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 const 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  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  76  * is 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 btrfs_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         enum btrfs_wq_endio_type metadata;
  86         struct list_head list;
  87         struct btrfs_work work;
  88 };
  89
  90 static struct kmem_cache *btrfs_end_io_wq_cache;
  91
  92 int __init btrfs_end_io_wq_init(void)
  93 {
  94         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  95                                         sizeof(struct btrfs_end_io_wq),
  96                                         0,
  97                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  98                                         NULL);
  99         if (!btrfs_end_io_wq_cache)
 100                 return -ENOMEM;
 101         return 0;
 102 }
 103
 104 void btrfs_end_io_wq_exit(void)
 105 {
 106         if (btrfs_end_io_wq_cache)
 107                 kmem_cache_destroy(btrfs_end_io_wq_cache);
 108 }
 109
 110 /*
 111  * async submit bios are used to offload expensive checksumming
 112  * onto the worker threads.  They checksum file and metadata bios
 113  * just before they are sent down the IO stack.
 114  */
 115 struct async_submit_bio {
 116         struct inode *inode;
 117         struct bio *bio;
 118         struct list_head list;
 119         extent_submit_bio_hook_t *submit_bio_start;
 120         extent_submit_bio_hook_t *submit_bio_done;
 121         int rw;
 122         int mirror_num;
 123         unsigned long bio_flags;
 124         /*
 125          * bio_offset is optional, can be used if the pages in the bio
 126          * can't tell us where in the file the bio should go
 127          */
 128         u64 bio_offset;
 129         struct btrfs_work work;
 130         int error;
 131 };
 132
 133 /*
 134  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 135  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 136  * the level the eb occupies in the tree.
 137  *
 138  * Different roots are used for different purposes and may nest inside each
 139  * other and they require separate keysets.  As lockdep keys should be
 140  * static, assign keysets according to the purpose of the root as indicated
 141  * by btrfs_root->objectid.  This ensures that all special purpose roots
 142  * have separate keysets.
 143  *
 144  * Lock-nesting across peer nodes is always done with the immediate parent
 145  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 146  * subclass to avoid triggering lockdep warning in such cases.
 147  *
 148  * The key is set by the readpage_end_io_hook after the buffer has passed
 149  * csum validation but before the pages are unlocked.  It is also set by
 150  * btrfs_init_new_buffer on freshly allocated blocks.
 151  *
 152  * We also add a check to make sure the highest level of the tree is the
 153  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 154  * needs update as well.
 155  */
 156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 157 # if BTRFS_MAX_LEVEL != 8
 158 #  error
 159 # endif
 160
 161 static struct btrfs_lockdep_keyset {
 162         u64                     id;             /* root objectid */
 163         const char              *name_stem;     /* lock name stem */
 164         char                    names[BTRFS_MAX_LEVEL + 1][20];
 165         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 166 } btrfs_lockdep_keysets[] = {
 167         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 168         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 169         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 170         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 171         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 172         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 173         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 174         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 175         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 176         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 177         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 178         { .id = 0,                              .name_stem = "tree"     },
 179 };
 180
 181 void __init btrfs_init_lockdep(void)
 182 {
 183         int i, j;
 184
 185         /* initialize lockdep class names */
 186         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 187                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 188
 189                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 190                         snprintf(ks->names[j], sizeof(ks->names[j]),
 191                                  "btrfs-%s-%02d", ks->name_stem, j);
 192         }
 193 }
 194
 195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 196                                     int level)
 197 {
 198         struct btrfs_lockdep_keyset *ks;
 199
 200         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 201
 202         /* find the matching keyset, id 0 is the default entry */
 203         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 204                 if (ks->id == objectid)
 205                         break;
 206
 207         lockdep_set_class_and_name(&eb->lock,
 208                                    &ks->keys[level], ks->names[level]);
 209 }
 210
 211 #endif
 212
 213 /*
 214  * extents on the btree inode are pretty simple, there's one extent
 215  * that covers the entire device
 216  */
 217 static struct extent_map *btree_get_extent(struct inode *inode,
 218                 struct page *page, size_t pg_offset, u64 start, u64 len,
 219                 int create)
 220 {
 221         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 222         struct extent_map *em;
 223         int ret;
 224
 225         read_lock(&em_tree->lock);
 226         em = lookup_extent_mapping(em_tree, start, len);
 227         if (em) {
 228                 em->bdev =
 229                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 230                 read_unlock(&em_tree->lock);
 231                 goto out;
 232         }
 233         read_unlock(&em_tree->lock);
 234
 235         em = alloc_extent_map();
 236         if (!em) {
 237                 em = ERR_PTR(-ENOMEM);
 238                 goto out;
 239         }
 240         em->start = 0;
 241         em->len = (u64)-1;
 242         em->block_len = (u64)-1;
 243         em->block_start = 0;
 244         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 245
 246         write_lock(&em_tree->lock);
 247         ret = add_extent_mapping(em_tree, em, 0);
 248         if (ret == -EEXIST) {
 249                 free_extent_map(em);
 250                 em = lookup_extent_mapping(em_tree, start, len);
 251                 if (!em)
 252                         em = ERR_PTR(-EIO);
 253         } else if (ret) {
 254                 free_extent_map(em);
 255                 em = ERR_PTR(ret);
 256         }
 257         write_unlock(&em_tree->lock);
 258
 259 out:
 260         return em;
 261 }
 262
 263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 264 {
 265         return btrfs_crc32c(seed, data, len);
 266 }
 267
 268 void btrfs_csum_final(u32 crc, char *result)
 269 {
 270         put_unaligned_le32(~crc, result);
 271 }
 272
 273 /*
 274  * compute the csum for a btree block, and either verify it or write it
 275  * into the csum field of the block.
 276  */
 277 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 278                            struct extent_buffer *buf,
 279                            int verify)
 280 {
 281         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 282         char *result = NULL;
 283         unsigned long len;
 284         unsigned long cur_len;
 285         unsigned long offset = BTRFS_CSUM_SIZE;
 286         char *kaddr;
 287         unsigned long map_start;
 288         unsigned long map_len;
 289         int err;
 290         u32 crc = ~(u32)0;
 291         unsigned long inline_result;
 292
 293         len = buf->len - offset;
 294         while (len > 0) {
 295                 err = map_private_extent_buffer(buf, offset, 32,
 296                                         &kaddr, &map_start, &map_len);
 297                 if (err)
 298                         return 1;
 299                 cur_len = min(len, map_len - (offset - map_start));
 300                 crc = btrfs_csum_data(kaddr + offset - map_start,
 301                                       crc, cur_len);
 302                 len -= cur_len;
 303                 offset += cur_len;
 304         }
 305         if (csum_size > sizeof(inline_result)) {
 306                 result = kzalloc(csum_size, GFP_NOFS);
 307                 if (!result)
 308                         return 1;
 309         } else {
 310                 result = (char *)&inline_result;
 311         }
 312
 313         btrfs_csum_final(crc, result);
 314
 315         if (verify) {
 316                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 317                         u32 val;
 318                         u32 found = 0;
 319                         memcpy(&found, result, csum_size);
 320
 321                         read_extent_buffer(buf, &val, 0, csum_size);
 322                         btrfs_warn_rl(fs_info,
 323                                 "%s checksum verify failed on %llu wanted %X found %X "
 324                                 "level %d",
 325                                 fs_info->sb->s_id, buf->start,
 326                                 val, found, btrfs_header_level(buf));
 327                         if (result != (char *)&inline_result)
 328                                 kfree(result);
 329                         return 1;
 330                 }
 331         } else {
 332                 write_extent_buffer(buf, result, 0, csum_size);
 333         }
 334         if (result != (char *)&inline_result)
 335                 kfree(result);
 336         return 0;
 337 }
 338
 339 /*
 340  * we can't consider a given block up to date unless the transid of the
 341  * block matches the transid in the parent node's pointer.  This is how we
 342  * detect blocks that either didn't get written at all or got written
 343  * in the wrong place.
 344  */
 345 static int verify_parent_transid(struct extent_io_tree *io_tree,
 346                                  struct extent_buffer *eb, u64 parent_transid,
 347                                  int atomic)
 348 {
 349         struct extent_state *cached_state = NULL;
 350         int ret;
 351         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 352
 353         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 354                 return 0;
 355
 356         if (atomic)
 357                 return -EAGAIN;
 358
 359         if (need_lock) {
 360                 btrfs_tree_read_lock(eb);
 361                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 362         }
 363
 364         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 365                          0, &cached_state);
 366         if (extent_buffer_uptodate(eb) &&
 367             btrfs_header_generation(eb) == parent_transid) {
 368                 ret = 0;
 369                 goto out;
 370         }
 371         btrfs_err_rl(eb->fs_info,
 372                 "parent transid verify failed on %llu wanted %llu found %llu",
 373                         eb->start,
 374                         parent_transid, btrfs_header_generation(eb));
 375         ret = 1;
 376
 377         /*
 378          * Things reading via commit roots that don't have normal protection,
 379          * like send, can have a really old block in cache that may point at a
 380          * block that has been free'd and re-allocated.  So don't clear uptodate
 381          * if we find an eb that is under IO (dirty/writeback) because we could
 382          * end up reading in the stale data and then writing it back out and
 383          * making everybody very sad.
 384          */
 385         if (!extent_buffer_under_io(eb))
 386                 clear_extent_buffer_uptodate(eb);
 387 out:
 388         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 389                              &cached_state, GFP_NOFS);
 390         if (need_lock)
 391                 btrfs_tree_read_unlock_blocking(eb);
 392         return ret;
 393 }
 394
 395 /*
 396  * Return 0 if the superblock checksum type matches the checksum value of that
 397  * algorithm. Pass the raw disk superblock data.
 398  */
 399 static int btrfs_check_super_csum(char *raw_disk_sb)
 400 {
 401         struct btrfs_super_block *disk_sb =
 402                 (struct btrfs_super_block *)raw_disk_sb;
 403         u16 csum_type = btrfs_super_csum_type(disk_sb);
 404         int ret = 0;
 405
 406         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 407                 u32 crc = ~(u32)0;
 408                 const int csum_size = sizeof(crc);
 409                 char result[csum_size];
 410
 411                 /*
 412                  * The super_block structure does not span the whole
 413                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 414                  * is filled with zeros and is included in the checkum.
 415                  */
 416                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 417                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 418                 btrfs_csum_final(crc, result);
 419
 420                 if (memcmp(raw_disk_sb, result, csum_size))
 421                         ret = 1;
 422         }
 423
 424         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 425                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 426                                 csum_type);
 427                 ret = 1;
 428         }
 429
 430         return ret;
 431 }
 432
 433 /*
 434  * helper to read a given tree block, doing retries as required when
 435  * the checksums don't match and we have alternate mirrors to try.
 436  */
 437 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 438                                           struct extent_buffer *eb,
 439                                           u64 start, u64 parent_transid)
 440 {
 441         struct extent_io_tree *io_tree;
 442         int failed = 0;
 443         int ret;
 444         int num_copies = 0;
 445         int mirror_num = 0;
 446         int failed_mirror = 0;
 447
 448         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 449         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 450         while (1) {
 451                 ret = read_extent_buffer_pages(io_tree, eb, start,
 452                                                WAIT_COMPLETE,
 453                                                btree_get_extent, mirror_num);
 454                 if (!ret) {
 455                         if (!verify_parent_transid(io_tree, eb,
 456                                                    parent_transid, 0))
 457                                 break;
 458                         else
 459                                 ret = -EIO;
 460                 }
 461
 462                 /*
 463                  * This buffer's crc is fine, but its contents are corrupted, so
 464                  * there is no reason to read the other copies, they won't be
 465                  * any less wrong.
 466                  */
 467                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 468                         break;
 469
 470                 num_copies = btrfs_num_copies(root->fs_info,
 471                                               eb->start, eb->len);
 472                 if (num_copies == 1)
 473                         break;
 474
 475                 if (!failed_mirror) {
 476                         failed = 1;
 477                         failed_mirror = eb->read_mirror;
 478                 }
 479
 480                 mirror_num++;
 481                 if (mirror_num == failed_mirror)
 482                         mirror_num++;
 483
 484                 if (mirror_num > num_copies)
 485                         break;
 486         }
 487
 488         if (failed && !ret && failed_mirror)
 489                 repair_eb_io_failure(root, eb, failed_mirror);
 490
 491         return ret;
 492 }
 493
 494 /*
 495  * checksum a dirty tree block before IO.  This has extra checks to make sure
 496  * we only fill in the checksum field in the first page of a multi-page block
 497  */
 498
 499 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 500 {
 501         u64 start = page_offset(page);
 502         u64 found_start;
 503         struct extent_buffer *eb;
 504
 505         eb = (struct extent_buffer *)page->private;
 506         if (page != eb->pages[0])
 507                 return 0;
 508         found_start = btrfs_header_bytenr(eb);
 509         if (WARN_ON(found_start != start || !PageUptodate(page)))
 510                 return 0;
 511         csum_tree_block(fs_info, eb, 0);
 512         return 0;
 513 }
 514
 515 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 516                                  struct extent_buffer *eb)
 517 {
 518         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 519         u8 fsid[BTRFS_UUID_SIZE];
 520         int ret = 1;
 521
 522         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 523         while (fs_devices) {
 524                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 525                         ret = 0;
 526                         break;
 527                 }
 528                 fs_devices = fs_devices->seed;
 529         }
 530         return ret;
 531 }
 532
 533 #define CORRUPT(reason, eb, root, slot)                         \
 534         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 535                    "root=%llu, slot=%d", reason,                        \
 536                btrfs_header_bytenr(eb), root->objectid, slot)
 537
 538 static noinline int check_leaf(struct btrfs_root *root,
 539                                struct extent_buffer *leaf)
 540 {
 541         struct btrfs_key key;
 542         struct btrfs_key leaf_key;
 543         u32 nritems = btrfs_header_nritems(leaf);
 544         int slot;
 545
 546         if (nritems == 0)
 547                 return 0;
 548
 549         /* Check the 0 item */
 550         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 551             BTRFS_LEAF_DATA_SIZE(root)) {
 552                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 553                 return -EIO;
 554         }
 555
 556         /*
 557          * Check to make sure each items keys are in the correct order and their
 558          * offsets make sense.  We only have to loop through nritems-1 because
 559          * we check the current slot against the next slot, which verifies the
 560          * next slot's offset+size makes sense and that the current's slot
 561          * offset is correct.
 562          */
 563         for (slot = 0; slot < nritems - 1; slot++) {
 564                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 565                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 566
 567                 /* Make sure the keys are in the right order */
 568                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 569                         CORRUPT("bad key order", leaf, root, slot);
 570                         return -EIO;
 571                 }
 572
 573                 /*
 574                  * Make sure the offset and ends are right, remember that the
 575                  * item data starts at the end of the leaf and grows towards the
 576                  * front.
 577                  */
 578                 if (btrfs_item_offset_nr(leaf, slot) !=
 579                         btrfs_item_end_nr(leaf, slot + 1)) {
 580                         CORRUPT("slot offset bad", leaf, root, slot);
 581                         return -EIO;
 582                 }
 583
 584                 /*
 585                  * Check to make sure that we don't point outside of the leaf,
 586                  * just incase all the items are consistent to eachother, but
 587                  * all point outside of the leaf.
 588                  */
 589                 if (btrfs_item_end_nr(leaf, slot) >
 590                     BTRFS_LEAF_DATA_SIZE(root)) {
 591                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 592                         return -EIO;
 593                 }
 594         }
 595
 596         return 0;
 597 }
 598
 599 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 600                                       u64 phy_offset, struct page *page,
 601                                       u64 start, u64 end, int mirror)
 602 {
 603         u64 found_start;
 604         int found_level;
 605         struct extent_buffer *eb;
 606         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 607         int ret = 0;
 608         int reads_done;
 609
 610         if (!page->private)
 611                 goto out;
 612
 613         eb = (struct extent_buffer *)page->private;
 614
 615         /* the pending IO might have been the only thing that kept this buffer
 616          * in memory.  Make sure we have a ref for all this other checks
 617          */
 618         extent_buffer_get(eb);
 619
 620         reads_done = atomic_dec_and_test(&eb->io_pages);
 621         if (!reads_done)
 622                 goto err;
 623
 624         eb->read_mirror = mirror;
 625         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 626                 ret = -EIO;
 627                 goto err;
 628         }
 629
 630         found_start = btrfs_header_bytenr(eb);
 631         if (found_start != eb->start) {
 632                 btrfs_err_rl(eb->fs_info, "bad tree block start %llu %llu",
 633                                found_start, eb->start);
 634                 ret = -EIO;
 635                 goto err;
 636         }
 637         if (check_tree_block_fsid(root->fs_info, eb)) {
 638                 btrfs_err_rl(eb->fs_info, "bad fsid on block %llu",
 639                                eb->start);
 640                 ret = -EIO;
 641                 goto err;
 642         }
 643         found_level = btrfs_header_level(eb);
 644         if (found_level >= BTRFS_MAX_LEVEL) {
 645                 btrfs_err(root->fs_info, "bad tree block level %d",
 646                            (int)btrfs_header_level(eb));
 647                 ret = -EIO;
 648                 goto err;
 649         }
 650
 651         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 652                                        eb, found_level);
 653
 654         ret = csum_tree_block(root->fs_info, eb, 1);
 655         if (ret) {
 656                 ret = -EIO;
 657                 goto err;
 658         }
 659
 660         /*
 661          * If this is a leaf block and it is corrupt, set the corrupt bit so
 662          * that we don't try and read the other copies of this block, just
 663          * return -EIO.
