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 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  69 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  70                                       struct btrfs_fs_info *fs_info);
  71 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  72 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  73                                         struct extent_io_tree *dirty_pages,
  74                                         int mark);
  75 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  76                                        struct extent_io_tree *pinned_extents);
  77 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  78 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  79
  80 /*
  81  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  82  * is complete.  This is used during reads to verify checksums, and it is used
  83  * by writes to insert metadata for new file extents after IO is complete.
  84  */
  85 struct btrfs_end_io_wq {
  86         struct bio *bio;
  87         bio_end_io_t *end_io;
  88         void *private;
  89         struct btrfs_fs_info *info;
  90         int error;
  91         enum btrfs_wq_endio_type metadata;
  92         struct btrfs_work work;
  93 };
  94
  95 static struct kmem_cache *btrfs_end_io_wq_cache;
  96
  97 int __init btrfs_end_io_wq_init(void)
  98 {
  99         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 100                                         sizeof(struct btrfs_end_io_wq),
 101                                         0,
 102                                         SLAB_MEM_SPREAD,
 103                                         NULL);
 104         if (!btrfs_end_io_wq_cache)
 105                 return -ENOMEM;
 106         return 0;
 107 }
 108
 109 void btrfs_end_io_wq_exit(void)
 110 {
 111         kmem_cache_destroy(btrfs_end_io_wq_cache);
 112 }
 113
 114 /*
 115  * async submit bios are used to offload expensive checksumming
 116  * onto the worker threads.  They checksum file and metadata bios
 117  * just before they are sent down the IO stack.
 118  */
 119 struct async_submit_bio {
 120         void *private_data;
 121         struct btrfs_fs_info *fs_info;
 122         struct bio *bio;
 123         extent_submit_bio_hook_t *submit_bio_start;
 124         extent_submit_bio_hook_t *submit_bio_done;
 125         int mirror_num;
 126         unsigned long bio_flags;
 127         /*
 128          * bio_offset is optional, can be used if the pages in the bio
 129          * can't tell us where in the file the bio should go
 130          */
 131         u64 bio_offset;
 132         struct btrfs_work work;
 133         int error;
 134 };
 135
 136 /*
 137  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 138  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 139  * the level the eb occupies in the tree.
 140  *
 141  * Different roots are used for different purposes and may nest inside each
 142  * other and they require separate keysets.  As lockdep keys should be
 143  * static, assign keysets according to the purpose of the root as indicated
 144  * by btrfs_root->objectid.  This ensures that all special purpose roots
 145  * have separate keysets.
 146  *
 147  * Lock-nesting across peer nodes is always done with the immediate parent
 148  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 149  * subclass to avoid triggering lockdep warning in such cases.
 150  *
 151  * The key is set by the readpage_end_io_hook after the buffer has passed
 152  * csum validation but before the pages are unlocked.  It is also set by
 153  * btrfs_init_new_buffer on freshly allocated blocks.
 154  *
 155  * We also add a check to make sure the highest level of the tree is the
 156  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 157  * needs update as well.
 158  */
 159 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 160 # if BTRFS_MAX_LEVEL != 8
 161 #  error
 162 # endif
 163
 164 static struct btrfs_lockdep_keyset {
 165         u64                     id;             /* root objectid */
 166         const char              *name_stem;     /* lock name stem */
 167         char                    names[BTRFS_MAX_LEVEL + 1][20];
 168         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 169 } btrfs_lockdep_keysets[] = {
 170         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 171         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 172         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 173         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 174         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 175         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 176         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 177         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 178         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 179         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 180         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 181         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 182         { .id = 0,                              .name_stem = "tree"     },
 183 };
 184
 185 void __init btrfs_init_lockdep(void)
 186 {
 187         int i, j;
 188
 189         /* initialize lockdep class names */
 190         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 191                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 192
 193                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 194                         snprintf(ks->names[j], sizeof(ks->names[j]),
 195                                  "btrfs-%s-%02d", ks->name_stem, j);
 196         }
 197 }
 198
 199 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 200                                     int level)
 201 {
 202         struct btrfs_lockdep_keyset *ks;
 203
 204         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 205
 206         /* find the matching keyset, id 0 is the default entry */
 207         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 208                 if (ks->id == objectid)
 209                         break;
 210
 211         lockdep_set_class_and_name(&eb->lock,
 212                                    &ks->keys[level], ks->names[level]);
 213 }
 214
 215 #endif
 216
 217 /*
 218  * extents on the btree inode are pretty simple, there's one extent
 219  * that covers the entire device
 220  */
 221 static struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 222                 struct page *page, size_t pg_offset, u64 start, u64 len,
 223                 int create)
 224 {
 225         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 226         struct extent_map_tree *em_tree = &inode->extent_tree;
 227         struct extent_map *em;
 228         int ret;
 229
 230         read_lock(&em_tree->lock);
 231         em = lookup_extent_mapping(em_tree, start, len);
 232         if (em) {
 233                 em->bdev = fs_info->fs_devices->latest_bdev;
 234                 read_unlock(&em_tree->lock);
 235                 goto out;
 236         }
 237         read_unlock(&em_tree->lock);
 238
 239         em = alloc_extent_map();
 240         if (!em) {
 241                 em = ERR_PTR(-ENOMEM);
 242                 goto out;
 243         }
 244         em->start = 0;
 245         em->len = (u64)-1;
 246         em->block_len = (u64)-1;
 247         em->block_start = 0;
 248         em->bdev = fs_info->fs_devices->latest_bdev;
 249
 250         write_lock(&em_tree->lock);
 251         ret = add_extent_mapping(em_tree, em, 0);
 252         if (ret == -EEXIST) {
 253                 free_extent_map(em);
 254                 em = lookup_extent_mapping(em_tree, start, len);
 255                 if (!em)
 256                         em = ERR_PTR(-EIO);
 257         } else if (ret) {
 258                 free_extent_map(em);
 259                 em = ERR_PTR(ret);
 260         }
 261         write_unlock(&em_tree->lock);
 262
 263 out:
 264         return em;
 265 }
 266
 267 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 268 {
 269         return btrfs_crc32c(seed, data, len);
 270 }
 271
 272 void btrfs_csum_final(u32 crc, u8 *result)
 273 {
 274         put_unaligned_le32(~crc, result);
 275 }
 276
 277 /*
 278  * compute the csum for a btree block, and either verify it or write it
 279  * into the csum field of the block.
 280  */
 281 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 282                            struct extent_buffer *buf,
 283                            int verify)
 284 {
 285         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 286         char *result = NULL;
 287         unsigned long len;
 288         unsigned long cur_len;
 289         unsigned long offset = BTRFS_CSUM_SIZE;
 290         char *kaddr;
 291         unsigned long map_start;
 292         unsigned long map_len;
 293         int err;
 294         u32 crc = ~(u32)0;
 295         unsigned long inline_result;
 296
 297         len = buf->len - offset;
 298         while (len > 0) {
 299                 err = map_private_extent_buffer(buf, offset, 32,
 300                                         &kaddr, &map_start, &map_len);
 301                 if (err)
 302                         return err;
 303                 cur_len = min(len, map_len - (offset - map_start));
 304                 crc = btrfs_csum_data(kaddr + offset - map_start,
 305                                       crc, cur_len);
 306                 len -= cur_len;
 307                 offset += cur_len;
 308         }
 309         if (csum_size > sizeof(inline_result)) {
 310                 result = kzalloc(csum_size, GFP_NOFS);
 311                 if (!result)
 312                         return -ENOMEM;
 313         } else {
 314                 result = (char *)&inline_result;
 315         }
 316
 317         btrfs_csum_final(crc, result);
 318
 319         if (verify) {
 320                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 321                         u32 val;
 322                         u32 found = 0;
 323                         memcpy(&found, result, csum_size);
 324
 325                         read_extent_buffer(buf, &val, 0, csum_size);
 326                         btrfs_warn_rl(fs_info,
 327                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 328                                 fs_info->sb->s_id, buf->start,
 329                                 val, found, btrfs_header_level(buf));
 330                         if (result != (char *)&inline_result)
 331                                 kfree(result);
 332                         return -EUCLEAN;
 333                 }
 334         } else {
 335                 write_extent_buffer(buf, result, 0, csum_size);
 336         }
 337         if (result != (char *)&inline_result)
 338                 kfree(result);
 339         return 0;
 340 }
 341
 342 /*
 343  * we can't consider a given block up to date unless the transid of the
 344  * block matches the transid in the parent node's pointer.  This is how we
 345  * detect blocks that either didn't get written at all or got written
 346  * in the wrong place.
 347  */
 348 static int verify_parent_transid(struct extent_io_tree *io_tree,
 349                                  struct extent_buffer *eb, u64 parent_transid,
 350                                  int atomic)
 351 {
 352         struct extent_state *cached_state = NULL;
 353         int ret;
 354         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 355
 356         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 357                 return 0;
 358
 359         if (atomic)
 360                 return -EAGAIN;
 361
 362         if (need_lock) {
 363                 btrfs_tree_read_lock(eb);
 364                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 365         }
 366
 367         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 368                          &cached_state);
 369         if (extent_buffer_uptodate(eb) &&
 370             btrfs_header_generation(eb) == parent_transid) {
 371                 ret = 0;
 372                 goto out;
 373         }
 374         btrfs_err_rl(eb->fs_info,
 375                 "parent transid verify failed on %llu wanted %llu found %llu",
 376                         eb->start,
 377                         parent_transid, btrfs_header_generation(eb));
 378         ret = 1;
 379
 380         /*
 381          * Things reading via commit roots that don't have normal protection,
 382          * like send, can have a really old block in cache that may point at a
 383          * block that has been freed and re-allocated.  So don't clear uptodate
 384          * if we find an eb that is under IO (dirty/writeback) because we could
 385          * end up reading in the stale data and then writing it back out and
 386          * making everybody very sad.
 387          */
 388         if (!extent_buffer_under_io(eb))
 389                 clear_extent_buffer_uptodate(eb);
 390 out:
 391         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 392                              &cached_state, GFP_NOFS);
 393         if (need_lock)
 394                 btrfs_tree_read_unlock_blocking(eb);
 395         return ret;
 396 }
 397
 398 /*
 399  * Return 0 if the superblock checksum type matches the checksum value of that
 400  * algorithm. Pass the raw disk superblock data.
 401  */
 402 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 403                                   char *raw_disk_sb)
 404 {
 405         struct btrfs_super_block *disk_sb =
 406                 (struct btrfs_super_block *)raw_disk_sb;
 407         u16 csum_type = btrfs_super_csum_type(disk_sb);
 408         int ret = 0;
 409
 410         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 411                 u32 crc = ~(u32)0;
 412                 const int csum_size = sizeof(crc);
 413                 char result[csum_size];
 414
 415                 /*
 416                  * The super_block structure does not span the whole
 417                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 418                  * is filled with zeros and is included in the checksum.
 419                  */
 420                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 421                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 422                 btrfs_csum_final(crc, result);
 423
 424                 if (memcmp(raw_disk_sb, result, csum_size))
 425                         ret = 1;
 426         }
 427
 428         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 429                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 430                                 csum_type);
 431                 ret = 1;
 432         }
 433
 434         return ret;
 435 }
 436
 437 /*
 438  * helper to read a given tree block, doing retries as required when
 439  * the checksums don't match and we have alternate mirrors to try.
 440  */
 441 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 442                                           struct extent_buffer *eb,
 443                                           u64 parent_transid)
 444 {
 445         struct extent_io_tree *io_tree;
 446         int failed = 0;
 447         int ret;
 448         int num_copies = 0;
 449         int mirror_num = 0;
 450         int failed_mirror = 0;
 451
 452         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 453         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 454         while (1) {
 455                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 456                                                btree_get_extent, mirror_num);
 457                 if (!ret) {
 458                         if (!verify_parent_transid(io_tree, eb,
 459                                                    parent_transid, 0))
 460                                 break;
 461                         else
 462                                 ret = -EIO;
 463                 }
 464
 465                 /*
 466                  * This buffer's crc is fine, but its contents are corrupted, so
 467                  * there is no reason to read the other copies, they won't be
 468                  * any less wrong.
 469                  */
 470                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 471                         break;
 472
 473                 num_copies = btrfs_num_copies(fs_info,
 474                                               eb->start, eb->len);
 475                 if (num_copies == 1)
 476                         break;
 477
 478                 if (!failed_mirror) {
 479                         failed = 1;
 480                         failed_mirror = eb->read_mirror;
 481                 }
 482
 483                 mirror_num++;
 484                 if (mirror_num == failed_mirror)
 485                         mirror_num++;
 486
 487                 if (mirror_num > num_copies)
 488                         break;
 489         }
 490
 491         if (failed && !ret && failed_mirror)
 492                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 493
 494         return ret;
 495 }
 496
 497 /*
 498  * checksum a dirty tree block before IO.  This has extra checks to make sure
 499  * we only fill in the checksum field in the first page of a multi-page block
 500  */
 501
 502 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 503 {
 504         u64 start = page_offset(page);
 505         u64 found_start;
 506         struct extent_buffer *eb;
 507
 508         eb = (struct extent_buffer *)page->private;
 509         if (page != eb->pages[0])
 510                 return 0;
 511
 512         found_start = btrfs_header_bytenr(eb);
 513         /*
 514          * Please do not consolidate these warnings into a single if.
 515          * It is useful to know what went wrong.
 516          */
 517         if (WARN_ON(found_start != start))
 518                 return -EUCLEAN;
 519         if (WARN_ON(!PageUptodate(page)))
 520                 return -EUCLEAN;
 521
 522         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 523                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 524
 525         return csum_tree_block(fs_info, eb, 0);
 526 }
 527
 528 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 529                                  struct extent_buffer *eb)
 530 {
 531         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 532         u8 fsid[BTRFS_UUID_SIZE];
 533         int ret = 1;
 534
 535         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 536         while (fs_devices) {
 537                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 538                         ret = 0;
 539                         break;
 540                 }
 541                 fs_devices = fs_devices->seed;
 542         }
 543         return ret;
 544 }
 545
 546 #define CORRUPT(reason, eb, root, slot)                                 \
 547         btrfs_crit(root->fs_info,                                       \
 548                    "corrupt %s, %s: block=%llu, root=%llu, slot=%d",    \
 549                    btrfs_header_level(eb) == 0 ? "leaf" : "node",       \
 550                    reason, btrfs_header_bytenr(eb), root->objectid, slot)
 551
 552 static noinline int check_leaf(struct btrfs_root *root,
 553                                struct extent_buffer *leaf)
 554 {
 555         struct btrfs_fs_info *fs_info = root->fs_info;
 556         struct btrfs_key key;
 557         struct btrfs_key leaf_key;
 558         u32 nritems = btrfs_header_nritems(leaf);
 559         int slot;
 560
 561         /*
 562          * Extent buffers from a relocation tree have a owner field that
 563          * corresponds to the subvolume tree they are based on. So just from an
 564          * extent buffer alone we can not find out what is the id of the
 565          * corresponding subvolume tree, so we can not figure out if the extent
 566          * buffer corresponds to the root of the relocation tree or not. So skip
 567          * this check for relocation trees.
 568          */
 569         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
 570                 struct btrfs_root *check_root;
 571
 572                 key.objectid = btrfs_header_owner(leaf);
 573                 key.type = BTRFS_ROOT_ITEM_KEY;
 574                 key.offset = (u64)-1;
 575
 576                 check_root = btrfs_get_fs_root(fs_info, &key, false);
 577                 /*
 578                  * The only reason we also check NULL here is that during
 579                  * open_ctree() some roots has not yet been set up.
 580                  */
 581                 if (!IS_ERR_OR_NULL(check_root)) {
 582                         struct extent_buffer *eb;
 583
 584                         eb = btrfs_root_node(check_root);
 585                         /* if leaf is the root, then it's fine */
 586                         if (leaf != eb) {
 587                                 CORRUPT("non-root leaf's nritems is 0",
 588                                         leaf, check_root, 0);
 589                                 free_extent_buffer(eb);
 590                                 return -EIO;
 591                         }
 592                         free_extent_buffer(eb);
 593                 }
 594                 return 0;
 595         }
 596
 597         if (nritems == 0)
 598                 return 0;
 599
 600         /* Check the 0 item */
 601         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 602             BTRFS_LEAF_DATA_SIZE(fs_info)) {
 603                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 604                 return -EIO;
 605         }
 606
 607         /*
 608          * Check to make sure each items keys are in the correct order and their
 609          * offsets make sense.  We only have to loop through nritems-1 because
 610          * we check the current slot against the next slot, which verifies the
 611          * next slot's offset+size makes sense and that the current's slot
 612          * offset is correct.
 613          */
 614         for (slot = 0; slot < nritems - 1; slot++) {
 615                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 616                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 617
 618                 /* Make sure the keys are in the right order */
 619                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 620                         CORRUPT("bad key order", leaf, root, slot);
 621                         return -EIO;
 622                 }
 623
 624                 /*
 625                  * Make sure the offset and ends are right, remember that the
 626                  * item data starts at the end of the leaf and grows towards the
 627                  * front.
 628                  */
 629                 if (btrfs_item_offset_nr(leaf, slot) !=
 630                         btrfs_item_end_nr(leaf, slot + 1)) {
 631                         CORRUPT("slot offset bad", leaf, root, slot);
 632                         return -EIO;
 633                 }
 634
 635                 /*
 636                  * Check to make sure that we don't point outside of the leaf,
 637                  * just in case all the items are consistent to each other, but
 638                  * all point outside of the leaf.
