fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/slab.h>
  29 #include <linux/migrate.h>
  30 #include <linux/ratelimit.h>
  31 #include <linux/uuid.h>
  32 #include <linux/semaphore.h>
  33 #include <asm/unaligned.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "hash.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "free-space-tree.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52 #include "compression.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68                                     int read_only);
  69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71                                       struct btrfs_fs_info *fs_info);
  72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  74                                         struct extent_io_tree *dirty_pages,
  75                                         int mark);
  76 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  77                                        struct extent_io_tree *pinned_extents);
  78 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  79 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  80
  81 /*
  82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83  * is complete.  This is used during reads to verify checksums, and it is used
  84  * by writes to insert metadata for new file extents after IO is complete.
  85  */
  86 struct btrfs_end_io_wq {
  87         struct bio *bio;
  88         bio_end_io_t *end_io;
  89         void *private;
  90         struct btrfs_fs_info *info;
  91         int error;
  92         enum btrfs_wq_endio_type metadata;
  93         struct list_head list;
  94         struct btrfs_work work;
  95 };
  96
  97 static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99 int __init btrfs_end_io_wq_init(void)
 100 {
 101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102                                         sizeof(struct btrfs_end_io_wq),
 103                                         0,
 104                                         SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         kmem_cache_destroy(btrfs_end_io_wq_cache);
 114 }
 115
 116 /*
 117  * async submit bios are used to offload expensive checksumming
 118  * onto the worker threads.  They checksum file and metadata bios
 119  * just before they are sent down the IO stack.
 120  */
 121 struct async_submit_bio {
 122         struct inode *inode;
 123         struct bio *bio;
 124         struct list_head list;
 125         extent_submit_bio_hook_t *submit_bio_start;
 126         extent_submit_bio_hook_t *submit_bio_done;
 127         int mirror_num;
 128         unsigned long bio_flags;
 129         /*
 130          * bio_offset is optional, can be used if the pages in the bio
 131          * can't tell us where in the file the bio should go
 132          */
 133         u64 bio_offset;
 134         struct btrfs_work work;
 135         int error;
 136 };
 137
 138 /*
 139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 141  * the level the eb occupies in the tree.
 142  *
 143  * Different roots are used for different purposes and may nest inside each
 144  * other and they require separate keysets.  As lockdep keys should be
 145  * static, assign keysets according to the purpose of the root as indicated
 146  * by btrfs_root->objectid.  This ensures that all special purpose roots
 147  * have separate keysets.
 148  *
 149  * Lock-nesting across peer nodes is always done with the immediate parent
 150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 151  * subclass to avoid triggering lockdep warning in such cases.
 152  *
 153  * The key is set by the readpage_end_io_hook after the buffer has passed
 154  * csum validation but before the pages are unlocked.  It is also set by
 155  * btrfs_init_new_buffer on freshly allocated blocks.
 156  *
 157  * We also add a check to make sure the highest level of the tree is the
 158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 159  * needs update as well.
 160  */
 161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 162 # if BTRFS_MAX_LEVEL != 8
 163 #  error
 164 # endif
 165
 166 static struct btrfs_lockdep_keyset {
 167         u64                     id;             /* root objectid */
 168         const char              *name_stem;     /* lock name stem */
 169         char                    names[BTRFS_MAX_LEVEL + 1][20];
 170         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 171 } btrfs_lockdep_keysets[] = {
 172         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 173         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 174         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 175         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 176         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 177         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 178         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 179         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 180         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 181         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 182         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 183         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 184         { .id = 0,                              .name_stem = "tree"     },
 185 };
 186
 187 void __init btrfs_init_lockdep(void)
 188 {
 189         int i, j;
 190
 191         /* initialize lockdep class names */
 192         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 193                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 194
 195                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 196                         snprintf(ks->names[j], sizeof(ks->names[j]),
 197                                  "btrfs-%s-%02d", ks->name_stem, j);
 198         }
 199 }
 200
 201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 202                                     int level)
 203 {
 204         struct btrfs_lockdep_keyset *ks;
 205
 206         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 207
 208         /* find the matching keyset, id 0 is the default entry */
 209         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 210                 if (ks->id == objectid)
 211                         break;
 212
 213         lockdep_set_class_and_name(&eb->lock,
 214                                    &ks->keys[level], ks->names[level]);
 215 }
 216
 217 #endif
 218
 219 /*
 220  * extents on the btree inode are pretty simple, there's one extent
 221  * that covers the entire device
 222  */
 223 static struct extent_map *btree_get_extent(struct inode *inode,
 224                 struct page *page, size_t pg_offset, u64 start, u64 len,
 225                 int create)
 226 {
 227         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 228         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 229         struct extent_map *em;
 230         int ret;
 231
 232         read_lock(&em_tree->lock);
 233         em = lookup_extent_mapping(em_tree, start, len);
 234         if (em) {
 235                 em->bdev = fs_info->fs_devices->latest_bdev;
 236                 read_unlock(&em_tree->lock);
 237                 goto out;
 238         }
 239         read_unlock(&em_tree->lock);
 240
 241         em = alloc_extent_map();
 242         if (!em) {
 243                 em = ERR_PTR(-ENOMEM);
 244                 goto out;
 245         }
 246         em->start = 0;
 247         em->len = (u64)-1;
 248         em->block_len = (u64)-1;
 249         em->block_start = 0;
 250         em->bdev = fs_info->fs_devices->latest_bdev;
 251
 252         write_lock(&em_tree->lock);
 253         ret = add_extent_mapping(em_tree, em, 0);
 254         if (ret == -EEXIST) {
 255                 free_extent_map(em);
 256                 em = lookup_extent_mapping(em_tree, start, len);
 257                 if (!em)
 258                         em = ERR_PTR(-EIO);
 259         } else if (ret) {
 260                 free_extent_map(em);
 261                 em = ERR_PTR(ret);
 262         }
 263         write_unlock(&em_tree->lock);
 264
 265 out:
 266         return em;
 267 }
 268
 269 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 270 {
 271         return btrfs_crc32c(seed, data, len);
 272 }
 273
 274 void btrfs_csum_final(u32 crc, u8 *result)
 275 {
 276         put_unaligned_le32(~crc, result);
 277 }
 278
 279 /*
 280  * compute the csum for a btree block, and either verify it or write it
 281  * into the csum field of the block.
 282  */
 283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 284                            struct extent_buffer *buf,
 285                            int verify)
 286 {
 287         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 288         char *result = NULL;
 289         unsigned long len;
 290         unsigned long cur_len;
 291         unsigned long offset = BTRFS_CSUM_SIZE;
 292         char *kaddr;
 293         unsigned long map_start;
 294         unsigned long map_len;
 295         int err;
 296         u32 crc = ~(u32)0;
 297         unsigned long inline_result;
 298
 299         len = buf->len - offset;
 300         while (len > 0) {
 301                 err = map_private_extent_buffer(buf, offset, 32,
 302                                         &kaddr, &map_start, &map_len);
 303                 if (err)
 304                         return err;
 305                 cur_len = min(len, map_len - (offset - map_start));
 306                 crc = btrfs_csum_data(kaddr + offset - map_start,
 307                                       crc, cur_len);
 308                 len -= cur_len;
 309                 offset += cur_len;
 310         }
 311         if (csum_size > sizeof(inline_result)) {
 312                 result = kzalloc(csum_size, GFP_NOFS);
 313                 if (!result)
 314                         return -ENOMEM;
 315         } else {
 316                 result = (char *)&inline_result;
 317         }
 318
 319         btrfs_csum_final(crc, result);
 320
 321         if (verify) {
 322                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 323                         u32 val;
 324                         u32 found = 0;
 325                         memcpy(&found, result, csum_size);
 326
 327                         read_extent_buffer(buf, &val, 0, csum_size);
 328                         btrfs_warn_rl(fs_info,
 329                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 330                                 fs_info->sb->s_id, buf->start,
 331                                 val, found, btrfs_header_level(buf));
 332                         if (result != (char *)&inline_result)
 333                                 kfree(result);
 334                         return -EUCLEAN;
 335                 }
 336         } else {
 337                 write_extent_buffer(buf, result, 0, csum_size);
 338         }
 339         if (result != (char *)&inline_result)
 340                 kfree(result);
 341         return 0;
 342 }
 343
 344 /*
 345  * we can't consider a given block up to date unless the transid of the
 346  * block matches the transid in the parent node's pointer.  This is how we
 347  * detect blocks that either didn't get written at all or got written
 348  * in the wrong place.
 349  */
 350 static int verify_parent_transid(struct extent_io_tree *io_tree,
 351                                  struct extent_buffer *eb, u64 parent_transid,
 352                                  int atomic)
 353 {
 354         struct extent_state *cached_state = NULL;
 355         int ret;
 356         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 357
 358         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 359                 return 0;
 360
 361         if (atomic)
 362                 return -EAGAIN;
 363
 364         if (need_lock) {
 365                 btrfs_tree_read_lock(eb);
 366                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 367         }
 368
 369         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 370                          &cached_state);
 371         if (extent_buffer_uptodate(eb) &&
 372             btrfs_header_generation(eb) == parent_transid) {
 373                 ret = 0;
 374                 goto out;
 375         }
 376         btrfs_err_rl(eb->fs_info,
 377                 "parent transid verify failed on %llu wanted %llu found %llu",
 378                         eb->start,
 379                         parent_transid, btrfs_header_generation(eb));
 380         ret = 1;
 381
 382         /*
 383          * Things reading via commit roots that don't have normal protection,
 384          * like send, can have a really old block in cache that may point at a
 385          * block that has been freed and re-allocated.  So don't clear uptodate
 386          * if we find an eb that is under IO (dirty/writeback) because we could
 387          * end up reading in the stale data and then writing it back out and
 388          * making everybody very sad.
 389          */
 390         if (!extent_buffer_under_io(eb))
 391                 clear_extent_buffer_uptodate(eb);
 392 out:
 393         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 394                              &cached_state, GFP_NOFS);
 395         if (need_lock)
 396                 btrfs_tree_read_unlock_blocking(eb);
 397         return ret;
 398 }
 399
 400 /*
 401  * Return 0 if the superblock checksum type matches the checksum value of that
 402  * algorithm. Pass the raw disk superblock data.
 403  */
 404 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 405                                   char *raw_disk_sb)
 406 {
 407         struct btrfs_super_block *disk_sb =
 408                 (struct btrfs_super_block *)raw_disk_sb;
 409         u16 csum_type = btrfs_super_csum_type(disk_sb);
 410         int ret = 0;
 411
 412         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 413                 u32 crc = ~(u32)0;
 414                 const int csum_size = sizeof(crc);
 415                 char result[csum_size];
 416
 417                 /*
 418                  * The super_block structure does not span the whole
 419                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 420                  * is filled with zeros and is included in the checksum.
 421                  */
 422                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 423                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 424                 btrfs_csum_final(crc, result);
 425
 426                 if (memcmp(raw_disk_sb, result, csum_size))
 427                         ret = 1;
 428         }
 429
 430         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 431                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 432                                 csum_type);
 433                 ret = 1;
 434         }
 435
 436         return ret;
 437 }
 438
 439 /*
 440  * helper to read a given tree block, doing retries as required when
 441  * the checksums don't match and we have alternate mirrors to try.
 442  */
 443 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 444                                           struct extent_buffer *eb,
 445                                           u64 parent_transid)
 446 {
 447         struct extent_io_tree *io_tree;
 448         int failed = 0;
 449         int ret;
 450         int num_copies = 0;
 451         int mirror_num = 0;
 452         int failed_mirror = 0;
 453
 454         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 455         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 456         while (1) {
 457                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 458                                                btree_get_extent, mirror_num);
 459                 if (!ret) {
 460                         if (!verify_parent_transid(io_tree, eb,
 461                                                    parent_transid, 0))
 462                                 break;
 463                         else
 464                                 ret = -EIO;
 465                 }
 466
 467                 /*
 468                  * This buffer's crc is fine, but its contents are corrupted, so
 469                  * there is no reason to read the other copies, they won't be
 470                  * any less wrong.
 471                  */
 472                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 473                         break;
 474
 475                 num_copies = btrfs_num_copies(fs_info,
 476                                               eb->start, eb->len);
 477                 if (num_copies == 1)
 478                         break;
 479
 480                 if (!failed_mirror) {
 481                         failed = 1;
 482                         failed_mirror = eb->read_mirror;
 483                 }
 484
 485                 mirror_num++;
 486                 if (mirror_num == failed_mirror)
 487                         mirror_num++;
 488
 489                 if (mirror_num > num_copies)
 490                         break;
 491         }
 492
 493         if (failed && !ret && failed_mirror)
 494                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 495
 496         return ret;
 497 }
 498
 499 /*
 500  * checksum a dirty tree block before IO.  This has extra checks to make sure
 501  * we only fill in the checksum field in the first page of a multi-page block
 502  */
 503
 504 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 505 {
 506         u64 start = page_offset(page);
 507         u64 found_start;
 508         struct extent_buffer *eb;
 509
 510         eb = (struct extent_buffer *)page->private;
 511         if (page != eb->pages[0])
 512                 return 0;
 513
 514         found_start = btrfs_header_bytenr(eb);
 515         /*
 516          * Please do not consolidate these warnings into a single if.
 517          * It is useful to know what went wrong.
 518          */
 519         if (WARN_ON(found_start != start))
 520                 return -EUCLEAN;
 521         if (WARN_ON(!PageUptodate(page)))
 522                 return -EUCLEAN;
 523
 524         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 525                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 526
 527         return csum_tree_block(fs_info, eb, 0);
 528 }
 529
 530 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 531                                  struct extent_buffer *eb)
 532 {
 533         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 534         u8 fsid[BTRFS_UUID_SIZE];
 535         int ret = 1;
 536
 537         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 538         while (fs_devices) {
 539                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 540                         ret = 0;
 541                         break;
 542                 }
 543                 fs_devices = fs_devices->seed;
 544         }
 545         return ret;
 546 }
 547
 548 #define CORRUPT(reason, eb, root, slot)                                 \
 549         btrfs_crit(root->fs_info,                                       \
 550                    "corrupt %s, %s: block=%llu, root=%llu, slot=%d",    \
 551                    btrfs_header_level(eb) == 0 ? "leaf" : "node",       \
 552                    reason, btrfs_header_bytenr(eb), root->objectid, slot)
 553
 554 static noinline int check_leaf(struct btrfs_root *root,
 555                                struct extent_buffer *leaf)
 556 {
 557         struct btrfs_fs_info *fs_info = root->fs_info;
 558         struct btrfs_key key;
 559         struct btrfs_key leaf_key;
 560         u32 nritems = btrfs_header_nritems(leaf);
 561         int slot;
 562
 563         /*
 564          * Extent buffers from a relocation tree have a owner field that
 565          * corresponds to the subvolume tree they are based on. So just from an
 566          * extent buffer alone we can not find out what is the id of the
 567          * corresponding subvolume tree, so we can not figure out if the extent
 568          * buffer corresponds to the root of the relocation tree or not. So skip
 569          * this check for relocation trees.
 570          */
 571         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
 572                 struct btrfs_root *check_root;
 573
 574                 key.objectid = btrfs_header_owner(leaf);
 575                 key.type = BTRFS_ROOT_ITEM_KEY;
 576                 key.offset = (u64)-1;
 577
 578                 check_root = btrfs_get_fs_root(fs_info, &key, false);
 579                 /*
 580                  * The only reason we also check NULL here is that during
 581                  * open_ctree() some roots has not yet been set up.
 582                  */
 583                 if (!IS_ERR_OR_NULL(check_root)) {
 584                         struct extent_buffer *eb;
 585
 586                         eb = btrfs_root_node(check_root);
 587                         /* if leaf is the root, then it's fine */
 588                         if (leaf != eb) {
 589                                 CORRUPT("non-root leaf's nritems is 0",
 590                                         leaf, check_root, 0);
 591                                 free_extent_buffer(eb);
 592                                 return -EIO;
 593                         }
 594                         free_extent_buffer(eb);
 595                 }
 596                 return 0;
 597         }
 598
 599         if (nritems == 0)
 600                 return 0;
 601
 602         /* Check the 0 item */
 603         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 604             BTRFS_LEAF_DATA_SIZE(fs_info)) {
 605                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 606                 return -EIO;
 607         }
 608
 609         /*
 610          * Check to make sure each items keys are in the correct order and their
 611          * offsets make sense.  We only have to loop through nritems-1 because
 612          * we check the current slot against the next slot, which verifies the
 613          * next slot's offset+size makes sense and that the current's slot
 614          * offset is correct.
 615          */
 616         for (slot = 0; slot < nritems - 1; slot++) {
 617                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 618                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 619
 620                 /* Make sure the keys are in the right order */
 621                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 622                         CORRUPT("bad key order", leaf, root, slot);
 623                         return -EIO;
 624                 }
 625
 626                 /*
 627                  * Make sure the offset and ends are right, remember that the
 628                  * item data starts at the end of the leaf and grows towards the
 629                  * front.
 630                  */
 631                 if (btrfs_item_offset_nr(leaf, slot) !=
 632                         btrfs_item_end_nr(leaf, slot + 1)) {
 633                         CORRUPT("slot offset bad", leaf, root, slot);
 634                         return -EIO;
 635                 }
 636
 637                 /*
 638                  * Check to make sure that we don't point outside of the leaf,
 639                  * just in case all the items are consistent to each other, but
 640                  * all point outside of the leaf.
