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 list_head list;
  93         struct btrfs_work work;
  94 };
  95
  96 static struct kmem_cache *btrfs_end_io_wq_cache;
  97
  98 int __init btrfs_end_io_wq_init(void)
  99 {
 100         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 101                                         sizeof(struct btrfs_end_io_wq),
 102                                         0,
 103                                         SLAB_MEM_SPREAD,
 104                                         NULL);
 105         if (!btrfs_end_io_wq_cache)
 106                 return -ENOMEM;
 107         return 0;
 108 }
 109
 110 void btrfs_end_io_wq_exit(void)
 111 {
 112         kmem_cache_destroy(btrfs_end_io_wq_cache);
 113 }
 114
 115 /*
 116  * async submit bios are used to offload expensive checksumming
 117  * onto the worker threads.  They checksum file and metadata bios
 118  * just before they are sent down the IO stack.
 119  */
 120 struct async_submit_bio {
 121         struct inode *inode;
 122         struct bio *bio;
 123         struct list_head list;
 124         extent_submit_bio_hook_t *submit_bio_start;
 125         extent_submit_bio_hook_t *submit_bio_done;
 126         int mirror_num;
 127         unsigned long bio_flags;
 128         /*
 129          * bio_offset is optional, can be used if the pages in the bio
 130          * can't tell us where in the file the bio should go
 131          */
 132         u64 bio_offset;
 133         struct btrfs_work work;
 134         int error;
 135 };
 136
 137 /*
 138  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 139  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 140  * the level the eb occupies in the tree.
 141  *
 142  * Different roots are used for different purposes and may nest inside each
 143  * other and they require separate keysets.  As lockdep keys should be
 144  * static, assign keysets according to the purpose of the root as indicated
 145  * by btrfs_root->objectid.  This ensures that all special purpose roots
 146  * have separate keysets.
 147  *
 148  * Lock-nesting across peer nodes is always done with the immediate parent
 149  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 150  * subclass to avoid triggering lockdep warning in such cases.
 151  *
 152  * The key is set by the readpage_end_io_hook after the buffer has passed
 153  * csum validation but before the pages are unlocked.  It is also set by
 154  * btrfs_init_new_buffer on freshly allocated blocks.
 155  *
 156  * We also add a check to make sure the highest level of the tree is the
 157  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 158  * needs update as well.
 159  */
 160 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 161 # if BTRFS_MAX_LEVEL != 8
 162 #  error
 163 # endif
 164
 165 static struct btrfs_lockdep_keyset {
 166         u64                     id;             /* root objectid */
 167         const char              *name_stem;     /* lock name stem */
 168         char                    names[BTRFS_MAX_LEVEL + 1][20];
 169         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 170 } btrfs_lockdep_keysets[] = {
 171         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 172         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 173         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 174         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 175         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 176         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 177         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 178         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 179         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 180         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 181         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 182         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 183         { .id = 0,                              .name_stem = "tree"     },
 184 };
 185
 186 void __init btrfs_init_lockdep(void)
 187 {
 188         int i, j;
 189
 190         /* initialize lockdep class names */
 191         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 192                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 193
 194                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 195                         snprintf(ks->names[j], sizeof(ks->names[j]),
 196                                  "btrfs-%s-%02d", ks->name_stem, j);
 197         }
 198 }
 199
 200 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 201                                     int level)
 202 {
 203         struct btrfs_lockdep_keyset *ks;
 204
 205         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 206
 207         /* find the matching keyset, id 0 is the default entry */
 208         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 209                 if (ks->id == objectid)
 210                         break;
 211
 212         lockdep_set_class_and_name(&eb->lock,
 213                                    &ks->keys[level], ks->names[level]);
 214 }
 215
 216 #endif
 217
 218 /*
 219  * extents on the btree inode are pretty simple, there's one extent
 220  * that covers the entire device
 221  */
 222 static struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 223                 struct page *page, size_t pg_offset, u64 start, u64 len,
 224                 int create)
 225 {
 226         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
 227         struct extent_map_tree *em_tree = &inode->extent_tree;
 228         struct extent_map *em;
 229         int ret;
 230
 231         read_lock(&em_tree->lock);
 232         em = lookup_extent_mapping(em_tree, start, len);
 233         if (em) {
 234                 em->bdev = fs_info->fs_devices->latest_bdev;
 235                 read_unlock(&em_tree->lock);
 236                 goto out;
 237         }
 238         read_unlock(&em_tree->lock);
 239
 240         em = alloc_extent_map();
 241         if (!em) {
 242                 em = ERR_PTR(-ENOMEM);
 243                 goto out;
 244         }
 245         em->start = 0;
 246         em->len = (u64)-1;
 247         em->block_len = (u64)-1;
 248         em->block_start = 0;
 249         em->bdev = fs_info->fs_devices->latest_bdev;
 250
 251         write_lock(&em_tree->lock);
 252         ret = add_extent_mapping(em_tree, em, 0);
 253         if (ret == -EEXIST) {
 254                 free_extent_map(em);
 255                 em = lookup_extent_mapping(em_tree, start, len);
 256                 if (!em)
 257                         em = ERR_PTR(-EIO);
 258         } else if (ret) {
 259                 free_extent_map(em);
 260                 em = ERR_PTR(ret);
 261         }
 262         write_unlock(&em_tree->lock);
 263
 264 out:
 265         return em;
 266 }
 267
 268 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 269 {
 270         return btrfs_crc32c(seed, data, len);
 271 }
 272
 273 void btrfs_csum_final(u32 crc, u8 *result)
 274 {
 275         put_unaligned_le32(~crc, result);
 276 }
 277
 278 /*
 279  * compute the csum for a btree block, and either verify it or write it
 280  * into the csum field of the block.
 281  */
 282 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 283                            struct extent_buffer *buf,
 284                            int verify)
 285 {
 286         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 287         char *result = NULL;
 288         unsigned long len;
 289         unsigned long cur_len;
 290         unsigned long offset = BTRFS_CSUM_SIZE;
 291         char *kaddr;
 292         unsigned long map_start;
 293         unsigned long map_len;
 294         int err;
 295         u32 crc = ~(u32)0;
 296         unsigned long inline_result;
 297
 298         len = buf->len - offset;
 299         while (len > 0) {
 300                 err = map_private_extent_buffer(buf, offset, 32,
 301                                         &kaddr, &map_start, &map_len);
 302                 if (err)
 303                         return err;
 304                 cur_len = min(len, map_len - (offset - map_start));
 305                 crc = btrfs_csum_data(kaddr + offset - map_start,
 306                                       crc, cur_len);
 307                 len -= cur_len;
 308                 offset += cur_len;
 309         }
 310         if (csum_size > sizeof(inline_result)) {
 311                 result = kzalloc(csum_size, GFP_NOFS);
 312                 if (!result)
 313                         return -ENOMEM;
 314         } else {
 315                 result = (char *)&inline_result;
 316         }
 317
 318         btrfs_csum_final(crc, result);
 319
 320         if (verify) {
 321                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 322                         u32 val;
 323                         u32 found = 0;
 324                         memcpy(&found, result, csum_size);
 325
 326                         read_extent_buffer(buf, &val, 0, csum_size);
 327                         btrfs_warn_rl(fs_info,
 328                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 329                                 fs_info->sb->s_id, buf->start,
 330                                 val, found, btrfs_header_level(buf));
 331                         if (result != (char *)&inline_result)
 332                                 kfree(result);
 333                         return -EUCLEAN;
 334                 }
 335         } else {
 336                 write_extent_buffer(buf, result, 0, csum_size);
 337         }
 338         if (result != (char *)&inline_result)
 339                 kfree(result);
 340         return 0;
 341 }
 342
 343 /*
 344  * we can't consider a given block up to date unless the transid of the
 345  * block matches the transid in the parent node's pointer.  This is how we
 346  * detect blocks that either didn't get written at all or got written
 347  * in the wrong place.
 348  */
 349 static int verify_parent_transid(struct extent_io_tree *io_tree,
 350                                  struct extent_buffer *eb, u64 parent_transid,
 351                                  int atomic)
 352 {
 353         struct extent_state *cached_state = NULL;
 354         int ret;
 355         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 356
 357         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 358                 return 0;
 359
 360         if (atomic)
 361                 return -EAGAIN;
 362
 363         if (need_lock) {
 364                 btrfs_tree_read_lock(eb);
 365                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 366         }
 367
 368         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 369                          &cached_state);
 370         if (extent_buffer_uptodate(eb) &&
 371             btrfs_header_generation(eb) == parent_transid) {
 372                 ret = 0;
 373                 goto out;
 374         }
 375         btrfs_err_rl(eb->fs_info,
 376                 "parent transid verify failed on %llu wanted %llu found %llu",
 377                         eb->start,
 378                         parent_transid, btrfs_header_generation(eb));
 379         ret = 1;
 380
 381         /*
 382          * Things reading via commit roots that don't have normal protection,
 383          * like send, can have a really old block in cache that may point at a
 384          * block that has been freed and re-allocated.  So don't clear uptodate
 385          * if we find an eb that is under IO (dirty/writeback) because we could
 386          * end up reading in the stale data and then writing it back out and
 387          * making everybody very sad.
 388          */
 389         if (!extent_buffer_under_io(eb))
 390                 clear_extent_buffer_uptodate(eb);
 391 out:
 392         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 393                              &cached_state, GFP_NOFS);
 394         if (need_lock)
 395                 btrfs_tree_read_unlock_blocking(eb);
 396         return ret;
 397 }
 398
 399 /*
 400  * Return 0 if the superblock checksum type matches the checksum value of that
 401  * algorithm. Pass the raw disk superblock data.
 402  */
 403 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 404                                   char *raw_disk_sb)
 405 {
 406         struct btrfs_super_block *disk_sb =
 407                 (struct btrfs_super_block *)raw_disk_sb;
 408         u16 csum_type = btrfs_super_csum_type(disk_sb);
 409         int ret = 0;
 410
 411         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 412                 u32 crc = ~(u32)0;
 413                 const int csum_size = sizeof(crc);
 414                 char result[csum_size];
 415
 416                 /*
 417                  * The super_block structure does not span the whole
 418                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 419                  * is filled with zeros and is included in the checksum.
 420                  */
 421                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 422                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 423                 btrfs_csum_final(crc, result);
 424
 425                 if (memcmp(raw_disk_sb, result, csum_size))
 426                         ret = 1;
 427         }
 428
 429         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 430                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 431                                 csum_type);
 432                 ret = 1;
 433         }
 434
 435         return ret;
 436 }
 437
 438 /*
 439  * helper to read a given tree block, doing retries as required when
 440  * the checksums don't match and we have alternate mirrors to try.
 441  */
 442 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 443                                           struct extent_buffer *eb,
 444                                           u64 parent_transid)
 445 {
 446         struct extent_io_tree *io_tree;
 447         int failed = 0;
 448         int ret;
 449         int num_copies = 0;
 450         int mirror_num = 0;
 451         int failed_mirror = 0;
 452
 453         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 454         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 455         while (1) {
 456                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 457                                                btree_get_extent, mirror_num);
 458                 if (!ret) {
 459                         if (!verify_parent_transid(io_tree, eb,
 460                                                    parent_transid, 0))
 461                                 break;
 462                         else
 463                                 ret = -EIO;
 464                 }
 465
 466                 /*
 467                  * This buffer's crc is fine, but its contents are corrupted, so
 468                  * there is no reason to read the other copies, they won't be
 469                  * any less wrong.
 470                  */
 471                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 472                         break;
 473
 474                 num_copies = btrfs_num_copies(fs_info,
 475                                               eb->start, eb->len);
 476                 if (num_copies == 1)
 477                         break;
 478
 479                 if (!failed_mirror) {
 480                         failed = 1;
 481                         failed_mirror = eb->read_mirror;
 482                 }
 483
 484                 mirror_num++;
 485                 if (mirror_num == failed_mirror)
 486                         mirror_num++;
 487
 488                 if (mirror_num > num_copies)
 489                         break;
 490         }
 491
 492         if (failed && !ret && failed_mirror)
 493                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 494
 495         return ret;
 496 }
 497
 498 /*
 499  * checksum a dirty tree block before IO.  This has extra checks to make sure
 500  * we only fill in the checksum field in the first page of a multi-page block
 501  */
 502
 503 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 504 {
 505         u64 start = page_offset(page);
 506         u64 found_start;
 507         struct extent_buffer *eb;
 508
 509         eb = (struct extent_buffer *)page->private;
 510         if (page != eb->pages[0])
 511                 return 0;
 512
 513         found_start = btrfs_header_bytenr(eb);
 514         /*
 515          * Please do not consolidate these warnings into a single if.
 516          * It is useful to know what went wrong.