 639                  */
 640                 if (btrfs_item_end_nr(leaf, slot) >
 641                     BTRFS_LEAF_DATA_SIZE(fs_info)) {
 642                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 643                         return -EIO;
 644                 }
 645         }
 646
 647         return 0;
 648 }
 649
 650 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 651 {
 652         unsigned long nr = btrfs_header_nritems(node);
 653         struct btrfs_key key, next_key;
 654         int slot;
 655         u64 bytenr;
 656         int ret = 0;
 657
 658         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
 659                 btrfs_crit(root->fs_info,
 660                            "corrupt node: block %llu root %llu nritems %lu",
 661                            node->start, root->objectid, nr);
 662                 return -EIO;
 663         }
 664
 665         for (slot = 0; slot < nr - 1; slot++) {
 666                 bytenr = btrfs_node_blockptr(node, slot);
 667                 btrfs_node_key_to_cpu(node, &key, slot);
 668                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
 669
 670                 if (!bytenr) {
 671                         CORRUPT("invalid item slot", node, root, slot);
 672                         ret = -EIO;
 673                         goto out;
 674                 }
 675
 676                 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
 677                         CORRUPT("bad key order", node, root, slot);
 678                         ret = -EIO;
 679                         goto out;
 680                 }
 681         }
 682 out:
 683         return ret;
 684 }
 685
 686 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 687                                       u64 phy_offset, struct page *page,
 688                                       u64 start, u64 end, int mirror)
 689 {
 690         u64 found_start;
 691         int found_level;
 692         struct extent_buffer *eb;
 693         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 694         struct btrfs_fs_info *fs_info = root->fs_info;
 695         int ret = 0;
 696         int reads_done;
 697
 698         if (!page->private)
 699                 goto out;
 700
 701         eb = (struct extent_buffer *)page->private;
 702
 703         /* the pending IO might have been the only thing that kept this buffer
 704          * in memory.  Make sure we have a ref for all this other checks
 705          */
 706         extent_buffer_get(eb);
 707
 708         reads_done = atomic_dec_and_test(&eb->io_pages);
 709         if (!reads_done)
 710                 goto err;
 711
 712         eb->read_mirror = mirror;
 713         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 714                 ret = -EIO;
 715                 goto err;
 716         }
 717
 718         found_start = btrfs_header_bytenr(eb);
 719         if (found_start != eb->start) {
 720                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 721                              found_start, eb->start);
 722                 ret = -EIO;
 723                 goto err;
 724         }
 725         if (check_tree_block_fsid(fs_info, eb)) {
 726                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 727                              eb->start);
 728                 ret = -EIO;
 729                 goto err;
 730         }
 731         found_level = btrfs_header_level(eb);
 732         if (found_level >= BTRFS_MAX_LEVEL) {
 733                 btrfs_err(fs_info, "bad tree block level %d",
 734                           (int)btrfs_header_level(eb));
 735                 ret = -EIO;
 736                 goto err;
 737         }
 738
 739         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 740                                        eb, found_level);
 741
 742         ret = csum_tree_block(fs_info, eb, 1);
 743         if (ret)
 744                 goto err;
 745
 746         /*
 747          * If this is a leaf block and it is corrupt, set the corrupt bit so
 748          * that we don't try and read the other copies of this block, just
 749          * return -EIO.
 750          */
 751         if (found_level == 0 && check_leaf(root, eb)) {
 752                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 753                 ret = -EIO;
 754         }
 755
 756         if (found_level > 0 && check_node(root, eb))
 757                 ret = -EIO;
 758
 759         if (!ret)
 760                 set_extent_buffer_uptodate(eb);
 761 err:
 762         if (reads_done &&
 763             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 764                 btree_readahead_hook(eb, ret);
 765
 766         if (ret) {
 767                 /*
 768                  * our io error hook is going to dec the io pages
 769                  * again, we have to make sure it has something
 770                  * to decrement
 771                  */
 772                 atomic_inc(&eb->io_pages);
 773                 clear_extent_buffer_uptodate(eb);
 774         }
 775         free_extent_buffer(eb);
 776 out:
 777         return ret;
 778 }
 779
 780 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 781 {
 782         struct extent_buffer *eb;
 783
 784         eb = (struct extent_buffer *)page->private;
 785         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 786         eb->read_mirror = failed_mirror;
 787         atomic_dec(&eb->io_pages);
 788         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 789                 btree_readahead_hook(eb, -EIO);
 790         return -EIO;    /* we fixed nothing */
 791 }
 792
 793 static void end_workqueue_bio(struct bio *bio)
 794 {
 795         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 796         struct btrfs_fs_info *fs_info;
 797         struct btrfs_workqueue *wq;
 798         btrfs_work_func_t func;
 799
 800         fs_info = end_io_wq->info;
 801         end_io_wq->error = bio->bi_error;
 802
 803         if (bio_op(bio) == REQ_OP_WRITE) {
 804                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 805                         wq = fs_info->endio_meta_write_workers;
 806                         func = btrfs_endio_meta_write_helper;
 807                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 808                         wq = fs_info->endio_freespace_worker;
 809                         func = btrfs_freespace_write_helper;
 810                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 811                         wq = fs_info->endio_raid56_workers;
 812                         func = btrfs_endio_raid56_helper;
 813                 } else {
 814                         wq = fs_info->endio_write_workers;
 815                         func = btrfs_endio_write_helper;
 816                 }
 817         } else {
 818                 if (unlikely(end_io_wq->metadata ==
 819                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 820                         wq = fs_info->endio_repair_workers;
 821                         func = btrfs_endio_repair_helper;
 822                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 823                         wq = fs_info->endio_raid56_workers;
 824                         func = btrfs_endio_raid56_helper;
 825                 } else if (end_io_wq->metadata) {
 826                         wq = fs_info->endio_meta_workers;
 827                         func = btrfs_endio_meta_helper;
 828                 } else {
 829                         wq = fs_info->endio_workers;
 830                         func = btrfs_endio_helper;
 831                 }
 832         }
 833
 834         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 835         btrfs_queue_work(wq, &end_io_wq->work);
 836 }
 837
 838 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 839                         enum btrfs_wq_endio_type metadata)
 840 {
 841         struct btrfs_end_io_wq *end_io_wq;
 842
 843         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 844         if (!end_io_wq)
 845                 return -ENOMEM;
 846
 847         end_io_wq->private = bio->bi_private;
 848         end_io_wq->end_io = bio->bi_end_io;
 849         end_io_wq->info = info;
 850         end_io_wq->error = 0;
 851         end_io_wq->bio = bio;
 852         end_io_wq->metadata = metadata;
 853
 854         bio->bi_private = end_io_wq;
 855         bio->bi_end_io = end_workqueue_bio;
 856         return 0;
 857 }
 858
 859 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 860 {
 861         unsigned long limit = min_t(unsigned long,
 862                                     info->thread_pool_size,
 863                                     info->fs_devices->open_devices);
 864         return 256 * limit;
 865 }
 866
 867 static void run_one_async_start(struct btrfs_work *work)
 868 {
 869         struct async_submit_bio *async;
 870         int ret;
 871
 872         async = container_of(work, struct  async_submit_bio, work);
 873         ret = async->submit_bio_start(async->private_data, async->bio,
 874                                       async->mirror_num, async->bio_flags,
 875                                       async->bio_offset);
 876         if (ret)
 877                 async->error = ret;
 878 }
 879
 880 static void run_one_async_done(struct btrfs_work *work)
 881 {
 882         struct btrfs_fs_info *fs_info;
 883         struct async_submit_bio *async;
 884         int limit;
 885
 886         async = container_of(work, struct  async_submit_bio, work);
 887         fs_info = async->fs_info;
 888
 889         limit = btrfs_async_submit_limit(fs_info);
 890         limit = limit * 2 / 3;
 891
 892         /*
 893          * atomic_dec_return implies a barrier for waitqueue_active
 894          */
 895         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 896             waitqueue_active(&fs_info->async_submit_wait))
 897                 wake_up(&fs_info->async_submit_wait);
 898
 899         /* If an error occurred we just want to clean up the bio and move on */
 900         if (async->error) {
 901                 async->bio->bi_error = async->error;
 902                 bio_endio(async->bio);
 903                 return;
 904         }
 905
 906         async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
 907                                async->bio_flags, async->bio_offset);
 908 }
 909
 910 static void run_one_async_free(struct btrfs_work *work)
 911 {
 912         struct async_submit_bio *async;
 913
 914         async = container_of(work, struct  async_submit_bio, work);
 915         kfree(async);
 916 }
 917
 918 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 919                         int mirror_num, unsigned long bio_flags,
 920                         u64 bio_offset, void *private_data,
 921                         extent_submit_bio_hook_t *submit_bio_start,
 922                         extent_submit_bio_hook_t *submit_bio_done)
 923 {
 924         struct async_submit_bio *async;
 925
 926         async = kmalloc(sizeof(*async), GFP_NOFS);
 927         if (!async)
 928                 return -ENOMEM;
 929
 930         async->private_data = private_data;
 931         async->fs_info = fs_info;
 932         async->bio = bio;
 933         async->mirror_num = mirror_num;
 934         async->submit_bio_start = submit_bio_start;
 935         async->submit_bio_done = submit_bio_done;
 936
 937         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 938                         run_one_async_done, run_one_async_free);
 939
 940         async->bio_flags = bio_flags;
 941         async->bio_offset = bio_offset;
 942
 943         async->error = 0;
 944
 945         atomic_inc(&fs_info->nr_async_submits);
 946
 947         if (op_is_sync(bio->bi_opf))
 948                 btrfs_set_work_high_priority(&async->work);
 949
 950         btrfs_queue_work(fs_info->workers, &async->work);
 951
 952         while (atomic_read(&fs_info->async_submit_draining) &&
 953               atomic_read(&fs_info->nr_async_submits)) {
 954                 wait_event(fs_info->async_submit_wait,
 955                            (atomic_read(&fs_info->nr_async_submits) == 0));
 956         }
 957
 958         return 0;
 959 }
 960
 961 static int btree_csum_one_bio(struct bio *bio)
 962 {
 963         struct bio_vec *bvec;
 964         struct btrfs_root *root;
 965         int i, ret = 0;
 966
 967         bio_for_each_segment_all(bvec, bio, i) {
 968                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 969                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 970                 if (ret)
 971                         break;
 972         }
 973
 974         return ret;
 975 }
 976
 977 static int __btree_submit_bio_start(void *private_data, struct bio *bio,
 978                                     int mirror_num, unsigned long bio_flags,
 979                                     u64 bio_offset)
 980 {
 981         /*
 982          * when we're called for a write, we're already in the async
 983          * submission context.  Just jump into btrfs_map_bio
 984          */
 985         return btree_csum_one_bio(bio);
 986 }
 987
 988 static int __btree_submit_bio_done(void *private_data, struct bio *bio,
 989                                  int mirror_num, unsigned long bio_flags,
 990                                  u64 bio_offset)
 991 {
 992         struct inode *inode = private_data;
 993         int ret;
 994
 995         /*
 996          * when we're called for a write, we're already in the async
 997          * submission context.  Just jump into btrfs_map_bio
 998          */
 999         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1000         if (ret) {
1001                 bio->bi_error = ret;
1002                 bio_endio(bio);
1003         }
1004         return ret;
1005 }
1006
1007 static int check_async_write(unsigned long bio_flags)
1008 {
1009         if (bio_flags & EXTENT_BIO_TREE_LOG)
1010                 return 0;
1011 #ifdef CONFIG_X86
1012         if (static_cpu_has(X86_FEATURE_XMM4_2))
1013                 return 0;
1014 #endif
1015         return 1;
1016 }
1017
1018 static int btree_submit_bio_hook(void *private_data, struct bio *bio,
1019                                  int mirror_num, unsigned long bio_flags,
1020                                  u64 bio_offset)
1021 {
1022         struct inode *inode = private_data;
1023         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1024         int async = check_async_write(bio_flags);
1025         int ret;
1026
1027         if (bio_op(bio) != REQ_OP_WRITE) {
1028                 /*
1029                  * called for a read, do the setup so that checksum validation
1030                  * can happen in the async kernel threads
1031                  */
1032                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1033                                           BTRFS_WQ_ENDIO_METADATA);
1034                 if (ret)
1035                         goto out_w_error;
1036                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1037         } else if (!async) {
1038                 ret = btree_csum_one_bio(bio);
1039                 if (ret)
1040                         goto out_w_error;
1041                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1042         } else {
1043                 /*
1044                  * kthread helpers are used to submit writes so that
1045                  * checksumming can happen in parallel across all CPUs
1046                  */
1047                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
1048                                           bio_offset, private_data,
1049                                           __btree_submit_bio_start,
1050                                           __btree_submit_bio_done);
1051         }
1052
1053         if (ret)
1054                 goto out_w_error;
1055         return 0;
1056
1057 out_w_error:
1058         bio->bi_error = ret;
1059         bio_endio(bio);
1060         return ret;
1061 }
1062
1063 #ifdef CONFIG_MIGRATION
1064 static int btree_migratepage(struct address_space *mapping,
1065                         struct page *newpage, struct page *page,
1066                         enum migrate_mode mode)
1067 {
1068         /*
1069          * we can't safely write a btree page from here,
1070          * we haven't done the locking hook
1071          */
1072         if (PageDirty(page))
1073                 return -EAGAIN;
1074         /*
1075          * Buffers may be managed in a filesystem specific way.
1076          * We must have no buffers or drop them.
