fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/slab.h>
  29 #include <linux/migrate.h>
  30 #include <linux/ratelimit.h>
  31 #include <linux/uuid.h>
  32 #include <linux/semaphore.h>
  33 #include <asm/unaligned.h>
  34 #include "ctree.h"
  35 #include "disk-io.h"
  36 #include "hash.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "locking.h"
  42 #include "tree-log.h"
  43 #include "free-space-cache.h"
  44 #include "free-space-tree.h"
  45 #include "inode-map.h"
  46 #include "check-integrity.h"
  47 #include "rcu-string.h"
  48 #include "dev-replace.h"
  49 #include "raid56.h"
  50 #include "sysfs.h"
  51 #include "qgroup.h"
  52 #include "compression.h"
  53
  54 #ifdef CONFIG_X86
  55 #include <asm/cpufeature.h>
  56 #endif
  57
  58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  59                                  BTRFS_HEADER_FLAG_RELOC |\
  60                                  BTRFS_SUPER_FLAG_ERROR |\
  61                                  BTRFS_SUPER_FLAG_SEEDING |\
  62                                  BTRFS_SUPER_FLAG_METADUMP)
  63
  64 static const struct extent_io_ops btree_extent_io_ops;
  65 static void end_workqueue_fn(struct btrfs_work *work);
  66 static void free_fs_root(struct btrfs_root *root);
  67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  68                                     int read_only);
  69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  71                                       struct btrfs_root *root);
  72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  74                                         struct extent_io_tree *dirty_pages,
  75                                         int mark);
  76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  77                                        struct extent_io_tree *pinned_extents);
  78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  79 static void btrfs_error_commit_super(struct btrfs_root *root);
  80
  81 /*
  82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  83  * is complete.  This is used during reads to verify checksums, and it is used
  84  * by writes to insert metadata for new file extents after IO is complete.
  85  */
  86 struct btrfs_end_io_wq {
  87         struct bio *bio;
  88         bio_end_io_t *end_io;
  89         void *private;
  90         struct btrfs_fs_info *info;
  91         int error;
  92         enum btrfs_wq_endio_type metadata;
  93         struct list_head list;
  94         struct btrfs_work work;
  95 };
  96
  97 static struct kmem_cache *btrfs_end_io_wq_cache;
  98
  99 int __init btrfs_end_io_wq_init(void)
 100 {
 101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
 102                                         sizeof(struct btrfs_end_io_wq),
 103                                         0,
 104                                         SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         kmem_cache_destroy(btrfs_end_io_wq_cache);
 114 }
 115
 116 /*
 117  * async submit bios are used to offload expensive checksumming
 118  * onto the worker threads.  They checksum file and metadata bios
 119  * just before they are sent down the IO stack.
 120  */
 121 struct async_submit_bio {
 122         struct inode *inode;
 123         struct bio *bio;
 124         struct list_head list;
 125         extent_submit_bio_hook_t *submit_bio_start;
 126         extent_submit_bio_hook_t *submit_bio_done;
 127         int mirror_num;
 128         unsigned long bio_flags;
 129         /*
 130          * bio_offset is optional, can be used if the pages in the bio
 131          * can't tell us where in the file the bio should go
 132          */
 133         u64 bio_offset;
 134         struct btrfs_work work;
 135         int error;
 136 };
 137
 138 /*
 139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 141  * the level the eb occupies in the tree.
 142  *
 143  * Different roots are used for different purposes and may nest inside each
 144  * other and they require separate keysets.  As lockdep keys should be
 145  * static, assign keysets according to the purpose of the root as indicated
 146  * by btrfs_root->objectid.  This ensures that all special purpose roots
 147  * have separate keysets.
 148  *
 149  * Lock-nesting across peer nodes is always done with the immediate parent
 150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 151  * subclass to avoid triggering lockdep warning in such cases.
 152  *
 153  * The key is set by the readpage_end_io_hook after the buffer has passed
 154  * csum validation but before the pages are unlocked.  It is also set by
 155  * btrfs_init_new_buffer on freshly allocated blocks.
 156  *
 157  * We also add a check to make sure the highest level of the tree is the
 158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 159  * needs update as well.
 160  */
 161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 162 # if BTRFS_MAX_LEVEL != 8
 163 #  error
 164 # endif
 165
 166 static struct btrfs_lockdep_keyset {
 167         u64                     id;             /* root objectid */
 168         const char              *name_stem;     /* lock name stem */
 169         char                    names[BTRFS_MAX_LEVEL + 1][20];
 170         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 171 } btrfs_lockdep_keysets[] = {
 172         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 173         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 174         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 175         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 176         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 177         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 178         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 179         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 180         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 181         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 182         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 183         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 184         { .id = 0,                              .name_stem = "tree"     },
 185 };
 186
 187 void __init btrfs_init_lockdep(void)
 188 {
 189         int i, j;
 190
 191         /* initialize lockdep class names */
 192         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 193                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 194
 195                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 196                         snprintf(ks->names[j], sizeof(ks->names[j]),
 197                                  "btrfs-%s-%02d", ks->name_stem, j);
 198         }
 199 }
 200
 201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 202                                     int level)
 203 {
 204         struct btrfs_lockdep_keyset *ks;
 205
 206         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 207
 208         /* find the matching keyset, id 0 is the default entry */
 209         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 210                 if (ks->id == objectid)
 211                         break;
 212
 213         lockdep_set_class_and_name(&eb->lock,
 214                                    &ks->keys[level], ks->names[level]);
 215 }
 216
 217 #endif
 218
 219 /*
 220  * extents on the btree inode are pretty simple, there's one extent
 221  * that covers the entire device
 222  */
 223 static struct extent_map *btree_get_extent(struct inode *inode,
 224                 struct page *page, size_t pg_offset, u64 start, u64 len,
 225                 int create)
 226 {
 227         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 228         struct extent_map *em;
 229         int ret;
 230
 231         read_lock(&em_tree->lock);
 232         em = lookup_extent_mapping(em_tree, start, len);
 233         if (em) {
 234                 em->bdev =
 235                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 236                 read_unlock(&em_tree->lock);
 237                 goto out;
 238         }
 239         read_unlock(&em_tree->lock);
 240
 241         em = alloc_extent_map();
 242         if (!em) {
 243                 em = ERR_PTR(-ENOMEM);
 244                 goto out;
 245         }
 246         em->start = 0;
 247         em->len = (u64)-1;
 248         em->block_len = (u64)-1;
 249         em->block_start = 0;
 250         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 251
 252         write_lock(&em_tree->lock);
 253         ret = add_extent_mapping(em_tree, em, 0);
 254         if (ret == -EEXIST) {
 255                 free_extent_map(em);
 256                 em = lookup_extent_mapping(em_tree, start, len);
 257                 if (!em)
 258                         em = ERR_PTR(-EIO);
 259         } else if (ret) {
 260                 free_extent_map(em);
 261                 em = ERR_PTR(ret);
 262         }
 263         write_unlock(&em_tree->lock);
 264
 265 out:
 266         return em;
 267 }
 268
 269 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 270 {
 271         return btrfs_crc32c(seed, data, len);
 272 }
 273
 274 void btrfs_csum_final(u32 crc, char *result)
 275 {
 276         put_unaligned_le32(~crc, result);
 277 }
 278
 279 /*
 280  * compute the csum for a btree block, and either verify it or write it
 281  * into the csum field of the block.
 282  */
 283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 284                            struct extent_buffer *buf,
 285                            int verify)
 286 {
 287         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 288         char *result = NULL;
 289         unsigned long len;
 290         unsigned long cur_len;
 291         unsigned long offset = BTRFS_CSUM_SIZE;
 292         char *kaddr;
 293         unsigned long map_start;
 294         unsigned long map_len;
 295         int err;
 296         u32 crc = ~(u32)0;
 297         unsigned long inline_result;
 298
 299         len = buf->len - offset;
 300         while (len > 0) {
 301                 err = map_private_extent_buffer(buf, offset, 32,
 302                                         &kaddr, &map_start, &map_len);
 303                 if (err)
 304                         return err;
 305                 cur_len = min(len, map_len - (offset - map_start));
 306                 crc = btrfs_csum_data(kaddr + offset - map_start,
 307                                       crc, cur_len);
 308                 len -= cur_len;
 309                 offset += cur_len;
 310         }
 311         if (csum_size > sizeof(inline_result)) {
 312                 result = kzalloc(csum_size, GFP_NOFS);
 313                 if (!result)
 314                         return -ENOMEM;
 315         } else {
 316                 result = (char *)&inline_result;
 317         }
 318
 319         btrfs_csum_final(crc, result);
 320
 321         if (verify) {
 322                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 323                         u32 val;
 324                         u32 found = 0;
 325                         memcpy(&found, result, csum_size);
 326
 327                         read_extent_buffer(buf, &val, 0, csum_size);
 328                         btrfs_warn_rl(fs_info,
 329                                 "%s checksum verify failed on %llu wanted %X found %X "
 330                                 "level %d",
 331                                 fs_info->sb->s_id, buf->start,
 332                                 val, found, btrfs_header_level(buf));
 333                         if (result != (char *)&inline_result)
 334                                 kfree(result);
 335                         return -EUCLEAN;
 336                 }
 337         } else {
 338                 write_extent_buffer(buf, result, 0, csum_size);
 339         }
 340         if (result != (char *)&inline_result)
 341                 kfree(result);
 342         return 0;
 343 }
 344
 345 /*
 346  * we can't consider a given block up to date unless the transid of the
 347  * block matches the transid in the parent node's pointer.  This is how we
 348  * detect blocks that either didn't get written at all or got written
 349  * in the wrong place.
 350  */
 351 static int verify_parent_transid(struct extent_io_tree *io_tree,
 352                                  struct extent_buffer *eb, u64 parent_transid,
 353                                  int atomic)
 354 {
 355         struct extent_state *cached_state = NULL;
 356         int ret;
 357         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 358
 359         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 360                 return 0;
 361
 362         if (atomic)
 363                 return -EAGAIN;
 364
 365         if (need_lock) {
 366                 btrfs_tree_read_lock(eb);
 367                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
 368         }
 369
 370         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 371                          &cached_state);
 372         if (extent_buffer_uptodate(eb) &&
 373             btrfs_header_generation(eb) == parent_transid) {
 374                 ret = 0;
 375                 goto out;
 376         }
 377         btrfs_err_rl(eb->fs_info,
 378                 "parent transid verify failed on %llu wanted %llu found %llu",
 379                         eb->start,
 380                         parent_transid, btrfs_header_generation(eb));
 381         ret = 1;
 382
 383         /*
 384          * Things reading via commit roots that don't have normal protection,
 385          * like send, can have a really old block in cache that may point at a
 386          * block that has been freed and re-allocated.  So don't clear uptodate
 387          * if we find an eb that is under IO (dirty/writeback) because we could
 388          * end up reading in the stale data and then writing it back out and
 389          * making everybody very sad.
 390          */
 391         if (!extent_buffer_under_io(eb))
 392                 clear_extent_buffer_uptodate(eb);
 393 out:
 394         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 395                              &cached_state, GFP_NOFS);
 396         if (need_lock)
 397                 btrfs_tree_read_unlock_blocking(eb);
 398         return ret;
 399 }
 400
 401 /*
 402  * Return 0 if the superblock checksum type matches the checksum value of that
 403  * algorithm. Pass the raw disk superblock data.
 404  */
 405 static int btrfs_check_super_csum(char *raw_disk_sb)
 406 {
 407         struct btrfs_super_block *disk_sb =
 408                 (struct btrfs_super_block *)raw_disk_sb;
 409         u16 csum_type = btrfs_super_csum_type(disk_sb);
 410         int ret = 0;
 411
 412         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 413                 u32 crc = ~(u32)0;
 414                 const int csum_size = sizeof(crc);
 415                 char result[csum_size];
 416
 417                 /*
 418                  * The super_block structure does not span the whole
 419                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 420                  * is filled with zeros and is included in the checksum.
 421                  */
 422                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 423                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 424                 btrfs_csum_final(crc, result);
 425
 426                 if (memcmp(raw_disk_sb, result, csum_size))
 427                         ret = 1;
 428         }
 429
 430         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 431                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 432                                 csum_type);
 433                 ret = 1;
 434         }
 435
 436         return ret;
 437 }
 438
 439 /*
 440  * helper to read a given tree block, doing retries as required when
 441  * the checksums don't match and we have alternate mirrors to try.
 442  */
 443 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 444                                           struct extent_buffer *eb,
 445                                           u64 start, u64 parent_transid)
 446 {
 447         struct extent_io_tree *io_tree;
 448         int failed = 0;
 449         int ret;
 450         int num_copies = 0;
 451         int mirror_num = 0;
 452         int failed_mirror = 0;
 453
 454         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 455         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 456         while (1) {
 457                 ret = read_extent_buffer_pages(io_tree, eb, start,
 458                                                WAIT_COMPLETE,
 459                                                btree_get_extent, mirror_num);
 460                 if (!ret) {
 461                         if (!verify_parent_transid(io_tree, eb,
 462                                                    parent_transid, 0))
 463                                 break;
 464                         else
 465                                 ret = -EIO;
 466                 }
 467
 468                 /*
 469                  * This buffer's crc is fine, but its contents are corrupted, so
 470                  * there is no reason to read the other copies, they won't be
 471                  * any less wrong.
 472                  */
 473                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 474                         break;
 475
 476                 num_copies = btrfs_num_copies(root->fs_info,
 477                                               eb->start, eb->len);
 478                 if (num_copies == 1)
 479                         break;
 480
 481                 if (!failed_mirror) {
 482                         failed = 1;
 483                         failed_mirror = eb->read_mirror;
 484                 }
 485
 486                 mirror_num++;
 487                 if (mirror_num == failed_mirror)
 488                         mirror_num++;
 489
 490                 if (mirror_num > num_copies)
 491                         break;
 492         }
 493
 494         if (failed && !ret && failed_mirror)
 495                 repair_eb_io_failure(root, eb, failed_mirror);
 496
 497         return ret;
 498 }
 499
 500 /*
 501  * checksum a dirty tree block before IO.  This has extra checks to make sure
 502  * we only fill in the checksum field in the first page of a multi-page block
 503  */
 504
 505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 506 {
 507         u64 start = page_offset(page);
 508         u64 found_start;
 509         struct extent_buffer *eb;
 510
 511         eb = (struct extent_buffer *)page->private;
 512         if (page != eb->pages[0])
 513                 return 0;
 514
 515         found_start = btrfs_header_bytenr(eb);
 516         /*
 517          * Please do not consolidate these warnings into a single if.
 518          * It is useful to know what went wrong.
 519          */
 520         if (WARN_ON(found_start != start))
 521                 return -EUCLEAN;
 522         if (WARN_ON(!PageUptodate(page)))
 523                 return -EUCLEAN;
 524
 525         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
 526                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 527
 528         return csum_tree_block(fs_info, eb, 0);
 529 }
 530
 531 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 532                                  struct extent_buffer *eb)
 533 {
 534         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 535         u8 fsid[BTRFS_UUID_SIZE];
 536         int ret = 1;
 537
 538         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 539         while (fs_devices) {
 540                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 541                         ret = 0;
 542                         break;
 543                 }
 544                 fs_devices = fs_devices->seed;
 545         }
 546         return ret;
 547 }
 548
 549 #define CORRUPT(reason, eb, root, slot)                         \
 550         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 551                    "root=%llu, slot=%d", reason,                        \
 552                btrfs_header_bytenr(eb), root->objectid, slot)
 553
 554 static noinline int check_leaf(struct btrfs_root *root,
 555                                struct extent_buffer *leaf)
 556 {
 557         struct btrfs_key key;
 558         struct btrfs_key leaf_key;
 559         u32 nritems = btrfs_header_nritems(leaf);
 560         int slot;
 561
 562         if (nritems == 0)
 563                 return 0;
 564
 565         /* Check the 0 item */
 566         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 567             BTRFS_LEAF_DATA_SIZE(root)) {
 568                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 569                 return -EIO;
 570         }
 571
 572         /*
 573          * Check to make sure each items keys are in the correct order and their
 574          * offsets make sense.  We only have to loop through nritems-1 because
 575          * we check the current slot against the next slot, which verifies the
 576          * next slot's offset+size makes sense and that the current's slot
 577          * offset is correct.
 578          */
 579         for (slot = 0; slot < nritems - 1; slot++) {
 580                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 581                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 582
 583                 /* Make sure the keys are in the right order */
 584                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 585                         CORRUPT("bad key order", leaf, root, slot);
 586                         return -EIO;
 587                 }
 588
 589                 /*
 590                  * Make sure the offset and ends are right, remember that the
 591                  * item data starts at the end of the leaf and grows towards the
 592                  * front.
 593                  */
 594                 if (btrfs_item_offset_nr(leaf, slot) !=
 595                         btrfs_item_end_nr(leaf, slot + 1)) {
 596                         CORRUPT("slot offset bad", leaf, root, slot);
 597                         return -EIO;
 598                 }
 599
 600                 /*
 601                  * Check to make sure that we don't point outside of the leaf,
 602                  * just in case all the items are consistent to each other, but
 603                  * all point outside of the leaf.