 664          */
 665         if (found_level == 0 && check_leaf(root, eb)) {
 666                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 667                 ret = -EIO;
 668         }
 669
 670         if (!ret)
 671                 set_extent_buffer_uptodate(eb);
 672 err:
 673         if (reads_done &&
 674             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 675                 btree_readahead_hook(root, eb, eb->start, ret);
 676
 677         if (ret) {
 678                 /*
 679                  * our io error hook is going to dec the io pages
 680                  * again, we have to make sure it has something
 681                  * to decrement
 682                  */
 683                 atomic_inc(&eb->io_pages);
 684                 clear_extent_buffer_uptodate(eb);
 685         }
 686         free_extent_buffer(eb);
 687 out:
 688         return ret;
 689 }
 690
 691 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 692 {
 693         struct extent_buffer *eb;
 694         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 695
 696         eb = (struct extent_buffer *)page->private;
 697         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 698         eb->read_mirror = failed_mirror;
 699         atomic_dec(&eb->io_pages);
 700         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 701                 btree_readahead_hook(root, eb, eb->start, -EIO);
 702         return -EIO;    /* we fixed nothing */
 703 }
 704
 705 static void end_workqueue_bio(struct bio *bio)
 706 {
 707         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 708         struct btrfs_fs_info *fs_info;
 709         struct btrfs_workqueue *wq;
 710         btrfs_work_func_t func;
 711
 712         fs_info = end_io_wq->info;
 713         end_io_wq->error = bio->bi_error;
 714
 715         if (bio->bi_rw & REQ_WRITE) {
 716                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 717                         wq = fs_info->endio_meta_write_workers;
 718                         func = btrfs_endio_meta_write_helper;
 719                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 720                         wq = fs_info->endio_freespace_worker;
 721                         func = btrfs_freespace_write_helper;
 722                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 723                         wq = fs_info->endio_raid56_workers;
 724                         func = btrfs_endio_raid56_helper;
 725                 } else {
 726                         wq = fs_info->endio_write_workers;
 727                         func = btrfs_endio_write_helper;
 728                 }
 729         } else {
 730                 if (unlikely(end_io_wq->metadata ==
 731                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 732                         wq = fs_info->endio_repair_workers;
 733                         func = btrfs_endio_repair_helper;
 734                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 735                         wq = fs_info->endio_raid56_workers;
 736                         func = btrfs_endio_raid56_helper;
 737                 } else if (end_io_wq->metadata) {
 738                         wq = fs_info->endio_meta_workers;
 739                         func = btrfs_endio_meta_helper;
 740                 } else {
 741                         wq = fs_info->endio_workers;
 742                         func = btrfs_endio_helper;
 743                 }
 744         }
 745
 746         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 747         btrfs_queue_work(wq, &end_io_wq->work);
 748 }
 749
 750 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 751                         enum btrfs_wq_endio_type metadata)
 752 {
 753         struct btrfs_end_io_wq *end_io_wq;
 754
 755         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 756         if (!end_io_wq)
 757                 return -ENOMEM;
 758
 759         end_io_wq->private = bio->bi_private;
 760         end_io_wq->end_io = bio->bi_end_io;
 761         end_io_wq->info = info;
 762         end_io_wq->error = 0;
 763         end_io_wq->bio = bio;
 764         end_io_wq->metadata = metadata;
 765
 766         bio->bi_private = end_io_wq;
 767         bio->bi_end_io = end_workqueue_bio;
 768         return 0;
 769 }
 770
 771 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 772 {
 773         unsigned long limit = min_t(unsigned long,
 774                                     info->thread_pool_size,
 775                                     info->fs_devices->open_devices);
 776         return 256 * limit;
 777 }
 778
 779 static void run_one_async_start(struct btrfs_work *work)
 780 {
 781         struct async_submit_bio *async;
 782         int ret;
 783
 784         async = container_of(work, struct  async_submit_bio, work);
 785         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 786                                       async->mirror_num, async->bio_flags,
 787                                       async->bio_offset);
 788         if (ret)
 789                 async->error = ret;
 790 }
 791
 792 static void run_one_async_done(struct btrfs_work *work)
 793 {
 794         struct btrfs_fs_info *fs_info;
 795         struct async_submit_bio *async;
 796         int limit;
 797
 798         async = container_of(work, struct  async_submit_bio, work);
 799         fs_info = BTRFS_I(async->inode)->root->fs_info;
 800
 801         limit = btrfs_async_submit_limit(fs_info);
 802         limit = limit * 2 / 3;
 803
 804         /*
 805          * atomic_dec_return implies a barrier for waitqueue_active
 806          */
 807         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 808             waitqueue_active(&fs_info->async_submit_wait))
 809                 wake_up(&fs_info->async_submit_wait);
 810
 811         /* If an error occured we just want to clean up the bio and move on */
 812         if (async->error) {
 813                 async->bio->bi_error = async->error;
 814                 bio_endio(async->bio);
 815                 return;
 816         }
 817
 818         async->submit_bio_done(async->inode, async->rw, async->bio,
 819                                async->mirror_num, async->bio_flags,
 820                                async->bio_offset);
 821 }
 822
 823 static void run_one_async_free(struct btrfs_work *work)
 824 {
 825         struct async_submit_bio *async;
 826
 827         async = container_of(work, struct  async_submit_bio, work);
 828         kfree(async);
 829 }
 830
 831 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 832                         int rw, struct bio *bio, int mirror_num,
 833                         unsigned long bio_flags,
 834                         u64 bio_offset,
 835                         extent_submit_bio_hook_t *submit_bio_start,
 836                         extent_submit_bio_hook_t *submit_bio_done)
 837 {
 838         struct async_submit_bio *async;
 839
 840         async = kmalloc(sizeof(*async), GFP_NOFS);
 841         if (!async)
 842                 return -ENOMEM;
 843
 844         async->inode = inode;
 845         async->rw = rw;
 846         async->bio = bio;
 847         async->mirror_num = mirror_num;
 848         async->submit_bio_start = submit_bio_start;
 849         async->submit_bio_done = submit_bio_done;
 850
 851         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 852                         run_one_async_done, run_one_async_free);
 853
 854         async->bio_flags = bio_flags;
 855         async->bio_offset = bio_offset;
 856
 857         async->error = 0;
 858
 859         atomic_inc(&fs_info->nr_async_submits);
 860
 861         if (rw & REQ_SYNC)
 862                 btrfs_set_work_high_priority(&async->work);
 863
 864         btrfs_queue_work(fs_info->workers, &async->work);
 865
 866         while (atomic_read(&fs_info->async_submit_draining) &&
 867               atomic_read(&fs_info->nr_async_submits)) {
 868                 wait_event(fs_info->async_submit_wait,
 869                            (atomic_read(&fs_info->nr_async_submits) == 0));
 870         }
 871
 872         return 0;
 873 }
 874
 875 static int btree_csum_one_bio(struct bio *bio)
 876 {
 877         struct bio_vec *bvec;
 878         struct btrfs_root *root;
 879         int i, ret = 0;
 880
 881         bio_for_each_segment_all(bvec, bio, i) {
 882                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 883                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 884                 if (ret)
 885                         break;
 886         }
 887
 888         return ret;
 889 }
 890
 891 static int __btree_submit_bio_start(struct inode *inode, int rw,
 892                                     struct bio *bio, int mirror_num,
 893                                     unsigned long bio_flags,
 894                                     u64 bio_offset)
 895 {
 896         /*
 897          * when we're called for a write, we're already in the async
 898          * submission context.  Just jump into btrfs_map_bio
 899          */
 900         return btree_csum_one_bio(bio);
 901 }
 902
 903 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 904                                  int mirror_num, unsigned long bio_flags,
 905                                  u64 bio_offset)
 906 {
 907         int ret;
 908
 909         /*
 910          * when we're called for a write, we're already in the async
 911          * submission context.  Just jump into btrfs_map_bio
 912          */
 913         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 914         if (ret) {
 915                 bio->bi_error = ret;
 916                 bio_endio(bio);
 917         }
 918         return ret;
 919 }
 920
 921 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 922 {
 923         if (bio_flags & EXTENT_BIO_TREE_LOG)
 924                 return 0;
 925 #ifdef CONFIG_X86
 926         if (cpu_has_xmm4_2)
 927                 return 0;
 928 #endif
 929         return 1;
 930 }
 931
 932 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 933                                  int mirror_num, unsigned long bio_flags,
 934                                  u64 bio_offset)
 935 {
 936         int async = check_async_write(inode, bio_flags);
 937         int ret;
 938
 939         if (!(rw & REQ_WRITE)) {
 940                 /*
 941                  * called for a read, do the setup so that checksum validation
 942                  * can happen in the async kernel threads
 943                  */
 944                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 945                                           bio, BTRFS_WQ_ENDIO_METADATA);
 946                 if (ret)
 947                         goto out_w_error;
 948                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 949                                     mirror_num, 0);
 950         } else if (!async) {
 951                 ret = btree_csum_one_bio(bio);
 952                 if (ret)
 953                         goto out_w_error;
 954                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 955                                     mirror_num, 0);
 956         } else {
 957                 /*
 958                  * kthread helpers are used to submit writes so that
 959                  * checksumming can happen in parallel across all CPUs
 960                  */
 961                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 962                                           inode, rw, bio, mirror_num, 0,
 963                                           bio_offset,
 964                                           __btree_submit_bio_start,
 965                                           __btree_submit_bio_done);
 966         }
 967
 968         if (ret)
 969                 goto out_w_error;
 970         return 0;
 971
 972 out_w_error:
 973         bio->bi_error = ret;
 974         bio_endio(bio);
 975         return ret;
 976 }
 977
 978 #ifdef CONFIG_MIGRATION
 979 static int btree_migratepage(struct address_space *mapping,
 980                         struct page *newpage, struct page *page,
 981                         enum migrate_mode mode)
 982 {
 983         /*
 984          * we can't safely write a btree page from here,
 985          * we haven't done the locking hook
 986          */
 987         if (PageDirty(page))
 988                 return -EAGAIN;
 989         /*
 990          * Buffers may be managed in a filesystem specific way.
 991          * We must have no buffers or drop them.
 992          */
 993         if (page_has_private(page) &&
 994             !try_to_release_page(page, GFP_KERNEL))
 995                 return -EAGAIN;
 996         return migrate_page(mapping, newpage, page, mode);
 997 }
 998 #endif
 999
1000
1001 static int btree_writepages(struct address_space *mapping,
1002                             struct writeback_control *wbc)
1003 {
1004         struct btrfs_fs_info *fs_info;
1005         int ret;
1006
1007         if (wbc->sync_mode == WB_SYNC_NONE) {
1008
1009                 if (wbc->for_kupdate)
1010                         return 0;
1011
1012                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1013                 /* this is a bit racy, but that's ok */
1014                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1015                                              BTRFS_DIRTY_METADATA_THRESH);
1016                 if (ret < 0)
1017                         return 0;
1018         }
1019         return btree_write_cache_pages(mapping, wbc);
1020 }
1021
1022 static int btree_readpage(struct file *file, struct page *page)
1023 {
1024         struct extent_io_tree *tree;
1025         tree = &BTRFS_I(page->mapping->host)->io_tree;
1026         return extent_read_full_page(tree, page, btree_get_extent, 0);
1027 }
1028
1029 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1030 {
1031         if (PageWriteback(page) || PageDirty(page))
1032                 return 0;
1033
1034         return try_release_extent_buffer(page);
1035 }
1036
1037 static void btree_invalidatepage(struct page *page, unsigned int offset,
1038                                  unsigned int length)
1039 {
1040         struct extent_io_tree *tree;
1041         tree = &BTRFS_I(page->mapping->host)->io_tree;
1042         extent_invalidatepage(tree, page, offset);
1043         btree_releasepage(page, GFP_NOFS);
1044         if (PagePrivate(page)) {
1045                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1046                            "page private not zero on page %llu",
1047                            (unsigned long long)page_offset(page));
1048                 ClearPagePrivate(page);
1049                 set_page_private(page, 0);
1050                 page_cache_release(page);
1051         }
1052 }
1053
1054 static int btree_set_page_dirty(struct page *page)
1055 {
1056 #ifdef DEBUG
1057         struct extent_buffer *eb;
1058
1059         BUG_ON(!PagePrivate(page));
1060         eb = (struct extent_buffer *)page->private;
1061         BUG_ON(!eb);
1062         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063         BUG_ON(!atomic_read(&eb->refs));
1064         btrfs_assert_tree_locked(eb);
1065 #endif
1066         return __set_page_dirty_nobuffers(page);
1067 }
1068
1069 static const struct address_space_operations btree_aops = {
1070         .readpage       = btree_readpage,
1071         .writepages     = btree_writepages,
1072         .releasepage    = btree_releasepage,
1073         .invalidatepage = btree_invalidatepage,
1074 #ifdef CONFIG_MIGRATION
1075         .migratepage    = btree_migratepage,
1076 #endif
1077         .set_page_dirty = btree_set_page_dirty,
1078 };
1079
1080 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1081 {
1082         struct extent_buffer *buf = NULL;
1083         struct inode *btree_inode = root->fs_info->btree_inode;
1084
1085         buf = btrfs_find_create_tree_block(root, bytenr);
1086         if (!buf)
1087                 return;
1088         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1089                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1090         free_extent_buffer(buf);
1091 }
1092
1093 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1094                          int mirror_num, struct extent_buffer **eb)
1095 {
1096         struct extent_buffer *buf = NULL;
1097         struct inode *btree_inode = root->fs_info->btree_inode;
1098         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1099         int ret;
1100
1101         buf = btrfs_find_create_tree_block(root, bytenr);
1102         if (!buf)
1103                 return 0;
1104
1105         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1106
1107         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1108                                        btree_get_extent, mirror_num);
1109         if (ret) {
1110                 free_extent_buffer(buf);
1111                 return ret;
1112         }
1113
1114         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1115                 free_extent_buffer(buf);
1116                 return -EIO;
1117         } else if (extent_buffer_uptodate(buf)) {
1118                 *eb = buf;
1119         } else {
1120                 free_extent_buffer(buf);
1121         }
1122         return 0;
1123 }
1124
1125 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1126                                             u64 bytenr)
1127 {
1128         return find_extent_buffer(fs_info, bytenr);
1129 }
1130
1131 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1132                                                  u64 bytenr)
1133 {
1134         if (btrfs_test_is_dummy_root(root))
1135                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1136         return alloc_extent_buffer(root->fs_info, bytenr);
1137 }
1138
1139
1140 int btrfs_write_tree_block(struct extent_buffer *buf)
1141 {
1142         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1143                                         buf->start + buf->len - 1);
1144 }
1145
1146 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1147 {
1148         return filemap_fdatawait_range(buf->pages[0]->mapping,
1149                                        buf->start, buf->start + buf->len - 1);
1150 }
1151
1152 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1153                                       u64 parent_transid)
1154 {
1155         struct extent_buffer *buf = NULL;
1156         int ret;
1157
1158         buf = btrfs_find_create_tree_block(root, bytenr);
1159         if (!buf)
1160                 return ERR_PTR(-ENOMEM);
1161
1162         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1163         if (ret) {
1164                 free_extent_buffer(buf);
1165                 return ERR_PTR(ret);
1166         }
1167         return buf;
1168
1169 }
1170
1171 void clean_tree_block(struct btrfs_trans_handle *trans,
1172                       struct btrfs_fs_info *fs_info,
1173                       struct extent_buffer *buf)
1174 {
1175         if (btrfs_header_generation(buf) ==
1176             fs_info->running_transaction->transid) {
1177                 btrfs_assert_tree_locked(buf);
1178
1179                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1180                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1181                                              -buf->len,
1182                                              fs_info->dirty_metadata_batch);
1183                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1184                         btrfs_set_lock_blocking(buf);
1185                         clear_extent_buffer_dirty(buf);
1186                 }
1187         }
1188 }
1189
1190 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1191 {
1192         struct btrfs_subvolume_writers *writers;
1193         int ret;
1194
1195         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1196         if (!writers)
1197                 return ERR_PTR(-ENOMEM);
1198
1199         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1200         if (ret < 0) {
1201                 kfree(writers);
1202                 return ERR_PTR(ret);
1203         }
1204
1205         init_waitqueue_head(&writers->wait);
1206         return writers;
1207 }
1208
1209 static void
1210 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1211 {
1212         percpu_counter_destroy(&writers->counter);
1213         kfree(writers);
1214 }
1215
1216 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1217                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1218                          u64 objectid)
1219 {
1220         root->node = NULL;
1221         root->commit_root = NULL;
1222         root->sectorsize = sectorsize;
1223         root->nodesize = nodesize;
