fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/slab.h>
  30 #include <linux/migrate.h>
  31 #include <linux/ratelimit.h>
  32 #include <linux/uuid.h>
  33 #include <linux/semaphore.h>
  34 #include <asm/unaligned.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "hash.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "volumes.h"
  41 #include "print-tree.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "free-space-tree.h"
  46 #include "inode-map.h"
  47 #include "check-integrity.h"
  48 #include "rcu-string.h"
  49 #include "dev-replace.h"
  50 #include "raid56.h"
  51 #include "sysfs.h"
  52 #include "qgroup.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68                                     int read_only);
  69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71                                       struct btrfs_root *root);
  72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  74                                         struct extent_io_tree *dirty_pages,
  75                                         int mark);
  76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  77                                        struct extent_io_tree *pinned_extents);
  78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  79 static void btrfs_error_commit_super(struct btrfs_root *root);
  80
  81 /*
  82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83  * is complete.  This is used during reads to verify checksums, and it is used
  84  * by writes to insert metadata for new file extents after IO is complete.
  85  */
  86 struct btrfs_end_io_wq {
  87         struct bio *bio;
  88         bio_end_io_t *end_io;
  89         void *private;
  90         struct btrfs_fs_info *info;
  91         int error;
  92         enum btrfs_wq_endio_type metadata;
  93         struct list_head list;
  94         struct btrfs_work work;
  95 };
  96
  97 static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99 int __init btrfs_end_io_wq_init(void)
 100 {
 101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102                                         sizeof(struct btrfs_end_io_wq),
 103                                         0,
 104                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         if (btrfs_end_io_wq_cache)
 114                 kmem_cache_destroy(btrfs_end_io_wq_cache);
 115 }
 116
 117 /*
 118  * async submit bios are used to offload expensive checksumming
 119  * onto the worker threads.  They checksum file and metadata bios
 120  * just before they are sent down the IO stack.
 121  */
 122 struct async_submit_bio {
 123         struct inode *inode;
 124         struct bio *bio;
 125         struct list_head list;
 126         extent_submit_bio_hook_t *submit_bio_start;
 127         extent_submit_bio_hook_t *submit_bio_done;
 128         int rw;
 129         int mirror_num;
 130         unsigned long bio_flags;
 131         /*
 132          * bio_offset is optional, can be used if the pages in the bio
 133          * can't tell us where in the file the bio should go
 134          */
 135         u64 bio_offset;
 136         struct btrfs_work work;
 137         int error;
 138 };
 139
 140 /*
 141  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 142  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 143  * the level the eb occupies in the tree.
 144  *
 145  * Different roots are used for different purposes and may nest inside each
 146  * other and they require separate keysets.  As lockdep keys should be
 147  * static, assign keysets according to the purpose of the root as indicated
 148  * by btrfs_root->objectid.  This ensures that all special purpose roots
 149  * have separate keysets.
 150  *
 151  * Lock-nesting across peer nodes is always done with the immediate parent
 152  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 153  * subclass to avoid triggering lockdep warning in such cases.
 154  *
 155  * The key is set by the readpage_end_io_hook after the buffer has passed
 156  * csum validation but before the pages are unlocked.  It is also set by
 157  * btrfs_init_new_buffer on freshly allocated blocks.
 158  *
 159  * We also add a check to make sure the highest level of the tree is the
 160  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 161  * needs update as well.
 162  */
 163 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 164 # if BTRFS_MAX_LEVEL != 8
 165 #  error
 166 # endif
 167
 168 static struct btrfs_lockdep_keyset {
 169         u64                     id;             /* root objectid */
 170         const char              *name_stem;     /* lock name stem */
 171         char                    names[BTRFS_MAX_LEVEL + 1][20];
 172         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 173 } btrfs_lockdep_keysets[] = {
 174         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 175         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 176         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 177         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 178         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 179         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 180         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 181         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 182         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 183         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 184         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 185         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 186         { .id = 0,                              .name_stem = "tree"     },
 187 };
 188
 189 void __init btrfs_init_lockdep(void)
 190 {
 191         int i, j;
 192
 193         /* initialize lockdep class names */
 194         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 195                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 196
 197                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 198                         snprintf(ks->names[j], sizeof(ks->names[j]),
 199                                  "btrfs-%s-%02d", ks->name_stem, j);
 200         }
 201 }
 202
 203 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 204                                     int level)
 205 {
 206         struct btrfs_lockdep_keyset *ks;
 207
 208         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 209
 210         /* find the matching keyset, id 0 is the default entry */
 211         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 212                 if (ks->id == objectid)
 213                         break;
 214
 215         lockdep_set_class_and_name(&eb->lock,
 216                                    &ks->keys[level], ks->names[level]);
 217 }
 218
 219 #endif
 220
 221 /*
 222  * extents on the btree inode are pretty simple, there's one extent
 223  * that covers the entire device
 224  */
 225 static struct extent_map *btree_get_extent(struct inode *inode,
 226                 struct page *page, size_t pg_offset, u64 start, u64 len,
 227                 int create)
 228 {
 229         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 230         struct extent_map *em;
 231         int ret;
 232
 233         read_lock(&em_tree->lock);
 234         em = lookup_extent_mapping(em_tree, start, len);
 235         if (em) {
 236                 em->bdev =
 237                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 238                 read_unlock(&em_tree->lock);
 239                 goto out;
 240         }
 241         read_unlock(&em_tree->lock);
 242
 243         em = alloc_extent_map();
 244         if (!em) {
 245                 em = ERR_PTR(-ENOMEM);
 246                 goto out;
 247         }
 248         em->start = 0;
 249         em->len = (u64)-1;
 250         em->block_len = (u64)-1;
 251         em->block_start = 0;
 252         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 253
 254         write_lock(&em_tree->lock);
 255         ret = add_extent_mapping(em_tree, em, 0);
 256         if (ret == -EEXIST) {
 257                 free_extent_map(em);
 258                 em = lookup_extent_mapping(em_tree, start, len);
 259                 if (!em)
 260                         em = ERR_PTR(-EIO);
 261         } else if (ret) {
 262                 free_extent_map(em);
 263                 em = ERR_PTR(ret);
 264         }
 265         write_unlock(&em_tree->lock);
 266
 267 out:
 268         return em;
 269 }
 270
 271 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 272 {
 273         return btrfs_crc32c(seed, data, len);
 274 }
 275
 276 void btrfs_csum_final(u32 crc, char *result)
 277 {
 278         put_unaligned_le32(~crc, result);
 279 }
 280
 281 /*
 282  * compute the csum for a btree block, and either verify it or write it
 283  * into the csum field of the block.
 284  */
 285 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 286                            struct extent_buffer *buf,
 287                            int verify)
 288 {
 289         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 290         char *result = NULL;
 291         unsigned long len;
 292         unsigned long cur_len;
 293         unsigned long offset = BTRFS_CSUM_SIZE;
 294         char *kaddr;
 295         unsigned long map_start;
 296         unsigned long map_len;
 297         int err;
 298         u32 crc = ~(u32)0;
 299         unsigned long inline_result;
 300
 301         len = buf->len - offset;
 302         while (len > 0) {
 303                 err = map_private_extent_buffer(buf, offset, 32,
 304                                         &kaddr, &map_start, &map_len);
 305                 if (err)
 306                         return 1;
 307                 cur_len = min(len, map_len - (offset - map_start));
 308                 crc = btrfs_csum_data(kaddr + offset - map_start,
 309                                       crc, cur_len);
 310                 len -= cur_len;
 311                 offset += cur_len;
 312         }
 313         if (csum_size > sizeof(inline_result)) {
 314                 result = kzalloc(csum_size, GFP_NOFS);
 315                 if (!result)
 316                         return 1;
 317         } else {
 318                 result = (char *)&inline_result;
 319         }
 320
 321         btrfs_csum_final(crc, result);
 322
 323         if (verify) {
 324                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 325                         u32 val;
 326                         u32 found = 0;
 327                         memcpy(&found, result, csum_size);
 328
 329                         read_extent_buffer(buf, &val, 0, csum_size);
 330                         btrfs_warn_rl(fs_info,
 331                                 "%s checksum verify failed on %llu wanted %X found %X "
 332                                 "level %d",
 333                                 fs_info->sb->s_id, buf->start,
 334                                 val, found, btrfs_header_level(buf));
 335                         if (result != (char *)&inline_result)
 336                                 kfree(result);
 337                         return 1;
 338                 }
 339         } else {
 340                 write_extent_buffer(buf, result, 0, csum_size);
 341         }
 342         if (result != (char *)&inline_result)
 343                 kfree(result);
 344         return 0;
 345 }
 346
 347 /*
 348  * we can't consider a given block up to date unless the transid of the
 349  * block matches the transid in the parent node's pointer.  This is how we
 350  * detect blocks that either didn't get written at all or got written
 351  * in the wrong place.
 352  */
 353 static int verify_parent_transid(struct extent_io_tree *io_tree,
 354                                  struct extent_buffer *eb, u64 parent_transid,
 355                                  int atomic)
 356 {
 357         struct extent_state *cached_state = NULL;
 358         int ret;
 359         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 360
 361         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 362                 return 0;
 363
 364         if (atomic)
 365                 return -EAGAIN;
 366
 367         if (need_lock) {
 368                 btrfs_tree_read_lock(eb);
 369                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 370         }
 371
 372         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 373                          &cached_state);
 374         if (extent_buffer_uptodate(eb) &&
 375             btrfs_header_generation(eb) == parent_transid) {
 376                 ret = 0;
 377                 goto out;
 378         }
 379         btrfs_err_rl(eb->fs_info,
 380                 "parent transid verify failed on %llu wanted %llu found %llu",
 381                         eb->start,
 382                         parent_transid, btrfs_header_generation(eb));
 383         ret = 1;
 384
 385         /*
 386          * Things reading via commit roots that don't have normal protection,
 387          * like send, can have a really old block in cache that may point at a
 388          * block that has been free'd and re-allocated.  So don't clear uptodate
 389          * if we find an eb that is under IO (dirty/writeback) because we could
 390          * end up reading in the stale data and then writing it back out and
 391          * making everybody very sad.
 392          */
 393         if (!extent_buffer_under_io(eb))
 394                 clear_extent_buffer_uptodate(eb);
 395 out:
 396         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 397                              &cached_state, GFP_NOFS);
 398         if (need_lock)
 399                 btrfs_tree_read_unlock_blocking(eb);
 400         return ret;
 401 }
 402
 403 /*
 404  * Return 0 if the superblock checksum type matches the checksum value of that
 405  * algorithm. Pass the raw disk superblock data.
 406  */
 407 static int btrfs_check_super_csum(char *raw_disk_sb)
 408 {
 409         struct btrfs_super_block *disk_sb =
 410                 (struct btrfs_super_block *)raw_disk_sb;
 411         u16 csum_type = btrfs_super_csum_type(disk_sb);
 412         int ret = 0;
 413
 414         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 415                 u32 crc = ~(u32)0;
 416                 const int csum_size = sizeof(crc);
 417                 char result[csum_size];
 418
 419                 /*
 420                  * The super_block structure does not span the whole
 421                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 422                  * is filled with zeros and is included in the checkum.
 423                  */
 424                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 425                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 426                 btrfs_csum_final(crc, result);
 427
 428                 if (memcmp(raw_disk_sb, result, csum_size))
 429                         ret = 1;
 430         }
 431
 432         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 433                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 434                                 csum_type);
 435                 ret = 1;
 436         }
 437
 438         return ret;
 439 }
 440
 441 /*
 442  * helper to read a given tree block, doing retries as required when
 443  * the checksums don't match and we have alternate mirrors to try.
 444  */
 445 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 446                                           struct extent_buffer *eb,
 447                                           u64 start, u64 parent_transid)
 448 {
 449         struct extent_io_tree *io_tree;
 450         int failed = 0;
 451         int ret;
 452         int num_copies = 0;
 453         int mirror_num = 0;
 454         int failed_mirror = 0;
 455
 456         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 457         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 458         while (1) {
 459                 ret = read_extent_buffer_pages(io_tree, eb, start,
 460                                                WAIT_COMPLETE,
 461                                                btree_get_extent, mirror_num);
 462                 if (!ret) {
 463                         if (!verify_parent_transid(io_tree, eb,
 464                                                    parent_transid, 0))
 465                                 break;
 466                         else
 467                                 ret = -EIO;
 468                 }
 469
 470                 /*
 471                  * This buffer's crc is fine, but its contents are corrupted, so
 472                  * there is no reason to read the other copies, they won't be
 473                  * any less wrong.
 474                  */
 475                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 476                         break;
 477
 478                 num_copies = btrfs_num_copies(root->fs_info,
 479                                               eb->start, eb->len);
 480                 if (num_copies == 1)
 481                         break;
 482
 483                 if (!failed_mirror) {
 484                         failed = 1;
 485                         failed_mirror = eb->read_mirror;
 486                 }
 487
 488                 mirror_num++;
 489                 if (mirror_num == failed_mirror)
 490                         mirror_num++;
 491
 492                 if (mirror_num > num_copies)
 493                         break;
 494         }
 495
 496         if (failed && !ret && failed_mirror)
 497                 repair_eb_io_failure(root, eb, failed_mirror);
 498
 499         return ret;
 500 }
 501
 502 /*
 503  * checksum a dirty tree block before IO.  This has extra checks to make sure
 504  * we only fill in the checksum field in the first page of a multi-page block
 505  */
 506
 507 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 508 {
 509         u64 start = page_offset(page);
 510         u64 found_start;
 511         struct extent_buffer *eb;
 512
 513         eb = (struct extent_buffer *)page->private;
 514         if (page != eb->pages[0])
 515                 return 0;
 516         found_start = btrfs_header_bytenr(eb);
 517         if (WARN_ON(found_start != start || !PageUptodate(page)))
 518                 return 0;
 519         csum_tree_block(fs_info, eb, 0);
 520         return 0;
 521 }
 522
 523 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 524                                  struct extent_buffer *eb)
 525 {
 526         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 527         u8 fsid[BTRFS_UUID_SIZE];
 528         int ret = 1;
 529
 530         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 531         while (fs_devices) {
 532                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 533                         ret = 0;
 534                         break;
 535                 }
 536                 fs_devices = fs_devices->seed;
 537         }
 538         return ret;
 539 }
 540
 541 #define CORRUPT(reason, eb, root, slot)                         \
 542         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 543                    "root=%llu, slot=%d", reason,                        \
 544                btrfs_header_bytenr(eb), root->objectid, slot)
 545
 546 static noinline int check_leaf(struct btrfs_root *root,
 547                                struct extent_buffer *leaf)
 548 {
 549         struct btrfs_key key;
 550         struct btrfs_key leaf_key;
 551         u32 nritems = btrfs_header_nritems(leaf);
 552         int slot;
 553
 554         if (nritems == 0)
 555                 return 0;
 556
 557         /* Check the 0 item */
 558         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 559             BTRFS_LEAF_DATA_SIZE(root)) {
 560                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 561                 return -EIO;
 562         }
 563
 564         /*
 565          * Check to make sure each items keys are in the correct order and their
 566          * offsets make sense.  We only have to loop through nritems-1 because
 567          * we check the current slot against the next slot, which verifies the
 568          * next slot's offset+size makes sense and that the current's slot
 569          * offset is correct.
 570          */
 571         for (slot = 0; slot < nritems - 1; slot++) {
 572                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 573                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 574
 575                 /* Make sure the keys are in the right order */
 576                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 577                         CORRUPT("bad key order", leaf, root, slot);
 578                         return -EIO;
 579                 }
 580
 581                 /*
 582                  * Make sure the offset and ends are right, remember that the
 583                  * item data starts at the end of the leaf and grows towards the
 584                  * front.
 585                  */
 586                 if (btrfs_item_offset_nr(leaf, slot) !=
 587                         btrfs_item_end_nr(leaf, slot + 1)) {
 588                         CORRUPT("slot offset bad", leaf, root, slot);
 589                         return -EIO;
 590                 }
 591
 592                 /*
 593                  * Check to make sure that we don't point outside of the leaf,
 594                  * just incase all the items are consistent to eachother, but
 595                  * all point outside of the leaf.