 641                  */
 642                 if (btrfs_item_end_nr(leaf, slot) >
 643                     BTRFS_LEAF_DATA_SIZE(fs_info)) {
 644                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 645                         return -EIO;
 646                 }
 647         }
 648
 649         return 0;
 650 }
 651
 652 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
 653 {
 654         unsigned long nr = btrfs_header_nritems(node);
 655         struct btrfs_key key, next_key;
 656         int slot;
 657         u64 bytenr;
 658         int ret = 0;
 659
 660         if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root->fs_info)) {
 661                 btrfs_crit(root->fs_info,
 662                            "corrupt node: block %llu root %llu nritems %lu",
 663                            node->start, root->objectid, nr);
 664                 return -EIO;
 665         }
 666
 667         for (slot = 0; slot < nr - 1; slot++) {
 668                 bytenr = btrfs_node_blockptr(node, slot);
 669                 btrfs_node_key_to_cpu(node, &key, slot);
 670                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
 671
 672                 if (!bytenr) {
 673                         CORRUPT("invalid item slot", node, root, slot);
 674                         ret = -EIO;
 675                         goto out;
 676                 }
 677
 678                 if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
 679                         CORRUPT("bad key order", node, root, slot);
 680                         ret = -EIO;
 681                         goto out;
 682                 }
 683         }
 684 out:
 685         return ret;
 686 }
 687
 688 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 689                                       u64 phy_offset, struct page *page,
 690                                       u64 start, u64 end, int mirror)
 691 {
 692         u64 found_start;
 693         int found_level;
 694         struct extent_buffer *eb;
 695         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 696         struct btrfs_fs_info *fs_info = root->fs_info;
 697         int ret = 0;
 698         int reads_done;
 699
 700         if (!page->private)
 701                 goto out;
 702
 703         eb = (struct extent_buffer *)page->private;
 704
 705         /* the pending IO might have been the only thing that kept this buffer
 706          * in memory.  Make sure we have a ref for all this other checks
 707          */
 708         extent_buffer_get(eb);
 709
 710         reads_done = atomic_dec_and_test(&eb->io_pages);
 711         if (!reads_done)
 712                 goto err;
 713
 714         eb->read_mirror = mirror;
 715         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 716                 ret = -EIO;
 717                 goto err;
 718         }
 719
 720         found_start = btrfs_header_bytenr(eb);
 721         if (found_start != eb->start) {
 722                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 723                              found_start, eb->start);
 724                 ret = -EIO;
 725                 goto err;
 726         }
 727         if (check_tree_block_fsid(fs_info, eb)) {
 728                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 729                              eb->start);
 730                 ret = -EIO;
 731                 goto err;
 732         }
 733         found_level = btrfs_header_level(eb);
 734         if (found_level >= BTRFS_MAX_LEVEL) {
 735                 btrfs_err(fs_info, "bad tree block level %d",
 736                           (int)btrfs_header_level(eb));
 737                 ret = -EIO;
 738                 goto err;
 739         }
 740
 741         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 742                                        eb, found_level);
 743
 744         ret = csum_tree_block(fs_info, eb, 1);
 745         if (ret)
 746                 goto err;
 747
 748         /*
 749          * If this is a leaf block and it is corrupt, set the corrupt bit so
 750          * that we don't try and read the other copies of this block, just
 751          * return -EIO.
 752          */
 753         if (found_level == 0 && check_leaf(root, eb)) {
 754                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 755                 ret = -EIO;
 756         }
 757
 758         if (found_level > 0 && check_node(root, eb))
 759                 ret = -EIO;
 760
 761         if (!ret)
 762                 set_extent_buffer_uptodate(eb);
 763 err:
 764         if (reads_done &&
 765             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 766                 btree_readahead_hook(fs_info, eb, ret);
 767
 768         if (ret) {
 769                 /*
 770                  * our io error hook is going to dec the io pages
 771                  * again, we have to make sure it has something
 772                  * to decrement
 773                  */
 774                 atomic_inc(&eb->io_pages);
 775                 clear_extent_buffer_uptodate(eb);
 776         }
 777         free_extent_buffer(eb);
 778 out:
 779         return ret;
 780 }
 781
 782 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 783 {
 784         struct extent_buffer *eb;
 785
 786         eb = (struct extent_buffer *)page->private;
 787         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 788         eb->read_mirror = failed_mirror;
 789         atomic_dec(&eb->io_pages);
 790         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 791                 btree_readahead_hook(eb->fs_info, eb, -EIO);
 792         return -EIO;    /* we fixed nothing */
 793 }
 794
 795 static void end_workqueue_bio(struct bio *bio)
 796 {
 797         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 798         struct btrfs_fs_info *fs_info;
 799         struct btrfs_workqueue *wq;
 800         btrfs_work_func_t func;
 801
 802         fs_info = end_io_wq->info;
 803         end_io_wq->error = bio->bi_error;
 804
 805         if (bio_op(bio) == REQ_OP_WRITE) {
 806                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 807                         wq = fs_info->endio_meta_write_workers;
 808                         func = btrfs_endio_meta_write_helper;
 809                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 810                         wq = fs_info->endio_freespace_worker;
 811                         func = btrfs_freespace_write_helper;
 812                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 813                         wq = fs_info->endio_raid56_workers;
 814                         func = btrfs_endio_raid56_helper;
 815                 } else {
 816                         wq = fs_info->endio_write_workers;
 817                         func = btrfs_endio_write_helper;
 818                 }
 819         } else {
 820                 if (unlikely(end_io_wq->metadata ==
 821                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 822                         wq = fs_info->endio_repair_workers;
 823                         func = btrfs_endio_repair_helper;
 824                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 825                         wq = fs_info->endio_raid56_workers;
 826                         func = btrfs_endio_raid56_helper;
 827                 } else if (end_io_wq->metadata) {
 828                         wq = fs_info->endio_meta_workers;
 829                         func = btrfs_endio_meta_helper;
 830                 } else {
 831                         wq = fs_info->endio_workers;
 832                         func = btrfs_endio_helper;
 833                 }
 834         }
 835
 836         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 837         btrfs_queue_work(wq, &end_io_wq->work);
 838 }
 839
 840 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 841                         enum btrfs_wq_endio_type metadata)
 842 {
 843         struct btrfs_end_io_wq *end_io_wq;
 844
 845         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 846         if (!end_io_wq)
 847                 return -ENOMEM;
 848
 849         end_io_wq->private = bio->bi_private;
 850         end_io_wq->end_io = bio->bi_end_io;
 851         end_io_wq->info = info;
 852         end_io_wq->error = 0;
 853         end_io_wq->bio = bio;
 854         end_io_wq->metadata = metadata;
 855
 856         bio->bi_private = end_io_wq;
 857         bio->bi_end_io = end_workqueue_bio;
 858         return 0;
 859 }
 860
 861 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 862 {
 863         unsigned long limit = min_t(unsigned long,
 864                                     info->thread_pool_size,
 865                                     info->fs_devices->open_devices);
 866         return 256 * limit;
 867 }
 868
 869 static void run_one_async_start(struct btrfs_work *work)
 870 {
 871         struct async_submit_bio *async;
 872         int ret;
 873
 874         async = container_of(work, struct  async_submit_bio, work);
 875         ret = async->submit_bio_start(async->inode, async->bio,
 876                                       async->mirror_num, async->bio_flags,
 877                                       async->bio_offset);
 878         if (ret)
 879                 async->error = ret;
 880 }
 881
 882 static void run_one_async_done(struct btrfs_work *work)
 883 {
 884         struct btrfs_fs_info *fs_info;
 885         struct async_submit_bio *async;
 886         int limit;
 887
 888         async = container_of(work, struct  async_submit_bio, work);
 889         fs_info = BTRFS_I(async->inode)->root->fs_info;
 890
 891         limit = btrfs_async_submit_limit(fs_info);
 892         limit = limit * 2 / 3;
 893
 894         /*
 895          * atomic_dec_return implies a barrier for waitqueue_active
 896          */
 897         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 898             waitqueue_active(&fs_info->async_submit_wait))
 899                 wake_up(&fs_info->async_submit_wait);
 900
 901         /* If an error occurred we just want to clean up the bio and move on */
 902         if (async->error) {
 903                 async->bio->bi_error = async->error;
 904                 bio_endio(async->bio);
 905                 return;
 906         }
 907
 908         async->submit_bio_done(async->inode, async->bio, async->mirror_num,
 909                                async->bio_flags, async->bio_offset);
 910 }
 911
 912 static void run_one_async_free(struct btrfs_work *work)
 913 {
 914         struct async_submit_bio *async;
 915
 916         async = container_of(work, struct  async_submit_bio, work);
 917         kfree(async);
 918 }
 919
 920 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 921                         struct bio *bio, int mirror_num,
 922                         unsigned long bio_flags,
 923                         u64 bio_offset,
 924                         extent_submit_bio_hook_t *submit_bio_start,
 925                         extent_submit_bio_hook_t *submit_bio_done)
 926 {
 927         struct async_submit_bio *async;
 928
 929         async = kmalloc(sizeof(*async), GFP_NOFS);
 930         if (!async)
 931                 return -ENOMEM;
 932
 933         async->inode = inode;
 934         async->bio = bio;
 935         async->mirror_num = mirror_num;
 936         async->submit_bio_start = submit_bio_start;
 937         async->submit_bio_done = submit_bio_done;
 938
 939         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 940                         run_one_async_done, run_one_async_free);
 941
 942         async->bio_flags = bio_flags;
 943         async->bio_offset = bio_offset;
 944
 945         async->error = 0;
 946
 947         atomic_inc(&fs_info->nr_async_submits);
 948
 949         if (op_is_sync(bio->bi_opf))
 950                 btrfs_set_work_high_priority(&async->work);
 951
 952         btrfs_queue_work(fs_info->workers, &async->work);
 953
 954         while (atomic_read(&fs_info->async_submit_draining) &&
 955               atomic_read(&fs_info->nr_async_submits)) {
 956                 wait_event(fs_info->async_submit_wait,
 957                            (atomic_read(&fs_info->nr_async_submits) == 0));
 958         }
 959
 960         return 0;
 961 }
 962
 963 static int btree_csum_one_bio(struct bio *bio)
 964 {
 965         struct bio_vec *bvec;
 966         struct btrfs_root *root;
 967         int i, ret = 0;
 968
 969         bio_for_each_segment_all(bvec, bio, i) {
 970                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 971                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 972                 if (ret)
 973                         break;
 974         }
 975
 976         return ret;
 977 }
 978
 979 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
 980                                     int mirror_num, unsigned long bio_flags,
 981                                     u64 bio_offset)
 982 {
 983         /*
 984          * when we're called for a write, we're already in the async
 985          * submission context.  Just jump into btrfs_map_bio
 986          */
 987         return btree_csum_one_bio(bio);
 988 }
 989
 990 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
 991                                  int mirror_num, unsigned long bio_flags,
 992                                  u64 bio_offset)
 993 {
 994         int ret;
 995
 996         /*
 997          * when we're called for a write, we're already in the async
 998          * submission context.  Just jump into btrfs_map_bio
 999          */
1000         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
1001         if (ret) {
1002                 bio->bi_error = ret;
1003                 bio_endio(bio);
1004         }
1005         return ret;
1006 }
1007
1008 static int check_async_write(struct inode *inode, unsigned long bio_flags)
1009 {
1010         if (bio_flags & EXTENT_BIO_TREE_LOG)
1011                 return 0;
1012 #ifdef CONFIG_X86
1013         if (static_cpu_has(X86_FEATURE_XMM4_2))
1014                 return 0;
1015 #endif
1016         return 1;
1017 }
1018
1019 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1020                                  int mirror_num, unsigned long bio_flags,
1021                                  u64 bio_offset)
1022 {
1023         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1024         int async = check_async_write(inode, bio_flags);
1025         int ret;
1026
1027         if (bio_op(bio) != REQ_OP_WRITE) {
1028                 /*
1029                  * called for a read, do the setup so that checksum validation
1030                  * can happen in the async kernel threads
1031                  */
1032                 ret = btrfs_bio_wq_end_io(fs_info, bio,
1033                                           BTRFS_WQ_ENDIO_METADATA);
1034                 if (ret)
1035                         goto out_w_error;
1036                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1037         } else if (!async) {
1038                 ret = btree_csum_one_bio(bio);
1039                 if (ret)
1040                         goto out_w_error;
1041                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1042         } else {
1043                 /*
1044                  * kthread helpers are used to submit writes so that
1045                  * checksumming can happen in parallel across all CPUs
1046                  */
1047                 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, 0,
1048                                           bio_offset,
1049                                           __btree_submit_bio_start,
1050                                           __btree_submit_bio_done);
1051         }
1052
1053         if (ret)
1054                 goto out_w_error;
1055         return 0;
1056
1057 out_w_error:
1058         bio->bi_error = ret;
1059         bio_endio(bio);
1060         return ret;
1061 }
1062
1063 #ifdef CONFIG_MIGRATION
1064 static int btree_migratepage(struct address_space *mapping,
1065                         struct page *newpage, struct page *page,
1066                         enum migrate_mode mode)
1067 {
1068         /*
1069          * we can't safely write a btree page from here,
1070          * we haven't done the locking hook
1071          */
1072         if (PageDirty(page))
1073                 return -EAGAIN;
1074         /*
1075          * Buffers may be managed in a filesystem specific way.
1076          * We must have no buffers or drop them.