 517          */
 518         if (WARN_ON(found_start != start))
 519                 return -EUCLEAN;
 520         if (WARN_ON(!PageUptodate(page)))
 521                 return -EUCLEAN;
 522
 523         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 524                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 525
 526         return csum_tree_block(fs_info, eb, 0);
 527 }
 528
 529 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 530                                  struct extent_buffer *eb)
 531 {
 532         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 533         u8 fsid[BTRFS_UUID_SIZE];
 534         int ret = 1;
 535
 536         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 537         while (fs_devices) {
 538                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 539                         ret = 0;
 540                         break;
 541                 }
 542                 fs_devices = fs_devices->seed;
 543         }
 544         return ret;
 545 }
 546
 547 #define CORRUPT(reason, eb, root, slot)                                 \
 548         btrfs_crit(root->fs_info,                                       \
 549                    "corrupt %s, %s: block=%llu, root=%llu, slot=%d",    \
 550                    btrfs_header_level(eb) == 0 ? "leaf" : "node",       \
 551                    reason, btrfs_header_bytenr(eb), root->objectid, slot)
 552
 553 static noinline int check_leaf(struct btrfs_root *root,
 554                                struct extent_buffer *leaf)
 555 {
 556         struct btrfs_fs_info *fs_info = root->fs_info;
 557         struct btrfs_key key;
 558         struct btrfs_key leaf_key;
 559         u32 nritems = btrfs_header_nritems(leaf);
 560         int slot;
 561
 562         /*
 563          * Extent buffers from a relocation tree have a owner field that
 564          * corresponds to the subvolume tree they are based on. So just from an
 565          * extent buffer alone we can not find out what is the id of the
 566          * corresponding subvolume tree, so we can not figure out if the extent
 567          * buffer corresponds to the root of the relocation tree or not. So skip
 568          * this check for relocation trees.
 569          */
 570         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
 571                 struct btrfs_root *check_root;
 572
 573                 key.objectid = btrfs_header_owner(leaf);
 574                 key.type = BTRFS_ROOT_ITEM_KEY;
 575                 key.offset = (u64)-1;
 576
 577                 check_root = btrfs_get_fs_root(fs_info, &key, false);
 578                 /*
 579                  * The only reason we also check NULL here is that during
 580                  * open_ctree() some roots has not yet been set up.
 581                  */
 582                 if (!IS_ERR_OR_NULL(check_root)) {
 583                         struct extent_buffer *eb;
 584
 585                         eb = btrfs_root_node(check_root);
 586                         /* if leaf is the root, then it's fine */
 587                         if (leaf != eb) {
 588                                 CORRUPT("non-root leaf's nritems is 0",
 589                                         leaf, check_root, 0);
 590                                 free_extent_buffer(eb);
 591                                 return -EIO;
 592                         }
 593                         free_extent_buffer(eb);
 594                 }
 595                 return 0;
 596         }
 597
 598         if (nritems == 0)
 599                 return 0;
 600
 601         /* Check the 0 item */
 602         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 603             BTRFS_LEAF_DATA_SIZE(fs_info)) {
 604                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 605                 return -EIO;
 606         }
 607
 608         /*
 609          * Check to make sure each items keys are in the correct order and their
 610          * offsets make sense.  We only have to loop through nritems-1 because
 611          * we check the current slot against the next slot, which verifies the
 612          * next slot's offset+size makes sense and that the current's slot
 613          * offset is correct.
 614          */
 615         for (slot = 0; slot < nritems - 1; slot++) {
 616                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 617                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 618
 619                 /* Make sure the keys are in the right order */
 620                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 621                         CORRUPT("bad key order", leaf, root, slot);
 622                         return -EIO;
 623                 }
 624
 625                 /*
 626                  * Make sure the offset and ends are right, remember that the
 627                  * item data starts at the end of the leaf and grows towards the
 628                  * front.
 629                  */
 630                 if (btrfs_item_offset_nr(leaf, slot) !=
 631                         btrfs_item_end_nr(leaf, slot + 1)) {
 632                         CORRUPT("slot offset bad", leaf, root, slot);
 633                         return -EIO;
 634                 }
 635
 636                 /*
 637                  * Check to make sure that we don't point outside of the leaf,
 638                  * just in case all the items are consistent to each other, but
 639                  * all point outside of the leaf.
 640                  */
 641                 if (btrfs_item_end_nr(leaf, slot) >
 642                     BTRFS_LEAF_DATA_SIZE(fs_info)) {
 643                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 644                         return -EIO;
 645                 }
 646         }
 647
 648         return 0;
 649 }
 650
 651 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 652 {
 653         unsigned long nr = btrfs_header_nritems(node);
 654         struct btrfs_key key, next_key;
 655         int slot;
 656         u64 bytenr;
 657         int ret = 0;
 658
 659         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
 660                 btrfs_crit(root->fs_info,
 661                            "corrupt node: block %llu root %llu nritems %lu",
 662                            node->start, root->objectid, nr);
 663                 return -EIO;
 664         }
 665
 666         for (slot = 0; slot < nr - 1; slot++) {
 667                 bytenr = btrfs_node_blockptr(node, slot);
 668                 btrfs_node_key_to_cpu(node, &key, slot);
 669                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
 670
 671                 if (!bytenr) {
 672                         CORRUPT("invalid item slot", node, root, slot);
 673                         ret = -EIO;
 674                         goto out;
 675                 }
 676
 677                 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
 678                         CORRUPT("bad key order", node, root, slot);
 679                         ret = -EIO;
 680                         goto out;
 681                 }
 682         }
 683 out:
 684         return ret;
 685 }
 686
 687 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 688                                       u64 phy_offset, struct page *page,
 689                                       u64 start, u64 end, int mirror)
 690 {
 691         u64 found_start;
 692         int found_level;
 693         struct extent_buffer *eb;
 694         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 695         struct btrfs_fs_info *fs_info = root->fs_info;
 696         int ret = 0;
 697         int reads_done;
 698
 699         if (!page->private)
 700                 goto out;
 701
 702         eb = (struct extent_buffer *)page->private;
 703
 704         /* the pending IO might have been the only thing that kept this buffer
 705          * in memory.  Make sure we have a ref for all this other checks
 706          */
 707         extent_buffer_get(eb);
 708
 709         reads_done = atomic_dec_and_test(&eb->io_pages);
 710         if (!reads_done)
 711                 goto err;
 712
 713         eb->read_mirror = mirror;
 714         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 715                 ret = -EIO;
 716                 goto err;
 717         }
 718
 719         found_start = btrfs_header_bytenr(eb);
 720         if (found_start != eb->start) {
 721                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 722                              found_start, eb->start);
 723                 ret = -EIO;
 724                 goto err;
 725         }
 726         if (check_tree_block_fsid(fs_info, eb)) {
 727                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 728                              eb->start);
 729                 ret = -EIO;
 730                 goto err;
 731         }
 732         found_level = btrfs_header_level(eb);
 733         if (found_level >= BTRFS_MAX_LEVEL) {
 734                 btrfs_err(fs_info, "bad tree block level %d",
 735                           (int)btrfs_header_level(eb));
 736                 ret = -EIO;
 737                 goto err;
 738         }
 739
 740         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 741                                        eb, found_level);
 742
 743         ret = csum_tree_block(fs_info, eb, 1);
 744         if (ret)
 745                 goto err;
 746
 747         /*
 748          * If this is a leaf block and it is corrupt, set the corrupt bit so
 749          * that we don't try and read the other copies of this block, just
 750          * return -EIO.
 751          */
 752         if (found_level == 0 && check_leaf(root, eb)) {
 753                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 754                 ret = -EIO;
 755         }
 756
 757         if (found_level > 0 && check_node(root, eb))
 758                 ret = -EIO;
 759
 760         if (!ret)
 761                 set_extent_buffer_uptodate(eb);
 762 err:
 763         if (reads_done &&
 764             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 765                 btree_readahead_hook(eb, ret);
 766
 767         if (ret) {
 768                 /*
 769                  * our io error hook is going to dec the io pages
 770                  * again, we have to make sure it has something
 771                  * to decrement
 772                  */
 773                 atomic_inc(&eb->io_pages);
 774                 clear_extent_buffer_uptodate(eb);
 775         }
 776         free_extent_buffer(eb);
 777 out:
 778         return ret;
 779 }
 780
 781 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 782 {
 783         struct extent_buffer *eb;
 784
 785         eb = (struct extent_buffer *)page->private;
 786         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 787         eb->read_mirror = failed_mirror;
 788         atomic_dec(&eb->io_pages);
 789         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 790                 btree_readahead_hook(eb, -EIO);
 791         return -EIO;    /* we fixed nothing */
 792 }
 793
 794 static void end_workqueue_bio(struct bio *bio)
 795 {
 796         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 797         struct btrfs_fs_info *fs_info;
 798         struct btrfs_workqueue *wq;
 799         btrfs_work_func_t func;
 800
 801         fs_info = end_io_wq->info;
 802         end_io_wq->error = bio->bi_error;
 803
 804         if (bio_op(bio) == REQ_OP_WRITE) {
 805                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 806                         wq = fs_info->endio_meta_write_workers;
 807                         func = btrfs_endio_meta_write_helper;
 808                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 809                         wq = fs_info->endio_freespace_worker;
 810                         func = btrfs_freespace_write_helper;
 811                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 812                         wq = fs_info->endio_raid56_workers;
 813                         func = btrfs_endio_raid56_helper;
 814                 } else {
 815                         wq = fs_info->endio_write_workers;
 816                         func = btrfs_endio_write_helper;
 817                 }
 818         } else {
 819                 if (unlikely(end_io_wq->metadata ==
 820                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 821                         wq = fs_info->endio_repair_workers;
 822                         func = btrfs_endio_repair_helper;
 823                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 824                         wq = fs_info->endio_raid56_workers;
 825                         func = btrfs_endio_raid56_helper;
 826                 } else if (end_io_wq->metadata) {
 827                         wq = fs_info->endio_meta_workers;
 828                         func = btrfs_endio_meta_helper;
 829                 } else {
 830                         wq = fs_info->endio_workers;
 831                         func = btrfs_endio_helper;
 832                 }
 833         }
 834
 835         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 836         btrfs_queue_work(wq, &end_io_wq->work);
 837 }
 838
 839 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 840                         enum btrfs_wq_endio_type metadata)
 841 {
 842         struct btrfs_end_io_wq *end_io_wq;
 843
 844         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 845         if (!end_io_wq)
 846                 return -ENOMEM;
 847
 848         end_io_wq->private = bio->bi_private;
 849         end_io_wq->end_io = bio->bi_end_io;
 850         end_io_wq->info = info;
 851         end_io_wq->error = 0;
 852         end_io_wq->bio = bio;
 853         end_io_wq->metadata = metadata;
 854
 855         bio->bi_private = end_io_wq;
 856         bio->bi_end_io = end_workqueue_bio;
 857         return 0;
 858 }
 859
 860 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 861 {
 862         unsigned long limit = min_t(unsigned long,
 863                                     info->thread_pool_size,
 864                                     info->fs_devices->open_devices);
 865         return 256 * limit;
 866 }
 867
 868 static void run_one_async_start(struct btrfs_work *work)
 869 {
 870         struct async_submit_bio *async;
 871         int ret;
 872
 873         async = container_of(work, struct  async_submit_bio, work);
 874         ret = async->submit_bio_start(async->inode, async->bio,
 875                                       async->mirror_num, async->bio_flags,
 876                                       async->bio_offset);
 877         if (ret)
 878                 async->error = ret;
 879 }
 880
 881 static void run_one_async_done(struct btrfs_work *work)
 882 {
 883         struct btrfs_fs_info *fs_info;
 884         struct async_submit_bio *async;
 885         int limit;
 886
 887         async = container_of(work, struct  async_submit_bio, work);
 888         fs_info = BTRFS_I(async->inode)->root->fs_info;
 889
 890         limit = btrfs_async_submit_limit(fs_info);
 891         limit = limit * 2 / 3;
 892
 893         /*
 894          * atomic_dec_return implies a barrier for waitqueue_active
 895          */
 896         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 897             waitqueue_active(&fs_info->async_submit_wait))
 898                 wake_up(&fs_info->async_submit_wait);
 899
 900         /* If an error occurred we just want to clean up the bio and move on */
 901         if (async->error) {
 902                 async->bio->bi_error = async->error;
 903                 bio_endio(async->bio);
 904                 return;
 905         }
 906
 907         async->submit_bio_done(async->inode, async->bio, async->mirror_num,
 908                                async->bio_flags, async->bio_offset);
 909 }
 910
 911 static void run_one_async_free(struct btrfs_work *work)
 912 {
 913         struct async_submit_bio *async;
 914
 915         async = container_of(work, struct  async_submit_bio, work);
 916         kfree(async);
 917 }
 918
 919 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 920                         struct bio *bio, int mirror_num,
 921                         unsigned long bio_flags,
 922                         u64 bio_offset,
 923                         extent_submit_bio_hook_t *submit_bio_start,
 924                         extent_submit_bio_hook_t *submit_bio_done)
 925 {
 926         struct async_submit_bio *async;
 927
 928         async = kmalloc(sizeof(*async), GFP_NOFS);
 929         if (!async)
 930                 return -ENOMEM;
 931
 932         async->inode = inode;
 933         async->bio = bio;
 934         async->mirror_num = mirror_num;
 935         async->submit_bio_start = submit_bio_start;
 936         async->submit_bio_done = submit_bio_done;
 937
 938         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 939                         run_one_async_done, run_one_async_free);
 940
 941         async->bio_flags = bio_flags;
 942         async->bio_offset = bio_offset;
 943
 944         async->error = 0;
 945
 946         atomic_inc(&fs_info->nr_async_submits);
 947
 948         if (op_is_sync(bio->bi_opf))
 949                 btrfs_set_work_high_priority(&async->work);
 950
 951         btrfs_queue_work(fs_info->workers, &async->work);
 952
 953         while (atomic_read(&fs_info->async_submit_draining) &&
 954               atomic_read(&fs_info->nr_async_submits)) {
 955                 wait_event(fs_info->async_submit_wait,
 956                            (atomic_read(&fs_info->nr_async_submits) == 0));
 957         }
 958
 959         return 0;
 960 }
 961
 962 static int btree_csum_one_bio(struct bio *bio)
 963 {
 964         struct bio_vec *bvec;
 965         struct btrfs_root *root;
 966         int i, ret = 0;
 967
 968         bio_for_each_segment_all(bvec, bio, i) {
 969                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 970                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 971                 if (ret)
 972                         break;
 973         }
 974
 975         return ret;
 976 }
 977
 978 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
 979                                     int mirror_num, unsigned long bio_flags,
 980                                     u64 bio_offset)
 981 {
 982         /*
 983          * when we're called for a write, we're already in the async
 984          * submission context.  Just jump into btrfs_map_bio
 985          */
 986         return btree_csum_one_bio(bio);
 987 }
 988
 989 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
 990                                  int mirror_num, unsigned long bio_flags,
 991                                  u64 bio_offset)
 992 {
 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(struct inode *inode, struct bio *bio,
1019                                  int mirror_num, unsigned long bio_flags,
1020                                  u64 bio_offset)
1021 {
1022         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1023         int async = check_async_write(bio_flags);
1024         int ret;
1025
1026         if (bio_op(bio) != REQ_OP_WRITE) {
1027                 /*
1028                  * called for a read, do the setup so that checksum validation
1029                  * can happen in the async kernel threads
1030                  */
1031                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1032                                           BTRFS_WQ_ENDIO_METADATA);
1033                 if (ret)
1034                         goto out_w_error;
1035                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1036         } else if (!async) {
1037                 ret = btree_csum_one_bio(bio);
1038                 if (ret)
1039                         goto out_w_error;
1040                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1041         } else {
1042                 /*
1043                  * kthread helpers are used to submit writes so that
1044                  * checksumming can happen in parallel across all CPUs
1045                  */
1046                 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1047                                           bio_offset,
1048                                           __btree_submit_bio_start,
1049                                           __btree_submit_bio_done);
1050         }
1051
1052         if (ret)
1053                 goto out_w_error;
1054         return 0;
1055
1056 out_w_error:
1057         bio->bi_error = ret;
1058         bio_endio(bio);
1059         return ret;
1060 }
1061
1062 #ifdef CONFIG_MIGRATION
1063 static int btree_migratepage(struct address_space *mapping,
1064                         struct page *newpage, struct page *page,
1065                         enum migrate_mode mode)
1066 {
1067         /*
1068          * we can't safely write a btree page from here,
1069          * we haven't done the locking hook
1070          */
1071         if (PageDirty(page))
1072                 return -EAGAIN;
1073         /*
1074          * Buffers may be managed in a filesystem specific way.
