_info);
3239
3240 fail_tree_roots:
3241         free_root_pointers(fs_info, 1);
3242         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3243
3244 fail_sb_buffer:
3245         btrfs_stop_all_workers(fs_info);
3246 fail_alloc:
3247 fail_iput:
3248         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3249
3250         iput(fs_info->btree_inode);
3251 fail_bio_counter:
3252         percpu_counter_destroy(&fs_info->bio_counter);
3253 fail_delalloc_bytes:
3254         percpu_counter_destroy(&fs_info->delalloc_bytes);
3255 fail_dirty_metadata_bytes:
3256         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3257 fail_bdi:
3258         bdi_destroy(&fs_info->bdi);
3259 fail_srcu:
3260         cleanup_srcu_struct(&fs_info->subvol_srcu);
3261 fail:
3262         btrfs_free_stripe_hash_table(fs_info);
3263         btrfs_close_devices(fs_info->fs_devices);
3264         return err;
3265
3266 recovery_tree_root:
3267         if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3268                 goto fail_tree_roots;
3269
3270         free_root_pointers(fs_info, 0);
3271
3272         /* don't use the log in recovery mode, it won't be valid */
3273         btrfs_set_super_log_root(disk_super, 0);
3274
3275         /* we can't trust the free space cache either */
3276         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3277
3278         ret = next_root_backup(fs_info, fs_info->super_copy,
3279                                &num_backups_tried, &backup_index);
3280         if (ret == -1)
3281                 goto fail_block_groups;
3282         goto retry_root_backup;
3283 }
3284
3285 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3286 {
3287         if (uptodate) {
3288                 set_buffer_uptodate(bh);
3289         } else {
3290                 struct btrfs_device *device = (struct btrfs_device *)
3291                         bh->b_private;
3292
3293                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3294                                 "lost page write due to IO error on %s",
3295                                           rcu_str_deref(device->name));
3296                 /* note, we don't set_buffer_write_io_error because we have
3297                  * our own ways of dealing with the IO errors
3298                  */
3299                 clear_buffer_uptodate(bh);
3300                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3301         }
3302         unlock_buffer(bh);
3303         put_bh(bh);
3304 }
3305
3306 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3307                         struct buffer_head **bh_ret)
3308 {
3309         struct buffer_head *bh;
3310         struct btrfs_super_block *super;
3311         u64 bytenr;
3312
3313         bytenr = btrfs_sb_offset(copy_num);
3314         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3315                 return -EINVAL;
3316
3317         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3318         /*
3319          * If we fail to read from the underlying devices, as of now
3320          * the best option we have is to mark it EIO.
3321          */
3322         if (!bh)
3323                 return -EIO;
3324
3325         super = (struct btrfs_super_block *)bh->b_data;
3326         if (btrfs_super_bytenr(super) != bytenr ||
3327                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3328                 brelse(bh);
3329                 return -EINVAL;
3330         }
3331
3332         *bh_ret = bh;
3333         return 0;
3334 }
3335
3336
3337 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3338 {
3339         struct buffer_head *bh;
3340         struct buffer_head *latest = NULL;
3341         struct btrfs_super_block *super;
3342         int i;
3343         u64 transid = 0;
3344         int ret = -EINVAL;
3345
3346         /* we would like to check all the supers, but that would make
3347          * a btrfs mount succeed after a mkfs from a different FS.
3348          * So, we need to add a special mount option to scan for
3349          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3350          */
3351         for (i = 0; i < 1; i++) {
3352                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3353                 if (ret)
3354                         continue;
3355
3356                 super = (struct btrfs_super_block *)bh->b_data;
3357
3358                 if (!latest || btrfs_super_generation(super) > transid) {
3359                         brelse(latest);
3360                         latest = bh;
3361                         transid = btrfs_super_generation(super);
3362                 } else {
3363                         brelse(bh);
3364                 }
3365         }
3366
3367         if (!latest)
3368                 return ERR_PTR(ret);
3369
3370         return latest;
3371 }
3372
3373 /*
3374  * this should be called twice, once with wait == 0 and
3375  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3376  * we write are pinned.
3377  *
3378  * They are released when wait == 1 is done.
3379  * max_mirrors must be the same for both runs, and it indicates how
3380  * many supers on this one device should be written.
3381  *
3382  * max_mirrors == 0 means to write them all.
