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_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
 105                                         NULL);
 106         if (!btrfs_end_io_wq_cache)
 107                 return -ENOMEM;
 108         return 0;
 109 }
 110
 111 void btrfs_end_io_wq_exit(void)
 112 {
 113         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_test_is_dummy_root(root))
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         root->node = NULL;
1231         root->commit_root = NULL;
1232         root->sectorsize = sectorsize;
1233         root->nodesize = nodesize;
1234         root->stripesize = stripesize;
1235         root->state = 0;
1236         root->orphan_cleanup_state = 0;
1237
1238         root->objectid = objectid;
1239         root->last_trans = 0;
1240         root->highest_objectid = 0;
1241         root->nr_delalloc_inodes = 0;
1242         root->nr_ordered_extents = 0;
1243         root->name = NULL;
1244         root->inode_tree = RB_ROOT;
1245         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1246         root->block_rsv = NULL;
1247         root->orphan_block_rsv = NULL;
1248
1249         INIT_LIST_HEAD(&root->dirty_list);
1250         INIT_LIST_HEAD(&root->root_list);
1251         INIT_LIST_HEAD(&root->delalloc_inodes);
1252         INIT_LIST_HEAD(&root->delalloc_root);
1253         INIT_LIST_HEAD(&root->ordered_extents);
1254         INIT_LIST_HEAD(&root->ordered_root);
1255         INIT_LIST_HEAD(&root->logged_list[0]);
1256         INIT_LIST_HEAD(&root->logged_list[1]);
1257         spin_lock_init(&root->orphan_lock);
1258         spin_lock_init(&root->inode_lock);
1259         spin_lock_init(&root->delalloc_lock);
1260         spin_lock_init(&root->ordered_extent_lock);
1261         spin_lock_init(&root->accounting_lock);
1262         spin_lock_init(&root->log_extents_lock[0]);
1263         spin_lock_init(&root->log_extents_lock[1]);
1264         mutex_init(&root->objectid_mutex);
1265         mutex_init(&root->log_mutex);
1266         mutex_init(&root->ordered_extent_mutex);
1267         mutex_init(&root->delalloc_mutex);
1268         init_waitqueue_head(&root->log_writer_wait);
1269         init_waitqueue_head(&root->log_commit_wait[0]);
1270         init_waitqueue_head(&root->log_commit_wait[1]);
1271         INIT_LIST_HEAD(&root->log_ctxs[0]);
1272         INIT_LIST_HEAD(&root->log_ctxs[1]);
1273         atomic_set(&root->log_commit[0], 0);
1274         atomic_set(&root->log_commit[1], 0);
1275         atomic_set(&root->log_writers, 0);
1276         atomic_set(&root->log_batch, 0);
1277         atomic_set(&root->orphan_inodes, 0);
1278         atomic_set(&root->refs, 1);
1279         atomic_set(&root->will_be_snapshoted, 0);
1280         atomic_set(&root->qgroup_meta_rsv, 0);
1281         root->log_transid = 0;
1282         root->log_transid_committed = -1;
1283         root->last_log_commit = 0;
1284         if (fs_info)
1285                 extent_io_tree_init(&root->dirty_log_pages,
1286                                      fs_info->btree_inode->i_mapping);
1287
1288         memset(&root->root_key, 0, sizeof(root->root_key));
1289         memset(&root->root_item, 0, sizeof(root->root_item));
1290         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1291         if (fs_info)
1292                 root->defrag_trans_start = fs_info->generation;
1293         else
1294                 root->defrag_trans_start = 0;
1295         root->root_key.objectid = objectid;
1296         root->anon_dev = 0;
1297
1298         spin_lock_init(&root->root_item_lock);
1299 }
1300
1301 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1302                 gfp_t flags)
1303 {
1304         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1305         if (root)
1306                 root->fs_info = fs_info;
1307         return root;
1308 }
1309
1310 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1311 /* Should only be used by the testing infrastructure */
1312 struct btrfs_root *btrfs_alloc_dummy_root(u32 sectorsize, u32 nodesize)
1313 {
1314         struct btrfs_root *root;
1315
1316         root = btrfs_alloc_root(NULL, GFP_KERNEL);
1317         if (!root)
1318                 return ERR_PTR(-ENOMEM);
1319         /* We don't use the stripesize in selftest, set it as sectorsize */
1320         __setup_root(nodesize, sectorsize, sectorsize, root, NULL,
1321                         BTRFS_ROOT_TREE_OBJECTID);
1322         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1323         root->alloc_bytenr = 0;
1324
1325         return root;
1326 }
1327 #endif
1328
1329 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1330                                      struct btrfs_fs_info *fs_info,
1331                                      u64 objectid)
1332 {
1333         struct extent_buffer *leaf;
1334         struct btrfs_root *tree_root = fs_info->tree_root;
1335         struct btrfs_root *root;
1336         struct btrfs_key key;
1337         int ret = 0;
1338         uuid_le uuid;
1339
1340         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1341         if (!root)
1342                 return ERR_PTR(-ENOMEM);
1343
1344         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1345                 tree_root->stripesize, root, fs_info, objectid);
1346         root->root_key.objectid = objectid;
1347         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1348         root->root_key.offset = 0;
1349
1350         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1351         if (IS_ERR(leaf)) {
1352                 ret = PTR_ERR(leaf);
1353                 leaf = NULL;
1354                 goto fail;
1355         }
1356
1357         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1358         btrfs_set_header_bytenr(leaf, leaf->start);
1359         btrfs_set_header_generation(leaf, trans->transid);
1360         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1361         btrfs_set_header_owner(leaf, objectid);
1362         root->node = leaf;
1363
1364         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1365                             BTRFS_FSID_SIZE);
1366         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1367                             btrfs_header_chunk_tree_uuid(leaf),
1368                             BTRFS_UUID_SIZE);
1369         btrfs_mark_buffer_dirty(leaf);
1370
1371         root->commit_root = btrfs_root_node(root);
1372         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1373
1374         root->root_item.flags = 0;
1375         root->root_item.byte_limit = 0;
1376         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1377         btrfs_set_root_generation(&root->root_item, trans->transid);
1378         btrfs_set_root_level(&root->root_item, 0);
1379         btrfs_set_root_refs(&root->root_item, 1);
1380         btrfs_set_root_used(&root->root_item, leaf->len);
1381         btrfs_set_root_last_snapshot(&root->root_item, 0);
1382         btrfs_set_root_dirid(&root->root_item, 0);
1383         uuid_le_gen(&uuid);
1384         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1385         root->root_item.drop_level = 0;
1386
1387         key.objectid = objectid;
1388         key.type = BTRFS_ROOT_ITEM_KEY;
1389         key.offset = 0;
1390         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1391         if (ret)
1392                 goto fail;
1393
1394         btrfs_tree_unlock(leaf);
1395
1396         return root;
1397
1398 fail:
1399         if (leaf) {
1400                 btrfs_tree_unlock(leaf);
1401                 free_extent_buffer(root->commit_root);
1402                 free_extent_buffer(leaf);
1403         }
1404         kfree(root);
1405
1406         return ERR_PTR(ret);
1407 }
1408
1409 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1410                                          struct btrfs_fs_info *fs_info)
1411 {
1412         struct btrfs_root *root;
1413         struct btrfs_root *tree_root = fs_info->tree_root;
1414         struct extent_buffer *leaf;
1415
1416         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1417         if (!root)
1418                 return ERR_PTR(-ENOMEM);
1419
1420         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1421                      tree_root->stripesize, root, fs_info,
1422                      BTRFS_TREE_LOG_OBJECTID);
1423
1424         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1425         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1426         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1427
1428         /*
1429          * DON'T set REF_COWS for log trees
1430          *
1431          * log trees do not get reference counted because they go away
1432          * before a real commit is actually done.  They do store pointers
1433          * to file data extents, and those reference counts still get
1434          * updated (along with back refs to the log tree).
1435          */
1436
1437         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1438                         NULL, 0, 0, 0);
1439         if (IS_ERR(leaf)) {
1440                 kfree(root);
1441                 return ERR_CAST(leaf);
1442         }
1443
1444         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1445         btrfs_set_header_bytenr(leaf, leaf->start);
1446         btrfs_set_header_generation(leaf, trans->transid);
1447         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1448         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1449         root->node = leaf;
1450
1451         write_extent_buffer(root->node, root->fs_info->fsid,
1452                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1453         btrfs_mark_buffer_dirty(root->node);
1454         btrfs_tree_unlock(root->node);
1455         return root;
1456 }
1457
1458 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1459                              struct btrfs_fs_info *fs_info)
1460 {
1461         struct btrfs_root *log_root;
1462
1463         log_root = alloc_log_tree(trans, fs_info);
1464         if (IS_ERR(log_root))
1465                 return PTR_ERR(log_root);
1466         WARN_ON(fs_info->log_root_tree);
1467         fs_info->log_root_tree = log_root;
1468         return 0;
1469 }
1470
1471 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1472                        struct btrfs_root *root)
1473 {
1474         struct btrfs_root *log_root;
1475         struct btrfs_inode_item *inode_item;
1476
1477         log_root = alloc_log_tree(trans, root->fs_info);
1478         if (IS_ERR(log_root))
1479                 return PTR_ERR(log_root);
1480
1481         log_root->last_trans = trans->transid;
1482         log_root->root_key.offset = root->root_key.objectid;
1483
1484         inode_item = &log_root->root_item.inode;
1485         btrfs_set_stack_inode_generation(inode_item, 1);
1486         btrfs_set_stack_inode_size(inode_item, 3);
1487         btrfs_set_stack_inode_nlink(inode_item, 1);
1488         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1489         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1490
1491         btrfs_set_root_node(&log_root->root_item, log_root->node);
1492
1493         WARN_ON(root->log_root);
1494         root->log_root = log_root;
1495         root->log_transid = 0;
1496         root->log_transid_committed = -1;
1497         root->last_log_commit = 0;
1498         return 0;
1499 }
1500
1501 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1502                                                struct btrfs_key *key)
1503 {
1504         struct btrfs_root *root;
1505         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1506         struct btrfs_path *path;
1507         u64 generation;
1508         int ret;
1509
1510         path = btrfs_alloc_path();
1511         if (!path)
1512                 return ERR_PTR(-ENOMEM);
1513
1514         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1515         if (!root) {
1516                 ret = -ENOMEM;
1517                 goto alloc_fail;
1518         }
1519
1520         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1521                 tree_root->stripesize, root, fs_info, key->objectid);
1522
1523         ret = btrfs_find_root(tree_root, key, path,
1524                               &root->root_item, &root->root_key);
1525         if (ret) {
1526                 if (ret > 0)
1527                         ret = -ENOENT;
1528                 goto find_fail;
1529         }
1530
1531         generation = btrfs_root_generation(&root->root_item);
1532         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1533                                      generation);
1534         if (IS_ERR(root->node)) {
1535                 ret = PTR_ERR(root->node);
1536                 goto find_fail;
1537         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1538                 ret = -EIO;
1539                 free_extent_buffer(root->node);
1540                 goto find_fail;
1541         }
1542         root->commit_root = btrfs_root_node(root);
1543 out:
1544         btrfs_free_path(path);
1545         return root;
1546
1547 find_fail:
1548         kfree(root);
1549 alloc_fail:
1550         root = ERR_PTR(ret);
1551         goto out;
1552 }
1553
1554 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1555                                       struct btrfs_key *location)
1556 {
1557         struct btrfs_root *root;
1558
1559         root = btrfs_read_tree_root(tree_root, location);
1560         if (IS_ERR(root))
1561                 return root;
1562
1563         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1564                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1565                 btrfs_check_and_init_root_item(&root->root_item);
1566         }
1567
1568         return root;
1569 }
1570
1571 int btrfs_init_fs_root(struct btrfs_root *root)
1572 {
1573         int ret;
1574         struct btrfs_subvolume_writers *writers;
1575
1576         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1577         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1578                                         GFP_NOFS);
1579         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1580                 ret = -ENOMEM;
1581                 goto fail;
1582         }
1583
1584         writers = btrfs_alloc_subvolume_writers();
1585         if (IS_ERR(writers)) {
1586                 ret = PTR_ERR(writers);
1587                 goto fail;
1588         }
1589         root->subv_writers = writers;
1590
1591         btrfs_init_free_ino_ctl(root);
1592         spin_lock_init(&root->ino_cache_lock);
1593         init_waitqueue_head(&root->ino_cache_wait);
1594
1595         ret = get_anon_bdev(&root->anon_dev);
1596         if (ret)
1597                 goto free_writers;
1598
1599         mutex_lock(&root->objectid_mutex);
1600         ret = btrfs_find_highest_objectid(root,
1601                                         &root->highest_objectid);
1602         if (ret) {
1603                 mutex_unlock(&root->objectid_mutex);
1604                 goto free_root_dev;
1605         }
1606
1607         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1608
1609         mutex_unlock(&root->objectid_mutex);
1610
1611         return 0;
1612
1613 free_root_dev:
1614         free_anon_bdev(root->anon_dev);
1615 free_writers:
1616         btrfs_free_subvolume_writers(root->subv_writers);
1617 fail:
1618         kfree(root->free_ino_ctl);
1619         kfree(root->free_ino_pinned);
1620         return ret;
1621 }
1622
1623 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1624                                                u64 root_id)
1625 {
1626         struct btrfs_root *root;
1627
1628         spin_lock(&fs_info->fs_roots_radix_lock);
1629         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1630                                  (unsigned long)root_id);
1631         spin_unlock(&fs_info->fs_roots_radix_lock);
1632         return root;
1633 }
1634
1635 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1636                          struct btrfs_root *root)
1637 {
1638         int ret;
1639
1640         ret = radix_tree_preload(GFP_NOFS);
1641         if (ret)
1642                 return ret;
1643
1644         spin_lock(&fs_info->fs_roots_radix_lock);
1645         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1646                                 (unsigned long)root->root_key.objectid,
1647                                 root);
1648         if (ret == 0)