1077          */
1078         if (page_has_private(page) &&
1079             !try_to_release_page(page, GFP_KERNEL))
1080                 return -EAGAIN;
1081         return migrate_page(mapping, newpage, page, mode);
1082 }
1083 #endif
1084
1085
1086 static int btree_writepages(struct address_space *mapping,
1087                             struct writeback_control *wbc)
1088 {
1089         struct btrfs_fs_info *fs_info;
1090         int ret;
1091
1092         if (wbc->sync_mode == WB_SYNC_NONE) {
1093
1094                 if (wbc->for_kupdate)
1095                         return 0;
1096
1097                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1098                 /* this is a bit racy, but that's ok */
1099                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1100                                              BTRFS_DIRTY_METADATA_THRESH);
1101                 if (ret < 0)
1102                         return 0;
1103         }
1104         return btree_write_cache_pages(mapping, wbc);
1105 }
1106
1107 static int btree_readpage(struct file *file, struct page *page)
1108 {
1109         struct extent_io_tree *tree;
1110         tree = &BTRFS_I(page->mapping->host)->io_tree;
1111         return extent_read_full_page(tree, page, btree_get_extent, 0);
1112 }
1113
1114 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1115 {
1116         if (PageWriteback(page) || PageDirty(page))
1117                 return 0;
1118
1119         return try_release_extent_buffer(page);
1120 }
1121
1122 static void btree_invalidatepage(struct page *page, unsigned int offset,
1123                                  unsigned int length)
1124 {
1125         struct extent_io_tree *tree;
1126         tree = &BTRFS_I(page->mapping->host)->io_tree;
1127         extent_invalidatepage(tree, page, offset);
1128         btree_releasepage(page, GFP_NOFS);
1129         if (PagePrivate(page)) {
1130                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1131                            "page private not zero on page %llu",
1132                            (unsigned long long)page_offset(page));
1133                 ClearPagePrivate(page);
1134                 set_page_private(page, 0);
1135                 put_page(page);
1136         }
1137 }
1138
1139 static int btree_set_page_dirty(struct page *page)
1140 {
1141 #ifdef DEBUG
1142         struct extent_buffer *eb;
1143
1144         BUG_ON(!PagePrivate(page));
1145         eb = (struct extent_buffer *)page->private;
1146         BUG_ON(!eb);
1147         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1148         BUG_ON(!atomic_read(&eb->refs));
1149         btrfs_assert_tree_locked(eb);
1150 #endif
1151         return __set_page_dirty_nobuffers(page);
1152 }
1153
1154 static const struct address_space_operations btree_aops = {
1155         .readpage       = btree_readpage,
1156         .writepages     = btree_writepages,
1157         .releasepage    = btree_releasepage,
1158         .invalidatepage = btree_invalidatepage,
1159 #ifdef CONFIG_MIGRATION
1160         .migratepage    = btree_migratepage,
1161 #endif
1162         .set_page_dirty = btree_set_page_dirty,
1163 };
1164
1165 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1166 {
1167         struct extent_buffer *buf = NULL;
1168         struct inode *btree_inode = fs_info->btree_inode;
1169
1170         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1171         if (IS_ERR(buf))
1172                 return;
1173         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1174                                  buf, WAIT_NONE, btree_get_extent, 0);
1175         free_extent_buffer(buf);
1176 }
1177
1178 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1179                          int mirror_num, struct extent_buffer **eb)
1180 {
1181         struct extent_buffer *buf = NULL;
1182         struct inode *btree_inode = fs_info->btree_inode;
1183         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1184         int ret;
1185
1186         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1187         if (IS_ERR(buf))
1188                 return 0;
1189
1190         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1191
1192         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1193                                        btree_get_extent, mirror_num);
1194         if (ret) {
1195                 free_extent_buffer(buf);
1196                 return ret;
1197         }
1198
1199         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1200                 free_extent_buffer(buf);
1201                 return -EIO;
1202         } else if (extent_buffer_uptodate(buf)) {
1203                 *eb = buf;
1204         } else {
1205                 free_extent_buffer(buf);
1206         }
1207         return 0;
1208 }
1209
1210 struct extent_buffer *btrfs_find_create_tree_block(
1211                                                 struct btrfs_fs_info *fs_info,
1212                                                 u64 bytenr)
1213 {
1214         if (btrfs_is_testing(fs_info))
1215                 return alloc_test_extent_buffer(fs_info, bytenr);
1216         return alloc_extent_buffer(fs_info, bytenr);
1217 }
1218
1219
1220 int btrfs_write_tree_block(struct extent_buffer *buf)
1221 {
1222         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1223                                         buf->start + buf->len - 1);
1224 }
1225
1226 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1227 {
1228         return filemap_fdatawait_range(buf->pages[0]->mapping,
1229                                        buf->start, buf->start + buf->len - 1);
1230 }
1231
1232 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1233                                       u64 parent_transid)
1234 {
1235         struct extent_buffer *buf = NULL;
1236         int ret;
1237
1238         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1239         if (IS_ERR(buf))
1240                 return buf;
1241
1242         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1243         if (ret) {
1244                 free_extent_buffer(buf);
1245                 return ERR_PTR(ret);
1246         }
1247         return buf;
1248
1249 }
1250
1251 void clean_tree_block(struct btrfs_trans_handle *trans,
1252                       struct btrfs_fs_info *fs_info,
1253                       struct extent_buffer *buf)
1254 {
1255         if (btrfs_header_generation(buf) ==
1256             fs_info->running_transaction->transid) {
1257                 btrfs_assert_tree_locked(buf);
1258
1259                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1260                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1261                                              -buf->len,
1262                                              fs_info->dirty_metadata_batch);
1263                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1264                         btrfs_set_lock_blocking(buf);
1265                         clear_extent_buffer_dirty(buf);
1266                 }
1267         }
1268 }
1269
1270 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1271 {
1272         struct btrfs_subvolume_writers *writers;
1273         int ret;
1274
1275         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1276         if (!writers)
1277                 return ERR_PTR(-ENOMEM);
1278
1279         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1280         if (ret < 0) {
1281                 kfree(writers);
1282                 return ERR_PTR(ret);
1283         }
1284
1285         init_waitqueue_head(&writers->wait);
1286         return writers;
1287 }
1288
1289 static void
1290 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1291 {
1292         percpu_counter_destroy(&writers->counter);
1293         kfree(writers);
1294 }
1295
1296 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1297                          u64 objectid)
1298 {
1299         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1300         root->node = NULL;
1301         root->commit_root = NULL;
1302         root->state = 0;
1303         root->orphan_cleanup_state = 0;
1304
1305         root->objectid = objectid;
1306         root->last_trans = 0;
1307         root->highest_objectid = 0;
1308         root->nr_delalloc_inodes = 0;
1309         root->nr_ordered_extents = 0;
1310         root->name = NULL;
1311         root->inode_tree = RB_ROOT;
1312         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1313         root->block_rsv = NULL;
1314         root->orphan_block_rsv = NULL;
1315
1316         INIT_LIST_HEAD(&root->dirty_list);
1317         INIT_LIST_HEAD(&root->root_list);
1318         INIT_LIST_HEAD(&root->delalloc_inodes);
1319         INIT_LIST_HEAD(&root->delalloc_root);
1320         INIT_LIST_HEAD(&root->ordered_extents);
1321         INIT_LIST_HEAD(&root->ordered_root);
1322         INIT_LIST_HEAD(&root->logged_list[0]);
1323         INIT_LIST_HEAD(&root->logged_list[1]);
1324         spin_lock_init(&root->orphan_lock);
1325         spin_lock_init(&root->inode_lock);
1326         spin_lock_init(&root->delalloc_lock);
1327         spin_lock_init(&root->ordered_extent_lock);
1328         spin_lock_init(&root->accounting_lock);
1329         spin_lock_init(&root->log_extents_lock[0]);
1330         spin_lock_init(&root->log_extents_lock[1]);
1331         mutex_init(&root->objectid_mutex);
1332         mutex_init(&root->log_mutex);
1333         mutex_init(&root->ordered_extent_mutex);
1334         mutex_init(&root->delalloc_mutex);
1335         init_waitqueue_head(&root->log_writer_wait);
1336         init_waitqueue_head(&root->log_commit_wait[0]);
1337         init_waitqueue_head(&root->log_commit_wait[1]);
1338         INIT_LIST_HEAD(&root->log_ctxs[0]);
1339         INIT_LIST_HEAD(&root->log_ctxs[1]);
1340         atomic_set(&root->log_commit[0], 0);
1341         atomic_set(&root->log_commit[1], 0);
1342         atomic_set(&root->log_writers, 0);
1343         atomic_set(&root->log_batch, 0);
1344         atomic_set(&root->orphan_inodes, 0);
1345         atomic_set(&root->refs, 1);
1346         atomic_set(&root->will_be_snapshoted, 0);
1347         atomic_set(&root->qgroup_meta_rsv, 0);
1348         root->log_transid = 0;
1349         root->log_transid_committed = -1;
1350         root->last_log_commit = 0;
1351         if (!dummy)
1352                 extent_io_tree_init(&root->dirty_log_pages,
1353                                      fs_info->btree_inode->i_mapping);
1354
1355         memset(&root->root_key, 0, sizeof(root->root_key));
1356         memset(&root->root_item, 0, sizeof(root->root_item));
1357         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1358         if (!dummy)
1359                 root->defrag_trans_start = fs_info->generation;
1360         else
1361                 root->defrag_trans_start = 0;
1362         root->root_key.objectid = objectid;
1363         root->anon_dev = 0;
1364
1365         spin_lock_init(&root->root_item_lock);
1366 }
1367
1368 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1369                 gfp_t flags)
1370 {
1371         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1372         if (root)
1373                 root->fs_info = fs_info;
1374         return root;
1375 }
1376
1377 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1378 /* Should only be used by the testing infrastructure */
1379 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1380 {
1381         struct btrfs_root *root;
1382
1383         if (!fs_info)
1384                 return ERR_PTR(-EINVAL);
1385
1386         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1387         if (!root)
1388                 return ERR_PTR(-ENOMEM);
1389
1390         /* We don't use the stripesize in selftest, set it as sectorsize */
1391         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1392         root->alloc_bytenr = 0;
1393
1394         return root;
1395 }
1396 #endif
1397
1398 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1399                                      struct btrfs_fs_info *fs_info,
1400                                      u64 objectid)
1401 {
1402         struct extent_buffer *leaf;
1403         struct btrfs_root *tree_root = fs_info->tree_root;
1404         struct btrfs_root *root;
1405         struct btrfs_key key;
1406         int ret = 0;
1407         uuid_le uuid;
1408
1409         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1410         if (!root)
1411                 return ERR_PTR(-ENOMEM);
1412
1413         __setup_root(root, fs_info, objectid);
1414         root->root_key.objectid = objectid;
1415         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1416         root->root_key.offset = 0;
1417
1418         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1419         if (IS_ERR(leaf)) {
1420                 ret = PTR_ERR(leaf);
1421                 leaf = NULL;
1422                 goto fail;
1423         }
1424
1425         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1426         btrfs_set_header_bytenr(leaf, leaf->start);
1427         btrfs_set_header_generation(leaf, trans->transid);
1428         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1429         btrfs_set_header_owner(leaf, objectid);
1430         root->node = leaf;
1431
1432         write_extent_buffer_fsid(leaf, fs_info->fsid);
1433         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1434         btrfs_mark_buffer_dirty(leaf);
1435
1436         root->commit_root = btrfs_root_node(root);
1437         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1438
1439         root->root_item.flags = 0;
1440         root->root_item.byte_limit = 0;
1441         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1442         btrfs_set_root_generation(&root->root_item, trans->transid);
1443         btrfs_set_root_level(&root->root_item, 0);
1444         btrfs_set_root_refs(&root->root_item, 1);
1445         btrfs_set_root_used(&root->root_item, leaf->len);
1446         btrfs_set_root_last_snapshot(&root->root_item, 0);
1447         btrfs_set_root_dirid(&root->root_item, 0);
1448         uuid_le_gen(&uuid);
1449         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1450         root->root_item.drop_level = 0;
1451
1452         key.objectid = objectid;
1453         key.type = BTRFS_ROOT_ITEM_KEY;
1454         key.offset = 0;
1455         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1456         if (ret)
1457                 goto fail;
1458
1459         btrfs_tree_unlock(leaf);
1460
1461         return root;
1462
1463 fail:
1464         if (leaf) {
1465                 btrfs_tree_unlock(leaf);
1466                 free_extent_buffer(root->commit_root);
1467                 free_extent_buffer(leaf);
1468         }
1469         kfree(root);
1470
1471         return ERR_PTR(ret);
1472 }
1473
1474 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1475                                          struct btrfs_fs_info *fs_info)
1476 {
1477         struct btrfs_root *root;
1478         struct extent_buffer *leaf;
1479
1480         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1481         if (!root)
1482                 return ERR_PTR(-ENOMEM);
1483
1484         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1485
1486         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1487         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1488         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1489
1490         /*
1491          * DON'T set REF_COWS for log trees
1492          *
1493          * log trees do not get reference counted because they go away
1494          * before a real commit is actually done.  They do store pointers
1495          * to file data extents, and those reference counts still get
1496          * updated (along with back refs to the log tree).
1497          */
1498
1499         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1500                         NULL, 0, 0, 0);
1501         if (IS_ERR(leaf)) {
1502                 kfree(root);
1503                 return ERR_CAST(leaf);
1504         }
1505
1506         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1507         btrfs_set_header_bytenr(leaf, leaf->start);
1508         btrfs_set_header_generation(leaf, trans->transid);
1509         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1510         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1511         root->node = leaf;
1512
1513         write_extent_buffer_fsid(root->node, fs_info->fsid);
1514         btrfs_mark_buffer_dirty(root->node);
1515         btrfs_tree_unlock(root->node);
1516         return root;
1517 }
1518
1519 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1520                              struct btrfs_fs_info *fs_info)
1521 {
1522         struct btrfs_root *log_root;
1523
1524         log_root = alloc_log_tree(trans, fs_info);
1525         if (IS_ERR(log_root))
1526                 return PTR_ERR(log_root);
1527         WARN_ON(fs_info->log_root_tree);
1528         fs_info->log_root_tree = log_root;
1529         return 0;
1530 }
1531
1532 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1533                        struct btrfs_root *root)
1534 {
1535         struct btrfs_fs_info *fs_info = root->fs_info;
1536         struct btrfs_root *log_root;
1537         struct btrfs_inode_item *inode_item;
1538
1539         log_root = alloc_log_tree(trans, fs_info);
1540         if (IS_ERR(log_root))
1541                 return PTR_ERR(log_root);
1542
1543         log_root->last_trans = trans->transid;
1544         log_root->root_key.offset = root->root_key.objectid;
1545
1546         inode_item = &log_root->root_item.inode;
1547         btrfs_set_stack_inode_generation(inode_item, 1);
1548         btrfs_set_stack_inode_size(inode_item, 3);
1549         btrfs_set_stack_inode_nlink(inode_item, 1);
1550         btrfs_set_stack_inode_nbytes(inode_item,
1551                                      fs_info->nodesize);
1552         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1553
1554         btrfs_set_root_node(&log_root->root_item, log_root->node);
1555
1556         WARN_ON(root->log_root);
1557         root->log_root = log_root;
1558         root->log_transid = 0;
1559         root->log_transid_committed = -1;
1560         root->last_log_commit = 0;
1561         return 0;
1562 }
1563
1564 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1565                                                struct btrfs_key *key)
1566 {
1567         struct btrfs_root *root;
1568         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1569         struct btrfs_path *path;
1570         u64 generation;
1571         int ret;
1572
1573         path = btrfs_alloc_path();
1574         if (!path)
1575                 return ERR_PTR(-ENOMEM);
1576
1577         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1578         if (!root) {
1579                 ret = -ENOMEM;
1580                 goto alloc_fail;
1581         }
1582
1583         __setup_root(root, fs_info, key->objectid);
1584
1585         ret = btrfs_find_root(tree_root, key, path,
1586                               &root->root_item, &root->root_key);
1587         if (ret) {
1588                 if (ret > 0)
1589                         ret = -ENOENT;
1590                 goto find_fail;
1591         }
1592
1593         generation = btrfs_root_generation(&root->root_item);
1594         root->node = read_tree_block(fs_info,
1595                                      btrfs_root_bytenr(&root->root_item),
1596                                      generation);
1597         if (IS_ERR(root->node)) {
1598                 ret = PTR_ERR(root->node);
1599                 goto find_fail;
1600         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1601                 ret = -EIO;
1602                 free_extent_buffer(root->node);
1603                 goto find_fail;
1604         }
1605         root->commit_root = btrfs_root_node(root);
1606 out:
1607         btrfs_free_path(path);
1608         return root;
1609
1610 find_fail:
1611         kfree(root);
1612 alloc_fail:
1613         root = ERR_PTR(ret);
1614         goto out;
1615 }
1616
1617 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1618                                       struct btrfs_key *location)
1619 {
1620         struct btrfs_root *root;
1621
1622         root = btrfs_read_tree_root(tree_root, location);
1623         if (IS_ERR(root))
1624                 return root;
1625
1626         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1627                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1628                 btrfs_check_and_init_root_item(&root->root_item);
1629         }
1630
1631         return root;
1632 }
1633
1634 int btrfs_init_fs_root(struct btrfs_root *root)
1635 {
1636         int ret;
1637         struct btrfs_subvolume_writers *writers;
1638
1639         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1640         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1641                                         GFP_NOFS);
1642         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1643                 ret = -ENOMEM;
1644                 goto fail;
1645         }
1646
1647         writers = btrfs_alloc_subvolume_writers();
1648         if (IS_ERR(writers)) {
1649                 ret = PTR_ERR(writers);
1650                 goto fail;
1651         }
1652         root->subv_writers = writers;
1653
1654         btrfs_init_free_ino_ctl(root);
1655         spin_lock_init(&root->ino_cache_lock);
1656         init_waitqueue_head(&root->ino_cache_wait);
1657
1658         ret = get_anon_bdev(&root->anon_dev);
1659         if (ret)
1660                 goto fail;
1661
1662         mutex_lock(&root->objectid_mutex);
1663         ret = btrfs_find_highest_objectid(root,
1664                                         &root->highest_objectid);
1665         if (ret) {
1666                 mutex_unlock(&root->objectid_mutex);
1667                 goto fail;
1668         }
1669
1670         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1671
1672         mutex_unlock(&root->objectid_mutex);
1673
1674         return 0;
1675 fail:
1676         /* the caller is responsible to call free_fs_root */
1677         return ret;
1678 }
1679
1680 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1681                                         u64 root_id)
1682 {
1683         struct btrfs_root *root;
1684
1685         spin_lock(&fs_info->fs_roots_radix_lock);
1686         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1687                                  (unsigned long)root_id);
1688         spin_unlock(&fs_info->fs_roots_radix_lock);
1689         return root;
1690 }
1691
1692 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1693                          struct btrfs_root *root)
1694 {
1695         int ret;
1696
1697         ret = radix_tree_preload(GFP_NOFS);
1698         if (ret)
1699                 return ret;
1700
1701         spin_lock(&fs_info->fs_roots_radix_lock);
1702         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1703                                 (unsigned long)root->root_key.objectid,
1704                                 root);
1705         if (ret == 0)
1706                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1707         spin_unlock(&fs_info->fs_roots_radix_lock);
1708         radix_tree_preload_end();
1709
1710         return ret;
1711 }
1712
1713 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1714                                      struct btrfs_key *location,
1715                                      bool check_ref)
1716 {
1717         struct btrfs_root *root;
1718         struct btrfs_path *path;
1719         struct btrfs_key key;
1720         int ret;
1721
1722         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1723                 return fs_info->tree_root;
1724         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1725                 return fs_info->extent_root;
1726         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1727                 return fs_info->chunk_root;
1728         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1729                 return fs_info->dev_root;
1730         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1731                 return fs_info->csum_root;
1732         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1733                 return fs_info->quota_root ? fs_info->quota_root :
1734                                              ERR_PTR(-ENOENT);
1735         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1736                 return fs_info->uuid_root ? fs_info->uuid_root :
1737                                             ERR_PTR(-ENOENT);
1738         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1739                 return fs_info->free_space_root ? fs_info->free_space_root :
1740                                                   ERR_PTR(-ENOENT);
1741 again:
1742         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1743         if (root) {
1744                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1745                         return ERR_PTR(-ENOENT);
1746                 return root;
1747         }
1748
1749         root = btrfs_read_fs_root(fs_info->tree_root, location);
1750         if (IS_ERR(root))
1751                 return root;
1752
1753         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1754                 ret = -ENOENT;
1755                 goto fail;
1756         }
1757
1758         ret = btrfs_init_fs_root(root);
1759         if (ret)
1760                 goto fail;
1761
1762         path = btrfs_alloc_path();
1763         if (!path) {
1764                 ret = -ENOMEM;
1765                 goto fail;
1766         }
1767         key.objectid = BTRFS_ORPHAN_OBJECTID;
1768         key.type = BTRFS_ORPHAN_ITEM_KEY;
1769         key.offset = location->objectid;
1770
1771         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1772         btrfs_free_path(path);
1773         if (ret < 0)
1774                 goto fail;
1775         if (ret == 0)
1776                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1777
1778         ret = btrfs_insert_fs_root(fs_info, root);
1779         if (ret) {
1780                 if (ret == -EEXIST) {
1781                         free_fs_root(root);
1782                         goto again;
1783                 }
1784                 goto fail;
1785         }
1786         return root;
1787 fail:
1788         free_fs_root(root);
1789         return ERR_PTR(ret);
1790 }
1791
1792 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1793 {
1794         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1795         int ret = 0;
1796         struct btrfs_device *device;
1797         struct backing_dev_info *bdi;
1798
1799         rcu_read_lock();
1800         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1801                 if (!device->bdev)
1802                         continue;
1803                 bdi = blk_get_backing_dev_info(device->bdev);
1804                 if (bdi_congested(bdi, bdi_bits)) {
1805                         ret = 1;
1806                         break;
1807                 }
1808         }
1809         rcu_read_unlock();
1810         return ret;
1811 }
1812
1813 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1814 {
1815         int err;
1816
1817         err = bdi_setup_and_register(bdi, "btrfs");
1818         if (err)
1819                 return err;
1820
1821         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1822         bdi->congested_fn       = btrfs_congested_fn;
1823         bdi->congested_data     = info;
1824         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1825         return 0;
1826 }
1827
1828 /*
1829  * called by the kthread helper functions to finally call the bio end_io
1830  * functions.  This is where read checksum verification actually happens
1831  */
1832 static void end_workqueue_fn(struct btrfs_work *work)
1833 {
1834         struct bio *bio;
1835         struct btrfs_end_io_wq *end_io_wq;
1836
1837         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1838         bio = end_io_wq->bio;
1839
1840         bio->bi_error = end_io_wq->error;
1841         bio->bi_private = end_io_wq->private;
1842         bio->bi_end_io = end_io_wq->end_io;
1843         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1844         bio_endio(bio);
1845 }
1846
1847 static int cleaner_kthread(void *arg)
1848 {
1849         struct btrfs_root *root = arg;
1850         struct btrfs_fs_info *fs_info = root->fs_info;
1851         int again;
1852         struct btrfs_trans_handle *trans;
1853
1854         do {
1855                 again = 0;
1856
1857                 /* Make the cleaner go to sleep early. */
1858                 if (btrfs_need_cleaner_sleep(fs_info))
1859                         goto sleep;
1860
1861                 /*
1862                  * Do not do anything if we might cause open_ctree() to block
1863                  * before we have finished mounting the filesystem.
