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 "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  64                                       struct btrfs_root *root);
  65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  67                                         struct extent_io_tree *dirty_pages,
  68                                         int mark);
  69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  70                                        struct extent_io_tree *pinned_extents);
  71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  72 static void btrfs_error_commit_super(struct btrfs_root *root);
  73
  74 /*
  75  * end_io_wq structs are used to do processing in task context when an IO is
  76  * complete.  This is used during reads to verify checksums, and it is used
  77  * by writes to insert metadata for new file extents after IO is complete.
  78  */
  79 struct end_io_wq {
  80         struct bio *bio;
  81         bio_end_io_t *end_io;
  82         void *private;
  83         struct btrfs_fs_info *info;
  84         int error;
  85         int metadata;
  86         struct list_head list;
  87         struct btrfs_work work;
  88 };
  89
  90 /*
  91  * async submit bios are used to offload expensive checksumming
  92  * onto the worker threads.  They checksum file and metadata bios
  93  * just before they are sent down the IO stack.
  94  */
  95 struct async_submit_bio {
  96         struct inode *inode;
  97         struct bio *bio;
  98         struct list_head list;
  99         extent_submit_bio_hook_t *submit_bio_start;
 100         extent_submit_bio_hook_t *submit_bio_done;
 101         int rw;
 102         int mirror_num;
 103         unsigned long bio_flags;
 104         /*
 105          * bio_offset is optional, can be used if the pages in the bio
 106          * can't tell us where in the file the bio should go
 107          */
 108         u64 bio_offset;
 109         struct btrfs_work work;
 110         int error;
 111 };
 112
 113 /*
 114  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 115  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 116  * the level the eb occupies in the tree.
 117  *
 118  * Different roots are used for different purposes and may nest inside each
 119  * other and they require separate keysets.  As lockdep keys should be
 120  * static, assign keysets according to the purpose of the root as indicated
 121  * by btrfs_root->objectid.  This ensures that all special purpose roots
 122  * have separate keysets.
 123  *
 124  * Lock-nesting across peer nodes is always done with the immediate parent
 125  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 126  * subclass to avoid triggering lockdep warning in such cases.
 127  *
 128  * The key is set by the readpage_end_io_hook after the buffer has passed
 129  * csum validation but before the pages are unlocked.  It is also set by
 130  * btrfs_init_new_buffer on freshly allocated blocks.
 131  *
 132  * We also add a check to make sure the highest level of the tree is the
 133  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 134  * needs update as well.
 135  */
 136 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 137 # if BTRFS_MAX_LEVEL != 8
 138 #  error
 139 # endif
 140
 141 static struct btrfs_lockdep_keyset {
 142         u64                     id;             /* root objectid */
 143         const char              *name_stem;     /* lock name stem */
 144         char                    names[BTRFS_MAX_LEVEL + 1][20];
 145         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 146 } btrfs_lockdep_keysets[] = {
 147         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 148         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 149         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 150         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 151         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 152         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 153         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 154         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 155         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 156         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 157         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 158         { .id = 0,                              .name_stem = "tree"     },
 159 };
 160
 161 void __init btrfs_init_lockdep(void)
 162 {
 163         int i, j;
 164
 165         /* initialize lockdep class names */
 166         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 167                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 168
 169                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 170                         snprintf(ks->names[j], sizeof(ks->names[j]),
 171                                  "btrfs-%s-%02d", ks->name_stem, j);
 172         }
 173 }
 174
 175 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 176                                     int level)
 177 {
 178         struct btrfs_lockdep_keyset *ks;
 179
 180         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 181
 182         /* find the matching keyset, id 0 is the default entry */
 183         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 184                 if (ks->id == objectid)
 185                         break;
 186
 187         lockdep_set_class_and_name(&eb->lock,
 188                                    &ks->keys[level], ks->names[level]);
 189 }
 190
 191 #endif
 192
 193 /*
 194  * extents on the btree inode are pretty simple, there's one extent
 195  * that covers the entire device
 196  */
 197 static struct extent_map *btree_get_extent(struct inode *inode,
 198                 struct page *page, size_t pg_offset, u64 start, u64 len,
 199                 int create)
 200 {
 201         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 202         struct extent_map *em;
 203         int ret;
 204
 205         read_lock(&em_tree->lock);
 206         em = lookup_extent_mapping(em_tree, start, len);
 207         if (em) {
 208                 em->bdev =
 209                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 210                 read_unlock(&em_tree->lock);
 211                 goto out;
 212         }
 213         read_unlock(&em_tree->lock);
 214
 215         em = alloc_extent_map();
 216         if (!em) {
 217                 em = ERR_PTR(-ENOMEM);
 218                 goto out;
 219         }
 220         em->start = 0;
 221         em->len = (u64)-1;
 222         em->block_len = (u64)-1;
 223         em->block_start = 0;
 224         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 225
 226         write_lock(&em_tree->lock);
 227         ret = add_extent_mapping(em_tree, em, 0);
 228         if (ret == -EEXIST) {
 229                 free_extent_map(em);
 230                 em = lookup_extent_mapping(em_tree, start, len);
 231                 if (!em)
 232                         em = ERR_PTR(-EIO);
 233         } else if (ret) {
 234                 free_extent_map(em);
 235                 em = ERR_PTR(ret);
 236         }
 237         write_unlock(&em_tree->lock);
 238
 239 out:
 240         return em;
 241 }
 242
 243 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 244 {
 245         return btrfs_crc32c(seed, data, len);
 246 }
 247
 248 void btrfs_csum_final(u32 crc, char *result)
 249 {
 250         put_unaligned_le32(~crc, result);
 251 }
 252
 253 /*
 254  * compute the csum for a btree block, and either verify it or write it
 255  * into the csum field of the block.
 256  */
 257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 258                            int verify)
 259 {
 260         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 261         char *result = NULL;
 262         unsigned long len;
 263         unsigned long cur_len;
 264         unsigned long offset = BTRFS_CSUM_SIZE;
 265         char *kaddr;
 266         unsigned long map_start;
 267         unsigned long map_len;
 268         int err;
 269         u32 crc = ~(u32)0;
 270         unsigned long inline_result;
 271
 272         len = buf->len - offset;
 273         while (len > 0) {
 274                 err = map_private_extent_buffer(buf, offset, 32,
 275                                         &kaddr, &map_start, &map_len);
 276                 if (err)
 277                         return 1;
 278                 cur_len = min(len, map_len - (offset - map_start));
 279                 crc = btrfs_csum_data(kaddr + offset - map_start,
 280                                       crc, cur_len);
 281                 len -= cur_len;
 282                 offset += cur_len;
 283         }
 284         if (csum_size > sizeof(inline_result)) {
 285                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 286                 if (!result)
 287                         return 1;
 288         } else {
 289                 result = (char *)&inline_result;
 290         }
 291
 292         btrfs_csum_final(crc, result);
 293
 294         if (verify) {
 295                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 296                         u32 val;
 297                         u32 found = 0;
 298                         memcpy(&found, result, csum_size);
 299
 300                         read_extent_buffer(buf, &val, 0, csum_size);
 301                         printk_ratelimited(KERN_INFO
 302                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 303                                 "level %d\n",
 304                                 root->fs_info->sb->s_id, buf->start,
 305                                 val, found, btrfs_header_level(buf));
 306                         if (result != (char *)&inline_result)
 307                                 kfree(result);
 308                         return 1;
 309                 }
 310         } else {
 311                 write_extent_buffer(buf, result, 0, csum_size);
 312         }
 313         if (result != (char *)&inline_result)
 314                 kfree(result);
 315         return 0;
 316 }
 317
 318 /*
 319  * we can't consider a given block up to date unless the transid of the
 320  * block matches the transid in the parent node's pointer.  This is how we
 321  * detect blocks that either didn't get written at all or got written
 322  * in the wrong place.
 323  */
 324 static int verify_parent_transid(struct extent_io_tree *io_tree,
 325                                  struct extent_buffer *eb, u64 parent_transid,
 326                                  int atomic)
 327 {
 328         struct extent_state *cached_state = NULL;
 329         int ret;
 330         bool need_lock = (current->journal_info ==
 331                           (void *)BTRFS_SEND_TRANS_STUB);
 332
 333         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 334                 return 0;
 335
 336         if (atomic)
 337                 return -EAGAIN;
 338
 339         if (need_lock) {
 340                 btrfs_tree_read_lock(eb);
 341                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 342         }
 343
 344         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 345                          0, &cached_state);
 346         if (extent_buffer_uptodate(eb) &&
 347             btrfs_header_generation(eb) == parent_transid) {
 348                 ret = 0;
 349                 goto out;
 350         }
 351         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 352                        "found %llu\n",
 353                        eb->start, parent_transid, btrfs_header_generation(eb));
 354         ret = 1;
 355
 356         /*
 357          * Things reading via commit roots that don't have normal protection,
 358          * like send, can have a really old block in cache that may point at a
 359          * block that has been free'd and re-allocated.  So don't clear uptodate
 360          * if we find an eb that is under IO (dirty/writeback) because we could
 361          * end up reading in the stale data and then writing it back out and
 362          * making everybody very sad.
 363          */
 364         if (!extent_buffer_under_io(eb))
 365                 clear_extent_buffer_uptodate(eb);
 366 out:
 367         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 368                              &cached_state, GFP_NOFS);
 369         if (need_lock)
 370                 btrfs_tree_read_unlock_blocking(eb);
 371         return ret;
 372 }
 373
 374 /*
 375  * Return 0 if the superblock checksum type matches the checksum value of that
 376  * algorithm. Pass the raw disk superblock data.
 377  */
 378 static int btrfs_check_super_csum(char *raw_disk_sb)
 379 {
 380         struct btrfs_super_block *disk_sb =
 381                 (struct btrfs_super_block *)raw_disk_sb;
 382         u16 csum_type = btrfs_super_csum_type(disk_sb);
 383         int ret = 0;
 384
 385         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 386                 u32 crc = ~(u32)0;
 387                 const int csum_size = sizeof(crc);
 388                 char result[csum_size];
 389
 390                 /*
 391                  * The super_block structure does not span the whole
 392                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 393                  * is filled with zeros and is included in the checkum.
 394                  */
 395                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 396                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 397                 btrfs_csum_final(crc, result);
 398
 399                 if (memcmp(raw_disk_sb, result, csum_size))
 400                         ret = 1;
 401
 402                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 403                         printk(KERN_WARNING
 404                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 405                         ret = 0;
 406                 }
 407         }
 408
 409         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 410                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 411                                 csum_type);
 412                 ret = 1;
 413         }
 414
 415         return ret;
 416 }
 417
 418 /*
 419  * helper to read a given tree block, doing retries as required when
 420  * the checksums don't match and we have alternate mirrors to try.
 421  */
 422 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 423                                           struct extent_buffer *eb,
 424                                           u64 start, u64 parent_transid)
 425 {
 426         struct extent_io_tree *io_tree;
 427         int failed = 0;
 428         int ret;
 429         int num_copies = 0;
 430         int mirror_num = 0;
 431         int failed_mirror = 0;
 432
 433         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 434         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 435         while (1) {
 436                 ret = read_extent_buffer_pages(io_tree, eb, start,
 437                                                WAIT_COMPLETE,
 438                                                btree_get_extent, mirror_num);
 439                 if (!ret) {
 440                         if (!verify_parent_transid(io_tree, eb,
 441                                                    parent_transid, 0))
 442                                 break;
 443                         else
 444                                 ret = -EIO;
 445                 }
 446
 447                 /*
 448                  * This buffer's crc is fine, but its contents are corrupted, so
 449                  * there is no reason to read the other copies, they won't be
 450                  * any less wrong.
 451                  */
 452                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 453                         break;
 454
 455                 num_copies = btrfs_num_copies(root->fs_info,
 456                                               eb->start, eb->len);
 457                 if (num_copies == 1)
 458                         break;
 459
 460                 if (!failed_mirror) {
 461                         failed = 1;
 462                         failed_mirror = eb->read_mirror;
 463                 }
 464
 465                 mirror_num++;
 466                 if (mirror_num == failed_mirror)
 467                         mirror_num++;
 468
 469                 if (mirror_num > num_copies)
 470                         break;
 471         }
 472
 473         if (failed && !ret && failed_mirror)
 474                 repair_eb_io_failure(root, eb, failed_mirror);
 475
 476         return ret;
 477 }
 478
 479 /*
 480  * checksum a dirty tree block before IO.  This has extra checks to make sure
 481  * we only fill in the checksum field in the first page of a multi-page block
 482  */
 483
 484 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 485 {
 486         u64 start = page_offset(page);
 487         u64 found_start;
 488         struct extent_buffer *eb;
 489
 490         eb = (struct extent_buffer *)page->private;
 491         if (page != eb->pages[0])
 492                 return 0;
 493         found_start = btrfs_header_bytenr(eb);
 494         if (WARN_ON(found_start != start || !PageUptodate(page)))
 495                 return 0;
 496         csum_tree_block(root, eb, 0);
 497         return 0;
 498 }
 499
 500 static int check_tree_block_fsid(struct btrfs_root *root,
 501                                  struct extent_buffer *eb)
 502 {
 503         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 504         u8 fsid[BTRFS_UUID_SIZE];
 505         int ret = 1;
 506
 507         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 508         while (fs_devices) {
 509                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 510                         ret = 0;
 511                         break;
 512                 }
 513                 fs_devices = fs_devices->seed;
 514         }
 515         return ret;
 516 }
 517
 518 #define CORRUPT(reason, eb, root, slot)                         \
 519         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 520                    "root=%llu, slot=%d", reason,                        \
 521                btrfs_header_bytenr(eb), root->objectid, slot)
 522
 523 static noinline int check_leaf(struct btrfs_root *root,
 524                                struct extent_buffer *leaf)
 525 {
 526         struct btrfs_key key;
 527         struct btrfs_key leaf_key;
 528         u32 nritems = btrfs_header_nritems(leaf);
 529         int slot;
 530
 531         if (nritems == 0)
 532                 return 0;
 533
 534         /* Check the 0 item */
 535         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 536             BTRFS_LEAF_DATA_SIZE(root)) {
 537                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 538                 return -EIO;
 539         }
 540
 541         /*
 542          * Check to make sure each items keys are in the correct order and their
 543          * offsets make sense.  We only have to loop through nritems-1 because
 544          * we check the current slot against the next slot, which verifies the
 545          * next slot's offset+size makes sense and that the current's slot
 546          * offset is correct.
 547          */
 548         for (slot = 0; slot < nritems - 1; slot++) {
 549                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 550                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 551
 552                 /* Make sure the keys are in the right order */
 553                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 554                         CORRUPT("bad key order", leaf, root, slot);
 555                         return -EIO;
 556                 }
 557
 558                 /*
 559                  * Make sure the offset and ends are right, remember that the
 560                  * item data starts at the end of the leaf and grows towards the
 561                  * front.
 562                  */
 563                 if (btrfs_item_offset_nr(leaf, slot) !=
 564                         btrfs_item_end_nr(leaf, slot + 1)) {
 565                         CORRUPT("slot offset bad", leaf, root, slot);
 566                         return -EIO;
 567                 }
 568
 569                 /*
 570                  * Check to make sure that we don't point outside of the leaf,
 571                  * just incase all the items are consistent to eachother, but
 572                  * all point outside of the leaf.