1649                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1650         spin_unlock(&fs_info->fs_roots_radix_lock);
1651         radix_tree_preload_end();
1652
1653         return ret;
1654 }
1655
1656 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1657                                      struct btrfs_key *location,
1658                                      bool check_ref)
1659 {
1660         struct btrfs_root *root;
1661         struct btrfs_path *path;
1662         struct btrfs_key key;
1663         int ret;
1664
1665         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1666                 return fs_info->tree_root;
1667         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1668                 return fs_info->extent_root;
1669         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1670                 return fs_info->chunk_root;
1671         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1672                 return fs_info->dev_root;
1673         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1674                 return fs_info->csum_root;
1675         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1676                 return fs_info->quota_root ? fs_info->quota_root :
1677                                              ERR_PTR(-ENOENT);
1678         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1679                 return fs_info->uuid_root ? fs_info->uuid_root :
1680                                             ERR_PTR(-ENOENT);
1681         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1682                 return fs_info->free_space_root ? fs_info->free_space_root :
1683                                                   ERR_PTR(-ENOENT);
1684 again:
1685         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1686         if (root) {
1687                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1688                         return ERR_PTR(-ENOENT);
1689                 return root;
1690         }
1691
1692         root = btrfs_read_fs_root(fs_info->tree_root, location);
1693         if (IS_ERR(root))
1694                 return root;
1695
1696         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1697                 ret = -ENOENT;
1698                 goto fail;
1699         }
1700
1701         ret = btrfs_init_fs_root(root);
1702         if (ret)
1703                 goto fail;
1704
1705         path = btrfs_alloc_path();
1706         if (!path) {
1707                 ret = -ENOMEM;
1708                 goto fail;
1709         }
1710         key.objectid = BTRFS_ORPHAN_OBJECTID;
1711         key.type = BTRFS_ORPHAN_ITEM_KEY;
1712         key.offset = location->objectid;
1713
1714         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1715         btrfs_free_path(path);
1716         if (ret < 0)
1717                 goto fail;
1718         if (ret == 0)
1719                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1720
1721         ret = btrfs_insert_fs_root(fs_info, root);
1722         if (ret) {
1723                 if (ret == -EEXIST) {
1724                         free_fs_root(root);
1725                         goto again;
1726                 }
1727                 goto fail;
1728         }
1729         return root;
1730 fail:
1731         free_fs_root(root);
1732         return ERR_PTR(ret);
1733 }
1734
1735 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1736 {
1737         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1738         int ret = 0;
1739         struct btrfs_device *device;
1740         struct backing_dev_info *bdi;
1741
1742         rcu_read_lock();
1743         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1744                 if (!device->bdev)
1745                         continue;
1746                 bdi = blk_get_backing_dev_info(device->bdev);
1747                 if (bdi_congested(bdi, bdi_bits)) {
1748                         ret = 1;
1749                         break;
1750                 }
1751         }
1752         rcu_read_unlock();
1753         return ret;
1754 }
1755
1756 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1757 {
1758         int err;
1759
1760         err = bdi_setup_and_register(bdi, "btrfs");
1761         if (err)
1762                 return err;
1763
1764         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1765         bdi->congested_fn       = btrfs_congested_fn;
1766         bdi->congested_data     = info;
1767         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1768         return 0;
1769 }
1770
1771 /*
1772  * called by the kthread helper functions to finally call the bio end_io
1773  * functions.  This is where read checksum verification actually happens
1774  */
1775 static void end_workqueue_fn(struct btrfs_work *work)
1776 {
1777         struct bio *bio;
1778         struct btrfs_end_io_wq *end_io_wq;
1779
1780         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1781         bio = end_io_wq->bio;
1782
1783         bio->bi_error = end_io_wq->error;
1784         bio->bi_private = end_io_wq->private;
1785         bio->bi_end_io = end_io_wq->end_io;
1786         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1787         bio_endio(bio);
1788 }
1789
1790 static int cleaner_kthread(void *arg)
1791 {
1792         struct btrfs_root *root = arg;
1793         int again;
1794         struct btrfs_trans_handle *trans;
1795
1796         do {
1797                 again = 0;
1798
1799                 /* Make the cleaner go to sleep early. */
1800                 if (btrfs_need_cleaner_sleep(root))
1801                         goto sleep;
1802
1803                 /*
1804                  * Do not do anything if we might cause open_ctree() to block
1805                  * before we have finished mounting the filesystem.
1806                  */
1807                 if (!root->fs_info->open)
1808                         goto sleep;
1809
1810                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1811                         goto sleep;
1812
1813                 /*
1814                  * Avoid the problem that we change the status of the fs
1815                  * during the above check and trylock.
1816                  */
1817                 if (btrfs_need_cleaner_sleep(root)) {
1818                         mutex_unlock(&root->fs_info->cleaner_mutex);
1819                         goto sleep;
1820                 }
1821
1822                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1823                 btrfs_run_delayed_iputs(root);
1824                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1825
1826                 again = btrfs_clean_one_deleted_snapshot(root);
1827                 mutex_unlock(&root->fs_info->cleaner_mutex);
1828
1829                 /*
1830                  * The defragger has dealt with the R/O remount and umount,
1831                  * needn't do anything special here.
1832                  */
1833                 btrfs_run_defrag_inodes(root->fs_info);
1834
1835                 /*
1836                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1837                  * with relocation (btrfs_relocate_chunk) and relocation
1838                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1839                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1840                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1841                  * unused block groups.
1842                  */
1843                 btrfs_delete_unused_bgs(root->fs_info);
1844 sleep:
1845                 if (!again) {
1846                         set_current_state(TASK_INTERRUPTIBLE);
1847                         if (!kthread_should_stop())
1848                                 schedule();
1849                         __set_current_state(TASK_RUNNING);
1850                 }
1851         } while (!kthread_should_stop());
1852
1853         /*
1854          * Transaction kthread is stopped before us and wakes us up.
1855          * However we might have started a new transaction and COWed some
1856          * tree blocks when deleting unused block groups for example. So
1857          * make sure we commit the transaction we started to have a clean
1858          * shutdown when evicting the btree inode - if it has dirty pages
1859          * when we do the final iput() on it, eviction will trigger a
1860          * writeback for it which will fail with null pointer dereferences
1861          * since work queues and other resources were already released and
1862          * destroyed by the time the iput/eviction/writeback is made.
1863          */
1864         trans = btrfs_attach_transaction(root);
1865         if (IS_ERR(trans)) {
1866                 if (PTR_ERR(trans) != -ENOENT)
1867                         btrfs_err(root->fs_info,
1868                                   "cleaner transaction attach returned %ld",
1869                                   PTR_ERR(trans));
1870         } else {
1871                 int ret;
1872
1873                 ret = btrfs_commit_transaction(trans, root);
1874                 if (ret)
1875                         btrfs_err(root->fs_info,
1876                                   "cleaner open transaction commit returned %d",
1877                                   ret);
1878         }
1879
1880         return 0;
1881 }
1882
1883 static int transaction_kthread(void *arg)
1884 {
1885         struct btrfs_root *root = arg;
1886         struct btrfs_trans_handle *trans;
1887         struct btrfs_transaction *cur;
1888         u64 transid;
1889         unsigned long now;
1890         unsigned long delay;
1891         bool cannot_commit;
1892
1893         do {
1894                 cannot_commit = false;
1895                 delay = HZ * root->fs_info->commit_interval;
1896                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1897
1898                 spin_lock(&root->fs_info->trans_lock);
1899                 cur = root->fs_info->running_transaction;
1900                 if (!cur) {
1901                         spin_unlock(&root->fs_info->trans_lock);
1902                         goto sleep;
1903                 }
1904
1905                 now = get_seconds();
1906                 if (cur->state < TRANS_STATE_BLOCKED &&
1907                     (now < cur->start_time ||
1908                      now - cur->start_time < root->fs_info->commit_interval)) {
1909                         spin_unlock(&root->fs_info->trans_lock);
1910                         delay = HZ * 5;
1911                         goto sleep;
1912                 }
1913                 transid = cur->transid;
1914                 spin_unlock(&root->fs_info->trans_lock);
1915
1916                 /* If the file system is aborted, this will always fail. */
1917                 trans = btrfs_attach_transaction(root);
1918                 if (IS_ERR(trans)) {
1919                         if (PTR_ERR(trans) != -ENOENT)
1920                                 cannot_commit = true;
1921                         goto sleep;
1922                 }
1923                 if (transid == trans->transid) {
1924                         btrfs_commit_transaction(trans, root);
1925                 } else {
1926                         btrfs_end_transaction(trans, root);
1927                 }
1928 sleep:
1929                 wake_up_process(root->fs_info->cleaner_kthread);
1930                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1931
1932                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1933                                       &root->fs_info->fs_state)))
1934                         btrfs_cleanup_transaction(root);
1935                 set_current_state(TASK_INTERRUPTIBLE);
1936                 if (!kthread_should_stop() &&
1937                                 (!btrfs_transaction_blocked(root->fs_info) ||
1938                                  cannot_commit))
1939                         schedule_timeout(delay);
1940                 __set_current_state(TASK_RUNNING);
1941         } while (!kthread_should_stop());
1942         return 0;
1943 }
1944
1945 /*
1946  * this will find the highest generation in the array of
1947  * root backups.  The index of the highest array is returned,
1948  * or -1 if we can't find anything.
1949  *
1950  * We check to make sure the array is valid by comparing the
1951  * generation of the latest  root in the array with the generation
1952  * in the super block.  If they don't match we pitch it.
1953  */
1954 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1955 {
1956         u64 cur;
1957         int newest_index = -1;
1958         struct btrfs_root_backup *root_backup;
1959         int i;
1960
1961         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1962                 root_backup = info->super_copy->super_roots + i;
1963                 cur = btrfs_backup_tree_root_gen(root_backup);
1964                 if (cur == newest_gen)
1965                         newest_index = i;
1966         }
1967
1968         /* check to see if we actually wrapped around */
1969         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1970                 root_backup = info->super_copy->super_roots;
1971                 cur = btrfs_backup_tree_root_gen(root_backup);
1972                 if (cur == newest_gen)
1973                         newest_index = 0;
1974         }
1975         return newest_index;
1976 }
1977
1978
1979 /*
1980  * find the oldest backup so we know where to store new entries
1981  * in the backup array.  This will set the backup_root_index
1982  * field in the fs_info struct
1983  */
1984 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1985                                      u64 newest_gen)
1986 {
1987         int newest_index = -1;
1988
1989         newest_index = find_newest_super_backup(info, newest_gen);
1990         /* if there was garbage in there, just move along */
1991         if (newest_index == -1) {
1992                 info->backup_root_index = 0;
1993         } else {
1994                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1995         }
1996 }
1997
1998 /*
1999  * copy all the root pointers into the super backup array.
2000  * this will bump the backup pointer by one when it is
2001  * done
2002  */
2003 static void backup_super_roots(struct btrfs_fs_info *info)
2004 {
2005         int next_backup;
2006         struct btrfs_root_backup *root_backup;
2007         int last_backup;
2008
2009         next_backup = info->backup_root_index;
2010         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2011                 BTRFS_NUM_BACKUP_ROOTS;
2012
2013         /*
2014          * just overwrite the last backup if we're at the same generation
2015          * this happens only at umount
2016          */
2017         root_backup = info->super_for_commit->super_roots + last_backup;
2018         if (btrfs_backup_tree_root_gen(root_backup) ==
2019             btrfs_header_generation(info->tree_root->node))
2020                 next_backup = last_backup;
2021
2022         root_backup = info->super_for_commit->super_roots + next_backup;
2023
2024         /*
2025          * make sure all of our padding and empty slots get zero filled
2026          * regardless of which ones we use today
2027          */
2028         memset(root_backup, 0, sizeof(*root_backup));
2029
2030         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2031
2032         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2033         btrfs_set_backup_tree_root_gen(root_backup,
2034                                btrfs_header_generation(info->tree_root->node));
2035
2036         btrfs_set_backup_tree_root_level(root_backup,
2037                                btrfs_header_level(info->tree_root->node));
2038
2039         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2040         btrfs_set_backup_chunk_root_gen(root_backup,
2041                                btrfs_header_generation(info->chunk_root->node));
2042         btrfs_set_backup_chunk_root_level(root_backup,
2043                                btrfs_header_level(info->chunk_root->node));
2044
2045         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2046         btrfs_set_backup_extent_root_gen(root_backup,
2047                                btrfs_header_generation(info->extent_root->node));
2048         btrfs_set_backup_extent_root_level(root_backup,
2049                                btrfs_header_level(info->extent_root->node));
2050
2051         /*
2052          * we might commit during log recovery, which happens before we set
2053          * the fs_root.  Make sure it is valid before we fill it in.