1864                  */
1865                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1866                         goto sleep;
1867
1868                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1869                         goto sleep;
1870
1871                 /*
1872                  * Avoid the problem that we change the status of the fs
1873                  * during the above check and trylock.
1874                  */
1875                 if (btrfs_need_cleaner_sleep(fs_info)) {
1876                         mutex_unlock(&fs_info->cleaner_mutex);
1877                         goto sleep;
1878                 }
1879
1880                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1881                 btrfs_run_delayed_iputs(fs_info);
1882                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1883
1884                 again = btrfs_clean_one_deleted_snapshot(root);
1885                 mutex_unlock(&fs_info->cleaner_mutex);
1886
1887                 /*
1888                  * The defragger has dealt with the R/O remount and umount,
1889                  * needn't do anything special here.
1890                  */
1891                 btrfs_run_defrag_inodes(fs_info);
1892
1893                 /*
1894                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1895                  * with relocation (btrfs_relocate_chunk) and relocation
1896                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1897                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1898                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1899                  * unused block groups.
1900                  */
1901                 btrfs_delete_unused_bgs(fs_info);
1902 sleep:
1903                 if (!again) {
1904                         set_current_state(TASK_INTERRUPTIBLE);
1905                         if (!kthread_should_stop())
1906                                 schedule();
1907                         __set_current_state(TASK_RUNNING);
1908                 }
1909         } while (!kthread_should_stop());
1910
1911         /*
1912          * Transaction kthread is stopped before us and wakes us up.
1913          * However we might have started a new transaction and COWed some
1914          * tree blocks when deleting unused block groups for example. So
1915          * make sure we commit the transaction we started to have a clean
1916          * shutdown when evicting the btree inode - if it has dirty pages
1917          * when we do the final iput() on it, eviction will trigger a
1918          * writeback for it which will fail with null pointer dereferences
1919          * since work queues and other resources were already released and
1920          * destroyed by the time the iput/eviction/writeback is made.
1921          */
1922         trans = btrfs_attach_transaction(root);
1923         if (IS_ERR(trans)) {
1924                 if (PTR_ERR(trans) != -ENOENT)
1925                         btrfs_err(fs_info,
1926                                   "cleaner transaction attach returned %ld",
1927                                   PTR_ERR(trans));
1928         } else {
1929                 int ret;
1930
1931                 ret = btrfs_commit_transaction(trans);
1932                 if (ret)
1933                         btrfs_err(fs_info,
1934                                   "cleaner open transaction commit returned %d",
1935                                   ret);
1936         }
1937
1938         return 0;
1939 }
1940
1941 static int transaction_kthread(void *arg)
1942 {
1943         struct btrfs_root *root = arg;
1944         struct btrfs_fs_info *fs_info = root->fs_info;
1945         struct btrfs_trans_handle *trans;
1946         struct btrfs_transaction *cur;
1947         u64 transid;
1948         unsigned long now;
1949         unsigned long delay;
1950         bool cannot_commit;
1951
1952         do {
1953                 cannot_commit = false;
1954                 delay = HZ * fs_info->commit_interval;
1955                 mutex_lock(&fs_info->transaction_kthread_mutex);
1956
1957                 spin_lock(&fs_info->trans_lock);
1958                 cur = fs_info->running_transaction;
1959                 if (!cur) {
1960                         spin_unlock(&fs_info->trans_lock);
1961                         goto sleep;
1962                 }
1963
1964                 now = get_seconds();
1965                 if (cur->state < TRANS_STATE_BLOCKED &&
1966                     (now < cur->start_time ||
1967                      now - cur->start_time < fs_info->commit_interval)) {
1968                         spin_unlock(&fs_info->trans_lock);
1969                         delay = HZ * 5;
1970                         goto sleep;
1971                 }
1972                 transid = cur->transid;
1973                 spin_unlock(&fs_info->trans_lock);
1974
1975                 /* If the file system is aborted, this will always fail. */
1976                 trans = btrfs_attach_transaction(root);
1977                 if (IS_ERR(trans)) {
1978                         if (PTR_ERR(trans) != -ENOENT)
1979                                 cannot_commit = true;
1980                         goto sleep;
1981                 }
1982                 if (transid == trans->transid) {
1983                         btrfs_commit_transaction(trans);
1984                 } else {
1985                         btrfs_end_transaction(trans);
1986                 }
1987 sleep:
1988                 wake_up_process(fs_info->cleaner_kthread);
1989                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1990
1991                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1992                                       &fs_info->fs_state)))
1993                         btrfs_cleanup_transaction(fs_info);
1994                 set_current_state(TASK_INTERRUPTIBLE);
1995                 if (!kthread_should_stop() &&
1996                                 (!btrfs_transaction_blocked(fs_info) ||
1997                                  cannot_commit))
1998                         schedule_timeout(delay);
1999                 __set_current_state(TASK_RUNNING);
2000         } while (!kthread_should_stop());
2001         return 0;
2002 }
2003
2004 /*
2005  * this will find the highest generation in the array of
2006  * root backups.  The index of the highest array is returned,
2007  * or -1 if we can't find anything.
2008  *
2009  * We check to make sure the array is valid by comparing the
2010  * generation of the latest  root in the array with the generation
2011  * in the super block.  If they don't match we pitch it.
2012  */
2013 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
2014 {
2015         u64 cur;
2016         int newest_index = -1;
2017         struct btrfs_root_backup *root_backup;
2018         int i;
2019
2020         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2021                 root_backup = info->super_copy->super_roots + i;
2022                 cur = btrfs_backup_tree_root_gen(root_backup);
2023                 if (cur == newest_gen)
2024                         newest_index = i;
2025         }
2026
2027         /* check to see if we actually wrapped around */
2028         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2029                 root_backup = info->super_copy->super_roots;
2030                 cur = btrfs_backup_tree_root_gen(root_backup);
2031                 if (cur == newest_gen)
2032                         newest_index = 0;
2033         }
2034         return newest_index;
2035 }
2036
2037
2038 /*
2039  * find the oldest backup so we know where to store new entries
2040  * in the backup array.  This will set the backup_root_index
2041  * field in the fs_info struct
2042  */
2043 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2044                                      u64 newest_gen)
2045 {
2046         int newest_index = -1;
2047
2048         newest_index = find_newest_super_backup(info, newest_gen);
2049         /* if there was garbage in there, just move along */
2050         if (newest_index == -1) {
2051                 info->backup_root_index = 0;
2052         } else {
2053                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2054         }
2055 }
2056
2057 /*
2058  * copy all the root pointers into the super backup array.
2059  * this will bump the backup pointer by one when it is
2060  * done
2061  */
2062 static void backup_super_roots(struct btrfs_fs_info *info)
2063 {
2064         int next_backup;
2065         struct btrfs_root_backup *root_backup;
2066         int last_backup;
2067
2068         next_backup = info->backup_root_index;
2069         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2070                 BTRFS_NUM_BACKUP_ROOTS;
2071
2072         /*
2073          * just overwrite the last backup if we're at the same generation
2074          * this happens only at umount
2075          */
2076         root_backup = info->super_for_commit->super_roots + last_backup;
2077         if (btrfs_backup_tree_root_gen(root_backup) ==
2078             btrfs_header_generation(info->tree_root->node))
2079                 next_backup = last_backup;
2080
2081         root_backup = info->super_for_commit->super_roots + next_backup;
2082
2083         /*
2084          * make sure all of our padding and empty slots get zero filled
2085          * regardless of which ones we use today
2086          */
2087         memset(root_backup, 0, sizeof(*root_backup));
2088
2089         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2090
2091         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2092         btrfs_set_backup_tree_root_gen(root_backup,
2093                                btrfs_header_generation(info->tree_root->node));
2094
2095         btrfs_set_backup_tree_root_level(root_backup,
2096                                btrfs_header_level(info->tree_root->node));
2097
2098         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2099         btrfs_set_backup_chunk_root_gen(root_backup,
2100                                btrfs_header_generation(info->chunk_root->node));
2101         btrfs_set_backup_chunk_root_level(root_backup,
2102                                btrfs_header_level(info->chunk_root->node));
2103
2104         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2105         btrfs_set_backup_extent_root_gen(root_backup,
2106                                btrfs_header_generation(info->extent_root->node));
2107         btrfs_set_backup_extent_root_level(root_backup,
2108                                btrfs_header_level(info->extent_root->node));
2109
2110         /*
2111          * we might commit during log recovery, which happens before we set
2112          * the fs_root.  Make sure it is valid before we fill it in.
2113          */
2114         if (info->fs_root && info->fs_root->node) {
2115                 btrfs_set_backup_fs_root(root_backup,
2116                                          info->fs_root->node->start);
2117                 btrfs_set_backup_fs_root_gen(root_backup,
2118                                btrfs_header_generation(info->fs_root->node));
2119                 btrfs_set_backup_fs_root_level(root_backup,
2120                                btrfs_header_level(info->fs_root->node));
2121         }
2122
2123         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2124         btrfs_set_backup_dev_root_gen(root_backup,
2125                                btrfs_header_generation(info->dev_root->node));
2126         btrfs_set_backup_dev_root_level(root_backup,
2127                                        btrfs_header_level(info->dev_root->node));
2128
2129         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2130         btrfs_set_backup_csum_root_gen(root_backup,
2131                                btrfs_header_generation(info->csum_root->node));
2132         btrfs_set_backup_csum_root_level(root_backup,
2133                                btrfs_header_level(info->csum_root->node));
2134
2135         btrfs_set_backup_total_bytes(root_backup,
2136                              btrfs_super_total_bytes(info->super_copy));
2137         btrfs_set_backup_bytes_used(root_backup,
2138                              btrfs_super_bytes_used(info->super_copy));
2139         btrfs_set_backup_num_devices(root_backup,
2140                              btrfs_super_num_devices(info->super_copy));
2141
2142         /*
2143          * if we don't copy this out to the super_copy, it won't get remembered
2144          * for the next commit
2145          */
2146         memcpy(&info->super_copy->super_roots,
2147                &info->super_for_commit->super_roots,
2148                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2149 }
2150
2151 /*
2152  * this copies info out of the root backup array and back into
2153  * the in-memory super block.  It is meant to help iterate through
2154  * the array, so you send it the number of backups you've already
2155  * tried and the last backup index you used.
2156  *
2157  * this returns -1 when it has tried all the backups
2158  */
2159 static noinline int next_root_backup(struct btrfs_fs_info *info,
2160                                      struct btrfs_super_block *super,
2161                                      int *num_backups_tried, int *backup_index)
2162 {
2163         struct btrfs_root_backup *root_backup;
2164         int newest = *backup_index;
2165
2166         if (*num_backups_tried == 0) {
2167                 u64 gen = btrfs_super_generation(super);
2168
2169                 newest = find_newest_super_backup(info, gen);
2170                 if (newest == -1)
2171                         return -1;
2172
2173                 *backup_index = newest;
2174                 *num_backups_tried = 1;
2175         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2176                 /* we've tried all the backups, all done */
2177                 return -1;
2178         } else {
2179                 /* jump to the next oldest backup */
2180                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2181                         BTRFS_NUM_BACKUP_ROOTS;
2182                 *backup_index = newest;
2183                 *num_backups_tried += 1;
2184         }
2185         root_backup = super->super_roots + newest;
2186
2187         btrfs_set_super_generation(super,
2188                                    btrfs_backup_tree_root_gen(root_backup));
2189         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2190         btrfs_set_super_root_level(super,
2191                                    btrfs_backup_tree_root_level(root_backup));
2192         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2193
2194         /*
2195          * fixme: the total bytes and num_devices need to match or we should
2196          * need a fsck
2197          */
2198         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2199         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2200         return 0;
2201 }
2202
2203 /* helper to cleanup workers */
2204 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2205 {
2206         btrfs_destroy_workqueue(fs_info->fixup_workers);
2207         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2208         btrfs_destroy_workqueue(fs_info->workers);
2209         btrfs_destroy_workqueue(fs_info->endio_workers);
2210         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2211         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2212         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2213         btrfs_destroy_workqueue(fs_info->rmw_workers);
2214         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2215         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2216         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2217         btrfs_destroy_workqueue(fs_info->submit_workers);
2218         btrfs_destroy_workqueue(fs_info->delayed_workers);
2219         btrfs_destroy_workqueue(fs_info->caching_workers);
2220         btrfs_destroy_workqueue(fs_info->readahead_workers);
2221         btrfs_destroy_workqueue(fs_info->flush_workers);
2222         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2223         btrfs_destroy_workqueue(fs_info->extent_workers);
2224 }
2225
2226 static void free_root_extent_buffers(struct btrfs_root *root)
2227 {
2228         if (root) {
2229                 free_extent_buffer(root->node);
2230                 free_extent_buffer(root->commit_root);
2231                 root->node = NULL;
2232                 root->commit_root = NULL;
2233         }
2234 }
2235
2236 /* helper to cleanup tree roots */
2237 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2238 {
2239         free_root_extent_buffers(info->tree_root);
2240
2241         free_root_extent_buffers(info->dev_root);
2242         free_root_extent_buffers(info->extent_root);
2243         free_root_extent_buffers(info->csum_root);
2244         free_root_extent_buffers(info->quota_root);
2245         free_root_extent_buffers(info->uuid_root);
2246         if (chunk_root)
2247                 free_root_extent_buffers(info->chunk_root);
2248         free_root_extent_buffers(info->free_space_root);
2249 }
2250
2251 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2252 {
2253         int ret;
2254         struct btrfs_root *gang[8];
2255         int i;
2256
2257         while (!list_empty(&fs_info->dead_roots)) {
2258                 gang[0] = list_entry(fs_info->dead_roots.next,
2259                                      struct btrfs_root, root_list);
2260                 list_del(&gang[0]->root_list);
2261
2262                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2263                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2264                 } else {
2265                         free_extent_buffer(gang[0]->node);
2266                         free_extent_buffer(gang[0]->commit_root);
2267                         btrfs_put_fs_root(gang[0]);
2268                 }
2269         }
2270
2271         while (1) {
2272                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2273                                              (void **)gang, 0,
2274                                              ARRAY_SIZE(gang));
2275                 if (!ret)
2276                         break;
2277                 for (i = 0; i < ret; i++)
2278                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2279         }
2280
2281         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2282                 btrfs_free_log_root_tree(NULL, fs_info);
2283                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2284         }
2285 }
2286
2287 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2288 {
2289         mutex_init(&fs_info->scrub_lock);
2290         atomic_set(&fs_info->scrubs_running, 0);
2291         atomic_set(&fs_info->scrub_pause_req, 0);
2292         atomic_set(&fs_info->scrubs_paused, 0);
2293         atomic_set(&fs_info->scrub_cancel_req, 0);
2294         init_waitqueue_head(&fs_info->scrub_pause_wait);
2295         fs_info->scrub_workers_refcnt = 0;
2296 }
2297
2298 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2299 {
2300         spin_lock_init(&fs_info->balance_lock);
2301         mutex_init(&fs_info->balance_mutex);
2302         atomic_set(&fs_info->balance_running, 0);
2303         atomic_set(&fs_info->balance_pause_req, 0);
2304         atomic_set(&fs_info->balance_cancel_req, 0);
2305         fs_info->balance_ctl = NULL;
2306         init_waitqueue_head(&fs_info->balance_wait_q);
2307 }
2308
2309 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2310 {
2311         struct inode *inode = fs_info->btree_inode;
2312
2313         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2314         set_nlink(inode, 1);
2315         /*
2316          * we set the i_size on the btree inode to the max possible int.