 596                  */
 597                 if (btrfs_item_end_nr(leaf, slot) >
 598                     BTRFS_LEAF_DATA_SIZE(root)) {
 599                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 600                         return -EIO;
 601                 }
 602         }
 603
 604         return 0;
 605 }
 606
 607 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 608                                       u64 phy_offset, struct page *page,
 609                                       u64 start, u64 end, int mirror)
 610 {
 611         u64 found_start;
 612         int found_level;
 613         struct extent_buffer *eb;
 614         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 615         int ret = 0;
 616         int reads_done;
 617
 618         if (!page->private)
 619                 goto out;
 620
 621         eb = (struct extent_buffer *)page->private;
 622
 623         /* the pending IO might have been the only thing that kept this buffer
 624          * in memory.  Make sure we have a ref for all this other checks
 625          */
 626         extent_buffer_get(eb);
 627
 628         reads_done = atomic_dec_and_test(&eb->io_pages);
 629         if (!reads_done)
 630                 goto err;
 631
 632         eb->read_mirror = mirror;
 633         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 634                 ret = -EIO;
 635                 goto err;
 636         }
 637
 638         found_start = btrfs_header_bytenr(eb);
 639         if (found_start != eb->start) {
 640                 btrfs_err_rl(eb->fs_info, "bad tree block start %llu %llu",
 641                                found_start, eb->start);
 642                 ret = -EIO;
 643                 goto err;
 644         }
 645         if (check_tree_block_fsid(root->fs_info, eb)) {
 646                 btrfs_err_rl(eb->fs_info, "bad fsid on block %llu",
 647                                eb->start);
 648                 ret = -EIO;
 649                 goto err;
 650         }
 651         found_level = btrfs_header_level(eb);
 652         if (found_level >= BTRFS_MAX_LEVEL) {
 653                 btrfs_err(root->fs_info, "bad tree block level %d",
 654                            (int)btrfs_header_level(eb));
 655                 ret = -EIO;
 656                 goto err;
 657         }
 658
 659         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 660                                        eb, found_level);
 661
 662         ret = csum_tree_block(root->fs_info, eb, 1);
 663         if (ret) {
 664                 ret = -EIO;
 665                 goto err;
 666         }
 667
 668         /*
 669          * If this is a leaf block and it is corrupt, set the corrupt bit so
 670          * that we don't try and read the other copies of this block, just
 671          * return -EIO.
 672          */
 673         if (found_level == 0 && check_leaf(root, eb)) {
 674                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 675                 ret = -EIO;
 676         }
 677
 678         if (!ret)
 679                 set_extent_buffer_uptodate(eb);
 680 err:
 681         if (reads_done &&
 682             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 683                 btree_readahead_hook(root, eb, eb->start, ret);
 684
 685         if (ret) {
 686                 /*
 687                  * our io error hook is going to dec the io pages
 688                  * again, we have to make sure it has something
 689                  * to decrement
 690                  */
 691                 atomic_inc(&eb->io_pages);
 692                 clear_extent_buffer_uptodate(eb);
 693         }
 694         free_extent_buffer(eb);
 695 out:
 696         return ret;
 697 }
 698
 699 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 700 {
 701         struct extent_buffer *eb;
 702         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 703
 704         eb = (struct extent_buffer *)page->private;
 705         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 706         eb->read_mirror = failed_mirror;
 707         atomic_dec(&eb->io_pages);
 708         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 709                 btree_readahead_hook(root, eb, eb->start, -EIO);
 710         return -EIO;    /* we fixed nothing */
 711 }
 712
 713 static void end_workqueue_bio(struct bio *bio)
 714 {
 715         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 716         struct btrfs_fs_info *fs_info;
 717         struct btrfs_workqueue *wq;
 718         btrfs_work_func_t func;
 719
 720         fs_info = end_io_wq->info;
 721         end_io_wq->error = bio->bi_error;
 722
 723         if (bio->bi_rw & REQ_WRITE) {
 724                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 725                         wq = fs_info->endio_meta_write_workers;
 726                         func = btrfs_endio_meta_write_helper;
 727                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 728                         wq = fs_info->endio_freespace_worker;
 729                         func = btrfs_freespace_write_helper;
 730                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 731                         wq = fs_info->endio_raid56_workers;
 732                         func = btrfs_endio_raid56_helper;
 733                 } else {
 734                         wq = fs_info->endio_write_workers;
 735                         func = btrfs_endio_write_helper;
 736                 }
 737         } else {
 738                 if (unlikely(end_io_wq->metadata ==
 739                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 740                         wq = fs_info->endio_repair_workers;
 741                         func = btrfs_endio_repair_helper;
 742                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 743                         wq = fs_info->endio_raid56_workers;
 744                         func = btrfs_endio_raid56_helper;
 745                 } else if (end_io_wq->metadata) {
 746                         wq = fs_info->endio_meta_workers;
 747                         func = btrfs_endio_meta_helper;
 748                 } else {
 749                         wq = fs_info->endio_workers;
 750                         func = btrfs_endio_helper;
 751                 }
 752         }
 753
 754         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 755         btrfs_queue_work(wq, &end_io_wq->work);
 756 }
 757
 758 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 759                         enum btrfs_wq_endio_type metadata)
 760 {
 761         struct btrfs_end_io_wq *end_io_wq;
 762
 763         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 764         if (!end_io_wq)
 765                 return -ENOMEM;
 766
 767         end_io_wq->private = bio->bi_private;
 768         end_io_wq->end_io = bio->bi_end_io;
 769         end_io_wq->info = info;
 770         end_io_wq->error = 0;
 771         end_io_wq->bio = bio;
 772         end_io_wq->metadata = metadata;
 773
 774         bio->bi_private = end_io_wq;
 775         bio->bi_end_io = end_workqueue_bio;
 776         return 0;
 777 }
 778
 779 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 780 {
 781         unsigned long limit = min_t(unsigned long,
 782                                     info->thread_pool_size,
 783                                     info->fs_devices->open_devices);
 784         return 256 * limit;
 785 }
 786
 787 static void run_one_async_start(struct btrfs_work *work)
 788 {
 789         struct async_submit_bio *async;
 790         int ret;
 791
 792         async = container_of(work, struct  async_submit_bio, work);
 793         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 794                                       async->mirror_num, async->bio_flags,
 795                                       async->bio_offset);
 796         if (ret)
 797                 async->error = ret;
 798 }
 799
 800 static void run_one_async_done(struct btrfs_work *work)
 801 {
 802         struct btrfs_fs_info *fs_info;
 803         struct async_submit_bio *async;
 804         int limit;
 805
 806         async = container_of(work, struct  async_submit_bio, work);
 807         fs_info = BTRFS_I(async->inode)->root->fs_info;
 808
 809         limit = btrfs_async_submit_limit(fs_info);
 810         limit = limit * 2 / 3;
 811
 812         /*
 813          * atomic_dec_return implies a barrier for waitqueue_active
 814          */
 815         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 816             waitqueue_active(&fs_info->async_submit_wait))
 817                 wake_up(&fs_info->async_submit_wait);
 818
 819         /* If an error occured we just want to clean up the bio and move on */
 820         if (async->error) {
 821                 async->bio->bi_error = async->error;
 822                 bio_endio(async->bio);
 823                 return;
 824         }
 825
 826         async->submit_bio_done(async->inode, async->rw, async->bio,
 827                                async->mirror_num, async->bio_flags,
 828                                async->bio_offset);
 829 }
 830
 831 static void run_one_async_free(struct btrfs_work *work)
 832 {
 833         struct async_submit_bio *async;
 834
 835         async = container_of(work, struct  async_submit_bio, work);
 836         kfree(async);
 837 }
 838
 839 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 840                         int rw, struct bio *bio, int mirror_num,
 841                         unsigned long bio_flags,
 842                         u64 bio_offset,
 843                         extent_submit_bio_hook_t *submit_bio_start,
 844                         extent_submit_bio_hook_t *submit_bio_done)
 845 {
 846         struct async_submit_bio *async;
 847
 848         async = kmalloc(sizeof(*async), GFP_NOFS);
 849         if (!async)
 850                 return -ENOMEM;
 851
 852         async->inode = inode;
 853         async->rw = rw;
 854         async->bio = bio;
 855         async->mirror_num = mirror_num;
 856         async->submit_bio_start = submit_bio_start;
 857         async->submit_bio_done = submit_bio_done;
 858
 859         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 860                         run_one_async_done, run_one_async_free);
 861
 862         async->bio_flags = bio_flags;
 863         async->bio_offset = bio_offset;
 864
 865         async->error = 0;
 866
 867         atomic_inc(&fs_info->nr_async_submits);
 868
 869         if (rw & REQ_SYNC)
 870                 btrfs_set_work_high_priority(&async->work);
 871
 872         btrfs_queue_work(fs_info->workers, &async->work);
 873
 874         while (atomic_read(&fs_info->async_submit_draining) &&
 875               atomic_read(&fs_info->nr_async_submits)) {
 876                 wait_event(fs_info->async_submit_wait,
 877                            (atomic_read(&fs_info->nr_async_submits) == 0));
 878         }
 879
 880         return 0;
 881 }
 882
 883 static int btree_csum_one_bio(struct bio *bio)
 884 {
 885         struct bio_vec *bvec;
 886         struct btrfs_root *root;
 887         int i, ret = 0;
 888
 889         bio_for_each_segment_all(bvec, bio, i) {
 890                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 891                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 892                 if (ret)
 893                         break;
 894         }
 895
 896         return ret;
 897 }
 898
 899 static int __btree_submit_bio_start(struct inode *inode, int rw,
 900                                     struct bio *bio, int mirror_num,
 901                                     unsigned long bio_flags,
 902                                     u64 bio_offset)
 903 {
 904         /*
 905          * when we're called for a write, we're already in the async
 906          * submission context.  Just jump into btrfs_map_bio
 907          */
 908         return btree_csum_one_bio(bio);
 909 }
 910
 911 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 912                                  int mirror_num, unsigned long bio_flags,
 913                                  u64 bio_offset)
 914 {
 915         int ret;
 916
 917         /*
 918          * when we're called for a write, we're already in the async
 919          * submission context.  Just jump into btrfs_map_bio
 920          */
 921         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 922         if (ret) {
 923                 bio->bi_error = ret;
 924                 bio_endio(bio);
 925         }
 926         return ret;
 927 }
 928
 929 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 930 {
 931         if (bio_flags & EXTENT_BIO_TREE_LOG)
 932                 return 0;
 933 #ifdef CONFIG_X86
 934         if (cpu_has_xmm4_2)
 935                 return 0;
 936 #endif
 937         return 1;
 938 }
 939
 940 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 941                                  int mirror_num, unsigned long bio_flags,
 942                                  u64 bio_offset)
 943 {
 944         int async = check_async_write(inode, bio_flags);
 945         int ret;
 946
 947         if (!(rw & REQ_WRITE)) {
 948                 /*
 949                  * called for a read, do the setup so that checksum validation
 950                  * can happen in the async kernel threads
 951                  */
 952                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 953                                           bio, BTRFS_WQ_ENDIO_METADATA);
 954                 if (ret)
 955                         goto out_w_error;
 956                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 957                                     mirror_num, 0);
 958         } else if (!async) {
 959                 ret = btree_csum_one_bio(bio);
 960                 if (ret)
 961                         goto out_w_error;
 962                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 963                                     mirror_num, 0);
 964         } else {
 965                 /*
 966                  * kthread helpers are used to submit writes so that
 967                  * checksumming can happen in parallel across all CPUs
 968                  */
 969                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 970                                           inode, rw, bio, mirror_num, 0,
 971                                           bio_offset,
 972                                           __btree_submit_bio_start,
 973                                           __btree_submit_bio_done);
 974         }
 975
 976         if (ret)
 977                 goto out_w_error;
 978         return 0;
 979
 980 out_w_error:
 981         bio->bi_error = ret;
 982         bio_endio(bio);
 983         return ret;
 984 }
 985
 986 #ifdef CONFIG_MIGRATION
 987 static int btree_migratepage(struct address_space *mapping,
 988                         struct page *newpage, struct page *page,
 989                         enum migrate_mode mode)
 990 {
 991         /*
 992          * we can't safely write a btree page from here,
 993          * we haven't done the locking hook
 994          */
 995         if (PageDirty(page))
 996                 return -EAGAIN;
 997         /*
 998          * Buffers may be managed in a filesystem specific way.
 999          * We must have no buffers or drop them.
1000          */
1001         if (page_has_private(page) &&
1002             !try_to_release_page(page, GFP_KERNEL))
1003                 return -EAGAIN;
1004         return migrate_page(mapping, newpage, page, mode);
1005 }
1006 #endif
1007
1008
1009 static int btree_writepages(struct address_space *mapping,
1010                             struct writeback_control *wbc)
1011 {
1012         struct btrfs_fs_info *fs_info;
1013         int ret;
1014
1015         if (wbc->sync_mode == WB_SYNC_NONE) {
1016
1017                 if (wbc->for_kupdate)
1018                         return 0;
1019
1020                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1021                 /* this is a bit racy, but that's ok */
1022                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1023                                              BTRFS_DIRTY_METADATA_THRESH);
1024                 if (ret < 0)
1025                         return 0;
1026         }
1027         return btree_write_cache_pages(mapping, wbc);
1028 }
1029
1030 static int btree_readpage(struct file *file, struct page *page)
1031 {
1032         struct extent_io_tree *tree;
1033         tree = &BTRFS_I(page->mapping->host)->io_tree;
1034         return extent_read_full_page(tree, page, btree_get_extent, 0);
1035 }
1036
1037 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1038 {
1039         if (PageWriteback(page) || PageDirty(page))
1040                 return 0;
1041
1042         return try_release_extent_buffer(page);
1043 }
1044
1045 static void btree_invalidatepage(struct page *page, unsigned int offset,
1046                                  unsigned int length)
1047 {
1048         struct extent_io_tree *tree;
1049         tree = &BTRFS_I(page->mapping->host)->io_tree;
1050         extent_invalidatepage(tree, page, offset);
1051         btree_releasepage(page, GFP_NOFS);
1052         if (PagePrivate(page)) {
1053                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1054                            "page private not zero on page %llu",
1055                            (unsigned long long)page_offset(page));
1056                 ClearPagePrivate(page);
1057                 set_page_private(page, 0);
1058                 page_cache_release(page);
1059         }
1060 }
1061
1062 static int btree_set_page_dirty(struct page *page)
1063 {
1064 #ifdef DEBUG
1065         struct extent_buffer *eb;
1066
1067         BUG_ON(!PagePrivate(page));
1068         eb = (struct extent_buffer *)page->private;
1069         BUG_ON(!eb);
1070         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1071         BUG_ON(!atomic_read(&eb->refs));
1072         btrfs_assert_tree_locked(eb);
1073 #endif
1074         return __set_page_dirty_nobuffers(page);
1075 }
1076
1077 static const struct address_space_operations btree_aops = {
1078         .readpage       = btree_readpage,
1079         .writepages     = btree_writepages,
1080         .releasepage    = btree_releasepage,
1081         .invalidatepage = btree_invalidatepage,
1082 #ifdef CONFIG_MIGRATION
1083         .migratepage    = btree_migratepage,
1084 #endif
1085         .set_page_dirty = btree_set_page_dirty,
1086 };
1087
1088 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1089 {
1090         struct extent_buffer *buf = NULL;
1091         struct inode *btree_inode = root->fs_info->btree_inode;
1092
1093         buf = btrfs_find_create_tree_block(root, bytenr);
1094         if (!buf)
1095                 return;
1096         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1097                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1098         free_extent_buffer(buf);
1099 }
1100
1101 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1102                          int mirror_num, struct extent_buffer **eb)
1103 {
1104         struct extent_buffer *buf = NULL;
1105         struct inode *btree_inode = root->fs_info->btree_inode;
1106         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1107         int ret;
1108
1109         buf = btrfs_find_create_tree_block(root, bytenr);
1110         if (!buf)
1111                 return 0;
1112
1113         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1114
1115         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1116                                        btree_get_extent, mirror_num);
1117         if (ret) {
1118                 free_extent_buffer(buf);
1119                 return ret;
1120         }
1121
1122         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1123                 free_extent_buffer(buf);
1124                 return -EIO;
1125         } else if (extent_buffer_uptodate(buf)) {
1126                 *eb = buf;
1127         } else {
1128                 free_extent_buffer(buf);
1129         }
1130         return 0;
1131 }
1132
1133 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1134                                             u64 bytenr)
1135 {
1136         return find_extent_buffer(fs_info, bytenr);
1137 }
1138
1139 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1140                                                  u64 bytenr)
1141 {
1142         if (btrfs_test_is_dummy_root(root))
1143                 return alloc_test_extent_buffer(root->fs_info, bytenr);
1144         return alloc_extent_buffer(root->fs_info, bytenr);
1145 }
1146
1147
1148 int btrfs_write_tree_block(struct extent_buffer *buf)
1149 {
1150         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1151                                         buf->start + buf->len - 1);
1152 }
1153
1154 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1155 {
1156         return filemap_fdatawait_range(buf->pages[0]->mapping,
1157                                        buf->start, buf->start + buf->len - 1);
1158 }
1159
1160 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1161                                       u64 parent_transid)
1162 {
1163         struct extent_buffer *buf = NULL;
1164         int ret;
1165
1166         buf = btrfs_find_create_tree_block(root, bytenr);
1167         if (!buf)
1168                 return ERR_PTR(-ENOMEM);
1169
1170         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1171         if (ret) {
1172                 free_extent_buffer(buf);
1173                 return ERR_PTR(ret);
1174         }
1175         return buf;
1176
1177 }
1178
1179 void clean_tree_block(struct btrfs_trans_handle *trans,
1180                       struct btrfs_fs_info *fs_info,
1181                       struct extent_buffer *buf)
1182 {
1183         if (btrfs_header_generation(buf) ==
1184             fs_info->running_transaction->transid) {
1185                 btrfs_assert_tree_locked(buf);
1186
1187                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1188                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1189                                              -buf->len,
1190                                              fs_info->dirty_metadata_batch);
1191                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1192                         btrfs_set_lock_blocking(buf);
1193                         clear_extent_buffer_dirty(buf);
1194                 }
1195         }
1196 }
1197