1075          * We must have no buffers or drop them.
1076          */
1077         if (page_has_private(page) &&
1078             !try_to_release_page(page, GFP_KERNEL))
1079                 return -EAGAIN;
1080         return migrate_page(mapping, newpage, page, mode);
1081 }
1082 #endif
1083
1084
1085 static int btree_writepages(struct address_space *mapping,
1086                             struct writeback_control *wbc)
1087 {
1088         struct btrfs_fs_info *fs_info;
1089         int ret;
1090
1091         if (wbc->sync_mode == WB_SYNC_NONE) {
1092
1093                 if (wbc->for_kupdate)
1094                         return 0;
1095
1096                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1097                 /* this is a bit racy, but that's ok */
1098                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1099                                              BTRFS_DIRTY_METADATA_THRESH);
1100                 if (ret < 0)
1101                         return 0;
1102         }
1103         return btree_write_cache_pages(mapping, wbc);
1104 }
1105
1106 static int btree_readpage(struct file *file, struct page *page)
1107 {
1108         struct extent_io_tree *tree;
1109         tree = &BTRFS_I(page->mapping->host)->io_tree;
1110         return extent_read_full_page(tree, page, btree_get_extent, 0);
1111 }
1112
1113 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1114 {
1115         if (PageWriteback(page) || PageDirty(page))
1116                 return 0;
1117
1118         return try_release_extent_buffer(page);
1119 }
1120
1121 static void btree_invalidatepage(struct page *page, unsigned int offset,
1122                                  unsigned int length)
1123 {
1124         struct extent_io_tree *tree;
1125         tree = &BTRFS_I(page->mapping->host)->io_tree;
1126         extent_invalidatepage(tree, page, offset);
1127         btree_releasepage(page, GFP_NOFS);
1128         if (PagePrivate(page)) {
1129                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1130                            "page private not zero on page %llu",
1131                            (unsigned long long)page_offset(page));
1132                 ClearPagePrivate(page);
1133                 set_page_private(page, 0);
1134                 put_page(page);
1135         }
1136 }
1137
1138 static int btree_set_page_dirty(struct page *page)
1139 {
1140 #ifdef DEBUG
1141         struct extent_buffer *eb;
1142
1143         BUG_ON(!PagePrivate(page));
1144         eb = (struct extent_buffer *)page->private;
1145         BUG_ON(!eb);
1146         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1147         BUG_ON(!atomic_read(&eb->refs));
1148         btrfs_assert_tree_locked(eb);
1149 #endif
1150         return __set_page_dirty_nobuffers(page);
1151 }
1152
1153 static const struct address_space_operations btree_aops = {
1154         .readpage       = btree_readpage,
1155         .writepages     = btree_writepages,
1156         .releasepage    = btree_releasepage,
1157         .invalidatepage = btree_invalidatepage,
1158 #ifdef CONFIG_MIGRATION
1159         .migratepage    = btree_migratepage,
1160 #endif
1161         .set_page_dirty = btree_set_page_dirty,
1162 };
1163
1164 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1165 {
1166         struct extent_buffer *buf = NULL;
1167         struct inode *btree_inode = fs_info->btree_inode;
1168
1169         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1170         if (IS_ERR(buf))
1171                 return;
1172         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1173                                  buf, WAIT_NONE, btree_get_extent, 0);
1174         free_extent_buffer(buf);
1175 }
1176
1177 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1178                          int mirror_num, struct extent_buffer **eb)
1179 {
1180         struct extent_buffer *buf = NULL;
1181         struct inode *btree_inode = fs_info->btree_inode;
1182         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1183         int ret;
1184
1185         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1186         if (IS_ERR(buf))
1187                 return 0;
1188
1189         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1190
1191         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1192                                        btree_get_extent, mirror_num);
1193         if (ret) {
1194                 free_extent_buffer(buf);
1195                 return ret;
1196         }
1197
1198         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1199                 free_extent_buffer(buf);
1200                 return -EIO;
1201         } else if (extent_buffer_uptodate(buf)) {
1202                 *eb = buf;
1203         } else {
1204                 free_extent_buffer(buf);
1205         }
1206         return 0;
1207 }
1208
1209 struct extent_buffer *btrfs_find_create_tree_block(
1210                                                 struct btrfs_fs_info *fs_info,
1211                                                 u64 bytenr)
1212 {
1213         if (btrfs_is_testing(fs_info))
1214                 return alloc_test_extent_buffer(fs_info, bytenr);
1215         return alloc_extent_buffer(fs_info, bytenr);
1216 }
1217
1218
1219 int btrfs_write_tree_block(struct extent_buffer *buf)
1220 {
1221         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1222                                         buf->start + buf->len - 1);
1223 }
1224
1225 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1226 {
1227         return filemap_fdatawait_range(buf->pages[0]->mapping,
1228                                        buf->start, buf->start + buf->len - 1);
1229 }
1230
1231 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1232                                       u64 parent_transid)
1233 {
1234         struct extent_buffer *buf = NULL;
1235         int ret;
1236
1237         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1238         if (IS_ERR(buf))
1239                 return buf;
1240
1241         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1242         if (ret) {
1243                 free_extent_buffer(buf);
1244                 return ERR_PTR(ret);
1245         }
1246         return buf;
1247
1248 }
1249
1250 void clean_tree_block(struct btrfs_fs_info *fs_info,
1251                       struct extent_buffer *buf)
1252 {
1253         if (btrfs_header_generation(buf) ==
1254             fs_info->running_transaction->transid) {
1255                 btrfs_assert_tree_locked(buf);
1256
1257                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1258                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1259                                              -buf->len,
1260                                              fs_info->dirty_metadata_batch);
1261                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1262                         btrfs_set_lock_blocking(buf);
1263                         clear_extent_buffer_dirty(buf);
1264                 }
1265         }
1266 }
1267
1268 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1269 {
1270         struct btrfs_subvolume_writers *writers;
1271         int ret;
1272
1273         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1274         if (!writers)
1275                 return ERR_PTR(-ENOMEM);
1276
1277         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1278         if (ret < 0) {
1279                 kfree(writers);
1280                 return ERR_PTR(ret);
1281         }
1282
1283         init_waitqueue_head(&writers->wait);
1284         return writers;
1285 }
1286
1287 static void
1288 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1289 {
1290         percpu_counter_destroy(&writers->counter);
1291         kfree(writers);
1292 }
1293
1294 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1295                          u64 objectid)
1296 {
1297         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1298         root->node = NULL;
1299         root->commit_root = NULL;
1300         root->state = 0;
1301         root->orphan_cleanup_state = 0;
1302
1303         root->objectid = objectid;
1304         root->last_trans = 0;
1305         root->highest_objectid = 0;
1306         root->nr_delalloc_inodes = 0;
1307         root->nr_ordered_extents = 0;
1308         root->name = NULL;
1309         root->inode_tree = RB_ROOT;
1310         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1311         root->block_rsv = NULL;
1312         root->orphan_block_rsv = NULL;
1313
1314         INIT_LIST_HEAD(&root->dirty_list);
1315         INIT_LIST_HEAD(&root->root_list);
1316         INIT_LIST_HEAD(&root->delalloc_inodes);
1317         INIT_LIST_HEAD(&root->delalloc_root);
1318         INIT_LIST_HEAD(&root->ordered_extents);
1319         INIT_LIST_HEAD(&root->ordered_root);
1320         INIT_LIST_HEAD(&root->logged_list[0]);
1321         INIT_LIST_HEAD(&root->logged_list[1]);
1322         spin_lock_init(&root->orphan_lock);
1323         spin_lock_init(&root->inode_lock);
1324         spin_lock_init(&root->delalloc_lock);
1325         spin_lock_init(&root->ordered_extent_lock);
1326         spin_lock_init(&root->accounting_lock);
1327         spin_lock_init(&root->log_extents_lock[0]);
1328         spin_lock_init(&root->log_extents_lock[1]);
1329         mutex_init(&root->objectid_mutex);
1330         mutex_init(&root->log_mutex);
1331         mutex_init(&root->ordered_extent_mutex);
1332         mutex_init(&root->delalloc_mutex);
1333         init_waitqueue_head(&root->log_writer_wait);
1334         init_waitqueue_head(&root->log_commit_wait[0]);
1335         init_waitqueue_head(&root->log_commit_wait[1]);
1336         INIT_LIST_HEAD(&root->log_ctxs[0]);
1337         INIT_LIST_HEAD(&root->log_ctxs[1]);
1338         atomic_set(&root->log_commit[0], 0);
1339         atomic_set(&root->log_commit[1], 0);
1340         atomic_set(&root->log_writers, 0);
1341         atomic_set(&root->log_batch, 0);
1342         atomic_set(&root->orphan_inodes, 0);
1343         refcount_set(&root->refs, 1);
1344         atomic_set(&root->will_be_snapshoted, 0);
1345         atomic64_set(&root->qgroup_meta_rsv, 0);
1346         root->log_transid = 0;
1347         root->log_transid_committed = -1;
1348         root->last_log_commit = 0;
1349         if (!dummy)
1350                 extent_io_tree_init(&root->dirty_log_pages,
1351                                      fs_info->btree_inode->i_mapping);
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->i_mapping);
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->free_chunk_lock);
2630         spin_lock_init(&fs_info->tree_mod_seq_lock);
2631         spin_lock_init(&fs_info->super_lock);
2632         spin_lock_init(&fs_info->qgroup_op_lock);
2633         spin_lock_init(&fs_info->buffer_lock);
2634         spin_lock_init(&fs_info->unused_bgs_lock);
2635         rwlock_init(&fs_info->tree_mod_log_lock);
2636         mutex_init(&fs_info->unused_bg_unpin_mutex);
2637         mutex_init(&fs_info->delete_unused_bgs_mutex);
2638         mutex_init(&fs_info->reloc_mutex);
2639         mutex_init(&fs_info->delalloc_root_mutex);
2640         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2641         seqlock_init(&fs_info->profiles_lock);
2642
2643         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2644         INIT_LIST_HEAD(&fs_info->space_info);
2645         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2646         INIT_LIST_HEAD(&fs_info->unused_bgs);
2647         btrfs_mapping_init(&fs_info->mapping_tree);
2648         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2649                              BTRFS_BLOCK_RSV_GLOBAL);
2650         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2651                              BTRFS_BLOCK_RSV_DELALLOC);
2652         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2653         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2654         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2655         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2656                              BTRFS_BLOCK_RSV_DELOPS);
2657         atomic_set(&fs_info->nr_async_submits, 0);
2658         atomic_set(&fs_info->async_delalloc_pages, 0);
2659         atomic_set(&fs_info->async_submit_draining, 0);
2660         atomic_set(&fs_info->nr_async_bios, 0);
2661         atomic_set(&fs_info->defrag_running, 0);
2662         atomic_set(&fs_info->qgroup_op_seq, 0);
2663         atomic_set(&fs_info->reada_works_cnt, 0);
2664         atomic64_set(&fs_info->tree_mod_seq, 0);
2665         fs_info->fs_frozen = 0;
2666         fs_info->sb = sb;
2667         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2668         fs_info->metadata_ratio = 0;
2669         fs_info->defrag_inodes = RB_ROOT;
2670         fs_info->free_chunk_space = 0;
2671         fs_info->tree_mod_log = RB_ROOT;
2672         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2673         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2674         /* readahead state */
2675         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2676         spin_lock_init(&fs_info->reada_lock);
2677
2678         fs_info->thread_pool_size = min_t(unsigned long,
2679                                           num_online_cpus() + 2, 8);
2680
2681         INIT_LIST_HEAD(&fs_info->ordered_roots);
2682         spin_lock_init(&fs_info->ordered_root_lock);
2683         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2684                                         GFP_KERNEL);
2685         if (!fs_info->delayed_root) {
2686                 err = -ENOMEM;
2687                 goto fail_iput;
2688         }
2689         btrfs_init_delayed_root(fs_info->delayed_root);
2690
2691         btrfs_init_scrub(fs_info);
2692 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2693         fs_info->check_integrity_print_mask = 0;
2694 #endif
2695         btrfs_init_balance(fs_info);