1224         root->stripesize = stripesize;
1225         root->state = 0;
1226         root->orphan_cleanup_state = 0;
1227
1228         root->objectid = objectid;
1229         root->last_trans = 0;
1230         root->highest_objectid = 0;
1231         root->nr_delalloc_inodes = 0;
1232         root->nr_ordered_extents = 0;
1233         root->name = NULL;
1234         root->inode_tree = RB_ROOT;
1235         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1236         root->block_rsv = NULL;
1237         root->orphan_block_rsv = NULL;
1238
1239         INIT_LIST_HEAD(&root->dirty_list);
1240         INIT_LIST_HEAD(&root->root_list);
1241         INIT_LIST_HEAD(&root->delalloc_inodes);
1242         INIT_LIST_HEAD(&root->delalloc_root);
1243         INIT_LIST_HEAD(&root->ordered_extents);
1244         INIT_LIST_HEAD(&root->ordered_root);
1245         INIT_LIST_HEAD(&root->logged_list[0]);
1246         INIT_LIST_HEAD(&root->logged_list[1]);
1247         spin_lock_init(&root->orphan_lock);
1248         spin_lock_init(&root->inode_lock);
1249         spin_lock_init(&root->delalloc_lock);
1250         spin_lock_init(&root->ordered_extent_lock);
1251         spin_lock_init(&root->accounting_lock);
1252         spin_lock_init(&root->log_extents_lock[0]);
1253         spin_lock_init(&root->log_extents_lock[1]);
1254         mutex_init(&root->objectid_mutex);
1255         mutex_init(&root->log_mutex);
1256         mutex_init(&root->ordered_extent_mutex);
1257         mutex_init(&root->delalloc_mutex);
1258         init_waitqueue_head(&root->log_writer_wait);
1259         init_waitqueue_head(&root->log_commit_wait[0]);
1260         init_waitqueue_head(&root->log_commit_wait[1]);
1261         INIT_LIST_HEAD(&root->log_ctxs[0]);
1262         INIT_LIST_HEAD(&root->log_ctxs[1]);
1263         atomic_set(&root->log_commit[0], 0);
1264         atomic_set(&root->log_commit[1], 0);
1265         atomic_set(&root->log_writers, 0);
1266         atomic_set(&root->log_batch, 0);
1267         atomic_set(&root->orphan_inodes, 0);
1268         atomic_set(&root->refs, 1);
1269         atomic_set(&root->will_be_snapshoted, 0);
1270         atomic_set(&root->qgroup_meta_rsv, 0);
1271         root->log_transid = 0;
1272         root->log_transid_committed = -1;
1273         root->last_log_commit = 0;
1274         if (fs_info)
1275                 extent_io_tree_init(&root->dirty_log_pages,
1276                                      fs_info->btree_inode->i_mapping);
1277
1278         memset(&root->root_key, 0, sizeof(root->root_key));
1279         memset(&root->root_item, 0, sizeof(root->root_item));
1280         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1281         if (fs_info)
1282                 root->defrag_trans_start = fs_info->generation;
1283         else
1284                 root->defrag_trans_start = 0;
1285         root->root_key.objectid = objectid;
1286         root->anon_dev = 0;
1287
1288         spin_lock_init(&root->root_item_lock);
1289 }
1290
1291 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1292 {
1293         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1294         if (root)
1295                 root->fs_info = fs_info;
1296         return root;
1297 }
1298
1299 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1300 /* Should only be used by the testing infrastructure */
1301 struct btrfs_root *btrfs_alloc_dummy_root(void)
1302 {
1303         struct btrfs_root *root;
1304
1305         root = btrfs_alloc_root(NULL);
1306         if (!root)
1307                 return ERR_PTR(-ENOMEM);
1308         __setup_root(4096, 4096, 4096, root, NULL, 1);
1309         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1310         root->alloc_bytenr = 0;
1311
1312         return root;
1313 }
1314 #endif
1315
1316 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1317                                      struct btrfs_fs_info *fs_info,
1318                                      u64 objectid)
1319 {
1320         struct extent_buffer *leaf;
1321         struct btrfs_root *tree_root = fs_info->tree_root;
1322         struct btrfs_root *root;
1323         struct btrfs_key key;
1324         int ret = 0;
1325         uuid_le uuid;
1326
1327         root = btrfs_alloc_root(fs_info);
1328         if (!root)
1329                 return ERR_PTR(-ENOMEM);
1330
1331         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1332                 tree_root->stripesize, root, fs_info, objectid);
1333         root->root_key.objectid = objectid;
1334         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1335         root->root_key.offset = 0;
1336
1337         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1338         if (IS_ERR(leaf)) {
1339                 ret = PTR_ERR(leaf);
1340                 leaf = NULL;
1341                 goto fail;
1342         }
1343
1344         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1345         btrfs_set_header_bytenr(leaf, leaf->start);
1346         btrfs_set_header_generation(leaf, trans->transid);
1347         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1348         btrfs_set_header_owner(leaf, objectid);
1349         root->node = leaf;
1350
1351         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1352                             BTRFS_FSID_SIZE);
1353         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1354                             btrfs_header_chunk_tree_uuid(leaf),
1355                             BTRFS_UUID_SIZE);
1356         btrfs_mark_buffer_dirty(leaf);
1357
1358         root->commit_root = btrfs_root_node(root);
1359         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1360
1361         root->root_item.flags = 0;
1362         root->root_item.byte_limit = 0;
1363         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1364         btrfs_set_root_generation(&root->root_item, trans->transid);
1365         btrfs_set_root_level(&root->root_item, 0);
1366         btrfs_set_root_refs(&root->root_item, 1);
1367         btrfs_set_root_used(&root->root_item, leaf->len);
1368         btrfs_set_root_last_snapshot(&root->root_item, 0);
1369         btrfs_set_root_dirid(&root->root_item, 0);
1370         uuid_le_gen(&uuid);
1371         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1372         root->root_item.drop_level = 0;
1373
1374         key.objectid = objectid;
1375         key.type = BTRFS_ROOT_ITEM_KEY;
1376         key.offset = 0;
1377         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1378         if (ret)
1379                 goto fail;
1380
1381         btrfs_tree_unlock(leaf);
1382
1383         return root;
1384
1385 fail:
1386         if (leaf) {
1387                 btrfs_tree_unlock(leaf);
1388                 free_extent_buffer(root->commit_root);
1389                 free_extent_buffer(leaf);
1390         }
1391         kfree(root);
1392
1393         return ERR_PTR(ret);
1394 }
1395
1396 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1397                                          struct btrfs_fs_info *fs_info)
1398 {
1399         struct btrfs_root *root;
1400         struct btrfs_root *tree_root = fs_info->tree_root;
1401         struct extent_buffer *leaf;
1402
1403         root = btrfs_alloc_root(fs_info);
1404         if (!root)
1405                 return ERR_PTR(-ENOMEM);
1406
1407         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1408                      tree_root->stripesize, root, fs_info,
1409                      BTRFS_TREE_LOG_OBJECTID);
1410
1411         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1412         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1413         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1414
1415         /*
1416          * DON'T set REF_COWS for log trees
1417          *
1418          * log trees do not get reference counted because they go away
1419          * before a real commit is actually done.  They do store pointers
1420          * to file data extents, and those reference counts still get
1421          * updated (along with back refs to the log tree).
1422          */
1423
1424         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1425                         NULL, 0, 0, 0);
1426         if (IS_ERR(leaf)) {
1427                 kfree(root);
1428                 return ERR_CAST(leaf);
1429         }
1430
1431         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1432         btrfs_set_header_bytenr(leaf, leaf->start);
1433         btrfs_set_header_generation(leaf, trans->transid);
1434         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1435         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1436         root->node = leaf;
1437
1438         write_extent_buffer(root->node, root->fs_info->fsid,
1439                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1440         btrfs_mark_buffer_dirty(root->node);
1441         btrfs_tree_unlock(root->node);
1442         return root;
1443 }
1444
1445 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1446                              struct btrfs_fs_info *fs_info)
1447 {
1448         struct btrfs_root *log_root;
1449
1450         log_root = alloc_log_tree(trans, fs_info);
1451         if (IS_ERR(log_root))
1452                 return PTR_ERR(log_root);
1453         WARN_ON(fs_info->log_root_tree);
1454         fs_info->log_root_tree = log_root;
1455         return 0;
1456 }
1457
1458 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1459                        struct btrfs_root *root)
1460 {
1461         struct btrfs_root *log_root;
1462         struct btrfs_inode_item *inode_item;
1463
1464         log_root = alloc_log_tree(trans, root->fs_info);
1465         if (IS_ERR(log_root))
1466                 return PTR_ERR(log_root);
1467
1468         log_root->last_trans = trans->transid;
1469         log_root->root_key.offset = root->root_key.objectid;
1470
1471         inode_item = &log_root->root_item.inode;
1472         btrfs_set_stack_inode_generation(inode_item, 1);
1473         btrfs_set_stack_inode_size(inode_item, 3);
1474         btrfs_set_stack_inode_nlink(inode_item, 1);
1475         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1476         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1477
1478         btrfs_set_root_node(&log_root->root_item, log_root->node);
1479
1480         WARN_ON(root->log_root);
1481         root->log_root = log_root;
1482         root->log_transid = 0;
1483         root->log_transid_committed = -1;
1484         root->last_log_commit = 0;
1485         return 0;
1486 }
1487
1488 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1489                                                struct btrfs_key *key)
1490 {
1491         struct btrfs_root *root;
1492         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1493         struct btrfs_path *path;
1494         u64 generation;
1495         int ret;
1496
1497         path = btrfs_alloc_path();
1498         if (!path)
1499                 return ERR_PTR(-ENOMEM);
1500
1501         root = btrfs_alloc_root(fs_info);
1502         if (!root) {
1503                 ret = -ENOMEM;
1504                 goto alloc_fail;
1505         }
1506
1507         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1508                 tree_root->stripesize, root, fs_info, key->objectid);
1509
1510         ret = btrfs_find_root(tree_root, key, path,
1511                               &root->root_item, &root->root_key);
1512         if (ret) {
1513                 if (ret > 0)
1514                         ret = -ENOENT;
1515                 goto find_fail;
1516         }
1517
1518         generation = btrfs_root_generation(&root->root_item);
1519         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1520                                      generation);
1521         if (IS_ERR(root->node)) {
1522                 ret = PTR_ERR(root->node);
1523                 goto find_fail;
1524         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1525                 ret = -EIO;
1526                 free_extent_buffer(root->node);
1527                 goto find_fail;
1528         }
1529         root->commit_root = btrfs_root_node(root);
1530 out:
1531         btrfs_free_path(path);
1532         return root;
1533
1534 find_fail:
1535         kfree(root);
1536 alloc_fail:
1537         root = ERR_PTR(ret);
1538         goto out;
1539 }
1540
1541 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1542                                       struct btrfs_key *location)
1543 {
1544         struct btrfs_root *root;
1545
1546         root = btrfs_read_tree_root(tree_root, location);
1547         if (IS_ERR(root))
1548                 return root;
1549
1550         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1551                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1552                 btrfs_check_and_init_root_item(&root->root_item);
1553         }
1554
1555         return root;
1556 }
1557
1558 int btrfs_init_fs_root(struct btrfs_root *root)
1559 {
1560         int ret;
1561         struct btrfs_subvolume_writers *writers;
1562
1563         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1564         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1565                                         GFP_NOFS);
1566         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1567                 ret = -ENOMEM;
1568                 goto fail;
1569         }
1570
1571         writers = btrfs_alloc_subvolume_writers();
1572         if (IS_ERR(writers)) {
1573                 ret = PTR_ERR(writers);
1574                 goto fail;
1575         }
1576         root->subv_writers = writers;
1577
1578         btrfs_init_free_ino_ctl(root);
1579         spin_lock_init(&root->ino_cache_lock);
1580         init_waitqueue_head(&root->ino_cache_wait);
1581
1582         ret = get_anon_bdev(&root->anon_dev);
1583         if (ret)
1584                 goto free_writers;
1585         return 0;
1586
1587 free_writers:
1588         btrfs_free_subvolume_writers(root->subv_writers);
1589 fail:
1590         kfree(root->free_ino_ctl);
1591         kfree(root->free_ino_pinned);
1592         return ret;
1593 }
1594
1595 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1596                                                u64 root_id)
1597 {
1598         struct btrfs_root *root;
1599
1600         spin_lock(&fs_info->fs_roots_radix_lock);
1601         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1602                                  (unsigned long)root_id);
1603         spin_unlock(&fs_info->fs_roots_radix_lock);
1604         return root;
1605 }
1606
1607 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1608                          struct btrfs_root *root)
1609 {
1610         int ret;
1611
1612         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1613         if (ret)
1614                 return ret;
1615
1616         spin_lock(&fs_info->fs_roots_radix_lock);
1617         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1618                                 (unsigned long)root->root_key.objectid,
1619                                 root);
1620         if (ret == 0)
1621                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1622         spin_unlock(&fs_info->fs_roots_radix_lock);
1623         radix_tree_preload_end();
1624
1625         return ret;
1626 }
1627
1628 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1629                                      struct btrfs_key *location,
1630                                      bool check_ref)
1631 {
1632         struct btrfs_root *root;
1633         struct btrfs_path *path;
1634         struct btrfs_key key;
1635         int ret;
1636
1637         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1638                 return fs_info->tree_root;
1639         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1640                 return fs_info->extent_root;
1641         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1642                 return fs_info->chunk_root;
1643         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1644                 return fs_info->dev_root;
1645         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1646                 return fs_info->csum_root;
1647         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1648                 return fs_info->quota_root ? fs_info->quota_root :
1649                                              ERR_PTR(-ENOENT);
1650         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1651                 return fs_info->uuid_root ? fs_info->uuid_root :
1652                                             ERR_PTR(-ENOENT);
1653 again:
1654         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1655         if (root) {
1656                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1657                         return ERR_PTR(-ENOENT);
1658                 return root;
1659         }
1660
1661         root = btrfs_read_fs_root(fs_info->tree_root, location);
1662         if (IS_ERR(root))
1663                 return root;
1664
1665         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1666                 ret = -ENOENT;
1667                 goto fail;
1668         }
1669
1670         ret = btrfs_init_fs_root(root);
1671         if (ret)
1672                 goto fail;
1673
1674         path = btrfs_alloc_path();
1675         if (!path) {
1676                 ret = -ENOMEM;
1677                 goto fail;
1678         }
1679         key.objectid = BTRFS_ORPHAN_OBJECTID;
1680         key.type = BTRFS_ORPHAN_ITEM_KEY;
1681         key.offset = location->objectid;
1682
1683         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1684         btrfs_free_path(path);
1685         if (ret < 0)
1686                 goto fail;
1687         if (ret == 0)
1688                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1689
1690         ret = btrfs_insert_fs_root(fs_info, root);
1691         if (ret) {
1692                 if (ret == -EEXIST) {
1693                         free_fs_root(root);
1694                         goto again;
1695                 }
1696                 goto fail;
1697         }
1698         return root;
1699 fail:
1700         free_fs_root(root);
1701         return ERR_PTR(ret);
1702 }
1703
1704 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1705 {
1706         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1707         int ret = 0;
1708         struct btrfs_device *device;
1709         struct backing_dev_info *bdi;
1710
1711         rcu_read_lock();
1712         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1713                 if (!device->bdev)
1714                         continue;
1715                 bdi = blk_get_backing_dev_info(device->bdev);
1716                 if (bdi_congested(bdi, bdi_bits)) {
1717                         ret = 1;
1718                         break;
1719                 }
1720         }
1721         rcu_read_unlock();
1722         return ret;
1723 }
1724
1725 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1726 {
1727         int err;
1728
1729         err = bdi_setup_and_register(bdi, "btrfs");
1730         if (err)
1731                 return err;
1732
1733         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1734         bdi->congested_fn       = btrfs_congested_fn;
1735         bdi->congested_data     = info;
1736         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1737         return 0;
1738 }
1739
1740 /*
1741  * called by the kthread helper functions to finally call the bio end_io
1742  * functions.  This is where read checksum verification actually happens
1743  */
1744 static void end_workqueue_fn(struct btrfs_work *work)
1745 {
1746         struct bio *bio;
1747         struct btrfs_end_io_wq *end_io_wq;
1748
1749         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1750         bio = end_io_wq->bio;
1751
1752         bio->bi_error = end_io_wq->error;
1753         bio->bi_private = end_io_wq->private;
1754         bio->bi_end_io = end_io_wq->end_io;
1755         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1756         bio_endio(bio);
1757 }
1758
1759 static int cleaner_kthread(void *arg)
1760 {
1761         struct btrfs_root *root = arg;
1762         int again;
1763         struct btrfs_trans_handle *trans;
1764
1765         set_freezable();
1766         do {
1767                 again = 0;
1768
1769                 /* Make the cleaner go to sleep early. */
1770                 if (btrfs_need_cleaner_sleep(root))
1771                         goto sleep;
1772
1773                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1774                         goto sleep;
1775
1776                 /*
1777                  * Avoid the problem that we change the status of the fs
1778                  * during the above check and trylock.