 573                  */
 574                 if (btrfs_item_end_nr(leaf, slot) >
 575                     BTRFS_LEAF_DATA_SIZE(root)) {
 576                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 577                         return -EIO;
 578                 }
 579         }
 580
 581         return 0;
 582 }
 583
 584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 585                                       u64 phy_offset, struct page *page,
 586                                       u64 start, u64 end, int mirror)
 587 {
 588         u64 found_start;
 589         int found_level;
 590         struct extent_buffer *eb;
 591         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 592         int ret = 0;
 593         int reads_done;
 594
 595         if (!page->private)
 596                 goto out;
 597
 598         eb = (struct extent_buffer *)page->private;
 599
 600         /* the pending IO might have been the only thing that kept this buffer
 601          * in memory.  Make sure we have a ref for all this other checks
 602          */
 603         extent_buffer_get(eb);
 604
 605         reads_done = atomic_dec_and_test(&eb->io_pages);
 606         if (!reads_done)
 607                 goto err;
 608
 609         eb->read_mirror = mirror;
 610         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
 611                 ret = -EIO;
 612                 goto err;
 613         }
 614
 615         found_start = btrfs_header_bytenr(eb);
 616         if (found_start != eb->start) {
 617                 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
 618                                "%llu %llu\n",
 619                                found_start, eb->start);
 620                 ret = -EIO;
 621                 goto err;
 622         }
 623         if (check_tree_block_fsid(root, eb)) {
 624                 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
 625                                eb->start);
 626                 ret = -EIO;
 627                 goto err;
 628         }
 629         found_level = btrfs_header_level(eb);
 630         if (found_level >= BTRFS_MAX_LEVEL) {
 631                 btrfs_info(root->fs_info, "bad tree block level %d",
 632                            (int)btrfs_header_level(eb));
 633                 ret = -EIO;
 634                 goto err;
 635         }
 636
 637         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 638                                        eb, found_level);
 639
 640         ret = csum_tree_block(root, eb, 1);
 641         if (ret) {
 642                 ret = -EIO;
 643                 goto err;
 644         }
 645
 646         /*
 647          * If this is a leaf block and it is corrupt, set the corrupt bit so
 648          * that we don't try and read the other copies of this block, just
 649          * return -EIO.
 650          */
 651         if (found_level == 0 && check_leaf(root, eb)) {
 652                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 653                 ret = -EIO;
 654         }
 655
 656         if (!ret)
 657                 set_extent_buffer_uptodate(eb);
 658 err:
 659         if (reads_done &&
 660             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 661                 btree_readahead_hook(root, eb, eb->start, ret);
 662
 663         if (ret) {
 664                 /*
 665                  * our io error hook is going to dec the io pages
 666                  * again, we have to make sure it has something
 667                  * to decrement
 668                  */
 669                 atomic_inc(&eb->io_pages);
 670                 clear_extent_buffer_uptodate(eb);
 671         }
 672         free_extent_buffer(eb);
 673 out:
 674         return ret;
 675 }
 676
 677 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 678 {
 679         struct extent_buffer *eb;
 680         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 681
 682         eb = (struct extent_buffer *)page->private;
 683         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
 684         eb->read_mirror = failed_mirror;
 685         atomic_dec(&eb->io_pages);
 686         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 687                 btree_readahead_hook(root, eb, eb->start, -EIO);
 688         return -EIO;    /* we fixed nothing */
 689 }
 690
 691 static void end_workqueue_bio(struct bio *bio, int err)
 692 {
 693         struct end_io_wq *end_io_wq = bio->bi_private;
 694         struct btrfs_fs_info *fs_info;
 695         struct btrfs_workqueue *wq;
 696         btrfs_work_func_t func;
 697
 698         fs_info = end_io_wq->info;
 699         end_io_wq->error = err;
 700
 701         if (bio->bi_rw & REQ_WRITE) {
 702                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 703                         wq = fs_info->endio_meta_write_workers;
 704                         func = btrfs_endio_meta_write_helper;
 705                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 706                         wq = fs_info->endio_freespace_worker;
 707                         func = btrfs_freespace_write_helper;
 708                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 709                         wq = fs_info->endio_raid56_workers;
 710                         func = btrfs_endio_raid56_helper;
 711                 } else {
 712                         wq = fs_info->endio_write_workers;
 713                         func = btrfs_endio_write_helper;
 714                 }
 715         } else {
 716                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 717                         wq = fs_info->endio_raid56_workers;
 718                         func = btrfs_endio_raid56_helper;
 719                 } else if (end_io_wq->metadata) {
 720                         wq = fs_info->endio_meta_workers;
 721                         func = btrfs_endio_meta_helper;
 722                 } else {
 723                         wq = fs_info->endio_workers;
 724                         func = btrfs_endio_helper;
 725                 }
 726         }
 727
 728         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 729         btrfs_queue_work(wq, &end_io_wq->work);
 730 }
 731
 732 /*
 733  * For the metadata arg you want
 734  *
 735  * 0 - if data
 736  * 1 - if normal metadta
 737  * 2 - if writing to the free space cache area
 738  * 3 - raid parity work
 739  */
 740 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 741                         int metadata)
 742 {
 743         struct end_io_wq *end_io_wq;
 744         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 745         if (!end_io_wq)
 746                 return -ENOMEM;
 747
 748         end_io_wq->private = bio->bi_private;
 749         end_io_wq->end_io = bio->bi_end_io;
 750         end_io_wq->info = info;
 751         end_io_wq->error = 0;
 752         end_io_wq->bio = bio;
 753         end_io_wq->metadata = metadata;
 754
 755         bio->bi_private = end_io_wq;
 756         bio->bi_end_io = end_workqueue_bio;
 757         return 0;
 758 }
 759
 760 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 761 {
 762         unsigned long limit = min_t(unsigned long,
 763                                     info->thread_pool_size,
 764                                     info->fs_devices->open_devices);
 765         return 256 * limit;
 766 }
 767
 768 static void run_one_async_start(struct btrfs_work *work)
 769 {
 770         struct async_submit_bio *async;
 771         int ret;
 772
 773         async = container_of(work, struct  async_submit_bio, work);
 774         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 775                                       async->mirror_num, async->bio_flags,
 776                                       async->bio_offset);
 777         if (ret)
 778                 async->error = ret;
 779 }
 780
 781 static void run_one_async_done(struct btrfs_work *work)
 782 {
 783         struct btrfs_fs_info *fs_info;
 784         struct async_submit_bio *async;
 785         int limit;
 786
 787         async = container_of(work, struct  async_submit_bio, work);
 788         fs_info = BTRFS_I(async->inode)->root->fs_info;
 789
 790         limit = btrfs_async_submit_limit(fs_info);
 791         limit = limit * 2 / 3;
 792
 793         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 794             waitqueue_active(&fs_info->async_submit_wait))
 795                 wake_up(&fs_info->async_submit_wait);
 796
 797         /* If an error occured we just want to clean up the bio and move on */
 798         if (async->error) {
 799                 bio_endio(async->bio, async->error);
 800                 return;
 801         }
 802
 803         async->submit_bio_done(async->inode, async->rw, async->bio,
 804                                async->mirror_num, async->bio_flags,
 805                                async->bio_offset);
 806 }
 807
 808 static void run_one_async_free(struct btrfs_work *work)
 809 {
 810         struct async_submit_bio *async;
 811
 812         async = container_of(work, struct  async_submit_bio, work);
 813         kfree(async);
 814 }
 815
 816 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 817                         int rw, struct bio *bio, int mirror_num,
 818                         unsigned long bio_flags,
 819                         u64 bio_offset,
 820                         extent_submit_bio_hook_t *submit_bio_start,
 821                         extent_submit_bio_hook_t *submit_bio_done)
 822 {
 823         struct async_submit_bio *async;
 824
 825         async = kmalloc(sizeof(*async), GFP_NOFS);
 826         if (!async)
 827                 return -ENOMEM;
 828
 829         async->inode = inode;
 830         async->rw = rw;
 831         async->bio = bio;
 832         async->mirror_num = mirror_num;
 833         async->submit_bio_start = submit_bio_start;
 834         async->submit_bio_done = submit_bio_done;
 835
 836         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 837                         run_one_async_done, run_one_async_free);
 838
 839         async->bio_flags = bio_flags;
 840         async->bio_offset = bio_offset;
 841
 842         async->error = 0;
 843
 844         atomic_inc(&fs_info->nr_async_submits);
 845
 846         if (rw & REQ_SYNC)
 847                 btrfs_set_work_high_priority(&async->work);
 848
 849         btrfs_queue_work(fs_info->workers, &async->work);
 850
 851         while (atomic_read(&fs_info->async_submit_draining) &&
 852               atomic_read(&fs_info->nr_async_submits)) {
 853                 wait_event(fs_info->async_submit_wait,
 854                            (atomic_read(&fs_info->nr_async_submits) == 0));
 855         }
 856
 857         return 0;
 858 }
 859
 860 static int btree_csum_one_bio(struct bio *bio)
 861 {
 862         struct bio_vec *bvec;
 863         struct btrfs_root *root;
 864         int i, ret = 0;
 865
 866         bio_for_each_segment_all(bvec, bio, i) {
 867                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 868                 ret = csum_dirty_buffer(root, bvec->bv_page);
 869                 if (ret)
 870                         break;
 871         }
 872
 873         return ret;
 874 }
 875
 876 static int __btree_submit_bio_start(struct inode *inode, int rw,
 877                                     struct bio *bio, int mirror_num,
 878                                     unsigned long bio_flags,
 879                                     u64 bio_offset)
 880 {
 881         /*
 882          * when we're called for a write, we're already in the async
 883          * submission context.  Just jump into btrfs_map_bio
 884          */
 885         return btree_csum_one_bio(bio);
 886 }
 887
 888 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 889                                  int mirror_num, unsigned long bio_flags,
 890                                  u64 bio_offset)
 891 {
 892         int ret;
 893
 894         /*
 895          * when we're called for a write, we're already in the async
 896          * submission context.  Just jump into btrfs_map_bio
 897          */
 898         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 899         if (ret)
 900                 bio_endio(bio, ret);
 901         return ret;
 902 }
 903
 904 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 905 {
 906         if (bio_flags & EXTENT_BIO_TREE_LOG)
 907                 return 0;
 908 #ifdef CONFIG_X86
 909         if (cpu_has_xmm4_2)
 910                 return 0;
 911 #endif
 912         return 1;
 913 }
 914
 915 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 916                                  int mirror_num, unsigned long bio_flags,
 917                                  u64 bio_offset)
 918 {
 919         int async = check_async_write(inode, bio_flags);
 920         int ret;
 921
 922         if (!(rw & REQ_WRITE)) {
 923                 /*
 924                  * called for a read, do the setup so that checksum validation
 925                  * can happen in the async kernel threads
 926                  */
 927                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 928                                           bio, 1);
 929                 if (ret)
 930                         goto out_w_error;
 931                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 932                                     mirror_num, 0);
 933         } else if (!async) {
 934                 ret = btree_csum_one_bio(bio);
 935                 if (ret)
 936                         goto out_w_error;
 937                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 938                                     mirror_num, 0);
 939         } else {
 940                 /*
 941                  * kthread helpers are used to submit writes so that
 942                  * checksumming can happen in parallel across all CPUs
 943                  */
 944                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 945                                           inode, rw, bio, mirror_num, 0,
 946                                           bio_offset,
 947                                           __btree_submit_bio_start,
 948                                           __btree_submit_bio_done);
 949         }
 950
 951         if (ret) {
 952 out_w_error:
 953                 bio_endio(bio, ret);
 954         }
 955         return ret;
 956 }
 957
 958 #ifdef CONFIG_MIGRATION
 959 static int btree_migratepage(struct address_space *mapping,
 960                         struct page *newpage, struct page *page,
 961                         enum migrate_mode mode)
 962 {
 963         /*
 964          * we can't safely write a btree page from here,
 965          * we haven't done the locking hook
 966          */
 967         if (PageDirty(page))
 968                 return -EAGAIN;
 969         /*
 970          * Buffers may be managed in a filesystem specific way.
 971          * We must have no buffers or drop them.