1077          */
1078         if (page_has_private(page) &&
1079             !try_to_release_page(page, GFP_KERNEL))
1080                 return -EAGAIN;
1081         return migrate_page(mapping, newpage, page, mode);
1082 }
1083 #endif
1084
1085
1086 static int btree_writepages(struct address_space *mapping,
1087                             struct writeback_control *wbc)
1088 {
1089         struct btrfs_fs_info *fs_info;
1090         int ret;
1091
1092         if (wbc->sync_mode == WB_SYNC_NONE) {
1093
1094                 if (wbc->for_kupdate)
1095                         return 0;
1096
1097                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1098                 /* this is a bit racy, but that's ok */
1099                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1100                                              BTRFS_DIRTY_METADATA_THRESH);
1101                 if (ret < 0)
1102                         return 0;
1103         }
1104         return btree_write_cache_pages(mapping, wbc);
1105 }
1106
1107 static int btree_readpage(struct file *file, struct page *page)
1108 {
1109         struct extent_io_tree *tree;
1110         tree = &BTRFS_I(page->mapping->host)->io_tree;
1111         return extent_read_full_page(tree, page, btree_get_extent, 0);
1112 }
1113
1114 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1115 {
1116         if (PageWriteback(page) || PageDirty(page))
1117                 return 0;
1118
1119         return try_release_extent_buffer(page);
1120 }
1121
1122 static void btree_invalidatepage(struct page *page, unsigned int offset,
1123                                  unsigned int length)
1124 {
1125         struct extent_io_tree *tree;
1126         tree = &BTRFS_I(page->mapping->host)->io_tree;
1127         extent_invalidatepage(tree, page, offset);
1128         btree_releasepage(page, GFP_NOFS);
1129         if (PagePrivate(page)) {
1130                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1131                            "page private not zero on page %llu",
1132                            (unsigned long long)page_offset(page));
1133                 ClearPagePrivate(page);
1134                 set_page_private(page, 0);
1135                 put_page(page);
1136         }
1137 }
1138
1139 static int btree_set_page_dirty(struct page *page)
1140 {
1141 #ifdef DEBUG
1142         struct extent_buffer *eb;
1143
1144         BUG_ON(!PagePrivate(page));
1145         eb = (struct extent_buffer *)page->private;
1146         BUG_ON(!eb);
1147         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1148         BUG_ON(!atomic_read(&eb->refs));
1149         btrfs_assert_tree_locked(eb);
1150 #endif
1151         return __set_page_dirty_nobuffers(page);
1152 }
1153
1154 static const struct address_space_operations btree_aops = {
1155         .readpage       = btree_readpage,
1156         .writepages     = btree_writepages,
1157         .releasepage    = btree_releasepage,
1158         .invalidatepage = btree_invalidatepage,
1159 #ifdef CONFIG_MIGRATION
1160         .migratepage    = btree_migratepage,
1161 #endif
1162         .set_page_dirty = btree_set_page_dirty,
1163 };
1164
1165 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1166 {
1167         struct extent_buffer *buf = NULL;
1168         struct inode *btree_inode = fs_info->btree_inode;
1169
1170         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1171         if (IS_ERR(buf))
1172                 return;
1173         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1174                                  buf, WAIT_NONE, btree_get_extent, 0);
1175         free_extent_buffer(buf);
1176 }
1177
1178 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1179                          int mirror_num, struct extent_buffer **eb)
1180 {
1181         struct extent_buffer *buf = NULL;
1182         struct inode *btree_inode = fs_info->btree_inode;
1183         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1184         int ret;
1185
1186         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1187         if (IS_ERR(buf))
1188                 return 0;
1189
1190         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1191
1192         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1193                                        btree_get_extent, mirror_num);
1194         if (ret) {
1195                 free_extent_buffer(buf);
1196                 return ret;
1197         }
1198
1199         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1200                 free_extent_buffer(buf);
1201                 return -EIO;
1202         } else if (extent_buffer_uptodate(buf)) {
1203                 *eb = buf;
1204         } else {
1205                 free_extent_buffer(buf);
1206         }
1207         return 0;
1208 }
1209
1210 struct extent_buffer *btrfs_find_create_tree_block(
1211                                                 struct btrfs_fs_info *fs_info,
1212                                                 u64 bytenr)
1213 {
1214         if (btrfs_is_testing(fs_info))
1215                 return alloc_test_extent_buffer(fs_info, bytenr);
1216         return alloc_extent_buffer(fs_info, bytenr);
1217 }
1218
1219
1220 int btrfs_write_tree_block(struct extent_buffer *buf)
1221 {
1222         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1223                                         buf->start + buf->len - 1);
1224 }
1225
1226 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1227 {
1228         filemap_fdatawait_range(buf->pages[0]->mapping,
1229                                 buf->start, buf->start + buf->len - 1);
1230 }
1231
1232 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1233                                       u64 parent_transid)
1234 {
1235         struct extent_buffer *buf = NULL;
1236         int ret;
1237
1238         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1239         if (IS_ERR(buf))
1240                 return buf;
1241
1242         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1243         if (ret) {
1244                 free_extent_buffer(buf);
1245                 return ERR_PTR(ret);
1246         }
1247         return buf;
1248
1249 }
1250
1251 void clean_tree_block(struct btrfs_fs_info *fs_info,
1252                       struct extent_buffer *buf)
1253 {
1254         if (btrfs_header_generation(buf) ==
1255             fs_info->running_transaction->transid) {
1256                 btrfs_assert_tree_locked(buf);
1257
1258                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1259                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1260                                              -buf->len,
1261                                              fs_info->dirty_metadata_batch);
1262                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1263                         btrfs_set_lock_blocking(buf);
1264                         clear_extent_buffer_dirty(buf);
1265                 }
1266         }
1267 }
1268
1269 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1270 {
1271         struct btrfs_subvolume_writers *writers;
1272         int ret;
1273
1274         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1275         if (!writers)
1276                 return ERR_PTR(-ENOMEM);
1277
1278         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1279         if (ret < 0) {
1280                 kfree(writers);
1281                 return ERR_PTR(ret);
1282         }
1283
1284         init_waitqueue_head(&writers->wait);
1285         return writers;
1286 }
1287
1288 static void
1289 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1290 {
1291         percpu_counter_destroy(&writers->counter);
1292         kfree(writers);
1293 }
1294
1295 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1296                          u64 objectid)
1297 {
1298         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1299         root->node = NULL;
1300         root->commit_root = NULL;
1301         root->state = 0;
1302         root->orphan_cleanup_state = 0;
1303
1304         root->objectid = objectid;
1305         root->last_trans = 0;
1306         root->highest_objectid = 0;
1307         root->nr_delalloc_inodes = 0;
1308         root->nr_ordered_extents = 0;
1309         root->name = NULL;
1310         root->inode_tree = RB_ROOT;
1311         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1312         root->block_rsv = NULL;
1313         root->orphan_block_rsv = NULL;
1314
1315         INIT_LIST_HEAD(&root->dirty_list);
1316         INIT_LIST_HEAD(&root->root_list);
1317         INIT_LIST_HEAD(&root->delalloc_inodes);
1318         INIT_LIST_HEAD(&root->delalloc_root);
1319         INIT_LIST_HEAD(&root->ordered_extents);
1320         INIT_LIST_HEAD(&root->ordered_root);
1321         INIT_LIST_HEAD(&root->logged_list[0]);
1322         INIT_LIST_HEAD(&root->logged_list[1]);
1323         spin_lock_init(&root->orphan_lock);
1324         spin_lock_init(&root->inode_lock);
1325         spin_lock_init(&root->delalloc_lock);
1326         spin_lock_init(&root->ordered_extent_lock);
1327         spin_lock_init(&root->accounting_lock);
1328         spin_lock_init(&root->log_extents_lock[0]);
1329         spin_lock_init(&root->log_extents_lock[1]);
1330         mutex_init(&root->objectid_mutex);
1331         mutex_init(&root->log_mutex);
1332         mutex_init(&root->ordered_extent_mutex);
1333         mutex_init(&root->delalloc_mutex);
1334         init_waitqueue_head(&root->log_writer_wait);
1335         init_waitqueue_head(&root->log_commit_wait[0]);
1336         init_waitqueue_head(&root->log_commit_wait[1]);
1337         INIT_LIST_HEAD(&root->log_ctxs[0]);
1338         INIT_LIST_HEAD(&root->log_ctxs[1]);
1339         atomic_set(&root->log_commit[0], 0);
1340         atomic_set(&root->log_commit[1], 0);
1341         atomic_set(&root->log_writers, 0);
1342         atomic_set(&root->log_batch, 0);
1343         atomic_set(&root->orphan_inodes, 0);
1344         refcount_set(&root->refs, 1);
1345         atomic_set(&root->will_be_snapshoted, 0);
1346         atomic64_set(&root->qgroup_meta_rsv, 0);
1347         root->log_transid = 0;
1348         root->log_transid_committed = -1;
1349         root->last_log_commit = 0;
1350         if (!dummy)
1351                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1352
1353         memset(&root->root_key, 0, sizeof(root->root_key));
1354         memset(&root->root_item, 0, sizeof(root->root_item));
1355         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1356         if (!dummy)
1357                 root->defrag_trans_start = fs_info->generation;
1358         else
1359                 root->defrag_trans_start = 0;
1360         root->root_key.objectid = objectid;
1361         root->anon_dev = 0;
1362
1363         spin_lock_init(&root->root_item_lock);
1364 }
1365
1366 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1367                 gfp_t flags)
1368 {
1369         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1370         if (root)
1371                 root->fs_info = fs_info;
1372         return root;
1373 }
1374
1375 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1376 /* Should only be used by the testing infrastructure */
1377 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1378 {
1379         struct btrfs_root *root;
1380
1381         if (!fs_info)
1382                 return ERR_PTR(-EINVAL);
1383
1384         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1385         if (!root)
1386                 return ERR_PTR(-ENOMEM);
1387
1388         /* We don't use the stripesize in selftest, set it as sectorsize */
1389         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1390         root->alloc_bytenr = 0;
1391
1392         return root;
1393 }
1394 #endif
1395
1396 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1397                                      struct btrfs_fs_info *fs_info,
1398                                      u64 objectid)
1399 {
1400         struct extent_buffer *leaf;
1401         struct btrfs_root *tree_root = fs_info->tree_root;
1402         struct btrfs_root *root;
1403         struct btrfs_key key;
1404         int ret = 0;
1405         uuid_le uuid;
1406
1407         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1408         if (!root)
1409                 return ERR_PTR(-ENOMEM);
1410
1411         __setup_root(root, fs_info, objectid);
1412         root->root_key.objectid = objectid;
1413         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1414         root->root_key.offset = 0;
1415
1416         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1417         if (IS_ERR(leaf)) {
1418                 ret = PTR_ERR(leaf);
1419                 leaf = NULL;
1420                 goto fail;
1421         }
1422
1423         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1424         btrfs_set_header_bytenr(leaf, leaf->start);
1425         btrfs_set_header_generation(leaf, trans->transid);
1426         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1427         btrfs_set_header_owner(leaf, objectid);
1428         root->node = leaf;
1429
1430         write_extent_buffer_fsid(leaf, fs_info->fsid);
1431         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1432         btrfs_mark_buffer_dirty(leaf);
1433
1434         root->commit_root = btrfs_root_node(root);
1435         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1436
1437         root->root_item.flags = 0;
1438         root->root_item.byte_limit = 0;
1439         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1440         btrfs_set_root_generation(&root->root_item, trans->transid);
1441         btrfs_set_root_level(&root->root_item, 0);
1442         btrfs_set_root_refs(&root->root_item, 1);
1443         btrfs_set_root_used(&root->root_item, leaf->len);
1444         btrfs_set_root_last_snapshot(&root->root_item, 0);
1445         btrfs_set_root_dirid(&root->root_item, 0);
1446         uuid_le_gen(&uuid);
1447         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1448         root->root_item.drop_level = 0;
1449
1450         key.objectid = objectid;
1451         key.type = BTRFS_ROOT_ITEM_KEY;
1452         key.offset = 0;
1453         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1454         if (ret)
1455                 goto fail;
1456
1457         btrfs_tree_unlock(leaf);
1458
1459         return root;
1460
1461 fail:
1462         if (leaf) {
1463                 btrfs_tree_unlock(leaf);
1464                 free_extent_buffer(root->commit_root);
1465                 free_extent_buffer(leaf);
1466         }
1467         kfree(root);
1468
1469         return ERR_PTR(ret);
1470 }
1471
1472 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1473                                          struct btrfs_fs_info *fs_info)
1474 {
1475         struct btrfs_root *root;
1476         struct extent_buffer *leaf;
1477
1478         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1479         if (!root)
1480                 return ERR_PTR(-ENOMEM);
1481
1482         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1483
1484         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1485         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1486         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1487
1488         /*
1489          * DON'T set REF_COWS for log trees
1490          *
1491          * log trees do not get reference counted because they go away
1492          * before a real commit is actually done.  They do store pointers
1493          * to file data extents, and those reference counts still get
1494          * updated (along with back refs to the log tree).
1495          */
1496
1497         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1498                         NULL, 0, 0, 0);
1499         if (IS_ERR(leaf)) {
1500                 kfree(root);
1501                 return ERR_CAST(leaf);
1502         }
1503
1504         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1505         btrfs_set_header_bytenr(leaf, leaf->start);
1506         btrfs_set_header_generation(leaf, trans->transid);
1507         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1508         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1509         root->node = leaf;
1510
1511         write_extent_buffer_fsid(root->node, fs_info->fsid);
1512         btrfs_mark_buffer_dirty(root->node);
1513         btrfs_tree_unlock(root->node);
1514         return root;
1515 }
1516
1517 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1518                              struct btrfs_fs_info *fs_info)
1519 {
1520         struct btrfs_root *log_root;
1521
1522         log_root = alloc_log_tree(trans, fs_info);
1523         if (IS_ERR(log_root))
1524                 return PTR_ERR(log_root);
1525         WARN_ON(fs_info->log_root_tree);
1526         fs_info->log_root_tree = log_root;
1527         return 0;
1528 }
1529
1530 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1531                        struct btrfs_root *root)
1532 {
1533         struct btrfs_fs_info *fs_info = root->fs_info;
1534         struct btrfs_root *log_root;
1535         struct btrfs_inode_item *inode_item;
1536
1537         log_root = alloc_log_tree(trans, fs_info);
1538         if (IS_ERR(log_root))
1539                 return PTR_ERR(log_root);
1540
1541         log_root->last_trans = trans->transid;
1542         log_root->root_key.offset = root->root_key.objectid;
1543
1544         inode_item = &log_root->root_item.inode;
1545         btrfs_set_stack_inode_generation(inode_item, 1);
1546         btrfs_set_stack_inode_size(inode_item, 3);
1547         btrfs_set_stack_inode_nlink(inode_item, 1);
1548         btrfs_set_stack_inode_nbytes(inode_item,
1549                                      fs_info->nodesize);
1550         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1551
1552         btrfs_set_root_node(&log_root->root_item, log_root->node);
1553
1554         WARN_ON(root->log_root);
1555         root->log_root = log_root;
1556         root->log_transid = 0;
1557         root->log_transid_committed = -1;
1558         root->last_log_commit = 0;
1559         return 0;
1560 }
1561
1562 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1563                                                struct btrfs_key *key)
1564 {
1565         struct btrfs_root *root;
1566         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1567         struct btrfs_path *path;
1568         u64 generation;
1569         int ret;
1570
1571         path = btrfs_alloc_path();
1572         if (!path)
1573                 return ERR_PTR(-ENOMEM);
1574
1575         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1576         if (!root) {
1577                 ret = -ENOMEM;
1578                 goto alloc_fail;
1579         }
1580
1581         __setup_root(root, fs_info, key->objectid);
1582
1583         ret = btrfs_find_root(tree_root, key, path,
1584                               &root->root_item, &root->root_key);
1585         if (ret) {
1586                 if (ret > 0)
1587                         ret = -ENOENT;
1588                 goto find_fail;
1589         }
1590
1591         generation = btrfs_root_generation(&root->root_item);
1592         root->node = read_tree_block(fs_info,
1593                                      btrfs_root_bytenr(&root->root_item),
1594                                      generation);
1595         if (IS_ERR(root->node)) {
1596                 ret = PTR_ERR(root->node);
1597                 goto find_fail;
1598         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1599                 ret = -EIO;
1600                 free_extent_buffer(root->node);
1601                 goto find_fail;
1602         }
1603         root->commit_root = btrfs_root_node(root);
1604 out:
1605         btrfs_free_path(path);
1606         return root;
1607
1608 find_fail:
1609         kfree(root);
1610 alloc_fail:
1611         root = ERR_PTR(ret);
1612         goto out;
1613 }
1614
1615 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1616                                       struct btrfs_key *location)
1617 {
1618         struct btrfs_root *root;
1619
1620         root = btrfs_read_tree_root(tree_root, location);
1621         if (IS_ERR(root))
1622                 return root;
1623
1624         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1625                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1626                 btrfs_check_and_init_root_item(&root->root_item);
1627         }
1628
1629         return root;
1630 }
1631
1632 int btrfs_init_fs_root(struct btrfs_root *root)
1633 {
1634         int ret;
1635         struct btrfs_subvolume_writers *writers;
1636
1637         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1638         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1639                                         GFP_NOFS);
1640         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1641                 ret = -ENOMEM;
1642                 goto fail;
1643         }
1644
1645         writers = btrfs_alloc_subvolume_writers();
1646         if (IS_ERR(writers)) {
1647                 ret = PTR_ERR(writers);
1648                 goto fail;
1649         }
1650         root->subv_writers = writers;
1651
1652         btrfs_init_free_ino_ctl(root);
1653         spin_lock_init(&root->ino_cache_lock);
1654         init_waitqueue_head(&root->ino_cache_wait);
1655
1656         ret = get_anon_bdev(&root->anon_dev);
1657         if (ret)
1658                 goto fail;
1659
1660         mutex_lock(&root->objectid_mutex);
1661         ret = btrfs_find_highest_objectid(root,
1662                                         &root->highest_objectid);
1663         if (ret) {
1664                 mutex_unlock(&root->objectid_mutex);
1665                 goto fail;
1666         }
1667
1668         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1669
1670         mutex_unlock(&root->objectid_mutex);
1671
1672         return 0;
1673 fail:
1674         /* the caller is responsible to call free_fs_root */
1675         return ret;
1676 }
1677
1678 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1679                                         u64 root_id)
1680 {
1681         struct btrfs_root *root;
1682
1683         spin_lock(&fs_info->fs_roots_radix_lock);
1684         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1685                                  (unsigned long)root_id);
1686         spin_unlock(&fs_info->fs_roots_radix_lock);
1687         return root;
1688 }
1689
1690 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1691                          struct btrfs_root *root)
1692 {
1693         int ret;
1694
1695         ret = radix_tree_preload(GFP_NOFS);
1696         if (ret)
1697                 return ret;
1698
1699         spin_lock(&fs_info->fs_roots_radix_lock);
1700         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1701                                 (unsigned long)root->root_key.objectid,
1702                                 root);
1703         if (ret == 0)
1704                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1705         spin_unlock(&fs_info->fs_roots_radix_lock);
1706         radix_tree_preload_end();
1707
1708         return ret;
1709 }
1710
1711 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1712                                      struct btrfs_key *location,
1713                                      bool check_ref)
1714 {
1715         struct btrfs_root *root;
1716         struct btrfs_path *path;
1717         struct btrfs_key key;
1718         int ret;
1719
1720         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1721                 return fs_info->tree_root;
1722         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1723                 return fs_info->extent_root;
1724         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1725                 return fs_info->chunk_root;
1726         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1727                 return fs_info->dev_root;
1728         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1729                 return fs_info->csum_root;
1730         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1731                 return fs_info->quota_root ? fs_info->quota_root :
1732                                              ERR_PTR(-ENOENT);
1733         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1734                 return fs_info->uuid_root ? fs_info->uuid_root :
1735                                             ERR_PTR(-ENOENT);
1736         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1737                 return fs_info->free_space_root ? fs_info->free_space_root :
1738                                                   ERR_PTR(-ENOENT);
1739 again:
1740         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1741         if (root) {
1742                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1743                         return ERR_PTR(-ENOENT);
1744                 return root;
1745         }
1746
1747         root = btrfs_read_fs_root(fs_info->tree_root, location);
1748         if (IS_ERR(root))
1749                 return root;
1750
1751         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1752                 ret = -ENOENT;
1753                 goto fail;
1754         }
1755
1756         ret = btrfs_init_fs_root(root);
1757         if (ret)
1758                 goto fail;
1759
1760         path = btrfs_alloc_path();
1761         if (!path) {
1762                 ret = -ENOMEM;
1763                 goto fail;
1764         }
1765         key.objectid = BTRFS_ORPHAN_OBJECTID;
1766         key.type = BTRFS_ORPHAN_ITEM_KEY;
1767         key.offset = location->objectid;
1768
1769         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1770         btrfs_free_path(path);
1771         if (ret < 0)
1772                 goto fail;
1773         if (ret == 0)
1774                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1775
1776         ret = btrfs_insert_fs_root(fs_info, root);
1777         if (ret) {
1778                 if (ret == -EEXIST) {
1779                         free_fs_root(root);
1780                         goto again;
1781                 }
1782                 goto fail;
1783         }
1784         return root;
1785 fail:
1786         free_fs_root(root);
1787         return ERR_PTR(ret);
1788 }
1789
1790 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1791 {
1792         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1793         int ret = 0;
1794         struct btrfs_device *device;
1795         struct backing_dev_info *bdi;
1796
1797         rcu_read_lock();
1798         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1799                 if (!device->bdev)
1800                         continue;
1801                 bdi = device->bdev->bd_bdi;
1802                 if (bdi_congested(bdi, bdi_bits)) {
1803                         ret = 1;
1804                         break;
1805                 }
1806         }
1807         rcu_read_unlock();
1808         return ret;
1809 }
1810
1811 /*
1812  * called by the kthread helper functions to finally call the bio end_io
1813  * functions.  This is where read checksum verification actually happens
1814  */
1815 static void end_workqueue_fn(struct btrfs_work *work)
1816 {
1817         struct bio *bio;
1818         struct btrfs_end_io_wq *end_io_wq;
1819
1820         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1821         bio = end_io_wq->bio;
1822
1823         bio->bi_error = end_io_wq->error;
1824         bio->bi_private = end_io_wq->private;
1825         bio->bi_end_io = end_io_wq->end_io;
1826         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1827         bio_endio(bio);
1828 }
1829
1830 static int cleaner_kthread(void *arg)
1831 {
1832         struct btrfs_root *root = arg;
1833         struct btrfs_fs_info *fs_info = root->fs_info;
1834         int again;
1835         struct btrfs_trans_handle *trans;
1836
1837         do {
1838                 again = 0;
1839
1840                 /* Make the cleaner go to sleep early. */
1841                 if (btrfs_need_cleaner_sleep(fs_info))
1842                         goto sleep;
1843
1844                 /*
1845                  * Do not do anything if we might cause open_ctree() to block
1846                  * before we have finished mounting the filesystem.
