fs/btrfs/disk-io.c

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