1779                  */
1780                 if (btrfs_need_cleaner_sleep(root)) {
1781                         mutex_unlock(&root->fs_info->cleaner_mutex);
1782                         goto sleep;
1783                 }
1784
1785                 btrfs_run_delayed_iputs(root);
1786                 again = btrfs_clean_one_deleted_snapshot(root);
1787                 mutex_unlock(&root->fs_info->cleaner_mutex);
1788
1789                 /*
1790                  * The defragger has dealt with the R/O remount and umount,
1791                  * needn't do anything special here.
1792                  */
1793                 btrfs_run_defrag_inodes(root->fs_info);
1794
1795                 /*
1796                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1797                  * with relocation (btrfs_relocate_chunk) and relocation
1798                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1799                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1800                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1801                  * unused block groups.
1802                  */
1803                 btrfs_delete_unused_bgs(root->fs_info);
1804 sleep:
1805                 if (!try_to_freeze() && !again) {
1806                         set_current_state(TASK_INTERRUPTIBLE);
1807                         if (!kthread_should_stop())
1808                                 schedule();
1809                         __set_current_state(TASK_RUNNING);
1810                 }
1811         } while (!kthread_should_stop());
1812
1813         /*
1814          * Transaction kthread is stopped before us and wakes us up.
1815          * However we might have started a new transaction and COWed some
1816          * tree blocks when deleting unused block groups for example. So
1817          * make sure we commit the transaction we started to have a clean
1818          * shutdown when evicting the btree inode - if it has dirty pages
1819          * when we do the final iput() on it, eviction will trigger a
1820          * writeback for it which will fail with null pointer dereferences
1821          * since work queues and other resources were already released and
1822          * destroyed by the time the iput/eviction/writeback is made.
1823          */
1824         trans = btrfs_attach_transaction(root);
1825         if (IS_ERR(trans)) {
1826                 if (PTR_ERR(trans) != -ENOENT)
1827                         btrfs_err(root->fs_info,
1828                                   "cleaner transaction attach returned %ld",
1829                                   PTR_ERR(trans));
1830         } else {
1831                 int ret;
1832
1833                 ret = btrfs_commit_transaction(trans, root);
1834                 if (ret)
1835                         btrfs_err(root->fs_info,
1836                                   "cleaner open transaction commit returned %d",
1837                                   ret);
1838         }
1839
1840         return 0;
1841 }
1842
1843 static int transaction_kthread(void *arg)
1844 {
1845         struct btrfs_root *root = arg;
1846         struct btrfs_trans_handle *trans;
1847         struct btrfs_transaction *cur;
1848         u64 transid;
1849         unsigned long now;
1850         unsigned long delay;
1851         bool cannot_commit;
1852
1853         do {
1854                 cannot_commit = false;
1855                 delay = HZ * root->fs_info->commit_interval;
1856                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1857
1858                 spin_lock(&root->fs_info->trans_lock);
1859                 cur = root->fs_info->running_transaction;
1860                 if (!cur) {
1861                         spin_unlock(&root->fs_info->trans_lock);
1862                         goto sleep;
1863                 }
1864
1865                 now = get_seconds();
1866                 if (cur->state < TRANS_STATE_BLOCKED &&
1867                     (now < cur->start_time ||
1868                      now - cur->start_time < root->fs_info->commit_interval)) {
1869                         spin_unlock(&root->fs_info->trans_lock);
1870                         delay = HZ * 5;
1871                         goto sleep;
1872                 }
1873                 transid = cur->transid;
1874                 spin_unlock(&root->fs_info->trans_lock);
1875
1876                 /* If the file system is aborted, this will always fail. */
1877                 trans = btrfs_attach_transaction(root);
1878                 if (IS_ERR(trans)) {
1879                         if (PTR_ERR(trans) != -ENOENT)
1880                                 cannot_commit = true;
1881                         goto sleep;
1882                 }
1883                 if (transid == trans->transid) {
1884                         btrfs_commit_transaction(trans, root);
1885                 } else {
1886                         btrfs_end_transaction(trans, root);
1887                 }
1888 sleep:
1889                 wake_up_process(root->fs_info->cleaner_kthread);
1890                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1891
1892                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1893                                       &root->fs_info->fs_state)))
1894                         btrfs_cleanup_transaction(root);
1895                 if (!try_to_freeze()) {
1896                         set_current_state(TASK_INTERRUPTIBLE);
1897                         if (!kthread_should_stop() &&
1898                             (!btrfs_transaction_blocked(root->fs_info) ||
1899                              cannot_commit))
1900                                 schedule_timeout(delay);
1901                         __set_current_state(TASK_RUNNING);
1902                 }
1903         } while (!kthread_should_stop());
1904         return 0;
1905 }
1906
1907 /*
1908  * this will find the highest generation in the array of
1909  * root backups.  The index of the highest array is returned,
1910  * or -1 if we can't find anything.
1911  *
1912  * We check to make sure the array is valid by comparing the
1913  * generation of the latest  root in the array with the generation
1914  * in the super block.  If they don't match we pitch it.
1915  */
1916 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1917 {
1918         u64 cur;
1919         int newest_index = -1;
1920         struct btrfs_root_backup *root_backup;
1921         int i;
1922
1923         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1924                 root_backup = info->super_copy->super_roots + i;
1925                 cur = btrfs_backup_tree_root_gen(root_backup);
1926                 if (cur == newest_gen)
1927                         newest_index = i;
1928         }
1929
1930         /* check to see if we actually wrapped around */
1931         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1932                 root_backup = info->super_copy->super_roots;
1933                 cur = btrfs_backup_tree_root_gen(root_backup);
1934                 if (cur == newest_gen)
1935                         newest_index = 0;
1936         }
1937         return newest_index;
1938 }
1939
1940
1941 /*
1942  * find the oldest backup so we know where to store new entries
1943  * in the backup array.  This will set the backup_root_index
1944  * field in the fs_info struct
1945  */
1946 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1947                                      u64 newest_gen)
1948 {
1949         int newest_index = -1;
1950
1951         newest_index = find_newest_super_backup(info, newest_gen);
1952         /* if there was garbage in there, just move along */
1953         if (newest_index == -1) {
1954                 info->backup_root_index = 0;
1955         } else {
1956                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1957         }
1958 }
1959
1960 /*
1961  * copy all the root pointers into the super backup array.
1962  * this will bump the backup pointer by one when it is
1963  * done
1964  */
1965 static void backup_super_roots(struct btrfs_fs_info *info)
1966 {
1967         int next_backup;
1968         struct btrfs_root_backup *root_backup;
1969         int last_backup;
1970
1971         next_backup = info->backup_root_index;
1972         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1973                 BTRFS_NUM_BACKUP_ROOTS;
1974
1975         /*
1976          * just overwrite the last backup if we're at the same generation
1977          * this happens only at umount
1978          */
1979         root_backup = info->super_for_commit->super_roots + last_backup;
1980         if (btrfs_backup_tree_root_gen(root_backup) ==
1981             btrfs_header_generation(info->tree_root->node))
1982                 next_backup = last_backup;
1983
1984         root_backup = info->super_for_commit->super_roots + next_backup;
1985
1986         /*
1987          * make sure all of our padding and empty slots get zero filled
1988          * regardless of which ones we use today
1989          */
1990         memset(root_backup, 0, sizeof(*root_backup));
1991
1992         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1993
1994         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1995         btrfs_set_backup_tree_root_gen(root_backup,
1996                                btrfs_header_generation(info->tree_root->node));
1997
1998         btrfs_set_backup_tree_root_level(root_backup,
1999                                btrfs_header_level(info->tree_root->node));
2000
2001         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2002         btrfs_set_backup_chunk_root_gen(root_backup,
2003                                btrfs_header_generation(info->chunk_root->node));
2004         btrfs_set_backup_chunk_root_level(root_backup,
2005                                btrfs_header_level(info->chunk_root->node));
2006
2007         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2008         btrfs_set_backup_extent_root_gen(root_backup,
2009                                btrfs_header_generation(info->extent_root->node));
2010         btrfs_set_backup_extent_root_level(root_backup,
2011                                btrfs_header_level(info->extent_root->node));
2012
2013         /*
2014          * we might commit during log recovery, which happens before we set
2015          * the fs_root.  Make sure it is valid before we fill it in.
2016          */
2017         if (info->fs_root && info->fs_root->node) {
2018                 btrfs_set_backup_fs_root(root_backup,
2019                                          info->fs_root->node->start);
2020                 btrfs_set_backup_fs_root_gen(root_backup,
2021                                btrfs_header_generation(info->fs_root->node));
2022                 btrfs_set_backup_fs_root_level(root_backup,
2023                                btrfs_header_level(info->fs_root->node));
2024         }
2025
2026         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2027         btrfs_set_backup_dev_root_gen(root_backup,
2028                                btrfs_header_generation(info->dev_root->node));
2029         btrfs_set_backup_dev_root_level(root_backup,
2030                                        btrfs_header_level(info->dev_root->node));
2031
2032         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2033         btrfs_set_backup_csum_root_gen(root_backup,
2034                                btrfs_header_generation(info->csum_root->node));
2035         btrfs_set_backup_csum_root_level(root_backup,
2036                                btrfs_header_level(info->csum_root->node));
2037
2038         btrfs_set_backup_total_bytes(root_backup,
2039                              btrfs_super_total_bytes(info->super_copy));
2040         btrfs_set_backup_bytes_used(root_backup,
2041                              btrfs_super_bytes_used(info->super_copy));
2042         btrfs_set_backup_num_devices(root_backup,
2043                              btrfs_super_num_devices(info->super_copy));
2044
2045         /*
2046          * if we don't copy this out to the super_copy, it won't get remembered
2047          * for the next commit
2048          */
2049         memcpy(&info->super_copy->super_roots,
2050                &info->super_for_commit->super_roots,
2051                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2052 }
2053
2054 /*
2055  * this copies info out of the root backup array and back into
2056  * the in-memory super block.  It is meant to help iterate through
2057  * the array, so you send it the number of backups you've already
2058  * tried and the last backup index you used.
2059  *
2060  * this returns -1 when it has tried all the backups
2061  */
2062 static noinline int next_root_backup(struct btrfs_fs_info *info,
2063                                      struct btrfs_super_block *super,
2064                                      int *num_backups_tried, int *backup_index)
2065 {
2066         struct btrfs_root_backup *root_backup;
2067         int newest = *backup_index;
2068
2069         if (*num_backups_tried == 0) {
2070                 u64 gen = btrfs_super_generation(super);
2071
2072                 newest = find_newest_super_backup(info, gen);
2073                 if (newest == -1)
2074                         return -1;
2075
2076                 *backup_index = newest;
2077                 *num_backups_tried = 1;
2078         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2079                 /* we've tried all the backups, all done */
2080                 return -1;
2081         } else {
2082                 /* jump to the next oldest backup */
2083                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2084                         BTRFS_NUM_BACKUP_ROOTS;
2085                 *backup_index = newest;
2086                 *num_backups_tried += 1;
2087         }
2088         root_backup = super->super_roots + newest;
2089
2090         btrfs_set_super_generation(super,
2091                                    btrfs_backup_tree_root_gen(root_backup));
2092         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2093         btrfs_set_super_root_level(super,
2094                                    btrfs_backup_tree_root_level(root_backup));
2095         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2096
2097         /*
2098          * fixme: the total bytes and num_devices need to match or we should
2099          * need a fsck
2100          */
2101         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2102         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2103         return 0;
2104 }
2105
2106 /* helper to cleanup workers */
2107 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2108 {
2109         btrfs_destroy_workqueue(fs_info->fixup_workers);
2110         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2111         btrfs_destroy_workqueue(fs_info->workers);
2112         btrfs_destroy_workqueue(fs_info->endio_workers);
2113         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2114         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2115         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2116         btrfs_destroy_workqueue(fs_info->rmw_workers);
2117         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2118         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2119         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2120         btrfs_destroy_workqueue(fs_info->submit_workers);
2121         btrfs_destroy_workqueue(fs_info->delayed_workers);
2122         btrfs_destroy_workqueue(fs_info->caching_workers);
2123         btrfs_destroy_workqueue(fs_info->readahead_workers);
2124         btrfs_destroy_workqueue(fs_info->flush_workers);
2125         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2126         btrfs_destroy_workqueue(fs_info->extent_workers);
2127 }
2128
2129 static void free_root_extent_buffers(struct btrfs_root *root)
2130 {
2131         if (root) {
2132                 free_extent_buffer(root->node);
2133                 free_extent_buffer(root->commit_root);
2134                 root->node = NULL;
2135                 root->commit_root = NULL;
2136         }
2137 }
2138
2139 /* helper to cleanup tree roots */
2140 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2141 {
2142         free_root_extent_buffers(info->tree_root);
2143
2144         free_root_extent_buffers(info->dev_root);
2145         free_root_extent_buffers(info->extent_root);
2146         free_root_extent_buffers(info->csum_root);
2147         free_root_extent_buffers(info->quota_root);
2148         free_root_extent_buffers(info->uuid_root);
2149         if (chunk_root)
2150                 free_root_extent_buffers(info->chunk_root);
2151 }
2152
2153 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2154 {
2155         int ret;
2156         struct btrfs_root *gang[8];
2157         int i;
2158
2159         while (!list_empty(&fs_info->dead_roots)) {
2160                 gang[0] = list_entry(fs_info->dead_roots.next,
2161                                      struct btrfs_root, root_list);
2162                 list_del(&gang[0]->root_list);
2163
2164                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2165                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2166                 } else {
2167                         free_extent_buffer(gang[0]->node);
2168                         free_extent_buffer(gang[0]->commit_root);
2169                         btrfs_put_fs_root(gang[0]);
2170                 }
2171         }
2172
2173         while (1) {
2174                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2175                                              (void **)gang, 0,
2176                                              ARRAY_SIZE(gang));
2177                 if (!ret)
2178                         break;
2179                 for (i = 0; i < ret; i++)
2180                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2181         }
2182
2183         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2184                 btrfs_free_log_root_tree(NULL, fs_info);
2185                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2186                                             fs_info->pinned_extents);
2187         }
2188 }
2189
2190 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2191 {
2192         mutex_init(&fs_info->scrub_lock);
2193         atomic_set(&fs_info->scrubs_running, 0);
2194         atomic_set(&fs_info->scrub_pause_req, 0);
2195         atomic_set(&fs_info->scrubs_paused, 0);
2196         atomic_set(&fs_info->scrub_cancel_req, 0);
2197         init_waitqueue_head(&fs_info->scrub_pause_wait);
2198         fs_info->scrub_workers_refcnt = 0;
2199 }
2200
2201 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2202 {
2203         spin_lock_init(&fs_info->balance_lock);
2204         mutex_init(&fs_info->balance_mutex);
2205         atomic_set(&fs_info->balance_running, 0);
2206         atomic_set(&fs_info->balance_pause_req, 0);
2207         atomic_set(&fs_info->balance_cancel_req, 0);
2208         fs_info->balance_ctl = NULL;
2209         init_waitqueue_head(&fs_info->balance_wait_q);
2210 }
2211
2212 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2213                                    struct btrfs_root *tree_root)
2214 {
2215         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2216         set_nlink(fs_info->btree_inode, 1);
2217         /*
2218          * we set the i_size on the btree inode to the max possible int.