1847                  */
1848                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1849                         goto sleep;
1850
1851                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1852                         goto sleep;
1853
1854                 /*
1855                  * Avoid the problem that we change the status of the fs
1856                  * during the above check and trylock.
1857                  */
1858                 if (btrfs_need_cleaner_sleep(fs_info)) {
1859                         mutex_unlock(&fs_info->cleaner_mutex);
1860                         goto sleep;
1861                 }
1862
1863                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1864                 btrfs_run_delayed_iputs(fs_info);
1865                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1866
1867                 again = btrfs_clean_one_deleted_snapshot(root);
1868                 mutex_unlock(&fs_info->cleaner_mutex);
1869
1870                 /*
1871                  * The defragger has dealt with the R/O remount and umount,
1872                  * needn't do anything special here.
1873                  */
1874                 btrfs_run_defrag_inodes(fs_info);
1875
1876                 /*
1877                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1878                  * with relocation (btrfs_relocate_chunk) and relocation
1879                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1880                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1881                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1882                  * unused block groups.
1883                  */
1884                 btrfs_delete_unused_bgs(fs_info);
1885 sleep:
1886                 if (!again) {
1887                         set_current_state(TASK_INTERRUPTIBLE);
1888                         if (!kthread_should_stop())
1889                                 schedule();
1890                         __set_current_state(TASK_RUNNING);
1891                 }
1892         } while (!kthread_should_stop());
1893
1894         /*
1895          * Transaction kthread is stopped before us and wakes us up.
1896          * However we might have started a new transaction and COWed some
1897          * tree blocks when deleting unused block groups for example. So
1898          * make sure we commit the transaction we started to have a clean
1899          * shutdown when evicting the btree inode - if it has dirty pages
1900          * when we do the final iput() on it, eviction will trigger a
1901          * writeback for it which will fail with null pointer dereferences
1902          * since work queues and other resources were already released and
1903          * destroyed by the time the iput/eviction/writeback is made.
1904          */
1905         trans = btrfs_attach_transaction(root);
1906         if (IS_ERR(trans)) {
1907                 if (PTR_ERR(trans) != -ENOENT)
1908                         btrfs_err(fs_info,
1909                                   "cleaner transaction attach returned %ld",
1910                                   PTR_ERR(trans));
1911         } else {
1912                 int ret;
1913
1914                 ret = btrfs_commit_transaction(trans);
1915                 if (ret)
1916                         btrfs_err(fs_info,
1917                                   "cleaner open transaction commit returned %d",
1918                                   ret);
1919         }
1920
1921         return 0;
1922 }
1923
1924 static int transaction_kthread(void *arg)
1925 {
1926         struct btrfs_root *root = arg;
1927         struct btrfs_fs_info *fs_info = root->fs_info;
1928         struct btrfs_trans_handle *trans;
1929         struct btrfs_transaction *cur;
1930         u64 transid;
1931         unsigned long now;
1932         unsigned long delay;
1933         bool cannot_commit;
1934
1935         do {
1936                 cannot_commit = false;
1937                 delay = HZ * fs_info->commit_interval;
1938                 mutex_lock(&fs_info->transaction_kthread_mutex);
1939
1940                 spin_lock(&fs_info->trans_lock);
1941                 cur = fs_info->running_transaction;
1942                 if (!cur) {
1943                         spin_unlock(&fs_info->trans_lock);
1944                         goto sleep;
1945                 }
1946
1947                 now = get_seconds();
1948                 if (cur->state < TRANS_STATE_BLOCKED &&
1949                     (now < cur->start_time ||
1950                      now - cur->start_time < fs_info->commit_interval)) {
1951                         spin_unlock(&fs_info->trans_lock);
1952                         delay = HZ * 5;
1953                         goto sleep;
1954                 }
1955                 transid = cur->transid;
1956                 spin_unlock(&fs_info->trans_lock);
1957
1958                 /* If the file system is aborted, this will always fail. */
1959                 trans = btrfs_attach_transaction(root);
1960                 if (IS_ERR(trans)) {
1961                         if (PTR_ERR(trans) != -ENOENT)
1962                                 cannot_commit = true;
1963                         goto sleep;
1964                 }
1965                 if (transid == trans->transid) {
1966                         btrfs_commit_transaction(trans);
1967                 } else {
1968                         btrfs_end_transaction(trans);
1969                 }
1970 sleep:
1971                 wake_up_process(fs_info->cleaner_kthread);
1972                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1973
1974                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1975                                       &fs_info->fs_state)))
1976                         btrfs_cleanup_transaction(fs_info);
1977                 set_current_state(TASK_INTERRUPTIBLE);
1978                 if (!kthread_should_stop() &&
1979                                 (!btrfs_transaction_blocked(fs_info) ||
1980                                  cannot_commit))
1981                         schedule_timeout(delay);
1982                 __set_current_state(TASK_RUNNING);
1983         } while (!kthread_should_stop());
1984         return 0;
1985 }
1986
1987 /*
1988  * this will find the highest generation in the array of
1989  * root backups.  The index of the highest array is returned,
1990  * or -1 if we can't find anything.
1991  *
1992  * We check to make sure the array is valid by comparing the
1993  * generation of the latest  root in the array with the generation
1994  * in the super block.  If they don't match we pitch it.
1995  */
1996 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1997 {
1998         u64 cur;
1999         int newest_index = -1;
2000         struct btrfs_root_backup *root_backup;
2001         int i;
2002
2003         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2004                 root_backup = info->super_copy->super_roots + i;
2005                 cur = btrfs_backup_tree_root_gen(root_backup);
2006                 if (cur == newest_gen)
2007                         newest_index = i;
2008         }
2009
2010         /* check to see if we actually wrapped around */
2011         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2012                 root_backup = info->super_copy->super_roots;
2013                 cur = btrfs_backup_tree_root_gen(root_backup);
2014                 if (cur == newest_gen)
2015                         newest_index = 0;
2016         }
2017         return newest_index;
2018 }
2019
2020
2021 /*
2022  * find the oldest backup so we know where to store new entries
2023  * in the backup array.  This will set the backup_root_index
2024  * field in the fs_info struct
2025  */
2026 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2027                                      u64 newest_gen)
2028 {
2029         int newest_index = -1;
2030
2031         newest_index = find_newest_super_backup(info, newest_gen);
2032         /* if there was garbage in there, just move along */
2033         if (newest_index == -1) {
2034                 info->backup_root_index = 0;
2035         } else {
2036                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2037         }
2038 }
2039
2040 /*
2041  * copy all the root pointers into the super backup array.
2042  * this will bump the backup pointer by one when it is
2043  * done
2044  */
2045 static void backup_super_roots(struct btrfs_fs_info *info)
2046 {
2047         int next_backup;
2048         struct btrfs_root_backup *root_backup;
2049         int last_backup;
2050
2051         next_backup = info->backup_root_index;
2052         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2053                 BTRFS_NUM_BACKUP_ROOTS;
2054
2055         /*
2056          * just overwrite the last backup if we're at the same generation
2057          * this happens only at umount
2058          */
2059         root_backup = info->super_for_commit->super_roots + last_backup;
2060         if (btrfs_backup_tree_root_gen(root_backup) ==
2061             btrfs_header_generation(info->tree_root->node))
2062                 next_backup = last_backup;
2063
2064         root_backup = info->super_for_commit->super_roots + next_backup;
2065
2066         /*
2067          * make sure all of our padding and empty slots get zero filled
2068          * regardless of which ones we use today
2069          */
2070         memset(root_backup, 0, sizeof(*root_backup));
2071
2072         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2073
2074         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2075         btrfs_set_backup_tree_root_gen(root_backup,
2076                                btrfs_header_generation(info->tree_root->node));
2077
2078         btrfs_set_backup_tree_root_level(root_backup,
2079                                btrfs_header_level(info->tree_root->node));
2080
2081         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2082         btrfs_set_backup_chunk_root_gen(root_backup,
2083                                btrfs_header_generation(info->chunk_root->node));
2084         btrfs_set_backup_chunk_root_level(root_backup,
2085                                btrfs_header_level(info->chunk_root->node));
2086
2087         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2088         btrfs_set_backup_extent_root_gen(root_backup,
2089                                btrfs_header_generation(info->extent_root->node));
2090         btrfs_set_backup_extent_root_level(root_backup,
2091                                btrfs_header_level(info->extent_root->node));
2092
2093         /*
2094          * we might commit during log recovery, which happens before we set
2095          * the fs_root.  Make sure it is valid before we fill it in.
2096          */
2097         if (info->fs_root && info->fs_root->node) {
2098                 btrfs_set_backup_fs_root(root_backup,
2099                                          info->fs_root->node->start);
2100                 btrfs_set_backup_fs_root_gen(root_backup,
2101                                btrfs_header_generation(info->fs_root->node));
2102                 btrfs_set_backup_fs_root_level(root_backup,
2103                                btrfs_header_level(info->fs_root->node));
2104         }
2105
2106         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2107         btrfs_set_backup_dev_root_gen(root_backup,
2108                                btrfs_header_generation(info->dev_root->node));
2109         btrfs_set_backup_dev_root_level(root_backup,
2110                                        btrfs_header_level(info->dev_root->node));
2111
2112         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2113         btrfs_set_backup_csum_root_gen(root_backup,
2114                                btrfs_header_generation(info->csum_root->node));
2115         btrfs_set_backup_csum_root_level(root_backup,
2116                                btrfs_header_level(info->csum_root->node));
2117
2118         btrfs_set_backup_total_bytes(root_backup,
2119                              btrfs_super_total_bytes(info->super_copy));
2120         btrfs_set_backup_bytes_used(root_backup,
2121                              btrfs_super_bytes_used(info->super_copy));
2122         btrfs_set_backup_num_devices(root_backup,
2123                              btrfs_super_num_devices(info->super_copy));
2124
2125         /*
2126          * if we don't copy this out to the super_copy, it won't get remembered
2127          * for the next commit
2128          */
2129         memcpy(&info->super_copy->super_roots,
2130                &info->super_for_commit->super_roots,
2131                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2132 }
2133
2134 /*
2135  * this copies info out of the root backup array and back into
2136  * the in-memory super block.  It is meant to help iterate through
2137  * the array, so you send it the number of backups you've already
2138  * tried and the last backup index you used.
2139  *
2140  * this returns -1 when it has tried all the backups
2141  */
2142 static noinline int next_root_backup(struct btrfs_fs_info *info,
2143                                      struct btrfs_super_block *super,
2144                                      int *num_backups_tried, int *backup_index)
2145 {
2146         struct btrfs_root_backup *root_backup;
2147         int newest = *backup_index;
2148
2149         if (*num_backups_tried == 0) {
2150                 u64 gen = btrfs_super_generation(super);
2151
2152                 newest = find_newest_super_backup(info, gen);
2153                 if (newest == -1)
2154                         return -1;
2155
2156                 *backup_index = newest;
2157                 *num_backups_tried = 1;
2158         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2159                 /* we've tried all the backups, all done */
2160                 return -1;
2161         } else {
2162                 /* jump to the next oldest backup */
2163                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2164                         BTRFS_NUM_BACKUP_ROOTS;
2165                 *backup_index = newest;
2166                 *num_backups_tried += 1;
2167         }
2168         root_backup = super->super_roots + newest;
2169
2170         btrfs_set_super_generation(super,
2171                                    btrfs_backup_tree_root_gen(root_backup));
2172         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2173         btrfs_set_super_root_level(super,
2174                                    btrfs_backup_tree_root_level(root_backup));
2175         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2176
2177         /*
2178          * fixme: the total bytes and num_devices need to match or we should
2179          * need a fsck
2180          */
2181         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2182         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2183         return 0;
2184 }
2185
2186 /* helper to cleanup workers */
2187 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2188 {
2189         btrfs_destroy_workqueue(fs_info->fixup_workers);
2190         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2191         btrfs_destroy_workqueue(fs_info->workers);
2192         btrfs_destroy_workqueue(fs_info->endio_workers);
2193         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2194         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2195         btrfs_destroy_workqueue(fs_info->rmw_workers);
2196         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2197         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2198         btrfs_destroy_workqueue(fs_info->submit_workers);
2199         btrfs_destroy_workqueue(fs_info->delayed_workers);
2200         btrfs_destroy_workqueue(fs_info->caching_workers);
2201         btrfs_destroy_workqueue(fs_info->readahead_workers);
2202         btrfs_destroy_workqueue(fs_info->flush_workers);
2203         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2204         btrfs_destroy_workqueue(fs_info->extent_workers);
2205         /*
2206          * Now that all other work queues are destroyed, we can safely destroy
2207          * the queues used for metadata I/O, since tasks from those other work
2208          * queues can do metadata I/O operations.
2209          */
2210         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2211         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2212 }
2213
2214 static void free_root_extent_buffers(struct btrfs_root *root)
2215 {
2216         if (root) {
2217                 free_extent_buffer(root->node);
2218                 free_extent_buffer(root->commit_root);
2219                 root->node = NULL;
2220                 root->commit_root = NULL;
2221         }
2222 }
2223
2224 /* helper to cleanup tree roots */
2225 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2226 {
2227         free_root_extent_buffers(info->tree_root);
2228
2229         free_root_extent_buffers(info->dev_root);
2230         free_root_extent_buffers(info->extent_root);
2231         free_root_extent_buffers(info->csum_root);
2232         free_root_extent_buffers(info->quota_root);
2233         free_root_extent_buffers(info->uuid_root);
2234         if (chunk_root)
2235                 free_root_extent_buffers(info->chunk_root);
2236         free_root_extent_buffers(info->free_space_root);
2237 }
2238
2239 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2240 {
2241         int ret;
2242         struct btrfs_root *gang[8];
2243         int i;
2244
2245         while (!list_empty(&fs_info->dead_roots)) {
2246                 gang[0] = list_entry(fs_info->dead_roots.next,
2247                                      struct btrfs_root, root_list);
2248                 list_del(&gang[0]->root_list);
2249
2250                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2251                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2252                 } else {
2253                         free_extent_buffer(gang[0]->node);
2254                         free_extent_buffer(gang[0]->commit_root);
2255                         btrfs_put_fs_root(gang[0]);
2256                 }
2257         }
2258
2259         while (1) {
2260                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2261                                              (void **)gang, 0,
2262                                              ARRAY_SIZE(gang));
2263                 if (!ret)
2264                         break;
2265                 for (i = 0; i < ret; i++)
2266                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2267         }
2268
2269         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2270                 btrfs_free_log_root_tree(NULL, fs_info);
2271                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2272         }
2273 }
2274
2275 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2276 {
2277         mutex_init(&fs_info->scrub_lock);
2278         atomic_set(&fs_info->scrubs_running, 0);
2279         atomic_set(&fs_info->scrub_pause_req, 0);
2280         atomic_set(&fs_info->scrubs_paused, 0);
2281         atomic_set(&fs_info->scrub_cancel_req, 0);
2282         init_waitqueue_head(&fs_info->scrub_pause_wait);
2283         fs_info->scrub_workers_refcnt = 0;
2284 }
2285
2286 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2287 {
2288         spin_lock_init(&fs_info->balance_lock);
2289         mutex_init(&fs_info->balance_mutex);
2290         atomic_set(&fs_info->balance_running, 0);
2291         atomic_set(&fs_info->balance_pause_req, 0);
2292         atomic_set(&fs_info->balance_cancel_req, 0);
2293         fs_info->balance_ctl = NULL;
2294         init_waitqueue_head(&fs_info->balance_wait_q);
2295 }
2296
2297 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2298 {
2299         struct inode *inode = fs_info->btree_inode;
2300
2301         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2302         set_nlink(inode, 1);
2303         /*
2304          * we set the i_size on the btree inode to the max possible int.