3383  */
3384 static int write_dev_supers(struct btrfs_device *device,
3385                             struct btrfs_super_block *sb,
3386                             int do_barriers, int wait, int max_mirrors)
3387 {
3388         struct buffer_head *bh;
3389         int i;
3390         int ret;
3391         int errors = 0;
3392         u32 crc;
3393         u64 bytenr;
3394
3395         if (max_mirrors == 0)
3396                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3397
3398         for (i = 0; i < max_mirrors; i++) {
3399                 bytenr = btrfs_sb_offset(i);
3400                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3401                     device->commit_total_bytes)
3402                         break;
3403
3404                 if (wait) {
3405                         bh = __find_get_block(device->bdev, bytenr / 4096,
3406                                               BTRFS_SUPER_INFO_SIZE);
3407                         if (!bh) {
3408                                 errors++;
3409                                 continue;
3410                         }
3411                         wait_on_buffer(bh);
3412                         if (!buffer_uptodate(bh))
3413                                 errors++;
3414
3415                         /* drop our reference */
3416                         brelse(bh);
3417
3418                         /* drop the reference from the wait == 0 run */
3419                         brelse(bh);
3420                         continue;
3421                 } else {
3422                         btrfs_set_super_bytenr(sb, bytenr);
3423
3424                         crc = ~(u32)0;
3425                         crc = btrfs_csum_data((char *)sb +
3426                                               BTRFS_CSUM_SIZE, crc,
3427                                               BTRFS_SUPER_INFO_SIZE -
3428                                               BTRFS_CSUM_SIZE);
3429                         btrfs_csum_final(crc, sb->csum);
3430
3431                         /*
3432                          * one reference for us, and we leave it for the
3433                          * caller
3434                          */
3435                         bh = __getblk(device->bdev, bytenr / 4096,
3436                                       BTRFS_SUPER_INFO_SIZE);
3437                         if (!bh) {
3438                                 btrfs_err(device->dev_root->fs_info,
3439                                     "couldn't get super buffer head for bytenr %llu",
3440                                     bytenr);
3441                                 errors++;
3442                                 continue;
3443                         }
3444
3445                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3446
3447                         /* one reference for submit_bh */
3448                         get_bh(bh);
3449
3450                         set_buffer_uptodate(bh);
3451                         lock_buffer(bh);
3452                         bh->b_end_io = btrfs_end_buffer_write_sync;
3453                         bh->b_private = device;
3454                 }
3455
3456                 /*
3457                  * we fua the first super.  The others we allow
3458                  * to go down lazy.
3459                  */
3460                 if (i == 0)
3461                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_FUA, bh);
3462                 else
3463                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_SYNC, bh);
3464                 if (ret)
3465                         errors++;
3466         }
3467         return errors < i ? 0 : -1;
3468 }
3469
3470 /*
3471  * endio for the write_dev_flush, this will wake anyone waiting
3472  * for the barrier when it is done
3473  */
3474 static void btrfs_end_empty_barrier(struct bio *bio)
3475 {
3476         if (bio->bi_private)
3477                 complete(bio->bi_private);
3478         bio_put(bio);
3479 }
3480
3481 /*
3482  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3483  * sent down.  With wait == 1, it waits for the previous flush.
3484  *
3485  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3486  * capable
3487  */
3488 static int write_dev_flush(struct btrfs_device *device, int wait)
3489 {
3490         struct bio *bio;
3491         int ret = 0;
3492
3493         if (device->nobarriers)
3494                 return 0;
3495
3496         if (wait) {
3497                 bio = device->flush_bio;
3498                 if (!bio)
3499                         return 0;
3500
3501                 wait_for_completion(&device->flush_wait);
3502
3503                 if (bio->bi_error) {
3504                         ret = bio->bi_error;
3505                         btrfs_dev_stat_inc_and_print(device,
3506                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3507                 }
3508
3509                 /* drop the reference from the wait == 0 run */
3510                 bio_put(bio);
3511                 device->flush_bio = NULL;
3512
3513                 return ret;
3514         }
3515
3516         /*
3517          * one reference for us, and we leave it for the
3518          * caller
3519          */
3520         device->flush_bio = NULL;
3521         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3522         if (!bio)
3523                 return -ENOMEM;
3524
3525         bio->bi_end_io = btrfs_end_empty_barrier;
3526         bio->bi_bdev = device->bdev;
3527         bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
3528         init_completion(&device->flush_wait);
3529         bio->bi_private = &device->flush_wait;
3530         device->flush_bio = bio;
3531
3532         bio_get(bio);
3533         btrfsic_submit_bio(bio);
3534
3535         return 0;
3536 }
3537
3538 /*
3539  * send an empty flush down to each device in parallel,
3540  * then wait for them
3541  */
3542 static int barrier_all_devices(struct btrfs_fs_info *info)
3543 {
3544         struct list_head *head;
3545         struct btrfs_device *dev;
3546         int errors_send = 0;
3547         int errors_wait = 0;
3548         int ret;
3549
3550         /* send down all the barriers */
3551         head = &info->fs_devices->devices;
3552         list_for_each_entry_rcu(dev, head, dev_list) {
3553                 if (dev->missing)
3554                         continue;
3555                 if (!dev->bdev) {
3556                         errors_send++;
3557                         continue;
3558                 }
3559                 if (!dev->in_fs_metadata || !dev->writeable)
3560                         continue;
3561
3562                 ret = write_dev_flush(dev, 0);
3563                 if (ret)
3564                         errors_send++;
3565         }
3566
3567         /* wait for all the barriers */
3568         list_for_each_entry_rcu(dev, head, dev_list) {
3569                 if (dev->missing)
3570                         continue;
3571                 if (!dev->bdev) {
3572                         errors_wait++;
3573                         continue;
3574                 }
3575                 if (!dev->in_fs_metadata || !dev->writeable)
3576                         continue;
3577
3578                 ret = write_dev_flush(dev, 1);
3579                 if (ret)
3580                         errors_wait++;
3581         }
3582         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3583             errors_wait > info->num_tolerated_disk_barrier_failures)
3584                 return -EIO;
3585         return 0;
3586 }
3587
3588 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3589 {
3590         int raid_type;
3591         int min_tolerated = INT_MAX;
3592
3593         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3594             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3595                 min_tolerated = min(min_tolerated,
3596                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3597                                     tolerated_failures);
3598
3599         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3600                 if (raid_type == BTRFS_RAID_SINGLE)
3601                         continue;
3602                 if (!(flags & btrfs_raid_group[raid_type]))
3603                         continue;
3604                 min_tolerated = min(min_tolerated,
3605                                     btrfs_raid_array[raid_type].
