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         blk_status_t status;
  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         blk_status_t status;
 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->status = bio->bi_status;
 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 blk_status_t 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 BLK_STS_RESOURCE;
 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->status = 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         blk_status_t 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->status = 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->status) {
 902                 async->bio->bi_status = async->status;
 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 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info,
 920                 struct inode *inode, struct bio *bio, int mirror_num,
 921                 unsigned long bio_flags, u64 bio_offset,
 922                 extent_submit_bio_hook_t *submit_bio_start,
 923                 extent_submit_bio_hook_t *submit_bio_done)
 924 {
 925         struct async_submit_bio *async;
 926
 927         async = kmalloc(sizeof(*async), GFP_NOFS);
 928         if (!async)
 929                 return BLK_STS_RESOURCE;
 930
 931         async->inode = inode;
 932         async->bio = bio;
 933         async->mirror_num = mirror_num;
 934         async->submit_bio_start = submit_bio_start;
 935         async->submit_bio_done = submit_bio_done;
 936
 937         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 938                         run_one_async_done, run_one_async_free);
 939
 940         async->bio_flags = bio_flags;
 941         async->bio_offset = bio_offset;
 942
 943         async->status = 0;
 944
 945         atomic_inc(&fs_info->nr_async_submits);
 946
 947         if (op_is_sync(bio->bi_opf))
 948                 btrfs_set_work_high_priority(&async->work);
 949
 950         btrfs_queue_work(fs_info->workers, &async->work);
 951
 952         while (atomic_read(&fs_info->async_submit_draining) &&
 953               atomic_read(&fs_info->nr_async_submits)) {
 954                 wait_event(fs_info->async_submit_wait,
 955                            (atomic_read(&fs_info->nr_async_submits) == 0));
 956         }
 957
 958         return 0;
 959 }
 960
 961 static blk_status_t btree_csum_one_bio(struct bio *bio)
 962 {
 963         struct bio_vec *bvec;
 964         struct btrfs_root *root;
 965         int i, ret = 0;
 966
 967         bio_for_each_segment_all(bvec, bio, i) {
 968                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 969                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 970                 if (ret)
 971                         break;
 972         }
 973
 974         return errno_to_blk_status(ret);
 975 }
 976
 977 static blk_status_t __btree_submit_bio_start(struct inode *inode,
 978                 struct bio *bio, int mirror_num, unsigned long bio_flags,
 979                 u64 bio_offset)
 980 {
 981         /*
 982          * when we're called for a write, we're already in the async
 983          * submission context.  Just jump into btrfs_map_bio
 984          */
 985         return btree_csum_one_bio(bio);
 986 }
 987
 988 static blk_status_t __btree_submit_bio_done(struct inode *inode,
 989                 struct bio *bio, int mirror_num, unsigned long bio_flags,
 990                 u64 bio_offset)
 991 {
 992         blk_status_t ret;
 993
 994         /*
 995          * when we're called for a write, we're already in the async
 996          * submission context.  Just jump into btrfs_map_bio
 997          */
 998         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
 999         if (ret) {
1000                 bio->bi_status = ret;
1001                 bio_endio(bio);
1002         }
1003         return ret;
1004 }
1005
1006 static int check_async_write(unsigned long bio_flags)
1007 {
1008         if (bio_flags & EXTENT_BIO_TREE_LOG)
1009                 return 0;
1010 #ifdef CONFIG_X86
1011         if (static_cpu_has(X86_FEATURE_XMM4_2))
1012                 return 0;
1013 #endif
1014         return 1;
1015 }
1016
1017 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1018                                  int mirror_num, unsigned long bio_flags,
1019                                  u64 bio_offset)
1020 {
1021         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1022         int async = check_async_write(bio_flags);
1023         blk_status_t ret;
1024
1025         if (bio_op(bio) != REQ_OP_WRITE) {
1026                 /*
1027                  * called for a read, do the setup so that checksum validation
1028                  * can happen in the async kernel threads
1029                  */
1030                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1031                                           BTRFS_WQ_ENDIO_METADATA);
1032                 if (ret)
1033                         goto out_w_error;
1034                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1035         } else if (!async) {
1036                 ret = btree_csum_one_bio(bio);
1037                 if (ret)
1038                         goto out_w_error;
1039                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1040         } else {
1041                 /*
1042                  * kthread helpers are used to submit writes so that
1043                  * checksumming can happen in parallel across all CPUs
1044                  */
1045                 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1046                                           bio_offset,
1047                                           __btree_submit_bio_start,
1048                                           __btree_submit_bio_done);
1049         }
1050
1051         if (ret)
1052                 goto out_w_error;
1053         return 0;
1054
1055 out_w_error:
1056         bio->bi_status = ret;
1057         bio_endio(bio);
1058         return ret;
1059 }
1060
1061 #ifdef CONFIG_MIGRATION
1062 static int btree_migratepage(struct address_space *mapping,
1063                         struct page *newpage, struct page *page,
1064                         enum migrate_mode mode)
1065 {
1066         /*
1067          * we can't safely write a btree page from here,
1068          * we haven't done the locking hook
1069          */
1070         if (PageDirty(page))
1071                 return -EAGAIN;
1072         /*
1073          * Buffers may be managed in a filesystem specific way.
1074          * We must have no buffers or drop them.
1075          */
1076         if (page_has_private(page) &&
1077             !try_to_release_page(page, GFP_KERNEL))
1078                 return -EAGAIN;
1079         return migrate_page(mapping, newpage, page, mode);
1080 }
1081 #endif
1082
1083
1084 static int btree_writepages(struct address_space *mapping,
1085                             struct writeback_control *wbc)
1086 {
1087         struct btrfs_fs_info *fs_info;
1088         int ret;
1089
1090         if (wbc->sync_mode == WB_SYNC_NONE) {
1091
1092                 if (wbc->for_kupdate)
1093                         return 0;
1094
1095                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1096                 /* this is a bit racy, but that's ok */
1097                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1098                                              BTRFS_DIRTY_METADATA_THRESH);
1099                 if (ret < 0)
1100                         return 0;
1101         }
1102         return btree_write_cache_pages(mapping, wbc);
1103 }
1104
1105 static int btree_readpage(struct file *file, struct page *page)
1106 {
1107         struct extent_io_tree *tree;
1108         tree = &BTRFS_I(page->mapping->host)->io_tree;
1109         return extent_read_full_page(tree, page, btree_get_extent, 0);
1110 }
1111
1112 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1113 {
1114         if (PageWriteback(page) || PageDirty(page))
1115                 return 0;
1116
1117         return try_release_extent_buffer(page);
1118 }
1119
1120 static void btree_invalidatepage(struct page *page, unsigned int offset,
1121                                  unsigned int length)
1122 {
1123         struct extent_io_tree *tree;
1124         tree = &BTRFS_I(page->mapping->host)->io_tree;
1125         extent_invalidatepage(tree, page, offset);
1126         btree_releasepage(page, GFP_NOFS);
1127         if (PagePrivate(page)) {
1128                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1129                            "page private not zero on page %llu",
1130                            (unsigned long long)page_offset(page));
1131                 ClearPagePrivate(page);
1132                 set_page_private(page, 0);
1133                 put_page(page);
1134         }
1135 }
1136
1137 static int btree_set_page_dirty(struct page *page)
1138 {
1139 #ifdef DEBUG
1140         struct extent_buffer *eb;
1141
1142         BUG_ON(!PagePrivate(page));
1143         eb = (struct extent_buffer *)page->private;
1144         BUG_ON(!eb);
1145         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1146         BUG_ON(!atomic_read(&eb->refs));
1147         btrfs_assert_tree_locked(eb);
1148 #endif
1149         return __set_page_dirty_nobuffers(page);
1150 }
1151
1152 static const struct address_space_operations btree_aops = {
1153         .readpage       = btree_readpage,
1154         .writepages     = btree_writepages,
1155         .releasepage    = btree_releasepage,
1156         .invalidatepage = btree_invalidatepage,
1157 #ifdef CONFIG_MIGRATION
1158         .migratepage    = btree_migratepage,
1159 #endif
1160         .set_page_dirty = btree_set_page_dirty,
1161 };
1162
1163 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1164 {
1165         struct extent_buffer *buf = NULL;
1166         struct inode *btree_inode = fs_info->btree_inode;
1167
1168         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1169         if (IS_ERR(buf))
1170                 return;
1171         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1172                                  buf, WAIT_NONE, btree_get_extent, 0);
1173         free_extent_buffer(buf);
1174 }
1175
1176 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1177                          int mirror_num, struct extent_buffer **eb)
1178 {
1179         struct extent_buffer *buf = NULL;
1180         struct inode *btree_inode = fs_info->btree_inode;
1181         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1182         int ret;
1183
1184         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1185         if (IS_ERR(buf))
1186                 return 0;
1187
1188         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1189
1190         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1191                                        btree_get_extent, mirror_num);
1192         if (ret) {
1193                 free_extent_buffer(buf);
1194                 return ret;
1195         }
1196
1197         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1198                 free_extent_buffer(buf);
1199                 return -EIO;
1200         } else if (extent_buffer_uptodate(buf)) {
1201                 *eb = buf;
1202         } else {
1203                 free_extent_buffer(buf);
1204         }
1205         return 0;
1206 }
1207
1208 struct extent_buffer *btrfs_find_create_tree_block(
1209                                                 struct btrfs_fs_info *fs_info,
1210                                                 u64 bytenr)
1211 {
1212         if (btrfs_is_testing(fs_info))
1213                 return alloc_test_extent_buffer(fs_info, bytenr);
1214         return alloc_extent_buffer(fs_info, bytenr);
1215 }
1216
1217
1218 int btrfs_write_tree_block(struct extent_buffer *buf)
1219 {
1220         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1221                                         buf->start + buf->len - 1);
1222 }
1223
1224 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1225 {
1226         return filemap_fdatawait_range(buf->pages[0]->mapping,
1227                                        buf->start, buf->start + buf->len - 1);
1228 }
1229
1230 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1231                                       u64 parent_transid)
1232 {
1233         struct extent_buffer *buf = NULL;
1234         int ret;
1235
1236         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1237         if (IS_ERR(buf))
1238                 return buf;
1239
1240         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1241         if (ret) {
1242                 free_extent_buffer(buf);
1243                 return ERR_PTR(ret);
1244         }
1245         return buf;
1246
1247 }
1248
1249 void clean_tree_block(struct btrfs_fs_info *fs_info,
1250                       struct extent_buffer *buf)
1251 {
1252         if (btrfs_header_generation(buf) ==
1253             fs_info->running_transaction->transid) {
1254                 btrfs_assert_tree_locked(buf);
1255
1256                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1257                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1258                                              -buf->len,
1259                                              fs_info->dirty_metadata_batch);
1260                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1261                         btrfs_set_lock_blocking(buf);
1262                         clear_extent_buffer_dirty(buf);
1263                 }
1264         }
1265 }
1266
1267 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1268 {
1269         struct btrfs_subvolume_writers *writers;
1270         int ret;
1271
1272         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1273         if (!writers)
1274                 return ERR_PTR(-ENOMEM);
1275
1276         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1277         if (ret < 0) {
1278                 kfree(writers);
1279                 return ERR_PTR(ret);
1280         }
1281
1282         init_waitqueue_head(&writers->wait);
1283         return writers;
1284 }
1285
1286 static void
1287 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1288 {
1289         percpu_counter_destroy(&writers->counter);
1290         kfree(writers);
1291 }
1292
1293 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1294                          u64 objectid)
1295 {
1296         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1297         root->node = NULL;
1298         root->commit_root = NULL;
1299         root->state = 0;
1300         root->orphan_cleanup_state = 0;
1301
1302         root->objectid = objectid;
1303         root->last_trans = 0;
1304         root->highest_objectid = 0;
1305         root->nr_delalloc_inodes = 0;
1306         root->nr_ordered_extents = 0;
1307         root->name = NULL;
1308         root->inode_tree = RB_ROOT;
1309         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1310         root->block_rsv = NULL;
1311         root->orphan_block_rsv = NULL;
1312
1313         INIT_LIST_HEAD(&root->dirty_list);
1314         INIT_LIST_HEAD(&root->root_list);
1315         INIT_LIST_HEAD(&root->delalloc_inodes);
1316         INIT_LIST_HEAD(&root->delalloc_root);
1317         INIT_LIST_HEAD(&root->ordered_extents);
1318         INIT_LIST_HEAD(&root->ordered_root);
1319         INIT_LIST_HEAD(&root->logged_list[0]);
1320         INIT_LIST_HEAD(&root->logged_list[1]);
1321         spin_lock_init(&root->orphan_lock);
1322         spin_lock_init(&root->inode_lock);
1323         spin_lock_init(&root->delalloc_lock);
1324         spin_lock_init(&root->ordered_extent_lock);
1325         spin_lock_init(&root->accounting_lock);
1326         spin_lock_init(&root->log_extents_lock[0]);
1327         spin_lock_init(&root->log_extents_lock[1]);
1328         mutex_init(&root->objectid_mutex);
1329         mutex_init(&root->log_mutex);
1330         mutex_init(&root->ordered_extent_mutex);
1331         mutex_init(&root->delalloc_mutex);
1332         init_waitqueue_head(&root->log_writer_wait);
1333         init_waitqueue_head(&root->log_commit_wait[0]);
1334         init_waitqueue_head(&root->log_commit_wait[1]);
1335         INIT_LIST_HEAD(&root->log_ctxs[0]);
1336         INIT_LIST_HEAD(&root->log_ctxs[1]);
1337         atomic_set(&root->log_commit[0], 0);
1338         atomic_set(&root->log_commit[1], 0);
1339         atomic_set(&root->log_writers, 0);
1340         atomic_set(&root->log_batch, 0);
1341         atomic_set(&root->orphan_inodes, 0);
1342         refcount_set(&root->refs, 1);
1343         atomic_set(&root->will_be_snapshoted, 0);
1344         atomic64_set(&root->qgroup_meta_rsv, 0);
1345         root->log_transid = 0;
1346         root->log_transid_committed = -1;
1347         root->last_log_commit = 0;
1348         if (!dummy)
1349                 extent_io_tree_init(&root->dirty_log_pages,
1350                                      fs_info->btree_inode->i_mapping);
1351
1352         memset(&root->root_key, 0, sizeof(root->root_key));
1353         memset(&root->root_item, 0, sizeof(root->root_item));
1354         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1355         if (!dummy)
1356                 root->defrag_trans_start = fs_info->generation;
1357         else
1358                 root->defrag_trans_start = 0;
1359         root->root_key.objectid = objectid;
1360         root->anon_dev = 0;
1361
1362         spin_lock_init(&root->root_item_lock);
1363 }
1364
1365 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1366                 gfp_t flags)
1367 {
1368         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1369         if (root)
1370                 root->fs_info = fs_info;
1371         return root;
1372 }
1373
1374 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1375 /* Should only be used by the testing infrastructure */
1376 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1377 {
1378         struct btrfs_root *root;
1379
1380         if (!fs_info)
1381                 return ERR_PTR(-EINVAL);
1382
1383         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1384         if (!root)
1385                 return ERR_PTR(-ENOMEM);
1386
1387         /* We don't use the stripesize in selftest, set it as sectorsize */
1388         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1389         root->alloc_bytenr = 0;
1390
1391         return root;
1392 }
1393 #endif
1394
1395 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1396                                      struct btrfs_fs_info *fs_info,
1397                                      u64 objectid)
1398 {
1399         struct extent_buffer *leaf;
1400         struct btrfs_root *tree_root = fs_info->tree_root;
1401         struct btrfs_root *root;
1402         struct btrfs_key key;
1403         int ret = 0;
1404         uuid_le uuid;
1405
1406         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1407         if (!root)
1408                 return ERR_PTR(-ENOMEM);
1409
1410         __setup_root(root, fs_info, objectid);
1411         root->root_key.objectid = objectid;
1412         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1413         root->root_key.offset = 0;
1414
1415         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1416         if (IS_ERR(leaf)) {
1417                 ret = PTR_ERR(leaf);
1418                 leaf = NULL;
1419                 goto fail;
1420         }
1421
1422         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1423         btrfs_set_header_bytenr(leaf, leaf->start);
1424         btrfs_set_header_generation(leaf, trans->transid);
1425         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1426         btrfs_set_header_owner(leaf, objectid);
1427         root->node = leaf;
1428
1429         write_extent_buffer_fsid(leaf, fs_info->fsid);
1430         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1431         btrfs_mark_buffer_dirty(leaf);
1432
1433         root->commit_root = btrfs_root_node(root);
1434         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1435
1436         root->root_item.flags = 0;
1437         root->root_item.byte_limit = 0;
1438         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1439         btrfs_set_root_generation(&root->root_item, trans->transid);
1440         btrfs_set_root_level(&root->root_item, 0);
1441         btrfs_set_root_refs(&root->root_item, 1);
1442         btrfs_set_root_used(&root->root_item, leaf->len);
1443         btrfs_set_root_last_snapshot(&root->root_item, 0);
1444         btrfs_set_root_dirid(&root->root_item, 0);
1445         uuid_le_gen(&uuid);
1446         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1447         root->root_item.drop_level = 0;
1448
1449         key.objectid = objectid;
1450         key.type = BTRFS_ROOT_ITEM_KEY;
1451         key.offset = 0;
1452         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1453         if (ret)
1454                 goto fail;
1455
1456         btrfs_tree_unlock(leaf);
1457
1458         return root;
1459
1460 fail:
1461         if (leaf) {
1462                 btrfs_tree_unlock(leaf);
1463                 free_extent_buffer(root->commit_root);
1464                 free_extent_buffer(leaf);
1465         }
1466         kfree(root);
1467
1468         return ERR_PTR(ret);
1469 }
1470
1471 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1472                                          struct btrfs_fs_info *fs_info)
1473 {
1474         struct btrfs_root *root;
1475         struct extent_buffer *leaf;
1476
1477         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1478         if (!root)
1479                 return ERR_PTR(-ENOMEM);
1480
1481         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1482
1483         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1484         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1485         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1486
1487         /*
1488          * DON'T set REF_COWS for log trees
1489          *
1490          * log trees do not get reference counted because they go away
1491          * before a real commit is actually done.  They do store pointers
1492          * to file data extents, and those reference counts still get
1493          * updated (along with back refs to the log tree).