2696         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2697
2698         sb->s_blocksize = 4096;
2699         sb->s_blocksize_bits = blksize_bits(4096);
2700
2701         btrfs_init_btree_inode(fs_info);
2702
2703         spin_lock_init(&fs_info->block_group_cache_lock);
2704         fs_info->block_group_cache_tree = RB_ROOT;
2705         fs_info->first_logical_byte = (u64)-1;
2706
2707         extent_io_tree_init(&fs_info->freed_extents[0],
2708                              fs_info->btree_inode->i_mapping);
2709         extent_io_tree_init(&fs_info->freed_extents[1],
2710                              fs_info->btree_inode->i_mapping);
2711         fs_info->pinned_extents = &fs_info->freed_extents[0];
2712         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2713
2714         mutex_init(&fs_info->ordered_operations_mutex);
2715         mutex_init(&fs_info->tree_log_mutex);
2716         mutex_init(&fs_info->chunk_mutex);
2717         mutex_init(&fs_info->transaction_kthread_mutex);
2718         mutex_init(&fs_info->cleaner_mutex);
2719         mutex_init(&fs_info->volume_mutex);
2720         mutex_init(&fs_info->ro_block_group_mutex);
2721         init_rwsem(&fs_info->commit_root_sem);
2722         init_rwsem(&fs_info->cleanup_work_sem);
2723         init_rwsem(&fs_info->subvol_sem);
2724         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2725
2726         btrfs_init_dev_replace_locks(fs_info);
2727         btrfs_init_qgroup(fs_info);
2728
2729         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2730         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2731
2732         init_waitqueue_head(&fs_info->transaction_throttle);
2733         init_waitqueue_head(&fs_info->transaction_wait);
2734         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2735         init_waitqueue_head(&fs_info->async_submit_wait);
2736
2737         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2738
2739         /* Usable values until the real ones are cached from the superblock */
2740         fs_info->nodesize = 4096;
2741         fs_info->sectorsize = 4096;
2742         fs_info->stripesize = 4096;
2743
2744         ret = btrfs_alloc_stripe_hash_table(fs_info);
2745         if (ret) {
2746                 err = ret;
2747                 goto fail_alloc;
2748         }
2749
2750         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2751
2752         invalidate_bdev(fs_devices->latest_bdev);
2753
2754         /*
2755          * Read super block and check the signature bytes only
2756          */
2757         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2758         if (IS_ERR(bh)) {
2759                 err = PTR_ERR(bh);
2760                 goto fail_alloc;
2761         }
2762
2763         /*
2764          * We want to check superblock checksum, the type is stored inside.
2765          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2766          */
2767         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2768                 btrfs_err(fs_info, "superblock checksum mismatch");
2769                 err = -EINVAL;
2770                 brelse(bh);
2771                 goto fail_alloc;
2772         }
2773
2774         /*
2775          * super_copy is zeroed at allocation time and we never touch the
2776          * following bytes up to INFO_SIZE, the checksum is calculated from
2777          * the whole block of INFO_SIZE
2778          */
2779         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2780         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2781                sizeof(*fs_info->super_for_commit));
2782         brelse(bh);
2783
2784         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2785
2786         ret = btrfs_check_super_valid(fs_info);
2787         if (ret) {
2788                 btrfs_err(fs_info, "superblock contains fatal errors");
2789                 err = -EINVAL;
2790                 goto fail_alloc;
2791         }
2792
2793         disk_super = fs_info->super_copy;
2794         if (!btrfs_super_root(disk_super))
2795                 goto fail_alloc;
2796
2797         /* check FS state, whether FS is broken. */
2798         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2799                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2800
2801         /*
2802          * run through our array of backup supers and setup
2803          * our ring pointer to the oldest one
2804          */
2805         generation = btrfs_super_generation(disk_super);
2806         find_oldest_super_backup(fs_info, generation);
2807
2808         /*
2809          * In the long term, we'll store the compression type in the super
2810          * block, and it'll be used for per file compression control.
2811          */
2812         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2813
2814         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2815         if (ret) {
2816                 err = ret;
2817                 goto fail_alloc;
2818         }
2819
2820         features = btrfs_super_incompat_flags(disk_super) &
2821                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2822         if (features) {
2823                 btrfs_err(fs_info,
2824                     "cannot mount because of unsupported optional features (%llx)",
2825                     features);
2826                 err = -EINVAL;
2827                 goto fail_alloc;
2828         }
2829
2830         features = btrfs_super_incompat_flags(disk_super);
2831         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2832         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2833                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2834
2835         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2836                 btrfs_info(fs_info, "has skinny extents");
2837
2838         /*
2839          * flag our filesystem as having big metadata blocks if
2840          * they are bigger than the page size
2841          */
2842         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2843                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2844                         btrfs_info(fs_info,
2845                                 "flagging fs with big metadata feature");
2846                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2847         }
2848
2849         nodesize = btrfs_super_nodesize(disk_super);
2850         sectorsize = btrfs_super_sectorsize(disk_super);
2851         stripesize = sectorsize;
2852         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2853         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2854
2855         /* Cache block sizes */
2856         fs_info->nodesize = nodesize;
2857         fs_info->sectorsize = sectorsize;
2858         fs_info->stripesize = stripesize;
2859
2860         /*
2861          * mixed block groups end up with duplicate but slightly offset
2862          * extent buffers for the same range.  It leads to corruptions
2863          */
2864         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2865             (sectorsize != nodesize)) {
2866                 btrfs_err(fs_info,
2867 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2868                         nodesize, sectorsize);
2869                 goto fail_alloc;
2870         }
2871
2872         /*
2873          * Needn't use the lock because there is no other task which will
2874          * update the flag.
2875          */
2876         btrfs_set_super_incompat_flags(disk_super, features);
2877
2878         features = btrfs_super_compat_ro_flags(disk_super) &
2879                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2880         if (!(sb->s_flags & MS_RDONLY) && features) {
2881                 btrfs_err(fs_info,
2882         "cannot mount read-write because of unsupported optional features (%llx)",
2883                        features);
2884                 err = -EINVAL;
2885                 goto fail_alloc;
2886         }
2887
2888         max_active = fs_info->thread_pool_size;
2889
2890         ret = btrfs_init_workqueues(fs_info, fs_devices);
2891         if (ret) {
2892                 err = ret;
2893                 goto fail_sb_buffer;
2894         }
2895
2896         sb->s_bdi->congested_fn = btrfs_congested_fn;
2897         sb->s_bdi->congested_data = fs_info;
2898         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2899         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2900         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2901         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2902
2903         sb->s_blocksize = sectorsize;
2904         sb->s_blocksize_bits = blksize_bits(sectorsize);
2905
2906         mutex_lock(&fs_info->chunk_mutex);
2907         ret = btrfs_read_sys_array(fs_info);
2908         mutex_unlock(&fs_info->chunk_mutex);
2909         if (ret) {
2910                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2911                 goto fail_sb_buffer;
2912         }
2913
2914         generation = btrfs_super_chunk_root_generation(disk_super);
2915
2916         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2917
2918         chunk_root->node = read_tree_block(fs_info,
2919                                            btrfs_super_chunk_root(disk_super),
2920                                            generation);
2921         if (IS_ERR(chunk_root->node) ||
2922             !extent_buffer_uptodate(chunk_root->node)) {
2923                 btrfs_err(fs_info, "failed to read chunk root");
2924                 if (!IS_ERR(chunk_root->node))
2925                         free_extent_buffer(chunk_root->node);
2926                 chunk_root->node = NULL;
2927                 goto fail_tree_roots;
2928         }
2929         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2930         chunk_root->commit_root = btrfs_root_node(chunk_root);
2931
2932         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2933            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2934
2935         ret = btrfs_read_chunk_tree(fs_info);
2936         if (ret) {
2937                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2938                 goto fail_tree_roots;
2939         }
2940
2941         /*
2942          * keep the device that is marked to be the target device for the
2943          * dev_replace procedure
2944          */
2945         btrfs_close_extra_devices(fs_devices, 0);
2946
2947         if (!fs_devices->latest_bdev) {
2948                 btrfs_err(fs_info, "failed to read devices");
2949                 goto fail_tree_roots;
2950         }
2951
2952 retry_root_backup:
2953         generation = btrfs_super_generation(disk_super);
2954
2955         tree_root->node = read_tree_block(fs_info,
2956                                           btrfs_super_root(disk_super),
2957                                           generation);
2958         if (IS_ERR(tree_root->node) ||
2959             !extent_buffer_uptodate(tree_root->node)) {
2960                 btrfs_warn(fs_info, "failed to read tree root");
2961                 if (!IS_ERR(tree_root->node))
2962                         free_extent_buffer(tree_root->node);
2963                 tree_root->node = NULL;
2964                 goto recovery_tree_root;
2965         }
2966
2967         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2968         tree_root->commit_root = btrfs_root_node(tree_root);
2969         btrfs_set_root_refs(&tree_root->root_item, 1);
2970
2971         mutex_lock(&tree_root->objectid_mutex);
2972         ret = btrfs_find_highest_objectid(tree_root,
2973                                         &tree_root->highest_objectid);
2974         if (ret) {
2975                 mutex_unlock(&tree_root->objectid_mutex);
2976                 goto recovery_tree_root;
2977         }
2978
2979         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2980
2981         mutex_unlock(&tree_root->objectid_mutex);
2982
2983         ret = btrfs_read_roots(fs_info);
2984         if (ret)
2985                 goto recovery_tree_root;
2986
2987         fs_info->generation = generation;
2988         fs_info->last_trans_committed = generation;
2989
2990         ret = btrfs_recover_balance(fs_info);
2991         if (ret) {
2992                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2993                 goto fail_block_groups;
2994         }
2995
2996         ret = btrfs_init_dev_stats(fs_info);
2997         if (ret) {
2998                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2999                 goto fail_block_groups;
3000         }
3001
3002         ret = btrfs_init_dev_replace(fs_info);
3003         if (ret) {
3004                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3005                 goto fail_block_groups;
3006         }
3007
3008         btrfs_close_extra_devices(fs_devices, 1);
3009
3010         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3011         if (ret) {
3012                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3013                                 ret);
3014                 goto fail_block_groups;
3015         }
3016
3017         ret = btrfs_sysfs_add_device(fs_devices);
3018         if (ret) {
3019                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3020                                 ret);
3021                 goto fail_fsdev_sysfs;
3022         }
3023
3024         ret = btrfs_sysfs_add_mounted(fs_info);
3025         if (ret) {
3026                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3027                 goto fail_fsdev_sysfs;
3028         }
3029
3030         ret = btrfs_init_space_info(fs_info);
3031         if (ret) {
3032                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3033                 goto fail_sysfs;
3034         }
3035
3036         ret = btrfs_read_block_groups(fs_info);
3037         if (ret) {
3038                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3039                 goto fail_sysfs;