 972          */
 973         if (page_has_private(page) &&
 974             !try_to_release_page(page, GFP_KERNEL))
 975                 return -EAGAIN;
 976         return migrate_page(mapping, newpage, page, mode);
 977 }
 978 #endif
 979
 980
 981 static int btree_writepages(struct address_space *mapping,
 982                             struct writeback_control *wbc)
 983 {
 984         struct btrfs_fs_info *fs_info;
 985         int ret;
 986
 987         if (wbc->sync_mode == WB_SYNC_NONE) {
 988
 989                 if (wbc->for_kupdate)
 990                         return 0;
 991
 992                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 993                 /* this is a bit racy, but that's ok */
 994                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 995                                              BTRFS_DIRTY_METADATA_THRESH);
 996                 if (ret < 0)
 997                         return 0;
 998         }
 999         return btree_write_cache_pages(mapping, wbc);
1000 }
1001
1002 static int btree_readpage(struct file *file, struct page *page)
1003 {
1004         struct extent_io_tree *tree;
1005         tree = &BTRFS_I(page->mapping->host)->io_tree;
1006         return extent_read_full_page(tree, page, btree_get_extent, 0);
1007 }
1008
1009 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1010 {
1011         if (PageWriteback(page) || PageDirty(page))
1012                 return 0;
1013
1014         return try_release_extent_buffer(page);
1015 }
1016
1017 static void btree_invalidatepage(struct page *page, unsigned int offset,
1018                                  unsigned int length)
1019 {
1020         struct extent_io_tree *tree;
1021         tree = &BTRFS_I(page->mapping->host)->io_tree;
1022         extent_invalidatepage(tree, page, offset);
1023         btree_releasepage(page, GFP_NOFS);
1024         if (PagePrivate(page)) {
1025                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1026                            "page private not zero on page %llu",
1027                            (unsigned long long)page_offset(page));
1028                 ClearPagePrivate(page);
1029                 set_page_private(page, 0);
1030                 page_cache_release(page);
1031         }
1032 }
1033
1034 static int btree_set_page_dirty(struct page *page)
1035 {
1036 #ifdef DEBUG
1037         struct extent_buffer *eb;
1038
1039         BUG_ON(!PagePrivate(page));
1040         eb = (struct extent_buffer *)page->private;
1041         BUG_ON(!eb);
1042         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1043         BUG_ON(!atomic_read(&eb->refs));
1044         btrfs_assert_tree_locked(eb);
1045 #endif
1046         return __set_page_dirty_nobuffers(page);
1047 }
1048
1049 static const struct address_space_operations btree_aops = {
1050         .readpage       = btree_readpage,
1051         .writepages     = btree_writepages,
1052         .releasepage    = btree_releasepage,
1053         .invalidatepage = btree_invalidatepage,
1054 #ifdef CONFIG_MIGRATION
1055         .migratepage    = btree_migratepage,
1056 #endif
1057         .set_page_dirty = btree_set_page_dirty,
1058 };
1059
1060 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1061                          u64 parent_transid)
1062 {
1063         struct extent_buffer *buf = NULL;
1064         struct inode *btree_inode = root->fs_info->btree_inode;
1065         int ret = 0;
1066
1067         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1068         if (!buf)
1069                 return 0;
1070         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1071                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1072         free_extent_buffer(buf);
1073         return ret;
1074 }
1075
1076 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077                          int mirror_num, struct extent_buffer **eb)
1078 {
1079         struct extent_buffer *buf = NULL;
1080         struct inode *btree_inode = root->fs_info->btree_inode;
1081         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1082         int ret;
1083
1084         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1085         if (!buf)
1086                 return 0;
1087
1088         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1089
1090         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1091                                        btree_get_extent, mirror_num);
1092         if (ret) {
1093                 free_extent_buffer(buf);
1094                 return ret;
1095         }
1096
1097         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1098                 free_extent_buffer(buf);
1099                 return -EIO;
1100         } else if (extent_buffer_uptodate(buf)) {
1101                 *eb = buf;
1102         } else {
1103                 free_extent_buffer(buf);
1104         }
1105         return 0;
1106 }
1107
1108 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1109                                             u64 bytenr, u32 blocksize)
1110 {
1111         return find_extent_buffer(root->fs_info, bytenr);
1112 }
1113
1114 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1115                                                  u64 bytenr, u32 blocksize)
1116 {
1117 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1118         if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1119                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1120                                                 blocksize);
1121 #endif
1122         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1123 }
1124
1125
1126 int btrfs_write_tree_block(struct extent_buffer *buf)
1127 {
1128         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1129                                         buf->start + buf->len - 1);
1130 }
1131
1132 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1133 {
1134         return filemap_fdatawait_range(buf->pages[0]->mapping,
1135                                        buf->start, buf->start + buf->len - 1);
1136 }
1137
1138 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1139                                       u32 blocksize, u64 parent_transid)
1140 {
1141         struct extent_buffer *buf = NULL;
1142         int ret;
1143
1144         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1145         if (!buf)
1146                 return NULL;
1147
1148         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1149         if (ret) {
1150                 free_extent_buffer(buf);
1151                 return NULL;
1152         }
1153         return buf;
1154
1155 }
1156
1157 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1158                       struct extent_buffer *buf)
1159 {
1160         struct btrfs_fs_info *fs_info = root->fs_info;
1161
1162         if (btrfs_header_generation(buf) ==
1163             fs_info->running_transaction->transid) {
1164                 btrfs_assert_tree_locked(buf);
1165
1166                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1167                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1168                                              -buf->len,
1169                                              fs_info->dirty_metadata_batch);
1170                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1171                         btrfs_set_lock_blocking(buf);
1172                         clear_extent_buffer_dirty(buf);
1173                 }
1174         }
1175 }
1176
1177 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1178 {
1179         struct btrfs_subvolume_writers *writers;
1180         int ret;
1181
1182         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1183         if (!writers)
1184                 return ERR_PTR(-ENOMEM);
1185
1186         ret = percpu_counter_init(&writers->counter, 0);
1187         if (ret < 0) {
1188                 kfree(writers);
1189                 return ERR_PTR(ret);
1190         }
1191
1192         init_waitqueue_head(&writers->wait);
1193         return writers;
1194 }
1195
1196 static void
1197 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1198 {
1199         percpu_counter_destroy(&writers->counter);
1200         kfree(writers);
1201 }
1202
1203 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1204                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1205                          u64 objectid)
1206 {
1207         root->node = NULL;
1208         root->commit_root = NULL;
1209         root->sectorsize = sectorsize;
1210         root->nodesize = nodesize;
1211         root->stripesize = stripesize;
1212         root->state = 0;
1213         root->orphan_cleanup_state = 0;
1214
1215         root->objectid = objectid;
1216         root->last_trans = 0;
1217         root->highest_objectid = 0;
1218         root->nr_delalloc_inodes = 0;
1219         root->nr_ordered_extents = 0;
1220         root->name = NULL;
1221         root->inode_tree = RB_ROOT;
1222         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1223         root->block_rsv = NULL;
1224         root->orphan_block_rsv = NULL;
1225
1226         INIT_LIST_HEAD(&root->dirty_list);
1227         INIT_LIST_HEAD(&root->root_list);
1228         INIT_LIST_HEAD(&root->delalloc_inodes);
1229         INIT_LIST_HEAD(&root->delalloc_root);
1230         INIT_LIST_HEAD(&root->ordered_extents);
1231         INIT_LIST_HEAD(&root->ordered_root);
1232         INIT_LIST_HEAD(&root->logged_list[0]);
1233         INIT_LIST_HEAD(&root->logged_list[1]);
1234         spin_lock_init(&root->orphan_lock);
1235         spin_lock_init(&root->inode_lock);
1236         spin_lock_init(&root->delalloc_lock);
1237         spin_lock_init(&root->ordered_extent_lock);
1238         spin_lock_init(&root->accounting_lock);
1239         spin_lock_init(&root->log_extents_lock[0]);
1240         spin_lock_init(&root->log_extents_lock[1]);
1241         mutex_init(&root->objectid_mutex);
1242         mutex_init(&root->log_mutex);
1243         mutex_init(&root->ordered_extent_mutex);
1244         mutex_init(&root->delalloc_mutex);
1245         init_waitqueue_head(&root->log_writer_wait);
1246         init_waitqueue_head(&root->log_commit_wait[0]);
1247         init_waitqueue_head(&root->log_commit_wait[1]);
1248         INIT_LIST_HEAD(&root->log_ctxs[0]);
1249         INIT_LIST_HEAD(&root->log_ctxs[1]);
1250         atomic_set(&root->log_commit[0], 0);
1251         atomic_set(&root->log_commit[1], 0);
1252         atomic_set(&root->log_writers, 0);
1253         atomic_set(&root->log_batch, 0);
1254         atomic_set(&root->orphan_inodes, 0);
1255         atomic_set(&root->refs, 1);
1256         atomic_set(&root->will_be_snapshoted, 0);
1257         root->log_transid = 0;
1258         root->log_transid_committed = -1;
1259         root->last_log_commit = 0;
1260         if (fs_info)
1261                 extent_io_tree_init(&root->dirty_log_pages,
1262                                      fs_info->btree_inode->i_mapping);
1263
1264         memset(&root->root_key, 0, sizeof(root->root_key));
1265         memset(&root->root_item, 0, sizeof(root->root_item));
1266         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1267         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1268         if (fs_info)
1269                 root->defrag_trans_start = fs_info->generation;
1270         else
1271                 root->defrag_trans_start = 0;
1272         init_completion(&root->kobj_unregister);
1273         root->root_key.objectid = objectid;
1274         root->anon_dev = 0;
1275
1276         spin_lock_init(&root->root_item_lock);
1277 }
1278
1279 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1280 {
1281         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1282         if (root)
1283                 root->fs_info = fs_info;
1284         return root;
1285 }
1286
1287 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1288 /* Should only be used by the testing infrastructure */
1289 struct btrfs_root *btrfs_alloc_dummy_root(void)
1290 {
1291         struct btrfs_root *root;
1292
1293         root = btrfs_alloc_root(NULL);
1294         if (!root)
1295                 return ERR_PTR(-ENOMEM);
1296         __setup_root(4096, 4096, 4096, root, NULL, 1);
1297         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1298         root->alloc_bytenr = 0;
1299
1300         return root;
1301 }
1302 #endif
1303
1304 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1305                                      struct btrfs_fs_info *fs_info,
1306                                      u64 objectid)
1307 {
1308         struct extent_buffer *leaf;
1309         struct btrfs_root *tree_root = fs_info->tree_root;
1310         struct btrfs_root *root;
1311         struct btrfs_key key;
1312         int ret = 0;
1313         uuid_le uuid;
1314
1315         root = btrfs_alloc_root(fs_info);
1316         if (!root)
1317                 return ERR_PTR(-ENOMEM);
1318
1319         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1320                 tree_root->stripesize, root, fs_info, objectid);
1321         root->root_key.objectid = objectid;
1322         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1323         root->root_key.offset = 0;
1324
1325         leaf = btrfs_alloc_free_block(trans, root, root->nodesize,
1326                                       0, objectid, NULL, 0, 0, 0);
1327         if (IS_ERR(leaf)) {
1328                 ret = PTR_ERR(leaf);
1329                 leaf = NULL;
1330                 goto fail;
1331         }
1332
1333         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1334         btrfs_set_header_bytenr(leaf, leaf->start);
1335         btrfs_set_header_generation(leaf, trans->transid);
1336         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1337         btrfs_set_header_owner(leaf, objectid);
1338         root->node = leaf;
1339
1340         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1341                             BTRFS_FSID_SIZE);
1342         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1343                             btrfs_header_chunk_tree_uuid(leaf),
1344                             BTRFS_UUID_SIZE);
1345         btrfs_mark_buffer_dirty(leaf);
1346
1347         root->commit_root = btrfs_root_node(root);
1348         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1349
1350         root->root_item.flags = 0;
1351         root->root_item.byte_limit = 0;
1352         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1353         btrfs_set_root_generation(&root->root_item, trans->transid);
1354         btrfs_set_root_level(&root->root_item, 0);
1355         btrfs_set_root_refs(&root->root_item, 1);
1356         btrfs_set_root_used(&root->root_item, leaf->len);
1357         btrfs_set_root_last_snapshot(&root->root_item, 0);
1358         btrfs_set_root_dirid(&root->root_item, 0);
1359         uuid_le_gen(&uuid);
1360         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1361         root->root_item.drop_level = 0;
1362
1363         key.objectid = objectid;
1364         key.type = BTRFS_ROOT_ITEM_KEY;
1365         key.offset = 0;
1366         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1367         if (ret)
1368                 goto fail;
1369
1370         btrfs_tree_unlock(leaf);
1371
1372         return root;
1373
1374 fail:
1375         if (leaf) {
1376                 btrfs_tree_unlock(leaf);
1377                 free_extent_buffer(root->commit_root);
1378                 free_extent_buffer(leaf);
1379         }
1380         kfree(root);
1381
1382         return ERR_PTR(ret);
1383 }
1384
1385 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1386                                          struct btrfs_fs_info *fs_info)
1387 {
1388         struct btrfs_root *root;
1389         struct btrfs_root *tree_root = fs_info->tree_root;
1390         struct extent_buffer *leaf;
1391
1392         root = btrfs_alloc_root(fs_info);
1393         if (!root)
1394                 return ERR_PTR(-ENOMEM);
1395
1396         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1397                      tree_root->stripesize, root, fs_info,
1398                      BTRFS_TREE_LOG_OBJECTID);
1399
1400         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1401         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1402         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1403
1404         /*
1405          * DON'T set REF_COWS for log trees
1406          *
1407          * log trees do not get reference counted because they go away
1408          * before a real commit is actually done.  They do store pointers
1409          * to file data extents, and those reference counts still get
1410          * updated (along with back refs to the log tree).
1411          */
1412
1413         leaf = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1414                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1415                                       0, 0, 0);
1416         if (IS_ERR(leaf)) {
1417                 kfree(root);
1418                 return ERR_CAST(leaf);
1419         }
1420
1421         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1422         btrfs_set_header_bytenr(leaf, leaf->start);
1423         btrfs_set_header_generation(leaf, trans->transid);
1424         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1425         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1426         root->node = leaf;
1427
1428         write_extent_buffer(root->node, root->fs_info->fsid,
1429                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1430         btrfs_mark_buffer_dirty(root->node);
1431         btrfs_tree_unlock(root->node);
1432         return root;
1433 }
1434
1435 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1436                              struct btrfs_fs_info *fs_info)
1437 {
1438         struct btrfs_root *log_root;
1439
1440         log_root = alloc_log_tree(trans, fs_info);
1441         if (IS_ERR(log_root))
1442                 return PTR_ERR(log_root);
1443         WARN_ON(fs_info->log_root_tree);
1444         fs_info->log_root_tree = log_root;
1445         return 0;
1446 }
1447
1448 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1449                        struct btrfs_root *root)
1450 {
1451         struct btrfs_root *log_root;
1452         struct btrfs_inode_item *inode_item;
1453
1454         log_root = alloc_log_tree(trans, root->fs_info);
1455         if (IS_ERR(log_root))
1456                 return PTR_ERR(log_root);
1457
1458         log_root->last_trans = trans->transid;
1459         log_root->root_key.offset = root->root_key.objectid;
1460
1461         inode_item = &log_root->root_item.inode;
1462         btrfs_set_stack_inode_generation(inode_item, 1);
1463         btrfs_set_stack_inode_size(inode_item, 3);
1464         btrfs_set_stack_inode_nlink(inode_item, 1);
1465         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1466         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1467
1468         btrfs_set_root_node(&log_root->root_item, log_root->node);
1469
1470         WARN_ON(root->log_root);
1471         root->log_root = log_root;
1472         root->log_transid = 0;
1473         root->log_transid_committed = -1;
1474         root->last_log_commit = 0;
1475         return 0;
1476 }
1477
1478 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1479                                                struct btrfs_key *key)
1480 {
1481         struct btrfs_root *root;
1482         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1483         struct btrfs_path *path;
1484         u64 generation;
1485         u32 blocksize;
1486         int ret;
1487
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return ERR_PTR(-ENOMEM);
1491
1492         root = btrfs_alloc_root(fs_info);
1493         if (!root) {
1494                 ret = -ENOMEM;
1495                 goto alloc_fail;
1496         }
1497
1498         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1499                 tree_root->stripesize, root, fs_info, key->objectid);
1500
1501         ret = btrfs_find_root(tree_root, key, path,
1502                               &root->root_item, &root->root_key);
1503         if (ret) {
1504                 if (ret > 0)
1505                         ret = -ENOENT;
1506                 goto find_fail;
1507         }
1508
1509         generation = btrfs_root_generation(&root->root_item);
1510         blocksize = root->nodesize;
1511         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1512                                      blocksize, generation);
1513         if (!root->node) {
1514                 ret = -ENOMEM;
1515                 goto find_fail;
1516         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1517                 ret = -EIO;
1518                 goto read_fail;
1519         }
1520         root->commit_root = btrfs_root_node(root);
1521 out:
1522         btrfs_free_path(path);
1523         return root;
1524
1525 read_fail:
1526         free_extent_buffer(root->node);
1527 find_fail:
1528         kfree(root);
1529 alloc_fail:
1530         root = ERR_PTR(ret);
1531         goto out;
1532 }
1533
1534 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1535                                       struct btrfs_key *location)
1536 {
1537         struct btrfs_root *root;
1538
1539         root = btrfs_read_tree_root(tree_root, location);
1540         if (IS_ERR(root))
1541                 return root;
1542
1543         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1544                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1545                 btrfs_check_and_init_root_item(&root->root_item);
1546         }
1547
1548         return root;
1549 }
1550
1551 int btrfs_init_fs_root(struct btrfs_root *root)
1552 {
1553         int ret;
1554         struct btrfs_subvolume_writers *writers;
1555
1556         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1557         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1558                                         GFP_NOFS);
1559         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1560                 ret = -ENOMEM;
1561                 goto fail;
1562         }
1563
1564         writers = btrfs_alloc_subvolume_writers();
1565         if (IS_ERR(writers)) {
1566                 ret = PTR_ERR(writers);
1567                 goto fail;
1568         }
1569         root->subv_writers = writers;
1570
1571         btrfs_init_free_ino_ctl(root);
1572         spin_lock_init(&root->ino_cache_lock);
1573         init_waitqueue_head(&root->ino_cache_wait);
1574
1575         ret = get_anon_bdev(&root->anon_dev);
1576         if (ret)
1577                 goto free_writers;
1578         return 0;
1579
1580 free_writers:
1581         btrfs_free_subvolume_writers(root->subv_writers);
1582 fail:
1583         kfree(root->free_ino_ctl);
1584         kfree(root->free_ino_pinned);
1585         return ret;
1586 }
1587
1588 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1589                                                u64 root_id)
1590 {
1591         struct btrfs_root *root;
1592
1593         spin_lock(&fs_info->fs_roots_radix_lock);
1594         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1595                                  (unsigned long)root_id);
1596         spin_unlock(&fs_info->fs_roots_radix_lock);
1597         return root;
1598 }
1599
1600 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1601                          struct btrfs_root *root)
1602 {
1603         int ret;
1604
1605         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1606         if (ret)
1607                 return ret;
1608
1609         spin_lock(&fs_info->fs_roots_radix_lock);
1610         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1611                                 (unsigned long)root->root_key.objectid,
1612                                 root);
1613         if (ret == 0)
1614                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1615         spin_unlock(&fs_info->fs_roots_radix_lock);
1616         radix_tree_preload_end();
1617
1618         return ret;
1619 }
1620
1621 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1622                                      struct btrfs_key *location,
1623                                      bool check_ref)
1624 {
1625         struct btrfs_root *root;
1626         int ret;
1627
1628         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1629                 return fs_info->tree_root;
1630         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1631                 return fs_info->extent_root;
1632         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1633                 return fs_info->chunk_root;
1634         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1635                 return fs_info->dev_root;
1636         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1637                 return fs_info->csum_root;
1638         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1639                 return fs_info->quota_root ? fs_info->quota_root :
1640                                              ERR_PTR(-ENOENT);
1641         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1642                 return fs_info->uuid_root ? fs_info->uuid_root :
1643                                             ERR_PTR(-ENOENT);
1644 again:
1645         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1646         if (root) {
1647                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1648                         return ERR_PTR(-ENOENT);
1649                 return root;
1650         }
1651
1652         root = btrfs_read_fs_root(fs_info->tree_root, location);
1653         if (IS_ERR(root))
1654                 return root;
1655
1656         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1657                 ret = -ENOENT;
1658                 goto fail;
1659         }
1660
1661         ret = btrfs_init_fs_root(root);
1662         if (ret)
1663                 goto fail;
1664
1665         ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1666                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1667         if (ret < 0)
1668                 goto fail;
1669         if (ret == 0)
1670                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1671
1672         ret = btrfs_insert_fs_root(fs_info, root);
1673         if (ret) {
1674                 if (ret == -EEXIST) {
1675                         free_fs_root(root);
1676                         goto again;
1677                 }
1678                 goto fail;
1679         }
1680         return root;
1681 fail:
1682         free_fs_root(root);
1683         return ERR_PTR(ret);
1684 }
1685
1686 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1687 {
1688         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1689         int ret = 0;
1690         struct btrfs_device *device;
1691         struct backing_dev_info *bdi;
1692
1693         rcu_read_lock();
1694         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1695                 if (!device->bdev)
1696                         continue;
1697                 bdi = blk_get_backing_dev_info(device->bdev);
1698                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1699                         ret = 1;
1700                         break;
1701                 }
1702         }
1703         rcu_read_unlock();
1704         return ret;
1705 }
1706
1707 /*
1708  * If this fails, caller must call bdi_destroy() to get rid of the
1709  * bdi again.