1494          */
1495
1496         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1497                         NULL, 0, 0, 0);
1498         if (IS_ERR(leaf)) {
1499                 kfree(root);
1500                 return ERR_CAST(leaf);
1501         }
1502
1503         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1504         btrfs_set_header_bytenr(leaf, leaf->start);
1505         btrfs_set_header_generation(leaf, trans->transid);
1506         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1507         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1508         root->node = leaf;
1509
1510         write_extent_buffer_fsid(root->node, fs_info->fsid);
1511         btrfs_mark_buffer_dirty(root->node);
1512         btrfs_tree_unlock(root->node);
1513         return root;
1514 }
1515
1516 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1517                              struct btrfs_fs_info *fs_info)
1518 {
1519         struct btrfs_root *log_root;
1520
1521         log_root = alloc_log_tree(trans, fs_info);
1522         if (IS_ERR(log_root))
1523                 return PTR_ERR(log_root);
1524         WARN_ON(fs_info->log_root_tree);
1525         fs_info->log_root_tree = log_root;
1526         return 0;
1527 }
1528
1529 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1530                        struct btrfs_root *root)
1531 {
1532         struct btrfs_fs_info *fs_info = root->fs_info;
1533         struct btrfs_root *log_root;
1534         struct btrfs_inode_item *inode_item;
1535
1536         log_root = alloc_log_tree(trans, fs_info);
1537         if (IS_ERR(log_root))
1538                 return PTR_ERR(log_root);
1539
1540         log_root->last_trans = trans->transid;
1541         log_root->root_key.offset = root->root_key.objectid;
1542
1543         inode_item = &log_root->root_item.inode;
1544         btrfs_set_stack_inode_generation(inode_item, 1);
1545         btrfs_set_stack_inode_size(inode_item, 3);
1546         btrfs_set_stack_inode_nlink(inode_item, 1);
1547         btrfs_set_stack_inode_nbytes(inode_item,
1548                                      fs_info->nodesize);
1549         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1550
1551         btrfs_set_root_node(&log_root->root_item, log_root->node);
1552
1553         WARN_ON(root->log_root);
1554         root->log_root = log_root;
1555         root->log_transid = 0;
1556         root->log_transid_committed = -1;
1557         root->last_log_commit = 0;
1558         return 0;
1559 }
1560
1561 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1562                                                struct btrfs_key *key)
1563 {
1564         struct btrfs_root *root;
1565         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1566         struct btrfs_path *path;
1567         u64 generation;
1568         int ret;
1569
1570         path = btrfs_alloc_path();
1571         if (!path)
1572                 return ERR_PTR(-ENOMEM);
1573
1574         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1575         if (!root) {
1576                 ret = -ENOMEM;
1577                 goto alloc_fail;
1578         }
1579
1580         __setup_root(root, fs_info, key->objectid);
1581
1582         ret = btrfs_find_root(tree_root, key, path,
1583                               &root->root_item, &root->root_key);
1584         if (ret) {
1585                 if (ret > 0)
1586                         ret = -ENOENT;
1587                 goto find_fail;
1588         }
1589
1590         generation = btrfs_root_generation(&root->root_item);
1591         root->node = read_tree_block(fs_info,
1592                                      btrfs_root_bytenr(&root->root_item),
1593                                      generation);
1594         if (IS_ERR(root->node)) {
1595                 ret = PTR_ERR(root->node);
1596                 goto find_fail;
1597         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1598                 ret = -EIO;
1599                 free_extent_buffer(root->node);
1600                 goto find_fail;
1601         }
1602         root->commit_root = btrfs_root_node(root);
1603 out:
1604         btrfs_free_path(path);
1605         return root;
1606
1607 find_fail:
1608         kfree(root);
1609 alloc_fail:
1610         root = ERR_PTR(ret);
1611         goto out;
1612 }
1613
1614 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1615                                       struct btrfs_key *location)
1616 {
1617         struct btrfs_root *root;
1618
1619         root = btrfs_read_tree_root(tree_root, location);
1620         if (IS_ERR(root))
1621                 return root;
1622
1623         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1624                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1625                 btrfs_check_and_init_root_item(&root->root_item);
1626         }
1627
1628         return root;
1629 }
1630
1631 int btrfs_init_fs_root(struct btrfs_root *root)
1632 {
1633         int ret;
1634         struct btrfs_subvolume_writers *writers;
1635
1636         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1637         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1638                                         GFP_NOFS);
1639         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1640                 ret = -ENOMEM;
1641                 goto fail;
1642         }
1643
1644         writers = btrfs_alloc_subvolume_writers();
1645         if (IS_ERR(writers)) {
1646                 ret = PTR_ERR(writers);
1647                 goto fail;
1648         }
1649         root->subv_writers = writers;
1650
1651         btrfs_init_free_ino_ctl(root);
1652         spin_lock_init(&root->ino_cache_lock);
1653         init_waitqueue_head(&root->ino_cache_wait);
1654
1655         ret = get_anon_bdev(&root->anon_dev);
1656         if (ret)
1657                 goto fail;
1658
1659         mutex_lock(&root->objectid_mutex);
1660         ret = btrfs_find_highest_objectid(root,
1661                                         &root->highest_objectid);
1662         if (ret) {
1663                 mutex_unlock(&root->objectid_mutex);
1664                 goto fail;
1665         }
1666
1667         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1668
1669         mutex_unlock(&root->objectid_mutex);
1670
1671         return 0;
1672 fail:
1673         /* the caller is responsible to call free_fs_root */
1674         return ret;
1675 }
1676
1677 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1678                                         u64 root_id)
1679 {
1680         struct btrfs_root *root;
1681
1682         spin_lock(&fs_info->fs_roots_radix_lock);
1683         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1684                                  (unsigned long)root_id);
1685         spin_unlock(&fs_info->fs_roots_radix_lock);
1686         return root;
1687 }
1688
1689 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1690                          struct btrfs_root *root)
1691 {
1692         int ret;
1693
1694         ret = radix_tree_preload(GFP_NOFS);
1695         if (ret)
1696                 return ret;
1697
1698         spin_lock(&fs_info->fs_roots_radix_lock);
1699         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1700                                 (unsigned long)root->root_key.objectid,
1701                                 root);
1702         if (ret == 0)
1703                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1704         spin_unlock(&fs_info->fs_roots_radix_lock);
1705         radix_tree_preload_end();
1706
1707         return ret;
1708 }
1709
1710 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1711                                      struct btrfs_key *location,
1712                                      bool check_ref)
1713 {
1714         struct btrfs_root *root;
1715         struct btrfs_path *path;
1716         struct btrfs_key key;
1717         int ret;
1718
1719         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1720                 return fs_info->tree_root;
1721         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1722                 return fs_info->extent_root;
1723         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1724                 return fs_info->chunk_root;
1725         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1726                 return fs_info->dev_root;
1727         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1728                 return fs_info->csum_root;
1729         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1730                 return fs_info->quota_root ? fs_info->quota_root :
1731                                              ERR_PTR(-ENOENT);
1732         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1733                 return fs_info->uuid_root ? fs_info->uuid_root :
1734                                             ERR_PTR(-ENOENT);
1735         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1736                 return fs_info->free_space_root ? fs_info->free_space_root :
1737                                                   ERR_PTR(-ENOENT);
1738 again:
1739         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1740         if (root) {
1741                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1742                         return ERR_PTR(-ENOENT);
1743                 return root;
1744         }
1745
1746         root = btrfs_read_fs_root(fs_info->tree_root, location);
1747         if (IS_ERR(root))
1748                 return root;
1749
1750         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1751                 ret = -ENOENT;
1752                 goto fail;
1753         }
1754
1755         ret = btrfs_init_fs_root(root);
1756         if (ret)
1757                 goto fail;
1758
1759         path = btrfs_alloc_path();
1760         if (!path) {
1761                 ret = -ENOMEM;
1762                 goto fail;
1763         }
1764         key.objectid = BTRFS_ORPHAN_OBJECTID;
1765         key.type = BTRFS_ORPHAN_ITEM_KEY;
1766         key.offset = location->objectid;
1767
1768         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1769         btrfs_free_path(path);
1770         if (ret < 0)
1771                 goto fail;
1772         if (ret == 0)
1773                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1774
1775         ret = btrfs_insert_fs_root(fs_info, root);
1776         if (ret) {
1777                 if (ret == -EEXIST) {
1778                         free_fs_root(root);
1779                         goto again;
1780                 }
1781                 goto fail;
1782         }
1783         return root;
1784 fail:
1785         free_fs_root(root);
1786         return ERR_PTR(ret);
1787 }
1788
1789 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1790 {
1791         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1792         int ret = 0;
1793         struct btrfs_device *device;
1794         struct backing_dev_info *bdi;
1795
1796         rcu_read_lock();
1797         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1798                 if (!device->bdev)
1799                         continue;
1800                 bdi = device->bdev->bd_bdi;
1801                 if (bdi_congested(bdi, bdi_bits)) {
1802                         ret = 1;
1803                         break;
1804                 }
1805         }
1806         rcu_read_unlock();
1807         return ret;
1808 }
1809
1810 /*
1811  * called by the kthread helper functions to finally call the bio end_io
1812  * functions.  This is where read checksum verification actually happens
1813  */
1814 static void end_workqueue_fn(struct btrfs_work *work)
1815 {
1816         struct bio *bio;
1817         struct btrfs_end_io_wq *end_io_wq;
1818
1819         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1820         bio = end_io_wq->bio;
1821
1822         bio->bi_status = end_io_wq->status;
1823         bio->bi_private = end_io_wq->private;
1824         bio->bi_end_io = end_io_wq->end_io;
1825         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1826         bio_endio(bio);
1827 }
1828
1829 static int cleaner_kthread(void *arg)
1830 {
1831         struct btrfs_root *root = arg;
1832         struct btrfs_fs_info *fs_info = root->fs_info;
1833         int again;
1834         struct btrfs_trans_handle *trans;
1835
1836         do {
1837                 again = 0;
1838
1839                 /* Make the cleaner go to sleep early. */
1840                 if (btrfs_need_cleaner_sleep(fs_info))
1841                         goto sleep;
1842
1843                 /*
1844                  * Do not do anything if we might cause open_ctree() to block
1845                  * before we have finished mounting the filesystem.
1846                  */
1847                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1848                         goto sleep;
1849
1850                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1851                         goto sleep;
1852
1853                 /*
1854                  * Avoid the problem that we change the status of the fs
1855                  * during the above check and trylock.
1856                  */
1857                 if (btrfs_need_cleaner_sleep(fs_info)) {
1858                         mutex_unlock(&fs_info->cleaner_mutex);
1859                         goto sleep;
1860                 }
1861
1862                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1863                 btrfs_run_delayed_iputs(fs_info);
1864                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1865
1866                 again = btrfs_clean_one_deleted_snapshot(root);
1867                 mutex_unlock(&fs_info->cleaner_mutex);
1868
1869                 /*
1870                  * The defragger has dealt with the R/O remount and umount,
1871                  * needn't do anything special here.
1872                  */
1873                 btrfs_run_defrag_inodes(fs_info);
1874
1875                 /*
1876                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1877                  * with relocation (btrfs_relocate_chunk) and relocation
1878                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1879                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1880                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1881                  * unused block groups.