3040         }
3041         fs_info->num_tolerated_disk_barrier_failures =
3042                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3043         if (fs_info->fs_devices->missing_devices >
3044              fs_info->num_tolerated_disk_barrier_failures &&
3045             !(sb->s_flags & MS_RDONLY)) {
3046                 btrfs_warn(fs_info,
3047 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3048                         fs_info->fs_devices->missing_devices,
3049                         fs_info->num_tolerated_disk_barrier_failures);
3050                 goto fail_sysfs;
3051         }
3052
3053         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3054                                                "btrfs-cleaner");
3055         if (IS_ERR(fs_info->cleaner_kthread))
3056                 goto fail_sysfs;
3057
3058         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3059                                                    tree_root,
3060                                                    "btrfs-transaction");
3061         if (IS_ERR(fs_info->transaction_kthread))
3062                 goto fail_cleaner;
3063
3064         if (!btrfs_test_opt(fs_info, SSD) &&
3065             !btrfs_test_opt(fs_info, NOSSD) &&
3066             !fs_info->fs_devices->rotating) {
3067                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3068                 btrfs_set_opt(fs_info->mount_opt, SSD);
3069         }
3070
3071         /*
3072          * Mount does not set all options immediately, we can do it now and do
3073          * not have to wait for transaction commit
3074          */
3075         btrfs_apply_pending_changes(fs_info);
3076
3077 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3078         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3079                 ret = btrfsic_mount(fs_info, fs_devices,
3080                                     btrfs_test_opt(fs_info,
3081                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3082                                     1 : 0,
3083                                     fs_info->check_integrity_print_mask);
3084                 if (ret)
3085                         btrfs_warn(fs_info,
3086                                 "failed to initialize integrity check module: %d",
3087                                 ret);
3088         }
3089 #endif
3090         ret = btrfs_read_qgroup_config(fs_info);
3091         if (ret)
3092                 goto fail_trans_kthread;
3093
3094         /* do not make disk changes in broken FS or nologreplay is given */
3095         if (btrfs_super_log_root(disk_super) != 0 &&
3096             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3097                 ret = btrfs_replay_log(fs_info, fs_devices);
3098                 if (ret) {
3099                         err = ret;
3100                         goto fail_qgroup;
3101                 }
3102         }
3103
3104         ret = btrfs_find_orphan_roots(fs_info);
3105         if (ret)
3106                 goto fail_qgroup;
3107
3108         if (!(sb->s_flags & MS_RDONLY)) {
3109                 ret = btrfs_cleanup_fs_roots(fs_info);
3110                 if (ret)
3111                         goto fail_qgroup;
3112
3113                 mutex_lock(&fs_info->cleaner_mutex);
3114                 ret = btrfs_recover_relocation(tree_root);
3115                 mutex_unlock(&fs_info->cleaner_mutex);
3116                 if (ret < 0) {
3117                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3118                                         ret);
3119                         err = -EINVAL;
3120                         goto fail_qgroup;
3121                 }
3122         }
3123
3124         location.objectid = BTRFS_FS_TREE_OBJECTID;
3125         location.type = BTRFS_ROOT_ITEM_KEY;
3126         location.offset = 0;
3127
3128         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3129         if (IS_ERR(fs_info->fs_root)) {
3130                 err = PTR_ERR(fs_info->fs_root);
3131                 goto fail_qgroup;
3132         }
3133
3134         if (sb->s_flags & MS_RDONLY)
3135                 return 0;
3136
3137         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3138             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3139                 clear_free_space_tree = 1;
3140         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3141                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3142                 btrfs_warn(fs_info, "free space tree is invalid");
3143                 clear_free_space_tree = 1;
3144         }
3145
3146         if (clear_free_space_tree) {
3147                 btrfs_info(fs_info, "clearing free space tree");
3148                 ret = btrfs_clear_free_space_tree(fs_info);
3149                 if (ret) {
3150                         btrfs_warn(fs_info,
3151                                    "failed to clear free space tree: %d", ret);
3152                         close_ctree(fs_info);
3153                         return ret;
3154                 }
3155         }
3156
3157         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3158             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3159                 btrfs_info(fs_info, "creating free space tree");
3160                 ret = btrfs_create_free_space_tree(fs_info);
3161                 if (ret) {
3162                         btrfs_warn(fs_info,
3163                                 "failed to create free space tree: %d", ret);
3164                         close_ctree(fs_info);
3165                         return ret;
3166                 }
3167         }
3168
3169         down_read(&fs_info->cleanup_work_sem);
3170         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3171             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3172                 up_read(&fs_info->cleanup_work_sem);
3173                 close_ctree(fs_info);
3174                 return ret;
3175         }
3176         up_read(&fs_info->cleanup_work_sem);
3177
3178         ret = btrfs_resume_balance_async(fs_info);
3179         if (ret) {
3180                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3181                 close_ctree(fs_info);
3182                 return ret;
3183         }
3184
3185         ret = btrfs_resume_dev_replace_async(fs_info);
3186         if (ret) {
3187                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3188                 close_ctree(fs_info);
3189                 return ret;
3190         }
3191
3192         btrfs_qgroup_rescan_resume(fs_info);
3193
3194         if (!fs_info->uuid_root) {
3195                 btrfs_info(fs_info, "creating UUID tree");
3196                 ret = btrfs_create_uuid_tree(fs_info);
3197                 if (ret) {
3198                         btrfs_warn(fs_info,
3199                                 "failed to create the UUID tree: %d", ret);
3200                         close_ctree(fs_info);
3201                         return ret;
3202                 }
3203         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3204                    fs_info->generation !=
3205                                 btrfs_super_uuid_tree_generation(disk_super)) {
3206                 btrfs_info(fs_info, "checking UUID tree");
3207                 ret = btrfs_check_uuid_tree(fs_info);
3208                 if (ret) {
3209                         btrfs_warn(fs_info,
3210                                 "failed to check the UUID tree: %d", ret);
3211                         close_ctree(fs_info);
3212                         return ret;
3213                 }
3214         } else {
3215                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3216         }
3217         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3218
3219         /*
3220          * backuproot only affect mount behavior, and if open_ctree succeeded,
3221          * no need to keep the flag
3222          */
3223         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3224
3225         return 0;
3226
3227 fail_qgroup:
3228         btrfs_free_qgroup_config(fs_info);
3229 fail_trans_kthread:
3230         kthread_stop(fs_info->transaction_kthread);
3231         btrfs_cleanup_transaction(fs_info);
3232         btrfs_free_fs_roots(fs_info);
3233 fail_cleaner:
3234         kthread_stop(fs_info->cleaner_kthread);
3235
3236         /*
3237          * make sure we're done with the btree inode before we stop our
3238          * kthreads
3239          */
3240         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3241
3242 fail_sysfs:
3243         btrfs_sysfs_remove_mounted(fs_info);
3244
3245 fail_fsdev_sysfs:
3246         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3247
3248 fail_block_groups:
3249         btrfs_put_block_group_cache(fs_info);
3250
3251 fail_tree_roots:
3252         free_root_pointers(fs_info, 1);
3253         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3254
3255 fail_sb_buffer:
3256         btrfs_stop_all_workers(fs_info);
3257         btrfs_free_block_groups(fs_info);
3258 fail_alloc:
3259 fail_iput:
3260         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3261
3262         iput(fs_info->btree_inode);
3263 fail_bio_counter:
3264         percpu_counter_destroy(&fs_info->bio_counter);
3265 fail_delalloc_bytes:
3266         percpu_counter_destroy(&fs_info->delalloc_bytes);
3267 fail_dirty_metadata_bytes:
3268         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3269 fail_srcu:
3270         cleanup_srcu_struct(&fs_info->subvol_srcu);
3271 fail:
3272         btrfs_free_stripe_hash_table(fs_info);
3273         btrfs_close_devices(fs_info->fs_devices);
3274         return err;
3275
3276 recovery_tree_root:
3277         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3278                 goto fail_tree_roots;
3279
3280         free_root_pointers(fs_info, 0);
3281
3282         /* don't use the log in recovery mode, it won't be valid */
3283         btrfs_set_super_log_root(disk_super, 0);
3284
3285         /* we can't trust the free space cache either */
3286         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3287
3288         ret = next_root_backup(fs_info, fs_info->super_copy,
3289                                &num_backups_tried, &backup_index);
3290         if (ret == -1)
3291                 goto fail_block_groups;
3292         goto retry_root_backup;
3293 }
3294
3295 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3296 {
3297         if (uptodate) {
3298                 set_buffer_uptodate(bh);
3299         } else {
3300                 struct btrfs_device *device = (struct btrfs_device *)
3301                         bh->b_private;
3302
3303                 btrfs_warn_rl_in_rcu(device->fs_info,
3304                                 "lost page write due to IO error on %s",
3305                                           rcu_str_deref(device->name));
3306                 /* note, we don't set_buffer_write_io_error because we have
3307                  * our own ways of dealing with the IO errors
3308                  */
3309                 clear_buffer_uptodate(bh);
3310                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3311         }
3312         unlock_buffer(bh);
3313         put_bh(bh);
3314 }
3315
3316 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3317                         struct buffer_head **bh_ret)
3318 {
3319         struct buffer_head *bh;
3320         struct btrfs_super_block *super;
3321         u64 bytenr;
3322
3323         bytenr = btrfs_sb_offset(copy_num);
3324         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3325                 return -EINVAL;
3326
3327         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3328         /*
3329          * If we fail to read from the underlying devices, as of now
3330          * the best option we have is to mark it EIO.
3331          */
3332         if (!bh)
3333                 return -EIO;
3334
3335         super = (struct btrfs_super_block *)bh->b_data;
3336         if (btrfs_super_bytenr(super) != bytenr ||
3337                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3338                 brelse(bh);
3339                 return -EINVAL;
3340         }
3341
3342         *bh_ret = bh;
3343         return 0;
3344 }
3345
3346
3347 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3348 {
3349         struct buffer_head *bh;
3350         struct buffer_head *latest = NULL;
3351         struct btrfs_super_block *super;
3352         int i;
3353         u64 transid = 0;
3354         int ret = -EINVAL;
3355
3356         /* we would like to check all the supers, but that would make
3357          * a btrfs mount succeed after a mkfs from a different FS.
3358          * So, we need to add a special mount option to scan for
3359          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3360          */
3361         for (i = 0; i < 1; i++) {
3362                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3363                 if (ret)
3364                         continue;
3365
3366                 super = (struct btrfs_super_block *)bh->b_data;
3367
3368                 if (!latest || btrfs_super_generation(super) > transid) {
3369                         brelse(latest);
3370                         latest = bh;
3371                         transid = btrfs_super_generation(super);
3372                 } else {
3373                         brelse(bh);
3374                 }
3375         }
3376
3377         if (!latest)
3378                 return ERR_PTR(ret);
3379
3380         return latest;
3381 }
3382
3383 /*
3384  * this should be called twice, once with wait == 0 and
3385  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3386  * we write are pinned.
3387  *
3388  * They are released when wait == 1 is done.
3389  * max_mirrors must be the same for both runs, and it indicates how
3390  * many supers on this one device should be written.
3391  *
3392  * max_mirrors == 0 means to write them all.