 604                  */
 605                 if (btrfs_item_end_nr(leaf, slot) >
 606                     BTRFS_LEAF_DATA_SIZE(root)) {
 607                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 608                         return -EIO;
 609                 }
 610         }
 611
 612         return 0;
 613 }
 614
 615 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 616                                       u64 phy_offset, struct page *page,
 617                                       u64 start, u64 end, int mirror)
 618 {
 619         u64 found_start;
 620         int found_level;
 621         struct extent_buffer *eb;
 622         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 623         struct btrfs_fs_info *fs_info = root->fs_info;
 624         int ret = 0;
 625         int reads_done;
 626
 627         if (!page->private)
 628                 goto out;
 629
 630         eb = (struct extent_buffer *)page->private;
 631
 632         /* the pending IO might have been the only thing that kept this buffer
 633          * in memory.  Make sure we have a ref for all this other checks
 634          */
 635         extent_buffer_get(eb);
 636
 637         reads_done = atomic_dec_and_test(&eb->io_pages);
 638         if (!reads_done)
 639                 goto err;
 640
 641         eb->read_mirror = mirror;
 642         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 643                 ret = -EIO;
 644                 goto err;
 645         }
 646
 647         found_start = btrfs_header_bytenr(eb);
 648         if (found_start != eb->start) {
 649                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
 650                              found_start, eb->start);
 651                 ret = -EIO;
 652                 goto err;
 653         }
 654         if (check_tree_block_fsid(fs_info, eb)) {
 655                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 656                              eb->start);
 657                 ret = -EIO;
 658                 goto err;
 659         }
 660         found_level = btrfs_header_level(eb);
 661         if (found_level >= BTRFS_MAX_LEVEL) {
 662                 btrfs_err(fs_info, "bad tree block level %d",
 663                           (int)btrfs_header_level(eb));
 664                 ret = -EIO;
 665                 goto err;
 666         }
 667
 668         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 669                                        eb, found_level);
 670
 671         ret = csum_tree_block(fs_info, eb, 1);
 672         if (ret)
 673                 goto err;
 674
 675         /*
 676          * If this is a leaf block and it is corrupt, set the corrupt bit so
 677          * that we don't try and read the other copies of this block, just
 678          * return -EIO.
 679          */
 680         if (found_level == 0 && check_leaf(root, eb)) {
 681                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 682                 ret = -EIO;
 683         }
 684
 685         if (!ret)
 686                 set_extent_buffer_uptodate(eb);
 687 err:
 688         if (reads_done &&
 689             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 690                 btree_readahead_hook(fs_info, eb, eb->start, ret);
 691
 692         if (ret) {
 693                 /*
 694                  * our io error hook is going to dec the io pages
 695                  * again, we have to make sure it has something
 696                  * to decrement
 697                  */
 698                 atomic_inc(&eb->io_pages);
 699                 clear_extent_buffer_uptodate(eb);
 700         }
 701         free_extent_buffer(eb);
 702 out:
 703         return ret;
 704 }
 705
 706 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 707 {
 708         struct extent_buffer *eb;
 709
 710         eb = (struct extent_buffer *)page->private;
 711         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
 712         eb->read_mirror = failed_mirror;
 713         atomic_dec(&eb->io_pages);
 714         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 715                 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
 716         return -EIO;    /* we fixed nothing */
 717 }
 718
 719 static void end_workqueue_bio(struct bio *bio)
 720 {
 721         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 722         struct btrfs_fs_info *fs_info;
 723         struct btrfs_workqueue *wq;
 724         btrfs_work_func_t func;
 725
 726         fs_info = end_io_wq->info;
 727         end_io_wq->error = bio->bi_error;
 728
 729         if (bio_op(bio) == REQ_OP_WRITE) {
 730                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 731                         wq = fs_info->endio_meta_write_workers;
 732                         func = btrfs_endio_meta_write_helper;
 733                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 734                         wq = fs_info->endio_freespace_worker;
 735                         func = btrfs_freespace_write_helper;
 736                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 737                         wq = fs_info->endio_raid56_workers;
 738                         func = btrfs_endio_raid56_helper;
 739                 } else {
 740                         wq = fs_info->endio_write_workers;
 741                         func = btrfs_endio_write_helper;
 742                 }
 743         } else {
 744                 if (unlikely(end_io_wq->metadata ==
 745                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 746                         wq = fs_info->endio_repair_workers;
 747                         func = btrfs_endio_repair_helper;
 748                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 749                         wq = fs_info->endio_raid56_workers;
 750                         func = btrfs_endio_raid56_helper;
 751                 } else if (end_io_wq->metadata) {
 752                         wq = fs_info->endio_meta_workers;
 753                         func = btrfs_endio_meta_helper;
 754                 } else {
 755                         wq = fs_info->endio_workers;
 756                         func = btrfs_endio_helper;
 757                 }
 758         }
 759
 760         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 761         btrfs_queue_work(wq, &end_io_wq->work);
 762 }
 763
 764 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 765                         enum btrfs_wq_endio_type metadata)
 766 {
 767         struct btrfs_end_io_wq *end_io_wq;
 768
 769         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 770         if (!end_io_wq)
 771                 return -ENOMEM;
 772
 773         end_io_wq->private = bio->bi_private;
 774         end_io_wq->end_io = bio->bi_end_io;
 775         end_io_wq->info = info;
 776         end_io_wq->error = 0;
 777         end_io_wq->bio = bio;
 778         end_io_wq->metadata = metadata;
 779
 780         bio->bi_private = end_io_wq;
 781         bio->bi_end_io = end_workqueue_bio;
 782         return 0;
 783 }
 784
 785 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 786 {
 787         unsigned long limit = min_t(unsigned long,
 788                                     info->thread_pool_size,
 789                                     info->fs_devices->open_devices);
 790         return 256 * limit;
 791 }
 792
 793 static void run_one_async_start(struct btrfs_work *work)
 794 {
 795         struct async_submit_bio *async;
 796         int ret;
 797
 798         async = container_of(work, struct  async_submit_bio, work);
 799         ret = async->submit_bio_start(async->inode, async->bio,
 800                                       async->mirror_num, async->bio_flags,
 801                                       async->bio_offset);
 802         if (ret)
 803                 async->error = ret;
 804 }
 805
 806 static void run_one_async_done(struct btrfs_work *work)
 807 {
 808         struct btrfs_fs_info *fs_info;
 809         struct async_submit_bio *async;
 810         int limit;
 811
 812         async = container_of(work, struct  async_submit_bio, work);
 813         fs_info = BTRFS_I(async->inode)->root->fs_info;
 814
 815         limit = btrfs_async_submit_limit(fs_info);
 816         limit = limit * 2 / 3;
 817
 818         /*
 819          * atomic_dec_return implies a barrier for waitqueue_active
 820          */
 821         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 822             waitqueue_active(&fs_info->async_submit_wait))
 823                 wake_up(&fs_info->async_submit_wait);
 824
 825         /* If an error occurred we just want to clean up the bio and move on */
 826         if (async->error) {
 827                 async->bio->bi_error = async->error;
 828                 bio_endio(async->bio);
 829                 return;
 830         }
 831
 832         async->submit_bio_done(async->inode, async->bio, async->mirror_num,
 833                                async->bio_flags, async->bio_offset);
 834 }
 835
 836 static void run_one_async_free(struct btrfs_work *work)
 837 {
 838         struct async_submit_bio *async;
 839
 840         async = container_of(work, struct  async_submit_bio, work);
 841         kfree(async);
 842 }
 843
 844 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 845                         struct bio *bio, int mirror_num,
 846                         unsigned long bio_flags,
 847                         u64 bio_offset,
 848                         extent_submit_bio_hook_t *submit_bio_start,
 849                         extent_submit_bio_hook_t *submit_bio_done)
 850 {
 851         struct async_submit_bio *async;
 852
 853         async = kmalloc(sizeof(*async), GFP_NOFS);
 854         if (!async)
 855                 return -ENOMEM;
 856
 857         async->inode = inode;
 858         async->bio = bio;
 859         async->mirror_num = mirror_num;
 860         async->submit_bio_start = submit_bio_start;
 861         async->submit_bio_done = submit_bio_done;
 862
 863         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 864                         run_one_async_done, run_one_async_free);
 865
 866         async->bio_flags = bio_flags;
 867         async->bio_offset = bio_offset;
 868
 869         async->error = 0;
 870
 871         atomic_inc(&fs_info->nr_async_submits);
 872
 873         if (bio->bi_rw & REQ_SYNC)
 874                 btrfs_set_work_high_priority(&async->work);
 875
 876         btrfs_queue_work(fs_info->workers, &async->work);
 877
 878         while (atomic_read(&fs_info->async_submit_draining) &&
 879               atomic_read(&fs_info->nr_async_submits)) {
 880                 wait_event(fs_info->async_submit_wait,
 881                            (atomic_read(&fs_info->nr_async_submits) == 0));
 882         }
 883
 884         return 0;
 885 }
 886
 887 static int btree_csum_one_bio(struct bio *bio)
 888 {
 889         struct bio_vec *bvec;
 890         struct btrfs_root *root;
 891         int i, ret = 0;
 892
 893         bio_for_each_segment_all(bvec, bio, i) {
 894                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 895                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 896                 if (ret)
 897                         break;
 898         }
 899
 900         return ret;
 901 }
 902
 903 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
 904                                     int mirror_num, unsigned long bio_flags,
 905                                     u64 bio_offset)
 906 {
 907         /*
 908          * when we're called for a write, we're already in the async
 909          * submission context.  Just jump into btrfs_map_bio
 910          */
 911         return btree_csum_one_bio(bio);
 912 }
 913
 914 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
 915                                  int mirror_num, unsigned long bio_flags,
 916                                  u64 bio_offset)
 917 {
 918         int ret;
 919
 920         /*
 921          * when we're called for a write, we're already in the async
 922          * submission context.  Just jump into btrfs_map_bio
 923          */
 924         ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 1);
 925         if (ret) {
 926                 bio->bi_error = ret;
 927                 bio_endio(bio);
 928         }
 929         return ret;
 930 }
 931
 932 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 933 {
 934         if (bio_flags & EXTENT_BIO_TREE_LOG)
 935                 return 0;
 936 #ifdef CONFIG_X86
 937         if (static_cpu_has(X86_FEATURE_XMM4_2))
 938                 return 0;
 939 #endif
 940         return 1;
 941 }
 942
 943 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
 944                                  int mirror_num, unsigned long bio_flags,
 945                                  u64 bio_offset)
 946 {
 947         int async = check_async_write(inode, bio_flags);
 948         int ret;
 949
 950         if (bio_op(bio) != REQ_OP_WRITE) {
 951                 /*
 952                  * called for a read, do the setup so that checksum validation
 953                  * can happen in the async kernel threads
 954                  */
 955                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 956                                           bio, BTRFS_WQ_ENDIO_METADATA);
 957                 if (ret)
 958                         goto out_w_error;
 959                 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
 960         } else if (!async) {
 961                 ret = btree_csum_one_bio(bio);
 962                 if (ret)
 963                         goto out_w_error;
 964                 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
 965         } else {
 966                 /*
 967                  * kthread helpers are used to submit writes so that
 968                  * checksumming can happen in parallel across all CPUs
 969                  */
 970                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 971                                           inode, bio, mirror_num, 0,
 972                                           bio_offset,
 973                                           __btree_submit_bio_start,
 974                                           __btree_submit_bio_done);
 975         }
 976
 977         if (ret)
 978                 goto out_w_error;
 979         return 0;
 980
 981 out_w_error:
 982         bio->bi_error = ret;
 983         bio_endio(bio);
 984         return ret;
 985 }
 986
 987 #ifdef CONFIG_MIGRATION
 988 static int btree_migratepage(struct address_space *mapping,
 989                         struct page *newpage, struct page *page,
 990                         enum migrate_mode mode)
 991 {
 992         /*
 993          * we can't safely write a btree page from here,
 994          * we haven't done the locking hook
 995          */
 996         if (PageDirty(page))
 997                 return -EAGAIN;
 998         /*
 999          * Buffers may be managed in a filesystem specific way.
1000          * We must have no buffers or drop them.
1001          */
1002         if (page_has_private(page) &&
1003             !try_to_release_page(page, GFP_KERNEL))
1004                 return -EAGAIN;
1005         return migrate_page(mapping, newpage, page, mode);
1006 }
1007 #endif
1008
1009
1010 static int btree_writepages(struct address_space *mapping,
1011                             struct writeback_control *wbc)
1012 {
1013         struct btrfs_fs_info *fs_info;
1014         int ret;
1015
1016         if (wbc->sync_mode == WB_SYNC_NONE) {
1017
1018                 if (wbc->for_kupdate)
1019                         return 0;
1020
1021                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1022                 /* this is a bit racy, but that's ok */
1023                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1024                                              BTRFS_DIRTY_METADATA_THRESH);
1025                 if (ret < 0)
1026                         return 0;
1027         }
1028         return btree_write_cache_pages(mapping, wbc);
1029 }
1030
1031 static int btree_readpage(struct file *file, struct page *page)
1032 {
1033         struct extent_io_tree *tree;
1034         tree = &BTRFS_I(page->mapping->host)->io_tree;
1035         return extent_read_full_page(tree, page, btree_get_extent, 0);
1036 }
1037
1038 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1039 {
1040         if (PageWriteback(page) || PageDirty(page))
1041                 return 0;
1042
1043         return try_release_extent_buffer(page);
1044 }
1045
1046 static void btree_invalidatepage(struct page *page, unsigned int offset,
1047                                  unsigned int length)
1048 {
1049         struct extent_io_tree *tree;
1050         tree = &BTRFS_I(page->mapping->host)->io_tree;
1051         extent_invalidatepage(tree, page, offset);
1052         btree_releasepage(page, GFP_NOFS);
1053         if (PagePrivate(page)) {
1054                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1055                            "page private not zero on page %llu",
1056                            (unsigned long long)page_offset(page));
1057                 ClearPagePrivate(page);
1058                 set_page_private(page, 0);
1059                 put_page(page);
1060         }
1061 }
1062
1063 static int btree_set_page_dirty(struct page *page)
1064 {
1065 #ifdef DEBUG
1066         struct extent_buffer *eb;
1067
1068         BUG_ON(!PagePrivate(page));
1069         eb = (struct extent_buffer *)page->private;
1070         BUG_ON(!eb);
1071         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1072         BUG_ON(!atomic_read(&eb->refs));
1073         btrfs_assert_tree_locked(eb);
1074 #endif
1075         return __set_page_dirty_nobuffers(page);
1076 }
1077
1078 static const struct address_space_operations btree_aops = {
1079         .readpage       = btree_readpage,
1080         .writepages     = btree_writepages,
1081         .releasepage    = btree_releasepage,
1082         .invalidatepage = btree_invalidatepage,
1083 #ifdef CONFIG_MIGRATION
1084         .migratepage    = btree_migratepage,
1085 #endif
1086         .set_page_dirty = btree_set_page_dirty,
1087 };
1088
1089 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1090 {
1091         struct extent_buffer *buf = NULL;
1092         struct inode *btree_inode = root->fs_info->btree_inode;
1093
1094         buf = btrfs_find_create_tree_block(root, bytenr);
1095         if (IS_ERR(buf))
1096                 return;
1097         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1098                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1099         free_extent_buffer(buf);
1100 }
1101
1102 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1103                          int mirror_num, struct extent_buffer **eb)
1104 {
1105         struct extent_buffer *buf = NULL;
1106         struct inode *btree_inode = root->fs_info->btree_inode;
1107         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1108         int ret;
1109
1110         buf = btrfs_find_create_tree_block(root, bytenr);
1111         if (IS_ERR(buf))
1112                 return 0;
1113
1114         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1115
1116         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1117                                        btree_get_extent, mirror_num);
1118         if (ret) {
1119                 free_extent_buffer(buf);
1120                 return ret;
1121         }
1122
1123         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1124                 free_extent_buffer(buf);
1125                 return -EIO;
1126         } else if (extent_buffer_uptodate(buf)) {
1127                 *eb = buf;
1128         } else {
1129                 free_extent_buffer(buf);
1130         }
1131         return 0;
1132 }
1133
1134 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1135                                             u64 bytenr)
1136 {
1137         return find_extent_buffer(fs_info, bytenr);
1138 }
1139
1140 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1141                                                  u64 bytenr)
1142 {
1143         if (btrfs_is_testing(root->fs_info))
1144                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1145                                 root->nodesize);
1146         return alloc_extent_buffer(root->fs_info, bytenr);
1147 }
1148
1149
1150 int btrfs_write_tree_block(struct extent_buffer *buf)
1151 {
1152         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1153                                         buf->start + buf->len - 1);
1154 }
1155
1156 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1157 {
1158         return filemap_fdatawait_range(buf->pages[0]->mapping,
1159                                        buf->start, buf->start + buf->len - 1);
1160 }
1161
1162 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1163                                       u64 parent_transid)
1164 {
1165         struct extent_buffer *buf = NULL;
1166         int ret;
1167
1168         buf = btrfs_find_create_tree_block(root, bytenr);
1169         if (IS_ERR(buf))
1170                 return buf;
1171
1172         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1173         if (ret) {
1174                 free_extent_buffer(buf);
1175                 return ERR_PTR(ret);
1176         }
1177         return buf;
1178
1179 }
1180
1181 void clean_tree_block(struct btrfs_trans_handle *trans,
1182                       struct btrfs_fs_info *fs_info,
1183                       struct extent_buffer *buf)
1184 {
1185         if (btrfs_header_generation(buf) ==
1186             fs_info->running_transaction->transid) {
1187                 btrfs_assert_tree_locked(buf);
1188
1189                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1190                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1191                                              -buf->len,
1192                                              fs_info->dirty_metadata_batch);
1193                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1194                         btrfs_set_lock_blocking(buf);
1195                         clear_extent_buffer_dirty(buf);
1196                 }
1197         }
1198 }
1199
1200 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1201 {
1202         struct btrfs_subvolume_writers *writers;
1203         int ret;
1204