1882                  */
1883                 btrfs_delete_unused_bgs(fs_info);
1884 sleep:
1885                 if (!again) {
1886                         set_current_state(TASK_INTERRUPTIBLE);
1887                         if (!kthread_should_stop())
1888                                 schedule();
1889                         __set_current_state(TASK_RUNNING);
1890                 }
1891         } while (!kthread_should_stop());
1892
1893         /*
1894          * Transaction kthread is stopped before us and wakes us up.
1895          * However we might have started a new transaction and COWed some
1896          * tree blocks when deleting unused block groups for example. So
1897          * make sure we commit the transaction we started to have a clean
1898          * shutdown when evicting the btree inode - if it has dirty pages
1899          * when we do the final iput() on it, eviction will trigger a
1900          * writeback for it which will fail with null pointer dereferences
1901          * since work queues and other resources were already released and
1902          * destroyed by the time the iput/eviction/writeback is made.
1903          */
1904         trans = btrfs_attach_transaction(root);
1905         if (IS_ERR(trans)) {
1906                 if (PTR_ERR(trans) != -ENOENT)
1907                         btrfs_err(fs_info,
1908                                   "cleaner transaction attach returned %ld",
1909                                   PTR_ERR(trans));
1910         } else {
1911                 int ret;
1912
1913                 ret = btrfs_commit_transaction(trans);
1914                 if (ret)
1915                         btrfs_err(fs_info,
1916                                   "cleaner open transaction commit returned %d",
1917                                   ret);
1918         }
1919
1920         return 0;
1921 }
1922
1923 static int transaction_kthread(void *arg)
1924 {
1925         struct btrfs_root *root = arg;
1926         struct btrfs_fs_info *fs_info = root->fs_info;
1927         struct btrfs_trans_handle *trans;
1928         struct btrfs_transaction *cur;
1929         u64 transid;
1930         unsigned long now;
1931         unsigned long delay;
1932         bool cannot_commit;
1933
1934         do {
1935                 cannot_commit = false;
1936                 delay = HZ * fs_info->commit_interval;
1937                 mutex_lock(&fs_info->transaction_kthread_mutex);
1938
1939                 spin_lock(&fs_info->trans_lock);
1940                 cur = fs_info->running_transaction;
1941                 if (!cur) {
1942                         spin_unlock(&fs_info->trans_lock);
1943                         goto sleep;
1944                 }
1945
1946                 now = get_seconds();
1947                 if (cur->state < TRANS_STATE_BLOCKED &&
1948                     (now < cur->start_time ||
1949                      now - cur->start_time < fs_info->commit_interval)) {
1950                         spin_unlock(&fs_info->trans_lock);
1951                         delay = HZ * 5;
1952                         goto sleep;
1953                 }
1954                 transid = cur->transid;
1955                 spin_unlock(&fs_info->trans_lock);
1956
1957                 /* If the file system is aborted, this will always fail. */
1958                 trans = btrfs_attach_transaction(root);
1959                 if (IS_ERR(trans)) {
1960                         if (PTR_ERR(trans) != -ENOENT)
1961                                 cannot_commit = true;
1962                         goto sleep;
1963                 }
1964                 if (transid == trans->transid) {
1965                         btrfs_commit_transaction(trans);
1966                 } else {
1967                         btrfs_end_transaction(trans);
1968                 }
1969 sleep:
1970                 wake_up_process(fs_info->cleaner_kthread);
1971                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1972
1973                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1974                                       &fs_info->fs_state)))
1975                         btrfs_cleanup_transaction(fs_info);
1976                 set_current_state(TASK_INTERRUPTIBLE);
1977                 if (!kthread_should_stop() &&
1978                                 (!btrfs_transaction_blocked(fs_info) ||
1979                                  cannot_commit))
1980                         schedule_timeout(delay);
1981                 __set_current_state(TASK_RUNNING);
1982         } while (!kthread_should_stop());
1983         return 0;
1984 }
1985
1986 /*
1987  * this will find the highest generation in the array of
1988  * root backups.  The index of the highest array is returned,
1989  * or -1 if we can't find anything.
1990  *
1991  * We check to make sure the array is valid by comparing the
1992  * generation of the latest  root in the array with the generation
1993  * in the super block.  If they don't match we pitch it.
1994  */
1995 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1996 {
1997         u64 cur;
1998         int newest_index = -1;
1999         struct btrfs_root_backup *root_backup;
2000         int i;
2001
2002         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2003                 root_backup = info->super_copy->super_roots + i;
2004                 cur = btrfs_backup_tree_root_gen(root_backup);
2005                 if (cur == newest_gen)
2006                         newest_index = i;
2007         }
2008
2009         /* check to see if we actually wrapped around */
2010         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2011                 root_backup = info->super_copy->super_roots;
2012                 cur = btrfs_backup_tree_root_gen(root_backup);
2013                 if (cur == newest_gen)
2014                         newest_index = 0;
2015         }
2016         return newest_index;
2017 }
2018
2019
2020 /*
2021  * find the oldest backup so we know where to store new entries
2022  * in the backup array.  This will set the backup_root_index
2023  * field in the fs_info struct
2024  */
2025 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2026                                      u64 newest_gen)
2027 {
2028         int newest_index = -1;
2029
2030         newest_index = find_newest_super_backup(info, newest_gen);
2031         /* if there was garbage in there, just move along */
2032         if (newest_index == -1) {
2033                 info->backup_root_index = 0;
2034         } else {
2035                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2036         }
2037 }
2038
2039 /*
2040  * copy all the root pointers into the super backup array.
2041  * this will bump the backup pointer by one when it is
2042  * done
2043  */
2044 static void backup_super_roots(struct btrfs_fs_info *info)
2045 {
2046         int next_backup;
2047         struct btrfs_root_backup *root_backup;
2048         int last_backup;
2049
2050         next_backup = info->backup_root_index;
2051         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2052                 BTRFS_NUM_BACKUP_ROOTS;
2053
2054         /*
2055          * just overwrite the last backup if we're at the same generation
2056          * this happens only at umount
2057          */
2058         root_backup = info->super_for_commit->super_roots + last_backup;
2059         if (btrfs_backup_tree_root_gen(root_backup) ==
2060             btrfs_header_generation(info->tree_root->node))
2061                 next_backup = last_backup;
2062
2063         root_backup = info->super_for_commit->super_roots + next_backup;
2064
2065         /*
2066          * make sure all of our padding and empty slots get zero filled
2067          * regardless of which ones we use today
2068          */
2069         memset(root_backup, 0, sizeof(*root_backup));
2070
2071         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2072
2073         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2074         btrfs_set_backup_tree_root_gen(root_backup,
2075                                btrfs_header_generation(info->tree_root->node));
2076
2077         btrfs_set_backup_tree_root_level(root_backup,
2078                                btrfs_header_level(info->tree_root->node));
2079
2080         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2081         btrfs_set_backup_chunk_root_gen(root_backup,
2082                                btrfs_header_generation(info->chunk_root->node));
2083         btrfs_set_backup_chunk_root_level(root_backup,
2084                                btrfs_header_level(info->chunk_root->node));
2085
2086         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2087         btrfs_set_backup_extent_root_gen(root_backup,
2088                                btrfs_header_generation(info->extent_root->node));
2089         btrfs_set_backup_extent_root_level(root_backup,
2090                                btrfs_header_level(info->extent_root->node));
2091
2092         /*
2093          * we might commit during log recovery, which happens before we set
2094          * the fs_root.  Make sure it is valid before we fill it in.
2095          */
2096         if (info->fs_root && info->fs_root->node) {
2097                 btrfs_set_backup_fs_root(root_backup,
2098                                          info->fs_root->node->start);
2099                 btrfs_set_backup_fs_root_gen(root_backup,
2100                                btrfs_header_generation(info->fs_root->node));
2101                 btrfs_set_backup_fs_root_level(root_backup,
2102                                btrfs_header_level(info->fs_root->node));
2103         }
2104
2105         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2106         btrfs_set_backup_dev_root_gen(root_backup,
2107                                btrfs_header_generation(info->dev_root->node));
2108         btrfs_set_backup_dev_root_level(root_backup,
2109                                        btrfs_header_level(info->dev_root->node));
2110
2111         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2112         btrfs_set_backup_csum_root_gen(root_backup,
2113                                btrfs_header_generation(info->csum_root->node));
2114         btrfs_set_backup_csum_root_level(root_backup,
2115                                btrfs_header_level(info->csum_root->node));
2116
2117         btrfs_set_backup_total_bytes(root_backup,
2118                              btrfs_super_total_bytes(info->super_copy));
2119         btrfs_set_backup_bytes_used(root_backup,
2120                              btrfs_super_bytes_used(info->super_copy));
2121         btrfs_set_backup_num_devices(root_backup,
2122                              btrfs_super_num_devices(info->super_copy));
2123
2124         /*
2125          * if we don't copy this out to the super_copy, it won't get remembered
2126          * for the next commit
2127          */
2128         memcpy(&info->super_copy->super_roots,
2129                &info->super_for_commit->super_roots,
2130                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2131 }
2132
2133 /*
2134  * this copies info out of the root backup array and back into
2135  * the in-memory super block.  It is meant to help iterate through
2136  * the array, so you send it the number of backups you've already
2137  * tried and the last backup index you used.
2138  *
2139  * this returns -1 when it has tried all the backups
2140  */
2141 static noinline int next_root_backup(struct btrfs_fs_info *info,
2142                                      struct btrfs_super_block *super,
2143                                      int *num_backups_tried, int *backup_index)
2144 {
2145         struct btrfs_root_backup *root_backup;
2146         int newest = *backup_index;
2147
2148         if (*num_backups_tried == 0) {
2149                 u64 gen = btrfs_super_generation(super);
2150
2151                 newest = find_newest_super_backup(info, gen);
2152                 if (newest == -1)
2153                         return -1;
2154
2155                 *backup_index = newest;
2156                 *num_backups_tried = 1;
2157         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2158                 /* we've tried all the backups, all done */
2159                 return -1;
2160         } else {
2161                 /* jump to the next oldest backup */
2162                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2163                         BTRFS_NUM_BACKUP_ROOTS;
2164                 *backup_index = newest;
2165                 *num_backups_tried += 1;
2166         }
2167         root_backup = super->super_roots + newest;
2168
2169         btrfs_set_super_generation(super,
2170                                    btrfs_backup_tree_root_gen(root_backup));
2171         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2172         btrfs_set_super_root_level(super,
2173                                    btrfs_backup_tree_root_level(root_backup));
2174         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2175
2176         /*
2177          * fixme: the total bytes and num_devices need to match or we should
2178          * need a fsck
2179          */
2180         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2181         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2182         return 0;
2183 }
2184
2185 /* helper to cleanup workers */
2186 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2187 {
2188         btrfs_destroy_workqueue(fs_info->fixup_workers);
2189         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2190         btrfs_destroy_workqueue(fs_info->workers);
2191         btrfs_destroy_workqueue(fs_info->endio_workers);
2192         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2193         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2194         btrfs_destroy_workqueue(fs_info->rmw_workers);
2195         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2196         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2197         btrfs_destroy_workqueue(fs_info->submit_workers);
2198         btrfs_destroy_workqueue(fs_info->delayed_workers);
2199         btrfs_destroy_workqueue(fs_info->caching_workers);
2200         btrfs_destroy_workqueue(fs_info->readahead_workers);
2201         btrfs_destroy_workqueue(fs_info->flush_workers);
2202         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2203         btrfs_destroy_workqueue(fs_info->extent_workers);
2204         /*
2205          * Now that all other work queues are destroyed, we can safely destroy
2206          * the queues used for metadata I/O, since tasks from those other work
2207          * queues can do metadata I/O operations.
2208          */
2209         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2210         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2211 }
2212
2213 static void free_root_extent_buffers(struct btrfs_root *root)
2214 {
2215         if (root) {
2216                 free_extent_buffer(root->node);
2217                 free_extent_buffer(root->commit_root);
2218                 root->node = NULL;
2219                 root->commit_root = NULL;
2220         }
2221 }
2222
2223 /* helper to cleanup tree roots */
2224 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2225 {
2226         free_root_extent_buffers(info->tree_root);
2227
2228         free_root_extent_buffers(info->dev_root);
2229         free_root_extent_buffers(info->extent_root);
2230         free_root_extent_buffers(info->csum_root);
2231         free_root_extent_buffers(info->quota_root);
2232         free_root_extent_buffers(info->uuid_root);
2233         if (chunk_root)
2234                 free_root_extent_buffers(info->chunk_root);
2235         free_root_extent_buffers(info->free_space_root);
2236 }
2237
2238 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2239 {
2240         int ret;
2241         struct btrfs_root *gang[8];
2242         int i;
2243
2244         while (!list_empty(&fs_info->dead_roots)) {
2245                 gang[0] = list_entry(fs_info->dead_roots.next,
2246                                      struct btrfs_root, root_list);
2247                 list_del(&gang[0]->root_list);
2248
2249                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2250                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2251                 } else {
2252                         free_extent_buffer(gang[0]->node);
2253                         free_extent_buffer(gang[0]->commit_root);
2254                         btrfs_put_fs_root(gang[0]);
2255                 }
2256         }
2257
2258         while (1) {
2259                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2260                                              (void **)gang, 0,
2261                                              ARRAY_SIZE(gang));
2262                 if (!ret)
2263                         break;
2264                 for (i = 0; i < ret; i++)
2265                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2266         }
2267
2268         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2269                 btrfs_free_log_root_tree(NULL, fs_info);
2270                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2271         }
2272 }
2273
2274 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2275 {
2276         mutex_init(&fs_info->scrub_lock);
2277         atomic_set(&fs_info->scrubs_running, 0);
2278         atomic_set(&fs_info->scrub_pause_req, 0);
2279         atomic_set(&fs_info->scrubs_paused, 0);
2280         atomic_set(&fs_info->scrub_cancel_req, 0);
2281         init_waitqueue_head(&fs_info->scrub_pause_wait);
2282         fs_info->scrub_workers_refcnt = 0;
2283 }
2284
2285 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2286 {
2287         spin_lock_init(&fs_info->balance_lock);
2288         mutex_init(&fs_info->balance_mutex);
2289         atomic_set(&fs_info->balance_running, 0);
2290         atomic_set(&fs_info->balance_pause_req, 0);
2291         atomic_set(&fs_info->balance_cancel_req, 0);
2292         fs_info->balance_ctl = NULL;
2293         init_waitqueue_head(&fs_info->balance_wait_q);
2294 }
2295
2296 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2297 {
2298         struct inode *inode = fs_info->btree_inode;
2299
2300         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2301         set_nlink(inode, 1);
2302         /*
2303          * we set the i_size on the btree inode to the max possible int.
