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