2317          * the real end of the address space is determined by all of
2318          * the devices in the system
2319          */
2320         inode->i_size = OFFSET_MAX;
2321         inode->i_mapping->a_ops = &btree_aops;
2322
2323         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2324         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode->i_mapping);
2325         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2326         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2327
2328         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2329
2330         BTRFS_I(inode)->root = fs_info->tree_root;
2331         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2332         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2333         btrfs_insert_inode_hash(inode);
2334 }
2335
2336 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2337 {
2338         fs_info->dev_replace.lock_owner = 0;
2339         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2340         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2341         rwlock_init(&fs_info->dev_replace.lock);
2342         atomic_set(&fs_info->dev_replace.read_locks, 0);
2343         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2344         init_waitqueue_head(&fs_info->replace_wait);
2345         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2346 }
2347
2348 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2349 {
2350         spin_lock_init(&fs_info->qgroup_lock);
2351         mutex_init(&fs_info->qgroup_ioctl_lock);
2352         fs_info->qgroup_tree = RB_ROOT;
2353         fs_info->qgroup_op_tree = RB_ROOT;
2354         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2355         fs_info->qgroup_seq = 1;
2356         fs_info->qgroup_ulist = NULL;
2357         fs_info->qgroup_rescan_running = false;
2358         mutex_init(&fs_info->qgroup_rescan_lock);
2359 }
2360
2361 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2362                 struct btrfs_fs_devices *fs_devices)
2363 {
2364         int max_active = fs_info->thread_pool_size;
2365         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2366
2367         fs_info->workers =
2368                 btrfs_alloc_workqueue(fs_info, "worker",
2369                                       flags | WQ_HIGHPRI, max_active, 16);
2370
2371         fs_info->delalloc_workers =
2372                 btrfs_alloc_workqueue(fs_info, "delalloc",
2373                                       flags, max_active, 2);
2374
2375         fs_info->flush_workers =
2376                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2377                                       flags, max_active, 0);
2378
2379         fs_info->caching_workers =
2380                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2381
2382         /*
2383          * a higher idle thresh on the submit workers makes it much more
2384          * likely that bios will be send down in a sane order to the
2385          * devices
2386          */
2387         fs_info->submit_workers =
2388                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2389                                       min_t(u64, fs_devices->num_devices,
2390                                             max_active), 64);
2391
2392         fs_info->fixup_workers =
2393                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2394
2395         /*
2396          * endios are largely parallel and should have a very
2397          * low idle thresh
2398          */
2399         fs_info->endio_workers =
2400                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2401         fs_info->endio_meta_workers =
2402                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2403                                       max_active, 4);
2404         fs_info->endio_meta_write_workers =
2405                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2406                                       max_active, 2);
2407         fs_info->endio_raid56_workers =
2408                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2409                                       max_active, 4);
2410         fs_info->endio_repair_workers =
2411                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2412         fs_info->rmw_workers =
2413                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2414         fs_info->endio_write_workers =
2415                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2416                                       max_active, 2);
2417         fs_info->endio_freespace_worker =
2418                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2419                                       max_active, 0);
2420         fs_info->delayed_workers =
2421                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2422                                       max_active, 0);
2423         fs_info->readahead_workers =
2424                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2425                                       max_active, 2);
2426         fs_info->qgroup_rescan_workers =
2427                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2428         fs_info->extent_workers =
2429                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2430                                       min_t(u64, fs_devices->num_devices,
2431                                             max_active), 8);
2432
2433         if (!(fs_info->workers && fs_info->delalloc_workers &&
2434               fs_info->submit_workers && fs_info->flush_workers &&
2435               fs_info->endio_workers && fs_info->endio_meta_workers &&
2436               fs_info->endio_meta_write_workers &&
2437               fs_info->endio_repair_workers &&
2438               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2439               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2440               fs_info->caching_workers && fs_info->readahead_workers &&
2441               fs_info->fixup_workers && fs_info->delayed_workers &&
2442               fs_info->extent_workers &&
2443               fs_info->qgroup_rescan_workers)) {
2444                 return -ENOMEM;
2445         }
2446
2447         return 0;
2448 }
2449
2450 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2451                             struct btrfs_fs_devices *fs_devices)
2452 {
2453         int ret;
2454         struct btrfs_root *log_tree_root;
2455         struct btrfs_super_block *disk_super = fs_info->super_copy;
2456         u64 bytenr = btrfs_super_log_root(disk_super);
2457
2458         if (fs_devices->rw_devices == 0) {
2459                 btrfs_warn(fs_info, "log replay required on RO media");
2460                 return -EIO;
2461         }
2462
2463         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2464         if (!log_tree_root)
2465                 return -ENOMEM;
2466
2467         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2468
2469         log_tree_root->node = read_tree_block(fs_info, bytenr,
2470                                               fs_info->generation + 1);
2471         if (IS_ERR(log_tree_root->node)) {
2472                 btrfs_warn(fs_info, "failed to read log tree");
2473                 ret = PTR_ERR(log_tree_root->node);
2474                 kfree(log_tree_root);
2475                 return ret;
2476         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2477                 btrfs_err(fs_info, "failed to read log tree");
2478                 free_extent_buffer(log_tree_root->node);
2479                 kfree(log_tree_root);
2480                 return -EIO;
2481         }
2482         /* returns with log_tree_root freed on success */
2483         ret = btrfs_recover_log_trees(log_tree_root);
2484         if (ret) {
2485                 btrfs_handle_fs_error(fs_info, ret,
2486                                       "Failed to recover log tree");
2487                 free_extent_buffer(log_tree_root->node);
2488                 kfree(log_tree_root);
2489                 return ret;
2490         }
2491
2492         if (fs_info->sb->s_flags & MS_RDONLY) {
2493                 ret = btrfs_commit_super(fs_info);
2494                 if (ret)
2495                         return ret;
2496         }
2497
2498         return 0;
2499 }
2500
2501 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2502 {
2503         struct btrfs_root *tree_root = fs_info->tree_root;
2504         struct btrfs_root *root;
2505         struct btrfs_key location;
2506         int ret;
2507
2508         BUG_ON(!fs_info->tree_root);
2509
2510         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2511         location.type = BTRFS_ROOT_ITEM_KEY;
2512         location.offset = 0;
2513
2514         root = btrfs_read_tree_root(tree_root, &location);
2515         if (IS_ERR(root))
2516                 return PTR_ERR(root);
2517         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2518         fs_info->extent_root = root;
2519
2520         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2521         root = btrfs_read_tree_root(tree_root, &location);
2522         if (IS_ERR(root))
2523                 return PTR_ERR(root);
2524         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2525         fs_info->dev_root = root;
2526         btrfs_init_devices_late(fs_info);
2527
2528         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2529         root = btrfs_read_tree_root(tree_root, &location);
2530         if (IS_ERR(root))
2531                 return PTR_ERR(root);
2532         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2533         fs_info->csum_root = root;
2534
2535         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2536         root = btrfs_read_tree_root(tree_root, &location);
2537         if (!IS_ERR(root)) {
2538                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2539                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2540                 fs_info->quota_root = root;
2541         }
2542
2543         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2544         root = btrfs_read_tree_root(tree_root, &location);
2545         if (IS_ERR(root)) {
2546                 ret = PTR_ERR(root);
2547                 if (ret != -ENOENT)
2548                         return ret;
2549         } else {
2550                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2551                 fs_info->uuid_root = root;
2552         }
2553
2554         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2555                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2556                 root = btrfs_read_tree_root(tree_root, &location);
2557                 if (IS_ERR(root))
2558                         return PTR_ERR(root);
2559                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2560                 fs_info->free_space_root = root;
2561         }
2562
2563         return 0;
2564 }
2565
2566 int open_ctree(struct super_block *sb,
2567                struct btrfs_fs_devices *fs_devices,
2568                char *options)
2569 {
2570         u32 sectorsize;
2571         u32 nodesize;
2572         u32 stripesize;
2573         u64 generation;
2574         u64 features;
2575         struct btrfs_key location;
2576         struct buffer_head *bh;
2577         struct btrfs_super_block *disk_super;
2578         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2579         struct btrfs_root *tree_root;
2580         struct btrfs_root *chunk_root;
2581         int ret;
2582         int err = -EINVAL;
2583         int num_backups_tried = 0;
2584         int backup_index = 0;
2585         int max_active;
2586         int clear_free_space_tree = 0;
2587
2588         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2589         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2590         if (!tree_root || !chunk_root) {
2591                 err = -ENOMEM;
2592                 goto fail;
2593         }
2594
2595         ret = init_srcu_struct(&fs_info->subvol_srcu);
2596         if (ret) {
2597                 err = ret;
2598                 goto fail;
2599         }
2600
2601         ret = setup_bdi(fs_info, &fs_info->bdi);
2602         if (ret) {
2603                 err = ret;
2604                 goto fail_srcu;
2605         }
2606
2607         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2608         if (ret) {
2609                 err = ret;
2610                 goto fail_bdi;
2611         }
2612         fs_info->dirty_metadata_batch = PAGE_SIZE *
2613                                         (1 + ilog2(nr_cpu_ids));
2614
2615         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2616         if (ret) {
2617                 err = ret;
2618                 goto fail_dirty_metadata_bytes;
2619         }
2620
2621         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2622         if (ret) {
2623                 err = ret;
2624                 goto fail_delalloc_bytes;
2625         }
2626
2627         fs_info->btree_inode = new_inode(sb);
2628         if (!fs_info->btree_inode) {
2629                 err = -ENOMEM;
2630                 goto fail_bio_counter;
2631         }
2632
2633         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2634
2635         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2636         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2637         INIT_LIST_HEAD(&fs_info->trans_list);
2638         INIT_LIST_HEAD(&fs_info->dead_roots);
2639         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2640         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2641         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2642         spin_lock_init(&fs_info->delalloc_root_lock);
2643         spin_lock_init(&fs_info->trans_lock);
2644         spin_lock_init(&fs_info->fs_roots_radix_lock);
2645         spin_lock_init(&fs_info->delayed_iput_lock);
2646         spin_lock_init(&fs_info->defrag_inodes_lock);
2647         spin_lock_init(&fs_info->free_chunk_lock);
2648         spin_lock_init(&fs_info->tree_mod_seq_lock);
2649         spin_lock_init(&fs_info->super_lock);
2650         spin_lock_init(&fs_info->qgroup_op_lock);
2651         spin_lock_init(&fs_info->buffer_lock);
2652         spin_lock_init(&fs_info->unused_bgs_lock);
2653         rwlock_init(&fs_info->tree_mod_log_lock);
2654         mutex_init(&fs_info->unused_bg_unpin_mutex);
2655         mutex_init(&fs_info->delete_unused_bgs_mutex);
2656         mutex_init(&fs_info->reloc_mutex);
2657         mutex_init(&fs_info->delalloc_root_mutex);
2658         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2659         seqlock_init(&fs_info->profiles_lock);
2660
2661         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2662         INIT_LIST_HEAD(&fs_info->space_info);
2663         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2664         INIT_LIST_HEAD(&fs_info->unused_bgs);
2665         btrfs_mapping_init(&fs_info->mapping_tree);
2666         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2667                              BTRFS_BLOCK_RSV_GLOBAL);
2668         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2669                              BTRFS_BLOCK_RSV_DELALLOC);
2670         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2671         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2672         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2673         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2674                              BTRFS_BLOCK_RSV_DELOPS);
2675         atomic_set(&fs_info->nr_async_submits, 0);
2676         atomic_set(&fs_info->async_delalloc_pages, 0);
2677         atomic_set(&fs_info->async_submit_draining, 0);
2678         atomic_set(&fs_info->nr_async_bios, 0);
2679         atomic_set(&fs_info->defrag_running, 0);
2680         atomic_set(&fs_info->qgroup_op_seq, 0);
2681         atomic_set(&fs_info->reada_works_cnt, 0);
2682         atomic64_set(&fs_info->tree_mod_seq, 0);
2683         fs_info->fs_frozen = 0;
2684         fs_info->sb = sb;
2685         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2686         fs_info->metadata_ratio = 0;
2687         fs_info->defrag_inodes = RB_ROOT;
2688         fs_info->free_chunk_space = 0;
2689         fs_info->tree_mod_log = RB_ROOT;
2690         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2691         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2692         /* readahead state */
2693         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2694         spin_lock_init(&fs_info->reada_lock);
2695
2696         fs_info->thread_pool_size = min_t(unsigned long,
2697                                           num_online_cpus() + 2, 8);
2698
2699         INIT_LIST_HEAD(&fs_info->ordered_roots);
2700         spin_lock_init(&fs_info->ordered_root_lock);
2701         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2702                                         GFP_KERNEL);
2703         if (!fs_info->delayed_root) {
2704                 err = -ENOMEM;
2705                 goto fail_iput;
2706         }
2707         btrfs_init_delayed_root(fs_info->delayed_root);
2708
2709         btrfs_init_scrub(fs_info);
2710 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2711         fs_info->check_integrity_print_mask = 0;
2712 #endif
2713         btrfs_init_balance(fs_info);
2714         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2715
2716         sb->s_blocksize = 4096;
2717         sb->s_blocksize_bits = blksize_bits(4096);
2718         sb->s_bdi = &fs_info->bdi;
2719
2720         btrfs_init_btree_inode(fs_info);
2721
2722         spin_lock_init(&fs_info->block_group_cache_lock);
2723         fs_info->block_group_cache_tree = RB_ROOT;
2724         fs_info->first_logical_byte = (u64)-1;
2725
2726         extent_io_tree_init(&fs_info->freed_extents[0],
2727                              fs_info->btree_inode->i_mapping);
2728         extent_io_tree_init(&fs_info->freed_extents[1],
2729                              fs_info->btree_inode->i_mapping);
2730         fs_info->pinned_extents = &fs_info->freed_extents[0];
2731         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2732
2733         mutex_init(&fs_info->ordered_operations_mutex);
2734         mutex_init(&fs_info->tree_log_mutex);
2735         mutex_init(&fs_info->chunk_mutex);
2736         mutex_init(&fs_info->transaction_kthread_mutex);
2737         mutex_init(&fs_info->cleaner_mutex);
2738         mutex_init(&fs_info->volume_mutex);
2739         mutex_init(&fs_info->ro_block_group_mutex);
2740         init_rwsem(&fs_info->commit_root_sem);
2741         init_rwsem(&fs_info->cleanup_work_sem);
2742         init_rwsem(&fs_info->subvol_sem);
2743         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2744
2745         btrfs_init_dev_replace_locks(fs_info);
2746         btrfs_init_qgroup(fs_info);
2747
2748         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2749         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2750
2751         init_waitqueue_head(&fs_info->transaction_throttle);
2752         init_waitqueue_head(&fs_info->transaction_wait);
2753         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2754         init_waitqueue_head(&fs_info->async_submit_wait);
2755
2756         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2757
2758         /* Usable values until the real ones are cached from the superblock */
2759         fs_info->nodesize = 4096;
2760         fs_info->sectorsize = 4096;
2761         fs_info->stripesize = 4096;
2762
2763         ret = btrfs_alloc_stripe_hash_table(fs_info);
2764         if (ret) {
2765                 err = ret;
2766                 goto fail_alloc;
2767         }
2768
2769         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2770
2771         invalidate_bdev(fs_devices->latest_bdev);
2772
2773         /*
2774          * Read super block and check the signature bytes only
2775          */
2776         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2777         if (IS_ERR(bh)) {
2778                 err = PTR_ERR(bh);
2779                 goto fail_alloc;
2780         }
2781
2782         /*
2783          * We want to check superblock checksum, the type is stored inside.
2784          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2785          */
2786         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2787                 btrfs_err(fs_info, "superblock checksum mismatch");
2788                 err = -EINVAL;
2789                 brelse(bh);
2790                 goto fail_alloc;
2791         }
2792
2793         /*
2794          * super_copy is zeroed at allocation time and we never touch the
2795          * following bytes up to INFO_SIZE, the checksum is calculated from
2796          * the whole block of INFO_SIZE
2797          */
2798         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2799         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2800                sizeof(*fs_info->super_for_commit));
2801         brelse(bh);
2802
2803         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2804
2805         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2806         if (ret) {
2807                 btrfs_err(fs_info, "superblock contains fatal errors");
2808                 err = -EINVAL;
2809                 goto fail_alloc;
2810         }
2811
2812         disk_super = fs_info->super_copy;
2813         if (!btrfs_super_root(disk_super))
2814                 goto fail_alloc;
2815
2816         /* check FS state, whether FS is broken. */
2817         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2818                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2819
2820         /*
2821          * run through our array of backup supers and setup
2822          * our ring pointer to the oldest one
2823          */
2824         generation = btrfs_super_generation(disk_super);
2825         find_oldest_super_backup(fs_info, generation);
2826
2827         /*
2828          * In the long term, we'll store the compression type in the super
2829          * block, and it'll be used for per file compression control.
2830          */
2831         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2832
2833         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2834         if (ret) {
2835                 err = ret;
2836                 goto fail_alloc;
2837         }
2838
2839         features = btrfs_super_incompat_flags(disk_super) &
2840                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2841         if (features) {
2842                 btrfs_err(fs_info,
2843                     "cannot mount because of unsupported optional features (%llx)",
2844                     features);
2845                 err = -EINVAL;
2846                 goto fail_alloc;
2847         }
2848
2849         features = btrfs_super_incompat_flags(disk_super);
2850         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2851         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2852                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2853
2854         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2855                 btrfs_info(fs_info, "has skinny extents");
2856
2857         /*
2858          * flag our filesystem as having big metadata blocks if
2859          * they are bigger than the page size
2860          */
2861         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2862                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2863                         btrfs_info(fs_info,
2864                                 "flagging fs with big metadata feature");
2865                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2866         }
2867
2868         nodesize = btrfs_super_nodesize(disk_super);
2869         sectorsize = btrfs_super_sectorsize(disk_super);
2870         stripesize = sectorsize;
2871         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2872         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2873
2874         /* Cache block sizes */
2875         fs_info->nodesize = nodesize;
2876         fs_info->sectorsize = sectorsize;
2877         fs_info->stripesize = stripesize;
2878
2879         /*
2880          * mixed block groups end up with duplicate but slightly offset
2881          * extent buffers for the same range.  It leads to corruptions
2882          */
2883         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2884             (sectorsize != nodesize)) {
2885                 btrfs_err(fs_info,
2886 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2887                         nodesize, sectorsize);
2888                 goto fail_alloc;
2889         }
2890
2891         /*
2892          * Needn't use the lock because there is no other task which will
2893          * update the flag.