1205         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1206         if (!writers)
1207                 return ERR_PTR(-ENOMEM);
1208
1209         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1210         if (ret < 0) {
1211                 kfree(writers);
1212                 return ERR_PTR(ret);
1213         }
1214
1215         init_waitqueue_head(&writers->wait);
1216         return writers;
1217 }
1218
1219 static void
1220 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1221 {
1222         percpu_counter_destroy(&writers->counter);
1223         kfree(writers);
1224 }
1225
1226 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1227                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1228                          u64 objectid)
1229 {
1230         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1231         root->node = NULL;
1232         root->commit_root = NULL;
1233         root->sectorsize = sectorsize;
1234         root->nodesize = nodesize;
1235         root->stripesize = stripesize;
1236         root->state = 0;
1237         root->orphan_cleanup_state = 0;
1238
1239         root->objectid = objectid;
1240         root->last_trans = 0;
1241         root->highest_objectid = 0;
1242         root->nr_delalloc_inodes = 0;
1243         root->nr_ordered_extents = 0;
1244         root->name = NULL;
1245         root->inode_tree = RB_ROOT;
1246         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1247         root->block_rsv = NULL;
1248         root->orphan_block_rsv = NULL;
1249
1250         INIT_LIST_HEAD(&root->dirty_list);
1251         INIT_LIST_HEAD(&root->root_list);
1252         INIT_LIST_HEAD(&root->delalloc_inodes);
1253         INIT_LIST_HEAD(&root->delalloc_root);
1254         INIT_LIST_HEAD(&root->ordered_extents);
1255         INIT_LIST_HEAD(&root->ordered_root);
1256         INIT_LIST_HEAD(&root->logged_list[0]);
1257         INIT_LIST_HEAD(&root->logged_list[1]);
1258         spin_lock_init(&root->orphan_lock);
1259         spin_lock_init(&root->inode_lock);
1260         spin_lock_init(&root->delalloc_lock);
1261         spin_lock_init(&root->ordered_extent_lock);
1262         spin_lock_init(&root->accounting_lock);
1263         spin_lock_init(&root->log_extents_lock[0]);
1264         spin_lock_init(&root->log_extents_lock[1]);
1265         mutex_init(&root->objectid_mutex);
1266         mutex_init(&root->log_mutex);
1267         mutex_init(&root->ordered_extent_mutex);
1268         mutex_init(&root->delalloc_mutex);
1269         init_waitqueue_head(&root->log_writer_wait);
1270         init_waitqueue_head(&root->log_commit_wait[0]);
1271         init_waitqueue_head(&root->log_commit_wait[1]);
1272         INIT_LIST_HEAD(&root->log_ctxs[0]);
1273         INIT_LIST_HEAD(&root->log_ctxs[1]);
1274         atomic_set(&root->log_commit[0], 0);
1275         atomic_set(&root->log_commit[1], 0);
1276         atomic_set(&root->log_writers, 0);
1277         atomic_set(&root->log_batch, 0);
1278         atomic_set(&root->orphan_inodes, 0);
1279         atomic_set(&root->refs, 1);
1280         atomic_set(&root->will_be_snapshoted, 0);
1281         atomic_set(&root->qgroup_meta_rsv, 0);
1282         root->log_transid = 0;
1283         root->log_transid_committed = -1;
1284         root->last_log_commit = 0;
1285         if (!dummy)
1286                 extent_io_tree_init(&root->dirty_log_pages,
1287                                      fs_info->btree_inode->i_mapping);
1288
1289         memset(&root->root_key, 0, sizeof(root->root_key));
1290         memset(&root->root_item, 0, sizeof(root->root_item));
1291         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1292         if (!dummy)
1293                 root->defrag_trans_start = fs_info->generation;
1294         else
1295                 root->defrag_trans_start = 0;
1296         root->root_key.objectid = objectid;
1297         root->anon_dev = 0;
1298
1299         spin_lock_init(&root->root_item_lock);
1300 }
1301
1302 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1303                 gfp_t flags)
1304 {
1305         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1306         if (root)
1307                 root->fs_info = fs_info;
1308         return root;
1309 }
1310
1311 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1312 /* Should only be used by the testing infrastructure */
1313 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1314                                           u32 sectorsize, u32 nodesize)
1315 {
1316         struct btrfs_root *root;
1317
1318         if (!fs_info)
1319                 return ERR_PTR(-EINVAL);
1320
1321         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1322         if (!root)
1323                 return ERR_PTR(-ENOMEM);
1324         /* We don't use the stripesize in selftest, set it as sectorsize */
1325         __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1326                         BTRFS_ROOT_TREE_OBJECTID);
1327         root->alloc_bytenr = 0;
1328
1329         return root;
1330 }
1331 #endif
1332
1333 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1334                                      struct btrfs_fs_info *fs_info,
1335                                      u64 objectid)
1336 {
1337         struct extent_buffer *leaf;
1338         struct btrfs_root *tree_root = fs_info->tree_root;
1339         struct btrfs_root *root;
1340         struct btrfs_key key;
1341         int ret = 0;
1342         uuid_le uuid;
1343
1344         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1345         if (!root)
1346                 return ERR_PTR(-ENOMEM);
1347
1348         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1349                 tree_root->stripesize, root, fs_info, objectid);
1350         root->root_key.objectid = objectid;
1351         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1352         root->root_key.offset = 0;
1353
1354         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1355         if (IS_ERR(leaf)) {
1356                 ret = PTR_ERR(leaf);
1357                 leaf = NULL;
1358                 goto fail;
1359         }
1360
1361         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1362         btrfs_set_header_bytenr(leaf, leaf->start);
1363         btrfs_set_header_generation(leaf, trans->transid);
1364         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1365         btrfs_set_header_owner(leaf, objectid);
1366         root->node = leaf;
1367
1368         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1369                             BTRFS_FSID_SIZE);
1370         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1371                             btrfs_header_chunk_tree_uuid(leaf),
1372                             BTRFS_UUID_SIZE);
1373         btrfs_mark_buffer_dirty(leaf);
1374
1375         root->commit_root = btrfs_root_node(root);
1376         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1377
1378         root->root_item.flags = 0;
1379         root->root_item.byte_limit = 0;
1380         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1381         btrfs_set_root_generation(&root->root_item, trans->transid);
1382         btrfs_set_root_level(&root->root_item, 0);
1383         btrfs_set_root_refs(&root->root_item, 1);
1384         btrfs_set_root_used(&root->root_item, leaf->len);
1385         btrfs_set_root_last_snapshot(&root->root_item, 0);
1386         btrfs_set_root_dirid(&root->root_item, 0);
1387         uuid_le_gen(&uuid);
1388         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1389         root->root_item.drop_level = 0;
1390
1391         key.objectid = objectid;
1392         key.type = BTRFS_ROOT_ITEM_KEY;
1393         key.offset = 0;
1394         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1395         if (ret)
1396                 goto fail;
1397
1398         btrfs_tree_unlock(leaf);
1399
1400         return root;
1401
1402 fail:
1403         if (leaf) {
1404                 btrfs_tree_unlock(leaf);
1405                 free_extent_buffer(root->commit_root);
1406                 free_extent_buffer(leaf);
1407         }
1408         kfree(root);
1409
1410         return ERR_PTR(ret);
1411 }
1412
1413 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1414                                          struct btrfs_fs_info *fs_info)
1415 {
1416         struct btrfs_root *root;
1417         struct btrfs_root *tree_root = fs_info->tree_root;
1418         struct extent_buffer *leaf;
1419
1420         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1421         if (!root)
1422                 return ERR_PTR(-ENOMEM);
1423
1424         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1425                      tree_root->stripesize, root, fs_info,
1426                      BTRFS_TREE_LOG_OBJECTID);
1427
1428         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1429         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1430         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1431
1432         /*
1433          * DON'T set REF_COWS for log trees
1434          *
1435          * log trees do not get reference counted because they go away
1436          * before a real commit is actually done.  They do store pointers
1437          * to file data extents, and those reference counts still get
1438          * updated (along with back refs to the log tree).
1439          */
1440
1441         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1442                         NULL, 0, 0, 0);
1443         if (IS_ERR(leaf)) {
1444                 kfree(root);
1445                 return ERR_CAST(leaf);
1446         }
1447
1448         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1449         btrfs_set_header_bytenr(leaf, leaf->start);
1450         btrfs_set_header_generation(leaf, trans->transid);
1451         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1452         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1453         root->node = leaf;
1454
1455         write_extent_buffer(root->node, root->fs_info->fsid,
1456                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1457         btrfs_mark_buffer_dirty(root->node);
1458         btrfs_tree_unlock(root->node);
1459         return root;
1460 }
1461
1462 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1463                              struct btrfs_fs_info *fs_info)
1464 {
1465         struct btrfs_root *log_root;
1466
1467         log_root = alloc_log_tree(trans, fs_info);
1468         if (IS_ERR(log_root))
1469                 return PTR_ERR(log_root);
1470         WARN_ON(fs_info->log_root_tree);
1471         fs_info->log_root_tree = log_root;
1472         return 0;
1473 }
1474
1475 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1476                        struct btrfs_root *root)
1477 {
1478         struct btrfs_root *log_root;
1479         struct btrfs_inode_item *inode_item;
1480
1481         log_root = alloc_log_tree(trans, root->fs_info);
1482         if (IS_ERR(log_root))
1483                 return PTR_ERR(log_root);
1484
1485         log_root->last_trans = trans->transid;
1486         log_root->root_key.offset = root->root_key.objectid;
1487
1488         inode_item = &log_root->root_item.inode;
1489         btrfs_set_stack_inode_generation(inode_item, 1);
1490         btrfs_set_stack_inode_size(inode_item, 3);
1491         btrfs_set_stack_inode_nlink(inode_item, 1);
1492         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1493         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1494
1495         btrfs_set_root_node(&log_root->root_item, log_root->node);
1496
1497         WARN_ON(root->log_root);
1498         root->log_root = log_root;
1499         root->log_transid = 0;
1500         root->log_transid_committed = -1;
1501         root->last_log_commit = 0;
1502         return 0;
1503 }
1504
1505 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1506                                                struct btrfs_key *key)
1507 {
1508         struct btrfs_root *root;
1509         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1510         struct btrfs_path *path;
1511         u64 generation;
1512         int ret;
1513
1514         path = btrfs_alloc_path();
1515         if (!path)
1516                 return ERR_PTR(-ENOMEM);
1517
1518         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1519         if (!root) {
1520                 ret = -ENOMEM;
1521                 goto alloc_fail;
1522         }
1523
1524         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1525                 tree_root->stripesize, root, fs_info, key->objectid);
1526
1527         ret = btrfs_find_root(tree_root, key, path,
1528                               &root->root_item, &root->root_key);
1529         if (ret) {
1530                 if (ret > 0)
1531                         ret = -ENOENT;
1532                 goto find_fail;
1533         }
1534
1535         generation = btrfs_root_generation(&root->root_item);
1536         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1537                                      generation);
1538         if (IS_ERR(root->node)) {
1539                 ret = PTR_ERR(root->node);
1540                 goto find_fail;
1541         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1542                 ret = -EIO;
1543                 free_extent_buffer(root->node);
1544                 goto find_fail;
1545         }
1546         root->commit_root = btrfs_root_node(root);
1547 out:
1548         btrfs_free_path(path);
1549         return root;
1550
1551 find_fail:
1552         kfree(root);
1553 alloc_fail:
1554         root = ERR_PTR(ret);
1555         goto out;
1556 }
1557
1558 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1559                                       struct btrfs_key *location)
1560 {
1561         struct btrfs_root *root;
1562
1563         root = btrfs_read_tree_root(tree_root, location);
1564         if (IS_ERR(root))
1565                 return root;
1566
1567         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1568                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1569                 btrfs_check_and_init_root_item(&root->root_item);
1570         }
1571
1572         return root;
1573 }
1574
1575 int btrfs_init_fs_root(struct btrfs_root *root)
1576 {
1577         int ret;
1578         struct btrfs_subvolume_writers *writers;
1579
1580         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1581         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1582                                         GFP_NOFS);
1583         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1584                 ret = -ENOMEM;
1585                 goto fail;
1586         }
1587
1588         writers = btrfs_alloc_subvolume_writers();
1589         if (IS_ERR(writers)) {
1590                 ret = PTR_ERR(writers);
1591                 goto fail;
1592         }
1593         root->subv_writers = writers;
1594
1595         btrfs_init_free_ino_ctl(root);
1596         spin_lock_init(&root->ino_cache_lock);
1597         init_waitqueue_head(&root->ino_cache_wait);
1598
1599         ret = get_anon_bdev(&root->anon_dev);
1600         if (ret)
1601                 goto fail;
1602
1603         mutex_lock(&root->objectid_mutex);
1604         ret = btrfs_find_highest_objectid(root,
1605                                         &root->highest_objectid);
1606         if (ret) {
1607                 mutex_unlock(&root->objectid_mutex);
1608                 goto fail;
1609         }
1610
1611         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1612
1613         mutex_unlock(&root->objectid_mutex);
1614
1615         return 0;
1616 fail:
1617         /* the caller is responsible to call free_fs_root */
1618         return ret;
1619 }
1620
1621 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1622                                                u64 root_id)
1623 {
1624         struct btrfs_root *root;
1625
1626         spin_lock(&fs_info->fs_roots_radix_lock);
1627         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1628                                  (unsigned long)root_id);
1629         spin_unlock(&fs_info->fs_roots_radix_lock);
1630         return root;
1631 }
1632
1633 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1634                          struct btrfs_root *root)
1635 {
1636         int ret;
1637
1638         ret = radix_tree_preload(GFP_NOFS);
1639         if (ret)
1640                 return ret;
1641
1642         spin_lock(&fs_info->fs_roots_radix_lock);
1643         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1644                                 (unsigned long)root->root_key.objectid,
1645                                 root);
1646         if (ret == 0)
1647                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1648         spin_unlock(&fs_info->fs_roots_radix_lock);
1649         radix_tree_preload_end();
1650
1651         return ret;
1652 }
1653
1654 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1655                                      struct btrfs_key *location,
1656                                      bool check_ref)
1657 {
1658         struct btrfs_root *root;
1659         struct btrfs_path *path;
1660         struct btrfs_key key;
1661         int ret;
1662
1663         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1664                 return fs_info->tree_root;
1665         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1666                 return fs_info->extent_root;
1667         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1668                 return fs_info->chunk_root;
1669         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1670                 return fs_info->dev_root;
1671         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1672                 return fs_info->csum_root;
1673         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1674                 return fs_info->quota_root ? fs_info->quota_root :
1675                                              ERR_PTR(-ENOENT);
1676         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1677                 return fs_info->uuid_root ? fs_info->uuid_root :
1678                                             ERR_PTR(-ENOENT);
1679         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1680                 return fs_info->free_space_root ? fs_info->free_space_root :
1681                                                   ERR_PTR(-ENOENT);
1682 again:
1683         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1684         if (root) {
1685                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1686                         return ERR_PTR(-ENOENT);
1687                 return root;
1688         }
1689
1690         root = btrfs_read_fs_root(fs_info->tree_root, location);
1691         if (IS_ERR(root))
1692                 return root;
1693
1694         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1695                 ret = -ENOENT;
1696                 goto fail;
1697         }
1698
1699         ret = btrfs_init_fs_root(root);
1700         if (ret)
1701                 goto fail;
1702
1703         path = btrfs_alloc_path();
1704         if (!path) {
1705                 ret = -ENOMEM;
1706                 goto fail;
1707         }
1708         key.objectid = BTRFS_ORPHAN_OBJECTID;
1709         key.type = BTRFS_ORPHAN_ITEM_KEY;
1710         key.offset = location->objectid;
1711
1712         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1713         btrfs_free_path(path);
1714         if (ret < 0)
1715                 goto fail;
1716         if (ret == 0)
1717                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1718
1719         ret = btrfs_insert_fs_root(fs_info, root);
1720         if (ret) {
1721                 if (ret == -EEXIST) {
1722                         free_fs_root(root);
1723                         goto again;
1724                 }
1725                 goto fail;
1726         }
1727         return root;
1728 fail:
1729         free_fs_root(root);
1730         return ERR_PTR(ret);
1731 }
1732
1733 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1734 {
1735         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1736         int ret = 0;
1737         struct btrfs_device *device;
1738         struct backing_dev_info *bdi;
1739
1740         rcu_read_lock();
1741         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1742                 if (!device->bdev)
1743                         continue;
1744                 bdi = blk_get_backing_dev_info(device->bdev);
1745                 if (bdi_congested(bdi, bdi_bits)) {
1746                         ret = 1;
1747                         break;
1748                 }
1749         }
1750         rcu_read_unlock();
1751         return ret;
1752 }
1753
1754 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1755 {
1756         int err;
1757
1758         err = bdi_setup_and_register(bdi, "btrfs");
1759         if (err)
1760                 return err;
1761
1762         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1763         bdi->congested_fn       = btrfs_congested_fn;
1764         bdi->congested_data     = info;
1765         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1766         return 0;
1767 }
1768
1769 /*
1770  * called by the kthread helper functions to finally call the bio end_io
1771  * functions.  This is where read checksum verification actually happens
1772  */
1773 static void end_workqueue_fn(struct btrfs_work *work)
1774 {
1775         struct bio *bio;
1776         struct btrfs_end_io_wq *end_io_wq;
1777
1778         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1779         bio = end_io_wq->bio;
1780
1781         bio->bi_error = end_io_wq->error;
1782         bio->bi_private = end_io_wq->private;
1783         bio->bi_end_io = end_io_wq->end_io;
1784         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1785         bio_endio(bio);
1786 }
1787
1788 static int cleaner_kthread(void *arg)
1789 {
1790         struct btrfs_root *root = arg;
1791         int again;
1792         struct btrfs_trans_handle *trans;
1793
1794         do {
1795                 again = 0;
1796
1797                 /* Make the cleaner go to sleep early. */
1798                 if (btrfs_need_cleaner_sleep(root))
1799                         goto sleep;
1800
1801                 /*
1802                  * Do not do anything if we might cause open_ctree() to block
1803                  * before we have finished mounting the filesystem.