1710  */
1711 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1712 {
1713         int err;
1714
1715         bdi->capabilities = BDI_CAP_MAP_COPY;
1716         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1717         if (err)
1718                 return err;
1719
1720         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1721         bdi->congested_fn       = btrfs_congested_fn;
1722         bdi->congested_data     = info;
1723         return 0;
1724 }
1725
1726 /*
1727  * called by the kthread helper functions to finally call the bio end_io
1728  * functions.  This is where read checksum verification actually happens
1729  */
1730 static void end_workqueue_fn(struct btrfs_work *work)
1731 {
1732         struct bio *bio;
1733         struct end_io_wq *end_io_wq;
1734         int error;
1735
1736         end_io_wq = container_of(work, struct end_io_wq, work);
1737         bio = end_io_wq->bio;
1738
1739         error = end_io_wq->error;
1740         bio->bi_private = end_io_wq->private;
1741         bio->bi_end_io = end_io_wq->end_io;
1742         kfree(end_io_wq);
1743         bio_endio_nodec(bio, error);
1744 }
1745
1746 static int cleaner_kthread(void *arg)
1747 {
1748         struct btrfs_root *root = arg;
1749         int again;
1750
1751         do {
1752                 again = 0;
1753
1754                 /* Make the cleaner go to sleep early. */
1755                 if (btrfs_need_cleaner_sleep(root))
1756                         goto sleep;
1757
1758                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1759                         goto sleep;
1760
1761                 /*
1762                  * Avoid the problem that we change the status of the fs
1763                  * during the above check and trylock.
1764                  */
1765                 if (btrfs_need_cleaner_sleep(root)) {
1766                         mutex_unlock(&root->fs_info->cleaner_mutex);
1767                         goto sleep;
1768                 }
1769
1770                 btrfs_run_delayed_iputs(root);
1771                 again = btrfs_clean_one_deleted_snapshot(root);
1772                 mutex_unlock(&root->fs_info->cleaner_mutex);
1773
1774                 /*
1775                  * The defragger has dealt with the R/O remount and umount,
1776                  * needn't do anything special here.
1777                  */
1778                 btrfs_run_defrag_inodes(root->fs_info);
1779 sleep:
1780                 if (!try_to_freeze() && !again) {
1781                         set_current_state(TASK_INTERRUPTIBLE);
1782                         if (!kthread_should_stop())
1783                                 schedule();
1784                         __set_current_state(TASK_RUNNING);
1785                 }
1786         } while (!kthread_should_stop());
1787         return 0;
1788 }
1789
1790 static int transaction_kthread(void *arg)
1791 {
1792         struct btrfs_root *root = arg;
1793         struct btrfs_trans_handle *trans;
1794         struct btrfs_transaction *cur;
1795         u64 transid;
1796         unsigned long now;
1797         unsigned long delay;
1798         bool cannot_commit;
1799
1800         do {
1801                 cannot_commit = false;
1802                 delay = HZ * root->fs_info->commit_interval;
1803                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1804
1805                 spin_lock(&root->fs_info->trans_lock);
1806                 cur = root->fs_info->running_transaction;
1807                 if (!cur) {
1808                         spin_unlock(&root->fs_info->trans_lock);
1809                         goto sleep;
1810                 }
1811
1812                 now = get_seconds();
1813                 if (cur->state < TRANS_STATE_BLOCKED &&
1814                     (now < cur->start_time ||
1815                      now - cur->start_time < root->fs_info->commit_interval)) {
1816                         spin_unlock(&root->fs_info->trans_lock);
1817                         delay = HZ * 5;
1818                         goto sleep;
1819                 }
1820                 transid = cur->transid;
1821                 spin_unlock(&root->fs_info->trans_lock);
1822
1823                 /* If the file system is aborted, this will always fail. */
1824                 trans = btrfs_attach_transaction(root);
1825                 if (IS_ERR(trans)) {
1826                         if (PTR_ERR(trans) != -ENOENT)
1827                                 cannot_commit = true;
1828                         goto sleep;
1829                 }
1830                 if (transid == trans->transid) {
1831                         btrfs_commit_transaction(trans, root);
1832                 } else {
1833                         btrfs_end_transaction(trans, root);
1834                 }
1835 sleep:
1836                 wake_up_process(root->fs_info->cleaner_kthread);
1837                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1838
1839                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1840                                       &root->fs_info->fs_state)))
1841                         btrfs_cleanup_transaction(root);
1842                 if (!try_to_freeze()) {
1843                         set_current_state(TASK_INTERRUPTIBLE);
1844                         if (!kthread_should_stop() &&
1845                             (!btrfs_transaction_blocked(root->fs_info) ||
1846                              cannot_commit))
1847                                 schedule_timeout(delay);
1848                         __set_current_state(TASK_RUNNING);
1849                 }
1850         } while (!kthread_should_stop());
1851         return 0;
1852 }
1853
1854 /*
1855  * this will find the highest generation in the array of
1856  * root backups.  The index of the highest array is returned,
1857  * or -1 if we can't find anything.
1858  *
1859  * We check to make sure the array is valid by comparing the
1860  * generation of the latest  root in the array with the generation
1861  * in the super block.  If they don't match we pitch it.
1862  */
1863 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1864 {
1865         u64 cur;
1866         int newest_index = -1;
1867         struct btrfs_root_backup *root_backup;
1868         int i;
1869
1870         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1871                 root_backup = info->super_copy->super_roots + i;
1872                 cur = btrfs_backup_tree_root_gen(root_backup);
1873                 if (cur == newest_gen)
1874                         newest_index = i;
1875         }
1876
1877         /* check to see if we actually wrapped around */
1878         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1879                 root_backup = info->super_copy->super_roots;
1880                 cur = btrfs_backup_tree_root_gen(root_backup);
1881                 if (cur == newest_gen)
1882                         newest_index = 0;
1883         }
1884         return newest_index;
1885 }
1886
1887
1888 /*
1889  * find the oldest backup so we know where to store new entries
1890  * in the backup array.  This will set the backup_root_index
1891  * field in the fs_info struct
1892  */
1893 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1894                                      u64 newest_gen)
1895 {
1896         int newest_index = -1;
1897
1898         newest_index = find_newest_super_backup(info, newest_gen);
1899         /* if there was garbage in there, just move along */
1900         if (newest_index == -1) {
1901                 info->backup_root_index = 0;
1902         } else {
1903                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1904         }
1905 }
1906
1907 /*
1908  * copy all the root pointers into the super backup array.
1909  * this will bump the backup pointer by one when it is
1910  * done
1911  */
1912 static void backup_super_roots(struct btrfs_fs_info *info)
1913 {
1914         int next_backup;
1915         struct btrfs_root_backup *root_backup;
1916         int last_backup;
1917
1918         next_backup = info->backup_root_index;
1919         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1920                 BTRFS_NUM_BACKUP_ROOTS;
1921
1922         /*
1923          * just overwrite the last backup if we're at the same generation
1924          * this happens only at umount
1925          */
1926         root_backup = info->super_for_commit->super_roots + last_backup;
1927         if (btrfs_backup_tree_root_gen(root_backup) ==
1928             btrfs_header_generation(info->tree_root->node))
1929                 next_backup = last_backup;
1930
1931         root_backup = info->super_for_commit->super_roots + next_backup;
1932
1933         /*
1934          * make sure all of our padding and empty slots get zero filled
1935          * regardless of which ones we use today
1936          */
1937         memset(root_backup, 0, sizeof(*root_backup));
1938
1939         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1940
1941         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1942         btrfs_set_backup_tree_root_gen(root_backup,
1943                                btrfs_header_generation(info->tree_root->node));
1944
1945         btrfs_set_backup_tree_root_level(root_backup,
1946                                btrfs_header_level(info->tree_root->node));
1947
1948         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1949         btrfs_set_backup_chunk_root_gen(root_backup,
1950                                btrfs_header_generation(info->chunk_root->node));
1951         btrfs_set_backup_chunk_root_level(root_backup,
1952                                btrfs_header_level(info->chunk_root->node));
1953
1954         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1955         btrfs_set_backup_extent_root_gen(root_backup,
1956                                btrfs_header_generation(info->extent_root->node));
1957         btrfs_set_backup_extent_root_level(root_backup,
1958                                btrfs_header_level(info->extent_root->node));
1959
1960         /*
1961          * we might commit during log recovery, which happens before we set
1962          * the fs_root.  Make sure it is valid before we fill it in.
1963          */
1964         if (info->fs_root && info->fs_root->node) {
1965                 btrfs_set_backup_fs_root(root_backup,
1966                                          info->fs_root->node->start);
1967                 btrfs_set_backup_fs_root_gen(root_backup,
1968                                btrfs_header_generation(info->fs_root->node));
1969                 btrfs_set_backup_fs_root_level(root_backup,
1970                                btrfs_header_level(info->fs_root->node));
1971         }
1972
1973         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1974         btrfs_set_backup_dev_root_gen(root_backup,
1975                                btrfs_header_generation(info->dev_root->node));
1976         btrfs_set_backup_dev_root_level(root_backup,
1977                                        btrfs_header_level(info->dev_root->node));
1978
1979         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1980         btrfs_set_backup_csum_root_gen(root_backup,
1981                                btrfs_header_generation(info->csum_root->node));
1982         btrfs_set_backup_csum_root_level(root_backup,
1983                                btrfs_header_level(info->csum_root->node));
1984
1985         btrfs_set_backup_total_bytes(root_backup,
1986                              btrfs_super_total_bytes(info->super_copy));
1987         btrfs_set_backup_bytes_used(root_backup,
1988                              btrfs_super_bytes_used(info->super_copy));
1989         btrfs_set_backup_num_devices(root_backup,
1990                              btrfs_super_num_devices(info->super_copy));
1991
1992         /*
1993          * if we don't copy this out to the super_copy, it won't get remembered
1994          * for the next commit
1995          */
1996         memcpy(&info->super_copy->super_roots,
1997                &info->super_for_commit->super_roots,
1998                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1999 }
2000
2001 /*
2002  * this copies info out of the root backup array and back into
2003  * the in-memory super block.  It is meant to help iterate through
2004  * the array, so you send it the number of backups you've already
2005  * tried and the last backup index you used.