2894          */
2895         btrfs_set_super_incompat_flags(disk_super, features);
2896
2897         features = btrfs_super_compat_ro_flags(disk_super) &
2898                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2899         if (!(sb->s_flags & MS_RDONLY) && features) {
2900                 btrfs_err(fs_info,
2901         "cannot mount read-write because of unsupported optional features (%llx)",
2902                        features);
2903                 err = -EINVAL;
2904                 goto fail_alloc;
2905         }
2906
2907         max_active = fs_info->thread_pool_size;
2908
2909         ret = btrfs_init_workqueues(fs_info, fs_devices);
2910         if (ret) {
2911                 err = ret;
2912                 goto fail_sb_buffer;
2913         }
2914
2915         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2916         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2917                                     SZ_4M / PAGE_SIZE);
2918
2919         sb->s_blocksize = sectorsize;
2920         sb->s_blocksize_bits = blksize_bits(sectorsize);
2921
2922         mutex_lock(&fs_info->chunk_mutex);
2923         ret = btrfs_read_sys_array(fs_info);
2924         mutex_unlock(&fs_info->chunk_mutex);
2925         if (ret) {
2926                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2927                 goto fail_sb_buffer;
2928         }
2929
2930         generation = btrfs_super_chunk_root_generation(disk_super);
2931
2932         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2933
2934         chunk_root->node = read_tree_block(fs_info,
2935                                            btrfs_super_chunk_root(disk_super),
2936                                            generation);
2937         if (IS_ERR(chunk_root->node) ||
2938             !extent_buffer_uptodate(chunk_root->node)) {
2939                 btrfs_err(fs_info, "failed to read chunk root");
2940                 if (!IS_ERR(chunk_root->node))
2941                         free_extent_buffer(chunk_root->node);
2942                 chunk_root->node = NULL;
2943                 goto fail_tree_roots;
2944         }
2945         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2946         chunk_root->commit_root = btrfs_root_node(chunk_root);
2947
2948         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2949            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2950
2951         ret = btrfs_read_chunk_tree(fs_info);
2952         if (ret) {
2953                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2954                 goto fail_tree_roots;
2955         }
2956
2957         /*
2958          * keep the device that is marked to be the target device for the
2959          * dev_replace procedure
2960          */
2961         btrfs_close_extra_devices(fs_devices, 0);
2962
2963         if (!fs_devices->latest_bdev) {
2964                 btrfs_err(fs_info, "failed to read devices");
2965                 goto fail_tree_roots;
2966         }
2967
2968 retry_root_backup:
2969         generation = btrfs_super_generation(disk_super);
2970
2971         tree_root->node = read_tree_block(fs_info,
2972                                           btrfs_super_root(disk_super),
2973                                           generation);
2974         if (IS_ERR(tree_root->node) ||
2975             !extent_buffer_uptodate(tree_root->node)) {
2976                 btrfs_warn(fs_info, "failed to read tree root");
2977                 if (!IS_ERR(tree_root->node))
2978                         free_extent_buffer(tree_root->node);
2979                 tree_root->node = NULL;
2980                 goto recovery_tree_root;
2981         }
2982
2983         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2984         tree_root->commit_root = btrfs_root_node(tree_root);
2985         btrfs_set_root_refs(&tree_root->root_item, 1);
2986
2987         mutex_lock(&tree_root->objectid_mutex);
2988         ret = btrfs_find_highest_objectid(tree_root,
2989                                         &tree_root->highest_objectid);
2990         if (ret) {
2991                 mutex_unlock(&tree_root->objectid_mutex);
2992                 goto recovery_tree_root;
2993         }
2994
2995         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2996
2997         mutex_unlock(&tree_root->objectid_mutex);
2998
2999         ret = btrfs_read_roots(fs_info);
3000         if (ret)
3001                 goto recovery_tree_root;
3002
3003         fs_info->generation = generation;
3004         fs_info->last_trans_committed = generation;
3005
3006         ret = btrfs_recover_balance(fs_info);
3007         if (ret) {
3008                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3009                 goto fail_block_groups;
3010         }
3011
3012         ret = btrfs_init_dev_stats(fs_info);
3013         if (ret) {
3014                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3015                 goto fail_block_groups;
3016         }
3017
3018         ret = btrfs_init_dev_replace(fs_info);
3019         if (ret) {
3020                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3021                 goto fail_block_groups;
3022         }
3023
3024         btrfs_close_extra_devices(fs_devices, 1);
3025
3026         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3027         if (ret) {
3028                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3029                                 ret);
3030                 goto fail_block_groups;
3031         }
3032
3033         ret = btrfs_sysfs_add_device(fs_devices);
3034         if (ret) {
3035                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3036                                 ret);
3037                 goto fail_fsdev_sysfs;
3038         }
3039
3040         ret = btrfs_sysfs_add_mounted(fs_info);
3041         if (ret) {
3042                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3043                 goto fail_fsdev_sysfs;
3044         }
3045
3046         ret = btrfs_init_space_info(fs_info);
3047         if (ret) {
3048                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3049                 goto fail_sysfs;
3050         }
3051
3052         ret = btrfs_read_block_groups(fs_info);
3053         if (ret) {
3054                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3055                 goto fail_sysfs;
3056         }
3057         fs_info->num_tolerated_disk_barrier_failures =
3058                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3059         if (fs_info->fs_devices->missing_devices >
3060              fs_info->num_tolerated_disk_barrier_failures &&
3061             !(sb->s_flags & MS_RDONLY)) {
3062                 btrfs_warn(fs_info,
3063 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3064                         fs_info->fs_devices->missing_devices,
3065                         fs_info->num_tolerated_disk_barrier_failures);
3066                 goto fail_sysfs;
3067         }
3068
3069         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3070                                                "btrfs-cleaner");
3071         if (IS_ERR(fs_info->cleaner_kthread))
3072                 goto fail_sysfs;
3073
3074         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3075                                                    tree_root,
3076                                                    "btrfs-transaction");
3077         if (IS_ERR(fs_info->transaction_kthread))
3078                 goto fail_cleaner;
3079
3080         if (!btrfs_test_opt(fs_info, SSD) &&
3081             !btrfs_test_opt(fs_info, NOSSD) &&
3082             !fs_info->fs_devices->rotating) {
3083                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3084                 btrfs_set_opt(fs_info->mount_opt, SSD);
3085         }
3086
3087         /*
3088          * Mount does not set all options immediately, we can do it now and do
3089          * not have to wait for transaction commit
3090          */
3091         btrfs_apply_pending_changes(fs_info);
3092
3093 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3094         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3095                 ret = btrfsic_mount(fs_info, fs_devices,
3096                                     btrfs_test_opt(fs_info,
3097                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3098                                     1 : 0,
3099                                     fs_info->check_integrity_print_mask);
3100                 if (ret)
3101                         btrfs_warn(fs_info,
3102                                 "failed to initialize integrity check module: %d",
3103                                 ret);
3104         }
3105 #endif
3106         ret = btrfs_read_qgroup_config(fs_info);
3107         if (ret)
3108                 goto fail_trans_kthread;
3109
3110         /* do not make disk changes in broken FS or nologreplay is given */
3111         if (btrfs_super_log_root(disk_super) != 0 &&
3112             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3113                 ret = btrfs_replay_log(fs_info, fs_devices);
3114                 if (ret) {
3115                         err = ret;
3116                         goto fail_qgroup;
3117                 }
3118         }
3119
3120         ret = btrfs_find_orphan_roots(fs_info);
3121         if (ret)
3122                 goto fail_qgroup;
3123
3124         if (!(sb->s_flags & MS_RDONLY)) {
3125                 ret = btrfs_cleanup_fs_roots(fs_info);
3126                 if (ret)
3127                         goto fail_qgroup;
3128
3129                 mutex_lock(&fs_info->cleaner_mutex);
3130                 ret = btrfs_recover_relocation(tree_root);
3131                 mutex_unlock(&fs_info->cleaner_mutex);
3132                 if (ret < 0) {
3133                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3134                                         ret);
3135                         err = -EINVAL;
3136                         goto fail_qgroup;
3137                 }
3138         }
3139
3140         location.objectid = BTRFS_FS_TREE_OBJECTID;
3141         location.type = BTRFS_ROOT_ITEM_KEY;
3142         location.offset = 0;
3143
3144         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3145         if (IS_ERR(fs_info->fs_root)) {
3146                 err = PTR_ERR(fs_info->fs_root);
3147                 goto fail_qgroup;
3148         }
3149
3150         if (sb->s_flags & MS_RDONLY)
3151                 return 0;
3152
3153         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3154             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3155                 clear_free_space_tree = 1;
3156         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3157                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3158                 btrfs_warn(fs_info, "free space tree is invalid");
3159                 clear_free_space_tree = 1;
3160         }
3161
3162         if (clear_free_space_tree) {
3163                 btrfs_info(fs_info, "clearing free space tree");
3164                 ret = btrfs_clear_free_space_tree(fs_info);
3165                 if (ret) {
3166                         btrfs_warn(fs_info,
3167                                    "failed to clear free space tree: %d", ret);
3168                         close_ctree(fs_info);
3169                         return ret;
3170                 }
3171         }
3172
3173         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3174             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3175                 btrfs_info(fs_info, "creating free space tree");
3176                 ret = btrfs_create_free_space_tree(fs_info);
3177                 if (ret) {
3178                         btrfs_warn(fs_info,
3179                                 "failed to create free space tree: %d", ret);
3180                         close_ctree(fs_info);
3181                         return ret;
3182                 }
3183         }
3184
3185         down_read(&fs_info->cleanup_work_sem);
3186         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3187             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3188                 up_read(&fs_info->cleanup_work_sem);
3189                 close_ctree(fs_info);
3190                 return ret;
3191         }
3192         up_read(&fs_info->cleanup_work_sem);
3193
3194         ret = btrfs_resume_balance_async(fs_info);
3195         if (ret) {
3196                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3197                 close_ctree(fs_info);
3198                 return ret;
3199         }
3200
3201         ret = btrfs_resume_dev_replace_async(fs_info);
3202         if (ret) {
3203                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3204                 close_ctree(fs_info);
3205                 return ret;
3206         }
3207
3208         btrfs_qgroup_rescan_resume(fs_info);
3209
3210         if (!fs_info->uuid_root) {
3211                 btrfs_info(fs_info, "creating UUID tree");
3212                 ret = btrfs_create_uuid_tree(fs_info);
3213                 if (ret) {
3214                         btrfs_warn(fs_info,
3215                                 "failed to create the UUID tree: %d", ret);
3216                         close_ctree(fs_info);
3217                         return ret;
3218                 }
3219         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3220                    fs_info->generation !=
3221                                 btrfs_super_uuid_tree_generation(disk_super)) {
3222                 btrfs_info(fs_info, "checking UUID tree");
3223                 ret = btrfs_check_uuid_tree(fs_info);
3224                 if (ret) {
3225                         btrfs_warn(fs_info,
3226                                 "failed to check the UUID tree: %d", ret);
3227                         close_ctree(fs_info);
3228                         return ret;
3229                 }
3230         } else {
3231                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3232         }
3233         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3234
3235         /*
3236          * backuproot only affect mount behavior, and if open_ctree succeeded,
3237          * no need to keep the flag
3238          */
3239         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3240
3241         return 0;
3242
3243 fail_qgroup:
3244         btrfs_free_qgroup_config(fs_info);
3245 fail_trans_kthread:
3246         kthread_stop(fs_info->transaction_kthread);
3247         btrfs_cleanup_transaction(fs_info);
3248         btrfs_free_fs_roots(fs_info);
3249 fail_cleaner:
3250         kthread_stop(fs_info->cleaner_kthread);
3251
3252         /*
3253          * make sure we're done with the btree inode before we stop our
3254          * kthreads
3255          */
3256         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3257
3258 fail_sysfs:
3259         btrfs_sysfs_remove_mounted(fs_info);
3260
3261 fail_fsdev_sysfs:
3262         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3263
3264 fail_block_groups:
3265         btrfs_put_block_group_cache(fs_info);
3266         btrfs_free_block_groups(fs_info);
3267
3268 fail_tree_roots:
3269         free_root_pointers(fs_info, 1);
3270         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3271
3272 fail_sb_buffer:
3273         btrfs_stop_all_workers(fs_info);
3274 fail_alloc:
3275 fail_iput:
3276         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3277
3278         iput(fs_info->btree_inode);
3279 fail_bio_counter:
3280         percpu_counter_destroy(&fs_info->bio_counter);
3281 fail_delalloc_bytes:
3282         percpu_counter_destroy(&fs_info->delalloc_bytes);
3283 fail_dirty_metadata_bytes:
3284         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3285 fail_bdi:
3286         bdi_destroy(&fs_info->bdi);
3287 fail_srcu:
3288         cleanup_srcu_struct(&fs_info->subvol_srcu);
3289 fail:
3290         btrfs_free_stripe_hash_table(fs_info);
3291         btrfs_close_devices(fs_info->fs_devices);
3292         return err;
3293
3294 recovery_tree_root:
3295         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3296                 goto fail_tree_roots;
3297
3298         free_root_pointers(fs_info, 0);
3299
3300         /* don't use the log in recovery mode, it won't be valid */
3301         btrfs_set_super_log_root(disk_super, 0);
3302
3303         /* we can't trust the free space cache either */
3304         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3305
3306         ret = next_root_backup(fs_info, fs_info->super_copy,
3307                                &num_backups_tried, &backup_index);
3308         if (ret == -1)
3309                 goto fail_block_groups;
3310         goto retry_root_backup;
3311 }
3312
3313 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3314 {
3315         if (uptodate) {
3316                 set_buffer_uptodate(bh);
3317         } else {
3318                 struct btrfs_device *device = (struct btrfs_device *)
3319                         bh->b_private;
3320
3321                 btrfs_warn_rl_in_rcu(device->fs_info,
3322                                 "lost page write due to IO error on %s",
3323                                           rcu_str_deref(device->name));
3324                 /* note, we don't set_buffer_write_io_error because we have
3325                  * our own ways of dealing with the IO errors
3326                  */
3327                 clear_buffer_uptodate(bh);
3328                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3329         }
3330         unlock_buffer(bh);
3331         put_bh(bh);
3332 }
3333
3334 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3335                         struct buffer_head **bh_ret)
3336 {
3337         struct buffer_head *bh;
3338         struct btrfs_super_block *super;
3339         u64 bytenr;
3340
3341         bytenr = btrfs_sb_offset(copy_num);
3342         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3343                 return -EINVAL;
3344
3345         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3346         /*
3347          * If we fail to read from the underlying devices, as of now
3348          * the best option we have is to mark it EIO.
3349          */
3350         if (!bh)
3351                 return -EIO;
3352
3353         super = (struct btrfs_super_block *)bh->b_data;
3354         if (btrfs_super_bytenr(super) != bytenr ||
3355                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3356                 brelse(bh);
3357                 return -EINVAL;
3358         }
3359
3360         *bh_ret = bh;
3361         return 0;
3362 }
3363
3364
3365 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3366 {
3367         struct buffer_head *bh;
3368         struct buffer_head *latest = NULL;
3369         struct btrfs_super_block *super;
3370         int i;
3371         u64 transid = 0;
3372         int ret = -EINVAL;
3373
3374         /* we would like to check all the supers, but that would make
3375          * a btrfs mount succeed after a mkfs from a different FS.
3376          * So, we need to add a special mount option to scan for
3377          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3378          */
3379         for (i = 0; i < 1; i++) {
3380                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3381                 if (ret)
3382                         continue;
3383
3384                 super = (struct btrfs_super_block *)bh->b_data;
3385
3386                 if (!latest || btrfs_super_generation(super) > transid) {
3387                         brelse(latest);
3388                         latest = bh;
3389                         transid = btrfs_super_generation(super);
3390                 } else {
3391                         brelse(bh);
3392                 }
3393         }
3394
3395         if (!latest)
3396                 return ERR_PTR(ret);
3397
3398         return latest;
3399 }
3400
3401 /*
3402  * this should be called twice, once with wait == 0 and
3403  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3404  * we write are pinned.
3405  *
3406  * They are released when wait == 1 is done.
3407  * max_mirrors must be the same for both runs, and it indicates how
3408  * many supers on this one device should be written.
3409  *
3410  * max_mirrors == 0 means to write them all.
3411  */
3412 static int write_dev_supers(struct btrfs_device *device,
3413                             struct btrfs_super_block *sb,
3414                             int do_barriers, int wait, int max_mirrors)
3415 {
3416         struct buffer_head *bh;
3417         int i;
3418         int ret;
3419         int errors = 0;
3420         u32 crc;
3421         u64 bytenr;
3422
3423         if (max_mirrors == 0)
3424                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3425
3426         for (i = 0; i < max_mirrors; i++) {
3427                 bytenr = btrfs_sb_offset(i);
3428                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3429                     device->commit_total_bytes)
3430                         break;
3431
3432                 if (wait) {
3433                         bh = __find_get_block(device->bdev, bytenr / 4096,
3434                                               BTRFS_SUPER_INFO_SIZE);
3435                         if (!bh) {
3436                                 errors++;
3437                                 continue;
3438                         }
3439                         wait_on_buffer(bh);
3440                         if (!buffer_uptodate(bh))
3441                                 errors++;
3442
3443                         /* drop our reference */
3444                         brelse(bh);
3445
3446                         /* drop the reference from the wait == 0 run */
3447                         brelse(bh);
3448                         continue;
3449                 } else {
3450                         btrfs_set_super_bytenr(sb, bytenr);
3451
3452                         crc = ~(u32)0;
3453                         crc = btrfs_csum_data((char *)sb +
3454                                               BTRFS_CSUM_SIZE, crc,
3455                                               BTRFS_SUPER_INFO_SIZE -
3456                                               BTRFS_CSUM_SIZE);
3457                         btrfs_csum_final(crc, sb->csum);
3458
3459                         /*
3460                          * one reference for us, and we leave it for the
3461                          * caller
3462                          */
3463                         bh = __getblk(device->bdev, bytenr / 4096,
3464                                       BTRFS_SUPER_INFO_SIZE);
3465                         if (!bh) {
3466                                 btrfs_err(device->fs_info,
3467                                     "couldn't get super buffer head for bytenr %llu",
3468                                     bytenr);
3469                                 errors++;
3470                                 continue;
3471                         }
3472
3473                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3474
3475                         /* one reference for submit_bh */
3476                         get_bh(bh);
3477
3478                         set_buffer_uptodate(bh);
3479                         lock_buffer(bh);
3480                         bh->b_end_io = btrfs_end_buffer_write_sync;
3481                         bh->b_private = device;
3482                 }
3483
3484                 /*
3485                  * we fua the first super.  The others we allow
3486                  * to go down lazy.
3487                  */
3488                 if (i == 0)
3489                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_FUA, bh);
3490                 else
3491                         ret = btrfsic_submit_bh(REQ_OP_WRITE, REQ_SYNC, bh);
3492                 if (ret)
3493                         errors++;
3494         }
3495         return errors < i ? 0 : -1;
3496 }
3497
3498 /*
3499  * endio for the write_dev_flush, this will wake anyone waiting
3500  * for the barrier when it is done
3501  */
3502 static void btrfs_end_empty_barrier(struct bio *bio)
3503 {
3504         if (bio->bi_private)
3505                 complete(bio->bi_private);
3506         bio_put(bio);
3507 }
3508
3509 /*
3510  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3511  * sent down.  With wait == 1, it waits for the previous flush.