3606                                     tolerated_failures);
3607         }
3608
3609         if (min_tolerated == INT_MAX) {
3610                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3611                 min_tolerated = 0;
3612         }
3613
3614         return min_tolerated;
3615 }
3616
3617 int btrfs_calc_num_tolerated_disk_barrier_failures(
3618         struct btrfs_fs_info *fs_info)
3619 {
3620         struct btrfs_ioctl_space_info space;
3621         struct btrfs_space_info *sinfo;
3622         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3623                        BTRFS_BLOCK_GROUP_SYSTEM,
3624                        BTRFS_BLOCK_GROUP_METADATA,
3625                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3626         int i;
3627         int c;
3628         int num_tolerated_disk_barrier_failures =
3629                 (int)fs_info->fs_devices->num_devices;
3630
3631         for (i = 0; i < ARRAY_SIZE(types); i++) {
3632                 struct btrfs_space_info *tmp;
3633
3634                 sinfo = NULL;
3635                 rcu_read_lock();
3636                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3637                         if (tmp->flags == types[i]) {
3638                                 sinfo = tmp;
3639                                 break;
3640                         }
3641                 }
3642                 rcu_read_unlock();
3643
3644                 if (!sinfo)
3645                         continue;
3646
3647                 down_read(&sinfo->groups_sem);
3648                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3649                         u64 flags;
3650
3651                         if (list_empty(&sinfo->block_groups[c]))
3652                                 continue;
3653
3654                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3655                                                    &space);
3656                         if (space.total_bytes == 0 || space.used_bytes == 0)
3657                                 continue;
3658                         flags = space.flags;
3659
3660                         num_tolerated_disk_barrier_failures = min(
3661                                 num_tolerated_disk_barrier_failures,
3662                                 btrfs_get_num_tolerated_disk_barrier_failures(
3663                                         flags));
3664                 }
3665                 up_read(&sinfo->groups_sem);
3666         }
3667
3668         return num_tolerated_disk_barrier_failures;
3669 }
3670
3671 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3672 {
3673         struct list_head *head;
3674         struct btrfs_device *dev;
3675         struct btrfs_super_block *sb;
3676         struct btrfs_dev_item *dev_item;
3677         int ret;
3678         int do_barriers;
3679         int max_errors;
3680         int total_errors = 0;
3681         u64 flags;
3682
3683         do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3684         backup_super_roots(root->fs_info);
3685
3686         sb = root->fs_info->super_for_commit;
3687         dev_item = &sb->dev_item;
3688
3689         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3690         head = &root->fs_info->fs_devices->devices;
3691         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3692
3693         if (do_barriers) {
3694                 ret = barrier_all_devices(root->fs_info);
3695                 if (ret) {
3696                         mutex_unlock(
3697                                 &root->fs_info->fs_devices->device_list_mutex);
3698                         btrfs_handle_fs_error(root->fs_info, ret,
3699                                     "errors while submitting device barriers.");
3700                         return ret;
3701                 }
3702         }
3703
3704         list_for_each_entry_rcu(dev, head, dev_list) {
3705                 if (!dev->bdev) {
3706                         total_errors++;
3707                         continue;
3708                 }
3709                 if (!dev->in_fs_metadata || !dev->writeable)
3710                         continue;
3711
3712                 btrfs_set_stack_device_generation(dev_item, 0);
3713                 btrfs_set_stack_device_type(dev_item, dev->type);
3714                 btrfs_set_stack_device_id(dev_item, dev->devid);
3715                 btrfs_set_stack_device_total_bytes(dev_item,
3716                                                    dev->commit_total_bytes);
3717                 btrfs_set_stack_device_bytes_used(dev_item,
3718                                                   dev->commit_bytes_used);
3719                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3720                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3721                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3722                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3723                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3724
3725                 flags = btrfs_super_flags(sb);
3726                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3727
3728                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3729                 if (ret)
3730                         total_errors++;
3731         }
3732         if (total_errors > max_errors) {
3733                 btrfs_err(root->fs_info, "%d errors while writing supers",
3734                        total_errors);
3735                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3736
3737                 /* FUA is masked off if unsupported and can't be the reason */
3738                 btrfs_handle_fs_error(root->fs_info, -EIO,
3739                             "%d errors while writing supers", total_errors);
3740                 return -EIO;
3741         }
3742
3743         total_errors = 0;
3744         list_for_each_entry_rcu(dev, head, dev_list) {
3745                 if (!dev->bdev)
3746                         continue;
3747                 if (!dev->in_fs_metadata || !dev->writeable)
3748                         continue;
3749
3750                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3751                 if (ret)
3752                         total_errors++;
3753         }
3754         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3755         if (total_errors > max_errors) {
3756                 btrfs_handle_fs_error(root->fs_info, -EIO,
3757                             "%d errors while writing supers", total_errors);
3758                 return -EIO;
3759         }
3760         return 0;
3761 }
3762