2006  *
2007  * this returns -1 when it has tried all the backups
2008  */
2009 static noinline int next_root_backup(struct btrfs_fs_info *info,
2010                                      struct btrfs_super_block *super,
2011                                      int *num_backups_tried, int *backup_index)
2012 {
2013         struct btrfs_root_backup *root_backup;
2014         int newest = *backup_index;
2015
2016         if (*num_backups_tried == 0) {
2017                 u64 gen = btrfs_super_generation(super);
2018
2019                 newest = find_newest_super_backup(info, gen);
2020                 if (newest == -1)
2021                         return -1;
2022
2023                 *backup_index = newest;
2024                 *num_backups_tried = 1;
2025         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2026                 /* we've tried all the backups, all done */
2027                 return -1;
2028         } else {
2029                 /* jump to the next oldest backup */
2030                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2031                         BTRFS_NUM_BACKUP_ROOTS;
2032                 *backup_index = newest;
2033                 *num_backups_tried += 1;
2034         }
2035         root_backup = super->super_roots + newest;
2036
2037         btrfs_set_super_generation(super,
2038                                    btrfs_backup_tree_root_gen(root_backup));
2039         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2040         btrfs_set_super_root_level(super,
2041                                    btrfs_backup_tree_root_level(root_backup));
2042         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2043
2044         /*
2045          * fixme: the total bytes and num_devices need to match or we should
2046          * need a fsck
2047          */
2048         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2049         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2050         return 0;
2051 }
2052
2053 /* helper to cleanup workers */
2054 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2055 {
2056         btrfs_destroy_workqueue(fs_info->fixup_workers);
2057         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2058         btrfs_destroy_workqueue(fs_info->workers);
2059         btrfs_destroy_workqueue(fs_info->endio_workers);
2060         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2061         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2062         btrfs_destroy_workqueue(fs_info->rmw_workers);
2063         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2064         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2065         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2066         btrfs_destroy_workqueue(fs_info->submit_workers);
2067         btrfs_destroy_workqueue(fs_info->delayed_workers);
2068         btrfs_destroy_workqueue(fs_info->caching_workers);
2069         btrfs_destroy_workqueue(fs_info->readahead_workers);
2070         btrfs_destroy_workqueue(fs_info->flush_workers);
2071         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2072         btrfs_destroy_workqueue(fs_info->extent_workers);
2073 }
2074
2075 static void free_root_extent_buffers(struct btrfs_root *root)
2076 {
2077         if (root) {
2078                 free_extent_buffer(root->node);
2079                 free_extent_buffer(root->commit_root);
2080                 root->node = NULL;
2081                 root->commit_root = NULL;
2082         }
2083 }
2084
2085 /* helper to cleanup tree roots */
2086 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2087 {
2088         free_root_extent_buffers(info->tree_root);
2089
2090         free_root_extent_buffers(info->dev_root);
2091         free_root_extent_buffers(info->extent_root);
2092         free_root_extent_buffers(info->csum_root);
2093         free_root_extent_buffers(info->quota_root);
2094         free_root_extent_buffers(info->uuid_root);
2095         if (chunk_root)
2096                 free_root_extent_buffers(info->chunk_root);
2097 }
2098
2099 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2100 {
2101         int ret;
2102         struct btrfs_root *gang[8];
2103         int i;
2104
2105         while (!list_empty(&fs_info->dead_roots)) {
2106                 gang[0] = list_entry(fs_info->dead_roots.next,
2107                                      struct btrfs_root, root_list);
2108                 list_del(&gang[0]->root_list);
2109
2110                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2111                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2112                 } else {
2113                         free_extent_buffer(gang[0]->node);
2114                         free_extent_buffer(gang[0]->commit_root);
2115                         btrfs_put_fs_root(gang[0]);
2116                 }
2117         }
2118
2119         while (1) {
2120                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2121                                              (void **)gang, 0,
2122                                              ARRAY_SIZE(gang));
2123                 if (!ret)
2124                         break;
2125                 for (i = 0; i < ret; i++)
2126                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2127         }
2128
2129         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2130                 btrfs_free_log_root_tree(NULL, fs_info);
2131                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2132                                             fs_info->pinned_extents);
2133         }
2134 }
2135
2136 int open_ctree(struct super_block *sb,
2137                struct btrfs_fs_devices *fs_devices,
2138                char *options)
2139 {
2140         u32 sectorsize;
2141         u32 nodesize;
2142         u32 blocksize;
2143         u32 stripesize;
2144         u64 generation;
2145         u64 features;
2146         struct btrfs_key location;
2147         struct buffer_head *bh;
2148         struct btrfs_super_block *disk_super;
2149         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2150         struct btrfs_root *tree_root;
2151         struct btrfs_root *extent_root;
2152         struct btrfs_root *csum_root;
2153         struct btrfs_root *chunk_root;
2154         struct btrfs_root *dev_root;
2155         struct btrfs_root *quota_root;
2156         struct btrfs_root *uuid_root;
2157         struct btrfs_root *log_tree_root;
2158         int ret;
2159         int err = -EINVAL;
2160         int num_backups_tried = 0;
2161         int backup_index = 0;
2162         int max_active;
2163         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2164         bool create_uuid_tree;
2165         bool check_uuid_tree;
2166
2167         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2168         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2169         if (!tree_root || !chunk_root) {
2170                 err = -ENOMEM;
2171                 goto fail;
2172         }
2173
2174         ret = init_srcu_struct(&fs_info->subvol_srcu);
2175         if (ret) {
2176                 err = ret;
2177                 goto fail;
2178         }
2179
2180         ret = setup_bdi(fs_info, &fs_info->bdi);
2181         if (ret) {
2182                 err = ret;
2183                 goto fail_srcu;
2184         }
2185
2186         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2187         if (ret) {
2188                 err = ret;
2189                 goto fail_bdi;
2190         }
2191         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2192                                         (1 + ilog2(nr_cpu_ids));
2193
2194         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2195         if (ret) {
2196                 err = ret;
2197                 goto fail_dirty_metadata_bytes;
2198         }
2199
2200         ret = percpu_counter_init(&fs_info->bio_counter, 0);
2201         if (ret) {
2202                 err = ret;
2203                 goto fail_delalloc_bytes;
2204         }
2205
2206         fs_info->btree_inode = new_inode(sb);
2207         if (!fs_info->btree_inode) {
2208                 err = -ENOMEM;
2209                 goto fail_bio_counter;
2210         }
2211
2212         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2213
2214         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2215         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2216         INIT_LIST_HEAD(&fs_info->trans_list);
2217         INIT_LIST_HEAD(&fs_info->dead_roots);
2218         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2219         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2220         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2221         spin_lock_init(&fs_info->delalloc_root_lock);
2222         spin_lock_init(&fs_info->trans_lock);
2223         spin_lock_init(&fs_info->fs_roots_radix_lock);
2224         spin_lock_init(&fs_info->delayed_iput_lock);
2225         spin_lock_init(&fs_info->defrag_inodes_lock);
2226         spin_lock_init(&fs_info->free_chunk_lock);
2227         spin_lock_init(&fs_info->tree_mod_seq_lock);
2228         spin_lock_init(&fs_info->super_lock);
2229         spin_lock_init(&fs_info->qgroup_op_lock);
2230         spin_lock_init(&fs_info->buffer_lock);
2231         rwlock_init(&fs_info->tree_mod_log_lock);
2232         mutex_init(&fs_info->reloc_mutex);
2233         mutex_init(&fs_info->delalloc_root_mutex);
2234         seqlock_init(&fs_info->profiles_lock);
2235
2236         init_completion(&fs_info->kobj_unregister);
2237         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2238         INIT_LIST_HEAD(&fs_info->space_info);
2239         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2240         btrfs_mapping_init(&fs_info->mapping_tree);
2241         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2242                              BTRFS_BLOCK_RSV_GLOBAL);
2243         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2244                              BTRFS_BLOCK_RSV_DELALLOC);
2245         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2246         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2247         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2248         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2249                              BTRFS_BLOCK_RSV_DELOPS);
2250         atomic_set(&fs_info->nr_async_submits, 0);
2251         atomic_set(&fs_info->async_delalloc_pages, 0);
2252         atomic_set(&fs_info->async_submit_draining, 0);
2253         atomic_set(&fs_info->nr_async_bios, 0);
2254         atomic_set(&fs_info->defrag_running, 0);
2255         atomic_set(&fs_info->qgroup_op_seq, 0);
2256         atomic64_set(&fs_info->tree_mod_seq, 0);
2257         fs_info->sb = sb;
2258         fs_info->max_inline = 8192 * 1024;
2259         fs_info->metadata_ratio = 0;
2260         fs_info->defrag_inodes = RB_ROOT;
2261         fs_info->free_chunk_space = 0;
2262         fs_info->tree_mod_log = RB_ROOT;
2263         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2264         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2265         /* readahead state */
2266         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2267         spin_lock_init(&fs_info->reada_lock);
2268
2269         fs_info->thread_pool_size = min_t(unsigned long,
2270                                           num_online_cpus() + 2, 8);
2271
2272         INIT_LIST_HEAD(&fs_info->ordered_roots);
2273         spin_lock_init(&fs_info->ordered_root_lock);
2274         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2275                                         GFP_NOFS);
2276         if (!fs_info->delayed_root) {
2277                 err = -ENOMEM;
2278                 goto fail_iput;
2279         }
2280         btrfs_init_delayed_root(fs_info->delayed_root);
2281
2282         mutex_init(&fs_info->scrub_lock);
2283         atomic_set(&fs_info->scrubs_running, 0);
2284         atomic_set(&fs_info->scrub_pause_req, 0);
2285         atomic_set(&fs_info->scrubs_paused, 0);
2286         atomic_set(&fs_info->scrub_cancel_req, 0);
2287         init_waitqueue_head(&fs_info->replace_wait);
2288         init_waitqueue_head(&fs_info->scrub_pause_wait);
2289         fs_info->scrub_workers_refcnt = 0;
2290 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2291         fs_info->check_integrity_print_mask = 0;
2292 #endif
2293
2294         spin_lock_init(&fs_info->balance_lock);
2295         mutex_init(&fs_info->balance_mutex);
2296         atomic_set(&fs_info->balance_running, 0);
2297         atomic_set(&fs_info->balance_pause_req, 0);
2298         atomic_set(&fs_info->balance_cancel_req, 0);
2299         fs_info->balance_ctl = NULL;
2300         init_waitqueue_head(&fs_info->balance_wait_q);
2301         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2302
2303         sb->s_blocksize = 4096;
2304         sb->s_blocksize_bits = blksize_bits(4096);
2305         sb->s_bdi = &fs_info->bdi;
2306
2307         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2308         set_nlink(fs_info->btree_inode, 1);
2309         /*
2310          * we set the i_size on the btree inode to the max possible int.
2311          * the real end of the address space is determined by all of
2312          * the devices in the system
2313          */
2314         fs_info->btree_inode->i_size = OFFSET_MAX;
2315         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2316         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2317
2318         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2319         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2320                              fs_info->btree_inode->i_mapping);
2321         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2322         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2323
2324         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2325
2326         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2327         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2328                sizeof(struct btrfs_key));
2329         set_bit(BTRFS_INODE_DUMMY,
2330                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2331         btrfs_insert_inode_hash(fs_info->btree_inode);
2332
2333         spin_lock_init(&fs_info->block_group_cache_lock);
2334         fs_info->block_group_cache_tree = RB_ROOT;
2335         fs_info->first_logical_byte = (u64)-1;
2336
2337         extent_io_tree_init(&fs_info->freed_extents[0],
2338                              fs_info->btree_inode->i_mapping);
2339         extent_io_tree_init(&fs_info->freed_extents[1],
2340                              fs_info->btree_inode->i_mapping);
2341         fs_info->pinned_extents = &fs_info->freed_extents[0];
2342         fs_info->do_barriers = 1;
2343
2344
2345         mutex_init(&fs_info->ordered_operations_mutex);
2346         mutex_init(&fs_info->ordered_extent_flush_mutex);
2347         mutex_init(&fs_info->tree_log_mutex);
2348         mutex_init(&fs_info->chunk_mutex);
2349         mutex_init(&fs_info->transaction_kthread_mutex);
2350         mutex_init(&fs_info->cleaner_mutex);
2351         mutex_init(&fs_info->volume_mutex);
2352         init_rwsem(&fs_info->commit_root_sem);
2353         init_rwsem(&fs_info->cleanup_work_sem);
2354         init_rwsem(&fs_info->subvol_sem);
2355         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2356         fs_info->dev_replace.lock_owner = 0;
2357         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2358         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2359         mutex_init(&fs_info->dev_replace.lock_management_lock);
2360         mutex_init(&fs_info->dev_replace.lock);
2361
2362         spin_lock_init(&fs_info->qgroup_lock);
2363         mutex_init(&fs_info->qgroup_ioctl_lock);
2364         fs_info->qgroup_tree = RB_ROOT;
2365         fs_info->qgroup_op_tree = RB_ROOT;
2366         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2367         fs_info->qgroup_seq = 1;
2368         fs_info->quota_enabled = 0;
2369         fs_info->pending_quota_state = 0;
2370         fs_info->qgroup_ulist = NULL;
2371         mutex_init(&fs_info->qgroup_rescan_lock);
2372
2373         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2374         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2375
2376         init_waitqueue_head(&fs_info->transaction_throttle);
2377         init_waitqueue_head(&fs_info->transaction_wait);
2378         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2379         init_waitqueue_head(&fs_info->async_submit_wait);
2380
2381         ret = btrfs_alloc_stripe_hash_table(fs_info);
2382         if (ret) {
2383                 err = ret;
2384                 goto fail_alloc;
2385         }
2386
2387         __setup_root(4096, 4096, 4096, tree_root,
2388                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2389
2390         invalidate_bdev(fs_devices->latest_bdev);
2391
2392         /*
2393          * Read super block and check the signature bytes only
2394          */
2395         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2396         if (!bh) {
2397                 err = -EINVAL;
2398                 goto fail_alloc;
2399         }
2400
2401         /*
2402          * We want to check superblock checksum, the type is stored inside.
2403          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2404          */
2405         if (btrfs_check_super_csum(bh->b_data)) {
2406                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2407                 err = -EINVAL;
2408                 goto fail_alloc;
2409         }
2410
2411         /*
2412          * super_copy is zeroed at allocation time and we never touch the
2413          * following bytes up to INFO_SIZE, the checksum is calculated from
2414          * the whole block of INFO_SIZE
2415          */
2416         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2417         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2418                sizeof(*fs_info->super_for_commit));
2419         brelse(bh);
2420
2421         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2422
2423         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2424         if (ret) {
2425                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2426                 err = -EINVAL;
2427                 goto fail_alloc;
2428         }
2429
2430         disk_super = fs_info->super_copy;
2431         if (!btrfs_super_root(disk_super))
2432                 goto fail_alloc;
2433
2434         /* check FS state, whether FS is broken. */
2435         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2436                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2437
2438         /*
2439          * run through our array of backup supers and setup
2440          * our ring pointer to the oldest one
2441          */
2442         generation = btrfs_super_generation(disk_super);
2443         find_oldest_super_backup(fs_info, generation);
2444
2445         /*
2446          * In the long term, we'll store the compression type in the super
2447          * block, and it'll be used for per file compression control.
2448          */
2449         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2450
2451         ret = btrfs_parse_options(tree_root, options);
2452         if (ret) {
2453                 err = ret;
2454                 goto fail_alloc;
2455         }
2456
2457         features = btrfs_super_incompat_flags(disk_super) &
2458                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2459         if (features) {
2460                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2461                        "unsupported optional features (%Lx).\n",
2462                        features);
2463                 err = -EINVAL;
2464                 goto fail_alloc;
2465         }
2466
2467         /*
2468          * Leafsize and nodesize were always equal, this is only a sanity check.
2469          */
2470         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2471             btrfs_super_nodesize(disk_super)) {
2472                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2473                        "blocksizes don't match.  node %d leaf %d\n",
2474                        btrfs_super_nodesize(disk_super),
2475                        le32_to_cpu(disk_super->__unused_leafsize));
2476                 err = -EINVAL;
2477                 goto fail_alloc;
2478         }
2479         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2480                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2481                        "blocksize (%d) was too large\n",
2482                        btrfs_super_nodesize(disk_super));
2483                 err = -EINVAL;
2484                 goto fail_alloc;
2485         }
2486
2487         features = btrfs_super_incompat_flags(disk_super);
2488         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2489         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2490                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2491
2492         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2493                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2494
2495         /*
2496          * flag our filesystem as having big metadata blocks if
2497          * they are bigger than the page size
2498          */
2499         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2500                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2501                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2502                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2503         }
2504
2505         nodesize = btrfs_super_nodesize(disk_super);
2506         sectorsize = btrfs_super_sectorsize(disk_super);
2507         stripesize = btrfs_super_stripesize(disk_super);
2508         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2509         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2510
2511         /*
2512          * mixed block groups end up with duplicate but slightly offset
2513          * extent buffers for the same range.  It leads to corruptions
2514          */
2515         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2516             (sectorsize != nodesize)) {
2517                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2518                                 "are not allowed for mixed block groups on %s\n",
2519                                 sb->s_id);
2520                 goto fail_alloc;
2521         }
2522
2523         /*
2524          * Needn't use the lock because there is no other task which will
2525          * update the flag.