1804                  */
1805                 if (!root->fs_info->open)
1806                         goto sleep;
1807
1808                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1809                         goto sleep;
1810
1811                 /*
1812                  * Avoid the problem that we change the status of the fs
1813                  * during the above check and trylock.
1814                  */
1815                 if (btrfs_need_cleaner_sleep(root)) {
1816                         mutex_unlock(&root->fs_info->cleaner_mutex);
1817                         goto sleep;
1818                 }
1819
1820                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1821                 btrfs_run_delayed_iputs(root);
1822                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1823
1824                 again = btrfs_clean_one_deleted_snapshot(root);
1825                 mutex_unlock(&root->fs_info->cleaner_mutex);
1826
1827                 /*
1828                  * The defragger has dealt with the R/O remount and umount,
1829                  * needn't do anything special here.
1830                  */
1831                 btrfs_run_defrag_inodes(root->fs_info);
1832
1833                 /*
1834                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1835                  * with relocation (btrfs_relocate_chunk) and relocation
1836                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1837                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1838                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1839                  * unused block groups.
1840                  */
1841                 btrfs_delete_unused_bgs(root->fs_info);
1842 sleep:
1843                 if (!again) {
1844                         set_current_state(TASK_INTERRUPTIBLE);
1845                         if (!kthread_should_stop())
1846                                 schedule();
1847                         __set_current_state(TASK_RUNNING);
1848                 }
1849         } while (!kthread_should_stop());
1850
1851         /*
1852          * Transaction kthread is stopped before us and wakes us up.
1853          * However we might have started a new transaction and COWed some
1854          * tree blocks when deleting unused block groups for example. So
1855          * make sure we commit the transaction we started to have a clean
1856          * shutdown when evicting the btree inode - if it has dirty pages
1857          * when we do the final iput() on it, eviction will trigger a
1858          * writeback for it which will fail with null pointer dereferences
1859          * since work queues and other resources were already released and
1860          * destroyed by the time the iput/eviction/writeback is made.
1861          */
1862         trans = btrfs_attach_transaction(root);
1863         if (IS_ERR(trans)) {
1864                 if (PTR_ERR(trans) != -ENOENT)
1865                         btrfs_err(root->fs_info,
1866                                   "cleaner transaction attach returned %ld",
1867                                   PTR_ERR(trans));
1868         } else {
1869                 int ret;
1870
1871                 ret = btrfs_commit_transaction(trans, root);
1872                 if (ret)
1873                         btrfs_err(root->fs_info,
1874                                   "cleaner open transaction commit returned %d",
1875                                   ret);
1876         }
1877
1878         return 0;
1879 }
1880
1881 static int transaction_kthread(void *arg)
1882 {
1883         struct btrfs_root *root = arg;
1884         struct btrfs_trans_handle *trans;
1885         struct btrfs_transaction *cur;
1886         u64 transid;
1887         unsigned long now;
1888         unsigned long delay;
1889         bool cannot_commit;
1890
1891         do {
1892                 cannot_commit = false;
1893                 delay = HZ * root->fs_info->commit_interval;
1894                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1895
1896                 spin_lock(&root->fs_info->trans_lock);
1897                 cur = root->fs_info->running_transaction;
1898                 if (!cur) {
1899                         spin_unlock(&root->fs_info->trans_lock);
1900                         goto sleep;
1901                 }
1902
1903                 now = get_seconds();
1904                 if (cur->state < TRANS_STATE_BLOCKED &&
1905                     (now < cur->start_time ||
1906                      now - cur->start_time < root->fs_info->commit_interval)) {
1907                         spin_unlock(&root->fs_info->trans_lock);
1908                         delay = HZ * 5;
1909                         goto sleep;
1910                 }
1911                 transid = cur->transid;
1912                 spin_unlock(&root->fs_info->trans_lock);
1913
1914                 /* If the file system is aborted, this will always fail. */
1915                 trans = btrfs_attach_transaction(root);
1916                 if (IS_ERR(trans)) {
1917                         if (PTR_ERR(trans) != -ENOENT)
1918                                 cannot_commit = true;
1919                         goto sleep;
1920                 }
1921                 if (transid == trans->transid) {
1922                         btrfs_commit_transaction(trans, root);
1923                 } else {
1924                         btrfs_end_transaction(trans, root);
1925                 }
1926 sleep:
1927                 wake_up_process(root->fs_info->cleaner_kthread);
1928                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1929
1930                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1931                                       &root->fs_info->fs_state)))
1932                         btrfs_cleanup_transaction(root);
1933                 set_current_state(TASK_INTERRUPTIBLE);
1934                 if (!kthread_should_stop() &&
1935                                 (!btrfs_transaction_blocked(root->fs_info) ||
1936                                  cannot_commit))
1937                         schedule_timeout(delay);
1938                 __set_current_state(TASK_RUNNING);
1939         } while (!kthread_should_stop());
1940         return 0;
1941 }
1942
1943 /*
1944  * this will find the highest generation in the array of
1945  * root backups.  The index of the highest array is returned,
1946  * or -1 if we can't find anything.
1947  *
1948  * We check to make sure the array is valid by comparing the
1949  * generation of the latest  root in the array with the generation
1950  * in the super block.  If they don't match we pitch it.
1951  */
1952 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1953 {
1954         u64 cur;
1955         int newest_index = -1;
1956         struct btrfs_root_backup *root_backup;
1957         int i;
1958
1959         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1960                 root_backup = info->super_copy->super_roots + i;
1961                 cur = btrfs_backup_tree_root_gen(root_backup);
1962                 if (cur == newest_gen)
1963                         newest_index = i;
1964         }
1965
1966         /* check to see if we actually wrapped around */
1967         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1968                 root_backup = info->super_copy->super_roots;
1969                 cur = btrfs_backup_tree_root_gen(root_backup);
1970                 if (cur == newest_gen)
1971                         newest_index = 0;
1972         }
1973         return newest_index;
1974 }
1975
1976
1977 /*
1978  * find the oldest backup so we know where to store new entries
1979  * in the backup array.  This will set the backup_root_index
1980  * field in the fs_info struct
1981  */
1982 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1983                                      u64 newest_gen)
1984 {
1985         int newest_index = -1;
1986
1987         newest_index = find_newest_super_backup(info, newest_gen);
1988         /* if there was garbage in there, just move along */
1989         if (newest_index == -1) {
1990                 info->backup_root_index = 0;
1991         } else {
1992                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1993         }
1994 }
1995
1996 /*
1997  * copy all the root pointers into the super backup array.
1998  * this will bump the backup pointer by one when it is
1999  * done
2000  */
2001 static void backup_super_roots(struct btrfs_fs_info *info)
2002 {
2003         int next_backup;
2004         struct btrfs_root_backup *root_backup;
2005         int last_backup;
2006
2007         next_backup = info->backup_root_index;
2008         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2009                 BTRFS_NUM_BACKUP_ROOTS;
2010
2011         /*
2012          * just overwrite the last backup if we're at the same generation
2013          * this happens only at umount
2014          */
2015         root_backup = info->super_for_commit->super_roots + last_backup;
2016         if (btrfs_backup_tree_root_gen(root_backup) ==
2017             btrfs_header_generation(info->tree_root->node))
2018                 next_backup = last_backup;
2019
2020         root_backup = info->super_for_commit->super_roots + next_backup;
2021
2022         /*
2023          * make sure all of our padding and empty slots get zero filled
2024          * regardless of which ones we use today
2025          */
2026         memset(root_backup, 0, sizeof(*root_backup));
2027
2028         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2029
2030         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2031         btrfs_set_backup_tree_root_gen(root_backup,
2032                                btrfs_header_generation(info->tree_root->node));
2033
2034         btrfs_set_backup_tree_root_level(root_backup,
2035                                btrfs_header_level(info->tree_root->node));
2036
2037         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2038         btrfs_set_backup_chunk_root_gen(root_backup,
2039                                btrfs_header_generation(info->chunk_root->node));
2040         btrfs_set_backup_chunk_root_level(root_backup,
2041                                btrfs_header_level(info->chunk_root->node));
2042
2043         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2044         btrfs_set_backup_extent_root_gen(root_backup,
2045                                btrfs_header_generation(info->extent_root->node));
2046         btrfs_set_backup_extent_root_level(root_backup,
2047                                btrfs_header_level(info->extent_root->node));
2048
2049         /*
2050          * we might commit during log recovery, which happens before we set
2051          * the fs_root.  Make sure it is valid before we fill it in.
2052          */
2053         if (info->fs_root && info->fs_root->node) {
2054                 btrfs_set_backup_fs_root(root_backup,
2055                                          info->fs_root->node->start);
2056                 btrfs_set_backup_fs_root_gen(root_backup,
2057                                btrfs_header_generation(info->fs_root->node));
2058                 btrfs_set_backup_fs_root_level(root_backup,
2059                                btrfs_header_level(info->fs_root->node));
2060         }
2061
2062         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2063         btrfs_set_backup_dev_root_gen(root_backup,
2064                                btrfs_header_generation(info->dev_root->node));
2065         btrfs_set_backup_dev_root_level(root_backup,
2066                                        btrfs_header_level(info->dev_root->node));
2067
2068         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2069         btrfs_set_backup_csum_root_gen(root_backup,
2070                                btrfs_header_generation(info->csum_root->node));
2071         btrfs_set_backup_csum_root_level(root_backup,
2072                                btrfs_header_level(info->csum_root->node));
2073
2074         btrfs_set_backup_total_bytes(root_backup,
2075                              btrfs_super_total_bytes(info->super_copy));
2076         btrfs_set_backup_bytes_used(root_backup,
2077                              btrfs_super_bytes_used(info->super_copy));
2078         btrfs_set_backup_num_devices(root_backup,
2079                              btrfs_super_num_devices(info->super_copy));
2080
2081         /*
2082          * if we don't copy this out to the super_copy, it won't get remembered
2083          * for the next commit
2084          */
2085         memcpy(&info->super_copy->super_roots,
2086                &info->super_for_commit->super_roots,
2087                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2088 }
2089
2090 /*
2091  * this copies info out of the root backup array and back into
2092  * the in-memory super block.  It is meant to help iterate through
2093  * the array, so you send it the number of backups you've already
2094  * tried and the last backup index you used.
2095  *
2096  * this returns -1 when it has tried all the backups
2097  */
2098 static noinline int next_root_backup(struct btrfs_fs_info *info,
2099                                      struct btrfs_super_block *super,
2100                                      int *num_backups_tried, int *backup_index)
2101 {
2102         struct btrfs_root_backup *root_backup;
2103         int newest = *backup_index;
2104
2105         if (*num_backups_tried == 0) {
2106                 u64 gen = btrfs_super_generation(super);
2107
2108                 newest = find_newest_super_backup(info, gen);
2109                 if (newest == -1)
2110                         return -1;
2111
2112                 *backup_index = newest;
2113                 *num_backups_tried = 1;
2114         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2115                 /* we've tried all the backups, all done */
2116                 return -1;
2117         } else {
2118                 /* jump to the next oldest backup */
2119                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2120                         BTRFS_NUM_BACKUP_ROOTS;
2121                 *backup_index = newest;
2122                 *num_backups_tried += 1;
2123         }
2124         root_backup = super->super_roots + newest;
2125
2126         btrfs_set_super_generation(super,
2127                                    btrfs_backup_tree_root_gen(root_backup));
2128         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2129         btrfs_set_super_root_level(super,
2130                                    btrfs_backup_tree_root_level(root_backup));
2131         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2132
2133         /*
2134          * fixme: the total bytes and num_devices need to match or we should
2135          * need a fsck
2136          */
2137         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2138         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2139         return 0;
2140 }
2141
2142 /* helper to cleanup workers */
2143 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2144 {
2145         btrfs_destroy_workqueue(fs_info->fixup_workers);
2146         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2147         btrfs_destroy_workqueue(fs_info->workers);
2148         btrfs_destroy_workqueue(fs_info->endio_workers);
2149         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2150         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2151         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2152         btrfs_destroy_workqueue(fs_info->rmw_workers);
2153         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2154         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2155         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2156         btrfs_destroy_workqueue(fs_info->submit_workers);
2157         btrfs_destroy_workqueue(fs_info->delayed_workers);
2158         btrfs_destroy_workqueue(fs_info->caching_workers);
2159         btrfs_destroy_workqueue(fs_info->readahead_workers);
2160         btrfs_destroy_workqueue(fs_info->flush_workers);
2161         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2162         btrfs_destroy_workqueue(fs_info->extent_workers);
2163 }
2164
2165 static void free_root_extent_buffers(struct btrfs_root *root)
2166 {
2167         if (root) {
2168                 free_extent_buffer(root->node);
2169                 free_extent_buffer(root->commit_root);
2170                 root->node = NULL;
2171                 root->commit_root = NULL;
2172         }
2173 }
2174
2175 /* helper to cleanup tree roots */
2176 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2177 {
2178         free_root_extent_buffers(info->tree_root);
2179
2180         free_root_extent_buffers(info->dev_root);
2181         free_root_extent_buffers(info->extent_root);
2182         free_root_extent_buffers(info->csum_root);
2183         free_root_extent_buffers(info->quota_root);
2184         free_root_extent_buffers(info->uuid_root);
2185         if (chunk_root)
2186                 free_root_extent_buffers(info->chunk_root);
2187         free_root_extent_buffers(info->free_space_root);
2188 }
2189
2190 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2191 {
2192         int ret;
2193         struct btrfs_root *gang[8];
2194         int i;
2195
2196         while (!list_empty(&fs_info->dead_roots)) {
2197                 gang[0] = list_entry(fs_info->dead_roots.next,
2198                                      struct btrfs_root, root_list);
2199                 list_del(&gang[0]->root_list);
2200
2201                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2202                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2203                 } else {
2204                         free_extent_buffer(gang[0]->node);
2205                         free_extent_buffer(gang[0]->commit_root);
2206                         btrfs_put_fs_root(gang[0]);
2207                 }
2208         }
2209
2210         while (1) {
2211                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2212                                              (void **)gang, 0,
2213                                              ARRAY_SIZE(gang));
2214                 if (!ret)
2215                         break;
2216                 for (i = 0; i < ret; i++)
2217                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2218         }
2219
2220         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2221                 btrfs_free_log_root_tree(NULL, fs_info);
2222                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2223                                             fs_info->pinned_extents);
2224         }
2225 }
2226
2227 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2228 {
2229         mutex_init(&fs_info->scrub_lock);
2230         atomic_set(&fs_info->scrubs_running, 0);
2231         atomic_set(&fs_info->scrub_pause_req, 0);
2232         atomic_set(&fs_info->scrubs_paused, 0);
2233         atomic_set(&fs_info->scrub_cancel_req, 0);
2234         init_waitqueue_head(&fs_info->scrub_pause_wait);
2235         fs_info->scrub_workers_refcnt = 0;
2236 }
2237
2238 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2239 {
2240         spin_lock_init(&fs_info->balance_lock);
2241         mutex_init(&fs_info->balance_mutex);
2242         atomic_set(&fs_info->balance_running, 0);
2243         atomic_set(&fs_info->balance_pause_req, 0);
2244         atomic_set(&fs_info->balance_cancel_req, 0);
2245         fs_info->balance_ctl = NULL;
2246         init_waitqueue_head(&fs_info->balance_wait_q);
2247 }
2248
2249 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2250                                    struct btrfs_root *tree_root)
2251 {
2252         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2253         set_nlink(fs_info->btree_inode, 1);
2254         /*
2255          * we set the i_size on the btree inode to the max possible int.