2219          * the real end of the address space is determined by all of
2220          * the devices in the system
2221          */
2222         fs_info->btree_inode->i_size = OFFSET_MAX;
2223         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2224
2225         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2226         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2227                              fs_info->btree_inode->i_mapping);
2228         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2229         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2230
2231         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2232
2233         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2234         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2235                sizeof(struct btrfs_key));
2236         set_bit(BTRFS_INODE_DUMMY,
2237                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2238         btrfs_insert_inode_hash(fs_info->btree_inode);
2239 }
2240
2241 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2242 {
2243         fs_info->dev_replace.lock_owner = 0;
2244         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2245         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2246         mutex_init(&fs_info->dev_replace.lock_management_lock);
2247         mutex_init(&fs_info->dev_replace.lock);
2248         init_waitqueue_head(&fs_info->replace_wait);
2249 }
2250
2251 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2252 {
2253         spin_lock_init(&fs_info->qgroup_lock);
2254         mutex_init(&fs_info->qgroup_ioctl_lock);
2255         fs_info->qgroup_tree = RB_ROOT;
2256         fs_info->qgroup_op_tree = RB_ROOT;
2257         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2258         fs_info->qgroup_seq = 1;
2259         fs_info->quota_enabled = 0;
2260         fs_info->pending_quota_state = 0;
2261         fs_info->qgroup_ulist = NULL;
2262         mutex_init(&fs_info->qgroup_rescan_lock);
2263 }
2264
2265 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2266                 struct btrfs_fs_devices *fs_devices)
2267 {
2268         int max_active = fs_info->thread_pool_size;
2269         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2270
2271         fs_info->workers =
2272                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2273                                       max_active, 16);
2274
2275         fs_info->delalloc_workers =
2276                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2277
2278         fs_info->flush_workers =
2279                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2280
2281         fs_info->caching_workers =
2282                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2283
2284         /*
2285          * a higher idle thresh on the submit workers makes it much more
2286          * likely that bios will be send down in a sane order to the
2287          * devices
2288          */
2289         fs_info->submit_workers =
2290                 btrfs_alloc_workqueue("submit", flags,
2291                                       min_t(u64, fs_devices->num_devices,
2292                                             max_active), 64);
2293
2294         fs_info->fixup_workers =
2295                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2296
2297         /*
2298          * endios are largely parallel and should have a very
2299          * low idle thresh
2300          */
2301         fs_info->endio_workers =
2302                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2303         fs_info->endio_meta_workers =
2304                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2305         fs_info->endio_meta_write_workers =
2306                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2307         fs_info->endio_raid56_workers =
2308                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2309         fs_info->endio_repair_workers =
2310                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2311         fs_info->rmw_workers =
2312                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2313         fs_info->endio_write_workers =
2314                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2315         fs_info->endio_freespace_worker =
2316                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2317         fs_info->delayed_workers =
2318                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2319         fs_info->readahead_workers =
2320                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2321         fs_info->qgroup_rescan_workers =
2322                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2323         fs_info->extent_workers =
2324                 btrfs_alloc_workqueue("extent-refs", flags,
2325                                       min_t(u64, fs_devices->num_devices,
2326                                             max_active), 8);
2327
2328         if (!(fs_info->workers && fs_info->delalloc_workers &&
2329               fs_info->submit_workers && fs_info->flush_workers &&
2330               fs_info->endio_workers && fs_info->endio_meta_workers &&
2331               fs_info->endio_meta_write_workers &&
2332               fs_info->endio_repair_workers &&
2333               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2334               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2335               fs_info->caching_workers && fs_info->readahead_workers &&
2336               fs_info->fixup_workers && fs_info->delayed_workers &&
2337               fs_info->extent_workers &&
2338               fs_info->qgroup_rescan_workers)) {
2339                 return -ENOMEM;
2340         }
2341
2342         return 0;
2343 }
2344
2345 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2346                             struct btrfs_fs_devices *fs_devices)
2347 {
2348         int ret;
2349         struct btrfs_root *tree_root = fs_info->tree_root;
2350         struct btrfs_root *log_tree_root;
2351         struct btrfs_super_block *disk_super = fs_info->super_copy;
2352         u64 bytenr = btrfs_super_log_root(disk_super);
2353
2354         if (fs_devices->rw_devices == 0) {
2355                 btrfs_warn(fs_info, "log replay required on RO media");
2356                 return -EIO;
2357         }
2358
2359         log_tree_root = btrfs_alloc_root(fs_info);
2360         if (!log_tree_root)
2361                 return -ENOMEM;
2362
2363         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2364                         tree_root->stripesize, log_tree_root, fs_info,
2365                         BTRFS_TREE_LOG_OBJECTID);
2366
2367         log_tree_root->node = read_tree_block(tree_root, bytenr,
2368                         fs_info->generation + 1);
2369         if (IS_ERR(log_tree_root->node)) {
2370                 btrfs_warn(fs_info, "failed to read log tree");
2371                 ret = PTR_ERR(log_tree_root->node);
2372                 kfree(log_tree_root);
2373                 return ret;
2374         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2375                 btrfs_err(fs_info, "failed to read log tree");
2376                 free_extent_buffer(log_tree_root->node);
2377                 kfree(log_tree_root);
2378                 return -EIO;
2379         }
2380         /* returns with log_tree_root freed on success */
2381         ret = btrfs_recover_log_trees(log_tree_root);
2382         if (ret) {
2383                 btrfs_std_error(tree_root->fs_info, ret,
2384                             "Failed to recover log tree");
2385                 free_extent_buffer(log_tree_root->node);
2386                 kfree(log_tree_root);
2387                 return ret;
2388         }
2389
2390         if (fs_info->sb->s_flags & MS_RDONLY) {
2391                 ret = btrfs_commit_super(tree_root);
2392                 if (ret)
2393                         return ret;
2394         }
2395
2396         return 0;
2397 }
2398
2399 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2400                             struct btrfs_root *tree_root)
2401 {
2402         struct btrfs_root *root;
2403         struct btrfs_key location;
2404         int ret;
2405
2406         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2407         location.type = BTRFS_ROOT_ITEM_KEY;
2408         location.offset = 0;
2409
2410         root = btrfs_read_tree_root(tree_root, &location);
2411         if (IS_ERR(root))
2412                 return PTR_ERR(root);
2413         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2414         fs_info->extent_root = root;
2415
2416         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2417         root = btrfs_read_tree_root(tree_root, &location);
2418         if (IS_ERR(root))
2419                 return PTR_ERR(root);
2420         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2421         fs_info->dev_root = root;
2422         btrfs_init_devices_late(fs_info);
2423
2424         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2425         root = btrfs_read_tree_root(tree_root, &location);
2426         if (IS_ERR(root))
2427                 return PTR_ERR(root);
2428         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2429         fs_info->csum_root = root;
2430
2431         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2432         root = btrfs_read_tree_root(tree_root, &location);
2433         if (!IS_ERR(root)) {
2434                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2435                 fs_info->quota_enabled = 1;
2436                 fs_info->pending_quota_state = 1;
2437                 fs_info->quota_root = root;
2438         }
2439
2440         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2441         root = btrfs_read_tree_root(tree_root, &location);
2442         if (IS_ERR(root)) {
2443                 ret = PTR_ERR(root);
2444                 if (ret != -ENOENT)
2445                         return ret;
2446         } else {
2447                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2448                 fs_info->uuid_root = root;
2449         }
2450
2451         return 0;
2452 }
2453
2454 int open_ctree(struct super_block *sb,
2455                struct btrfs_fs_devices *fs_devices,
2456                char *options)
2457 {
2458         u32 sectorsize;
2459         u32 nodesize;
2460         u32 stripesize;
2461         u64 generation;
2462         u64 features;
2463         struct btrfs_key location;
2464         struct buffer_head *bh;
2465         struct btrfs_super_block *disk_super;
2466         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2467         struct btrfs_root *tree_root;
2468         struct btrfs_root *chunk_root;
2469         int ret;
2470         int err = -EINVAL;
2471         int num_backups_tried = 0;
2472         int backup_index = 0;
2473         int max_active;
2474
2475         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2476         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2477         if (!tree_root || !chunk_root) {
2478                 err = -ENOMEM;
2479                 goto fail;
2480         }
2481
2482         ret = init_srcu_struct(&fs_info->subvol_srcu);
2483         if (ret) {
2484                 err = ret;
2485                 goto fail;
2486         }
2487
2488         ret = setup_bdi(fs_info, &fs_info->bdi);
2489         if (ret) {
2490                 err = ret;
2491                 goto fail_srcu;
2492         }
2493
2494         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2495         if (ret) {
2496                 err = ret;
2497                 goto fail_bdi;
2498         }
2499         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2500                                         (1 + ilog2(nr_cpu_ids));
2501
2502         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2503         if (ret) {
2504                 err = ret;
2505                 goto fail_dirty_metadata_bytes;
2506         }
2507
2508         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2509         if (ret) {
2510                 err = ret;
2511                 goto fail_delalloc_bytes;
2512         }
2513
2514         fs_info->btree_inode = new_inode(sb);
2515         if (!fs_info->btree_inode) {
2516                 err = -ENOMEM;
2517                 goto fail_bio_counter;
2518         }
2519
2520         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2521
2522         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2523         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2524         INIT_LIST_HEAD(&fs_info->trans_list);
2525         INIT_LIST_HEAD(&fs_info->dead_roots);
2526         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2527         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2528         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2529         spin_lock_init(&fs_info->delalloc_root_lock);
2530         spin_lock_init(&fs_info->trans_lock);
2531         spin_lock_init(&fs_info->fs_roots_radix_lock);
2532         spin_lock_init(&fs_info->delayed_iput_lock);
2533         spin_lock_init(&fs_info->defrag_inodes_lock);
2534         spin_lock_init(&fs_info->free_chunk_lock);
2535         spin_lock_init(&fs_info->tree_mod_seq_lock);
2536         spin_lock_init(&fs_info->super_lock);
2537         spin_lock_init(&fs_info->qgroup_op_lock);
2538         spin_lock_init(&fs_info->buffer_lock);
2539         spin_lock_init(&fs_info->unused_bgs_lock);
2540         rwlock_init(&fs_info->tree_mod_log_lock);
2541         mutex_init(&fs_info->unused_bg_unpin_mutex);
2542         mutex_init(&fs_info->delete_unused_bgs_mutex);
2543         mutex_init(&fs_info->reloc_mutex);
2544         mutex_init(&fs_info->delalloc_root_mutex);
2545         seqlock_init(&fs_info->profiles_lock);
2546         init_rwsem(&fs_info->delayed_iput_sem);
2547
2548         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2549         INIT_LIST_HEAD(&fs_info->space_info);
2550         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2551         INIT_LIST_HEAD(&fs_info->unused_bgs);
2552         btrfs_mapping_init(&fs_info->mapping_tree);
2553         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2554                              BTRFS_BLOCK_RSV_GLOBAL);
2555         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2556                              BTRFS_BLOCK_RSV_DELALLOC);
2557         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2558         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2559         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2560         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2561                              BTRFS_BLOCK_RSV_DELOPS);
2562         atomic_set(&fs_info->nr_async_submits, 0);
2563         atomic_set(&fs_info->async_delalloc_pages, 0);
2564         atomic_set(&fs_info->async_submit_draining, 0);
2565         atomic_set(&fs_info->nr_async_bios, 0);
2566         atomic_set(&fs_info->defrag_running, 0);
2567         atomic_set(&fs_info->qgroup_op_seq, 0);
2568         atomic64_set(&fs_info->tree_mod_seq, 0);
2569         fs_info->sb = sb;
2570         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2571         fs_info->metadata_ratio = 0;
2572         fs_info->defrag_inodes = RB_ROOT;
2573         fs_info->free_chunk_space = 0;
2574         fs_info->tree_mod_log = RB_ROOT;
2575         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2576         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2577         /* readahead state */
2578         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2579         spin_lock_init(&fs_info->reada_lock);
2580
2581         fs_info->thread_pool_size = min_t(unsigned long,
2582                                           num_online_cpus() + 2, 8);
2583
2584         INIT_LIST_HEAD(&fs_info->ordered_roots);
2585         spin_lock_init(&fs_info->ordered_root_lock);
2586         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2587                                         GFP_NOFS);
2588         if (!fs_info->delayed_root) {
2589                 err = -ENOMEM;
2590                 goto fail_iput;
2591         }
2592         btrfs_init_delayed_root(fs_info->delayed_root);
2593
2594         btrfs_init_scrub(fs_info);
2595 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2596         fs_info->check_integrity_print_mask = 0;
2597 #endif
2598         btrfs_init_balance(fs_info);
2599         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2600
2601         sb->s_blocksize = 4096;
2602         sb->s_blocksize_bits = blksize_bits(4096);
2603         sb->s_bdi = &fs_info->bdi;
2604
2605         btrfs_init_btree_inode(fs_info, tree_root);
2606
2607         spin_lock_init(&fs_info->block_group_cache_lock);
2608         fs_info->block_group_cache_tree = RB_ROOT;
2609         fs_info->first_logical_byte = (u64)-1;
2610
2611         extent_io_tree_init(&fs_info->freed_extents[0],
2612                              fs_info->btree_inode->i_mapping);
2613         extent_io_tree_init(&fs_info->freed_extents[1],
2614                              fs_info->btree_inode->i_mapping);
2615         fs_info->pinned_extents = &fs_info->freed_extents[0];
2616         fs_info->do_barriers = 1;
2617
2618
2619         mutex_init(&fs_info->ordered_operations_mutex);
2620         mutex_init(&fs_info->tree_log_mutex);
2621         mutex_init(&fs_info->chunk_mutex);
2622         mutex_init(&fs_info->transaction_kthread_mutex);
2623         mutex_init(&fs_info->cleaner_mutex);
2624         mutex_init(&fs_info->volume_mutex);
2625         mutex_init(&fs_info->ro_block_group_mutex);
2626         init_rwsem(&fs_info->commit_root_sem);
2627         init_rwsem(&fs_info->cleanup_work_sem);
2628         init_rwsem(&fs_info->subvol_sem);
2629         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2630
2631         btrfs_init_dev_replace_locks(fs_info);
2632         btrfs_init_qgroup(fs_info);
2633
2634         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2635         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2636
2637         init_waitqueue_head(&fs_info->transaction_throttle);
2638         init_waitqueue_head(&fs_info->transaction_wait);
2639         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2640         init_waitqueue_head(&fs_info->async_submit_wait);
2641
2642         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2643
2644         ret = btrfs_alloc_stripe_hash_table(fs_info);
2645         if (ret) {
2646                 err = ret;
2647                 goto fail_alloc;
2648         }
2649
2650         __setup_root(4096, 4096, 4096, tree_root,
2651                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2652
2653         invalidate_bdev(fs_devices->latest_bdev);
2654
2655         /*
2656          * Read super block and check the signature bytes only
2657          */
2658         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2659         if (IS_ERR(bh)) {
2660                 err = PTR_ERR(bh);
2661                 goto fail_alloc;
2662         }
2663
2664         /*
2665          * We want to check superblock checksum, the type is stored inside.
2666          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2667          */
2668         if (btrfs_check_super_csum(bh->b_data)) {
2669                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2670                 err = -EINVAL;
2671                 brelse(bh);
2672                 goto fail_alloc;
2673         }
2674
2675         /*
2676          * super_copy is zeroed at allocation time and we never touch the
2677          * following bytes up to INFO_SIZE, the checksum is calculated from
2678          * the whole block of INFO_SIZE
2679          */
2680         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2681         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2682                sizeof(*fs_info->super_for_commit));
2683         brelse(bh);
2684
2685         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2686
2687         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2688         if (ret) {
2689                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2690                 err = -EINVAL;
2691                 goto fail_alloc;
2692         }
2693
2694         disk_super = fs_info->super_copy;
2695         if (!btrfs_super_root(disk_super))
2696                 goto fail_alloc;
2697
2698         /* check FS state, whether FS is broken. */
2699         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2700                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2701
2702         /*
2703          * run through our array of backup supers and setup
2704          * our ring pointer to the oldest one
2705          */
2706         generation = btrfs_super_generation(disk_super);
2707         find_oldest_super_backup(fs_info, generation);
2708
2709         /*
2710          * In the long term, we'll store the compression type in the super
2711          * block, and it'll be used for per file compression control.
2712          */
2713         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2714
2715         ret = btrfs_parse_options(tree_root, options);
2716         if (ret) {
2717                 err = ret;
2718                 goto fail_alloc;
2719         }
2720
2721         features = btrfs_super_incompat_flags(disk_super) &
2722                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2723         if (features) {
2724                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2725                        "unsupported optional features (%Lx).\n",
2726                        features);
2727                 err = -EINVAL;
2728                 goto fail_alloc;
2729         }
2730
2731         /*
2732          * Leafsize and nodesize were always equal, this is only a sanity check.
2733          */
2734         if (le32_to_cpu(disk_super->__unused_leafsize) !=
2735             btrfs_super_nodesize(disk_super)) {
2736                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2737                        "blocksizes don't match.  node %d leaf %d\n",
2738                        btrfs_super_nodesize(disk_super),
2739                        le32_to_cpu(disk_super->__unused_leafsize));
2740                 err = -EINVAL;
2741                 goto fail_alloc;
2742         }
2743         if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2744                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2745                        "blocksize (%d) was too large\n",
2746                        btrfs_super_nodesize(disk_super));
2747                 err = -EINVAL;
2748                 goto fail_alloc;
2749         }
2750
2751         features = btrfs_super_incompat_flags(disk_super);
2752         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2753         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2754                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2755
2756         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2757                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2758
2759         /*
2760          * flag our filesystem as having big metadata blocks if
2761          * they are bigger than the page size
2762          */
2763         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2764                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2765                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2766                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2767         }
2768
2769         nodesize = btrfs_super_nodesize(disk_super);
2770         sectorsize = btrfs_super_sectorsize(disk_super);
2771         stripesize = btrfs_super_stripesize(disk_super);
2772         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2773         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2774
2775         /*
2776          * mixed block groups end up with duplicate but slightly offset
2777          * extent buffers for the same range.  It leads to corruptions
2778          */
2779         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2780             (sectorsize != nodesize)) {
2781                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2782                                 "are not allowed for mixed block groups on %s\n",
2783                                 sb->s_id);
2784                 goto fail_alloc;
2785         }
2786
2787         /*
2788          * Needn't use the lock because there is no other task which will
2789          * update the flag.