1198 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1199 {
1200         struct btrfs_subvolume_writers *writers;
1201         int ret;
1202
1203         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1204         if (!writers)
1205                 return ERR_PTR(-ENOMEM);
1206
1207         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1208         if (ret < 0) {
1209                 kfree(writers);
1210                 return ERR_PTR(ret);
1211         }
1212
1213         init_waitqueue_head(&writers->wait);
1214         return writers;
1215 }
1216
1217 static void
1218 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1219 {
1220         percpu_counter_destroy(&writers->counter);
1221         kfree(writers);
1222 }
1223
1224 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1225                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1226                          u64 objectid)
1227 {
1228         root->node = NULL;
1229         root->commit_root = NULL;
1230         root->sectorsize = sectorsize;
1231         root->nodesize = nodesize;
1232         root->stripesize = stripesize;
1233         root->state = 0;
1234         root->orphan_cleanup_state = 0;
1235
1236         root->objectid = objectid;
1237         root->last_trans = 0;
1238         root->highest_objectid = 0;
1239         root->nr_delalloc_inodes = 0;
1240         root->nr_ordered_extents = 0;
1241         root->name = NULL;
1242         root->inode_tree = RB_ROOT;
1243         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1244         root->block_rsv = NULL;
1245         root->orphan_block_rsv = NULL;
1246
1247         INIT_LIST_HEAD(&root->dirty_list);
1248         INIT_LIST_HEAD(&root->root_list);
1249         INIT_LIST_HEAD(&root->delalloc_inodes);
1250         INIT_LIST_HEAD(&root->delalloc_root);
1251         INIT_LIST_HEAD(&root->ordered_extents);
1252         INIT_LIST_HEAD(&root->ordered_root);
1253         INIT_LIST_HEAD(&root->logged_list[0]);
1254         INIT_LIST_HEAD(&root->logged_list[1]);
1255         spin_lock_init(&root->orphan_lock);
1256         spin_lock_init(&root->inode_lock);
1257         spin_lock_init(&root->delalloc_lock);
1258         spin_lock_init(&root->ordered_extent_lock);
1259         spin_lock_init(&root->accounting_lock);
1260         spin_lock_init(&root->log_extents_lock[0]);
1261         spin_lock_init(&root->log_extents_lock[1]);
1262         mutex_init(&root->objectid_mutex);
1263         mutex_init(&root->log_mutex);
1264         mutex_init(&root->ordered_extent_mutex);
1265         mutex_init(&root->delalloc_mutex);
1266         init_waitqueue_head(&root->log_writer_wait);
1267         init_waitqueue_head(&root->log_commit_wait[0]);
1268         init_waitqueue_head(&root->log_commit_wait[1]);
1269         INIT_LIST_HEAD(&root->log_ctxs[0]);
1270         INIT_LIST_HEAD(&root->log_ctxs[1]);
1271         atomic_set(&root->log_commit[0], 0);
1272         atomic_set(&root->log_commit[1], 0);
1273         atomic_set(&root->log_writers, 0);
1274         atomic_set(&root->log_batch, 0);
1275         atomic_set(&root->orphan_inodes, 0);
1276         atomic_set(&root->refs, 1);
1277         atomic_set(&root->will_be_snapshoted, 0);
1278         atomic_set(&root->qgroup_meta_rsv, 0);
1279         root->log_transid = 0;
1280         root->log_transid_committed = -1;
1281         root->last_log_commit = 0;
1282         if (fs_info)
1283                 extent_io_tree_init(&root->dirty_log_pages,
1284                                      fs_info->btree_inode->i_mapping);
1285
1286         memset(&root->root_key, 0, sizeof(root->root_key));
1287         memset(&root->root_item, 0, sizeof(root->root_item));
1288         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1289         if (fs_info)
1290                 root->defrag_trans_start = fs_info->generation;
1291         else
1292                 root->defrag_trans_start = 0;
1293         root->root_key.objectid = objectid;
1294         root->anon_dev = 0;
1295
1296         spin_lock_init(&root->root_item_lock);
1297 }
1298
1299 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1300 {
1301         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1302         if (root)
1303                 root->fs_info = fs_info;
1304         return root;
1305 }
1306
1307 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1308 /* Should only be used by the testing infrastructure */
1309 struct btrfs_root *btrfs_alloc_dummy_root(void)
1310 {
1311         struct btrfs_root *root;
1312
1313         root = btrfs_alloc_root(NULL);
1314         if (!root)
1315                 return ERR_PTR(-ENOMEM);
1316         __setup_root(4096, 4096, 4096, root, NULL, 1);
1317         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1318         root->alloc_bytenr = 0;
1319
1320         return root;
1321 }
1322 #endif
1323
1324 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1325                                      struct btrfs_fs_info *fs_info,
1326                                      u64 objectid)
1327 {
1328         struct extent_buffer *leaf;
1329         struct btrfs_root *tree_root = fs_info->tree_root;
1330         struct btrfs_root *root;
1331         struct btrfs_key key;
1332         int ret = 0;
1333         uuid_le uuid;
1334
1335         root = btrfs_alloc_root(fs_info);
1336         if (!root)
1337                 return ERR_PTR(-ENOMEM);
1338
1339         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1340                 tree_root->stripesize, root, fs_info, objectid);
1341         root->root_key.objectid = objectid;
1342         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1343         root->root_key.offset = 0;
1344
1345         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1346         if (IS_ERR(leaf)) {
1347                 ret = PTR_ERR(leaf);
1348                 leaf = NULL;
1349                 goto fail;
1350         }
1351
1352         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1353         btrfs_set_header_bytenr(leaf, leaf->start);
1354         btrfs_set_header_generation(leaf, trans->transid);
1355         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1356         btrfs_set_header_owner(leaf, objectid);
1357         root->node = leaf;
1358
1359         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1360                             BTRFS_FSID_SIZE);
1361         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1362                             btrfs_header_chunk_tree_uuid(leaf),
1363                             BTRFS_UUID_SIZE);
1364         btrfs_mark_buffer_dirty(leaf);
1365
1366         root->commit_root = btrfs_root_node(root);
1367         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1368
1369         root->root_item.flags = 0;
1370         root->root_item.byte_limit = 0;
1371         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1372         btrfs_set_root_generation(&root->root_item, trans->transid);
1373         btrfs_set_root_level(&root->root_item, 0);
1374         btrfs_set_root_refs(&root->root_item, 1);
1375         btrfs_set_root_used(&root->root_item, leaf->len);
1376         btrfs_set_root_last_snapshot(&root->root_item, 0);
1377         btrfs_set_root_dirid(&root->root_item, 0);
1378         uuid_le_gen(&uuid);
1379         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1380         root->root_item.drop_level = 0;
1381
1382         key.objectid = objectid;
1383         key.type = BTRFS_ROOT_ITEM_KEY;
1384         key.offset = 0;
1385         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1386         if (ret)
1387                 goto fail;
1388
1389         btrfs_tree_unlock(leaf);
1390
1391         return root;
1392
1393 fail:
1394         if (leaf) {
1395                 btrfs_tree_unlock(leaf);
1396                 free_extent_buffer(root->commit_root);
1397                 free_extent_buffer(leaf);
1398         }
1399         kfree(root);
1400
1401         return ERR_PTR(ret);
1402 }
1403
1404 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1405                                          struct btrfs_fs_info *fs_info)
1406 {
1407         struct btrfs_root *root;
1408         struct btrfs_root *tree_root = fs_info->tree_root;
1409         struct extent_buffer *leaf;
1410
1411         root = btrfs_alloc_root(fs_info);
1412         if (!root)
1413                 return ERR_PTR(-ENOMEM);
1414
1415         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1416                      tree_root->stripesize, root, fs_info,
1417                      BTRFS_TREE_LOG_OBJECTID);
1418
1419         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1420         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1421         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1422
1423         /*
1424          * DON'T set REF_COWS for log trees
1425          *
1426          * log trees do not get reference counted because they go away
1427          * before a real commit is actually done.  They do store pointers
1428          * to file data extents, and those reference counts still get
1429          * updated (along with back refs to the log tree).
1430          */
1431
1432         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1433                         NULL, 0, 0, 0);
1434         if (IS_ERR(leaf)) {
1435                 kfree(root);
1436                 return ERR_CAST(leaf);
1437         }
1438
1439         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1440         btrfs_set_header_bytenr(leaf, leaf->start);
1441         btrfs_set_header_generation(leaf, trans->transid);
1442         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1443         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1444         root->node = leaf;
1445
1446         write_extent_buffer(root->node, root->fs_info->fsid,
1447                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1448         btrfs_mark_buffer_dirty(root->node);
1449         btrfs_tree_unlock(root->node);
1450         return root;
1451 }
1452
1453 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1454                              struct btrfs_fs_info *fs_info)
1455 {
1456         struct btrfs_root *log_root;
1457
1458         log_root = alloc_log_tree(trans, fs_info);
1459         if (IS_ERR(log_root))
1460                 return PTR_ERR(log_root);
1461         WARN_ON(fs_info->log_root_tree);
1462         fs_info->log_root_tree = log_root;
1463         return 0;
1464 }
1465
1466 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1467                        struct btrfs_root *root)
1468 {
1469         struct btrfs_root *log_root;
1470         struct btrfs_inode_item *inode_item;
1471
1472         log_root = alloc_log_tree(trans, root->fs_info);
1473         if (IS_ERR(log_root))
1474                 return PTR_ERR(log_root);
1475
1476         log_root->last_trans = trans->transid;
1477         log_root->root_key.offset = root->root_key.objectid;
1478
1479         inode_item = &log_root->root_item.inode;
1480         btrfs_set_stack_inode_generation(inode_item, 1);
1481         btrfs_set_stack_inode_size(inode_item, 3);
1482         btrfs_set_stack_inode_nlink(inode_item, 1);
1483         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1484         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1485
1486         btrfs_set_root_node(&log_root->root_item, log_root->node);
1487
1488         WARN_ON(root->log_root);
1489         root->log_root = log_root;
1490         root->log_transid = 0;
1491         root->log_transid_committed = -1;
1492         root->last_log_commit = 0;
1493         return 0;
1494 }
1495
1496 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1497                                                struct btrfs_key *key)
1498 {
1499         struct btrfs_root *root;
1500         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1501         struct btrfs_path *path;
1502         u64 generation;
1503         int ret;
1504
1505         path = btrfs_alloc_path();
1506         if (!path)
1507                 return ERR_PTR(-ENOMEM);
1508
1509         root = btrfs_alloc_root(fs_info);
1510         if (!root) {
1511                 ret = -ENOMEM;
1512                 goto alloc_fail;
1513         }
1514
1515         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1516                 tree_root->stripesize, root, fs_info, key->objectid);
1517
1518         ret = btrfs_find_root(tree_root, key, path,
1519                               &root->root_item, &root->root_key);
1520         if (ret) {
1521                 if (ret > 0)
1522                         ret = -ENOENT;
1523                 goto find_fail;
1524         }
1525
1526         generation = btrfs_root_generation(&root->root_item);
1527         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1528                                      generation);
1529         if (IS_ERR(root->node)) {
1530                 ret = PTR_ERR(root->node);
1531                 goto find_fail;
1532         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1533                 ret = -EIO;
1534                 free_extent_buffer(root->node);
1535                 goto find_fail;
1536         }
1537         root->commit_root = btrfs_root_node(root);
1538 out:
1539         btrfs_free_path(path);
1540         return root;
1541
1542 find_fail:
1543         kfree(root);
1544 alloc_fail:
1545         root = ERR_PTR(ret);
1546         goto out;
1547 }
1548
1549 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1550                                       struct btrfs_key *location)
1551 {
1552         struct btrfs_root *root;
1553
1554         root = btrfs_read_tree_root(tree_root, location);
1555         if (IS_ERR(root))
1556                 return root;
1557
1558         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1559                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1560                 btrfs_check_and_init_root_item(&root->root_item);
1561         }
1562
1563         return root;
1564 }
1565
1566 int btrfs_init_fs_root(struct btrfs_root *root)
1567 {
1568         int ret;
1569         struct btrfs_subvolume_writers *writers;
1570
1571         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1572         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1573                                         GFP_NOFS);
1574         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1575                 ret = -ENOMEM;
1576                 goto fail;
1577         }
1578
1579         writers = btrfs_alloc_subvolume_writers();
1580         if (IS_ERR(writers)) {
1581                 ret = PTR_ERR(writers);
1582                 goto fail;
1583         }
1584         root->subv_writers = writers;
1585
1586         btrfs_init_free_ino_ctl(root);
1587         spin_lock_init(&root->ino_cache_lock);
1588         init_waitqueue_head(&root->ino_cache_wait);
1589
1590         ret = get_anon_bdev(&root->anon_dev);
1591         if (ret)
1592                 goto free_writers;
1593
1594         mutex_lock(&root->objectid_mutex);
1595         ret = btrfs_find_highest_objectid(root,
1596                                         &root->highest_objectid);
1597         if (ret) {
1598                 mutex_unlock(&root->objectid_mutex);
1599                 goto free_root_dev;
1600         }
1601
1602         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1603
1604         mutex_unlock(&root->objectid_mutex);
1605
1606         return 0;
1607
1608 free_root_dev:
1609         free_anon_bdev(root->anon_dev);
1610 free_writers:
1611         btrfs_free_subvolume_writers(root->subv_writers);
1612 fail:
1613         kfree(root->free_ino_ctl);
1614         kfree(root->free_ino_pinned);
1615         return ret;
1616 }
1617
1618 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1619                                                u64 root_id)
1620 {
1621         struct btrfs_root *root;
1622
1623         spin_lock(&fs_info->fs_roots_radix_lock);
1624         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1625                                  (unsigned long)root_id);
1626         spin_unlock(&fs_info->fs_roots_radix_lock);
1627         return root;
1628 }
1629
1630 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1631                          struct btrfs_root *root)
1632 {
1633         int ret;
1634
1635         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1636         if (ret)
1637                 return ret;
1638
1639         spin_lock(&fs_info->fs_roots_radix_lock);
1640         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1641                                 (unsigned long)root->root_key.objectid,
1642                                 root);
1643         if (ret == 0)
1644                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1645         spin_unlock(&fs_info->fs_roots_radix_lock);
1646         radix_tree_preload_end();
1647
1648         return ret;
1649 }
1650
1651 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1652                                      struct btrfs_key *location,
1653                                      bool check_ref)
1654 {
1655         struct btrfs_root *root;
1656         struct btrfs_path *path;
1657         struct btrfs_key key;
1658         int ret;
1659
1660         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1661                 return fs_info->tree_root;
1662         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1663                 return fs_info->extent_root;
1664         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1665                 return fs_info->chunk_root;
1666         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1667                 return fs_info->dev_root;
1668         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1669                 return fs_info->csum_root;
1670         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1671                 return fs_info->quota_root ? fs_info->quota_root :
1672                                              ERR_PTR(-ENOENT);
1673         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1674                 return fs_info->uuid_root ? fs_info->uuid_root :
1675                                             ERR_PTR(-ENOENT);
1676         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1677                 return fs_info->free_space_root ? fs_info->free_space_root :
1678                                                   ERR_PTR(-ENOENT);
1679 again:
1680         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1681         if (root) {
1682                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1683                         return ERR_PTR(-ENOENT);
1684                 return root;
1685         }
1686
1687         root = btrfs_read_fs_root(fs_info->tree_root, location);
1688         if (IS_ERR(root))
1689                 return root;
1690
1691         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1692                 ret = -ENOENT;
1693                 goto fail;
1694         }
1695
1696         ret = btrfs_init_fs_root(root);
1697         if (ret)
1698                 goto fail;
1699
1700         path = btrfs_alloc_path();
1701         if (!path) {
1702                 ret = -ENOMEM;
1703                 goto fail;
1704         }
1705         key.objectid = BTRFS_ORPHAN_OBJECTID;
1706         key.type = BTRFS_ORPHAN_ITEM_KEY;
1707         key.offset = location->objectid;
1708
1709         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1710         btrfs_free_path(path);
1711         if (ret < 0)
1712                 goto fail;
1713         if (ret == 0)
1714                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1715
1716         ret = btrfs_insert_fs_root(fs_info, root);
1717         if (ret) {
1718                 if (ret == -EEXIST) {
1719                         free_fs_root(root);
1720                         goto again;
1721                 }
1722                 goto fail;
1723         }
1724         return root;
1725 fail:
1726         free_fs_root(root);
1727         return ERR_PTR(ret);
1728 }
1729
1730 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1731 {
1732         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1733         int ret = 0;
1734         struct btrfs_device *device;
1735         struct backing_dev_info *bdi;
1736
1737         rcu_read_lock();
1738         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1739                 if (!device->bdev)
1740                         continue;
1741                 bdi = blk_get_backing_dev_info(device->bdev);
1742                 if (bdi_congested(bdi, bdi_bits)) {
1743                         ret = 1;
1744                         break;
1745                 }
1746         }
1747         rcu_read_unlock();
1748         return ret;
1749 }
1750
1751 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1752 {
1753         int err;
1754
1755         err = bdi_setup_and_register(bdi, "btrfs");
1756         if (err)
1757                 return err;
1758
1759         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1760         bdi->congested_fn       = btrfs_congested_fn;
1761         bdi->congested_data     = info;
1762         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1763         return 0;
1764 }
1765
1766 /*
1767  * called by the kthread helper functions to finally call the bio end_io
1768  * functions.  This is where read checksum verification actually happens
1769  */
1770 static void end_workqueue_fn(struct btrfs_work *work)
1771 {
1772         struct bio *bio;
1773         struct btrfs_end_io_wq *end_io_wq;
1774
1775         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1776         bio = end_io_wq->bio;
1777
1778         bio->bi_error = end_io_wq->error;
1779         bio->bi_private = end_io_wq->private;
1780         bio->bi_end_io = end_io_wq->end_io;
1781         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1782         bio_endio(bio);
1783 }
1784
1785 static int cleaner_kthread(void *arg)
1786 {
1787         struct btrfs_root *root = arg;
1788         int again;
1789         struct btrfs_trans_handle *trans;
1790
1791         do {
1792                 again = 0;
1793
1794                 /* Make the cleaner go to sleep early. */
1795                 if (btrfs_need_cleaner_sleep(root))
1796                         goto sleep;
1797
1798                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1799                         goto sleep;
1800
1801                 /*
1802                  * Avoid the problem that we change the status of the fs
1803                  * during the above check and trylock.