2526          */
2527         btrfs_set_super_incompat_flags(disk_super, features);
2528
2529         features = btrfs_super_compat_ro_flags(disk_super) &
2530                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2531         if (!(sb->s_flags & MS_RDONLY) && features) {
2532                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2533                        "unsupported option features (%Lx).\n",
2534                        features);
2535                 err = -EINVAL;
2536                 goto fail_alloc;
2537         }
2538
2539         max_active = fs_info->thread_pool_size;
2540
2541         fs_info->workers =
2542                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2543                                       max_active, 16);
2544
2545         fs_info->delalloc_workers =
2546                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2547
2548         fs_info->flush_workers =
2549                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2550
2551         fs_info->caching_workers =
2552                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2553
2554         /*
2555          * a higher idle thresh on the submit workers makes it much more
2556          * likely that bios will be send down in a sane order to the
2557          * devices
2558          */
2559         fs_info->submit_workers =
2560                 btrfs_alloc_workqueue("submit", flags,
2561                                       min_t(u64, fs_devices->num_devices,
2562                                             max_active), 64);
2563
2564         fs_info->fixup_workers =
2565                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2566
2567         /*
2568          * endios are largely parallel and should have a very
2569          * low idle thresh
2570          */
2571         fs_info->endio_workers =
2572                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2573         fs_info->endio_meta_workers =
2574                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2575         fs_info->endio_meta_write_workers =
2576                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2577         fs_info->endio_raid56_workers =
2578                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2579         fs_info->rmw_workers =
2580                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2581         fs_info->endio_write_workers =
2582                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2583         fs_info->endio_freespace_worker =
2584                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2585         fs_info->delayed_workers =
2586                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2587         fs_info->readahead_workers =
2588                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2589         fs_info->qgroup_rescan_workers =
2590                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2591         fs_info->extent_workers =
2592                 btrfs_alloc_workqueue("extent-refs", flags,
2593                                       min_t(u64, fs_devices->num_devices,
2594                                             max_active), 8);
2595
2596         if (!(fs_info->workers && fs_info->delalloc_workers &&
2597               fs_info->submit_workers && fs_info->flush_workers &&
2598               fs_info->endio_workers && fs_info->endio_meta_workers &&
2599               fs_info->endio_meta_write_workers &&
2600               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2601               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2602               fs_info->caching_workers && fs_info->readahead_workers &&
2603               fs_info->fixup_workers && fs_info->delayed_workers &&
2604               fs_info->fixup_workers && fs_info->extent_workers &&
2605               fs_info->qgroup_rescan_workers)) {
2606                 err = -ENOMEM;
2607                 goto fail_sb_buffer;
2608         }
2609
2610         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2611         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2612                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2613
2614         tree_root->nodesize = nodesize;
2615         tree_root->sectorsize = sectorsize;
2616         tree_root->stripesize = stripesize;
2617
2618         sb->s_blocksize = sectorsize;
2619         sb->s_blocksize_bits = blksize_bits(sectorsize);
2620
2621         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2622                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2623                 goto fail_sb_buffer;
2624         }
2625
2626         if (sectorsize != PAGE_SIZE) {
2627                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2628                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2629                 goto fail_sb_buffer;
2630         }
2631
2632         mutex_lock(&fs_info->chunk_mutex);
2633         ret = btrfs_read_sys_array(tree_root);
2634         mutex_unlock(&fs_info->chunk_mutex);
2635         if (ret) {
2636                 printk(KERN_WARNING "BTRFS: failed to read the system "
2637                        "array on %s\n", sb->s_id);
2638                 goto fail_sb_buffer;
2639         }
2640
2641         blocksize = tree_root->nodesize;
2642         generation = btrfs_super_chunk_root_generation(disk_super);
2643
2644         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2645                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2646
2647         chunk_root->node = read_tree_block(chunk_root,
2648                                            btrfs_super_chunk_root(disk_super),
2649                                            blocksize, generation);
2650         if (!chunk_root->node ||
2651             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2652                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2653                        sb->s_id);
2654                 goto fail_tree_roots;
2655         }
2656         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2657         chunk_root->commit_root = btrfs_root_node(chunk_root);
2658
2659         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2660            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2661
2662         ret = btrfs_read_chunk_tree(chunk_root);
2663         if (ret) {
2664                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2665                        sb->s_id);
2666                 goto fail_tree_roots;
2667         }
2668
2669         /*
2670          * keep the device that is marked to be the target device for the
2671          * dev_replace procedure
2672          */
2673         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2674
2675         if (!fs_devices->latest_bdev) {
2676                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2677                        sb->s_id);
2678                 goto fail_tree_roots;
2679         }
2680
2681 retry_root_backup:
2682         blocksize = tree_root->nodesize;
2683         generation = btrfs_super_generation(disk_super);
2684
2685         tree_root->node = read_tree_block(tree_root,
2686                                           btrfs_super_root(disk_super),
2687                                           blocksize, generation);
2688         if (!tree_root->node ||
2689             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2690                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2691                        sb->s_id);
2692
2693                 goto recovery_tree_root;
2694         }
2695
2696         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2697         tree_root->commit_root = btrfs_root_node(tree_root);
2698         btrfs_set_root_refs(&tree_root->root_item, 1);
2699
2700         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2701         location.type = BTRFS_ROOT_ITEM_KEY;
2702         location.offset = 0;
2703
2704         extent_root = btrfs_read_tree_root(tree_root, &location);
2705         if (IS_ERR(extent_root)) {
2706                 ret = PTR_ERR(extent_root);
2707                 goto recovery_tree_root;
2708         }
2709         set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2710         fs_info->extent_root = extent_root;
2711
2712         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2713         dev_root = btrfs_read_tree_root(tree_root, &location);
2714         if (IS_ERR(dev_root)) {
2715                 ret = PTR_ERR(dev_root);
2716                 goto recovery_tree_root;
2717         }
2718         set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2719         fs_info->dev_root = dev_root;
2720         btrfs_init_devices_late(fs_info);
2721
2722         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2723         csum_root = btrfs_read_tree_root(tree_root, &location);
2724         if (IS_ERR(csum_root)) {
2725                 ret = PTR_ERR(csum_root);
2726                 goto recovery_tree_root;
2727         }
2728         set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2729         fs_info->csum_root = csum_root;
2730
2731         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2732         quota_root = btrfs_read_tree_root(tree_root, &location);
2733         if (!IS_ERR(quota_root)) {
2734                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2735                 fs_info->quota_enabled = 1;
2736                 fs_info->pending_quota_state = 1;
2737                 fs_info->quota_root = quota_root;
2738         }
2739
2740         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2741         uuid_root = btrfs_read_tree_root(tree_root, &location);
2742         if (IS_ERR(uuid_root)) {
2743                 ret = PTR_ERR(uuid_root);
2744                 if (ret != -ENOENT)
2745                         goto recovery_tree_root;
2746                 create_uuid_tree = true;
2747                 check_uuid_tree = false;
2748         } else {
2749                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2750                 fs_info->uuid_root = uuid_root;
2751                 create_uuid_tree = false;
2752                 check_uuid_tree =
2753                     generation != btrfs_super_uuid_tree_generation(disk_super);
2754         }
2755
2756         fs_info->generation = generation;
2757         fs_info->last_trans_committed = generation;
2758
2759         ret = btrfs_recover_balance(fs_info);
2760         if (ret) {
2761                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2762                 goto fail_block_groups;
2763         }
2764
2765         ret = btrfs_init_dev_stats(fs_info);
2766         if (ret) {
2767                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2768                        ret);
2769                 goto fail_block_groups;
2770         }
2771
2772         ret = btrfs_init_dev_replace(fs_info);
2773         if (ret) {
2774                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2775                 goto fail_block_groups;
2776         }
2777
2778         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2779
2780         ret = btrfs_sysfs_add_one(fs_info);
2781         if (ret) {
2782                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2783                 goto fail_block_groups;
2784         }
2785
2786         ret = btrfs_init_space_info(fs_info);
2787         if (ret) {
2788                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2789                 goto fail_sysfs;
2790         }
2791
2792         ret = btrfs_read_block_groups(extent_root);
2793         if (ret) {
2794                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2795                 goto fail_sysfs;
2796         }
2797         fs_info->num_tolerated_disk_barrier_failures =
2798                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2799         if (fs_info->fs_devices->missing_devices >
2800              fs_info->num_tolerated_disk_barrier_failures &&
2801             !(sb->s_flags & MS_RDONLY)) {
2802                 printk(KERN_WARNING "BTRFS: "
2803                         "too many missing devices, writeable mount is not allowed\n");
2804                 goto fail_sysfs;
2805         }
2806
2807         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2808                                                "btrfs-cleaner");
2809         if (IS_ERR(fs_info->cleaner_kthread))
2810                 goto fail_sysfs;
2811
2812         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2813                                                    tree_root,
2814                                                    "btrfs-transaction");
2815         if (IS_ERR(fs_info->transaction_kthread))
2816                 goto fail_cleaner;
2817
2818         if (!btrfs_test_opt(tree_root, SSD) &&
2819             !btrfs_test_opt(tree_root, NOSSD) &&
2820             !fs_info->fs_devices->rotating) {
2821                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2822                        "mode\n");
2823                 btrfs_set_opt(fs_info->mount_opt, SSD);
2824         }
2825
2826         /* Set the real inode map cache flag */
2827         if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2828                 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2829
2830 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2831         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2832                 ret = btrfsic_mount(tree_root, fs_devices,
2833                                     btrfs_test_opt(tree_root,
2834                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2835                                     1 : 0,
2836                                     fs_info->check_integrity_print_mask);
2837                 if (ret)
2838                         printk(KERN_WARNING "BTRFS: failed to initialize"
2839                                " integrity check module %s\n", sb->s_id);
2840         }
2841 #endif
2842         ret = btrfs_read_qgroup_config(fs_info);
2843         if (ret)
2844                 goto fail_trans_kthread;
2845
2846         /* do not make disk changes in broken FS */
2847         if (btrfs_super_log_root(disk_super) != 0) {
2848                 u64 bytenr = btrfs_super_log_root(disk_super);
2849
2850                 if (fs_devices->rw_devices == 0) {
2851                         printk(KERN_WARNING "BTRFS: log replay required "
2852                                "on RO media\n");
2853                         err = -EIO;
2854                         goto fail_qgroup;
2855                 }
2856                 blocksize = tree_root->nodesize;
2857
2858                 log_tree_root = btrfs_alloc_root(fs_info);
2859                 if (!log_tree_root) {
2860                         err = -ENOMEM;
2861                         goto fail_qgroup;
2862                 }
2863
2864                 __setup_root(nodesize, sectorsize, stripesize,
2865                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2866
2867                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2868                                                       blocksize,
2869                                                       generation + 1);
2870                 if (!log_tree_root->node ||
2871                     !extent_buffer_uptodate(log_tree_root->node)) {
2872                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2873                         free_extent_buffer(log_tree_root->node);
2874                         kfree(log_tree_root);
2875                         goto fail_qgroup;
2876                 }
2877                 /* returns with log_tree_root freed on success */
2878                 ret = btrfs_recover_log_trees(log_tree_root);
2879                 if (ret) {
2880                         btrfs_error(tree_root->fs_info, ret,
2881                                     "Failed to recover log tree");
2882                         free_extent_buffer(log_tree_root->node);
2883                         kfree(log_tree_root);
2884                         goto fail_qgroup;
2885                 }
2886
2887                 if (sb->s_flags & MS_RDONLY) {
2888                         ret = btrfs_commit_super(tree_root);
2889                         if (ret)
2890                                 goto fail_qgroup;
2891                 }
2892         }
2893
2894         ret = btrfs_find_orphan_roots(tree_root);
2895         if (ret)
2896                 goto fail_qgroup;
2897
2898         if (!(sb->s_flags & MS_RDONLY)) {
2899                 ret = btrfs_cleanup_fs_roots(fs_info);
2900                 if (ret)
2901                         goto fail_qgroup;
2902
2903                 mutex_lock(&fs_info->cleaner_mutex);
2904                 ret = btrfs_recover_relocation(tree_root);
2905                 mutex_unlock(&fs_info->cleaner_mutex);
2906                 if (ret < 0) {
2907                         printk(KERN_WARNING
2908                                "BTRFS: failed to recover relocation\n");
2909                         err = -EINVAL;
2910                         goto fail_qgroup;
2911                 }
2912         }
2913
2914         location.objectid = BTRFS_FS_TREE_OBJECTID;
2915         location.type = BTRFS_ROOT_ITEM_KEY;
2916         location.offset = 0;
2917
2918         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2919         if (IS_ERR(fs_info->fs_root)) {
2920                 err = PTR_ERR(fs_info->fs_root);
2921                 goto fail_qgroup;
2922         }
2923
2924         if (sb->s_flags & MS_RDONLY)
2925                 return 0;
2926
2927         down_read(&fs_info->cleanup_work_sem);
2928         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2929             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2930                 up_read(&fs_info->cleanup_work_sem);
2931                 close_ctree(tree_root);
2932                 return ret;
2933         }
2934         up_read(&fs_info->cleanup_work_sem);
2935
2936         ret = btrfs_resume_balance_async(fs_info);
2937         if (ret) {
2938                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2939                 close_ctree(tree_root);
2940                 return ret;
2941         }
2942
2943         ret = btrfs_resume_dev_replace_async(fs_info);
2944         if (ret) {
2945                 pr_warn("BTRFS: failed to resume dev_replace\n");
2946                 close_ctree(tree_root);
2947                 return ret;
2948         }
2949
2950         btrfs_qgroup_rescan_resume(fs_info);
2951
2952         if (create_uuid_tree) {
2953                 pr_info("BTRFS: creating UUID tree\n");
2954                 ret = btrfs_create_uuid_tree(fs_info);
2955                 if (ret) {
2956                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2957                                 ret);
2958                         close_ctree(tree_root);
2959                         return ret;
2960                 }
2961         } else if (check_uuid_tree ||
2962                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2963                 pr_info("BTRFS: checking UUID tree\n");
2964                 ret = btrfs_check_uuid_tree(fs_info);
2965                 if (ret) {
2966                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2967                                 ret);
2968                         close_ctree(tree_root);
2969                         return ret;
2970                 }
2971         } else {
2972                 fs_info->update_uuid_tree_gen = 1;
2973         }
2974
2975         return 0;
2976
2977 fail_qgroup:
2978         btrfs_free_qgroup_config(fs_info);
2979 fail_trans_kthread:
2980         kthread_stop(fs_info->transaction_kthread);
2981         btrfs_cleanup_transaction(fs_info->tree_root);
2982         btrfs_free_fs_roots(fs_info);
2983 fail_cleaner:
2984         kthread_stop(fs_info->cleaner_kthread);
2985
2986         /*
2987          * make sure we're done with the btree inode before we stop our
2988          * kthreads
2989          */
2990         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2991
2992 fail_sysfs:
2993         btrfs_sysfs_remove_one(fs_info);
2994
2995 fail_block_groups:
2996         btrfs_put_block_group_cache(fs_info);
2997         btrfs_free_block_groups(fs_info);
2998
2999 fail_tree_roots:
3000         free_root_pointers(fs_info, 1);
3001         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3002
3003 fail_sb_buffer:
3004         btrfs_stop_all_workers(fs_info);
3005 fail_alloc:
3006 fail_iput:
3007         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3008
3009         iput(fs_info->btree_inode);
3010 fail_bio_counter:
3011         percpu_counter_destroy(&fs_info->bio_counter);
3012 fail_delalloc_bytes:
3013         percpu_counter_destroy(&fs_info->delalloc_bytes);
3014 fail_dirty_metadata_bytes:
3015         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3016 fail_bdi:
3017         bdi_destroy(&fs_info->bdi);
3018 fail_srcu:
3019         cleanup_srcu_struct(&fs_info->subvol_srcu);
3020 fail:
3021         btrfs_free_stripe_hash_table(fs_info);
3022         btrfs_close_devices(fs_info->fs_devices);
3023         return err;
3024
3025 recovery_tree_root:
3026         if (!btrfs_test_opt(tree_root, RECOVERY))
3027                 goto fail_tree_roots;
3028
3029         free_root_pointers(fs_info, 0);
3030
3031         /* don't use the log in recovery mode, it won't be valid */
3032         btrfs_set_super_log_root(disk_super, 0);
3033
3034         /* we can't trust the free space cache either */
3035         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3036
3037         ret = next_root_backup(fs_info, fs_info->super_copy,
3038                                &num_backups_tried, &backup_index);
3039         if (ret == -1)
3040                 goto fail_block_groups;
3041         goto retry_root_backup;
3042 }
3043
3044 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3045 {
3046         if (uptodate) {
3047                 set_buffer_uptodate(bh);
3048         } else {
3049                 struct btrfs_device *device = (struct btrfs_device *)
3050                         bh->b_private;
3051
3052                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3053                                           "I/O error on %s\n",
3054                                           rcu_str_deref(device->name));
3055                 /* note, we dont' set_buffer_write_io_error because we have
3056                  * our own ways of dealing with the IO errors
3057                  */
3058                 clear_buffer_uptodate(bh);
3059                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3060         }
3061         unlock_buffer(bh);
3062         put_bh(bh);
3063 }
3064
3065 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3066 {
3067         struct buffer_head *bh;
3068         struct buffer_head *latest = NULL;
3069         struct btrfs_super_block *super;
3070         int i;
3071         u64 transid = 0;
3072         u64 bytenr;
3073
3074         /* we would like to check all the supers, but that would make
3075          * a btrfs mount succeed after a mkfs from a different FS.
3076          * So, we need to add a special mount option to scan for
3077          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3078          */
3079         for (i = 0; i < 1; i++) {
3080                 bytenr = btrfs_sb_offset(i);
3081                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3082                                         i_size_read(bdev->bd_inode))
3083                         break;
3084                 bh = __bread(bdev, bytenr / 4096,
3085                                         BTRFS_SUPER_INFO_SIZE);
3086                 if (!bh)
3087                         continue;
3088
3089                 super = (struct btrfs_super_block *)bh->b_data;
3090                 if (btrfs_super_bytenr(super) != bytenr ||
3091                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3092                         brelse(bh);
3093                         continue;
3094                 }
3095
3096                 if (!latest || btrfs_super_generation(super) > transid) {
3097                         brelse(latest);
3098                         latest = bh;
3099                         transid = btrfs_super_generation(super);
3100                 } else {
3101                         brelse(bh);
3102                 }
3103         }
3104         return latest;
3105 }
3106
3107 /*
3108  * this should be called twice, once with wait == 0 and
3109  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3110  * we write are pinned.
3111  *
3112  * They are released when wait == 1 is done.
3113  * max_mirrors must be the same for both runs, and it indicates how
3114  * many supers on this one device should be written.
3115  *
3116  * max_mirrors == 0 means to write them all.