2790          */
2791         btrfs_set_super_incompat_flags(disk_super, features);
2792
2793         features = btrfs_super_compat_ro_flags(disk_super) &
2794                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2795         if (!(sb->s_flags & MS_RDONLY) && features) {
2796                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2797                        "unsupported option features (%Lx).\n",
2798                        features);
2799                 err = -EINVAL;
2800                 goto fail_alloc;
2801         }
2802
2803         max_active = fs_info->thread_pool_size;
2804
2805         ret = btrfs_init_workqueues(fs_info, fs_devices);
2806         if (ret) {
2807                 err = ret;
2808                 goto fail_sb_buffer;
2809         }
2810
2811         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2812         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2813                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2814
2815         tree_root->nodesize = nodesize;
2816         tree_root->sectorsize = sectorsize;
2817         tree_root->stripesize = stripesize;
2818
2819         sb->s_blocksize = sectorsize;
2820         sb->s_blocksize_bits = blksize_bits(sectorsize);
2821
2822         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2823                 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2824                 goto fail_sb_buffer;
2825         }
2826
2827         if (sectorsize != PAGE_SIZE) {
2828                 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2829                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2830                 goto fail_sb_buffer;
2831         }
2832
2833         mutex_lock(&fs_info->chunk_mutex);
2834         ret = btrfs_read_sys_array(tree_root);
2835         mutex_unlock(&fs_info->chunk_mutex);
2836         if (ret) {
2837                 printk(KERN_ERR "BTRFS: failed to read the system "
2838                        "array on %s\n", sb->s_id);
2839                 goto fail_sb_buffer;
2840         }
2841
2842         generation = btrfs_super_chunk_root_generation(disk_super);
2843
2844         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2845                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2846
2847         chunk_root->node = read_tree_block(chunk_root,
2848                                            btrfs_super_chunk_root(disk_super),
2849                                            generation);
2850         if (IS_ERR(chunk_root->node) ||
2851             !extent_buffer_uptodate(chunk_root->node)) {
2852                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2853                        sb->s_id);
2854                 if (!IS_ERR(chunk_root->node))
2855                         free_extent_buffer(chunk_root->node);
2856                 chunk_root->node = NULL;
2857                 goto fail_tree_roots;
2858         }
2859         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2860         chunk_root->commit_root = btrfs_root_node(chunk_root);
2861
2862         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2863            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2864
2865         ret = btrfs_read_chunk_tree(chunk_root);
2866         if (ret) {
2867                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2868                        sb->s_id);
2869                 goto fail_tree_roots;
2870         }
2871
2872         /*
2873          * keep the device that is marked to be the target device for the
2874          * dev_replace procedure
2875          */
2876         btrfs_close_extra_devices(fs_devices, 0);
2877
2878         if (!fs_devices->latest_bdev) {
2879                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2880                        sb->s_id);
2881                 goto fail_tree_roots;
2882         }
2883
2884 retry_root_backup:
2885         generation = btrfs_super_generation(disk_super);
2886
2887         tree_root->node = read_tree_block(tree_root,
2888                                           btrfs_super_root(disk_super),
2889                                           generation);
2890         if (IS_ERR(tree_root->node) ||
2891             !extent_buffer_uptodate(tree_root->node)) {
2892                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2893                        sb->s_id);
2894                 if (!IS_ERR(tree_root->node))
2895                         free_extent_buffer(tree_root->node);
2896                 tree_root->node = NULL;
2897                 goto recovery_tree_root;
2898         }
2899
2900         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2901         tree_root->commit_root = btrfs_root_node(tree_root);
2902         btrfs_set_root_refs(&tree_root->root_item, 1);
2903
2904         ret = btrfs_read_roots(fs_info, tree_root);
2905         if (ret)
2906                 goto recovery_tree_root;
2907
2908         fs_info->generation = generation;
2909         fs_info->last_trans_committed = generation;
2910
2911         ret = btrfs_recover_balance(fs_info);
2912         if (ret) {
2913                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2914                 goto fail_block_groups;
2915         }
2916
2917         ret = btrfs_init_dev_stats(fs_info);
2918         if (ret) {
2919                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2920                        ret);
2921                 goto fail_block_groups;
2922         }
2923
2924         ret = btrfs_init_dev_replace(fs_info);
2925         if (ret) {
2926                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2927                 goto fail_block_groups;
2928         }
2929
2930         btrfs_close_extra_devices(fs_devices, 1);
2931
2932         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2933         if (ret) {
2934                 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2935                 goto fail_block_groups;
2936         }
2937
2938         ret = btrfs_sysfs_add_device(fs_devices);
2939         if (ret) {
2940                 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2941                 goto fail_fsdev_sysfs;
2942         }
2943
2944         ret = btrfs_sysfs_add_mounted(fs_info);
2945         if (ret) {
2946                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2947                 goto fail_fsdev_sysfs;
2948         }
2949
2950         ret = btrfs_init_space_info(fs_info);
2951         if (ret) {
2952                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2953                 goto fail_sysfs;
2954         }
2955
2956         ret = btrfs_read_block_groups(fs_info->extent_root);
2957         if (ret) {
2958                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2959                 goto fail_sysfs;
2960         }
2961         fs_info->num_tolerated_disk_barrier_failures =
2962                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2963         if (fs_info->fs_devices->missing_devices >
2964              fs_info->num_tolerated_disk_barrier_failures &&
2965             !(sb->s_flags & MS_RDONLY)) {
2966                 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2967                         fs_info->fs_devices->missing_devices,
2968                         fs_info->num_tolerated_disk_barrier_failures);
2969                 goto fail_sysfs;
2970         }
2971
2972         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2973                                                "btrfs-cleaner");
2974         if (IS_ERR(fs_info->cleaner_kthread))
2975                 goto fail_sysfs;
2976
2977         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2978                                                    tree_root,
2979                                                    "btrfs-transaction");
2980         if (IS_ERR(fs_info->transaction_kthread))
2981                 goto fail_cleaner;
2982
2983         if (!btrfs_test_opt(tree_root, SSD) &&
2984             !btrfs_test_opt(tree_root, NOSSD) &&
2985             !fs_info->fs_devices->rotating) {
2986                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2987                        "mode\n");
2988                 btrfs_set_opt(fs_info->mount_opt, SSD);
2989         }
2990
2991         /*
2992          * Mount does not set all options immediatelly, we can do it now and do
2993          * not have to wait for transaction commit
2994          */
2995         btrfs_apply_pending_changes(fs_info);
2996
2997 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2998         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2999                 ret = btrfsic_mount(tree_root, fs_devices,
3000                                     btrfs_test_opt(tree_root,
3001                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3002                                     1 : 0,
3003                                     fs_info->check_integrity_print_mask);
3004                 if (ret)
3005                         printk(KERN_WARNING "BTRFS: failed to initialize"
3006                                " integrity check module %s\n", sb->s_id);
3007         }
3008 #endif
3009         ret = btrfs_read_qgroup_config(fs_info);
3010         if (ret)
3011                 goto fail_trans_kthread;
3012
3013         /* do not make disk changes in broken FS */
3014         if (btrfs_super_log_root(disk_super) != 0) {
3015                 ret = btrfs_replay_log(fs_info, fs_devices);
3016                 if (ret) {
3017                         err = ret;
3018                         goto fail_qgroup;
3019                 }
3020         }
3021
3022         ret = btrfs_find_orphan_roots(tree_root);
3023         if (ret)
3024                 goto fail_qgroup;
3025
3026         if (!(sb->s_flags & MS_RDONLY)) {
3027                 ret = btrfs_cleanup_fs_roots(fs_info);
3028                 if (ret)
3029                         goto fail_qgroup;
3030
3031                 mutex_lock(&fs_info->cleaner_mutex);
3032                 ret = btrfs_recover_relocation(tree_root);
3033                 mutex_unlock(&fs_info->cleaner_mutex);
3034                 if (ret < 0) {
3035                         printk(KERN_WARNING
3036                                "BTRFS: failed to recover relocation\n");
3037                         err = -EINVAL;
3038                         goto fail_qgroup;
3039                 }
3040         }
3041
3042         location.objectid = BTRFS_FS_TREE_OBJECTID;
3043         location.type = BTRFS_ROOT_ITEM_KEY;
3044         location.offset = 0;
3045
3046         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3047         if (IS_ERR(fs_info->fs_root)) {
3048                 err = PTR_ERR(fs_info->fs_root);
3049                 goto fail_qgroup;
3050         }
3051
3052         if (sb->s_flags & MS_RDONLY)
3053                 return 0;
3054
3055         down_read(&fs_info->cleanup_work_sem);
3056         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3057             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3058                 up_read(&fs_info->cleanup_work_sem);
3059                 close_ctree(tree_root);
3060                 return ret;
3061         }
3062         up_read(&fs_info->cleanup_work_sem);
3063
3064         ret = btrfs_resume_balance_async(fs_info);
3065         if (ret) {
3066                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3067                 close_ctree(tree_root);
3068                 return ret;
3069         }
3070
3071         ret = btrfs_resume_dev_replace_async(fs_info);
3072         if (ret) {
3073                 pr_warn("BTRFS: failed to resume dev_replace\n");
3074                 close_ctree(tree_root);
3075                 return ret;
3076         }
3077
3078         btrfs_qgroup_rescan_resume(fs_info);
3079
3080         if (!fs_info->uuid_root) {
3081                 pr_info("BTRFS: creating UUID tree\n");
3082                 ret = btrfs_create_uuid_tree(fs_info);
3083                 if (ret) {
3084                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3085                                 ret);
3086                         close_ctree(tree_root);
3087                         return ret;
3088                 }
3089         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3090                    fs_info->generation !=
3091                                 btrfs_super_uuid_tree_generation(disk_super)) {
3092                 pr_info("BTRFS: checking UUID tree\n");
3093                 ret = btrfs_check_uuid_tree(fs_info);
3094                 if (ret) {
3095                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3096                                 ret);
3097                         close_ctree(tree_root);
3098                         return ret;
3099                 }
3100         } else {
3101                 fs_info->update_uuid_tree_gen = 1;
3102         }
3103
3104         fs_info->open = 1;
3105
3106         return 0;
3107
3108 fail_qgroup:
3109         btrfs_free_qgroup_config(fs_info);
3110 fail_trans_kthread:
3111         kthread_stop(fs_info->transaction_kthread);
3112         btrfs_cleanup_transaction(fs_info->tree_root);
3113         btrfs_free_fs_roots(fs_info);
3114 fail_cleaner:
3115         kthread_stop(fs_info->cleaner_kthread);
3116
3117         /*
3118          * make sure we're done with the btree inode before we stop our
3119          * kthreads
3120          */
3121         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3122
3123 fail_sysfs:
3124         btrfs_sysfs_remove_mounted(fs_info);
3125
3126 fail_fsdev_sysfs:
3127         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3128
3129 fail_block_groups:
3130         btrfs_put_block_group_cache(fs_info);
3131         btrfs_free_block_groups(fs_info);
3132
3133 fail_tree_roots:
3134         free_root_pointers(fs_info, 1);
3135         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3136
3137 fail_sb_buffer:
3138         btrfs_stop_all_workers(fs_info);
3139 fail_alloc:
3140 fail_iput:
3141         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3142
3143         iput(fs_info->btree_inode);
3144 fail_bio_counter:
3145         percpu_counter_destroy(&fs_info->bio_counter);
3146 fail_delalloc_bytes:
3147         percpu_counter_destroy(&fs_info->delalloc_bytes);
3148 fail_dirty_metadata_bytes:
3149         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3150 fail_bdi:
3151         bdi_destroy(&fs_info->bdi);
3152 fail_srcu:
3153         cleanup_srcu_struct(&fs_info->subvol_srcu);
3154 fail:
3155         btrfs_free_stripe_hash_table(fs_info);
3156         btrfs_close_devices(fs_info->fs_devices);
3157         return err;
3158
3159 recovery_tree_root:
3160         if (!btrfs_test_opt(tree_root, RECOVERY))
3161                 goto fail_tree_roots;
3162
3163         free_root_pointers(fs_info, 0);
3164
3165         /* don't use the log in recovery mode, it won't be valid */
3166         btrfs_set_super_log_root(disk_super, 0);
3167
3168         /* we can't trust the free space cache either */
3169         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3170
3171         ret = next_root_backup(fs_info, fs_info->super_copy,
3172                                &num_backups_tried, &backup_index);
3173         if (ret == -1)
3174                 goto fail_block_groups;
3175         goto retry_root_backup;
3176 }
3177
3178 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3179 {
3180         if (uptodate) {
3181                 set_buffer_uptodate(bh);
3182         } else {
3183                 struct btrfs_device *device = (struct btrfs_device *)
3184                         bh->b_private;
3185
3186                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3187                                 "lost page write due to IO error on %s",
3188                                           rcu_str_deref(device->name));
3189                 /* note, we dont' set_buffer_write_io_error because we have
3190                  * our own ways of dealing with the IO errors
3191                  */
3192                 clear_buffer_uptodate(bh);
3193                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3194         }
3195         unlock_buffer(bh);
3196         put_bh(bh);
3197 }
3198
3199 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3200                         struct buffer_head **bh_ret)
3201 {
3202         struct buffer_head *bh;
3203         struct btrfs_super_block *super;
3204         u64 bytenr;
3205
3206         bytenr = btrfs_sb_offset(copy_num);
3207         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3208                 return -EINVAL;
3209
3210         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3211         /*
3212          * If we fail to read from the underlying devices, as of now
3213          * the best option we have is to mark it EIO.
3214          */
3215         if (!bh)
3216                 return -EIO;
3217
3218         super = (struct btrfs_super_block *)bh->b_data;
3219         if (btrfs_super_bytenr(super) != bytenr ||
3220                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3221                 brelse(bh);
3222                 return -EINVAL;
3223         }
3224
3225         *bh_ret = bh;
3226         return 0;
3227 }
3228
3229
3230 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3231 {
3232         struct buffer_head *bh;
3233         struct buffer_head *latest = NULL;
3234         struct btrfs_super_block *super;
3235         int i;
3236         u64 transid = 0;
3237         int ret = -EINVAL;
3238
3239         /* we would like to check all the supers, but that would make
3240          * a btrfs mount succeed after a mkfs from a different FS.
3241          * So, we need to add a special mount option to scan for
3242          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3243          */
3244         for (i = 0; i < 1; i++) {
3245                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3246                 if (ret)
3247                         continue;
3248
3249                 super = (struct btrfs_super_block *)bh->b_data;
3250
3251                 if (!latest || btrfs_super_generation(super) > transid) {
3252                         brelse(latest);
3253                         latest = bh;
3254                         transid = btrfs_super_generation(super);
3255                 } else {
3256                         brelse(bh);
3257                 }
3258         }
3259
3260         if (!latest)
3261                 return ERR_PTR(ret);
3262
3263         return latest;
3264 }
3265
3266 /*
3267  * this should be called twice, once with wait == 0 and
3268  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3269  * we write are pinned.
3270  *
3271  * They are released when wait == 1 is done.
3272  * max_mirrors must be the same for both runs, and it indicates how
3273  * many supers on this one device should be written.
3274  *
3275  * max_mirrors == 0 means to write them all.
3276  */
3277 static int write_dev_supers(struct btrfs_device *device,
3278                             struct btrfs_super_block *sb,
3279                             int do_barriers, int wait, int max_mirrors)
3280 {
3281         struct buffer_head *bh;
3282         int i;
3283         int ret;
3284         int errors = 0;
3285         u32 crc;
3286         u64 bytenr;
3287
3288         if (max_mirrors == 0)
3289                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3290
3291         for (i = 0; i < max_mirrors; i++) {
3292                 bytenr = btrfs_sb_offset(i);
3293                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3294                     device->commit_total_bytes)
3295                         break;
3296
3297                 if (wait) {
3298                         bh = __find_get_block(device->bdev, bytenr / 4096,
3299                                               BTRFS_SUPER_INFO_SIZE);
3300                         if (!bh) {
3301                                 errors++;
3302                                 continue;
3303                         }
3304                         wait_on_buffer(bh);
3305                         if (!buffer_uptodate(bh))
3306                                 errors++;
3307
3308                         /* drop our reference */
3309                         brelse(bh);
3310
3311                         /* drop the reference from the wait == 0 run */
3312                         brelse(bh);
3313                         continue;
3314                 } else {
3315                         btrfs_set_super_bytenr(sb, bytenr);
3316
3317                         crc = ~(u32)0;
3318                         crc = btrfs_csum_data((char *)sb +
3319                                               BTRFS_CSUM_SIZE, crc,
3320                                               BTRFS_SUPER_INFO_SIZE -
3321                                               BTRFS_CSUM_SIZE);
3322                         btrfs_csum_final(crc, sb->csum);
3323
3324                         /*
3325                          * one reference for us, and we leave it for the
3326                          * caller
3327                          */
3328                         bh = __getblk(device->bdev, bytenr / 4096,
3329                                       BTRFS_SUPER_INFO_SIZE);
3330                         if (!bh) {
3331                                 btrfs_err(device->dev_root->fs_info,
3332                                     "couldn't get super buffer head for bytenr %llu",
3333                                     bytenr);
3334                                 errors++;
3335                                 continue;
3336                         }
3337
3338                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3339
3340                         /* one reference for submit_bh */
3341                         get_bh(bh);
3342
3343                         set_buffer_uptodate(bh);
3344                         lock_buffer(bh);
3345                         bh->b_end_io = btrfs_end_buffer_write_sync;
3346                         bh->b_private = device;
3347                 }
3348
3349                 /*
3350                  * we fua the first super.  The others we allow
3351                  * to go down lazy.