3763 int write_ctree_super(struct btrfs_trans_handle *trans,
3764                       struct btrfs_root *root, int max_mirrors)
3765 {
3766         return write_all_supers(root, max_mirrors);
3767 }
3768
3769 /* Drop a fs root from the radix tree and free it. */
3770 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3771                                   struct btrfs_root *root)
3772 {
3773         spin_lock(&fs_info->fs_roots_radix_lock);
3774         radix_tree_delete(&fs_info->fs_roots_radix,
3775                           (unsigned long)root->root_key.objectid);
3776         spin_unlock(&fs_info->fs_roots_radix_lock);
3777
3778         if (btrfs_root_refs(&root->root_item) == 0)
3779                 synchronize_srcu(&fs_info->subvol_srcu);
3780
3781         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3782                 btrfs_free_log(NULL, root);
3783                 if (root->reloc_root) {
3784                         free_extent_buffer(root->reloc_root->node);
3785                         free_extent_buffer(root->reloc_root->commit_root);
3786                         btrfs_put_fs_root(root->reloc_root);
3787                         root->reloc_root = NULL;
3788                 }
3789         }
3790
3791         if (root->free_ino_pinned)
3792                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3793         if (root->free_ino_ctl)
3794                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3795         free_fs_root(root);
3796 }
3797
3798 static void free_fs_root(struct btrfs_root *root)
3799 {
3800         iput(root->ino_cache_inode);
3801         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3802         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3803         root->orphan_block_rsv = NULL;
3804         if (root->anon_dev)
3805                 free_anon_bdev(root->anon_dev);
3806         if (root->subv_writers)
3807                 btrfs_free_subvolume_writers(root->subv_writers);
3808         free_extent_buffer(root->node);
3809         free_extent_buffer(root->commit_root);
3810         kfree(root->free_ino_ctl);
3811         kfree(root->free_ino_pinned);
3812         kfree(root->name);
3813         btrfs_put_fs_root(root);
3814 }
3815
3816 void btrfs_free_fs_root(struct btrfs_root *root)
3817 {
3818         free_fs_root(root);
3819 }
3820
3821 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3822 {
3823         u64 root_objectid = 0;
3824         struct btrfs_root *gang[8];
3825         int i = 0;
3826         int err = 0;
3827         unsigned int ret = 0;
3828         int index;
3829
3830         while (1) {
3831                 index = srcu_read_lock(&fs_info->subvol_srcu);
3832                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3833                                              (void **)gang, root_objectid,
3834                                              ARRAY_SIZE(gang));
3835                 if (!ret) {
3836                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3837                         break;
3838                 }
3839                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3840
3841                 for (i = 0; i < ret; i++) {
3842                         /* Avoid to grab roots in dead_roots */
3843                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3844                                 gang[i] = NULL;
3845                                 continue;
3846                         }
3847                         /* grab all the search result for later use */
3848                         gang[i] = btrfs_grab_fs_root(gang[i]);
3849                 }
3850                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3851
3852                 for (i = 0; i < ret; i++) {
3853                         if (!gang[i])
3854                                 continue;
3855                         root_objectid = gang[i]->root_key.objectid;
3856                         err = btrfs_orphan_cleanup(gang[i]);
3857                         if (err)
3858                                 break;
3859                         btrfs_put_fs_root(gang[i]);
3860                 }
3861                 root_objectid++;
3862         }
3863
3864         /* release the uncleaned roots due to error */
3865         for (; i < ret; i++) {
3866                 if (gang[i])
3867                         btrfs_put_fs_root(gang[i]);
3868         }
3869         return err;
3870 }
3871
3872 int btrfs_commit_super(struct btrfs_root *root)
3873 {
3874         struct btrfs_trans_handle *trans;
3875
3876         mutex_lock(&root->fs_info->cleaner_mutex);
3877         btrfs_run_delayed_iputs(root);
3878         mutex_unlock(&root->fs_info->cleaner_mutex);
3879         wake_up_process(root->fs_info->cleaner_kthread);
3880
3881         /* wait until ongoing cleanup work done */
3882         down_write(&root->fs_info->cleanup_work_sem);
3883         up_write(&root->fs_info->cleanup_work_sem);
3884
3885         trans = btrfs_join_transaction(root);
3886         if (IS_ERR(trans))
3887                 return PTR_ERR(trans);
3888         return btrfs_commit_transaction(trans, root);
3889 }
3890
3891 void close_ctree(struct btrfs_root *root)
3892 {
3893         struct btrfs_fs_info *fs_info = root->fs_info;
3894         int ret;
3895
3896         fs_info->closing = 1;
3897         smp_mb();
3898
3899         /* wait for the qgroup rescan worker to stop */
3900         btrfs_qgroup_wait_for_completion(fs_info, false);
3901
3902         /* wait for the uuid_scan task to finish */
3903         down(&fs_info->uuid_tree_rescan_sem);
3904         /* avoid complains from lockdep et al., set sem back to initial state */
3905         up(&fs_info->uuid_tree_rescan_sem);
3906
3907         /* pause restriper - we want to resume on mount */
3908         btrfs_pause_balance(fs_info);
3909
3910         btrfs_dev_replace_suspend_for_unmount(fs_info);
3911
3912         btrfs_scrub_cancel(fs_info);
3913
3914         /* wait for any defraggers to finish */
3915         wait_event(fs_info->transaction_wait,
3916                    (atomic_read(&fs_info->defrag_running) == 0));
3917
3918         /* clear out the rbtree of defraggable inodes */
3919         btrfs_cleanup_defrag_inodes(fs_info);
3920
3921         cancel_work_sync(&fs_info->async_reclaim_work);
3922
3923         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3924                 /*
3925                  * If the cleaner thread is stopped and there are
3926                  * block groups queued for removal, the deletion will be
3927                  * skipped when we quit the cleaner thread.