2305          * the real end of the address space is determined by all of
2306          * the devices in the system
2307          */
2308         inode->i_size = OFFSET_MAX;
2309         inode->i_mapping->a_ops = &btree_aops;
2310
2311         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2312         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2313         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2314         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2315
2316         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2317
2318         BTRFS_I(inode)->root = fs_info->tree_root;
2319         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2320         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2321         btrfs_insert_inode_hash(inode);
2322 }
2323
2324 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2325 {
2326         fs_info->dev_replace.lock_owner = 0;
2327         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2328         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2329         rwlock_init(&fs_info->dev_replace.lock);
2330         atomic_set(&fs_info->dev_replace.read_locks, 0);
2331         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2332         init_waitqueue_head(&fs_info->replace_wait);
2333         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2334 }
2335
2336 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2337 {
2338         spin_lock_init(&fs_info->qgroup_lock);
2339         mutex_init(&fs_info->qgroup_ioctl_lock);
2340         fs_info->qgroup_tree = RB_ROOT;
2341         fs_info->qgroup_op_tree = RB_ROOT;
2342         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2343         fs_info->qgroup_seq = 1;
2344         fs_info->qgroup_ulist = NULL;
2345         fs_info->qgroup_rescan_running = false;
2346         mutex_init(&fs_info->qgroup_rescan_lock);
2347 }
2348
2349 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2350                 struct btrfs_fs_devices *fs_devices)
2351 {
2352         int max_active = fs_info->thread_pool_size;
2353         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2354
2355         fs_info->workers =
2356                 btrfs_alloc_workqueue(fs_info, "worker",
2357                                       flags | WQ_HIGHPRI, max_active, 16);
2358
2359         fs_info->delalloc_workers =
2360                 btrfs_alloc_workqueue(fs_info, "delalloc",
2361                                       flags, max_active, 2);
2362
2363         fs_info->flush_workers =
2364                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2365                                       flags, max_active, 0);
2366
2367         fs_info->caching_workers =
2368                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2369
2370         /*
2371          * a higher idle thresh on the submit workers makes it much more
2372          * likely that bios will be send down in a sane order to the
2373          * devices
2374          */
2375         fs_info->submit_workers =
2376                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2377                                       min_t(u64, fs_devices->num_devices,
2378                                             max_active), 64);
2379
2380         fs_info->fixup_workers =
2381                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2382
2383         /*
2384          * endios are largely parallel and should have a very
2385          * low idle thresh
2386          */
2387         fs_info->endio_workers =
2388                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2389         fs_info->endio_meta_workers =
2390                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2391                                       max_active, 4);
2392         fs_info->endio_meta_write_workers =
2393                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2394                                       max_active, 2);
2395         fs_info->endio_raid56_workers =
2396                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2397                                       max_active, 4);
2398         fs_info->endio_repair_workers =
2399                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2400         fs_info->rmw_workers =
2401                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2402         fs_info->endio_write_workers =
2403                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2404                                       max_active, 2);
2405         fs_info->endio_freespace_worker =
2406                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2407                                       max_active, 0);
2408         fs_info->delayed_workers =
2409                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2410                                       max_active, 0);
2411         fs_info->readahead_workers =
2412                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2413                                       max_active, 2);
2414         fs_info->qgroup_rescan_workers =
2415                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2416         fs_info->extent_workers =
2417                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2418                                       min_t(u64, fs_devices->num_devices,
2419                                             max_active), 8);
2420
2421         if (!(fs_info->workers && fs_info->delalloc_workers &&
2422               fs_info->submit_workers && fs_info->flush_workers &&
2423               fs_info->endio_workers && fs_info->endio_meta_workers &&
2424               fs_info->endio_meta_write_workers &&
2425               fs_info->endio_repair_workers &&
2426               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2427               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2428               fs_info->caching_workers && fs_info->readahead_workers &&
2429               fs_info->fixup_workers && fs_info->delayed_workers &&
2430               fs_info->extent_workers &&
2431               fs_info->qgroup_rescan_workers)) {
2432                 return -ENOMEM;
2433         }
2434
2435         return 0;
2436 }
2437
2438 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2439                             struct btrfs_fs_devices *fs_devices)
2440 {
2441         int ret;
2442         struct btrfs_root *log_tree_root;
2443         struct btrfs_super_block *disk_super = fs_info->super_copy;
2444         u64 bytenr = btrfs_super_log_root(disk_super);
2445
2446         if (fs_devices->rw_devices == 0) {
2447                 btrfs_warn(fs_info, "log replay required on RO media");
2448                 return -EIO;
2449         }
2450
2451         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2452         if (!log_tree_root)
2453                 return -ENOMEM;
2454
2455         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2456
2457         log_tree_root->node = read_tree_block(fs_info, bytenr,
2458                                               fs_info->generation + 1);
2459         if (IS_ERR(log_tree_root->node)) {
2460                 btrfs_warn(fs_info, "failed to read log tree");
2461                 ret = PTR_ERR(log_tree_root->node);
2462                 kfree(log_tree_root);
2463                 return ret;
2464         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2465                 btrfs_err(fs_info, "failed to read log tree");
2466                 free_extent_buffer(log_tree_root->node);
2467                 kfree(log_tree_root);
2468                 return -EIO;
2469         }
2470         /* returns with log_tree_root freed on success */
2471         ret = btrfs_recover_log_trees(log_tree_root);
2472         if (ret) {
2473                 btrfs_handle_fs_error(fs_info, ret,
2474                                       "Failed to recover log tree");
2475                 free_extent_buffer(log_tree_root->node);
2476                 kfree(log_tree_root);
2477                 return ret;
2478         }
2479
2480         if (fs_info->sb->s_flags & MS_RDONLY) {
2481                 ret = btrfs_commit_super(fs_info);
2482                 if (ret)
2483                         return ret;
2484         }
2485
2486         return 0;
2487 }
2488
2489 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2490 {
2491         struct btrfs_root *tree_root = fs_info->tree_root;
2492         struct btrfs_root *root;
2493         struct btrfs_key location;
2494         int ret;
2495
2496         BUG_ON(!fs_info->tree_root);
2497
2498         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2499         location.type = BTRFS_ROOT_ITEM_KEY;
2500         location.offset = 0;
2501
2502         root = btrfs_read_tree_root(tree_root, &location);
2503         if (IS_ERR(root))
2504                 return PTR_ERR(root);
2505         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2506         fs_info->extent_root = root;
2507
2508         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2509         root = btrfs_read_tree_root(tree_root, &location);
2510         if (IS_ERR(root))
2511                 return PTR_ERR(root);
2512         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2513         fs_info->dev_root = root;
2514         btrfs_init_devices_late(fs_info);
2515
2516         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2517         root = btrfs_read_tree_root(tree_root, &location);
2518         if (IS_ERR(root))
2519                 return PTR_ERR(root);
2520         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2521         fs_info->csum_root = root;
2522
2523         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2524         root = btrfs_read_tree_root(tree_root, &location);
2525         if (!IS_ERR(root)) {
2526                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2527                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2528                 fs_info->quota_root = root;
2529         }
2530
2531         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2532         root = btrfs_read_tree_root(tree_root, &location);
2533         if (IS_ERR(root)) {
2534                 ret = PTR_ERR(root);
2535                 if (ret != -ENOENT)
2536                         return ret;
2537         } else {
2538                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2539                 fs_info->uuid_root = root;
2540         }
2541
2542         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2543                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2544                 root = btrfs_read_tree_root(tree_root, &location);
2545                 if (IS_ERR(root))
2546                         return PTR_ERR(root);
2547                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2548                 fs_info->free_space_root = root;
2549         }
2550
2551         return 0;
2552 }
2553
2554 int open_ctree(struct super_block *sb,
2555                struct btrfs_fs_devices *fs_devices,
2556                char *options)
2557 {
2558         u32 sectorsize;
2559         u32 nodesize;
2560         u32 stripesize;
2561         u64 generation;
2562         u64 features;
2563         struct btrfs_key location;
2564         struct buffer_head *bh;
2565         struct btrfs_super_block *disk_super;
2566         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2567         struct btrfs_root *tree_root;
2568         struct btrfs_root *chunk_root;
2569         int ret;
2570         int err = -EINVAL;
2571         int num_backups_tried = 0;
2572         int backup_index = 0;
2573         int max_active;
2574         int clear_free_space_tree = 0;
2575
2576         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2577         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2578         if (!tree_root || !chunk_root) {
2579                 err = -ENOMEM;
2580                 goto fail;
2581         }
2582
2583         ret = init_srcu_struct(&fs_info->subvol_srcu);
2584         if (ret) {
2585                 err = ret;
2586                 goto fail;
2587         }
2588
2589         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2590         if (ret) {
2591                 err = ret;
2592                 goto fail_srcu;
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->tree_mod_seq_lock);
2630         spin_lock_init(&fs_info->super_lock);
2631         spin_lock_init(&fs_info->qgroup_op_lock);
2632         spin_lock_init(&fs_info->buffer_lock);
2633         spin_lock_init(&fs_info->unused_bgs_lock);
2634         rwlock_init(&fs_info->tree_mod_log_lock);
2635         mutex_init(&fs_info->unused_bg_unpin_mutex);
2636         mutex_init(&fs_info->delete_unused_bgs_mutex);
2637         mutex_init(&fs_info->reloc_mutex);
2638         mutex_init(&fs_info->delalloc_root_mutex);
2639         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2640         seqlock_init(&fs_info->profiles_lock);
2641
2642         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2643         INIT_LIST_HEAD(&fs_info->space_info);
2644         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2645         INIT_LIST_HEAD(&fs_info->unused_bgs);
2646         btrfs_mapping_init(&fs_info->mapping_tree);
2647         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2648                              BTRFS_BLOCK_RSV_GLOBAL);
2649         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2650                              BTRFS_BLOCK_RSV_DELALLOC);
2651         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2652         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2653         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2654         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2655                              BTRFS_BLOCK_RSV_DELOPS);
2656         atomic_set(&fs_info->nr_async_submits, 0);
2657         atomic_set(&fs_info->async_delalloc_pages, 0);
2658         atomic_set(&fs_info->async_submit_draining, 0);
2659         atomic_set(&fs_info->nr_async_bios, 0);
2660         atomic_set(&fs_info->defrag_running, 0);
2661         atomic_set(&fs_info->qgroup_op_seq, 0);
2662         atomic_set(&fs_info->reada_works_cnt, 0);
2663         atomic64_set(&fs_info->tree_mod_seq, 0);
2664         fs_info->sb = sb;
2665         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2666         fs_info->metadata_ratio = 0;
2667         fs_info->defrag_inodes = RB_ROOT;
2668         atomic64_set(&fs_info->free_chunk_space, 0);
2669         fs_info->tree_mod_log = RB_ROOT;
2670         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2671         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2672         /* readahead state */
2673         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2674         spin_lock_init(&fs_info->reada_lock);
2675
2676         fs_info->thread_pool_size = min_t(unsigned long,
2677                                           num_online_cpus() + 2, 8);
2678
2679         INIT_LIST_HEAD(&fs_info->ordered_roots);
2680         spin_lock_init(&fs_info->ordered_root_lock);
2681         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2682                                         GFP_KERNEL);
2683         if (!fs_info->delayed_root) {
2684                 err = -ENOMEM;
2685                 goto fail_iput;
2686         }
2687         btrfs_init_delayed_root(fs_info->delayed_root);
2688
2689         btrfs_init_scrub(fs_info);
2690 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2691         fs_info->check_integrity_print_mask = 0;
2692 #endif
2693         btrfs_init_balance(fs_info);
2694         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2695
2696         sb->s_blocksize = 4096;
2697         sb->s_blocksize_bits = blksize_bits(4096);
2698
2699         btrfs_init_btree_inode(fs_info);
2700
2701         spin_lock_init(&fs_info->block_group_cache_lock);
2702         fs_info->block_group_cache_tree = RB_ROOT;
2703         fs_info->first_logical_byte = (u64)-1;
2704
2705         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2706         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2707         fs_info->pinned_extents = &fs_info->freed_extents[0];
2708         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2709
2710         mutex_init(&fs_info->ordered_operations_mutex);
2711         mutex_init(&fs_info->tree_log_mutex);
2712         mutex_init(&fs_info->chunk_mutex);
2713         mutex_init(&fs_info->transaction_kthread_mutex);
2714         mutex_init(&fs_info->cleaner_mutex);
2715         mutex_init(&fs_info->volume_mutex);
2716         mutex_init(&fs_info->ro_block_group_mutex);
2717         init_rwsem(&fs_info->commit_root_sem);
2718         init_rwsem(&fs_info->cleanup_work_sem);
2719         init_rwsem(&fs_info->subvol_sem);
2720         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2721
2722         btrfs_init_dev_replace_locks(fs_info);
2723         btrfs_init_qgroup(fs_info);
2724
2725         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2726         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2727
2728         init_waitqueue_head(&fs_info->transaction_throttle);
2729         init_waitqueue_head(&fs_info->transaction_wait);
2730         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2731         init_waitqueue_head(&fs_info->async_submit_wait);
2732
2733         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2734
2735         /* Usable values until the real ones are cached from the superblock */
2736         fs_info->nodesize = 4096;
2737         fs_info->sectorsize = 4096;
2738         fs_info->stripesize = 4096;
2739
2740         ret = btrfs_alloc_stripe_hash_table(fs_info);
2741         if (ret) {
2742                 err = ret;
2743                 goto fail_alloc;
2744         }
2745
2746         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2747
2748         invalidate_bdev(fs_devices->latest_bdev);
2749
2750         /*
2751          * Read super block and check the signature bytes only
2752          */
2753         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2754         if (IS_ERR(bh)) {
2755                 err = PTR_ERR(bh);
2756                 goto fail_alloc;
2757         }
2758
2759         /*
2760          * We want to check superblock checksum, the type is stored inside.
2761          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2762          */
2763         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2764                 btrfs_err(fs_info, "superblock checksum mismatch");
2765                 err = -EINVAL;
2766                 brelse(bh);
2767                 goto fail_alloc;
2768         }
2769
2770         /*
2771          * super_copy is zeroed at allocation time and we never touch the
2772          * following bytes up to INFO_SIZE, the checksum is calculated from
2773          * the whole block of INFO_SIZE
2774          */
2775         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2776         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2777                sizeof(*fs_info->super_for_commit));
2778         brelse(bh);
2779
2780         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2781
2782         ret = btrfs_check_super_valid(fs_info);
2783         if (ret) {
2784                 btrfs_err(fs_info, "superblock contains fatal errors");
2785                 err = -EINVAL;
2786                 goto fail_alloc;
2787         }
2788
2789         disk_super = fs_info->super_copy;
2790         if (!btrfs_super_root(disk_super))
2791                 goto fail_alloc;
2792
2793         /* check FS state, whether FS is broken. */
2794         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2795                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2796
2797         /*
2798          * run through our array of backup supers and setup
2799          * our ring pointer to the oldest one
2800          */
2801         generation = btrfs_super_generation(disk_super);
2802         find_oldest_super_backup(fs_info, generation);
2803
2804         /*
2805          * In the long term, we'll store the compression type in the super
2806          * block, and it'll be used for per file compression control.
2807          */
2808         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2809
2810         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2811         if (ret) {
2812                 err = ret;
2813                 goto fail_alloc;
2814         }
2815
2816         features = btrfs_super_incompat_flags(disk_super) &
2817                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2818         if (features) {
2819                 btrfs_err(fs_info,
2820                     "cannot mount because of unsupported optional features (%llx)",
2821                     features);
2822                 err = -EINVAL;
2823                 goto fail_alloc;
2824         }
2825
2826         features = btrfs_super_incompat_flags(disk_super);
2827         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2828         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2829                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2830
2831         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2832                 btrfs_info(fs_info, "has skinny extents");
2833
2834         /*
2835          * flag our filesystem as having big metadata blocks if
2836          * they are bigger than the page size
2837          */
2838         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2839                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2840                         btrfs_info(fs_info,
2841                                 "flagging fs with big metadata feature");
2842                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2843         }
2844
2845         nodesize = btrfs_super_nodesize(disk_super);
2846         sectorsize = btrfs_super_sectorsize(disk_super);
2847         stripesize = sectorsize;
2848         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2849         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2850
2851         /* Cache block sizes */
2852         fs_info->nodesize = nodesize;
2853         fs_info->sectorsize = sectorsize;
2854         fs_info->stripesize = stripesize;
2855
2856         /*
2857          * mixed block groups end up with duplicate but slightly offset
2858          * extent buffers for the same range.  It leads to corruptions
2859          */
2860         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2861             (sectorsize != nodesize)) {
2862                 btrfs_err(fs_info,
2863 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2864                         nodesize, sectorsize);
2865                 goto fail_alloc;
2866         }
2867
2868         /*
2869          * Needn't use the lock because there is no other task which will
2870          * update the flag.