3117  */
3118 static int write_dev_supers(struct btrfs_device *device,
3119                             struct btrfs_super_block *sb,
3120                             int do_barriers, int wait, int max_mirrors)
3121 {
3122         struct buffer_head *bh;
3123         int i;
3124         int ret;
3125         int errors = 0;
3126         u32 crc;
3127         u64 bytenr;
3128
3129         if (max_mirrors == 0)
3130                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3131
3132         for (i = 0; i < max_mirrors; i++) {
3133                 bytenr = btrfs_sb_offset(i);
3134                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3135                         break;
3136
3137                 if (wait) {
3138                         bh = __find_get_block(device->bdev, bytenr / 4096,
3139                                               BTRFS_SUPER_INFO_SIZE);
3140                         if (!bh) {
3141                                 errors++;
3142                                 continue;
3143                         }
3144                         wait_on_buffer(bh);
3145                         if (!buffer_uptodate(bh))
3146                                 errors++;
3147
3148                         /* drop our reference */
3149                         brelse(bh);
3150
3151                         /* drop the reference from the wait == 0 run */
3152                         brelse(bh);
3153                         continue;
3154                 } else {
3155                         btrfs_set_super_bytenr(sb, bytenr);
3156
3157                         crc = ~(u32)0;
3158                         crc = btrfs_csum_data((char *)sb +
3159                                               BTRFS_CSUM_SIZE, crc,
3160                                               BTRFS_SUPER_INFO_SIZE -
3161                                               BTRFS_CSUM_SIZE);
3162                         btrfs_csum_final(crc, sb->csum);
3163
3164                         /*
3165                          * one reference for us, and we leave it for the
3166                          * caller
3167                          */
3168                         bh = __getblk(device->bdev, bytenr / 4096,
3169                                       BTRFS_SUPER_INFO_SIZE);
3170                         if (!bh) {
3171                                 printk(KERN_ERR "BTRFS: couldn't get super "
3172                                        "buffer head for bytenr %Lu\n", bytenr);
3173                                 errors++;
3174                                 continue;
3175                         }
3176
3177                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3178
3179                         /* one reference for submit_bh */
3180                         get_bh(bh);
3181
3182                         set_buffer_uptodate(bh);
3183                         lock_buffer(bh);
3184                         bh->b_end_io = btrfs_end_buffer_write_sync;
3185                         bh->b_private = device;
3186                 }
3187
3188                 /*
3189                  * we fua the first super.  The others we allow
3190                  * to go down lazy.
3191                  */
3192                 if (i == 0)
3193                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3194                 else
3195                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3196                 if (ret)
3197                         errors++;
3198         }
3199         return errors < i ? 0 : -1;
3200 }
3201
3202 /*
3203  * endio for the write_dev_flush, this will wake anyone waiting
3204  * for the barrier when it is done
3205  */
3206 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3207 {
3208         if (err) {
3209                 if (err == -EOPNOTSUPP)
3210                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3211                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3212         }
3213         if (bio->bi_private)
3214                 complete(bio->bi_private);
3215         bio_put(bio);
3216 }
3217
3218 /*
3219  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3220  * sent down.  With wait == 1, it waits for the previous flush.
3221  *
3222  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3223  * capable
3224  */
3225 static int write_dev_flush(struct btrfs_device *device, int wait)
3226 {
3227         struct bio *bio;
3228         int ret = 0;
3229
3230         if (device->nobarriers)
3231                 return 0;
3232
3233         if (wait) {
3234                 bio = device->flush_bio;
3235                 if (!bio)
3236                         return 0;
3237
3238                 wait_for_completion(&device->flush_wait);
3239
3240                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3241                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3242                                       rcu_str_deref(device->name));
3243                         device->nobarriers = 1;
3244                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3245                         ret = -EIO;
3246                         btrfs_dev_stat_inc_and_print(device,
3247                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3248                 }
3249
3250                 /* drop the reference from the wait == 0 run */
3251                 bio_put(bio);
3252                 device->flush_bio = NULL;
3253
3254                 return ret;
3255         }
3256
3257         /*
3258          * one reference for us, and we leave it for the
3259          * caller
3260          */
3261         device->flush_bio = NULL;
3262         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3263         if (!bio)
3264                 return -ENOMEM;
3265
3266         bio->bi_end_io = btrfs_end_empty_barrier;
3267         bio->bi_bdev = device->bdev;
3268         init_completion(&device->flush_wait);
3269         bio->bi_private = &device->flush_wait;
3270         device->flush_bio = bio;
3271
3272         bio_get(bio);
3273         btrfsic_submit_bio(WRITE_FLUSH, bio);
3274
3275         return 0;
3276 }
3277
3278 /*
3279  * send an empty flush down to each device in parallel,
3280  * then wait for them
3281  */
3282 static int barrier_all_devices(struct btrfs_fs_info *info)
3283 {
3284         struct list_head *head;
3285         struct btrfs_device *dev;
3286         int errors_send = 0;
3287         int errors_wait = 0;
3288         int ret;
3289
3290         /* send down all the barriers */
3291         head = &info->fs_devices->devices;
3292         list_for_each_entry_rcu(dev, head, dev_list) {
3293                 if (dev->missing)
3294                         continue;
3295                 if (!dev->bdev) {
3296                         errors_send++;
3297                         continue;
3298                 }
3299                 if (!dev->in_fs_metadata || !dev->writeable)
3300                         continue;
3301
3302                 ret = write_dev_flush(dev, 0);
3303                 if (ret)
3304                         errors_send++;
3305         }
3306
3307         /* wait for all the barriers */
3308         list_for_each_entry_rcu(dev, head, dev_list) {
3309                 if (dev->missing)
3310                         continue;
3311                 if (!dev->bdev) {
3312                         errors_wait++;
3313                         continue;
3314                 }
3315                 if (!dev->in_fs_metadata || !dev->writeable)
3316                         continue;
3317
3318                 ret = write_dev_flush(dev, 1);
3319                 if (ret)
3320                         errors_wait++;
3321         }
3322         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3323             errors_wait > info->num_tolerated_disk_barrier_failures)
3324                 return -EIO;
3325         return 0;
3326 }
3327
3328 int btrfs_calc_num_tolerated_disk_barrier_failures(
3329         struct btrfs_fs_info *fs_info)
3330 {
3331         struct btrfs_ioctl_space_info space;
3332         struct btrfs_space_info *sinfo;
3333         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3334                        BTRFS_BLOCK_GROUP_SYSTEM,
3335                        BTRFS_BLOCK_GROUP_METADATA,
3336                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3337         int num_types = 4;
3338         int i;
3339         int c;
3340         int num_tolerated_disk_barrier_failures =
3341                 (int)fs_info->fs_devices->num_devices;
3342
3343         for (i = 0; i < num_types; i++) {
3344                 struct btrfs_space_info *tmp;
3345
3346                 sinfo = NULL;
3347                 rcu_read_lock();
3348                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3349                         if (tmp->flags == types[i]) {
3350                                 sinfo = tmp;
3351                                 break;
3352                         }
3353                 }
3354                 rcu_read_unlock();
3355
3356                 if (!sinfo)
3357                         continue;
3358
3359                 down_read(&sinfo->groups_sem);
3360                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3361                         if (!list_empty(&sinfo->block_groups[c])) {
3362                                 u64 flags;
3363
3364                                 btrfs_get_block_group_info(
3365                                         &sinfo->block_groups[c], &space);
3366                                 if (space.total_bytes == 0 ||
3367                                     space.used_bytes == 0)
3368                                         continue;
3369                                 flags = space.flags;
3370                                 /*
3371                                  * return
3372                                  * 0: if dup, single or RAID0 is configured for
3373                                  *    any of metadata, system or data, else
3374                                  * 1: if RAID5 is configured, or if RAID1 or
3375                                  *    RAID10 is configured and only two mirrors
3376                                  *    are used, else
3377                                  * 2: if RAID6 is configured, else
3378                                  * num_mirrors - 1: if RAID1 or RAID10 is
3379                                  *                  configured and more than
3380                                  *                  2 mirrors are used.
3381                                  */
3382                                 if (num_tolerated_disk_barrier_failures > 0 &&
3383                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3384                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3385                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3386                                       == 0)))
3387                                         num_tolerated_disk_barrier_failures = 0;
3388                                 else if (num_tolerated_disk_barrier_failures > 1) {
3389                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3390                                             BTRFS_BLOCK_GROUP_RAID5 |
3391                                             BTRFS_BLOCK_GROUP_RAID10)) {
3392                                                 num_tolerated_disk_barrier_failures = 1;
3393                                         } else if (flags &
3394                                                    BTRFS_BLOCK_GROUP_RAID6) {
3395                                                 num_tolerated_disk_barrier_failures = 2;
3396                                         }
3397                                 }
3398                         }
3399                 }
3400                 up_read(&sinfo->groups_sem);
3401         }
3402
3403         return num_tolerated_disk_barrier_failures;
3404 }
3405
3406 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3407 {
3408         struct list_head *head;
3409         struct btrfs_device *dev;
3410         struct btrfs_super_block *sb;
3411         struct btrfs_dev_item *dev_item;
3412         int ret;
3413         int do_barriers;
3414         int max_errors;
3415         int total_errors = 0;
3416         u64 flags;
3417
3418         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3419         backup_super_roots(root->fs_info);
3420
3421         sb = root->fs_info->super_for_commit;
3422         dev_item = &sb->dev_item;
3423
3424         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3425         head = &root->fs_info->fs_devices->devices;
3426         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3427
3428         if (do_barriers) {
3429                 ret = barrier_all_devices(root->fs_info);
3430                 if (ret) {
3431                         mutex_unlock(
3432                                 &root->fs_info->fs_devices->device_list_mutex);
3433                         btrfs_error(root->fs_info, ret,
3434                                     "errors while submitting device barriers.");
3435                         return ret;
3436                 }
3437         }
3438
3439         list_for_each_entry_rcu(dev, head, dev_list) {
3440                 if (!dev->bdev) {
3441                         total_errors++;
3442                         continue;
3443                 }
3444                 if (!dev->in_fs_metadata || !dev->writeable)
3445                         continue;
3446
3447                 btrfs_set_stack_device_generation(dev_item, 0);
3448                 btrfs_set_stack_device_type(dev_item, dev->type);
3449                 btrfs_set_stack_device_id(dev_item, dev->devid);
3450                 btrfs_set_stack_device_total_bytes(dev_item,
3451                                                    dev->disk_total_bytes);
3452                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3453                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3454                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3455                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3456                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3457                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3458
3459                 flags = btrfs_super_flags(sb);
3460                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3461
3462                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3463                 if (ret)
3464                         total_errors++;
3465         }
3466         if (total_errors > max_errors) {
3467                 btrfs_err(root->fs_info, "%d errors while writing supers",
3468                        total_errors);
3469                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3470
3471                 /* FUA is masked off if unsupported and can't be the reason */
3472                 btrfs_error(root->fs_info, -EIO,
3473                             "%d errors while writing supers", total_errors);
3474                 return -EIO;
3475         }
3476
3477         total_errors = 0;
3478         list_for_each_entry_rcu(dev, head, dev_list) {
3479                 if (!dev->bdev)
3480                         continue;
3481                 if (!dev->in_fs_metadata || !dev->writeable)
3482                         continue;
3483
3484                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3485                 if (ret)
3486                         total_errors++;
3487         }
3488         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3489         if (total_errors > max_errors) {
3490                 btrfs_error(root->fs_info, -EIO,
3491                             "%d errors while writing supers", total_errors);
3492                 return -EIO;
3493         }
3494         return 0;
3495 }
3496
3497 int write_ctree_super(struct btrfs_trans_handle *trans,
3498                       struct btrfs_root *root, int max_mirrors)
3499 {
3500         return write_all_supers(root, max_mirrors);
3501 }
3502
3503 /* Drop a fs root from the radix tree and free it. */
3504 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3505                                   struct btrfs_root *root)
3506 {
3507         spin_lock(&fs_info->fs_roots_radix_lock);
3508         radix_tree_delete(&fs_info->fs_roots_radix,
3509                           (unsigned long)root->root_key.objectid);
3510         spin_unlock(&fs_info->fs_roots_radix_lock);
3511
3512         if (btrfs_root_refs(&root->root_item) == 0)
3513                 synchronize_srcu(&fs_info->subvol_srcu);
3514
3515         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3516                 btrfs_free_log(NULL, root);
3517
3518         if (root->free_ino_pinned)
3519                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3520         if (root->free_ino_ctl)
3521                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3522         free_fs_root(root);
3523 }
3524
3525 static void free_fs_root(struct btrfs_root *root)
3526 {
3527         iput(root->ino_cache_inode);
3528         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3529         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3530         root->orphan_block_rsv = NULL;
3531         if (root->anon_dev)
3532                 free_anon_bdev(root->anon_dev);
3533         if (root->subv_writers)
3534                 btrfs_free_subvolume_writers(root->subv_writers);
3535         free_extent_buffer(root->node);
3536         free_extent_buffer(root->commit_root);
3537         kfree(root->free_ino_ctl);
3538         kfree(root->free_ino_pinned);
3539         kfree(root->name);
3540         btrfs_put_fs_root(root);
3541 }
3542
3543 void btrfs_free_fs_root(struct btrfs_root *root)
3544 {
3545         free_fs_root(root);
3546 }
3547
3548 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3549 {
3550         u64 root_objectid = 0;
3551         struct btrfs_root *gang[8];
3552         int i = 0;
3553         int err = 0;
3554         unsigned int ret = 0;
3555         int index;
3556
3557         while (1) {
3558                 index = srcu_read_lock(&fs_info->subvol_srcu);
3559                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3560                                              (void **)gang, root_objectid,
3561                                              ARRAY_SIZE(gang));
3562                 if (!ret) {
3563                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3564                         break;
3565                 }
3566                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3567
3568                 for (i = 0; i < ret; i++) {
3569                         /* Avoid to grab roots in dead_roots */
3570                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3571                                 gang[i] = NULL;
3572                                 continue;
3573                         }
3574                         /* grab all the search result for later use */
3575                         gang[i] = btrfs_grab_fs_root(gang[i]);
3576                 }
3577                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3578
3579                 for (i = 0; i < ret; i++) {
3580                         if (!gang[i])
3581                                 continue;
3582                         root_objectid = gang[i]->root_key.objectid;
3583                         err = btrfs_orphan_cleanup(gang[i]);
3584                         if (err)
3585                                 break;
3586                         btrfs_put_fs_root(gang[i]);
3587                 }
3588                 root_objectid++;
3589         }
3590
3591         /* release the uncleaned roots due to error */
3592         for (; i < ret; i++) {
3593                 if (gang[i])
3594                         btrfs_put_fs_root(gang[i]);
3595         }
3596         return err;
3597 }
3598
3599 int btrfs_commit_super(struct btrfs_root *root)
3600 {
3601         struct btrfs_trans_handle *trans;
3602
3603         mutex_lock(&root->fs_info->cleaner_mutex);
3604         btrfs_run_delayed_iputs(root);
3605         mutex_unlock(&root->fs_info->cleaner_mutex);
3606         wake_up_process(root->fs_info->cleaner_kthread);
3607
3608         /* wait until ongoing cleanup work done */
3609         down_write(&root->fs_info->cleanup_work_sem);
3610         up_write(&root->fs_info->cleanup_work_sem);
3611
3612         trans = btrfs_join_transaction(root);
3613         if (IS_ERR(trans))
3614                 return PTR_ERR(trans);
3615         return btrfs_commit_transaction(trans, root);
3616 }
3617
3618 void close_ctree(struct btrfs_root *root)
3619 {
3620         struct btrfs_fs_info *fs_info = root->fs_info;
3621         int ret;
3622
3623         fs_info->closing = 1;
3624         smp_mb();
3625
3626         /* wait for the uuid_scan task to finish */
3627         down(&fs_info->uuid_tree_rescan_sem);
3628         /* avoid complains from lockdep et al., set sem back to initial state */
3629         up(&fs_info->uuid_tree_rescan_sem);
3630
3631         /* pause restriper - we want to resume on mount */
3632         btrfs_pause_balance(fs_info);
3633
3634         btrfs_dev_replace_suspend_for_unmount(fs_info);
3635
3636         btrfs_scrub_cancel(fs_info);
3637
3638         /* wait for any defraggers to finish */
3639         wait_event(fs_info->transaction_wait,
3640                    (atomic_read(&fs_info->defrag_running) == 0));
3641
3642         /* clear out the rbtree of defraggable inodes */
3643         btrfs_cleanup_defrag_inodes(fs_info);
3644
3645         cancel_work_sync(&fs_info->async_reclaim_work);
3646
3647         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3648                 ret = btrfs_commit_super(root);
3649                 if (ret)
3650                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3651         }
3652
3653         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3654                 btrfs_error_commit_super(root);
3655
3656         kthread_stop(fs_info->transaction_kthread);
3657         kthread_stop(fs_info->cleaner_kthread);
3658
3659         fs_info->closing = 2;
3660         smp_mb();
3661
3662         btrfs_free_qgroup_config(root->fs_info);
3663
3664         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3665                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3666                        percpu_counter_sum(&fs_info->delalloc_bytes));
3667         }
3668
3669         btrfs_sysfs_remove_one(fs_info);
3670
3671         btrfs_free_fs_roots(fs_info);
3672
3673         btrfs_put_block_group_cache(fs_info);
3674
3675         btrfs_free_block_groups(fs_info);
3676
3677         /*
3678          * we must make sure there is not any read request to
3679          * submit after we stopping all workers.