3928                  */
3929                 btrfs_delete_unused_bgs(root->fs_info);
3930
3931                 ret = btrfs_commit_super(root);
3932                 if (ret)
3933                         btrfs_err(fs_info, "commit super ret %d", ret);
3934         }
3935
3936         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3937                 btrfs_error_commit_super(root);
3938
3939         kthread_stop(fs_info->transaction_kthread);
3940         kthread_stop(fs_info->cleaner_kthread);
3941
3942         fs_info->closing = 2;
3943         smp_mb();
3944
3945         btrfs_free_qgroup_config(fs_info);
3946
3947         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3948                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3949                        percpu_counter_sum(&fs_info->delalloc_bytes));
3950         }
3951
3952         btrfs_sysfs_remove_mounted(fs_info);
3953         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3954
3955         btrfs_free_fs_roots(fs_info);
3956
3957         btrfs_put_block_group_cache(fs_info);
3958
3959         btrfs_free_block_groups(fs_info);
3960
3961         /*
3962          * we must make sure there is not any read request to
3963          * submit after we stopping all workers.
3964          */
3965         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3966         btrfs_stop_all_workers(fs_info);
3967
3968         fs_info->open = 0;
3969         free_root_pointers(fs_info, 1);
3970
3971         iput(fs_info->btree_inode);
3972
3973 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3974         if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
3975                 btrfsic_unmount(root, fs_info->fs_devices);
3976 #endif
3977
3978         btrfs_close_devices(fs_info->fs_devices);
3979         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3980
3981         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3982         percpu_counter_destroy(&fs_info->delalloc_bytes);
3983         percpu_counter_destroy(&fs_info->bio_counter);
3984         bdi_destroy(&fs_info->bdi);
3985         cleanup_srcu_struct(&fs_info->subvol_srcu);
3986
3987         btrfs_free_stripe_hash_table(fs_info);
3988
3989         __btrfs_free_block_rsv(root->orphan_block_rsv);
3990         root->orphan_block_rsv = NULL;
3991
3992         lock_chunks(root);
3993         while (!list_empty(&fs_info->pinned_chunks)) {
3994                 struct extent_map *em;
3995
3996                 em = list_first_entry(&fs_info->pinned_chunks,
3997                                       struct extent_map, list);
3998                 list_del_init(&em->list);
3999                 free_extent_map(em);
4000         }
4001         unlock_chunks(root);
4002 }
4003
4004 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4005                           int atomic)
4006 {
4007         int ret;
4008         struct inode *btree_inode = buf->pages[0]->mapping->host;
4009
4010         ret = extent_buffer_uptodate(buf);
4011         if (!ret)
4012                 return ret;
4013
4014         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4015                                     parent_transid, atomic);
4016         if (ret == -EAGAIN)
4017                 return ret;
4018         return !ret;
4019 }
4020
4021 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4022 {
4023         struct btrfs_root *root;
4024         u64 transid = btrfs_header_generation(buf);
4025         int was_dirty;
4026
4027 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4028         /*
4029          * This is a fast path so only do this check if we have sanity tests
4030          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4031          * outside of the sanity tests.
4032          */
4033         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4034                 return;
4035 #endif
4036         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4037         btrfs_assert_tree_locked(buf);
4038         if (transid != root->fs_info->generation)
4039                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
4040                        "found %llu running %llu\n",
4041                         buf->start, transid, root->fs_info->generation);
4042         was_dirty = set_extent_buffer_dirty(buf);
4043         if (!was_dirty)
4044                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4045                                      buf->len,
4046                                      root->fs_info->dirty_metadata_batch);
4047 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4048         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4049                 btrfs_print_leaf(root, buf);
4050                 ASSERT(0);
4051         }
4052 #endif
4053 }
4054
4055 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4056                                         int flush_delayed)
4057 {
4058         /*
4059          * looks as though older kernels can get into trouble with
4060          * this code, they end up stuck in balance_dirty_pages forever
4061          */
4062         int ret;
4063
4064         if (current->flags & PF_MEMALLOC)
4065                 return;
4066
4067         if (flush_delayed)
4068                 btrfs_balance_delayed_items(root);
4069
4070         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4071                                      BTRFS_DIRTY_METADATA_THRESH);
4072         if (ret > 0) {
4073                 balance_dirty_pages_ratelimited(
4074                                    root->fs_info->btree_inode->i_mapping);
4075         }
4076 }
4077
4078 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4079 {
4080         __btrfs_btree_balance_dirty(root, 1);
4081 }
4082
4083 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4084 {
4085         __btrfs_btree_balance_dirty(root, 0);
4086 }
4087
4088 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4089 {
4090         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4091         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4092 }
4093
4094 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4095                               int read_only)
4096 {
4097         struct btrfs_super_block *sb = fs_info->super_copy;
4098         u64 nodesize = btrfs_super_nodesize(sb);
4099         u64 sectorsize = btrfs_super_sectorsize(sb);
4100         int ret = 0;
4101
4102         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4103                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4104                 ret = -EINVAL;
4105         }
4106         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4107                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4108                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4109         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4110                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4111                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4112                 ret = -EINVAL;
4113         }
4114         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4115                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4116                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4117                 ret = -EINVAL;
4118         }
4119         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4120                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4121                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4122                 ret = -EINVAL;
4123         }
4124
4125         /*
4126          * Check sectorsize and nodesize first, other check will need it.