2871          */
2872         btrfs_set_super_incompat_flags(disk_super, features);
2873
2874         features = btrfs_super_compat_ro_flags(disk_super) &
2875                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2876         if (!(sb->s_flags & MS_RDONLY) && features) {
2877                 btrfs_err(fs_info,
2878         "cannot mount read-write because of unsupported optional features (%llx)",
2879                        features);
2880                 err = -EINVAL;
2881                 goto fail_alloc;
2882         }
2883
2884         max_active = fs_info->thread_pool_size;
2885
2886         ret = btrfs_init_workqueues(fs_info, fs_devices);
2887         if (ret) {
2888                 err = ret;
2889                 goto fail_sb_buffer;
2890         }
2891
2892         sb->s_bdi->congested_fn = btrfs_congested_fn;
2893         sb->s_bdi->congested_data = fs_info;
2894         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2895         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2896         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2897         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2898
2899         sb->s_blocksize = sectorsize;
2900         sb->s_blocksize_bits = blksize_bits(sectorsize);
2901
2902         mutex_lock(&fs_info->chunk_mutex);
2903         ret = btrfs_read_sys_array(fs_info);
2904         mutex_unlock(&fs_info->chunk_mutex);
2905         if (ret) {
2906                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2907                 goto fail_sb_buffer;
2908         }
2909
2910         generation = btrfs_super_chunk_root_generation(disk_super);
2911
2912         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2913
2914         chunk_root->node = read_tree_block(fs_info,
2915                                            btrfs_super_chunk_root(disk_super),
2916                                            generation);
2917         if (IS_ERR(chunk_root->node) ||
2918             !extent_buffer_uptodate(chunk_root->node)) {
2919                 btrfs_err(fs_info, "failed to read chunk root");
2920                 if (!IS_ERR(chunk_root->node))
2921                         free_extent_buffer(chunk_root->node);
2922                 chunk_root->node = NULL;
2923                 goto fail_tree_roots;
2924         }
2925         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2926         chunk_root->commit_root = btrfs_root_node(chunk_root);
2927
2928         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2929            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2930
2931         ret = btrfs_read_chunk_tree(fs_info);
2932         if (ret) {
2933                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2934                 goto fail_tree_roots;
2935         }
2936
2937         /*
2938          * keep the device that is marked to be the target device for the
2939          * dev_replace procedure
2940          */
2941         btrfs_close_extra_devices(fs_devices, 0);
2942
2943         if (!fs_devices->latest_bdev) {
2944                 btrfs_err(fs_info, "failed to read devices");
2945                 goto fail_tree_roots;
2946         }
2947
2948 retry_root_backup:
2949         generation = btrfs_super_generation(disk_super);
2950
2951         tree_root->node = read_tree_block(fs_info,
2952                                           btrfs_super_root(disk_super),
2953                                           generation);
2954         if (IS_ERR(tree_root->node) ||
2955             !extent_buffer_uptodate(tree_root->node)) {
2956                 btrfs_warn(fs_info, "failed to read tree root");
2957                 if (!IS_ERR(tree_root->node))
2958                         free_extent_buffer(tree_root->node);
2959                 tree_root->node = NULL;
2960                 goto recovery_tree_root;
2961         }
2962
2963         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2964         tree_root->commit_root = btrfs_root_node(tree_root);
2965         btrfs_set_root_refs(&tree_root->root_item, 1);
2966
2967         mutex_lock(&tree_root->objectid_mutex);
2968         ret = btrfs_find_highest_objectid(tree_root,
2969                                         &tree_root->highest_objectid);
2970         if (ret) {
2971                 mutex_unlock(&tree_root->objectid_mutex);
2972                 goto recovery_tree_root;
2973         }
2974
2975         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2976
2977         mutex_unlock(&tree_root->objectid_mutex);
2978
2979         ret = btrfs_read_roots(fs_info);
2980         if (ret)
2981                 goto recovery_tree_root;
2982
2983         fs_info->generation = generation;
2984         fs_info->last_trans_committed = generation;
2985
2986         ret = btrfs_recover_balance(fs_info);
2987         if (ret) {
2988                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2989                 goto fail_block_groups;
2990         }
2991
2992         ret = btrfs_init_dev_stats(fs_info);
2993         if (ret) {
2994                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2995                 goto fail_block_groups;
2996         }
2997
2998         ret = btrfs_init_dev_replace(fs_info);
2999         if (ret) {
3000                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3001                 goto fail_block_groups;
3002         }
3003
3004         btrfs_close_extra_devices(fs_devices, 1);
3005
3006         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3007         if (ret) {
3008                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3009                                 ret);
3010                 goto fail_block_groups;
3011         }
3012
3013         ret = btrfs_sysfs_add_device(fs_devices);
3014         if (ret) {
3015                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3016                                 ret);
3017                 goto fail_fsdev_sysfs;
3018         }
3019
3020         ret = btrfs_sysfs_add_mounted(fs_info);
3021         if (ret) {
3022                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3023                 goto fail_fsdev_sysfs;
3024         }
3025
3026         ret = btrfs_init_space_info(fs_info);
3027         if (ret) {
3028                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3029                 goto fail_sysfs;
3030         }
3031
3032         ret = btrfs_read_block_groups(fs_info);
3033         if (ret) {
3034                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3035                 goto fail_sysfs;
3036         }
3037         fs_info->num_tolerated_disk_barrier_failures =
3038                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3039         if (fs_info->fs_devices->missing_devices >
3040              fs_info->num_tolerated_disk_barrier_failures &&
3041             !(sb->s_flags & MS_RDONLY)) {
3042                 btrfs_warn(fs_info,
3043 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3044                         fs_info->fs_devices->missing_devices,
3045                         fs_info->num_tolerated_disk_barrier_failures);
3046                 goto fail_sysfs;
3047         }
3048
3049         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3050                                                "btrfs-cleaner");
3051         if (IS_ERR(fs_info->cleaner_kthread))
3052                 goto fail_sysfs;
3053
3054         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3055                                                    tree_root,
3056                                                    "btrfs-transaction");
3057         if (IS_ERR(fs_info->transaction_kthread))
3058                 goto fail_cleaner;
3059
3060         if (!btrfs_test_opt(fs_info, SSD) &&
3061             !btrfs_test_opt(fs_info, NOSSD) &&
3062             !fs_info->fs_devices->rotating) {
3063                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3064                 btrfs_set_opt(fs_info->mount_opt, SSD);
3065         }
3066
3067         /*
3068          * Mount does not set all options immediately, we can do it now and do
3069          * not have to wait for transaction commit
3070          */
3071         btrfs_apply_pending_changes(fs_info);
3072
3073 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3074         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3075                 ret = btrfsic_mount(fs_info, fs_devices,
3076                                     btrfs_test_opt(fs_info,
3077                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3078                                     1 : 0,
3079                                     fs_info->check_integrity_print_mask);
3080                 if (ret)
3081                         btrfs_warn(fs_info,
3082                                 "failed to initialize integrity check module: %d",
3083                                 ret);
3084         }
3085 #endif
3086         ret = btrfs_read_qgroup_config(fs_info);
3087         if (ret)
3088                 goto fail_trans_kthread;
3089
3090         /* do not make disk changes in broken FS or nologreplay is given */
3091         if (btrfs_super_log_root(disk_super) != 0 &&
3092             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3093                 ret = btrfs_replay_log(fs_info, fs_devices);
3094                 if (ret) {
3095                         err = ret;
3096                         goto fail_qgroup;
3097                 }
3098         }
3099
3100         ret = btrfs_find_orphan_roots(fs_info);
3101         if (ret)
3102                 goto fail_qgroup;
3103
3104         if (!(sb->s_flags & MS_RDONLY)) {
3105                 ret = btrfs_cleanup_fs_roots(fs_info);
3106                 if (ret)
3107                         goto fail_qgroup;
3108
3109                 mutex_lock(&fs_info->cleaner_mutex);
3110                 ret = btrfs_recover_relocation(tree_root);
3111                 mutex_unlock(&fs_info->cleaner_mutex);
3112                 if (ret < 0) {
3113                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3114                                         ret);
3115                         err = -EINVAL;
3116                         goto fail_qgroup;
3117                 }
3118         }
3119
3120         location.objectid = BTRFS_FS_TREE_OBJECTID;
3121         location.type = BTRFS_ROOT_ITEM_KEY;
3122         location.offset = 0;
3123
3124         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3125         if (IS_ERR(fs_info->fs_root)) {
3126                 err = PTR_ERR(fs_info->fs_root);
3127                 goto fail_qgroup;
3128         }
3129
3130         if (sb->s_flags & MS_RDONLY)
3131                 return 0;
3132
3133         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3134             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3135                 clear_free_space_tree = 1;
3136         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3137                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3138                 btrfs_warn(fs_info, "free space tree is invalid");
3139                 clear_free_space_tree = 1;
3140         }
3141
3142         if (clear_free_space_tree) {
3143                 btrfs_info(fs_info, "clearing free space tree");
3144                 ret = btrfs_clear_free_space_tree(fs_info);
3145                 if (ret) {
3146                         btrfs_warn(fs_info,
3147                                    "failed to clear free space tree: %d", ret);
3148                         close_ctree(fs_info);
3149                         return ret;
3150                 }
3151         }
3152
3153         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3154             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3155                 btrfs_info(fs_info, "creating free space tree");
3156                 ret = btrfs_create_free_space_tree(fs_info);
3157                 if (ret) {
3158                         btrfs_warn(fs_info,
3159                                 "failed to create free space tree: %d", ret);
3160                         close_ctree(fs_info);
3161                         return ret;
3162                 }
3163         }
3164
3165         down_read(&fs_info->cleanup_work_sem);
3166         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3167             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3168                 up_read(&fs_info->cleanup_work_sem);
3169                 close_ctree(fs_info);
3170                 return ret;
3171         }
3172         up_read(&fs_info->cleanup_work_sem);
3173
3174         ret = btrfs_resume_balance_async(fs_info);
3175         if (ret) {
3176                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3177                 close_ctree(fs_info);
3178                 return ret;
3179         }
3180
3181         ret = btrfs_resume_dev_replace_async(fs_info);
3182         if (ret) {
3183                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3184                 close_ctree(fs_info);
3185                 return ret;
3186         }
3187
3188         btrfs_qgroup_rescan_resume(fs_info);
3189
3190         if (!fs_info->uuid_root) {
3191                 btrfs_info(fs_info, "creating UUID tree");
3192                 ret = btrfs_create_uuid_tree(fs_info);
3193                 if (ret) {
3194                         btrfs_warn(fs_info,
3195                                 "failed to create the UUID tree: %d", ret);
3196                         close_ctree(fs_info);
3197                         return ret;
3198                 }
3199         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3200                    fs_info->generation !=
3201                                 btrfs_super_uuid_tree_generation(disk_super)) {
3202                 btrfs_info(fs_info, "checking UUID tree");
3203                 ret = btrfs_check_uuid_tree(fs_info);
3204                 if (ret) {
3205                         btrfs_warn(fs_info,
3206                                 "failed to check the UUID tree: %d", ret);
3207                         close_ctree(fs_info);
3208                         return ret;
3209                 }
3210         } else {
3211                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3212         }
3213         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3214
3215         /*
3216          * backuproot only affect mount behavior, and if open_ctree succeeded,
3217          * no need to keep the flag
3218          */
3219         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3220
3221         return 0;
3222
3223 fail_qgroup:
3224         btrfs_free_qgroup_config(fs_info);
3225 fail_trans_kthread:
3226         kthread_stop(fs_info->transaction_kthread);
3227         btrfs_cleanup_transaction(fs_info);
3228         btrfs_free_fs_roots(fs_info);
3229 fail_cleaner:
3230         kthread_stop(fs_info->cleaner_kthread);
3231
3232         /*
3233          * make sure we're done with the btree inode before we stop our
3234          * kthreads
3235          */
3236         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3237
3238 fail_sysfs:
3239         btrfs_sysfs_remove_mounted(fs_info);
3240
3241 fail_fsdev_sysfs:
3242         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3243
3244 fail_block_groups:
3245         btrfs_put_block_group_cache(fs_info);
3246
3247 fail_tree_roots:
3248         free_root_pointers(fs_info, 1);
3249         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3250
3251 fail_sb_buffer:
3252         btrfs_stop_all_workers(fs_info);
3253         btrfs_free_block_groups(fs_info);
3254 fail_alloc:
3255 fail_iput:
3256         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3257
3258         iput(fs_info->btree_inode);
3259 fail_bio_counter:
3260         percpu_counter_destroy(&fs_info->bio_counter);
3261 fail_delalloc_bytes:
3262         percpu_counter_destroy(&fs_info->delalloc_bytes);
3263 fail_dirty_metadata_bytes:
3264         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3265 fail_srcu:
3266         cleanup_srcu_struct(&fs_info->subvol_srcu);
3267 fail:
3268         btrfs_free_stripe_hash_table(fs_info);
3269         btrfs_close_devices(fs_info->fs_devices);
3270         return err;
3271
3272 recovery_tree_root:
3273         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3274                 goto fail_tree_roots;
3275
3276         free_root_pointers(fs_info, 0);
3277
3278         /* don't use the log in recovery mode, it won't be valid */
3279         btrfs_set_super_log_root(disk_super, 0);
3280
3281         /* we can't trust the free space cache either */
3282         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3283
3284         ret = next_root_backup(fs_info, fs_info->super_copy,
3285                                &num_backups_tried, &backup_index);
3286         if (ret == -1)
3287                 goto fail_block_groups;
3288         goto retry_root_backup;
3289 }
3290
3291 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3292 {
3293         if (uptodate) {
3294                 set_buffer_uptodate(bh);
3295         } else {
3296                 struct btrfs_device *device = (struct btrfs_device *)
3297                         bh->b_private;
3298
3299                 btrfs_warn_rl_in_rcu(device->fs_info,
3300                                 "lost page write due to IO error on %s",
3301                                           rcu_str_deref(device->name));
3302                 /* note, we don't set_buffer_write_io_error because we have
3303                  * our own ways of dealing with the IO errors
3304                  */
3305                 clear_buffer_uptodate(bh);
3306                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3307         }
3308         unlock_buffer(bh);
3309         put_bh(bh);
3310 }
3311
3312 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3313                         struct buffer_head **bh_ret)
3314 {
3315         struct buffer_head *bh;
3316         struct btrfs_super_block *super;
3317         u64 bytenr;
3318
3319         bytenr = btrfs_sb_offset(copy_num);
3320         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3321                 return -EINVAL;
3322
3323         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3324         /*
3325          * If we fail to read from the underlying devices, as of now
3326          * the best option we have is to mark it EIO.
3327          */
3328         if (!bh)
3329                 return -EIO;
3330
3331         super = (struct btrfs_super_block *)bh->b_data;
3332         if (btrfs_super_bytenr(super) != bytenr ||
3333                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3334                 brelse(bh);
3335                 return -EINVAL;
3336         }
3337
3338         *bh_ret = bh;
3339         return 0;
3340 }
3341
3342
3343 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3344 {
3345         struct buffer_head *bh;
3346         struct buffer_head *latest = NULL;
3347         struct btrfs_super_block *super;
3348         int i;
3349         u64 transid = 0;
3350         int ret = -EINVAL;
3351
3352         /* we would like to check all the supers, but that would make
3353          * a btrfs mount succeed after a mkfs from a different FS.
3354          * So, we need to add a special mount option to scan for
3355          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3356          */
3357         for (i = 0; i < 1; i++) {
3358                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3359                 if (ret)
3360                         continue;
3361
3362                 super = (struct btrfs_super_block *)bh->b_data;
3363
3364                 if (!latest || btrfs_super_generation(super) > transid) {
3365                         brelse(latest);
3366                         latest = bh;
3367                         transid = btrfs_super_generation(super);
3368                 } else {
3369                         brelse(bh);
3370                 }
3371         }
3372
3373         if (!latest)
3374                 return ERR_PTR(ret);
3375
3376         return latest;
3377 }
3378
3379 /*
3380  * this should be called twice, once with wait == 0 and
3381  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3382  * we write are pinned.
3383  *
3384  * They are released when wait == 1 is done.
3385  * max_mirrors must be the same for both runs, and it indicates how
3386  * many supers on this one device should be written.
3387  *
3388  * max_mirrors == 0 means to write them all.
3389  */
3390 static int write_dev_supers(struct btrfs_device *device,
3391                             struct btrfs_super_block *sb,
3392                             int wait, int max_mirrors)
3393 {
3394         struct buffer_head *bh;
3395         int i;
3396         int ret;
3397         int errors = 0;
3398         u32 crc;
3399         u64 bytenr;
3400
3401         if (max_mirrors == 0)
3402                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3403
3404         for (i = 0; i < max_mirrors; i++) {
3405                 bytenr = btrfs_sb_offset(i);
3406                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3407                     device->commit_total_bytes)
3408                         break;
3409
3410                 if (wait) {
3411                         bh = __find_get_block(device->bdev, bytenr / 4096,
3412                                               BTRFS_SUPER_INFO_SIZE);
3413                         if (!bh) {
3414                                 errors++;
3415                                 continue;
3416                         }
3417                         wait_on_buffer(bh);
3418                         if (!buffer_uptodate(bh))
3419                                 errors++;
3420
3421                         /* drop our reference */
3422                         brelse(bh);
3423
3424                         /* drop the reference from the wait == 0 run */
3425                         brelse(bh);
3426                         continue;
3427                 } else {
3428                         btrfs_set_super_bytenr(sb, bytenr);
3429
3430                         crc = ~(u32)0;
3431                         crc = btrfs_csum_data((const char *)sb +
3432                                               BTRFS_CSUM_SIZE, crc,
3433                                               BTRFS_SUPER_INFO_SIZE -
3434                                               BTRFS_CSUM_SIZE);
3435                         btrfs_csum_final(crc, sb->csum);
3436
3437                         /*
3438                          * one reference for us, and we leave it for the
3439                          * caller
3440                          */
3441                         bh = __getblk(device->bdev, bytenr / 4096,
3442                                       BTRFS_SUPER_INFO_SIZE);
3443                         if (!bh) {
3444                                 btrfs_err(device->fs_info,
3445                                     "couldn't get super buffer head for bytenr %llu",
3446                                     bytenr);
3447                                 errors++;
3448                                 continue;
3449                         }
3450
3451                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3452
3453                         /* one reference for submit_bh */
3454                         get_bh(bh);
3455
3456                         set_buffer_uptodate(bh);
3457                         lock_buffer(bh);
3458                         bh->b_end_io = btrfs_end_buffer_write_sync;
3459                         bh->b_private = device;
3460                 }
3461
3462                 /*
3463                  * we fua the first super.  The others we allow
3464                  * to go down lazy.
3465                  */
3466                 if (i == 0) {
3467                         ret = btrfsic_submit_bh(REQ_OP_WRITE,
3468                                                 REQ_SYNC | REQ_FUA, bh);
3469                 } else {
3470                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3471                 }
3472                 if (ret)
3473                         errors++;
3474         }
3475         return errors < i ? 0 : -1;
3476 }
3477
3478 /*
3479  * endio for the write_dev_flush, this will wake anyone waiting
3480  * for the barrier when it is done
3481  */
3482 static void btrfs_end_empty_barrier(struct bio *bio)
3483 {
3484         complete(bio->bi_private);
3485 }
3486
3487 /*
3488  * Submit a flush request to the device if it supports it. Error handling is
3489  * done in the waiting counterpart.
3490  */
3491 static void write_dev_flush(struct btrfs_device *device)
3492 {
3493         struct request_queue *q = bdev_get_queue(device->bdev);
3494         struct bio *bio = device->flush_bio;
3495
3496         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3497                 return;
3498
3499         bio_reset(bio);
3500         bio->bi_end_io = btrfs_end_empty_barrier;
3501         bio->bi_bdev = device->bdev;
3502         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3503         init_completion(&device->flush_wait);
3504         bio->bi_private = &device->flush_wait;
3505
3506         submit_bio(bio);
3507         device->flush_bio_sent = 1;
3508 }
3509
3510 /*
3511  * If the flush bio has been submitted by write_dev_flush, wait for it.