3680          */
3681         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3682         btrfs_stop_all_workers(fs_info);
3683
3684         free_root_pointers(fs_info, 1);
3685
3686         iput(fs_info->btree_inode);
3687
3688 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3689         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3690                 btrfsic_unmount(root, fs_info->fs_devices);
3691 #endif
3692
3693         btrfs_close_devices(fs_info->fs_devices);
3694         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3695
3696         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3697         percpu_counter_destroy(&fs_info->delalloc_bytes);
3698         percpu_counter_destroy(&fs_info->bio_counter);
3699         bdi_destroy(&fs_info->bdi);
3700         cleanup_srcu_struct(&fs_info->subvol_srcu);
3701
3702         btrfs_free_stripe_hash_table(fs_info);
3703
3704         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3705         root->orphan_block_rsv = NULL;
3706 }
3707
3708 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3709                           int atomic)
3710 {
3711         int ret;
3712         struct inode *btree_inode = buf->pages[0]->mapping->host;
3713
3714         ret = extent_buffer_uptodate(buf);
3715         if (!ret)
3716                 return ret;
3717
3718         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3719                                     parent_transid, atomic);
3720         if (ret == -EAGAIN)
3721                 return ret;
3722         return !ret;
3723 }
3724
3725 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3726 {
3727         return set_extent_buffer_uptodate(buf);
3728 }
3729
3730 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3731 {
3732         struct btrfs_root *root;
3733         u64 transid = btrfs_header_generation(buf);
3734         int was_dirty;
3735
3736 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3737         /*
3738          * This is a fast path so only do this check if we have sanity tests
3739          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3740          * outside of the sanity tests.
3741          */
3742         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3743                 return;
3744 #endif
3745         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3746         btrfs_assert_tree_locked(buf);
3747         if (transid != root->fs_info->generation)
3748                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3749                        "found %llu running %llu\n",
3750                         buf->start, transid, root->fs_info->generation);
3751         was_dirty = set_extent_buffer_dirty(buf);
3752         if (!was_dirty)
3753                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3754                                      buf->len,
3755                                      root->fs_info->dirty_metadata_batch);
3756 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3757         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3758                 btrfs_print_leaf(root, buf);
3759                 ASSERT(0);
3760         }
3761 #endif
3762 }
3763
3764 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3765                                         int flush_delayed)
3766 {
3767         /*
3768          * looks as though older kernels can get into trouble with
3769          * this code, they end up stuck in balance_dirty_pages forever
3770          */
3771         int ret;
3772
3773         if (current->flags & PF_MEMALLOC)
3774                 return;
3775
3776         if (flush_delayed)
3777                 btrfs_balance_delayed_items(root);
3778
3779         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3780                                      BTRFS_DIRTY_METADATA_THRESH);
3781         if (ret > 0) {
3782                 balance_dirty_pages_ratelimited(
3783                                    root->fs_info->btree_inode->i_mapping);
3784         }
3785         return;
3786 }
3787
3788 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3789 {
3790         __btrfs_btree_balance_dirty(root, 1);
3791 }
3792
3793 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3794 {
3795         __btrfs_btree_balance_dirty(root, 0);
3796 }
3797
3798 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3799 {
3800         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3801         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3802 }
3803
3804 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3805                               int read_only)
3806 {
3807         /*
3808          * Placeholder for checks
3809          */
3810         return 0;
3811 }
3812
3813 static void btrfs_error_commit_super(struct btrfs_root *root)
3814 {
3815         mutex_lock(&root->fs_info->cleaner_mutex);
3816         btrfs_run_delayed_iputs(root);
3817         mutex_unlock(&root->fs_info->cleaner_mutex);
3818
3819         down_write(&root->fs_info->cleanup_work_sem);
3820         up_write(&root->fs_info->cleanup_work_sem);
3821
3822         /* cleanup FS via transaction */
3823         btrfs_cleanup_transaction(root);
3824 }
3825
3826 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3827 {
3828         struct btrfs_ordered_extent *ordered;
3829
3830         spin_lock(&root->ordered_extent_lock);
3831         /*
3832          * This will just short circuit the ordered completion stuff which will
3833          * make sure the ordered extent gets properly cleaned up.
3834          */
3835         list_for_each_entry(ordered, &root->ordered_extents,
3836                             root_extent_list)
3837                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3838         spin_unlock(&root->ordered_extent_lock);
3839 }
3840
3841 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3842 {
3843         struct btrfs_root *root;
3844         struct list_head splice;
3845
3846         INIT_LIST_HEAD(&splice);
3847
3848         spin_lock(&fs_info->ordered_root_lock);
3849         list_splice_init(&fs_info->ordered_roots, &splice);
3850         while (!list_empty(&splice)) {
3851                 root = list_first_entry(&splice, struct btrfs_root,
3852                                         ordered_root);
3853                 list_move_tail(&root->ordered_root,
3854                                &fs_info->ordered_roots);
3855
3856                 spin_unlock(&fs_info->ordered_root_lock);
3857                 btrfs_destroy_ordered_extents(root);
3858
3859                 cond_resched();
3860                 spin_lock(&fs_info->ordered_root_lock);
3861         }
3862         spin_unlock(&fs_info->ordered_root_lock);
3863 }
3864
3865 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3866                                       struct btrfs_root *root)
3867 {
3868         struct rb_node *node;
3869         struct btrfs_delayed_ref_root *delayed_refs;
3870         struct btrfs_delayed_ref_node *ref;
3871         int ret = 0;
3872
3873         delayed_refs = &trans->delayed_refs;
3874
3875         spin_lock(&delayed_refs->lock);
3876         if (atomic_read(&delayed_refs->num_entries) == 0) {
3877                 spin_unlock(&delayed_refs->lock);
3878                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3879                 return ret;
3880         }
3881
3882         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3883                 struct btrfs_delayed_ref_head *head;
3884                 bool pin_bytes = false;
3885
3886                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3887                                 href_node);
3888                 if (!mutex_trylock(&head->mutex)) {
3889                         atomic_inc(&head->node.refs);
3890                         spin_unlock(&delayed_refs->lock);
3891
3892                         mutex_lock(&head->mutex);
3893                         mutex_unlock(&head->mutex);
3894                         btrfs_put_delayed_ref(&head->node);
3895                         spin_lock(&delayed_refs->lock);
3896                         continue;
3897                 }
3898                 spin_lock(&head->lock);
3899                 while ((node = rb_first(&head->ref_root)) != NULL) {
3900                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3901                                        rb_node);
3902                         ref->in_tree = 0;
3903                         rb_erase(&ref->rb_node, &head->ref_root);
3904                         atomic_dec(&delayed_refs->num_entries);
3905                         btrfs_put_delayed_ref(ref);
3906                 }
3907                 if (head->must_insert_reserved)
3908                         pin_bytes = true;
3909                 btrfs_free_delayed_extent_op(head->extent_op);
3910                 delayed_refs->num_heads--;
3911                 if (head->processing == 0)
3912                         delayed_refs->num_heads_ready--;
3913                 atomic_dec(&delayed_refs->num_entries);
3914                 head->node.in_tree = 0;
3915                 rb_erase(&head->href_node, &delayed_refs->href_root);
3916                 spin_unlock(&head->lock);
3917                 spin_unlock(&delayed_refs->lock);
3918                 mutex_unlock(&head->mutex);
3919
3920                 if (pin_bytes)
3921                         btrfs_pin_extent(root, head->node.bytenr,
3922                                          head->node.num_bytes, 1);
3923                 btrfs_put_delayed_ref(&head->node);
3924                 cond_resched();
3925                 spin_lock(&delayed_refs->lock);
3926         }
3927
3928         spin_unlock(&delayed_refs->lock);
3929
3930         return ret;
3931 }
3932
3933 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3934 {
3935         struct btrfs_inode *btrfs_inode;
3936         struct list_head splice;
3937
3938         INIT_LIST_HEAD(&splice);
3939
3940         spin_lock(&root->delalloc_lock);
3941         list_splice_init(&root->delalloc_inodes, &splice);
3942
3943         while (!list_empty(&splice)) {
3944                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3945                                                delalloc_inodes);
3946
3947                 list_del_init(&btrfs_inode->delalloc_inodes);
3948                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3949                           &btrfs_inode->runtime_flags);
3950                 spin_unlock(&root->delalloc_lock);
3951
3952                 btrfs_invalidate_inodes(btrfs_inode->root);
3953
3954                 spin_lock(&root->delalloc_lock);
3955         }
3956
3957         spin_unlock(&root->delalloc_lock);
3958 }
3959
3960 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3961 {
3962         struct btrfs_root *root;
3963         struct list_head splice;
3964
3965         INIT_LIST_HEAD(&splice);
3966
3967         spin_lock(&fs_info->delalloc_root_lock);
3968         list_splice_init(&fs_info->delalloc_roots, &splice);
3969         while (!list_empty(&splice)) {
3970                 root = list_first_entry(&splice, struct btrfs_root,
3971                                          delalloc_root);
3972                 list_del_init(&root->delalloc_root);
3973                 root = btrfs_grab_fs_root(root);
3974                 BUG_ON(!root);
3975                 spin_unlock(&fs_info->delalloc_root_lock);
3976
3977                 btrfs_destroy_delalloc_inodes(root);
3978                 btrfs_put_fs_root(root);
3979
3980                 spin_lock(&fs_info->delalloc_root_lock);
3981         }
3982         spin_unlock(&fs_info->delalloc_root_lock);
3983 }
3984
3985 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3986                                         struct extent_io_tree *dirty_pages,
3987                                         int mark)
3988 {
3989         int ret;
3990         struct extent_buffer *eb;
3991         u64 start = 0;
3992         u64 end;
3993
3994         while (1) {
3995                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3996                                             mark, NULL);
3997                 if (ret)
3998                         break;
3999
4000                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4001                 while (start <= end) {
4002                         eb = btrfs_find_tree_block(root, start,
4003                                                    root->nodesize);
4004                         start += root->nodesize;
4005                         if (!eb)
4006                                 continue;
4007                         wait_on_extent_buffer_writeback(eb);
4008
4009                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4010                                                &eb->bflags))
4011                                 clear_extent_buffer_dirty(eb);
4012                         free_extent_buffer_stale(eb);
4013                 }
4014         }
4015
4016         return ret;
4017 }
4018
4019 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4020                                        struct extent_io_tree *pinned_extents)
4021 {
4022         struct extent_io_tree *unpin;
4023         u64 start;
4024         u64 end;
4025         int ret;
4026         bool loop = true;
4027
4028         unpin = pinned_extents;
4029 again:
4030         while (1) {
4031                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4032                                             EXTENT_DIRTY, NULL);
4033                 if (ret)
4034                         break;
4035
4036                 /* opt_discard */
4037                 if (btrfs_test_opt(root, DISCARD))
4038                         ret = btrfs_error_discard_extent(root, start,
4039                                                          end + 1 - start,
4040                                                          NULL);
4041
4042                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4043                 btrfs_error_unpin_extent_range(root, start, end);
4044                 cond_resched();
4045         }
4046
4047         if (loop) {
4048                 if (unpin == &root->fs_info->freed_extents[0])
4049                         unpin = &root->fs_info->freed_extents[1];
4050                 else
4051                         unpin = &root->fs_info->freed_extents[0];
4052                 loop = false;
4053                 goto again;
4054         }
4055
4056         return 0;
4057 }
4058
4059 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4060                                    struct btrfs_root *root)
4061 {
4062         btrfs_destroy_delayed_refs(cur_trans, root);
4063
4064         cur_trans->state = TRANS_STATE_COMMIT_START;
4065         wake_up(&root->fs_info->transaction_blocked_wait);
4066
4067         cur_trans->state = TRANS_STATE_UNBLOCKED;
4068         wake_up(&root->fs_info->transaction_wait);
4069
4070         btrfs_destroy_delayed_inodes(root);
4071         btrfs_assert_delayed_root_empty(root);
4072
4073         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4074                                      EXTENT_DIRTY);
4075         btrfs_destroy_pinned_extent(root,
4076                                     root->fs_info->pinned_extents);
4077
4078         cur_trans->state =TRANS_STATE_COMPLETED;
4079         wake_up(&cur_trans->commit_wait);
4080
4081         /*
4082         memset(cur_trans, 0, sizeof(*cur_trans));
4083         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4084         */
4085 }
4086
4087 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4088 {
4089         struct btrfs_transaction *t;
4090
4091         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4092
4093         spin_lock(&root->fs_info->trans_lock);
4094         while (!list_empty(&root->fs_info->trans_list)) {
4095                 t = list_first_entry(&root->fs_info->trans_list,
4096                                      struct btrfs_transaction, list);
4097                 if (t->state >= TRANS_STATE_COMMIT_START) {
4098                         atomic_inc(&t->use_count);
4099                         spin_unlock(&root->fs_info->trans_lock);
4100                         btrfs_wait_for_commit(root, t->transid);
4101                         btrfs_put_transaction(t);
4102                         spin_lock(&root->fs_info->trans_lock);
4103                         continue;
4104                 }
4105                 if (t == root->fs_info->running_transaction) {
4106                         t->state = TRANS_STATE_COMMIT_DOING;
4107                         spin_unlock(&root->fs_info->trans_lock);
4108                         /*
4109                          * We wait for 0 num_writers since we don't hold a trans
4110                          * handle open currently for this transaction.
4111                          */
4112                         wait_event(t->writer_wait,
4113                                    atomic_read(&t->num_writers) == 0);
4114                 } else {
4115                         spin_unlock(&root->fs_info->trans_lock);
4116                 }
4117                 btrfs_cleanup_one_transaction(t, root);
4118
4119                 spin_lock(&root->fs_info->trans_lock);
4120                 if (t == root->fs_info->running_transaction)
4121                         root->fs_info->running_transaction = NULL;
4122                 list_del_init(&t->list);
4123                 spin_unlock(&root->fs_info->trans_lock);
4124
4125                 btrfs_put_transaction(t);
4126                 trace_btrfs_transaction_commit(root);
4127                 spin_lock(&root->fs_info->trans_lock);
4128         }
4129         spin_unlock(&root->fs_info->trans_lock);
4130         btrfs_destroy_all_ordered_extents(root->fs_info);
4131         btrfs_destroy_delayed_inodes(root);
4132         btrfs_assert_delayed_root_empty(root);
4133         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4134         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4135         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4136
4137         return 0;
4138 }
4139
4140 static struct extent_io_ops btree_extent_io_ops = {
4141         .readpage_end_io_hook = btree_readpage_end_io_hook,
4142         .readpage_io_failed_hook = btree_io_failed_hook,
4143         .submit_bio_hook = btree_submit_bio_hook,
4144         /* note we're sharing with inode.c for the merge bio hook */
4145         .merge_bio_hook = btrfs_merge_bio_hook,
4146 };