4127          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4128          */
4129         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4130             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4131                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4132                 ret = -EINVAL;
4133         }
4134         /* Only PAGE SIZE is supported yet */
4135         if (sectorsize != PAGE_SIZE) {
4136                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4137                                 sectorsize, PAGE_SIZE);
4138                 ret = -EINVAL;
4139         }
4140         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4141             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4142                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4143                 ret = -EINVAL;
4144         }
4145         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4146                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4147                                 le32_to_cpu(sb->__unused_leafsize),
4148                                 nodesize);
4149                 ret = -EINVAL;
4150         }
4151
4152         /* Root alignment check */
4153         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4154                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4155                                 btrfs_super_root(sb));
4156                 ret = -EINVAL;
4157         }
4158         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4159                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4160                                 btrfs_super_chunk_root(sb));
4161                 ret = -EINVAL;
4162         }
4163         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4164                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4165                                 btrfs_super_log_root(sb));
4166                 ret = -EINVAL;
4167         }
4168
4169         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4170                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4171                                 fs_info->fsid, sb->dev_item.fsid);
4172                 ret = -EINVAL;
4173         }
4174
4175         /*
4176          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4177          * done later
4178          */
4179         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4180                 btrfs_err(fs_info, "bytes_used is too small %llu",
4181                        btrfs_super_bytes_used(sb));
4182                 ret = -EINVAL;
4183         }
4184         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4185                 btrfs_err(fs_info, "invalid stripesize %u",
4186                        btrfs_super_stripesize(sb));
4187                 ret = -EINVAL;
4188         }
4189         if (btrfs_super_num_devices(sb) > (1UL << 31))
4190                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4191                                 btrfs_super_num_devices(sb));
4192         if (btrfs_super_num_devices(sb) == 0) {
4193                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4194                 ret = -EINVAL;
4195         }
4196
4197         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4198                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4199                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4200                 ret = -EINVAL;
4201         }
4202
4203         /*
4204          * Obvious sys_chunk_array corruptions, it must hold at least one key
4205          * and one chunk
4206          */
4207         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4208                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4209                                 btrfs_super_sys_array_size(sb),
4210                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4211                 ret = -EINVAL;
4212         }
4213         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4214                         + sizeof(struct btrfs_chunk)) {
4215                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4216                                 btrfs_super_sys_array_size(sb),
4217                                 sizeof(struct btrfs_disk_key)
4218                                 + sizeof(struct btrfs_chunk));
4219                 ret = -EINVAL;
4220         }
4221
4222         /*
4223          * The generation is a global counter, we'll trust it more than the others
4224          * but it's still possible that it's the one that's wrong.
4225          */
4226         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4227                 printk(KERN_WARNING
4228                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4229                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4230         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4231             && btrfs_super_cache_generation(sb) != (u64)-1)
4232                 printk(KERN_WARNING
4233                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4234                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4235
4236         return ret;
4237 }
4238
4239 static void btrfs_error_commit_super(struct btrfs_root *root)
4240 {
4241         mutex_lock(&root->fs_info->cleaner_mutex);
4242         btrfs_run_delayed_iputs(root);
4243         mutex_unlock(&root->fs_info->cleaner_mutex);
4244
4245         down_write(&root->fs_info->cleanup_work_sem);
4246         up_write(&root->fs_info->cleanup_work_sem);
4247
4248         /* cleanup FS via transaction */
4249         btrfs_cleanup_transaction(root);
4250 }
4251
4252 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4253 {
4254         struct btrfs_ordered_extent *ordered;
4255
4256         spin_lock(&root->ordered_extent_lock);
4257         /*
4258          * This will just short circuit the ordered completion stuff which will
4259          * make sure the ordered extent gets properly cleaned up.