3393  */
3394 static int write_dev_supers(struct btrfs_device *device,
3395                             struct btrfs_super_block *sb,
3396                             int wait, int max_mirrors)
3397 {
3398         struct buffer_head *bh;
3399         int i;
3400         int ret;
3401         int errors = 0;
3402         u32 crc;
3403         u64 bytenr;
3404
3405         if (max_mirrors == 0)
3406                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3407
3408         for (i = 0; i < max_mirrors; i++) {
3409                 bytenr = btrfs_sb_offset(i);
3410                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3411                     device->commit_total_bytes)
3412                         break;
3413
3414                 if (wait) {
3415                         bh = __find_get_block(device->bdev, bytenr / 4096,
3416                                               BTRFS_SUPER_INFO_SIZE);
3417                         if (!bh) {
3418                                 errors++;
3419                                 continue;
3420                         }
3421                         wait_on_buffer(bh);
3422                         if (!buffer_uptodate(bh))
3423                                 errors++;
3424
3425                         /* drop our reference */
3426                         brelse(bh);
3427
3428                         /* drop the reference from the wait == 0 run */
3429                         brelse(bh);
3430                         continue;
3431                 } else {
3432                         btrfs_set_super_bytenr(sb, bytenr);
3433
3434                         crc = ~(u32)0;
3435                         crc = btrfs_csum_data((const char *)sb +
3436                                               BTRFS_CSUM_SIZE, crc,
3437                                               BTRFS_SUPER_INFO_SIZE -
3438                                               BTRFS_CSUM_SIZE);
3439                         btrfs_csum_final(crc, sb->csum);
3440
3441                         /*
3442                          * one reference for us, and we leave it for the
3443                          * caller
3444                          */
3445                         bh = __getblk(device->bdev, bytenr / 4096,
3446                                       BTRFS_SUPER_INFO_SIZE);
3447                         if (!bh) {
3448                                 btrfs_err(device->fs_info,
3449                                     "couldn't get super buffer head for bytenr %llu",
3450                                     bytenr);
3451                                 errors++;
3452                                 continue;
3453                         }
3454
3455                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3456
3457                         /* one reference for submit_bh */
3458                         get_bh(bh);
3459
3460                         set_buffer_uptodate(bh);
3461                         lock_buffer(bh);
3462                         bh->b_end_io = btrfs_end_buffer_write_sync;
3463                         bh->b_private = device;
3464                 }
3465
3466                 /*
3467                  * we fua the first super.  The others we allow
3468                  * to go down lazy.
3469                  */
3470                 if (i == 0) {
3471                         ret = btrfsic_submit_bh(REQ_OP_WRITE,
3472                                                 REQ_SYNC | REQ_FUA, bh);
3473                 } else {
3474                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3475                 }
3476                 if (ret)
3477                         errors++;
3478         }
3479         return errors < i ? 0 : -1;
3480 }
3481
3482 /*
3483  * endio for the write_dev_flush, this will wake anyone waiting
3484  * for the barrier when it is done
3485  */
3486 static void btrfs_end_empty_barrier(struct bio *bio)
3487 {
3488         if (bio->bi_private)
3489                 complete(bio->bi_private);
3490         bio_put(bio);
3491 }
3492
3493 /*
3494  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3495  * sent down.  With wait == 1, it waits for the previous flush.
3496  *
3497  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3498  * capable
3499  */
3500 static int write_dev_flush(struct btrfs_device *device, int wait)
3501 {
3502         struct request_queue *q = bdev_get_queue(device->bdev);
3503         struct bio *bio;
3504         int ret = 0;
3505
3506         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3507                 return 0;
3508
3509         if (wait) {
3510                 bio = device->flush_bio;
3511                 if (!bio)
3512                         return 0;
3513
3514                 wait_for_completion(&device->flush_wait);
3515
3516                 if (bio->bi_error) {
3517                         ret = bio->bi_error;
3518                         btrfs_dev_stat_inc_and_print(device,
3519                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3520                 }
3521
3522                 /* drop the reference from the wait == 0 run */
3523                 bio_put(bio);
3524                 device->flush_bio = NULL;
3525
3526                 return ret;
3527         }
3528
3529         /*
3530          * one reference for us, and we leave it for the
3531          * caller
3532          */
3533         device->flush_bio = NULL;
3534         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3535         if (!bio)
3536                 return -ENOMEM;
3537
3538         bio->bi_end_io = btrfs_end_empty_barrier;
3539         bio->bi_bdev = device->bdev;
3540         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3541         init_completion(&device->flush_wait);
3542         bio->bi_private = &device->flush_wait;
3543         device->flush_bio = bio;
3544
3545         bio_get(bio);
3546         btrfsic_submit_bio(bio);
3547
3548         return 0;
3549 }
3550
3551 /*
3552  * send an empty flush down to each device in parallel,
3553  * then wait for them
3554  */
3555 static int barrier_all_devices(struct btrfs_fs_info *info)
3556 {
3557         struct list_head *head;
3558         struct btrfs_device *dev;
3559         int errors_send = 0;
3560         int errors_wait = 0;
3561         int ret;
3562
3563         /* send down all the barriers */
3564         head = &info->fs_devices->devices;
3565         list_for_each_entry_rcu(dev, head, dev_list) {
3566                 if (dev->missing)
3567                         continue;
3568                 if (!dev->bdev) {
3569                         errors_send++;
3570                         continue;
3571                 }
3572                 if (!dev->in_fs_metadata || !dev->writeable)
3573                         continue;
3574
3575                 ret = write_dev_flush(dev, 0);
3576                 if (ret)
3577                         errors_send++;
3578         }
3579
3580         /* wait for all the barriers */
3581         list_for_each_entry_rcu(dev, head, dev_list) {
3582                 if (dev->missing)
3583                         continue;
3584                 if (!dev->bdev) {
3585                         errors_wait++;
3586                         continue;
3587                 }
3588                 if (!dev->in_fs_metadata || !dev->writeable)
3589                         continue;
3590
3591                 ret = write_dev_flush(dev, 1);
3592                 if (ret)
3593                         errors_wait++;
3594         }
3595         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3596             errors_wait > info->num_tolerated_disk_barrier_failures)
3597                 return -EIO;
3598         return 0;
3599 }
3600
3601 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3602 {
3603         int raid_type;
3604         int min_tolerated = INT_MAX;
3605
3606         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3607             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3608                 min_tolerated = min(min_tolerated,
3609                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3610                                     tolerated_failures);
3611
3612         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3613                 if (raid_type == BTRFS_RAID_SINGLE)
3614                         continue;
3615                 if (!(flags & btrfs_raid_group[raid_type]))
3616                         continue;
3617                 min_tolerated = min(min_tolerated,
3618                                     btrfs_raid_array[raid_type].
3619                                     tolerated_failures);
3620         }
3621
3622         if (min_tolerated == INT_MAX) {
3623                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3624                 min_tolerated = 0;
3625         }
3626
3627         return min_tolerated;
3628 }
3629
3630 int btrfs_calc_num_tolerated_disk_barrier_failures(
3631         struct btrfs_fs_info *fs_info)
3632 {
3633         struct btrfs_ioctl_space_info space;
3634         struct btrfs_space_info *sinfo;
3635         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3636                        BTRFS_BLOCK_GROUP_SYSTEM,
3637                        BTRFS_BLOCK_GROUP_METADATA,
3638                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3639         int i;
3640         int c;
3641         int num_tolerated_disk_barrier_failures =
3642                 (int)fs_info->fs_devices->num_devices;
3643
3644         for (i = 0; i < ARRAY_SIZE(types); i++) {
3645                 struct btrfs_space_info *tmp;
3646
3647                 sinfo = NULL;
3648                 rcu_read_lock();
3649                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3650                         if (tmp->flags == types[i]) {
3651                                 sinfo = tmp;
3652                                 break;
3653                         }
3654                 }
3655                 rcu_read_unlock();
3656
3657                 if (!sinfo)
3658                         continue;
3659
3660                 down_read(&sinfo->groups_sem);
3661                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3662                         u64 flags;
3663
3664                         if (list_empty(&sinfo->block_groups[c]))
3665                                 continue;
3666
3667                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3668                                                    &space);
3669                         if (space.total_bytes == 0 || space.used_bytes == 0)
3670                                 continue;
3671                         flags = space.flags;
3672
3673                         num_tolerated_disk_barrier_failures = min(
3674                                 num_tolerated_disk_barrier_failures,
3675                                 btrfs_get_num_tolerated_disk_barrier_failures(
3676                                         flags));
3677                 }
3678                 up_read(&sinfo->groups_sem);
3679         }
3680
3681         return num_tolerated_disk_barrier_failures;
3682 }
3683
3684 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3685 {
3686         struct list_head *head;
3687         struct btrfs_device *dev;
3688         struct btrfs_super_block *sb;
3689         struct btrfs_dev_item *dev_item;
3690         int ret;
3691         int do_barriers;
3692         int max_errors;
3693         int total_errors = 0;
3694         u64 flags;
3695
3696         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3697         backup_super_roots(fs_info);
3698
3699         sb = fs_info->super_for_commit;
3700         dev_item = &sb->dev_item;
3701
3702         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3703         head = &fs_info->fs_devices->devices;
3704         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3705
3706         if (do_barriers) {
3707                 ret = barrier_all_devices(fs_info);
3708                 if (ret) {
3709                         mutex_unlock(
3710                                 &fs_info->fs_devices->device_list_mutex);
3711                         btrfs_handle_fs_error(fs_info, ret,
3712                                               "errors while submitting device barriers.");
3713                         return ret;
3714                 }
3715         }
3716
3717         list_for_each_entry_rcu(dev, head, dev_list) {
3718                 if (!dev->bdev) {
3719                         total_errors++;
3720                         continue;
3721                 }
3722                 if (!dev->in_fs_metadata || !dev->writeable)
3723                         continue;
3724
3725                 btrfs_set_stack_device_generation(dev_item, 0);
3726                 btrfs_set_stack_device_type(dev_item, dev->type);
3727                 btrfs_set_stack_device_id(dev_item, dev->devid);
3728                 btrfs_set_stack_device_total_bytes(dev_item,
3729                                                    dev->commit_total_bytes);
3730                 btrfs_set_stack_device_bytes_used(dev_item,
3731                                                   dev->commit_bytes_used);
3732                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3733                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3734                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3735                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3736                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3737
3738                 flags = btrfs_super_flags(sb);
3739                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3740
3741                 ret = write_dev_supers(dev, sb, 0, max_mirrors);
3742                 if (ret)
3743                         total_errors++;
3744         }
3745         if (total_errors > max_errors) {
3746                 btrfs_err(fs_info, "%d errors while writing supers",
3747                           total_errors);
3748                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3749
3750                 /* FUA is masked off if unsupported and can't be the reason */
3751                 btrfs_handle_fs_error(fs_info, -EIO,
3752                                       "%d errors while writing supers",
3753                                       total_errors);
3754                 return -EIO;
3755         }
3756
3757         total_errors = 0;
3758         list_for_each_entry_rcu(dev, head, dev_list) {
3759                 if (!dev->bdev)
3760                         continue;
3761                 if (!dev->in_fs_metadata || !dev->writeable)
3762                         continue;
3763
3764                 ret = write_dev_supers(dev, sb, 1, max_mirrors);
3765                 if (ret)
3766                         total_errors++;
3767         }
3768         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3769         if (total_errors > max_errors) {
3770                 btrfs_handle_fs_error(fs_info, -EIO,
3771                                       "%d errors while writing supers",
3772                                       total_errors);
3773                 return -EIO;
3774         }
3775         return 0;
3776 }
3777
3778 /* Drop a fs root from the radix tree and free it. */
3779 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3780                                   struct btrfs_root *root)
3781 {
3782         spin_lock(&fs_info->fs_roots_radix_lock);
3783         radix_tree_delete(&fs_info->fs_roots_radix,
3784                           (unsigned long)root->root_key.objectid);
3785         spin_unlock(&fs_info->fs_roots_radix_lock);
3786
3787         if (btrfs_root_refs(&root->root_item) == 0)
3788                 synchronize_srcu(&fs_info->subvol_srcu);
3789
3790         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3791                 btrfs_free_log(NULL, root);
3792                 if (root->reloc_root) {
3793                         free_extent_buffer(root->reloc_root->node);
3794                         free_extent_buffer(root->reloc_root->commit_root);
3795                         btrfs_put_fs_root(root->reloc_root);
3796                         root->reloc_root = NULL;
3797                 }
3798         }
3799
3800         if (root->free_ino_pinned)
3801                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3802         if (root->free_ino_ctl)
3803                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3804         free_fs_root(root);
3805 }
3806
3807 static void free_fs_root(struct btrfs_root *root)
3808 {
3809         iput(root->ino_cache_inode);
3810         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3811         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3812         root->orphan_block_rsv = NULL;
3813         if (root->anon_dev)
3814                 free_anon_bdev(root->anon_dev);
3815         if (root->subv_writers)
3816                 btrfs_free_subvolume_writers(root->subv_writers);
3817         free_extent_buffer(root->node);
3818         free_extent_buffer(root->commit_root);
3819         kfree(root->free_ino_ctl);
3820         kfree(root->free_ino_pinned);
3821         kfree(root->name);
3822         btrfs_put_fs_root(root);
3823 }
3824
3825 void btrfs_free_fs_root(struct btrfs_root *root)
3826 {
3827         free_fs_root(root);
3828 }
3829
3830 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3831 {
3832         u64 root_objectid = 0;
3833         struct btrfs_root *gang[8];
3834         int i = 0;
3835         int err = 0;
3836         unsigned int ret = 0;
3837         int index;
3838
3839         while (1) {
3840                 index = srcu_read_lock(&fs_info->subvol_srcu);
3841                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3842                                              (void **)gang, root_objectid,
3843                                              ARRAY_SIZE(gang));
3844                 if (!ret) {
3845                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3846                         break;
3847                 }
3848                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3849
3850                 for (i = 0; i < ret; i++) {
3851                         /* Avoid to grab roots in dead_roots */
3852                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3853                                 gang[i] = NULL;
3854                                 continue;
3855                         }
3856                         /* grab all the search result for later use */
3857                         gang[i] = btrfs_grab_fs_root(gang[i]);
3858                 }
3859                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3860
3861                 for (i = 0; i < ret; i++) {
3862                         if (!gang[i])
3863                                 continue;
3864                         root_objectid = gang[i]->root_key.objectid;
3865                         err = btrfs_orphan_cleanup(gang[i]);
3866                         if (err)
3867                                 break;
3868                         btrfs_put_fs_root(gang[i]);
3869                 }
3870                 root_objectid++;
3871         }
3872
3873         /* release the uncleaned roots due to error */
3874         for (; i < ret; i++) {
3875                 if (gang[i])
3876                         btrfs_put_fs_root(gang[i]);
3877         }
3878         return err;
3879 }
3880
3881 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3882 {
3883         struct btrfs_root *root = fs_info->tree_root;
3884         struct btrfs_trans_handle *trans;
3885
3886         mutex_lock(&fs_info->cleaner_mutex);
3887         btrfs_run_delayed_iputs(fs_info);
3888         mutex_unlock(&fs_info->cleaner_mutex);
3889         wake_up_process(fs_info->cleaner_kthread);
3890
3891         /* wait until ongoing cleanup work done */
3892         down_write(&fs_info->cleanup_work_sem);
3893         up_write(&fs_info->cleanup_work_sem);
3894
3895         trans = btrfs_join_transaction(root);
3896         if (IS_ERR(trans))
3897                 return PTR_ERR(trans);
3898         return btrfs_commit_transaction(trans);
3899 }
3900
3901 void close_ctree(struct btrfs_fs_info *fs_info)
3902 {
3903         struct btrfs_root *root = fs_info->tree_root;
3904         int ret;
3905
3906         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3907
3908         /* wait for the qgroup rescan worker to stop */
3909         btrfs_qgroup_wait_for_completion(fs_info, false);
3910
3911         /* wait for the uuid_scan task to finish */
3912         down(&fs_info->uuid_tree_rescan_sem);
3913         /* avoid complains from lockdep et al., set sem back to initial state */
3914         up(&fs_info->uuid_tree_rescan_sem);
3915
3916         /* pause restriper - we want to resume on mount */
3917         btrfs_pause_balance(fs_info);
3918
3919         btrfs_dev_replace_suspend_for_unmount(fs_info);
3920
3921         btrfs_scrub_cancel(fs_info);
3922
3923         /* wait for any defraggers to finish */
3924         wait_event(fs_info->transaction_wait,
3925                    (atomic_read(&fs_info->defrag_running) == 0));
3926
3927         /* clear out the rbtree of defraggable inodes */
3928         btrfs_cleanup_defrag_inodes(fs_info);
3929
3930         cancel_work_sync(&fs_info->async_reclaim_work);
3931
3932         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3933                 /*
3934                  * If the cleaner thread is stopped and there are
3935                  * block groups queued for removal, the deletion will be
3936                  * skipped when we quit the cleaner thread.