2304          * the real end of the address space is determined by all of
2305          * the devices in the system
2306          */
2307         inode->i_size = OFFSET_MAX;
2308         inode->i_mapping->a_ops = &btree_aops;
2309
2310         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2311         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping);
2312         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2313         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2314
2315         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2316
2317         BTRFS_I(inode)->root = fs_info->tree_root;
2318         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2319         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2320         btrfs_insert_inode_hash(inode);
2321 }
2322
2323 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2324 {
2325         fs_info->dev_replace.lock_owner = 0;
2326         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2327         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2328         rwlock_init(&fs_info->dev_replace.lock);
2329         atomic_set(&fs_info->dev_replace.read_locks, 0);
2330         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2331         init_waitqueue_head(&fs_info->replace_wait);
2332         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2333 }
2334
2335 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2336 {
2337         spin_lock_init(&fs_info->qgroup_lock);
2338         mutex_init(&fs_info->qgroup_ioctl_lock);
2339         fs_info->qgroup_tree = RB_ROOT;
2340         fs_info->qgroup_op_tree = RB_ROOT;
2341         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2342         fs_info->qgroup_seq = 1;
2343         fs_info->qgroup_ulist = NULL;
2344         fs_info->qgroup_rescan_running = false;
2345         mutex_init(&fs_info->qgroup_rescan_lock);
2346 }
2347
2348 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2349                 struct btrfs_fs_devices *fs_devices)
2350 {
2351         int max_active = fs_info->thread_pool_size;
2352         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2353
2354         fs_info->workers =
2355                 btrfs_alloc_workqueue(fs_info, "worker",
2356                                       flags | WQ_HIGHPRI, max_active, 16);
2357
2358         fs_info->delalloc_workers =
2359                 btrfs_alloc_workqueue(fs_info, "delalloc",
2360                                       flags, max_active, 2);
2361
2362         fs_info->flush_workers =
2363                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2364                                       flags, max_active, 0);
2365
2366         fs_info->caching_workers =
2367                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2368
2369         /*
2370          * a higher idle thresh on the submit workers makes it much more
2371          * likely that bios will be send down in a sane order to the
2372          * devices
2373          */
2374         fs_info->submit_workers =
2375                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2376                                       min_t(u64, fs_devices->num_devices,
2377                                             max_active), 64);
2378
2379         fs_info->fixup_workers =
2380                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2381
2382         /*
2383          * endios are largely parallel and should have a very
2384          * low idle thresh
2385          */
2386         fs_info->endio_workers =
2387                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2388         fs_info->endio_meta_workers =
2389                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2390                                       max_active, 4);
2391         fs_info->endio_meta_write_workers =
2392                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2393                                       max_active, 2);
2394         fs_info->endio_raid56_workers =
2395                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2396                                       max_active, 4);
2397         fs_info->endio_repair_workers =
2398                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2399         fs_info->rmw_workers =
2400                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2401         fs_info->endio_write_workers =
2402                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2403                                       max_active, 2);
2404         fs_info->endio_freespace_worker =
2405                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2406                                       max_active, 0);
2407         fs_info->delayed_workers =
2408                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2409                                       max_active, 0);
2410         fs_info->readahead_workers =
2411                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2412                                       max_active, 2);
2413         fs_info->qgroup_rescan_workers =
2414                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2415         fs_info->extent_workers =
2416                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2417                                       min_t(u64, fs_devices->num_devices,
2418                                             max_active), 8);
2419
2420         if (!(fs_info->workers && fs_info->delalloc_workers &&
2421               fs_info->submit_workers && fs_info->flush_workers &&
2422               fs_info->endio_workers && fs_info->endio_meta_workers &&
2423               fs_info->endio_meta_write_workers &&
2424               fs_info->endio_repair_workers &&
2425               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2426               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2427               fs_info->caching_workers && fs_info->readahead_workers &&
2428               fs_info->fixup_workers && fs_info->delayed_workers &&
2429               fs_info->extent_workers &&
2430               fs_info->qgroup_rescan_workers)) {
2431                 return -ENOMEM;
2432         }
2433
2434         return 0;
2435 }
2436
2437 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2438                             struct btrfs_fs_devices *fs_devices)
2439 {
2440         int ret;
2441         struct btrfs_root *log_tree_root;
2442         struct btrfs_super_block *disk_super = fs_info->super_copy;
2443         u64 bytenr = btrfs_super_log_root(disk_super);
2444
2445         if (fs_devices->rw_devices == 0) {
2446                 btrfs_warn(fs_info, "log replay required on RO media");
2447                 return -EIO;
2448         }
2449
2450         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2451         if (!log_tree_root)
2452                 return -ENOMEM;
2453
2454         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2455
2456         log_tree_root->node = read_tree_block(fs_info, bytenr,
2457                                               fs_info->generation + 1);
2458         if (IS_ERR(log_tree_root->node)) {
2459                 btrfs_warn(fs_info, "failed to read log tree");
2460                 ret = PTR_ERR(log_tree_root->node);
2461                 kfree(log_tree_root);
2462                 return ret;
2463         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2464                 btrfs_err(fs_info, "failed to read log tree");
2465                 free_extent_buffer(log_tree_root->node);
2466                 kfree(log_tree_root);
2467                 return -EIO;
2468         }
2469         /* returns with log_tree_root freed on success */
2470         ret = btrfs_recover_log_trees(log_tree_root);
2471         if (ret) {
2472                 btrfs_handle_fs_error(fs_info, ret,
2473                                       "Failed to recover log tree");
2474                 free_extent_buffer(log_tree_root->node);
2475                 kfree(log_tree_root);
2476                 return ret;
2477         }
2478
2479         if (fs_info->sb->s_flags & MS_RDONLY) {
2480                 ret = btrfs_commit_super(fs_info);
2481                 if (ret)
2482                         return ret;
2483         }
2484
2485         return 0;
2486 }
2487
2488 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2489 {
2490         struct btrfs_root *tree_root = fs_info->tree_root;
2491         struct btrfs_root *root;
2492         struct btrfs_key location;
2493         int ret;
2494
2495         BUG_ON(!fs_info->tree_root);
2496
2497         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2498         location.type = BTRFS_ROOT_ITEM_KEY;
2499         location.offset = 0;
2500
2501         root = btrfs_read_tree_root(tree_root, &location);
2502         if (IS_ERR(root))
2503                 return PTR_ERR(root);
2504         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2505         fs_info->extent_root = root;
2506
2507         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2508         root = btrfs_read_tree_root(tree_root, &location);
2509         if (IS_ERR(root))
2510                 return PTR_ERR(root);
2511         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2512         fs_info->dev_root = root;
2513         btrfs_init_devices_late(fs_info);
2514
2515         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2516         root = btrfs_read_tree_root(tree_root, &location);
2517         if (IS_ERR(root))
2518                 return PTR_ERR(root);
2519         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2520         fs_info->csum_root = root;
2521
2522         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2523         root = btrfs_read_tree_root(tree_root, &location);
2524         if (!IS_ERR(root)) {
2525                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2526                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2527                 fs_info->quota_root = root;
2528         }
2529
2530         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2531         root = btrfs_read_tree_root(tree_root, &location);
2532         if (IS_ERR(root)) {
2533                 ret = PTR_ERR(root);
2534                 if (ret != -ENOENT)
2535                         return ret;
2536         } else {
2537                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2538                 fs_info->uuid_root = root;
2539         }
2540
2541         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2542                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2543                 root = btrfs_read_tree_root(tree_root, &location);
2544                 if (IS_ERR(root))
2545                         return PTR_ERR(root);
2546                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2547                 fs_info->free_space_root = root;
2548         }
2549
2550         return 0;
2551 }
2552
2553 int open_ctree(struct super_block *sb,
2554                struct btrfs_fs_devices *fs_devices,
2555                char *options)
2556 {
2557         u32 sectorsize;
2558         u32 nodesize;
2559         u32 stripesize;
2560         u64 generation;
2561         u64 features;
2562         struct btrfs_key location;
2563         struct buffer_head *bh;
2564         struct btrfs_super_block *disk_super;
2565         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2566         struct btrfs_root *tree_root;
2567         struct btrfs_root *chunk_root;
2568         int ret;
2569         int err = -EINVAL;
2570         int num_backups_tried = 0;
2571         int backup_index = 0;
2572         int max_active;
2573         int clear_free_space_tree = 0;
2574
2575         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2576         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2577         if (!tree_root || !chunk_root) {
2578                 err = -ENOMEM;
2579                 goto fail;
2580         }
2581
2582         ret = init_srcu_struct(&fs_info->subvol_srcu);
2583         if (ret) {
2584                 err = ret;
2585                 goto fail;
2586         }
2587
2588         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2589         if (ret) {
2590                 err = ret;
2591                 goto fail_srcu;
2592         }
2593         fs_info->dirty_metadata_batch = PAGE_SIZE *
2594                                         (1 + ilog2(nr_cpu_ids));
2595
2596         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2597         if (ret) {
2598                 err = ret;
2599                 goto fail_dirty_metadata_bytes;
2600         }
2601
2602         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2603         if (ret) {
2604                 err = ret;
2605                 goto fail_delalloc_bytes;
2606         }
2607
2608         fs_info->btree_inode = new_inode(sb);
2609         if (!fs_info->btree_inode) {
2610                 err = -ENOMEM;
2611                 goto fail_bio_counter;
2612         }
2613
2614         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2615
2616         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2617         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2618         INIT_LIST_HEAD(&fs_info->trans_list);
2619         INIT_LIST_HEAD(&fs_info->dead_roots);
2620         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2621         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2622         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2623         spin_lock_init(&fs_info->delalloc_root_lock);
2624         spin_lock_init(&fs_info->trans_lock);
2625         spin_lock_init(&fs_info->fs_roots_radix_lock);
2626         spin_lock_init(&fs_info->delayed_iput_lock);
2627         spin_lock_init(&fs_info->defrag_inodes_lock);
2628         spin_lock_init(&fs_info->free_chunk_lock);
2629         spin_lock_init(&fs_info->tree_mod_seq_lock);
2630         spin_lock_init(&fs_info->super_lock);
2631         spin_lock_init(&fs_info->qgroup_op_lock);
2632         spin_lock_init(&fs_info->buffer_lock);
2633         spin_lock_init(&fs_info->unused_bgs_lock);
2634         rwlock_init(&fs_info->tree_mod_log_lock);
2635         mutex_init(&fs_info->unused_bg_unpin_mutex);
2636         mutex_init(&fs_info->delete_unused_bgs_mutex);
2637         mutex_init(&fs_info->reloc_mutex);
2638         mutex_init(&fs_info->delalloc_root_mutex);
2639         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2640         seqlock_init(&fs_info->profiles_lock);
2641
2642         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2643         INIT_LIST_HEAD(&fs_info->space_info);
2644         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2645         INIT_LIST_HEAD(&fs_info->unused_bgs);
2646         btrfs_mapping_init(&fs_info->mapping_tree);
2647         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2648                              BTRFS_BLOCK_RSV_GLOBAL);
2649         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2650                              BTRFS_BLOCK_RSV_DELALLOC);
2651         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2652         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2653         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2654         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2655                              BTRFS_BLOCK_RSV_DELOPS);
2656         atomic_set(&fs_info->nr_async_submits, 0);
2657         atomic_set(&fs_info->async_delalloc_pages, 0);
2658         atomic_set(&fs_info->async_submit_draining, 0);
2659         atomic_set(&fs_info->nr_async_bios, 0);
2660         atomic_set(&fs_info->defrag_running, 0);
2661         atomic_set(&fs_info->qgroup_op_seq, 0);
2662         atomic_set(&fs_info->reada_works_cnt, 0);
2663         atomic64_set(&fs_info->tree_mod_seq, 0);
2664         fs_info->fs_frozen = 0;
2665         fs_info->sb = sb;
2666         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2667         fs_info->metadata_ratio = 0;
2668         fs_info->defrag_inodes = RB_ROOT;
2669         fs_info->free_chunk_space = 0;
2670         fs_info->tree_mod_log = RB_ROOT;
2671         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2672         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2673         /* readahead state */
2674         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2675         spin_lock_init(&fs_info->reada_lock);
2676
2677         fs_info->thread_pool_size = min_t(unsigned long,
2678                                           num_online_cpus() + 2, 8);
2679
2680         INIT_LIST_HEAD(&fs_info->ordered_roots);
2681         spin_lock_init(&fs_info->ordered_root_lock);
2682         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2683                                         GFP_KERNEL);
2684         if (!fs_info->delayed_root) {
2685                 err = -ENOMEM;
2686                 goto fail_iput;
2687         }
2688         btrfs_init_delayed_root(fs_info->delayed_root);
2689
2690         btrfs_init_scrub(fs_info);
2691 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2692         fs_info->check_integrity_print_mask = 0;
2693 #endif
2694         btrfs_init_balance(fs_info);
2695         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2696
2697         sb->s_blocksize = 4096;
2698         sb->s_blocksize_bits = blksize_bits(4096);
2699
2700         btrfs_init_btree_inode(fs_info);
2701
2702         spin_lock_init(&fs_info->block_group_cache_lock);
2703         fs_info->block_group_cache_tree = RB_ROOT;
2704         fs_info->first_logical_byte = (u64)-1;
2705
2706         extent_io_tree_init(&fs_info->freed_extents[0],
2707                              fs_info->btree_inode->i_mapping);
2708         extent_io_tree_init(&fs_info->freed_extents[1],
2709                              fs_info->btree_inode->i_mapping);
2710         fs_info->pinned_extents = &fs_info->freed_extents[0];
2711         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2712
2713         mutex_init(&fs_info->ordered_operations_mutex);
2714         mutex_init(&fs_info->tree_log_mutex);
2715         mutex_init(&fs_info->chunk_mutex);
2716         mutex_init(&fs_info->transaction_kthread_mutex);
2717         mutex_init(&fs_info->cleaner_mutex);
2718         mutex_init(&fs_info->volume_mutex);
2719         mutex_init(&fs_info->ro_block_group_mutex);
2720         init_rwsem(&fs_info->commit_root_sem);
2721         init_rwsem(&fs_info->cleanup_work_sem);
2722         init_rwsem(&fs_info->subvol_sem);
2723         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2724
2725         btrfs_init_dev_replace_locks(fs_info);
2726         btrfs_init_qgroup(fs_info);
2727
2728         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2729         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2730
2731         init_waitqueue_head(&fs_info->transaction_throttle);
2732         init_waitqueue_head(&fs_info->transaction_wait);
2733         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2734         init_waitqueue_head(&fs_info->async_submit_wait);
2735
2736         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2737
2738         /* Usable values until the real ones are cached from the superblock */
2739         fs_info->nodesize = 4096;
2740         fs_info->sectorsize = 4096;
2741         fs_info->stripesize = 4096;
2742
2743         ret = btrfs_alloc_stripe_hash_table(fs_info);
2744         if (ret) {
2745                 err = ret;
2746                 goto fail_alloc;
2747         }
2748
2749         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2750
2751         invalidate_bdev(fs_devices->latest_bdev);
2752
2753         /*
2754          * Read super block and check the signature bytes only
2755          */
2756         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2757         if (IS_ERR(bh)) {
2758                 err = PTR_ERR(bh);
2759                 goto fail_alloc;
2760         }
2761
2762         /*
2763          * We want to check superblock checksum, the type is stored inside.