4260          */
4261         list_for_each_entry(ordered, &root->ordered_extents,
4262                             root_extent_list)
4263                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4264         spin_unlock(&root->ordered_extent_lock);
4265 }
4266
4267 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4268 {
4269         struct btrfs_root *root;
4270         struct list_head splice;
4271
4272         INIT_LIST_HEAD(&splice);
4273
4274         spin_lock(&fs_info->ordered_root_lock);
4275         list_splice_init(&fs_info->ordered_roots, &splice);
4276         while (!list_empty(&splice)) {
4277                 root = list_first_entry(&splice, struct btrfs_root,
4278                                         ordered_root);
4279                 list_move_tail(&root->ordered_root,
4280                                &fs_info->ordered_roots);
4281
4282                 spin_unlock(&fs_info->ordered_root_lock);
4283                 btrfs_destroy_ordered_extents(root);
4284
4285                 cond_resched();
4286                 spin_lock(&fs_info->ordered_root_lock);
4287         }
4288         spin_unlock(&fs_info->ordered_root_lock);
4289 }
4290
4291 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4292                                       struct btrfs_root *root)
4293 {
4294         struct rb_node *node;
4295         struct btrfs_delayed_ref_root *delayed_refs;
4296         struct btrfs_delayed_ref_node *ref;
4297         int ret = 0;
4298
4299         delayed_refs = &trans->delayed_refs;
4300
4301         spin_lock(&delayed_refs->lock);
4302         if (atomic_read(&delayed_refs->num_entries) == 0) {
4303                 spin_unlock(&delayed_refs->lock);
4304                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4305                 return ret;
4306         }
4307
4308         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4309                 struct btrfs_delayed_ref_head *head;
4310                 struct btrfs_delayed_ref_node *tmp;
4311                 bool pin_bytes = false;
4312
4313                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4314                                 href_node);
4315                 if (!mutex_trylock(&head->mutex)) {
4316                         atomic_inc(&head->node.refs);
4317                         spin_unlock(&delayed_refs->lock);
4318
4319                         mutex_lock(&head->mutex);
4320                         mutex_unlock(&head->mutex);
4321                         btrfs_put_delayed_ref(&head->node);
4322                         spin_lock(&delayed_refs->lock);
4323                         continue;
4324                 }
4325                 spin_lock(&head->lock);
4326                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4327                                                  list) {
4328                         ref->in_tree = 0;
4329                         list_del(&ref->list);
4330                         atomic_dec(&delayed_refs->num_entries);
4331                         btrfs_put_delayed_ref(ref);
4332                 }
4333                 if (head->must_insert_reserved)
4334                         pin_bytes = true;
4335                 btrfs_free_delayed_extent_op(head->extent_op);
4336                 delayed_refs->num_heads--;
4337                 if (head->processing == 0)
4338                         delayed_refs->num_heads_ready--;
4339                 atomic_dec(&delayed_refs->num_entries);
4340                 head->node.in_tree = 0;
4341                 rb_erase(&head->href_node, &delayed_refs->href_root);
4342                 spin_unlock(&head->lock);
4343                 spin_unlock(&delayed_refs->lock);
4344                 mutex_unlock(&head->mutex);
4345
4346                 if (pin_bytes)
4347                         btrfs_pin_extent(root, head->node.bytenr,
4348                                          head->node.num_bytes, 1);
4349                 btrfs_put_delayed_ref(&head->node);
4350                 cond_resched();
4351                 spin_lock(&delayed_refs->lock);
4352         }
4353
4354         spin_unlock(&delayed_refs->lock);
4355
4356         return ret;
4357 }
4358
4359 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4360 {
4361         struct btrfs_inode *btrfs_inode;
4362         struct list_head splice;
4363
4364         INIT_LIST_HEAD(&splice);
4365
4366         spin_lock(&root->delalloc_lock);
4367         list_splice_init(&root->delalloc_inodes, &splice);
4368
4369         while (!list_empty(&splice)) {
4370                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4371                                                delalloc_inodes);
4372
4373                 list_del_init(&btrfs_inode->delalloc_inodes);
4374                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4375                           &btrfs_inode->runtime_flags);
4376                 spin_unlock(&root->delalloc_lock);
4377
4378                 btrfs_invalidate_inodes(btrfs_inode->root);
4379
4380                 spin_lock(&root->delalloc_lock);
4381         }
4382
4383         spin_unlock(&root->delalloc_lock);
4384 }
4385
4386 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4387 {
4388         struct btrfs_root *root;
4389         struct list_head splice;
4390
4391         INIT_LIST_HEAD(&splice);
4392
4393         spin_lock(&fs_info->delalloc_root_lock);
4394         list_splice_init(&fs_info->delalloc_roots, &splice);
4395         while (!list_empty(&splice)) {
4396                 root = list_first_entry(&splice, struct btrfs_root,
4397                                          delalloc_root);
4398                 list_del_init(&root->delalloc_root);
4399                 root = btrfs_grab_fs_root(root);
4400                 BUG_ON(!root);
4401                 spin_unlock(&fs_info->delalloc_root_lock);
4402
4403                 btrfs_destroy_delalloc_inodes(root);
4404                 btrfs_put_fs_root(root);
4405
4406                 spin_lock(&fs_info->delalloc_root_lock);
4407         }
4408         spin_unlock(&fs_info->delalloc_root_lock);
4409 }
4410
4411 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4412                                         struct extent_io_tree *dirty_pages,
4413                                         int mark)