1804                  */
1805                 if (btrfs_need_cleaner_sleep(root)) {
1806                         mutex_unlock(&root->fs_info->cleaner_mutex);
1807                         goto sleep;
1808                 }
1809
1810                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1811                 btrfs_run_delayed_iputs(root);
1812                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1813
1814                 again = btrfs_clean_one_deleted_snapshot(root);
1815                 mutex_unlock(&root->fs_info->cleaner_mutex);
1816
1817                 /*
1818                  * The defragger has dealt with the R/O remount and umount,
1819                  * needn't do anything special here.
1820                  */
1821                 btrfs_run_defrag_inodes(root->fs_info);
1822
1823                 /*
1824                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1825                  * with relocation (btrfs_relocate_chunk) and relocation
1826                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1827                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1828                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1829                  * unused block groups.
1830                  */
1831                 btrfs_delete_unused_bgs(root->fs_info);
1832 sleep:
1833                 if (!try_to_freeze() && !again) {
1834                         set_current_state(TASK_INTERRUPTIBLE);
1835                         if (!kthread_should_stop())
1836                                 schedule();
1837                         __set_current_state(TASK_RUNNING);
1838                 }
1839         } while (!kthread_should_stop());
1840
1841         /*
1842          * Transaction kthread is stopped before us and wakes us up.
1843          * However we might have started a new transaction and COWed some
1844          * tree blocks when deleting unused block groups for example. So
1845          * make sure we commit the transaction we started to have a clean
1846          * shutdown when evicting the btree inode - if it has dirty pages
1847          * when we do the final iput() on it, eviction will trigger a
1848          * writeback for it which will fail with null pointer dereferences
1849          * since work queues and other resources were already released and
1850          * destroyed by the time the iput/eviction/writeback is made.
1851          */
1852         trans = btrfs_attach_transaction(root);
1853         if (IS_ERR(trans)) {
1854                 if (PTR_ERR(trans) != -ENOENT)
1855                         btrfs_err(root->fs_info,
1856                                   "cleaner transaction attach returned %ld",
1857                                   PTR_ERR(trans));
1858         } else {
1859                 int ret;
1860
1861                 ret = btrfs_commit_transaction(trans, root);
1862                 if (ret)
1863                         btrfs_err(root->fs_info,
1864                                   "cleaner open transaction commit returned %d",
1865                                   ret);
1866         }
1867
1868         return 0;
1869 }
1870
1871 static int transaction_kthread(void *arg)
1872 {
1873         struct btrfs_root *root = arg;
1874         struct btrfs_trans_handle *trans;
1875         struct btrfs_transaction *cur;
1876         u64 transid;
1877         unsigned long now;
1878         unsigned long delay;
1879         bool cannot_commit;
1880
1881         do {
1882                 cannot_commit = false;
1883                 delay = HZ * root->fs_info->commit_interval;
1884                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1885
1886                 spin_lock(&root->fs_info->trans_lock);
1887                 cur = root->fs_info->running_transaction;
1888                 if (!cur) {
1889                         spin_unlock(&root->fs_info->trans_lock);
1890                         goto sleep;
1891                 }
1892
1893                 now = get_seconds();
1894                 if (cur->state < TRANS_STATE_BLOCKED &&
1895                     (now < cur->start_time ||
1896                      now - cur->start_time < root->fs_info->commit_interval)) {
1897                         spin_unlock(&root->fs_info->trans_lock);
1898                         delay = HZ * 5;
1899                         goto sleep;
1900                 }
1901                 transid = cur->transid;
1902                 spin_unlock(&root->fs_info->trans_lock);
1903
1904                 /* If the file system is aborted, this will always fail. */
1905                 trans = btrfs_attach_transaction(root);
1906                 if (IS_ERR(trans)) {
1907                         if (PTR_ERR(trans) != -ENOENT)
1908                                 cannot_commit = true;
1909                         goto sleep;
1910                 }
1911                 if (transid == trans->transid) {
1912                         btrfs_commit_transaction(trans, root);
1913                 } else {
1914                         btrfs_end_transaction(trans, root);
1915                 }
1916 sleep:
1917                 wake_up_process(root->fs_info->cleaner_kthread);
1918                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1919
1920                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1921                                       &root->fs_info->fs_state)))
1922                         btrfs_cleanup_transaction(root);
1923                 if (!try_to_freeze()) {
1924                         set_current_state(TASK_INTERRUPTIBLE);
1925                         if (!kthread_should_stop() &&
1926                             (!btrfs_transaction_blocked(root->fs_info) ||
1927                              cannot_commit))
1928                                 schedule_timeout(delay);
1929                         __set_current_state(TASK_RUNNING);
1930                 }
1931         } while (!kthread_should_stop());
1932         return 0;
1933 }
1934
1935 /*
1936  * this will find the highest generation in the array of
1937  * root backups.  The index of the highest array is returned,
1938  * or -1 if we can't find anything.
1939  *
1940  * We check to make sure the array is valid by comparing the
1941  * generation of the latest  root in the array with the generation
1942  * in the super block.  If they don't match we pitch it.
1943  */
1944 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1945 {
1946         u64 cur;
1947         int newest_index = -1;
1948         struct btrfs_root_backup *root_backup;
1949         int i;
1950
1951         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1952                 root_backup = info->super_copy->super_roots + i;
1953                 cur = btrfs_backup_tree_root_gen(root_backup);
1954                 if (cur == newest_gen)
1955                         newest_index = i;
1956         }
1957
1958         /* check to see if we actually wrapped around */
1959         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1960                 root_backup = info->super_copy->super_roots;
1961                 cur = btrfs_backup_tree_root_gen(root_backup);
1962                 if (cur == newest_gen)
1963                         newest_index = 0;
1964         }
1965         return newest_index;
1966 }
1967
1968
1969 /*
1970  * find the oldest backup so we know where to store new entries
1971  * in the backup array.  This will set the backup_root_index
1972  * field in the fs_info struct
1973  */
1974 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1975                                      u64 newest_gen)
1976 {
1977         int newest_index = -1;
1978
1979         newest_index = find_newest_super_backup(info, newest_gen);
1980         /* if there was garbage in there, just move along */
1981         if (newest_index == -1) {
1982                 info->backup_root_index = 0;
1983         } else {
1984                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1985         }
1986 }
1987
1988 /*
1989  * copy all the root pointers into the super backup array.
1990  * this will bump the backup pointer by one when it is
1991  * done
1992  */
1993 static void backup_super_roots(struct btrfs_fs_info *info)
1994 {
1995         int next_backup;
1996         struct btrfs_root_backup *root_backup;
1997         int last_backup;
1998
1999         next_backup = info->backup_root_index;
2000         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2001                 BTRFS_NUM_BACKUP_ROOTS;
2002
2003         /*
2004          * just overwrite the last backup if we're at the same generation
2005          * this happens only at umount
2006          */
2007         root_backup = info->super_for_commit->super_roots + last_backup;
2008         if (btrfs_backup_tree_root_gen(root_backup) ==
2009             btrfs_header_generation(info->tree_root->node))
2010                 next_backup = last_backup;
2011
2012         root_backup = info->super_for_commit->super_roots + next_backup;
2013
2014         /*
2015          * make sure all of our padding and empty slots get zero filled
2016          * regardless of which ones we use today
2017          */
2018         memset(root_backup, 0, sizeof(*root_backup));
2019
2020         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2021
2022         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2023         btrfs_set_backup_tree_root_gen(root_backup,
2024                                btrfs_header_generation(info->tree_root->node));
2025
2026         btrfs_set_backup_tree_root_level(root_backup,
2027                                btrfs_header_level(info->tree_root->node));
2028
2029         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2030         btrfs_set_backup_chunk_root_gen(root_backup,
2031                                btrfs_header_generation(info->chunk_root->node));
2032         btrfs_set_backup_chunk_root_level(root_backup,
2033                                btrfs_header_level(info->chunk_root->node));
2034
2035         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2036         btrfs_set_backup_extent_root_gen(root_backup,
2037                                btrfs_header_generation(info->extent_root->node));
2038         btrfs_set_backup_extent_root_level(root_backup,
2039                                btrfs_header_level(info->extent_root->node));
2040
2041         /*
2042          * we might commit during log recovery, which happens before we set
2043          * the fs_root.  Make sure it is valid before we fill it in.
2044          */
2045         if (info->fs_root && info->fs_root->node) {
2046                 btrfs_set_backup_fs_root(root_backup,
2047                                          info->fs_root->node->start);
2048                 btrfs_set_backup_fs_root_gen(root_backup,
2049                                btrfs_header_generation(info->fs_root->node));
2050                 btrfs_set_backup_fs_root_level(root_backup,
2051                                btrfs_header_level(info->fs_root->node));
2052         }
2053
2054         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2055         btrfs_set_backup_dev_root_gen(root_backup,
2056                                btrfs_header_generation(info->dev_root->node));
2057         btrfs_set_backup_dev_root_level(root_backup,
2058                                        btrfs_header_level(info->dev_root->node));
2059
2060         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2061         btrfs_set_backup_csum_root_gen(root_backup,
2062                                btrfs_header_generation(info->csum_root->node));
2063         btrfs_set_backup_csum_root_level(root_backup,
2064                                btrfs_header_level(info->csum_root->node));
2065
2066         btrfs_set_backup_total_bytes(root_backup,
2067                              btrfs_super_total_bytes(info->super_copy));
2068         btrfs_set_backup_bytes_used(root_backup,
2069                              btrfs_super_bytes_used(info->super_copy));
2070         btrfs_set_backup_num_devices(root_backup,
2071                              btrfs_super_num_devices(info->super_copy));
2072
2073         /*
2074          * if we don't copy this out to the super_copy, it won't get remembered
2075          * for the next commit
2076          */
2077         memcpy(&info->super_copy->super_roots,
2078                &info->super_for_commit->super_roots,
2079                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2080 }
2081
2082 /*
2083  * this copies info out of the root backup array and back into
2084  * the in-memory super block.  It is meant to help iterate through
2085  * the array, so you send it the number of backups you've already
2086  * tried and the last backup index you used.
2087  *
2088  * this returns -1 when it has tried all the backups
2089  */
2090 static noinline int next_root_backup(struct btrfs_fs_info *info,
2091                                      struct btrfs_super_block *super,
2092                                      int *num_backups_tried, int *backup_index)
2093 {
2094         struct btrfs_root_backup *root_backup;
2095         int newest = *backup_index;
2096
2097         if (*num_backups_tried == 0) {
2098                 u64 gen = btrfs_super_generation(super);
2099
2100                 newest = find_newest_super_backup(info, gen);
2101                 if (newest == -1)
2102                         return -1;
2103
2104                 *backup_index = newest;
2105                 *num_backups_tried = 1;
2106         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2107                 /* we've tried all the backups, all done */
2108                 return -1;
2109         } else {
2110                 /* jump to the next oldest backup */
2111                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2112                         BTRFS_NUM_BACKUP_ROOTS;
2113                 *backup_index = newest;
2114                 *num_backups_tried += 1;
2115         }
2116         root_backup = super->super_roots + newest;
2117
2118         btrfs_set_super_generation(super,
2119                                    btrfs_backup_tree_root_gen(root_backup));
2120         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2121         btrfs_set_super_root_level(super,
2122                                    btrfs_backup_tree_root_level(root_backup));
2123         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2124
2125         /*
2126          * fixme: the total bytes and num_devices need to match or we should
2127          * need a fsck
2128          */
2129         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2130         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2131         return 0;
2132 }
2133
2134 /* helper to cleanup workers */
2135 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2136 {
2137         btrfs_destroy_workqueue(fs_info->fixup_workers);
2138         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2139         btrfs_destroy_workqueue(fs_info->workers);
2140         btrfs_destroy_workqueue(fs_info->endio_workers);
2141         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2142         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2143         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2144         btrfs_destroy_workqueue(fs_info->rmw_workers);
2145         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2146         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2147         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2148         btrfs_destroy_workqueue(fs_info->submit_workers);
2149         btrfs_destroy_workqueue(fs_info->delayed_workers);
2150         btrfs_destroy_workqueue(fs_info->caching_workers);
2151         btrfs_destroy_workqueue(fs_info->readahead_workers);
2152         btrfs_destroy_workqueue(fs_info->flush_workers);
2153         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2154         btrfs_destroy_workqueue(fs_info->extent_workers);
2155 }
2156
2157 static void free_root_extent_buffers(struct btrfs_root *root)
2158 {
2159         if (root) {
2160                 free_extent_buffer(root->node);
2161                 free_extent_buffer(root->commit_root);
2162                 root->node = NULL;
2163                 root->commit_root = NULL;
2164         }
2165 }
2166
2167 /* helper to cleanup tree roots */
2168 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2169 {
2170         free_root_extent_buffers(info->tree_root);
2171
2172         free_root_extent_buffers(info->dev_root);
2173         free_root_extent_buffers(info->extent_root);
2174         free_root_extent_buffers(info->csum_root);
2175         free_root_extent_buffers(info->quota_root);
2176         free_root_extent_buffers(info->uuid_root);
2177         if (chunk_root)
2178                 free_root_extent_buffers(info->chunk_root);
2179         free_root_extent_buffers(info->free_space_root);
2180 }
2181
2182 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2183 {
2184         int ret;
2185         struct btrfs_root *gang[8];
2186         int i;
2187
2188         while (!list_empty(&fs_info->dead_roots)) {
2189                 gang[0] = list_entry(fs_info->dead_roots.next,
2190                                      struct btrfs_root, root_list);
2191                 list_del(&gang[0]->root_list);
2192
2193                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2194                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2195                 } else {
2196                         free_extent_buffer(gang[0]->node);
2197                         free_extent_buffer(gang[0]->commit_root);
2198                         btrfs_put_fs_root(gang[0]);
2199                 }
2200         }
2201
2202         while (1) {
2203                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2204                                              (void **)gang, 0,
2205                                              ARRAY_SIZE(gang));
2206                 if (!ret)
2207                         break;
2208                 for (i = 0; i < ret; i++)
2209                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2210         }
2211
2212         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2213                 btrfs_free_log_root_tree(NULL, fs_info);
2214                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2215                                             fs_info->pinned_extents);
2216         }
2217 }
2218
2219 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2220 {
2221         mutex_init(&fs_info->scrub_lock);
2222         atomic_set(&fs_info->scrubs_running, 0);
2223         atomic_set(&fs_info->scrub_pause_req, 0);
2224         atomic_set(&fs_info->scrubs_paused, 0);
2225         atomic_set(&fs_info->scrub_cancel_req, 0);
2226         init_waitqueue_head(&fs_info->scrub_pause_wait);
2227         fs_info->scrub_workers_refcnt = 0;
2228 }
2229
2230 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2231 {
2232         spin_lock_init(&fs_info->balance_lock);
2233         mutex_init(&fs_info->balance_mutex);
2234         atomic_set(&fs_info->balance_running, 0);
2235         atomic_set(&fs_info->balance_pause_req, 0);
2236         atomic_set(&fs_info->balance_cancel_req, 0);
2237         fs_info->balance_ctl = NULL;
2238         init_waitqueue_head(&fs_info->balance_wait_q);
2239 }
2240
2241 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2242                                    struct btrfs_root *tree_root)
2243 {
2244         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2245         set_nlink(fs_info->btree_inode, 1);
2246         /*
2247          * we set the i_size on the btree inode to the max possible int.