2054          */
2055         if (info->fs_root && info->fs_root->node) {
2056                 btrfs_set_backup_fs_root(root_backup,
2057                                          info->fs_root->node->start);
2058                 btrfs_set_backup_fs_root_gen(root_backup,
2059                                btrfs_header_generation(info->fs_root->node));
2060                 btrfs_set_backup_fs_root_level(root_backup,
2061                                btrfs_header_level(info->fs_root->node));
2062         }
2063
2064         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2065         btrfs_set_backup_dev_root_gen(root_backup,
2066                                btrfs_header_generation(info->dev_root->node));
2067         btrfs_set_backup_dev_root_level(root_backup,
2068                                        btrfs_header_level(info->dev_root->node));
2069
2070         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2071         btrfs_set_backup_csum_root_gen(root_backup,
2072                                btrfs_header_generation(info->csum_root->node));
2073         btrfs_set_backup_csum_root_level(root_backup,
2074                                btrfs_header_level(info->csum_root->node));
2075
2076         btrfs_set_backup_total_bytes(root_backup,
2077                              btrfs_super_total_bytes(info->super_copy));
2078         btrfs_set_backup_bytes_used(root_backup,
2079                              btrfs_super_bytes_used(info->super_copy));
2080         btrfs_set_backup_num_devices(root_backup,
2081                              btrfs_super_num_devices(info->super_copy));
2082
2083         /*
2084          * if we don't copy this out to the super_copy, it won't get remembered
2085          * for the next commit
2086          */
2087         memcpy(&info->super_copy->super_roots,
2088                &info->super_for_commit->super_roots,
2089                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2090 }
2091
2092 /*
2093  * this copies info out of the root backup array and back into
2094  * the in-memory super block.  It is meant to help iterate through
2095  * the array, so you send it the number of backups you've already
2096  * tried and the last backup index you used.
2097  *
2098  * this returns -1 when it has tried all the backups
2099  */
2100 static noinline int next_root_backup(struct btrfs_fs_info *info,
2101                                      struct btrfs_super_block *super,
2102                                      int *num_backups_tried, int *backup_index)
2103 {
2104         struct btrfs_root_backup *root_backup;
2105         int newest = *backup_index;
2106
2107         if (*num_backups_tried == 0) {
2108                 u64 gen = btrfs_super_generation(super);
2109
2110                 newest = find_newest_super_backup(info, gen);
2111                 if (newest == -1)
2112                         return -1;
2113
2114                 *backup_index = newest;
2115                 *num_backups_tried = 1;
2116         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2117                 /* we've tried all the backups, all done */
2118                 return -1;
2119         } else {
2120                 /* jump to the next oldest backup */
2121                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2122                         BTRFS_NUM_BACKUP_ROOTS;
2123                 *backup_index = newest;
2124                 *num_backups_tried += 1;
2125         }
2126         root_backup = super->super_roots + newest;
2127
2128         btrfs_set_super_generation(super,
2129                                    btrfs_backup_tree_root_gen(root_backup));
2130         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2131         btrfs_set_super_root_level(super,
2132                                    btrfs_backup_tree_root_level(root_backup));
2133         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2134
2135         /*
2136          * fixme: the total bytes and num_devices need to match or we should
2137          * need a fsck
2138          */
2139         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2140         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2141         return 0;
2142 }
2143
2144 /* helper to cleanup workers */
2145 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2146 {
2147         btrfs_destroy_workqueue(fs_info->fixup_workers);
2148         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2149         btrfs_destroy_workqueue(fs_info->workers);
2150         btrfs_destroy_workqueue(fs_info->endio_workers);
2151         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2152         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2153         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2154         btrfs_destroy_workqueue(fs_info->rmw_workers);
2155         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2156         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2157         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2158         btrfs_destroy_workqueue(fs_info->submit_workers);
2159         btrfs_destroy_workqueue(fs_info->delayed_workers);
2160         btrfs_destroy_workqueue(fs_info->caching_workers);
2161         btrfs_destroy_workqueue(fs_info->readahead_workers);
2162         btrfs_destroy_workqueue(fs_info->flush_workers);
2163         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2164         btrfs_destroy_workqueue(fs_info->extent_workers);
2165 }
2166
2167 static void free_root_extent_buffers(struct btrfs_root *root)
2168 {
2169         if (root) {
2170                 free_extent_buffer(root->node);
2171                 free_extent_buffer(root->commit_root);
2172                 root->node = NULL;
2173                 root->commit_root = NULL;
2174         }
2175 }
2176
2177 /* helper to cleanup tree roots */
2178 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2179 {
2180         free_root_extent_buffers(info->tree_root);
2181
2182         free_root_extent_buffers(info->dev_root);
2183         free_root_extent_buffers(info->extent_root);
2184         free_root_extent_buffers(info->csum_root);
2185         free_root_extent_buffers(info->quota_root);
2186         free_root_extent_buffers(info->uuid_root);
2187         if (chunk_root)
2188                 free_root_extent_buffers(info->chunk_root);
2189         free_root_extent_buffers(info->free_space_root);
2190 }
2191
2192 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2193 {
2194         int ret;
2195         struct btrfs_root *gang[8];
2196         int i;
2197
2198         while (!list_empty(&fs_info->dead_roots)) {
2199                 gang[0] = list_entry(fs_info->dead_roots.next,
2200                                      struct btrfs_root, root_list);
2201                 list_del(&gang[0]->root_list);
2202
2203                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2204                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2205                 } else {
2206                         free_extent_buffer(gang[0]->node);
2207                         free_extent_buffer(gang[0]->commit_root);
2208                         btrfs_put_fs_root(gang[0]);
2209                 }
2210         }
2211
2212         while (1) {
2213                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2214                                              (void **)gang, 0,
2215                                              ARRAY_SIZE(gang));
2216                 if (!ret)
2217                         break;
2218                 for (i = 0; i < ret; i++)
2219                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2220         }
2221
2222         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2223                 btrfs_free_log_root_tree(NULL, fs_info);
2224                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2225                                             fs_info->pinned_extents);
2226         }
2227 }
2228
2229 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2230 {
2231         mutex_init(&fs_info->scrub_lock);
2232         atomic_set(&fs_info->scrubs_running, 0);
2233         atomic_set(&fs_info->scrub_pause_req, 0);
2234         atomic_set(&fs_info->scrubs_paused, 0);
2235         atomic_set(&fs_info->scrub_cancel_req, 0);
2236         init_waitqueue_head(&fs_info->scrub_pause_wait);
2237         fs_info->scrub_workers_refcnt = 0;
2238 }
2239
2240 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2241 {
2242         spin_lock_init(&fs_info->balance_lock);
2243         mutex_init(&fs_info->balance_mutex);
2244         atomic_set(&fs_info->balance_running, 0);
2245         atomic_set(&fs_info->balance_pause_req, 0);
2246         atomic_set(&fs_info->balance_cancel_req, 0);
2247         fs_info->balance_ctl = NULL;
2248         init_waitqueue_head(&fs_info->balance_wait_q);
2249 }
2250
2251 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2252                                    struct btrfs_root *tree_root)
2253 {
2254         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2255         set_nlink(fs_info->btree_inode, 1);
2256         /*
2257          * we set the i_size on the btree inode to the max possible int.
2258          * the real end of the address space is determined by all of
2259          * the devices in the system
2260          */
2261         fs_info->btree_inode->i_size = OFFSET_MAX;
2262         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2263
2264         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2265         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2266                              fs_info->btree_inode->i_mapping);
2267         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2268         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2269
2270         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2271
2272         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2273         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2274                sizeof(struct btrfs_key));
2275         set_bit(BTRFS_INODE_DUMMY,
2276                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2277         btrfs_insert_inode_hash(fs_info->btree_inode);
2278 }
2279
2280 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2281 {
2282         fs_info->dev_replace.lock_owner = 0;
2283         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2284         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2285         rwlock_init(&fs_info->dev_replace.lock);
2286         atomic_set(&fs_info->dev_replace.read_locks, 0);
2287         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2288         init_waitqueue_head(&fs_info->replace_wait);
2289         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2290 }
2291
2292 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2293 {
2294         spin_lock_init(&fs_info->qgroup_lock);
2295         mutex_init(&fs_info->qgroup_ioctl_lock);
2296         fs_info->qgroup_tree = RB_ROOT;
2297         fs_info->qgroup_op_tree = RB_ROOT;
2298         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2299         fs_info->qgroup_seq = 1;
2300         fs_info->quota_enabled = 0;
2301         fs_info->pending_quota_state = 0;
2302         fs_info->qgroup_ulist = NULL;
2303         mutex_init(&fs_info->qgroup_rescan_lock);
2304 }
2305
2306 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2307                 struct btrfs_fs_devices *fs_devices)
2308 {
2309         int max_active = fs_info->thread_pool_size;
2310         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2311
2312         fs_info->workers =
2313                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2314                                       max_active, 16);
2315
2316         fs_info->delalloc_workers =
2317                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2318
2319         fs_info->flush_workers =
2320                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2321
2322         fs_info->caching_workers =
2323                 btrfs_alloc_workqueue("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("submit", flags,
2332                                       min_t(u64, fs_devices->num_devices,
2333                                             max_active), 64);
2334
2335         fs_info->fixup_workers =
2336                 btrfs_alloc_workqueue("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("endio", flags, max_active, 4);
2344         fs_info->endio_meta_workers =
2345                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2346         fs_info->endio_meta_write_workers =
2347                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2348         fs_info->endio_raid56_workers =
2349                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2350         fs_info->endio_repair_workers =
2351                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2352         fs_info->rmw_workers =
2353                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2354         fs_info->endio_write_workers =
2355                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2356         fs_info->endio_freespace_worker =
2357                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2358         fs_info->delayed_workers =
2359                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2360         fs_info->readahead_workers =
2361                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2362         fs_info->qgroup_rescan_workers =
2363                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2364         fs_info->extent_workers =
2365                 btrfs_alloc_workqueue("extent-refs", flags,
2366                                       min_t(u64, fs_devices->num_devices,
2367                                             max_active), 8);
2368
2369         if (!(fs_info->workers && fs_info->delalloc_workers &&
2370               fs_info->submit_workers && fs_info->flush_workers &&
2371               fs_info->endio_workers && fs_info->endio_meta_workers &&
2372               fs_info->endio_meta_write_workers &&
2373               fs_info->endio_repair_workers &&
2374               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2375               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2376               fs_info->caching_workers && fs_info->readahead_workers &&
2377               fs_info->fixup_workers && fs_info->delayed_workers &&
2378               fs_info->extent_workers &&
2379               fs_info->qgroup_rescan_workers)) {
2380                 return -ENOMEM;
2381         }
2382
2383         return 0;
2384 }
2385
2386 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2387                             struct btrfs_fs_devices *fs_devices)
2388 {
2389         int ret;
2390         struct btrfs_root *tree_root = fs_info->tree_root;
2391         struct btrfs_root *log_tree_root;
2392         struct btrfs_super_block *disk_super = fs_info->super_copy;
2393         u64 bytenr = btrfs_super_log_root(disk_super);
2394
2395         if (fs_devices->rw_devices == 0) {
2396                 btrfs_warn(fs_info, "log replay required on RO media");
2397                 return -EIO;
2398         }
2399
2400         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2401         if (!log_tree_root)
2402                 return -ENOMEM;
2403
2404         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2405                         tree_root->stripesize, log_tree_root, fs_info,
2406                         BTRFS_TREE_LOG_OBJECTID);
2407
2408         log_tree_root->node = read_tree_block(tree_root, bytenr,
2409                         fs_info->generation + 1);
2410         if (IS_ERR(log_tree_root->node)) {
2411                 btrfs_warn(fs_info, "failed to read log tree");
2412                 ret = PTR_ERR(log_tree_root->node);
2413                 kfree(log_tree_root);
2414                 return ret;
2415         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2416                 btrfs_err(fs_info, "failed to read log tree");
2417                 free_extent_buffer(log_tree_root->node);
2418                 kfree(log_tree_root);
2419                 return -EIO;
2420         }
2421         /* returns with log_tree_root freed on success */
2422         ret = btrfs_recover_log_trees(log_tree_root);
2423         if (ret) {
2424                 btrfs_handle_fs_error(tree_root->fs_info, ret,
2425                             "Failed to recover log tree");