3512  */
3513 static int wait_dev_flush(struct btrfs_device *device)
3514 {
3515         struct bio *bio = device->flush_bio;
3516
3517         if (!device->flush_bio_sent)
3518                 return 0;
3519
3520         device->flush_bio_sent = 0;
3521         wait_for_completion_io(&device->flush_wait);
3522
3523         return bio->bi_error;
3524 }
3525
3526 static int check_barrier_error(struct btrfs_fs_devices *fsdevs)
3527 {
3528         int dev_flush_error = 0;
3529         struct btrfs_device *dev;
3530
3531         list_for_each_entry_rcu(dev, &fsdevs->devices, dev_list) {
3532                 if (!dev->bdev || dev->last_flush_error)
3533                         dev_flush_error++;
3534         }
3535
3536         if (dev_flush_error >
3537             fsdevs->fs_info->num_tolerated_disk_barrier_failures)
3538                 return -EIO;
3539
3540         return 0;
3541 }
3542
3543 /*
3544  * send an empty flush down to each device in parallel,
3545  * then wait for them
3546  */
3547 static int barrier_all_devices(struct btrfs_fs_info *info)
3548 {
3549         struct list_head *head;
3550         struct btrfs_device *dev;
3551         int errors_wait = 0;
3552         int ret;
3553
3554         /* send down all the barriers */
3555         head = &info->fs_devices->devices;
3556         list_for_each_entry_rcu(dev, head, dev_list) {
3557                 if (dev->missing)
3558                         continue;
3559                 if (!dev->bdev)
3560                         continue;
3561                 if (!dev->in_fs_metadata || !dev->writeable)
3562                         continue;
3563
3564                 write_dev_flush(dev);
3565                 dev->last_flush_error = 0;
3566         }
3567
3568         /* wait for all the barriers */
3569         list_for_each_entry_rcu(dev, head, dev_list) {
3570                 if (dev->missing)
3571                         continue;
3572                 if (!dev->bdev) {
3573                         errors_wait++;
3574                         continue;
3575                 }
3576                 if (!dev->in_fs_metadata || !dev->writeable)
3577                         continue;
3578
3579                 ret = wait_dev_flush(dev);
3580                 if (ret) {
3581                         dev->last_flush_error = ret;
3582                         btrfs_dev_stat_inc_and_print(dev,
3583                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3584                         errors_wait++;
3585                 }
3586         }
3587
3588         if (errors_wait) {
3589                 /*
3590                  * At some point we need the status of all disks
3591                  * to arrive at the volume status. So error checking
3592                  * is being pushed to a separate loop.
3593                  */
3594                 return check_barrier_error(info->fs_devices);
3595         }
3596         return 0;
3597 }
3598
3599 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3600 {
3601         int raid_type;
3602         int min_tolerated = INT_MAX;
3603
3604         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3605             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3606                 min_tolerated = min(min_tolerated,
3607                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3608                                     tolerated_failures);
3609
3610         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3611                 if (raid_type == BTRFS_RAID_SINGLE)
3612                         continue;
3613                 if (!(flags & btrfs_raid_group[raid_type]))
3614                         continue;
3615                 min_tolerated = min(min_tolerated,
3616                                     btrfs_raid_array[raid_type].
3617                                     tolerated_failures);
3618         }
3619
3620         if (min_tolerated == INT_MAX) {
3621                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3622                 min_tolerated = 0;
3623         }
3624
3625         return min_tolerated;
3626 }
3627
3628 int btrfs_calc_num_tolerated_disk_barrier_failures(
3629         struct btrfs_fs_info *fs_info)
3630 {
3631         struct btrfs_ioctl_space_info space;
3632         struct btrfs_space_info *sinfo;
3633         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3634                        BTRFS_BLOCK_GROUP_SYSTEM,
3635                        BTRFS_BLOCK_GROUP_METADATA,
3636                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3637         int i;
3638         int c;
3639         int num_tolerated_disk_barrier_failures =
3640                 (int)fs_info->fs_devices->num_devices;
3641
3642         for (i = 0; i < ARRAY_SIZE(types); i++) {
3643                 struct btrfs_space_info *tmp;
3644
3645                 sinfo = NULL;
3646                 rcu_read_lock();
3647                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3648                         if (tmp->flags == types[i]) {
3649                                 sinfo = tmp;
3650                                 break;
3651                         }
3652                 }
3653                 rcu_read_unlock();
3654
3655                 if (!sinfo)
3656                         continue;
3657
3658                 down_read(&sinfo->groups_sem);
3659                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3660                         u64 flags;
3661
3662                         if (list_empty(&sinfo->block_groups[c]))
3663                                 continue;
3664
3665                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3666                                                    &space);
3667                         if (space.total_bytes == 0 || space.used_bytes == 0)
3668                                 continue;
3669                         flags = space.flags;
3670
3671                         num_tolerated_disk_barrier_failures = min(
3672                                 num_tolerated_disk_barrier_failures,
3673                                 btrfs_get_num_tolerated_disk_barrier_failures(
3674                                         flags));
3675                 }
3676                 up_read(&sinfo->groups_sem);
3677         }
3678
3679         return num_tolerated_disk_barrier_failures;
3680 }
3681
3682 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3683 {
3684         struct list_head *head;
3685         struct btrfs_device *dev;
3686         struct btrfs_super_block *sb;
3687         struct btrfs_dev_item *dev_item;
3688         int ret;
3689         int do_barriers;
3690         int max_errors;
3691         int total_errors = 0;
3692         u64 flags;
3693
3694         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3695         backup_super_roots(fs_info);
3696
3697         sb = fs_info->super_for_commit;
3698         dev_item = &sb->dev_item;
3699
3700         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3701         head = &fs_info->fs_devices->devices;
3702         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3703
3704         if (do_barriers) {
3705                 ret = barrier_all_devices(fs_info);
3706                 if (ret) {
3707                         mutex_unlock(
3708                                 &fs_info->fs_devices->device_list_mutex);
3709                         btrfs_handle_fs_error(fs_info, ret,
3710                                               "errors while submitting device barriers.");
3711                         return ret;
3712                 }
3713         }
3714
3715         list_for_each_entry_rcu(dev, head, dev_list) {
3716                 if (!dev->bdev) {
3717                         total_errors++;
3718                         continue;
3719                 }
3720                 if (!dev->in_fs_metadata || !dev->writeable)
3721                         continue;
3722
3723                 btrfs_set_stack_device_generation(dev_item, 0);
3724                 btrfs_set_stack_device_type(dev_item, dev->type);
3725                 btrfs_set_stack_device_id(dev_item, dev->devid);
3726                 btrfs_set_stack_device_total_bytes(dev_item,
3727                                                    dev->commit_total_bytes);
3728                 btrfs_set_stack_device_bytes_used(dev_item,
3729                                                   dev->commit_bytes_used);
3730                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3731                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3732                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3733                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3734                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3735
3736                 flags = btrfs_super_flags(sb);
3737                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3738
3739                 ret = write_dev_supers(dev, sb, 0, max_mirrors);
3740                 if (ret)
3741                         total_errors++;
3742         }
3743         if (total_errors > max_errors) {
3744                 btrfs_err(fs_info, "%d errors while writing supers",
3745                           total_errors);
3746                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3747
3748                 /* FUA is masked off if unsupported and can't be the reason */
3749                 btrfs_handle_fs_error(fs_info, -EIO,
3750                                       "%d errors while writing supers",
3751                                       total_errors);
3752                 return -EIO;
3753         }
3754
3755         total_errors = 0;
3756         list_for_each_entry_rcu(dev, head, dev_list) {
3757                 if (!dev->bdev)
3758                         continue;
3759                 if (!dev->in_fs_metadata || !dev->writeable)
3760                         continue;
3761
3762                 ret = write_dev_supers(dev, sb, 1, max_mirrors);
3763                 if (ret)
3764                         total_errors++;
3765         }
3766         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3767         if (total_errors > max_errors) {
3768                 btrfs_handle_fs_error(fs_info, -EIO,
3769                                       "%d errors while writing supers",
3770                                       total_errors);
3771                 return -EIO;
3772         }
3773         return 0;
3774 }
3775
3776 /* Drop a fs root from the radix tree and free it. */
3777 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3778                                   struct btrfs_root *root)
3779 {
3780         spin_lock(&fs_info->fs_roots_radix_lock);
3781         radix_tree_delete(&fs_info->fs_roots_radix,
3782                           (unsigned long)root->root_key.objectid);
3783         spin_unlock(&fs_info->fs_roots_radix_lock);
3784
3785         if (btrfs_root_refs(&root->root_item) == 0)
3786                 synchronize_srcu(&fs_info->subvol_srcu);
3787
3788         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3789                 btrfs_free_log(NULL, root);
3790                 if (root->reloc_root) {
3791                         free_extent_buffer(root->reloc_root->node);
3792                         free_extent_buffer(root->reloc_root->commit_root);
3793                         btrfs_put_fs_root(root->reloc_root);
3794                         root->reloc_root = NULL;
3795                 }
3796         }
3797
3798         if (root->free_ino_pinned)
3799                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3800         if (root->free_ino_ctl)
3801                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3802         free_fs_root(root);
3803 }
3804
3805 static void free_fs_root(struct btrfs_root *root)
3806 {
3807         iput(root->ino_cache_inode);
3808         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3809         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3810         root->orphan_block_rsv = NULL;
3811         if (root->anon_dev)
3812                 free_anon_bdev(root->anon_dev);
3813         if (root->subv_writers)
3814                 btrfs_free_subvolume_writers(root->subv_writers);
3815         free_extent_buffer(root->node);
3816         free_extent_buffer(root->commit_root);
3817         kfree(root->free_ino_ctl);
3818         kfree(root->free_ino_pinned);
3819         kfree(root->name);
3820         btrfs_put_fs_root(root);
3821 }
3822
3823 void btrfs_free_fs_root(struct btrfs_root *root)
3824 {
3825         free_fs_root(root);
3826 }
3827
3828 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3829 {
3830         u64 root_objectid = 0;
3831         struct btrfs_root *gang[8];
3832         int i = 0;
3833         int err = 0;
3834         unsigned int ret = 0;
3835         int index;
3836
3837         while (1) {
3838                 index = srcu_read_lock(&fs_info->subvol_srcu);
3839                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3840                                              (void **)gang, root_objectid,
3841                                              ARRAY_SIZE(gang));
3842                 if (!ret) {
3843                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3844                         break;
3845                 }
3846                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3847
3848                 for (i = 0; i < ret; i++) {
3849                         /* Avoid to grab roots in dead_roots */
3850                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3851                                 gang[i] = NULL;
3852                                 continue;
3853                         }
3854                         /* grab all the search result for later use */
3855                         gang[i] = btrfs_grab_fs_root(gang[i]);
3856                 }
3857                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3858
3859                 for (i = 0; i < ret; i++) {
3860                         if (!gang[i])
3861                                 continue;
3862                         root_objectid = gang[i]->root_key.objectid;
3863                         err = btrfs_orphan_cleanup(gang[i]);
3864                         if (err)
3865                                 break;
3866                         btrfs_put_fs_root(gang[i]);
3867                 }
3868                 root_objectid++;
3869         }
3870
3871         /* release the uncleaned roots due to error */
3872         for (; i < ret; i++) {
3873                 if (gang[i])
3874                         btrfs_put_fs_root(gang[i]);
3875         }
3876         return err;
3877 }
3878
3879 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3880 {
3881         struct btrfs_root *root = fs_info->tree_root;
3882         struct btrfs_trans_handle *trans;
3883
3884         mutex_lock(&fs_info->cleaner_mutex);
3885         btrfs_run_delayed_iputs(fs_info);
3886         mutex_unlock(&fs_info->cleaner_mutex);
3887         wake_up_process(fs_info->cleaner_kthread);
3888
3889         /* wait until ongoing cleanup work done */
3890         down_write(&fs_info->cleanup_work_sem);
3891         up_write(&fs_info->cleanup_work_sem);
3892
3893         trans = btrfs_join_transaction(root);
3894         if (IS_ERR(trans))
3895                 return PTR_ERR(trans);
3896         return btrfs_commit_transaction(trans);
3897 }
3898
3899 void close_ctree(struct btrfs_fs_info *fs_info)
3900 {
3901         struct btrfs_root *root = fs_info->tree_root;
3902         int ret;
3903
3904         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3905
3906         /* wait for the qgroup rescan worker to stop */
3907         btrfs_qgroup_wait_for_completion(fs_info, false);
3908
3909         /* wait for the uuid_scan task to finish */
3910         down(&fs_info->uuid_tree_rescan_sem);
3911         /* avoid complains from lockdep et al., set sem back to initial state */
3912         up(&fs_info->uuid_tree_rescan_sem);
3913
3914         /* pause restriper - we want to resume on mount */
3915         btrfs_pause_balance(fs_info);
3916
3917         btrfs_dev_replace_suspend_for_unmount(fs_info);
3918
3919         btrfs_scrub_cancel(fs_info);
3920
3921         /* wait for any defraggers to finish */
3922         wait_event(fs_info->transaction_wait,
3923                    (atomic_read(&fs_info->defrag_running) == 0));
3924
3925         /* clear out the rbtree of defraggable inodes */
3926         btrfs_cleanup_defrag_inodes(fs_info);
3927
3928         cancel_work_sync(&fs_info->async_reclaim_work);
3929
3930         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3931                 /*
3932                  * If the cleaner thread is stopped and there are
3933                  * block groups queued for removal, the deletion will be
3934                  * skipped when we quit the cleaner thread.
3935                  */
3936                 btrfs_delete_unused_bgs(fs_info);
3937
3938                 ret = btrfs_commit_super(fs_info);
3939                 if (ret)
3940                         btrfs_err(fs_info, "commit super ret %d", ret);
3941         }
3942
3943         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3944                 btrfs_error_commit_super(fs_info);
3945
3946         kthread_stop(fs_info->transaction_kthread);
3947         kthread_stop(fs_info->cleaner_kthread);
3948
3949         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3950
3951         btrfs_free_qgroup_config(fs_info);
3952
3953         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3954                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3955                        percpu_counter_sum(&fs_info->delalloc_bytes));
3956         }
3957
3958         btrfs_sysfs_remove_mounted(fs_info);
3959         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3960
3961         btrfs_free_fs_roots(fs_info);
3962
3963         btrfs_put_block_group_cache(fs_info);
3964
3965         /*
3966          * we must make sure there is not any read request to
3967          * submit after we stopping all workers.
3968          */
3969         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3970         btrfs_stop_all_workers(fs_info);
3971
3972         btrfs_free_block_groups(fs_info);
3973
3974         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3975         free_root_pointers(fs_info, 1);
3976
3977         iput(fs_info->btree_inode);
3978
3979 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3980         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3981                 btrfsic_unmount(fs_info->fs_devices);
3982 #endif
3983
3984         btrfs_close_devices(fs_info->fs_devices);
3985         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3986
3987         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3988         percpu_counter_destroy(&fs_info->delalloc_bytes);
3989         percpu_counter_destroy(&fs_info->bio_counter);
3990         cleanup_srcu_struct(&fs_info->subvol_srcu);
3991
3992         btrfs_free_stripe_hash_table(fs_info);
3993
3994         __btrfs_free_block_rsv(root->orphan_block_rsv);
3995         root->orphan_block_rsv = NULL;
3996
3997         mutex_lock(&fs_info->chunk_mutex);
3998         while (!list_empty(&fs_info->pinned_chunks)) {
3999                 struct extent_map *em;
4000
4001                 em = list_first_entry(&fs_info->pinned_chunks,
4002                                       struct extent_map, list);
4003                 list_del_init(&em->list);
4004                 free_extent_map(em);
4005         }
4006         mutex_unlock(&fs_info->chunk_mutex);
4007 }
4008
4009 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4010                           int atomic)
4011 {
4012         int ret;
4013         struct inode *btree_inode = buf->pages[0]->mapping->host;
4014
4015         ret = extent_buffer_uptodate(buf);
4016         if (!ret)
4017                 return ret;
4018
4019         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4020                                     parent_transid, atomic);
4021         if (ret == -EAGAIN)
4022                 return ret;
4023         return !ret;
4024 }
4025
4026 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4027 {
4028         struct btrfs_fs_info *fs_info;
4029         struct btrfs_root *root;
4030         u64 transid = btrfs_header_generation(buf);
4031         int was_dirty;
4032
4033 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4034         /*
4035          * This is a fast path so only do this check if we have sanity tests
4036          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4037          * outside of the sanity tests.
4038          */
4039         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4040                 return;
4041 #endif
4042         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4043         fs_info = root->fs_info;
4044         btrfs_assert_tree_locked(buf);
4045         if (transid != fs_info->generation)
4046                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4047                         buf->start, transid, fs_info->generation);
4048         was_dirty = set_extent_buffer_dirty(buf);
4049         if (!was_dirty)
4050                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
4051                                      buf->len,
4052                                      fs_info->dirty_metadata_batch);
4053 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4054         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4055                 btrfs_print_leaf(fs_info, buf);
4056                 ASSERT(0);
4057         }
4058 #endif
4059 }
4060
4061 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4062                                         int flush_delayed)
4063 {
4064         /*
4065          * looks as though older kernels can get into trouble with
4066          * this code, they end up stuck in balance_dirty_pages forever
4067          */
4068         int ret;
4069
4070         if (current->flags & PF_MEMALLOC)
4071                 return;
4072
4073         if (flush_delayed)
4074                 btrfs_balance_delayed_items(fs_info);
4075
4076         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4077                                      BTRFS_DIRTY_METADATA_THRESH);
4078         if (ret > 0) {
4079                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4080         }
4081 }
4082
4083 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4084 {
4085         __btrfs_btree_balance_dirty(fs_info, 1);
4086 }
4087
4088 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4089 {
4090         __btrfs_btree_balance_dirty(fs_info, 0);
4091 }
4092
4093 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4094 {
4095         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4096         struct btrfs_fs_info *fs_info = root->fs_info;
4097
4098         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4099 }
4100
4101 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
4102 {
4103         struct btrfs_super_block *sb = fs_info->super_copy;
4104         u64 nodesize = btrfs_super_nodesize(sb);
4105         u64 sectorsize = btrfs_super_sectorsize(sb);
4106         int ret = 0;
4107
4108         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4109                 btrfs_err(fs_info, "no valid FS found");
4110                 ret = -EINVAL;
4111         }
4112         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4113                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4114                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4115         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4116                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4117                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4118                 ret = -EINVAL;
4119         }
4120         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4121                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4122                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4123                 ret = -EINVAL;
4124         }
4125         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4126                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4127                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4128                 ret = -EINVAL;
4129         }
4130
4131         /*
4132          * Check sectorsize and nodesize first, other check will need it.