4414 {
4415         int ret;
4416         struct extent_buffer *eb;
4417         u64 start = 0;
4418         u64 end;
4419
4420         while (1) {
4421                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4422                                             mark, NULL);
4423                 if (ret)
4424                         break;
4425
4426                 clear_extent_bits(dirty_pages, start, end, mark);
4427                 while (start <= end) {
4428                         eb = btrfs_find_tree_block(root->fs_info, start);
4429                         start += root->nodesize;
4430                         if (!eb)
4431                                 continue;
4432                         wait_on_extent_buffer_writeback(eb);
4433
4434                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4435                                                &eb->bflags))
4436                                 clear_extent_buffer_dirty(eb);
4437                         free_extent_buffer_stale(eb);
4438                 }
4439         }
4440
4441         return ret;
4442 }
4443
4444 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4445                                        struct extent_io_tree *pinned_extents)
4446 {
4447         struct extent_io_tree *unpin;
4448         u64 start;
4449         u64 end;
4450         int ret;
4451         bool loop = true;
4452
4453         unpin = pinned_extents;
4454 again:
4455         while (1) {
4456                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4457                                             EXTENT_DIRTY, NULL);
4458                 if (ret)
4459                         break;
4460
4461                 clear_extent_dirty(unpin, start, end);
4462                 btrfs_error_unpin_extent_range(root, start, end);
4463                 cond_resched();
4464         }
4465
4466         if (loop) {
4467                 if (unpin == &root->fs_info->freed_extents[0])
4468                         unpin = &root->fs_info->freed_extents[1];
4469                 else
4470                         unpin = &root->fs_info->freed_extents[0];
4471                 loop = false;
4472                 goto again;
4473         }
4474
4475         return 0;
4476 }
4477
4478 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4479                                    struct btrfs_root *root)
4480 {
4481         btrfs_destroy_delayed_refs(cur_trans, root);
4482
4483         cur_trans->state = TRANS_STATE_COMMIT_START;
4484         wake_up(&root->fs_info->transaction_blocked_wait);
4485
4486         cur_trans->state = TRANS_STATE_UNBLOCKED;
4487         wake_up(&root->fs_info->transaction_wait);
4488
4489         btrfs_destroy_delayed_inodes(root);
4490         btrfs_assert_delayed_root_empty(root);
4491
4492         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4493                                      EXTENT_DIRTY);
4494         btrfs_destroy_pinned_extent(root,
4495                                     root->fs_info->pinned_extents);
4496
4497         cur_trans->state =TRANS_STATE_COMPLETED;
4498         wake_up(&cur_trans->commit_wait);
4499
4500         /*
4501         memset(cur_trans, 0, sizeof(*cur_trans));
4502         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4503         */
4504 }
4505
4506 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4507 {
4508         struct btrfs_transaction *t;
4509
4510         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4511
4512         spin_lock(&root->fs_info->trans_lock);
4513         while (!list_empty(&root->fs_info->trans_list)) {
4514                 t = list_first_entry(&root->fs_info->trans_list,
4515                                      struct btrfs_transaction, list);
4516                 if (t->state >= TRANS_STATE_COMMIT_START) {
4517                         atomic_inc(&t->use_count);
4518                         spin_unlock(&root->fs_info->trans_lock);
4519                         btrfs_wait_for_commit(root, t->transid);
4520                         btrfs_put_transaction(t);
4521                         spin_lock(&root->fs_info->trans_lock);
4522                         continue;
4523                 }
4524                 if (t == root->fs_info->running_transaction) {
4525                         t->state = TRANS_STATE_COMMIT_DOING;
4526                         spin_unlock(&root->fs_info->trans_lock);
4527                         /*
4528                          * We wait for 0 num_writers since we don't hold a trans
4529                          * handle open currently for this transaction.
4530                          */
4531                         wait_event(t->writer_wait,
4532                                    atomic_read(&t->num_writers) == 0);
4533                 } else {
4534                         spin_unlock(&root->fs_info->trans_lock);
4535                 }
4536                 btrfs_cleanup_one_transaction(t, root);
4537
4538                 spin_lock(&root->fs_info->trans_lock);
4539                 if (t == root->fs_info->running_transaction)
4540                         root->fs_info->running_transaction = NULL;
4541                 list_del_init(&t->list);
4542                 spin_unlock(&root->fs_info->trans_lock);
4543
4544                 btrfs_put_transaction(t);
4545                 trace_btrfs_transaction_commit(root);
4546                 spin_lock(&root->fs_info->trans_lock);
4547         }
4548         spin_unlock(&root->fs_info->trans_lock);
4549         btrfs_destroy_all_ordered_extents(root->fs_info);
4550         btrfs_destroy_delayed_inodes(root);
4551         btrfs_assert_delayed_root_empty(root);
4552         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4553         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4554         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4555
4556         return 0;
4557 }
4558
4559 static const struct extent_io_ops btree_extent_io_ops = {
4560         .readpage_end_io_hook = btree_readpage_end_io_hook,
4561         .readpage_io_failed_hook = btree_io_failed_hook,
4562         .submit_bio_hook = btree_submit_bio_hook,
4563         /* note we're sharing with inode.c for the merge bio hook */
4564         .merge_bio_hook = btrfs_merge_bio_hook,
4565 };