2426                 free_extent_buffer(log_tree_root->node);
2427                 kfree(log_tree_root);
2428                 return ret;
2429         }
2430
2431         if (fs_info->sb->s_flags & MS_RDONLY) {
2432                 ret = btrfs_commit_super(tree_root);
2433                 if (ret)
2434                         return ret;
2435         }
2436
2437         return 0;
2438 }
2439
2440 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2441                             struct btrfs_root *tree_root)
2442 {
2443         struct btrfs_root *root;
2444         struct btrfs_key location;
2445         int ret;
2446
2447         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2448         location.type = BTRFS_ROOT_ITEM_KEY;
2449         location.offset = 0;
2450
2451         root = btrfs_read_tree_root(tree_root, &location);
2452         if (IS_ERR(root))
2453                 return PTR_ERR(root);
2454         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2455         fs_info->extent_root = root;
2456
2457         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2458         root = btrfs_read_tree_root(tree_root, &location);
2459         if (IS_ERR(root))
2460                 return PTR_ERR(root);
2461         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2462         fs_info->dev_root = root;
2463         btrfs_init_devices_late(fs_info);
2464
2465         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2466         root = btrfs_read_tree_root(tree_root, &location);
2467         if (IS_ERR(root))
2468                 return PTR_ERR(root);
2469         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2470         fs_info->csum_root = root;
2471
2472         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2473         root = btrfs_read_tree_root(tree_root, &location);
2474         if (!IS_ERR(root)) {
2475                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2476                 fs_info->quota_enabled = 1;
2477                 fs_info->pending_quota_state = 1;
2478                 fs_info->quota_root = root;
2479         }
2480
2481         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2482         root = btrfs_read_tree_root(tree_root, &location);
2483         if (IS_ERR(root)) {
2484                 ret = PTR_ERR(root);
2485                 if (ret != -ENOENT)
2486                         return ret;
2487         } else {
2488                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2489                 fs_info->uuid_root = root;
2490         }
2491
2492         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2493                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2494                 root = btrfs_read_tree_root(tree_root, &location);
2495                 if (IS_ERR(root))
2496                         return PTR_ERR(root);
2497                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2498                 fs_info->free_space_root = root;
2499         }
2500
2501         return 0;
2502 }
2503
2504 int open_ctree(struct super_block *sb,
2505                struct btrfs_fs_devices *fs_devices,
2506                char *options)
2507 {
2508         u32 sectorsize;
2509         u32 nodesize;
2510         u32 stripesize;
2511         u64 generation;
2512         u64 features;
2513         struct btrfs_key location;
2514         struct buffer_head *bh;
2515         struct btrfs_super_block *disk_super;
2516         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2517         struct btrfs_root *tree_root;
2518         struct btrfs_root *chunk_root;
2519         int ret;
2520         int err = -EINVAL;
2521         int num_backups_tried = 0;
2522         int backup_index = 0;
2523         int max_active;
2524
2525         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2526         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2527         if (!tree_root || !chunk_root) {
2528                 err = -ENOMEM;
2529                 goto fail;
2530         }
2531
2532         ret = init_srcu_struct(&fs_info->subvol_srcu);
2533         if (ret) {
2534                 err = ret;
2535                 goto fail;
2536         }
2537
2538         ret = setup_bdi(fs_info, &fs_info->bdi);
2539         if (ret) {
2540                 err = ret;
2541                 goto fail_srcu;
2542         }
2543
2544         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2545         if (ret) {
2546                 err = ret;
2547                 goto fail_bdi;
2548         }
2549         fs_info->dirty_metadata_batch = PAGE_SIZE *
2550                                         (1 + ilog2(nr_cpu_ids));
2551
2552         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2553         if (ret) {
2554                 err = ret;
2555                 goto fail_dirty_metadata_bytes;
2556         }
2557
2558         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2559         if (ret) {
2560                 err = ret;
2561                 goto fail_delalloc_bytes;
2562         }
2563
2564         fs_info->btree_inode = new_inode(sb);
2565         if (!fs_info->btree_inode) {
2566                 err = -ENOMEM;
2567                 goto fail_bio_counter;
2568         }
2569
2570         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2571
2572         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2573         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2574         INIT_LIST_HEAD(&fs_info->trans_list);
2575         INIT_LIST_HEAD(&fs_info->dead_roots);
2576         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2577         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2578         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2579         spin_lock_init(&fs_info->delalloc_root_lock);
2580         spin_lock_init(&fs_info->trans_lock);
2581         spin_lock_init(&fs_info->fs_roots_radix_lock);
2582         spin_lock_init(&fs_info->delayed_iput_lock);
2583         spin_lock_init(&fs_info->defrag_inodes_lock);
2584         spin_lock_init(&fs_info->free_chunk_lock);
2585         spin_lock_init(&fs_info->tree_mod_seq_lock);
2586         spin_lock_init(&fs_info->super_lock);
2587         spin_lock_init(&fs_info->qgroup_op_lock);
2588         spin_lock_init(&fs_info->buffer_lock);
2589         spin_lock_init(&fs_info->unused_bgs_lock);
2590         rwlock_init(&fs_info->tree_mod_log_lock);
2591         mutex_init(&fs_info->unused_bg_unpin_mutex);
2592         mutex_init(&fs_info->delete_unused_bgs_mutex);
2593         mutex_init(&fs_info->reloc_mutex);
2594         mutex_init(&fs_info->delalloc_root_mutex);
2595         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2596         seqlock_init(&fs_info->profiles_lock);
2597
2598         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2599         INIT_LIST_HEAD(&fs_info->space_info);
2600         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2601         INIT_LIST_HEAD(&fs_info->unused_bgs);
2602         btrfs_mapping_init(&fs_info->mapping_tree);
2603         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2604                              BTRFS_BLOCK_RSV_GLOBAL);
2605         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2606                              BTRFS_BLOCK_RSV_DELALLOC);
2607         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2608         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2609         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2610         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2611                              BTRFS_BLOCK_RSV_DELOPS);
2612         atomic_set(&fs_info->nr_async_submits, 0);
2613         atomic_set(&fs_info->async_delalloc_pages, 0);
2614         atomic_set(&fs_info->async_submit_draining, 0);
2615         atomic_set(&fs_info->nr_async_bios, 0);
2616         atomic_set(&fs_info->defrag_running, 0);
2617         atomic_set(&fs_info->qgroup_op_seq, 0);
2618         atomic_set(&fs_info->reada_works_cnt, 0);
2619         atomic64_set(&fs_info->tree_mod_seq, 0);
2620         fs_info->sb = sb;
2621         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2622         fs_info->metadata_ratio = 0;
2623         fs_info->defrag_inodes = RB_ROOT;
2624         fs_info->free_chunk_space = 0;
2625         fs_info->tree_mod_log = RB_ROOT;
2626         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2627         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2628         /* readahead state */
2629         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2630         spin_lock_init(&fs_info->reada_lock);
2631
2632         fs_info->thread_pool_size = min_t(unsigned long,
2633                                           num_online_cpus() + 2, 8);
2634
2635         INIT_LIST_HEAD(&fs_info->ordered_roots);
2636         spin_lock_init(&fs_info->ordered_root_lock);
2637         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2638                                         GFP_KERNEL);
2639         if (!fs_info->delayed_root) {
2640                 err = -ENOMEM;
2641                 goto fail_iput;
2642         }
2643         btrfs_init_delayed_root(fs_info->delayed_root);
2644
2645         btrfs_init_scrub(fs_info);
2646 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2647         fs_info->check_integrity_print_mask = 0;
2648 #endif
2649         btrfs_init_balance(fs_info);
2650         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2651
2652         sb->s_blocksize = 4096;
2653         sb->s_blocksize_bits = blksize_bits(4096);
2654         sb->s_bdi = &fs_info->bdi;
2655
2656         btrfs_init_btree_inode(fs_info, tree_root);
2657
2658         spin_lock_init(&fs_info->block_group_cache_lock);
2659         fs_info->block_group_cache_tree = RB_ROOT;
2660         fs_info->first_logical_byte = (u64)-1;
2661
2662         extent_io_tree_init(&fs_info->freed_extents[0],
2663                              fs_info->btree_inode->i_mapping);
2664         extent_io_tree_init(&fs_info->freed_extents[1],
2665                              fs_info->btree_inode->i_mapping);
2666         fs_info->pinned_extents = &fs_info->freed_extents[0];
2667         fs_info->do_barriers = 1;
2668
2669
2670         mutex_init(&fs_info->ordered_operations_mutex);
2671         mutex_init(&fs_info->tree_log_mutex);
2672         mutex_init(&fs_info->chunk_mutex);
2673         mutex_init(&fs_info->transaction_kthread_mutex);
2674         mutex_init(&fs_info->cleaner_mutex);
2675         mutex_init(&fs_info->volume_mutex);
2676         mutex_init(&fs_info->ro_block_group_mutex);
2677         init_rwsem(&fs_info->commit_root_sem);
2678         init_rwsem(&fs_info->cleanup_work_sem);
2679         init_rwsem(&fs_info->subvol_sem);
2680         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2681
2682         btrfs_init_dev_replace_locks(fs_info);
2683         btrfs_init_qgroup(fs_info);
2684
2685         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2686         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2687
2688         init_waitqueue_head(&fs_info->transaction_throttle);
2689         init_waitqueue_head(&fs_info->transaction_wait);
2690         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2691         init_waitqueue_head(&fs_info->async_submit_wait);
2692
2693         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2694
2695         ret = btrfs_alloc_stripe_hash_table(fs_info);
2696         if (ret) {
2697                 err = ret;
2698                 goto fail_alloc;
2699         }
2700
2701         __setup_root(4096, 4096, 4096, tree_root,
2702                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2703
2704         invalidate_bdev(fs_devices->latest_bdev);
2705
2706         /*
2707          * Read super block and check the signature bytes only
2708          */
2709         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2710         if (IS_ERR(bh)) {
2711                 err = PTR_ERR(bh);
2712                 goto fail_alloc;
2713         }
2714
2715         /*
2716          * We want to check superblock checksum, the type is stored inside.
2717          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2718          */
2719         if (btrfs_check_super_csum(bh->b_data)) {
2720                 btrfs_err(fs_info, "superblock checksum mismatch");
2721                 err = -EINVAL;
2722                 brelse(bh);
2723                 goto fail_alloc;
2724         }
2725
2726         /*
2727          * super_copy is zeroed at allocation time and we never touch the
2728          * following bytes up to INFO_SIZE, the checksum is calculated from
2729          * the whole block of INFO_SIZE
2730          */
2731         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2732         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2733                sizeof(*fs_info->super_for_commit));
2734         brelse(bh);
2735
2736         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2737
2738         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2739         if (ret) {
2740                 btrfs_err(fs_info, "superblock contains fatal errors");
2741                 err = -EINVAL;
2742                 goto fail_alloc;
2743         }
2744
2745         disk_super = fs_info->super_copy;
2746         if (!btrfs_super_root(disk_super))
2747                 goto fail_alloc;
2748
2749         /* check FS state, whether FS is broken. */
2750         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2751                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2752
2753         /*
2754          * run through our array of backup supers and setup
2755          * our ring pointer to the oldest one
2756          */
2757         generation = btrfs_super_generation(disk_super);
2758         find_oldest_super_backup(fs_info, generation);
2759
2760         /*
2761          * In the long term, we'll store the compression type in the super
2762          * block, and it'll be used for per file compression control.
2763          */
2764         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2765
2766         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2767         if (ret) {
2768                 err = ret;
2769                 goto fail_alloc;
2770         }
2771
2772         features = btrfs_super_incompat_flags(disk_super) &
2773                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2774         if (features) {
2775                 btrfs_err(fs_info,
2776                     "cannot mount because of unsupported optional features (%llx)",
2777                     features);
2778                 err = -EINVAL;
2779                 goto fail_alloc;
2780         }
2781
2782         features = btrfs_super_incompat_flags(disk_super);
2783         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2784         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2785                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2786
2787         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2788                 btrfs_info(fs_info, "has skinny extents");
2789
2790         /*
2791          * flag our filesystem as having big metadata blocks if
2792          * they are bigger than the page size
2793          */
2794         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2795                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2796                         btrfs_info(fs_info,
2797                                 "flagging fs with big metadata feature");
2798                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2799         }
2800
2801         nodesize = btrfs_super_nodesize(disk_super);
2802         sectorsize = btrfs_super_sectorsize(disk_super);
2803         stripesize = sectorsize;
2804         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2805         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2806
2807         /*
2808          * mixed block groups end up with duplicate but slightly offset
2809          * extent buffers for the same range.  It leads to corruptions
2810          */
2811         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2812             (sectorsize != nodesize)) {
2813                 btrfs_err(fs_info,
2814 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2815                         nodesize, sectorsize);
2816                 goto fail_alloc;
2817         }
2818
2819         /*
2820          * Needn't use the lock because there is no other task which will
2821          * update the flag.