2764          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2765          */
2766         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2767                 btrfs_err(fs_info, "superblock checksum mismatch");
2768                 err = -EINVAL;
2769                 brelse(bh);
2770                 goto fail_alloc;
2771         }
2772
2773         /*
2774          * super_copy is zeroed at allocation time and we never touch the
2775          * following bytes up to INFO_SIZE, the checksum is calculated from
2776          * the whole block of INFO_SIZE
2777          */
2778         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2779         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2780                sizeof(*fs_info->super_for_commit));
2781         brelse(bh);
2782
2783         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2784
2785         ret = btrfs_check_super_valid(fs_info);
2786         if (ret) {
2787                 btrfs_err(fs_info, "superblock contains fatal errors");
2788                 err = -EINVAL;
2789                 goto fail_alloc;
2790         }
2791
2792         disk_super = fs_info->super_copy;
2793         if (!btrfs_super_root(disk_super))
2794                 goto fail_alloc;
2795
2796         /* check FS state, whether FS is broken. */
2797         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2798                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2799
2800         /*
2801          * run through our array of backup supers and setup
2802          * our ring pointer to the oldest one
2803          */
2804         generation = btrfs_super_generation(disk_super);
2805         find_oldest_super_backup(fs_info, generation);
2806
2807         /*
2808          * In the long term, we'll store the compression type in the super
2809          * block, and it'll be used for per file compression control.
2810          */
2811         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2812
2813         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2814         if (ret) {
2815                 err = ret;
2816                 goto fail_alloc;
2817         }
2818
2819         features = btrfs_super_incompat_flags(disk_super) &
2820                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2821         if (features) {
2822                 btrfs_err(fs_info,
2823                     "cannot mount because of unsupported optional features (%llx)",
2824                     features);
2825                 err = -EINVAL;
2826                 goto fail_alloc;
2827         }
2828
2829         features = btrfs_super_incompat_flags(disk_super);
2830         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2831         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2832                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2833
2834         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2835                 btrfs_info(fs_info, "has skinny extents");
2836
2837         /*
2838          * flag our filesystem as having big metadata blocks if
2839          * they are bigger than the page size
2840          */
2841         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2842                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2843                         btrfs_info(fs_info,
2844                                 "flagging fs with big metadata feature");
2845                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2846         }
2847
2848         nodesize = btrfs_super_nodesize(disk_super);
2849         sectorsize = btrfs_super_sectorsize(disk_super);
2850         stripesize = sectorsize;
2851         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2852         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2853
2854         /* Cache block sizes */
2855         fs_info->nodesize = nodesize;
2856         fs_info->sectorsize = sectorsize;
2857         fs_info->stripesize = stripesize;
2858
2859         /*
2860          * mixed block groups end up with duplicate but slightly offset
2861          * extent buffers for the same range.  It leads to corruptions
2862          */
2863         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2864             (sectorsize != nodesize)) {
2865                 btrfs_err(fs_info,
2866 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2867                         nodesize, sectorsize);
2868                 goto fail_alloc;
2869         }
2870
2871         /*
2872          * Needn't use the lock because there is no other task which will
2873          * update the flag.
2874          */
2875         btrfs_set_super_incompat_flags(disk_super, features);
2876
2877         features = btrfs_super_compat_ro_flags(disk_super) &
2878                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2879         if (!(sb->s_flags & MS_RDONLY) && features) {
2880                 btrfs_err(fs_info,
2881         "cannot mount read-write because of unsupported optional features (%llx)",
2882                        features);
2883                 err = -EINVAL;
2884                 goto fail_alloc;
2885         }
2886
2887         max_active = fs_info->thread_pool_size;
2888
2889         ret = btrfs_init_workqueues(fs_info, fs_devices);
2890         if (ret) {
2891                 err = ret;
2892                 goto fail_sb_buffer;
2893         }
2894
2895         sb->s_bdi->congested_fn = btrfs_congested_fn;
2896         sb->s_bdi->congested_data = fs_info;
2897         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2898         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
2899         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2900         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2901
2902         sb->s_blocksize = sectorsize;
2903         sb->s_blocksize_bits = blksize_bits(sectorsize);
2904
2905         mutex_lock(&fs_info->chunk_mutex);
2906         ret = btrfs_read_sys_array(fs_info);
2907         mutex_unlock(&fs_info->chunk_mutex);
2908         if (ret) {
2909                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2910                 goto fail_sb_buffer;
2911         }
2912
2913         generation = btrfs_super_chunk_root_generation(disk_super);
2914
2915         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2916
2917         chunk_root->node = read_tree_block(fs_info,
2918                                            btrfs_super_chunk_root(disk_super),
2919                                            generation);
2920         if (IS_ERR(chunk_root->node) ||
2921             !extent_buffer_uptodate(chunk_root->node)) {
2922                 btrfs_err(fs_info, "failed to read chunk root");
2923                 if (!IS_ERR(chunk_root->node))
2924                         free_extent_buffer(chunk_root->node);
2925                 chunk_root->node = NULL;
2926                 goto fail_tree_roots;
2927         }
2928         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2929         chunk_root->commit_root = btrfs_root_node(chunk_root);
2930
2931         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2932            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2933
2934         ret = btrfs_read_chunk_tree(fs_info);
2935         if (ret) {
2936                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2937                 goto fail_tree_roots;
2938         }
2939
2940         /*
2941          * keep the device that is marked to be the target device for the
2942          * dev_replace procedure
2943          */
2944         btrfs_close_extra_devices(fs_devices, 0);
2945
2946         if (!fs_devices->latest_bdev) {
2947                 btrfs_err(fs_info, "failed to read devices");
2948                 goto fail_tree_roots;
2949         }
2950
2951 retry_root_backup:
2952         generation = btrfs_super_generation(disk_super);
2953
2954         tree_root->node = read_tree_block(fs_info,
2955                                           btrfs_super_root(disk_super),
2956                                           generation);
2957         if (IS_ERR(tree_root->node) ||
2958             !extent_buffer_uptodate(tree_root->node)) {
2959                 btrfs_warn(fs_info, "failed to read tree root");
2960                 if (!IS_ERR(tree_root->node))
2961                         free_extent_buffer(tree_root->node);
2962                 tree_root->node = NULL;
2963                 goto recovery_tree_root;
2964         }
2965
2966         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2967         tree_root->commit_root = btrfs_root_node(tree_root);
2968         btrfs_set_root_refs(&tree_root->root_item, 1);
2969
2970         mutex_lock(&tree_root->objectid_mutex);
2971         ret = btrfs_find_highest_objectid(tree_root,
2972                                         &tree_root->highest_objectid);
2973         if (ret) {
2974                 mutex_unlock(&tree_root->objectid_mutex);
2975                 goto recovery_tree_root;
2976         }
2977
2978         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2979
2980         mutex_unlock(&tree_root->objectid_mutex);
2981
2982         ret = btrfs_read_roots(fs_info);
2983         if (ret)
2984                 goto recovery_tree_root;
2985
2986         fs_info->generation = generation;
2987         fs_info->last_trans_committed = generation;
2988
2989         ret = btrfs_recover_balance(fs_info);
2990         if (ret) {
2991                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2992                 goto fail_block_groups;
2993         }
2994
2995         ret = btrfs_init_dev_stats(fs_info);
2996         if (ret) {
2997                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2998                 goto fail_block_groups;
2999         }
3000
3001         ret = btrfs_init_dev_replace(fs_info);
3002         if (ret) {
3003                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3004                 goto fail_block_groups;
3005         }
3006
3007         btrfs_close_extra_devices(fs_devices, 1);
3008
3009         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3010         if (ret) {
3011                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3012                                 ret);
3013                 goto fail_block_groups;
3014         }
3015
3016         ret = btrfs_sysfs_add_device(fs_devices);
3017         if (ret) {
3018                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3019                                 ret);
3020                 goto fail_fsdev_sysfs;
3021         }
3022
3023         ret = btrfs_sysfs_add_mounted(fs_info);
3024         if (ret) {
3025                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3026                 goto fail_fsdev_sysfs;
3027         }
3028
3029         ret = btrfs_init_space_info(fs_info);
3030         if (ret) {
3031                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3032                 goto fail_sysfs;
3033         }
3034
3035         ret = btrfs_read_block_groups(fs_info);
3036         if (ret) {
3037                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3038                 goto fail_sysfs;
3039         }
3040         fs_info->num_tolerated_disk_barrier_failures =
3041                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3042         if (fs_info->fs_devices->missing_devices >
3043              fs_info->num_tolerated_disk_barrier_failures &&
3044             !(sb->s_flags & MS_RDONLY)) {
3045                 btrfs_warn(fs_info,
3046 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3047                         fs_info->fs_devices->missing_devices,
3048                         fs_info->num_tolerated_disk_barrier_failures);
3049                 goto fail_sysfs;
3050         }
3051
3052         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3053                                                "btrfs-cleaner");
3054         if (IS_ERR(fs_info->cleaner_kthread))
3055                 goto fail_sysfs;
3056
3057         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3058                                                    tree_root,
3059                                                    "btrfs-transaction");
3060         if (IS_ERR(fs_info->transaction_kthread))
3061                 goto fail_cleaner;
3062
3063         if (!btrfs_test_opt(fs_info, SSD) &&
3064             !btrfs_test_opt(fs_info, NOSSD) &&
3065             !fs_info->fs_devices->rotating) {
3066                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3067                 btrfs_set_opt(fs_info->mount_opt, SSD);
3068         }
3069
3070         /*
3071          * Mount does not set all options immediately, we can do it now and do
3072          * not have to wait for transaction commit
3073          */
3074         btrfs_apply_pending_changes(fs_info);
3075
3076 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3077         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3078                 ret = btrfsic_mount(fs_info, fs_devices,
3079                                     btrfs_test_opt(fs_info,
3080                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3081                                     1 : 0,
3082                                     fs_info->check_integrity_print_mask);
3083                 if (ret)
3084                         btrfs_warn(fs_info,
3085                                 "failed to initialize integrity check module: %d",
3086                                 ret);
3087         }
3088 #endif
3089         ret = btrfs_read_qgroup_config(fs_info);
3090         if (ret)
3091                 goto fail_trans_kthread;
3092
3093         /* do not make disk changes in broken FS or nologreplay is given */
3094         if (btrfs_super_log_root(disk_super) != 0 &&
3095             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3096                 ret = btrfs_replay_log(fs_info, fs_devices);
3097                 if (ret) {
3098                         err = ret;
3099                         goto fail_qgroup;
3100                 }
3101         }
3102
3103         ret = btrfs_find_orphan_roots(fs_info);
3104         if (ret)
3105                 goto fail_qgroup;
3106
3107         if (!(sb->s_flags & MS_RDONLY)) {
3108                 ret = btrfs_cleanup_fs_roots(fs_info);
3109                 if (ret)
3110                         goto fail_qgroup;
3111
3112                 mutex_lock(&fs_info->cleaner_mutex);
3113                 ret = btrfs_recover_relocation(tree_root);
3114                 mutex_unlock(&fs_info->cleaner_mutex);
3115                 if (ret < 0) {
3116                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3117                                         ret);
3118                         err = -EINVAL;
3119                         goto fail_qgroup;
3120                 }
3121         }
3122
3123         location.objectid = BTRFS_FS_TREE_OBJECTID;
3124         location.type = BTRFS_ROOT_ITEM_KEY;
3125         location.offset = 0;
3126
3127         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3128         if (IS_ERR(fs_info->fs_root)) {
3129                 err = PTR_ERR(fs_info->fs_root);
3130                 goto fail_qgroup;
3131         }
3132
3133         if (sb->s_flags & MS_RDONLY)
3134                 return 0;
3135
3136         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3137             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3138                 clear_free_space_tree = 1;
3139         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3140                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3141                 btrfs_warn(fs_info, "free space tree is invalid");
3142                 clear_free_space_tree = 1;
3143         }
3144
3145         if (clear_free_space_tree) {
3146                 btrfs_info(fs_info, "clearing free space tree");
3147                 ret = btrfs_clear_free_space_tree(fs_info);
3148                 if (ret) {
3149                         btrfs_warn(fs_info,
3150                                    "failed to clear free space tree: %d", ret);
3151                         close_ctree(fs_info);
3152                         return ret;
3153                 }
3154         }
3155
3156         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3157             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3158                 btrfs_info(fs_info, "creating free space tree");
3159                 ret = btrfs_create_free_space_tree(fs_info);
3160                 if (ret) {
3161                         btrfs_warn(fs_info,
3162                                 "failed to create free space tree: %d", ret);
3163                         close_ctree(fs_info);
3164                         return ret;
3165                 }
3166         }
3167
3168         down_read(&fs_info->cleanup_work_sem);
3169         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3170             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3171                 up_read(&fs_info->cleanup_work_sem);
3172                 close_ctree(fs_info);
3173                 return ret;
3174         }
3175         up_read(&fs_info->cleanup_work_sem);
3176
3177         ret = btrfs_resume_balance_async(fs_info);
3178         if (ret) {
3179                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3180                 close_ctree(fs_info);
3181                 return ret;
3182         }
3183
3184         ret = btrfs_resume_dev_replace_async(fs_info);
3185         if (ret) {
3186                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3187                 close_ctree(fs_info);
3188                 return ret;
3189         }
3190
3191         btrfs_qgroup_rescan_resume(fs_info);
3192
3193         if (!fs_info->uuid_root) {
3194                 btrfs_info(fs_info, "creating UUID tree");
3195                 ret = btrfs_create_uuid_tree(fs_info);
3196                 if (ret) {
3197                         btrfs_warn(fs_info,
3198                                 "failed to create the UUID tree: %d", ret);
3199                         close_ctree(fs_info);
3200                         return ret;
3201                 }
3202         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3203                    fs_info->generation !=
3204                                 btrfs_super_uuid_tree_generation(disk_super)) {
3205                 btrfs_info(fs_info, "checking UUID tree");
3206                 ret = btrfs_check_uuid_tree(fs_info);
3207                 if (ret) {
3208                         btrfs_warn(fs_info,
3209                                 "failed to check the UUID tree: %d", ret);
3210                         close_ctree(fs_info);
3211                         return ret;
3212                 }
3213         } else {
3214                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3215         }
3216         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3217
3218         /*
3219          * backuproot only affect mount behavior, and if open_ctree succeeded,
3220          * no need to keep the flag
3221          */
3222         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3223
3224         return 0;
3225
3226 fail_qgroup:
3227         btrfs_free_qgroup_config(fs_info);
3228 fail_trans_kthread:
3229         kthread_stop(fs_info->transaction_kthread);
3230         btrfs_cleanup_transaction(fs_info);
3231         btrfs_free_fs_roots(fs_info);
3232 fail_cleaner:
3233         kthread_stop(fs_info->cleaner_kthread);
3234
3235         /*
3236          * make sure we're done with the btree inode before we stop our
3237          * kthreads
3238          */
3239         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3240
3241 fail_sysfs:
3242         btrfs_sysfs_remove_mounted(fs_info);
3243
3244 fail_fsdev_sysfs:
3245         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3246
3247 fail_block_groups:
3248         btrfs_put_block_group_cache(fs_info);
3249
3250 fail_tree_roots:
3251         free_root_pointers(fs_info, 1);
3252         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3253
3254 fail_sb_buffer:
3255         btrfs_stop_all_workers(fs_info);
3256         btrfs_free_block_groups(fs_info);
3257 fail_alloc:
3258 fail_iput:
3259         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3260
3261         iput(fs_info->btree_inode);
3262 fail_bio_counter:
3263         percpu_counter_destroy(&fs_info->bio_counter);
3264 fail_delalloc_bytes:
3265         percpu_counter_destroy(&fs_info->delalloc_bytes);
3266 fail_dirty_metadata_bytes:
3267         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3268 fail_srcu:
3269         cleanup_srcu_struct(&fs_info->subvol_srcu);
3270 fail:
3271         btrfs_free_stripe_hash_table(fs_info);
3272         btrfs_close_devices(fs_info->fs_devices);
3273         return err;
3274
3275 recovery_tree_root:
3276         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3277                 goto fail_tree_roots;
3278
3279         free_root_pointers(fs_info, 0);
3280
3281         /* don't use the log in recovery mode, it won't be valid */
3282         btrfs_set_super_log_root(disk_super, 0);
3283
3284         /* we can't trust the free space cache either */
3285         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3286
3287         ret = next_root_backup(fs_info, fs_info->super_copy,
3288                                &num_backups_tried, &backup_index);
3289         if (ret == -1)
3290                 goto fail_block_groups;
3291         goto retry_root_backup;
3292 }
3293
3294 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3295 {
3296         if (uptodate) {
3297                 set_buffer_uptodate(bh);
3298         } else {
3299                 struct btrfs_device *device = (struct btrfs_device *)
3300                         bh->b_private;
3301
3302                 btrfs_warn_rl_in_rcu(device->fs_info,
3303                                 "lost page write due to IO error on %s",
3304                                           rcu_str_deref(device->name));
3305                 /* note, we don't set_buffer_write_io_error because we have
3306                  * our own ways of dealing with the IO errors
3307                  */
3308                 clear_buffer_uptodate(bh);
3309                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3310         }
3311         unlock_buffer(bh);
3312         put_bh(bh);
3313 }
3314
3315 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3316                         struct buffer_head **bh_ret)
3317 {
3318         struct buffer_head *bh;
3319         struct btrfs_super_block *super;
3320         u64 bytenr;
3321
3322         bytenr = btrfs_sb_offset(copy_num);
3323         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3324                 return -EINVAL;
3325
3326         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3327         /*
3328          * If we fail to read from the underlying devices, as of now
3329          * the best option we have is to mark it EIO.