2248          * the real end of the address space is determined by all of
2249          * the devices in the system
2250          */
2251         fs_info->btree_inode->i_size = OFFSET_MAX;
2252         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2253
2254         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2255         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2256                              fs_info->btree_inode->i_mapping);
2257         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2258         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2259
2260         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2261
2262         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2263         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2264                sizeof(struct btrfs_key));
2265         set_bit(BTRFS_INODE_DUMMY,
2266                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2267         btrfs_insert_inode_hash(fs_info->btree_inode);
2268 }
2269
2270 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2271 {
2272         fs_info->dev_replace.lock_owner = 0;
2273         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2274         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2275         mutex_init(&fs_info->dev_replace.lock_management_lock);
2276         mutex_init(&fs_info->dev_replace.lock);
2277         init_waitqueue_head(&fs_info->replace_wait);
2278 }
2279
2280 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2281 {
2282         spin_lock_init(&fs_info->qgroup_lock);
2283         mutex_init(&fs_info->qgroup_ioctl_lock);
2284         fs_info->qgroup_tree = RB_ROOT;
2285         fs_info->qgroup_op_tree = RB_ROOT;
2286         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2287         fs_info->qgroup_seq = 1;
2288         fs_info->quota_enabled = 0;
2289         fs_info->pending_quota_state = 0;
2290         fs_info->qgroup_ulist = NULL;
2291         mutex_init(&fs_info->qgroup_rescan_lock);
2292 }
2293
2294 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2295                 struct btrfs_fs_devices *fs_devices)
2296 {
2297         int max_active = fs_info->thread_pool_size;
2298         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2299
2300         fs_info->workers =
2301                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2302                                       max_active, 16);
2303
2304         fs_info->delalloc_workers =
2305                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2306
2307         fs_info->flush_workers =
2308                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2309
2310         fs_info->caching_workers =
2311                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2312
2313         /*
2314          * a higher idle thresh on the submit workers makes it much more
2315          * likely that bios will be send down in a sane order to the
2316          * devices
2317          */
2318         fs_info->submit_workers =
2319                 btrfs_alloc_workqueue("submit", flags,
2320                                       min_t(u64, fs_devices->num_devices,
2321                                             max_active), 64);
2322
2323         fs_info->fixup_workers =
2324                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2325
2326         /*
2327          * endios are largely parallel and should have a very
2328          * low idle thresh
2329          */
2330         fs_info->endio_workers =
2331                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2332         fs_info->endio_meta_workers =
2333                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2334         fs_info->endio_meta_write_workers =
2335                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2336         fs_info->endio_raid56_workers =
2337                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2338         fs_info->endio_repair_workers =
2339                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2340         fs_info->rmw_workers =
2341                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2342         fs_info->endio_write_workers =
2343                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2344         fs_info->endio_freespace_worker =
2345                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2346         fs_info->delayed_workers =
2347                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2348         fs_info->readahead_workers =
2349                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2350         fs_info->qgroup_rescan_workers =
2351                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2352         fs_info->extent_workers =
2353                 btrfs_alloc_workqueue("extent-refs", flags,
2354                                       min_t(u64, fs_devices->num_devices,
2355                                             max_active), 8);
2356
2357         if (!(fs_info->workers && fs_info->delalloc_workers &&
2358               fs_info->submit_workers && fs_info->flush_workers &&
2359               fs_info->endio_workers && fs_info->endio_meta_workers &&
2360               fs_info->endio_meta_write_workers &&
2361               fs_info->endio_repair_workers &&
2362               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2363               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2364               fs_info->caching_workers && fs_info->readahead_workers &&
2365               fs_info->fixup_workers && fs_info->delayed_workers &&
2366               fs_info->extent_workers &&
2367               fs_info->qgroup_rescan_workers)) {
2368                 return -ENOMEM;
2369         }
2370
2371         return 0;
2372 }
2373
2374 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2375                             struct btrfs_fs_devices *fs_devices)
2376 {
2377         int ret;
2378         struct btrfs_root *tree_root = fs_info->tree_root;
2379         struct btrfs_root *log_tree_root;
2380         struct btrfs_super_block *disk_super = fs_info->super_copy;
2381         u64 bytenr = btrfs_super_log_root(disk_super);
2382
2383         if (fs_devices->rw_devices == 0) {
2384                 btrfs_warn(fs_info, "log replay required on RO media");
2385                 return -EIO;
2386         }
2387
2388         log_tree_root = btrfs_alloc_root(fs_info);
2389         if (!log_tree_root)
2390                 return -ENOMEM;
2391
2392         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2393                         tree_root->stripesize, log_tree_root, fs_info,
2394                         BTRFS_TREE_LOG_OBJECTID);
2395
2396         log_tree_root->node = read_tree_block(tree_root, bytenr,
2397                         fs_info->generation + 1);
2398         if (IS_ERR(log_tree_root->node)) {
2399                 btrfs_warn(fs_info, "failed to read log tree");
2400                 ret = PTR_ERR(log_tree_root->node);
2401                 kfree(log_tree_root);
2402                 return ret;
2403         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2404                 btrfs_err(fs_info, "failed to read log tree");
2405                 free_extent_buffer(log_tree_root->node);
2406                 kfree(log_tree_root);
2407                 return -EIO;
2408         }
2409         /* returns with log_tree_root freed on success */
2410         ret = btrfs_recover_log_trees(log_tree_root);
2411         if (ret) {
2412                 btrfs_std_error(tree_root->fs_info, ret,
2413                             "Failed to recover log tree");
2414                 free_extent_buffer(log_tree_root->node);
2415                 kfree(log_tree_root);
2416                 return ret;
2417         }
2418
2419         if (fs_info->sb->s_flags & MS_RDONLY) {
2420                 ret = btrfs_commit_super(tree_root);
2421                 if (ret)
2422                         return ret;
2423         }
2424
2425         return 0;
2426 }
2427
2428 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2429                             struct btrfs_root *tree_root)
2430 {
2431         struct btrfs_root *root;
2432         struct btrfs_key location;
2433         int ret;
2434
2435         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2436         location.type = BTRFS_ROOT_ITEM_KEY;
2437         location.offset = 0;
2438
2439         root = btrfs_read_tree_root(tree_root, &location);
2440         if (IS_ERR(root))
2441                 return PTR_ERR(root);
2442         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2443         fs_info->extent_root = root;
2444
2445         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2446         root = btrfs_read_tree_root(tree_root, &location);
2447         if (IS_ERR(root))
2448                 return PTR_ERR(root);
2449         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2450         fs_info->dev_root = root;
2451         btrfs_init_devices_late(fs_info);
2452
2453         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2454         root = btrfs_read_tree_root(tree_root, &location);
2455         if (IS_ERR(root))
2456                 return PTR_ERR(root);
2457         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2458         fs_info->csum_root = root;
2459
2460         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2461         root = btrfs_read_tree_root(tree_root, &location);
2462         if (!IS_ERR(root)) {
2463                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2464                 fs_info->quota_enabled = 1;
2465                 fs_info->pending_quota_state = 1;
2466                 fs_info->quota_root = root;
2467         }
2468
2469         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2470         root = btrfs_read_tree_root(tree_root, &location);
2471         if (IS_ERR(root)) {
2472                 ret = PTR_ERR(root);
2473                 if (ret != -ENOENT)
2474                         return ret;
2475         } else {
2476                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2477                 fs_info->uuid_root = root;
2478         }
2479
2480         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2481                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2482                 root = btrfs_read_tree_root(tree_root, &location);
2483                 if (IS_ERR(root))
2484                         return PTR_ERR(root);
2485                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2486                 fs_info->free_space_root = root;
2487         }
2488
2489         return 0;
2490 }
2491
2492 int open_ctree(struct super_block *sb,
2493                struct btrfs_fs_devices *fs_devices,
2494                char *options)
2495 {
2496         u32 sectorsize;
2497         u32 nodesize;
2498         u32 stripesize;
2499         u64 generation;
2500         u64 features;
2501         struct btrfs_key location;
2502         struct buffer_head *bh;
2503         struct btrfs_super_block *disk_super;
2504         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2505         struct btrfs_root *tree_root;
2506         struct btrfs_root *chunk_root;
2507         int ret;
2508         int err = -EINVAL;
2509         int num_backups_tried = 0;
2510         int backup_index = 0;
2511         int max_active;
2512
2513         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2514         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2515         if (!tree_root || !chunk_root) {
2516                 err = -ENOMEM;
2517                 goto fail;
2518         }
2519
2520         ret = init_srcu_struct(&fs_info->subvol_srcu);
2521         if (ret) {
2522                 err = ret;
2523                 goto fail;
2524         }
2525
2526         ret = setup_bdi(fs_info, &fs_info->bdi);
2527         if (ret) {
2528                 err = ret;
2529                 goto fail_srcu;
2530         }
2531
2532         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2533         if (ret) {
2534                 err = ret;
2535                 goto fail_bdi;
2536         }
2537         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2538                                         (1 + ilog2(nr_cpu_ids));
2539
2540         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2541         if (ret) {
2542                 err = ret;
2543                 goto fail_dirty_metadata_bytes;
2544         }
2545
2546         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2547         if (ret) {
2548                 err = ret;
2549                 goto fail_delalloc_bytes;
2550         }
2551
2552         fs_info->btree_inode = new_inode(sb);
2553         if (!fs_info->btree_inode) {
2554                 err = -ENOMEM;
2555                 goto fail_bio_counter;
2556         }
2557
2558         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2559
2560         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2561         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2562         INIT_LIST_HEAD(&fs_info->trans_list);
2563         INIT_LIST_HEAD(&fs_info->dead_roots);
2564         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2565         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2566         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2567         spin_lock_init(&fs_info->delalloc_root_lock);
2568         spin_lock_init(&fs_info->trans_lock);
2569         spin_lock_init(&fs_info->fs_roots_radix_lock);
2570         spin_lock_init(&fs_info->delayed_iput_lock);
2571         spin_lock_init(&fs_info->defrag_inodes_lock);
2572         spin_lock_init(&fs_info->free_chunk_lock);
2573         spin_lock_init(&fs_info->tree_mod_seq_lock);
2574         spin_lock_init(&fs_info->super_lock);
2575         spin_lock_init(&fs_info->qgroup_op_lock);
2576         spin_lock_init(&fs_info->buffer_lock);
2577         spin_lock_init(&fs_info->unused_bgs_lock);
2578         rwlock_init(&fs_info->tree_mod_log_lock);
2579         mutex_init(&fs_info->unused_bg_unpin_mutex);
2580         mutex_init(&fs_info->delete_unused_bgs_mutex);
2581         mutex_init(&fs_info->reloc_mutex);
2582         mutex_init(&fs_info->delalloc_root_mutex);
2583         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2584         seqlock_init(&fs_info->profiles_lock);
2585
2586         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2587         INIT_LIST_HEAD(&fs_info->space_info);
2588         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2589         INIT_LIST_HEAD(&fs_info->unused_bgs);
2590         btrfs_mapping_init(&fs_info->mapping_tree);
2591         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2592                              BTRFS_BLOCK_RSV_GLOBAL);
2593         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2594                              BTRFS_BLOCK_RSV_DELALLOC);
2595         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2596         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2597         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2598         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2599                              BTRFS_BLOCK_RSV_DELOPS);
2600         atomic_set(&fs_info->nr_async_submits, 0);
2601         atomic_set(&fs_info->async_delalloc_pages, 0);
2602         atomic_set(&fs_info->async_submit_draining, 0);
2603         atomic_set(&fs_info->nr_async_bios, 0);
2604         atomic_set(&fs_info->defrag_running, 0);
2605         atomic_set(&fs_info->qgroup_op_seq, 0);
2606         atomic64_set(&fs_info->tree_mod_seq, 0);
2607         fs_info->sb = sb;
2608         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2609         fs_info->metadata_ratio = 0;
2610         fs_info->defrag_inodes = RB_ROOT;
2611         fs_info->free_chunk_space = 0;
2612         fs_info->tree_mod_log = RB_ROOT;
2613         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2614         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2615         /* readahead state */
2616         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2617         spin_lock_init(&fs_info->reada_lock);
2618
2619         fs_info->thread_pool_size = min_t(unsigned long,
2620                                           num_online_cpus() + 2, 8);
2621
2622         INIT_LIST_HEAD(&fs_info->ordered_roots);
2623         spin_lock_init(&fs_info->ordered_root_lock);
2624         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2625                                         GFP_NOFS);
2626         if (!fs_info->delayed_root) {
2627                 err = -ENOMEM;
2628                 goto fail_iput;
2629         }
2630         btrfs_init_delayed_root(fs_info->delayed_root);
2631
2632         btrfs_init_scrub(fs_info);
2633 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2634         fs_info->check_integrity_print_mask = 0;
2635 #endif
2636         btrfs_init_balance(fs_info);
2637         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2638
2639         sb->s_blocksize = 4096;
2640         sb->s_blocksize_bits = blksize_bits(4096);
2641         sb->s_bdi = &fs_info->bdi;
2642
2643         btrfs_init_btree_inode(fs_info, tree_root);
2644
2645         spin_lock_init(&fs_info->block_group_cache_lock);
2646         fs_info->block_group_cache_tree = RB_ROOT;
2647         fs_info->first_logical_byte = (u64)-1;
2648
2649         extent_io_tree_init(&fs_info->freed_extents[0],
2650                              fs_info->btree_inode->i_mapping);
2651         extent_io_tree_init(&fs_info->freed_extents[1],
2652                              fs_info->btree_inode->i_mapping);
2653         fs_info->pinned_extents = &fs_info->freed_extents[0];
2654         fs_info->do_barriers = 1;
2655
2656
2657         mutex_init(&fs_info->ordered_operations_mutex);
2658         mutex_init(&fs_info->tree_log_mutex);
2659         mutex_init(&fs_info->chunk_mutex);
2660         mutex_init(&fs_info->transaction_kthread_mutex);
2661         mutex_init(&fs_info->cleaner_mutex);
2662         mutex_init(&fs_info->volume_mutex);
2663         mutex_init(&fs_info->ro_block_group_mutex);
2664         init_rwsem(&fs_info->commit_root_sem);
2665         init_rwsem(&fs_info->cleanup_work_sem);
2666         init_rwsem(&fs_info->subvol_sem);
2667         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2668
2669         btrfs_init_dev_replace_locks(fs_info);
2670         btrfs_init_qgroup(fs_info);
2671
2672         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2673         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2674
2675         init_waitqueue_head(&fs_info->transaction_throttle);
2676         init_waitqueue_head(&fs_info->transaction_wait);
2677         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2678         init_waitqueue_head(&fs_info->async_submit_wait);
2679
2680         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2681
2682         ret = btrfs_alloc_stripe_hash_table(fs_info);
2683         if (ret) {
2684                 err = ret;
2685                 goto fail_alloc;
2686         }
2687
2688         __setup_root(4096, 4096, 4096, tree_root,
2689                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2690
2691         invalidate_bdev(fs_devices->latest_bdev);
2692
2693         /*
2694          * Read super block and check the signature bytes only
2695          */
2696         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2697         if (IS_ERR(bh)) {
2698                 err = PTR_ERR(bh);
2699                 goto fail_alloc;
2700         }
2701
2702         /*
2703          * We want to check superblock checksum, the type is stored inside.
2704          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2705          */
2706         if (btrfs_check_super_csum(bh->b_data)) {
2707                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2708                 err = -EINVAL;
2709                 brelse(bh);
2710                 goto fail_alloc;
2711         }
2712
2713         /*
2714          * super_copy is zeroed at allocation time and we never touch the
2715          * following bytes up to INFO_SIZE, the checksum is calculated from
2716          * the whole block of INFO_SIZE
2717          */
2718         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2719         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2720                sizeof(*fs_info->super_for_commit));
2721         brelse(bh);
2722
2723         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2724
2725         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2726         if (ret) {
2727                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2728                 err = -EINVAL;
2729                 goto fail_alloc;
2730         }
2731
2732         disk_super = fs_info->super_copy;
2733         if (!btrfs_super_root(disk_super))
2734                 goto fail_alloc;
2735
2736         /* check FS state, whether FS is broken. */
2737         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2738                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2739
2740         /*
2741          * run through our array of backup supers and setup
2742          * our ring pointer to the oldest one
2743          */
2744         generation = btrfs_super_generation(disk_super);
2745         find_oldest_super_backup(fs_info, generation);
2746
2747         /*
2748          * In the long term, we'll store the compression type in the super
2749          * block, and it'll be used for per file compression control.
2750          */
2751         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2752
2753         ret = btrfs_parse_options(tree_root, options);
2754         if (ret) {
2755                 err = ret;
2756                 goto fail_alloc;
2757         }
2758
2759         features = btrfs_super_incompat_flags(disk_super) &
2760                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2761         if (features) {
2762                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2763                        "unsupported optional features (%Lx).\n",
2764                        features);
2765                 err = -EINVAL;
2766                 goto fail_alloc;
2767         }
2768
2769         features = btrfs_super_incompat_flags(disk_super);
2770         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2771         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2772                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2773
2774         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2775                 printk(KERN_INFO "BTRFS: has skinny extents\n");
2776
2777         /*
2778          * flag our filesystem as having big metadata blocks if
2779          * they are bigger than the page size
2780          */
2781         if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2782                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2783                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2784                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2785         }
2786
2787         nodesize = btrfs_super_nodesize(disk_super);
2788         sectorsize = btrfs_super_sectorsize(disk_super);
2789         stripesize = btrfs_super_stripesize(disk_super);
2790         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2791         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2792
2793         /*
2794          * mixed block groups end up with duplicate but slightly offset
2795          * extent buffers for the same range.  It leads to corruptions
2796          */
2797         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2798             (sectorsize != nodesize)) {
2799                 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2800                                 "are not allowed for mixed block groups on %s\n",
2801                                 sb->s_id);
2802                 goto fail_alloc;
2803         }
2804
2805         /*
2806          * Needn't use the lock because there is no other task which will
2807          * update the flag.