2822          */
2823         btrfs_set_super_incompat_flags(disk_super, features);
2824
2825         features = btrfs_super_compat_ro_flags(disk_super) &
2826                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2827         if (!(sb->s_flags & MS_RDONLY) && features) {
2828                 btrfs_err(fs_info,
2829         "cannot mount read-write because of unsupported optional features (%llx)",
2830                        features);
2831                 err = -EINVAL;
2832                 goto fail_alloc;
2833         }
2834
2835         max_active = fs_info->thread_pool_size;
2836
2837         ret = btrfs_init_workqueues(fs_info, fs_devices);
2838         if (ret) {
2839                 err = ret;
2840                 goto fail_sb_buffer;
2841         }
2842
2843         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2844         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2845                                     SZ_4M / PAGE_SIZE);
2846
2847         tree_root->nodesize = nodesize;
2848         tree_root->sectorsize = sectorsize;
2849         tree_root->stripesize = stripesize;
2850
2851         sb->s_blocksize = sectorsize;
2852         sb->s_blocksize_bits = blksize_bits(sectorsize);
2853
2854         mutex_lock(&fs_info->chunk_mutex);
2855         ret = btrfs_read_sys_array(tree_root);
2856         mutex_unlock(&fs_info->chunk_mutex);
2857         if (ret) {
2858                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2859                 goto fail_sb_buffer;
2860         }
2861
2862         generation = btrfs_super_chunk_root_generation(disk_super);
2863
2864         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2865                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2866
2867         chunk_root->node = read_tree_block(chunk_root,
2868                                            btrfs_super_chunk_root(disk_super),
2869                                            generation);
2870         if (IS_ERR(chunk_root->node) ||
2871             !extent_buffer_uptodate(chunk_root->node)) {
2872                 btrfs_err(fs_info, "failed to read chunk root");
2873                 if (!IS_ERR(chunk_root->node))
2874                         free_extent_buffer(chunk_root->node);
2875                 chunk_root->node = NULL;
2876                 goto fail_tree_roots;
2877         }
2878         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2879         chunk_root->commit_root = btrfs_root_node(chunk_root);
2880
2881         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2882            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2883
2884         ret = btrfs_read_chunk_tree(chunk_root);
2885         if (ret) {
2886                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2887                 goto fail_tree_roots;
2888         }
2889
2890         /*
2891          * keep the device that is marked to be the target device for the
2892          * dev_replace procedure
2893          */
2894         btrfs_close_extra_devices(fs_devices, 0);
2895
2896         if (!fs_devices->latest_bdev) {
2897                 btrfs_err(fs_info, "failed to read devices");
2898                 goto fail_tree_roots;
2899         }
2900
2901 retry_root_backup:
2902         generation = btrfs_super_generation(disk_super);
2903
2904         tree_root->node = read_tree_block(tree_root,
2905                                           btrfs_super_root(disk_super),
2906                                           generation);
2907         if (IS_ERR(tree_root->node) ||
2908             !extent_buffer_uptodate(tree_root->node)) {
2909                 btrfs_warn(fs_info, "failed to read tree root");
2910                 if (!IS_ERR(tree_root->node))
2911                         free_extent_buffer(tree_root->node);
2912                 tree_root->node = NULL;
2913                 goto recovery_tree_root;
2914         }
2915
2916         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2917         tree_root->commit_root = btrfs_root_node(tree_root);
2918         btrfs_set_root_refs(&tree_root->root_item, 1);
2919
2920         mutex_lock(&tree_root->objectid_mutex);
2921         ret = btrfs_find_highest_objectid(tree_root,
2922                                         &tree_root->highest_objectid);
2923         if (ret) {
2924                 mutex_unlock(&tree_root->objectid_mutex);
2925                 goto recovery_tree_root;
2926         }
2927
2928         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2929
2930         mutex_unlock(&tree_root->objectid_mutex);
2931
2932         ret = btrfs_read_roots(fs_info, tree_root);
2933         if (ret)
2934                 goto recovery_tree_root;
2935
2936         fs_info->generation = generation;
2937         fs_info->last_trans_committed = generation;
2938
2939         ret = btrfs_recover_balance(fs_info);
2940         if (ret) {
2941                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2942                 goto fail_block_groups;
2943         }
2944
2945         ret = btrfs_init_dev_stats(fs_info);
2946         if (ret) {
2947                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2948                 goto fail_block_groups;
2949         }
2950
2951         ret = btrfs_init_dev_replace(fs_info);
2952         if (ret) {
2953                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2954                 goto fail_block_groups;
2955         }
2956
2957         btrfs_close_extra_devices(fs_devices, 1);
2958
2959         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2960         if (ret) {
2961                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2962                                 ret);
2963                 goto fail_block_groups;
2964         }
2965
2966         ret = btrfs_sysfs_add_device(fs_devices);
2967         if (ret) {
2968                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2969                                 ret);
2970                 goto fail_fsdev_sysfs;
2971         }
2972
2973         ret = btrfs_sysfs_add_mounted(fs_info);
2974         if (ret) {
2975                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2976                 goto fail_fsdev_sysfs;
2977         }
2978
2979         ret = btrfs_init_space_info(fs_info);
2980         if (ret) {
2981                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2982                 goto fail_sysfs;
2983         }
2984
2985         ret = btrfs_read_block_groups(fs_info->extent_root);
2986         if (ret) {
2987                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2988                 goto fail_sysfs;
2989         }
2990         fs_info->num_tolerated_disk_barrier_failures =
2991                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2992         if (fs_info->fs_devices->missing_devices >
2993              fs_info->num_tolerated_disk_barrier_failures &&
2994             !(sb->s_flags & MS_RDONLY)) {
2995                 btrfs_warn(fs_info,
2996 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
2997                         fs_info->fs_devices->missing_devices,
2998                         fs_info->num_tolerated_disk_barrier_failures);
2999                 goto fail_sysfs;
3000         }
3001
3002         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3003                                                "btrfs-cleaner");
3004         if (IS_ERR(fs_info->cleaner_kthread))
3005                 goto fail_sysfs;
3006
3007         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3008                                                    tree_root,
3009                                                    "btrfs-transaction");
3010         if (IS_ERR(fs_info->transaction_kthread))
3011                 goto fail_cleaner;
3012
3013         if (!btrfs_test_opt(tree_root, SSD) &&
3014             !btrfs_test_opt(tree_root, NOSSD) &&
3015             !fs_info->fs_devices->rotating) {
3016                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3017                 btrfs_set_opt(fs_info->mount_opt, SSD);
3018         }
3019
3020         /*
3021          * Mount does not set all options immediately, we can do it now and do
3022          * not have to wait for transaction commit
3023          */
3024         btrfs_apply_pending_changes(fs_info);
3025
3026 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3027         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3028                 ret = btrfsic_mount(tree_root, fs_devices,
3029                                     btrfs_test_opt(tree_root,
3030                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3031                                     1 : 0,
3032                                     fs_info->check_integrity_print_mask);
3033                 if (ret)
3034                         btrfs_warn(fs_info,
3035                                 "failed to initialize integrity check module: %d",
3036                                 ret);
3037         }
3038 #endif
3039         ret = btrfs_read_qgroup_config(fs_info);
3040         if (ret)
3041                 goto fail_trans_kthread;
3042
3043         /* do not make disk changes in broken FS or nologreplay is given */
3044         if (btrfs_super_log_root(disk_super) != 0 &&
3045             !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3046                 ret = btrfs_replay_log(fs_info, fs_devices);
3047                 if (ret) {
3048                         err = ret;
3049                         goto fail_qgroup;
3050                 }
3051         }
3052
3053         ret = btrfs_find_orphan_roots(tree_root);
3054         if (ret)
3055                 goto fail_qgroup;
3056
3057         if (!(sb->s_flags & MS_RDONLY)) {
3058                 ret = btrfs_cleanup_fs_roots(fs_info);
3059                 if (ret)
3060                         goto fail_qgroup;
3061
3062                 mutex_lock(&fs_info->cleaner_mutex);
3063                 ret = btrfs_recover_relocation(tree_root);
3064                 mutex_unlock(&fs_info->cleaner_mutex);
3065                 if (ret < 0) {
3066                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3067                                         ret);
3068                         err = -EINVAL;
3069                         goto fail_qgroup;
3070                 }
3071         }
3072
3073         location.objectid = BTRFS_FS_TREE_OBJECTID;
3074         location.type = BTRFS_ROOT_ITEM_KEY;
3075         location.offset = 0;
3076
3077         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3078         if (IS_ERR(fs_info->fs_root)) {
3079                 err = PTR_ERR(fs_info->fs_root);
3080                 goto fail_qgroup;
3081         }
3082
3083         if (sb->s_flags & MS_RDONLY)
3084                 return 0;
3085
3086         if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3087             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3088                 btrfs_info(fs_info, "creating free space tree");
3089                 ret = btrfs_create_free_space_tree(fs_info);
3090                 if (ret) {
3091                         btrfs_warn(fs_info,
3092                                 "failed to create free space tree: %d", ret);
3093                         close_ctree(tree_root);
3094                         return ret;
3095                 }
3096         }
3097
3098         down_read(&fs_info->cleanup_work_sem);
3099         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3100             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3101                 up_read(&fs_info->cleanup_work_sem);
3102                 close_ctree(tree_root);
3103                 return ret;
3104         }
3105         up_read(&fs_info->cleanup_work_sem);
3106
3107         ret = btrfs_resume_balance_async(fs_info);
3108         if (ret) {
3109                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3110                 close_ctree(tree_root);
3111                 return ret;
3112         }
3113
3114         ret = btrfs_resume_dev_replace_async(fs_info);
3115         if (ret) {
3116                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3117                 close_ctree(tree_root);
3118                 return ret;
3119         }
3120
3121         btrfs_qgroup_rescan_resume(fs_info);
3122
3123         if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3124             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3125                 btrfs_info(fs_info, "clearing free space tree");
3126                 ret = btrfs_clear_free_space_tree(fs_info);
3127                 if (ret) {
3128                         btrfs_warn(fs_info,
3129                                 "failed to clear free space tree: %d", ret);
3130                         close_ctree(tree_root);
3131                         return ret;
3132                 }
3133         }
3134
3135         if (!fs_info->uuid_root) {
3136                 btrfs_info(fs_info, "creating UUID tree");
3137                 ret = btrfs_create_uuid_tree(fs_info);
3138                 if (ret) {
3139                         btrfs_warn(fs_info,
3140                                 "failed to create the UUID tree: %d", ret);
3141                         close_ctree(tree_root);
3142                         return ret;
3143                 }
3144         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3145                    fs_info->generation !=
3146                                 btrfs_super_uuid_tree_generation(disk_super)) {
3147                 btrfs_info(fs_info, "checking UUID tree");
3148                 ret = btrfs_check_uuid_tree(fs_info);
3149                 if (ret) {
3150                         btrfs_warn(fs_info,
3151                                 "failed to check the UUID tree: %d", ret);
3152                         close_ctree(tree_root);
3153                         return ret;
3154                 }
3155         } else {
3156                 fs_info->update_uuid_tree_gen = 1;
3157         }
3158
3159         fs_info->open = 1;
3160
3161         /*
3162          * backuproot only affect mount behavior, and if open_ctree succeeded,
3163          * no need to keep the flag
3164          */
3165         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3166
3167         return 0;
3168
3169 fail_qgroup:
3170         btrfs_free_qgroup_config(fs_info);
3171 fail_trans_kthread:
3172         kthread_stop(fs_info->transaction_kthread);
3173         btrfs_cleanup_transaction(fs_info->tree_root);
3174         btrfs_free_fs_roots(fs_info);
3175 fail_cleaner:
3176         kthread_stop(fs_info->cleaner_kthread);
3177
3178         /*
3179          * make sure we're done with the btree inode before we stop our
3180          * kthreads
3181          */
3182         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3183
3184 fail_sysfs:
3185         btrfs_sysfs_remove_mounted(fs_info);
3186
3187 fail_fsdev_sysfs:
3188         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3189
3190 fail_block_groups:
3191         btrfs_put_block_group_cache(fs_info);
3192         btrfs_free_block_groups(fs_info);
3193
3194 fail_tree_roots:
3195         free_root_pointers(fs_info, 1);
3196         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3197
3198 fail_sb_buffer:
3199         btrfs_stop_all_workers(fs_info);
3200 fail_alloc:
3201 fail_iput:
3202         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3203
3204         iput(fs_info->btree_inode);
3205 fail_bio_counter:
3206         percpu_counter_destroy(&fs_info->bio_counter);
3207 fail_delalloc_bytes:
3208         percpu_counter_destroy(&fs_info->delalloc_bytes);
3209 fail_dirty_metadata_bytes:
3210         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3211 fail_bdi:
3212         bdi_destroy(&fs_info->bdi);
3213 fail_srcu:
3214         cleanup_srcu_struct(&fs_info->subvol_srcu);
3215 fail:
3216         btrfs_free_stripe_hash_table(fs_info);
3217         btrfs_close_devices(fs_info->fs_devices);
3218         return err;
3219
3220 recovery_tree_root:
3221         if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3222                 goto fail_tree_roots;
3223
3224         free_root_pointers(fs_info, 0);
3225
3226         /* don't use the log in recovery mode, it won't be valid */
3227         btrfs_set_super_log_root(disk_super, 0);
3228
3229         /* we can't trust the free space cache either */
3230         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3231
3232         ret = next_root_backup(fs_info, fs_info->super_copy,
3233                                &num_backups_tried, &backup_index);
3234         if (ret == -1)
3235                 goto fail_block_groups;
3236         goto retry_root_backup;
3237 }
3238
3239 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3240 {
3241         if (uptodate) {
3242                 set_buffer_uptodate(bh);
3243         } else {
3244                 struct btrfs_device *device = (struct btrfs_device *)
3245                         bh->b_private;
3246
3247                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3248                                 "lost page write due to IO error on %s",
3249                                           rcu_str_deref(device->name));
3250                 /* note, we don't set_buffer_write_io_error because we have
3251                  * our own ways of dealing with the IO errors
3252                  */
3253                 clear_buffer_uptodate(bh);
3254                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3255         }
3256         unlock_buffer(bh);
3257         put_bh(bh);
3258 }
3259
3260 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3261                         struct buffer_head **bh_ret)
3262 {
3263         struct buffer_head *bh;
3264         struct btrfs_super_block *super;
3265         u64 bytenr;
3266
3267         bytenr = btrfs_sb_offset(copy_num);
3268         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3269                 return -EINVAL;
3270
3271         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3272         /*
3273          * If we fail to read from the underlying devices, as of now
3274          * the best option we have is to mark it EIO.