3330          */
3331         if (!bh)
3332                 return -EIO;
3333
3334         super = (struct btrfs_super_block *)bh->b_data;
3335         if (btrfs_super_bytenr(super) != bytenr ||
3336                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3337                 brelse(bh);
3338                 return -EINVAL;
3339         }
3340
3341         *bh_ret = bh;
3342         return 0;
3343 }
3344
3345
3346 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3347 {
3348         struct buffer_head *bh;
3349         struct buffer_head *latest = NULL;
3350         struct btrfs_super_block *super;
3351         int i;
3352         u64 transid = 0;
3353         int ret = -EINVAL;
3354
3355         /* we would like to check all the supers, but that would make
3356          * a btrfs mount succeed after a mkfs from a different FS.
3357          * So, we need to add a special mount option to scan for
3358          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3359          */
3360         for (i = 0; i < 1; i++) {
3361                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3362                 if (ret)
3363                         continue;
3364
3365                 super = (struct btrfs_super_block *)bh->b_data;
3366
3367                 if (!latest || btrfs_super_generation(super) > transid) {
3368                         brelse(latest);
3369                         latest = bh;
3370                         transid = btrfs_super_generation(super);
3371                 } else {
3372                         brelse(bh);
3373                 }
3374         }
3375
3376         if (!latest)
3377                 return ERR_PTR(ret);
3378
3379         return latest;
3380 }
3381
3382 /*
3383  * this should be called twice, once with wait == 0 and
3384  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3385  * we write are pinned.
3386  *
3387  * They are released when wait == 1 is done.
3388  * max_mirrors must be the same for both runs, and it indicates how
3389  * many supers on this one device should be written.
3390  *
3391  * max_mirrors == 0 means to write them all.
3392  */
3393 static int write_dev_supers(struct btrfs_device *device,
3394                             struct btrfs_super_block *sb,
3395                             int wait, int max_mirrors)
3396 {
3397         struct buffer_head *bh;
3398         int i;
3399         int ret;
3400         int errors = 0;
3401         u32 crc;
3402         u64 bytenr;
3403
3404         if (max_mirrors == 0)
3405                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3406
3407         for (i = 0; i < max_mirrors; i++) {
3408                 bytenr = btrfs_sb_offset(i);
3409                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3410                     device->commit_total_bytes)
3411                         break;
3412
3413                 if (wait) {
3414                         bh = __find_get_block(device->bdev, bytenr / 4096,
3415                                               BTRFS_SUPER_INFO_SIZE);
3416                         if (!bh) {
3417                                 errors++;
3418                                 continue;
3419                         }
3420                         wait_on_buffer(bh);
3421                         if (!buffer_uptodate(bh))
3422                                 errors++;
3423
3424                         /* drop our reference */
3425                         brelse(bh);
3426
3427                         /* drop the reference from the wait == 0 run */
3428                         brelse(bh);
3429                         continue;
3430                 } else {
3431                         btrfs_set_super_bytenr(sb, bytenr);
3432
3433                         crc = ~(u32)0;
3434                         crc = btrfs_csum_data((const char *)sb +
3435                                               BTRFS_CSUM_SIZE, crc,
3436                                               BTRFS_SUPER_INFO_SIZE -
3437                                               BTRFS_CSUM_SIZE);
3438                         btrfs_csum_final(crc, sb->csum);
3439
3440                         /*
3441                          * one reference for us, and we leave it for the
3442                          * caller
3443                          */
3444                         bh = __getblk(device->bdev, bytenr / 4096,
3445                                       BTRFS_SUPER_INFO_SIZE);
3446                         if (!bh) {
3447                                 btrfs_err(device->fs_info,
3448                                     "couldn't get super buffer head for bytenr %llu",
3449                                     bytenr);
3450                                 errors++;
3451                                 continue;
3452                         }
3453
3454                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3455
3456                         /* one reference for submit_bh */
3457                         get_bh(bh);
3458
3459                         set_buffer_uptodate(bh);
3460                         lock_buffer(bh);
3461                         bh->b_end_io = btrfs_end_buffer_write_sync;
3462                         bh->b_private = device;
3463                 }
3464
3465                 /*
3466                  * we fua the first super.  The others we allow
3467                  * to go down lazy.
3468                  */
3469                 if (i == 0)
3470                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_FUA, bh);
3471                 else
3472                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3473                 if (ret)
3474                         errors++;
3475         }
3476         return errors < i ? 0 : -1;
3477 }
3478
3479 /*
3480  * endio for the write_dev_flush, this will wake anyone waiting
3481  * for the barrier when it is done
3482  */
3483 static void btrfs_end_empty_barrier(struct bio *bio)
3484 {
3485         if (bio->bi_private)
3486                 complete(bio->bi_private);
3487         bio_put(bio);
3488 }
3489
3490 /*
3491  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3492  * sent down.  With wait == 1, it waits for the previous flush.
3493  *
3494  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3495  * capable
3496  */
3497 static blk_status_t write_dev_flush(struct btrfs_device *device, int wait)
3498 {
3499         struct request_queue *q = bdev_get_queue(device->bdev);
3500         struct bio *bio;
3501         blk_status_t ret = 0;
3502
3503         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3504                 return 0;
3505
3506         if (wait) {
3507                 bio = device->flush_bio;
3508                 if (!bio)
3509                         return 0;
3510
3511                 wait_for_completion(&device->flush_wait);
3512
3513                 if (bio->bi_status) {
3514                         ret = bio->bi_status;
3515                         btrfs_dev_stat_inc_and_print(device,
3516                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3517                 }
3518
3519                 /* drop the reference from the wait == 0 run */
3520                 bio_put(bio);
3521                 device->flush_bio = NULL;
3522
3523                 return ret;
3524         }
3525
3526         /*
3527          * one reference for us, and we leave it for the
3528          * caller
3529          */
3530         device->flush_bio = NULL;
3531         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3532         if (!bio)
3533                 return BLK_STS_RESOURCE;
3534
3535         bio->bi_end_io = btrfs_end_empty_barrier;
3536         bio->bi_bdev = device->bdev;
3537         bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3538         init_completion(&device->flush_wait);
3539         bio->bi_private = &device->flush_wait;
3540         device->flush_bio = bio;
3541
3542         bio_get(bio);
3543         btrfsic_submit_bio(bio);
3544
3545         return 0;
3546 }
3547
3548 /*
3549  * send an empty flush down to each device in parallel,
3550  * then wait for them
3551  */
3552 static int barrier_all_devices(struct btrfs_fs_info *info)
3553 {
3554         struct list_head *head;
3555         struct btrfs_device *dev;
3556         int errors_send = 0;
3557         int errors_wait = 0;
3558         blk_status_t ret;
3559
3560         /* send down all the barriers */
3561         head = &info->fs_devices->devices;
3562         list_for_each_entry_rcu(dev, head, dev_list) {
3563                 if (dev->missing)
3564                         continue;
3565                 if (!dev->bdev) {
3566                         errors_send++;
3567                         continue;
3568                 }
3569                 if (!dev->in_fs_metadata || !dev->writeable)
3570                         continue;
3571
3572                 ret = write_dev_flush(dev, 0);
3573                 if (ret)
3574                         errors_send++;
3575         }
3576
3577         /* wait for all the barriers */
3578         list_for_each_entry_rcu(dev, head, dev_list) {
3579                 if (dev->missing)
3580                         continue;
3581                 if (!dev->bdev) {
3582                         errors_wait++;
3583                         continue;
3584                 }
3585                 if (!dev->in_fs_metadata || !dev->writeable)
3586                         continue;
3587
3588                 ret = write_dev_flush(dev, 1);
3589                 if (ret)
3590                         errors_wait++;
3591         }
3592         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3593             errors_wait > info->num_tolerated_disk_barrier_failures)
3594                 return -EIO;
3595         return 0;
3596 }
3597
3598 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3599 {
3600         int raid_type;
3601         int min_tolerated = INT_MAX;
3602
3603         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3604             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3605                 min_tolerated = min(min_tolerated,
3606                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3607                                     tolerated_failures);
3608
3609         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3610                 if (raid_type == BTRFS_RAID_SINGLE)
3611                         continue;
3612                 if (!(flags & btrfs_raid_group[raid_type]))
3613                         continue;
3614                 min_tolerated = min(min_tolerated,
3615                                     btrfs_raid_array[raid_type].