2808          */
2809         btrfs_set_super_incompat_flags(disk_super, features);
2810
2811         features = btrfs_super_compat_ro_flags(disk_super) &
2812                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2813         if (!(sb->s_flags & MS_RDONLY) && features) {
2814                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2815                        "unsupported option features (%Lx).\n",
2816                        features);
2817                 err = -EINVAL;
2818                 goto fail_alloc;
2819         }
2820
2821         max_active = fs_info->thread_pool_size;
2822
2823         ret = btrfs_init_workqueues(fs_info, fs_devices);
2824         if (ret) {
2825                 err = ret;
2826                 goto fail_sb_buffer;
2827         }
2828
2829         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2830         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2831                                     SZ_4M / PAGE_CACHE_SIZE);
2832
2833         tree_root->nodesize = nodesize;
2834         tree_root->sectorsize = sectorsize;
2835         tree_root->stripesize = stripesize;
2836
2837         sb->s_blocksize = sectorsize;
2838         sb->s_blocksize_bits = blksize_bits(sectorsize);
2839
2840         mutex_lock(&fs_info->chunk_mutex);
2841         ret = btrfs_read_sys_array(tree_root);
2842         mutex_unlock(&fs_info->chunk_mutex);
2843         if (ret) {
2844                 printk(KERN_ERR "BTRFS: failed to read the system "
2845                        "array on %s\n", sb->s_id);
2846                 goto fail_sb_buffer;
2847         }
2848
2849         generation = btrfs_super_chunk_root_generation(disk_super);
2850
2851         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2852                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2853
2854         chunk_root->node = read_tree_block(chunk_root,
2855                                            btrfs_super_chunk_root(disk_super),
2856                                            generation);
2857         if (IS_ERR(chunk_root->node) ||
2858             !extent_buffer_uptodate(chunk_root->node)) {
2859                 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2860                        sb->s_id);
2861                 if (!IS_ERR(chunk_root->node))
2862                         free_extent_buffer(chunk_root->node);
2863                 chunk_root->node = NULL;
2864                 goto fail_tree_roots;
2865         }
2866         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2867         chunk_root->commit_root = btrfs_root_node(chunk_root);
2868
2869         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2870            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2871
2872         ret = btrfs_read_chunk_tree(chunk_root);
2873         if (ret) {
2874                 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2875                        sb->s_id);
2876                 goto fail_tree_roots;
2877         }
2878
2879         /*
2880          * keep the device that is marked to be the target device for the
2881          * dev_replace procedure
2882          */
2883         btrfs_close_extra_devices(fs_devices, 0);
2884
2885         if (!fs_devices->latest_bdev) {
2886                 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2887                        sb->s_id);
2888                 goto fail_tree_roots;
2889         }
2890
2891 retry_root_backup:
2892         generation = btrfs_super_generation(disk_super);
2893
2894         tree_root->node = read_tree_block(tree_root,
2895                                           btrfs_super_root(disk_super),
2896                                           generation);
2897         if (IS_ERR(tree_root->node) ||
2898             !extent_buffer_uptodate(tree_root->node)) {
2899                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2900                        sb->s_id);
2901                 if (!IS_ERR(tree_root->node))
2902                         free_extent_buffer(tree_root->node);
2903                 tree_root->node = NULL;
2904                 goto recovery_tree_root;
2905         }
2906
2907         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2908         tree_root->commit_root = btrfs_root_node(tree_root);
2909         btrfs_set_root_refs(&tree_root->root_item, 1);
2910
2911         mutex_lock(&tree_root->objectid_mutex);
2912         ret = btrfs_find_highest_objectid(tree_root,
2913                                         &tree_root->highest_objectid);
2914         if (ret) {
2915                 mutex_unlock(&tree_root->objectid_mutex);
2916                 goto recovery_tree_root;
2917         }
2918
2919         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2920
2921         mutex_unlock(&tree_root->objectid_mutex);
2922
2923         ret = btrfs_read_roots(fs_info, tree_root);
2924         if (ret)
2925                 goto recovery_tree_root;
2926
2927         fs_info->generation = generation;
2928         fs_info->last_trans_committed = generation;
2929
2930         ret = btrfs_recover_balance(fs_info);
2931         if (ret) {
2932                 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2933                 goto fail_block_groups;
2934         }
2935
2936         ret = btrfs_init_dev_stats(fs_info);
2937         if (ret) {
2938                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2939                        ret);
2940                 goto fail_block_groups;
2941         }
2942
2943         ret = btrfs_init_dev_replace(fs_info);
2944         if (ret) {
2945                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2946                 goto fail_block_groups;
2947         }
2948
2949         btrfs_close_extra_devices(fs_devices, 1);
2950
2951         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2952         if (ret) {
2953                 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2954                 goto fail_block_groups;
2955         }
2956
2957         ret = btrfs_sysfs_add_device(fs_devices);
2958         if (ret) {
2959                 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2960                 goto fail_fsdev_sysfs;
2961         }
2962
2963         ret = btrfs_sysfs_add_mounted(fs_info);
2964         if (ret) {
2965                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2966                 goto fail_fsdev_sysfs;
2967         }
2968
2969         ret = btrfs_init_space_info(fs_info);
2970         if (ret) {
2971                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2972                 goto fail_sysfs;
2973         }
2974
2975         ret = btrfs_read_block_groups(fs_info->extent_root);
2976         if (ret) {
2977                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2978                 goto fail_sysfs;
2979         }
2980         fs_info->num_tolerated_disk_barrier_failures =
2981                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2982         if (fs_info->fs_devices->missing_devices >
2983              fs_info->num_tolerated_disk_barrier_failures &&
2984             !(sb->s_flags & MS_RDONLY)) {
2985                 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2986                         fs_info->fs_devices->missing_devices,
2987                         fs_info->num_tolerated_disk_barrier_failures);
2988                 goto fail_sysfs;
2989         }
2990
2991         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2992                                                "btrfs-cleaner");
2993         if (IS_ERR(fs_info->cleaner_kthread))
2994                 goto fail_sysfs;
2995
2996         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2997                                                    tree_root,
2998                                                    "btrfs-transaction");
2999         if (IS_ERR(fs_info->transaction_kthread))
3000                 goto fail_cleaner;
3001
3002         if (!btrfs_test_opt(tree_root, SSD) &&
3003             !btrfs_test_opt(tree_root, NOSSD) &&
3004             !fs_info->fs_devices->rotating) {
3005                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3006                        "mode\n");
3007                 btrfs_set_opt(fs_info->mount_opt, SSD);
3008         }
3009
3010         /*
3011          * Mount does not set all options immediatelly, we can do it now and do
3012          * not have to wait for transaction commit
3013          */
3014         btrfs_apply_pending_changes(fs_info);
3015
3016 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3017         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3018                 ret = btrfsic_mount(tree_root, fs_devices,
3019                                     btrfs_test_opt(tree_root,
3020                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3021                                     1 : 0,
3022                                     fs_info->check_integrity_print_mask);
3023                 if (ret)
3024                         printk(KERN_WARNING "BTRFS: failed to initialize"
3025                                " integrity check module %s\n", sb->s_id);
3026         }
3027 #endif
3028         ret = btrfs_read_qgroup_config(fs_info);
3029         if (ret)
3030                 goto fail_trans_kthread;
3031
3032         /* do not make disk changes in broken FS */
3033         if (btrfs_super_log_root(disk_super) != 0) {
3034                 ret = btrfs_replay_log(fs_info, fs_devices);
3035                 if (ret) {
3036                         err = ret;
3037                         goto fail_qgroup;
3038                 }
3039         }
3040
3041         ret = btrfs_find_orphan_roots(tree_root);
3042         if (ret)
3043                 goto fail_qgroup;
3044
3045         if (!(sb->s_flags & MS_RDONLY)) {
3046                 ret = btrfs_cleanup_fs_roots(fs_info);
3047                 if (ret)
3048                         goto fail_qgroup;
3049
3050                 mutex_lock(&fs_info->cleaner_mutex);
3051                 ret = btrfs_recover_relocation(tree_root);
3052                 mutex_unlock(&fs_info->cleaner_mutex);
3053                 if (ret < 0) {
3054                         printk(KERN_WARNING
3055                                "BTRFS: failed to recover relocation\n");
3056                         err = -EINVAL;
3057                         goto fail_qgroup;
3058                 }
3059         }
3060
3061         location.objectid = BTRFS_FS_TREE_OBJECTID;
3062         location.type = BTRFS_ROOT_ITEM_KEY;
3063         location.offset = 0;
3064
3065         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3066         if (IS_ERR(fs_info->fs_root)) {
3067                 err = PTR_ERR(fs_info->fs_root);
3068                 goto fail_qgroup;
3069         }
3070
3071         if (sb->s_flags & MS_RDONLY)
3072                 return 0;
3073
3074         if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3075             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3076                 pr_info("BTRFS: creating free space tree\n");
3077                 ret = btrfs_create_free_space_tree(fs_info);
3078                 if (ret) {
3079                         pr_warn("BTRFS: failed to create free space tree %d\n",
3080                                 ret);
3081                         close_ctree(tree_root);
3082                         return ret;
3083                 }
3084         }
3085
3086         down_read(&fs_info->cleanup_work_sem);
3087         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3088             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3089                 up_read(&fs_info->cleanup_work_sem);
3090                 close_ctree(tree_root);
3091                 return ret;
3092         }
3093         up_read(&fs_info->cleanup_work_sem);
3094
3095         ret = btrfs_resume_balance_async(fs_info);
3096         if (ret) {
3097                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3098                 close_ctree(tree_root);
3099                 return ret;
3100         }
3101
3102         ret = btrfs_resume_dev_replace_async(fs_info);
3103         if (ret) {
3104                 pr_warn("BTRFS: failed to resume dev_replace\n");
3105                 close_ctree(tree_root);
3106                 return ret;
3107         }
3108
3109         btrfs_qgroup_rescan_resume(fs_info);
3110
3111         if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3112             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3113                 pr_info("BTRFS: clearing free space tree\n");
3114                 ret = btrfs_clear_free_space_tree(fs_info);
3115                 if (ret) {
3116                         pr_warn("BTRFS: failed to clear free space tree %d\n",
3117                                 ret);
3118                         close_ctree(tree_root);
3119                         return ret;
3120                 }
3121         }
3122
3123         if (!fs_info->uuid_root) {
3124                 pr_info("BTRFS: creating UUID tree\n");
3125                 ret = btrfs_create_uuid_tree(fs_info);
3126                 if (ret) {
3127                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
3128                                 ret);
3129                         close_ctree(tree_root);
3130                         return ret;
3131                 }
3132         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3133                    fs_info->generation !=
3134                                 btrfs_super_uuid_tree_generation(disk_super)) {
3135                 pr_info("BTRFS: checking UUID tree\n");
3136                 ret = btrfs_check_uuid_tree(fs_info);
3137                 if (ret) {
3138                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
3139                                 ret);
3140                         close_ctree(tree_root);
3141                         return ret;
3142                 }
3143         } else {
3144                 fs_info->update_uuid_tree_gen = 1;
3145         }
3146
3147         fs_info->open = 1;
3148
3149         return 0;
3150
3151 fail_qgroup:
3152         btrfs_free_qgroup_config(fs_info);
3153 fail_trans_kthread:
3154         kthread_stop(fs_info->transaction_kthread);
3155         btrfs_cleanup_transaction(fs_info->tree_root);
3156         btrfs_free_fs_roots(fs_info);
3157 fail_cleaner:
3158         kthread_stop(fs_info->cleaner_kthread);
3159
3160         /*
3161          * make sure we're done with the btree inode before we stop our
3162          * kthreads
3163          */
3164         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3165
3166 fail_sysfs:
3167         btrfs_sysfs_remove_mounted(fs_info);
3168
3169 fail_fsdev_sysfs:
3170         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3171
3172 fail_block_groups:
3173         btrfs_put_block_group_cache(fs_info);
3174         btrfs_free_block_groups(fs_info);
3175
3176 fail_tree_roots:
3177         free_root_pointers(fs_info, 1);
3178         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3179
3180 fail_sb_buffer:
3181         btrfs_stop_all_workers(fs_info);
3182 fail_alloc:
3183 fail_iput:
3184         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3185
3186         iput(fs_info->btree_inode);
3187 fail_bio_counter:
3188         percpu_counter_destroy(&fs_info->bio_counter);
3189 fail_delalloc_bytes:
3190         percpu_counter_destroy(&fs_info->delalloc_bytes);
3191 fail_dirty_metadata_bytes:
3192         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3193 fail_bdi:
3194         bdi_destroy(&fs_info->bdi);
3195 fail_srcu:
3196         cleanup_srcu_struct(&fs_info->subvol_srcu);
3197 fail:
3198         btrfs_free_stripe_hash_table(fs_info);
3199         btrfs_close_devices(fs_info->fs_devices);
3200         return err;
3201
3202 recovery_tree_root:
3203         if (!btrfs_test_opt(tree_root, RECOVERY))
3204                 goto fail_tree_roots;
3205
3206         free_root_pointers(fs_info, 0);
3207
3208         /* don't use the log in recovery mode, it won't be valid */
3209         btrfs_set_super_log_root(disk_super, 0);
3210
3211         /* we can't trust the free space cache either */
3212         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3213
3214         ret = next_root_backup(fs_info, fs_info->super_copy,
3215                                &num_backups_tried, &backup_index);
3216         if (ret == -1)
3217                 goto fail_block_groups;
3218         goto retry_root_backup;
3219 }
3220
3221 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3222 {
3223         if (uptodate) {
3224                 set_buffer_uptodate(bh);
3225         } else {
3226                 struct btrfs_device *device = (struct btrfs_device *)
3227                         bh->b_private;
3228
3229                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3230                                 "lost page write due to IO error on %s",
3231                                           rcu_str_deref(device->name));
3232                 /* note, we dont' set_buffer_write_io_error because we have
3233                  * our own ways of dealing with the IO errors
3234                  */
3235                 clear_buffer_uptodate(bh);
3236                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3237         }
3238         unlock_buffer(bh);
3239         put_bh(bh);
3240 }
3241
3242 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3243                         struct buffer_head **bh_ret)
3244 {
3245         struct buffer_head *bh;
3246         struct btrfs_super_block *super;
3247         u64 bytenr;
3248
3249         bytenr = btrfs_sb_offset(copy_num);
3250         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3251                 return -EINVAL;
3252
3253         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3254         /*
3255          * If we fail to read from the underlying devices, as of now
3256          * the best option we have is to mark it EIO.
3257          */
3258         if (!bh)
3259                 return -EIO;
3260
3261         super = (struct btrfs_super_block *)bh->b_data;
3262         if (btrfs_super_bytenr(super) != bytenr ||
3263                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3264                 brelse(bh);
3265                 return -EINVAL;
3266         }
3267
3268         *bh_ret = bh;
3269         return 0;
3270 }
3271
3272
3273 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3274 {
3275         struct buffer_head *bh;
3276         struct buffer_head *latest = NULL;
3277         struct btrfs_super_block *super;
3278         int i;
3279         u64 transid = 0;
3280         int ret = -EINVAL;
3281
3282         /* we would like to check all the supers, but that would make
3283          * a btrfs mount succeed after a mkfs from a different FS.
3284          * So, we need to add a special mount option to scan for
3285          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3286          */
3287         for (i = 0; i < 1; i++) {
3288                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3289                 if (ret)
3290                         continue;
3291
3292                 super = (struct btrfs_super_block *)bh->b_data;
3293
3294                 if (!latest || btrfs_super_generation(super) > transid) {
3295                         brelse(latest);
3296                         latest = bh;
3297                         transid = btrfs_super_generation(super);
3298                 } else {
3299                         brelse(bh);
3300                 }
3301         }
3302
3303         if (!latest)
3304                 return ERR_PTR(ret);
3305
3306         return latest;
3307 }
3308
3309 /*
3310  * this should be called twice, once with wait == 0 and
3311  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3312  * we write are pinned.
3313  *
3314  * They are released when wait == 1 is done.
3315  * max_mirrors must be the same for both runs, and it indicates how
3316  * many supers on this one device should be written.
3317  *
3318  * max_mirrors == 0 means to write them all.