3275          */
3276         if (!bh)
3277                 return -EIO;
3278
3279         super = (struct btrfs_super_block *)bh->b_data;
3280         if (btrfs_super_bytenr(super) != bytenr ||
3281                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3282                 brelse(bh);
3283                 return -EINVAL;
3284         }
3285
3286         *bh_ret = bh;
3287         return 0;
3288 }
3289
3290
3291 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3292 {
3293         struct buffer_head *bh;
3294         struct buffer_head *latest = NULL;
3295         struct btrfs_super_block *super;
3296         int i;
3297         u64 transid = 0;
3298         int ret = -EINVAL;
3299
3300         /* we would like to check all the supers, but that would make
3301          * a btrfs mount succeed after a mkfs from a different FS.
3302          * So, we need to add a special mount option to scan for
3303          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3304          */
3305         for (i = 0; i < 1; i++) {
3306                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3307                 if (ret)
3308                         continue;
3309
3310                 super = (struct btrfs_super_block *)bh->b_data;
3311
3312                 if (!latest || btrfs_super_generation(super) > transid) {
3313                         brelse(latest);
3314                         latest = bh;
3315                         transid = btrfs_super_generation(super);
3316                 } else {
3317                         brelse(bh);
3318                 }
3319         }
3320
3321         if (!latest)
3322                 return ERR_PTR(ret);
3323
3324         return latest;
3325 }
3326
3327 /*
3328  * this should be called twice, once with wait == 0 and
3329  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3330  * we write are pinned.
3331  *
3332  * They are released when wait == 1 is done.
3333  * max_mirrors must be the same for both runs, and it indicates how
3334  * many supers on this one device should be written.
3335  *
3336  * max_mirrors == 0 means to write them all.
3337  */
3338 static int write_dev_supers(struct btrfs_device *device,
3339                             struct btrfs_super_block *sb,
3340                             int do_barriers, int wait, int max_mirrors)
3341 {
3342         struct buffer_head *bh;
3343         int i;
3344         int ret;
3345         int errors = 0;
3346         u32 crc;
3347         u64 bytenr;
3348
3349         if (max_mirrors == 0)
3350                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3351
3352         for (i = 0; i < max_mirrors; i++) {
3353                 bytenr = btrfs_sb_offset(i);
3354                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3355                     device->commit_total_bytes)
3356                         break;
3357
3358                 if (wait) {
3359                         bh = __find_get_block(device->bdev, bytenr / 4096,
3360                                               BTRFS_SUPER_INFO_SIZE);
3361                         if (!bh) {
3362                                 errors++;
3363                                 continue;
3364                         }
3365                         wait_on_buffer(bh);
3366                         if (!buffer_uptodate(bh))
3367                                 errors++;
3368
3369                         /* drop our reference */
3370                         brelse(bh);
3371
3372                         /* drop the reference from the wait == 0 run */
3373                         brelse(bh);
3374                         continue;
3375                 } else {
3376                         btrfs_set_super_bytenr(sb, bytenr);
3377
3378                         crc = ~(u32)0;
3379                         crc = btrfs_csum_data((char *)sb +
3380                                               BTRFS_CSUM_SIZE, crc,
3381                                               BTRFS_SUPER_INFO_SIZE -
3382                                               BTRFS_CSUM_SIZE);
3383                         btrfs_csum_final(crc, sb->csum);
3384
3385                         /*
3386                          * one reference for us, and we leave it for the
3387                          * caller
3388                          */
3389                         bh = __getblk(device->bdev, bytenr / 4096,
3390                                       BTRFS_SUPER_INFO_SIZE);
3391                         if (!bh) {
3392                                 btrfs_err(device->dev_root->fs_info,
3393                                     "couldn't get super buffer head for bytenr %llu",
3394                                     bytenr);
3395                                 errors++;
3396                                 continue;
3397                         }
3398
3399                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3400
3401                         /* one reference for submit_bh */
3402                         get_bh(bh);
3403
3404                         set_buffer_uptodate(bh);
3405                         lock_buffer(bh);
3406                         bh->b_end_io = btrfs_end_buffer_write_sync;
3407                         bh->b_private = device;
3408                 }
3409
3410                 /*
3411                  * we fua the first super.  The others we allow
3412                  * to go down lazy.
3413                  */
3414                 if (i == 0)
3415                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_FUA, bh);
3416                 else
3417                         ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_SYNC, bh);
3418                 if (ret)
3419                         errors++;
3420         }
3421         return errors < i ? 0 : -1;
3422 }
3423
3424 /*
3425  * endio for the write_dev_flush, this will wake anyone waiting
3426  * for the barrier when it is done
3427  */
3428 static void btrfs_end_empty_barrier(struct bio *bio)
3429 {
3430         if (bio->bi_private)
3431                 complete(bio->bi_private);
3432         bio_put(bio);
3433 }
3434
3435 /*
3436  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3437  * sent down.  With wait == 1, it waits for the previous flush.
3438  *
3439  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3440  * capable
3441  */
3442 static int write_dev_flush(struct btrfs_device *device, int wait)
3443 {
3444         struct bio *bio;
3445         int ret = 0;
3446
3447         if (device->nobarriers)
3448                 return 0;
3449
3450         if (wait) {
3451                 bio = device->flush_bio;
3452                 if (!bio)
3453                         return 0;
3454
3455                 wait_for_completion(&device->flush_wait);
3456
3457                 if (bio->bi_error) {
3458                         ret = bio->bi_error;
3459                         btrfs_dev_stat_inc_and_print(device,
3460                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3461                 }
3462
3463                 /* drop the reference from the wait == 0 run */
3464                 bio_put(bio);
3465                 device->flush_bio = NULL;
3466
3467                 return ret;
3468         }
3469
3470         /*
3471          * one reference for us, and we leave it for the
3472          * caller
3473          */
3474         device->flush_bio = NULL;
3475         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3476         if (!bio)
3477                 return -ENOMEM;
3478
3479         bio->bi_end_io = btrfs_end_empty_barrier;
3480         bio->bi_bdev = device->bdev;
3481         bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
3482         init_completion(&device->flush_wait);
3483         bio->bi_private = &device->flush_wait;
3484         device->flush_bio = bio;
3485
3486         bio_get(bio);
3487         btrfsic_submit_bio(bio);
3488
3489         return 0;
3490 }
3491
3492 /*
3493  * send an empty flush down to each device in parallel,
3494  * then wait for them
3495  */
3496 static int barrier_all_devices(struct btrfs_fs_info *info)
3497 {
3498         struct list_head *head;
3499         struct btrfs_device *dev;
3500         int errors_send = 0;
3501         int errors_wait = 0;
3502         int ret;
3503
3504         /* send down all the barriers */
3505         head = &info->fs_devices->devices;
3506         list_for_each_entry_rcu(dev, head, dev_list) {
3507                 if (dev->missing)
3508                         continue;
3509                 if (!dev->bdev) {
3510                         errors_send++;
3511                         continue;
3512                 }
3513                 if (!dev->in_fs_metadata || !dev->writeable)
3514                         continue;
3515
3516                 ret = write_dev_flush(dev, 0);
3517                 if (ret)
3518                         errors_send++;
3519         }
3520
3521         /* wait for all the barriers */
3522         list_for_each_entry_rcu(dev, head, dev_list) {
3523                 if (dev->missing)
3524                         continue;
3525                 if (!dev->bdev) {
3526                         errors_wait++;
3527                         continue;
3528                 }
3529                 if (!dev->in_fs_metadata || !dev->writeable)
3530                         continue;
3531
3532                 ret = write_dev_flush(dev, 1);
3533                 if (ret)
3534                         errors_wait++;
3535         }
3536         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3537             errors_wait > info->num_tolerated_disk_barrier_failures)
3538                 return -EIO;
3539         return 0;
3540 }
3541
3542 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3543 {
3544         int raid_type;
3545         int min_tolerated = INT_MAX;
3546
3547         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3548             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3549                 min_tolerated = min(min_tolerated,
3550                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3551                                     tolerated_failures);
3552
3553         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3554                 if (raid_type == BTRFS_RAID_SINGLE)
3555                         continue;
3556                 if (!(flags & btrfs_raid_group[raid_type]))
3557                         continue;
3558                 min_tolerated = min(min_tolerated,
3559                                     btrfs_raid_array[raid_type].