2256          * the real end of the address space is determined by all of
2257          * the devices in the system
2258          */
2259         fs_info->btree_inode->i_size = OFFSET_MAX;
2260         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2261
2262         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2263         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2264                              fs_info->btree_inode->i_mapping);
2265         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2266         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2267
2268         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2269
2270         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2271         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2272                sizeof(struct btrfs_key));
2273         set_bit(BTRFS_INODE_DUMMY,
2274                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2275         btrfs_insert_inode_hash(fs_info->btree_inode);
2276 }
2277
2278 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2279 {
2280         fs_info->dev_replace.lock_owner = 0;
2281         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2282         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2283         rwlock_init(&fs_info->dev_replace.lock);
2284         atomic_set(&fs_info->dev_replace.read_locks, 0);
2285         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2286         init_waitqueue_head(&fs_info->replace_wait);
2287         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2288 }
2289
2290 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2291 {
2292         spin_lock_init(&fs_info->qgroup_lock);
2293         mutex_init(&fs_info->qgroup_ioctl_lock);
2294         fs_info->qgroup_tree = RB_ROOT;
2295         fs_info->qgroup_op_tree = RB_ROOT;
2296         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2297         fs_info->qgroup_seq = 1;
2298         fs_info->quota_enabled = 0;
2299         fs_info->pending_quota_state = 0;
2300         fs_info->qgroup_ulist = NULL;
2301         mutex_init(&fs_info->qgroup_rescan_lock);
2302 }
2303
2304 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2305                 struct btrfs_fs_devices *fs_devices)
2306 {
2307         int max_active = fs_info->thread_pool_size;
2308         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2309
2310         fs_info->workers =
2311                 btrfs_alloc_workqueue(fs_info, "worker",
2312                                       flags | WQ_HIGHPRI, max_active, 16);
2313
2314         fs_info->delalloc_workers =
2315                 btrfs_alloc_workqueue(fs_info, "delalloc",
2316                                       flags, max_active, 2);
2317
2318         fs_info->flush_workers =
2319                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2320                                       flags, max_active, 0);
2321
2322         fs_info->caching_workers =
2323                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2324
2325         /*
2326          * a higher idle thresh on the submit workers makes it much more
2327          * likely that bios will be send down in a sane order to the
2328          * devices
2329          */
2330         fs_info->submit_workers =
2331                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2332                                       min_t(u64, fs_devices->num_devices,
2333                                             max_active), 64);
2334
2335         fs_info->fixup_workers =
2336                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2337
2338         /*
2339          * endios are largely parallel and should have a very
2340          * low idle thresh
2341          */
2342         fs_info->endio_workers =
2343                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2344         fs_info->endio_meta_workers =
2345                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2346                                       max_active, 4);
2347         fs_info->endio_meta_write_workers =
2348                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2349                                       max_active, 2);
2350         fs_info->endio_raid56_workers =
2351                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2352                                       max_active, 4);
2353         fs_info->endio_repair_workers =
2354                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2355         fs_info->rmw_workers =
2356                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2357         fs_info->endio_write_workers =
2358                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2359                                       max_active, 2);
2360         fs_info->endio_freespace_worker =
2361                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2362                                       max_active, 0);
2363         fs_info->delayed_workers =
2364                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2365                                       max_active, 0);
2366         fs_info->readahead_workers =
2367                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2368                                       max_active, 2);
2369         fs_info->qgroup_rescan_workers =
2370                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2371         fs_info->extent_workers =
2372                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2373                                       min_t(u64, fs_devices->num_devices,
2374                                             max_active), 8);
2375
2376         if (!(fs_info->workers && fs_info->delalloc_workers &&
2377               fs_info->submit_workers && fs_info->flush_workers &&
2378               fs_info->endio_workers && fs_info->endio_meta_workers &&
2379               fs_info->endio_meta_write_workers &&
2380               fs_info->endio_repair_workers &&
2381               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2382               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2383               fs_info->caching_workers && fs_info->readahead_workers &&
2384               fs_info->fixup_workers && fs_info->delayed_workers &&
2385               fs_info->extent_workers &&
2386               fs_info->qgroup_rescan_workers)) {
2387                 return -ENOMEM;
2388         }
2389
2390         return 0;
2391 }
2392
2393 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2394                             struct btrfs_fs_devices *fs_devices)
2395 {
2396         int ret;
2397         struct btrfs_root *tree_root = fs_info->tree_root;
2398         struct btrfs_root *log_tree_root;
2399         struct btrfs_super_block *disk_super = fs_info->super_copy;
2400         u64 bytenr = btrfs_super_log_root(disk_super);
2401
2402         if (fs_devices->rw_devices == 0) {
2403                 btrfs_warn(fs_info, "log replay required on RO media");
2404                 return -EIO;
2405         }
2406
2407         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2408         if (!log_tree_root)
2409                 return -ENOMEM;
2410
2411         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2412                         tree_root->stripesize, log_tree_root, fs_info,
2413                         BTRFS_TREE_LOG_OBJECTID);
2414
2415         log_tree_root->node = read_tree_block(tree_root, bytenr,
2416                         fs_info->generation + 1);
2417         if (IS_ERR(log_tree_root->node)) {
2418                 btrfs_warn(fs_info, "failed to read log tree");
2419                 ret = PTR_ERR(log_tree_root->node);
2420                 kfree(log_tree_root);
2421                 return ret;
2422         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2423                 btrfs_err(fs_info, "failed to read log tree");
2424                 free_extent_buffer(log_tree_root->node);
2425                 kfree(log_tree_root);
2426                 return -EIO;
2427         }
2428         /* returns with log_tree_root freed on success */
2429         ret = btrfs_recover_log_trees(log_tree_root);
2430         if (ret) {
2431                 btrfs_handle_fs_error(tree_root->fs_info, ret,
2432                             "Failed to recover log tree");
2433                 free_extent_buffer(log_tree_root->node);
2434                 kfree(log_tree_root);
2435                 return ret;
2436         }
2437
2438         if (fs_info->sb->s_flags & MS_RDONLY) {
2439                 ret = btrfs_commit_super(tree_root);
2440                 if (ret)
2441                         return ret;
2442         }
2443
2444         return 0;
2445 }
2446
2447 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2448                             struct btrfs_root *tree_root)
2449 {
2450         struct btrfs_root *root;
2451         struct btrfs_key location;
2452         int ret;
2453
2454         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2455         location.type = BTRFS_ROOT_ITEM_KEY;
2456         location.offset = 0;
2457
2458         root = btrfs_read_tree_root(tree_root, &location);
2459         if (IS_ERR(root))
2460                 return PTR_ERR(root);
2461         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2462         fs_info->extent_root = root;
2463
2464         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2465         root = btrfs_read_tree_root(tree_root, &location);
2466         if (IS_ERR(root))
2467                 return PTR_ERR(root);
2468         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2469         fs_info->dev_root = root;
2470         btrfs_init_devices_late(fs_info);
2471
2472         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2473         root = btrfs_read_tree_root(tree_root, &location);
2474         if (IS_ERR(root))
2475                 return PTR_ERR(root);
2476         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2477         fs_info->csum_root = root;
2478
2479         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2480         root = btrfs_read_tree_root(tree_root, &location);
2481         if (!IS_ERR(root)) {
2482                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2483                 fs_info->quota_enabled = 1;
2484                 fs_info->pending_quota_state = 1;
2485                 fs_info->quota_root = root;
2486         }
2487
2488         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2489         root = btrfs_read_tree_root(tree_root, &location);
2490         if (IS_ERR(root)) {
2491                 ret = PTR_ERR(root);
2492                 if (ret != -ENOENT)
2493                         return ret;
2494         } else {
2495                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496                 fs_info->uuid_root = root;
2497         }
2498
2499         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2500                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2501                 root = btrfs_read_tree_root(tree_root, &location);
2502                 if (IS_ERR(root))
2503                         return PTR_ERR(root);
2504                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2505                 fs_info->free_space_root = root;
2506         }
2507
2508         return 0;
2509 }
2510
2511 int open_ctree(struct super_block *sb,
2512                struct btrfs_fs_devices *fs_devices,
2513                char *options)
2514 {
2515         u32 sectorsize;
2516         u32 nodesize;
2517         u32 stripesize;
2518         u64 generation;
2519         u64 features;
2520         struct btrfs_key location;
2521         struct buffer_head *bh;
2522         struct btrfs_super_block *disk_super;
2523         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2524         struct btrfs_root *tree_root;
2525         struct btrfs_root *chunk_root;
2526         int ret;
2527         int err = -EINVAL;
2528         int num_backups_tried = 0;
2529         int backup_index = 0;
2530         int max_active;
2531
2532         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2533         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2534         if (!tree_root || !chunk_root) {
2535                 err = -ENOMEM;
2536                 goto fail;
2537         }
2538
2539         ret = init_srcu_struct(&fs_info->subvol_srcu);
2540         if (ret) {
2541                 err = ret;
2542                 goto fail;
2543         }
2544
2545         ret = setup_bdi(fs_info, &fs_info->bdi);
2546         if (ret) {
2547                 err = ret;
2548                 goto fail_srcu;
2549         }
2550
2551         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2552         if (ret) {
2553                 err = ret;
2554                 goto fail_bdi;
2555         }
2556         fs_info->dirty_metadata_batch = PAGE_SIZE *
2557                                         (1 + ilog2(nr_cpu_ids));
2558
2559         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2560         if (ret) {
2561                 err = ret;
2562                 goto fail_dirty_metadata_bytes;
2563         }
2564
2565         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2566         if (ret) {
2567                 err = ret;
2568                 goto fail_delalloc_bytes;
2569         }
2570
2571         fs_info->btree_inode = new_inode(sb);
2572         if (!fs_info->btree_inode) {
2573                 err = -ENOMEM;
2574                 goto fail_bio_counter;
2575         }
2576
2577         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2578
2579         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2580         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2581         INIT_LIST_HEAD(&fs_info->trans_list);
2582         INIT_LIST_HEAD(&fs_info->dead_roots);
2583         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2584         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2585         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2586         spin_lock_init(&fs_info->delalloc_root_lock);
2587         spin_lock_init(&fs_info->trans_lock);
2588         spin_lock_init(&fs_info->fs_roots_radix_lock);
2589         spin_lock_init(&fs_info->delayed_iput_lock);
2590         spin_lock_init(&fs_info->defrag_inodes_lock);
2591         spin_lock_init(&fs_info->free_chunk_lock);
2592         spin_lock_init(&fs_info->tree_mod_seq_lock);
2593         spin_lock_init(&fs_info->super_lock);
2594         spin_lock_init(&fs_info->qgroup_op_lock);
2595         spin_lock_init(&fs_info->buffer_lock);
2596         spin_lock_init(&fs_info->unused_bgs_lock);
2597         rwlock_init(&fs_info->tree_mod_log_lock);
2598         mutex_init(&fs_info->unused_bg_unpin_mutex);
2599         mutex_init(&fs_info->delete_unused_bgs_mutex);
2600         mutex_init(&fs_info->reloc_mutex);
2601         mutex_init(&fs_info->delalloc_root_mutex);
2602         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2603         seqlock_init(&fs_info->profiles_lock);
2604
2605         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2606         INIT_LIST_HEAD(&fs_info->space_info);
2607         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2608         INIT_LIST_HEAD(&fs_info->unused_bgs);
2609         btrfs_mapping_init(&fs_info->mapping_tree);
2610         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2611                              BTRFS_BLOCK_RSV_GLOBAL);
2612         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2613                              BTRFS_BLOCK_RSV_DELALLOC);
2614         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2615         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2616         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2617         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2618                              BTRFS_BLOCK_RSV_DELOPS);
2619         atomic_set(&fs_info->nr_async_submits, 0);
2620         atomic_set(&fs_info->async_delalloc_pages, 0);
2621         atomic_set(&fs_info->async_submit_draining, 0);
2622         atomic_set(&fs_info->nr_async_bios, 0);
2623         atomic_set(&fs_info->defrag_running, 0);
2624         atomic_set(&fs_info->qgroup_op_seq, 0);
2625         atomic_set(&fs_info->reada_works_cnt, 0);
2626         atomic64_set(&fs_info->tree_mod_seq, 0);
2627         fs_info->sb = sb;
2628         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2629         fs_info->metadata_ratio = 0;
2630         fs_info->defrag_inodes = RB_ROOT;
2631         fs_info->free_chunk_space = 0;
2632         fs_info->tree_mod_log = RB_ROOT;
2633         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2634         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2635         /* readahead state */
2636         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2637         spin_lock_init(&fs_info->reada_lock);
2638
2639         fs_info->thread_pool_size = min_t(unsigned long,
2640                                           num_online_cpus() + 2, 8);
2641
2642         INIT_LIST_HEAD(&fs_info->ordered_roots);
2643         spin_lock_init(&fs_info->ordered_root_lock);
2644         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2645                                         GFP_KERNEL);
2646         if (!fs_info->delayed_root) {
2647                 err = -ENOMEM;
2648                 goto fail_iput;
2649         }
2650         btrfs_init_delayed_root(fs_info->delayed_root);
2651
2652         btrfs_init_scrub(fs_info);
2653 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2654         fs_info->check_integrity_print_mask = 0;
2655 #endif
2656         btrfs_init_balance(fs_info);
2657         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2658
2659         sb->s_blocksize = 4096;
2660         sb->s_blocksize_bits = blksize_bits(4096);
2661         sb->s_bdi = &fs_info->bdi;
2662
2663         btrfs_init_btree_inode(fs_info, tree_root);
2664
2665         spin_lock_init(&fs_info->block_group_cache_lock);
2666         fs_info->block_group_cache_tree = RB_ROOT;
2667         fs_info->first_logical_byte = (u64)-1;
2668
2669         extent_io_tree_init(&fs_info->freed_extents[0],
2670                              fs_info->btree_inode->i_mapping);
2671         extent_io_tree_init(&fs_info->freed_extents[1],
2672                              fs_info->btree_inode->i_mapping);
2673         fs_info->pinned_extents = &fs_info->freed_extents[0];
2674         fs_info->do_barriers = 1;
2675
2676
2677         mutex_init(&fs_info->ordered_operations_mutex);
2678         mutex_init(&fs_info->tree_log_mutex);
2679         mutex_init(&fs_info->chunk_mutex);
2680         mutex_init(&fs_info->transaction_kthread_mutex);
2681         mutex_init(&fs_info->cleaner_mutex);
2682         mutex_init(&fs_info->volume_mutex);
2683         mutex_init(&fs_info->ro_block_group_mutex);
2684         init_rwsem(&fs_info->commit_root_sem);
2685         init_rwsem(&fs_info->cleanup_work_sem);
2686         init_rwsem(&fs_info->subvol_sem);
2687         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2688
2689         btrfs_init_dev_replace_locks(fs_info);
2690         btrfs_init_qgroup(fs_info);
2691
2692         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2693         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2694
2695         init_waitqueue_head(&fs_info->transaction_throttle);
2696         init_waitqueue_head(&fs_info->transaction_wait);
2697         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2698         init_waitqueue_head(&fs_info->async_submit_wait);
2699
2700         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2701
2702         ret = btrfs_alloc_stripe_hash_table(fs_info);
2703         if (ret) {
2704                 err = ret;
2705                 goto fail_alloc;
2706         }
2707
2708         __setup_root(4096, 4096, 4096, tree_root,
2709                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2710
2711         invalidate_bdev(fs_devices->latest_bdev);
2712
2713         /*
2714          * Read super block and check the signature bytes only
2715          */
2716         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2717         if (IS_ERR(bh)) {
2718                 err = PTR_ERR(bh);
2719                 goto fail_alloc;
2720         }
2721
2722         /*
2723          * We want to check superblock checksum, the type is stored inside.
2724          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2725          */
2726         if (btrfs_check_super_csum(bh->b_data)) {
2727                 btrfs_err(fs_info, "superblock checksum mismatch");
2728                 err = -EINVAL;
2729                 brelse(bh);
2730                 goto fail_alloc;
2731         }
2732
2733         /*
2734          * super_copy is zeroed at allocation time and we never touch the
2735          * following bytes up to INFO_SIZE, the checksum is calculated from
2736          * the whole block of INFO_SIZE
2737          */
2738         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2739         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2740                sizeof(*fs_info->super_for_commit));
2741         brelse(bh);
2742
2743         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2744
2745         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2746         if (ret) {
2747                 btrfs_err(fs_info, "superblock contains fatal errors");
2748                 err = -EINVAL;
2749                 goto fail_alloc;
2750         }
2751
2752         disk_super = fs_info->super_copy;
2753         if (!btrfs_super_root(disk_super))
2754                 goto fail_alloc;
2755
2756         /* check FS state, whether FS is broken. */
2757         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2758                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2759
2760         /*
2761          * run through our array of backup supers and setup
2762          * our ring pointer to the oldest one
2763          */
2764         generation = btrfs_super_generation(disk_super);
2765         find_oldest_super_backup(fs_info, generation);
2766
2767         /*
2768          * In the long term, we'll store the compression type in the super
2769          * block, and it'll be used for per file compression control.
2770          */
2771         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2772
2773         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2774         if (ret) {
2775                 err = ret;
2776                 goto fail_alloc;
2777         }
2778
2779         features = btrfs_super_incompat_flags(disk_super) &
2780                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2781         if (features) {
2782                 btrfs_err(fs_info,
2783                     "cannot mount because of unsupported optional features (%llx)",
2784                     features);
2785                 err = -EINVAL;
2786                 goto fail_alloc;
2787         }
2788
2789         features = btrfs_super_incompat_flags(disk_super);
2790         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2791         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2792                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2793
2794         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2795                 btrfs_info(fs_info, "has skinny extents");
2796
2797         /*
2798          * flag our filesystem as having big metadata blocks if
2799          * they are bigger than the page size
2800          */
2801         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2802                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2803                         btrfs_info(fs_info,
2804                                 "flagging fs with big metadata feature");
2805                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2806         }
2807
2808         nodesize = btrfs_super_nodesize(disk_super);
2809         sectorsize = btrfs_super_sectorsize(disk_super);
2810         stripesize = sectorsize;
2811         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2812         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2813
2814         /*
2815          * mixed block groups end up with duplicate but slightly offset
2816          * extent buffers for the same range.  It leads to corruptions
2817          */
2818         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2819             (sectorsize != nodesize)) {
2820                 btrfs_err(fs_info,
2821 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2822                         nodesize, sectorsize);
2823                 goto fail_alloc;
2824         }
2825
2826         /*
2827          * Needn't use the lock because there is no other task which will
2828          * update the flag.