3319  */
3320 static int write_dev_supers(struct btrfs_device *device,
3321                             struct btrfs_super_block *sb,
3322                             int do_barriers, int wait, int max_mirrors)
3323 {
3324         struct buffer_head *bh;
3325         int i;
3326         int ret;
3327         int errors = 0;
3328         u32 crc;
3329         u64 bytenr;
3330
3331         if (max_mirrors == 0)
3332                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3333
3334         for (i = 0; i < max_mirrors; i++) {
3335                 bytenr = btrfs_sb_offset(i);
3336                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3337                     device->commit_total_bytes)
3338                         break;
3339
3340                 if (wait) {
3341                         bh = __find_get_block(device->bdev, bytenr / 4096,
3342                                               BTRFS_SUPER_INFO_SIZE);
3343                         if (!bh) {
3344                                 errors++;
3345                                 continue;
3346                         }
3347                         wait_on_buffer(bh);
3348                         if (!buffer_uptodate(bh))
3349                                 errors++;
3350
3351                         /* drop our reference */
3352                         brelse(bh);
3353
3354                         /* drop the reference from the wait == 0 run */
3355                         brelse(bh);
3356                         continue;
3357                 } else {
3358                         btrfs_set_super_bytenr(sb, bytenr);
3359
3360                         crc = ~(u32)0;
3361                         crc = btrfs_csum_data((char *)sb +
3362                                               BTRFS_CSUM_SIZE, crc,
3363                                               BTRFS_SUPER_INFO_SIZE -
3364                                               BTRFS_CSUM_SIZE);
3365                         btrfs_csum_final(crc, sb->csum);
3366
3367                         /*
3368                          * one reference for us, and we leave it for the
3369                          * caller
3370                          */
3371                         bh = __getblk(device->bdev, bytenr / 4096,
3372                                       BTRFS_SUPER_INFO_SIZE);
3373                         if (!bh) {
3374                                 btrfs_err(device->dev_root->fs_info,
3375                                     "couldn't get super buffer head for bytenr %llu",
3376                                     bytenr);
3377                                 errors++;
3378                                 continue;
3379                         }
3380
3381                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3382
3383                         /* one reference for submit_bh */
3384                         get_bh(bh);
3385
3386                         set_buffer_uptodate(bh);
3387                         lock_buffer(bh);
3388                         bh->b_end_io = btrfs_end_buffer_write_sync;
3389                         bh->b_private = device;
3390                 }
3391
3392                 /*
3393                  * we fua the first super.  The others we allow
3394                  * to go down lazy.
3395                  */
3396                 if (i == 0)
3397                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3398                 else
3399                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3400                 if (ret)
3401                         errors++;
3402         }
3403         return errors < i ? 0 : -1;
3404 }
3405
3406 /*
3407  * endio for the write_dev_flush, this will wake anyone waiting
3408  * for the barrier when it is done
3409  */
3410 static void btrfs_end_empty_barrier(struct bio *bio)
3411 {
3412         if (bio->bi_private)
3413                 complete(bio->bi_private);
3414         bio_put(bio);
3415 }
3416
3417 /*
3418  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3419  * sent down.  With wait == 1, it waits for the previous flush.
3420  *
3421  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3422  * capable
3423  */
3424 static int write_dev_flush(struct btrfs_device *device, int wait)
3425 {
3426         struct bio *bio;
3427         int ret = 0;
3428
3429         if (device->nobarriers)
3430                 return 0;
3431
3432         if (wait) {
3433                 bio = device->flush_bio;
3434                 if (!bio)
3435                         return 0;
3436
3437                 wait_for_completion(&device->flush_wait);
3438
3439                 if (bio->bi_error) {
3440                         ret = bio->bi_error;
3441                         btrfs_dev_stat_inc_and_print(device,
3442                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3443                 }
3444
3445                 /* drop the reference from the wait == 0 run */
3446                 bio_put(bio);
3447                 device->flush_bio = NULL;
3448
3449                 return ret;
3450         }
3451
3452         /*
3453          * one reference for us, and we leave it for the
3454          * caller
3455          */
3456         device->flush_bio = NULL;
3457         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3458         if (!bio)
3459                 return -ENOMEM;
3460
3461         bio->bi_end_io = btrfs_end_empty_barrier;
3462         bio->bi_bdev = device->bdev;
3463         init_completion(&device->flush_wait);
3464         bio->bi_private = &device->flush_wait;
3465         device->flush_bio = bio;
3466
3467         bio_get(bio);
3468         btrfsic_submit_bio(WRITE_FLUSH, bio);
3469
3470         return 0;
3471 }
3472
3473 /*
3474  * send an empty flush down to each device in parallel,
3475  * then wait for them
3476  */
3477 static int barrier_all_devices(struct btrfs_fs_info *info)
3478 {
3479         struct list_head *head;
3480         struct btrfs_device *dev;
3481         int errors_send = 0;
3482         int errors_wait = 0;
3483         int ret;
3484
3485         /* send down all the barriers */
3486         head = &info->fs_devices->devices;
3487         list_for_each_entry_rcu(dev, head, dev_list) {
3488                 if (dev->missing)
3489                         continue;
3490                 if (!dev->bdev) {
3491                         errors_send++;
3492                         continue;
3493                 }
3494                 if (!dev->in_fs_metadata || !dev->writeable)
3495                         continue;
3496
3497                 ret = write_dev_flush(dev, 0);
3498                 if (ret)
3499                         errors_send++;
3500         }
3501
3502         /* wait for all the barriers */
3503         list_for_each_entry_rcu(dev, head, dev_list) {
3504                 if (dev->missing)
3505                         continue;
3506                 if (!dev->bdev) {
3507                         errors_wait++;
3508                         continue;
3509                 }
3510                 if (!dev->in_fs_metadata || !dev->writeable)
3511                         continue;
3512
3513                 ret = write_dev_flush(dev, 1);
3514                 if (ret)
3515                         errors_wait++;
3516         }
3517         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3518             errors_wait > info->num_tolerated_disk_barrier_failures)
3519                 return -EIO;
3520         return 0;
3521 }
3522
3523 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3524 {
3525         int raid_type;
3526         int min_tolerated = INT_MAX;
3527
3528         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3529             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3530                 min_tolerated = min(min_tolerated,
3531                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3532                                     tolerated_failures);
3533
3534         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3535                 if (raid_type == BTRFS_RAID_SINGLE)
3536                         continue;
3537                 if (!(flags & btrfs_raid_group[raid_type]))
3538                         continue;
3539                 min_tolerated = min(min_tolerated,
3540                                     btrfs_raid_array[raid_type].
3541                                     tolerated_failures);
3542         }
3543
3544         if (min_tolerated == INT_MAX) {
3545                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3546                 min_tolerated = 0;
3547         }
3548
3549         return min_tolerated;
3550 }
3551
3552 int btrfs_calc_num_tolerated_disk_barrier_failures(
3553         struct btrfs_fs_info *fs_info)
3554 {
3555         struct btrfs_ioctl_space_info space;
3556         struct btrfs_space_info *sinfo;
3557         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3558                        BTRFS_BLOCK_GROUP_SYSTEM,
3559                        BTRFS_BLOCK_GROUP_METADATA,
3560                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3561         int i;
3562         int c;
3563         int num_tolerated_disk_barrier_failures =
3564                 (int)fs_info->fs_devices->num_devices;
3565
3566         for (i = 0; i < ARRAY_SIZE(types); i++) {
3567                 struct btrfs_space_info *tmp;
3568
3569                 sinfo = NULL;
3570                 rcu_read_lock();
3571                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3572                         if (tmp->flags == types[i]) {
3573                                 sinfo = tmp;
3574                                 break;
3575                         }
3576                 }
3577                 rcu_read_unlock();
3578
3579                 if (!sinfo)
3580                         continue;
3581
3582                 down_read(&sinfo->groups_sem);
3583                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3584                         u64 flags;
3585
3586                         if (list_empty(&sinfo->block_groups[c]))
3587                                 continue;
3588
3589                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3590                                                    &space);
3591                         if (space.total_bytes == 0 || space.used_bytes == 0)
3592                                 continue;
3593                         flags = space.flags;
3594
3595                         num_tolerated_disk_barrier_failures = min(
3596                                 num_tolerated_disk_barrier_failures,
3597                                 btrfs_get_num_tolerated_disk_barrier_failures(
3598                                         flags));
3599                 }
3600                 up_read(&sinfo->groups_sem);
3601         }
3602
3603         return num_tolerated_disk_barrier_failures;
3604 }
3605
3606 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3607 {
3608         struct list_head *head;
3609         struct btrfs_device *dev;
3610         struct btrfs_super_block *sb;
3611         struct btrfs_dev_item *dev_item;
3612         int ret;
3613         int do_barriers;
3614         int max_errors;
3615         int total_errors = 0;
3616         u64 flags;
3617
3618         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3619         backup_super_roots(root->fs_info);
3620
3621         sb = root->fs_info->super_for_commit;
3622         dev_item = &sb->dev_item;
3623
3624         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3625         head = &root->fs_info->fs_devices->devices;
3626         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3627
3628         if (do_barriers) {
3629                 ret = barrier_all_devices(root->fs_info);
3630                 if (ret) {
3631                         mutex_unlock(
3632                                 &root->fs_info->fs_devices->device_list_mutex);
3633                         btrfs_std_error(root->fs_info, ret,
3634                                     "errors while submitting device barriers.");
3635                         return ret;
3636                 }
3637         }
3638
3639         list_for_each_entry_rcu(dev, head, dev_list) {
3640                 if (!dev->bdev) {
3641                         total_errors++;
3642                         continue;
3643                 }
3644                 if (!dev->in_fs_metadata || !dev->writeable)
3645                         continue;
3646
3647                 btrfs_set_stack_device_generation(dev_item, 0);
3648                 btrfs_set_stack_device_type(dev_item, dev->type);
3649                 btrfs_set_stack_device_id(dev_item, dev->devid);
3650                 btrfs_set_stack_device_total_bytes(dev_item,
3651                                                    dev->commit_total_bytes);
3652                 btrfs_set_stack_device_bytes_used(dev_item,
3653                                                   dev->commit_bytes_used);
3654                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3655                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3656                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3657                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3658                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3659
3660                 flags = btrfs_super_flags(sb);
3661                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3662
3663                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3664                 if (ret)
3665                         total_errors++;
3666         }
3667         if (total_errors > max_errors) {
3668                 btrfs_err(root->fs_info, "%d errors while writing supers",
3669                        total_errors);
3670                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3671
3672                 /* FUA is masked off if unsupported and can't be the reason */
3673                 btrfs_std_error(root->fs_info, -EIO,
3674                             "%d errors while writing supers", total_errors);
3675                 return -EIO;
3676         }
3677
3678         total_errors = 0;
3679         list_for_each_entry_rcu(dev, head, dev_list) {
3680                 if (!dev->bdev)
3681                         continue;
3682                 if (!dev->in_fs_metadata || !dev->writeable)
3683                         continue;
3684
3685                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3686                 if (ret)
3687                         total_errors++;
3688         }
3689         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3690         if (total_errors > max_errors) {
3691                 btrfs_std_error(root->fs_info, -EIO,
3692                             "%d errors while writing supers", total_errors);
3693                 return -EIO;
3694         }
3695         return 0;
3696 }
3697
3698 int write_ctree_super(struct btrfs_trans_handle *trans,
3699                       struct btrfs_root *root, int max_mirrors)
3700 {
3701         return write_all_supers(root, max_mirrors);
3702 }
3703
3704 /* Drop a fs root from the radix tree and free it. */
3705 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3706                                   struct btrfs_root *root)
3707 {
3708         spin_lock(&fs_info->fs_roots_radix_lock);
3709         radix_tree_delete(&fs_info->fs_roots_radix,
3710                           (unsigned long)root->root_key.objectid);
3711         spin_unlock(&fs_info->fs_roots_radix_lock);
3712
3713         if (btrfs_root_refs(&root->root_item) == 0)
3714                 synchronize_srcu(&fs_info->subvol_srcu);
3715
3716         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3717                 btrfs_free_log(NULL, root);
3718
3719         if (root->free_ino_pinned)
3720                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3721         if (root->free_ino_ctl)
3722                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3723         free_fs_root(root);
3724 }
3725
3726 static void free_fs_root(struct btrfs_root *root)
3727 {
3728         iput(root->ino_cache_inode);
3729         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3730         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3731         root->orphan_block_rsv = NULL;
3732         if (root->anon_dev)
3733                 free_anon_bdev(root->anon_dev);
3734         if (root->subv_writers)
3735                 btrfs_free_subvolume_writers(root->subv_writers);
3736         free_extent_buffer(root->node);
3737         free_extent_buffer(root->commit_root);
3738         kfree(root->free_ino_ctl);
3739         kfree(root->free_ino_pinned);
3740         kfree(root->name);
3741         btrfs_put_fs_root(root);
3742 }
3743
3744 void btrfs_free_fs_root(struct btrfs_root *root)
3745 {
3746         free_fs_root(root);
3747 }
3748
3749 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3750 {
3751         u64 root_objectid = 0;
3752         struct btrfs_root *gang[8];
3753         int i = 0;
3754         int err = 0;
3755         unsigned int ret = 0;
3756         int index;
3757
3758         while (1) {
3759                 index = srcu_read_lock(&fs_info->subvol_srcu);
3760                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3761                                              (void **)gang, root_objectid,
3762                                              ARRAY_SIZE(gang));
3763                 if (!ret) {
3764                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3765                         break;
3766                 }
3767                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3768
3769                 for (i = 0; i < ret; i++) {
3770                         /* Avoid to grab roots in dead_roots */
3771                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3772                                 gang[i] = NULL;
3773                                 continue;
3774                         }
3775                         /* grab all the search result for later use */
3776                         gang[i] = btrfs_grab_fs_root(gang[i]);
3777                 }
3778                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3779
3780                 for (i = 0; i < ret; i++) {
3781                         if (!gang[i])
3782                                 continue;
3783                         root_objectid = gang[i]->root_key.objectid;
3784                         err = btrfs_orphan_cleanup(gang[i]);
3785                         if (err)
3786                                 break;
3787                         btrfs_put_fs_root(gang[i]);
3788                 }
3789                 root_objectid++;
3790         }
3791
3792         /* release the uncleaned roots due to error */
3793         for (; i < ret; i++) {
3794                 if (gang[i])
3795                         btrfs_put_fs_root(gang[i]);
3796         }
3797         return err;
3798 }
3799
3800 int btrfs_commit_super(struct btrfs_root *root)
3801 {
3802         struct btrfs_trans_handle *trans;
3803
3804         mutex_lock(&root->fs_info->cleaner_mutex);
3805         btrfs_run_delayed_iputs(root);
3806         mutex_unlock(&root->fs_info->cleaner_mutex);
3807         wake_up_process(root->fs_info->cleaner_kthread);
3808
3809         /* wait until ongoing cleanup work done */
3810         down_write(&root->fs_info->cleanup_work_sem);
3811         up_write(&root->fs_info->cleanup_work_sem);
3812
3813         trans = btrfs_join_transaction(root);
3814         if (IS_ERR(trans))
3815                 return PTR_ERR(trans);
3816         return btrfs_commit_transaction(trans, root);
3817 }
3818
3819 void close_ctree(struct btrfs_root *root)
3820 {
3821         struct btrfs_fs_info *fs_info = root->fs_info;
3822         int ret;
3823
3824         fs_info->closing = 1;
3825         smp_mb();
3826
3827         /* wait for the qgroup rescan worker to stop */
3828         btrfs_qgroup_wait_for_completion(fs_info);
3829
3830         /* wait for the uuid_scan task to finish */
3831         down(&fs_info->uuid_tree_rescan_sem);
3832         /* avoid complains from lockdep et al., set sem back to initial state */
3833         up(&fs_info->uuid_tree_rescan_sem);
3834
3835         /* pause restriper - we want to resume on mount */
3836         btrfs_pause_balance(fs_info);
3837
3838         btrfs_dev_replace_suspend_for_unmount(fs_info);
3839
3840         btrfs_scrub_cancel(fs_info);
3841
3842         /* wait for any defraggers to finish */
3843         wait_event(fs_info->transaction_wait,
3844                    (atomic_read(&fs_info->defrag_running) == 0));
3845
3846         /* clear out the rbtree of defraggable inodes */
3847         btrfs_cleanup_defrag_inodes(fs_info);
3848
3849         cancel_work_sync(&fs_info->async_reclaim_work);
3850
3851         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3852                 /*
3853                  * If the cleaner thread is stopped and there are
3854                  * block groups queued for removal, the deletion will be
3855                  * skipped when we quit the cleaner thread.
3856                  */
3857                 btrfs_delete_unused_bgs(root->fs_info);
3858
3859                 ret = btrfs_commit_super(root);
3860                 if (ret)
3861                         btrfs_err(fs_info, "commit super ret %d", ret);
3862         }
3863
3864         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3865                 btrfs_error_commit_super(root);
3866
3867         kthread_stop(fs_info->transaction_kthread);
3868         kthread_stop(fs_info->cleaner_kthread);
3869
3870         fs_info->closing = 2;
3871         smp_mb();
3872
3873         btrfs_free_qgroup_config(fs_info);
3874
3875         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3876                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3877                        percpu_counter_sum(&fs_info->delalloc_bytes));
3878         }
3879
3880         btrfs_sysfs_remove_mounted(fs_info);
3881         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3882
3883         btrfs_free_fs_roots(fs_info);
3884
3885         btrfs_put_block_group_cache(fs_info);
3886
3887         btrfs_free_block_groups(fs_info);
3888
3889         /*
3890          * we must make sure there is not any read request to
3891          * submit after we stopping all workers.