3352                  */
3353                 if (i == 0)
3354                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3355                 else
3356                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3357                 if (ret)
3358                         errors++;
3359         }
3360         return errors < i ? 0 : -1;
3361 }
3362
3363 /*
3364  * endio for the write_dev_flush, this will wake anyone waiting
3365  * for the barrier when it is done
3366  */
3367 static void btrfs_end_empty_barrier(struct bio *bio)
3368 {
3369         if (bio->bi_private)
3370                 complete(bio->bi_private);
3371         bio_put(bio);
3372 }
3373
3374 /*
3375  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3376  * sent down.  With wait == 1, it waits for the previous flush.
3377  *
3378  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3379  * capable
3380  */
3381 static int write_dev_flush(struct btrfs_device *device, int wait)
3382 {
3383         struct bio *bio;
3384         int ret = 0;
3385
3386         if (device->nobarriers)
3387                 return 0;
3388
3389         if (wait) {
3390                 bio = device->flush_bio;
3391                 if (!bio)
3392                         return 0;
3393
3394                 wait_for_completion(&device->flush_wait);
3395
3396                 if (bio->bi_error) {
3397                         ret = bio->bi_error;
3398                         btrfs_dev_stat_inc_and_print(device,
3399                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3400                 }
3401
3402                 /* drop the reference from the wait == 0 run */
3403                 bio_put(bio);
3404                 device->flush_bio = NULL;
3405
3406                 return ret;
3407         }
3408
3409         /*
3410          * one reference for us, and we leave it for the
3411          * caller
3412          */
3413         device->flush_bio = NULL;
3414         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3415         if (!bio)
3416                 return -ENOMEM;
3417
3418         bio->bi_end_io = btrfs_end_empty_barrier;
3419         bio->bi_bdev = device->bdev;
3420         init_completion(&device->flush_wait);
3421         bio->bi_private = &device->flush_wait;
3422         device->flush_bio = bio;
3423
3424         bio_get(bio);
3425         btrfsic_submit_bio(WRITE_FLUSH, bio);
3426
3427         return 0;
3428 }
3429
3430 /*
3431  * send an empty flush down to each device in parallel,
3432  * then wait for them
3433  */
3434 static int barrier_all_devices(struct btrfs_fs_info *info)
3435 {
3436         struct list_head *head;
3437         struct btrfs_device *dev;
3438         int errors_send = 0;
3439         int errors_wait = 0;
3440         int ret;
3441
3442         /* send down all the barriers */
3443         head = &info->fs_devices->devices;
3444         list_for_each_entry_rcu(dev, head, dev_list) {
3445                 if (dev->missing)
3446                         continue;
3447                 if (!dev->bdev) {
3448                         errors_send++;
3449                         continue;
3450                 }
3451                 if (!dev->in_fs_metadata || !dev->writeable)
3452                         continue;
3453
3454                 ret = write_dev_flush(dev, 0);
3455                 if (ret)
3456                         errors_send++;
3457         }
3458
3459         /* wait for all the barriers */
3460         list_for_each_entry_rcu(dev, head, dev_list) {
3461                 if (dev->missing)
3462                         continue;
3463                 if (!dev->bdev) {
3464                         errors_wait++;
3465                         continue;
3466                 }
3467                 if (!dev->in_fs_metadata || !dev->writeable)
3468                         continue;
3469
3470                 ret = write_dev_flush(dev, 1);
3471                 if (ret)
3472                         errors_wait++;
3473         }
3474         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3475             errors_wait > info->num_tolerated_disk_barrier_failures)
3476                 return -EIO;
3477         return 0;
3478 }
3479
3480 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3481 {
3482         int raid_type;
3483         int min_tolerated = INT_MAX;
3484
3485         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3486             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3487                 min_tolerated = min(min_tolerated,
3488                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3489                                     tolerated_failures);
3490
3491         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3492                 if (raid_type == BTRFS_RAID_SINGLE)
3493                         continue;
3494                 if (!(flags & btrfs_raid_group[raid_type]))
3495                         continue;
3496                 min_tolerated = min(min_tolerated,
3497                                     btrfs_raid_array[raid_type].
3498                                     tolerated_failures);
3499         }
3500
3501         if (min_tolerated == INT_MAX) {
3502                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3503                 min_tolerated = 0;
3504         }
3505
3506         return min_tolerated;
3507 }
3508
3509 int btrfs_calc_num_tolerated_disk_barrier_failures(
3510         struct btrfs_fs_info *fs_info)
3511 {
3512         struct btrfs_ioctl_space_info space;
3513         struct btrfs_space_info *sinfo;
3514         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3515                        BTRFS_BLOCK_GROUP_SYSTEM,
3516                        BTRFS_BLOCK_GROUP_METADATA,
3517                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3518         int i;
3519         int c;
3520         int num_tolerated_disk_barrier_failures =
3521                 (int)fs_info->fs_devices->num_devices;
3522
3523         for (i = 0; i < ARRAY_SIZE(types); i++) {
3524                 struct btrfs_space_info *tmp;
3525
3526                 sinfo = NULL;
3527                 rcu_read_lock();
3528                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3529                         if (tmp->flags == types[i]) {
3530                                 sinfo = tmp;
3531                                 break;
3532                         }
3533                 }
3534                 rcu_read_unlock();
3535
3536                 if (!sinfo)
3537                         continue;
3538
3539                 down_read(&sinfo->groups_sem);
3540                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3541                         u64 flags;
3542
3543                         if (list_empty(&sinfo->block_groups[c]))
3544                                 continue;
3545
3546                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3547                                                    &space);
3548                         if (space.total_bytes == 0 || space.used_bytes == 0)
3549                                 continue;
3550                         flags = space.flags;
3551
3552                         num_tolerated_disk_barrier_failures = min(
3553                                 num_tolerated_disk_barrier_failures,
3554                                 btrfs_get_num_tolerated_disk_barrier_failures(
3555                                         flags));
3556                 }
3557                 up_read(&sinfo->groups_sem);
3558         }
3559
3560         return num_tolerated_disk_barrier_failures;
3561 }
3562
3563 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3564 {
3565         struct list_head *head;
3566         struct btrfs_device *dev;
3567         struct btrfs_super_block *sb;
3568         struct btrfs_dev_item *dev_item;
3569         int ret;
3570         int do_barriers;
3571         int max_errors;
3572         int total_errors = 0;
3573         u64 flags;
3574
3575         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3576         backup_super_roots(root->fs_info);
3577
3578         sb = root->fs_info->super_for_commit;
3579         dev_item = &sb->dev_item;
3580
3581         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3582         head = &root->fs_info->fs_devices->devices;
3583         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3584
3585         if (do_barriers) {
3586                 ret = barrier_all_devices(root->fs_info);
3587                 if (ret) {
3588                         mutex_unlock(
3589                                 &root->fs_info->fs_devices->device_list_mutex);
3590                         btrfs_std_error(root->fs_info, ret,
3591                                     "errors while submitting device barriers.");
3592                         return ret;
3593                 }
3594         }
3595
3596         list_for_each_entry_rcu(dev, head, dev_list) {
3597                 if (!dev->bdev) {
3598                         total_errors++;
3599                         continue;
3600                 }
3601                 if (!dev->in_fs_metadata || !dev->writeable)
3602                         continue;
3603
3604                 btrfs_set_stack_device_generation(dev_item, 0);
3605                 btrfs_set_stack_device_type(dev_item, dev->type);
3606                 btrfs_set_stack_device_id(dev_item, dev->devid);
3607                 btrfs_set_stack_device_total_bytes(dev_item,
3608                                                    dev->commit_total_bytes);
3609                 btrfs_set_stack_device_bytes_used(dev_item,
3610                                                   dev->commit_bytes_used);
3611                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3612                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3613                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3614                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3615                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3616
3617                 flags = btrfs_super_flags(sb);
3618                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3619
3620                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3621                 if (ret)
3622                         total_errors++;
3623         }
3624         if (total_errors > max_errors) {
3625                 btrfs_err(root->fs_info, "%d errors while writing supers",
3626                        total_errors);
3627                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3628
3629                 /* FUA is masked off if unsupported and can't be the reason */
3630                 btrfs_std_error(root->fs_info, -EIO,
3631                             "%d errors while writing supers", total_errors);
3632                 return -EIO;
3633         }
3634
3635         total_errors = 0;
3636         list_for_each_entry_rcu(dev, head, dev_list) {
3637                 if (!dev->bdev)
3638                         continue;
3639                 if (!dev->in_fs_metadata || !dev->writeable)
3640                         continue;
3641
3642                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3643                 if (ret)
3644                         total_errors++;
3645         }
3646         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3647         if (total_errors > max_errors) {
3648                 btrfs_std_error(root->fs_info, -EIO,
3649                             "%d errors while writing supers", total_errors);
3650                 return -EIO;
3651         }
3652         return 0;
3653 }
3654
3655 int write_ctree_super(struct btrfs_trans_handle *trans,
3656                       struct btrfs_root *root, int max_mirrors)
3657 {
3658         return write_all_supers(root, max_mirrors);
3659 }
3660
3661 /* Drop a fs root from the radix tree and free it. */
3662 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3663                                   struct btrfs_root *root)
3664 {
3665         spin_lock(&fs_info->fs_roots_radix_lock);
3666         radix_tree_delete(&fs_info->fs_roots_radix,
3667                           (unsigned long)root->root_key.objectid);
3668         spin_unlock(&fs_info->fs_roots_radix_lock);
3669
3670         if (btrfs_root_refs(&root->root_item) == 0)
3671                 synchronize_srcu(&fs_info->subvol_srcu);
3672
3673         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3674                 btrfs_free_log(NULL, root);
3675
3676         if (root->free_ino_pinned)
3677                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3678         if (root->free_ino_ctl)
3679                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3680         free_fs_root(root);
3681 }
3682
3683 static void free_fs_root(struct btrfs_root *root)
3684 {
3685         iput(root->ino_cache_inode);
3686         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3687         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3688         root->orphan_block_rsv = NULL;
3689         if (root->anon_dev)
3690                 free_anon_bdev(root->anon_dev);
3691         if (root->subv_writers)
3692                 btrfs_free_subvolume_writers(root->subv_writers);
3693         free_extent_buffer(root->node);
3694         free_extent_buffer(root->commit_root);
3695         kfree(root->free_ino_ctl);
3696         kfree(root->free_ino_pinned);
3697         kfree(root->name);
3698         btrfs_put_fs_root(root);
3699 }
3700
3701 void btrfs_free_fs_root(struct btrfs_root *root)
3702 {
3703         free_fs_root(root);
3704 }
3705
3706 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3707 {
3708         u64 root_objectid = 0;
3709         struct btrfs_root *gang[8];
3710         int i = 0;
3711         int err = 0;
3712         unsigned int ret = 0;
3713         int index;
3714
3715         while (1) {
3716                 index = srcu_read_lock(&fs_info->subvol_srcu);
3717                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3718                                              (void **)gang, root_objectid,
3719                                              ARRAY_SIZE(gang));
3720                 if (!ret) {
3721                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3722                         break;
3723                 }
3724                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3725
3726                 for (i = 0; i < ret; i++) {
3727                         /* Avoid to grab roots in dead_roots */
3728                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3729                                 gang[i] = NULL;
3730                                 continue;
3731                         }
3732                         /* grab all the search result for later use */
3733                         gang[i] = btrfs_grab_fs_root(gang[i]);
3734                 }
3735                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3736
3737                 for (i = 0; i < ret; i++) {
3738                         if (!gang[i])
3739                                 continue;
3740                         root_objectid = gang[i]->root_key.objectid;
3741                         err = btrfs_orphan_cleanup(gang[i]);
3742                         if (err)
3743                                 break;
3744                         btrfs_put_fs_root(gang[i]);
3745                 }
3746                 root_objectid++;
3747         }
3748
3749         /* release the uncleaned roots due to error */
3750         for (; i < ret; i++) {
3751                 if (gang[i])
3752                         btrfs_put_fs_root(gang[i]);
3753         }
3754         return err;
3755 }
3756
3757 int btrfs_commit_super(struct btrfs_root *root)
3758 {
3759         struct btrfs_trans_handle *trans;
3760
3761         mutex_lock(&root->fs_info->cleaner_mutex);
3762         btrfs_run_delayed_iputs(root);
3763         mutex_unlock(&root->fs_info->cleaner_mutex);
3764         wake_up_process(root->fs_info->cleaner_kthread);
3765
3766         /* wait until ongoing cleanup work done */
3767         down_write(&root->fs_info->cleanup_work_sem);
3768         up_write(&root->fs_info->cleanup_work_sem);
3769
3770         trans = btrfs_join_transaction(root);
3771         if (IS_ERR(trans))
3772                 return PTR_ERR(trans);
3773         return btrfs_commit_transaction(trans, root);
3774 }
3775
3776 void close_ctree(struct btrfs_root *root)
3777 {
3778         struct btrfs_fs_info *fs_info = root->fs_info;
3779         int ret;
3780
3781         fs_info->closing = 1;
3782         smp_mb();
3783
3784         /* wait for the qgroup rescan worker to stop */
3785         btrfs_qgroup_wait_for_completion(fs_info);
3786
3787         /* wait for the uuid_scan task to finish */
3788         down(&fs_info->uuid_tree_rescan_sem);
3789         /* avoid complains from lockdep et al., set sem back to initial state */
3790         up(&fs_info->uuid_tree_rescan_sem);
3791
3792         /* pause restriper - we want to resume on mount */
3793         btrfs_pause_balance(fs_info);
3794
3795         btrfs_dev_replace_suspend_for_unmount(fs_info);
3796
3797         btrfs_scrub_cancel(fs_info);
3798
3799         /* wait for any defraggers to finish */
3800         wait_event(fs_info->transaction_wait,
3801                    (atomic_read(&fs_info->defrag_running) == 0));
3802
3803         /* clear out the rbtree of defraggable inodes */
3804         btrfs_cleanup_defrag_inodes(fs_info);
3805
3806         cancel_work_sync(&fs_info->async_reclaim_work);
3807
3808         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3809                 /*
3810                  * If the cleaner thread is stopped and there are
3811                  * block groups queued for removal, the deletion will be
3812                  * skipped when we quit the cleaner thread.
3813                  */
3814                 btrfs_delete_unused_bgs(root->fs_info);
3815
3816                 ret = btrfs_commit_super(root);
3817                 if (ret)
3818                         btrfs_err(fs_info, "commit super ret %d", ret);
3819         }
3820
3821         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3822                 btrfs_error_commit_super(root);
3823
3824         kthread_stop(fs_info->transaction_kthread);
3825         kthread_stop(fs_info->cleaner_kthread);
3826
3827         fs_info->closing = 2;
3828         smp_mb();
3829
3830         btrfs_free_qgroup_config(fs_info);
3831
3832         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3833                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3834                        percpu_counter_sum(&fs_info->delalloc_bytes));
3835         }
3836
3837         btrfs_sysfs_remove_mounted(fs_info);
3838         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3839
3840         btrfs_free_fs_roots(fs_info);
3841
3842         btrfs_put_block_group_cache(fs_info);
3843
3844         btrfs_free_block_groups(fs_info);
3845
3846         /*
3847          * we must make sure there is not any read request to
3848          * submit after we stopping all workers.