3560                                     tolerated_failures);
3561         }
3562
3563         if (min_tolerated == INT_MAX) {
3564                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3565                 min_tolerated = 0;
3566         }
3567
3568         return min_tolerated;
3569 }
3570
3571 int btrfs_calc_num_tolerated_disk_barrier_failures(
3572         struct btrfs_fs_info *fs_info)
3573 {
3574         struct btrfs_ioctl_space_info space;
3575         struct btrfs_space_info *sinfo;
3576         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3577                        BTRFS_BLOCK_GROUP_SYSTEM,
3578                        BTRFS_BLOCK_GROUP_METADATA,
3579                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3580         int i;
3581         int c;
3582         int num_tolerated_disk_barrier_failures =
3583                 (int)fs_info->fs_devices->num_devices;
3584
3585         for (i = 0; i < ARRAY_SIZE(types); i++) {
3586                 struct btrfs_space_info *tmp;
3587
3588                 sinfo = NULL;
3589                 rcu_read_lock();
3590                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3591                         if (tmp->flags == types[i]) {
3592                                 sinfo = tmp;
3593                                 break;
3594                         }
3595                 }
3596                 rcu_read_unlock();
3597
3598                 if (!sinfo)
3599                         continue;
3600
3601                 down_read(&sinfo->groups_sem);
3602                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3603                         u64 flags;
3604
3605                         if (list_empty(&sinfo->block_groups[c]))
3606                                 continue;
3607
3608                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3609                                                    &space);
3610                         if (space.total_bytes == 0 || space.used_bytes == 0)
3611                                 continue;
3612                         flags = space.flags;
3613
3614                         num_tolerated_disk_barrier_failures = min(
3615                                 num_tolerated_disk_barrier_failures,
3616                                 btrfs_get_num_tolerated_disk_barrier_failures(
3617                                         flags));
3618                 }
3619                 up_read(&sinfo->groups_sem);
3620         }
3621
3622         return num_tolerated_disk_barrier_failures;
3623 }
3624
3625 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3626 {
3627         struct list_head *head;
3628         struct btrfs_device *dev;
3629         struct btrfs_super_block *sb;
3630         struct btrfs_dev_item *dev_item;
3631         int ret;
3632         int do_barriers;
3633         int max_errors;
3634         int total_errors = 0;
3635         u64 flags;
3636
3637         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3638         backup_super_roots(root->fs_info);
3639
3640         sb = root->fs_info->super_for_commit;
3641         dev_item = &sb->dev_item;
3642
3643         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3644         head = &root->fs_info->fs_devices->devices;
3645         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3646
3647         if (do_barriers) {
3648                 ret = barrier_all_devices(root->fs_info);
3649                 if (ret) {
3650                         mutex_unlock(
3651                                 &root->fs_info->fs_devices->device_list_mutex);
3652                         btrfs_handle_fs_error(root->fs_info, ret,
3653                                     "errors while submitting device barriers.");
3654                         return ret;
3655                 }
3656         }
3657
3658         list_for_each_entry_rcu(dev, head, dev_list) {
3659                 if (!dev->bdev) {
3660                         total_errors++;
3661                         continue;
3662                 }
3663                 if (!dev->in_fs_metadata || !dev->writeable)
3664                         continue;
3665
3666                 btrfs_set_stack_device_generation(dev_item, 0);
3667                 btrfs_set_stack_device_type(dev_item, dev->type);
3668                 btrfs_set_stack_device_id(dev_item, dev->devid);
3669                 btrfs_set_stack_device_total_bytes(dev_item,
3670                                                    dev->commit_total_bytes);
3671                 btrfs_set_stack_device_bytes_used(dev_item,
3672                                                   dev->commit_bytes_used);
3673                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3674                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3675                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3676                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3677                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3678
3679                 flags = btrfs_super_flags(sb);
3680                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3681
3682                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3683                 if (ret)
3684                         total_errors++;
3685         }
3686         if (total_errors > max_errors) {
3687                 btrfs_err(root->fs_info, "%d errors while writing supers",
3688                        total_errors);
3689                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3690
3691                 /* FUA is masked off if unsupported and can't be the reason */
3692                 btrfs_handle_fs_error(root->fs_info, -EIO,
3693                             "%d errors while writing supers", total_errors);
3694                 return -EIO;
3695         }
3696
3697         total_errors = 0;
3698         list_for_each_entry_rcu(dev, head, dev_list) {
3699                 if (!dev->bdev)
3700                         continue;
3701                 if (!dev->in_fs_metadata || !dev->writeable)
3702                         continue;
3703
3704                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3705                 if (ret)
3706                         total_errors++;
3707         }
3708         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3709         if (total_errors > max_errors) {
3710                 btrfs_handle_fs_error(root->fs_info, -EIO,
3711                             "%d errors while writing supers", total_errors);
3712                 return -EIO;
3713         }
3714         return 0;
3715 }
3716
3717 int write_ctree_super(struct btrfs_trans_handle *trans,
3718                       struct btrfs_root *root, int max_mirrors)
3719 {
3720         return write_all_supers(root, max_mirrors);
3721 }
3722
3723 /* Drop a fs root from the radix tree and free it. */
3724 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3725                                   struct btrfs_root *root)
3726 {
3727         spin_lock(&fs_info->fs_roots_radix_lock);
3728         radix_tree_delete(&fs_info->fs_roots_radix,
3729                           (unsigned long)root->root_key.objectid);
3730         spin_unlock(&fs_info->fs_roots_radix_lock);
3731
3732         if (btrfs_root_refs(&root->root_item) == 0)
3733                 synchronize_srcu(&fs_info->subvol_srcu);
3734
3735         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3736                 btrfs_free_log(NULL, root);
3737
3738         if (root->free_ino_pinned)
3739                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3740         if (root->free_ino_ctl)
3741                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3742         free_fs_root(root);
3743 }
3744
3745 static void free_fs_root(struct btrfs_root *root)
3746 {
3747         iput(root->ino_cache_inode);
3748         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3749         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3750         root->orphan_block_rsv = NULL;
3751         if (root->anon_dev)
3752                 free_anon_bdev(root->anon_dev);
3753         if (root->subv_writers)
3754                 btrfs_free_subvolume_writers(root->subv_writers);
3755         free_extent_buffer(root->node);
3756         free_extent_buffer(root->commit_root);
3757         kfree(root->free_ino_ctl);
3758         kfree(root->free_ino_pinned);
3759         kfree(root->name);
3760         btrfs_put_fs_root(root);
3761 }
3762
3763 void btrfs_free_fs_root(struct btrfs_root *root)
3764 {
3765         free_fs_root(root);
3766 }
3767
3768 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3769 {
3770         u64 root_objectid = 0;
3771         struct btrfs_root *gang[8];
3772         int i = 0;
3773         int err = 0;
3774         unsigned int ret = 0;
3775         int index;
3776
3777         while (1) {
3778                 index = srcu_read_lock(&fs_info->subvol_srcu);
3779                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3780                                              (void **)gang, root_objectid,
3781                                              ARRAY_SIZE(gang));
3782                 if (!ret) {
3783                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3784                         break;
3785                 }
3786                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3787
3788                 for (i = 0; i < ret; i++) {
3789                         /* Avoid to grab roots in dead_roots */
3790                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3791                                 gang[i] = NULL;
3792                                 continue;
3793                         }
3794                         /* grab all the search result for later use */
3795                         gang[i] = btrfs_grab_fs_root(gang[i]);
3796                 }
3797                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3798
3799                 for (i = 0; i < ret; i++) {
3800                         if (!gang[i])
3801                                 continue;
3802                         root_objectid = gang[i]->root_key.objectid;
3803                         err = btrfs_orphan_cleanup(gang[i]);
3804                         if (err)
3805                                 break;
3806                         btrfs_put_fs_root(gang[i]);
3807                 }
3808                 root_objectid++;
3809         }
3810
3811         /* release the uncleaned roots due to error */
3812         for (; i < ret; i++) {
3813                 if (gang[i])
3814                         btrfs_put_fs_root(gang[i]);
3815         }
3816         return err;
3817 }
3818
3819 int btrfs_commit_super(struct btrfs_root *root)
3820 {
3821         struct btrfs_trans_handle *trans;
3822
3823         mutex_lock(&root->fs_info->cleaner_mutex);
3824         btrfs_run_delayed_iputs(root);
3825         mutex_unlock(&root->fs_info->cleaner_mutex);
3826         wake_up_process(root->fs_info->cleaner_kthread);
3827
3828         /* wait until ongoing cleanup work done */
3829         down_write(&root->fs_info->cleanup_work_sem);
3830         up_write(&root->fs_info->cleanup_work_sem);
3831
3832         trans = btrfs_join_transaction(root);
3833         if (IS_ERR(trans))
3834                 return PTR_ERR(trans);
3835         return btrfs_commit_transaction(trans, root);
3836 }
3837
3838 void close_ctree(struct btrfs_root *root)
3839 {
3840         struct btrfs_fs_info *fs_info = root->fs_info;
3841         int ret;
3842
3843         fs_info->closing = 1;
3844         smp_mb();
3845
3846         /* wait for the qgroup rescan worker to stop */
3847         btrfs_qgroup_wait_for_completion(fs_info);
3848
3849         /* wait for the uuid_scan task to finish */
3850         down(&fs_info->uuid_tree_rescan_sem);
3851         /* avoid complains from lockdep et al., set sem back to initial state */
3852         up(&fs_info->uuid_tree_rescan_sem);
3853
3854         /* pause restriper - we want to resume on mount */
3855         btrfs_pause_balance(fs_info);
3856
3857         btrfs_dev_replace_suspend_for_unmount(fs_info);
3858
3859         btrfs_scrub_cancel(fs_info);
3860
3861         /* wait for any defraggers to finish */
3862         wait_event(fs_info->transaction_wait,
3863                    (atomic_read(&fs_info->defrag_running) == 0));
3864
3865         /* clear out the rbtree of defraggable inodes */
3866         btrfs_cleanup_defrag_inodes(fs_info);
3867
3868         cancel_work_sync(&fs_info->async_reclaim_work);
3869
3870         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3871                 /*
3872                  * If the cleaner thread is stopped and there are
3873                  * block groups queued for removal, the deletion will be
3874                  * skipped when we quit the cleaner thread.
3875                  */
3876                 btrfs_delete_unused_bgs(root->fs_info);
3877
3878                 ret = btrfs_commit_super(root);
3879                 if (ret)
3880                         btrfs_err(fs_info, "commit super ret %d", ret);
3881         }
3882
3883         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3884                 btrfs_error_commit_super(root);
3885
3886         kthread_stop(fs_info->transaction_kthread);
3887         kthread_stop(fs_info->cleaner_kthread);
3888
3889         fs_info->closing = 2;
3890         smp_mb();
3891
3892         btrfs_free_qgroup_config(fs_info);
3893
3894         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3895                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3896                        percpu_counter_sum(&fs_info->delalloc_bytes));
3897         }
3898
3899         btrfs_sysfs_remove_mounted(fs_info);
3900         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3901
3902         btrfs_free_fs_roots(fs_info);
3903
3904         btrfs_put_block_group_cache(fs_info);
3905
3906         btrfs_free_block_groups(fs_info);
3907
3908         /*
3909          * we must make sure there is not any read request to
3910          * submit after we stopping all workers.
3911          */
3912         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3913         btrfs_stop_all_workers(fs_info);
3914
3915         fs_info->open = 0;
3916         free_root_pointers(fs_info, 1);
3917
3918         iput(fs_info->btree_inode);
3919
3920 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3921         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3922                 btrfsic_unmount(root, fs_info->fs_devices);
3923 #endif
3924
3925         btrfs_close_devices(fs_info->fs_devices);
3926         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3927
3928         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3929         percpu_counter_destroy(&fs_info->delalloc_bytes);
3930         percpu_counter_destroy(&fs_info->bio_counter);
3931         bdi_destroy(&fs_info->bdi);
3932         cleanup_srcu_struct(&fs_info->subvol_srcu);
3933
3934         btrfs_free_stripe_hash_table(fs_info);
3935
3936         __btrfs_free_block_rsv(root->orphan_block_rsv);
3937         root->orphan_block_rsv = NULL;
3938
3939         lock_chunks(root);
3940         while (!list_empty(&fs_info->pinned_chunks)) {
3941                 struct extent_map *em;
3942
3943                 em = list_first_entry(&fs_info->pinned_chunks,
3944                                       struct extent_map, list);
3945                 list_del_init(&em->list);
3946                 free_extent_map(em);
3947         }
3948         unlock_chunks(root);
3949 }
3950
3951 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3952                           int atomic)
3953 {
3954         int ret;
3955         struct inode *btree_inode = buf->pages[0]->mapping->host;
3956
3957         ret = extent_buffer_uptodate(buf);
3958         if (!ret)
3959                 return ret;
3960
3961         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3962                                     parent_transid, atomic);
3963         if (ret == -EAGAIN)
3964                 return ret;
3965         return !ret;
3966 }
3967
3968 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3969 {
3970         struct btrfs_root *root;
3971         u64 transid = btrfs_header_generation(buf);
3972         int was_dirty;
3973
3974 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3975         /*
3976          * This is a fast path so only do this check if we have sanity tests
3977          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3978          * outside of the sanity tests.
3979          */
3980         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3981                 return;
3982 #endif
3983         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3984         btrfs_assert_tree_locked(buf);
3985         if (transid != root->fs_info->generation)
3986                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3987                        "found %llu running %llu\n",
3988                         buf->start, transid, root->fs_info->generation);
3989         was_dirty = set_extent_buffer_dirty(buf);
3990         if (!was_dirty)
3991                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3992                                      buf->len,
3993                                      root->fs_info->dirty_metadata_batch);
3994 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3995         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3996                 btrfs_print_leaf(root, buf);
3997                 ASSERT(0);
3998         }
3999 #endif
4000 }
4001
4002 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4003                                         int flush_delayed)
4004 {
4005         /*
4006          * looks as though older kernels can get into trouble with
4007          * this code, they end up stuck in balance_dirty_pages forever
4008          */
4009         int ret;
4010
4011         if (current->flags & PF_MEMALLOC)
4012                 return;
4013
4014         if (flush_delayed)
4015                 btrfs_balance_delayed_items(root);
4016
4017         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4018                                      BTRFS_DIRTY_METADATA_THRESH);
4019         if (ret > 0) {
4020                 balance_dirty_pages_ratelimited(
4021                                    root->fs_info->btree_inode->i_mapping);
4022         }
4023 }
4024
4025 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4026 {
4027         __btrfs_btree_balance_dirty(root, 1);
4028 }
4029
4030 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4031 {
4032         __btrfs_btree_balance_dirty(root, 0);
4033 }
4034
4035 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4036 {
4037         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4038         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4039 }
4040
4041 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4042                               int read_only)
4043 {
4044         struct btrfs_super_block *sb = fs_info->super_copy;
4045         u64 nodesize = btrfs_super_nodesize(sb);
4046         u64 sectorsize = btrfs_super_sectorsize(sb);
4047         int ret = 0;
4048
4049         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4050                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4051                 ret = -EINVAL;
4052         }
4053         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4054                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4055                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4056         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4057                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4058                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4059                 ret = -EINVAL;
4060         }
4061         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4062                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4063                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4064                 ret = -EINVAL;
4065         }
4066         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4067                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4068                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4069                 ret = -EINVAL;
4070         }
4071
4072         /*
4073          * Check sectorsize and nodesize first, other check will need it.