2829          */
2830         btrfs_set_super_incompat_flags(disk_super, features);
2831
2832         features = btrfs_super_compat_ro_flags(disk_super) &
2833                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2834         if (!(sb->s_flags & MS_RDONLY) && features) {
2835                 btrfs_err(fs_info,
2836         "cannot mount read-write because of unsupported optional features (%llx)",
2837                        features);
2838                 err = -EINVAL;
2839                 goto fail_alloc;
2840         }
2841
2842         max_active = fs_info->thread_pool_size;
2843
2844         ret = btrfs_init_workqueues(fs_info, fs_devices);
2845         if (ret) {
2846                 err = ret;
2847                 goto fail_sb_buffer;
2848         }
2849
2850         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2851         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2852                                     SZ_4M / PAGE_SIZE);
2853
2854         tree_root->nodesize = nodesize;
2855         tree_root->sectorsize = sectorsize;
2856         tree_root->stripesize = stripesize;
2857
2858         sb->s_blocksize = sectorsize;
2859         sb->s_blocksize_bits = blksize_bits(sectorsize);
2860
2861         mutex_lock(&fs_info->chunk_mutex);
2862         ret = btrfs_read_sys_array(tree_root);
2863         mutex_unlock(&fs_info->chunk_mutex);
2864         if (ret) {
2865                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2866                 goto fail_sb_buffer;
2867         }
2868
2869         generation = btrfs_super_chunk_root_generation(disk_super);
2870
2871         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2872                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2873
2874         chunk_root->node = read_tree_block(chunk_root,
2875                                            btrfs_super_chunk_root(disk_super),
2876                                            generation);
2877         if (IS_ERR(chunk_root->node) ||
2878             !extent_buffer_uptodate(chunk_root->node)) {
2879                 btrfs_err(fs_info, "failed to read chunk root");
2880                 if (!IS_ERR(chunk_root->node))
2881                         free_extent_buffer(chunk_root->node);
2882                 chunk_root->node = NULL;
2883                 goto fail_tree_roots;
2884         }
2885         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2886         chunk_root->commit_root = btrfs_root_node(chunk_root);
2887
2888         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2889            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2890
2891         ret = btrfs_read_chunk_tree(chunk_root);
2892         if (ret) {
2893                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2894                 goto fail_tree_roots;
2895         }
2896
2897         /*
2898          * keep the device that is marked to be the target device for the
2899          * dev_replace procedure
2900          */
2901         btrfs_close_extra_devices(fs_devices, 0);
2902
2903         if (!fs_devices->latest_bdev) {
2904                 btrfs_err(fs_info, "failed to read devices");
2905                 goto fail_tree_roots;
2906         }
2907
2908 retry_root_backup:
2909         generation = btrfs_super_generation(disk_super);
2910
2911         tree_root->node = read_tree_block(tree_root,
2912                                           btrfs_super_root(disk_super),
2913                                           generation);
2914         if (IS_ERR(tree_root->node) ||
2915             !extent_buffer_uptodate(tree_root->node)) {
2916                 btrfs_warn(fs_info, "failed to read tree root");
2917                 if (!IS_ERR(tree_root->node))
2918                         free_extent_buffer(tree_root->node);
2919                 tree_root->node = NULL;
2920                 goto recovery_tree_root;
2921         }
2922
2923         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2924         tree_root->commit_root = btrfs_root_node(tree_root);
2925         btrfs_set_root_refs(&tree_root->root_item, 1);
2926
2927         mutex_lock(&tree_root->objectid_mutex);
2928         ret = btrfs_find_highest_objectid(tree_root,
2929                                         &tree_root->highest_objectid);
2930         if (ret) {
2931                 mutex_unlock(&tree_root->objectid_mutex);
2932                 goto recovery_tree_root;
2933         }
2934
2935         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2936
2937         mutex_unlock(&tree_root->objectid_mutex);
2938
2939         ret = btrfs_read_roots(fs_info, tree_root);
2940         if (ret)
2941                 goto recovery_tree_root;
2942
2943         fs_info->generation = generation;
2944         fs_info->last_trans_committed = generation;
2945
2946         ret = btrfs_recover_balance(fs_info);
2947         if (ret) {
2948                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2949                 goto fail_block_groups;
2950         }
2951
2952         ret = btrfs_init_dev_stats(fs_info);
2953         if (ret) {
2954                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2955                 goto fail_block_groups;
2956         }
2957
2958         ret = btrfs_init_dev_replace(fs_info);
2959         if (ret) {
2960                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2961                 goto fail_block_groups;
2962         }
2963
2964         btrfs_close_extra_devices(fs_devices, 1);
2965
2966         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2967         if (ret) {
2968                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2969                                 ret);
2970                 goto fail_block_groups;
2971         }
2972
2973         ret = btrfs_sysfs_add_device(fs_devices);
2974         if (ret) {
2975                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2976                                 ret);
2977                 goto fail_fsdev_sysfs;
2978         }
2979
2980         ret = btrfs_sysfs_add_mounted(fs_info);
2981         if (ret) {
2982                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2983                 goto fail_fsdev_sysfs;
2984         }
2985
2986         ret = btrfs_init_space_info(fs_info);
2987         if (ret) {
2988                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2989                 goto fail_sysfs;
2990         }
2991
2992         ret = btrfs_read_block_groups(fs_info->extent_root);
2993         if (ret) {
2994                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2995                 goto fail_sysfs;
2996         }
2997         fs_info->num_tolerated_disk_barrier_failures =
2998                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2999         if (fs_info->fs_devices->missing_devices >
3000              fs_info->num_tolerated_disk_barrier_failures &&
3001             !(sb->s_flags & MS_RDONLY)) {
3002                 btrfs_warn(fs_info,
3003 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3004                         fs_info->fs_devices->missing_devices,
3005                         fs_info->num_tolerated_disk_barrier_failures);
3006                 goto fail_sysfs;
3007         }
3008
3009         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3010                                                "btrfs-cleaner");
3011         if (IS_ERR(fs_info->cleaner_kthread))
3012                 goto fail_sysfs;
3013
3014         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3015                                                    tree_root,
3016                                                    "btrfs-transaction");
3017         if (IS_ERR(fs_info->transaction_kthread))
3018                 goto fail_cleaner;
3019
3020         if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3021             !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3022             !fs_info->fs_devices->rotating) {
3023                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3024                 btrfs_set_opt(fs_info->mount_opt, SSD);
3025         }
3026
3027         /*
3028          * Mount does not set all options immediately, we can do it now and do
3029          * not have to wait for transaction commit
3030          */
3031         btrfs_apply_pending_changes(fs_info);
3032
3033 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3034         if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3035                 ret = btrfsic_mount(tree_root, fs_devices,
3036                                     btrfs_test_opt(tree_root->fs_info,
3037                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3038                                     1 : 0,
3039                                     fs_info->check_integrity_print_mask);
3040                 if (ret)
3041                         btrfs_warn(fs_info,
3042                                 "failed to initialize integrity check module: %d",
3043                                 ret);
3044         }
3045 #endif
3046         ret = btrfs_read_qgroup_config(fs_info);
3047         if (ret)
3048                 goto fail_trans_kthread;
3049
3050         /* do not make disk changes in broken FS or nologreplay is given */
3051         if (btrfs_super_log_root(disk_super) != 0 &&
3052             !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3053                 ret = btrfs_replay_log(fs_info, fs_devices);
3054                 if (ret) {
3055                         err = ret;
3056                         goto fail_qgroup;
3057                 }
3058         }
3059
3060         ret = btrfs_find_orphan_roots(tree_root);
3061         if (ret)
3062                 goto fail_qgroup;
3063
3064         if (!(sb->s_flags & MS_RDONLY)) {
3065                 ret = btrfs_cleanup_fs_roots(fs_info);
3066                 if (ret)
3067                         goto fail_qgroup;
3068
3069                 mutex_lock(&fs_info->cleaner_mutex);
3070                 ret = btrfs_recover_relocation(tree_root);
3071                 mutex_unlock(&fs_info->cleaner_mutex);
3072                 if (ret < 0) {
3073                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3074                                         ret);
3075                         err = -EINVAL;
3076                         goto fail_qgroup;
3077                 }
3078         }
3079
3080         location.objectid = BTRFS_FS_TREE_OBJECTID;
3081         location.type = BTRFS_ROOT_ITEM_KEY;
3082         location.offset = 0;
3083
3084         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3085         if (IS_ERR(fs_info->fs_root)) {
3086                 err = PTR_ERR(fs_info->fs_root);
3087                 goto fail_qgroup;
3088         }
3089
3090         if (sb->s_flags & MS_RDONLY)
3091                 return 0;
3092
3093         if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3094             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3095                 btrfs_info(fs_info, "creating free space tree");
3096                 ret = btrfs_create_free_space_tree(fs_info);
3097                 if (ret) {
3098                         btrfs_warn(fs_info,
3099                                 "failed to create free space tree: %d", ret);
3100                         close_ctree(tree_root);
3101                         return ret;
3102                 }
3103         }
3104
3105         down_read(&fs_info->cleanup_work_sem);
3106         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3107             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3108                 up_read(&fs_info->cleanup_work_sem);
3109                 close_ctree(tree_root);
3110                 return ret;
3111         }
3112         up_read(&fs_info->cleanup_work_sem);
3113
3114         ret = btrfs_resume_balance_async(fs_info);
3115         if (ret) {
3116                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3117                 close_ctree(tree_root);
3118                 return ret;
3119         }
3120
3121         ret = btrfs_resume_dev_replace_async(fs_info);
3122         if (ret) {
3123                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3124                 close_ctree(tree_root);
3125                 return ret;
3126         }
3127
3128         btrfs_qgroup_rescan_resume(fs_info);
3129
3130         if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3131             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3132                 btrfs_info(fs_info, "clearing free space tree");
3133                 ret = btrfs_clear_free_space_tree(fs_info);
3134                 if (ret) {
3135                         btrfs_warn(fs_info,
3136                                 "failed to clear free space tree: %d", ret);
3137                         close_ctree(tree_root);
3138                         return ret;
3139                 }
3140         }
3141
3142         if (!fs_info->uuid_root) {
3143                 btrfs_info(fs_info, "creating UUID tree");
3144                 ret = btrfs_create_uuid_tree(fs_info);
3145                 if (ret) {
3146                         btrfs_warn(fs_info,
3147                                 "failed to create the UUID tree: %d", ret);
3148                         close_ctree(tree_root);
3149                         return ret;
3150                 }
3151         } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3152                    fs_info->generation !=
3153                                 btrfs_super_uuid_tree_generation(disk_super)) {
3154                 btrfs_info(fs_info, "checking UUID tree");
3155                 ret = btrfs_check_uuid_tree(fs_info);
3156                 if (ret) {
3157                         btrfs_warn(fs_info,
3158                                 "failed to check the UUID tree: %d", ret);
3159                         close_ctree(tree_root);
3160                         return ret;
3161                 }
3162         } else {
3163                 fs_info->update_uuid_tree_gen = 1;
3164         }
3165
3166         fs_info->open = 1;
3167
3168         /*
3169          * backuproot only affect mount behavior, and if open_ctree succeeded,
3170          * no need to keep the flag
3171          */
3172         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3173
3174         return 0;
3175
3176 fail_qgroup:
3177         btrfs_free_qgroup_config(fs_info);
3178 fail_trans_kthread:
3179         kthread_stop(fs_info->transaction_kthread);
3180         btrfs_cleanup_transaction(fs_info->tree_root);
3181         btrfs_free_fs_roots(fs_info);
3182 fail_cleaner:
3183         kthread_stop(fs_info->cleaner_kthread);
3184
3185         /*
3186          * make sure we're done with the btree inode before we stop our
3187          * kthreads
3188          */
3189         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3190
3191 fail_sysfs:
3192         btrfs_sysfs_remove_mounted(fs_info);
3193
3194 fail_fsdev_sysfs:
3195         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3196
3197 fail_block_groups:
3198         btrfs_put_block_group_cache(fs_info);
3199         btrfs_free_block_groups(fs_info);
3200
3201 fail_tree_roots:
3202         free_root_pointers(fs_info, 1);
3203         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3204
3205 fail_sb_buffer:
3206         btrfs_stop_all_workers(fs_info);
3207 fail_alloc:
3208 fail_iput:
3209         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3210
3211         iput(fs_info->btree_inode);
3212 fail_bio_counter:
3213         percpu_counter_destroy(&fs_info->bio_counter);
3214 fail_delalloc_bytes:
3215         percpu_counter_destroy(&fs_info->delalloc_bytes);
3216 fail_dirty_metadata_bytes:
3217         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3218 fail_bdi:
3219         bdi_destroy(&fs_info->bdi);
3220 fail_srcu:
3221         cleanup_srcu_struct(&fs_info->subvol_srcu);
3222 fail:
3223         btrfs_free_stripe_hash_table(fs_info);
3224         btrfs_close_devices(fs_info->fs_devices);
3225         return err;
3226
3227 recovery_tree_root:
3228         if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3229                 goto fail_tree_roots;
3230
3231         free_root_pointers(fs_info, 0);
3232
3233         /* don't use the log in recovery mode, it won't be valid */
3234         btrfs_set_super_log_root(disk_super, 0);
3235
3236         /* we can't trust the free space cache either */
3237         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3238
3239         ret = next_root_backup(fs_info, fs_info->super_copy,
3240                                &num_backups_tried, &backup_index);
3241         if (ret == -1)
3242                 goto fail_block_groups;
3243         goto retry_root_backup;
3244 }
3245
3246 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3247 {
3248         if (uptodate) {
3249                 set_buffer_uptodate(bh);
3250         } else {
3251                 struct btrfs_device *device = (struct btrfs_device *)
3252                         bh->b_private;
3253
3254                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3255                                 "lost page write due to IO error on %s",
3256                                           rcu_str_deref(device->name));
3257                 /* note, we don't set_buffer_write_io_error because we have
3258                  * our own ways of dealing with the IO errors
3259                  */
3260                 clear_buffer_uptodate(bh);
3261                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3262         }
3263         unlock_buffer(bh);
3264         put_bh(bh);
3265 }
3266
3267 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3268                         struct buffer_head **bh_ret)
3269 {
3270         struct buffer_head *bh;
3271         struct btrfs_super_block *super;
3272         u64 bytenr;
3273
3274         bytenr = btrfs_sb_offset(copy_num);
3275         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3276                 return -EINVAL;
3277
3278         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3279         /*
3280          * If we fail to read from the underlying devices, as of now
3281          * the best option we have is to mark it EIO.
3282          */
3283         if (!bh)
3284                 return -EIO;
3285
3286         super = (struct btrfs_super_block *)bh->b_data;
3287         if (btrfs_super_bytenr(super) != bytenr ||
3288                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3289                 brelse(bh);
3290                 return -EINVAL;
3291         }
3292
3293         *bh_ret = bh;
3294         return 0;
3295 }
3296
3297
3298 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3299 {
3300         struct buffer_head *bh;
3301         struct buffer_head *latest = NULL;
3302         struct btrfs_super_block *super;
3303         int i;
3304         u64 transid = 0;
3305         int ret = -EINVAL;
3306
3307         /* we would like to check all the supers, but that would make
3308          * a btrfs mount succeed after a mkfs from a different FS.
3309          * So, we need to add a special mount option to scan for
3310          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3311          */
3312         for (i = 0; i < 1; i++) {
3313                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3314                 if (ret)
3315                         continue;
3316
3317                 super = (struct btrfs_super_block *)bh->b_data;
3318
3319                 if (!latest || btrfs_super_generation(super) > transid) {
3320                         brelse(latest);
3321                         latest = bh;
3322                         transid = btrfs_super_generation(super);
3323                 } else {
3324                         brelse(bh);
3325                 }
3326         }
3327
3328         if (!latest)
3329                 return ERR_PTR(ret);
3330
3331         return latest;
3332 }
3333
3334 /*
3335  * this should be called twice, once with wait == 0 and
3336  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3337  * we write are pinned.
3338  *
3339  * They are released when wait == 1 is done.
3340  * max_mirrors must be the same for both runs, and it indicates how
3341  * many supers on this one device should be written.
3342  *
3343  * max_mirrors == 0 means to write them all.