3892          */
3893         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3894         btrfs_stop_all_workers(fs_info);
3895
3896         fs_info->open = 0;
3897         free_root_pointers(fs_info, 1);
3898
3899         iput(fs_info->btree_inode);
3900
3901 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3902         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3903                 btrfsic_unmount(root, fs_info->fs_devices);
3904 #endif
3905
3906         btrfs_close_devices(fs_info->fs_devices);
3907         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3908
3909         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3910         percpu_counter_destroy(&fs_info->delalloc_bytes);
3911         percpu_counter_destroy(&fs_info->bio_counter);
3912         bdi_destroy(&fs_info->bdi);
3913         cleanup_srcu_struct(&fs_info->subvol_srcu);
3914
3915         btrfs_free_stripe_hash_table(fs_info);
3916
3917         __btrfs_free_block_rsv(root->orphan_block_rsv);
3918         root->orphan_block_rsv = NULL;
3919
3920         lock_chunks(root);
3921         while (!list_empty(&fs_info->pinned_chunks)) {
3922                 struct extent_map *em;
3923
3924                 em = list_first_entry(&fs_info->pinned_chunks,
3925                                       struct extent_map, list);
3926                 list_del_init(&em->list);
3927                 free_extent_map(em);
3928         }
3929         unlock_chunks(root);
3930 }
3931
3932 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3933                           int atomic)
3934 {
3935         int ret;
3936         struct inode *btree_inode = buf->pages[0]->mapping->host;
3937
3938         ret = extent_buffer_uptodate(buf);
3939         if (!ret)
3940                 return ret;
3941
3942         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3943                                     parent_transid, atomic);
3944         if (ret == -EAGAIN)
3945                 return ret;
3946         return !ret;
3947 }
3948
3949 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3950 {
3951         struct btrfs_root *root;
3952         u64 transid = btrfs_header_generation(buf);
3953         int was_dirty;
3954
3955 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3956         /*
3957          * This is a fast path so only do this check if we have sanity tests
3958          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3959          * outside of the sanity tests.
3960          */
3961         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3962                 return;
3963 #endif
3964         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3965         btrfs_assert_tree_locked(buf);
3966         if (transid != root->fs_info->generation)
3967                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3968                        "found %llu running %llu\n",
3969                         buf->start, transid, root->fs_info->generation);
3970         was_dirty = set_extent_buffer_dirty(buf);
3971         if (!was_dirty)
3972                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3973                                      buf->len,
3974                                      root->fs_info->dirty_metadata_batch);
3975 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3976         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3977                 btrfs_print_leaf(root, buf);
3978                 ASSERT(0);
3979         }
3980 #endif
3981 }
3982
3983 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3984                                         int flush_delayed)
3985 {
3986         /*
3987          * looks as though older kernels can get into trouble with
3988          * this code, they end up stuck in balance_dirty_pages forever
3989          */
3990         int ret;
3991
3992         if (current->flags & PF_MEMALLOC)
3993                 return;
3994
3995         if (flush_delayed)
3996                 btrfs_balance_delayed_items(root);
3997
3998         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3999                                      BTRFS_DIRTY_METADATA_THRESH);
4000         if (ret > 0) {
4001                 balance_dirty_pages_ratelimited(
4002                                    root->fs_info->btree_inode->i_mapping);
4003         }
4004 }
4005
4006 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4007 {
4008         __btrfs_btree_balance_dirty(root, 1);
4009 }
4010
4011 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4012 {
4013         __btrfs_btree_balance_dirty(root, 0);
4014 }
4015
4016 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4017 {
4018         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4019         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4020 }
4021
4022 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4023                               int read_only)
4024 {
4025         struct btrfs_super_block *sb = fs_info->super_copy;
4026         u64 nodesize = btrfs_super_nodesize(sb);
4027         u64 sectorsize = btrfs_super_sectorsize(sb);
4028         int ret = 0;
4029
4030         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4031                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4032                 ret = -EINVAL;
4033         }
4034         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4035                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4036                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4037         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4038                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4039                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4040                 ret = -EINVAL;
4041         }
4042         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4043                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4044                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4045                 ret = -EINVAL;
4046         }
4047         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4048                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4049                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4050                 ret = -EINVAL;
4051         }
4052
4053         /*
4054          * Check sectorsize and nodesize first, other check will need it.
4055          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4056          */
4057         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4058             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4059                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4060                 ret = -EINVAL;
4061         }
4062         /* Only PAGE SIZE is supported yet */
4063         if (sectorsize != PAGE_CACHE_SIZE) {
4064                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4065                                 sectorsize, PAGE_CACHE_SIZE);
4066                 ret = -EINVAL;
4067         }
4068         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4069             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4070                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4071                 ret = -EINVAL;
4072         }
4073         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4074                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4075                                 le32_to_cpu(sb->__unused_leafsize),
4076                                 nodesize);
4077                 ret = -EINVAL;
4078         }
4079
4080         /* Root alignment check */
4081         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4082                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4083                                 btrfs_super_root(sb));
4084                 ret = -EINVAL;
4085         }
4086         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4087                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4088                                 btrfs_super_chunk_root(sb));
4089                 ret = -EINVAL;
4090         }
4091         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4092                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4093                                 btrfs_super_log_root(sb));
4094                 ret = -EINVAL;
4095         }
4096
4097         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4098                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4099                                 fs_info->fsid, sb->dev_item.fsid);
4100                 ret = -EINVAL;
4101         }
4102
4103         /*
4104          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4105          * done later
4106          */
4107         if (btrfs_super_num_devices(sb) > (1UL << 31))
4108                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4109                                 btrfs_super_num_devices(sb));
4110         if (btrfs_super_num_devices(sb) == 0) {
4111                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4112                 ret = -EINVAL;
4113         }
4114
4115         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4116                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4117                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4118                 ret = -EINVAL;
4119         }
4120
4121         /*
4122          * Obvious sys_chunk_array corruptions, it must hold at least one key
4123          * and one chunk
4124          */
4125         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4126                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4127                                 btrfs_super_sys_array_size(sb),
4128                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4129                 ret = -EINVAL;
4130         }
4131         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4132                         + sizeof(struct btrfs_chunk)) {
4133                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4134                                 btrfs_super_sys_array_size(sb),
4135                                 sizeof(struct btrfs_disk_key)
4136                                 + sizeof(struct btrfs_chunk));
4137                 ret = -EINVAL;
4138         }
4139
4140         /*
4141          * The generation is a global counter, we'll trust it more than the others
4142          * but it's still possible that it's the one that's wrong.
4143          */
4144         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4145                 printk(KERN_WARNING
4146                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4147                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4148         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4149             && btrfs_super_cache_generation(sb) != (u64)-1)
4150                 printk(KERN_WARNING
4151                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4152                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4153
4154         return ret;
4155 }
4156
4157 static void btrfs_error_commit_super(struct btrfs_root *root)
4158 {
4159         mutex_lock(&root->fs_info->cleaner_mutex);
4160         btrfs_run_delayed_iputs(root);
4161         mutex_unlock(&root->fs_info->cleaner_mutex);
4162
4163         down_write(&root->fs_info->cleanup_work_sem);
4164         up_write(&root->fs_info->cleanup_work_sem);
4165
4166         /* cleanup FS via transaction */
4167         btrfs_cleanup_transaction(root);
4168 }
4169
4170 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4171 {
4172         struct btrfs_ordered_extent *ordered;
4173
4174         spin_lock(&root->ordered_extent_lock);
4175         /*
4176          * This will just short circuit the ordered completion stuff which will
4177          * make sure the ordered extent gets properly cleaned up.
4178          */
4179         list_for_each_entry(ordered, &root->ordered_extents,
4180                             root_extent_list)
4181                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4182         spin_unlock(&root->ordered_extent_lock);
4183 }
4184
4185 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4186 {
4187         struct btrfs_root *root;
4188         struct list_head splice;
4189
4190         INIT_LIST_HEAD(&splice);
4191
4192         spin_lock(&fs_info->ordered_root_lock);
4193         list_splice_init(&fs_info->ordered_roots, &splice);
4194         while (!list_empty(&splice)) {
4195                 root = list_first_entry(&splice, struct btrfs_root,
4196                                         ordered_root);
4197                 list_move_tail(&root->ordered_root,
4198                                &fs_info->ordered_roots);
4199
4200                 spin_unlock(&fs_info->ordered_root_lock);
4201                 btrfs_destroy_ordered_extents(root);
4202
4203                 cond_resched();
4204                 spin_lock(&fs_info->ordered_root_lock);
4205         }
4206         spin_unlock(&fs_info->ordered_root_lock);
4207 }
4208
4209 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4210                                       struct btrfs_root *root)
4211 {
4212         struct rb_node *node;
4213         struct btrfs_delayed_ref_root *delayed_refs;
4214         struct btrfs_delayed_ref_node *ref;
4215         int ret = 0;
4216
4217         delayed_refs = &trans->delayed_refs;
4218
4219         spin_lock(&delayed_refs->lock);
4220         if (atomic_read(&delayed_refs->num_entries) == 0) {
4221                 spin_unlock(&delayed_refs->lock);
4222                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4223                 return ret;
4224         }
4225
4226         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4227                 struct btrfs_delayed_ref_head *head;
4228                 struct btrfs_delayed_ref_node *tmp;
4229                 bool pin_bytes = false;
4230
4231                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4232                                 href_node);
4233                 if (!mutex_trylock(&head->mutex)) {
4234                         atomic_inc(&head->node.refs);
4235                         spin_unlock(&delayed_refs->lock);
4236
4237                         mutex_lock(&head->mutex);
4238                         mutex_unlock(&head->mutex);
4239                         btrfs_put_delayed_ref(&head->node);
4240                         spin_lock(&delayed_refs->lock);
4241                         continue;
4242                 }
4243                 spin_lock(&head->lock);
4244                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4245                                                  list) {
4246                         ref->in_tree = 0;
4247                         list_del(&ref->list);
4248                         atomic_dec(&delayed_refs->num_entries);
4249                         btrfs_put_delayed_ref(ref);
4250                 }
4251                 if (head->must_insert_reserved)
4252                         pin_bytes = true;
4253                 btrfs_free_delayed_extent_op(head->extent_op);
4254                 delayed_refs->num_heads--;
4255                 if (head->processing == 0)
4256                         delayed_refs->num_heads_ready--;
4257                 atomic_dec(&delayed_refs->num_entries);
4258                 head->node.in_tree = 0;
4259                 rb_erase(&head->href_node, &delayed_refs->href_root);
4260                 spin_unlock(&head->lock);
4261                 spin_unlock(&delayed_refs->lock);
4262                 mutex_unlock(&head->mutex);
4263
4264                 if (pin_bytes)
4265                         btrfs_pin_extent(root, head->node.bytenr,
4266                                          head->node.num_bytes, 1);
4267                 btrfs_put_delayed_ref(&head->node);
4268                 cond_resched();
4269                 spin_lock(&delayed_refs->lock);
4270         }
4271
4272         spin_unlock(&delayed_refs->lock);
4273
4274         return ret;
4275 }
4276
4277 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4278 {
4279         struct btrfs_inode *btrfs_inode;
4280         struct list_head splice;
4281
4282         INIT_LIST_HEAD(&splice);
4283
4284         spin_lock(&root->delalloc_lock);
4285         list_splice_init(&root->delalloc_inodes, &splice);
4286
4287         while (!list_empty(&splice)) {
4288                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4289                                                delalloc_inodes);
4290
4291                 list_del_init(&btrfs_inode->delalloc_inodes);
4292                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4293                           &btrfs_inode->runtime_flags);
4294                 spin_unlock(&root->delalloc_lock);
4295
4296                 btrfs_invalidate_inodes(btrfs_inode->root);
4297
4298                 spin_lock(&root->delalloc_lock);
4299         }
4300
4301         spin_unlock(&root->delalloc_lock);
4302 }
4303
4304 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4305 {
4306         struct btrfs_root *root;
4307         struct list_head splice;
4308
4309         INIT_LIST_HEAD(&splice);
4310
4311         spin_lock(&fs_info->delalloc_root_lock);
4312         list_splice_init(&fs_info->delalloc_roots, &splice);
4313         while (!list_empty(&splice)) {
4314                 root = list_first_entry(&splice, struct btrfs_root,
4315                                          delalloc_root);
4316                 list_del_init(&root->delalloc_root);
4317                 root = btrfs_grab_fs_root(root);
4318                 BUG_ON(!root);
4319                 spin_unlock(&fs_info->delalloc_root_lock);
4320
4321                 btrfs_destroy_delalloc_inodes(root);
4322                 btrfs_put_fs_root(root);
4323
4324                 spin_lock(&fs_info->delalloc_root_lock);
4325         }
4326         spin_unlock(&fs_info->delalloc_root_lock);
4327 }
4328
4329 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4330                                         struct extent_io_tree *dirty_pages,
4331                                         int mark)
4332 {
4333         int ret;
4334         struct extent_buffer *eb;
4335         u64 start = 0;
4336         u64 end;
4337
4338         while (1) {
4339                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4340                                             mark, NULL);
4341                 if (ret)
4342                         break;
4343
4344                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4345                 while (start <= end) {
4346                         eb = btrfs_find_tree_block(root->fs_info, start);
4347                         start += root->nodesize;
4348                         if (!eb)
4349                                 continue;
4350                         wait_on_extent_buffer_writeback(eb);
4351
4352                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4353                                                &eb->bflags))
4354                                 clear_extent_buffer_dirty(eb);
4355                         free_extent_buffer_stale(eb);
4356                 }
4357         }
4358
4359         return ret;
4360 }
4361
4362 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4363                                        struct extent_io_tree *pinned_extents)
4364 {
4365         struct extent_io_tree *unpin;
4366         u64 start;
4367         u64 end;
4368         int ret;
4369         bool loop = true;
4370
4371         unpin = pinned_extents;
4372 again:
4373         while (1) {
4374                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4375                                             EXTENT_DIRTY, NULL);
4376                 if (ret)
4377                         break;
4378
4379                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4380                 btrfs_error_unpin_extent_range(root, start, end);
4381                 cond_resched();
4382         }
4383
4384         if (loop) {
4385                 if (unpin == &root->fs_info->freed_extents[0])
4386                         unpin = &root->fs_info->freed_extents[1];
4387                 else
4388                         unpin = &root->fs_info->freed_extents[0];
4389                 loop = false;
4390                 goto again;
4391         }
4392
4393         return 0;
4394 }
4395
4396 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4397                                    struct btrfs_root *root)
4398 {
4399         btrfs_destroy_delayed_refs(cur_trans, root);
4400
4401         cur_trans->state = TRANS_STATE_COMMIT_START;
4402         wake_up(&root->fs_info->transaction_blocked_wait);
4403
4404         cur_trans->state = TRANS_STATE_UNBLOCKED;
4405         wake_up(&root->fs_info->transaction_wait);
4406
4407         btrfs_destroy_delayed_inodes(root);
4408         btrfs_assert_delayed_root_empty(root);
4409
4410         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4411                                      EXTENT_DIRTY);
4412         btrfs_destroy_pinned_extent(root,
4413                                     root->fs_info->pinned_extents);
4414
4415         cur_trans->state =TRANS_STATE_COMPLETED;
4416         wake_up(&cur_trans->commit_wait);
4417
4418         /*
4419         memset(cur_trans, 0, sizeof(*cur_trans));
4420         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4421         */
4422 }
4423
4424 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4425 {
4426         struct btrfs_transaction *t;
4427
4428         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4429
4430         spin_lock(&root->fs_info->trans_lock);
4431         while (!list_empty(&root->fs_info->trans_list)) {
4432                 t = list_first_entry(&root->fs_info->trans_list,
4433                                      struct btrfs_transaction, list);
4434                 if (t->state >= TRANS_STATE_COMMIT_START) {
4435                         atomic_inc(&t->use_count);
4436                         spin_unlock(&root->fs_info->trans_lock);
4437                         btrfs_wait_for_commit(root, t->transid);
4438                         btrfs_put_transaction(t);
4439                         spin_lock(&root->fs_info->trans_lock);
4440                         continue;
4441                 }
4442                 if (t == root->fs_info->running_transaction) {
4443                         t->state = TRANS_STATE_COMMIT_DOING;
4444                         spin_unlock(&root->fs_info->trans_lock);
4445                         /*
4446                          * We wait for 0 num_writers since we don't hold a trans
4447                          * handle open currently for this transaction.
4448                          */
4449                         wait_event(t->writer_wait,
4450                                    atomic_read(&t->num_writers) == 0);
4451                 } else {
4452                         spin_unlock(&root->fs_info->trans_lock);
4453                 }
4454                 btrfs_cleanup_one_transaction(t, root);
4455
4456                 spin_lock(&root->fs_info->trans_lock);
4457                 if (t == root->fs_info->running_transaction)
4458                         root->fs_info->running_transaction = NULL;
4459                 list_del_init(&t->list);
4460                 spin_unlock(&root->fs_info->trans_lock);
4461
4462                 btrfs_put_transaction(t);
4463                 trace_btrfs_transaction_commit(root);
4464                 spin_lock(&root->fs_info->trans_lock);
4465         }
4466         spin_unlock(&root->fs_info->trans_lock);
4467         btrfs_destroy_all_ordered_extents(root->fs_info);
4468         btrfs_destroy_delayed_inodes(root);
4469         btrfs_assert_delayed_root_empty(root);
4470         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4471         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4472         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4473
4474         return 0;
4475 }
4476
4477 static const struct extent_io_ops btree_extent_io_ops = {
4478         .readpage_end_io_hook = btree_readpage_end_io_hook,
4479         .readpage_io_failed_hook = btree_io_failed_hook,
4480         .submit_bio_hook = btree_submit_bio_hook,
4481         /* note we're sharing with inode.c for the merge bio hook */
4482         .merge_bio_hook = btrfs_merge_bio_hook,
4483 };