3937                  */
3938                 btrfs_delete_unused_bgs(fs_info);
3939
3940                 ret = btrfs_commit_super(fs_info);
3941                 if (ret)
3942                         btrfs_err(fs_info, "commit super ret %d", ret);
3943         }
3944
3945         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3946                 btrfs_error_commit_super(fs_info);
3947
3948         kthread_stop(fs_info->transaction_kthread);
3949         kthread_stop(fs_info->cleaner_kthread);
3950
3951         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3952
3953         btrfs_free_qgroup_config(fs_info);
3954
3955         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3956                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3957                        percpu_counter_sum(&fs_info->delalloc_bytes));
3958         }
3959
3960         btrfs_sysfs_remove_mounted(fs_info);
3961         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3962
3963         btrfs_free_fs_roots(fs_info);
3964
3965         btrfs_put_block_group_cache(fs_info);
3966
3967         /*
3968          * we must make sure there is not any read request to
3969          * submit after we stopping all workers.
3970          */
3971         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3972         btrfs_stop_all_workers(fs_info);
3973
3974         btrfs_free_block_groups(fs_info);
3975
3976         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3977         free_root_pointers(fs_info, 1);
3978
3979         iput(fs_info->btree_inode);
3980
3981 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3982         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3983                 btrfsic_unmount(fs_info->fs_devices);
3984 #endif
3985
3986         btrfs_close_devices(fs_info->fs_devices);
3987         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3988
3989         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3990         percpu_counter_destroy(&fs_info->delalloc_bytes);
3991         percpu_counter_destroy(&fs_info->bio_counter);
3992         cleanup_srcu_struct(&fs_info->subvol_srcu);
3993
3994         btrfs_free_stripe_hash_table(fs_info);
3995
3996         __btrfs_free_block_rsv(root->orphan_block_rsv);
3997         root->orphan_block_rsv = NULL;
3998
3999         mutex_lock(&fs_info->chunk_mutex);
4000         while (!list_empty(&fs_info->pinned_chunks)) {
4001                 struct extent_map *em;
4002
4003                 em = list_first_entry(&fs_info->pinned_chunks,
4004                                       struct extent_map, list);
4005                 list_del_init(&em->list);
4006                 free_extent_map(em);
4007         }
4008         mutex_unlock(&fs_info->chunk_mutex);
4009 }
4010
4011 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4012                           int atomic)
4013 {
4014         int ret;
4015         struct inode *btree_inode = buf->pages[0]->mapping->host;
4016
4017         ret = extent_buffer_uptodate(buf);
4018         if (!ret)
4019                 return ret;
4020
4021         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4022                                     parent_transid, atomic);
4023         if (ret == -EAGAIN)
4024                 return ret;
4025         return !ret;
4026 }
4027
4028 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4029 {
4030         struct btrfs_fs_info *fs_info;
4031         struct btrfs_root *root;
4032         u64 transid = btrfs_header_generation(buf);
4033         int was_dirty;
4034
4035 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4036         /*
4037          * This is a fast path so only do this check if we have sanity tests
4038          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4039          * outside of the sanity tests.
4040          */
4041         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4042                 return;
4043 #endif
4044         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4045         fs_info = root->fs_info;
4046         btrfs_assert_tree_locked(buf);
4047         if (transid != fs_info->generation)
4048                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4049                         buf->start, transid, fs_info->generation);
4050         was_dirty = set_extent_buffer_dirty(buf);
4051         if (!was_dirty)
4052                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
4053                                      buf->len,
4054                                      fs_info->dirty_metadata_batch);
4055 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4056         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4057                 btrfs_print_leaf(fs_info, buf);
4058                 ASSERT(0);
4059         }
4060 #endif
4061 }
4062
4063 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4064                                         int flush_delayed)
4065 {
4066         /*
4067          * looks as though older kernels can get into trouble with
4068          * this code, they end up stuck in balance_dirty_pages forever
4069          */
4070         int ret;
4071
4072         if (current->flags & PF_MEMALLOC)
4073                 return;
4074
4075         if (flush_delayed)
4076                 btrfs_balance_delayed_items(fs_info);
4077
4078         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4079                                      BTRFS_DIRTY_METADATA_THRESH);
4080         if (ret > 0) {
4081                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4082         }
4083 }
4084
4085 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4086 {
4087         __btrfs_btree_balance_dirty(fs_info, 1);
4088 }
4089
4090 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4091 {
4092         __btrfs_btree_balance_dirty(fs_info, 0);
4093 }
4094
4095 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4096 {
4097         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4098         struct btrfs_fs_info *fs_info = root->fs_info;
4099
4100         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4101 }
4102
4103 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
4104 {
4105         struct btrfs_super_block *sb = fs_info->super_copy;
4106         u64 nodesize = btrfs_super_nodesize(sb);
4107         u64 sectorsize = btrfs_super_sectorsize(sb);
4108         int ret = 0;
4109
4110         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4111                 btrfs_err(fs_info, "no valid FS found");
4112                 ret = -EINVAL;
4113         }
4114         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4115                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4116                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4117         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4118                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4119                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4120                 ret = -EINVAL;
4121         }
4122         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4123                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4124                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4125                 ret = -EINVAL;
4126         }
4127         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4128                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4129                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4130                 ret = -EINVAL;
4131         }
4132
4133         /*
4134          * Check sectorsize and nodesize first, other check will need it.
4135          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4136          */
4137         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4138             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4139                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4140                 ret = -EINVAL;
4141         }
4142         /* Only PAGE SIZE is supported yet */
4143         if (sectorsize != PAGE_SIZE) {
4144                 btrfs_err(fs_info,
4145                         "sectorsize %llu not supported yet, only support %lu",
4146                         sectorsize, PAGE_SIZE);
4147                 ret = -EINVAL;
4148         }
4149         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4150             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4151                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4152                 ret = -EINVAL;
4153         }
4154         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4155                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4156                           le32_to_cpu(sb->__unused_leafsize), nodesize);
4157                 ret = -EINVAL;
4158         }
4159
4160         /* Root alignment check */
4161         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4162                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4163                            btrfs_super_root(sb));
4164                 ret = -EINVAL;
4165         }
4166         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4167                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4168                            btrfs_super_chunk_root(sb));
4169                 ret = -EINVAL;
4170         }
4171         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4172                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4173                            btrfs_super_log_root(sb));
4174                 ret = -EINVAL;
4175         }
4176
4177         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4178                 btrfs_err(fs_info,
4179                            "dev_item UUID does not match fsid: %pU != %pU",
4180                            fs_info->fsid, sb->dev_item.fsid);
4181                 ret = -EINVAL;
4182         }
4183
4184         /*
4185          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4186          * done later
4187          */
4188         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4189                 btrfs_err(fs_info, "bytes_used is too small %llu",
4190                           btrfs_super_bytes_used(sb));
4191                 ret = -EINVAL;
4192         }
4193         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4194                 btrfs_err(fs_info, "invalid stripesize %u",
4195                           btrfs_super_stripesize(sb));
4196                 ret = -EINVAL;
4197         }
4198         if (btrfs_super_num_devices(sb) > (1UL << 31))
4199                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4200                            btrfs_super_num_devices(sb));
4201         if (btrfs_super_num_devices(sb) == 0) {
4202                 btrfs_err(fs_info, "number of devices is 0");
4203                 ret = -EINVAL;
4204         }
4205
4206         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4207                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4208                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4209                 ret = -EINVAL;
4210         }
4211
4212         /*
4213          * Obvious sys_chunk_array corruptions, it must hold at least one key
4214          * and one chunk
4215          */
4216         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4217                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4218                           btrfs_super_sys_array_size(sb),
4219                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4220                 ret = -EINVAL;
4221         }
4222         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4223                         + sizeof(struct btrfs_chunk)) {
4224                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4225                           btrfs_super_sys_array_size(sb),
4226                           sizeof(struct btrfs_disk_key)
4227                           + sizeof(struct btrfs_chunk));
4228                 ret = -EINVAL;
4229         }
4230
4231         /*
4232          * The generation is a global counter, we'll trust it more than the others
4233          * but it's still possible that it's the one that's wrong.
4234          */
4235         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4236                 btrfs_warn(fs_info,
4237                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4238                         btrfs_super_generation(sb),
4239                         btrfs_super_chunk_root_generation(sb));
4240         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4241             && btrfs_super_cache_generation(sb) != (u64)-1)
4242                 btrfs_warn(fs_info,
4243                         "suspicious: generation < cache_generation: %llu < %llu",
4244                         btrfs_super_generation(sb),
4245                         btrfs_super_cache_generation(sb));
4246
4247         return ret;
4248 }
4249
4250 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4251 {
4252         mutex_lock(&fs_info->cleaner_mutex);
4253         btrfs_run_delayed_iputs(fs_info);
4254         mutex_unlock(&fs_info->cleaner_mutex);
4255
4256         down_write(&fs_info->cleanup_work_sem);
4257         up_write(&fs_info->cleanup_work_sem);
4258
4259         /* cleanup FS via transaction */
4260         btrfs_cleanup_transaction(fs_info);
4261 }
4262
4263 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4264 {
4265         struct btrfs_ordered_extent *ordered;
4266
4267         spin_lock(&root->ordered_extent_lock);
4268         /*
4269          * This will just short circuit the ordered completion stuff which will
4270          * make sure the ordered extent gets properly cleaned up.