3344  */
3345 static int write_dev_supers(struct btrfs_device *device,
3346                             struct btrfs_super_block *sb,
3347                             int do_barriers, int wait, int max_mirrors)
3348 {
3349         struct buffer_head *bh;
3350         int i;
3351         int ret;
3352         int errors = 0;
3353         u32 crc;
3354         u64 bytenr;
3355
3356         if (max_mirrors == 0)
3357                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3358
3359         for (i = 0; i < max_mirrors; i++) {
3360                 bytenr = btrfs_sb_offset(i);
3361                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3362                     device->commit_total_bytes)
3363                         break;
3364
3365                 if (wait) {
3366                         bh = __find_get_block(device->bdev, bytenr / 4096,
3367                                               BTRFS_SUPER_INFO_SIZE);
3368                         if (!bh) {
3369                                 errors++;
3370                                 continue;
3371                         }
3372                         wait_on_buffer(bh);
3373                         if (!buffer_uptodate(bh))
3374                                 errors++;
3375
3376                         /* drop our reference */
3377                         brelse(bh);
3378
3379                         /* drop the reference from the wait == 0 run */
3380                         brelse(bh);
3381                         continue;
3382                 } else {
3383                         btrfs_set_super_bytenr(sb, bytenr);
3384
3385                         crc = ~(u32)0;
3386                         crc = btrfs_csum_data((char *)sb +
3387                                               BTRFS_CSUM_SIZE, crc,
3388                                               BTRFS_SUPER_INFO_SIZE -
3389                                               BTRFS_CSUM_SIZE);
3390                         btrfs_csum_final(crc, sb->csum);
3391
3392                         /*
3393                          * one reference for us, and we leave it for the
3394                          * caller
3395                          */
3396                         bh = __getblk(device->bdev, bytenr / 4096,
3397                                       BTRFS_SUPER_INFO_SIZE);
3398                         if (!bh) {
3399                                 btrfs_err(device->dev_root->fs_info,
3400                                     "couldn't get super buffer head for bytenr %llu",
3401                                     bytenr);
3402                                 errors++;
3403                                 continue;
3404                         }
3405
3406                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3407
3408                         /* one reference for submit_bh */
3409                         get_bh(bh);
3410
3411                         set_buffer_uptodate(bh);
3412                         lock_buffer(bh);
3413                         bh->b_end_io = btrfs_end_buffer_write_sync;
3414                         bh->b_private = device;
3415                 }
3416
3417                 /*
3418                  * we fua the first super.  The others we allow
3419                  * to go down lazy.
3420                  */
3421                 if (i == 0)
3422                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_FUA, bh);
3423                 else
3424                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_SYNC, bh);
3425                 if (ret)
3426                         errors++;
3427         }
3428         return errors < i ? 0 : -1;
3429 }
3430
3431 /*
3432  * endio for the write_dev_flush, this will wake anyone waiting
3433  * for the barrier when it is done
3434  */
3435 static void btrfs_end_empty_barrier(struct bio *bio)
3436 {
3437         if (bio->bi_private)
3438                 complete(bio->bi_private);
3439         bio_put(bio);
3440 }
3441
3442 /*
3443  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3444  * sent down.  With wait == 1, it waits for the previous flush.
3445  *
3446  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3447  * capable
3448  */
3449 static int write_dev_flush(struct btrfs_device *device, int wait)
3450 {
3451         struct bio *bio;
3452         int ret = 0;
3453
3454         if (device->nobarriers)
3455                 return 0;
3456
3457         if (wait) {
3458                 bio = device->flush_bio;
3459                 if (!bio)
3460                         return 0;
3461
3462                 wait_for_completion(&device->flush_wait);
3463
3464                 if (bio->bi_error) {
3465                         ret = bio->bi_error;
3466                         btrfs_dev_stat_inc_and_print(device,
3467                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3468                 }
3469
3470                 /* drop the reference from the wait == 0 run */
3471                 bio_put(bio);
3472                 device->flush_bio = NULL;
3473
3474                 return ret;
3475         }
3476
3477         /*
3478          * one reference for us, and we leave it for the
3479          * caller
3480          */
3481         device->flush_bio = NULL;
3482         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3483         if (!bio)
3484                 return -ENOMEM;
3485
3486         bio->bi_end_io = btrfs_end_empty_barrier;
3487         bio->bi_bdev = device->bdev;
3488         bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
3489         init_completion(&device->flush_wait);
3490         bio->bi_private = &device->flush_wait;
3491         device->flush_bio = bio;
3492
3493         bio_get(bio);
3494         btrfsic_submit_bio(bio);
3495
3496         return 0;
3497 }
3498
3499 /*
3500  * send an empty flush down to each device in parallel,
3501  * then wait for them
3502  */
3503 static int barrier_all_devices(struct btrfs_fs_info *info)
3504 {
3505         struct list_head *head;
3506         struct btrfs_device *dev;
3507         int errors_send = 0;
3508         int errors_wait = 0;
3509         int ret;
3510
3511         /* send down all the barriers */
3512         head = &info->fs_devices->devices;
3513         list_for_each_entry_rcu(dev, head, dev_list) {
3514                 if (dev->missing)
3515                         continue;
3516                 if (!dev->bdev) {
3517                         errors_send++;
3518                         continue;
3519                 }
3520                 if (!dev->in_fs_metadata || !dev->writeable)
3521                         continue;
3522
3523                 ret = write_dev_flush(dev, 0);
3524                 if (ret)
3525                         errors_send++;
3526         }
3527
3528         /* wait for all the barriers */
3529         list_for_each_entry_rcu(dev, head, dev_list) {
3530                 if (dev->missing)
3531                         continue;
3532                 if (!dev->bdev) {
3533                         errors_wait++;
3534                         continue;
3535                 }
3536                 if (!dev->in_fs_metadata || !dev->writeable)
3537                         continue;
3538
3539                 ret = write_dev_flush(dev, 1);
3540                 if (ret)
3541                         errors_wait++;
3542         }
3543         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3544             errors_wait > info->num_tolerated_disk_barrier_failures)
3545                 return -EIO;
3546         return 0;
3547 }
3548
3549 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3550 {
3551         int raid_type;
3552         int min_tolerated = INT_MAX;
3553
3554         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3555             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3556                 min_tolerated = min(min_tolerated,
3557                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3558                                     tolerated_failures);
3559
3560         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3561                 if (raid_type == BTRFS_RAID_SINGLE)
3562                         continue;
3563                 if (!(flags & btrfs_raid_group[raid_type]))
3564                         continue;
3565                 min_tolerated = min(min_tolerated,
3566                                     btrfs_raid_array[raid_type].
3567                                     tolerated_failures);
3568         }
3569
3570         if (min_tolerated == INT_MAX) {
3571                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3572                 min_tolerated = 0;
3573         }
3574
3575         return min_tolerated;
3576 }
3577
3578 int btrfs_calc_num_tolerated_disk_barrier_failures(
3579         struct btrfs_fs_info *fs_info)
3580 {
3581         struct btrfs_ioctl_space_info space;
3582         struct btrfs_space_info *sinfo;
3583         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3584                        BTRFS_BLOCK_GROUP_SYSTEM,
3585                        BTRFS_BLOCK_GROUP_METADATA,
3586                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3587         int i;
3588         int c;
3589         int num_tolerated_disk_barrier_failures =
3590                 (int)fs_info->fs_devices->num_devices;
3591
3592         for (i = 0; i < ARRAY_SIZE(types); i++) {
3593                 struct btrfs_space_info *tmp;
3594
3595                 sinfo = NULL;
3596                 rcu_read_lock();
3597                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3598                         if (tmp->flags == types[i]) {
3599                                 sinfo = tmp;
3600                                 break;
3601                         }
3602                 }
3603                 rcu_read_unlock();
3604
3605                 if (!sinfo)
3606                         continue;
3607
3608                 down_read(&sinfo->groups_sem);
3609                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3610                         u64 flags;
3611
3612                         if (list_empty(&sinfo->block_groups[c]))
3613                                 continue;
3614
3615                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3616                                                    &space);
3617                         if (space.total_bytes == 0 || space.used_bytes == 0)
3618                                 continue;
3619                         flags = space.flags;
3620
3621                         num_tolerated_disk_barrier_failures = min(
3622                                 num_tolerated_disk_barrier_failures,
3623                                 btrfs_get_num_tolerated_disk_barrier_failures(
3624                                         flags));
3625                 }
3626                 up_read(&sinfo->groups_sem);
3627         }
3628
3629         return num_tolerated_disk_barrier_failures;
3630 }
3631
3632 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3633 {
3634         struct list_head *head;
3635         struct btrfs_device *dev;
3636         struct btrfs_super_block *sb;
3637         struct btrfs_dev_item *dev_item;
3638         int ret;
3639         int do_barriers;
3640         int max_errors;
3641         int total_errors = 0;
3642         u64 flags;
3643
3644         do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3645         backup_super_roots(root->fs_info);
3646
3647         sb = root->fs_info->super_for_commit;
3648         dev_item = &sb->dev_item;
3649
3650         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3651         head = &root->fs_info->fs_devices->devices;
3652         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3653
3654         if (do_barriers) {
3655                 ret = barrier_all_devices(root->fs_info);
3656                 if (ret) {
3657                         mutex_unlock(
3658                                 &root->fs_info->fs_devices->device_list_mutex);
3659                         btrfs_handle_fs_error(root->fs_info, ret,
3660                                     "errors while submitting device barriers.");
3661                         return ret;
3662                 }
3663         }
3664
3665         list_for_each_entry_rcu(dev, head, dev_list) {
3666                 if (!dev->bdev) {
3667                         total_errors++;
3668                         continue;
3669                 }
3670                 if (!dev->in_fs_metadata || !dev->writeable)
3671                         continue;
3672
3673                 btrfs_set_stack_device_generation(dev_item, 0);
3674                 btrfs_set_stack_device_type(dev_item, dev->type);
3675                 btrfs_set_stack_device_id(dev_item, dev->devid);
3676                 btrfs_set_stack_device_total_bytes(dev_item,
3677                                                    dev->commit_total_bytes);
3678                 btrfs_set_stack_device_bytes_used(dev_item,
3679                                                   dev->commit_bytes_used);
3680                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3681                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3682                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3683                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3684                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3685
3686                 flags = btrfs_super_flags(sb);
3687                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3688
3689                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3690                 if (ret)
3691                         total_errors++;
3692         }
3693         if (total_errors > max_errors) {
3694                 btrfs_err(root->fs_info, "%d errors while writing supers",
3695                        total_errors);
3696                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3697
3698                 /* FUA is masked off if unsupported and can't be the reason */
3699                 btrfs_handle_fs_error(root->fs_info, -EIO,
3700                             "%d errors while writing supers", total_errors);
3701                 return -EIO;
3702         }
3703
3704         total_errors = 0;
3705         list_for_each_entry_rcu(dev, head, dev_list) {
3706                 if (!dev->bdev)
3707                         continue;
3708                 if (!dev->in_fs_metadata || !dev->writeable)
3709                         continue;
3710
3711                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3712                 if (ret)
3713                         total_errors++;
3714         }
3715         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3716         if (total_errors > max_errors) {
3717                 btrfs_handle_fs_error(root->fs_info, -EIO,
3718                             "%d errors while writing supers", total_errors);
3719                 return -EIO;
3720         }
3721         return 0;
3722 }
3723
3724 int write_ctree_super(struct btrfs_trans_handle *trans,
3725                       struct btrfs_root *root, int max_mirrors)
3726 {
3727         return write_all_supers(root, max_mirrors);
3728 }
3729
3730 /* Drop a fs root from the radix tree and free it. */
3731 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3732                                   struct btrfs_root *root)
3733 {
3734         spin_lock(&fs_info->fs_roots_radix_lock);
3735         radix_tree_delete(&fs_info->fs_roots_radix,
3736                           (unsigned long)root->root_key.objectid);
3737         spin_unlock(&fs_info->fs_roots_radix_lock);
3738
3739         if (btrfs_root_refs(&root->root_item) == 0)
3740                 synchronize_srcu(&fs_info->subvol_srcu);
3741
3742         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3743                 btrfs_free_log(NULL, root);
3744
3745         if (root->free_ino_pinned)
3746                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3747         if (root->free_ino_ctl)
3748                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3749         free_fs_root(root);
3750 }
3751
3752 static void free_fs_root(struct btrfs_root *root)
3753 {
3754         iput(root->ino_cache_inode);
3755         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3756         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3757         root->orphan_block_rsv = NULL;
3758         if (root->anon_dev)
3759                 free_anon_bdev(root->anon_dev);
3760         if (root->subv_writers)
3761                 btrfs_free_subvolume_writers(root->subv_writers);
3762         free_extent_buffer(root->node);
3763         free_extent_buffer(root->commit_root);
3764         kfree(root->free_ino_ctl);
3765         kfree(root->free_ino_pinned);
3766         kfree(root->name);
3767         btrfs_put_fs_root(root);
3768 }
3769
3770 void btrfs_free_fs_root(struct btrfs_root *root)
3771 {
3772         free_fs_root(root);
3773 }
3774
3775 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3776 {
3777         u64 root_objectid = 0;
3778         struct btrfs_root *gang[8];
3779         int i = 0;
3780         int err = 0;
3781         unsigned int ret = 0;
3782         int index;
3783
3784         while (1) {
3785                 index = srcu_read_lock(&fs_info->subvol_srcu);
3786                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3787                                              (void **)gang, root_objectid,
3788                                              ARRAY_SIZE(gang));
3789                 if (!ret) {
3790                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3791                         break;
3792                 }
3793                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3794
3795                 for (i = 0; i < ret; i++) {
3796                         /* Avoid to grab roots in dead_roots */
3797                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3798                                 gang[i] = NULL;
3799                                 continue;
3800                         }
3801                         /* grab all the search result for later use */
3802                         gang[i] = btrfs_grab_fs_root(gang[i]);
3803                 }
3804                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3805
3806                 for (i = 0; i < ret; i++) {
3807                         if (!gang[i])
3808                                 continue;
3809                         root_objectid = gang[i]->root_key.objectid;
3810                         err = btrfs_orphan_cleanup(gang[i]);
3811                         if (err)
3812                                 break;
3813                         btrfs_put_fs_root(gang[i]);
3814                 }
3815                 root_objectid++;
3816         }
3817
3818         /* release the uncleaned roots due to error */
3819         for (; i < ret; i++) {
3820                 if (gang[i])
3821                         btrfs_put_fs_root(gang[i]);
3822         }
3823         return err;
3824 }
3825
3826 int btrfs_commit_super(struct btrfs_root *root)
3827 {
3828         struct btrfs_trans_handle *trans;
3829
3830         mutex_lock(&root->fs_info->cleaner_mutex);
3831         btrfs_run_delayed_iputs(root);
3832         mutex_unlock(&root->fs_info->cleaner_mutex);
3833         wake_up_process(root->fs_info->cleaner_kthread);
3834
3835         /* wait until ongoing cleanup work done */
3836         down_write(&root->fs_info->cleanup_work_sem);
3837         up_write(&root->fs_info->cleanup_work_sem);
3838
3839         trans = btrfs_join_transaction(root);
3840         if (IS_ERR(trans))
3841                 return PTR_ERR(trans);
3842         return btrfs_commit_transaction(trans, root);
3843 }
3844
3845 void close_ctree(struct btrfs_root *root)
3846 {
3847         struct btrfs_fs_info *fs_info = root->fs_info;
3848         int ret;
3849
3850         fs_info->closing = 1;
3851         smp_mb();
3852
3853         /* wait for the qgroup rescan worker to stop */
3854         btrfs_qgroup_wait_for_completion(fs_info);
3855
3856         /* wait for the uuid_scan task to finish */
3857         down(&fs_info->uuid_tree_rescan_sem);
3858         /* avoid complains from lockdep et al., set sem back to initial state */
3859         up(&fs_info->uuid_tree_rescan_sem);
3860
3861         /* pause restriper - we want to resume on mount */
3862         btrfs_pause_balance(fs_info);
3863
3864         btrfs_dev_replace_suspend_for_unmount(fs_info);
3865
3866         btrfs_scrub_cancel(fs_info);
3867
3868         /* wait for any defraggers to finish */
3869         wait_event(fs_info->transaction_wait,
3870                    (atomic_read(&fs_info->defrag_running) == 0));
3871
3872         /* clear out the rbtree of defraggable inodes */
3873         btrfs_cleanup_defrag_inodes(fs_info);
3874
3875         cancel_work_sync(&fs_info->async_reclaim_work);
3876
3877         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3878                 /*
3879                  * If the cleaner thread is stopped and there are
3880                  * block groups queued for removal, the deletion will be
3881                  * skipped when we quit the cleaner thread.
3882                  */
3883                 btrfs_delete_unused_bgs(root->fs_info);
3884
3885                 ret = btrfs_commit_super(root);
3886                 if (ret)
3887                         btrfs_err(fs_info, "commit super ret %d", ret);
3888         }
3889
3890         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3891                 btrfs_error_commit_super(root);
3892
3893         kthread_stop(fs_info->transaction_kthread);
3894         kthread_stop(fs_info->cleaner_kthread);
3895
3896         fs_info->closing = 2;
3897         smp_mb();
3898
3899         btrfs_free_qgroup_config(fs_info);
3900
3901         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3902                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3903                        percpu_counter_sum(&fs_info->delalloc_bytes));
3904         }
3905
3906         btrfs_sysfs_remove_mounted(fs_info);
3907         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3908
3909         btrfs_free_fs_roots(fs_info);
3910
3911         btrfs_put_block_group_cache(fs_info);
3912
3913         btrfs_free_block_groups(fs_info);
3914
3915         /*
3916          * we must make sure there is not any read request to
3917          * submit after we stopping all workers.