4074          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4075          */
4076         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4077             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4078                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4079                 ret = -EINVAL;
4080         }
4081         /* Only PAGE SIZE is supported yet */
4082         if (sectorsize != PAGE_SIZE) {
4083                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4084                                 sectorsize, PAGE_SIZE);
4085                 ret = -EINVAL;
4086         }
4087         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4088             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4089                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4090                 ret = -EINVAL;
4091         }
4092         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4093                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4094                                 le32_to_cpu(sb->__unused_leafsize),
4095                                 nodesize);
4096                 ret = -EINVAL;
4097         }
4098
4099         /* Root alignment check */
4100         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4101                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4102                                 btrfs_super_root(sb));
4103                 ret = -EINVAL;
4104         }
4105         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4106                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4107                                 btrfs_super_chunk_root(sb));
4108                 ret = -EINVAL;
4109         }
4110         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4111                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4112                                 btrfs_super_log_root(sb));
4113                 ret = -EINVAL;
4114         }
4115
4116         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4117                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4118                                 fs_info->fsid, sb->dev_item.fsid);
4119                 ret = -EINVAL;
4120         }
4121
4122         /*
4123          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4124          * done later
4125          */
4126         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4127                 btrfs_err(fs_info, "bytes_used is too small %llu",
4128                        btrfs_super_bytes_used(sb));
4129                 ret = -EINVAL;
4130         }
4131         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4132                 btrfs_err(fs_info, "invalid stripesize %u",
4133                        btrfs_super_stripesize(sb));
4134                 ret = -EINVAL;
4135         }
4136         if (btrfs_super_num_devices(sb) > (1UL << 31))
4137                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4138                                 btrfs_super_num_devices(sb));
4139         if (btrfs_super_num_devices(sb) == 0) {
4140                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4141                 ret = -EINVAL;
4142         }
4143
4144         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4145                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4146                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4147                 ret = -EINVAL;
4148         }
4149
4150         /*
4151          * Obvious sys_chunk_array corruptions, it must hold at least one key
4152          * and one chunk
4153          */
4154         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4155                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4156                                 btrfs_super_sys_array_size(sb),
4157                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4158                 ret = -EINVAL;
4159         }
4160         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4161                         + sizeof(struct btrfs_chunk)) {
4162                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4163                                 btrfs_super_sys_array_size(sb),
4164                                 sizeof(struct btrfs_disk_key)
4165                                 + sizeof(struct btrfs_chunk));
4166                 ret = -EINVAL;
4167         }
4168
4169         /*
4170          * The generation is a global counter, we'll trust it more than the others
4171          * but it's still possible that it's the one that's wrong.
4172          */
4173         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4174                 printk(KERN_WARNING
4175                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4176                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4177         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4178             && btrfs_super_cache_generation(sb) != (u64)-1)
4179                 printk(KERN_WARNING
4180                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4181                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4182
4183         return ret;
4184 }
4185
4186 static void btrfs_error_commit_super(struct btrfs_root *root)
4187 {
4188         mutex_lock(&root->fs_info->cleaner_mutex);
4189         btrfs_run_delayed_iputs(root);
4190         mutex_unlock(&root->fs_info->cleaner_mutex);
4191
4192         down_write(&root->fs_info->cleanup_work_sem);
4193         up_write(&root->fs_info->cleanup_work_sem);
4194
4195         /* cleanup FS via transaction */
4196         btrfs_cleanup_transaction(root);
4197 }
4198
4199 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4200 {
4201         struct btrfs_ordered_extent *ordered;
4202
4203         spin_lock(&root->ordered_extent_lock);
4204         /*
4205          * This will just short circuit the ordered completion stuff which will
4206          * make sure the ordered extent gets properly cleaned up.
4207          */
4208         list_for_each_entry(ordered, &root->ordered_extents,
4209                             root_extent_list)
4210                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4211         spin_unlock(&root->ordered_extent_lock);
4212 }
4213
4214 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4215 {
4216         struct btrfs_root *root;
4217         struct list_head splice;
4218
4219         INIT_LIST_HEAD(&splice);
4220
4221         spin_lock(&fs_info->ordered_root_lock);
4222         list_splice_init(&fs_info->ordered_roots, &splice);
4223         while (!list_empty(&splice)) {
4224                 root = list_first_entry(&splice, struct btrfs_root,
4225                                         ordered_root);
4226                 list_move_tail(&root->ordered_root,
4227                                &fs_info->ordered_roots);
4228
4229                 spin_unlock(&fs_info->ordered_root_lock);
4230                 btrfs_destroy_ordered_extents(root);
4231
4232                 cond_resched();
4233                 spin_lock(&fs_info->ordered_root_lock);
4234         }
4235         spin_unlock(&fs_info->ordered_root_lock);
4236 }
4237
4238 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4239                                       struct btrfs_root *root)
4240 {
4241         struct rb_node *node;
4242         struct btrfs_delayed_ref_root *delayed_refs;
4243         struct btrfs_delayed_ref_node *ref;
4244         int ret = 0;
4245
4246         delayed_refs = &trans->delayed_refs;
4247
4248         spin_lock(&delayed_refs->lock);
4249         if (atomic_read(&delayed_refs->num_entries) == 0) {
4250                 spin_unlock(&delayed_refs->lock);
4251                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4252                 return ret;
4253         }
4254
4255         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4256                 struct btrfs_delayed_ref_head *head;
4257                 struct btrfs_delayed_ref_node *tmp;
4258                 bool pin_bytes = false;
4259
4260                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4261                                 href_node);
4262                 if (!mutex_trylock(&head->mutex)) {
4263                         atomic_inc(&head->node.refs);
4264                         spin_unlock(&delayed_refs->lock);
4265
4266                         mutex_lock(&head->mutex);
4267                         mutex_unlock(&head->mutex);
4268                         btrfs_put_delayed_ref(&head->node);
4269                         spin_lock(&delayed_refs->lock);
4270                         continue;
4271                 }
4272                 spin_lock(&head->lock);
4273                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4274                                                  list) {
4275                         ref->in_tree = 0;
4276                         list_del(&ref->list);
4277                         atomic_dec(&delayed_refs->num_entries);
4278                         btrfs_put_delayed_ref(ref);
4279                 }
4280                 if (head->must_insert_reserved)
4281                         pin_bytes = true;
4282                 btrfs_free_delayed_extent_op(head->extent_op);
4283                 delayed_refs->num_heads--;
4284                 if (head->processing == 0)
4285                         delayed_refs->num_heads_ready--;
4286                 atomic_dec(&delayed_refs->num_entries);
4287                 head->node.in_tree = 0;
4288                 rb_erase(&head->href_node, &delayed_refs->href_root);
4289                 spin_unlock(&head->lock);
4290                 spin_unlock(&delayed_refs->lock);
4291                 mutex_unlock(&head->mutex);
4292
4293                 if (pin_bytes)
4294                         btrfs_pin_extent(root, head->node.bytenr,
4295                                          head->node.num_bytes, 1);
4296                 btrfs_put_delayed_ref(&head->node);
4297                 cond_resched();
4298                 spin_lock(&delayed_refs->lock);
4299         }
4300
4301         spin_unlock(&delayed_refs->lock);
4302
4303         return ret;
4304 }
4305
4306 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4307 {
4308         struct btrfs_inode *btrfs_inode;
4309         struct list_head splice;
4310
4311         INIT_LIST_HEAD(&splice);
4312
4313         spin_lock(&root->delalloc_lock);
4314         list_splice_init(&root->delalloc_inodes, &splice);
4315
4316         while (!list_empty(&splice)) {
4317                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4318                                                delalloc_inodes);
4319
4320                 list_del_init(&btrfs_inode->delalloc_inodes);
4321                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4322                           &btrfs_inode->runtime_flags);
4323                 spin_unlock(&root->delalloc_lock);
4324
4325                 btrfs_invalidate_inodes(btrfs_inode->root);
4326
4327                 spin_lock(&root->delalloc_lock);
4328         }
4329
4330         spin_unlock(&root->delalloc_lock);
4331 }
4332
4333 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4334 {
4335         struct btrfs_root *root;
4336         struct list_head splice;
4337
4338         INIT_LIST_HEAD(&splice);
4339
4340         spin_lock(&fs_info->delalloc_root_lock);
4341         list_splice_init(&fs_info->delalloc_roots, &splice);
4342         while (!list_empty(&splice)) {
4343                 root = list_first_entry(&splice, struct btrfs_root,
4344                                          delalloc_root);
4345                 list_del_init(&root->delalloc_root);
4346                 root = btrfs_grab_fs_root(root);
4347                 BUG_ON(!root);
4348                 spin_unlock(&fs_info->delalloc_root_lock);
4349
4350                 btrfs_destroy_delalloc_inodes(root);
4351                 btrfs_put_fs_root(root);
4352
4353                 spin_lock(&fs_info->delalloc_root_lock);
4354         }
4355         spin_unlock(&fs_info->delalloc_root_lock);
4356 }
4357
4358 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4359                                         struct extent_io_tree *dirty_pages,
4360                                         int mark)
4361 {
4362         int ret;
4363         struct extent_buffer *eb;
4364         u64 start = 0;
4365         u64 end;
4366
4367         while (1) {
4368                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4369                                             mark, NULL);
4370                 if (ret)
4371                         break;
4372
4373                 clear_extent_bits(dirty_pages, start, end, mark);
4374                 while (start <= end) {
4375                         eb = btrfs_find_tree_block(root->fs_info, start);
4376                         start += root->nodesize;
4377                         if (!eb)
4378                                 continue;
4379                         wait_on_extent_buffer_writeback(eb);
4380
4381                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4382                                                &eb->bflags))
4383                                 clear_extent_buffer_dirty(eb);
4384                         free_extent_buffer_stale(eb);
4385                 }
4386         }
4387
4388         return ret;
4389 }
4390
4391 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4392                                        struct extent_io_tree *pinned_extents)
4393 {
4394         struct extent_io_tree *unpin;
4395         u64 start;
4396         u64 end;
4397         int ret;
4398         bool loop = true;
4399
4400         unpin = pinned_extents;
4401 again:
4402         while (1) {
4403                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4404                                             EXTENT_DIRTY, NULL);
4405                 if (ret)
4406                         break;
4407
4408                 clear_extent_dirty(unpin, start, end);
4409                 btrfs_error_unpin_extent_range(root, start, end);
4410                 cond_resched();
4411         }
4412
4413         if (loop) {
4414                 if (unpin == &root->fs_info->freed_extents[0])
4415                         unpin = &root->fs_info->freed_extents[1];
4416                 else
4417                         unpin = &root->fs_info->freed_extents[0];
4418                 loop = false;
4419                 goto again;
4420         }
4421
4422         return 0;
4423 }
4424
4425 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4426                                    struct btrfs_root *root)
4427 {
4428         btrfs_destroy_delayed_refs(cur_trans, root);
4429
4430         cur_trans->state = TRANS_STATE_COMMIT_START;
4431         wake_up(&root->fs_info->transaction_blocked_wait);
4432
4433         cur_trans->state = TRANS_STATE_UNBLOCKED;
4434         wake_up(&root->fs_info->transaction_wait);
4435
4436         btrfs_destroy_delayed_inodes(root);
4437         btrfs_assert_delayed_root_empty(root);
4438
4439         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4440                                      EXTENT_DIRTY);
4441         btrfs_destroy_pinned_extent(root,
4442                                     root->fs_info->pinned_extents);
4443
4444         cur_trans->state =TRANS_STATE_COMPLETED;
4445         wake_up(&cur_trans->commit_wait);
4446
4447         /*
4448         memset(cur_trans, 0, sizeof(*cur_trans));
4449         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4450         */
4451 }
4452
4453 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4454 {
4455         struct btrfs_transaction *t;
4456
4457         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4458
4459         spin_lock(&root->fs_info->trans_lock);
4460         while (!list_empty(&root->fs_info->trans_list)) {
4461                 t = list_first_entry(&root->fs_info->trans_list,
4462                                      struct btrfs_transaction, list);
4463                 if (t->state >= TRANS_STATE_COMMIT_START) {
4464                         atomic_inc(&t->use_count);
4465                         spin_unlock(&root->fs_info->trans_lock);
4466                         btrfs_wait_for_commit(root, t->transid);
4467                         btrfs_put_transaction(t);
4468                         spin_lock(&root->fs_info->trans_lock);
4469                         continue;
4470                 }
4471                 if (t == root->fs_info->running_transaction) {
4472                         t->state = TRANS_STATE_COMMIT_DOING;
4473                         spin_unlock(&root->fs_info->trans_lock);
4474                         /*
4475                          * We wait for 0 num_writers since we don't hold a trans
4476                          * handle open currently for this transaction.
4477                          */
4478                         wait_event(t->writer_wait,
4479                                    atomic_read(&t->num_writers) == 0);
4480                 } else {
4481                         spin_unlock(&root->fs_info->trans_lock);
4482                 }
4483                 btrfs_cleanup_one_transaction(t, root);
4484
4485                 spin_lock(&root->fs_info->trans_lock);
4486                 if (t == root->fs_info->running_transaction)
4487                         root->fs_info->running_transaction = NULL;
4488                 list_del_init(&t->list);
4489                 spin_unlock(&root->fs_info->trans_lock);
4490
4491                 btrfs_put_transaction(t);
4492                 trace_btrfs_transaction_commit(root);
4493                 spin_lock(&root->fs_info->trans_lock);
4494         }
4495         spin_unlock(&root->fs_info->trans_lock);
4496         btrfs_destroy_all_ordered_extents(root->fs_info);
4497         btrfs_destroy_delayed_inodes(root);
4498         btrfs_assert_delayed_root_empty(root);
4499         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4500         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4501         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4502
4503         return 0;
4504 }
4505
4506 static const struct extent_io_ops btree_extent_io_ops = {
4507         .readpage_end_io_hook = btree_readpage_end_io_hook,
4508         .readpage_io_failed_hook = btree_io_failed_hook,
4509         .submit_bio_hook = btree_submit_bio_hook,
4510         /* note we're sharing with inode.c for the merge bio hook */
4511         .merge_bio_hook = btrfs_merge_bio_hook,
4512 };