4271          */
4272         list_for_each_entry(ordered, &root->ordered_extents,
4273                             root_extent_list)
4274                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4275         spin_unlock(&root->ordered_extent_lock);
4276 }
4277
4278 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4279 {
4280         struct btrfs_root *root;
4281         struct list_head splice;
4282
4283         INIT_LIST_HEAD(&splice);
4284
4285         spin_lock(&fs_info->ordered_root_lock);
4286         list_splice_init(&fs_info->ordered_roots, &splice);
4287         while (!list_empty(&splice)) {
4288                 root = list_first_entry(&splice, struct btrfs_root,
4289                                         ordered_root);
4290                 list_move_tail(&root->ordered_root,
4291                                &fs_info->ordered_roots);
4292
4293                 spin_unlock(&fs_info->ordered_root_lock);
4294                 btrfs_destroy_ordered_extents(root);
4295
4296                 cond_resched();
4297                 spin_lock(&fs_info->ordered_root_lock);
4298         }
4299         spin_unlock(&fs_info->ordered_root_lock);
4300 }
4301
4302 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4303                                       struct btrfs_fs_info *fs_info)
4304 {
4305         struct rb_node *node;
4306         struct btrfs_delayed_ref_root *delayed_refs;
4307         struct btrfs_delayed_ref_node *ref;
4308         int ret = 0;
4309
4310         delayed_refs = &trans->delayed_refs;
4311
4312         spin_lock(&delayed_refs->lock);
4313         if (atomic_read(&delayed_refs->num_entries) == 0) {
4314                 spin_unlock(&delayed_refs->lock);
4315                 btrfs_info(fs_info, "delayed_refs has NO entry");
4316                 return ret;
4317         }
4318
4319         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4320                 struct btrfs_delayed_ref_head *head;
4321                 struct btrfs_delayed_ref_node *tmp;
4322                 bool pin_bytes = false;
4323
4324                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4325                                 href_node);
4326                 if (!mutex_trylock(&head->mutex)) {
4327                         refcount_inc(&head->node.refs);
4328                         spin_unlock(&delayed_refs->lock);
4329
4330                         mutex_lock(&head->mutex);
4331                         mutex_unlock(&head->mutex);
4332                         btrfs_put_delayed_ref(&head->node);
4333                         spin_lock(&delayed_refs->lock);
4334                         continue;
4335                 }
4336                 spin_lock(&head->lock);
4337                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4338                                                  list) {
4339                         ref->in_tree = 0;
4340                         list_del(&ref->list);
4341                         if (!list_empty(&ref->add_list))
4342                                 list_del(&ref->add_list);
4343                         atomic_dec(&delayed_refs->num_entries);
4344                         btrfs_put_delayed_ref(ref);
4345                 }
4346                 if (head->must_insert_reserved)
4347                         pin_bytes = true;
4348                 btrfs_free_delayed_extent_op(head->extent_op);
4349                 delayed_refs->num_heads--;
4350                 if (head->processing == 0)
4351                         delayed_refs->num_heads_ready--;
4352                 atomic_dec(&delayed_refs->num_entries);
4353                 head->node.in_tree = 0;
4354                 rb_erase(&head->href_node, &delayed_refs->href_root);
4355                 spin_unlock(&head->lock);
4356                 spin_unlock(&delayed_refs->lock);
4357                 mutex_unlock(&head->mutex);
4358
4359                 if (pin_bytes)
4360                         btrfs_pin_extent(fs_info, head->node.bytenr,
4361                                          head->node.num_bytes, 1);
4362                 btrfs_put_delayed_ref(&head->node);
4363                 cond_resched();
4364                 spin_lock(&delayed_refs->lock);
4365         }
4366
4367         spin_unlock(&delayed_refs->lock);
4368
4369         return ret;
4370 }
4371
4372 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4373 {
4374         struct btrfs_inode *btrfs_inode;
4375         struct list_head splice;
4376
4377         INIT_LIST_HEAD(&splice);
4378
4379         spin_lock(&root->delalloc_lock);
4380         list_splice_init(&root->delalloc_inodes, &splice);
4381
4382         while (!list_empty(&splice)) {
4383                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4384                                                delalloc_inodes);
4385
4386                 list_del_init(&btrfs_inode->delalloc_inodes);
4387                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4388                           &btrfs_inode->runtime_flags);
4389                 spin_unlock(&root->delalloc_lock);
4390
4391                 btrfs_invalidate_inodes(btrfs_inode->root);
4392
4393                 spin_lock(&root->delalloc_lock);
4394         }
4395
4396         spin_unlock(&root->delalloc_lock);
4397 }
4398
4399 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4400 {
4401         struct btrfs_root *root;
4402         struct list_head splice;
4403
4404         INIT_LIST_HEAD(&splice);
4405
4406         spin_lock(&fs_info->delalloc_root_lock);
4407         list_splice_init(&fs_info->delalloc_roots, &splice);
4408         while (!list_empty(&splice)) {
4409                 root = list_first_entry(&splice, struct btrfs_root,
4410                                          delalloc_root);
4411                 list_del_init(&root->delalloc_root);
4412                 root = btrfs_grab_fs_root(root);
4413                 BUG_ON(!root);
4414                 spin_unlock(&fs_info->delalloc_root_lock);
4415
4416                 btrfs_destroy_delalloc_inodes(root);
4417                 btrfs_put_fs_root(root);
4418
4419                 spin_lock(&fs_info->delalloc_root_lock);
4420         }
4421         spin_unlock(&fs_info->delalloc_root_lock);
4422 }
4423
4424 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4425                                         struct extent_io_tree *dirty_pages,
4426                                         int mark)
4427 {
4428         int ret;
4429         struct extent_buffer *eb;
4430         u64 start = 0;
4431         u64 end;
4432
4433         while (1) {
4434                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4435                                             mark, NULL);
4436                 if (ret)
4437                         break;
4438
4439                 clear_extent_bits(dirty_pages, start, end, mark);
4440                 while (start <= end) {
4441                         eb = find_extent_buffer(fs_info, start);
4442                         start += fs_info->nodesize;
4443                         if (!eb)
4444                                 continue;
4445                         wait_on_extent_buffer_writeback(eb);
4446
4447                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4448                                                &eb->bflags))
4449                                 clear_extent_buffer_dirty(eb);
4450                         free_extent_buffer_stale(eb);
4451                 }
4452         }
4453
4454         return ret;
4455 }
4456
4457 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4458                                        struct extent_io_tree *pinned_extents)
4459 {
4460         struct extent_io_tree *unpin;
4461         u64 start;
4462         u64 end;
4463         int ret;
4464         bool loop = true;
4465
4466         unpin = pinned_extents;
4467 again:
4468         while (1) {
4469                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4470                                             EXTENT_DIRTY, NULL);
4471                 if (ret)
4472                         break;
4473
4474                 clear_extent_dirty(unpin, start, end);
4475                 btrfs_error_unpin_extent_range(fs_info, start, end);
4476                 cond_resched();
4477         }
4478
4479         if (loop) {
4480                 if (unpin == &fs_info->freed_extents[0])
4481                         unpin = &fs_info->freed_extents[1];
4482                 else
4483                         unpin = &fs_info->freed_extents[0];
4484                 loop = false;
4485                 goto again;
4486         }
4487
4488         return 0;
4489 }
4490
4491 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4492 {
4493         struct inode *inode;
4494
4495         inode = cache->io_ctl.inode;
4496         if (inode) {
4497                 invalidate_inode_pages2(inode->i_mapping);
4498                 BTRFS_I(inode)->generation = 0;
4499                 cache->io_ctl.inode = NULL;
4500                 iput(inode);
4501         }
4502         btrfs_put_block_group(cache);
4503 }
4504
4505 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4506                              struct btrfs_fs_info *fs_info)
4507 {
4508         struct btrfs_block_group_cache *cache;
4509
4510         spin_lock(&cur_trans->dirty_bgs_lock);
4511         while (!list_empty(&cur_trans->dirty_bgs)) {
4512                 cache = list_first_entry(&cur_trans->dirty_bgs,
4513                                          struct btrfs_block_group_cache,
4514                                          dirty_list);
4515                 if (!cache) {
4516                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4517                         spin_unlock(&cur_trans->dirty_bgs_lock);
4518                         return;
4519                 }
4520
4521                 if (!list_empty(&cache->io_list)) {
4522                         spin_unlock(&cur_trans->dirty_bgs_lock);
4523                         list_del_init(&cache->io_list);
4524                         btrfs_cleanup_bg_io(cache);
4525                         spin_lock(&cur_trans->dirty_bgs_lock);
4526                 }
4527
4528                 list_del_init(&cache->dirty_list);
4529                 spin_lock(&cache->lock);
4530                 cache->disk_cache_state = BTRFS_DC_ERROR;
4531                 spin_unlock(&cache->lock);
4532
4533                 spin_unlock(&cur_trans->dirty_bgs_lock);
4534                 btrfs_put_block_group(cache);
4535                 spin_lock(&cur_trans->dirty_bgs_lock);
4536         }
4537         spin_unlock(&cur_trans->dirty_bgs_lock);
4538
4539         while (!list_empty(&cur_trans->io_bgs)) {
4540                 cache = list_first_entry(&cur_trans->io_bgs,
4541                                          struct btrfs_block_group_cache,
4542                                          io_list);
4543                 if (!cache) {
4544                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4545                         return;
4546                 }
4547
4548                 list_del_init(&cache->io_list);
4549                 spin_lock(&cache->lock);
4550                 cache->disk_cache_state = BTRFS_DC_ERROR;
4551                 spin_unlock(&cache->lock);
4552                 btrfs_cleanup_bg_io(cache);
4553         }
4554 }
4555
4556 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4557                                    struct btrfs_fs_info *fs_info)
4558 {
4559         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4560         ASSERT(list_empty(&cur_trans->dirty_bgs));
4561         ASSERT(list_empty(&cur_trans->io_bgs));
4562
4563         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4564
4565         cur_trans->state = TRANS_STATE_COMMIT_START;
4566         wake_up(&fs_info->transaction_blocked_wait);
4567
4568         cur_trans->state = TRANS_STATE_UNBLOCKED;
4569         wake_up(&fs_info->transaction_wait);
4570
4571         btrfs_destroy_delayed_inodes(fs_info);
4572         btrfs_assert_delayed_root_empty(fs_info);
4573
4574         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4575                                      EXTENT_DIRTY);
4576         btrfs_destroy_pinned_extent(fs_info,
4577                                     fs_info->pinned_extents);
4578
4579         cur_trans->state =TRANS_STATE_COMPLETED;
4580         wake_up(&cur_trans->commit_wait);
4581
4582         /*
4583         memset(cur_trans, 0, sizeof(*cur_trans));
4584         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4585         */
4586 }
4587
4588 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4589 {
4590         struct btrfs_transaction *t;
4591
4592         mutex_lock(&fs_info->transaction_kthread_mutex);
4593
4594         spin_lock(&fs_info->trans_lock);
4595         while (!list_empty(&fs_info->trans_list)) {
4596                 t = list_first_entry(&fs_info->trans_list,
4597                                      struct btrfs_transaction, list);
4598                 if (t->state >= TRANS_STATE_COMMIT_START) {
4599                         refcount_inc(&t->use_count);
4600                         spin_unlock(&fs_info->trans_lock);
4601                         btrfs_wait_for_commit(fs_info, t->transid);
4602                         btrfs_put_transaction(t);
4603                         spin_lock(&fs_info->trans_lock);
4604                         continue;
4605                 }
4606                 if (t == fs_info->running_transaction) {
4607                         t->state = TRANS_STATE_COMMIT_DOING;
4608                         spin_unlock(&fs_info->trans_lock);
4609                         /*
4610                          * We wait for 0 num_writers since we don't hold a trans
4611                          * handle open currently for this transaction.
4612                          */
4613                         wait_event(t->writer_wait,
4614                                    atomic_read(&t->num_writers) == 0);
4615                 } else {
4616                         spin_unlock(&fs_info->trans_lock);
4617                 }
4618                 btrfs_cleanup_one_transaction(t, fs_info);
4619
4620                 spin_lock(&fs_info->trans_lock);
4621                 if (t == fs_info->running_transaction)
4622                         fs_info->running_transaction = NULL;
4623                 list_del_init(&t->list);
4624                 spin_unlock(&fs_info->trans_lock);
4625
4626                 btrfs_put_transaction(t);
4627                 trace_btrfs_transaction_commit(fs_info->tree_root);
4628                 spin_lock(&fs_info->trans_lock);
4629         }
4630         spin_unlock(&fs_info->trans_lock);
4631         btrfs_destroy_all_ordered_extents(fs_info);
4632         btrfs_destroy_delayed_inodes(fs_info);
4633         btrfs_assert_delayed_root_empty(fs_info);
4634         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4635         btrfs_destroy_all_delalloc_inodes(fs_info);
4636         mutex_unlock(&fs_info->transaction_kthread_mutex);
4637
4638         return 0;
4639 }
4640
4641 static const struct extent_io_ops btree_extent_io_ops = {
4642         /* mandatory callbacks */
4643         .submit_bio_hook = btree_submit_bio_hook,
4644         .readpage_end_io_hook = btree_readpage_end_io_hook,
4645         /* note we're sharing with inode.c for the merge bio hook */
4646         .merge_bio_hook = btrfs_merge_bio_hook,
4647         .readpage_io_failed_hook = btree_io_failed_hook,
4648
4649         /* optional callbacks */
4650 };