3918          */
3919         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3920         btrfs_stop_all_workers(fs_info);
3921
3922         fs_info->open = 0;
3923         free_root_pointers(fs_info, 1);
3924
3925         iput(fs_info->btree_inode);
3926
3927 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3928         if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
3929                 btrfsic_unmount(root, fs_info->fs_devices);
3930 #endif
3931
3932         btrfs_close_devices(fs_info->fs_devices);
3933         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3934
3935         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3936         percpu_counter_destroy(&fs_info->delalloc_bytes);
3937         percpu_counter_destroy(&fs_info->bio_counter);
3938         bdi_destroy(&fs_info->bdi);
3939         cleanup_srcu_struct(&fs_info->subvol_srcu);
3940
3941         btrfs_free_stripe_hash_table(fs_info);
3942
3943         __btrfs_free_block_rsv(root->orphan_block_rsv);
3944         root->orphan_block_rsv = NULL;
3945
3946         lock_chunks(root);
3947         while (!list_empty(&fs_info->pinned_chunks)) {
3948                 struct extent_map *em;
3949
3950                 em = list_first_entry(&fs_info->pinned_chunks,
3951                                       struct extent_map, list);
3952                 list_del_init(&em->list);
3953                 free_extent_map(em);
3954         }
3955         unlock_chunks(root);
3956 }
3957
3958 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3959                           int atomic)
3960 {
3961         int ret;
3962         struct inode *btree_inode = buf->pages[0]->mapping->host;
3963
3964         ret = extent_buffer_uptodate(buf);
3965         if (!ret)
3966                 return ret;
3967
3968         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3969                                     parent_transid, atomic);
3970         if (ret == -EAGAIN)
3971                 return ret;
3972         return !ret;
3973 }
3974
3975 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3976 {
3977         struct btrfs_root *root;
3978         u64 transid = btrfs_header_generation(buf);
3979         int was_dirty;
3980
3981 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3982         /*
3983          * This is a fast path so only do this check if we have sanity tests
3984          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3985          * outside of the sanity tests.
3986          */
3987         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3988                 return;
3989 #endif
3990         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3991         btrfs_assert_tree_locked(buf);
3992         if (transid != root->fs_info->generation)
3993                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3994                        "found %llu running %llu\n",
3995                         buf->start, transid, root->fs_info->generation);
3996         was_dirty = set_extent_buffer_dirty(buf);
3997         if (!was_dirty)
3998                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3999                                      buf->len,
4000                                      root->fs_info->dirty_metadata_batch);
4001 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4002         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4003                 btrfs_print_leaf(root, buf);
4004                 ASSERT(0);
4005         }
4006 #endif
4007 }
4008
4009 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4010                                         int flush_delayed)
4011 {
4012         /*
4013          * looks as though older kernels can get into trouble with
4014          * this code, they end up stuck in balance_dirty_pages forever
4015          */
4016         int ret;
4017
4018         if (current->flags & PF_MEMALLOC)
4019                 return;
4020
4021         if (flush_delayed)
4022                 btrfs_balance_delayed_items(root);
4023
4024         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4025                                      BTRFS_DIRTY_METADATA_THRESH);
4026         if (ret > 0) {
4027                 balance_dirty_pages_ratelimited(
4028                                    root->fs_info->btree_inode->i_mapping);
4029         }
4030 }
4031
4032 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4033 {
4034         __btrfs_btree_balance_dirty(root, 1);
4035 }
4036
4037 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4038 {
4039         __btrfs_btree_balance_dirty(root, 0);
4040 }
4041
4042 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4043 {
4044         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4045         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4046 }
4047
4048 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4049                               int read_only)
4050 {
4051         struct btrfs_super_block *sb = fs_info->super_copy;
4052         u64 nodesize = btrfs_super_nodesize(sb);
4053         u64 sectorsize = btrfs_super_sectorsize(sb);
4054         int ret = 0;
4055
4056         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4057                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4058                 ret = -EINVAL;
4059         }
4060         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4061                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4062                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4063         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4064                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4065                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4066                 ret = -EINVAL;
4067         }
4068         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4069                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4070                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4071                 ret = -EINVAL;
4072         }
4073         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4074                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4075                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4076                 ret = -EINVAL;
4077         }
4078
4079         /*
4080          * Check sectorsize and nodesize first, other check will need it.
4081          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4082          */
4083         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4084             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4085                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4086                 ret = -EINVAL;
4087         }
4088         /* Only PAGE SIZE is supported yet */
4089         if (sectorsize != PAGE_SIZE) {
4090                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4091                                 sectorsize, PAGE_SIZE);
4092                 ret = -EINVAL;
4093         }
4094         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4095             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4096                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4097                 ret = -EINVAL;
4098         }
4099         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4100                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4101                                 le32_to_cpu(sb->__unused_leafsize),
4102                                 nodesize);
4103                 ret = -EINVAL;
4104         }
4105
4106         /* Root alignment check */
4107         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4108                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4109                                 btrfs_super_root(sb));
4110                 ret = -EINVAL;
4111         }
4112         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4113                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4114                                 btrfs_super_chunk_root(sb));
4115                 ret = -EINVAL;
4116         }
4117         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4118                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4119                                 btrfs_super_log_root(sb));
4120                 ret = -EINVAL;
4121         }
4122
4123         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4124                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4125                                 fs_info->fsid, sb->dev_item.fsid);
4126                 ret = -EINVAL;
4127         }
4128
4129         /*
4130          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4131          * done later
4132          */
4133         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4134                 btrfs_err(fs_info, "bytes_used is too small %llu",
4135                        btrfs_super_bytes_used(sb));
4136                 ret = -EINVAL;
4137         }
4138         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4139                 btrfs_err(fs_info, "invalid stripesize %u",
4140                        btrfs_super_stripesize(sb));
4141                 ret = -EINVAL;
4142         }
4143         if (btrfs_super_num_devices(sb) > (1UL << 31))
4144                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4145                                 btrfs_super_num_devices(sb));
4146         if (btrfs_super_num_devices(sb) == 0) {
4147                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4148                 ret = -EINVAL;
4149         }
4150
4151         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4152                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4153                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4154                 ret = -EINVAL;
4155         }
4156
4157         /*
4158          * Obvious sys_chunk_array corruptions, it must hold at least one key
4159          * and one chunk
4160          */
4161         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4162                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4163                                 btrfs_super_sys_array_size(sb),
4164                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4165                 ret = -EINVAL;
4166         }
4167         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4168                         + sizeof(struct btrfs_chunk)) {
4169                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4170                                 btrfs_super_sys_array_size(sb),
4171                                 sizeof(struct btrfs_disk_key)
4172                                 + sizeof(struct btrfs_chunk));
4173                 ret = -EINVAL;
4174         }
4175
4176         /*
4177          * The generation is a global counter, we'll trust it more than the others
4178          * but it's still possible that it's the one that's wrong.
4179          */
4180         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4181                 printk(KERN_WARNING
4182                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4183                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4184         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4185             && btrfs_super_cache_generation(sb) != (u64)-1)
4186                 printk(KERN_WARNING
4187                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4188                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4189
4190         return ret;
4191 }
4192
4193 static void btrfs_error_commit_super(struct btrfs_root *root)
4194 {
4195         mutex_lock(&root->fs_info->cleaner_mutex);
4196         btrfs_run_delayed_iputs(root);
4197         mutex_unlock(&root->fs_info->cleaner_mutex);
4198
4199         down_write(&root->fs_info->cleanup_work_sem);
4200         up_write(&root->fs_info->cleanup_work_sem);
4201
4202         /* cleanup FS via transaction */
4203         btrfs_cleanup_transaction(root);
4204 }
4205
4206 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4207 {
4208         struct btrfs_ordered_extent *ordered;
4209
4210         spin_lock(&root->ordered_extent_lock);
4211         /*
4212          * This will just short circuit the ordered completion stuff which will
4213          * make sure the ordered extent gets properly cleaned up.
4214          */
4215         list_for_each_entry(ordered, &root->ordered_extents,
4216                             root_extent_list)
4217                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4218         spin_unlock(&root->ordered_extent_lock);
4219 }
4220
4221 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4222 {
4223         struct btrfs_root *root;
4224         struct list_head splice;
4225
4226         INIT_LIST_HEAD(&splice);
4227
4228         spin_lock(&fs_info->ordered_root_lock);
4229         list_splice_init(&fs_info->ordered_roots, &splice);
4230         while (!list_empty(&splice)) {
4231                 root = list_first_entry(&splice, struct btrfs_root,
4232                                         ordered_root);
4233                 list_move_tail(&root->ordered_root,
4234                                &fs_info->ordered_roots);
4235
4236                 spin_unlock(&fs_info->ordered_root_lock);
4237                 btrfs_destroy_ordered_extents(root);
4238
4239                 cond_resched();
4240                 spin_lock(&fs_info->ordered_root_lock);
4241         }
4242         spin_unlock(&fs_info->ordered_root_lock);
4243 }
4244
4245 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4246                                       struct btrfs_root *root)
4247 {
4248         struct rb_node *node;
4249         struct btrfs_delayed_ref_root *delayed_refs;
4250         struct btrfs_delayed_ref_node *ref;
4251         int ret = 0;
4252
4253         delayed_refs = &trans->delayed_refs;
4254
4255         spin_lock(&delayed_refs->lock);
4256         if (atomic_read(&delayed_refs->num_entries) == 0) {
4257                 spin_unlock(&delayed_refs->lock);
4258                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4259                 return ret;
4260         }
4261
4262         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4263                 struct btrfs_delayed_ref_head *head;
4264                 struct btrfs_delayed_ref_node *tmp;
4265                 bool pin_bytes = false;
4266
4267                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4268                                 href_node);
4269                 if (!mutex_trylock(&head->mutex)) {
4270                         atomic_inc(&head->node.refs);
4271                         spin_unlock(&delayed_refs->lock);
4272
4273                         mutex_lock(&head->mutex);
4274                         mutex_unlock(&head->mutex);
4275                         btrfs_put_delayed_ref(&head->node);
4276                         spin_lock(&delayed_refs->lock);
4277                         continue;
4278                 }
4279                 spin_lock(&head->lock);
4280                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4281                                                  list) {
4282                         ref->in_tree = 0;
4283                         list_del(&ref->list);
4284                         atomic_dec(&delayed_refs->num_entries);
4285                         btrfs_put_delayed_ref(ref);
4286                 }
4287                 if (head->must_insert_reserved)
4288                         pin_bytes = true;
4289                 btrfs_free_delayed_extent_op(head->extent_op);
4290                 delayed_refs->num_heads--;
4291                 if (head->processing == 0)
4292                         delayed_refs->num_heads_ready--;
4293                 atomic_dec(&delayed_refs->num_entries);
4294                 head->node.in_tree = 0;
4295                 rb_erase(&head->href_node, &delayed_refs->href_root);
4296                 spin_unlock(&head->lock);
4297                 spin_unlock(&delayed_refs->lock);
4298                 mutex_unlock(&head->mutex);
4299
4300                 if (pin_bytes)
4301                         btrfs_pin_extent(root, head->node.bytenr,
4302                                          head->node.num_bytes, 1);
4303                 btrfs_put_delayed_ref(&head->node);
4304                 cond_resched();
4305                 spin_lock(&delayed_refs->lock);
4306         }
4307
4308         spin_unlock(&delayed_refs->lock);
4309
4310         return ret;
4311 }
4312
4313 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4314 {
4315         struct btrfs_inode *btrfs_inode;
4316         struct list_head splice;
4317
4318         INIT_LIST_HEAD(&splice);
4319
4320         spin_lock(&root->delalloc_lock);
4321         list_splice_init(&root->delalloc_inodes, &splice);
4322
4323         while (!list_empty(&splice)) {
4324                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4325                                                delalloc_inodes);
4326
4327                 list_del_init(&btrfs_inode->delalloc_inodes);
4328                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4329                           &btrfs_inode->runtime_flags);
4330                 spin_unlock(&root->delalloc_lock);
4331
4332                 btrfs_invalidate_inodes(btrfs_inode->root);
4333
4334                 spin_lock(&root->delalloc_lock);
4335         }
4336
4337         spin_unlock(&root->delalloc_lock);
4338 }
4339
4340 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4341 {
4342         struct btrfs_root *root;
4343         struct list_head splice;
4344
4345         INIT_LIST_HEAD(&splice);
4346
4347         spin_lock(&fs_info->delalloc_root_lock);
4348         list_splice_init(&fs_info->delalloc_roots, &splice);
4349         while (!list_empty(&splice)) {
4350                 root = list_first_entry(&splice, struct btrfs_root,
4351                                          delalloc_root);
4352                 list_del_init(&root->delalloc_root);
4353                 root = btrfs_grab_fs_root(root);
4354                 BUG_ON(!root);
4355                 spin_unlock(&fs_info->delalloc_root_lock);
4356
4357                 btrfs_destroy_delalloc_inodes(root);
4358                 btrfs_put_fs_root(root);
4359
4360                 spin_lock(&fs_info->delalloc_root_lock);
4361         }
4362         spin_unlock(&fs_info->delalloc_root_lock);
4363 }
4364
4365 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4366                                         struct extent_io_tree *dirty_pages,
4367                                         int mark)
4368 {
4369         int ret;
4370         struct extent_buffer *eb;
4371         u64 start = 0;
4372         u64 end;
4373
4374         while (1) {
4375                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4376                                             mark, NULL);
4377                 if (ret)
4378                         break;
4379
4380                 clear_extent_bits(dirty_pages, start, end, mark);
4381                 while (start <= end) {
4382                         eb = btrfs_find_tree_block(root->fs_info, start);
4383                         start += root->nodesize;
4384                         if (!eb)
4385                                 continue;
4386                         wait_on_extent_buffer_writeback(eb);
4387
4388                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4389                                                &eb->bflags))
4390                                 clear_extent_buffer_dirty(eb);
4391                         free_extent_buffer_stale(eb);
4392                 }
4393         }
4394
4395         return ret;
4396 }
4397
4398 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4399                                        struct extent_io_tree *pinned_extents)
4400 {
4401         struct extent_io_tree *unpin;
4402         u64 start;
4403         u64 end;
4404         int ret;
4405         bool loop = true;
4406
4407         unpin = pinned_extents;
4408 again:
4409         while (1) {
4410                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4411                                             EXTENT_DIRTY, NULL);
4412                 if (ret)
4413                         break;
4414
4415                 clear_extent_dirty(unpin, start, end);
4416                 btrfs_error_unpin_extent_range(root, start, end);
4417                 cond_resched();
4418         }
4419
4420         if (loop) {
4421                 if (unpin == &root->fs_info->freed_extents[0])
4422                         unpin = &root->fs_info->freed_extents[1];
4423                 else
4424                         unpin = &root->fs_info->freed_extents[0];
4425                 loop = false;
4426                 goto again;
4427         }
4428
4429         return 0;
4430 }
4431
4432 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4433                                    struct btrfs_root *root)
4434 {
4435         btrfs_destroy_delayed_refs(cur_trans, root);
4436
4437         cur_trans->state = TRANS_STATE_COMMIT_START;
4438         wake_up(&root->fs_info->transaction_blocked_wait);
4439
4440         cur_trans->state = TRANS_STATE_UNBLOCKED;
4441         wake_up(&root->fs_info->transaction_wait);
4442
4443         btrfs_destroy_delayed_inodes(root);
4444         btrfs_assert_delayed_root_empty(root);
4445
4446         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4447                                      EXTENT_DIRTY);
4448         btrfs_destroy_pinned_extent(root,
4449                                     root->fs_info->pinned_extents);
4450
4451         cur_trans->state =TRANS_STATE_COMPLETED;
4452         wake_up(&cur_trans->commit_wait);
4453
4454         /*
4455         memset(cur_trans, 0, sizeof(*cur_trans));
4456         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4457         */
4458 }
4459
4460 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4461 {
4462         struct btrfs_transaction *t;
4463
4464         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4465
4466         spin_lock(&root->fs_info->trans_lock);
4467         while (!list_empty(&root->fs_info->trans_list)) {
4468                 t = list_first_entry(&root->fs_info->trans_list,
4469                                      struct btrfs_transaction, list);
4470                 if (t->state >= TRANS_STATE_COMMIT_START) {
4471                         atomic_inc(&t->use_count);
4472                         spin_unlock(&root->fs_info->trans_lock);
4473                         btrfs_wait_for_commit(root, t->transid);
4474                         btrfs_put_transaction(t);
4475                         spin_lock(&root->fs_info->trans_lock);
4476                         continue;
4477                 }
4478                 if (t == root->fs_info->running_transaction) {
4479                         t->state = TRANS_STATE_COMMIT_DOING;
4480                         spin_unlock(&root->fs_info->trans_lock);
4481                         /*
4482                          * We wait for 0 num_writers since we don't hold a trans
4483                          * handle open currently for this transaction.
4484                          */
4485                         wait_event(t->writer_wait,
4486                                    atomic_read(&t->num_writers) == 0);
4487                 } else {
4488                         spin_unlock(&root->fs_info->trans_lock);
4489                 }
4490                 btrfs_cleanup_one_transaction(t, root);
4491
4492                 spin_lock(&root->fs_info->trans_lock);
4493                 if (t == root->fs_info->running_transaction)
4494                         root->fs_info->running_transaction = NULL;
4495                 list_del_init(&t->list);
4496                 spin_unlock(&root->fs_info->trans_lock);
4497
4498                 btrfs_put_transaction(t);
4499                 trace_btrfs_transaction_commit(root);
4500                 spin_lock(&root->fs_info->trans_lock);
4501         }
4502         spin_unlock(&root->fs_info->trans_lock);
4503         btrfs_destroy_all_ordered_extents(root->fs_info);
4504         btrfs_destroy_delayed_inodes(root);
4505         btrfs_assert_delayed_root_empty(root);
4506         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4507         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4508         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4509
4510         return 0;
4511 }
4512
4513 static const struct extent_io_ops btree_extent_io_ops = {
4514         .readpage_end_io_hook = btree_readpage_end_io_hook,
4515         .readpage_io_failed_hook = btree_io_failed_hook,
4516         .submit_bio_hook = btree_submit_bio_hook,
4517         /* note we're sharing with inode.c for the merge bio hook */
4518         .merge_bio_hook = btrfs_merge_bio_hook,
4519 };