3616                                     tolerated_failures);
3617         }
3618
3619         if (min_tolerated == INT_MAX) {
3620                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3621                 min_tolerated = 0;
3622         }
3623
3624         return min_tolerated;
3625 }
3626
3627 int btrfs_calc_num_tolerated_disk_barrier_failures(
3628         struct btrfs_fs_info *fs_info)
3629 {
3630         struct btrfs_ioctl_space_info space;
3631         struct btrfs_space_info *sinfo;
3632         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3633                        BTRFS_BLOCK_GROUP_SYSTEM,
3634                        BTRFS_BLOCK_GROUP_METADATA,
3635                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3636         int i;
3637         int c;
3638         int num_tolerated_disk_barrier_failures =
3639                 (int)fs_info->fs_devices->num_devices;
3640
3641         for (i = 0; i < ARRAY_SIZE(types); i++) {
3642                 struct btrfs_space_info *tmp;
3643
3644                 sinfo = NULL;
3645                 rcu_read_lock();
3646                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3647                         if (tmp->flags == types[i]) {
3648                                 sinfo = tmp;
3649                                 break;
3650                         }
3651                 }
3652                 rcu_read_unlock();
3653
3654                 if (!sinfo)
3655                         continue;
3656
3657                 down_read(&sinfo->groups_sem);
3658                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3659                         u64 flags;
3660
3661                         if (list_empty(&sinfo->block_groups[c]))
3662                                 continue;
3663
3664                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3665                                                    &space);
3666                         if (space.total_bytes == 0 || space.used_bytes == 0)
3667                                 continue;
3668                         flags = space.flags;
3669
3670                         num_tolerated_disk_barrier_failures = min(
3671                                 num_tolerated_disk_barrier_failures,
3672                                 btrfs_get_num_tolerated_disk_barrier_failures(
3673                                         flags));
3674                 }
3675                 up_read(&sinfo->groups_sem);
3676         }
3677
3678         return num_tolerated_disk_barrier_failures;
3679 }
3680
3681 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3682 {
3683         struct list_head *head;
3684         struct btrfs_device *dev;
3685         struct btrfs_super_block *sb;
3686         struct btrfs_dev_item *dev_item;
3687         int ret;
3688         int do_barriers;
3689         int max_errors;
3690         int total_errors = 0;
3691         u64 flags;
3692
3693         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3694         backup_super_roots(fs_info);
3695
3696         sb = fs_info->super_for_commit;
3697         dev_item = &sb->dev_item;
3698
3699         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3700         head = &fs_info->fs_devices->devices;
3701         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3702
3703         if (do_barriers) {
3704                 ret = barrier_all_devices(fs_info);
3705                 if (ret) {
3706                         mutex_unlock(
3707                                 &fs_info->fs_devices->device_list_mutex);
3708                         btrfs_handle_fs_error(fs_info, ret,
3709                                               "errors while submitting device barriers.");
3710                         return ret;
3711                 }
3712         }
3713
3714         list_for_each_entry_rcu(dev, head, dev_list) {
3715                 if (!dev->bdev) {
3716                         total_errors++;
3717                         continue;
3718                 }
3719                 if (!dev->in_fs_metadata || !dev->writeable)
3720                         continue;
3721
3722                 btrfs_set_stack_device_generation(dev_item, 0);
3723                 btrfs_set_stack_device_type(dev_item, dev->type);
3724                 btrfs_set_stack_device_id(dev_item, dev->devid);
3725                 btrfs_set_stack_device_total_bytes(dev_item,
3726                                                    dev->commit_total_bytes);
3727                 btrfs_set_stack_device_bytes_used(dev_item,
3728                                                   dev->commit_bytes_used);
3729                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3730                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3731                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3732                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3733                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3734
3735                 flags = btrfs_super_flags(sb);
3736                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3737
3738                 ret = write_dev_supers(dev, sb, 0, max_mirrors);
3739                 if (ret)
3740                         total_errors++;
3741         }
3742         if (total_errors > max_errors) {
3743                 btrfs_err(fs_info, "%d errors while writing supers",
3744                           total_errors);
3745                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3746
3747                 /* FUA is masked off if unsupported and can't be the reason */
3748                 btrfs_handle_fs_error(fs_info, -EIO,
3749                                       "%d errors while writing supers",
3750                                       total_errors);
3751                 return -EIO;
3752         }
3753
3754         total_errors = 0;
3755         list_for_each_entry_rcu(dev, head, dev_list) {
3756                 if (!dev->bdev)
3757                         continue;
3758                 if (!dev->in_fs_metadata || !dev->writeable)
3759                         continue;
3760
3761                 ret = write_dev_supers(dev, sb, 1, max_mirrors);
3762                 if (ret)
3763                         total_errors++;
3764         }
3765         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3766         if (total_errors > max_errors) {
3767                 btrfs_handle_fs_error(fs_info, -EIO,
3768                                       "%d errors while writing supers",
3769                                       total_errors);
3770                 return -EIO;
3771         }
3772         return 0;
3773 }
3774
3775 /* Drop a fs root from the radix tree and free it. */
3776 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3777                                   struct btrfs_root *root)
3778 {
3779         spin_lock(&fs_info->fs_roots_radix_lock);
3780         radix_tree_delete(&fs_info->fs_roots_radix,
3781                           (unsigned long)root->root_key.objectid);
3782         spin_unlock(&fs_info->fs_roots_radix_lock);
3783
3784         if (btrfs_root_refs(&root->root_item) == 0)
3785                 synchronize_srcu(&fs_info->subvol_srcu);
3786
3787         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3788                 btrfs_free_log(NULL, root);
3789                 if (root->reloc_root) {
3790                         free_extent_buffer(root->reloc_root->node);
3791                         free_extent_buffer(root->reloc_root->commit_root);
3792                         btrfs_put_fs_root(root->reloc_root);
3793                         root->reloc_root = NULL;
3794                 }
3795         }
3796
3797         if (root->free_ino_pinned)
3798                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3799         if (root->free_ino_ctl)
3800                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3801         free_fs_root(root);
3802 }
3803
3804 static void free_fs_root(struct btrfs_root *root)
3805 {
3806         iput(root->ino_cache_inode);
3807         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3808         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3809         root->orphan_block_rsv = NULL;
3810         if (root->anon_dev)
3811                 free_anon_bdev(root->anon_dev);
3812         if (root->subv_writers)
3813                 btrfs_free_subvolume_writers(root->subv_writers);
3814         free_extent_buffer(root->node);
3815         free_extent_buffer(root->commit_root);
3816         kfree(root->free_ino_ctl);
3817         kfree(root->free_ino_pinned);
3818         kfree(root->name);
3819         btrfs_put_fs_root(root);
3820 }
3821
3822 void btrfs_free_fs_root(struct btrfs_root *root)
3823 {
3824         free_fs_root(root);
3825 }
3826
3827 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3828 {
3829         u64 root_objectid = 0;
3830         struct btrfs_root *gang[8];
3831         int i = 0;
3832         int err = 0;
3833         unsigned int ret = 0;
3834         int index;
3835
3836         while (1) {
3837                 index = srcu_read_lock(&fs_info->subvol_srcu);
3838                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3839                                              (void **)gang, root_objectid,
3840                                              ARRAY_SIZE(gang));
3841                 if (!ret) {
3842                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3843                         break;
3844                 }
3845                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3846
3847                 for (i = 0; i < ret; i++) {
3848                         /* Avoid to grab roots in dead_roots */
3849                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3850                                 gang[i] = NULL;
3851                                 continue;
3852                         }
3853                         /* grab all the search result for later use */
3854                         gang[i] = btrfs_grab_fs_root(gang[i]);
3855                 }
3856                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3857
3858                 for (i = 0; i < ret; i++) {
3859                         if (!gang[i])
3860                                 continue;
3861                         root_objectid = gang[i]->root_key.objectid;
3862                         err = btrfs_orphan_cleanup(gang[i]);
3863                         if (err)
3864                                 break;
3865                         btrfs_put_fs_root(gang[i]);
3866                 }
3867                 root_objectid++;
3868         }
3869
3870         /* release the uncleaned roots due to error */
3871         for (; i < ret; i++) {
3872                 if (gang[i])
3873                         btrfs_put_fs_root(gang[i]);
3874         }
3875         return err;
3876 }
3877
3878 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3879 {
3880         struct btrfs_root *root = fs_info->tree_root;
3881         struct btrfs_trans_handle *trans;
3882
3883         mutex_lock(&fs_info->cleaner_mutex);
3884         btrfs_run_delayed_iputs(fs_info);
3885         mutex_unlock(&fs_info->cleaner_mutex);
3886         wake_up_process(fs_info->cleaner_kthread);
3887
3888         /* wait until ongoing cleanup work done */
3889         down_write(&fs_info->cleanup_work_sem);
3890         up_write(&fs_info->cleanup_work_sem);
3891
3892         trans = btrfs_join_transaction(root);
3893         if (IS_ERR(trans))
3894                 return PTR_ERR(trans);
3895         return btrfs_commit_transaction(trans);
3896 }
3897
3898 void close_ctree(struct btrfs_fs_info *fs_info)
3899 {
3900         struct btrfs_root *root = fs_info->tree_root;
3901         int ret;
3902
3903         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3904
3905         /* wait for the qgroup rescan worker to stop */
3906         btrfs_qgroup_wait_for_completion(fs_info, false);
3907
3908         /* wait for the uuid_scan task to finish */
3909         down(&fs_info->uuid_tree_rescan_sem);
3910         /* avoid complains from lockdep et al., set sem back to initial state */
3911         up(&fs_info->uuid_tree_rescan_sem);
3912
3913         /* pause restriper - we want to resume on mount */
3914         btrfs_pause_balance(fs_info);
3915
3916         btrfs_dev_replace_suspend_for_unmount(fs_info);
3917
3918         btrfs_scrub_cancel(fs_info);
3919
3920         /* wait for any defraggers to finish */
3921         wait_event(fs_info->transaction_wait,
3922                    (atomic_read(&fs_info->defrag_running) == 0));
3923
3924         /* clear out the rbtree of defraggable inodes */
3925         btrfs_cleanup_defrag_inodes(fs_info);
3926
3927         cancel_work_sync(&fs_info->async_reclaim_work);
3928
3929         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3930                 /*
3931                  * If the cleaner thread is stopped and there are
3932                  * block groups queued for removal, the deletion will be
3933                  * skipped when we quit the cleaner thread.
3934                  */
3935                 btrfs_delete_unused_bgs(fs_info);
3936
3937                 ret = btrfs_commit_super(fs_info);
3938                 if (ret)
3939                         btrfs_err(fs_info, "commit super ret %d", ret);
3940         }
3941
3942         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3943                 btrfs_error_commit_super(fs_info);
3944
3945         kthread_stop(fs_info->transaction_kthread);
3946         kthread_stop(fs_info->cleaner_kthread);
3947
3948         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3949
3950         btrfs_free_qgroup_config(fs_info);
3951
3952         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3953                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3954                        percpu_counter_sum(&fs_info->delalloc_bytes));
3955         }
3956
3957         btrfs_sysfs_remove_mounted(fs_info);
3958         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3959
3960         btrfs_free_fs_roots(fs_info);
3961
3962         btrfs_put_block_group_cache(fs_info);
3963
3964         /*
3965          * we must make sure there is not any read request to
3966          * submit after we stopping all workers.
3967          */
3968         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3969         btrfs_stop_all_workers(fs_info);
3970
3971         btrfs_free_block_groups(fs_info);
3972
3973         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3974         free_root_pointers(fs_info, 1);
3975
3976         iput(fs_info->btree_inode);
3977
3978 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3979         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
3980                 btrfsic_unmount(fs_info->fs_devices);
3981 #endif
3982
3983         btrfs_close_devices(fs_info->fs_devices);
3984         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3985
3986         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3987         percpu_counter_destroy(&fs_info->delalloc_bytes);
3988         percpu_counter_destroy(&fs_info->bio_counter);
3989         cleanup_srcu_struct(&fs_info->subvol_srcu);
3990
3991         btrfs_free_stripe_hash_table(fs_info);
3992
3993         __btrfs_free_block_rsv(root->orphan_block_rsv);
3994         root->orphan_block_rsv = NULL;
3995
3996         mutex_lock(&fs_info->chunk_mutex);
3997         while (!list_empty(&fs_info->pinned_chunks)) {
3998                 struct extent_map *em;
3999
4000                 em = list_first_entry(&fs_info->pinned_chunks,
4001                                       struct extent_map, list);
4002                 list_del_init(&em->list);
4003                 free_extent_map(em);
4004         }
4005         mutex_unlock(&fs_info->chunk_mutex);
4006 }
4007
4008 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4009                           int atomic)
4010 {
4011         int ret;
4012         struct inode *btree_inode = buf->pages[0]->mapping->host;
4013
4014         ret = extent_buffer_uptodate(buf);
4015         if (!ret)
4016                 return ret;
4017
4018         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4019                                     parent_transid, atomic);
4020         if (ret == -EAGAIN)
4021                 return ret;
4022         return !ret;
4023 }
4024
4025 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4026 {
4027         struct btrfs_fs_info *fs_info;
4028         struct btrfs_root *root;
4029         u64 transid = btrfs_header_generation(buf);
4030         int was_dirty;
4031
4032 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4033         /*
4034          * This is a fast path so only do this check if we have sanity tests
4035          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4036          * outside of the sanity tests.
4037          */
4038         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4039                 return;
4040 #endif
4041         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4042         fs_info = root->fs_info;
4043         btrfs_assert_tree_locked(buf);
4044         if (transid != fs_info->generation)
4045                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4046                         buf->start, transid, fs_info->generation);
4047         was_dirty = set_extent_buffer_dirty(buf);
4048         if (!was_dirty)
4049                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
4050                                      buf->len,
4051                                      fs_info->dirty_metadata_batch);
4052 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4053         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4054                 btrfs_print_leaf(fs_info, buf);
4055                 ASSERT(0);
4056         }
4057 #endif
4058 }
4059
4060 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4061                                         int flush_delayed)
4062 {
4063         /*
4064          * looks as though older kernels can get into trouble with
4065          * this code, they end up stuck in balance_dirty_pages forever
4066          */
4067         int ret;
4068
4069         if (current->flags & PF_MEMALLOC)
4070                 return;
4071
4072         if (flush_delayed)
4073                 btrfs_balance_delayed_items(fs_info);
4074
4075         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4076                                      BTRFS_DIRTY_METADATA_THRESH);
4077         if (ret > 0) {
4078                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4079         }
4080 }
4081
4082 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4083 {
4084         __btrfs_btree_balance_dirty(fs_info, 1);
4085 }
4086
4087 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4088 {
4089         __btrfs_btree_balance_dirty(fs_info, 0);
4090 }
4091
4092 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4093 {
4094         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4095         struct btrfs_fs_info *fs_info = root->fs_info;
4096
4097         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4098 }
4099
4100 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info)
4101 {
4102         struct btrfs_super_block *sb = fs_info->super_copy;
4103         u64 nodesize = btrfs_super_nodesize(sb);
4104         u64 sectorsize = btrfs_super_sectorsize(sb);
4105         int ret = 0;
4106
4107         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4108                 btrfs_err(fs_info, "no valid FS found");
4109                 ret = -EINVAL;
4110         }
4111         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4112                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4113                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4114         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4115                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4116                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4117                 ret = -EINVAL;
4118         }
4119         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4120                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4121                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4122                 ret = -EINVAL;
4123         }
4124         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4125                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4126                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4127                 ret = -EINVAL;
4128         }
4129
4130         /*
4131          * Check sectorsize and nodesize first, other check will need it.
4132          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4133          */
4134         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4135             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4136                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4137                 ret = -EINVAL;
4138         }
4139         /* Only PAGE SIZE is supported yet */
4140         if (sectorsize != PAGE_SIZE) {
4141                 btrfs_err(fs_info,
4142                         "sectorsize %llu not supported yet, only support %lu",
4143                         sectorsize, PAGE_SIZE);
4144                 ret = -EINVAL;
4145         }
4146         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4147             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4148                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4149                 ret = -EINVAL;
4150         }
4151         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4152                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4153                           le32_to_cpu(sb->__unused_leafsize), nodesize);
4154                 ret = -EINVAL;
4155         }
4156
4157         /* Root alignment check */
4158         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4159                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4160                            btrfs_super_root(sb));
4161                 ret = -EINVAL;
4162         }
4163         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4164                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4165                            btrfs_super_chunk_root(sb));
4166                 ret = -EINVAL;
4167         }
4168         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4169                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4170                            btrfs_super_log_root(sb));
4171                 ret = -EINVAL;
4172         }
4173
4174         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4175                 btrfs_err(fs_info,
4176                            "dev_item UUID does not match fsid: %pU != %pU",
4177                            fs_info->fsid, sb->dev_item.fsid);
4178                 ret = -EINVAL;
4179         }
4180
4181         /*
4182          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4183          * done later
4184          */
4185         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4186                 btrfs_err(fs_info, "bytes_used is too small %llu",
4187                           btrfs_super_bytes_used(sb));
4188                 ret = -EINVAL;
4189         }
4190         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4191                 btrfs_err(fs_info, "invalid stripesize %u",
4192                           btrfs_super_stripesize(sb));
4193                 ret = -EINVAL;
4194         }
4195         if (btrfs_super_num_devices(sb) > (1UL << 31))
4196                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4197                            btrfs_super_num_devices(sb));
4198         if (btrfs_super_num_devices(sb) == 0) {
4199                 btrfs_err(fs_info, "number of devices is 0");
4200                 ret = -EINVAL;
4201         }
4202
4203         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4204                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4205                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4206                 ret = -EINVAL;
4207         }
4208
4209         /*
4210          * Obvious sys_chunk_array corruptions, it must hold at least one key
4211          * and one chunk
4212          */
4213         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4214                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4215                           btrfs_super_sys_array_size(sb),
4216                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4217                 ret = -EINVAL;
4218         }
4219         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4220                         + sizeof(struct btrfs_chunk)) {
4221                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4222                           btrfs_super_sys_array_size(sb),
4223                           sizeof(struct btrfs_disk_key)
4224                           + sizeof(struct btrfs_chunk));
4225                 ret = -EINVAL;
4226         }
4227
4228         /*
4229          * The generation is a global counter, we'll trust it more than the others
4230          * but it's still possible that it's the one that's wrong.