3512  *
3513  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3514  * capable
3515  */
3516 static int write_dev_flush(struct btrfs_device *device, int wait)
3517 {
3518         struct bio *bio;
3519         int ret = 0;
3520
3521         if (device->nobarriers)
3522                 return 0;
3523
3524         if (wait) {
3525                 bio = device->flush_bio;
3526                 if (!bio)
3527                         return 0;
3528
3529                 wait_for_completion(&device->flush_wait);
3530
3531                 if (bio->bi_error) {
3532                         ret = bio->bi_error;
3533                         btrfs_dev_stat_inc_and_print(device,
3534                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3535                 }
3536
3537                 /* drop the reference from the wait == 0 run */
3538                 bio_put(bio);
3539                 device->flush_bio = NULL;
3540
3541                 return ret;
3542         }
3543
3544         /*
3545          * one reference for us, and we leave it for the
3546          * caller
3547          */
3548         device->flush_bio = NULL;
3549         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3550         if (!bio)
3551                 return -ENOMEM;
3552
3553         bio->bi_end_io = btrfs_end_empty_barrier;
3554         bio->bi_bdev = device->bdev;
3555         bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
3556         init_completion(&device->flush_wait);
3557         bio->bi_private = &device->flush_wait;
3558         device->flush_bio = bio;
3559
3560         bio_get(bio);
3561         btrfsic_submit_bio(bio);
3562
3563         return 0;
3564 }
3565
3566 /*
3567  * send an empty flush down to each device in parallel,
3568  * then wait for them
3569  */
3570 static int barrier_all_devices(struct btrfs_fs_info *info)
3571 {
3572         struct list_head *head;
3573         struct btrfs_device *dev;
3574         int errors_send = 0;
3575         int errors_wait = 0;
3576         int ret;
3577
3578         /* send down all the barriers */
3579         head = &info->fs_devices->devices;
3580         list_for_each_entry_rcu(dev, head, dev_list) {
3581                 if (dev->missing)
3582                         continue;
3583                 if (!dev->bdev) {
3584                         errors_send++;
3585                         continue;
3586                 }
3587                 if (!dev->in_fs_metadata || !dev->writeable)
3588                         continue;
3589
3590                 ret = write_dev_flush(dev, 0);
3591                 if (ret)
3592                         errors_send++;
3593         }
3594
3595         /* wait for all the barriers */
3596         list_for_each_entry_rcu(dev, head, dev_list) {
3597                 if (dev->missing)
3598                         continue;
3599                 if (!dev->bdev) {
3600                         errors_wait++;
3601                         continue;
3602                 }
3603                 if (!dev->in_fs_metadata || !dev->writeable)
3604                         continue;
3605
3606                 ret = write_dev_flush(dev, 1);
3607                 if (ret)
3608                         errors_wait++;
3609         }
3610         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3611             errors_wait > info->num_tolerated_disk_barrier_failures)
3612                 return -EIO;
3613         return 0;
3614 }
3615
3616 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3617 {
3618         int raid_type;
3619         int min_tolerated = INT_MAX;
3620
3621         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3622             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3623                 min_tolerated = min(min_tolerated,
3624                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3625                                     tolerated_failures);
3626
3627         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3628                 if (raid_type == BTRFS_RAID_SINGLE)
3629                         continue;
3630                 if (!(flags & btrfs_raid_group[raid_type]))
3631                         continue;
3632                 min_tolerated = min(min_tolerated,
3633                                     btrfs_raid_array[raid_type].
3634                                     tolerated_failures);
3635         }
3636
3637         if (min_tolerated == INT_MAX) {
3638                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3639                 min_tolerated = 0;
3640         }
3641
3642         return min_tolerated;
3643 }
3644
3645 int btrfs_calc_num_tolerated_disk_barrier_failures(
3646         struct btrfs_fs_info *fs_info)
3647 {
3648         struct btrfs_ioctl_space_info space;
3649         struct btrfs_space_info *sinfo;
3650         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3651                        BTRFS_BLOCK_GROUP_SYSTEM,
3652                        BTRFS_BLOCK_GROUP_METADATA,
3653                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3654         int i;
3655         int c;
3656         int num_tolerated_disk_barrier_failures =
3657                 (int)fs_info->fs_devices->num_devices;
3658
3659         for (i = 0; i < ARRAY_SIZE(types); i++) {
3660                 struct btrfs_space_info *tmp;
3661
3662                 sinfo = NULL;
3663                 rcu_read_lock();
3664                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3665                         if (tmp->flags == types[i]) {
3666                                 sinfo = tmp;
3667                                 break;
3668                         }
3669                 }
3670                 rcu_read_unlock();
3671
3672                 if (!sinfo)
3673                         continue;
3674
3675                 down_read(&sinfo->groups_sem);
3676                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3677                         u64 flags;
3678
3679                         if (list_empty(&sinfo->block_groups[c]))
3680                                 continue;
3681
3682                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3683                                                    &space);
3684                         if (space.total_bytes == 0 || space.used_bytes == 0)
3685                                 continue;
3686                         flags = space.flags;
3687
3688                         num_tolerated_disk_barrier_failures = min(
3689                                 num_tolerated_disk_barrier_failures,
3690                                 btrfs_get_num_tolerated_disk_barrier_failures(
3691                                         flags));
3692                 }
3693                 up_read(&sinfo->groups_sem);
3694         }
3695
3696         return num_tolerated_disk_barrier_failures;
3697 }
3698
3699 static int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3700 {
3701         struct list_head *head;
3702         struct btrfs_device *dev;
3703         struct btrfs_super_block *sb;
3704         struct btrfs_dev_item *dev_item;
3705         int ret;
3706         int do_barriers;
3707         int max_errors;
3708         int total_errors = 0;
3709         u64 flags;
3710
3711         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3712         backup_super_roots(fs_info);
3713
3714         sb = fs_info->super_for_commit;
3715         dev_item = &sb->dev_item;
3716
3717         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3718         head = &fs_info->fs_devices->devices;
3719         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3720
3721         if (do_barriers) {
3722                 ret = barrier_all_devices(fs_info);
3723                 if (ret) {
3724                         mutex_unlock(
3725                                 &fs_info->fs_devices->device_list_mutex);
3726                         btrfs_handle_fs_error(fs_info, ret,
3727                                               "errors while submitting device barriers.");
3728                         return ret;
3729                 }
3730         }
3731
3732         list_for_each_entry_rcu(dev, head, dev_list) {
3733                 if (!dev->bdev) {
3734                         total_errors++;
3735                         continue;
3736                 }
3737                 if (!dev->in_fs_metadata || !dev->writeable)
3738                         continue;
3739
3740                 btrfs_set_stack_device_generation(dev_item, 0);
3741                 btrfs_set_stack_device_type(dev_item, dev->type);
3742                 btrfs_set_stack_device_id(dev_item, dev->devid);
3743                 btrfs_set_stack_device_total_bytes(dev_item,
3744                                                    dev->commit_total_bytes);
3745                 btrfs_set_stack_device_bytes_used(dev_item,
3746                                                   dev->commit_bytes_used);
3747                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3748                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3749                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3750                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3751                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3752
3753                 flags = btrfs_super_flags(sb);
3754                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3755
3756                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3757                 if (ret)
3758                         total_errors++;
3759         }
3760         if (total_errors > max_errors) {
3761                 btrfs_err(fs_info, "%d errors while writing supers",
3762                           total_errors);
3763                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3764
3765                 /* FUA is masked off if unsupported and can't be the reason */
3766                 btrfs_handle_fs_error(fs_info, -EIO,
3767                                       "%d errors while writing supers",
3768                                       total_errors);
3769                 return -EIO;
3770         }
3771
3772         total_errors = 0;
3773         list_for_each_entry_rcu(dev, head, dev_list) {
3774                 if (!dev->bdev)
3775                         continue;
3776                 if (!dev->in_fs_metadata || !dev->writeable)
3777                         continue;
3778
3779                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3780                 if (ret)
3781                         total_errors++;
3782         }
3783         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3784         if (total_errors > max_errors) {
3785                 btrfs_handle_fs_error(fs_info, -EIO,
3786                                       "%d errors while writing supers",
3787                                       total_errors);
3788                 return -EIO;
3789         }
3790         return 0;
3791 }
3792
3793 int write_ctree_super(struct btrfs_trans_handle *trans,
3794                       struct btrfs_fs_info *fs_info, int max_mirrors)
3795 {
3796         return write_all_supers(fs_info, max_mirrors);
3797 }
3798
3799 /* Drop a fs root from the radix tree and free it. */
3800 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3801                                   struct btrfs_root *root)
3802 {
3803         spin_lock(&fs_info->fs_roots_radix_lock);
3804         radix_tree_delete(&fs_info->fs_roots_radix,
3805                           (unsigned long)root->root_key.objectid);
3806         spin_unlock(&fs_info->fs_roots_radix_lock);
3807
3808         if (btrfs_root_refs(&root->root_item) == 0)
3809                 synchronize_srcu(&fs_info->subvol_srcu);
3810
3811         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3812                 btrfs_free_log(NULL, root);
3813                 if (root->reloc_root) {
3814                         free_extent_buffer(root->reloc_root->node);
3815                         free_extent_buffer(root->reloc_root->commit_root);
3816                         btrfs_put_fs_root(root->reloc_root);
3817                         root->reloc_root = NULL;
3818                 }
3819         }
3820
3821         if (root->free_ino_pinned)
3822                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3823         if (root->free_ino_ctl)
3824                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3825         free_fs_root(root);
3826 }
3827
3828 static void free_fs_root(struct btrfs_root *root)
3829 {
3830         iput(root->ino_cache_inode);
3831         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3832         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3833         root->orphan_block_rsv = NULL;
3834         if (root->anon_dev)
3835                 free_anon_bdev(root->anon_dev);
3836         if (root->subv_writers)
3837                 btrfs_free_subvolume_writers(root->subv_writers);
3838         free_extent_buffer(root->node);
3839         free_extent_buffer(root->commit_root);
3840         kfree(root->free_ino_ctl);
3841         kfree(root->free_ino_pinned);
3842         kfree(root->name);
3843         btrfs_put_fs_root(root);
3844 }
3845
3846 void btrfs_free_fs_root(struct btrfs_root *root)
3847 {
3848         free_fs_root(root);
3849 }
3850
3851 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3852 {
3853         u64 root_objectid = 0;
3854         struct btrfs_root *gang[8];
3855         int i = 0;
3856         int err = 0;
3857         unsigned int ret = 0;
3858         int index;
3859
3860         while (1) {
3861                 index = srcu_read_lock(&fs_info->subvol_srcu);
3862                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3863                                              (void **)gang, root_objectid,
3864                                              ARRAY_SIZE(gang));
3865                 if (!ret) {
3866                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3867                         break;
3868                 }
3869                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3870
3871                 for (i = 0; i < ret; i++) {
3872                         /* Avoid to grab roots in dead_roots */
3873                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3874                                 gang[i] = NULL;
3875                                 continue;
3876                         }
3877                         /* grab all the search result for later use */
3878                         gang[i] = btrfs_grab_fs_root(gang[i]);
3879                 }
3880                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3881
3882                 for (i = 0; i < ret; i++) {
3883                         if (!gang[i])
3884                                 continue;
3885                         root_objectid = gang[i]->root_key.objectid;
3886                         err = btrfs_orphan_cleanup(gang[i]);
3887                         if (err)
3888                                 break;
3889                         btrfs_put_fs_root(gang[i]);
3890                 }
3891                 root_objectid++;
3892         }
3893
3894         /* release the uncleaned roots due to error */
3895         for (; i < ret; i++) {
3896                 if (gang[i])
3897                         btrfs_put_fs_root(gang[i]);
3898         }
3899         return err;
3900 }
3901
3902 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3903 {
3904         struct btrfs_root *root = fs_info->tree_root;
3905         struct btrfs_trans_handle *trans;
3906
3907         mutex_lock(&fs_info->cleaner_mutex);
3908         btrfs_run_delayed_iputs(fs_info);
3909         mutex_unlock(&fs_info->cleaner_mutex);
3910         wake_up_process(fs_info->cleaner_kthread);
3911
3912         /* wait until ongoing cleanup work done */
3913         down_write(&fs_info->cleanup_work_sem);
3914         up_write(&fs_info->cleanup_work_sem);
3915
3916         trans = btrfs_join_transaction(root);
3917         if (IS_ERR(trans))
3918                 return PTR_ERR(trans);
3919         return btrfs_commit_transaction(trans);
3920 }
3921
3922 void close_ctree(struct btrfs_fs_info *fs_info)
3923 {
3924         struct btrfs_root *root = fs_info->tree_root;
3925         int ret;
3926
3927         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3928
3929         /* wait for the qgroup rescan worker to stop */
3930         btrfs_qgroup_wait_for_completion(fs_info, false);
3931
3932         /* wait for the uuid_scan task to finish */
3933         down(&fs_info->uuid_tree_rescan_sem);
3934         /* avoid complains from lockdep et al., set sem back to initial state */
3935         up(&fs_info->uuid_tree_rescan_sem);
3936
3937         /* pause restriper - we want to resume on mount */
3938         btrfs_pause_balance(fs_info);
3939
3940         btrfs_dev_replace_suspend_for_unmount(fs_info);
3941
3942         btrfs_scrub_cancel(fs_info);
3943
3944         /* wait for any defraggers to finish */
3945         wait_event(fs_info->transaction_wait,
3946                    (atomic_read(&fs_info->defrag_running) == 0));
3947
3948         /* clear out the rbtree of defraggable inodes */
3949         btrfs_cleanup_defrag_inodes(fs_info);
3950
3951         cancel_work_sync(&fs_info->async_reclaim_work);
3952
3953         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3954                 /*
3955                  * If the cleaner thread is stopped and there are
3956                  * block groups queued for removal, the deletion will be
3957                  * skipped when we quit the cleaner thread.
3958                  */
3959                 btrfs_delete_unused_bgs(fs_info);
3960
3961                 ret = btrfs_commit_super(fs_info);
3962                 if (ret)
3963                         btrfs_err(fs_info, "commit super ret %d", ret);
3964         }
3965
3966         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3967                 btrfs_error_commit_super(fs_info);
3968
3969         kthread_stop(fs_info->transaction_kthread);
3970         kthread_stop(fs_info->cleaner_kthread);
3971
3972         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3973
3974         btrfs_free_qgroup_config(fs_info);
3975
3976         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3977                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3978                        percpu_counter_sum(&fs_info->delalloc_bytes));
3979         }
3980
3981         btrfs_sysfs_remove_mounted(fs_info);
3982         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3983
3984         btrfs_free_fs_roots(fs_info);
3985
3986         btrfs_put_block_group_cache(fs_info);
3987
3988         btrfs_free_block_groups(fs_info);
3989
3990         /*
3991          * we must make sure there is not any read request to
3992          * submit after we stopping all workers.
3993          */
3994         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3995         btrfs_stop_all_workers(fs_info);
3996
3997         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
3998         free_root_pointers(fs_info, 1);
3999
4000         iput(fs_info->btree_inode);
4001
4002 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4003         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4004                 btrfsic_unmount(fs_info->fs_devices);
4005 #endif
4006
4007         btrfs_close_devices(fs_info->fs_devices);
4008         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4009
4010         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4011         percpu_counter_destroy(&fs_info->delalloc_bytes);
4012         percpu_counter_destroy(&fs_info->bio_counter);
4013         bdi_destroy(&fs_info->bdi);
4014         cleanup_srcu_struct(&fs_info->subvol_srcu);
4015
4016         btrfs_free_stripe_hash_table(fs_info);
4017
4018         __btrfs_free_block_rsv(root->orphan_block_rsv);
4019         root->orphan_block_rsv = NULL;
4020
4021         mutex_lock(&fs_info->chunk_mutex);
4022         while (!list_empty(&fs_info->pinned_chunks)) {
4023                 struct extent_map *em;
4024
4025                 em = list_first_entry(&fs_info->pinned_chunks,
4026                                       struct extent_map, list);
4027                 list_del_init(&em->list);
4028                 free_extent_map(em);
4029         }
4030         mutex_unlock(&fs_info->chunk_mutex);
4031 }
4032
4033 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4034                           int atomic)
4035 {
4036         int ret;
4037         struct inode *btree_inode = buf->pages[0]->mapping->host;
4038
4039         ret = extent_buffer_uptodate(buf);
4040         if (!ret)
4041                 return ret;
4042
4043         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4044                                     parent_transid, atomic);
4045         if (ret == -EAGAIN)
4046                 return ret;
4047         return !ret;
4048 }
4049
4050 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4051 {
4052         struct btrfs_fs_info *fs_info;
4053         struct btrfs_root *root;
4054         u64 transid = btrfs_header_generation(buf);
4055         int was_dirty;
4056
4057 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4058         /*
4059          * This is a fast path so only do this check if we have sanity tests
4060          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4061          * outside of the sanity tests.
4062          */
4063         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4064                 return;
4065 #endif
4066         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4067         fs_info = root->fs_info;
4068         btrfs_assert_tree_locked(buf);
4069         if (transid != fs_info->generation)
4070                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4071                         buf->start, transid, fs_info->generation);
4072         was_dirty = set_extent_buffer_dirty(buf);
4073         if (!was_dirty)
4074                 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
4075                                      buf->len,
4076                                      fs_info->dirty_metadata_batch);
4077 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4078         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4079                 btrfs_print_leaf(fs_info, buf);
4080                 ASSERT(0);
4081         }
4082 #endif
4083 }
4084
4085 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4086                                         int flush_delayed)
4087 {
4088         /*
4089          * looks as though older kernels can get into trouble with
4090          * this code, they end up stuck in balance_dirty_pages forever
4091          */
4092         int ret;
4093
4094         if (current->flags & PF_MEMALLOC)
4095                 return;
4096
4097         if (flush_delayed)
4098                 btrfs_balance_delayed_items(fs_info);
4099
4100         ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4101                                      BTRFS_DIRTY_METADATA_THRESH);
4102         if (ret > 0) {
4103                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4104         }
4105 }
4106
4107 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4108 {
4109         __btrfs_btree_balance_dirty(fs_info, 1);
4110 }
4111
4112 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4113 {
4114         __btrfs_btree_balance_dirty(fs_info, 0);
4115 }
4116
4117 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4118 {
4119         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4120         struct btrfs_fs_info *fs_info = root->fs_info;
4121
4122         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
4123 }
4124
4125 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4126                               int read_only)
4127 {
4128         struct btrfs_super_block *sb = fs_info->super_copy;
4129         u64 nodesize = btrfs_super_nodesize(sb);
4130         u64 sectorsize = btrfs_super_sectorsize(sb);
4131         int ret = 0;
4132
4133         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4134                 btrfs_err(fs_info, "no valid FS found");
4135                 ret = -EINVAL;
4136         }
4137         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4138                 btrfs_warn(fs_info, "unrecognized super flag: %llu",
4139                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4140         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4141                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4142                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4143                 ret = -EINVAL;
4144         }
4145         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4146                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4147                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4148                 ret = -EINVAL;
4149         }
4150         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4151                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
4152                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4153                 ret = -EINVAL;
4154         }
4155
4156         /*
4157          * Check sectorsize and nodesize first, other check will need it.