4133          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4134          */
4135         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4136             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4137                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4138                 ret = -EINVAL;
4139         }
4140         /* Only PAGE SIZE is supported yet */
4141         if (sectorsize != PAGE_SIZE) {
4142                 btrfs_err(fs_info,
4143                         "sectorsize %llu not supported yet, only support %lu",
4144                         sectorsize, PAGE_SIZE);
4145                 ret = -EINVAL;
4146         }
4147         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4148             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4149                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4150                 ret = -EINVAL;
4151         }
4152         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4153                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4154                           le32_to_cpu(sb->__unused_leafsize), nodesize);
4155                 ret = -EINVAL;
4156         }
4157
4158         /* Root alignment check */
4159         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4160                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4161                            btrfs_super_root(sb));
4162                 ret = -EINVAL;
4163         }
4164         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4165                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4166                            btrfs_super_chunk_root(sb));
4167                 ret = -EINVAL;
4168         }
4169         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4170                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4171                            btrfs_super_log_root(sb));
4172                 ret = -EINVAL;
4173         }
4174
4175         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4176                 btrfs_err(fs_info,
4177                            "dev_item UUID does not match fsid: %pU != %pU",
4178                            fs_info->fsid, sb->dev_item.fsid);
4179                 ret = -EINVAL;
4180         }
4181
4182         /*
4183          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4184          * done later
4185          */
4186         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4187                 btrfs_err(fs_info, "bytes_used is too small %llu",
4188                           btrfs_super_bytes_used(sb));
4189                 ret = -EINVAL;
4190         }
4191         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4192                 btrfs_err(fs_info, "invalid stripesize %u",
4193                           btrfs_super_stripesize(sb));
4194                 ret = -EINVAL;
4195         }
4196         if (btrfs_super_num_devices(sb) > (1UL << 31))
4197                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4198                            btrfs_super_num_devices(sb));
4199         if (btrfs_super_num_devices(sb) == 0) {
4200                 btrfs_err(fs_info, "number of devices is 0");
4201                 ret = -EINVAL;
4202         }
4203
4204         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4205                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4206                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4207                 ret = -EINVAL;
4208         }
4209
4210         /*
4211          * Obvious sys_chunk_array corruptions, it must hold at least one key
4212          * and one chunk
4213          */
4214         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4215                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4216                           btrfs_super_sys_array_size(sb),
4217                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4218                 ret = -EINVAL;
4219         }
4220         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4221                         + sizeof(struct btrfs_chunk)) {
4222                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4223                           btrfs_super_sys_array_size(sb),
4224                           sizeof(struct btrfs_disk_key)
4225                           + sizeof(struct btrfs_chunk));
4226                 ret = -EINVAL;
4227         }
4228
4229         /*
4230          * The generation is a global counter, we'll trust it more than the others
4231          * but it's still possible that it's the one that's wrong.
4232          */
4233         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4234                 btrfs_warn(fs_info,
4235                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4236                         btrfs_super_generation(sb),
4237                         btrfs_super_chunk_root_generation(sb));
4238         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4239             && btrfs_super_cache_generation(sb) != (u64)-1)
4240                 btrfs_warn(fs_info,
4241                         "suspicious: generation < cache_generation: %llu < %llu",
4242                         btrfs_super_generation(sb),
4243                         btrfs_super_cache_generation(sb));
4244
4245         return ret;
4246 }
4247
4248 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4249 {
4250         mutex_lock(&fs_info->cleaner_mutex);
4251         btrfs_run_delayed_iputs(fs_info);
4252         mutex_unlock(&fs_info->cleaner_mutex);
4253
4254         down_write(&fs_info->cleanup_work_sem);
4255         up_write(&fs_info->cleanup_work_sem);
4256
4257         /* cleanup FS via transaction */
4258         btrfs_cleanup_transaction(fs_info);
4259 }
4260
4261 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4262 {
4263         struct btrfs_ordered_extent *ordered;
4264
4265         spin_lock(&root->ordered_extent_lock);
4266         /*
4267          * This will just short circuit the ordered completion stuff which will
4268          * make sure the ordered extent gets properly cleaned up.
4269          */
4270         list_for_each_entry(ordered, &root->ordered_extents,
4271                             root_extent_list)
4272                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4273         spin_unlock(&root->ordered_extent_lock);
4274 }
4275
4276 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4277 {
4278         struct btrfs_root *root;
4279         struct list_head splice;
4280
4281         INIT_LIST_HEAD(&splice);
4282
4283         spin_lock(&fs_info->ordered_root_lock);
4284         list_splice_init(&fs_info->ordered_roots, &splice);
4285         while (!list_empty(&splice)) {
4286                 root = list_first_entry(&splice, struct btrfs_root,
4287                                         ordered_root);
4288                 list_move_tail(&root->ordered_root,
4289                                &fs_info->ordered_roots);
4290
4291                 spin_unlock(&fs_info->ordered_root_lock);
4292                 btrfs_destroy_ordered_extents(root);
4293
4294                 cond_resched();
4295                 spin_lock(&fs_info->ordered_root_lock);
4296         }
4297         spin_unlock(&fs_info->ordered_root_lock);
4298 }
4299
4300 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4301                                       struct btrfs_fs_info *fs_info)
4302 {
4303         struct rb_node *node;
4304         struct btrfs_delayed_ref_root *delayed_refs;
4305         struct btrfs_delayed_ref_node *ref;
4306         int ret = 0;
4307
4308         delayed_refs = &trans->delayed_refs;
4309
4310         spin_lock(&delayed_refs->lock);
4311         if (atomic_read(&delayed_refs->num_entries) == 0) {
4312                 spin_unlock(&delayed_refs->lock);
4313                 btrfs_info(fs_info, "delayed_refs has NO entry");
4314                 return ret;
4315         }
4316
4317         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4318                 struct btrfs_delayed_ref_head *head;
4319                 struct btrfs_delayed_ref_node *tmp;
4320                 bool pin_bytes = false;
4321
4322                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4323                                 href_node);
4324                 if (!mutex_trylock(&head->mutex)) {
4325                         refcount_inc(&head->node.refs);
4326                         spin_unlock(&delayed_refs->lock);
4327
4328                         mutex_lock(&head->mutex);
4329                         mutex_unlock(&head->mutex);
4330                         btrfs_put_delayed_ref(&head->node);
4331                         spin_lock(&delayed_refs->lock);
4332                         continue;
4333                 }
4334                 spin_lock(&head->lock);
4335                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4336                                                  list) {
4337                         ref->in_tree = 0;
4338                         list_del(&ref->list);
4339                         if (!list_empty(&ref->add_list))
4340                                 list_del(&ref->add_list);
4341                         atomic_dec(&delayed_refs->num_entries);
4342                         btrfs_put_delayed_ref(ref);
4343                 }
4344                 if (head->must_insert_reserved)
4345                         pin_bytes = true;
4346                 btrfs_free_delayed_extent_op(head->extent_op);
4347                 delayed_refs->num_heads--;
4348                 if (head->processing == 0)
4349                         delayed_refs->num_heads_ready--;
4350                 atomic_dec(&delayed_refs->num_entries);
4351                 head->node.in_tree = 0;
4352                 rb_erase(&head->href_node, &delayed_refs->href_root);
4353                 spin_unlock(&head->lock);
4354                 spin_unlock(&delayed_refs->lock);
4355                 mutex_unlock(&head->mutex);
4356
4357                 if (pin_bytes)
4358                         btrfs_pin_extent(fs_info, head->node.bytenr,
4359                                          head->node.num_bytes, 1);
4360                 btrfs_put_delayed_ref(&head->node);
4361                 cond_resched();
4362                 spin_lock(&delayed_refs->lock);
4363         }
4364
4365         spin_unlock(&delayed_refs->lock);
4366
4367         return ret;
4368 }
4369
4370 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4371 {
4372         struct btrfs_inode *btrfs_inode;
4373         struct list_head splice;
4374
4375         INIT_LIST_HEAD(&splice);
4376
4377         spin_lock(&root->delalloc_lock);
4378         list_splice_init(&root->delalloc_inodes, &splice);
4379
4380         while (!list_empty(&splice)) {
4381                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4382                                                delalloc_inodes);
4383
4384                 list_del_init(&btrfs_inode->delalloc_inodes);
4385                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4386                           &btrfs_inode->runtime_flags);
4387                 spin_unlock(&root->delalloc_lock);
4388
4389                 btrfs_invalidate_inodes(btrfs_inode->root);
4390
4391                 spin_lock(&root->delalloc_lock);
4392         }
4393
4394         spin_unlock(&root->delalloc_lock);
4395 }
4396
4397 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4398 {
4399         struct btrfs_root *root;
4400         struct list_head splice;
4401
4402         INIT_LIST_HEAD(&splice);
4403
4404         spin_lock(&fs_info->delalloc_root_lock);
4405         list_splice_init(&fs_info->delalloc_roots, &splice);
4406         while (!list_empty(&splice)) {
4407                 root = list_first_entry(&splice, struct btrfs_root,
4408                                          delalloc_root);
4409                 list_del_init(&root->delalloc_root);
4410                 root = btrfs_grab_fs_root(root);
4411                 BUG_ON(!root);
4412                 spin_unlock(&fs_info->delalloc_root_lock);
4413
4414                 btrfs_destroy_delalloc_inodes(root);
4415                 btrfs_put_fs_root(root);
4416
4417                 spin_lock(&fs_info->delalloc_root_lock);
4418         }
4419         spin_unlock(&fs_info->delalloc_root_lock);
4420 }
4421
4422 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4423                                         struct extent_io_tree *dirty_pages,
4424                                         int mark)
4425 {
4426         int ret;
4427         struct extent_buffer *eb;
4428         u64 start = 0;
4429         u64 end;
4430
4431         while (1) {
4432                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4433                                             mark, NULL);
4434                 if (ret)
4435                         break;
4436
4437                 clear_extent_bits(dirty_pages, start, end, mark);
4438                 while (start <= end) {
4439                         eb = find_extent_buffer(fs_info, start);
4440                         start += fs_info->nodesize;
4441                         if (!eb)
4442                                 continue;
4443                         wait_on_extent_buffer_writeback(eb);
4444
4445                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4446                                                &eb->bflags))
4447                                 clear_extent_buffer_dirty(eb);
4448                         free_extent_buffer_stale(eb);
4449                 }
4450         }
4451
4452         return ret;
4453 }
4454
4455 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4456                                        struct extent_io_tree *pinned_extents)
4457 {
4458         struct extent_io_tree *unpin;
4459         u64 start;
4460         u64 end;
4461         int ret;
4462         bool loop = true;
4463
4464         unpin = pinned_extents;
4465 again:
4466         while (1) {
4467                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4468                                             EXTENT_DIRTY, NULL);
4469                 if (ret)
4470                         break;
4471
4472                 clear_extent_dirty(unpin, start, end);
4473                 btrfs_error_unpin_extent_range(fs_info, start, end);
4474                 cond_resched();
4475         }
4476
4477         if (loop) {
4478                 if (unpin == &fs_info->freed_extents[0])
4479                         unpin = &fs_info->freed_extents[1];
4480                 else
4481                         unpin = &fs_info->freed_extents[0];
4482                 loop = false;
4483                 goto again;
4484         }
4485
4486         return 0;
4487 }
4488
4489 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4490 {
4491         struct inode *inode;
4492
4493         inode = cache->io_ctl.inode;
4494         if (inode) {
4495                 invalidate_inode_pages2(inode->i_mapping);
4496                 BTRFS_I(inode)->generation = 0;
4497                 cache->io_ctl.inode = NULL;
4498                 iput(inode);
4499         }
4500         btrfs_put_block_group(cache);
4501 }
4502
4503 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4504                              struct btrfs_fs_info *fs_info)
4505 {
4506         struct btrfs_block_group_cache *cache;
4507
4508         spin_lock(&cur_trans->dirty_bgs_lock);
4509         while (!list_empty(&cur_trans->dirty_bgs)) {
4510                 cache = list_first_entry(&cur_trans->dirty_bgs,
4511                                          struct btrfs_block_group_cache,
4512                                          dirty_list);
4513                 if (!cache) {
4514                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4515                         spin_unlock(&cur_trans->dirty_bgs_lock);
4516                         return;
4517                 }
4518
4519                 if (!list_empty(&cache->io_list)) {
4520                         spin_unlock(&cur_trans->dirty_bgs_lock);
4521                         list_del_init(&cache->io_list);
4522                         btrfs_cleanup_bg_io(cache);
4523                         spin_lock(&cur_trans->dirty_bgs_lock);
4524                 }
4525
4526                 list_del_init(&cache->dirty_list);
4527                 spin_lock(&cache->lock);
4528                 cache->disk_cache_state = BTRFS_DC_ERROR;
4529                 spin_unlock(&cache->lock);
4530
4531                 spin_unlock(&cur_trans->dirty_bgs_lock);
4532                 btrfs_put_block_group(cache);
4533                 spin_lock(&cur_trans->dirty_bgs_lock);
4534         }
4535         spin_unlock(&cur_trans->dirty_bgs_lock);
4536
4537         while (!list_empty(&cur_trans->io_bgs)) {
4538                 cache = list_first_entry(&cur_trans->io_bgs,
4539                                          struct btrfs_block_group_cache,
4540                                          io_list);
4541                 if (!cache) {
4542                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4543                         return;
4544                 }
4545
4546                 list_del_init(&cache->io_list);
4547                 spin_lock(&cache->lock);
4548                 cache->disk_cache_state = BTRFS_DC_ERROR;
4549                 spin_unlock(&cache->lock);
4550                 btrfs_cleanup_bg_io(cache);
4551         }
4552 }
4553
4554 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4555                                    struct btrfs_fs_info *fs_info)
4556 {
4557         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4558         ASSERT(list_empty(&cur_trans->dirty_bgs));
4559         ASSERT(list_empty(&cur_trans->io_bgs));
4560
4561         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4562
4563         cur_trans->state = TRANS_STATE_COMMIT_START;
4564         wake_up(&fs_info->transaction_blocked_wait);
4565
4566         cur_trans->state = TRANS_STATE_UNBLOCKED;
4567         wake_up(&fs_info->transaction_wait);
4568
4569         btrfs_destroy_delayed_inodes(fs_info);
4570         btrfs_assert_delayed_root_empty(fs_info);
4571
4572         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4573                                      EXTENT_DIRTY);
4574         btrfs_destroy_pinned_extent(fs_info,
4575                                     fs_info->pinned_extents);
4576
4577         cur_trans->state =TRANS_STATE_COMPLETED;
4578         wake_up(&cur_trans->commit_wait);
4579 }
4580
4581 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4582 {
4583         struct btrfs_transaction *t;
4584
4585         mutex_lock(&fs_info->transaction_kthread_mutex);
4586
4587         spin_lock(&fs_info->trans_lock);
4588         while (!list_empty(&fs_info->trans_list)) {
4589                 t = list_first_entry(&fs_info->trans_list,
4590                                      struct btrfs_transaction, list);
4591                 if (t->state >= TRANS_STATE_COMMIT_START) {
4592                         refcount_inc(&t->use_count);
4593                         spin_unlock(&fs_info->trans_lock);
4594                         btrfs_wait_for_commit(fs_info, t->transid);
4595                         btrfs_put_transaction(t);
4596                         spin_lock(&fs_info->trans_lock);
4597                         continue;
4598                 }
4599                 if (t == fs_info->running_transaction) {
4600                         t->state = TRANS_STATE_COMMIT_DOING;
4601                         spin_unlock(&fs_info->trans_lock);
4602                         /*
4603                          * We wait for 0 num_writers since we don't hold a trans
4604                          * handle open currently for this transaction.
4605                          */
4606                         wait_event(t->writer_wait,
4607                                    atomic_read(&t->num_writers) == 0);
4608                 } else {
4609                         spin_unlock(&fs_info->trans_lock);
4610                 }
4611                 btrfs_cleanup_one_transaction(t, fs_info);
4612
4613                 spin_lock(&fs_info->trans_lock);
4614                 if (t == fs_info->running_transaction)
4615                         fs_info->running_transaction = NULL;
4616                 list_del_init(&t->list);
4617                 spin_unlock(&fs_info->trans_lock);
4618
4619                 btrfs_put_transaction(t);
4620                 trace_btrfs_transaction_commit(fs_info->tree_root);
4621                 spin_lock(&fs_info->trans_lock);
4622         }
4623         spin_unlock(&fs_info->trans_lock);
4624         btrfs_destroy_all_ordered_extents(fs_info);
4625         btrfs_destroy_delayed_inodes(fs_info);
4626         btrfs_assert_delayed_root_empty(fs_info);
4627         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4628         btrfs_destroy_all_delalloc_inodes(fs_info);
4629         mutex_unlock(&fs_info->transaction_kthread_mutex);
4630
4631         return 0;
4632 }
4633
4634 static struct btrfs_fs_info *btree_fs_info(void *private_data)
4635 {
4636         struct inode *inode = private_data;
4637         return btrfs_sb(inode->i_sb);
4638 }
4639
4640 static const struct extent_io_ops btree_extent_io_ops = {
4641         /* mandatory callbacks */
4642         .submit_bio_hook = btree_submit_bio_hook,
4643         .readpage_end_io_hook = btree_readpage_end_io_hook,
4644         /* note we're sharing with inode.c for the merge bio hook */
4645         .merge_bio_hook = btrfs_merge_bio_hook,
4646         .readpage_io_failed_hook = btree_io_failed_hook,
4647         .set_range_writeback = btrfs_set_range_writeback,
4648         .tree_fs_info = btree_fs_info,
4649
4650         /* optional callbacks */
4651 };