3849          */
3850         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3851         btrfs_stop_all_workers(fs_info);
3852
3853         fs_info->open = 0;
3854         free_root_pointers(fs_info, 1);
3855
3856         iput(fs_info->btree_inode);
3857
3858 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3859         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3860                 btrfsic_unmount(root, fs_info->fs_devices);
3861 #endif
3862
3863         btrfs_close_devices(fs_info->fs_devices);
3864         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3865
3866         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3867         percpu_counter_destroy(&fs_info->delalloc_bytes);
3868         percpu_counter_destroy(&fs_info->bio_counter);
3869         bdi_destroy(&fs_info->bdi);
3870         cleanup_srcu_struct(&fs_info->subvol_srcu);
3871
3872         btrfs_free_stripe_hash_table(fs_info);
3873
3874         __btrfs_free_block_rsv(root->orphan_block_rsv);
3875         root->orphan_block_rsv = NULL;
3876
3877         lock_chunks(root);
3878         while (!list_empty(&fs_info->pinned_chunks)) {
3879                 struct extent_map *em;
3880
3881                 em = list_first_entry(&fs_info->pinned_chunks,
3882                                       struct extent_map, list);
3883                 list_del_init(&em->list);
3884                 free_extent_map(em);
3885         }
3886         unlock_chunks(root);
3887 }
3888
3889 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3890                           int atomic)
3891 {
3892         int ret;
3893         struct inode *btree_inode = buf->pages[0]->mapping->host;
3894
3895         ret = extent_buffer_uptodate(buf);
3896         if (!ret)
3897                 return ret;
3898
3899         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3900                                     parent_transid, atomic);
3901         if (ret == -EAGAIN)
3902                 return ret;
3903         return !ret;
3904 }
3905
3906 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3907 {
3908         return set_extent_buffer_uptodate(buf);
3909 }
3910
3911 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3912 {
3913         struct btrfs_root *root;
3914         u64 transid = btrfs_header_generation(buf);
3915         int was_dirty;
3916
3917 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3918         /*
3919          * This is a fast path so only do this check if we have sanity tests
3920          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3921          * outside of the sanity tests.
3922          */
3923         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3924                 return;
3925 #endif
3926         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3927         btrfs_assert_tree_locked(buf);
3928         if (transid != root->fs_info->generation)
3929                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3930                        "found %llu running %llu\n",
3931                         buf->start, transid, root->fs_info->generation);
3932         was_dirty = set_extent_buffer_dirty(buf);
3933         if (!was_dirty)
3934                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3935                                      buf->len,
3936                                      root->fs_info->dirty_metadata_batch);
3937 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3938         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3939                 btrfs_print_leaf(root, buf);
3940                 ASSERT(0);
3941         }
3942 #endif
3943 }
3944
3945 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3946                                         int flush_delayed)
3947 {
3948         /*
3949          * looks as though older kernels can get into trouble with
3950          * this code, they end up stuck in balance_dirty_pages forever
3951          */
3952         int ret;
3953
3954         if (current->flags & PF_MEMALLOC)
3955                 return;
3956
3957         if (flush_delayed)
3958                 btrfs_balance_delayed_items(root);
3959
3960         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3961                                      BTRFS_DIRTY_METADATA_THRESH);
3962         if (ret > 0) {
3963                 balance_dirty_pages_ratelimited(
3964                                    root->fs_info->btree_inode->i_mapping);
3965         }
3966         return;
3967 }
3968
3969 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3970 {
3971         __btrfs_btree_balance_dirty(root, 1);
3972 }
3973
3974 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3975 {
3976         __btrfs_btree_balance_dirty(root, 0);
3977 }
3978
3979 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3980 {
3981         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3982         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3983 }
3984
3985 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3986                               int read_only)
3987 {
3988         struct btrfs_super_block *sb = fs_info->super_copy;
3989         int ret = 0;
3990
3991         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3992                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3993                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3994                 ret = -EINVAL;
3995         }
3996         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3997                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3998                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3999                 ret = -EINVAL;
4000         }
4001         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4002                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4003                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4004                 ret = -EINVAL;
4005         }
4006
4007         /*
4008          * The common minimum, we don't know if we can trust the nodesize/sectorsize
4009          * items yet, they'll be verified later. Issue just a warning.
4010          */
4011         if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
4012                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4013                                 btrfs_super_root(sb));
4014         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
4015                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4016                                 btrfs_super_chunk_root(sb));
4017         if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
4018                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4019                                 btrfs_super_log_root(sb));
4020
4021         /*
4022          * Check the lower bound, the alignment and other constraints are
4023          * checked later.
4024          */
4025         if (btrfs_super_nodesize(sb) < 4096) {
4026                 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
4027                                 btrfs_super_nodesize(sb));
4028                 ret = -EINVAL;
4029         }
4030         if (btrfs_super_sectorsize(sb) < 4096) {
4031                 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
4032                                 btrfs_super_sectorsize(sb));
4033                 ret = -EINVAL;
4034         }
4035
4036         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4037                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4038                                 fs_info->fsid, sb->dev_item.fsid);
4039                 ret = -EINVAL;
4040         }
4041
4042         /*
4043          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4044          * done later
4045          */
4046         if (btrfs_super_num_devices(sb) > (1UL << 31))
4047                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4048                                 btrfs_super_num_devices(sb));
4049         if (btrfs_super_num_devices(sb) == 0) {
4050                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4051                 ret = -EINVAL;
4052         }
4053
4054         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4055                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4056                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4057                 ret = -EINVAL;
4058         }
4059
4060         /*
4061          * Obvious sys_chunk_array corruptions, it must hold at least one key
4062          * and one chunk
4063          */
4064         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4065                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4066                                 btrfs_super_sys_array_size(sb),
4067                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4068                 ret = -EINVAL;
4069         }
4070         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4071                         + sizeof(struct btrfs_chunk)) {
4072                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4073                                 btrfs_super_sys_array_size(sb),
4074                                 sizeof(struct btrfs_disk_key)
4075                                 + sizeof(struct btrfs_chunk));
4076                 ret = -EINVAL;
4077         }
4078
4079         /*
4080          * The generation is a global counter, we'll trust it more than the others
4081          * but it's still possible that it's the one that's wrong.
4082          */
4083         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4084                 printk(KERN_WARNING
4085                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4086                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4087         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4088             && btrfs_super_cache_generation(sb) != (u64)-1)
4089                 printk(KERN_WARNING
4090                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4091                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4092
4093         return ret;
4094 }
4095
4096 static void btrfs_error_commit_super(struct btrfs_root *root)
4097 {
4098         mutex_lock(&root->fs_info->cleaner_mutex);
4099         btrfs_run_delayed_iputs(root);
4100         mutex_unlock(&root->fs_info->cleaner_mutex);
4101
4102         down_write(&root->fs_info->cleanup_work_sem);
4103         up_write(&root->fs_info->cleanup_work_sem);
4104
4105         /* cleanup FS via transaction */
4106         btrfs_cleanup_transaction(root);
4107 }
4108
4109 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4110 {
4111         struct btrfs_ordered_extent *ordered;
4112
4113         spin_lock(&root->ordered_extent_lock);
4114         /*
4115          * This will just short circuit the ordered completion stuff which will
4116          * make sure the ordered extent gets properly cleaned up.
4117          */
4118         list_for_each_entry(ordered, &root->ordered_extents,
4119                             root_extent_list)
4120                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4121         spin_unlock(&root->ordered_extent_lock);
4122 }
4123
4124 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4125 {
4126         struct btrfs_root *root;
4127         struct list_head splice;
4128
4129         INIT_LIST_HEAD(&splice);
4130
4131         spin_lock(&fs_info->ordered_root_lock);
4132         list_splice_init(&fs_info->ordered_roots, &splice);
4133         while (!list_empty(&splice)) {
4134                 root = list_first_entry(&splice, struct btrfs_root,
4135                                         ordered_root);
4136                 list_move_tail(&root->ordered_root,
4137                                &fs_info->ordered_roots);
4138
4139                 spin_unlock(&fs_info->ordered_root_lock);
4140                 btrfs_destroy_ordered_extents(root);
4141
4142                 cond_resched();
4143                 spin_lock(&fs_info->ordered_root_lock);
4144         }
4145         spin_unlock(&fs_info->ordered_root_lock);
4146 }
4147
4148 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4149                                       struct btrfs_root *root)
4150 {
4151         struct rb_node *node;
4152         struct btrfs_delayed_ref_root *delayed_refs;
4153         struct btrfs_delayed_ref_node *ref;
4154         int ret = 0;
4155
4156         delayed_refs = &trans->delayed_refs;
4157
4158         spin_lock(&delayed_refs->lock);
4159         if (atomic_read(&delayed_refs->num_entries) == 0) {
4160                 spin_unlock(&delayed_refs->lock);
4161                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4162                 return ret;
4163         }
4164
4165         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4166                 struct btrfs_delayed_ref_head *head;
4167                 struct btrfs_delayed_ref_node *tmp;
4168                 bool pin_bytes = false;
4169
4170                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4171                                 href_node);
4172                 if (!mutex_trylock(&head->mutex)) {
4173                         atomic_inc(&head->node.refs);
4174                         spin_unlock(&delayed_refs->lock);
4175
4176                         mutex_lock(&head->mutex);
4177                         mutex_unlock(&head->mutex);
4178                         btrfs_put_delayed_ref(&head->node);
4179                         spin_lock(&delayed_refs->lock);
4180                         continue;
4181                 }
4182                 spin_lock(&head->lock);
4183                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4184                                                  list) {
4185                         ref->in_tree = 0;
4186                         list_del(&ref->list);
4187                         atomic_dec(&delayed_refs->num_entries);
4188                         btrfs_put_delayed_ref(ref);
4189                 }
4190                 if (head->must_insert_reserved)
4191                         pin_bytes = true;
4192                 btrfs_free_delayed_extent_op(head->extent_op);
4193                 delayed_refs->num_heads--;
4194                 if (head->processing == 0)
4195                         delayed_refs->num_heads_ready--;
4196                 atomic_dec(&delayed_refs->num_entries);
4197                 head->node.in_tree = 0;
4198                 rb_erase(&head->href_node, &delayed_refs->href_root);
4199                 spin_unlock(&head->lock);
4200                 spin_unlock(&delayed_refs->lock);
4201                 mutex_unlock(&head->mutex);
4202
4203                 if (pin_bytes)
4204                         btrfs_pin_extent(root, head->node.bytenr,
4205                                          head->node.num_bytes, 1);
4206                 btrfs_put_delayed_ref(&head->node);
4207                 cond_resched();
4208                 spin_lock(&delayed_refs->lock);
4209         }
4210
4211         spin_unlock(&delayed_refs->lock);
4212
4213         return ret;
4214 }
4215
4216 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4217 {
4218         struct btrfs_inode *btrfs_inode;
4219         struct list_head splice;
4220
4221         INIT_LIST_HEAD(&splice);
4222
4223         spin_lock(&root->delalloc_lock);
4224         list_splice_init(&root->delalloc_inodes, &splice);
4225
4226         while (!list_empty(&splice)) {
4227                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4228                                                delalloc_inodes);
4229
4230                 list_del_init(&btrfs_inode->delalloc_inodes);
4231                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4232                           &btrfs_inode->runtime_flags);
4233                 spin_unlock(&root->delalloc_lock);
4234
4235                 btrfs_invalidate_inodes(btrfs_inode->root);
4236
4237                 spin_lock(&root->delalloc_lock);
4238         }
4239
4240         spin_unlock(&root->delalloc_lock);
4241 }
4242
4243 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4244 {
4245         struct btrfs_root *root;
4246         struct list_head splice;
4247
4248         INIT_LIST_HEAD(&splice);
4249
4250         spin_lock(&fs_info->delalloc_root_lock);
4251         list_splice_init(&fs_info->delalloc_roots, &splice);
4252         while (!list_empty(&splice)) {
4253                 root = list_first_entry(&splice, struct btrfs_root,
4254                                          delalloc_root);
4255                 list_del_init(&root->delalloc_root);
4256                 root = btrfs_grab_fs_root(root);
4257                 BUG_ON(!root);
4258                 spin_unlock(&fs_info->delalloc_root_lock);
4259
4260                 btrfs_destroy_delalloc_inodes(root);
4261                 btrfs_put_fs_root(root);
4262
4263                 spin_lock(&fs_info->delalloc_root_lock);
4264         }
4265         spin_unlock(&fs_info->delalloc_root_lock);
4266 }
4267
4268 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4269                                         struct extent_io_tree *dirty_pages,
4270                                         int mark)
4271 {
4272         int ret;
4273         struct extent_buffer *eb;
4274         u64 start = 0;
4275         u64 end;
4276
4277         while (1) {
4278                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4279                                             mark, NULL);
4280                 if (ret)
4281                         break;
4282
4283                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4284                 while (start <= end) {
4285                         eb = btrfs_find_tree_block(root->fs_info, start);
4286                         start += root->nodesize;
4287                         if (!eb)
4288                                 continue;
4289                         wait_on_extent_buffer_writeback(eb);
4290
4291                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4292                                                &eb->bflags))
4293                                 clear_extent_buffer_dirty(eb);
4294                         free_extent_buffer_stale(eb);
4295                 }
4296         }
4297
4298         return ret;
4299 }
4300
4301 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4302                                        struct extent_io_tree *pinned_extents)
4303 {
4304         struct extent_io_tree *unpin;
4305         u64 start;
4306         u64 end;
4307         int ret;
4308         bool loop = true;
4309
4310         unpin = pinned_extents;
4311 again:
4312         while (1) {
4313                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4314                                             EXTENT_DIRTY, NULL);
4315                 if (ret)
4316                         break;
4317
4318                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4319                 btrfs_error_unpin_extent_range(root, start, end);
4320                 cond_resched();
4321         }
4322
4323         if (loop) {
4324                 if (unpin == &root->fs_info->freed_extents[0])
4325                         unpin = &root->fs_info->freed_extents[1];
4326                 else
4327                         unpin = &root->fs_info->freed_extents[0];
4328                 loop = false;
4329                 goto again;
4330         }
4331
4332         return 0;
4333 }
4334
4335 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4336                                    struct btrfs_root *root)
4337 {
4338         btrfs_destroy_delayed_refs(cur_trans, root);
4339
4340         cur_trans->state = TRANS_STATE_COMMIT_START;
4341         wake_up(&root->fs_info->transaction_blocked_wait);
4342
4343         cur_trans->state = TRANS_STATE_UNBLOCKED;
4344         wake_up(&root->fs_info->transaction_wait);
4345
4346         btrfs_destroy_delayed_inodes(root);
4347         btrfs_assert_delayed_root_empty(root);
4348
4349         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4350                                      EXTENT_DIRTY);
4351         btrfs_destroy_pinned_extent(root,
4352                                     root->fs_info->pinned_extents);
4353
4354         cur_trans->state =TRANS_STATE_COMPLETED;
4355         wake_up(&cur_trans->commit_wait);
4356
4357         /*
4358         memset(cur_trans, 0, sizeof(*cur_trans));
4359         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4360         */
4361 }
4362
4363 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4364 {
4365         struct btrfs_transaction *t;
4366
4367         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4368
4369         spin_lock(&root->fs_info->trans_lock);
4370         while (!list_empty(&root->fs_info->trans_list)) {
4371                 t = list_first_entry(&root->fs_info->trans_list,
4372                                      struct btrfs_transaction, list);
4373                 if (t->state >= TRANS_STATE_COMMIT_START) {
4374                         atomic_inc(&t->use_count);
4375                         spin_unlock(&root->fs_info->trans_lock);
4376                         btrfs_wait_for_commit(root, t->transid);
4377                         btrfs_put_transaction(t);
4378                         spin_lock(&root->fs_info->trans_lock);
4379                         continue;
4380                 }
4381                 if (t == root->fs_info->running_transaction) {
4382                         t->state = TRANS_STATE_COMMIT_DOING;
4383                         spin_unlock(&root->fs_info->trans_lock);
4384                         /*
4385                          * We wait for 0 num_writers since we don't hold a trans
4386                          * handle open currently for this transaction.
4387                          */
4388                         wait_event(t->writer_wait,
4389                                    atomic_read(&t->num_writers) == 0);
4390                 } else {
4391                         spin_unlock(&root->fs_info->trans_lock);
4392                 }
4393                 btrfs_cleanup_one_transaction(t, root);
4394
4395                 spin_lock(&root->fs_info->trans_lock);
4396                 if (t == root->fs_info->running_transaction)
4397                         root->fs_info->running_transaction = NULL;
4398                 list_del_init(&t->list);
4399                 spin_unlock(&root->fs_info->trans_lock);
4400
4401                 btrfs_put_transaction(t);
4402                 trace_btrfs_transaction_commit(root);
4403                 spin_lock(&root->fs_info->trans_lock);
4404         }
4405         spin_unlock(&root->fs_info->trans_lock);
4406         btrfs_destroy_all_ordered_extents(root->fs_info);
4407         btrfs_destroy_delayed_inodes(root);
4408         btrfs_assert_delayed_root_empty(root);
4409         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4410         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4411         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4412
4413         return 0;
4414 }
4415
4416 static const struct extent_io_ops btree_extent_io_ops = {
4417         .readpage_end_io_hook = btree_readpage_end_io_hook,
4418         .readpage_io_failed_hook = btree_io_failed_hook,
4419         .submit_bio_hook = btree_submit_bio_hook,
4420         /* note we're sharing with inode.c for the merge bio hook */
4421         .merge_bio_hook = btrfs_merge_bio_hook,
4422 };