4158          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4159          */
4160         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4161             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4162                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4163                 ret = -EINVAL;
4164         }
4165         /* Only PAGE SIZE is supported yet */
4166         if (sectorsize != PAGE_SIZE) {
4167                 btrfs_err(fs_info,
4168                         "sectorsize %llu not supported yet, only support %lu",
4169                         sectorsize, PAGE_SIZE);
4170                 ret = -EINVAL;
4171         }
4172         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4173             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4174                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4175                 ret = -EINVAL;
4176         }
4177         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4178                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4179                           le32_to_cpu(sb->__unused_leafsize), nodesize);
4180                 ret = -EINVAL;
4181         }
4182
4183         /* Root alignment check */
4184         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4185                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4186                            btrfs_super_root(sb));
4187                 ret = -EINVAL;
4188         }
4189         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4190                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4191                            btrfs_super_chunk_root(sb));
4192                 ret = -EINVAL;
4193         }
4194         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4195                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
4196                            btrfs_super_log_root(sb));
4197                 ret = -EINVAL;
4198         }
4199
4200         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4201                 btrfs_err(fs_info,
4202                            "dev_item UUID does not match fsid: %pU != %pU",
4203                            fs_info->fsid, sb->dev_item.fsid);
4204                 ret = -EINVAL;
4205         }
4206
4207         /*
4208          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4209          * done later
4210          */
4211         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4212                 btrfs_err(fs_info, "bytes_used is too small %llu",
4213                           btrfs_super_bytes_used(sb));
4214                 ret = -EINVAL;
4215         }
4216         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4217                 btrfs_err(fs_info, "invalid stripesize %u",
4218                           btrfs_super_stripesize(sb));
4219                 ret = -EINVAL;
4220         }
4221         if (btrfs_super_num_devices(sb) > (1UL << 31))
4222                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
4223                            btrfs_super_num_devices(sb));
4224         if (btrfs_super_num_devices(sb) == 0) {
4225                 btrfs_err(fs_info, "number of devices is 0");
4226                 ret = -EINVAL;
4227         }
4228
4229         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4230                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
4231                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4232                 ret = -EINVAL;
4233         }
4234
4235         /*
4236          * Obvious sys_chunk_array corruptions, it must hold at least one key
4237          * and one chunk
4238          */
4239         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4240                 btrfs_err(fs_info, "system chunk array too big %u > %u",
4241                           btrfs_super_sys_array_size(sb),
4242                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4243                 ret = -EINVAL;
4244         }
4245         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4246                         + sizeof(struct btrfs_chunk)) {
4247                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
4248                           btrfs_super_sys_array_size(sb),
4249                           sizeof(struct btrfs_disk_key)
4250                           + sizeof(struct btrfs_chunk));
4251                 ret = -EINVAL;
4252         }
4253
4254         /*
4255          * The generation is a global counter, we'll trust it more than the others
4256          * but it's still possible that it's the one that's wrong.
4257          */
4258         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4259                 btrfs_warn(fs_info,
4260                         "suspicious: generation < chunk_root_generation: %llu < %llu",
4261                         btrfs_super_generation(sb),
4262                         btrfs_super_chunk_root_generation(sb));
4263         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4264             && btrfs_super_cache_generation(sb) != (u64)-1)
4265                 btrfs_warn(fs_info,
4266                         "suspicious: generation < cache_generation: %llu < %llu",
4267                         btrfs_super_generation(sb),
4268                         btrfs_super_cache_generation(sb));
4269
4270         return ret;
4271 }
4272
4273 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4274 {
4275         mutex_lock(&fs_info->cleaner_mutex);
4276         btrfs_run_delayed_iputs(fs_info);
4277         mutex_unlock(&fs_info->cleaner_mutex);
4278
4279         down_write(&fs_info->cleanup_work_sem);
4280         up_write(&fs_info->cleanup_work_sem);
4281
4282         /* cleanup FS via transaction */
4283         btrfs_cleanup_transaction(fs_info);
4284 }
4285
4286 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4287 {
4288         struct btrfs_ordered_extent *ordered;
4289
4290         spin_lock(&root->ordered_extent_lock);
4291         /*
4292          * This will just short circuit the ordered completion stuff which will
4293          * make sure the ordered extent gets properly cleaned up.
4294          */
4295         list_for_each_entry(ordered, &root->ordered_extents,
4296                             root_extent_list)
4297                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4298         spin_unlock(&root->ordered_extent_lock);
4299 }
4300
4301 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4302 {
4303         struct btrfs_root *root;
4304         struct list_head splice;
4305
4306         INIT_LIST_HEAD(&splice);
4307
4308         spin_lock(&fs_info->ordered_root_lock);
4309         list_splice_init(&fs_info->ordered_roots, &splice);
4310         while (!list_empty(&splice)) {
4311                 root = list_first_entry(&splice, struct btrfs_root,
4312                                         ordered_root);
4313                 list_move_tail(&root->ordered_root,
4314                                &fs_info->ordered_roots);
4315
4316                 spin_unlock(&fs_info->ordered_root_lock);
4317                 btrfs_destroy_ordered_extents(root);
4318
4319                 cond_resched();
4320                 spin_lock(&fs_info->ordered_root_lock);
4321         }
4322         spin_unlock(&fs_info->ordered_root_lock);
4323 }
4324
4325 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4326                                       struct btrfs_fs_info *fs_info)
4327 {
4328         struct rb_node *node;
4329         struct btrfs_delayed_ref_root *delayed_refs;
4330         struct btrfs_delayed_ref_node *ref;
4331         int ret = 0;
4332
4333         delayed_refs = &trans->delayed_refs;
4334
4335         spin_lock(&delayed_refs->lock);
4336         if (atomic_read(&delayed_refs->num_entries) == 0) {
4337                 spin_unlock(&delayed_refs->lock);
4338                 btrfs_info(fs_info, "delayed_refs has NO entry");
4339                 return ret;
4340         }
4341
4342         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4343                 struct btrfs_delayed_ref_head *head;
4344                 struct btrfs_delayed_ref_node *tmp;
4345                 bool pin_bytes = false;
4346
4347                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4348                                 href_node);
4349                 if (!mutex_trylock(&head->mutex)) {
4350                         atomic_inc(&head->node.refs);
4351                         spin_unlock(&delayed_refs->lock);
4352
4353                         mutex_lock(&head->mutex);
4354                         mutex_unlock(&head->mutex);
4355                         btrfs_put_delayed_ref(&head->node);
4356                         spin_lock(&delayed_refs->lock);
4357                         continue;
4358                 }
4359                 spin_lock(&head->lock);
4360                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4361                                                  list) {
4362                         ref->in_tree = 0;
4363                         list_del(&ref->list);
4364                         if (!list_empty(&ref->add_list))
4365                                 list_del(&ref->add_list);
4366                         atomic_dec(&delayed_refs->num_entries);
4367                         btrfs_put_delayed_ref(ref);
4368                 }
4369                 if (head->must_insert_reserved)
4370                         pin_bytes = true;
4371                 btrfs_free_delayed_extent_op(head->extent_op);
4372                 delayed_refs->num_heads--;
4373                 if (head->processing == 0)
4374                         delayed_refs->num_heads_ready--;
4375                 atomic_dec(&delayed_refs->num_entries);
4376                 head->node.in_tree = 0;
4377                 rb_erase(&head->href_node, &delayed_refs->href_root);
4378                 spin_unlock(&head->lock);
4379                 spin_unlock(&delayed_refs->lock);
4380                 mutex_unlock(&head->mutex);
4381
4382                 if (pin_bytes)
4383                         btrfs_pin_extent(fs_info, head->node.bytenr,
4384                                          head->node.num_bytes, 1);
4385                 btrfs_put_delayed_ref(&head->node);
4386                 cond_resched();
4387                 spin_lock(&delayed_refs->lock);
4388         }
4389
4390         spin_unlock(&delayed_refs->lock);
4391
4392         return ret;
4393 }
4394
4395 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4396 {
4397         struct btrfs_inode *btrfs_inode;
4398         struct list_head splice;
4399
4400         INIT_LIST_HEAD(&splice);
4401
4402         spin_lock(&root->delalloc_lock);
4403         list_splice_init(&root->delalloc_inodes, &splice);
4404
4405         while (!list_empty(&splice)) {
4406                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4407                                                delalloc_inodes);
4408
4409                 list_del_init(&btrfs_inode->delalloc_inodes);
4410                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4411                           &btrfs_inode->runtime_flags);
4412                 spin_unlock(&root->delalloc_lock);
4413
4414                 btrfs_invalidate_inodes(btrfs_inode->root);
4415
4416                 spin_lock(&root->delalloc_lock);
4417         }
4418
4419         spin_unlock(&root->delalloc_lock);
4420 }
4421
4422 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4423 {
4424         struct btrfs_root *root;
4425         struct list_head splice;
4426
4427         INIT_LIST_HEAD(&splice);
4428
4429         spin_lock(&fs_info->delalloc_root_lock);
4430         list_splice_init(&fs_info->delalloc_roots, &splice);
4431         while (!list_empty(&splice)) {
4432                 root = list_first_entry(&splice, struct btrfs_root,
4433                                          delalloc_root);
4434                 list_del_init(&root->delalloc_root);
4435                 root = btrfs_grab_fs_root(root);
4436                 BUG_ON(!root);
4437                 spin_unlock(&fs_info->delalloc_root_lock);
4438
4439                 btrfs_destroy_delalloc_inodes(root);
4440                 btrfs_put_fs_root(root);
4441
4442                 spin_lock(&fs_info->delalloc_root_lock);
4443         }
4444         spin_unlock(&fs_info->delalloc_root_lock);
4445 }
4446
4447 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4448                                         struct extent_io_tree *dirty_pages,
4449                                         int mark)
4450 {
4451         int ret;
4452         struct extent_buffer *eb;
4453         u64 start = 0;
4454         u64 end;
4455
4456         while (1) {
4457                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4458                                             mark, NULL);
4459                 if (ret)
4460                         break;
4461
4462                 clear_extent_bits(dirty_pages, start, end, mark);
4463                 while (start <= end) {
4464                         eb = find_extent_buffer(fs_info, start);
4465                         start += fs_info->nodesize;
4466                         if (!eb)
4467                                 continue;
4468                         wait_on_extent_buffer_writeback(eb);
4469
4470                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4471                                                &eb->bflags))
4472                                 clear_extent_buffer_dirty(eb);
4473                         free_extent_buffer_stale(eb);
4474                 }
4475         }
4476
4477         return ret;
4478 }
4479
4480 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4481                                        struct extent_io_tree *pinned_extents)
4482 {
4483         struct extent_io_tree *unpin;
4484         u64 start;
4485         u64 end;
4486         int ret;
4487         bool loop = true;
4488
4489         unpin = pinned_extents;
4490 again:
4491         while (1) {
4492                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4493                                             EXTENT_DIRTY, NULL);
4494                 if (ret)
4495                         break;
4496
4497                 clear_extent_dirty(unpin, start, end);
4498                 btrfs_error_unpin_extent_range(fs_info, start, end);
4499                 cond_resched();
4500         }
4501
4502         if (loop) {
4503                 if (unpin == &fs_info->freed_extents[0])
4504                         unpin = &fs_info->freed_extents[1];
4505                 else
4506                         unpin = &fs_info->freed_extents[0];
4507                 loop = false;
4508                 goto again;
4509         }
4510
4511         return 0;
4512 }
4513
4514 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4515 {
4516         struct inode *inode;
4517
4518         inode = cache->io_ctl.inode;
4519         if (inode) {
4520                 invalidate_inode_pages2(inode->i_mapping);
4521                 BTRFS_I(inode)->generation = 0;
4522                 cache->io_ctl.inode = NULL;
4523                 iput(inode);
4524         }
4525         btrfs_put_block_group(cache);
4526 }
4527
4528 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4529                              struct btrfs_fs_info *fs_info)
4530 {
4531         struct btrfs_block_group_cache *cache;
4532
4533         spin_lock(&cur_trans->dirty_bgs_lock);
4534         while (!list_empty(&cur_trans->dirty_bgs)) {
4535                 cache = list_first_entry(&cur_trans->dirty_bgs,
4536                                          struct btrfs_block_group_cache,
4537                                          dirty_list);
4538                 if (!cache) {
4539                         btrfs_err(fs_info, "orphan block group dirty_bgs list");
4540                         spin_unlock(&cur_trans->dirty_bgs_lock);
4541                         return;
4542                 }
4543
4544                 if (!list_empty(&cache->io_list)) {
4545                         spin_unlock(&cur_trans->dirty_bgs_lock);
4546                         list_del_init(&cache->io_list);
4547                         btrfs_cleanup_bg_io(cache);
4548                         spin_lock(&cur_trans->dirty_bgs_lock);
4549                 }
4550
4551                 list_del_init(&cache->dirty_list);
4552                 spin_lock(&cache->lock);
4553                 cache->disk_cache_state = BTRFS_DC_ERROR;
4554                 spin_unlock(&cache->lock);
4555
4556                 spin_unlock(&cur_trans->dirty_bgs_lock);
4557                 btrfs_put_block_group(cache);
4558                 spin_lock(&cur_trans->dirty_bgs_lock);
4559         }
4560         spin_unlock(&cur_trans->dirty_bgs_lock);
4561
4562         while (!list_empty(&cur_trans->io_bgs)) {
4563                 cache = list_first_entry(&cur_trans->io_bgs,
4564                                          struct btrfs_block_group_cache,
4565                                          io_list);
4566                 if (!cache) {
4567                         btrfs_err(fs_info, "orphan block group on io_bgs list");
4568                         return;
4569                 }
4570
4571                 list_del_init(&cache->io_list);
4572                 spin_lock(&cache->lock);
4573                 cache->disk_cache_state = BTRFS_DC_ERROR;
4574                 spin_unlock(&cache->lock);
4575                 btrfs_cleanup_bg_io(cache);
4576         }
4577 }
4578
4579 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4580                                    struct btrfs_fs_info *fs_info)
4581 {
4582         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4583         ASSERT(list_empty(&cur_trans->dirty_bgs));
4584         ASSERT(list_empty(&cur_trans->io_bgs));
4585
4586         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4587
4588         cur_trans->state = TRANS_STATE_COMMIT_START;
4589         wake_up(&fs_info->transaction_blocked_wait);
4590
4591         cur_trans->state = TRANS_STATE_UNBLOCKED;
4592         wake_up(&fs_info->transaction_wait);
4593
4594         btrfs_destroy_delayed_inodes(fs_info);
4595         btrfs_assert_delayed_root_empty(fs_info);
4596
4597         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4598                                      EXTENT_DIRTY);
4599         btrfs_destroy_pinned_extent(fs_info,
4600                                     fs_info->pinned_extents);
4601
4602         cur_trans->state =TRANS_STATE_COMPLETED;
4603         wake_up(&cur_trans->commit_wait);
4604
4605         /*
4606         memset(cur_trans, 0, sizeof(*cur_trans));
4607         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4608         */
4609 }
4610
4611 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4612 {
4613         struct btrfs_transaction *t;
4614
4615         mutex_lock(&fs_info->transaction_kthread_mutex);
4616
4617         spin_lock(&fs_info->trans_lock);
4618         while (!list_empty(&fs_info->trans_list)) {
4619                 t = list_first_entry(&fs_info->trans_list,
4620                                      struct btrfs_transaction, list);
4621                 if (t->state >= TRANS_STATE_COMMIT_START) {
4622                         atomic_inc(&t->use_count);
4623                         spin_unlock(&fs_info->trans_lock);
4624                         btrfs_wait_for_commit(fs_info, t->transid);
4625                         btrfs_put_transaction(t);
4626                         spin_lock(&fs_info->trans_lock);
4627                         continue;
4628                 }
4629                 if (t == fs_info->running_transaction) {
4630                         t->state = TRANS_STATE_COMMIT_DOING;
4631                         spin_unlock(&fs_info->trans_lock);
4632                         /*
4633                          * We wait for 0 num_writers since we don't hold a trans
4634                          * handle open currently for this transaction.
4635                          */
4636                         wait_event(t->writer_wait,
4637                                    atomic_read(&t->num_writers) == 0);
4638                 } else {
4639                         spin_unlock(&fs_info->trans_lock);
4640                 }
4641                 btrfs_cleanup_one_transaction(t, fs_info);
4642
4643                 spin_lock(&fs_info->trans_lock);
4644                 if (t == fs_info->running_transaction)
4645                         fs_info->running_transaction = NULL;
4646                 list_del_init(&t->list);
4647                 spin_unlock(&fs_info->trans_lock);
4648
4649                 btrfs_put_transaction(t);
4650                 trace_btrfs_transaction_commit(fs_info->tree_root);
4651                 spin_lock(&fs_info->trans_lock);
4652         }
4653         spin_unlock(&fs_info->trans_lock);
4654         btrfs_destroy_all_ordered_extents(fs_info);
4655         btrfs_destroy_delayed_inodes(fs_info);
4656         btrfs_assert_delayed_root_empty(fs_info);
4657         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4658         btrfs_destroy_all_delalloc_inodes(fs_info);
4659         mutex_unlock(&fs_info->transaction_kthread_mutex);
4660
4661         return 0;
4662 }
4663
4664 static const struct extent_io_ops btree_extent_io_ops = {
4665         .readpage_end_io_hook = btree_readpage_end_io_hook,
4666         .readpage_io_failed_hook = btree_io_failed_hook,
4667         .submit_bio_hook = btree_submit_bio_hook,
4668         /* note we're sharing with inode.c for the merge bio hook */
4669         .merge_bio_hook = btrfs_merge_bio_hook,
4670 };