4231          */
4232         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4233                 btrfs_warn(fs_info,
4234                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4235                         btrfs_super_generation(sb),
4236                         btrfs_super_chunk_root_generation(sb));
4237         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4238             && btrfs_super_cache_generation(sb) != (u64)-1)
4239                 btrfs_warn(fs_info,
4240                         "suspicious: generation < cache_generation: %llu < %llu",
4241                         btrfs_super_generation(sb),
4242                         btrfs_super_cache_generation(sb));
4243
4244         return ret;
4245 }
4246
4247 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4248 {
4249         mutex_lock(&fs_info->cleaner_mutex);
4250         btrfs_run_delayed_iputs(fs_info);
4251         mutex_unlock(&fs_info->cleaner_mutex);
4252
4253         down_write(&fs_info->cleanup_work_sem);
4254         up_write(&fs_info->cleanup_work_sem);
4255
4256         /* cleanup FS via transaction */
4257         btrfs_cleanup_transaction(fs_info);
4258 }
4259
4260 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4261 {
4262         struct btrfs_ordered_extent *ordered;
4263
4264         spin_lock(&root->ordered_extent_lock);
4265         /*
4266          * This will just short circuit the ordered completion stuff which will
4267          * make sure the ordered extent gets properly cleaned up.
4268          */
4269         list_for_each_entry(ordered, &root->ordered_extents,
4270                             root_extent_list)
4271                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4272         spin_unlock(&root->ordered_extent_lock);
4273 }
4274
4275 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4276 {
4277         struct btrfs_root *root;
4278         struct list_head splice;
4279
4280         INIT_LIST_HEAD(&splice);
4281
4282         spin_lock(&fs_info->ordered_root_lock);
4283         list_splice_init(&fs_info->ordered_roots, &splice);
4284         while (!list_empty(&splice)) {
4285                 root = list_first_entry(&splice, struct btrfs_root,
4286                                         ordered_root);
4287                 list_move_tail(&root->ordered_root,
4288                                &fs_info->ordered_roots);
4289
4290                 spin_unlock(&fs_info->ordered_root_lock);
4291                 btrfs_destroy_ordered_extents(root);
4292
4293                 cond_resched();
4294                 spin_lock(&fs_info->ordered_root_lock);
4295         }
4296         spin_unlock(&fs_info->ordered_root_lock);
4297 }
4298
4299 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4300                                       struct btrfs_fs_info *fs_info)
4301 {
4302         struct rb_node *node;
4303         struct btrfs_delayed_ref_root *delayed_refs;
4304         struct btrfs_delayed_ref_node *ref;
4305         int ret = 0;
4306
4307         delayed_refs = &trans->delayed_refs;
4308
4309         spin_lock(&delayed_refs->lock);
4310         if (atomic_read(&delayed_refs->num_entries) == 0) {
4311                 spin_unlock(&delayed_refs->lock);
4312                 btrfs_info(fs_info, "delayed_refs has NO entry");
4313                 return ret;
4314         }
4315
4316         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4317                 struct btrfs_delayed_ref_head *head;
4318                 struct btrfs_delayed_ref_node *tmp;
4319                 bool pin_bytes = false;
4320
4321                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4322                                 href_node);
4323                 if (!mutex_trylock(&head->mutex)) {
4324                         refcount_inc(&head->node.refs);
4325                         spin_unlock(&delayed_refs->lock);
4326
4327                         mutex_lock(&head->mutex);
4328                         mutex_unlock(&head->mutex);
4329                         btrfs_put_delayed_ref(&head->node);
4330                         spin_lock(&delayed_refs->lock);
4331                         continue;
4332                 }
4333                 spin_lock(&head->lock);
4334                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4335                                                  list) {
4336                         ref->in_tree = 0;
4337                         list_del(&ref->list);
4338                         if (!list_empty(&ref->add_list))
4339                                 list_del(&ref->add_list);
4340                         atomic_dec(&delayed_refs->num_entries);
4341                         btrfs_put_delayed_ref(ref);
4342                 }
4343                 if (head->must_insert_reserved)
4344                         pin_bytes = true;
4345                 btrfs_free_delayed_extent_op(head->extent_op);
4346                 delayed_refs->num_heads--;
4347                 if (head->processing == 0)
4348                         delayed_refs->num_heads_ready--;
4349                 atomic_dec(&delayed_refs->num_entries);
4350                 head->node.in_tree = 0;
4351                 rb_erase(&head->href_node, &delayed_refs->href_root);
4352                 spin_unlock(&head->lock);
4353                 spin_unlock(&delayed_refs->lock);
4354                 mutex_unlock(&head->mutex);
4355
4356                 if (pin_bytes)
4357                         btrfs_pin_extent(fs_info, head->node.bytenr,
4358                                          head->node.num_bytes, 1);
4359                 btrfs_put_delayed_ref(&head->node);
4360                 cond_resched();
4361                 spin_lock(&delayed_refs->lock);
4362         }
4363
4364         spin_unlock(&delayed_refs->lock);
4365
4366         return ret;
4367 }
4368
4369 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4370 {
4371         struct btrfs_inode *btrfs_inode;
4372         struct list_head splice;
4373
4374         INIT_LIST_HEAD(&splice);
4375
4376         spin_lock(&root->delalloc_lock);
4377         list_splice_init(&root->delalloc_inodes, &splice);
4378
4379         while (!list_empty(&splice)) {
4380                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4381                                                delalloc_inodes);
4382
4383                 list_del_init(&btrfs_inode->delalloc_inodes);
4384                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4385                           &btrfs_inode->runtime_flags);
4386                 spin_unlock(&root->delalloc_lock);
4387
4388                 btrfs_invalidate_inodes(btrfs_inode->root);
4389
4390                 spin_lock(&root->delalloc_lock);
4391         }
4392
4393         spin_unlock(&root->delalloc_lock);
4394 }
4395
4396 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4397 {
4398         struct btrfs_root *root;
4399         struct list_head splice;
4400
4401         INIT_LIST_HEAD(&splice);
4402
4403         spin_lock(&fs_info->delalloc_root_lock);
4404         list_splice_init(&fs_info->delalloc_roots, &splice);
4405         while (!list_empty(&splice)) {
4406                 root = list_first_entry(&splice, struct btrfs_root,
4407                                          delalloc_root);
4408                 list_del_init(&root->delalloc_root);
4409                 root = btrfs_grab_fs_root(root);
4410                 BUG_ON(!root);
4411                 spin_unlock(&fs_info->delalloc_root_lock);
4412
4413                 btrfs_destroy_delalloc_inodes(root);
4414                 btrfs_put_fs_root(root);
4415
4416                 spin_lock(&fs_info->delalloc_root_lock);
4417         }
4418         spin_unlock(&fs_info->delalloc_root_lock);
4419 }
4420
4421 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4422                                         struct extent_io_tree *dirty_pages,
4423                                         int mark)
4424 {
4425         int ret;
4426         struct extent_buffer *eb;
4427         u64 start = 0;
4428         u64 end;
4429
4430         while (1) {
4431                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4432                                             mark, NULL);
4433                 if (ret)
4434                         break;
4435
4436                 clear_extent_bits(dirty_pages, start, end, mark);
4437                 while (start <= end) {
4438                         eb = find_extent_buffer(fs_info, start);
4439                         start += fs_info->nodesize;
4440                         if (!eb)
4441                                 continue;
4442                         wait_on_extent_buffer_writeback(eb);
4443
4444                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4445                                                &eb->bflags))
4446                                 clear_extent_buffer_dirty(eb);
4447                         free_extent_buffer_stale(eb);
4448                 }
4449         }
4450
4451         return ret;
4452 }
4453
4454 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4455                                        struct extent_io_tree *pinned_extents)
4456 {
4457         struct extent_io_tree *unpin;
4458         u64 start;
4459         u64 end;
4460         int ret;
4461         bool loop = true;
4462
4463         unpin = pinned_extents;
4464 again:
4465         while (1) {
4466                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4467                                             EXTENT_DIRTY, NULL);
4468                 if (ret)
4469                         break;
4470
4471                 clear_extent_dirty(unpin, start, end);
4472                 btrfs_error_unpin_extent_range(fs_info, start, end);
4473                 cond_resched();
4474         }
4475
4476         if (loop) {
4477                 if (unpin == &fs_info->freed_extents[0])
4478                         unpin = &fs_info->freed_extents[1];
4479                 else
4480                         unpin = &fs_info->freed_extents[0];
4481                 loop = false;
4482                 goto again;
4483         }
4484
4485         return 0;
4486 }
4487
4488 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4489 {
4490         struct inode *inode;
4491
4492         inode = cache->io_ctl.inode;
4493         if (inode) {
4494                 invalidate_inode_pages2(inode->i_mapping);
4495                 BTRFS_I(inode)->generation = 0;
4496                 cache->io_ctl.inode = NULL;
4497                 iput(inode);
4498         }
4499         btrfs_put_block_group(cache);
4500 }
4501
4502 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4503                              struct btrfs_fs_info *fs_info)
4504 {
4505         struct btrfs_block_group_cache *cache;
4506
4507         spin_lock(&cur_trans->dirty_bgs_lock);
4508         while (!list_empty(&cur_trans->dirty_bgs)) {
4509                 cache = list_first_entry(&cur_trans->dirty_bgs,
4510                                          struct btrfs_block_group_cache,
4511                                          dirty_list);
4512                 if (!cache) {
4513                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4514                         spin_unlock(&cur_trans->dirty_bgs_lock);
4515                         return;
4516                 }
4517
4518                 if (!list_empty(&cache->io_list)) {
4519                         spin_unlock(&cur_trans->dirty_bgs_lock);
4520                         list_del_init(&cache->io_list);
4521                         btrfs_cleanup_bg_io(cache);
4522                         spin_lock(&cur_trans->dirty_bgs_lock);
4523                 }
4524
4525                 list_del_init(&cache->dirty_list);
4526                 spin_lock(&cache->lock);
4527                 cache->disk_cache_state = BTRFS_DC_ERROR;
4528                 spin_unlock(&cache->lock);
4529
4530                 spin_unlock(&cur_trans->dirty_bgs_lock);
4531                 btrfs_put_block_group(cache);
4532                 spin_lock(&cur_trans->dirty_bgs_lock);
4533         }
4534         spin_unlock(&cur_trans->dirty_bgs_lock);
4535
4536         while (!list_empty(&cur_trans->io_bgs)) {
4537                 cache = list_first_entry(&cur_trans->io_bgs,
4538                                          struct btrfs_block_group_cache,
4539                                          io_list);
4540                 if (!cache) {
4541                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4542                         return;
4543                 }
4544
4545                 list_del_init(&cache->io_list);
4546                 spin_lock(&cache->lock);
4547                 cache->disk_cache_state = BTRFS_DC_ERROR;
4548                 spin_unlock(&cache->lock);
4549                 btrfs_cleanup_bg_io(cache);
4550         }
4551 }
4552
4553 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4554                                    struct btrfs_fs_info *fs_info)
4555 {
4556         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4557         ASSERT(list_empty(&cur_trans->dirty_bgs));
4558         ASSERT(list_empty(&cur_trans->io_bgs));
4559
4560         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4561
4562         cur_trans->state = TRANS_STATE_COMMIT_START;
4563         wake_up(&fs_info->transaction_blocked_wait);
4564
4565         cur_trans->state = TRANS_STATE_UNBLOCKED;
4566         wake_up(&fs_info->transaction_wait);
4567
4568         btrfs_destroy_delayed_inodes(fs_info);
4569         btrfs_assert_delayed_root_empty(fs_info);
4570
4571         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4572                                      EXTENT_DIRTY);
4573         btrfs_destroy_pinned_extent(fs_info,
4574                                     fs_info->pinned_extents);
4575
4576         cur_trans->state =TRANS_STATE_COMPLETED;
4577         wake_up(&cur_trans->commit_wait);
4578
4579         /*
4580         memset(cur_trans, 0, sizeof(*cur_trans));
4581         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4582         */
4583 }
4584
4585 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4586 {
4587         struct btrfs_transaction *t;
4588
4589         mutex_lock(&fs_info->transaction_kthread_mutex);
4590
4591         spin_lock(&fs_info->trans_lock);
4592         while (!list_empty(&fs_info->trans_list)) {
4593                 t = list_first_entry(&fs_info->trans_list,
4594                                      struct btrfs_transaction, list);
4595                 if (t->state >= TRANS_STATE_COMMIT_START) {
4596                         refcount_inc(&t->use_count);
4597                         spin_unlock(&fs_info->trans_lock);
4598                         btrfs_wait_for_commit(fs_info, t->transid);
4599                         btrfs_put_transaction(t);
4600                         spin_lock(&fs_info->trans_lock);
4601                         continue;
4602                 }
4603                 if (t == fs_info->running_transaction) {
4604                         t->state = TRANS_STATE_COMMIT_DOING;
4605                         spin_unlock(&fs_info->trans_lock);
4606                         /*
4607                          * We wait for 0 num_writers since we don't hold a trans
4608                          * handle open currently for this transaction.
4609                          */
4610                         wait_event(t->writer_wait,
4611                                    atomic_read(&t->num_writers) == 0);
4612                 } else {
4613                         spin_unlock(&fs_info->trans_lock);
4614                 }
4615                 btrfs_cleanup_one_transaction(t, fs_info);
4616
4617                 spin_lock(&fs_info->trans_lock);
4618                 if (t == fs_info->running_transaction)
4619                         fs_info->running_transaction = NULL;
4620                 list_del_init(&t->list);
4621                 spin_unlock(&fs_info->trans_lock);
4622
4623                 btrfs_put_transaction(t);
4624                 trace_btrfs_transaction_commit(fs_info->tree_root);
4625                 spin_lock(&fs_info->trans_lock);
4626         }
4627         spin_unlock(&fs_info->trans_lock);
4628         btrfs_destroy_all_ordered_extents(fs_info);
4629         btrfs_destroy_delayed_inodes(fs_info);
4630         btrfs_assert_delayed_root_empty(fs_info);
4631         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4632         btrfs_destroy_all_delalloc_inodes(fs_info);
4633         mutex_unlock(&fs_info->transaction_kthread_mutex);
4634
4635         return 0;
4636 }
4637
4638 static const struct extent_io_ops btree_extent_io_ops = {
4639         /* mandatory callbacks */
4640         .submit_bio_hook = btree_submit_bio_hook,
4641         .readpage_end_io_hook = btree_readpage_end_io_hook,
4642         /* note we're sharing with inode.c for the merge bio hook */
4643         .merge_bio_hook = btrfs_merge_bio_hook,
4644         .readpage_io_failed_hook = btree_io_failed_hook,
4645
4646         /* optional callbacks */
4647 };